2019-05-28 17:10:09 +00:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0-only */
|
2016-09-21 10:43:57 +00:00
|
|
|
/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
|
|
|
|
|
*/
|
|
|
|
|
#ifndef _LINUX_BPF_VERIFIER_H
|
|
|
|
|
#define _LINUX_BPF_VERIFIER_H 1
|
|
|
|
|
|
|
|
|
|
#include <linux/bpf.h> /* for enum bpf_reg_type */
|
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.
This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.
Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.
Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
2020-12-03 20:46:29 +00:00
|
|
|
#include <linux/btf.h> /* for struct btf and btf_id() */
|
2016-09-21 10:43:57 +00:00
|
|
|
#include <linux/filter.h> /* for MAX_BPF_STACK */
|
2017-08-07 14:26:19 +00:00
|
|
|
#include <linux/tnum.h>
|
2016-09-21 10:43:57 +00:00
|
|
|
|
2017-08-07 14:26:36 +00:00
|
|
|
/* Maximum variable offset umax_value permitted when resolving memory accesses.
|
|
|
|
|
* In practice this is far bigger than any realistic pointer offset; this limit
|
|
|
|
|
* ensures that umax_value + (int)off + (int)size cannot overflow a u64.
|
|
|
|
|
*/
|
2017-12-19 04:12:00 +00:00
|
|
|
#define BPF_MAX_VAR_OFF (1 << 29)
|
2017-08-07 14:26:36 +00:00
|
|
|
/* Maximum variable size permitted for ARG_CONST_SIZE[_OR_ZERO]. This ensures
|
|
|
|
|
* that converting umax_value to int cannot overflow.
|
|
|
|
|
*/
|
2017-12-19 04:12:00 +00:00
|
|
|
#define BPF_MAX_VAR_SIZ (1 << 29)
|
2023-05-05 04:33:11 +00:00
|
|
|
/* size of tmp_str_buf in bpf_verifier.
|
|
|
|
|
* we need at least 306 bytes to fit full stack mask representation
|
|
|
|
|
* (in the "-8,-16,...,-512" form)
|
|
|
|
|
*/
|
|
|
|
|
#define TMP_STR_BUF_LEN 320
|
2024-08-29 21:08:23 +00:00
|
|
|
/* Patch buffer size */
|
2024-09-04 18:08:44 +00:00
|
|
|
#define INSN_BUF_SIZE 32
|
2016-09-28 14:54:32 +00:00
|
|
|
|
2017-08-23 14:11:21 +00:00
|
|
|
/* Liveness marks, used for registers and spilled-regs (in stack slots).
|
|
|
|
|
* Read marks propagate upwards until they find a write mark; they record that
|
|
|
|
|
* "one of this state's descendants read this reg" (and therefore the reg is
|
|
|
|
|
* relevant for states_equal() checks).
|
|
|
|
|
* Write marks collect downwards and do not propagate; they record that "the
|
|
|
|
|
* straight-line code that reached this state (from its parent) wrote this reg"
|
|
|
|
|
* (and therefore that reads propagated from this state or its descendants
|
|
|
|
|
* should not propagate to its parent).
|
|
|
|
|
* A state with a write mark can receive read marks; it just won't propagate
|
|
|
|
|
* them to its parent, since the write mark is a property, not of the state,
|
|
|
|
|
* but of the link between it and its parent. See mark_reg_read() and
|
|
|
|
|
* mark_stack_slot_read() in kernel/bpf/verifier.c.
|
|
|
|
|
*/
|
2017-08-15 19:34:35 +00:00
|
|
|
enum bpf_reg_liveness {
|
|
|
|
|
REG_LIVE_NONE = 0, /* reg hasn't been read or written this branch */
|
bpf: verifier: mark verified-insn with sub-register zext flag
eBPF ISA specification requires high 32-bit cleared when low 32-bit
sub-register is written. This applies to destination register of ALU32 etc.
JIT back-ends must guarantee this semantic when doing code-gen. x86_64 and
AArch64 ISA has the same semantics, so the corresponding JIT back-end
doesn't need to do extra work.
However, 32-bit arches (arm, x86, nfp etc.) and some other 64-bit arches
(PowerPC, SPARC etc) need to do explicit zero extension to meet this
requirement, otherwise code like the following will fail.
u64_value = (u64) u32_value
... other uses of u64_value
This is because compiler could exploit the semantic described above and
save those zero extensions for extending u32_value to u64_value, these JIT
back-ends are expected to guarantee this through inserting extra zero
extensions which however could be a significant increase on the code size.
Some benchmarks show there could be ~40% sub-register writes out of total
insns, meaning at least ~40% extra code-gen.
One observation is these extra zero extensions are not always necessary.
Take above code snippet for example, it is possible u32_value will never be
casted into a u64, the value of high 32-bit of u32_value then could be
ignored and extra zero extension could be eliminated.
This patch implements this idea, insns defining sub-registers will be
marked when the high 32-bit of the defined sub-register matters. For
those unmarked insns, it is safe to eliminate high 32-bit clearnace for
them.
Algo:
- Split read flags into READ32 and READ64.
- Record index of insn that does sub-register write. Keep the index inside
reg state and update it during verifier insn walking.
- A full register read on a sub-register marks its definition insn as
needing zero extension on dst register.
A new sub-register write overrides the old one.
- When propagating read64 during path pruning, also mark any insn defining
a sub-register that is read in the pruned path as full-register.
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-05-24 22:25:12 +00:00
|
|
|
REG_LIVE_READ32 = 0x1, /* reg was read, so we're sensitive to initial value */
|
|
|
|
|
REG_LIVE_READ64 = 0x2, /* likewise, but full 64-bit content matters */
|
|
|
|
|
REG_LIVE_READ = REG_LIVE_READ32 | REG_LIVE_READ64,
|
|
|
|
|
REG_LIVE_WRITTEN = 0x4, /* reg was written first, screening off later reads */
|
|
|
|
|
REG_LIVE_DONE = 0x8, /* liveness won't be updating this register anymore */
|
2017-08-15 19:34:35 +00:00
|
|
|
};
|
|
|
|
|
|
bpf: add iterator kfuncs registration and validation logic
Add ability to register kfuncs that implement BPF open-coded iterator
contract and enforce naming and function proto convention. Enforcement
happens at the time of kfunc registration and significantly simplifies
the rest of iterators logic in the verifier.
More details follow in subsequent patches, but we enforce the following
conditions.
All kfuncs (constructor, next, destructor) have to be named consistenly
as bpf_iter_<type>_{new,next,destroy}(), respectively. <type> represents
iterator type, and iterator state should be represented as a matching
`struct bpf_iter_<type>` state type. Also, all iter kfuncs should have
a pointer to this `struct bpf_iter_<type>` as the very first argument.
Additionally:
- Constructor, i.e., bpf_iter_<type>_new(), can have arbitrary extra
number of arguments. Return type is not enforced either.
- Next method, i.e., bpf_iter_<type>_next(), has to return a pointer
type and should have exactly one argument: `struct bpf_iter_<type> *`
(const/volatile/restrict and typedefs are ignored).
- Destructor, i.e., bpf_iter_<type>_destroy(), should return void and
should have exactly one argument, similar to the next method.
- struct bpf_iter_<type> size is enforced to be positive and
a multiple of 8 bytes (to fit stack slots correctly).
Such strictness and consistency allows to build generic helpers
abstracting important, but boilerplate, details to be able to use
open-coded iterators effectively and ergonomically (see bpf_for_each()
in subsequent patches). It also simplifies the verifier logic in some
places. At the same time, this doesn't hurt generality of possible
iterator implementations. Win-win.
Constructor kfunc is marked with a new KF_ITER_NEW flags, next method is
marked with KF_ITER_NEXT (and should also have KF_RET_NULL, of course),
while destructor kfunc is marked as KF_ITER_DESTROY.
Additionally, we add a trivial kfunc name validation: it should be
a valid non-NULL and non-empty string.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230308184121.1165081-3-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 18:41:15 +00:00
|
|
|
#define ITER_PREFIX "bpf_iter_"
|
|
|
|
|
|
bpf: add support for open-coded iterator loops
Teach verifier about the concept of the open-coded (or inline) iterators.
This patch adds generic iterator loop verification logic, new STACK_ITER
stack slot type to contain iterator state, and necessary kfunc plumbing
for iterator's constructor, destructor and next methods. Next patch
implements first specific iterator (numbers iterator for implementing
for() loop logic). Such split allows to have more focused commits for
verifier logic and separate commit that we could point later to
demonstrating what does it take to add a new kind of iterator.
Each kind of iterator has its own associated struct bpf_iter_<type>,
where <type> denotes a specific type of iterator. struct bpf_iter_<type>
state is supposed to live on BPF program stack, so there will be no way
to change its size later on without breaking backwards compatibility, so
choose wisely! But given this struct is specific to a given <type> of
iterator, this allows a lot of flexibility: simple iterators could be
fine with just one stack slot (8 bytes), like numbers iterator in the
next patch, while some other more complicated iterators might need way
more to keep their iterator state. Either way, such design allows to
avoid runtime memory allocations, which otherwise would be necessary if
we fixed on-the-stack size and it turned out to be too small for a given
iterator implementation.
The way BPF verifier logic is implemented, there are no artificial
restrictions on a number of active iterators, it should work correctly
using multiple active iterators at the same time. This also means you
can have multiple nested iteration loops. struct bpf_iter_<type>
reference can be safely passed to subprograms as well.
General flow is easiest to demonstrate with a simple example using
number iterator implemented in next patch. Here's the simplest possible
loop:
struct bpf_iter_num it;
int *v;
bpf_iter_num_new(&it, 2, 5);
while ((v = bpf_iter_num_next(&it))) {
bpf_printk("X = %d", *v);
}
bpf_iter_num_destroy(&it);
Above snippet should output "X = 2", "X = 3", "X = 4". Note that 5 is
exclusive and is not returned. This matches similar APIs (e.g., slices
in Go or Rust) that implement a range of elements, where end index is
non-inclusive.
In the above example, we see a trio of function:
- constructor, bpf_iter_num_new(), which initializes iterator state
(struct bpf_iter_num it) on the stack. If any of the input arguments
are invalid, constructor should make sure to still initialize it such
that subsequent bpf_iter_num_next() calls will return NULL. I.e., on
error, return error and construct empty iterator.
- next method, bpf_iter_num_next(), which accepts pointer to iterator
state and produces an element. Next method should always return
a pointer. The contract between BPF verifier is that next method will
always eventually return NULL when elements are exhausted. Once NULL is
returned, subsequent next calls should keep returning NULL. In the
case of numbers iterator, bpf_iter_num_next() returns a pointer to an int
(storage for this integer is inside the iterator state itself),
which can be dereferenced after corresponding NULL check.
- once done with the iterator, it's mandated that user cleans up its
state with the call to destructor, bpf_iter_num_destroy() in this
case. Destructor frees up any resources and marks stack space used by
struct bpf_iter_num as usable for something else.
Any other iterator implementation will have to implement at least these
three methods. It is enforced that for any given type of iterator only
applicable constructor/destructor/next are callable. I.e., verifier
ensures you can't pass number iterator state into, say, cgroup
iterator's next method.
It is important to keep the naming pattern consistent to be able to
create generic macros to help with BPF iter usability. E.g., one
of the follow up patches adds generic bpf_for_each() macro to bpf_misc.h
in selftests, which allows to utilize iterator "trio" nicely without
having to code the above somewhat tedious loop explicitly every time.
This is enforced at kfunc registration point by one of the previous
patches in this series.
At the implementation level, iterator state tracking for verification
purposes is very similar to dynptr. We add STACK_ITER stack slot type,
reserve necessary number of slots, depending on
sizeof(struct bpf_iter_<type>), and keep track of necessary extra state
in the "main" slot, which is marked with non-zero ref_obj_id. Other
slots are also marked as STACK_ITER, but have zero ref_obj_id. This is
simpler than having a separate "is_first_slot" flag.
Another big distinction is that STACK_ITER is *always refcounted*, which
simplifies implementation without sacrificing usability. So no need for
extra "iter_id", no need to anticipate reuse of STACK_ITER slots for new
constructors, etc. Keeping it simple here.
As far as the verification logic goes, there are two extensive comments:
in process_iter_next_call() and iter_active_depths_differ() explaining
some important and sometimes subtle aspects. Please refer to them for
details.
But from 10,000-foot point of view, next methods are the points of
forking a verification state, which are conceptually similar to what
verifier is doing when validating conditional jump. We branch out at
a `call bpf_iter_<type>_next` instruction and simulate two outcomes:
NULL (iteration is done) and non-NULL (new element is returned). NULL is
simulated first and is supposed to reach exit without looping. After
that non-NULL case is validated and it either reaches exit (for trivial
examples with no real loop), or reaches another `call bpf_iter_<type>_next`
instruction with the state equivalent to already (partially) validated
one. State equivalency at that point means we technically are going to
be looping forever without "breaking out" out of established "state
envelope" (i.e., subsequent iterations don't add any new knowledge or
constraints to the verifier state, so running 1, 2, 10, or a million of
them doesn't matter). But taking into account the contract stating that
iterator next method *has to* return NULL eventually, we can conclude
that loop body is safe and will eventually terminate. Given we validated
logic outside of the loop (NULL case), and concluded that loop body is
safe (though potentially looping many times), verifier can claim safety
of the overall program logic.
The rest of the patch is necessary plumbing for state tracking, marking,
validation, and necessary further kfunc plumbing to allow implementing
iterator constructor, destructor, and next methods.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230308184121.1165081-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 18:41:16 +00:00
|
|
|
enum bpf_iter_state {
|
|
|
|
|
BPF_ITER_STATE_INVALID, /* for non-first slot */
|
|
|
|
|
BPF_ITER_STATE_ACTIVE,
|
|
|
|
|
BPF_ITER_STATE_DRAINED,
|
|
|
|
|
};
|
|
|
|
|
|
2016-09-21 10:43:57 +00:00
|
|
|
struct bpf_reg_state {
|
2018-08-22 19:02:19 +00:00
|
|
|
/* Ordering of fields matters. See states_equal() */
|
2016-09-21 10:43:57 +00:00
|
|
|
enum bpf_reg_type type;
|
2024-06-13 01:38:13 +00:00
|
|
|
/*
|
|
|
|
|
* Fixed part of pointer offset, pointer types only.
|
|
|
|
|
* Or constant delta between "linked" scalars with the same ID.
|
|
|
|
|
*/
|
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.
This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.
Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.
Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
2020-12-03 20:46:29 +00:00
|
|
|
s32 off;
|
2016-09-21 10:43:57 +00:00
|
|
|
union {
|
2017-08-07 14:26:19 +00:00
|
|
|
/* valid when type == PTR_TO_PACKET */
|
2020-11-11 03:12:11 +00:00
|
|
|
int range;
|
2016-09-21 10:43:57 +00:00
|
|
|
|
|
|
|
|
/* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
|
|
|
|
|
* PTR_TO_MAP_VALUE_OR_NULL
|
|
|
|
|
*/
|
2021-07-15 00:54:11 +00:00
|
|
|
struct {
|
|
|
|
|
struct bpf_map *map_ptr;
|
|
|
|
|
/* To distinguish map lookups from outer map
|
|
|
|
|
* the map_uid is non-zero for registers
|
|
|
|
|
* pointing to inner maps.
|
|
|
|
|
*/
|
|
|
|
|
u32 map_uid;
|
|
|
|
|
};
|
2018-10-31 23:05:52 +00:00
|
|
|
|
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.
This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.
Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.
Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
2020-12-03 20:46:29 +00:00
|
|
|
/* for PTR_TO_BTF_ID */
|
|
|
|
|
struct {
|
|
|
|
|
struct btf *btf;
|
|
|
|
|
u32 btf_id;
|
|
|
|
|
};
|
2019-10-16 03:25:00 +00:00
|
|
|
|
bpf: Invalidate slices on destruction of dynptrs on stack
The previous commit implemented destroy_if_dynptr_stack_slot. It
destroys the dynptr which given spi belongs to, but still doesn't
invalidate the slices that belong to such a dynptr. While for the case
of referenced dynptr, we don't allow their overwrite and return an error
early, we still allow it and destroy the dynptr for unreferenced dynptr.
To be able to enable precise and scoped invalidation of dynptr slices in
this case, we must be able to associate the source dynptr of slices that
have been obtained using bpf_dynptr_data. When doing destruction, only
slices belonging to the dynptr being destructed should be invalidated,
and nothing else. Currently, dynptr slices belonging to different
dynptrs are indistinguishible.
Hence, allocate a unique id to each dynptr (CONST_PTR_TO_DYNPTR and
those on stack). This will be stored as part of reg->id. Whenever using
bpf_dynptr_data, transfer this unique dynptr id to the returned
PTR_TO_MEM_OR_NULL slice pointer, and store it in a new per-PTR_TO_MEM
dynptr_id register state member.
Finally, after establishing such a relationship between dynptrs and
their slices, implement precise invalidation logic that only invalidates
slices belong to the destroyed dynptr in destroy_if_dynptr_stack_slot.
Acked-by: Joanne Koong <joannelkoong@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20230121002241.2113993-5-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-01-21 00:22:33 +00:00
|
|
|
struct { /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */
|
|
|
|
|
u32 mem_size;
|
|
|
|
|
u32 dynptr_id; /* for dynptr slices */
|
|
|
|
|
};
|
bpf: Implement BPF ring buffer and verifier support for it
This commit adds a new MPSC ring buffer implementation into BPF ecosystem,
which allows multiple CPUs to submit data to a single shared ring buffer. On
the consumption side, only single consumer is assumed.
Motivation
----------
There are two distinctive motivators for this work, which are not satisfied by
existing perf buffer, which prompted creation of a new ring buffer
implementation.
- more efficient memory utilization by sharing ring buffer across CPUs;
- preserving ordering of events that happen sequentially in time, even
across multiple CPUs (e.g., fork/exec/exit events for a task).
These two problems are independent, but perf buffer fails to satisfy both.
Both are a result of a choice to have per-CPU perf ring buffer. Both can be
also solved by having an MPSC implementation of ring buffer. The ordering
problem could technically be solved for perf buffer with some in-kernel
counting, but given the first one requires an MPSC buffer, the same solution
would solve the second problem automatically.
Semantics and APIs
------------------
Single ring buffer is presented to BPF programs as an instance of BPF map of
type BPF_MAP_TYPE_RINGBUF. Two other alternatives considered, but ultimately
rejected.
One way would be to, similar to BPF_MAP_TYPE_PERF_EVENT_ARRAY, make
BPF_MAP_TYPE_RINGBUF could represent an array of ring buffers, but not enforce
"same CPU only" rule. This would be more familiar interface compatible with
existing perf buffer use in BPF, but would fail if application needed more
advanced logic to lookup ring buffer by arbitrary key. HASH_OF_MAPS addresses
this with current approach. Additionally, given the performance of BPF
ringbuf, many use cases would just opt into a simple single ring buffer shared
among all CPUs, for which current approach would be an overkill.
Another approach could introduce a new concept, alongside BPF map, to
represent generic "container" object, which doesn't necessarily have key/value
interface with lookup/update/delete operations. This approach would add a lot
of extra infrastructure that has to be built for observability and verifier
support. It would also add another concept that BPF developers would have to
familiarize themselves with, new syntax in libbpf, etc. But then would really
provide no additional benefits over the approach of using a map.
BPF_MAP_TYPE_RINGBUF doesn't support lookup/update/delete operations, but so
doesn't few other map types (e.g., queue and stack; array doesn't support
delete, etc).
The approach chosen has an advantage of re-using existing BPF map
infrastructure (introspection APIs in kernel, libbpf support, etc), being
familiar concept (no need to teach users a new type of object in BPF program),
and utilizing existing tooling (bpftool). For common scenario of using
a single ring buffer for all CPUs, it's as simple and straightforward, as
would be with a dedicated "container" object. On the other hand, by being
a map, it can be combined with ARRAY_OF_MAPS and HASH_OF_MAPS map-in-maps to
implement a wide variety of topologies, from one ring buffer for each CPU
(e.g., as a replacement for perf buffer use cases), to a complicated
application hashing/sharding of ring buffers (e.g., having a small pool of
ring buffers with hashed task's tgid being a look up key to preserve order,
but reduce contention).
Key and value sizes are enforced to be zero. max_entries is used to specify
the size of ring buffer and has to be a power of 2 value.
There are a bunch of similarities between perf buffer
(BPF_MAP_TYPE_PERF_EVENT_ARRAY) and new BPF ring buffer semantics:
- variable-length records;
- if there is no more space left in ring buffer, reservation fails, no
blocking;
- memory-mappable data area for user-space applications for ease of
consumption and high performance;
- epoll notifications for new incoming data;
- but still the ability to do busy polling for new data to achieve the
lowest latency, if necessary.
BPF ringbuf provides two sets of APIs to BPF programs:
- bpf_ringbuf_output() allows to *copy* data from one place to a ring
buffer, similarly to bpf_perf_event_output();
- bpf_ringbuf_reserve()/bpf_ringbuf_commit()/bpf_ringbuf_discard() APIs
split the whole process into two steps. First, a fixed amount of space is
reserved. If successful, a pointer to a data inside ring buffer data area
is returned, which BPF programs can use similarly to a data inside
array/hash maps. Once ready, this piece of memory is either committed or
discarded. Discard is similar to commit, but makes consumer ignore the
record.
bpf_ringbuf_output() has disadvantage of incurring extra memory copy, because
record has to be prepared in some other place first. But it allows to submit
records of the length that's not known to verifier beforehand. It also closely
matches bpf_perf_event_output(), so will simplify migration significantly.
bpf_ringbuf_reserve() avoids the extra copy of memory by providing a memory
pointer directly to ring buffer memory. In a lot of cases records are larger
than BPF stack space allows, so many programs have use extra per-CPU array as
a temporary heap for preparing sample. bpf_ringbuf_reserve() avoid this needs
completely. But in exchange, it only allows a known constant size of memory to
be reserved, such that verifier can verify that BPF program can't access
memory outside its reserved record space. bpf_ringbuf_output(), while slightly
slower due to extra memory copy, covers some use cases that are not suitable
for bpf_ringbuf_reserve().
The difference between commit and discard is very small. Discard just marks
a record as discarded, and such records are supposed to be ignored by consumer
code. Discard is useful for some advanced use-cases, such as ensuring
all-or-nothing multi-record submission, or emulating temporary malloc()/free()
within single BPF program invocation.
Each reserved record is tracked by verifier through existing
reference-tracking logic, similar to socket ref-tracking. It is thus
impossible to reserve a record, but forget to submit (or discard) it.
bpf_ringbuf_query() helper allows to query various properties of ring buffer.
Currently 4 are supported:
- BPF_RB_AVAIL_DATA returns amount of unconsumed data in ring buffer;
- BPF_RB_RING_SIZE returns the size of ring buffer;
- BPF_RB_CONS_POS/BPF_RB_PROD_POS returns current logical possition of
consumer/producer, respectively.
Returned values are momentarily snapshots of ring buffer state and could be
off by the time helper returns, so this should be used only for
debugging/reporting reasons or for implementing various heuristics, that take
into account highly-changeable nature of some of those characteristics.
One such heuristic might involve more fine-grained control over poll/epoll
notifications about new data availability in ring buffer. Together with
BPF_RB_NO_WAKEUP/BPF_RB_FORCE_WAKEUP flags for output/commit/discard helpers,
it allows BPF program a high degree of control and, e.g., more efficient
batched notifications. Default self-balancing strategy, though, should be
adequate for most applications and will work reliable and efficiently already.
Design and implementation
-------------------------
This reserve/commit schema allows a natural way for multiple producers, either
on different CPUs or even on the same CPU/in the same BPF program, to reserve
independent records and work with them without blocking other producers. This
means that if BPF program was interruped by another BPF program sharing the
same ring buffer, they will both get a record reserved (provided there is
enough space left) and can work with it and submit it independently. This
applies to NMI context as well, except that due to using a spinlock during
reservation, in NMI context, bpf_ringbuf_reserve() might fail to get a lock,
in which case reservation will fail even if ring buffer is not full.
The ring buffer itself internally is implemented as a power-of-2 sized
circular buffer, with two logical and ever-increasing counters (which might
wrap around on 32-bit architectures, that's not a problem):
- consumer counter shows up to which logical position consumer consumed the
data;
- producer counter denotes amount of data reserved by all producers.
Each time a record is reserved, producer that "owns" the record will
successfully advance producer counter. At that point, data is still not yet
ready to be consumed, though. Each record has 8 byte header, which contains
the length of reserved record, as well as two extra bits: busy bit to denote
that record is still being worked on, and discard bit, which might be set at
commit time if record is discarded. In the latter case, consumer is supposed
to skip the record and move on to the next one. Record header also encodes
record's relative offset from the beginning of ring buffer data area (in
pages). This allows bpf_ringbuf_commit()/bpf_ringbuf_discard() to accept only
the pointer to the record itself, without requiring also the pointer to ring
buffer itself. Ring buffer memory location will be restored from record
metadata header. This significantly simplifies verifier, as well as improving
API usability.
Producer counter increments are serialized under spinlock, so there is
a strict ordering between reservations. Commits, on the other hand, are
completely lockless and independent. All records become available to consumer
in the order of reservations, but only after all previous records where
already committed. It is thus possible for slow producers to temporarily hold
off submitted records, that were reserved later.
Reservation/commit/consumer protocol is verified by litmus tests in
Documentation/litmus-test/bpf-rb.
One interesting implementation bit, that significantly simplifies (and thus
speeds up as well) implementation of both producers and consumers is how data
area is mapped twice contiguously back-to-back in the virtual memory. This
allows to not take any special measures for samples that have to wrap around
at the end of the circular buffer data area, because the next page after the
last data page would be first data page again, and thus the sample will still
appear completely contiguous in virtual memory. See comment and a simple ASCII
diagram showing this visually in bpf_ringbuf_area_alloc().
Another feature that distinguishes BPF ringbuf from perf ring buffer is
a self-pacing notifications of new data being availability.
bpf_ringbuf_commit() implementation will send a notification of new record
being available after commit only if consumer has already caught up right up
to the record being committed. If not, consumer still has to catch up and thus
will see new data anyways without needing an extra poll notification.
Benchmarks (see tools/testing/selftests/bpf/benchs/bench_ringbuf.c) show that
this allows to achieve a very high throughput without having to resort to
tricks like "notify only every Nth sample", which are necessary with perf
buffer. For extreme cases, when BPF program wants more manual control of
notifications, commit/discard/output helpers accept BPF_RB_NO_WAKEUP and
BPF_RB_FORCE_WAKEUP flags, which give full control over notifications of data
availability, but require extra caution and diligence in using this API.
Comparison to alternatives
--------------------------
Before considering implementing BPF ring buffer from scratch existing
alternatives in kernel were evaluated, but didn't seem to meet the needs. They
largely fell into few categores:
- per-CPU buffers (perf, ftrace, etc), which don't satisfy two motivations
outlined above (ordering and memory consumption);
- linked list-based implementations; while some were multi-producer designs,
consuming these from user-space would be very complicated and most
probably not performant; memory-mapping contiguous piece of memory is
simpler and more performant for user-space consumers;
- io_uring is SPSC, but also requires fixed-sized elements. Naively turning
SPSC queue into MPSC w/ lock would have subpar performance compared to
locked reserve + lockless commit, as with BPF ring buffer. Fixed sized
elements would be too limiting for BPF programs, given existing BPF
programs heavily rely on variable-sized perf buffer already;
- specialized implementations (like a new printk ring buffer, [0]) with lots
of printk-specific limitations and implications, that didn't seem to fit
well for intended use with BPF programs.
[0] https://lwn.net/Articles/779550/
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200529075424.3139988-2-andriin@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 07:54:20 +00:00
|
|
|
|
bpf: Add verifier support for dynptrs
This patch adds the bulk of the verifier work for supporting dynamic
pointers (dynptrs) in bpf.
A bpf_dynptr is opaque to the bpf program. It is a 16-byte structure
defined internally as:
struct bpf_dynptr_kern {
void *data;
u32 size;
u32 offset;
} __aligned(8);
The upper 8 bits of *size* is reserved (it contains extra metadata about
read-only status and dynptr type). Consequently, a dynptr only supports
memory less than 16 MB.
There are different types of dynptrs (eg malloc, ringbuf, ...). In this
patchset, the most basic one, dynptrs to a bpf program's local memory,
is added. For now only local memory that is of reg type PTR_TO_MAP_VALUE
is supported.
In the verifier, dynptr state information will be tracked in stack
slots. When the program passes in an uninitialized dynptr
(ARG_PTR_TO_DYNPTR | MEM_UNINIT), the stack slots corresponding
to the frame pointer where the dynptr resides at are marked
STACK_DYNPTR. For helper functions that take in initialized dynptrs (eg
bpf_dynptr_read + bpf_dynptr_write which are added later in this
patchset), the verifier enforces that the dynptr has been initialized
properly by checking that their corresponding stack slots have been
marked as STACK_DYNPTR.
The 6th patch in this patchset adds test cases that the verifier should
successfully reject, such as for example attempting to use a dynptr
after doing a direct write into it inside the bpf program.
Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/bpf/20220523210712.3641569-2-joannelkoong@gmail.com
2022-05-23 21:07:07 +00:00
|
|
|
/* For dynptr stack slots */
|
|
|
|
|
struct {
|
|
|
|
|
enum bpf_dynptr_type type;
|
|
|
|
|
/* A dynptr is 16 bytes so it takes up 2 stack slots.
|
|
|
|
|
* We need to track which slot is the first slot
|
|
|
|
|
* to protect against cases where the user may try to
|
|
|
|
|
* pass in an address starting at the second slot of the
|
|
|
|
|
* dynptr.
|
|
|
|
|
*/
|
|
|
|
|
bool first_slot;
|
|
|
|
|
} dynptr;
|
|
|
|
|
|
bpf: add support for open-coded iterator loops
Teach verifier about the concept of the open-coded (or inline) iterators.
This patch adds generic iterator loop verification logic, new STACK_ITER
stack slot type to contain iterator state, and necessary kfunc plumbing
for iterator's constructor, destructor and next methods. Next patch
implements first specific iterator (numbers iterator for implementing
for() loop logic). Such split allows to have more focused commits for
verifier logic and separate commit that we could point later to
demonstrating what does it take to add a new kind of iterator.
Each kind of iterator has its own associated struct bpf_iter_<type>,
where <type> denotes a specific type of iterator. struct bpf_iter_<type>
state is supposed to live on BPF program stack, so there will be no way
to change its size later on without breaking backwards compatibility, so
choose wisely! But given this struct is specific to a given <type> of
iterator, this allows a lot of flexibility: simple iterators could be
fine with just one stack slot (8 bytes), like numbers iterator in the
next patch, while some other more complicated iterators might need way
more to keep their iterator state. Either way, such design allows to
avoid runtime memory allocations, which otherwise would be necessary if
we fixed on-the-stack size and it turned out to be too small for a given
iterator implementation.
The way BPF verifier logic is implemented, there are no artificial
restrictions on a number of active iterators, it should work correctly
using multiple active iterators at the same time. This also means you
can have multiple nested iteration loops. struct bpf_iter_<type>
reference can be safely passed to subprograms as well.
General flow is easiest to demonstrate with a simple example using
number iterator implemented in next patch. Here's the simplest possible
loop:
struct bpf_iter_num it;
int *v;
bpf_iter_num_new(&it, 2, 5);
while ((v = bpf_iter_num_next(&it))) {
bpf_printk("X = %d", *v);
}
bpf_iter_num_destroy(&it);
Above snippet should output "X = 2", "X = 3", "X = 4". Note that 5 is
exclusive and is not returned. This matches similar APIs (e.g., slices
in Go or Rust) that implement a range of elements, where end index is
non-inclusive.
In the above example, we see a trio of function:
- constructor, bpf_iter_num_new(), which initializes iterator state
(struct bpf_iter_num it) on the stack. If any of the input arguments
are invalid, constructor should make sure to still initialize it such
that subsequent bpf_iter_num_next() calls will return NULL. I.e., on
error, return error and construct empty iterator.
- next method, bpf_iter_num_next(), which accepts pointer to iterator
state and produces an element. Next method should always return
a pointer. The contract between BPF verifier is that next method will
always eventually return NULL when elements are exhausted. Once NULL is
returned, subsequent next calls should keep returning NULL. In the
case of numbers iterator, bpf_iter_num_next() returns a pointer to an int
(storage for this integer is inside the iterator state itself),
which can be dereferenced after corresponding NULL check.
- once done with the iterator, it's mandated that user cleans up its
state with the call to destructor, bpf_iter_num_destroy() in this
case. Destructor frees up any resources and marks stack space used by
struct bpf_iter_num as usable for something else.
Any other iterator implementation will have to implement at least these
three methods. It is enforced that for any given type of iterator only
applicable constructor/destructor/next are callable. I.e., verifier
ensures you can't pass number iterator state into, say, cgroup
iterator's next method.
It is important to keep the naming pattern consistent to be able to
create generic macros to help with BPF iter usability. E.g., one
of the follow up patches adds generic bpf_for_each() macro to bpf_misc.h
in selftests, which allows to utilize iterator "trio" nicely without
having to code the above somewhat tedious loop explicitly every time.
This is enforced at kfunc registration point by one of the previous
patches in this series.
At the implementation level, iterator state tracking for verification
purposes is very similar to dynptr. We add STACK_ITER stack slot type,
reserve necessary number of slots, depending on
sizeof(struct bpf_iter_<type>), and keep track of necessary extra state
in the "main" slot, which is marked with non-zero ref_obj_id. Other
slots are also marked as STACK_ITER, but have zero ref_obj_id. This is
simpler than having a separate "is_first_slot" flag.
Another big distinction is that STACK_ITER is *always refcounted*, which
simplifies implementation without sacrificing usability. So no need for
extra "iter_id", no need to anticipate reuse of STACK_ITER slots for new
constructors, etc. Keeping it simple here.
As far as the verification logic goes, there are two extensive comments:
in process_iter_next_call() and iter_active_depths_differ() explaining
some important and sometimes subtle aspects. Please refer to them for
details.
But from 10,000-foot point of view, next methods are the points of
forking a verification state, which are conceptually similar to what
verifier is doing when validating conditional jump. We branch out at
a `call bpf_iter_<type>_next` instruction and simulate two outcomes:
NULL (iteration is done) and non-NULL (new element is returned). NULL is
simulated first and is supposed to reach exit without looping. After
that non-NULL case is validated and it either reaches exit (for trivial
examples with no real loop), or reaches another `call bpf_iter_<type>_next`
instruction with the state equivalent to already (partially) validated
one. State equivalency at that point means we technically are going to
be looping forever without "breaking out" out of established "state
envelope" (i.e., subsequent iterations don't add any new knowledge or
constraints to the verifier state, so running 1, 2, 10, or a million of
them doesn't matter). But taking into account the contract stating that
iterator next method *has to* return NULL eventually, we can conclude
that loop body is safe and will eventually terminate. Given we validated
logic outside of the loop (NULL case), and concluded that loop body is
safe (though potentially looping many times), verifier can claim safety
of the overall program logic.
The rest of the patch is necessary plumbing for state tracking, marking,
validation, and necessary further kfunc plumbing to allow implementing
iterator constructor, destructor, and next methods.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230308184121.1165081-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 18:41:16 +00:00
|
|
|
/* For bpf_iter stack slots */
|
|
|
|
|
struct {
|
|
|
|
|
/* BTF container and BTF type ID describing
|
|
|
|
|
* struct bpf_iter_<type> of an iterator state
|
|
|
|
|
*/
|
|
|
|
|
struct btf *btf;
|
|
|
|
|
u32 btf_id;
|
|
|
|
|
/* packing following two fields to fit iter state into 16 bytes */
|
|
|
|
|
enum bpf_iter_state state:2;
|
|
|
|
|
int depth:30;
|
|
|
|
|
} iter;
|
|
|
|
|
|
bpf: Implement verifier support for rqspinlock
Introduce verifier-side support for rqspinlock kfuncs. The first step is
allowing bpf_res_spin_lock type to be defined in map values and
allocated objects, so BTF-side is updated with a new BPF_RES_SPIN_LOCK
field to recognize and validate.
Any object cannot have both bpf_spin_lock and bpf_res_spin_lock, only
one of them (and at most one of them per-object, like before) must be
present. The bpf_res_spin_lock can also be used to protect objects that
require lock protection for their kfuncs, like BPF rbtree and linked
list.
The verifier plumbing to simulate success and failure cases when calling
the kfuncs is done by pushing a new verifier state to the verifier state
stack which will verify the failure case upon calling the kfunc. The
path where success is indicated creates all lock reference state and IRQ
state (if necessary for irqsave variants). In the case of failure, the
state clears the registers r0-r5, sets the return value, and skips kfunc
processing, proceeding to the next instruction.
When marking the return value for success case, the value is marked as
0, and for the failure case as [-MAX_ERRNO, -1]. Then, in the program,
whenever user checks the return value as 'if (ret)' or 'if (ret < 0)'
the verifier never traverses such branches for success cases, and would
be aware that the lock is not held in such cases.
We push the kfunc state in check_kfunc_call whenever rqspinlock kfuncs
are invoked. We introduce a kfunc_class state to avoid mixing lock
irqrestore kfuncs with IRQ state created by bpf_local_irq_save.
With all this infrastructure, these kfuncs become usable in programs
while satisfying all safety properties required by the kernel.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20250316040541.108729-24-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-03-16 04:05:39 +00:00
|
|
|
/* For irq stack slots */
|
|
|
|
|
struct {
|
|
|
|
|
enum {
|
|
|
|
|
IRQ_NATIVE_KFUNC,
|
|
|
|
|
IRQ_LOCK_KFUNC,
|
|
|
|
|
} kfunc_class;
|
|
|
|
|
} irq;
|
|
|
|
|
|
2018-10-31 23:05:52 +00:00
|
|
|
/* Max size from any of the above. */
|
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.
This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.
Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.
Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
2020-12-03 20:46:29 +00:00
|
|
|
struct {
|
|
|
|
|
unsigned long raw1;
|
|
|
|
|
unsigned long raw2;
|
|
|
|
|
} raw;
|
bpf: Add bpf_for_each_map_elem() helper
The bpf_for_each_map_elem() helper is introduced which
iterates all map elements with a callback function. The
helper signature looks like
long bpf_for_each_map_elem(map, callback_fn, callback_ctx, flags)
and for each map element, the callback_fn will be called. For example,
like hashmap, the callback signature may look like
long callback_fn(map, key, val, callback_ctx)
There are two known use cases for this. One is from upstream ([1]) where
a for_each_map_elem helper may help implement a timeout mechanism
in a more generic way. Another is from our internal discussion
for a firewall use case where a map contains all the rules. The packet
data can be compared to all these rules to decide allow or deny
the packet.
For array maps, users can already use a bounded loop to traverse
elements. Using this helper can avoid using bounded loop. For other
type of maps (e.g., hash maps) where bounded loop is hard or
impossible to use, this helper provides a convenient way to
operate on all elements.
For callback_fn, besides map and map element, a callback_ctx,
allocated on caller stack, is also passed to the callback
function. This callback_ctx argument can provide additional
input and allow to write to caller stack for output.
If the callback_fn returns 0, the helper will iterate through next
element if available. If the callback_fn returns 1, the helper
will stop iterating and returns to the bpf program. Other return
values are not used for now.
Currently, this helper is only available with jit. It is possible
to make it work with interpreter with so effort but I leave it
as the future work.
[1]: https://lore.kernel.org/bpf/20210122205415.113822-1-xiyou.wangcong@gmail.com/
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210226204925.3884923-1-yhs@fb.com
2021-02-26 20:49:25 +00:00
|
|
|
|
|
|
|
|
u32 subprogno; /* for PTR_TO_FUNC */
|
2016-09-21 10:43:57 +00:00
|
|
|
};
|
2022-12-23 05:49:16 +00:00
|
|
|
/* For scalar types (SCALAR_VALUE), this represents our knowledge of
|
|
|
|
|
* the actual value.
|
|
|
|
|
* For pointer types, this represents the variable part of the offset
|
|
|
|
|
* from the pointed-to object, and is shared with all bpf_reg_states
|
|
|
|
|
* with the same id as us.
|
|
|
|
|
*/
|
|
|
|
|
struct tnum var_off;
|
|
|
|
|
/* Used to determine if any memory access using this register will
|
|
|
|
|
* result in a bad access.
|
|
|
|
|
* These refer to the same value as var_off, not necessarily the actual
|
|
|
|
|
* contents of the register.
|
|
|
|
|
*/
|
|
|
|
|
s64 smin_value; /* minimum possible (s64)value */
|
|
|
|
|
s64 smax_value; /* maximum possible (s64)value */
|
|
|
|
|
u64 umin_value; /* minimum possible (u64)value */
|
|
|
|
|
u64 umax_value; /* maximum possible (u64)value */
|
|
|
|
|
s32 s32_min_value; /* minimum possible (s32)value */
|
|
|
|
|
s32 s32_max_value; /* maximum possible (s32)value */
|
|
|
|
|
u32 u32_min_value; /* minimum possible (u32)value */
|
|
|
|
|
u32 u32_max_value; /* maximum possible (u32)value */
|
2017-08-07 14:26:19 +00:00
|
|
|
/* For PTR_TO_PACKET, used to find other pointers with the same variable
|
|
|
|
|
* offset, so they can share range knowledge.
|
|
|
|
|
* For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
|
|
|
|
|
* came from, when one is tested for != NULL.
|
bpf: Implement BPF ring buffer and verifier support for it
This commit adds a new MPSC ring buffer implementation into BPF ecosystem,
which allows multiple CPUs to submit data to a single shared ring buffer. On
the consumption side, only single consumer is assumed.
Motivation
----------
There are two distinctive motivators for this work, which are not satisfied by
existing perf buffer, which prompted creation of a new ring buffer
implementation.
- more efficient memory utilization by sharing ring buffer across CPUs;
- preserving ordering of events that happen sequentially in time, even
across multiple CPUs (e.g., fork/exec/exit events for a task).
These two problems are independent, but perf buffer fails to satisfy both.
Both are a result of a choice to have per-CPU perf ring buffer. Both can be
also solved by having an MPSC implementation of ring buffer. The ordering
problem could technically be solved for perf buffer with some in-kernel
counting, but given the first one requires an MPSC buffer, the same solution
would solve the second problem automatically.
Semantics and APIs
------------------
Single ring buffer is presented to BPF programs as an instance of BPF map of
type BPF_MAP_TYPE_RINGBUF. Two other alternatives considered, but ultimately
rejected.
One way would be to, similar to BPF_MAP_TYPE_PERF_EVENT_ARRAY, make
BPF_MAP_TYPE_RINGBUF could represent an array of ring buffers, but not enforce
"same CPU only" rule. This would be more familiar interface compatible with
existing perf buffer use in BPF, but would fail if application needed more
advanced logic to lookup ring buffer by arbitrary key. HASH_OF_MAPS addresses
this with current approach. Additionally, given the performance of BPF
ringbuf, many use cases would just opt into a simple single ring buffer shared
among all CPUs, for which current approach would be an overkill.
Another approach could introduce a new concept, alongside BPF map, to
represent generic "container" object, which doesn't necessarily have key/value
interface with lookup/update/delete operations. This approach would add a lot
of extra infrastructure that has to be built for observability and verifier
support. It would also add another concept that BPF developers would have to
familiarize themselves with, new syntax in libbpf, etc. But then would really
provide no additional benefits over the approach of using a map.
BPF_MAP_TYPE_RINGBUF doesn't support lookup/update/delete operations, but so
doesn't few other map types (e.g., queue and stack; array doesn't support
delete, etc).
The approach chosen has an advantage of re-using existing BPF map
infrastructure (introspection APIs in kernel, libbpf support, etc), being
familiar concept (no need to teach users a new type of object in BPF program),
and utilizing existing tooling (bpftool). For common scenario of using
a single ring buffer for all CPUs, it's as simple and straightforward, as
would be with a dedicated "container" object. On the other hand, by being
a map, it can be combined with ARRAY_OF_MAPS and HASH_OF_MAPS map-in-maps to
implement a wide variety of topologies, from one ring buffer for each CPU
(e.g., as a replacement for perf buffer use cases), to a complicated
application hashing/sharding of ring buffers (e.g., having a small pool of
ring buffers with hashed task's tgid being a look up key to preserve order,
but reduce contention).
Key and value sizes are enforced to be zero. max_entries is used to specify
the size of ring buffer and has to be a power of 2 value.
There are a bunch of similarities between perf buffer
(BPF_MAP_TYPE_PERF_EVENT_ARRAY) and new BPF ring buffer semantics:
- variable-length records;
- if there is no more space left in ring buffer, reservation fails, no
blocking;
- memory-mappable data area for user-space applications for ease of
consumption and high performance;
- epoll notifications for new incoming data;
- but still the ability to do busy polling for new data to achieve the
lowest latency, if necessary.
BPF ringbuf provides two sets of APIs to BPF programs:
- bpf_ringbuf_output() allows to *copy* data from one place to a ring
buffer, similarly to bpf_perf_event_output();
- bpf_ringbuf_reserve()/bpf_ringbuf_commit()/bpf_ringbuf_discard() APIs
split the whole process into two steps. First, a fixed amount of space is
reserved. If successful, a pointer to a data inside ring buffer data area
is returned, which BPF programs can use similarly to a data inside
array/hash maps. Once ready, this piece of memory is either committed or
discarded. Discard is similar to commit, but makes consumer ignore the
record.
bpf_ringbuf_output() has disadvantage of incurring extra memory copy, because
record has to be prepared in some other place first. But it allows to submit
records of the length that's not known to verifier beforehand. It also closely
matches bpf_perf_event_output(), so will simplify migration significantly.
bpf_ringbuf_reserve() avoids the extra copy of memory by providing a memory
pointer directly to ring buffer memory. In a lot of cases records are larger
than BPF stack space allows, so many programs have use extra per-CPU array as
a temporary heap for preparing sample. bpf_ringbuf_reserve() avoid this needs
completely. But in exchange, it only allows a known constant size of memory to
be reserved, such that verifier can verify that BPF program can't access
memory outside its reserved record space. bpf_ringbuf_output(), while slightly
slower due to extra memory copy, covers some use cases that are not suitable
for bpf_ringbuf_reserve().
The difference between commit and discard is very small. Discard just marks
a record as discarded, and such records are supposed to be ignored by consumer
code. Discard is useful for some advanced use-cases, such as ensuring
all-or-nothing multi-record submission, or emulating temporary malloc()/free()
within single BPF program invocation.
Each reserved record is tracked by verifier through existing
reference-tracking logic, similar to socket ref-tracking. It is thus
impossible to reserve a record, but forget to submit (or discard) it.
bpf_ringbuf_query() helper allows to query various properties of ring buffer.
Currently 4 are supported:
- BPF_RB_AVAIL_DATA returns amount of unconsumed data in ring buffer;
- BPF_RB_RING_SIZE returns the size of ring buffer;
- BPF_RB_CONS_POS/BPF_RB_PROD_POS returns current logical possition of
consumer/producer, respectively.
Returned values are momentarily snapshots of ring buffer state and could be
off by the time helper returns, so this should be used only for
debugging/reporting reasons or for implementing various heuristics, that take
into account highly-changeable nature of some of those characteristics.
One such heuristic might involve more fine-grained control over poll/epoll
notifications about new data availability in ring buffer. Together with
BPF_RB_NO_WAKEUP/BPF_RB_FORCE_WAKEUP flags for output/commit/discard helpers,
it allows BPF program a high degree of control and, e.g., more efficient
batched notifications. Default self-balancing strategy, though, should be
adequate for most applications and will work reliable and efficiently already.
Design and implementation
-------------------------
This reserve/commit schema allows a natural way for multiple producers, either
on different CPUs or even on the same CPU/in the same BPF program, to reserve
independent records and work with them without blocking other producers. This
means that if BPF program was interruped by another BPF program sharing the
same ring buffer, they will both get a record reserved (provided there is
enough space left) and can work with it and submit it independently. This
applies to NMI context as well, except that due to using a spinlock during
reservation, in NMI context, bpf_ringbuf_reserve() might fail to get a lock,
in which case reservation will fail even if ring buffer is not full.
The ring buffer itself internally is implemented as a power-of-2 sized
circular buffer, with two logical and ever-increasing counters (which might
wrap around on 32-bit architectures, that's not a problem):
- consumer counter shows up to which logical position consumer consumed the
data;
- producer counter denotes amount of data reserved by all producers.
Each time a record is reserved, producer that "owns" the record will
successfully advance producer counter. At that point, data is still not yet
ready to be consumed, though. Each record has 8 byte header, which contains
the length of reserved record, as well as two extra bits: busy bit to denote
that record is still being worked on, and discard bit, which might be set at
commit time if record is discarded. In the latter case, consumer is supposed
to skip the record and move on to the next one. Record header also encodes
record's relative offset from the beginning of ring buffer data area (in
pages). This allows bpf_ringbuf_commit()/bpf_ringbuf_discard() to accept only
the pointer to the record itself, without requiring also the pointer to ring
buffer itself. Ring buffer memory location will be restored from record
metadata header. This significantly simplifies verifier, as well as improving
API usability.
Producer counter increments are serialized under spinlock, so there is
a strict ordering between reservations. Commits, on the other hand, are
completely lockless and independent. All records become available to consumer
in the order of reservations, but only after all previous records where
already committed. It is thus possible for slow producers to temporarily hold
off submitted records, that were reserved later.
Reservation/commit/consumer protocol is verified by litmus tests in
Documentation/litmus-test/bpf-rb.
One interesting implementation bit, that significantly simplifies (and thus
speeds up as well) implementation of both producers and consumers is how data
area is mapped twice contiguously back-to-back in the virtual memory. This
allows to not take any special measures for samples that have to wrap around
at the end of the circular buffer data area, because the next page after the
last data page would be first data page again, and thus the sample will still
appear completely contiguous in virtual memory. See comment and a simple ASCII
diagram showing this visually in bpf_ringbuf_area_alloc().
Another feature that distinguishes BPF ringbuf from perf ring buffer is
a self-pacing notifications of new data being availability.
bpf_ringbuf_commit() implementation will send a notification of new record
being available after commit only if consumer has already caught up right up
to the record being committed. If not, consumer still has to catch up and thus
will see new data anyways without needing an extra poll notification.
Benchmarks (see tools/testing/selftests/bpf/benchs/bench_ringbuf.c) show that
this allows to achieve a very high throughput without having to resort to
tricks like "notify only every Nth sample", which are necessary with perf
buffer. For extreme cases, when BPF program wants more manual control of
notifications, commit/discard/output helpers accept BPF_RB_NO_WAKEUP and
BPF_RB_FORCE_WAKEUP flags, which give full control over notifications of data
availability, but require extra caution and diligence in using this API.
Comparison to alternatives
--------------------------
Before considering implementing BPF ring buffer from scratch existing
alternatives in kernel were evaluated, but didn't seem to meet the needs. They
largely fell into few categores:
- per-CPU buffers (perf, ftrace, etc), which don't satisfy two motivations
outlined above (ordering and memory consumption);
- linked list-based implementations; while some were multi-producer designs,
consuming these from user-space would be very complicated and most
probably not performant; memory-mapping contiguous piece of memory is
simpler and more performant for user-space consumers;
- io_uring is SPSC, but also requires fixed-sized elements. Naively turning
SPSC queue into MPSC w/ lock would have subpar performance compared to
locked reserve + lockless commit, as with BPF ring buffer. Fixed sized
elements would be too limiting for BPF programs, given existing BPF
programs heavily rely on variable-sized perf buffer already;
- specialized implementations (like a new printk ring buffer, [0]) with lots
of printk-specific limitations and implications, that didn't seem to fit
well for intended use with BPF programs.
[0] https://lwn.net/Articles/779550/
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200529075424.3139988-2-andriin@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 07:54:20 +00:00
|
|
|
* For PTR_TO_MEM_OR_NULL this is used to identify memory allocation
|
|
|
|
|
* for the purpose of tracking that it's freed.
|
2018-10-02 20:35:33 +00:00
|
|
|
* For PTR_TO_SOCKET this is used to share which pointers retain the
|
|
|
|
|
* same reference to the socket, to determine proper reference freeing.
|
2022-05-23 21:07:09 +00:00
|
|
|
* For stack slots that are dynptrs, this is used to track references to
|
|
|
|
|
* the dynptr to determine proper reference freeing.
|
bpf: add support for open-coded iterator loops
Teach verifier about the concept of the open-coded (or inline) iterators.
This patch adds generic iterator loop verification logic, new STACK_ITER
stack slot type to contain iterator state, and necessary kfunc plumbing
for iterator's constructor, destructor and next methods. Next patch
implements first specific iterator (numbers iterator for implementing
for() loop logic). Such split allows to have more focused commits for
verifier logic and separate commit that we could point later to
demonstrating what does it take to add a new kind of iterator.
Each kind of iterator has its own associated struct bpf_iter_<type>,
where <type> denotes a specific type of iterator. struct bpf_iter_<type>
state is supposed to live on BPF program stack, so there will be no way
to change its size later on without breaking backwards compatibility, so
choose wisely! But given this struct is specific to a given <type> of
iterator, this allows a lot of flexibility: simple iterators could be
fine with just one stack slot (8 bytes), like numbers iterator in the
next patch, while some other more complicated iterators might need way
more to keep their iterator state. Either way, such design allows to
avoid runtime memory allocations, which otherwise would be necessary if
we fixed on-the-stack size and it turned out to be too small for a given
iterator implementation.
The way BPF verifier logic is implemented, there are no artificial
restrictions on a number of active iterators, it should work correctly
using multiple active iterators at the same time. This also means you
can have multiple nested iteration loops. struct bpf_iter_<type>
reference can be safely passed to subprograms as well.
General flow is easiest to demonstrate with a simple example using
number iterator implemented in next patch. Here's the simplest possible
loop:
struct bpf_iter_num it;
int *v;
bpf_iter_num_new(&it, 2, 5);
while ((v = bpf_iter_num_next(&it))) {
bpf_printk("X = %d", *v);
}
bpf_iter_num_destroy(&it);
Above snippet should output "X = 2", "X = 3", "X = 4". Note that 5 is
exclusive and is not returned. This matches similar APIs (e.g., slices
in Go or Rust) that implement a range of elements, where end index is
non-inclusive.
In the above example, we see a trio of function:
- constructor, bpf_iter_num_new(), which initializes iterator state
(struct bpf_iter_num it) on the stack. If any of the input arguments
are invalid, constructor should make sure to still initialize it such
that subsequent bpf_iter_num_next() calls will return NULL. I.e., on
error, return error and construct empty iterator.
- next method, bpf_iter_num_next(), which accepts pointer to iterator
state and produces an element. Next method should always return
a pointer. The contract between BPF verifier is that next method will
always eventually return NULL when elements are exhausted. Once NULL is
returned, subsequent next calls should keep returning NULL. In the
case of numbers iterator, bpf_iter_num_next() returns a pointer to an int
(storage for this integer is inside the iterator state itself),
which can be dereferenced after corresponding NULL check.
- once done with the iterator, it's mandated that user cleans up its
state with the call to destructor, bpf_iter_num_destroy() in this
case. Destructor frees up any resources and marks stack space used by
struct bpf_iter_num as usable for something else.
Any other iterator implementation will have to implement at least these
three methods. It is enforced that for any given type of iterator only
applicable constructor/destructor/next are callable. I.e., verifier
ensures you can't pass number iterator state into, say, cgroup
iterator's next method.
It is important to keep the naming pattern consistent to be able to
create generic macros to help with BPF iter usability. E.g., one
of the follow up patches adds generic bpf_for_each() macro to bpf_misc.h
in selftests, which allows to utilize iterator "trio" nicely without
having to code the above somewhat tedious loop explicitly every time.
This is enforced at kfunc registration point by one of the previous
patches in this series.
At the implementation level, iterator state tracking for verification
purposes is very similar to dynptr. We add STACK_ITER stack slot type,
reserve necessary number of slots, depending on
sizeof(struct bpf_iter_<type>), and keep track of necessary extra state
in the "main" slot, which is marked with non-zero ref_obj_id. Other
slots are also marked as STACK_ITER, but have zero ref_obj_id. This is
simpler than having a separate "is_first_slot" flag.
Another big distinction is that STACK_ITER is *always refcounted*, which
simplifies implementation without sacrificing usability. So no need for
extra "iter_id", no need to anticipate reuse of STACK_ITER slots for new
constructors, etc. Keeping it simple here.
As far as the verification logic goes, there are two extensive comments:
in process_iter_next_call() and iter_active_depths_differ() explaining
some important and sometimes subtle aspects. Please refer to them for
details.
But from 10,000-foot point of view, next methods are the points of
forking a verification state, which are conceptually similar to what
verifier is doing when validating conditional jump. We branch out at
a `call bpf_iter_<type>_next` instruction and simulate two outcomes:
NULL (iteration is done) and non-NULL (new element is returned). NULL is
simulated first and is supposed to reach exit without looping. After
that non-NULL case is validated and it either reaches exit (for trivial
examples with no real loop), or reaches another `call bpf_iter_<type>_next`
instruction with the state equivalent to already (partially) validated
one. State equivalency at that point means we technically are going to
be looping forever without "breaking out" out of established "state
envelope" (i.e., subsequent iterations don't add any new knowledge or
constraints to the verifier state, so running 1, 2, 10, or a million of
them doesn't matter). But taking into account the contract stating that
iterator next method *has to* return NULL eventually, we can conclude
that loop body is safe and will eventually terminate. Given we validated
logic outside of the loop (NULL case), and concluded that loop body is
safe (though potentially looping many times), verifier can claim safety
of the overall program logic.
The rest of the patch is necessary plumbing for state tracking, marking,
validation, and necessary further kfunc plumbing to allow implementing
iterator constructor, destructor, and next methods.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230308184121.1165081-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 18:41:16 +00:00
|
|
|
* Similarly to dynptrs, we use ID to track "belonging" of a reference
|
|
|
|
|
* to a specific instance of bpf_iter.
|
2017-08-07 14:26:19 +00:00
|
|
|
*/
|
2024-06-13 01:38:13 +00:00
|
|
|
/*
|
|
|
|
|
* Upper bit of ID is used to remember relationship between "linked"
|
|
|
|
|
* registers. Example:
|
|
|
|
|
* r1 = r2; both will have r1->id == r2->id == N
|
|
|
|
|
* r1 += 10; r1->id == N | BPF_ADD_CONST and r1->off == 10
|
|
|
|
|
*/
|
|
|
|
|
#define BPF_ADD_CONST (1U << 31)
|
2016-12-07 18:57:59 +00:00
|
|
|
u32 id;
|
bpf: Fix bpf_tcp_sock and bpf_sk_fullsock issue related to bpf_sk_release
Lorenz Bauer [thanks!] reported that a ptr returned by bpf_tcp_sock(sk)
can still be accessed after bpf_sk_release(sk).
Both bpf_tcp_sock() and bpf_sk_fullsock() have the same issue.
This patch addresses them together.
A simple reproducer looks like this:
sk = bpf_sk_lookup_tcp();
/* if (!sk) ... */
tp = bpf_tcp_sock(sk);
/* if (!tp) ... */
bpf_sk_release(sk);
snd_cwnd = tp->snd_cwnd; /* oops! The verifier does not complain. */
The problem is the verifier did not scrub the register's states of
the tcp_sock ptr (tp) after bpf_sk_release(sk).
[ Note that when calling bpf_tcp_sock(sk), the sk is not always
refcount-acquired. e.g. bpf_tcp_sock(skb->sk). The verifier works
fine for this case. ]
Currently, the verifier does not track if a helper's return ptr (in REG_0)
is "carry"-ing one of its argument's refcount status. To carry this info,
the reg1->id needs to be stored in reg0.
One approach was tried, like "reg0->id = reg1->id", when calling
"bpf_tcp_sock()". The main idea was to avoid adding another "ref_obj_id"
for the same reg. However, overlapping the NULL marking and ref
tracking purpose in one "id" does not work well:
ref_sk = bpf_sk_lookup_tcp();
fullsock = bpf_sk_fullsock(ref_sk);
tp = bpf_tcp_sock(ref_sk);
if (!fullsock) {
bpf_sk_release(ref_sk);
return 0;
}
/* fullsock_reg->id is marked for NOT-NULL.
* Same for tp_reg->id because they have the same id.
*/
/* oops. verifier did not complain about the missing !tp check */
snd_cwnd = tp->snd_cwnd;
Hence, a new "ref_obj_id" is needed in "struct bpf_reg_state".
With a new ref_obj_id, when bpf_sk_release(sk) is called, the verifier can
scrub all reg states which has a ref_obj_id match. It is done with the
changes in release_reg_references() in this patch.
While fixing it, sk_to_full_sk() is removed from bpf_tcp_sock() and
bpf_sk_fullsock() to avoid these helpers from returning
another ptr. It will make bpf_sk_release(tp) possible:
sk = bpf_sk_lookup_tcp();
/* if (!sk) ... */
tp = bpf_tcp_sock(sk);
/* if (!tp) ... */
bpf_sk_release(tp);
A separate helper "bpf_get_listener_sock()" will be added in a later
patch to do sk_to_full_sk().
Misc change notes:
- To allow bpf_sk_release(tp), the arg of bpf_sk_release() is changed
from ARG_PTR_TO_SOCKET to ARG_PTR_TO_SOCK_COMMON. ARG_PTR_TO_SOCKET
is removed from bpf.h since no helper is using it.
- arg_type_is_refcounted() is renamed to arg_type_may_be_refcounted()
because ARG_PTR_TO_SOCK_COMMON is the only one and skb->sk is not
refcounted. All bpf_sk_release(), bpf_sk_fullsock() and bpf_tcp_sock()
take ARG_PTR_TO_SOCK_COMMON.
- check_refcount_ok() ensures is_acquire_function() cannot take
arg_type_may_be_refcounted() as its argument.
- The check_func_arg() can only allow one refcount-ed arg. It is
guaranteed by check_refcount_ok() which ensures at most one arg can be
refcounted. Hence, it is a verifier internal error if >1 refcount arg
found in check_func_arg().
- In release_reference(), release_reference_state() is called
first to ensure a match on "reg->ref_obj_id" can be found before
scrubbing the reg states with release_reg_references().
- reg_is_refcounted() is no longer needed.
1. In mark_ptr_or_null_regs(), its usage is replaced by
"ref_obj_id && ref_obj_id == id" because,
when is_null == true, release_reference_state() should only be
called on the ref_obj_id obtained by a acquire helper (i.e.
is_acquire_function() == true). Otherwise, the following
would happen:
sk = bpf_sk_lookup_tcp();
/* if (!sk) { ... } */
fullsock = bpf_sk_fullsock(sk);
if (!fullsock) {
/*
* release_reference_state(fullsock_reg->ref_obj_id)
* where fullsock_reg->ref_obj_id == sk_reg->ref_obj_id.
*
* Hence, the following bpf_sk_release(sk) will fail
* because the ref state has already been released in the
* earlier release_reference_state(fullsock_reg->ref_obj_id).
*/
bpf_sk_release(sk);
}
2. In release_reg_references(), the current reg_is_refcounted() call
is unnecessary because the id check is enough.
- The type_is_refcounted() and type_is_refcounted_or_null()
are no longer needed also because reg_is_refcounted() is removed.
Fixes: 655a51e536c0 ("bpf: Add struct bpf_tcp_sock and BPF_FUNC_tcp_sock")
Reported-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-03-12 17:23:02 +00:00
|
|
|
/* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
|
|
|
|
|
* from a pointer-cast helper, bpf_sk_fullsock() and
|
|
|
|
|
* bpf_tcp_sock().
|
|
|
|
|
*
|
|
|
|
|
* Consider the following where "sk" is a reference counted
|
|
|
|
|
* pointer returned from "sk = bpf_sk_lookup_tcp();":
|
|
|
|
|
*
|
|
|
|
|
* 1: sk = bpf_sk_lookup_tcp();
|
|
|
|
|
* 2: if (!sk) { return 0; }
|
|
|
|
|
* 3: fullsock = bpf_sk_fullsock(sk);
|
|
|
|
|
* 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
|
|
|
|
|
* 5: tp = bpf_tcp_sock(fullsock);
|
|
|
|
|
* 6: if (!tp) { bpf_sk_release(sk); return 0; }
|
|
|
|
|
* 7: bpf_sk_release(sk);
|
|
|
|
|
* 8: snd_cwnd = tp->snd_cwnd; // verifier will complain
|
|
|
|
|
*
|
|
|
|
|
* After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
|
|
|
|
|
* "tp" ptr should be invalidated also. In order to do that,
|
|
|
|
|
* the reg holding "fullsock" and "sk" need to remember
|
|
|
|
|
* the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
|
|
|
|
|
* such that the verifier can reset all regs which have
|
|
|
|
|
* ref_obj_id matching the sk_reg->id.
|
|
|
|
|
*
|
|
|
|
|
* sk_reg->ref_obj_id is set to sk_reg->id at line 1.
|
|
|
|
|
* sk_reg->id will stay as NULL-marking purpose only.
|
|
|
|
|
* After NULL-marking is done, sk_reg->id can be reset to 0.
|
|
|
|
|
*
|
|
|
|
|
* After "fullsock = bpf_sk_fullsock(sk);" at line 3,
|
|
|
|
|
* fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
|
|
|
|
|
*
|
|
|
|
|
* After "tp = bpf_tcp_sock(fullsock);" at line 5,
|
|
|
|
|
* tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
|
|
|
|
|
* which is the same as sk_reg->ref_obj_id.
|
|
|
|
|
*
|
|
|
|
|
* From the verifier perspective, if sk, fullsock and tp
|
|
|
|
|
* are not NULL, they are the same ptr with different
|
|
|
|
|
* reg->type. In particular, bpf_sk_release(tp) is also
|
|
|
|
|
* allowed and has the same effect as bpf_sk_release(sk).
|
|
|
|
|
*/
|
|
|
|
|
u32 ref_obj_id;
|
2018-08-22 19:02:19 +00:00
|
|
|
/* parentage chain for liveness checking */
|
|
|
|
|
struct bpf_reg_state *parent;
|
2017-12-15 01:55:06 +00:00
|
|
|
/* Inside the callee two registers can be both PTR_TO_STACK like
|
|
|
|
|
* R1=fp-8 and R2=fp-8, but one of them points to this function stack
|
|
|
|
|
* while another to the caller's stack. To differentiate them 'frameno'
|
|
|
|
|
* is used which is an index in bpf_verifier_state->frame[] array
|
|
|
|
|
* pointing to bpf_func_state.
|
|
|
|
|
*/
|
|
|
|
|
u32 frameno;
|
bpf: verifier: mark verified-insn with sub-register zext flag
eBPF ISA specification requires high 32-bit cleared when low 32-bit
sub-register is written. This applies to destination register of ALU32 etc.
JIT back-ends must guarantee this semantic when doing code-gen. x86_64 and
AArch64 ISA has the same semantics, so the corresponding JIT back-end
doesn't need to do extra work.
However, 32-bit arches (arm, x86, nfp etc.) and some other 64-bit arches
(PowerPC, SPARC etc) need to do explicit zero extension to meet this
requirement, otherwise code like the following will fail.
u64_value = (u64) u32_value
... other uses of u64_value
This is because compiler could exploit the semantic described above and
save those zero extensions for extending u32_value to u64_value, these JIT
back-ends are expected to guarantee this through inserting extra zero
extensions which however could be a significant increase on the code size.
Some benchmarks show there could be ~40% sub-register writes out of total
insns, meaning at least ~40% extra code-gen.
One observation is these extra zero extensions are not always necessary.
Take above code snippet for example, it is possible u32_value will never be
casted into a u64, the value of high 32-bit of u32_value then could be
ignored and extra zero extension could be eliminated.
This patch implements this idea, insns defining sub-registers will be
marked when the high 32-bit of the defined sub-register matters. For
those unmarked insns, it is safe to eliminate high 32-bit clearnace for
them.
Algo:
- Split read flags into READ32 and READ64.
- Record index of insn that does sub-register write. Keep the index inside
reg state and update it during verifier insn walking.
- A full register read on a sub-register marks its definition insn as
needing zero extension on dst register.
A new sub-register write overrides the old one.
- When propagating read64 during path pruning, also mark any insn defining
a sub-register that is read in the pruned path as full-register.
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-05-24 22:25:12 +00:00
|
|
|
/* Tracks subreg definition. The stored value is the insn_idx of the
|
|
|
|
|
* writing insn. This is safe because subreg_def is used before any insn
|
|
|
|
|
* patching which only happens after main verification finished.
|
|
|
|
|
*/
|
|
|
|
|
s32 subreg_def;
|
2017-08-15 19:34:35 +00:00
|
|
|
enum bpf_reg_liveness live;
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-15 19:12:25 +00:00
|
|
|
/* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
|
|
|
|
|
bool precise;
|
2016-09-21 10:43:57 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
enum bpf_stack_slot_type {
|
|
|
|
|
STACK_INVALID, /* nothing was stored in this stack slot */
|
|
|
|
|
STACK_SPILL, /* register spilled into stack */
|
2017-12-15 01:55:08 +00:00
|
|
|
STACK_MISC, /* BPF program wrote some data into this slot */
|
|
|
|
|
STACK_ZERO, /* BPF program wrote constant zero */
|
bpf: Add verifier support for dynptrs
This patch adds the bulk of the verifier work for supporting dynamic
pointers (dynptrs) in bpf.
A bpf_dynptr is opaque to the bpf program. It is a 16-byte structure
defined internally as:
struct bpf_dynptr_kern {
void *data;
u32 size;
u32 offset;
} __aligned(8);
The upper 8 bits of *size* is reserved (it contains extra metadata about
read-only status and dynptr type). Consequently, a dynptr only supports
memory less than 16 MB.
There are different types of dynptrs (eg malloc, ringbuf, ...). In this
patchset, the most basic one, dynptrs to a bpf program's local memory,
is added. For now only local memory that is of reg type PTR_TO_MAP_VALUE
is supported.
In the verifier, dynptr state information will be tracked in stack
slots. When the program passes in an uninitialized dynptr
(ARG_PTR_TO_DYNPTR | MEM_UNINIT), the stack slots corresponding
to the frame pointer where the dynptr resides at are marked
STACK_DYNPTR. For helper functions that take in initialized dynptrs (eg
bpf_dynptr_read + bpf_dynptr_write which are added later in this
patchset), the verifier enforces that the dynptr has been initialized
properly by checking that their corresponding stack slots have been
marked as STACK_DYNPTR.
The 6th patch in this patchset adds test cases that the verifier should
successfully reject, such as for example attempting to use a dynptr
after doing a direct write into it inside the bpf program.
Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/bpf/20220523210712.3641569-2-joannelkoong@gmail.com
2022-05-23 21:07:07 +00:00
|
|
|
/* A dynptr is stored in this stack slot. The type of dynptr
|
|
|
|
|
* is stored in bpf_stack_state->spilled_ptr.dynptr.type
|
|
|
|
|
*/
|
|
|
|
|
STACK_DYNPTR,
|
bpf: add support for open-coded iterator loops
Teach verifier about the concept of the open-coded (or inline) iterators.
This patch adds generic iterator loop verification logic, new STACK_ITER
stack slot type to contain iterator state, and necessary kfunc plumbing
for iterator's constructor, destructor and next methods. Next patch
implements first specific iterator (numbers iterator for implementing
for() loop logic). Such split allows to have more focused commits for
verifier logic and separate commit that we could point later to
demonstrating what does it take to add a new kind of iterator.
Each kind of iterator has its own associated struct bpf_iter_<type>,
where <type> denotes a specific type of iterator. struct bpf_iter_<type>
state is supposed to live on BPF program stack, so there will be no way
to change its size later on without breaking backwards compatibility, so
choose wisely! But given this struct is specific to a given <type> of
iterator, this allows a lot of flexibility: simple iterators could be
fine with just one stack slot (8 bytes), like numbers iterator in the
next patch, while some other more complicated iterators might need way
more to keep their iterator state. Either way, such design allows to
avoid runtime memory allocations, which otherwise would be necessary if
we fixed on-the-stack size and it turned out to be too small for a given
iterator implementation.
The way BPF verifier logic is implemented, there are no artificial
restrictions on a number of active iterators, it should work correctly
using multiple active iterators at the same time. This also means you
can have multiple nested iteration loops. struct bpf_iter_<type>
reference can be safely passed to subprograms as well.
General flow is easiest to demonstrate with a simple example using
number iterator implemented in next patch. Here's the simplest possible
loop:
struct bpf_iter_num it;
int *v;
bpf_iter_num_new(&it, 2, 5);
while ((v = bpf_iter_num_next(&it))) {
bpf_printk("X = %d", *v);
}
bpf_iter_num_destroy(&it);
Above snippet should output "X = 2", "X = 3", "X = 4". Note that 5 is
exclusive and is not returned. This matches similar APIs (e.g., slices
in Go or Rust) that implement a range of elements, where end index is
non-inclusive.
In the above example, we see a trio of function:
- constructor, bpf_iter_num_new(), which initializes iterator state
(struct bpf_iter_num it) on the stack. If any of the input arguments
are invalid, constructor should make sure to still initialize it such
that subsequent bpf_iter_num_next() calls will return NULL. I.e., on
error, return error and construct empty iterator.
- next method, bpf_iter_num_next(), which accepts pointer to iterator
state and produces an element. Next method should always return
a pointer. The contract between BPF verifier is that next method will
always eventually return NULL when elements are exhausted. Once NULL is
returned, subsequent next calls should keep returning NULL. In the
case of numbers iterator, bpf_iter_num_next() returns a pointer to an int
(storage for this integer is inside the iterator state itself),
which can be dereferenced after corresponding NULL check.
- once done with the iterator, it's mandated that user cleans up its
state with the call to destructor, bpf_iter_num_destroy() in this
case. Destructor frees up any resources and marks stack space used by
struct bpf_iter_num as usable for something else.
Any other iterator implementation will have to implement at least these
three methods. It is enforced that for any given type of iterator only
applicable constructor/destructor/next are callable. I.e., verifier
ensures you can't pass number iterator state into, say, cgroup
iterator's next method.
It is important to keep the naming pattern consistent to be able to
create generic macros to help with BPF iter usability. E.g., one
of the follow up patches adds generic bpf_for_each() macro to bpf_misc.h
in selftests, which allows to utilize iterator "trio" nicely without
having to code the above somewhat tedious loop explicitly every time.
This is enforced at kfunc registration point by one of the previous
patches in this series.
At the implementation level, iterator state tracking for verification
purposes is very similar to dynptr. We add STACK_ITER stack slot type,
reserve necessary number of slots, depending on
sizeof(struct bpf_iter_<type>), and keep track of necessary extra state
in the "main" slot, which is marked with non-zero ref_obj_id. Other
slots are also marked as STACK_ITER, but have zero ref_obj_id. This is
simpler than having a separate "is_first_slot" flag.
Another big distinction is that STACK_ITER is *always refcounted*, which
simplifies implementation without sacrificing usability. So no need for
extra "iter_id", no need to anticipate reuse of STACK_ITER slots for new
constructors, etc. Keeping it simple here.
As far as the verification logic goes, there are two extensive comments:
in process_iter_next_call() and iter_active_depths_differ() explaining
some important and sometimes subtle aspects. Please refer to them for
details.
But from 10,000-foot point of view, next methods are the points of
forking a verification state, which are conceptually similar to what
verifier is doing when validating conditional jump. We branch out at
a `call bpf_iter_<type>_next` instruction and simulate two outcomes:
NULL (iteration is done) and non-NULL (new element is returned). NULL is
simulated first and is supposed to reach exit without looping. After
that non-NULL case is validated and it either reaches exit (for trivial
examples with no real loop), or reaches another `call bpf_iter_<type>_next`
instruction with the state equivalent to already (partially) validated
one. State equivalency at that point means we technically are going to
be looping forever without "breaking out" out of established "state
envelope" (i.e., subsequent iterations don't add any new knowledge or
constraints to the verifier state, so running 1, 2, 10, or a million of
them doesn't matter). But taking into account the contract stating that
iterator next method *has to* return NULL eventually, we can conclude
that loop body is safe and will eventually terminate. Given we validated
logic outside of the loop (NULL case), and concluded that loop body is
safe (though potentially looping many times), verifier can claim safety
of the overall program logic.
The rest of the patch is necessary plumbing for state tracking, marking,
validation, and necessary further kfunc plumbing to allow implementing
iterator constructor, destructor, and next methods.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230308184121.1165081-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 18:41:16 +00:00
|
|
|
STACK_ITER,
|
bpf: Introduce support for bpf_local_irq_{save,restore}
Teach the verifier about IRQ-disabled sections through the introduction
of two new kfuncs, bpf_local_irq_save, to save IRQ state and disable
them, and bpf_local_irq_restore, to restore IRQ state and enable them
back again.
For the purposes of tracking the saved IRQ state, the verifier is taught
about a new special object on the stack of type STACK_IRQ_FLAG. This is
a 8 byte value which saves the IRQ flags which are to be passed back to
the IRQ restore kfunc.
Renumber the enums for REF_TYPE_* to simplify the check in
find_lock_state, filtering out non-lock types as they grow will become
cumbersome and is unecessary.
To track a dynamic number of IRQ-disabled regions and their associated
saved states, a new resource type RES_TYPE_IRQ is introduced, which its
state management functions: acquire_irq_state and release_irq_state,
taking advantage of the refactoring and clean ups made in earlier
commits.
One notable requirement of the kernel's IRQ save and restore API is that
they cannot happen out of order. For this purpose, when releasing reference
we keep track of the prev_id we saw with REF_TYPE_IRQ. Since reference
states are inserted in increasing order of the index, this is used to
remember the ordering of acquisitions of IRQ saved states, so that we
maintain a logical stack in acquisition order of resource identities,
and can enforce LIFO ordering when restoring IRQ state. The top of the
stack is maintained using bpf_verifier_state's active_irq_id.
To maintain the stack property when releasing reference states, we need
to modify release_reference_state to instead shift the remaining array
left using memmove instead of swapping deleted element with last that
might break the ordering. A selftest to test this subtle behavior is
added in late patches.
The logic to detect initialized and unitialized irq flag slots, marking
and unmarking is similar to how it's done for iterators. No additional
checks are needed in refsafe for REF_TYPE_IRQ, apart from the usual
check_id satisfiability check on the ref[i].id. We have to perform the
same check_ids check on state->active_irq_id as well.
To ensure we don't get assigned REF_TYPE_PTR by default after
acquire_reference_state, if someone forgets to assign the type, let's
also renumber the enum ref_state_type. This way any unassigned types
get caught by refsafe's default switch statement, don't assume
REF_TYPE_PTR by default.
The kfuncs themselves are plain wrappers over local_irq_save and
local_irq_restore macros.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20241204030400.208005-5-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2024-12-04 03:03:57 +00:00
|
|
|
STACK_IRQ_FLAG,
|
2016-09-21 10:43:57 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
#define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
|
bpf: add support for open-coded iterator loops
Teach verifier about the concept of the open-coded (or inline) iterators.
This patch adds generic iterator loop verification logic, new STACK_ITER
stack slot type to contain iterator state, and necessary kfunc plumbing
for iterator's constructor, destructor and next methods. Next patch
implements first specific iterator (numbers iterator for implementing
for() loop logic). Such split allows to have more focused commits for
verifier logic and separate commit that we could point later to
demonstrating what does it take to add a new kind of iterator.
Each kind of iterator has its own associated struct bpf_iter_<type>,
where <type> denotes a specific type of iterator. struct bpf_iter_<type>
state is supposed to live on BPF program stack, so there will be no way
to change its size later on without breaking backwards compatibility, so
choose wisely! But given this struct is specific to a given <type> of
iterator, this allows a lot of flexibility: simple iterators could be
fine with just one stack slot (8 bytes), like numbers iterator in the
next patch, while some other more complicated iterators might need way
more to keep their iterator state. Either way, such design allows to
avoid runtime memory allocations, which otherwise would be necessary if
we fixed on-the-stack size and it turned out to be too small for a given
iterator implementation.
The way BPF verifier logic is implemented, there are no artificial
restrictions on a number of active iterators, it should work correctly
using multiple active iterators at the same time. This also means you
can have multiple nested iteration loops. struct bpf_iter_<type>
reference can be safely passed to subprograms as well.
General flow is easiest to demonstrate with a simple example using
number iterator implemented in next patch. Here's the simplest possible
loop:
struct bpf_iter_num it;
int *v;
bpf_iter_num_new(&it, 2, 5);
while ((v = bpf_iter_num_next(&it))) {
bpf_printk("X = %d", *v);
}
bpf_iter_num_destroy(&it);
Above snippet should output "X = 2", "X = 3", "X = 4". Note that 5 is
exclusive and is not returned. This matches similar APIs (e.g., slices
in Go or Rust) that implement a range of elements, where end index is
non-inclusive.
In the above example, we see a trio of function:
- constructor, bpf_iter_num_new(), which initializes iterator state
(struct bpf_iter_num it) on the stack. If any of the input arguments
are invalid, constructor should make sure to still initialize it such
that subsequent bpf_iter_num_next() calls will return NULL. I.e., on
error, return error and construct empty iterator.
- next method, bpf_iter_num_next(), which accepts pointer to iterator
state and produces an element. Next method should always return
a pointer. The contract between BPF verifier is that next method will
always eventually return NULL when elements are exhausted. Once NULL is
returned, subsequent next calls should keep returning NULL. In the
case of numbers iterator, bpf_iter_num_next() returns a pointer to an int
(storage for this integer is inside the iterator state itself),
which can be dereferenced after corresponding NULL check.
- once done with the iterator, it's mandated that user cleans up its
state with the call to destructor, bpf_iter_num_destroy() in this
case. Destructor frees up any resources and marks stack space used by
struct bpf_iter_num as usable for something else.
Any other iterator implementation will have to implement at least these
three methods. It is enforced that for any given type of iterator only
applicable constructor/destructor/next are callable. I.e., verifier
ensures you can't pass number iterator state into, say, cgroup
iterator's next method.
It is important to keep the naming pattern consistent to be able to
create generic macros to help with BPF iter usability. E.g., one
of the follow up patches adds generic bpf_for_each() macro to bpf_misc.h
in selftests, which allows to utilize iterator "trio" nicely without
having to code the above somewhat tedious loop explicitly every time.
This is enforced at kfunc registration point by one of the previous
patches in this series.
At the implementation level, iterator state tracking for verification
purposes is very similar to dynptr. We add STACK_ITER stack slot type,
reserve necessary number of slots, depending on
sizeof(struct bpf_iter_<type>), and keep track of necessary extra state
in the "main" slot, which is marked with non-zero ref_obj_id. Other
slots are also marked as STACK_ITER, but have zero ref_obj_id. This is
simpler than having a separate "is_first_slot" flag.
Another big distinction is that STACK_ITER is *always refcounted*, which
simplifies implementation without sacrificing usability. So no need for
extra "iter_id", no need to anticipate reuse of STACK_ITER slots for new
constructors, etc. Keeping it simple here.
As far as the verification logic goes, there are two extensive comments:
in process_iter_next_call() and iter_active_depths_differ() explaining
some important and sometimes subtle aspects. Please refer to them for
details.
But from 10,000-foot point of view, next methods are the points of
forking a verification state, which are conceptually similar to what
verifier is doing when validating conditional jump. We branch out at
a `call bpf_iter_<type>_next` instruction and simulate two outcomes:
NULL (iteration is done) and non-NULL (new element is returned). NULL is
simulated first and is supposed to reach exit without looping. After
that non-NULL case is validated and it either reaches exit (for trivial
examples with no real loop), or reaches another `call bpf_iter_<type>_next`
instruction with the state equivalent to already (partially) validated
one. State equivalency at that point means we technically are going to
be looping forever without "breaking out" out of established "state
envelope" (i.e., subsequent iterations don't add any new knowledge or
constraints to the verifier state, so running 1, 2, 10, or a million of
them doesn't matter). But taking into account the contract stating that
iterator next method *has to* return NULL eventually, we can conclude
that loop body is safe and will eventually terminate. Given we validated
logic outside of the loop (NULL case), and concluded that loop body is
safe (though potentially looping many times), verifier can claim safety
of the overall program logic.
The rest of the patch is necessary plumbing for state tracking, marking,
validation, and necessary further kfunc plumbing to allow implementing
iterator constructor, destructor, and next methods.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230308184121.1165081-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 18:41:16 +00:00
|
|
|
|
2023-05-05 04:33:10 +00:00
|
|
|
#define BPF_REGMASK_ARGS ((1 << BPF_REG_1) | (1 << BPF_REG_2) | \
|
|
|
|
|
(1 << BPF_REG_3) | (1 << BPF_REG_4) | \
|
|
|
|
|
(1 << BPF_REG_5))
|
|
|
|
|
|
bpf: Add verifier support for dynptrs
This patch adds the bulk of the verifier work for supporting dynamic
pointers (dynptrs) in bpf.
A bpf_dynptr is opaque to the bpf program. It is a 16-byte structure
defined internally as:
struct bpf_dynptr_kern {
void *data;
u32 size;
u32 offset;
} __aligned(8);
The upper 8 bits of *size* is reserved (it contains extra metadata about
read-only status and dynptr type). Consequently, a dynptr only supports
memory less than 16 MB.
There are different types of dynptrs (eg malloc, ringbuf, ...). In this
patchset, the most basic one, dynptrs to a bpf program's local memory,
is added. For now only local memory that is of reg type PTR_TO_MAP_VALUE
is supported.
In the verifier, dynptr state information will be tracked in stack
slots. When the program passes in an uninitialized dynptr
(ARG_PTR_TO_DYNPTR | MEM_UNINIT), the stack slots corresponding
to the frame pointer where the dynptr resides at are marked
STACK_DYNPTR. For helper functions that take in initialized dynptrs (eg
bpf_dynptr_read + bpf_dynptr_write which are added later in this
patchset), the verifier enforces that the dynptr has been initialized
properly by checking that their corresponding stack slots have been
marked as STACK_DYNPTR.
The 6th patch in this patchset adds test cases that the verifier should
successfully reject, such as for example attempting to use a dynptr
after doing a direct write into it inside the bpf program.
Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/bpf/20220523210712.3641569-2-joannelkoong@gmail.com
2022-05-23 21:07:07 +00:00
|
|
|
#define BPF_DYNPTR_SIZE sizeof(struct bpf_dynptr_kern)
|
|
|
|
|
#define BPF_DYNPTR_NR_SLOTS (BPF_DYNPTR_SIZE / BPF_REG_SIZE)
|
2016-09-21 10:43:57 +00:00
|
|
|
|
2017-11-01 01:16:05 +00:00
|
|
|
struct bpf_stack_state {
|
|
|
|
|
struct bpf_reg_state spilled_ptr;
|
|
|
|
|
u8 slot_type[BPF_REG_SIZE];
|
|
|
|
|
};
|
|
|
|
|
|
2018-10-02 20:35:35 +00:00
|
|
|
struct bpf_reference_state {
|
2024-11-09 23:14:29 +00:00
|
|
|
/* Each reference object has a type. Ensure REF_TYPE_PTR is zero to
|
|
|
|
|
* default to pointer reference on zero initialization of a state.
|
|
|
|
|
*/
|
|
|
|
|
enum ref_state_type {
|
bpf: Implement verifier support for rqspinlock
Introduce verifier-side support for rqspinlock kfuncs. The first step is
allowing bpf_res_spin_lock type to be defined in map values and
allocated objects, so BTF-side is updated with a new BPF_RES_SPIN_LOCK
field to recognize and validate.
Any object cannot have both bpf_spin_lock and bpf_res_spin_lock, only
one of them (and at most one of them per-object, like before) must be
present. The bpf_res_spin_lock can also be used to protect objects that
require lock protection for their kfuncs, like BPF rbtree and linked
list.
The verifier plumbing to simulate success and failure cases when calling
the kfuncs is done by pushing a new verifier state to the verifier state
stack which will verify the failure case upon calling the kfunc. The
path where success is indicated creates all lock reference state and IRQ
state (if necessary for irqsave variants). In the case of failure, the
state clears the registers r0-r5, sets the return value, and skips kfunc
processing, proceeding to the next instruction.
When marking the return value for success case, the value is marked as
0, and for the failure case as [-MAX_ERRNO, -1]. Then, in the program,
whenever user checks the return value as 'if (ret)' or 'if (ret < 0)'
the verifier never traverses such branches for success cases, and would
be aware that the lock is not held in such cases.
We push the kfunc state in check_kfunc_call whenever rqspinlock kfuncs
are invoked. We introduce a kfunc_class state to avoid mixing lock
irqrestore kfuncs with IRQ state created by bpf_local_irq_save.
With all this infrastructure, these kfuncs become usable in programs
while satisfying all safety properties required by the kernel.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20250316040541.108729-24-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-03-16 04:05:39 +00:00
|
|
|
REF_TYPE_PTR = (1 << 1),
|
|
|
|
|
REF_TYPE_IRQ = (1 << 2),
|
|
|
|
|
REF_TYPE_LOCK = (1 << 3),
|
|
|
|
|
REF_TYPE_RES_LOCK = (1 << 4),
|
|
|
|
|
REF_TYPE_RES_LOCK_IRQ = (1 << 5),
|
2025-03-16 04:05:40 +00:00
|
|
|
REF_TYPE_LOCK_MASK = REF_TYPE_LOCK | REF_TYPE_RES_LOCK | REF_TYPE_RES_LOCK_IRQ,
|
2024-11-09 23:14:29 +00:00
|
|
|
} type;
|
2018-10-02 20:35:35 +00:00
|
|
|
/* Track each reference created with a unique id, even if the same
|
|
|
|
|
* instruction creates the reference multiple times (eg, via CALL).
|
|
|
|
|
*/
|
|
|
|
|
int id;
|
|
|
|
|
/* Instruction where the allocation of this reference occurred. This
|
|
|
|
|
* is used purely to inform the user of a reference leak.
|
|
|
|
|
*/
|
|
|
|
|
int insn_idx;
|
2024-11-09 23:14:30 +00:00
|
|
|
/* Use to keep track of the source object of a lock, to ensure
|
|
|
|
|
* it matches on unlock.
|
|
|
|
|
*/
|
|
|
|
|
void *ptr;
|
2018-10-02 20:35:35 +00:00
|
|
|
};
|
|
|
|
|
|
2023-12-02 17:56:58 +00:00
|
|
|
struct bpf_retval_range {
|
|
|
|
|
s32 minval;
|
|
|
|
|
s32 maxval;
|
|
|
|
|
};
|
|
|
|
|
|
2016-09-21 10:43:57 +00:00
|
|
|
/* state of the program:
|
|
|
|
|
* type of all registers and stack info
|
|
|
|
|
*/
|
2017-12-15 01:55:06 +00:00
|
|
|
struct bpf_func_state {
|
2016-09-21 10:43:57 +00:00
|
|
|
struct bpf_reg_state regs[MAX_BPF_REG];
|
2017-12-15 01:55:06 +00:00
|
|
|
/* index of call instruction that called into this func */
|
|
|
|
|
int callsite;
|
|
|
|
|
/* stack frame number of this function state from pov of
|
|
|
|
|
* enclosing bpf_verifier_state.
|
|
|
|
|
* 0 = main function, 1 = first callee.
|
|
|
|
|
*/
|
|
|
|
|
u32 frameno;
|
bpf: Allow variable-offset stack access
Before this patch, variable offset access to the stack was dissalowed
for regular instructions, but was allowed for "indirect" accesses (i.e.
helpers). This patch removes the restriction, allowing reading and
writing to the stack through stack pointers with variable offsets. This
makes stack-allocated buffers more usable in programs, and brings stack
pointers closer to other types of pointers.
The motivation is being able to use stack-allocated buffers for data
manipulation. When the stack size limit is sufficient, allocating
buffers on the stack is simpler than per-cpu arrays, or other
alternatives.
In unpriviledged programs, variable-offset reads and writes are
disallowed (they were already disallowed for the indirect access case)
because the speculative execution checking code doesn't support them.
Additionally, when writing through a variable-offset stack pointer, if
any pointers are in the accessible range, there's possilibities of later
leaking pointers because the write cannot be tracked precisely.
Writes with variable offset mark the whole range as initialized, even
though we don't know which stack slots are actually written. This is in
order to not reject future reads to these slots. Note that this doesn't
affect writes done through helpers; like before, helpers need the whole
stack range to be initialized to begin with.
All the stack slots are in range are considered scalars after the write;
variable-offset register spills are not tracked.
For reads, all the stack slots in the variable range needs to be
initialized (but see above about what writes do), otherwise the read is
rejected. All register spilled in stack slots that might be read are
marked as having been read, however reads through such pointers don't do
register filling; the target register will always be either a scalar or
a constant zero.
Signed-off-by: Andrei Matei <andreimatei1@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210207011027.676572-2-andreimatei1@gmail.com
2021-02-07 01:10:24 +00:00
|
|
|
/* subprog number == index within subprog_info
|
2017-12-15 01:55:06 +00:00
|
|
|
* zero == main subprog
|
|
|
|
|
*/
|
|
|
|
|
u32 subprogno;
|
2021-07-15 00:54:14 +00:00
|
|
|
/* Every bpf_timer_start will increment async_entry_cnt.
|
|
|
|
|
* It's used to distinguish:
|
|
|
|
|
* void foo(void) { for(;;); }
|
|
|
|
|
* void foo(void) { bpf_timer_set_callback(,foo); }
|
|
|
|
|
*/
|
|
|
|
|
u32 async_entry_cnt;
|
2023-12-02 17:56:58 +00:00
|
|
|
struct bpf_retval_range callback_ret_range;
|
bpf: rearrange bpf_func_state fields to save a bit of memory
It's a trivial rearrangement saving 8 bytes. We have 4 bytes of padding
at the end which can be filled with another field without increasing
struct bpf_func_state.
copy_func_state() logic remains correct without any further changes.
BEFORE
======
struct bpf_func_state {
struct bpf_reg_state regs[11]; /* 0 1320 */
/* --- cacheline 20 boundary (1280 bytes) was 40 bytes ago --- */
int callsite; /* 1320 4 */
u32 frameno; /* 1324 4 */
u32 subprogno; /* 1328 4 */
u32 async_entry_cnt; /* 1332 4 */
bool in_callback_fn; /* 1336 1 */
/* XXX 7 bytes hole, try to pack */
/* --- cacheline 21 boundary (1344 bytes) --- */
struct tnum callback_ret_range; /* 1344 16 */
bool in_async_callback_fn; /* 1360 1 */
bool in_exception_callback_fn; /* 1361 1 */
/* XXX 2 bytes hole, try to pack */
int acquired_refs; /* 1364 4 */
struct bpf_reference_state * refs; /* 1368 8 */
int allocated_stack; /* 1376 4 */
/* XXX 4 bytes hole, try to pack */
struct bpf_stack_state * stack; /* 1384 8 */
/* size: 1392, cachelines: 22, members: 13 */
/* sum members: 1379, holes: 3, sum holes: 13 */
/* last cacheline: 48 bytes */
};
AFTER
=====
struct bpf_func_state {
struct bpf_reg_state regs[11]; /* 0 1320 */
/* --- cacheline 20 boundary (1280 bytes) was 40 bytes ago --- */
int callsite; /* 1320 4 */
u32 frameno; /* 1324 4 */
u32 subprogno; /* 1328 4 */
u32 async_entry_cnt; /* 1332 4 */
struct tnum callback_ret_range; /* 1336 16 */
/* --- cacheline 21 boundary (1344 bytes) was 8 bytes ago --- */
bool in_callback_fn; /* 1352 1 */
bool in_async_callback_fn; /* 1353 1 */
bool in_exception_callback_fn; /* 1354 1 */
/* XXX 1 byte hole, try to pack */
int acquired_refs; /* 1356 4 */
struct bpf_reference_state * refs; /* 1360 8 */
struct bpf_stack_state * stack; /* 1368 8 */
int allocated_stack; /* 1376 4 */
/* size: 1384, cachelines: 22, members: 13 */
/* sum members: 1379, holes: 1, sum holes: 1 */
/* padding: 4 */
/* last cacheline: 40 bytes */
};
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231202175705.885270-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-02 17:56:55 +00:00
|
|
|
bool in_callback_fn;
|
2021-07-15 00:54:14 +00:00
|
|
|
bool in_async_callback_fn;
|
bpf: Add support for custom exception callbacks
By default, the subprog generated by the verifier to handle a thrown
exception hardcodes a return value of 0. To allow user-defined logic
and modification of the return value when an exception is thrown,
introduce the 'exception_callback:' declaration tag, which marks a
callback as the default exception handler for the program.
The format of the declaration tag is 'exception_callback:<value>', where
<value> is the name of the exception callback. Each main program can be
tagged using this BTF declaratiion tag to associate it with an exception
callback. In case the tag is absent, the default callback is used.
As such, the exception callback cannot be modified at runtime, only set
during verification.
Allowing modification of the callback for the current program execution
at runtime leads to issues when the programs begin to nest, as any
per-CPU state maintaing this information will have to be saved and
restored. We don't want it to stay in bpf_prog_aux as this takes a
global effect for all programs. An alternative solution is spilling
the callback pointer at a known location on the program stack on entry,
and then passing this location to bpf_throw as a parameter.
However, since exceptions are geared more towards a use case where they
are ideally never invoked, optimizing for this use case and adding to
the complexity has diminishing returns.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20230912233214.1518551-7-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-09-12 23:32:03 +00:00
|
|
|
bool in_exception_callback_fn;
|
2023-11-21 02:07:00 +00:00
|
|
|
/* For callback calling functions that limit number of possible
|
|
|
|
|
* callback executions (e.g. bpf_loop) keeps track of current
|
|
|
|
|
* simulated iteration number.
|
|
|
|
|
* Value in frame N refers to number of times callback with frame
|
|
|
|
|
* N+1 was simulated, e.g. for the following call:
|
|
|
|
|
*
|
|
|
|
|
* bpf_loop(..., fn, ...); | suppose current frame is N
|
|
|
|
|
* | fn would be simulated in frame N+1
|
|
|
|
|
* | number of simulations is tracked in frame N
|
|
|
|
|
*/
|
|
|
|
|
u32 callback_depth;
|
2017-12-15 01:55:06 +00:00
|
|
|
|
2018-10-02 20:35:35 +00:00
|
|
|
/* The following fields should be last. See copy_func_state() */
|
2023-12-08 03:25:17 +00:00
|
|
|
/* The state of the stack. Each element of the array describes BPF_REG_SIZE
|
|
|
|
|
* (i.e. 8) bytes worth of stack memory.
|
|
|
|
|
* stack[0] represents bytes [*(r10-8)..*(r10-1)]
|
|
|
|
|
* stack[1] represents bytes [*(r10-16)..*(r10-9)]
|
|
|
|
|
* ...
|
|
|
|
|
* stack[allocated_stack/8 - 1] represents [*(r10-allocated_stack)..*(r10-allocated_stack+7)]
|
|
|
|
|
*/
|
2017-11-01 01:16:05 +00:00
|
|
|
struct bpf_stack_state *stack;
|
2023-12-08 03:25:17 +00:00
|
|
|
/* Size of the current stack, in bytes. The stack state is tracked below, in
|
|
|
|
|
* `stack`. allocated_stack is always a multiple of BPF_REG_SIZE.
|
|
|
|
|
*/
|
bpf: rearrange bpf_func_state fields to save a bit of memory
It's a trivial rearrangement saving 8 bytes. We have 4 bytes of padding
at the end which can be filled with another field without increasing
struct bpf_func_state.
copy_func_state() logic remains correct without any further changes.
BEFORE
======
struct bpf_func_state {
struct bpf_reg_state regs[11]; /* 0 1320 */
/* --- cacheline 20 boundary (1280 bytes) was 40 bytes ago --- */
int callsite; /* 1320 4 */
u32 frameno; /* 1324 4 */
u32 subprogno; /* 1328 4 */
u32 async_entry_cnt; /* 1332 4 */
bool in_callback_fn; /* 1336 1 */
/* XXX 7 bytes hole, try to pack */
/* --- cacheline 21 boundary (1344 bytes) --- */
struct tnum callback_ret_range; /* 1344 16 */
bool in_async_callback_fn; /* 1360 1 */
bool in_exception_callback_fn; /* 1361 1 */
/* XXX 2 bytes hole, try to pack */
int acquired_refs; /* 1364 4 */
struct bpf_reference_state * refs; /* 1368 8 */
int allocated_stack; /* 1376 4 */
/* XXX 4 bytes hole, try to pack */
struct bpf_stack_state * stack; /* 1384 8 */
/* size: 1392, cachelines: 22, members: 13 */
/* sum members: 1379, holes: 3, sum holes: 13 */
/* last cacheline: 48 bytes */
};
AFTER
=====
struct bpf_func_state {
struct bpf_reg_state regs[11]; /* 0 1320 */
/* --- cacheline 20 boundary (1280 bytes) was 40 bytes ago --- */
int callsite; /* 1320 4 */
u32 frameno; /* 1324 4 */
u32 subprogno; /* 1328 4 */
u32 async_entry_cnt; /* 1332 4 */
struct tnum callback_ret_range; /* 1336 16 */
/* --- cacheline 21 boundary (1344 bytes) was 8 bytes ago --- */
bool in_callback_fn; /* 1352 1 */
bool in_async_callback_fn; /* 1353 1 */
bool in_exception_callback_fn; /* 1354 1 */
/* XXX 1 byte hole, try to pack */
int acquired_refs; /* 1356 4 */
struct bpf_reference_state * refs; /* 1360 8 */
struct bpf_stack_state * stack; /* 1368 8 */
int allocated_stack; /* 1376 4 */
/* size: 1384, cachelines: 22, members: 13 */
/* sum members: 1379, holes: 1, sum holes: 1 */
/* padding: 4 */
/* last cacheline: 40 bytes */
};
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231202175705.885270-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-02 17:56:55 +00:00
|
|
|
int allocated_stack;
|
2016-09-21 10:43:57 +00:00
|
|
|
};
|
|
|
|
|
|
bpf: support non-r10 register spill/fill to/from stack in precision tracking
Use instruction (jump) history to record instructions that performed
register spill/fill to/from stack, regardless if this was done through
read-only r10 register, or any other register after copying r10 into it
*and* potentially adjusting offset.
To make this work reliably, we push extra per-instruction flags into
instruction history, encoding stack slot index (spi) and stack frame
number in extra 10 bit flags we take away from prev_idx in instruction
history. We don't touch idx field for maximum performance, as it's
checked most frequently during backtracking.
This change removes basically the last remaining practical limitation of
precision backtracking logic in BPF verifier. It fixes known
deficiencies, but also opens up new opportunities to reduce number of
verified states, explored in the subsequent patches.
There are only three differences in selftests' BPF object files
according to veristat, all in the positive direction (less states).
File Program Insns (A) Insns (B) Insns (DIFF) States (A) States (B) States (DIFF)
-------------------------------------- ------------- --------- --------- ------------- ---------- ---------- -------------
test_cls_redirect_dynptr.bpf.linked3.o cls_redirect 2987 2864 -123 (-4.12%) 240 231 -9 (-3.75%)
xdp_synproxy_kern.bpf.linked3.o syncookie_tc 82848 82661 -187 (-0.23%) 5107 5073 -34 (-0.67%)
xdp_synproxy_kern.bpf.linked3.o syncookie_xdp 85116 84964 -152 (-0.18%) 5162 5130 -32 (-0.62%)
Note, I avoided renaming jmp_history to more generic insn_hist to
minimize number of lines changed and potential merge conflicts between
bpf and bpf-next trees.
Notice also cur_hist_entry pointer reset to NULL at the beginning of
instruction verification loop. This pointer avoids the problem of
relying on last jump history entry's insn_idx to determine whether we
already have entry for current instruction or not. It can happen that we
added jump history entry because current instruction is_jmp_point(), but
also we need to add instruction flags for stack access. In this case, we
don't want to entries, so we need to reuse last added entry, if it is
present.
Relying on insn_idx comparison has the same ambiguity problem as the one
that was fixed recently in [0], so we avoid that.
[0] https://patchwork.kernel.org/project/netdevbpf/patch/20231110002638.4168352-3-andrii@kernel.org/
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Reported-by: Tao Lyu <tao.lyu@epfl.ch>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231205184248.1502704-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-05 18:42:39 +00:00
|
|
|
#define MAX_CALL_FRAMES 8
|
|
|
|
|
|
2025-06-11 20:08:26 +00:00
|
|
|
/* instruction history flags, used in bpf_jmp_history_entry.flags field */
|
bpf: support non-r10 register spill/fill to/from stack in precision tracking
Use instruction (jump) history to record instructions that performed
register spill/fill to/from stack, regardless if this was done through
read-only r10 register, or any other register after copying r10 into it
*and* potentially adjusting offset.
To make this work reliably, we push extra per-instruction flags into
instruction history, encoding stack slot index (spi) and stack frame
number in extra 10 bit flags we take away from prev_idx in instruction
history. We don't touch idx field for maximum performance, as it's
checked most frequently during backtracking.
This change removes basically the last remaining practical limitation of
precision backtracking logic in BPF verifier. It fixes known
deficiencies, but also opens up new opportunities to reduce number of
verified states, explored in the subsequent patches.
There are only three differences in selftests' BPF object files
according to veristat, all in the positive direction (less states).
File Program Insns (A) Insns (B) Insns (DIFF) States (A) States (B) States (DIFF)
-------------------------------------- ------------- --------- --------- ------------- ---------- ---------- -------------
test_cls_redirect_dynptr.bpf.linked3.o cls_redirect 2987 2864 -123 (-4.12%) 240 231 -9 (-3.75%)
xdp_synproxy_kern.bpf.linked3.o syncookie_tc 82848 82661 -187 (-0.23%) 5107 5073 -34 (-0.67%)
xdp_synproxy_kern.bpf.linked3.o syncookie_xdp 85116 84964 -152 (-0.18%) 5162 5130 -32 (-0.62%)
Note, I avoided renaming jmp_history to more generic insn_hist to
minimize number of lines changed and potential merge conflicts between
bpf and bpf-next trees.
Notice also cur_hist_entry pointer reset to NULL at the beginning of
instruction verification loop. This pointer avoids the problem of
relying on last jump history entry's insn_idx to determine whether we
already have entry for current instruction or not. It can happen that we
added jump history entry because current instruction is_jmp_point(), but
also we need to add instruction flags for stack access. In this case, we
don't want to entries, so we need to reuse last added entry, if it is
present.
Relying on insn_idx comparison has the same ambiguity problem as the one
that was fixed recently in [0], so we avoid that.
[0] https://patchwork.kernel.org/project/netdevbpf/patch/20231110002638.4168352-3-andrii@kernel.org/
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Reported-by: Tao Lyu <tao.lyu@epfl.ch>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231205184248.1502704-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-05 18:42:39 +00:00
|
|
|
enum {
|
|
|
|
|
/* instruction references stack slot through PTR_TO_STACK register;
|
|
|
|
|
* we also store stack's frame number in lower 3 bits (MAX_CALL_FRAMES is 8)
|
|
|
|
|
* and accessed stack slot's index in next 6 bits (MAX_BPF_STACK is 512,
|
|
|
|
|
* 8 bytes per slot, so slot index (spi) is [0, 63])
|
|
|
|
|
*/
|
|
|
|
|
INSN_F_FRAMENO_MASK = 0x7, /* 3 bits */
|
|
|
|
|
|
|
|
|
|
INSN_F_SPI_MASK = 0x3f, /* 6 bits */
|
|
|
|
|
INSN_F_SPI_SHIFT = 3, /* shifted 3 bits to the left */
|
|
|
|
|
|
bpf: Do not include stack ptr register in precision backtracking bookkeeping
Yi Lai reported an issue ([1]) where the following warning appears
in kernel dmesg:
[ 60.643604] verifier backtracking bug
[ 60.643635] WARNING: CPU: 10 PID: 2315 at kernel/bpf/verifier.c:4302 __mark_chain_precision+0x3a6c/0x3e10
[ 60.648428] Modules linked in: bpf_testmod(OE)
[ 60.650471] CPU: 10 UID: 0 PID: 2315 Comm: test_progs Tainted: G OE 6.15.0-rc4-gef11287f8289-dirty #327 PREEMPT(full)
[ 60.654385] Tainted: [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
[ 60.656682] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014
[ 60.660475] RIP: 0010:__mark_chain_precision+0x3a6c/0x3e10
[ 60.662814] Code: 5a 30 84 89 ea e8 c4 d9 01 00 80 3d 3e 7d d8 04 00 0f 85 60 fa ff ff c6 05 31 7d d8 04
01 48 c7 c7 00 58 30 84 e8 c4 06 a5 ff <0f> 0b e9 46 fa ff ff 48 ...
[ 60.668720] RSP: 0018:ffff888116cc7298 EFLAGS: 00010246
[ 60.671075] RAX: 54d70e82dfd31900 RBX: ffff888115b65e20 RCX: 0000000000000000
[ 60.673659] RDX: 0000000000000001 RSI: 0000000000000004 RDI: 00000000ffffffff
[ 60.676241] RBP: 0000000000000400 R08: ffff8881f6f23bd3 R09: 1ffff1103ede477a
[ 60.678787] R10: dffffc0000000000 R11: ffffed103ede477b R12: ffff888115b60ae8
[ 60.681420] R13: 1ffff11022b6cbc4 R14: 00000000fffffff2 R15: 0000000000000001
[ 60.684030] FS: 00007fc2aedd80c0(0000) GS:ffff88826fa8a000(0000) knlGS:0000000000000000
[ 60.686837] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 60.689027] CR2: 000056325369e000 CR3: 000000011088b002 CR4: 0000000000370ef0
[ 60.691623] Call Trace:
[ 60.692821] <TASK>
[ 60.693960] ? __pfx_verbose+0x10/0x10
[ 60.695656] ? __pfx_disasm_kfunc_name+0x10/0x10
[ 60.697495] check_cond_jmp_op+0x16f7/0x39b0
[ 60.699237] do_check+0x58fa/0xab10
...
Further analysis shows the warning is at line 4302 as below:
4294 /* static subprog call instruction, which
4295 * means that we are exiting current subprog,
4296 * so only r1-r5 could be still requested as
4297 * precise, r0 and r6-r10 or any stack slot in
4298 * the current frame should be zero by now
4299 */
4300 if (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) {
4301 verbose(env, "BUG regs %x\n", bt_reg_mask(bt));
4302 WARN_ONCE(1, "verifier backtracking bug");
4303 return -EFAULT;
4304 }
With the below test (also in the next patch):
__used __naked static void __bpf_jmp_r10(void)
{
asm volatile (
"r2 = 2314885393468386424 ll;"
"goto +0;"
"if r2 <= r10 goto +3;"
"if r1 >= -1835016 goto +0;"
"if r2 <= 8 goto +0;"
"if r3 <= 0 goto +0;"
"exit;"
::: __clobber_all);
}
SEC("?raw_tp")
__naked void bpf_jmp_r10(void)
{
asm volatile (
"r3 = 0 ll;"
"call __bpf_jmp_r10;"
"r0 = 0;"
"exit;"
::: __clobber_all);
}
The following is the verifier failure log:
0: (18) r3 = 0x0 ; R3_w=0
2: (85) call pc+2
caller:
R10=fp0
callee:
frame1: R1=ctx() R3_w=0 R10=fp0
5: frame1: R1=ctx() R3_w=0 R10=fp0
; asm volatile (" \ @ verifier_precision.c:184
5: (18) r2 = 0x20202000256c6c78 ; frame1: R2_w=0x20202000256c6c78
7: (05) goto pc+0
8: (bd) if r2 <= r10 goto pc+3 ; frame1: R2_w=0x20202000256c6c78 R10=fp0
9: (35) if r1 >= 0xffe3fff8 goto pc+0 ; frame1: R1=ctx()
10: (b5) if r2 <= 0x8 goto pc+0
mark_precise: frame1: last_idx 10 first_idx 0 subseq_idx -1
mark_precise: frame1: regs=r2 stack= before 9: (35) if r1 >= 0xffe3fff8 goto pc+0
mark_precise: frame1: regs=r2 stack= before 8: (bd) if r2 <= r10 goto pc+3
mark_precise: frame1: regs=r2,r10 stack= before 7: (05) goto pc+0
mark_precise: frame1: regs=r2,r10 stack= before 5: (18) r2 = 0x20202000256c6c78
mark_precise: frame1: regs=r10 stack= before 2: (85) call pc+2
BUG regs 400
The main failure reason is due to r10 in precision backtracking bookkeeping.
Actually r10 is always precise and there is no need to add it for the precision
backtracking bookkeeping.
One way to fix the issue is to prevent bt_set_reg() if any src/dst reg is
r10. Andrii suggested to go with push_insn_history() approach to avoid
explicitly checking r10 in backtrack_insn().
This patch added push_insn_history() support for cond_jmp like 'rX <op> rY'
operations. In check_cond_jmp_op(), if any of rX or rY is a stack pointer,
push_insn_history() will record such information, and later backtrack_insn()
will do bt_set_reg() properly for those register(s).
[1] https://lore.kernel.org/bpf/Z%2F8q3xzpU59CIYQE@ly-workstation/
Reported by: Yi Lai <yi1.lai@linux.intel.com>
Fixes: 407958a0e980 ("bpf: encapsulate precision backtracking bookkeeping")
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20250524041335.4046126-1-yonghong.song@linux.dev
2025-05-24 04:13:35 +00:00
|
|
|
INSN_F_STACK_ACCESS = BIT(9),
|
|
|
|
|
|
|
|
|
|
INSN_F_DST_REG_STACK = BIT(10), /* dst_reg is PTR_TO_STACK */
|
|
|
|
|
INSN_F_SRC_REG_STACK = BIT(11), /* src_reg is PTR_TO_STACK */
|
|
|
|
|
/* total 12 bits are used now. */
|
bpf: support non-r10 register spill/fill to/from stack in precision tracking
Use instruction (jump) history to record instructions that performed
register spill/fill to/from stack, regardless if this was done through
read-only r10 register, or any other register after copying r10 into it
*and* potentially adjusting offset.
To make this work reliably, we push extra per-instruction flags into
instruction history, encoding stack slot index (spi) and stack frame
number in extra 10 bit flags we take away from prev_idx in instruction
history. We don't touch idx field for maximum performance, as it's
checked most frequently during backtracking.
This change removes basically the last remaining practical limitation of
precision backtracking logic in BPF verifier. It fixes known
deficiencies, but also opens up new opportunities to reduce number of
verified states, explored in the subsequent patches.
There are only three differences in selftests' BPF object files
according to veristat, all in the positive direction (less states).
File Program Insns (A) Insns (B) Insns (DIFF) States (A) States (B) States (DIFF)
-------------------------------------- ------------- --------- --------- ------------- ---------- ---------- -------------
test_cls_redirect_dynptr.bpf.linked3.o cls_redirect 2987 2864 -123 (-4.12%) 240 231 -9 (-3.75%)
xdp_synproxy_kern.bpf.linked3.o syncookie_tc 82848 82661 -187 (-0.23%) 5107 5073 -34 (-0.67%)
xdp_synproxy_kern.bpf.linked3.o syncookie_xdp 85116 84964 -152 (-0.18%) 5162 5130 -32 (-0.62%)
Note, I avoided renaming jmp_history to more generic insn_hist to
minimize number of lines changed and potential merge conflicts between
bpf and bpf-next trees.
Notice also cur_hist_entry pointer reset to NULL at the beginning of
instruction verification loop. This pointer avoids the problem of
relying on last jump history entry's insn_idx to determine whether we
already have entry for current instruction or not. It can happen that we
added jump history entry because current instruction is_jmp_point(), but
also we need to add instruction flags for stack access. In this case, we
don't want to entries, so we need to reuse last added entry, if it is
present.
Relying on insn_idx comparison has the same ambiguity problem as the one
that was fixed recently in [0], so we avoid that.
[0] https://patchwork.kernel.org/project/netdevbpf/patch/20231110002638.4168352-3-andrii@kernel.org/
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Reported-by: Tao Lyu <tao.lyu@epfl.ch>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231205184248.1502704-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-05 18:42:39 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static_assert(INSN_F_FRAMENO_MASK + 1 >= MAX_CALL_FRAMES);
|
|
|
|
|
static_assert(INSN_F_SPI_MASK + 1 >= MAX_BPF_STACK / 8);
|
|
|
|
|
|
2025-06-11 20:08:26 +00:00
|
|
|
struct bpf_jmp_history_entry {
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-15 19:12:25 +00:00
|
|
|
u32 idx;
|
bpf: support non-r10 register spill/fill to/from stack in precision tracking
Use instruction (jump) history to record instructions that performed
register spill/fill to/from stack, regardless if this was done through
read-only r10 register, or any other register after copying r10 into it
*and* potentially adjusting offset.
To make this work reliably, we push extra per-instruction flags into
instruction history, encoding stack slot index (spi) and stack frame
number in extra 10 bit flags we take away from prev_idx in instruction
history. We don't touch idx field for maximum performance, as it's
checked most frequently during backtracking.
This change removes basically the last remaining practical limitation of
precision backtracking logic in BPF verifier. It fixes known
deficiencies, but also opens up new opportunities to reduce number of
verified states, explored in the subsequent patches.
There are only three differences in selftests' BPF object files
according to veristat, all in the positive direction (less states).
File Program Insns (A) Insns (B) Insns (DIFF) States (A) States (B) States (DIFF)
-------------------------------------- ------------- --------- --------- ------------- ---------- ---------- -------------
test_cls_redirect_dynptr.bpf.linked3.o cls_redirect 2987 2864 -123 (-4.12%) 240 231 -9 (-3.75%)
xdp_synproxy_kern.bpf.linked3.o syncookie_tc 82848 82661 -187 (-0.23%) 5107 5073 -34 (-0.67%)
xdp_synproxy_kern.bpf.linked3.o syncookie_xdp 85116 84964 -152 (-0.18%) 5162 5130 -32 (-0.62%)
Note, I avoided renaming jmp_history to more generic insn_hist to
minimize number of lines changed and potential merge conflicts between
bpf and bpf-next trees.
Notice also cur_hist_entry pointer reset to NULL at the beginning of
instruction verification loop. This pointer avoids the problem of
relying on last jump history entry's insn_idx to determine whether we
already have entry for current instruction or not. It can happen that we
added jump history entry because current instruction is_jmp_point(), but
also we need to add instruction flags for stack access. In this case, we
don't want to entries, so we need to reuse last added entry, if it is
present.
Relying on insn_idx comparison has the same ambiguity problem as the one
that was fixed recently in [0], so we avoid that.
[0] https://patchwork.kernel.org/project/netdevbpf/patch/20231110002638.4168352-3-andrii@kernel.org/
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Reported-by: Tao Lyu <tao.lyu@epfl.ch>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231205184248.1502704-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-05 18:42:39 +00:00
|
|
|
/* insn idx can't be bigger than 1 million */
|
bpf: Do not include stack ptr register in precision backtracking bookkeeping
Yi Lai reported an issue ([1]) where the following warning appears
in kernel dmesg:
[ 60.643604] verifier backtracking bug
[ 60.643635] WARNING: CPU: 10 PID: 2315 at kernel/bpf/verifier.c:4302 __mark_chain_precision+0x3a6c/0x3e10
[ 60.648428] Modules linked in: bpf_testmod(OE)
[ 60.650471] CPU: 10 UID: 0 PID: 2315 Comm: test_progs Tainted: G OE 6.15.0-rc4-gef11287f8289-dirty #327 PREEMPT(full)
[ 60.654385] Tainted: [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
[ 60.656682] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014
[ 60.660475] RIP: 0010:__mark_chain_precision+0x3a6c/0x3e10
[ 60.662814] Code: 5a 30 84 89 ea e8 c4 d9 01 00 80 3d 3e 7d d8 04 00 0f 85 60 fa ff ff c6 05 31 7d d8 04
01 48 c7 c7 00 58 30 84 e8 c4 06 a5 ff <0f> 0b e9 46 fa ff ff 48 ...
[ 60.668720] RSP: 0018:ffff888116cc7298 EFLAGS: 00010246
[ 60.671075] RAX: 54d70e82dfd31900 RBX: ffff888115b65e20 RCX: 0000000000000000
[ 60.673659] RDX: 0000000000000001 RSI: 0000000000000004 RDI: 00000000ffffffff
[ 60.676241] RBP: 0000000000000400 R08: ffff8881f6f23bd3 R09: 1ffff1103ede477a
[ 60.678787] R10: dffffc0000000000 R11: ffffed103ede477b R12: ffff888115b60ae8
[ 60.681420] R13: 1ffff11022b6cbc4 R14: 00000000fffffff2 R15: 0000000000000001
[ 60.684030] FS: 00007fc2aedd80c0(0000) GS:ffff88826fa8a000(0000) knlGS:0000000000000000
[ 60.686837] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 60.689027] CR2: 000056325369e000 CR3: 000000011088b002 CR4: 0000000000370ef0
[ 60.691623] Call Trace:
[ 60.692821] <TASK>
[ 60.693960] ? __pfx_verbose+0x10/0x10
[ 60.695656] ? __pfx_disasm_kfunc_name+0x10/0x10
[ 60.697495] check_cond_jmp_op+0x16f7/0x39b0
[ 60.699237] do_check+0x58fa/0xab10
...
Further analysis shows the warning is at line 4302 as below:
4294 /* static subprog call instruction, which
4295 * means that we are exiting current subprog,
4296 * so only r1-r5 could be still requested as
4297 * precise, r0 and r6-r10 or any stack slot in
4298 * the current frame should be zero by now
4299 */
4300 if (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) {
4301 verbose(env, "BUG regs %x\n", bt_reg_mask(bt));
4302 WARN_ONCE(1, "verifier backtracking bug");
4303 return -EFAULT;
4304 }
With the below test (also in the next patch):
__used __naked static void __bpf_jmp_r10(void)
{
asm volatile (
"r2 = 2314885393468386424 ll;"
"goto +0;"
"if r2 <= r10 goto +3;"
"if r1 >= -1835016 goto +0;"
"if r2 <= 8 goto +0;"
"if r3 <= 0 goto +0;"
"exit;"
::: __clobber_all);
}
SEC("?raw_tp")
__naked void bpf_jmp_r10(void)
{
asm volatile (
"r3 = 0 ll;"
"call __bpf_jmp_r10;"
"r0 = 0;"
"exit;"
::: __clobber_all);
}
The following is the verifier failure log:
0: (18) r3 = 0x0 ; R3_w=0
2: (85) call pc+2
caller:
R10=fp0
callee:
frame1: R1=ctx() R3_w=0 R10=fp0
5: frame1: R1=ctx() R3_w=0 R10=fp0
; asm volatile (" \ @ verifier_precision.c:184
5: (18) r2 = 0x20202000256c6c78 ; frame1: R2_w=0x20202000256c6c78
7: (05) goto pc+0
8: (bd) if r2 <= r10 goto pc+3 ; frame1: R2_w=0x20202000256c6c78 R10=fp0
9: (35) if r1 >= 0xffe3fff8 goto pc+0 ; frame1: R1=ctx()
10: (b5) if r2 <= 0x8 goto pc+0
mark_precise: frame1: last_idx 10 first_idx 0 subseq_idx -1
mark_precise: frame1: regs=r2 stack= before 9: (35) if r1 >= 0xffe3fff8 goto pc+0
mark_precise: frame1: regs=r2 stack= before 8: (bd) if r2 <= r10 goto pc+3
mark_precise: frame1: regs=r2,r10 stack= before 7: (05) goto pc+0
mark_precise: frame1: regs=r2,r10 stack= before 5: (18) r2 = 0x20202000256c6c78
mark_precise: frame1: regs=r10 stack= before 2: (85) call pc+2
BUG regs 400
The main failure reason is due to r10 in precision backtracking bookkeeping.
Actually r10 is always precise and there is no need to add it for the precision
backtracking bookkeeping.
One way to fix the issue is to prevent bt_set_reg() if any src/dst reg is
r10. Andrii suggested to go with push_insn_history() approach to avoid
explicitly checking r10 in backtrack_insn().
This patch added push_insn_history() support for cond_jmp like 'rX <op> rY'
operations. In check_cond_jmp_op(), if any of rX or rY is a stack pointer,
push_insn_history() will record such information, and later backtrack_insn()
will do bt_set_reg() properly for those register(s).
[1] https://lore.kernel.org/bpf/Z%2F8q3xzpU59CIYQE@ly-workstation/
Reported by: Yi Lai <yi1.lai@linux.intel.com>
Fixes: 407958a0e980 ("bpf: encapsulate precision backtracking bookkeeping")
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20250524041335.4046126-1-yonghong.song@linux.dev
2025-05-24 04:13:35 +00:00
|
|
|
u32 prev_idx : 20;
|
|
|
|
|
/* special INSN_F_xxx flags */
|
|
|
|
|
u32 flags : 12;
|
bpf: Track equal scalars history on per-instruction level
Use bpf_verifier_state->jmp_history to track which registers were
updated by find_equal_scalars() (renamed to collect_linked_regs())
when conditional jump was verified. Use recorded information in
backtrack_insn() to propagate precision.
E.g. for the following program:
while verifying instructions
1: r1 = r0 |
2: if r1 < 8 goto ... | push r0,r1 as linked registers in jmp_history
3: if r0 > 16 goto ... | push r0,r1 as linked registers in jmp_history
4: r2 = r10 |
5: r2 += r0 v mark_chain_precision(r0)
while doing mark_chain_precision(r0)
5: r2 += r0 | mark r0 precise
4: r2 = r10 |
3: if r0 > 16 goto ... | mark r0,r1 as precise
2: if r1 < 8 goto ... | mark r0,r1 as precise
1: r1 = r0 v
Technically, do this as follows:
- Use 10 bits to identify each register that gains range because of
sync_linked_regs():
- 3 bits for frame number;
- 6 bits for register or stack slot number;
- 1 bit to indicate if register is spilled.
- Use u64 as a vector of 6 such records + 4 bits for vector length.
- Augment struct bpf_jmp_history_entry with a field 'linked_regs'
representing such vector.
- When doing check_cond_jmp_op() remember up to 6 registers that
gain range because of sync_linked_regs() in such a vector.
- Don't propagate range information and reset IDs for registers that
don't fit in 6-value vector.
- Push a pair {instruction index, linked registers vector}
to bpf_verifier_state->jmp_history.
- When doing backtrack_insn() check if any of recorded linked
registers is currently marked precise, if so mark all linked
registers as precise.
This also requires fixes for two test_verifier tests:
- precise: test 1
- precise: test 2
Both tests contain the following instruction sequence:
19: (bf) r2 = r9 ; R2=scalar(id=3) R9=scalar(id=3)
20: (a5) if r2 < 0x8 goto pc+1 ; R2=scalar(id=3,umin=8)
21: (95) exit
22: (07) r2 += 1 ; R2_w=scalar(id=3+1,...)
23: (bf) r1 = r10 ; R1_w=fp0 R10=fp0
24: (07) r1 += -8 ; R1_w=fp-8
25: (b7) r3 = 0 ; R3_w=0
26: (85) call bpf_probe_read_kernel#113
The call to bpf_probe_read_kernel() at (26) forces r2 to be precise.
Previously, this forced all registers with same id to become precise
immediately when mark_chain_precision() is called.
After this change, the precision is propagated to registers sharing
same id only when 'if' instruction is backtracked.
Hence verification log for both tests is changed:
regs=r2,r9 -> regs=r2 for instructions 25..20.
Fixes: 904e6ddf4133 ("bpf: Use scalar ids in mark_chain_precision()")
Reported-by: Hao Sun <sunhao.th@gmail.com>
Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20240718202357.1746514-2-eddyz87@gmail.com
Closes: https://lore.kernel.org/bpf/CAEf4BzZ0xidVCqB47XnkXcNhkPWF6_nTV7yt+_Lf0kcFEut2Mg@mail.gmail.com/
2024-07-18 20:23:53 +00:00
|
|
|
/* additional registers that need precision tracking when this
|
|
|
|
|
* jump is backtracked, vector of six 10-bit records
|
|
|
|
|
*/
|
|
|
|
|
u64 linked_regs;
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-15 19:12:25 +00:00
|
|
|
};
|
|
|
|
|
|
2022-12-09 13:57:29 +00:00
|
|
|
/* Maximum number of register states that can exist at once */
|
|
|
|
|
#define BPF_ID_MAP_SIZE ((MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) * MAX_CALL_FRAMES)
|
2017-12-15 01:55:06 +00:00
|
|
|
struct bpf_verifier_state {
|
|
|
|
|
/* call stack tracking */
|
|
|
|
|
struct bpf_func_state *frame[MAX_CALL_FRAMES];
|
2019-06-15 19:12:20 +00:00
|
|
|
struct bpf_verifier_state *parent;
|
2024-12-04 03:03:54 +00:00
|
|
|
/* Acquired reference states */
|
|
|
|
|
struct bpf_reference_state *refs;
|
2019-06-15 19:12:20 +00:00
|
|
|
/*
|
|
|
|
|
* 'branches' field is the number of branches left to explore:
|
|
|
|
|
* 0 - all possible paths from this state reached bpf_exit or
|
|
|
|
|
* were safely pruned
|
|
|
|
|
* 1 - at least one path is being explored.
|
|
|
|
|
* This state hasn't reached bpf_exit
|
|
|
|
|
* 2 - at least two paths are being explored.
|
|
|
|
|
* This state is an immediate parent of two children.
|
|
|
|
|
* One is fallthrough branch with branches==1 and another
|
|
|
|
|
* state is pushed into stack (to be explored later) also with
|
|
|
|
|
* branches==1. The parent of this state has branches==1.
|
|
|
|
|
* The verifier state tree connected via 'parent' pointer looks like:
|
|
|
|
|
* 1
|
|
|
|
|
* 1
|
|
|
|
|
* 2 -> 1 (first 'if' pushed into stack)
|
|
|
|
|
* 1
|
|
|
|
|
* 2 -> 1 (second 'if' pushed into stack)
|
|
|
|
|
* 1
|
|
|
|
|
* 1
|
|
|
|
|
* 1 bpf_exit.
|
|
|
|
|
*
|
|
|
|
|
* Once do_check() reaches bpf_exit, it calls update_branch_counts()
|
|
|
|
|
* and the verifier state tree will look:
|
|
|
|
|
* 1
|
|
|
|
|
* 1
|
|
|
|
|
* 2 -> 1 (first 'if' pushed into stack)
|
|
|
|
|
* 1
|
|
|
|
|
* 1 -> 1 (second 'if' pushed into stack)
|
|
|
|
|
* 0
|
|
|
|
|
* 0
|
|
|
|
|
* 0 bpf_exit.
|
|
|
|
|
* After pop_stack() the do_check() will resume at second 'if'.
|
|
|
|
|
*
|
|
|
|
|
* If is_state_visited() sees a state with branches > 0 it means
|
|
|
|
|
* there is a loop. If such state is exactly equal to the current state
|
|
|
|
|
* it's an infinite loop. Note states_equal() checks for states
|
2022-06-13 21:16:33 +00:00
|
|
|
* equivalency, so two states being 'states_equal' does not mean
|
2019-06-15 19:12:20 +00:00
|
|
|
* infinite loop. The exact comparison is provided by
|
|
|
|
|
* states_maybe_looping() function. It's a stronger pre-check and
|
|
|
|
|
* much faster than states_equal().
|
|
|
|
|
*
|
|
|
|
|
* This algorithm may not find all possible infinite loops or
|
|
|
|
|
* loop iteration count may be too high.
|
|
|
|
|
* In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
|
|
|
|
|
*/
|
|
|
|
|
u32 branches;
|
2019-05-22 03:17:07 +00:00
|
|
|
u32 insn_idx;
|
2017-12-15 01:55:06 +00:00
|
|
|
u32 curframe;
|
bpf: Migrate release_on_unlock logic to non-owning ref semantics
This patch introduces non-owning reference semantics to the verifier,
specifically linked_list API kfunc handling. release_on_unlock logic for
refs is refactored - with small functional changes - to implement these
semantics, and bpf_list_push_{front,back} are migrated to use them.
When a list node is pushed to a list, the program still has a pointer to
the node:
n = bpf_obj_new(typeof(*n));
bpf_spin_lock(&l);
bpf_list_push_back(&l, n);
/* n still points to the just-added node */
bpf_spin_unlock(&l);
What the verifier considers n to be after the push, and thus what can be
done with n, are changed by this patch.
Common properties both before/after this patch:
* After push, n is only a valid reference to the node until end of
critical section
* After push, n cannot be pushed to any list
* After push, the program can read the node's fields using n
Before:
* After push, n retains the ref_obj_id which it received on
bpf_obj_new, but the associated bpf_reference_state's
release_on_unlock field is set to true
* release_on_unlock field and associated logic is used to implement
"n is only a valid ref until end of critical section"
* After push, n cannot be written to, the node must be removed from
the list before writing to its fields
* After push, n is marked PTR_UNTRUSTED
After:
* After push, n's ref is released and ref_obj_id set to 0. NON_OWN_REF
type flag is added to reg's type, indicating that it's a non-owning
reference.
* NON_OWN_REF flag and logic is used to implement "n is only a
valid ref until end of critical section"
* n can be written to (except for special fields e.g. bpf_list_node,
timer, ...)
Summary of specific implementation changes to achieve the above:
* release_on_unlock field, ref_set_release_on_unlock helper, and logic
to "release on unlock" based on that field are removed
* The anonymous active_lock struct used by bpf_verifier_state is
pulled out into a named struct bpf_active_lock.
* NON_OWN_REF type flag is introduced along with verifier logic
changes to handle non-owning refs
* Helpers are added to use NON_OWN_REF flag to implement non-owning
ref semantics as described above
* invalidate_non_owning_refs - helper to clobber all non-owning refs
matching a particular bpf_active_lock identity. Replaces
release_on_unlock logic in process_spin_lock.
* ref_set_non_owning - set NON_OWN_REF type flag after doing some
sanity checking
* ref_convert_owning_non_owning - convert owning reference w/
specified ref_obj_id to non-owning references. Set NON_OWN_REF
flag for each reg with that ref_obj_id and 0-out its ref_obj_id
* Update linked_list selftests to account for minor semantic
differences introduced by this patch
* Writes to a release_on_unlock node ref are not allowed, while
writes to non-owning reference pointees are. As a result the
linked_list "write after push" failure tests are no longer scenarios
that should fail.
* The test##missing_lock##op and test##incorrect_lock##op
macro-generated failure tests need to have a valid node argument in
order to have the same error output as before. Otherwise
verification will fail early and the expected error output won't be seen.
Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230212092715.1422619-2-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-02-12 09:27:07 +00:00
|
|
|
|
2024-12-04 03:03:54 +00:00
|
|
|
u32 acquired_refs;
|
|
|
|
|
u32 active_locks;
|
|
|
|
|
u32 active_preempt_locks;
|
bpf: Introduce support for bpf_local_irq_{save,restore}
Teach the verifier about IRQ-disabled sections through the introduction
of two new kfuncs, bpf_local_irq_save, to save IRQ state and disable
them, and bpf_local_irq_restore, to restore IRQ state and enable them
back again.
For the purposes of tracking the saved IRQ state, the verifier is taught
about a new special object on the stack of type STACK_IRQ_FLAG. This is
a 8 byte value which saves the IRQ flags which are to be passed back to
the IRQ restore kfunc.
Renumber the enums for REF_TYPE_* to simplify the check in
find_lock_state, filtering out non-lock types as they grow will become
cumbersome and is unecessary.
To track a dynamic number of IRQ-disabled regions and their associated
saved states, a new resource type RES_TYPE_IRQ is introduced, which its
state management functions: acquire_irq_state and release_irq_state,
taking advantage of the refactoring and clean ups made in earlier
commits.
One notable requirement of the kernel's IRQ save and restore API is that
they cannot happen out of order. For this purpose, when releasing reference
we keep track of the prev_id we saw with REF_TYPE_IRQ. Since reference
states are inserted in increasing order of the index, this is used to
remember the ordering of acquisitions of IRQ saved states, so that we
maintain a logical stack in acquisition order of resource identities,
and can enforce LIFO ordering when restoring IRQ state. The top of the
stack is maintained using bpf_verifier_state's active_irq_id.
To maintain the stack property when releasing reference states, we need
to modify release_reference_state to instead shift the remaining array
left using memmove instead of swapping deleted element with last that
might break the ordering. A selftest to test this subtle behavior is
added in late patches.
The logic to detect initialized and unitialized irq flag slots, marking
and unmarking is similar to how it's done for iterators. No additional
checks are needed in refsafe for REF_TYPE_IRQ, apart from the usual
check_id satisfiability check on the ref[i].id. We have to perform the
same check_ids check on state->active_irq_id as well.
To ensure we don't get assigned REF_TYPE_PTR by default after
acquire_reference_state, if someone forgets to assign the type, let's
also renumber the enum ref_state_type. This way any unassigned types
get caught by refsafe's default switch statement, don't assume
REF_TYPE_PTR by default.
The kfuncs themselves are plain wrappers over local_irq_save and
local_irq_restore macros.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20241204030400.208005-5-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2024-12-04 03:03:57 +00:00
|
|
|
u32 active_irq_id;
|
2025-03-16 04:05:40 +00:00
|
|
|
u32 active_lock_id;
|
|
|
|
|
void *active_lock_ptr;
|
2022-11-24 05:32:17 +00:00
|
|
|
bool active_rcu_lock;
|
2024-12-04 03:03:54 +00:00
|
|
|
|
|
|
|
|
bool speculative;
|
2024-04-20 09:09:13 +00:00
|
|
|
bool in_sleepable;
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-15 19:12:25 +00:00
|
|
|
|
|
|
|
|
/* first and last insn idx of this verifier state */
|
|
|
|
|
u32 first_insn_idx;
|
|
|
|
|
u32 last_insn_idx;
|
bpf: propagate read/precision marks over state graph backedges
Current loop_entry-based exact states comparison logic does not handle
the following case:
.-> A --. Assume the states are visited in the order A, B, C.
| | | Assume that state B reaches a state equivalent to state A.
| v v At this point, state C is not processed yet, so state A
'-- B C has not received any read or precision marks from C.
As a result, these marks won't be propagated to B.
If B has incomplete marks, it is unsafe to use it in states_equal()
checks.
This commit replaces the existing logic with the following:
- Strongly connected components (SCCs) are computed over the program's
control flow graph (intraprocedurally).
- When a verifier state enters an SCC, that state is recorded as the
SCC entry point.
- When a verifier state is found equivalent to another (e.g., B to A
in the example), it is recorded as a states graph backedge.
Backedges are accumulated per SCC.
- When an SCC entry state reaches `branches == 0`, read and precision
marks are propagated through the backedges (e.g., from A to B, from
C to A, and then again from A to B).
To support nested subprogram calls, the entry state and backedge list
are associated not with the SCC itself but with an object called
`bpf_scc_callchain`. A callchain is a tuple `(callsite*, scc_id)`,
where `callsite` is the index of a call instruction for each frame
except the last.
See the comments added in `is_state_visited()` and
`compute_scc_callchain()` for more details.
Fixes: 2a0992829ea3 ("bpf: correct loop detection for iterators convergence")
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250611200836.4135542-8-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-06-11 20:08:33 +00:00
|
|
|
/* if this state is a backedge state then equal_state
|
|
|
|
|
* records cached state to which this state is equal.
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_verifier_state *equal_state;
|
2025-06-11 20:08:26 +00:00
|
|
|
/* jmp history recorded from first to last.
|
|
|
|
|
* backtracking is using it to go from last to first.
|
|
|
|
|
* For most states jmp_history_cnt is [0-3].
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-15 19:12:25 +00:00
|
|
|
* For loops can go up to ~40.
|
|
|
|
|
*/
|
2025-06-11 20:08:26 +00:00
|
|
|
struct bpf_jmp_history_entry *jmp_history;
|
|
|
|
|
u32 jmp_history_cnt;
|
bpf: exact states comparison for iterator convergence checks
Convergence for open coded iterators is computed in is_state_visited()
by examining states with branches count > 1 and using states_equal().
states_equal() computes sub-state relation using read and precision marks.
Read and precision marks are propagated from children states,
thus are not guaranteed to be complete inside a loop when branches
count > 1. This could be demonstrated using the following unsafe program:
1. r7 = -16
2. r6 = bpf_get_prandom_u32()
3. while (bpf_iter_num_next(&fp[-8])) {
4. if (r6 != 42) {
5. r7 = -32
6. r6 = bpf_get_prandom_u32()
7. continue
8. }
9. r0 = r10
10. r0 += r7
11. r8 = *(u64 *)(r0 + 0)
12. r6 = bpf_get_prandom_u32()
13. }
Here verifier would first visit path 1-3, create a checkpoint at 3
with r7=-16, continue to 4-7,3 with r7=-32.
Because instructions at 9-12 had not been visitied yet existing
checkpoint at 3 does not have read or precision mark for r7.
Thus states_equal() would return true and verifier would discard
current state, thus unsafe memory access at 11 would not be caught.
This commit fixes this loophole by introducing exact state comparisons
for iterator convergence logic:
- registers are compared using regs_exact() regardless of read or
precision marks;
- stack slots have to have identical type.
Unfortunately, this is too strict even for simple programs like below:
i = 0;
while(iter_next(&it))
i++;
At each iteration step i++ would produce a new distinct state and
eventually instruction processing limit would be reached.
To avoid such behavior speculatively forget (widen) range for
imprecise scalar registers, if those registers were not precise at the
end of the previous iteration and do not match exactly.
This a conservative heuristic that allows to verify wide range of
programs, however it precludes verification of programs that conjure
an imprecise value on the first loop iteration and use it as precise
on the second.
Test case iter_task_vma_for_each() presents one of such cases:
unsigned int seen = 0;
...
bpf_for_each(task_vma, vma, task, 0) {
if (seen >= 1000)
break;
...
seen++;
}
Here clang generates the following code:
<LBB0_4>:
24: r8 = r6 ; stash current value of
... body ... 'seen'
29: r1 = r10
30: r1 += -0x8
31: call bpf_iter_task_vma_next
32: r6 += 0x1 ; seen++;
33: if r0 == 0x0 goto +0x2 <LBB0_6> ; exit on next() == NULL
34: r7 += 0x10
35: if r8 < 0x3e7 goto -0xc <LBB0_4> ; loop on seen < 1000
<LBB0_6>:
... exit ...
Note that counter in r6 is copied to r8 and then incremented,
conditional jump is done using r8. Because of this precision mark for
r6 lags one state behind of precision mark on r8 and widening logic
kicks in.
Adding barrier_var(seen) after conditional is sufficient to force
clang use the same register for both counting and conditional jump.
This issue was discussed in the thread [1] which was started by
Andrew Werner <awerner32@gmail.com> demonstrating a similar bug
in callback functions handling. The callbacks would be addressed
in a followup patch.
[1] https://lore.kernel.org/bpf/97a90da09404c65c8e810cf83c94ac703705dc0e.camel@gmail.com/
Co-developed-by: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Co-developed-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20231024000917.12153-4-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-10-24 00:09:13 +00:00
|
|
|
u32 dfs_depth;
|
2023-11-21 02:06:56 +00:00
|
|
|
u32 callback_unroll_depth;
|
bpf: Introduce may_goto instruction
Introduce may_goto instruction that from the verifier pov is similar to
open coded iterators bpf_for()/bpf_repeat() and bpf_loop() helper, but it
doesn't iterate any objects.
In assembly 'may_goto' is a nop most of the time until bpf runtime has to
terminate the program for whatever reason. In the current implementation
may_goto has a hidden counter, but other mechanisms can be used.
For programs written in C the later patch introduces 'cond_break' macro
that combines 'may_goto' with 'break' statement and has similar semantics:
cond_break is a nop until bpf runtime has to break out of this loop.
It can be used in any normal "for" or "while" loop, like
for (i = zero; i < cnt; cond_break, i++) {
The verifier recognizes that may_goto is used in the program, reserves
additional 8 bytes of stack, initializes them in subprog prologue, and
replaces may_goto instruction with:
aux_reg = *(u64 *)(fp - 40)
if aux_reg == 0 goto pc+off
aux_reg -= 1
*(u64 *)(fp - 40) = aux_reg
may_goto instruction can be used by LLVM to implement __builtin_memcpy,
__builtin_strcmp.
may_goto is not a full substitute for bpf_for() macro.
bpf_for() doesn't have induction variable that verifiers sees,
so 'i' in bpf_for(i, 0, 100) is seen as imprecise and bounded.
But when the code is written as:
for (i = 0; i < 100; cond_break, i++)
the verifier see 'i' as precise constant zero,
hence cond_break (aka may_goto) doesn't help to converge the loop.
A static or global variable can be used as a workaround:
static int zero = 0;
for (i = zero; i < 100; cond_break, i++) // works!
may_goto works well with arena pointers that don't need to be bounds
checked on access. Load/store from arena returns imprecise unbounded
scalar and loops with may_goto pass the verifier.
Reserve new opcode BPF_JMP | BPF_JCOND for may_goto insn.
JCOND stands for conditional pseudo jump.
Since goto_or_nop insn was proposed, it may use the same opcode.
may_goto vs goto_or_nop can be distinguished by src_reg:
code = BPF_JMP | BPF_JCOND
src_reg = 0 - may_goto
src_reg = 1 - goto_or_nop
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Tested-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20240306031929.42666-2-alexei.starovoitov@gmail.com
2024-03-06 03:19:26 +00:00
|
|
|
u32 may_goto_depth;
|
2017-12-15 01:55:06 +00:00
|
|
|
};
|
|
|
|
|
|
2023-10-18 06:17:43 +00:00
|
|
|
#define bpf_get_spilled_reg(slot, frame, mask) \
|
2018-10-02 20:35:29 +00:00
|
|
|
(((slot < frame->allocated_stack / BPF_REG_SIZE) && \
|
2024-01-08 20:51:58 +00:00
|
|
|
((1 << frame->stack[slot].slot_type[BPF_REG_SIZE - 1]) & (mask))) \
|
2018-10-02 20:35:29 +00:00
|
|
|
? &frame->stack[slot].spilled_ptr : NULL)
|
|
|
|
|
|
|
|
|
|
/* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
|
2023-10-18 06:17:43 +00:00
|
|
|
#define bpf_for_each_spilled_reg(iter, frame, reg, mask) \
|
|
|
|
|
for (iter = 0, reg = bpf_get_spilled_reg(iter, frame, mask); \
|
2018-10-02 20:35:29 +00:00
|
|
|
iter < frame->allocated_stack / BPF_REG_SIZE; \
|
2023-10-18 06:17:43 +00:00
|
|
|
iter++, reg = bpf_get_spilled_reg(iter, frame, mask))
|
2018-10-02 20:35:29 +00:00
|
|
|
|
2023-10-18 06:17:43 +00:00
|
|
|
#define bpf_for_each_reg_in_vstate_mask(__vst, __state, __reg, __mask, __expr) \
|
bpf: Add helper macro bpf_for_each_reg_in_vstate
For a lot of use cases in future patches, we will want to modify the
state of registers part of some same 'group' (e.g. same ref_obj_id). It
won't just be limited to releasing reference state, but setting a type
flag dynamically based on certain actions, etc.
Hence, we need a way to easily pass a callback to the function that
iterates over all registers in current bpf_verifier_state in all frames
upto (and including) the curframe.
While in C++ we would be able to easily use a lambda to pass state and
the callback together, sadly we aren't using C++ in the kernel. The next
best thing to avoid defining a function for each case seems like
statement expressions in GNU C. The kernel already uses them heavily,
hence they can passed to the macro in the style of a lambda. The
statement expression will then be substituted in the for loop bodies.
Variables __state and __reg are set to current bpf_func_state and reg
for each invocation of the expression inside the passed in verifier
state.
Then, convert mark_ptr_or_null_regs, clear_all_pkt_pointers,
release_reference, find_good_pkt_pointers, find_equal_scalars to
use bpf_for_each_reg_in_vstate.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220904204145.3089-16-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2022-09-04 20:41:28 +00:00
|
|
|
({ \
|
|
|
|
|
struct bpf_verifier_state *___vstate = __vst; \
|
|
|
|
|
int ___i, ___j; \
|
|
|
|
|
for (___i = 0; ___i <= ___vstate->curframe; ___i++) { \
|
|
|
|
|
struct bpf_reg_state *___regs; \
|
|
|
|
|
__state = ___vstate->frame[___i]; \
|
|
|
|
|
___regs = __state->regs; \
|
|
|
|
|
for (___j = 0; ___j < MAX_BPF_REG; ___j++) { \
|
|
|
|
|
__reg = &___regs[___j]; \
|
|
|
|
|
(void)(__expr); \
|
|
|
|
|
} \
|
2023-10-18 06:17:43 +00:00
|
|
|
bpf_for_each_spilled_reg(___j, __state, __reg, __mask) { \
|
bpf: Add helper macro bpf_for_each_reg_in_vstate
For a lot of use cases in future patches, we will want to modify the
state of registers part of some same 'group' (e.g. same ref_obj_id). It
won't just be limited to releasing reference state, but setting a type
flag dynamically based on certain actions, etc.
Hence, we need a way to easily pass a callback to the function that
iterates over all registers in current bpf_verifier_state in all frames
upto (and including) the curframe.
While in C++ we would be able to easily use a lambda to pass state and
the callback together, sadly we aren't using C++ in the kernel. The next
best thing to avoid defining a function for each case seems like
statement expressions in GNU C. The kernel already uses them heavily,
hence they can passed to the macro in the style of a lambda. The
statement expression will then be substituted in the for loop bodies.
Variables __state and __reg are set to current bpf_func_state and reg
for each invocation of the expression inside the passed in verifier
state.
Then, convert mark_ptr_or_null_regs, clear_all_pkt_pointers,
release_reference, find_good_pkt_pointers, find_equal_scalars to
use bpf_for_each_reg_in_vstate.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220904204145.3089-16-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2022-09-04 20:41:28 +00:00
|
|
|
if (!__reg) \
|
|
|
|
|
continue; \
|
|
|
|
|
(void)(__expr); \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
})
|
|
|
|
|
|
2023-10-18 06:17:43 +00:00
|
|
|
/* Invoke __expr over regsiters in __vst, setting __state and __reg */
|
|
|
|
|
#define bpf_for_each_reg_in_vstate(__vst, __state, __reg, __expr) \
|
|
|
|
|
bpf_for_each_reg_in_vstate_mask(__vst, __state, __reg, 1 << STACK_SPILL, __expr)
|
|
|
|
|
|
2016-09-21 10:43:57 +00:00
|
|
|
/* linked list of verifier states used to prune search */
|
|
|
|
|
struct bpf_verifier_state_list {
|
|
|
|
|
struct bpf_verifier_state state;
|
2025-02-15 11:03:59 +00:00
|
|
|
struct list_head node;
|
bpf: free verifier states when they are no longer referenced
When fixes from patches 1 and 3 are applied, Patrick Somaru reported
an increase in memory consumption for sched_ext iterator-based
programs hitting 1M instructions limit. For example, 2Gb VMs ran out
of memory while verifying a program. Similar behaviour could be
reproduced on current bpf-next master.
Here is an example of such program:
/* verification completes if given 16G or RAM,
* final env->free_list size is 369,960 entries.
*/
SEC("raw_tp")
__flag(BPF_F_TEST_STATE_FREQ)
__success
int free_list_bomb(const void *ctx)
{
volatile char buf[48] = {};
unsigned i, j;
j = 0;
bpf_for(i, 0, 10) {
/* this forks verifier state:
* - verification of current path continues and
* creates a checkpoint after 'if';
* - verification of forked path hits the
* checkpoint and marks it as loop_entry.
*/
if (bpf_get_prandom_u32())
asm volatile ("");
/* this marks 'j' as precise, thus any checkpoint
* created on current iteration would not be matched
* on the next iteration.
*/
buf[j++] = 42;
j %= ARRAY_SIZE(buf);
}
asm volatile (""::"r"(buf));
return 0;
}
Memory consumption increased due to more states being marked as loop
entries and eventually added to env->free_list.
This commit introduces logic to free states from env->free_list during
verification. A state in env->free_list can be freed if:
- it has no child states;
- it is not used as a loop_entry.
This commit:
- updates bpf_verifier_state->used_as_loop_entry to be a counter
that tracks how many states use this one as a loop entry;
- adds a function maybe_free_verifier_state(), which:
- frees a state if its ->branches and ->used_as_loop_entry counters
are both zero;
- if the state is freed, state->loop_entry->used_as_loop_entry is
decremented, and an attempt is made to free state->loop_entry.
In the example above, this approach reduces the maximum number of
states in the free list from 369,960 to 16,223.
However, this approach has its limitations. If the buf size in the
example above is modified to 64, state caching overflows: the state
for j=0 is evicted from the cache before it can be used to stop
traversal. As a result, states in the free list accumulate because
their branch counters do not reach zero.
The effect of this patch on the selftests looks as follows:
File Program Max free list (A) Max free list (B) Max free list (DIFF)
-------------------------------- ------------------------------------ ----------------- ----------------- --------------------
arena_list.bpf.o arena_list_add 17 3 -14 (-82.35%)
bpf_iter_task_stack.bpf.o dump_task_stack 39 9 -30 (-76.92%)
iters.bpf.o checkpoint_states_deletion 265 89 -176 (-66.42%)
iters.bpf.o clean_live_states 19 0 -19 (-100.00%)
profiler2.bpf.o tracepoint__syscalls__sys_enter_kill 102 1 -101 (-99.02%)
profiler3.bpf.o tracepoint__syscalls__sys_enter_kill 144 0 -144 (-100.00%)
pyperf600_iter.bpf.o on_event 15 0 -15 (-100.00%)
pyperf600_nounroll.bpf.o on_event 1170 1158 -12 (-1.03%)
setget_sockopt.bpf.o skops_sockopt 18 0 -18 (-100.00%)
strobemeta_nounroll1.bpf.o on_event 147 83 -64 (-43.54%)
strobemeta_nounroll2.bpf.o on_event 312 209 -103 (-33.01%)
strobemeta_subprogs.bpf.o on_event 124 86 -38 (-30.65%)
test_cls_redirect_subprogs.bpf.o cls_redirect 15 0 -15 (-100.00%)
timer.bpf.o test1 30 15 -15 (-50.00%)
Measured using "do-not-submit" patches from here:
https://github.com/eddyz87/bpf/tree/get-loop-entry-hungup
Reported-by: Patrick Somaru <patsomaru@meta.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250215110411.3236773-10-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-02-15 11:04:00 +00:00
|
|
|
u32 miss_cnt;
|
|
|
|
|
u32 hit_cnt:31;
|
|
|
|
|
u32 in_free_list:1;
|
2016-09-21 10:43:57 +00:00
|
|
|
};
|
|
|
|
|
|
2022-06-20 23:53:42 +00:00
|
|
|
struct bpf_loop_inline_state {
|
2022-07-11 08:12:00 +00:00
|
|
|
unsigned int initialized:1; /* set to true upon first entry */
|
|
|
|
|
unsigned int fit_for_inline:1; /* true if callback function is the same
|
|
|
|
|
* at each call and flags are always zero
|
|
|
|
|
*/
|
2022-06-20 23:53:42 +00:00
|
|
|
u32 callback_subprogno; /* valid when fit_for_inline is true */
|
|
|
|
|
};
|
|
|
|
|
|
2024-04-05 02:55:34 +00:00
|
|
|
/* pointer and state for maps */
|
|
|
|
|
struct bpf_map_ptr_state {
|
|
|
|
|
struct bpf_map *map_ptr;
|
|
|
|
|
bool poison;
|
|
|
|
|
bool unpriv;
|
|
|
|
|
};
|
|
|
|
|
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-02 23:58:34 +00:00
|
|
|
/* Possible states for alu_state member. */
|
bpf: Fix leakage of uninitialized bpf stack under speculation
The current implemented mechanisms to mitigate data disclosure under
speculation mainly address stack and map value oob access from the
speculative domain. However, Piotr discovered that uninitialized BPF
stack is not protected yet, and thus old data from the kernel stack,
potentially including addresses of kernel structures, could still be
extracted from that 512 bytes large window. The BPF stack is special
compared to map values since it's not zero initialized for every
program invocation, whereas map values /are/ zero initialized upon
their initial allocation and thus cannot leak any prior data in either
domain. In the non-speculative domain, the verifier ensures that every
stack slot read must have a prior stack slot write by the BPF program
to avoid such data leaking issue.
However, this is not enough: for example, when the pointer arithmetic
operation moves the stack pointer from the last valid stack offset to
the first valid offset, the sanitation logic allows for any intermediate
offsets during speculative execution, which could then be used to
extract any restricted stack content via side-channel.
Given for unprivileged stack pointer arithmetic the use of unknown
but bounded scalars is generally forbidden, we can simply turn the
register-based arithmetic operation into an immediate-based arithmetic
operation without the need for masking. This also gives the benefit
of reducing the needed instructions for the operation. Given after
the work in 7fedb63a8307 ("bpf: Tighten speculative pointer arithmetic
mask"), the aux->alu_limit already holds the final immediate value for
the offset register with the known scalar. Thus, a simple mov of the
immediate to AX register with using AX as the source for the original
instruction is sufficient and possible now in this case.
Reported-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
2021-04-29 15:19:37 +00:00
|
|
|
#define BPF_ALU_SANITIZE_SRC (1U << 0)
|
|
|
|
|
#define BPF_ALU_SANITIZE_DST (1U << 1)
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-02 23:58:34 +00:00
|
|
|
#define BPF_ALU_NEG_VALUE (1U << 2)
|
2019-01-05 23:54:37 +00:00
|
|
|
#define BPF_ALU_NON_POINTER (1U << 3)
|
bpf: Fix leakage of uninitialized bpf stack under speculation
The current implemented mechanisms to mitigate data disclosure under
speculation mainly address stack and map value oob access from the
speculative domain. However, Piotr discovered that uninitialized BPF
stack is not protected yet, and thus old data from the kernel stack,
potentially including addresses of kernel structures, could still be
extracted from that 512 bytes large window. The BPF stack is special
compared to map values since it's not zero initialized for every
program invocation, whereas map values /are/ zero initialized upon
their initial allocation and thus cannot leak any prior data in either
domain. In the non-speculative domain, the verifier ensures that every
stack slot read must have a prior stack slot write by the BPF program
to avoid such data leaking issue.
However, this is not enough: for example, when the pointer arithmetic
operation moves the stack pointer from the last valid stack offset to
the first valid offset, the sanitation logic allows for any intermediate
offsets during speculative execution, which could then be used to
extract any restricted stack content via side-channel.
Given for unprivileged stack pointer arithmetic the use of unknown
but bounded scalars is generally forbidden, we can simply turn the
register-based arithmetic operation into an immediate-based arithmetic
operation without the need for masking. This also gives the benefit
of reducing the needed instructions for the operation. Given after
the work in 7fedb63a8307 ("bpf: Tighten speculative pointer arithmetic
mask"), the aux->alu_limit already holds the final immediate value for
the offset register with the known scalar. Thus, a simple mov of the
immediate to AX register with using AX as the source for the original
instruction is sufficient and possible now in this case.
Reported-by: Piotr Krysiuk <piotras@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: Piotr Krysiuk <piotras@gmail.com>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
2021-04-29 15:19:37 +00:00
|
|
|
#define BPF_ALU_IMMEDIATE (1U << 4)
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-02 23:58:34 +00:00
|
|
|
#define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \
|
|
|
|
|
BPF_ALU_SANITIZE_DST)
|
|
|
|
|
|
2016-09-21 10:43:57 +00:00
|
|
|
struct bpf_insn_aux_data {
|
2017-03-16 01:26:42 +00:00
|
|
|
union {
|
|
|
|
|
enum bpf_reg_type ptr_type; /* pointer type for load/store insns */
|
2024-04-05 02:55:34 +00:00
|
|
|
struct bpf_map_ptr_state map_ptr_state;
|
2017-12-15 01:55:15 +00:00
|
|
|
s32 call_imm; /* saved imm field of call insn */
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-02 23:58:34 +00:00
|
|
|
u32 alu_limit; /* limit for add/sub register with pointer */
|
bpf: implement lookup-free direct value access for maps
This generic extension to BPF maps allows for directly loading
an address residing inside a BPF map value as a single BPF
ldimm64 instruction!
The idea is similar to what BPF_PSEUDO_MAP_FD does today, which
is a special src_reg flag for ldimm64 instruction that indicates
that inside the first part of the double insns's imm field is a
file descriptor which the verifier then replaces as a full 64bit
address of the map into both imm parts. For the newly added
BPF_PSEUDO_MAP_VALUE src_reg flag, the idea is the following:
the first part of the double insns's imm field is again a file
descriptor corresponding to the map, and the second part of the
imm field is an offset into the value. The verifier will then
replace both imm parts with an address that points into the BPF
map value at the given value offset for maps that support this
operation. Currently supported is array map with single entry.
It is possible to support more than just single map element by
reusing both 16bit off fields of the insns as a map index, so
full array map lookup could be expressed that way. It hasn't
been implemented here due to lack of concrete use case, but
could easily be done so in future in a compatible way, since
both off fields right now have to be 0 and would correctly
denote a map index 0.
The BPF_PSEUDO_MAP_VALUE is a distinct flag as otherwise with
BPF_PSEUDO_MAP_FD we could not differ offset 0 between load of
map pointer versus load of map's value at offset 0, and changing
BPF_PSEUDO_MAP_FD's encoding into off by one to differ between
regular map pointer and map value pointer would add unnecessary
complexity and increases barrier for debugability thus less
suitable. Using the second part of the imm field as an offset
into the value does /not/ come with limitations since maximum
possible value size is in u32 universe anyway.
This optimization allows for efficiently retrieving an address
to a map value memory area without having to issue a helper call
which needs to prepare registers according to calling convention,
etc, without needing the extra NULL test, and without having to
add the offset in an additional instruction to the value base
pointer. The verifier then treats the destination register as
PTR_TO_MAP_VALUE with constant reg->off from the user passed
offset from the second imm field, and guarantees that this is
within bounds of the map value. Any subsequent operations are
normally treated as typical map value handling without anything
extra needed from verification side.
The two map operations for direct value access have been added to
array map for now. In future other types could be supported as
well depending on the use case. The main use case for this commit
is to allow for BPF loader support for global variables that
reside in .data/.rodata/.bss sections such that we can directly
load the address of them with minimal additional infrastructure
required. Loader support has been added in subsequent commits for
libbpf library.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 21:20:03 +00:00
|
|
|
struct {
|
|
|
|
|
u32 map_index; /* index into used_maps[] */
|
|
|
|
|
u32 map_off; /* offset from value base address */
|
|
|
|
|
};
|
2020-09-29 23:50:44 +00:00
|
|
|
struct {
|
|
|
|
|
enum bpf_reg_type reg_type; /* type of pseudo_btf_id */
|
|
|
|
|
union {
|
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.
This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.
Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.
Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
2020-12-03 20:46:29 +00:00
|
|
|
struct {
|
|
|
|
|
struct btf *btf;
|
|
|
|
|
u32 btf_id; /* btf_id for struct typed var */
|
|
|
|
|
};
|
2020-09-29 23:50:44 +00:00
|
|
|
u32 mem_size; /* mem_size for non-struct typed var */
|
|
|
|
|
};
|
|
|
|
|
} btf_var;
|
2022-06-20 23:53:42 +00:00
|
|
|
/* if instruction is a call to bpf_loop this field tracks
|
|
|
|
|
* the state of the relevant registers to make decision about inlining
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_loop_inline_state loop_inline_state;
|
2017-03-16 01:26:42 +00:00
|
|
|
};
|
bpf: Migrate bpf_rbtree_add and bpf_list_push_{front,back} to possibly fail
Consider this code snippet:
struct node {
long key;
bpf_list_node l;
bpf_rb_node r;
bpf_refcount ref;
}
int some_bpf_prog(void *ctx)
{
struct node *n = bpf_obj_new(/*...*/), *m;
bpf_spin_lock(&glock);
bpf_rbtree_add(&some_tree, &n->r, /* ... */);
m = bpf_refcount_acquire(n);
bpf_rbtree_add(&other_tree, &m->r, /* ... */);
bpf_spin_unlock(&glock);
/* ... */
}
After bpf_refcount_acquire, n and m point to the same underlying memory,
and that node's bpf_rb_node field is being used by the some_tree insert,
so overwriting it as a result of the second insert is an error. In order
to properly support refcounted nodes, the rbtree and list insert
functions must be allowed to fail. This patch adds such support.
The kfuncs bpf_rbtree_add, bpf_list_push_{front,back} are modified to
return an int indicating success/failure, with 0 -> success, nonzero ->
failure.
bpf_obj_drop on failure
=======================
Currently the only reason an insert can fail is the example above: the
bpf_{list,rb}_node is already in use. When such a failure occurs, the
insert kfuncs will bpf_obj_drop the input node. This allows the insert
operations to logically fail without changing their verifier owning ref
behavior, namely the unconditional release_reference of the input
owning ref.
With insert that always succeeds, ownership of the node is always passed
to the collection, since the node always ends up in the collection.
With a possibly-failed insert w/ bpf_obj_drop, ownership of the node
is always passed either to the collection (success), or to bpf_obj_drop
(failure). Regardless, it's correct to continue unconditionally
releasing the input owning ref, as something is always taking ownership
from the calling program on insert.
Keeping owning ref behavior unchanged results in a nice default UX for
insert functions that can fail. If the program's reaction to a failed
insert is "fine, just get rid of this owning ref for me and let me go
on with my business", then there's no reason to check for failure since
that's default behavior. e.g.:
long important_failures = 0;
int some_bpf_prog(void *ctx)
{
struct node *n, *m, *o; /* all bpf_obj_new'd */
bpf_spin_lock(&glock);
bpf_rbtree_add(&some_tree, &n->node, /* ... */);
bpf_rbtree_add(&some_tree, &m->node, /* ... */);
if (bpf_rbtree_add(&some_tree, &o->node, /* ... */)) {
important_failures++;
}
bpf_spin_unlock(&glock);
}
If we instead chose to pass ownership back to the program on failed
insert - by returning NULL on success or an owning ref on failure -
programs would always have to do something with the returned ref on
failure. The most likely action is probably "I'll just get rid of this
owning ref and go about my business", which ideally would look like:
if (n = bpf_rbtree_add(&some_tree, &n->node, /* ... */))
bpf_obj_drop(n);
But bpf_obj_drop isn't allowed in a critical section and inserts must
occur within one, so in reality error handling would become a
hard-to-parse mess.
For refcounted nodes, we can replicate the "pass ownership back to
program on failure" logic with this patch's semantics, albeit in an ugly
way:
struct node *n = bpf_obj_new(/* ... */), *m;
bpf_spin_lock(&glock);
m = bpf_refcount_acquire(n);
if (bpf_rbtree_add(&some_tree, &n->node, /* ... */)) {
/* Do something with m */
}
bpf_spin_unlock(&glock);
bpf_obj_drop(m);
bpf_refcount_acquire is used to simulate "return owning ref on failure".
This should be an uncommon occurrence, though.
Addition of two verifier-fixup'd args to collection inserts
===========================================================
The actual bpf_obj_drop kfunc is
bpf_obj_drop_impl(void *, struct btf_struct_meta *), with bpf_obj_drop
macro populating the second arg with 0 and the verifier later filling in
the arg during insn fixup.
Because bpf_rbtree_add and bpf_list_push_{front,back} now might do
bpf_obj_drop, these kfuncs need a btf_struct_meta parameter that can be
passed to bpf_obj_drop_impl.
Similarly, because the 'node' param to those insert functions is the
bpf_{list,rb}_node within the node type, and bpf_obj_drop expects a
pointer to the beginning of the node, the insert functions need to be
able to find the beginning of the node struct. A second
verifier-populated param is necessary: the offset of {list,rb}_node within the
node type.
These two new params allow the insert kfuncs to correctly call
__bpf_obj_drop_impl:
beginning_of_node = bpf_rb_node_ptr - offset
if (already_inserted)
__bpf_obj_drop_impl(beginning_of_node, btf_struct_meta->record);
Similarly to other kfuncs with "hidden" verifier-populated params, the
insert functions are renamed with _impl prefix and a macro is provided
for common usage. For example, bpf_rbtree_add kfunc is now
bpf_rbtree_add_impl and bpf_rbtree_add is now a macro which sets
"hidden" args to 0.
Due to the two new args BPF progs will need to be recompiled to work
with the new _impl kfuncs.
This patch also rewrites the "hidden argument" explanation to more
directly say why the BPF program writer doesn't need to populate the
arguments with anything meaningful.
How does this new logic affect non-owning references?
=====================================================
Currently, non-owning refs are valid until the end of the critical
section in which they're created. We can make this guarantee because, if
a non-owning ref exists, the referent was added to some collection. The
collection will drop() its nodes when it goes away, but it can't go away
while our program is accessing it, so that's not a problem. If the
referent is removed from the collection in the same CS that it was added
in, it can't be bpf_obj_drop'd until after CS end. Those are the only
two ways to free the referent's memory and neither can happen until
after the non-owning ref's lifetime ends.
On first glance, having these collection insert functions potentially
bpf_obj_drop their input seems like it breaks the "can't be
bpf_obj_drop'd until after CS end" line of reasoning. But we care about
the memory not being _freed_ until end of CS end, and a previous patch
in the series modified bpf_obj_drop such that it doesn't free refcounted
nodes until refcount == 0. So the statement can be more accurately
rewritten as "can't be free'd until after CS end".
We can prove that this rewritten statement holds for any non-owning
reference produced by collection insert functions:
* If the input to the insert function is _not_ refcounted
* We have an owning reference to the input, and can conclude it isn't
in any collection
* Inserting a node in a collection turns owning refs into
non-owning, and since our input type isn't refcounted, there's no
way to obtain additional owning refs to the same underlying
memory
* Because our node isn't in any collection, the insert operation
cannot fail, so bpf_obj_drop will not execute
* If bpf_obj_drop is guaranteed not to execute, there's no risk of
memory being free'd
* Otherwise, the input to the insert function is refcounted
* If the insert operation fails due to the node's list_head or rb_root
already being in some collection, there was some previous successful
insert which passed refcount to the collection
* We have an owning reference to the input, it must have been
acquired via bpf_refcount_acquire, which bumped the refcount
* refcount must be >= 2 since there's a valid owning reference and the
node is already in a collection
* Insert triggering bpf_obj_drop will decr refcount to >= 1, never
resulting in a free
So although we may do bpf_obj_drop during the critical section, this
will never result in memory being free'd, and no changes to non-owning
ref logic are needed in this patch.
Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-6-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-04-15 20:18:07 +00:00
|
|
|
union {
|
|
|
|
|
/* remember the size of type passed to bpf_obj_new to rewrite R1 */
|
|
|
|
|
u64 obj_new_size;
|
|
|
|
|
/* remember the offset of node field within type to rewrite */
|
|
|
|
|
u64 insert_off;
|
|
|
|
|
};
|
bpf: Introduce bpf_obj_new
Introduce type safe memory allocator bpf_obj_new for BPF programs. The
kernel side kfunc is named bpf_obj_new_impl, as passing hidden arguments
to kfuncs still requires having them in prototype, unlike BPF helpers
which always take 5 arguments and have them checked using bpf_func_proto
in verifier, ignoring unset argument types.
Introduce __ign suffix to ignore a specific kfunc argument during type
checks, then use this to introduce support for passing type metadata to
the bpf_obj_new_impl kfunc.
The user passes BTF ID of the type it wants to allocates in program BTF,
the verifier then rewrites the first argument as the size of this type,
after performing some sanity checks (to ensure it exists and it is a
struct type).
The second argument is also fixed up and passed by the verifier. This is
the btf_struct_meta for the type being allocated. It would be needed
mostly for the offset array which is required for zero initializing
special fields while leaving the rest of storage in unitialized state.
It would also be needed in the next patch to perform proper destruction
of the object's special fields.
Under the hood, bpf_obj_new will call bpf_mem_alloc and bpf_mem_free,
using the any context BPF memory allocator introduced recently. To this
end, a global instance of the BPF memory allocator is initialized on
boot to be used for this purpose. This 'bpf_global_ma' serves all
allocations for bpf_obj_new. In the future, bpf_obj_new variants will
allow specifying a custom allocator.
Note that now that bpf_obj_new can be used to allocate objects that can
be linked to BPF linked list (when future linked list helpers are
available), we need to also free the elements using bpf_mem_free.
However, since the draining of elements is done outside the
bpf_spin_lock, we need to do migrate_disable around the call since
bpf_list_head_free can be called from map free path where migration is
enabled. Otherwise, when called from BPF programs migration is already
disabled.
A convenience macro is included in the bpf_experimental.h header to hide
over the ugly details of the implementation, leading to user code
looking similar to a language level extension which allocates and
constructs fields of a user type.
struct bar {
struct bpf_list_node node;
};
struct foo {
struct bpf_spin_lock lock;
struct bpf_list_head head __contains(bar, node);
};
void prog(void) {
struct foo *f;
f = bpf_obj_new(typeof(*f));
if (!f)
return;
...
}
A key piece of this story is still missing, i.e. the free function,
which will come in the next patch.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20221118015614.2013203-14-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2022-11-18 01:56:03 +00:00
|
|
|
struct btf_struct_meta *kptr_struct_meta;
|
2019-11-22 20:07:59 +00:00
|
|
|
u64 map_key_state; /* constant (32 bit) key tracking for maps */
|
2017-06-22 22:07:39 +00:00
|
|
|
int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
|
2020-01-10 06:41:20 +00:00
|
|
|
u32 seen; /* this insn was processed by the verifier at env->pass_cnt */
|
bpf: Fall back to nospec for Spectre v1
This implements the core of the series and causes the verifier to fall
back to mitigating Spectre v1 using speculation barriers. The approach
was presented at LPC'24 [1] and RAID'24 [2].
If we find any forbidden behavior on a speculative path, we insert a
nospec (e.g., lfence speculation barrier on x86) before the instruction
and stop verifying the path. While verifying a speculative path, we can
furthermore stop verification of that path whenever we encounter a
nospec instruction.
A minimal example program would look as follows:
A = true
B = true
if A goto e
f()
if B goto e
unsafe()
e: exit
There are the following speculative and non-speculative paths
(`cur->speculative` and `speculative` referring to the value of the
push_stack() parameters):
- A = true
- B = true
- if A goto e
- A && !cur->speculative && !speculative
- exit
- !A && !cur->speculative && speculative
- f()
- if B goto e
- B && cur->speculative && !speculative
- exit
- !B && cur->speculative && speculative
- unsafe()
If f() contains any unsafe behavior under Spectre v1 and the unsafe
behavior matches `state->speculative &&
error_recoverable_with_nospec(err)`, do_check() will now add a nospec
before f() instead of rejecting the program:
A = true
B = true
if A goto e
nospec
f()
if B goto e
unsafe()
e: exit
Alternatively, the algorithm also takes advantage of nospec instructions
inserted for other reasons (e.g., Spectre v4). Taking the program above
as an example, speculative path exploration can stop before f() if a
nospec was inserted there because of Spectre v4 sanitization.
In this example, all instructions after the nospec are dead code (and
with the nospec they are also dead code speculatively).
For this, it relies on the fact that speculation barriers generally
prevent all later instructions from executing if the speculation was not
correct:
* On Intel x86_64, lfence acts as full speculation barrier, not only as
a load fence [3]:
An LFENCE instruction or a serializing instruction will ensure that
no later instructions execute, even speculatively, until all prior
instructions complete locally. [...] Inserting an LFENCE instruction
after a bounds check prevents later operations from executing before
the bound check completes.
This was experimentally confirmed in [4].
* On AMD x86_64, lfence is dispatch-serializing [5] (requires MSR
C001_1029[1] to be set if the MSR is supported, this happens in
init_amd()). AMD further specifies "A dispatch serializing instruction
forces the processor to retire the serializing instruction and all
previous instructions before the next instruction is executed" [8]. As
dispatch is not specific to memory loads or branches, lfence therefore
also affects all instructions there. Also, if retiring a branch means
it's PC change becomes architectural (should be), this means any
"wrong" speculation is aborted as required for this series.
* ARM's SB speculation barrier instruction also affects "any instruction
that appears later in the program order than the barrier" [6].
* PowerPC's barrier also affects all subsequent instructions [7]:
[...] executing an ori R31,R31,0 instruction ensures that all
instructions preceding the ori R31,R31,0 instruction have completed
before the ori R31,R31,0 instruction completes, and that no
subsequent instructions are initiated, even out-of-order, until
after the ori R31,R31,0 instruction completes. The ori R31,R31,0
instruction may complete before storage accesses associated with
instructions preceding the ori R31,R31,0 instruction have been
performed
Regarding the example, this implies that `if B goto e` will not execute
before `if A goto e` completes. Once `if A goto e` completes, the CPU
should find that the speculation was wrong and continue with `exit`.
If there is any other path that leads to `if B goto e` (and therefore
`unsafe()`) without going through `if A goto e`, then a nospec will
still be needed there. However, this patch assumes this other path will
be explored separately and therefore be discovered by the verifier even
if the exploration discussed here stops at the nospec.
This patch furthermore has the unfortunate consequence that Spectre v1
mitigations now only support architectures which implement BPF_NOSPEC.
Before this commit, Spectre v1 mitigations prevented exploits by
rejecting the programs on all architectures. Because some JITs do not
implement BPF_NOSPEC, this patch therefore may regress unpriv BPF's
security to a limited extent:
* The regression is limited to systems vulnerable to Spectre v1, have
unprivileged BPF enabled, and do NOT emit insns for BPF_NOSPEC. The
latter is not the case for x86 64- and 32-bit, arm64, and powerpc
64-bit and they are therefore not affected by the regression.
According to commit a6f6a95f2580 ("LoongArch, bpf: Fix jit to skip
speculation barrier opcode"), LoongArch is not vulnerable to Spectre
v1 and therefore also not affected by the regression.
* To the best of my knowledge this regression may therefore only affect
MIPS. This is deemed acceptable because unpriv BPF is still disabled
there by default. As stated in a previous commit, BPF_NOSPEC could be
implemented for MIPS based on GCC's speculation_barrier
implementation.
* It is unclear which other architectures (besides x86 64- and 32-bit,
ARM64, PowerPC 64-bit, LoongArch, and MIPS) supported by the kernel
are vulnerable to Spectre v1. Also, it is not clear if barriers are
available on these architectures. Implementing BPF_NOSPEC on these
architectures therefore is non-trivial. Searching GCC and the kernel
for speculation barrier implementations for these architectures
yielded no result.
* If any of those regressed systems is also vulnerable to Spectre v4,
the system was already vulnerable to Spectre v4 attacks based on
unpriv BPF before this patch and the impact is therefore further
limited.
As an alternative to regressing security, one could still reject
programs if the architecture does not emit BPF_NOSPEC (e.g., by removing
the empty BPF_NOSPEC-case from all JITs except for LoongArch where it
appears justified). However, this will cause rejections on these archs
that are likely unfounded in the vast majority of cases.
In the tests, some are now successful where we previously had a
false-positive (i.e., rejection). Change them to reflect where the
nospec should be inserted (using __xlated_unpriv) and modify the error
message if the nospec is able to mitigate a problem that previously
shadowed another problem (in that case __xlated_unpriv does not work,
therefore just add a comment).
Define SPEC_V1 to avoid duplicating this ifdef whenever we check for
nospec insns using __xlated_unpriv, define it here once. This also
improves readability. PowerPC can probably also be added here. However,
omit it for now because the BPF CI currently does not include a test.
Limit it to EPERM, EACCES, and EINVAL (and not everything except for
EFAULT and ENOMEM) as it already has the desired effect for most
real-world programs. Briefly went through all the occurrences of EPERM,
EINVAL, and EACCESS in verifier.c to validate that catching them like
this makes sense.
Thanks to Dustin for their help in checking the vendor documentation.
[1] https://lpc.events/event/18/contributions/1954/ ("Mitigating
Spectre-PHT using Speculation Barriers in Linux eBPF")
[2] https://arxiv.org/pdf/2405.00078 ("VeriFence: Lightweight and
Precise Spectre Defenses for Untrusted Linux Kernel Extensions")
[3] https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/technical-documentation/runtime-speculative-side-channel-mitigations.html
("Managed Runtime Speculative Execution Side Channel Mitigations")
[4] https://dl.acm.org/doi/pdf/10.1145/3359789.3359837 ("Speculator: a
tool to analyze speculative execution attacks and mitigations" -
Section 4.6 "Stopping Speculative Execution")
[5] https://www.amd.com/content/dam/amd/en/documents/processor-tech-docs/programmer-references/software-techniques-for-managing-speculation.pdf
("White Paper - SOFTWARE TECHNIQUES FOR MANAGING SPECULATION ON AMD
PROCESSORS - REVISION 5.09.23")
[6] https://developer.arm.com/documentation/ddi0597/2020-12/Base-Instructions/SB--Speculation-Barrier-
("SB - Speculation Barrier - Arm Armv8-A A32/T32 Instruction Set
Architecture (2020-12)")
[7] https://wiki.raptorcs.com/w/images/5/5f/OPF_PowerISA_v3.1C.pdf
("Power ISA™ - Version 3.1C - May 26, 2024 - Section 9.2.1 of Book
III")
[8] https://www.amd.com/content/dam/amd/en/documents/processor-tech-docs/programmer-references/40332.pdf
("AMD64 Architecture Programmer’s Manual Volumes 1–5 - Revision 4.08
- April 2024 - 7.6.4 Serializing Instructions")
Signed-off-by: Luis Gerhorst <luis.gerhorst@fau.de>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Acked-by: Henriette Herzog <henriette.herzog@rub.de>
Cc: Dustin Nguyen <nguyen@cs.fau.de>
Cc: Maximilian Ott <ott@cs.fau.de>
Cc: Milan Stephan <milan.stephan@fau.de>
Link: https://lore.kernel.org/r/20250603212428.338473-1-luis.gerhorst@fau.de
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-06-03 21:24:28 +00:00
|
|
|
bool nospec; /* do not execute this instruction speculatively */
|
2025-06-03 21:20:24 +00:00
|
|
|
bool nospec_result; /* result is unsafe under speculation, nospec must follow */
|
bpf: verifier: mark verified-insn with sub-register zext flag
eBPF ISA specification requires high 32-bit cleared when low 32-bit
sub-register is written. This applies to destination register of ALU32 etc.
JIT back-ends must guarantee this semantic when doing code-gen. x86_64 and
AArch64 ISA has the same semantics, so the corresponding JIT back-end
doesn't need to do extra work.
However, 32-bit arches (arm, x86, nfp etc.) and some other 64-bit arches
(PowerPC, SPARC etc) need to do explicit zero extension to meet this
requirement, otherwise code like the following will fail.
u64_value = (u64) u32_value
... other uses of u64_value
This is because compiler could exploit the semantic described above and
save those zero extensions for extending u32_value to u64_value, these JIT
back-ends are expected to guarantee this through inserting extra zero
extensions which however could be a significant increase on the code size.
Some benchmarks show there could be ~40% sub-register writes out of total
insns, meaning at least ~40% extra code-gen.
One observation is these extra zero extensions are not always necessary.
Take above code snippet for example, it is possible u32_value will never be
casted into a u64, the value of high 32-bit of u32_value then could be
ignored and extra zero extension could be eliminated.
This patch implements this idea, insns defining sub-registers will be
marked when the high 32-bit of the defined sub-register matters. For
those unmarked insns, it is safe to eliminate high 32-bit clearnace for
them.
Algo:
- Split read flags into READ32 and READ64.
- Record index of insn that does sub-register write. Keep the index inside
reg state and update it during verifier insn walking.
- A full register read on a sub-register marks its definition insn as
needing zero extension on dst register.
A new sub-register write overrides the old one.
- When propagating read64 during path pruning, also mark any insn defining
a sub-register that is read in the pruned path as full-register.
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-05-24 22:25:12 +00:00
|
|
|
bool zext_dst; /* this insn zero extends dst reg */
|
2024-03-08 01:08:03 +00:00
|
|
|
bool needs_zext; /* alu op needs to clear upper bits */
|
2022-11-24 05:32:17 +00:00
|
|
|
bool storage_get_func_atomic; /* bpf_*_storage_get() with atomic memory alloc */
|
bpf: add support for open-coded iterator loops
Teach verifier about the concept of the open-coded (or inline) iterators.
This patch adds generic iterator loop verification logic, new STACK_ITER
stack slot type to contain iterator state, and necessary kfunc plumbing
for iterator's constructor, destructor and next methods. Next patch
implements first specific iterator (numbers iterator for implementing
for() loop logic). Such split allows to have more focused commits for
verifier logic and separate commit that we could point later to
demonstrating what does it take to add a new kind of iterator.
Each kind of iterator has its own associated struct bpf_iter_<type>,
where <type> denotes a specific type of iterator. struct bpf_iter_<type>
state is supposed to live on BPF program stack, so there will be no way
to change its size later on without breaking backwards compatibility, so
choose wisely! But given this struct is specific to a given <type> of
iterator, this allows a lot of flexibility: simple iterators could be
fine with just one stack slot (8 bytes), like numbers iterator in the
next patch, while some other more complicated iterators might need way
more to keep their iterator state. Either way, such design allows to
avoid runtime memory allocations, which otherwise would be necessary if
we fixed on-the-stack size and it turned out to be too small for a given
iterator implementation.
The way BPF verifier logic is implemented, there are no artificial
restrictions on a number of active iterators, it should work correctly
using multiple active iterators at the same time. This also means you
can have multiple nested iteration loops. struct bpf_iter_<type>
reference can be safely passed to subprograms as well.
General flow is easiest to demonstrate with a simple example using
number iterator implemented in next patch. Here's the simplest possible
loop:
struct bpf_iter_num it;
int *v;
bpf_iter_num_new(&it, 2, 5);
while ((v = bpf_iter_num_next(&it))) {
bpf_printk("X = %d", *v);
}
bpf_iter_num_destroy(&it);
Above snippet should output "X = 2", "X = 3", "X = 4". Note that 5 is
exclusive and is not returned. This matches similar APIs (e.g., slices
in Go or Rust) that implement a range of elements, where end index is
non-inclusive.
In the above example, we see a trio of function:
- constructor, bpf_iter_num_new(), which initializes iterator state
(struct bpf_iter_num it) on the stack. If any of the input arguments
are invalid, constructor should make sure to still initialize it such
that subsequent bpf_iter_num_next() calls will return NULL. I.e., on
error, return error and construct empty iterator.
- next method, bpf_iter_num_next(), which accepts pointer to iterator
state and produces an element. Next method should always return
a pointer. The contract between BPF verifier is that next method will
always eventually return NULL when elements are exhausted. Once NULL is
returned, subsequent next calls should keep returning NULL. In the
case of numbers iterator, bpf_iter_num_next() returns a pointer to an int
(storage for this integer is inside the iterator state itself),
which can be dereferenced after corresponding NULL check.
- once done with the iterator, it's mandated that user cleans up its
state with the call to destructor, bpf_iter_num_destroy() in this
case. Destructor frees up any resources and marks stack space used by
struct bpf_iter_num as usable for something else.
Any other iterator implementation will have to implement at least these
three methods. It is enforced that for any given type of iterator only
applicable constructor/destructor/next are callable. I.e., verifier
ensures you can't pass number iterator state into, say, cgroup
iterator's next method.
It is important to keep the naming pattern consistent to be able to
create generic macros to help with BPF iter usability. E.g., one
of the follow up patches adds generic bpf_for_each() macro to bpf_misc.h
in selftests, which allows to utilize iterator "trio" nicely without
having to code the above somewhat tedious loop explicitly every time.
This is enforced at kfunc registration point by one of the previous
patches in this series.
At the implementation level, iterator state tracking for verification
purposes is very similar to dynptr. We add STACK_ITER stack slot type,
reserve necessary number of slots, depending on
sizeof(struct bpf_iter_<type>), and keep track of necessary extra state
in the "main" slot, which is marked with non-zero ref_obj_id. Other
slots are also marked as STACK_ITER, but have zero ref_obj_id. This is
simpler than having a separate "is_first_slot" flag.
Another big distinction is that STACK_ITER is *always refcounted*, which
simplifies implementation without sacrificing usability. So no need for
extra "iter_id", no need to anticipate reuse of STACK_ITER slots for new
constructors, etc. Keeping it simple here.
As far as the verification logic goes, there are two extensive comments:
in process_iter_next_call() and iter_active_depths_differ() explaining
some important and sometimes subtle aspects. Please refer to them for
details.
But from 10,000-foot point of view, next methods are the points of
forking a verification state, which are conceptually similar to what
verifier is doing when validating conditional jump. We branch out at
a `call bpf_iter_<type>_next` instruction and simulate two outcomes:
NULL (iteration is done) and non-NULL (new element is returned). NULL is
simulated first and is supposed to reach exit without looping. After
that non-NULL case is validated and it either reaches exit (for trivial
examples with no real loop), or reaches another `call bpf_iter_<type>_next`
instruction with the state equivalent to already (partially) validated
one. State equivalency at that point means we technically are going to
be looping forever without "breaking out" out of established "state
envelope" (i.e., subsequent iterations don't add any new knowledge or
constraints to the verifier state, so running 1, 2, 10, or a million of
them doesn't matter). But taking into account the contract stating that
iterator next method *has to* return NULL eventually, we can conclude
that loop body is safe and will eventually terminate. Given we validated
logic outside of the loop (NULL case), and concluded that loop body is
safe (though potentially looping many times), verifier can claim safety
of the overall program logic.
The rest of the patch is necessary plumbing for state tracking, marking,
validation, and necessary further kfunc plumbing to allow implementing
iterator constructor, destructor, and next methods.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230308184121.1165081-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 18:41:16 +00:00
|
|
|
bool is_iter_next; /* bpf_iter_<type>_next() kfunc call */
|
2023-08-27 15:27:49 +00:00
|
|
|
bool call_with_percpu_alloc_ptr; /* {this,per}_cpu_ptr() with prog percpu alloc */
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-02 23:58:34 +00:00
|
|
|
u8 alu_state; /* used in combination with alu_limit */
|
bpf: no_caller_saved_registers attribute for helper calls
GCC and LLVM define a no_caller_saved_registers function attribute.
This attribute means that function scratches only some of
the caller saved registers defined by ABI.
For BPF the set of such registers could be defined as follows:
- R0 is scratched only if function is non-void;
- R1-R5 are scratched only if corresponding parameter type is defined
in the function prototype.
This commit introduces flag bpf_func_prot->allow_nocsr.
If this flag is set for some helper function, verifier assumes that
it follows no_caller_saved_registers calling convention.
The contract between kernel and clang allows to simultaneously use
such functions and maintain backwards compatibility with old
kernels that don't understand no_caller_saved_registers calls
(nocsr for short):
- clang generates a simple pattern for nocsr calls, e.g.:
r1 = 1;
r2 = 2;
*(u64 *)(r10 - 8) = r1;
*(u64 *)(r10 - 16) = r2;
call %[to_be_inlined]
r2 = *(u64 *)(r10 - 16);
r1 = *(u64 *)(r10 - 8);
r0 = r1;
r0 += r2;
exit;
- kernel removes unnecessary spills and fills, if called function is
inlined by verifier or current JIT (with assumption that patch
inserted by verifier or JIT honors nocsr contract, e.g. does not
scratch r3-r5 for the example above), e.g. the code above would be
transformed to:
r1 = 1;
r2 = 2;
call %[to_be_inlined]
r0 = r1;
r0 += r2;
exit;
Technically, the transformation is split into the following phases:
- function mark_nocsr_patterns(), called from bpf_check()
searches and marks potential patterns in instruction auxiliary data;
- upon stack read or write access,
function check_nocsr_stack_contract() is used to verify if
stack offsets, presumably reserved for nocsr patterns, are used
only from those patterns;
- function remove_nocsr_spills_fills(), called from bpf_check(),
applies the rewrite for valid patterns.
See comment in mark_nocsr_pattern_for_call() for more details.
Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20240722233844.1406874-3-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2024-07-22 23:38:36 +00:00
|
|
|
/* true if STX or LDX instruction is a part of a spill/fill
|
2024-08-22 08:41:07 +00:00
|
|
|
* pattern for a bpf_fastcall call.
|
bpf: no_caller_saved_registers attribute for helper calls
GCC and LLVM define a no_caller_saved_registers function attribute.
This attribute means that function scratches only some of
the caller saved registers defined by ABI.
For BPF the set of such registers could be defined as follows:
- R0 is scratched only if function is non-void;
- R1-R5 are scratched only if corresponding parameter type is defined
in the function prototype.
This commit introduces flag bpf_func_prot->allow_nocsr.
If this flag is set for some helper function, verifier assumes that
it follows no_caller_saved_registers calling convention.
The contract between kernel and clang allows to simultaneously use
such functions and maintain backwards compatibility with old
kernels that don't understand no_caller_saved_registers calls
(nocsr for short):
- clang generates a simple pattern for nocsr calls, e.g.:
r1 = 1;
r2 = 2;
*(u64 *)(r10 - 8) = r1;
*(u64 *)(r10 - 16) = r2;
call %[to_be_inlined]
r2 = *(u64 *)(r10 - 16);
r1 = *(u64 *)(r10 - 8);
r0 = r1;
r0 += r2;
exit;
- kernel removes unnecessary spills and fills, if called function is
inlined by verifier or current JIT (with assumption that patch
inserted by verifier or JIT honors nocsr contract, e.g. does not
scratch r3-r5 for the example above), e.g. the code above would be
transformed to:
r1 = 1;
r2 = 2;
call %[to_be_inlined]
r0 = r1;
r0 += r2;
exit;
Technically, the transformation is split into the following phases:
- function mark_nocsr_patterns(), called from bpf_check()
searches and marks potential patterns in instruction auxiliary data;
- upon stack read or write access,
function check_nocsr_stack_contract() is used to verify if
stack offsets, presumably reserved for nocsr patterns, are used
only from those patterns;
- function remove_nocsr_spills_fills(), called from bpf_check(),
applies the rewrite for valid patterns.
See comment in mark_nocsr_pattern_for_call() for more details.
Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20240722233844.1406874-3-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2024-07-22 23:38:36 +00:00
|
|
|
*/
|
2024-08-22 08:41:07 +00:00
|
|
|
u8 fastcall_pattern:1;
|
bpf: no_caller_saved_registers attribute for helper calls
GCC and LLVM define a no_caller_saved_registers function attribute.
This attribute means that function scratches only some of
the caller saved registers defined by ABI.
For BPF the set of such registers could be defined as follows:
- R0 is scratched only if function is non-void;
- R1-R5 are scratched only if corresponding parameter type is defined
in the function prototype.
This commit introduces flag bpf_func_prot->allow_nocsr.
If this flag is set for some helper function, verifier assumes that
it follows no_caller_saved_registers calling convention.
The contract between kernel and clang allows to simultaneously use
such functions and maintain backwards compatibility with old
kernels that don't understand no_caller_saved_registers calls
(nocsr for short):
- clang generates a simple pattern for nocsr calls, e.g.:
r1 = 1;
r2 = 2;
*(u64 *)(r10 - 8) = r1;
*(u64 *)(r10 - 16) = r2;
call %[to_be_inlined]
r2 = *(u64 *)(r10 - 16);
r1 = *(u64 *)(r10 - 8);
r0 = r1;
r0 += r2;
exit;
- kernel removes unnecessary spills and fills, if called function is
inlined by verifier or current JIT (with assumption that patch
inserted by verifier or JIT honors nocsr contract, e.g. does not
scratch r3-r5 for the example above), e.g. the code above would be
transformed to:
r1 = 1;
r2 = 2;
call %[to_be_inlined]
r0 = r1;
r0 += r2;
exit;
Technically, the transformation is split into the following phases:
- function mark_nocsr_patterns(), called from bpf_check()
searches and marks potential patterns in instruction auxiliary data;
- upon stack read or write access,
function check_nocsr_stack_contract() is used to verify if
stack offsets, presumably reserved for nocsr patterns, are used
only from those patterns;
- function remove_nocsr_spills_fills(), called from bpf_check(),
applies the rewrite for valid patterns.
See comment in mark_nocsr_pattern_for_call() for more details.
Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20240722233844.1406874-3-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2024-07-22 23:38:36 +00:00
|
|
|
/* for CALL instructions, a number of spill/fill pairs in the
|
2024-08-22 08:41:07 +00:00
|
|
|
* bpf_fastcall pattern.
|
bpf: no_caller_saved_registers attribute for helper calls
GCC and LLVM define a no_caller_saved_registers function attribute.
This attribute means that function scratches only some of
the caller saved registers defined by ABI.
For BPF the set of such registers could be defined as follows:
- R0 is scratched only if function is non-void;
- R1-R5 are scratched only if corresponding parameter type is defined
in the function prototype.
This commit introduces flag bpf_func_prot->allow_nocsr.
If this flag is set for some helper function, verifier assumes that
it follows no_caller_saved_registers calling convention.
The contract between kernel and clang allows to simultaneously use
such functions and maintain backwards compatibility with old
kernels that don't understand no_caller_saved_registers calls
(nocsr for short):
- clang generates a simple pattern for nocsr calls, e.g.:
r1 = 1;
r2 = 2;
*(u64 *)(r10 - 8) = r1;
*(u64 *)(r10 - 16) = r2;
call %[to_be_inlined]
r2 = *(u64 *)(r10 - 16);
r1 = *(u64 *)(r10 - 8);
r0 = r1;
r0 += r2;
exit;
- kernel removes unnecessary spills and fills, if called function is
inlined by verifier or current JIT (with assumption that patch
inserted by verifier or JIT honors nocsr contract, e.g. does not
scratch r3-r5 for the example above), e.g. the code above would be
transformed to:
r1 = 1;
r2 = 2;
call %[to_be_inlined]
r0 = r1;
r0 += r2;
exit;
Technically, the transformation is split into the following phases:
- function mark_nocsr_patterns(), called from bpf_check()
searches and marks potential patterns in instruction auxiliary data;
- upon stack read or write access,
function check_nocsr_stack_contract() is used to verify if
stack offsets, presumably reserved for nocsr patterns, are used
only from those patterns;
- function remove_nocsr_spills_fills(), called from bpf_check(),
applies the rewrite for valid patterns.
See comment in mark_nocsr_pattern_for_call() for more details.
Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20240722233844.1406874-3-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2024-07-22 23:38:36 +00:00
|
|
|
*/
|
2024-08-22 08:41:07 +00:00
|
|
|
u8 fastcall_spills_num:3;
|
2025-05-13 14:28:12 +00:00
|
|
|
u8 arg_prog:4;
|
2020-01-10 06:41:20 +00:00
|
|
|
|
|
|
|
|
/* below fields are initialized once */
|
2019-01-23 06:45:23 +00:00
|
|
|
unsigned int orig_idx; /* original instruction index */
|
2022-12-06 23:33:43 +00:00
|
|
|
bool jmp_point;
|
bpf: ensure state checkpointing at iter_next() call sites
State equivalence check and checkpointing performed in is_state_visited()
employs certain heuristics to try to save memory by avoiding state checkpoints
if not enough jumps and instructions happened since last checkpoint. This leads
to unpredictability of whether a particular instruction will be checkpointed
and how regularly. While normally this is not causing much problems (except
inconveniences for predictable verifier tests, which we overcome with
BPF_F_TEST_STATE_FREQ flag), turns out it's not the case for open-coded
iterators.
Checking and saving state checkpoints at iter_next() call is crucial for fast
convergence of open-coded iterator loop logic, so we need to force it. If we
don't do that, is_state_visited() might skip saving a checkpoint, causing
unnecessarily long sequence of not checkpointed instructions and jumps, leading
to exhaustion of jump history buffer, and potentially other undesired outcomes.
It is expected that with correct open-coded iterators convergence will happen
quickly, so we don't run a risk of exhausting memory.
This patch adds, in addition to prune and jump instruction marks, also a
"forced checkpoint" mark, and makes sure that any iter_next() call instruction
is marked as such.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230310060149.625887-1-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-10 06:01:49 +00:00
|
|
|
bool prune_point;
|
|
|
|
|
/* ensure we check state equivalence and save state checkpoint and
|
|
|
|
|
* this instruction, regardless of any heuristics
|
|
|
|
|
*/
|
|
|
|
|
bool force_checkpoint;
|
2023-11-21 02:06:56 +00:00
|
|
|
/* true if instruction is a call to a helper function that
|
|
|
|
|
* accepts callback function as a parameter.
|
|
|
|
|
*/
|
|
|
|
|
bool calls_callback;
|
2025-06-11 20:08:27 +00:00
|
|
|
/*
|
|
|
|
|
* CFG strongly connected component this instruction belongs to,
|
|
|
|
|
* zero if it is a singleton SCC.
|
|
|
|
|
*/
|
|
|
|
|
u32 scc;
|
bpf: simple DFA-based live registers analysis
Compute may-live registers before each instruction in the program.
The register is live before the instruction I if it is read by I or
some instruction S following I during program execution and is not
overwritten between I and S.
This information would be used in the next patch as a hint in
func_states_equal().
Use a simple algorithm described in [1] to compute this information:
- define the following:
- I.use : a set of all registers read by instruction I;
- I.def : a set of all registers written by instruction I;
- I.in : a set of all registers that may be alive before I execution;
- I.out : a set of all registers that may be alive after I execution;
- I.successors : a set of instructions S that might immediately
follow I for some program execution;
- associate separate empty sets 'I.in' and 'I.out' with each instruction;
- visit each instruction in a postorder and update corresponding
'I.in' and 'I.out' sets as follows:
I.out = U [S.in for S in I.successors]
I.in = (I.out / I.def) U I.use
(where U stands for set union, / stands for set difference)
- repeat the computation while I.{in,out} changes for any instruction.
On implementation side keep things as simple, as possible:
- check_cfg() already marks instructions EXPLORED in post-order,
modify it to save the index of each EXPLORED instruction in a vector;
- represent I.{in,out,use,def} as bitmasks;
- don't split the program into basic blocks and don't maintain the
work queue, instead:
- do fixed-point computation by visiting each instruction;
- maintain a simple 'changed' flag if I.{in,out} for any instruction
change;
Measurements show that even such simplistic implementation does not
add measurable verification time overhead (for selftests, at-least).
Note on check_cfg() ex_insn_beg/ex_done change:
To avoid out of bounds access to env->cfg.insn_postorder array,
it should be guaranteed that instruction transitions to EXPLORED state
only once. Previously this was not the fact for incorrect programs
with direct calls to exception callbacks.
The 'align' selftest needs adjustment to skip computed insn/live
registers printout. Otherwise it matches lines from the live registers
printout.
[1] https://en.wikipedia.org/wiki/Live-variable_analysis
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250304195024.2478889-4-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-03-04 19:50:22 +00:00
|
|
|
/* registers alive before this instruction. */
|
|
|
|
|
u16 live_regs_before;
|
2016-09-21 10:43:57 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
#define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
|
2021-01-12 07:55:18 +00:00
|
|
|
#define MAX_USED_BTFS 64 /* max number of BTFs accessed by one BPF program */
|
2016-09-21 10:43:57 +00:00
|
|
|
|
2017-10-09 17:30:15 +00:00
|
|
|
#define BPF_VERIFIER_TMP_LOG_SIZE 1024
|
|
|
|
|
|
2018-03-24 18:44:22 +00:00
|
|
|
struct bpf_verifier_log {
|
bpf: Switch BPF verifier log to be a rotating log by default
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
2023-04-06 23:41:49 +00:00
|
|
|
/* Logical start and end positions of a "log window" of the verifier log.
|
|
|
|
|
* start_pos == 0 means we haven't truncated anything.
|
|
|
|
|
* Once truncation starts to happen, start_pos + len_total == end_pos,
|
|
|
|
|
* except during log reset situations, in which (end_pos - start_pos)
|
|
|
|
|
* might get smaller than len_total (see bpf_vlog_reset()).
|
|
|
|
|
* Generally, (end_pos - start_pos) gives number of useful data in
|
|
|
|
|
* user log buffer.
|
|
|
|
|
*/
|
|
|
|
|
u64 start_pos;
|
|
|
|
|
u64 end_pos;
|
2017-10-09 17:30:10 +00:00
|
|
|
char __user *ubuf;
|
bpf: Switch BPF verifier log to be a rotating log by default
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
2023-04-06 23:41:49 +00:00
|
|
|
u32 level;
|
2017-10-09 17:30:10 +00:00
|
|
|
u32 len_total;
|
bpf: Keep track of total log content size in both fixed and rolling modes
Change how we do accounting in BPF_LOG_FIXED mode and adopt log->end_pos
as *logical* log position. This means that we can go beyond physical log
buffer size now and be able to tell what log buffer size should be to
fit entire log contents without -ENOSPC.
To do this for BPF_LOG_FIXED mode, we need to remove a short-circuiting
logic of not vsnprintf()'ing further log content once we filled up
user-provided buffer, which is done by bpf_verifier_log_needed() checks.
We modify these checks to always keep going if log->level is non-zero
(i.e., log is requested), even if log->ubuf was NULL'ed out due to
copying data to user-space, or if entire log buffer is physically full.
We adopt bpf_verifier_vlog() routine to work correctly with
log->ubuf == NULL condition, performing log formatting into temporary
kernel buffer, doing all the necessary accounting, but just avoiding
copying data out if buffer is full or NULL'ed out.
With these changes, it's now possible to do this sort of determination of
log contents size in both BPF_LOG_FIXED and default rolling log mode.
We need to keep in mind bpf_vlog_reset(), though, which shrinks log
contents after successful verification of a particular code path. This
log reset means that log->end_pos isn't always increasing, so to return
back to users what should be the log buffer size to fit all log content
without causing -ENOSPC even in the presence of log resetting, we need
to keep maximum over "lifetime" of logging. We do this accounting in
bpf_vlog_update_len_max() helper.
A related and subtle aspect is that with this logical log->end_pos even in
BPF_LOG_FIXED mode we could temporary "overflow" buffer, but then reset
it back with bpf_vlog_reset() to a position inside user-supplied
log_buf. In such situation we still want to properly maintain
terminating zero. We will eventually return -ENOSPC even if final log
buffer is small (we detect this through log->len_max check). This
behavior is simpler to reason about and is consistent with current
behavior of verifier log. Handling of this required a small addition to
bpf_vlog_reset() logic to avoid doing put_user() beyond physical log
buffer dimensions.
Another issue to keep in mind is that we limit log buffer size to 32-bit
value and keep such log length as u32, but theoretically verifier could
produce huge log stretching beyond 4GB. Instead of keeping (and later
returning) 64-bit log length, we cap it at UINT_MAX. Current UAPI makes
it impossible to specify log buffer size bigger than 4GB anyways, so we
don't really loose anything here and keep everything consistently 32-bit
in UAPI. This property will be utilized in next patch.
Doing the same determination of maximum log buffer for rolling mode is
trivial, as log->end_pos and log->start_pos are already logical
positions, so there is nothing new there.
These changes do incidentally fix one small issue with previous logging
logic. Previously, if use provided log buffer of size N, and actual log
output was exactly N-1 bytes + terminating \0, kernel logic coun't
distinguish this condition from log truncation scenario which would end
up with truncated log contents of N-1 bytes + terminating \0 as well.
But now with log->end_pos being logical position that could go beyond
actual log buffer size, we can distinguish these two conditions, which
we do in this patch. This plays nicely with returning log_size_actual
(implemented in UAPI in the next patch), as we can now guarantee that if
user takes such log_size_actual and provides log buffer of that exact
size, they will not get -ENOSPC in return.
All in all, all these changes do conceptually unify fixed and rolling
log modes much better, and allow a nice feature requested by users:
knowing what should be the size of the buffer to avoid -ENOSPC.
We'll plumb this through the UAPI and the code in the next patch.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-12-andrii@kernel.org
2023-04-06 23:41:57 +00:00
|
|
|
u32 len_max;
|
bpf: Switch BPF verifier log to be a rotating log by default
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
2023-04-06 23:41:49 +00:00
|
|
|
char kbuf[BPF_VERIFIER_TMP_LOG_SIZE];
|
2017-10-09 17:30:10 +00:00
|
|
|
};
|
|
|
|
|
|
2019-04-02 04:27:40 +00:00
|
|
|
#define BPF_LOG_LEVEL1 1
|
|
|
|
|
#define BPF_LOG_LEVEL2 2
|
|
|
|
|
#define BPF_LOG_STATS 4
|
bpf: Switch BPF verifier log to be a rotating log by default
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
2023-04-06 23:41:49 +00:00
|
|
|
#define BPF_LOG_FIXED 8
|
2019-04-02 04:27:40 +00:00
|
|
|
#define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
|
bpf: Switch BPF verifier log to be a rotating log by default
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
2023-04-06 23:41:49 +00:00
|
|
|
#define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS | BPF_LOG_FIXED)
|
2019-10-16 03:24:57 +00:00
|
|
|
#define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */
|
bpf: Right align verifier states in verifier logs.
Make the verifier logs more readable, print the verifier states
on the corresponding instruction line. If the previous line was
not a bpf instruction, then print the verifier states on its own
line.
Before:
Validating test_pkt_access_subprog3() func#3...
86: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R10=fp0
; int test_pkt_access_subprog3(int val, struct __sk_buff *skb)
86: (bf) r6 = r2
87: R2=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
87: (bc) w7 = w1
88: R1=invP(id=0) R7_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
88: (bf) r1 = r6
89: R1_w=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
89: (85) call pc+9
Func#4 is global and valid. Skipping.
90: R0_w=invP(id=0)
90: (bc) w8 = w0
91: R0_w=invP(id=0) R8_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
91: (b7) r1 = 123
92: R1_w=invP123
92: (85) call pc+65
Func#5 is global and valid. Skipping.
93: R0=invP(id=0)
After:
86: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R10=fp0
; int test_pkt_access_subprog3(int val, struct __sk_buff *skb)
86: (bf) r6 = r2 ; R2=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
87: (bc) w7 = w1 ; R1=invP(id=0) R7_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
88: (bf) r1 = r6 ; R1_w=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
89: (85) call pc+9
Func#4 is global and valid. Skipping.
90: R0_w=invP(id=0)
90: (bc) w8 = w0 ; R0_w=invP(id=0) R8_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
91: (b7) r1 = 123 ; R1_w=invP123
92: (85) call pc+65
Func#5 is global and valid. Skipping.
93: R0=invP(id=0)
Signed-off-by: Christy Lee <christylee@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2021-12-17 03:42:45 +00:00
|
|
|
#define BPF_LOG_MIN_ALIGNMENT 8U
|
|
|
|
|
#define BPF_LOG_ALIGNMENT 40U
|
2019-04-02 04:27:40 +00:00
|
|
|
|
2018-03-24 18:44:23 +00:00
|
|
|
static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
|
|
|
|
|
{
|
bpf: Keep track of total log content size in both fixed and rolling modes
Change how we do accounting in BPF_LOG_FIXED mode and adopt log->end_pos
as *logical* log position. This means that we can go beyond physical log
buffer size now and be able to tell what log buffer size should be to
fit entire log contents without -ENOSPC.
To do this for BPF_LOG_FIXED mode, we need to remove a short-circuiting
logic of not vsnprintf()'ing further log content once we filled up
user-provided buffer, which is done by bpf_verifier_log_needed() checks.
We modify these checks to always keep going if log->level is non-zero
(i.e., log is requested), even if log->ubuf was NULL'ed out due to
copying data to user-space, or if entire log buffer is physically full.
We adopt bpf_verifier_vlog() routine to work correctly with
log->ubuf == NULL condition, performing log formatting into temporary
kernel buffer, doing all the necessary accounting, but just avoiding
copying data out if buffer is full or NULL'ed out.
With these changes, it's now possible to do this sort of determination of
log contents size in both BPF_LOG_FIXED and default rolling log mode.
We need to keep in mind bpf_vlog_reset(), though, which shrinks log
contents after successful verification of a particular code path. This
log reset means that log->end_pos isn't always increasing, so to return
back to users what should be the log buffer size to fit all log content
without causing -ENOSPC even in the presence of log resetting, we need
to keep maximum over "lifetime" of logging. We do this accounting in
bpf_vlog_update_len_max() helper.
A related and subtle aspect is that with this logical log->end_pos even in
BPF_LOG_FIXED mode we could temporary "overflow" buffer, but then reset
it back with bpf_vlog_reset() to a position inside user-supplied
log_buf. In such situation we still want to properly maintain
terminating zero. We will eventually return -ENOSPC even if final log
buffer is small (we detect this through log->len_max check). This
behavior is simpler to reason about and is consistent with current
behavior of verifier log. Handling of this required a small addition to
bpf_vlog_reset() logic to avoid doing put_user() beyond physical log
buffer dimensions.
Another issue to keep in mind is that we limit log buffer size to 32-bit
value and keep such log length as u32, but theoretically verifier could
produce huge log stretching beyond 4GB. Instead of keeping (and later
returning) 64-bit log length, we cap it at UINT_MAX. Current UAPI makes
it impossible to specify log buffer size bigger than 4GB anyways, so we
don't really loose anything here and keep everything consistently 32-bit
in UAPI. This property will be utilized in next patch.
Doing the same determination of maximum log buffer for rolling mode is
trivial, as log->end_pos and log->start_pos are already logical
positions, so there is nothing new there.
These changes do incidentally fix one small issue with previous logging
logic. Previously, if use provided log buffer of size N, and actual log
output was exactly N-1 bytes + terminating \0, kernel logic coun't
distinguish this condition from log truncation scenario which would end
up with truncated log contents of N-1 bytes + terminating \0 as well.
But now with log->end_pos being logical position that could go beyond
actual log buffer size, we can distinguish these two conditions, which
we do in this patch. This plays nicely with returning log_size_actual
(implemented in UAPI in the next patch), as we can now guarantee that if
user takes such log_size_actual and provides log buffer of that exact
size, they will not get -ENOSPC in return.
All in all, all these changes do conceptually unify fixed and rolling
log modes much better, and allow a nice feature requested by users:
knowing what should be the size of the buffer to avoid -ENOSPC.
We'll plumb this through the UAPI and the code in the next patch.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-12-andrii@kernel.org
2023-04-06 23:41:57 +00:00
|
|
|
return log && log->level;
|
2018-03-24 18:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
2017-12-15 01:55:05 +00:00
|
|
|
#define BPF_MAX_SUBPROGS 256
|
|
|
|
|
|
bpf: abstract away global subprog arg preparation logic from reg state setup
btf_prepare_func_args() is used to understand expectations and
restrictions on global subprog arguments. But current implementation is
hard to extend, as it intermixes BTF-based func prototype parsing and
interpretation logic with setting up register state at subprog entry.
Worse still, those registers are not completely set up inside
btf_prepare_func_args(), requiring some more logic later in
do_check_common(). Like calling mark_reg_unknown() and similar
initialization operations.
This intermixing of BTF interpretation and register state setup is
problematic. First, it causes duplication of BTF parsing logic for global
subprog verification (to set up initial state of global subprog) and
global subprog call sites analysis (when we need to check that whatever
is being passed into global subprog matches expectations), performed in
btf_check_subprog_call().
Given we want to extend global func argument with tags later, this
duplication is problematic. So refactor btf_prepare_func_args() to do
only BTF-based func proto and args parsing, returning high-level
argument "expectations" only, with no regard to specifics of register
state. I.e., if it's a context argument, instead of setting register
state to PTR_TO_CTX, we return ARG_PTR_TO_CTX enum for that argument as
"an argument specification" for further processing inside
do_check_common(). Similarly for SCALAR arguments, PTR_TO_MEM, etc.
This allows to reuse btf_prepare_func_args() in following patches at
global subprog call site analysis time. It also keeps register setup
code consistently in one place, do_check_common().
Besides all this, we cache this argument specs information inside
env->subprog_info, eliminating the need to redo these potentially
expensive BTF traversals, especially if BPF program's BTF is big and/or
there are lots of global subprog calls.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231215011334.2307144-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-15 01:13:25 +00:00
|
|
|
struct bpf_subprog_arg_info {
|
|
|
|
|
enum bpf_arg_type arg_type;
|
|
|
|
|
union {
|
|
|
|
|
u32 mem_size;
|
2024-01-30 00:06:45 +00:00
|
|
|
u32 btf_id;
|
bpf: abstract away global subprog arg preparation logic from reg state setup
btf_prepare_func_args() is used to understand expectations and
restrictions on global subprog arguments. But current implementation is
hard to extend, as it intermixes BTF-based func prototype parsing and
interpretation logic with setting up register state at subprog entry.
Worse still, those registers are not completely set up inside
btf_prepare_func_args(), requiring some more logic later in
do_check_common(). Like calling mark_reg_unknown() and similar
initialization operations.
This intermixing of BTF interpretation and register state setup is
problematic. First, it causes duplication of BTF parsing logic for global
subprog verification (to set up initial state of global subprog) and
global subprog call sites analysis (when we need to check that whatever
is being passed into global subprog matches expectations), performed in
btf_check_subprog_call().
Given we want to extend global func argument with tags later, this
duplication is problematic. So refactor btf_prepare_func_args() to do
only BTF-based func proto and args parsing, returning high-level
argument "expectations" only, with no regard to specifics of register
state. I.e., if it's a context argument, instead of setting register
state to PTR_TO_CTX, we return ARG_PTR_TO_CTX enum for that argument as
"an argument specification" for further processing inside
do_check_common(). Similarly for SCALAR arguments, PTR_TO_MEM, etc.
This allows to reuse btf_prepare_func_args() in following patches at
global subprog call site analysis time. It also keeps register setup
code consistently in one place, do_check_common().
Besides all this, we cache this argument specs information inside
env->subprog_info, eliminating the need to redo these potentially
expensive BTF traversals, especially if BPF program's BTF is big and/or
there are lots of global subprog calls.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231215011334.2307144-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-15 01:13:25 +00:00
|
|
|
};
|
|
|
|
|
};
|
|
|
|
|
|
2024-11-12 16:39:07 +00:00
|
|
|
enum priv_stack_mode {
|
|
|
|
|
PRIV_STACK_UNKNOWN,
|
|
|
|
|
NO_PRIV_STACK,
|
|
|
|
|
PRIV_STACK_ADAPTIVE,
|
|
|
|
|
};
|
|
|
|
|
|
2018-05-02 20:17:18 +00:00
|
|
|
struct bpf_subprog_info {
|
2019-11-14 18:57:16 +00:00
|
|
|
/* 'start' has to be the first field otherwise find_subprog() won't work */
|
2018-05-02 20:17:18 +00:00
|
|
|
u32 start; /* insn idx of function entry point */
|
2018-12-08 00:42:25 +00:00
|
|
|
u32 linfo_idx; /* The idx to the main_prog->aux->linfo */
|
2018-05-02 20:17:18 +00:00
|
|
|
u16 stack_depth; /* max. stack depth used by this function */
|
bpf: Introduce may_goto instruction
Introduce may_goto instruction that from the verifier pov is similar to
open coded iterators bpf_for()/bpf_repeat() and bpf_loop() helper, but it
doesn't iterate any objects.
In assembly 'may_goto' is a nop most of the time until bpf runtime has to
terminate the program for whatever reason. In the current implementation
may_goto has a hidden counter, but other mechanisms can be used.
For programs written in C the later patch introduces 'cond_break' macro
that combines 'may_goto' with 'break' statement and has similar semantics:
cond_break is a nop until bpf runtime has to break out of this loop.
It can be used in any normal "for" or "while" loop, like
for (i = zero; i < cnt; cond_break, i++) {
The verifier recognizes that may_goto is used in the program, reserves
additional 8 bytes of stack, initializes them in subprog prologue, and
replaces may_goto instruction with:
aux_reg = *(u64 *)(fp - 40)
if aux_reg == 0 goto pc+off
aux_reg -= 1
*(u64 *)(fp - 40) = aux_reg
may_goto instruction can be used by LLVM to implement __builtin_memcpy,
__builtin_strcmp.
may_goto is not a full substitute for bpf_for() macro.
bpf_for() doesn't have induction variable that verifiers sees,
so 'i' in bpf_for(i, 0, 100) is seen as imprecise and bounded.
But when the code is written as:
for (i = 0; i < 100; cond_break, i++)
the verifier see 'i' as precise constant zero,
hence cond_break (aka may_goto) doesn't help to converge the loop.
A static or global variable can be used as a workaround:
static int zero = 0;
for (i = zero; i < 100; cond_break, i++) // works!
may_goto works well with arena pointers that don't need to be bounds
checked on access. Load/store from arena returns imprecise unbounded
scalar and loops with may_goto pass the verifier.
Reserve new opcode BPF_JMP | BPF_JCOND for may_goto insn.
JCOND stands for conditional pseudo jump.
Since goto_or_nop insn was proposed, it may use the same opcode.
may_goto vs goto_or_nop can be distinguished by src_reg:
code = BPF_JMP | BPF_JCOND
src_reg = 0 - may_goto
src_reg = 1 - goto_or_nop
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Tested-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20240306031929.42666-2-alexei.starovoitov@gmail.com
2024-03-06 03:19:26 +00:00
|
|
|
u16 stack_extra;
|
2024-08-22 08:41:07 +00:00
|
|
|
/* offsets in range [stack_depth .. fastcall_stack_off)
|
|
|
|
|
* are used for bpf_fastcall spills and fills.
|
bpf: no_caller_saved_registers attribute for helper calls
GCC and LLVM define a no_caller_saved_registers function attribute.
This attribute means that function scratches only some of
the caller saved registers defined by ABI.
For BPF the set of such registers could be defined as follows:
- R0 is scratched only if function is non-void;
- R1-R5 are scratched only if corresponding parameter type is defined
in the function prototype.
This commit introduces flag bpf_func_prot->allow_nocsr.
If this flag is set for some helper function, verifier assumes that
it follows no_caller_saved_registers calling convention.
The contract between kernel and clang allows to simultaneously use
such functions and maintain backwards compatibility with old
kernels that don't understand no_caller_saved_registers calls
(nocsr for short):
- clang generates a simple pattern for nocsr calls, e.g.:
r1 = 1;
r2 = 2;
*(u64 *)(r10 - 8) = r1;
*(u64 *)(r10 - 16) = r2;
call %[to_be_inlined]
r2 = *(u64 *)(r10 - 16);
r1 = *(u64 *)(r10 - 8);
r0 = r1;
r0 += r2;
exit;
- kernel removes unnecessary spills and fills, if called function is
inlined by verifier or current JIT (with assumption that patch
inserted by verifier or JIT honors nocsr contract, e.g. does not
scratch r3-r5 for the example above), e.g. the code above would be
transformed to:
r1 = 1;
r2 = 2;
call %[to_be_inlined]
r0 = r1;
r0 += r2;
exit;
Technically, the transformation is split into the following phases:
- function mark_nocsr_patterns(), called from bpf_check()
searches and marks potential patterns in instruction auxiliary data;
- upon stack read or write access,
function check_nocsr_stack_contract() is used to verify if
stack offsets, presumably reserved for nocsr patterns, are used
only from those patterns;
- function remove_nocsr_spills_fills(), called from bpf_check(),
applies the rewrite for valid patterns.
See comment in mark_nocsr_pattern_for_call() for more details.
Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20240722233844.1406874-3-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2024-07-22 23:38:36 +00:00
|
|
|
*/
|
2024-08-22 08:41:07 +00:00
|
|
|
s16 fastcall_stack_off;
|
2023-12-04 23:39:22 +00:00
|
|
|
bool has_tail_call: 1;
|
|
|
|
|
bool tail_call_reachable: 1;
|
|
|
|
|
bool has_ld_abs: 1;
|
|
|
|
|
bool is_cb: 1;
|
|
|
|
|
bool is_async_cb: 1;
|
|
|
|
|
bool is_exception_cb: 1;
|
bpf: abstract away global subprog arg preparation logic from reg state setup
btf_prepare_func_args() is used to understand expectations and
restrictions on global subprog arguments. But current implementation is
hard to extend, as it intermixes BTF-based func prototype parsing and
interpretation logic with setting up register state at subprog entry.
Worse still, those registers are not completely set up inside
btf_prepare_func_args(), requiring some more logic later in
do_check_common(). Like calling mark_reg_unknown() and similar
initialization operations.
This intermixing of BTF interpretation and register state setup is
problematic. First, it causes duplication of BTF parsing logic for global
subprog verification (to set up initial state of global subprog) and
global subprog call sites analysis (when we need to check that whatever
is being passed into global subprog matches expectations), performed in
btf_check_subprog_call().
Given we want to extend global func argument with tags later, this
duplication is problematic. So refactor btf_prepare_func_args() to do
only BTF-based func proto and args parsing, returning high-level
argument "expectations" only, with no regard to specifics of register
state. I.e., if it's a context argument, instead of setting register
state to PTR_TO_CTX, we return ARG_PTR_TO_CTX enum for that argument as
"an argument specification" for further processing inside
do_check_common(). Similarly for SCALAR arguments, PTR_TO_MEM, etc.
This allows to reuse btf_prepare_func_args() in following patches at
global subprog call site analysis time. It also keeps register setup
code consistently in one place, do_check_common().
Besides all this, we cache this argument specs information inside
env->subprog_info, eliminating the need to redo these potentially
expensive BTF traversals, especially if BPF program's BTF is big and/or
there are lots of global subprog calls.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231215011334.2307144-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-15 01:13:25 +00:00
|
|
|
bool args_cached: 1;
|
2024-08-22 08:41:07 +00:00
|
|
|
/* true if bpf_fastcall stack region is used by functions that can't be inlined */
|
|
|
|
|
bool keep_fastcall_stack: 1;
|
2024-12-10 04:10:55 +00:00
|
|
|
bool changes_pkt_data: 1;
|
2025-03-01 15:18:44 +00:00
|
|
|
bool might_sleep: 1;
|
bpf: abstract away global subprog arg preparation logic from reg state setup
btf_prepare_func_args() is used to understand expectations and
restrictions on global subprog arguments. But current implementation is
hard to extend, as it intermixes BTF-based func prototype parsing and
interpretation logic with setting up register state at subprog entry.
Worse still, those registers are not completely set up inside
btf_prepare_func_args(), requiring some more logic later in
do_check_common(). Like calling mark_reg_unknown() and similar
initialization operations.
This intermixing of BTF interpretation and register state setup is
problematic. First, it causes duplication of BTF parsing logic for global
subprog verification (to set up initial state of global subprog) and
global subprog call sites analysis (when we need to check that whatever
is being passed into global subprog matches expectations), performed in
btf_check_subprog_call().
Given we want to extend global func argument with tags later, this
duplication is problematic. So refactor btf_prepare_func_args() to do
only BTF-based func proto and args parsing, returning high-level
argument "expectations" only, with no regard to specifics of register
state. I.e., if it's a context argument, instead of setting register
state to PTR_TO_CTX, we return ARG_PTR_TO_CTX enum for that argument as
"an argument specification" for further processing inside
do_check_common(). Similarly for SCALAR arguments, PTR_TO_MEM, etc.
This allows to reuse btf_prepare_func_args() in following patches at
global subprog call site analysis time. It also keeps register setup
code consistently in one place, do_check_common().
Besides all this, we cache this argument specs information inside
env->subprog_info, eliminating the need to redo these potentially
expensive BTF traversals, especially if BPF program's BTF is big and/or
there are lots of global subprog calls.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231215011334.2307144-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-15 01:13:25 +00:00
|
|
|
|
2024-11-12 16:39:07 +00:00
|
|
|
enum priv_stack_mode priv_stack_mode;
|
bpf: abstract away global subprog arg preparation logic from reg state setup
btf_prepare_func_args() is used to understand expectations and
restrictions on global subprog arguments. But current implementation is
hard to extend, as it intermixes BTF-based func prototype parsing and
interpretation logic with setting up register state at subprog entry.
Worse still, those registers are not completely set up inside
btf_prepare_func_args(), requiring some more logic later in
do_check_common(). Like calling mark_reg_unknown() and similar
initialization operations.
This intermixing of BTF interpretation and register state setup is
problematic. First, it causes duplication of BTF parsing logic for global
subprog verification (to set up initial state of global subprog) and
global subprog call sites analysis (when we need to check that whatever
is being passed into global subprog matches expectations), performed in
btf_check_subprog_call().
Given we want to extend global func argument with tags later, this
duplication is problematic. So refactor btf_prepare_func_args() to do
only BTF-based func proto and args parsing, returning high-level
argument "expectations" only, with no regard to specifics of register
state. I.e., if it's a context argument, instead of setting register
state to PTR_TO_CTX, we return ARG_PTR_TO_CTX enum for that argument as
"an argument specification" for further processing inside
do_check_common(). Similarly for SCALAR arguments, PTR_TO_MEM, etc.
This allows to reuse btf_prepare_func_args() in following patches at
global subprog call site analysis time. It also keeps register setup
code consistently in one place, do_check_common().
Besides all this, we cache this argument specs information inside
env->subprog_info, eliminating the need to redo these potentially
expensive BTF traversals, especially if BPF program's BTF is big and/or
there are lots of global subprog calls.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231215011334.2307144-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-15 01:13:25 +00:00
|
|
|
u8 arg_cnt;
|
|
|
|
|
struct bpf_subprog_arg_info args[MAX_BPF_FUNC_REG_ARGS];
|
2018-05-02 20:17:18 +00:00
|
|
|
};
|
|
|
|
|
|
2023-05-05 04:33:10 +00:00
|
|
|
struct bpf_verifier_env;
|
|
|
|
|
|
|
|
|
|
struct backtrack_state {
|
|
|
|
|
struct bpf_verifier_env *env;
|
|
|
|
|
u32 frame;
|
|
|
|
|
u32 reg_masks[MAX_CALL_FRAMES];
|
|
|
|
|
u64 stack_masks[MAX_CALL_FRAMES];
|
|
|
|
|
};
|
|
|
|
|
|
bpf: Verify scalar ids mapping in regsafe() using check_ids()
Make sure that the following unsafe example is rejected by verifier:
1: r9 = ... some pointer with range X ...
2: r6 = ... unbound scalar ID=a ...
3: r7 = ... unbound scalar ID=b ...
4: if (r6 > r7) goto +1
5: r6 = r7
6: if (r6 > X) goto ...
--- checkpoint ---
7: r9 += r7
8: *(u64 *)r9 = Y
This example is unsafe because not all execution paths verify r7 range.
Because of the jump at (4) the verifier would arrive at (6) in two states:
I. r6{.id=b}, r7{.id=b} via path 1-6;
II. r6{.id=a}, r7{.id=b} via path 1-4, 6.
Currently regsafe() does not call check_ids() for scalar registers,
thus from POV of regsafe() states (I) and (II) are identical. If the
path 1-6 is taken by verifier first, and checkpoint is created at (6)
the path [1-4, 6] would be considered safe.
Changes in this commit:
- check_ids() is modified to disallow mapping multiple old_id to the
same cur_id.
- check_scalar_ids() is added, unlike check_ids() it treats ID zero as
a unique scalar ID.
- check_scalar_ids() needs to generate temporary unique IDs, field
'tmp_id_gen' is added to bpf_verifier_env::idmap_scratch to
facilitate this.
- regsafe() is updated to:
- use check_scalar_ids() for precise scalar registers.
- compare scalar registers using memcmp only for explore_alu_limits
branch. This simplifies control flow for scalar case, and has no
measurable performance impact.
- check_alu_op() is updated to avoid generating bpf_reg_state::id for
constant scalar values when processing BPF_MOV. ID is needed to
propagate range information for identical values, but there is
nothing to propagate for constants.
Fixes: 75748837b7e5 ("bpf: Propagate scalar ranges through register assignments.")
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230613153824.3324830-4-eddyz87@gmail.com
2023-06-13 15:38:23 +00:00
|
|
|
struct bpf_id_pair {
|
|
|
|
|
u32 old;
|
|
|
|
|
u32 cur;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct bpf_idmap {
|
|
|
|
|
u32 tmp_id_gen;
|
|
|
|
|
struct bpf_id_pair map[BPF_ID_MAP_SIZE];
|
|
|
|
|
};
|
|
|
|
|
|
2023-06-13 15:38:21 +00:00
|
|
|
struct bpf_idset {
|
|
|
|
|
u32 count;
|
|
|
|
|
u32 ids[BPF_ID_MAP_SIZE];
|
|
|
|
|
};
|
|
|
|
|
|
bpf: propagate read/precision marks over state graph backedges
Current loop_entry-based exact states comparison logic does not handle
the following case:
.-> A --. Assume the states are visited in the order A, B, C.
| | | Assume that state B reaches a state equivalent to state A.
| v v At this point, state C is not processed yet, so state A
'-- B C has not received any read or precision marks from C.
As a result, these marks won't be propagated to B.
If B has incomplete marks, it is unsafe to use it in states_equal()
checks.
This commit replaces the existing logic with the following:
- Strongly connected components (SCCs) are computed over the program's
control flow graph (intraprocedurally).
- When a verifier state enters an SCC, that state is recorded as the
SCC entry point.
- When a verifier state is found equivalent to another (e.g., B to A
in the example), it is recorded as a states graph backedge.
Backedges are accumulated per SCC.
- When an SCC entry state reaches `branches == 0`, read and precision
marks are propagated through the backedges (e.g., from A to B, from
C to A, and then again from A to B).
To support nested subprogram calls, the entry state and backedge list
are associated not with the SCC itself but with an object called
`bpf_scc_callchain`. A callchain is a tuple `(callsite*, scc_id)`,
where `callsite` is the index of a call instruction for each frame
except the last.
See the comments added in `is_state_visited()` and
`compute_scc_callchain()` for more details.
Fixes: 2a0992829ea3 ("bpf: correct loop detection for iterators convergence")
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250611200836.4135542-8-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-06-11 20:08:33 +00:00
|
|
|
/* see verifier.c:compute_scc_callchain() */
|
|
|
|
|
struct bpf_scc_callchain {
|
|
|
|
|
/* call sites from bpf_verifier_state->frame[*]->callsite leading to this SCC */
|
|
|
|
|
u32 callsites[MAX_CALL_FRAMES - 1];
|
|
|
|
|
/* last frame in a chain is identified by SCC id */
|
|
|
|
|
u32 scc;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* verifier state waiting for propagate_backedges() */
|
|
|
|
|
struct bpf_scc_backedge {
|
|
|
|
|
struct bpf_scc_backedge *next;
|
|
|
|
|
struct bpf_verifier_state state;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct bpf_scc_visit {
|
|
|
|
|
struct bpf_scc_callchain callchain;
|
|
|
|
|
/* first state in current verification path that entered SCC
|
|
|
|
|
* identified by the callchain
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_verifier_state *entry_state;
|
|
|
|
|
struct bpf_scc_backedge *backedges; /* list of backedges */
|
2025-06-11 20:08:35 +00:00
|
|
|
u32 num_backedges;
|
bpf: propagate read/precision marks over state graph backedges
Current loop_entry-based exact states comparison logic does not handle
the following case:
.-> A --. Assume the states are visited in the order A, B, C.
| | | Assume that state B reaches a state equivalent to state A.
| v v At this point, state C is not processed yet, so state A
'-- B C has not received any read or precision marks from C.
As a result, these marks won't be propagated to B.
If B has incomplete marks, it is unsafe to use it in states_equal()
checks.
This commit replaces the existing logic with the following:
- Strongly connected components (SCCs) are computed over the program's
control flow graph (intraprocedurally).
- When a verifier state enters an SCC, that state is recorded as the
SCC entry point.
- When a verifier state is found equivalent to another (e.g., B to A
in the example), it is recorded as a states graph backedge.
Backedges are accumulated per SCC.
- When an SCC entry state reaches `branches == 0`, read and precision
marks are propagated through the backedges (e.g., from A to B, from
C to A, and then again from A to B).
To support nested subprogram calls, the entry state and backedge list
are associated not with the SCC itself but with an object called
`bpf_scc_callchain`. A callchain is a tuple `(callsite*, scc_id)`,
where `callsite` is the index of a call instruction for each frame
except the last.
See the comments added in `is_state_visited()` and
`compute_scc_callchain()` for more details.
Fixes: 2a0992829ea3 ("bpf: correct loop detection for iterators convergence")
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250611200836.4135542-8-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-06-11 20:08:33 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* An array of bpf_scc_visit structs sharing tht same bpf_scc_callchain->scc
|
|
|
|
|
* but having different bpf_scc_callchain->callsites.
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_scc_info {
|
|
|
|
|
u32 num_visits;
|
|
|
|
|
struct bpf_scc_visit visits[];
|
|
|
|
|
};
|
|
|
|
|
|
2016-09-21 10:43:57 +00:00
|
|
|
/* single container for all structs
|
|
|
|
|
* one verifier_env per bpf_check() call
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_verifier_env {
|
2019-01-02 23:58:27 +00:00
|
|
|
u32 insn_idx;
|
|
|
|
|
u32 prev_insn_idx;
|
2016-09-21 10:43:57 +00:00
|
|
|
struct bpf_prog *prog; /* eBPF program being verified */
|
2017-10-16 23:40:54 +00:00
|
|
|
const struct bpf_verifier_ops *ops;
|
2024-01-19 22:50:00 +00:00
|
|
|
struct module *attach_btf_mod; /* The owner module of prog->aux->attach_btf */
|
2016-09-21 10:43:57 +00:00
|
|
|
struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
|
|
|
|
|
int stack_size; /* number of states to be processed */
|
2017-05-10 18:38:07 +00:00
|
|
|
bool strict_alignment; /* perform strict pointer alignment checks */
|
2019-08-23 05:52:12 +00:00
|
|
|
bool test_state_freq; /* test verifier with different pruning frequency */
|
2023-11-17 17:14:04 +00:00
|
|
|
bool test_reg_invariants; /* fail verification on register invariants violations */
|
2017-11-01 01:16:05 +00:00
|
|
|
struct bpf_verifier_state *cur_state; /* current verifier state */
|
2025-02-15 11:03:59 +00:00
|
|
|
/* Search pruning optimization, array of list_heads for
|
|
|
|
|
* lists of struct bpf_verifier_state_list.
|
|
|
|
|
*/
|
|
|
|
|
struct list_head *explored_states;
|
|
|
|
|
struct list_head free_list; /* list of struct bpf_verifier_state_list */
|
2016-09-21 10:43:57 +00:00
|
|
|
struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
|
2021-01-12 07:55:18 +00:00
|
|
|
struct btf_mod_pair used_btfs[MAX_USED_BTFS]; /* array of BTF's used by BPF program */
|
2016-09-21 10:43:57 +00:00
|
|
|
u32 used_map_cnt; /* number of used maps */
|
2021-01-12 07:55:18 +00:00
|
|
|
u32 used_btf_cnt; /* number of used BTF objects */
|
2016-09-21 10:43:57 +00:00
|
|
|
u32 id_gen; /* used to generate unique reg IDs */
|
bpf: Implement support for adding hidden subprogs
Introduce support in the verifier for generating a subprogram and
include it as part of a BPF program dynamically after the do_check phase
is complete. The first user will be the next patch which generates
default exception callbacks if none are set for the program. The phase
of invocation will be do_misc_fixups. Note that this is an internal
verifier function, and should be used with instruction blocks which
uphold the invariants stated in check_subprogs.
Since these subprogs are always appended to the end of the instruction
sequence of the program, it becomes relatively inexpensive to do the
related adjustments to the subprog_info of the program. Only the fake
exit subprogram is shifted forward, making room for our new subprog.
This is useful to insert a new subprogram, get it JITed, and obtain its
function pointer. The next patch will use this functionality to insert a
default exception callback which will be invoked after unwinding the
stack.
Note that these added subprograms are invisible to userspace, and never
reported in BPF_OBJ_GET_INFO_BY_ID etc. For now, only a single
subprogram is supported, but more can be easily supported in the future.
To this end, two function counts are introduced now, the existing
func_cnt, and real_func_cnt, the latter including hidden programs. This
allows us to conver the JIT code to use the real_func_cnt for management
of resources while syscall path continues working with existing
func_cnt.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20230912233214.1518551-4-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-09-12 23:32:00 +00:00
|
|
|
u32 hidden_subprog_cnt; /* number of hidden subprogs */
|
bpf: Implement BPF exceptions
This patch implements BPF exceptions, and introduces a bpf_throw kfunc
to allow programs to throw exceptions during their execution at runtime.
A bpf_throw invocation is treated as an immediate termination of the
program, returning back to its caller within the kernel, unwinding all
stack frames.
This allows the program to simplify its implementation, by testing for
runtime conditions which the verifier has no visibility into, and assert
that they are true. In case they are not, the program can simply throw
an exception from the other branch.
BPF exceptions are explicitly *NOT* an unlikely slowpath error handling
primitive, and this objective has guided design choices of the
implementation of the them within the kernel (with the bulk of the cost
for unwinding the stack offloaded to the bpf_throw kfunc).
The implementation of this mechanism requires use of add_hidden_subprog
mechanism introduced in the previous patch, which generates a couple of
instructions to move R1 to R0 and exit. The JIT then rewrites the
prologue of this subprog to take the stack pointer and frame pointer as
inputs and reset the stack frame, popping all callee-saved registers
saved by the main subprog. The bpf_throw function then walks the stack
at runtime, and invokes this exception subprog with the stack and frame
pointers as parameters.
Reviewers must take note that currently the main program is made to save
all callee-saved registers on x86_64 during entry into the program. This
is because we must do an equivalent of a lightweight context switch when
unwinding the stack, therefore we need the callee-saved registers of the
caller of the BPF program to be able to return with a sane state.
Note that we have to additionally handle r12, even though it is not used
by the program, because when throwing the exception the program makes an
entry into the kernel which could clobber r12 after saving it on the
stack. To be able to preserve the value we received on program entry, we
push r12 and restore it from the generated subprogram when unwinding the
stack.
For now, bpf_throw invocation fails when lingering resources or locks
exist in that path of the program. In a future followup, bpf_throw will
be extended to perform frame-by-frame unwinding to release lingering
resources for each stack frame, removing this limitation.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20230912233214.1518551-5-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-09-12 23:32:01 +00:00
|
|
|
int exception_callback_subprog;
|
bpf: Fix pointer arithmetic mask tightening under state pruning
In 7fedb63a8307 ("bpf: Tighten speculative pointer arithmetic mask") we
narrowed the offset mask for unprivileged pointer arithmetic in order to
mitigate a corner case where in the speculative domain it is possible to
advance, for example, the map value pointer by up to value_size-1 out-of-
bounds in order to leak kernel memory via side-channel to user space.
The verifier's state pruning for scalars leaves one corner case open
where in the first verification path R_x holds an unknown scalar with an
aux->alu_limit of e.g. 7, and in a second verification path that same
register R_x, here denoted as R_x', holds an unknown scalar which has
tighter bounds and would thus satisfy range_within(R_x, R_x') as well as
tnum_in(R_x, R_x') for state pruning, yielding an aux->alu_limit of 3:
Given the second path fits the register constraints for pruning, the final
generated mask from aux->alu_limit will remain at 7. While technically
not wrong for the non-speculative domain, it would however be possible
to craft similar cases where the mask would be too wide as in 7fedb63a8307.
One way to fix it is to detect the presence of unknown scalar map pointer
arithmetic and force a deeper search on unknown scalars to ensure that
we do not run into a masking mismatch.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
2021-07-16 09:18:21 +00:00
|
|
|
bool explore_alu_limits;
|
2016-09-21 10:43:57 +00:00
|
|
|
bool allow_ptr_leaks;
|
2023-12-08 03:25:17 +00:00
|
|
|
/* Allow access to uninitialized stack memory. Writes with fixed offset are
|
|
|
|
|
* always allowed, so this refers to reads (with fixed or variable offset),
|
|
|
|
|
* to writes with variable offset and to indirect (helper) accesses.
|
|
|
|
|
*/
|
bpf: Allow variable-offset stack access
Before this patch, variable offset access to the stack was dissalowed
for regular instructions, but was allowed for "indirect" accesses (i.e.
helpers). This patch removes the restriction, allowing reading and
writing to the stack through stack pointers with variable offsets. This
makes stack-allocated buffers more usable in programs, and brings stack
pointers closer to other types of pointers.
The motivation is being able to use stack-allocated buffers for data
manipulation. When the stack size limit is sufficient, allocating
buffers on the stack is simpler than per-cpu arrays, or other
alternatives.
In unpriviledged programs, variable-offset reads and writes are
disallowed (they were already disallowed for the indirect access case)
because the speculative execution checking code doesn't support them.
Additionally, when writing through a variable-offset stack pointer, if
any pointers are in the accessible range, there's possilibities of later
leaking pointers because the write cannot be tracked precisely.
Writes with variable offset mark the whole range as initialized, even
though we don't know which stack slots are actually written. This is in
order to not reject future reads to these slots. Note that this doesn't
affect writes done through helpers; like before, helpers need the whole
stack range to be initialized to begin with.
All the stack slots are in range are considered scalars after the write;
variable-offset register spills are not tracked.
For reads, all the stack slots in the variable range needs to be
initialized (but see above about what writes do), otherwise the read is
rejected. All register spilled in stack slots that might be read are
marked as having been read, however reads through such pointers don't do
register filling; the target register will always be either a scalar or
a constant zero.
Signed-off-by: Andrei Matei <andreimatei1@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210207011027.676572-2-andreimatei1@gmail.com
2021-02-07 01:10:24 +00:00
|
|
|
bool allow_uninit_stack;
|
2020-05-13 23:03:54 +00:00
|
|
|
bool bpf_capable;
|
|
|
|
|
bool bypass_spec_v1;
|
|
|
|
|
bool bypass_spec_v4;
|
2016-09-21 10:43:57 +00:00
|
|
|
bool seen_direct_write;
|
bpf: Implement BPF exceptions
This patch implements BPF exceptions, and introduces a bpf_throw kfunc
to allow programs to throw exceptions during their execution at runtime.
A bpf_throw invocation is treated as an immediate termination of the
program, returning back to its caller within the kernel, unwinding all
stack frames.
This allows the program to simplify its implementation, by testing for
runtime conditions which the verifier has no visibility into, and assert
that they are true. In case they are not, the program can simply throw
an exception from the other branch.
BPF exceptions are explicitly *NOT* an unlikely slowpath error handling
primitive, and this objective has guided design choices of the
implementation of the them within the kernel (with the bulk of the cost
for unwinding the stack offloaded to the bpf_throw kfunc).
The implementation of this mechanism requires use of add_hidden_subprog
mechanism introduced in the previous patch, which generates a couple of
instructions to move R1 to R0 and exit. The JIT then rewrites the
prologue of this subprog to take the stack pointer and frame pointer as
inputs and reset the stack frame, popping all callee-saved registers
saved by the main subprog. The bpf_throw function then walks the stack
at runtime, and invokes this exception subprog with the stack and frame
pointers as parameters.
Reviewers must take note that currently the main program is made to save
all callee-saved registers on x86_64 during entry into the program. This
is because we must do an equivalent of a lightweight context switch when
unwinding the stack, therefore we need the callee-saved registers of the
caller of the BPF program to be able to return with a sane state.
Note that we have to additionally handle r12, even though it is not used
by the program, because when throwing the exception the program makes an
entry into the kernel which could clobber r12 after saving it on the
stack. To be able to preserve the value we received on program entry, we
push r12 and restore it from the generated subprogram when unwinding the
stack.
For now, bpf_throw invocation fails when lingering resources or locks
exist in that path of the program. In a future followup, bpf_throw will
be extended to perform frame-by-frame unwinding to release lingering
resources for each stack frame, removing this limitation.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20230912233214.1518551-5-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-09-12 23:32:01 +00:00
|
|
|
bool seen_exception;
|
2016-09-21 10:43:57 +00:00
|
|
|
struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
|
2018-12-13 18:41:48 +00:00
|
|
|
const struct bpf_line_info *prev_linfo;
|
2018-03-24 18:44:22 +00:00
|
|
|
struct bpf_verifier_log log;
|
bpf: Implement support for adding hidden subprogs
Introduce support in the verifier for generating a subprogram and
include it as part of a BPF program dynamically after the do_check phase
is complete. The first user will be the next patch which generates
default exception callbacks if none are set for the program. The phase
of invocation will be do_misc_fixups. Note that this is an internal
verifier function, and should be used with instruction blocks which
uphold the invariants stated in check_subprogs.
Since these subprogs are always appended to the end of the instruction
sequence of the program, it becomes relatively inexpensive to do the
related adjustments to the subprog_info of the program. Only the fake
exit subprogram is shifted forward, making room for our new subprog.
This is useful to insert a new subprogram, get it JITed, and obtain its
function pointer. The next patch will use this functionality to insert a
default exception callback which will be invoked after unwinding the
stack.
Note that these added subprograms are invisible to userspace, and never
reported in BPF_OBJ_GET_INFO_BY_ID etc. For now, only a single
subprogram is supported, but more can be easily supported in the future.
To this end, two function counts are introduced now, the existing
func_cnt, and real_func_cnt, the latter including hidden programs. This
allows us to conver the JIT code to use the real_func_cnt for management
of resources while syscall path continues working with existing
func_cnt.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20230912233214.1518551-4-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-09-12 23:32:00 +00:00
|
|
|
struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 2]; /* max + 2 for the fake and exception subprogs */
|
2023-06-13 15:38:21 +00:00
|
|
|
union {
|
bpf: Verify scalar ids mapping in regsafe() using check_ids()
Make sure that the following unsafe example is rejected by verifier:
1: r9 = ... some pointer with range X ...
2: r6 = ... unbound scalar ID=a ...
3: r7 = ... unbound scalar ID=b ...
4: if (r6 > r7) goto +1
5: r6 = r7
6: if (r6 > X) goto ...
--- checkpoint ---
7: r9 += r7
8: *(u64 *)r9 = Y
This example is unsafe because not all execution paths verify r7 range.
Because of the jump at (4) the verifier would arrive at (6) in two states:
I. r6{.id=b}, r7{.id=b} via path 1-6;
II. r6{.id=a}, r7{.id=b} via path 1-4, 6.
Currently regsafe() does not call check_ids() for scalar registers,
thus from POV of regsafe() states (I) and (II) are identical. If the
path 1-6 is taken by verifier first, and checkpoint is created at (6)
the path [1-4, 6] would be considered safe.
Changes in this commit:
- check_ids() is modified to disallow mapping multiple old_id to the
same cur_id.
- check_scalar_ids() is added, unlike check_ids() it treats ID zero as
a unique scalar ID.
- check_scalar_ids() needs to generate temporary unique IDs, field
'tmp_id_gen' is added to bpf_verifier_env::idmap_scratch to
facilitate this.
- regsafe() is updated to:
- use check_scalar_ids() for precise scalar registers.
- compare scalar registers using memcmp only for explore_alu_limits
branch. This simplifies control flow for scalar case, and has no
measurable performance impact.
- check_alu_op() is updated to avoid generating bpf_reg_state::id for
constant scalar values when processing BPF_MOV. ID is needed to
propagate range information for identical values, but there is
nothing to propagate for constants.
Fixes: 75748837b7e5 ("bpf: Propagate scalar ranges through register assignments.")
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230613153824.3324830-4-eddyz87@gmail.com
2023-06-13 15:38:23 +00:00
|
|
|
struct bpf_idmap idmap_scratch;
|
2023-06-13 15:38:21 +00:00
|
|
|
struct bpf_idset idset_scratch;
|
|
|
|
|
};
|
2019-04-19 14:44:54 +00:00
|
|
|
struct {
|
|
|
|
|
int *insn_state;
|
|
|
|
|
int *insn_stack;
|
bpf: simple DFA-based live registers analysis
Compute may-live registers before each instruction in the program.
The register is live before the instruction I if it is read by I or
some instruction S following I during program execution and is not
overwritten between I and S.
This information would be used in the next patch as a hint in
func_states_equal().
Use a simple algorithm described in [1] to compute this information:
- define the following:
- I.use : a set of all registers read by instruction I;
- I.def : a set of all registers written by instruction I;
- I.in : a set of all registers that may be alive before I execution;
- I.out : a set of all registers that may be alive after I execution;
- I.successors : a set of instructions S that might immediately
follow I for some program execution;
- associate separate empty sets 'I.in' and 'I.out' with each instruction;
- visit each instruction in a postorder and update corresponding
'I.in' and 'I.out' sets as follows:
I.out = U [S.in for S in I.successors]
I.in = (I.out / I.def) U I.use
(where U stands for set union, / stands for set difference)
- repeat the computation while I.{in,out} changes for any instruction.
On implementation side keep things as simple, as possible:
- check_cfg() already marks instructions EXPLORED in post-order,
modify it to save the index of each EXPLORED instruction in a vector;
- represent I.{in,out,use,def} as bitmasks;
- don't split the program into basic blocks and don't maintain the
work queue, instead:
- do fixed-point computation by visiting each instruction;
- maintain a simple 'changed' flag if I.{in,out} for any instruction
change;
Measurements show that even such simplistic implementation does not
add measurable verification time overhead (for selftests, at-least).
Note on check_cfg() ex_insn_beg/ex_done change:
To avoid out of bounds access to env->cfg.insn_postorder array,
it should be guaranteed that instruction transitions to EXPLORED state
only once. Previously this was not the fact for incorrect programs
with direct calls to exception callbacks.
The 'align' selftest needs adjustment to skip computed insn/live
registers printout. Otherwise it matches lines from the live registers
printout.
[1] https://en.wikipedia.org/wiki/Live-variable_analysis
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250304195024.2478889-4-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-03-04 19:50:22 +00:00
|
|
|
/* vector of instruction indexes sorted in post-order */
|
|
|
|
|
int *insn_postorder;
|
2019-04-19 14:44:54 +00:00
|
|
|
int cur_stack;
|
bpf: simple DFA-based live registers analysis
Compute may-live registers before each instruction in the program.
The register is live before the instruction I if it is read by I or
some instruction S following I during program execution and is not
overwritten between I and S.
This information would be used in the next patch as a hint in
func_states_equal().
Use a simple algorithm described in [1] to compute this information:
- define the following:
- I.use : a set of all registers read by instruction I;
- I.def : a set of all registers written by instruction I;
- I.in : a set of all registers that may be alive before I execution;
- I.out : a set of all registers that may be alive after I execution;
- I.successors : a set of instructions S that might immediately
follow I for some program execution;
- associate separate empty sets 'I.in' and 'I.out' with each instruction;
- visit each instruction in a postorder and update corresponding
'I.in' and 'I.out' sets as follows:
I.out = U [S.in for S in I.successors]
I.in = (I.out / I.def) U I.use
(where U stands for set union, / stands for set difference)
- repeat the computation while I.{in,out} changes for any instruction.
On implementation side keep things as simple, as possible:
- check_cfg() already marks instructions EXPLORED in post-order,
modify it to save the index of each EXPLORED instruction in a vector;
- represent I.{in,out,use,def} as bitmasks;
- don't split the program into basic blocks and don't maintain the
work queue, instead:
- do fixed-point computation by visiting each instruction;
- maintain a simple 'changed' flag if I.{in,out} for any instruction
change;
Measurements show that even such simplistic implementation does not
add measurable verification time overhead (for selftests, at-least).
Note on check_cfg() ex_insn_beg/ex_done change:
To avoid out of bounds access to env->cfg.insn_postorder array,
it should be guaranteed that instruction transitions to EXPLORED state
only once. Previously this was not the fact for incorrect programs
with direct calls to exception callbacks.
The 'align' selftest needs adjustment to skip computed insn/live
registers printout. Otherwise it matches lines from the live registers
printout.
[1] https://en.wikipedia.org/wiki/Live-variable_analysis
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250304195024.2478889-4-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-03-04 19:50:22 +00:00
|
|
|
/* current position in the insn_postorder vector */
|
|
|
|
|
int cur_postorder;
|
2019-04-19 14:44:54 +00:00
|
|
|
} cfg;
|
2023-05-05 04:33:10 +00:00
|
|
|
struct backtrack_state bt;
|
2025-06-11 20:08:26 +00:00
|
|
|
struct bpf_jmp_history_entry *cur_hist_ent;
|
2020-01-10 06:41:20 +00:00
|
|
|
u32 pass_cnt; /* number of times do_check() was called */
|
2017-12-15 01:55:05 +00:00
|
|
|
u32 subprog_cnt;
|
2019-04-02 04:27:40 +00:00
|
|
|
/* number of instructions analyzed by the verifier */
|
2019-06-15 19:12:20 +00:00
|
|
|
u32 prev_insn_processed, insn_processed;
|
|
|
|
|
/* number of jmps, calls, exits analyzed so far */
|
|
|
|
|
u32 prev_jmps_processed, jmps_processed;
|
2019-04-02 04:27:40 +00:00
|
|
|
/* total verification time */
|
|
|
|
|
u64 verification_time;
|
|
|
|
|
/* maximum number of verifier states kept in 'branching' instructions */
|
|
|
|
|
u32 max_states_per_insn;
|
|
|
|
|
/* total number of allocated verifier states */
|
|
|
|
|
u32 total_states;
|
|
|
|
|
/* some states are freed during program analysis.
|
|
|
|
|
* this is peak number of states. this number dominates kernel
|
|
|
|
|
* memory consumption during verification
|
|
|
|
|
*/
|
|
|
|
|
u32 peak_states;
|
|
|
|
|
/* longest register parentage chain walked for liveness marking */
|
|
|
|
|
u32 longest_mark_read_walk;
|
2025-02-15 11:04:01 +00:00
|
|
|
u32 free_list_size;
|
|
|
|
|
u32 explored_states_size;
|
2025-06-11 20:08:35 +00:00
|
|
|
u32 num_backedges;
|
2021-05-14 00:36:10 +00:00
|
|
|
bpfptr_t fd_array;
|
bpf: Only print scratched registers and stack slots to verifier logs.
When printing verifier state for any log level, print full verifier
state only on function calls or on errors. Otherwise, only print the
registers and stack slots that were accessed.
Log size differences:
verif_scale_loop6 before: 234566564
verif_scale_loop6 after: 72143943
69% size reduction
kfree_skb before: 166406
kfree_skb after: 55386
69% size reduction
Before:
156: (61) r0 = *(u32 *)(r1 +0)
157: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=ctx(id=0,off=0,imm=0) R2_w=invP0 R10=fp0 fp-8_w=00000000 fp-16_w=00\
000000 fp-24_w=00000000 fp-32_w=00000000 fp-40_w=00000000 fp-48_w=00000000 fp-56_w=00000000 fp-64_w=00000000 fp-72_w=00000000 fp-80_w=00000\
000 fp-88_w=00000000 fp-96_w=00000000 fp-104_w=00000000 fp-112_w=00000000 fp-120_w=00000000 fp-128_w=00000000 fp-136_w=00000000 fp-144_w=00\
000000 fp-152_w=00000000 fp-160_w=00000000 fp-168_w=00000000 fp-176_w=00000000 fp-184_w=00000000 fp-192_w=00000000 fp-200_w=00000000 fp-208\
_w=00000000 fp-216_w=00000000 fp-224_w=00000000 fp-232_w=00000000 fp-240_w=00000000 fp-248_w=00000000 fp-256_w=00000000 fp-264_w=00000000 f\
p-272_w=00000000 fp-280_w=00000000 fp-288_w=00000000 fp-296_w=00000000 fp-304_w=00000000 fp-312_w=00000000 fp-320_w=00000000 fp-328_w=00000\
000 fp-336_w=00000000 fp-344_w=00000000 fp-352_w=00000000 fp-360_w=00000000 fp-368_w=00000000 fp-376_w=00000000 fp-384_w=00000000 fp-392_w=\
00000000 fp-400_w=00000000 fp-408_w=00000000 fp-416_w=00000000 fp-424_w=00000000 fp-432_w=00000000 fp-440_w=00000000 fp-448_w=00000000
; return skb->len;
157: (95) exit
Func#4 is safe for any args that match its prototype
Validating get_constant() func#5...
158: R1=invP(id=0) R10=fp0
; int get_constant(long val)
158: (bf) r0 = r1
159: R0_w=invP(id=1) R1=invP(id=1) R10=fp0
; return val - 122;
159: (04) w0 += -122
160: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=invP(id=1) R10=fp0
; return val - 122;
160: (95) exit
Func#5 is safe for any args that match its prototype
Validating get_skb_ifindex() func#6...
161: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R3=invP(id=0) R10=fp0
; int get_skb_ifindex(int val, struct __sk_buff *skb, int var)
161: (bc) w0 = w3
162: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R3=invP(id=0) R10=fp0
After:
156: (61) r0 = *(u32 *)(r1 +0)
157: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=ctx(id=0,off=0,imm=0)
; return skb->len;
157: (95) exit
Func#4 is safe for any args that match its prototype
Validating get_constant() func#5...
158: R1=invP(id=0) R10=fp0
; int get_constant(long val)
158: (bf) r0 = r1
159: R0_w=invP(id=1) R1=invP(id=1)
; return val - 122;
159: (04) w0 += -122
160: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return val - 122;
160: (95) exit
Func#5 is safe for any args that match its prototype
Validating get_skb_ifindex() func#6...
161: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R3=invP(id=0) R10=fp0
; int get_skb_ifindex(int val, struct __sk_buff *skb, int var)
161: (bc) w0 = w3
162: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R3=invP(id=0)
Signed-off-by: Christy Lee <christylee@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211216213358.3374427-2-christylee@fb.com
2021-12-16 21:33:56 +00:00
|
|
|
|
|
|
|
|
/* bit mask to keep track of whether a register has been accessed
|
|
|
|
|
* since the last time the function state was printed
|
|
|
|
|
*/
|
|
|
|
|
u32 scratched_regs;
|
|
|
|
|
/* Same as scratched_regs but for stack slots */
|
|
|
|
|
u64 scratched_stack_slots;
|
bpf: Switch BPF verifier log to be a rotating log by default
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
2023-04-06 23:41:49 +00:00
|
|
|
u64 prev_log_pos, prev_insn_print_pos;
|
bpf: Fix reg_set_min_max corruption of fake_reg
Juan reported that after doing some changes to buzzer [0] and implementing
a new fuzzing strategy guided by coverage, they noticed the following in
one of the probes:
[...]
13: (79) r6 = *(u64 *)(r0 +0) ; R0=map_value(ks=4,vs=8) R6_w=scalar()
14: (b7) r0 = 0 ; R0_w=0
15: (b4) w0 = -1 ; R0_w=0xffffffff
16: (74) w0 >>= 1 ; R0_w=0x7fffffff
17: (5c) w6 &= w0 ; R0_w=0x7fffffff R6_w=scalar(smin=smin32=0,smax=umax=umax32=0x7fffffff,var_off=(0x0; 0x7fffffff))
18: (44) w6 |= 2 ; R6_w=scalar(smin=umin=smin32=umin32=2,smax=umax=umax32=0x7fffffff,var_off=(0x2; 0x7ffffffd))
19: (56) if w6 != 0x7ffffffd goto pc+1
REG INVARIANTS VIOLATION (true_reg2): range bounds violation u64=[0x7fffffff, 0x7ffffffd] s64=[0x7fffffff, 0x7ffffffd] u32=[0x7fffffff, 0x7ffffffd] s32=[0x7fffffff, 0x7ffffffd] var_off=(0x7fffffff, 0x0)
REG INVARIANTS VIOLATION (false_reg1): range bounds violation u64=[0x7fffffff, 0x7ffffffd] s64=[0x7fffffff, 0x7ffffffd] u32=[0x7fffffff, 0x7ffffffd] s32=[0x7fffffff, 0x7ffffffd] var_off=(0x7fffffff, 0x0)
REG INVARIANTS VIOLATION (false_reg2): const tnum out of sync with range bounds u64=[0x0, 0xffffffffffffffff] s64=[0x8000000000000000, 0x7fffffffffffffff] u32=[0x0, 0xffffffff] s32=[0x80000000, 0x7fffffff] var_off=(0x7fffffff, 0x0)
19: R6_w=0x7fffffff
20: (95) exit
from 19 to 21: R0=0x7fffffff R6=scalar(smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x7ffffffe,var_off=(0x2; 0x7ffffffd)) R7=map_ptr(ks=4,vs=8) R9=ctx() R10=fp0 fp-24=map_ptr(ks=4,vs=8) fp-40=mmmmmmmm
21: R0=0x7fffffff R6=scalar(smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x7ffffffe,var_off=(0x2; 0x7ffffffd)) R7=map_ptr(ks=4,vs=8) R9=ctx() R10=fp0 fp-24=map_ptr(ks=4,vs=8) fp-40=mmmmmmmm
21: (14) w6 -= 2147483632 ; R6_w=scalar(smin=umin=umin32=2,smax=umax=0xffffffff,smin32=0x80000012,smax32=14,var_off=(0x2; 0xfffffffd))
22: (76) if w6 s>= 0xe goto pc+1 ; R6_w=scalar(smin=umin=umin32=2,smax=umax=0xffffffff,smin32=0x80000012,smax32=13,var_off=(0x2; 0xfffffffd))
23: (95) exit
from 22 to 24: R0=0x7fffffff R6_w=14 R7=map_ptr(ks=4,vs=8) R9=ctx() R10=fp0 fp-24=map_ptr(ks=4,vs=8) fp-40=mmmmmmmm
24: R0=0x7fffffff R6_w=14 R7=map_ptr(ks=4,vs=8) R9=ctx() R10=fp0 fp-24=map_ptr(ks=4,vs=8) fp-40=mmmmmmmm
24: (14) w6 -= 14 ; R6_w=0
[...]
What can be seen here is a register invariant violation on line 19. After
the binary-or in line 18, the verifier knows that bit 2 is set but knows
nothing about the rest of the content which was loaded from a map value,
meaning, range is [2,0x7fffffff] with var_off=(0x2; 0x7ffffffd). When in
line 19 the verifier analyzes the branch, it splits the register states
in reg_set_min_max() into the registers of the true branch (true_reg1,
true_reg2) and the registers of the false branch (false_reg1, false_reg2).
Since the test is w6 != 0x7ffffffd, the src_reg is a known constant.
Internally, the verifier creates a "fake" register initialized as scalar
to the value of 0x7ffffffd, and then passes it onto reg_set_min_max(). Now,
for line 19, it is mathematically impossible to take the false branch of
this program, yet the verifier analyzes it. It is impossible because the
second bit of r6 will be set due to the prior or operation and the
constant in the condition has that bit unset (hex(fd) == binary(1111 1101).
When the verifier first analyzes the false / fall-through branch, it will
compute an intersection between the var_off of r6 and of the constant. This
is because the verifier creates a "fake" register initialized to the value
of the constant. The intersection result later refines both registers in
regs_refine_cond_op():
[...]
t = tnum_intersect(tnum_subreg(reg1->var_off), tnum_subreg(reg2->var_off));
reg1->var_off = tnum_with_subreg(reg1->var_off, t);
reg2->var_off = tnum_with_subreg(reg2->var_off, t);
[...]
Since the verifier is analyzing the false branch of the conditional jump,
reg1 is equal to false_reg1 and reg2 is equal to false_reg2, i.e. the reg2
is the "fake" register that was meant to hold a constant value. The resulting
var_off of the intersection says that both registers now hold a known value
of var_off=(0x7fffffff, 0x0) or in other words: this operation manages to
make the verifier think that the "constant" value that was passed in the
jump operation now holds a different value.
Normally this would not be an issue since it should not influence the true
branch, however, false_reg2 and true_reg2 are pointers to the same "fake"
register. Meaning, the false branch can influence the results of the true
branch. In line 24, the verifier assumes R6_w=0, but the actual runtime
value in this case is 1. The fix is simply not passing in the same "fake"
register location as inputs to reg_set_min_max(), but instead making a
copy. Moving the fake_reg into the env also reduces stack consumption by
120 bytes. With this, the verifier successfully rejects invalid accesses
from the test program.
[0] https://github.com/google/buzzer
Fixes: 67420501e868 ("bpf: generalize reg_set_min_max() to handle non-const register comparisons")
Reported-by: Juan José López Jaimez <jjlopezjaimez@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/r/20240613115310.25383-1-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2024-06-13 11:53:08 +00:00
|
|
|
/* buffer used to temporary hold constants as scalar registers */
|
|
|
|
|
struct bpf_reg_state fake_reg[2];
|
2023-05-05 04:33:11 +00:00
|
|
|
/* buffer used to generate temporary string representations,
|
|
|
|
|
* e.g., in reg_type_str() to generate reg_type string
|
|
|
|
|
*/
|
|
|
|
|
char tmp_str_buf[TMP_STR_BUF_LEN];
|
2024-08-29 21:08:23 +00:00
|
|
|
struct bpf_insn insn_buf[INSN_BUF_SIZE];
|
2024-08-29 21:08:25 +00:00
|
|
|
struct bpf_insn epilogue_buf[INSN_BUF_SIZE];
|
2025-07-03 14:11:17 +00:00
|
|
|
struct bpf_scc_callchain callchain_buf;
|
bpf: propagate read/precision marks over state graph backedges
Current loop_entry-based exact states comparison logic does not handle
the following case:
.-> A --. Assume the states are visited in the order A, B, C.
| | | Assume that state B reaches a state equivalent to state A.
| v v At this point, state C is not processed yet, so state A
'-- B C has not received any read or precision marks from C.
As a result, these marks won't be propagated to B.
If B has incomplete marks, it is unsafe to use it in states_equal()
checks.
This commit replaces the existing logic with the following:
- Strongly connected components (SCCs) are computed over the program's
control flow graph (intraprocedurally).
- When a verifier state enters an SCC, that state is recorded as the
SCC entry point.
- When a verifier state is found equivalent to another (e.g., B to A
in the example), it is recorded as a states graph backedge.
Backedges are accumulated per SCC.
- When an SCC entry state reaches `branches == 0`, read and precision
marks are propagated through the backedges (e.g., from A to B, from
C to A, and then again from A to B).
To support nested subprogram calls, the entry state and backedge list
are associated not with the SCC itself but with an object called
`bpf_scc_callchain`. A callchain is a tuple `(callsite*, scc_id)`,
where `callsite` is the index of a call instruction for each frame
except the last.
See the comments added in `is_state_visited()` and
`compute_scc_callchain()` for more details.
Fixes: 2a0992829ea3 ("bpf: correct loop detection for iterators convergence")
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20250611200836.4135542-8-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2025-06-11 20:08:33 +00:00
|
|
|
/* array of pointers to bpf_scc_info indexed by SCC id */
|
|
|
|
|
struct bpf_scc_info **scc_info;
|
|
|
|
|
u32 scc_cnt;
|
2016-09-21 10:43:57 +00:00
|
|
|
};
|
|
|
|
|
|
2023-12-15 01:13:27 +00:00
|
|
|
static inline struct bpf_func_info_aux *subprog_aux(struct bpf_verifier_env *env, int subprog)
|
|
|
|
|
{
|
|
|
|
|
return &env->prog->aux->func_info_aux[subprog];
|
|
|
|
|
}
|
|
|
|
|
|
bpf: abstract away global subprog arg preparation logic from reg state setup
btf_prepare_func_args() is used to understand expectations and
restrictions on global subprog arguments. But current implementation is
hard to extend, as it intermixes BTF-based func prototype parsing and
interpretation logic with setting up register state at subprog entry.
Worse still, those registers are not completely set up inside
btf_prepare_func_args(), requiring some more logic later in
do_check_common(). Like calling mark_reg_unknown() and similar
initialization operations.
This intermixing of BTF interpretation and register state setup is
problematic. First, it causes duplication of BTF parsing logic for global
subprog verification (to set up initial state of global subprog) and
global subprog call sites analysis (when we need to check that whatever
is being passed into global subprog matches expectations), performed in
btf_check_subprog_call().
Given we want to extend global func argument with tags later, this
duplication is problematic. So refactor btf_prepare_func_args() to do
only BTF-based func proto and args parsing, returning high-level
argument "expectations" only, with no regard to specifics of register
state. I.e., if it's a context argument, instead of setting register
state to PTR_TO_CTX, we return ARG_PTR_TO_CTX enum for that argument as
"an argument specification" for further processing inside
do_check_common(). Similarly for SCALAR arguments, PTR_TO_MEM, etc.
This allows to reuse btf_prepare_func_args() in following patches at
global subprog call site analysis time. It also keeps register setup
code consistently in one place, do_check_common().
Besides all this, we cache this argument specs information inside
env->subprog_info, eliminating the need to redo these potentially
expensive BTF traversals, especially if BPF program's BTF is big and/or
there are lots of global subprog calls.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20231215011334.2307144-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-12-15 01:13:25 +00:00
|
|
|
static inline struct bpf_subprog_info *subprog_info(struct bpf_verifier_env *env, int subprog)
|
|
|
|
|
{
|
|
|
|
|
return &env->subprog_info[subprog];
|
|
|
|
|
}
|
|
|
|
|
|
2018-05-16 20:27:41 +00:00
|
|
|
__printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log,
|
|
|
|
|
const char *fmt, va_list args);
|
2018-01-10 12:26:06 +00:00
|
|
|
__printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
|
|
|
|
|
const char *fmt, ...);
|
2019-10-16 03:25:00 +00:00
|
|
|
__printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
|
|
|
|
|
const char *fmt, ...);
|
2023-04-06 23:41:59 +00:00
|
|
|
int bpf_vlog_init(struct bpf_verifier_log *log, u32 log_level,
|
|
|
|
|
char __user *log_buf, u32 log_size);
|
bpf: Switch BPF verifier log to be a rotating log by default
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
2023-04-06 23:41:49 +00:00
|
|
|
void bpf_vlog_reset(struct bpf_verifier_log *log, u64 new_pos);
|
2023-04-06 23:41:59 +00:00
|
|
|
int bpf_vlog_finalize(struct bpf_verifier_log *log, u32 *log_size_actual);
|
2018-01-10 12:26:06 +00:00
|
|
|
|
2023-11-18 03:46:16 +00:00
|
|
|
__printf(3, 4) void verbose_linfo(struct bpf_verifier_env *env,
|
|
|
|
|
u32 insn_off,
|
|
|
|
|
const char *prefix_fmt, ...);
|
|
|
|
|
|
2025-05-19 13:43:57 +00:00
|
|
|
#define verifier_bug_if(cond, env, fmt, args...) \
|
|
|
|
|
({ \
|
|
|
|
|
bool __cond = (cond); \
|
|
|
|
|
if (unlikely(__cond)) { \
|
|
|
|
|
BPF_WARN_ONCE(1, "verifier bug: " fmt "(" #cond ")\n", ##args); \
|
|
|
|
|
bpf_log(&env->log, "verifier bug: " fmt "(" #cond ")\n", ##args); \
|
|
|
|
|
} \
|
|
|
|
|
(__cond); \
|
|
|
|
|
})
|
|
|
|
|
#define verifier_bug(env, fmt, args...) verifier_bug_if(1, env, fmt, ##args)
|
|
|
|
|
|
2018-10-02 20:35:35 +00:00
|
|
|
static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env)
|
2017-11-01 01:16:05 +00:00
|
|
|
{
|
2017-12-15 01:55:06 +00:00
|
|
|
struct bpf_verifier_state *cur = env->cur_state;
|
|
|
|
|
|
2018-10-02 20:35:35 +00:00
|
|
|
return cur->frame[cur->curframe];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env)
|
|
|
|
|
{
|
|
|
|
|
return cur_func(env)->regs;
|
2017-11-01 01:16:05 +00:00
|
|
|
}
|
|
|
|
|
|
2018-11-09 13:03:31 +00:00
|
|
|
int bpf_prog_offload_verifier_prep(struct bpf_prog *prog);
|
2017-12-28 02:39:05 +00:00
|
|
|
int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
|
|
|
|
|
int insn_idx, int prev_insn_idx);
|
2018-10-07 11:56:47 +00:00
|
|
|
int bpf_prog_offload_finalize(struct bpf_verifier_env *env);
|
2019-01-23 06:45:24 +00:00
|
|
|
void
|
|
|
|
|
bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
|
|
|
|
|
struct bpf_insn *insn);
|
|
|
|
|
void
|
|
|
|
|
bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
|
2017-11-03 20:56:17 +00:00
|
|
|
|
2020-09-25 21:25:02 +00:00
|
|
|
/* this lives here instead of in bpf.h because it needs to dereference tgt_prog */
|
|
|
|
|
static inline u64 bpf_trampoline_compute_key(const struct bpf_prog *tgt_prog,
|
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.
This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.
Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.
Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
2020-12-03 20:46:29 +00:00
|
|
|
struct btf *btf, u32 btf_id)
|
2020-09-25 21:25:02 +00:00
|
|
|
{
|
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.
This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.
Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.
Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
2020-12-03 20:46:29 +00:00
|
|
|
if (tgt_prog)
|
|
|
|
|
return ((u64)tgt_prog->aux->id << 32) | btf_id;
|
|
|
|
|
else
|
|
|
|
|
return ((u64)btf_obj_id(btf) << 32) | 0x80000000 | btf_id;
|
2020-09-25 21:25:02 +00:00
|
|
|
}
|
|
|
|
|
|
2021-04-13 09:16:06 +00:00
|
|
|
/* unpack the IDs from the key as constructed above */
|
|
|
|
|
static inline void bpf_trampoline_unpack_key(u64 key, u32 *obj_id, u32 *btf_id)
|
|
|
|
|
{
|
|
|
|
|
if (obj_id)
|
|
|
|
|
*obj_id = key >> 32;
|
|
|
|
|
if (btf_id)
|
|
|
|
|
*btf_id = key & 0x7FFFFFFF;
|
|
|
|
|
}
|
|
|
|
|
|
2020-09-25 21:25:02 +00:00
|
|
|
int bpf_check_attach_target(struct bpf_verifier_log *log,
|
|
|
|
|
const struct bpf_prog *prog,
|
|
|
|
|
const struct bpf_prog *tgt_prog,
|
|
|
|
|
u32 btf_id,
|
|
|
|
|
struct bpf_attach_target_info *tgt_info);
|
bpf: Introduce BPF support for kernel module function calls
This change adds support on the kernel side to allow for BPF programs to
call kernel module functions. Userspace will prepare an array of module
BTF fds that is passed in during BPF_PROG_LOAD using fd_array parameter.
In the kernel, the module BTFs are placed in the auxilliary struct for
bpf_prog, and loaded as needed.
The verifier then uses insn->off to index into the fd_array. insn->off
0 is reserved for vmlinux BTF (for backwards compat), so userspace must
use an fd_array index > 0 for module kfunc support. kfunc_btf_tab is
sorted based on offset in an array, and each offset corresponds to one
descriptor, with a max limit up to 256 such module BTFs.
We also change existing kfunc_tab to distinguish each element based on
imm, off pair as each such call will now be distinct.
Another change is to check_kfunc_call callback, which now include a
struct module * pointer, this is to be used in later patch such that the
kfunc_id and module pointer are matched for dynamically registered BTF
sets from loadable modules, so that same kfunc_id in two modules doesn't
lead to check_kfunc_call succeeding. For the duration of the
check_kfunc_call, the reference to struct module exists, as it returns
the pointer stored in kfunc_btf_tab.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211002011757.311265-2-memxor@gmail.com
2021-10-02 01:17:49 +00:00
|
|
|
void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab);
|
|
|
|
|
|
2022-09-06 15:13:02 +00:00
|
|
|
int mark_chain_precision(struct bpf_verifier_env *env, int regno);
|
|
|
|
|
|
bpf: Introduce composable reg, ret and arg types.
There are some common properties shared between bpf reg, ret and arg
values. For instance, a value may be a NULL pointer, or a pointer to
a read-only memory. Previously, to express these properties, enumeration
was used. For example, in order to test whether a reg value can be NULL,
reg_type_may_be_null() simply enumerates all types that are possibly
NULL. The problem of this approach is that it's not scalable and causes
a lot of duplication. These properties can be combined, for example, a
type could be either MAYBE_NULL or RDONLY, or both.
This patch series rewrites the layout of reg_type, arg_type and
ret_type, so that common properties can be extracted and represented as
composable flag. For example, one can write
ARG_PTR_TO_MEM | PTR_MAYBE_NULL
which is equivalent to the previous
ARG_PTR_TO_MEM_OR_NULL
The type ARG_PTR_TO_MEM are called "base type" in this patch. Base
types can be extended with flags. A flag occupies the higher bits while
base types sits in the lower bits.
This patch in particular sets up a set of macro for this purpose. The
following patches will rewrite arg_types, ret_types and reg_types
respectively.
Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-2-haoluo@google.com
2021-12-17 00:31:44 +00:00
|
|
|
#define BPF_BASE_TYPE_MASK GENMASK(BPF_BASE_TYPE_BITS - 1, 0)
|
|
|
|
|
|
|
|
|
|
/* extract base type from bpf_{arg, return, reg}_type. */
|
|
|
|
|
static inline u32 base_type(u32 type)
|
|
|
|
|
{
|
|
|
|
|
return type & BPF_BASE_TYPE_MASK;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* extract flags from an extended type. See bpf_type_flag in bpf.h. */
|
|
|
|
|
static inline u32 type_flag(u32 type)
|
|
|
|
|
{
|
|
|
|
|
return type & ~BPF_BASE_TYPE_MASK;
|
|
|
|
|
}
|
2020-09-25 21:25:02 +00:00
|
|
|
|
2022-03-30 01:14:56 +00:00
|
|
|
/* only use after check_attach_btf_id() */
|
2022-10-25 18:45:16 +00:00
|
|
|
static inline enum bpf_prog_type resolve_prog_type(const struct bpf_prog *prog)
|
bpf: Add reference tracking support to kfunc
This patch adds verifier support for PTR_TO_BTF_ID return type of kfunc
to be a reference, by reusing acquire_reference_state/release_reference
support for existing in-kernel bpf helpers.
We make use of the three kfunc types:
- BTF_KFUNC_TYPE_ACQUIRE
Return true if kfunc_btf_id is an acquire kfunc. This will
acquire_reference_state for the returned PTR_TO_BTF_ID (this is the
only allow return value). Note that acquire kfunc must always return a
PTR_TO_BTF_ID{_OR_NULL}, otherwise the program is rejected.
- BTF_KFUNC_TYPE_RELEASE
Return true if kfunc_btf_id is a release kfunc. This will release the
reference to the passed in PTR_TO_BTF_ID which has a reference state
(from earlier acquire kfunc).
The btf_check_func_arg_match returns the regno (of argument register,
hence > 0) if the kfunc is a release kfunc, and a proper referenced
PTR_TO_BTF_ID is being passed to it.
This is similar to how helper call check uses bpf_call_arg_meta to
store the ref_obj_id that is later used to release the reference.
Similar to in-kernel helper, we only allow passing one referenced
PTR_TO_BTF_ID as an argument. It can also be passed in to normal
kfunc, but in case of release kfunc there must always be one
PTR_TO_BTF_ID argument that is referenced.
- BTF_KFUNC_TYPE_RET_NULL
For kfunc returning PTR_TO_BTF_ID, tells if it can be NULL, hence
force caller to mark the pointer not null (using check) before
accessing it. Note that taking into account the case fixed by commit
93c230e3f5bd ("bpf: Enforce id generation for all may-be-null register type")
we assign a non-zero id for mark_ptr_or_null_reg logic. Later, if more
return types are supported by kfunc, which have a _OR_NULL variant, it
might be better to move this id generation under a common
reg_type_may_be_null check, similar to the case in the commit.
Referenced PTR_TO_BTF_ID is currently only limited to kfunc, but can be
extended in the future to other BPF helpers as well. For now, we can
rely on the btf_struct_ids_match check to ensure we get the pointer to
the expected struct type. In the future, care needs to be taken to avoid
ambiguity for reference PTR_TO_BTF_ID passed to release function, in
case multiple candidates can release same BTF ID.
e.g. there might be two release kfuncs (or kfunc and helper):
foo(struct abc *p);
bar(struct abc *p);
... such that both release a PTR_TO_BTF_ID with btf_id of struct abc. In
this case we would need to track the acquire function corresponding to
the release function to avoid type confusion, and store this information
in the register state so that an incorrect program can be rejected. This
is not a problem right now, hence it is left as an exercise for the
future patch introducing such a case in the kernel.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220114163953.1455836-6-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2022-01-14 16:39:48 +00:00
|
|
|
{
|
2024-07-28 11:46:11 +00:00
|
|
|
return (prog->type == BPF_PROG_TYPE_EXT && prog->aux->saved_dst_prog_type) ?
|
|
|
|
|
prog->aux->saved_dst_prog_type : prog->type;
|
bpf: Add reference tracking support to kfunc
This patch adds verifier support for PTR_TO_BTF_ID return type of kfunc
to be a reference, by reusing acquire_reference_state/release_reference
support for existing in-kernel bpf helpers.
We make use of the three kfunc types:
- BTF_KFUNC_TYPE_ACQUIRE
Return true if kfunc_btf_id is an acquire kfunc. This will
acquire_reference_state for the returned PTR_TO_BTF_ID (this is the
only allow return value). Note that acquire kfunc must always return a
PTR_TO_BTF_ID{_OR_NULL}, otherwise the program is rejected.
- BTF_KFUNC_TYPE_RELEASE
Return true if kfunc_btf_id is a release kfunc. This will release the
reference to the passed in PTR_TO_BTF_ID which has a reference state
(from earlier acquire kfunc).
The btf_check_func_arg_match returns the regno (of argument register,
hence > 0) if the kfunc is a release kfunc, and a proper referenced
PTR_TO_BTF_ID is being passed to it.
This is similar to how helper call check uses bpf_call_arg_meta to
store the ref_obj_id that is later used to release the reference.
Similar to in-kernel helper, we only allow passing one referenced
PTR_TO_BTF_ID as an argument. It can also be passed in to normal
kfunc, but in case of release kfunc there must always be one
PTR_TO_BTF_ID argument that is referenced.
- BTF_KFUNC_TYPE_RET_NULL
For kfunc returning PTR_TO_BTF_ID, tells if it can be NULL, hence
force caller to mark the pointer not null (using check) before
accessing it. Note that taking into account the case fixed by commit
93c230e3f5bd ("bpf: Enforce id generation for all may-be-null register type")
we assign a non-zero id for mark_ptr_or_null_reg logic. Later, if more
return types are supported by kfunc, which have a _OR_NULL variant, it
might be better to move this id generation under a common
reg_type_may_be_null check, similar to the case in the commit.
Referenced PTR_TO_BTF_ID is currently only limited to kfunc, but can be
extended in the future to other BPF helpers as well. For now, we can
rely on the btf_struct_ids_match check to ensure we get the pointer to
the expected struct type. In the future, care needs to be taken to avoid
ambiguity for reference PTR_TO_BTF_ID passed to release function, in
case multiple candidates can release same BTF ID.
e.g. there might be two release kfuncs (or kfunc and helper):
foo(struct abc *p);
bar(struct abc *p);
... such that both release a PTR_TO_BTF_ID with btf_id of struct abc. In
this case we would need to track the acquire function corresponding to
the release function to avoid type confusion, and store this information
in the register state so that an incorrect program can be rejected. This
is not a problem right now, hence it is left as an exercise for the
future patch introducing such a case in the kernel.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220114163953.1455836-6-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2022-01-14 16:39:48 +00:00
|
|
|
}
|
|
|
|
|
|
2022-10-25 18:45:16 +00:00
|
|
|
static inline bool bpf_prog_check_recur(const struct bpf_prog *prog)
|
|
|
|
|
{
|
|
|
|
|
switch (resolve_prog_type(prog)) {
|
|
|
|
|
case BPF_PROG_TYPE_TRACING:
|
|
|
|
|
return prog->expected_attach_type != BPF_TRACE_ITER;
|
|
|
|
|
case BPF_PROG_TYPE_STRUCT_OPS:
|
2024-11-12 16:39:33 +00:00
|
|
|
return prog->aux->jits_use_priv_stack;
|
2022-10-25 18:45:16 +00:00
|
|
|
case BPF_PROG_TYPE_LSM:
|
|
|
|
|
return false;
|
|
|
|
|
default:
|
|
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
bpf: Consider non-owning refs trusted
Recent discussions around default kptr "trustedness" led to changes such
as commit 6fcd486b3a0a ("bpf: Refactor RCU enforcement in the
verifier."). One of the conclusions of those discussions, as expressed
in code and comments in that patch, is that we'd like to move away from
'raw' PTR_TO_BTF_ID without some type flag or other register state
indicating trustedness. Although PTR_TRUSTED and PTR_UNTRUSTED flags mark
this state explicitly, the verifier currently considers trustedness
implied by other register state. For example, owning refs to graph
collection nodes must have a nonzero ref_obj_id, so they pass the
is_trusted_reg check despite having no explicit PTR_{UN}TRUSTED flag.
This patch makes trustedness of non-owning refs to graph collection
nodes explicit as well.
By definition, non-owning refs are currently trusted. Although the ref
has no control over pointee lifetime, due to non-owning ref clobbering
rules (see invalidate_non_owning_refs) dereferencing a non-owning ref is
safe in the critical section controlled by bpf_spin_lock associated with
its owning collection.
Note that the previous statement does not hold true for nodes with shared
ownership due to the use-after-free issue that this series is
addressing. True shared ownership was disabled by commit 7deca5eae833
("bpf: Disable bpf_refcount_acquire kfunc calls until race conditions are fixed"),
though, so the statement holds for now. Further patches in the series will change
the trustedness state of non-owning refs before re-enabling
bpf_refcount_acquire.
Let's add NON_OWN_REF type flag to BPF_REG_TRUSTED_MODIFIERS such that a
non-owning ref reg state would pass is_trusted_reg check. Somewhat
surprisingly, this doesn't result in any change to user-visible
functionality elsewhere in the verifier: graph collection nodes are all
marked MEM_ALLOC, which tends to be handled in separate codepaths from
"raw" PTR_TO_BTF_ID. Regardless, let's be explicit here and document the
current state of things before changing it elsewhere in the series.
Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230821193311.3290257-3-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-08-21 19:33:06 +00:00
|
|
|
#define BPF_REG_TRUSTED_MODIFIERS (MEM_ALLOC | PTR_TRUSTED | NON_OWN_REF)
|
bpf: Allow trusted pointers to be passed to KF_TRUSTED_ARGS kfuncs
Kfuncs currently support specifying the KF_TRUSTED_ARGS flag to signal
to the verifier that it should enforce that a BPF program passes it a
"safe", trusted pointer. Currently, "safe" means that the pointer is
either PTR_TO_CTX, or is refcounted. There may be cases, however, where
the kernel passes a BPF program a safe / trusted pointer to an object
that the BPF program wishes to use as a kptr, but because the object
does not yet have a ref_obj_id from the perspective of the verifier, the
program would be unable to pass it to a KF_ACQUIRE | KF_TRUSTED_ARGS
kfunc.
The solution is to expand the set of pointers that are considered
trusted according to KF_TRUSTED_ARGS, so that programs can invoke kfuncs
with these pointers without getting rejected by the verifier.
There is already a PTR_UNTRUSTED flag that is set in some scenarios,
such as when a BPF program reads a kptr directly from a map
without performing a bpf_kptr_xchg() call. These pointers of course can
and should be rejected by the verifier. Unfortunately, however,
PTR_UNTRUSTED does not cover all the cases for safety that need to
be addressed to adequately protect kfuncs. Specifically, pointers
obtained by a BPF program "walking" a struct are _not_ considered
PTR_UNTRUSTED according to BPF. For example, say that we were to add a
kfunc called bpf_task_acquire(), with KF_ACQUIRE | KF_TRUSTED_ARGS, to
acquire a struct task_struct *. If we only used PTR_UNTRUSTED to signal
that a task was unsafe to pass to a kfunc, the verifier would mistakenly
allow the following unsafe BPF program to be loaded:
SEC("tp_btf/task_newtask")
int BPF_PROG(unsafe_acquire_task,
struct task_struct *task,
u64 clone_flags)
{
struct task_struct *acquired, *nested;
nested = task->last_wakee;
/* Would not be rejected by the verifier. */
acquired = bpf_task_acquire(nested);
if (!acquired)
return 0;
bpf_task_release(acquired);
return 0;
}
To address this, this patch defines a new type flag called PTR_TRUSTED
which tracks whether a PTR_TO_BTF_ID pointer is safe to pass to a
KF_TRUSTED_ARGS kfunc or a BPF helper function. PTR_TRUSTED pointers are
passed directly from the kernel as a tracepoint or struct_ops callback
argument. Any nested pointer that is obtained from walking a PTR_TRUSTED
pointer is no longer PTR_TRUSTED. From the example above, the struct
task_struct *task argument is PTR_TRUSTED, but the 'nested' pointer
obtained from 'task->last_wakee' is not PTR_TRUSTED.
A subsequent patch will add kfuncs for storing a task kfunc as a kptr,
and then another patch will add selftests to validate.
Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20221120051004.3605026-3-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2022-11-20 05:10:02 +00:00
|
|
|
|
|
|
|
|
static inline bool bpf_type_has_unsafe_modifiers(u32 type)
|
|
|
|
|
{
|
|
|
|
|
return type_flag(type) & ~BPF_REG_TRUSTED_MODIFIERS;
|
|
|
|
|
}
|
|
|
|
|
|
2023-11-18 03:46:17 +00:00
|
|
|
static inline bool type_is_ptr_alloc_obj(u32 type)
|
|
|
|
|
{
|
|
|
|
|
return base_type(type) == PTR_TO_BTF_ID && type_flag(type) & MEM_ALLOC;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline bool type_is_non_owning_ref(u32 type)
|
|
|
|
|
{
|
|
|
|
|
return type_is_ptr_alloc_obj(type) && type_flag(type) & NON_OWN_REF;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline bool type_is_pkt_pointer(enum bpf_reg_type type)
|
|
|
|
|
{
|
|
|
|
|
type = base_type(type);
|
|
|
|
|
return type == PTR_TO_PACKET ||
|
|
|
|
|
type == PTR_TO_PACKET_META;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline bool type_is_sk_pointer(enum bpf_reg_type type)
|
|
|
|
|
{
|
|
|
|
|
return type == PTR_TO_SOCKET ||
|
|
|
|
|
type == PTR_TO_SOCK_COMMON ||
|
|
|
|
|
type == PTR_TO_TCP_SOCK ||
|
|
|
|
|
type == PTR_TO_XDP_SOCK;
|
|
|
|
|
}
|
|
|
|
|
|
2024-09-05 05:52:32 +00:00
|
|
|
static inline bool type_may_be_null(u32 type)
|
|
|
|
|
{
|
|
|
|
|
return type & PTR_MAYBE_NULL;
|
|
|
|
|
}
|
|
|
|
|
|
2023-11-18 03:46:17 +00:00
|
|
|
static inline void mark_reg_scratched(struct bpf_verifier_env *env, u32 regno)
|
|
|
|
|
{
|
|
|
|
|
env->scratched_regs |= 1U << regno;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline void mark_stack_slot_scratched(struct bpf_verifier_env *env, u32 spi)
|
|
|
|
|
{
|
|
|
|
|
env->scratched_stack_slots |= 1ULL << spi;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline bool reg_scratched(const struct bpf_verifier_env *env, u32 regno)
|
|
|
|
|
{
|
|
|
|
|
return (env->scratched_regs >> regno) & 1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline bool stack_slot_scratched(const struct bpf_verifier_env *env, u64 regno)
|
|
|
|
|
{
|
|
|
|
|
return (env->scratched_stack_slots >> regno) & 1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline bool verifier_state_scratched(const struct bpf_verifier_env *env)
|
|
|
|
|
{
|
|
|
|
|
return env->scratched_regs || env->scratched_stack_slots;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline void mark_verifier_state_clean(struct bpf_verifier_env *env)
|
|
|
|
|
{
|
|
|
|
|
env->scratched_regs = 0U;
|
|
|
|
|
env->scratched_stack_slots = 0ULL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Used for printing the entire verifier state. */
|
|
|
|
|
static inline void mark_verifier_state_scratched(struct bpf_verifier_env *env)
|
|
|
|
|
{
|
|
|
|
|
env->scratched_regs = ~0U;
|
|
|
|
|
env->scratched_stack_slots = ~0ULL;
|
|
|
|
|
}
|
|
|
|
|
|
bpf: Preserve boundaries and track scalars on narrowing fill
When the width of a fill is smaller than the width of the preceding
spill, the information about scalar boundaries can still be preserved,
as long as it's coerced to the right width (done by coerce_reg_to_size).
Even further, if the actual value fits into the fill width, the ID can
be preserved as well for further tracking of equal scalars.
Implement the above improvements, which makes narrowing fills behave the
same as narrowing spills and MOVs between registers.
Two tests are adjusted to accommodate for endianness differences and to
take into account that it's now allowed to do a narrowing fill from the
least significant bits.
reg_bounds_sync is added to coerce_reg_to_size to correctly adjust
umin/umax boundaries after the var_off truncation, for example, a 64-bit
value 0xXXXXXXXX00000000, when read as a 32-bit, gets umin = 0, umax =
0xFFFFFFFF, var_off = (0x0; 0xffffffff00000000), which needs to be
synced down to umax = 0, otherwise reg_bounds_sanity_check doesn't pass.
Signed-off-by: Maxim Mikityanskiy <maxim@isovalent.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20240127175237.526726-4-maxtram95@gmail.com
2024-01-27 17:52:34 +00:00
|
|
|
static inline bool bpf_stack_narrow_access_ok(int off, int fill_size, int spill_size)
|
|
|
|
|
{
|
|
|
|
|
#ifdef __BIG_ENDIAN
|
|
|
|
|
off -= spill_size - fill_size;
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
return !(off % BPF_REG_SIZE);
|
|
|
|
|
}
|
|
|
|
|
|
2023-11-18 03:46:17 +00:00
|
|
|
const char *reg_type_str(struct bpf_verifier_env *env, enum bpf_reg_type type);
|
|
|
|
|
const char *dynptr_type_str(enum bpf_dynptr_type type);
|
|
|
|
|
const char *iter_type_str(const struct btf *btf, u32 btf_id);
|
|
|
|
|
const char *iter_state_str(enum bpf_iter_state state);
|
|
|
|
|
|
2024-12-04 03:03:54 +00:00
|
|
|
void print_verifier_state(struct bpf_verifier_env *env, const struct bpf_verifier_state *vstate,
|
|
|
|
|
u32 frameno, bool print_all);
|
|
|
|
|
void print_insn_state(struct bpf_verifier_env *env, const struct bpf_verifier_state *vstate,
|
|
|
|
|
u32 frameno);
|
2023-11-18 03:46:17 +00:00
|
|
|
|
2016-09-21 10:43:57 +00:00
|
|
|
#endif /* _LINUX_BPF_VERIFIER_H */
|
