When querying info for an offloaded BPF map or program,
bpf_map_offload_info_fill_ns() and bpf_prog_offload_info_fill_ns()
obtain the network namespace with get_net(dev_net(offmap->netdev)).
However, the associated netdev's netns may be racing with teardown
during netns destruction. If the netns refcount has already reached 0,
get_net() performs a refcount_t increment on 0, triggering:
refcount_t: addition on 0; use-after-free.
Although rtnl_lock and bpf_devs_lock ensure the netdev pointer remains
valid, they cannot prevent the netns refcount from reaching zero.
Fix this by using maybe_get_net() instead of get_net(). maybe_get_net()
uses refcount_inc_not_zero() and returns NULL if the refcount is already
zero, which causes ns_get_path_cb() to fail and the caller to return
-ENOENT -- the correct behavior when the netns is being destroyed.
Fixes: 675fc275a3 ("bpf: offload: report device information for offloaded programs")
Fixes: 52775b33bb ("bpf: offload: report device information about offloaded maps")
Reported-by: Yinhao Hu <dddddd@hust.edu.cn>
Reported-by: Kaiyan Mei <M202472210@hust.edu.cn>
Reviewed-by: Dongliang Mu <dzm91@hust.edu.cn>
Closes: https://lore.kernel.org/bpf/f0aa3678-79c9-47ae-9e8c-02a3d1df160a@hust.edu.cn/
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20260409023733.168050-1-jiayuan.chen@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When a pkt pointer acquires AT_PKT_END or BEYOND_PKT_END range from
a comparison, and then, known-constant arithmetic is performed,
adjust_ptr_min_max_vals() copies the stale range via dst_reg->raw =
ptr_reg->raw without clearing the negative reg->range sentinel values.
This lets is_pkt_ptr_branch_taken() choose one branch direction and
skip going through the other. Fix this by clearing negative pkt range
values (that is, AT_PKT_END and BEYOND_PKT_END) after arithmetic on
pkt pointers. This ensures is_pkt_ptr_branch_taken() returns unknown
and both branches are properly verified.
Fixes: 6d94e741a8 ("bpf: Support for pointers beyond pkt_end.")
Reported-by: STAR Labs SG <info@starlabs.sg>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20260409155016.536608-1-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The local subprog pointer in create_jt() and visit_abnormal_return_insn()
was declared static.
It is unconditionally assigned via bpf_find_containing_subprog() before
every use. Thus, the static qualifier serves no purpose and rather creates
confusion. Just remove it.
Fixes: e40f5a6bf8 ("bpf: correct stack liveness for tail calls")
Fixes: 493d9e0d60 ("bpf, x86: add support for indirect jumps")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Anton Protopopov <a.s.protopopov@gmail.com>
Link: https://lore.kernel.org/r/20260408191242.526279-3-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Usage of ld_{abs,ind} instructions got extended into subprogs some time
ago via commit 09b28d76ea ("bpf: Add abnormal return checks."). These
are only allowed in subprograms when the latter are BTF annotated and
have scalar return types.
The code generator in bpf_gen_ld_abs() has an abnormal exit path (r0=0 +
exit) from legacy cBPF times. While the enforcement is on scalar return
types, the verifier must also simulate the path of abnormal exit if the
packet data load via ld_{abs,ind} failed.
This is currently not the case. Fix it by having the verifier simulate
both success and failure paths, and extend it in similar ways as we do
for tail calls. The success path (r0=unknown, continue to next insn) is
pushed onto stack for later validation and the r0=0 and return to the
caller is done on the fall-through side.
Fixes: 09b28d76ea ("bpf: Add abnormal return checks.")
Reported-by: STAR Labs SG <info@starlabs.sg>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20260408191242.526279-2-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Move find_linfo() as bpf_find_linfo() into core.c to allow for its use
in the verifier in subsequent patches.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Link: https://lore.kernel.org/r/20260408021359.3786905-4-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Extract bpf_get_linfo_file_line as its own function so that the logic to
obtain the file, line, and line number for a given program can be shared
in subsequent patches.
Reviewed-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260408021359.3786905-3-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The verifier currently does not allow overwriting a referenced dynptr's
stack slot to prevent resource leak. This is because referenced dynptr
holds additional resources that requires calling specific helpers to
release. This limitation can be relaxed when there are multiple copies
of the same dynptr. Whether it is the orignial dynptr or one of its
clones, as long as there exists at least one other dynptr with the same
ref_obj_id (to be used to release the reference), its stack slot should
be allowed to be overwritten.
Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Amery Hung <ameryhung@gmail.com>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260406150548.1354271-2-ameryhung@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When a non-{add,sub} alu op such as xor is performed on a scalar
register that previously had a BPF_ADD_CONST delta, the else path
in adjust_reg_min_max_vals() only clears dst_reg->id but leaves
dst_reg->delta unchanged.
This stale delta can propagate via assign_scalar_id_before_mov()
when the register is later used in a mov. It gets a fresh id but
keeps the stale delta from the old (now-cleared) BPF_ADD_CONST.
This stale delta can later propagate leading to a verifier-vs-
runtime value mismatch.
The clear_id label already correctly clears both delta and id.
Make the else path consistent by also zeroing the delta when id
is cleared. More generally, this introduces a helper clear_scalar_id()
which internally takes care of zeroing. There are various other
locations in the verifier where only the id is cleared. By using
the helper we catch all current and future locations.
Fixes: 98d7ca374b ("bpf: Track delta between "linked" registers.")
Reported-by: STAR Labs SG <info@starlabs.sg>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20260407192421.508817-2-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Consider the case of rX += rX where src_reg and dst_reg are pointers to
the same bpf_reg_state in adjust_reg_min_max_vals(). The latter first
modifies the dst_reg in-place, and later in the delta tracking, the
subsequent is_reg_const(src_reg)/reg_const_value(src_reg) reads the
post-{add,sub} value instead of the original source.
This is problematic since it sets an incorrect delta, which sync_linked_regs()
then propagates to linked registers, thus creating a verifier-vs-runtime
mismatch. Fix it by just skipping this corner case.
Fixes: 98d7ca374b ("bpf: Track delta between "linked" registers.")
Reported-by: STAR Labs SG <info@starlabs.sg>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20260407192421.508817-1-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
RCU Tasks Trace grace period implies RCU grace period, and this
guarantee is expected to remain in the future. Only BPF is the user of
this predicate, hence retire the API and clean up all in-tree users.
RCU Tasks Trace is now implemented on SRCU-fast and its grace period
mechanism always has at least one call to synchronize_rcu() as it is
required for SRCU-fast's correctness (it replaces the smp_mb() that
SRCU-fast readers skip). So, RCU-tt GP will always imply RCU GP.
Reviewed-by: Puranjay Mohan <puranjay@kernel.org>
Reviewed-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260407162234.785270-1-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
bpf_lsm_task_to_inode() is called under rcu_read_lock() and
bpf_lsm_inet_conn_established() is called from softirq context, so
neither hook can be used by sleepable LSM programs.
Fixes: 423f16108c ("bpf: Augment the set of sleepable LSM hooks")
Reported-by: Quan Sun <2022090917019@std.uestc.edu.cn>
Reported-by: Yinhao Hu <dddddd@hust.edu.cn>
Reported-by: Kaiyan Mei <M202472210@hust.edu.cn>
Reported-by: Dongliang Mu <dzm91@hust.edu.cn>
Closes: https://lore.kernel.org/bpf/3ab69731-24d1-431a-a351-452aafaaf2a5@std.uestc.edu.cn/T/#u
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Link: https://lore.kernel.org/r/20260407122334.344072-1-jiayuan.chen@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When a pointer to PTR_TO_INSN is dereferenced, the offset field
of the BPF_LDX_MEM instruction can be nonzero. Patch the verifier
to not ignore this field.
Reported-by: Jiyong Yang <ksur673@gmail.com>
Fixes: 493d9e0d60 ("bpf, x86: add support for indirect jumps")
Signed-off-by: Anton Protopopov <a.s.protopopov@gmail.com>
Link: https://lore.kernel.org/r/20260406160141.36943-2-a.s.protopopov@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Apparently, struct bpf_empty_prog_array exists entirely to populate a
single element of "items" in a global variable. "null_prog" is only
used during the initializer.
None of this is needed; globals will be correctly sized with an array
initializer of a flexible-array member.
So, remove struct bpf_empty_prog_array and adjust the rest of the code,
accordingly.
With these changes, fix the following warnings:
./include/linux/bpf.h:2369:31: warning: structure containing a flexible
array member is not at the end of another structure [-Wflex-array-member-not-at-end]
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Link: https://lore.kernel.org/r/acr7Whmn0br3xeBP@kspp
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Don't reject usage of fixed unaligned offsets for syscall ctx. Tests
will be added in later commits. Unaligned offsets already work for
variable offsets.
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260406194403.1649608-3-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Allow accessing PTR_TO_CTX with variable offsets in syscall programs.
Fixed offsets are already enabled for all program types that do not
convert their ctx accesses, since the changes we made in the commit
de6c7d99f8 ("bpf: Relax fixed offset check for PTR_TO_CTX"). Note
that we also lift the restriction on passing syscall context into
helpers, which was not permitted before, and passing modified syscall
context into kfuncs.
The structure of check_mem_access can be mostly shared and preserved,
but we must use check_mem_region_access to correctly verify access with
variable offsets.
The check made in check_helper_mem_access is hardened to only allow
PTR_TO_CTX for syscall programs to be passed in as helper memory. This
was the original intention of the existing code anyway, and it makes
little sense for other program types' context to be utilized as a memory
buffer. In case a convincing example presents itself in the future, this
check can be relaxed further.
We also no longer use the last-byte access to simulate helper memory
access, but instead go through check_mem_region_access. Since this no
longer updates our max_ctx_offset, we must do so manually, to keep track
of the maximum offset at which the program ctx may be accessed.
Take care to ensure that when arg_type is ARG_PTR_TO_CTX, we do not
relax any fixed or variable offset constraints around PTR_TO_CTX even in
syscall programs, and require them to be passed unmodified. There are
several reasons why this is necessary. First, if we pass a modified ctx,
then the global subprog's accesses will not update the max_ctx_offset to
its true maximum offset, and can lead to out of bounds accesses. Second,
tail called program (or extension program replacing global subprog) where
their max_ctx_offset exceeds the program they are being called from can
also cause issues. For the latter, unmodified PTR_TO_CTX is the first
requirement for the fix, the second is ensuring max_ctx_offset >= the
program they are being called from, which has to be a separate change
not made in this commit.
All in all, we can hint using arg_type when we expect ARG_PTR_TO_CTX and
make our relaxation around offsets conditional on it.
Drop coverage of syscall tests from verifier_ctx.c temporarily for
negative cases until they are updated in subsequent commits.
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260406194403.1649608-2-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When a dev-bound-only BPF program (BPF_F_XDP_DEV_BOUND_ONLY) undergoes
JIT compilation with constant blinding enabled (bpf_jit_harden >= 2),
bpf_jit_blind_constants() clones the program. The original prog is then
freed in bpf_jit_prog_release_other(), which updates aux->prog to point
to the surviving clone, but fails to update offload->prog.
This leaves offload->prog pointing to the freed original program. When
the network namespace is subsequently destroyed, cleanup_net() triggers
bpf_dev_bound_netdev_unregister(), which iterates ondev->progs and calls
__bpf_prog_offload_destroy(offload->prog). Accessing the freed prog
causes a page fault:
BUG: unable to handle page fault for address: ffffc900085f1038
Workqueue: netns cleanup_net
RIP: 0010:__bpf_prog_offload_destroy+0xc/0x80
Call Trace:
__bpf_offload_dev_netdev_unregister+0x257/0x350
bpf_dev_bound_netdev_unregister+0x4a/0x90
unregister_netdevice_many_notify+0x2a2/0x660
...
cleanup_net+0x21a/0x320
The test sequence that triggers this reliably is:
1. Set net.core.bpf_jit_harden=2 (echo 2 > /proc/sys/net/core/bpf_jit_harden)
2. Run xdp_metadata selftest, which creates a dev-bound-only XDP
program on a veth inside a netns (./test_progs -t xdp_metadata)
3. cleanup_net -> page fault in __bpf_prog_offload_destroy
Dev-bound-only programs are unique in that they have an offload structure
but go through the normal JIT path instead of bpf_prog_offload_compile().
This means they are subject to constant blinding's prog clone-and-replace,
while also having offload->prog that must stay in sync.
Fix this by updating offload->prog in bpf_jit_prog_release_other(),
alongside the existing aux->prog update. Both are back-pointers to
the prog that must be kept in sync when the prog is replaced.
Fixes: 2b3486bc2d ("bpf: Introduce device-bound XDP programs")
Signed-off-by: MingTao Huang <mintaohuang@tencent.com>
Link: https://lore.kernel.org/r/tencent_BCF692F45859CCE6C22B7B0B64827947D406@qq.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
list_next_entry() never returns NULL -- when the current element is the
last entry it wraps to the list head via container_of(). The subsequent
NULL check is therefore dead code and get_next_key() never returns
-ENOENT for the last element, instead reading storage->key from a bogus
pointer that aliases internal map fields and copying the result to
userspace.
Replace it with list_entry_is_head() so the function correctly returns
-ENOENT when there are no more entries.
Fixes: de9cbbaadb ("bpf: introduce cgroup storage maps")
Reported-by: Xiang Mei <xmei5@asu.edu>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Reviewed-by: Sun Jian <sun.jian.kdev@gmail.com>
Acked-by: Paul Chaignon <paul.chaignon@gmail.com>
Link: https://lore.kernel.org/r/20260403132951.43533-2-bestswngs@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When a BPF_F_LOCK update races with a concurrent delete, the freed
element can be immediately recycled by alloc_htab_elem(). The fast path
in htab_map_update_elem() performs a lockless lookup and then calls
copy_map_value_locked() under the element's spin_lock. If
alloc_htab_elem() recycles the same memory, it overwrites the value
with plain copy_map_value(), without taking the spin_lock, causing
torn writes.
Use copy_map_value_locked() when BPF_F_LOCK is set so the new element's
value is written under the embedded spin_lock, serializing against any
stale lock holders.
Fixes: 96049f3afd ("bpf: introduce BPF_F_LOCK flag")
Reported-by: Aaron Esau <aaron1esau@gmail.com>
Closes: https://lore.kernel.org/all/CADucPGRvSRpkneb94dPP08YkOHgNgBnskTK6myUag_Mkjimihg@mail.gmail.com/
Signed-off-by: Mykyta Yatsenko <yatsenko@meta.com>
Link: https://lore.kernel.org/r/20260401-bpf_map_torn_writes-v1-1-782d071c55e7@meta.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The static stack liveness analysis needs to know how many bytes a
helper or kfunc accesses through a stack pointer argument, so it can
precisely mark the affected stack slots as stack 'def' or 'use'.
Add bpf_helper_stack_access_bytes() and bpf_kfunc_stack_access_bytes()
which resolve the access size for a given call argument.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260403024422.87231-7-alexei.starovoitov@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add two passes before the main verifier pass:
bpf_compute_const_regs() is a forward dataflow analysis that tracks
register values in R0-R9 across the program using fixed-point
iteration in reverse postorder. Each register is tracked with
a six-state lattice:
UNVISITED -> CONST(val) / MAP_PTR(map_index) /
MAP_VALUE(map_index, offset) / SUBPROG(num) -> UNKNOWN
At merge points, if two paths produce the same state and value for
a register, it stays; otherwise it becomes UNKNOWN.
The analysis handles:
- MOV, ADD, SUB, AND with immediate or register operands
- LD_IMM64 for plain constants, map FDs, map values, and subprogs
- LDX from read-only maps: constant-folds the load by reading the
map value directly via bpf_map_direct_read()
Results that fit in 32 bits are stored per-instruction in
insn_aux_data and bitmasks.
bpf_prune_dead_branches() uses the computed constants to evaluate
conditional branches. When both operands of a conditional jump are
known constants, the branch outcome is determined statically and the
instruction is rewritten to an unconditional jump.
The CFG postorder is then recomputed to reflect new control flow.
This eliminates dead edges so that subsequent liveness analysis
doesn't propagate through dead code.
Also add runtime sanity check to validate that precomputed
constants match the verifier's tracked state.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260403024422.87231-5-alexei.starovoitov@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add a pass that sorts subprogs in topological order so that iterating
subprog_topo_order[] walks leaf subprogs first, then their callers.
This is computed as a DFS post-order traversal of the CFG.
The pass runs after check_cfg() to ensure the CFG has been validated
before traversing and after postorder has been computed to avoid
walking dead code.
Reviewed-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260403024422.87231-3-alexei.starovoitov@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch fixes the invariant violations that can happen after we
refine ranges & tnum based on an incorrectly-detected branch condition.
For example, the branch is always true, but we miss it in
is_branch_taken; we then refine based on the branch being false and end
up with incoherent ranges (e.g. umax < umin).
To avoid this, we can simulate the refinement on both branches. More
specifically, this patch simulates both branches taken using
regs_refine_cond_op and reg_bounds_sync. If the resulting register
states are ill-formed on one of the branches, is_branch_taken can mark
that branch as "never taken".
On a more formal note, we can deduce a branch is not taken when
regs_refine_cond_op or reg_bounds_sync returns an ill-formed state
because the branch operators are sound (verified with Agni [1]).
Soundness means that the verifier is guaranteed to produce sound
outputs on the taken branches. On the non-taken branch (explored
because of imprecision in the bounds), the verifier is free to produce
any output. We use ill-formedness as a signal that the branch is dead
and prune that branch.
This patch moves the refinement logic for both branches from
reg_set_min_max to their own function, simulate_both_branches_taken,
which is called from is_scalar_branch_taken. As a result,
reg_set_min_max now only runs sanity checks and has been renamed to
reg_bounds_sanity_check_branches to reflect that.
We have had five patches fixing specific cases of invariant violations
in the past, all added with selftests:
- commit fbc7aef517 ("bpf: Fix u32/s32 bounds when ranges cross
min/max boundary")
- commit efc11a6678 ("bpf: Improve bounds when tnum has a single
possible value")
- commit f41345f47f ("bpf: Use tnums for JEQ/JNE is_branch_taken
logic")
- commit 00bf8d0c6c ("bpf: Improve bounds when s64 crosses sign
boundary")
- commit 6279846b9b ("bpf: Forget ranges when refining tnum after
JSET")
To confirm that this patch addresses all invariant violations, we have
also reverted those five commits and verified that their related
selftests don't cause any invariant violation warnings anymore. Those
selftests still fail but only because of misdetected branches or
less-precise bounds than expected. This demonstrates that the current
patch is enough to avoid the invariant violation warning AND that the
previous five patches are still useful to improve branch detection.
In addition to the selftests, this change was also tested with the
Cilium complexity test suite: all programs were successfully loaded and
it didn't change the number of processed instructions.
Link: https://github.com/bpfverif/agni [1]
Reported-by: syzbot+c950cc277150935cc0b5@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=c950cc277150935cc0b5
Co-developed-by: Paul Chaignon <paul.chaignon@gmail.com>
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
Co-developed-by: Srinivas Narayana <srinivas.narayana@rutgers.edu>
Signed-off-by: Srinivas Narayana <srinivas.narayana@rutgers.edu>
Co-developed-by: Santosh Nagarakatte <santosh.nagarakatte@rutgers.edu>
Signed-off-by: Santosh Nagarakatte <santosh.nagarakatte@rutgers.edu>
Signed-off-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/a166b54a3cbbbdbcdf8a87f53045f1097176218b.1775142354.git.paul.chaignon@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In the subsequent commit, to prune dead branches we will rely on
detecting ill-formed ranges using range_bounds_violations()
(e.g., umin > umax) after refining register bounds using
regs_refine_cond_op().
However, reg_bounds_sync() can sometimes "repair" ill-formed bounds,
potentially masking a violation that was produced by
regs_refine_cond_op().
This commit modifies reg_bounds_sync() to exit early if an invariant
violation is already present in the input.
This ensures ill-formed reg_states remain ill-formed after
reg_bounds_sync(), allowing simulate_both_branches_taken() to correctly
identify dead branches with a single check to range_bounds_violation().
Suggested-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/73127d628841c59cb7423d6bdcd204bf90bcdc80.1775142354.git.paul.chaignon@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In a subsequent patch, the regs_refine_cond_op and reg_bounds_sync
functions will be called in is_branch_taken instead of reg_set_min_max,
to simulate each branch's outcome. Since they will run before we branch
out, these two functions will need to work on temporary registers for
the two branches.
This refactoring patch prepares for that change, by introducing the
temporary registers on bpf_verifier_env and using them in
reg_set_min_max.
This change also allows us to save one fake_reg slot as we don't need to
allocate an additional temporary buffer in case of a BPF_K condition.
Finally, you may notice that this patch removes the check for
"false_reg1 == false_reg2" in reg_set_min_max. That check was introduced
in commit d43ad9da80 ("bpf: Skip bounds adjustment for conditional
jumps on same scalar register") to avoid an invariant violation. Given
that "env->false_reg1 == env->false_reg2" doesn't make sense and
invariant violations are addressed in a subsequent commit, this patch
just removes the check.
Suggested-by: Eduard Zingerman <eddyz87@gmail.com>
Co-developed-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Signed-off-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/260b0270052944a420e1c56e6a92df4d43cadf03.1775142354.git.paul.chaignon@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This commit refactors reg_bounds_sanity_check to factor out the logic
that performs the sanity check from the logic that does the reporting.
Signed-off-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/198ec3e69343e2c46dc9cbe2b1bc9be9ae2df5bd.1775142354.git.paul.chaignon@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When backtrack_insn encounters a BPF_STX instruction with BPF_ATOMIC
and BPF_FETCH, the src register (or r0 for BPF_CMPXCHG) also acts as
a destination, thus receiving the old value from the memory location.
The current backtracking logic does not account for this. It treats
atomic fetch operations the same as regular stores where the src
register is only an input. This leads the backtrack_insn to fail to
propagate precision to the stack location, which is then not marked
as precise!
Later, the verifier's path pruning can incorrectly consider two states
equivalent when they differ in terms of stack state. Meaning, two
branches can be treated as equivalent and thus get pruned when they
should not be seen as such.
Fix it as follows: Extend the BPF_LDX handling in backtrack_insn to
also cover atomic fetch operations via is_atomic_fetch_insn() helper.
When the fetch dst register is being tracked for precision, clear it,
and propagate precision over to the stack slot. For non-stack memory,
the precision walk stops at the atomic instruction, same as regular
BPF_LDX. This covers all fetch variants.
Before:
0: (b7) r1 = 8 ; R1=8
1: (7b) *(u64 *)(r10 -8) = r1 ; R1=8 R10=fp0 fp-8=8
2: (b7) r2 = 0 ; R2=0
3: (db) r2 = atomic64_fetch_add((u64 *)(r10 -8), r2) ; R2=8 R10=fp0 fp-8=mmmmmmmm
4: (bf) r3 = r10 ; R3=fp0 R10=fp0
5: (0f) r3 += r2
mark_precise: frame0: last_idx 5 first_idx 0 subseq_idx -1
mark_precise: frame0: regs=r2 stack= before 4: (bf) r3 = r10
mark_precise: frame0: regs=r2 stack= before 3: (db) r2 = atomic64_fetch_add((u64 *)(r10 -8), r2)
mark_precise: frame0: regs=r2 stack= before 2: (b7) r2 = 0
6: R2=8 R3=fp8
6: (b7) r0 = 0 ; R0=0
7: (95) exit
After:
0: (b7) r1 = 8 ; R1=8
1: (7b) *(u64 *)(r10 -8) = r1 ; R1=8 R10=fp0 fp-8=8
2: (b7) r2 = 0 ; R2=0
3: (db) r2 = atomic64_fetch_add((u64 *)(r10 -8), r2) ; R2=8 R10=fp0 fp-8=mmmmmmmm
4: (bf) r3 = r10 ; R3=fp0 R10=fp0
5: (0f) r3 += r2
mark_precise: frame0: last_idx 5 first_idx 0 subseq_idx -1
mark_precise: frame0: regs=r2 stack= before 4: (bf) r3 = r10
mark_precise: frame0: regs=r2 stack= before 3: (db) r2 = atomic64_fetch_add((u64 *)(r10 -8), r2)
mark_precise: frame0: regs= stack=-8 before 2: (b7) r2 = 0
mark_precise: frame0: regs= stack=-8 before 1: (7b) *(u64 *)(r10 -8) = r1
mark_precise: frame0: regs=r1 stack= before 0: (b7) r1 = 8
6: R2=8 R3=fp8
6: (b7) r0 = 0 ; R0=0
7: (95) exit
Fixes: 5ffa25502b ("bpf: Add instructions for atomic_[cmp]xchg")
Fixes: 5ca419f286 ("bpf: Add BPF_FETCH field / create atomic_fetch_add instruction")
Reported-by: STAR Labs SG <info@starlabs.sg>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20260331222020.401848-1-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
check_mem_access() matches PTR_TO_BUF via base_type() which strips
PTR_MAYBE_NULL, allowing direct dereference without a null check.
Map iterator ctx->key and ctx->value are PTR_TO_BUF | PTR_MAYBE_NULL.
On stop callbacks these are NULL, causing a kernel NULL dereference.
Add a type_may_be_null() guard to the PTR_TO_BUF branch, matching the
existing PTR_TO_BTF_ID pattern.
Fixes: 20b2aff4bc ("bpf: Introduce MEM_RDONLY flag")
Signed-off-by: Qi Tang <tpluszz77@gmail.com>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260402092923.38357-2-tpluszz77@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Replace bpf_mem_alloc/bpf_mem_free with kmalloc_nolock/kfree_nolock for
bpf_dynptr_file_impl, continuing the migration away from bpf_mem_alloc
now that kmalloc can be used from NMI context.
freader_cleanup() runs before kfree_nolock() while the dynptr still
holds exclusive access, so plain kfree_nolock() is safe — no concurrent
readers can access the object.
Signed-off-by: Mykyta Yatsenko <yatsenko@meta.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260330-kmalloc_special-v2-2-c90403f92ff0@meta.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Replace bpf_mem_alloc/bpf_mem_free with
kmalloc_nolock/kfree_rcu for bpf_task_work_ctx.
Replace guard(rcu_tasks_trace)() with guard(rcu)() in
bpf_task_work_irq(). The function only accesses ctx struct members
(not map values), so tasks trace protection is not needed - regular
RCU is sufficient since ctx is freed via kfree_rcu. The guard in
bpf_task_work_callback() remains as tasks trace since it accesses map
values from process context.
Sleepable BPF programs hold rcu_read_lock_trace but not
regular rcu_read_lock. Since kfree_rcu
waits for a regular RCU grace period, the ctx memory can be freed
while a sleepable program is still running. Add scoped_guard(rcu)
around the pointer read and refcount tryget in
bpf_task_work_acquire_ctx to close this race window.
Since kfree_rcu uses call_rcu internally which is not safe from
NMI context, defer destruction via irq_work when irqs are disabled.
For the lost-cmpxchg path the ctx was never published, so
kfree_nolock is safe.
Signed-off-by: Mykyta Yatsenko <yatsenko@meta.com>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260330-kmalloc_special-v2-1-c90403f92ff0@meta.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
uprobe programs are allowed to modify struct pt_regs.
Since the actual program type of uprobe is KPROBE, it can be abused to
modify struct pt_regs via kprobe+freplace when the kprobe attaches to
kernel functions.
For example,
SEC("?kprobe")
int kprobe(struct pt_regs *regs)
{
return 0;
}
SEC("?freplace")
int freplace_kprobe(struct pt_regs *regs)
{
regs->di = 0;
return 0;
}
freplace_kprobe prog will attach to kprobe prog.
kprobe prog will attach to a kernel function.
Without this patch, when the kernel function runs, its first arg will
always be set as 0 via the freplace_kprobe prog.
To fix the abuse of kprobe_write_ctx=true via kprobe+freplace, disallow
attaching freplace programs on kprobe programs with different
kprobe_write_ctx values.
Fixes: 7384893d97 ("bpf: Allow uprobe program to change context registers")
Acked-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
Link: https://lore.kernel.org/r/20260331145353.87606-2-leon.hwang@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Recently, tracepoints were switched from using disabled preemption
(which acts as RCU read section) to SRCU-fast when they are not
faultable. This means that to do a proper grace period wait for programs
running in such tracepoints, we must use SRCU's grace period wait.
This is only for non-faultable tracepoints, faultable ones continue
using RCU Tasks Trace.
However, bpf_link_free() currently does call_rcu() for all cases when
the link is non-sleepable (hence, for tracepoints, non-faultable). Fix
this by doing a call_srcu() grace period wait.
As far RCU Tasks Trace gp -> RCU gp chaining is concerned, it is deemed
unnecessary for tracepoint programs. The link and program are either
accessed under RCU Tasks Trace protection, or SRCU-fast protection now.
The earlier logic of chaining both RCU Tasks Trace and RCU gp waits was
to generalize the logic, even if it conceded an extra RCU gp wait,
however that is unnecessary for tracepoints even before this change.
In practice no cost was paid since rcu_trace_implies_rcu_gp() was always
true. Hence we need not chaining any RCU gp after the SRCU gp.
For instance, in the non-faultable raw tracepoint, the RCU read section
of the program in __bpf_trace_run() is enclosed in the SRCU gp, likewise
for faultable raw tracepoint, the program is under the RCU Tasks Trace
protection. Hence, the outermost scope can be waited upon to ensure
correctness.
Also, sleepable programs cannot be attached to non-faultable
tracepoints, so whenever program or link is sleepable, only RCU Tasks
Trace protection is being used for the link and prog.
Fixes: a46023d561 ("tracing: Guard __DECLARE_TRACE() use of __DO_TRACE_CALL() with SRCU-fast")
Reviewed-by: Sun Jian <sun.jian.kdev@gmail.com>
Reviewed-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Link: https://lore.kernel.org/r/20260331211021.1632902-2-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In case rold->reg->range == BEYOND_PKT_END && rcur->reg->range == N
regsafe() may return true which may lead to current state with
valid packet range not being explored. Fix the bug.
Fixes: 6d94e741a8 ("bpf: Support for pointers beyond pkt_end.")
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Amery Hung <ameryhung@gmail.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/bpf/20260331204228.26726-1-alexei.starovoitov@gmail.com
The following kfuncs currently accept void *meta__ign argument:
* bpf_obj_new_impl
* bpf_obj_drop_impl
* bpf_percpu_obj_new_impl
* bpf_percpu_obj_drop_impl
* bpf_refcount_acquire_impl
* bpf_list_push_back_impl
* bpf_list_push_front_impl
* bpf_rbtree_add_impl
The __ign suffix is an indicator for the verifier to skip the argument
in check_kfunc_args(). Then, in fixup_kfunc_call() the verifier may
set the value of this argument to struct btf_struct_meta *
kptr_struct_meta from insn_aux_data.
BPF programs must pass a dummy NULL value when calling these kfuncs.
Additionally, the list and rbtree _impl kfuncs also accept an implicit
u64 argument, which doesn't require __ign suffix because it's a
scalar, and BPF programs explicitly pass 0.
Add new kfuncs with KF_IMPLICIT_ARGS [1], that correspond to each
_impl kfunc accepting meta__ign. The existing _impl kfuncs remain
unchanged for backwards compatibility.
To support this, add "btf_struct_meta" to the list of recognized
implicit argument types in resolve_btfids.
Implement is_kfunc_arg_implicit() in the verifier, that determines
implicit args by inspecting both a non-_impl BTF prototype of the
kfunc.
Update the special_kfunc_list in the verifier and relevant checks to
support both the old _impl and the new KF_IMPLICIT_ARGS variants of
btf_struct_meta users.
[1] https://lore.kernel.org/bpf/20260120222638.3976562-1-ihor.solodrai@linux.dev/
Signed-off-by: Ihor Solodrai <ihor.solodrai@linux.dev>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20260327203241.3365046-1-ihor.solodrai@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
A bunch of new hooks for managing block devices were added a while ago
but they weren't actually appropriately classified.
* bpf_lsm_bdev_alloc() is called when the inode for the block
device is allocated. This happens from a sleepable context so mark the
function as sleepable. When this function is called the memory for the
block device storage embedded into the inode is zeroed. That block
device cannot be meaningfully reference or interacted with at this
point. So mark it as untrusted for now.
* bpf_lsm_bdev_free() is called when the inode for the block
device is freed. A bunch of memory associated with the block device
has already been freed and there's dangling pointers in there. So mark
it as untrusted. It cannot be meaningfully referenced or interacted
with anymore. It is also called from sb->s_op->free_inode:: which
means it runs in rcu context (most of the times). So leave it as
non-sleepable.
* bpf_lsm_bdev_setintegrity() is called when a dm-verity device
is instantiated (glossing over details for simplicity of the commit
message). The block device is very much alive so it remains a trusted
hook. It's also called with device mapper's suspend lock held and so
the hook is able to sleep so mark it sleepable.
Signed-off-by: Christian Brauner <brauner@kernel.org>
Link: https://lore.kernel.org/r/20260326-work-bpf-bdev-v2-1-5e3c58963987@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Validate layout if present, but because the kernel must be
strict in what it accepts, reject BTF with unsupported kinds,
even if they are in the layout information.
Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20260326145444.2076244-8-alan.maguire@oracle.com
The DEVMAP_HASH branch in dev_map_redirect_multi() uses
hlist_for_each_entry_safe() to iterate hash buckets, but this function
runs under RCU protection (called from xdp_do_generic_redirect_map()
in softirq context). Concurrent writers (__dev_map_hash_update_elem,
dev_map_hash_delete_elem) modify the list using RCU primitives
(hlist_add_head_rcu, hlist_del_rcu).
hlist_for_each_entry_safe() performs plain pointer dereferences without
rcu_dereference(), missing the acquire barrier needed to pair with
writers' rcu_assign_pointer(). On weakly-ordered architectures (ARM64,
POWER), a reader can observe a partially-constructed node. It also
defeats CONFIG_PROVE_RCU lockdep validation and KCSAN data-race
detection.
Replace with hlist_for_each_entry_rcu() using rcu_read_lock_bh_held()
as the lockdep condition, consistent with the rcu_dereference_check()
used in the DEVMAP (non-hash) branch of the same functions. Also fix
the same incorrect lockdep_is_held(&dtab->index_lock) condition in
dev_map_enqueue_multi(), where the lock is not held either.
Fixes: e624d4ed4a ("xdp: Extend xdp_redirect_map with broadcast support")
Signed-off-by: David Carlier <devnexen@gmail.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://patch.msgid.link/20260320072645.16731-1-devnexen@gmail.com
The function btf_check_kfunc_arg_match() was refactored into
check_kfunc_args() by commit 00b85860fe ("bpf: Rewrite kfunc
argument handling"). Update the comment accordingly.
Assisted-by: unnamed:deepseek-v3.2 coccinelle
Signed-off-by: Kexin Sun <kexinsun@smail.nju.edu.cn>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Link: https://lore.kernel.org/r/20260321105658.6006-1-kexinsun@smail.nju.edu.cn
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add BPF verifier support for single- and multi-level pointer
parameters and return values in BPF trampolines by treating these
parameters as SCALAR_VALUE.
This extends the existing support for int and void pointers that are
already treated as SCALAR_VALUE.
This provides consistent logic for single and multi-level pointers:
if a type is treated as SCALAR for a single-level pointer, the same
applies to multi-level pointers. The exception is pointer-to-struct,
which is currently PTR_TO_BTF_ID for single-level but treated as
scalar for multi-level pointers since the verifier lacks context
to infer the size of target memory regions.
Safety is ensured by existing BTF verification, which rejects invalid
pointer types at the BTF verification stage.
Signed-off-by: Slava Imameev <slava.imameev@crowdstrike.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260314082127.7939-2-slava.imameev@crowdstrike.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The current implementation only checks whether the first argument is
refcounted. Fix this by iterating over all arguments.
Signed-off-by: Keisuke Nishimura <keisuke.nishimura@inria.fr>
Fixes: 38f1e66abd ("bpf: Do not allow tail call in strcut_ops program with __ref argument")
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Amery Hung <ameryhung@gmail.com>
Link: https://lore.kernel.org/r/20260320130219.63711-1-keisuke.nishimura@inria.fr
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
kvmemdup_bpfptr() returns -EFAULT when the user pointer cannot be
copied, and -ENOMEM on allocation failure. The error path always
returned -ENOMEM, misreporting bad addresses as out-of-memory.
Return PTR_ERR(sig) so user space gets the correct errno.
Signed-off-by: Weixie Cui <cuiweixie@gmail.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/tencent_C9C5B2B28413D6303D505CD02BFEA4708C07@qq.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Simplify tnum_step() from a 10-variable algorithm into a straight
line sequence of bitwise operations.
Problem Reduction:
tnum_step(): Given a tnum `(tval, tmask)` where `tval & tmask == 0`,
and a value `z` with `tval ≤ z < (tval | tmask)`, find the smallest
`r > z`, a tnum-satisfying value, i.e., `r & ~tmask == tval`.
Every tnum-satisfying value has the form tval | s where s is a subset
of tmask bits (s & ~tmask == 0). Since tval and tmask are disjoint:
tval | s = tval + s
Similarly z = tval + d where d = z - tval, so r > z becomes:
tval + s > tval + d
s > d
The problem reduces to: find the smallest s, a subset of tmask, such
that s > d.
Notice that `s` must be a subset of tmask, the problem now is simplified.
Algorithm:
The mask bits of `d` form a "counter" that we want to increment by one,
but the counter has gaps at the fixed-bit positions. A normal +1 would
stop at the first 0-bit it meets; we need it to skip over fixed-bit
gaps and land on the next mask bit.
Step 1 -- plug the gaps:
d | carry_mask | ~tmask
- ~tmask fills all fixed-bit positions with 1.
- carry_mask = (1 << fls64(d & ~tmask)) - 1 fills all positions
(including mask positions) below the highest non-mask bit of d.
After this, the only remaining 0s are mask bits above the highest
non-mask bit of d where d is also 0 -- exactly the positions where
the carry can validly land.
Step 2 -- increment:
(d | carry_mask | ~tmask) + 1
Adding 1 flips all trailing 1s to 0 and sets the first 0 to 1. Since
every gap has been plugged, that first 0 is guaranteed to be a mask bit
above all non-mask bits of d.
Step 3 -- mask:
((d | carry_mask | ~tmask) + 1) & tmask
Strip the scaffolding, keeping only mask bits. Call the result inc.
Step 4 -- result:
tval | inc
Reattach the fixed bits.
A simple 8-bit example:
tmask: 1 1 0 1 0 1 1 0
d: 1 0 1 0 0 0 1 0 (d = 162)
^
non-mask 1 at bit 5
With carry_mask = 0b00111111 (smeared from bit 5):
d|carry|~tm 1 0 1 1 1 1 1 1
+ 1 1 1 0 0 0 0 0 0
& tmask 1 1 0 0 0 0 0 0
The patch passes my local test: test_verifier, test_progs for
`-t verifier` and `-t reg_bounds`.
CBMC shows the new code is equiv to original one[1], and
a lean4 proof of correctness is available[2]:
theorem tnumStep_correct (tval tmask z : BitVec 64)
-- Precondition: valid tnum and input z
(h_consistent : (tval &&& tmask) = 0)
(h_lo : tval ≤ z)
(h_hi : z < (tval ||| tmask)) :
-- Postcondition: r must be:
-- (1) tnum member
-- (2) z < r
-- (3) for any other member w > z, r <= w
let r := tnumStep tval tmask z
satisfiesTnum64 r tval tmask ∧
tval ≤ r ∧ r ≤ (tval ||| tmask) ∧
z < r ∧
∀ w, satisfiesTnum64 w tval tmask → z < w → r ≤ w := by
-- unfold definition
unfold tnumStep satisfiesTnum64
simp only []
refine ⟨?_, ?_, ?_, ?_, ?_⟩
-- the solver proves each conjunct
· bv_decide
· bv_decide
· bv_decide
· bv_decide
· intro w hw1 hw2; bv_decide
[1] https://github.com/eddyz87/tnum-step-verif/blob/master/main.c
[2] https://pastebin.com/raw/czHKiyY0
Signed-off-by: Hao Sun <hao.sun@inf.ethz.ch>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Reviewed-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Link: https://lore.kernel.org/r/20260320162336.166542-1-hao.sun@inf.ethz.ch
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This was renamed in commit 23ef9d4397 ("kcfi: Rename CONFIG_CFI_CLANG
to CONFIG_CFI") as it is now a compiler-agnostic option. Using the wrong
name results in the code getting compiled out. Meaning the CFI failures
for btf_dtor_kfunc_t would still trigger.
Fixes: 99fde4d062 ("bpf, btf: Enforce destructor kfunc type with CFI")
Signed-off-by: Carlos Llamas <cmllamas@google.com>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20260312183818.2721750-1-cmllamas@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Since commit 603b441623 ("bpf: Update the bpf_prog_calc_tag to use
SHA256") made BPF program tags use SHA-256 instead of SHA-1, the header
<crypto/sha1.h> no longer needs to be included. Remove the relevant
inclusions so that they no longer unnecessarily come up in searches for
which kernel code is still using the obsolete SHA-1 algorithm.
Since net/ipv6/addrconf.c was relying on the transitive inclusion of
<crypto/sha1.h> (for an unrelated purpose) via <linux/filter.h>, make it
include <crypto/sha1.h> explicitly in order to keep that file building.
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Acked-by: Paul Chaignon <paul.chaignon@gmail.com>
Link: https://lore.kernel.org/r/20260314214555.112386-1-ebiggers@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Cross-merge BPF and other fixes after downstream PR.
Minor conflicts in:
tools/testing/selftests/bpf/progs/exceptions_fail.c
tools/testing/selftests/bpf/progs/verifier_bounds.c
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Jenny reported that in sync_linked_regs() the BPF_ADD_CONST32 flag is
checked on known_reg (the register narrowed by a conditional branch)
instead of reg (the linked target register created by an alu32 operation).
Example case with reg:
1. r6 = bpf_get_prandom_u32()
2. r7 = r6 (linked, same id)
3. w7 += 5 (alu32 -- r7 gets BPF_ADD_CONST32, zero-extended by CPU)
4. if w6 < 0xFFFFFFFC goto safe (narrows r6 to [0xFFFFFFFC, 0xFFFFFFFF])
5. sync_linked_regs() propagates to r7 but does NOT call zext_32_to_64()
6. Verifier thinks r7 is [0x100000001, 0x100000004] instead of [1, 4]
Since known_reg above does not have BPF_ADD_CONST32 set above, zext_32_to_64()
is never called on alu32-derived linked registers. This causes the verifier
to track incorrect 64-bit bounds, while the CPU correctly zero-extends the
32-bit result.
The code checking known_reg->id was correct however (see scalars_alu32_wrap
selftest case), but the real fix needs to handle both directions - zext
propagation should be done when either register has BPF_ADD_CONST32, since
the linked relationship involves a 32-bit operation regardless of which
side has the flag.
Example case with known_reg (exercised also by scalars_alu32_wrap):
1. r1 = r0; w1 += 0x100 (alu32 -- r1 gets BPF_ADD_CONST32)
2. if r1 > 0x80 - known_reg = r1 (has BPF_ADD_CONST32), reg = r0 (doesn't)
Hence, fix it by checking for (reg->id | known_reg->id) & BPF_ADD_CONST32.
Moreover, sync_linked_regs() also has a soundness issue when two linked
registers used different ALU widths: one with BPF_ADD_CONST32 and the
other with BPF_ADD_CONST64. The delta relationship between linked registers
assumes the same arithmetic width though. When one register went through
alu32 (CPU zero-extends the 32-bit result) and the other went through
alu64 (no zero-extension), the propagation produces incorrect bounds.
Example:
r6 = bpf_get_prandom_u32() // fully unknown
if r6 >= 0x100000000 goto out // constrain r6 to [0, U32_MAX]
r7 = r6
w7 += 1 // alu32: r7.id = N | BPF_ADD_CONST32
r8 = r6
r8 += 2 // alu64: r8.id = N | BPF_ADD_CONST64
if r7 < 0xFFFFFFFF goto out // narrows r7 to [0xFFFFFFFF, 0xFFFFFFFF]
At the branch on r7, sync_linked_regs() runs with known_reg=r7
(BPF_ADD_CONST32) and reg=r8 (BPF_ADD_CONST64). The delta path
computes:
r8 = r7 + (delta_r8 - delta_r7) = 0xFFFFFFFF + (2 - 1) = 0x100000000
Then, because known_reg->id has BPF_ADD_CONST32, zext_32_to_64(r8) is
called, truncating r8 to [0, 0]. But r8 used a 64-bit ALU op -- the
CPU does NOT zero-extend it. The actual CPU value of r8 is
0xFFFFFFFE + 2 = 0x100000000, not 0. The verifier now underestimates
r8's 64-bit bounds, which is a soundness violation.
Fix sync_linked_regs() by skipping propagation when the two registers
have mixed ALU widths (one BPF_ADD_CONST32, the other BPF_ADD_CONST64).
Lastly, fix regsafe() used for path pruning: the existing checks used
"& BPF_ADD_CONST" to test for offset linkage, which treated
BPF_ADD_CONST32 and BPF_ADD_CONST64 as equivalent.
Fixes: 7a433e5193 ("bpf: Support negative offsets, BPF_SUB, and alu32 for linked register tracking")
Reported-by: Jenny Guanni Qu <qguanni@gmail.com>
Co-developed-by: Puranjay Mohan <puranjay@kernel.org>
Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260319211507.213816-1-daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
maybe_fork_scalars() is called for both BPF_AND and BPF_OR when the
source operand is a constant. When dst has signed range [-1, 0], it
forks the verifier state: the pushed path gets dst = 0, the current
path gets dst = -1.
For BPF_AND this is correct: 0 & K == 0.
For BPF_OR this is wrong: 0 | K == K, not 0.
The pushed path therefore tracks dst as 0 when the runtime value is K,
producing an exploitable verifier/runtime divergence that allows
out-of-bounds map access.
Fix this by passing env->insn_idx (instead of env->insn_idx + 1) to
push_stack(), so the pushed path re-executes the ALU instruction with
dst = 0 and naturally computes the correct result for any opcode.
Fixes: bffacdb80b ("bpf: Recognize special arithmetic shift in the verifier")
Signed-off-by: Daniel Wade <danjwade95@gmail.com>
Reviewed-by: Amery Hung <ameryhung@gmail.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260314021521.128361-2-danjwade95@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The BPF interpreter's signed 32-bit division and modulo handlers use
the kernel abs() macro on s32 operands. The abs() macro documentation
(include/linux/math.h) explicitly states the result is undefined when
the input is the type minimum. When DST contains S32_MIN (0x80000000),
abs((s32)DST) triggers undefined behavior and returns S32_MIN unchanged
on arm64/x86. This value is then sign-extended to u64 as
0xFFFFFFFF80000000, causing do_div() to compute the wrong result.
The verifier's abstract interpretation (scalar32_min_max_sdiv) computes
the mathematically correct result for range tracking, creating a
verifier/interpreter mismatch that can be exploited for out-of-bounds
map value access.
Introduce abs_s32() which handles S32_MIN correctly by casting to u32
before negating, avoiding signed overflow entirely. Replace all 8
abs((s32)...) call sites in the interpreter's sdiv32/smod32 handlers.
s32 is the only affected case -- the s64 division/modulo handlers do
not use abs().
Fixes: ec0e2da95f ("bpf: Support new signed div/mod instructions.")
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Signed-off-by: Jenny Guanni Qu <qguanni@gmail.com>
Link: https://lore.kernel.org/r/20260311011116.2108005-2-qguanni@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The sleepable context check for global function calls in
check_func_call() open-codes the same checks that in_sleepable_context()
already performs. Replace the open-coded check with a call to
in_sleepable_context() and use non_sleepable_context_description() for
the error message, consistent with check_helper_call() and
check_kfunc_call().
Note that in_sleepable_context() also checks active_locks, which
overlaps with the existing active_locks check above it. However, the two
checks serve different purposes: the active_locks check rejects all
global function calls while holding a lock (not just sleepable ones), so
it must remain as a separate guard.
Update the expected error messages in the irq and preempt_lock selftests
to match.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260318174327.3151925-4-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
check_kfunc_call() has multiple scattered checks that reject sleepable
kfuncs in various non-sleepable contexts (RCU, preempt-disabled, IRQ-
disabled). These are the same conditions already checked by
in_sleepable_context(), so replace them with a single consolidated
check.
This also simplifies the preempt lock tracking by flattening the nested
if/else structure into a linear chain: preempt_disable increments,
preempt_enable checks for underflow and decrements. The sleepable check
is kept as a separate block since it is logically distinct from the lock
accounting.
No functional change since in_sleepable_context() checks all the same
state (active_rcu_locks, active_preempt_locks, active_locks,
active_irq_id, in_sleepable).
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260318174327.3151925-3-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
check_helper_call() prints the error message for every
env->cur_state->active* element when calling a sleepable helper.
Consolidate all of them into a single print statement.
The check for env->cur_state->active_locks was not part of the removed
print statements and will not be triggered with the consolidated print
as well because it is checked in do_check() before check_helper_call()
is even reached.
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260318174327.3151925-2-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
process_bpf_exit_full() passes check_lock = !curframe to
check_resource_leak(), which is false in cases when bpf_throw() is
called from a static subprog. This makes check_resource_leak() to skip
validation of active_rcu_locks, active_preempt_locks, and
active_irq_id on exception exits from subprogs.
At runtime bpf_throw() unwinds the stack via ORC without releasing any
user-acquired locks, which may cause various issues as the result.
Fix by setting check_lock = true for exception exits regardless of
curframe, since exceptions bypass all intermediate frame
cleanup. Update the error message prefix to "bpf_throw" for exception
exits to distinguish them from normal BPF_EXIT.
Fix reject_subprog_with_rcu_read_lock test which was previously
passing for the wrong reason. Test program returned directly from the
subprog call without closing the RCU section, so the error was
triggered by the unclosed RCU lock on normal exit, not by
bpf_throw. Update __msg annotations for affected tests to match the
new "bpf_throw" error prefix.
The spin_lock case is not affected because they are already checked [1]
at the call site in do_check_insn() before bpf_throw can run.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/bpf/verifier.c?h=v7.0-rc4#n21098
Assisted-by: Claude:claude-opus-4-6
Fixes: f18b03faba ("bpf: Implement BPF exceptions")
Signed-off-by: Ihor Solodrai <ihor.solodrai@linux.dev>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260320000809.643798-1-ihor.solodrai@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
While very unlikely, local storage theoretically may leak memory of the
size of "struct bpf_local_storage" when destroy() fails to grab
local_storage->lock and initializes selem->local_storage before other
racing map_free() see it. Warn the user to allow debugging the issue
instead of leaking the memory silently.
Note that test_maps in bpf selftests already stress tested
bpf_selem_unlink_nofail() by creating 4096 sockets and then immediately
destroying them in multiple threads. With 64 threads, 64 x 4096 socket
local storages were created and destroyed during the test and no warning
in the function were triggered.
Signed-off-by: Amery Hung <ameryhung@gmail.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://patch.msgid.link/20260318224219.615105-1-ameryhung@gmail.com
Currently, local storage may deadlock when deferring freeing selem or
local storage through kfree_rcu(), call_rcu() or call_rcu_tasks_trace()
in NMI or reentrant. Since deleting selem in NMI is an unlikely use
case, partially mitigate it by returning error when calling from
bpf_xxx_storage_delete() helpers in NMI. Note that, it is still possible
to deadlock through reentrant. A full mitigation requires returning
error when irqs_disabled() is true, which, however is too heavy-handed
for bpf_xxx_storage_delete().
The long-term solution requires _nolock versions of call_rcu. Another
possible solution is to defer the free through irq_work [0], but it
would grow the size of selem, which is non-ideal.
The check is only needed in bpf_selem_unlink(), which is used by helpers
and syscalls. bpf_selem_unlink_nofail() is fine as it is called during
map and owner tear down that never run in NMI or reentrant.
[0] https://lore.kernel.org/bpf/20260205190233.912-1-alexei.starovoitov@gmail.com/
Fixes: a10787e6d5 ("bpf: Enable task local storage for tracing programs")
Signed-off-by: Amery Hung <ameryhung@gmail.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://patch.msgid.link/20260319025716.2361065-1-ameryhung@gmail.com
Gregory reported in [0] that the global_map_resize test when run in
repeatedly ends up failing during program load. This stems from the fact
that BTF reference has not dropped to zero after the previous run's
module is unloaded, and the older module's BTF is still discoverable and
visible. Later, in libbpf, load_module_btfs() will find the ID for this
stale BTF, open its fd, and then it will be used during program load
where later steps taking module reference using btf_try_get_module()
fail since the underlying module for the BTF is gone.
Logically, once a module is unloaded, it's associated BTF artifacts
should become hidden. The BTF object inside the kernel may still remain
alive as long its reference counts are alive, but it should no longer be
discoverable.
To fix this, let us call btf_free_id() from the MODULE_STATE_GOING case
for the module unload to free the BTF associated IDR entry, and disable
its discovery once module unload returns to user space. If a race
happens during unload, the outcome is non-deterministic anyway. However,
user space should be able to rely on the guarantee that once it has
synchronously established a successful module unload, no more stale
artifacts associated with this module can be obtained subsequently.
Note that we must be careful to not invoke btf_free_id() in btf_put()
when btf_is_module() is true now. There could be a window where the
module unload drops a non-terminal reference, frees the IDR, but the
same ID gets reused and the second unconditional btf_free_id() ends up
releasing an unrelated entry.
To avoid a special case for btf_is_module() case, set btf->id to zero to
make btf_free_id() idempotent, such that we can unconditionally invoke it
from btf_put(), and also from the MODULE_STATE_GOING case. Since zero is
an invalid IDR, the idr_remove() should be a noop.
Note that we can be sure that by the time we reach final btf_put() for
btf_is_module() case, the btf_free_id() is already done, since the
module itself holds the BTF reference, and it will call this function
for the BTF before dropping its own reference.
[0]: https://lore.kernel.org/bpf/cover.1773170190.git.grbell@redhat.com
Fixes: 36e68442d1 ("bpf: Load and verify kernel module BTFs")
Acked-by: Martin KaFai Lau <martin.lau@kernel.org>
Suggested-by: Martin KaFai Lau <martin.lau@kernel.org>
Reported-by: Gregory Bell <grbell@redhat.com>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260312205307.1346991-1-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The BPF verifier currently enforces a call stack depth of 8 frames,
regardless of the actual stack space consumption of those frames. The
limit is necessary for static call stacks, because the bookkeeping data
structures used by the verifier when stepping into static functions
during verification only support 8 stack frames. However, this
limitation only matters for static stack frames: Global subprogs are
verified by themselves and do not require limiting the call depth.
Relax this limitation to only apply to static stack frames. Verification
now only fails when there is a sequence of 8 calls to non-global
subprogs. Calling into a global subprog resets the counter. This allows
deeper call stacks, provided all frames still fit in the stack.
The change does not increase the maximum size of the call stack, only
the maximum number of frames we can place in it.
Also change the progs/test_global_func3.c selftest to use static
functions, since with the new patch it would otherwise unexpectedly
pass verification.
Acked-by: Mykyta Yatsenko <yatsenko@meta.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260316161225.128011-2-emil@etsalapatis.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In commit 5dbb19b16a ("bpf: Add third round of bounds deduction"), I
added a new round of bounds deduction because two rounds were not enough
to converge to a fixed point. This commit slightly refactor the bounds
deduction logic such that two rounds are enough.
In [1], Eduard noticed that after we improved the refinement logic, a
third call to the bounds deduction (__reg_deduce_bounds) was needed to
converge to a fixed point. More specifically, we needed this third call
to improve the s64 range using the s32 range. We added the third call
and postponed a more detailed analysis of the refinement logic.
I've been looking into this more recently. The register refinement
consists of the following calls.
__update_reg_bounds();
3 x __reg_deduce_bounds() {
deduce_bounds_32_from_64();
deduce_bounds_32_from_32();
deduce_bounds_64_from_64();
deduce_bounds_64_from_32();
};
__reg_bound_offset();
__update_reg_bounds();
From this, we can observe that we first improve the 32bit ranges from
the 64bit ranges in deduce_bounds_32_from_64, then improve the 64bit
ranges on their own in deduce_bounds_64_from_64. Intuitively, if we
were to improve the 64bit ranges on their own *before* we use them to
improve the 32bit ranges, we may reach a fixed point earlier.
In a similar manner, using CBMC, Eduard found that it's best to improve
the 32bit ranges on their own *after* we've improve them using the 64bit
ranges. That is, running deduce_bounds_32_from_32 after
deduce_bounds_32_from_64.
These changes allow us to lose one call to __reg_deduce_bounds. Without
this reordering, the test "verifier_bounds/bounds deduction cross sign
boundary, negative overlap" fails when removing one call to
__reg_deduce_bounds. In some cases, this change can even improve
precision a little bit, as illustrated in the new selftest in the next
patch.
As expected, this change didn't have any impact on the number of
instructions processed when running it through the Cilium complexity
test suite [2].
Link: https://lore.kernel.org/bpf/aIKtSK9LjQXB8FLY@mail.gmail.com/ [1]
Link: https://pchaigno.github.io/test-verifier-complexity.html [2]
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Co-developed-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
Link: https://lore.kernel.org/r/1b00d2749ec4c774c3ada84e265ac7fda72cfe56.1773401138.git.paul.chaignon@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
BPF_ST | BPF_PROBE_MEM32 immediate stores are not handled by
bpf_jit_blind_insn(), allowing user-controlled 32-bit immediates to
survive unblinded into JIT-compiled native code when bpf_jit_harden >= 1.
The root cause is that convert_ctx_accesses() rewrites BPF_ST|BPF_MEM
to BPF_ST|BPF_PROBE_MEM32 for arena pointer stores during verification,
before bpf_jit_blind_constants() runs during JIT compilation. The
blinding switch only matches BPF_ST|BPF_MEM (mode 0x60), not
BPF_ST|BPF_PROBE_MEM32 (mode 0xa0). The instruction falls through
unblinded.
Add BPF_ST|BPF_PROBE_MEM32 cases to bpf_jit_blind_insn() alongside the
existing BPF_ST|BPF_MEM cases. The blinding transformation is identical:
load the blinded immediate into BPF_REG_AX via mov+xor, then convert
the immediate store to a register store (BPF_STX).
The rewritten STX instruction must preserve the BPF_PROBE_MEM32 mode so
the architecture JIT emits the correct arena addressing (R12-based on
x86-64). Cannot use the BPF_STX_MEM() macro here because it hardcodes
BPF_MEM mode; construct the instruction directly instead.
Fixes: 6082b6c328 ("bpf: Recognize addr_space_cast instruction in the verifier.")
Reviewed-by: Puranjay Mohan <puranjay@kernel.org>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Sachin Kumar <xcyfun@protonmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/Y6IT5VvNRchPBLI5D7JZHBzZrU9rb0ycRJPJzJSXGj7kJlX8RJwZFSM2YZjcDxoQKABkxt1T8Os2gi23PYyFuQe6KkZGWVyfz8K5afdy9ak=@protonmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch adds support to validate a pointer as not null when its
value is compared to a register whose value the verifier knows to be
null.
Initial pattern only verifies against an immediate operand.
Signed-off-by: Cupertino Miranda <cupertino.miranda@oracle.com>
Cc: David Faust <david.faust@oracle.com>
Cc: Jose Marchesi <jose.marchesi@oracle.com>
Cc: Elena Zannoni <elena.zannoni@oracle.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260304195018.181396-3-cupertino.miranda@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When a register undergoes a BPF_END (byte swap) operation, its scalar
value is mutated in-place. If this register previously shared a scalar ID
with another register (e.g., after an `r1 = r0` assignment), this tie must
be broken.
Currently, the verifier misses resetting `dst_reg->id` to 0 for BPF_END.
Consequently, if a conditional jump checks the swapped register, the
verifier incorrectly propagates the learned bounds to the linked register,
leading to false confidence in the linked register's value and potentially
allowing out-of-bounds memory accesses.
Fix this by explicitly resetting `dst_reg->id` to 0 in the BPF_END case
to break the scalar tie, similar to how BPF_NEG handles it via
`__mark_reg_known`.
Fixes: 9d21199842 ("bpf: Add bitwise tracking for BPF_END")
Closes: https://lore.kernel.org/bpf/AMBPR06MB108683CFEB1CB8D9E02FC95ECF17EA@AMBPR06MB10868.eurprd06.prod.outlook.com/
Link: https://lore.kernel.org/bpf/4be25f7442a52244d0dd1abb47bc6750e57984c9.camel@gmail.com/
Reported-by: Guillaume Laporte <glapt.pro@outlook.com>
Co-developed-by: Tianci Cao <ziye@zju.edu.cn>
Signed-off-by: Tianci Cao <ziye@zju.edu.cn>
Co-developed-by: Shenghao Yuan <shenghaoyuan0928@163.com>
Signed-off-by: Shenghao Yuan <shenghaoyuan0928@163.com>
Signed-off-by: Yazhou Tang <tangyazhou518@outlook.com>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260304083228.142016-2-tangyazhou@zju.edu.cn
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
fmod_ret BPF programs can only be attached to selected functions. For
convenience, the error injection list was originally used (along with
functions prefixed with "security_"), which contains syscalls and
several other functions.
When error injection is disabled (CONFIG_FUNCTION_ERROR_INJECTION=n),
that list is empty and fmod_ret programs are effectively unavailable for
most of the functions. In such a case, at least enable fmod_ret programs
on syscalls.
Signed-off-by: Viktor Malik <vmalik@redhat.com>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Acked-by: Leon Hwang <leon.hwang@linux.dev>
Link: https://lore.kernel.org/r/472310f9a5f4944ad03214e4d943a4830fd8eb76.1773055375.git.vmalik@redhat.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Sleepable BPF programs can only be attached to selected functions. For
convenience, the error injection list was originally used, which
contains syscalls and several other functions.
When error injection is disabled (CONFIG_FUNCTION_ERROR_INJECTION=n),
that list is empty and sleepable tracing programs are effectively
unavailable. In such a case, at least enable sleepable programs on
syscalls. For discussion why syscalls were chosen, see [1].
To detect that a function is a syscall handler, we check for
arch-specific prefixes for the most common architectures. Unfortunately,
the prefixes are hard-coded in arch syscall code so we need to hard-code
them, too.
[1] https://lore.kernel.org/bpf/CAADnVQK6qP8izg+k9yV0vdcT-+=axtFQ2fKw7D-2Ei-V6WS5Dw@mail.gmail.com/
Signed-off-by: Viktor Malik <vmalik@redhat.com>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Acked-by: Leon Hwang <leon.hwang@linux.dev>
Link: https://lore.kernel.org/r/2704a8512746655037e3c02b471b31bd0d76c8db.1773055375.git.vmalik@redhat.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Fix an inconsistency between func_states_equal() and
collect_linked_regs():
- regsafe() uses check_ids() to verify that cached and current states
have identical register id mapping.
- func_states_equal() calls regsafe() only for registers computed as
live by compute_live_registers().
- clean_live_states() is supposed to remove dead registers from cached
states, but it can skip states belonging to an iterator-based loop.
- collect_linked_regs() collects all registers sharing the same id,
ignoring the marks computed by compute_live_registers().
Linked registers are stored in the state's jump history.
- backtrack_insn() marks all linked registers for an instruction
as precise whenever one of the linked registers is precise.
The above might lead to a scenario:
- There is an instruction I with register rY known to be dead at I.
- Instruction I is reached via two paths: first A, then B.
- On path A:
- There is an id link between registers rX and rY.
- Checkpoint C is created at I.
- Linked register set {rX, rY} is saved to the jump history.
- rX is marked as precise at I, causing both rX and rY
to be marked precise at C.
- On path B:
- There is no id link between registers rX and rY,
otherwise register states are sub-states of those in C.
- Because rY is dead at I, check_ids() returns true.
- Current state is considered equal to checkpoint C,
propagate_precision() propagates spurious precision
mark for register rY along the path B.
- Depending on a program, this might hit verifier_bug()
in the backtrack_insn(), e.g. if rY ∈ [r1..r5]
and backtrack_insn() spots a function call.
The reproducer program is in the next patch.
This was hit by sched_ext scx_lavd scheduler code.
Changes in tests:
- verifier_scalar_ids.c selftests need modification to preserve
some registers as live for __msg() checks.
- exceptions_assert.c adjusted to match changes in the verifier log,
R0 is dead after conditional instruction and thus does not get
range.
- precise.c adjusted to match changes in the verifier log, register r9
is dead after comparison and it's range is not important for test.
Reported-by: Emil Tsalapatis <emil@etsalapatis.com>
Fixes: 0fb3cf6110 ("bpf: use register liveness information for func_states_equal")
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260306-linked-regs-and-propagate-precision-v1-1-18e859be570d@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Same as in __reg64_deduce_bounds(), refine s32/u32 ranges
in __reg32_deduce_bounds() in the following situations:
- s32 range crosses U32_MAX/0 boundary, positive part of the s32 range
overlaps with u32 range:
0 U32_MAX
| [xxxxxxxxxxxxxx u32 range xxxxxxxxxxxxxx] |
|----------------------------|----------------------------|
|xxxxx s32 range xxxxxxxxx] [xxxxxxx|
0 S32_MAX S32_MIN -1
- s32 range crosses U32_MAX/0 boundary, negative part of the s32 range
overlaps with u32 range:
0 U32_MAX
| [xxxxxxxxxxxxxx u32 range xxxxxxxxxxxxxx] |
|----------------------------|----------------------------|
|xxxxxxxxx] [xxxxxxxxxxxx s32 range |
0 S32_MAX S32_MIN -1
- No refinement if ranges overlap in two intervals.
This helps for e.g. consider the following program:
call %[bpf_get_prandom_u32];
w0 &= 0xffffffff;
if w0 < 0x3 goto 1f; // on fall-through u32 range [3..U32_MAX]
if w0 s> 0x1 goto 1f; // on fall-through s32 range [S32_MIN..1]
if w0 s< 0x0 goto 1f; // range can be narrowed to [S32_MIN..-1]
r10 = 0;
1: ...;
The reg_bounds.c selftest is updated to incorporate identical logic,
refinement based on non-overflowing range halves:
((x ∩ [0, smax]) ∩ (y ∩ [0, smax])) ∪
((x ∩ [smin,-1]) ∩ (y ∩ [smin,-1]))
Reported-by: Andrea Righi <arighi@nvidia.com>
Reported-by: Emil Tsalapatis <emil@etsalapatis.com>
Closes: https://lore.kernel.org/bpf/aakqucg4vcujVwif@gpd4/T/
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260306-bpf-32-bit-range-overflow-v3-1-f7f67e060a6b@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
kthread_exit became a macro to do_exit in commit 28aaa9c399
("kthread: consolidate kthread exit paths to prevent use-after-free"),
so there is no kthread_exit function BTF ID to resolve. Remove it from
noreturn_deny to avoid resolve_btfids unresolved symbol warnings.
Signed-off-by: Christian Loehle <christian.loehle@arm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The root cause of this bug is that when 'bpf_link_put' reduces the
refcount of 'shim_link->link.link' to zero, the resource is considered
released but may still be referenced via 'tr->progs_hlist' in
'cgroup_shim_find'. The actual cleanup of 'tr->progs_hlist' in
'bpf_shim_tramp_link_release' is deferred. During this window, another
process can cause a use-after-free via 'bpf_trampoline_link_cgroup_shim'.
Based on Martin KaFai Lau's suggestions, I have created a simple patch.
To fix this:
Add an atomic non-zero check in 'bpf_trampoline_link_cgroup_shim'.
Only increment the refcount if it is not already zero.
Testing:
I verified the fix by adding a delay in
'bpf_shim_tramp_link_release' to make the bug easier to trigger:
static void bpf_shim_tramp_link_release(struct bpf_link *link)
{
/* ... */
if (!shim_link->trampoline)
return;
+ msleep(100);
WARN_ON_ONCE(bpf_trampoline_unlink_prog(&shim_link->link,
shim_link->trampoline, NULL));
bpf_trampoline_put(shim_link->trampoline);
}
Before the patch, running a PoC easily reproduced the crash(almost 100%)
with a call trace similar to KaiyanM's report.
After the patch, the bug no longer occurs even after millions of
iterations.
Fixes: 69fd337a97 ("bpf: per-cgroup lsm flavor")
Reported-by: Kaiyan Mei <M202472210@hust.edu.cn>
Closes: https://lore.kernel.org/bpf/3c4ebb0b.46ff8.19abab8abe2.Coremail.kaiyanm@hust.edu.cn/
Signed-off-by: Lang Xu <xulang@uniontech.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://patch.msgid.link/279EEE1BA1DDB49D+20260303095217.34436-1-xulang@uniontech.com
Global subprogs are currently not allowed to return void. Adjust
verifier logic to allow global functions with a void return type.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260228184759.108145-5-emil@etsalapatis.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
From: Eduard Zingerman <eddyz87@gmail.com>
Both main progs and subprogs use the same function in the verifier,
check_return_code, to verify the type and value range of the register
being returned. However, subprogs only need a subset of the logic in
check_return_code. this also goes the way - check_return_code explicitly
checks whether it is handling a subprogram in multiple places, complicating
the logic. Separate the handling of the two into separate functions.
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260228184759.108145-4-emil@etsalapatis.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
From: Eduard Zingerman <eddyz87@gmail.com>
The check_return_code function has explicit checks on whether
a program type can return void. Factor this logic out to reuse
it later for both main progs and subprogs.
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260228184759.108145-3-emil@etsalapatis.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Factor the return value range calculation logic in check_return_code
out of the function in preparation for separating the return value
validation logic for BPF_EXIT and bpf_throw().
No functional changes. The change made in return_retval_code's handling
of PROG_TRACING program types (not error'ing on the default case) is a
no-op because the match on the program attach type is exhaustive.
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260228184759.108145-2-emil@etsalapatis.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The limitation on fixed offsets stems from the fact that certain program
types rewrite the accesses to the context structure and translate them
to accesses to the real underlying type. Technically, in the past, we
could have stashed the register offset in insn_aux and made rewrites
work, but we've never needed it in the past since the offset for such
context structures easily fit in the s16 signed instruction offset.
Regardless, the consequence is that for program types where the program
type's verifier ops doesn't supply a convert_ctx_access callback, we
unnecessarily reject accesses with a modified ctx pointer (i.e., one
whose offset has been shifted) in check_ptr_off_reg. Make an exception
for such program types (like syscall, tracepoint, raw_tp, etc.).
Pass in fixed_off_ok as true to __check_ptr_off_reg for such cases, and
accumulate the register offset into insn->off passed to check_ctx_access.
In particular, the accumulation is critical since we need to correctly
track the max_ctx_offset which is used for bounds checking the buffer
for syscall programs at runtime.
Reported-by: Tejun Heo <tj@kernel.org>
Reported-by: Dan Schatzberg <dschatzberg@meta.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260227005725.1247305-2-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
cpu_map_bpf_prog_run_xdp() handles XDP_PASS, XDP_REDIRECT, and
XDP_DROP but is missing an XDP_ABORTED case. Without it, XDP_ABORTED
falls into the default case which logs a misleading "invalid XDP
action" warning instead of tracing the abort via trace_xdp_exception().
Add the missing XDP_ABORTED case with trace_xdp_exception(), matching
the handling already present in the skb path (cpu_map_bpf_prog_run_skb),
devmap (dev_map_bpf_prog_run), and the generic XDP path (do_xdp_generic).
Also pass xdpf->dev_rx instead of NULL to bpf_warn_invalid_xdp_action()
in the default case, so the warning includes the actual device name.
This aligns with the generic XDP path in net/core/dev.c which already
passes the real device.
Signed-off-by: Anand Kumar Shaw <anandkrshawheritage@gmail.com>
Link: https://lore.kernel.org/r/20260218042924.42931-1-anandkrshawheritage@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Implementing BPF version of preempt_count() requires accessing lowcore
from BPF. Since lowcore can be relocated, open-coding
(struct lowcore *)0 does not work, so add a kfunc.
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Link: https://lore.kernel.org/r/20260217160813.100855-2-iii@linux.ibm.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
We're hitting an invariant violation in Cilium that sometimes leads to
BPF programs being rejected and Cilium failing to start [1]. The
following extract from verifier logs shows what's happening:
from 201 to 236: R1=0 R6=ctx() R7=1 R9=scalar(smin=umin=smin32=umin32=3584,smax=umax=smax32=umax32=3840,var_off=(0xe00; 0x100)) R10=fp0
236: R1=0 R6=ctx() R7=1 R9=scalar(smin=umin=smin32=umin32=3584,smax=umax=smax32=umax32=3840,var_off=(0xe00; 0x100)) R10=fp0
; if (magic == MARK_MAGIC_HOST || magic == MARK_MAGIC_OVERLAY || magic == MARK_MAGIC_ENCRYPT) @ bpf_host.c:1337
236: (16) if w9 == 0xe00 goto pc+45 ; R9=scalar(smin=umin=smin32=umin32=3585,smax=umax=smax32=umax32=3840,var_off=(0xe00; 0x100))
237: (16) if w9 == 0xf00 goto pc+1
verifier bug: REG INVARIANTS VIOLATION (false_reg1): range bounds violation u64=[0xe01, 0xe00] s64=[0xe01, 0xe00] u32=[0xe01, 0xe00] s32=[0xe01, 0xe00] var_off=(0xe00, 0x0)
We reach instruction 236 with two possible values for R9, 0xe00 and
0xf00. This is perfectly reflected in the tnum, but of course the ranges
are less accurate and cover [0xe00; 0xf00]. Taking the fallthrough path
at instruction 236 allows the verifier to reduce the range to
[0xe01; 0xf00]. The tnum is however not updated.
With these ranges, at instruction 237, the verifier is not able to
deduce that R9 is always equal to 0xf00. Hence the fallthrough pass is
explored first, the verifier refines the bounds using the assumption
that R9 != 0xf00, and ends up with an invariant violation.
This pattern of impossible branch + bounds refinement is common to all
invariant violations seen so far. The long-term solution is likely to
rely on the refinement + invariant violation check to detect dead
branches, as started by Eduard. To fix the current issue, we need
something with less refactoring that we can backport.
This patch uses the tnum_step helper introduced in the previous patch to
detect the above situation. In particular, three cases are now detected
in the bounds refinement:
1. The u64 range and the tnum only overlap in umin.
u64: ---[xxxxxx]-----
tnum: --xx----------x-
2. The u64 range and the tnum only overlap in the maximum value
represented by the tnum, called tmax.
u64: ---[xxxxxx]-----
tnum: xx-----x--------
3. The u64 range and the tnum only overlap in between umin (excluded)
and umax.
u64: ---[xxxxxx]-----
tnum: xx----x-------x-
To detect these three cases, we call tnum_step(tnum, umin), which
returns the smallest member of the tnum greater than umin, called
tnum_next here. We're in case (1) if umin is part of the tnum and
tnum_next is greater than umax. We're in case (2) if umin is not part of
the tnum and tnum_next is equal to tmax. Finally, we're in case (3) if
umin is not part of the tnum, tnum_next is inferior or equal to umax,
and calling tnum_step a second time gives us a value past umax.
This change implements these three cases. With it, the above bytecode
looks as follows:
0: (85) call bpf_get_prandom_u32#7 ; R0=scalar()
1: (47) r0 |= 3584 ; R0=scalar(smin=0x8000000000000e00,umin=umin32=3584,smin32=0x80000e00,var_off=(0xe00; 0xfffffffffffff1ff))
2: (57) r0 &= 3840 ; R0=scalar(smin=umin=smin32=umin32=3584,smax=umax=smax32=umax32=3840,var_off=(0xe00; 0x100))
3: (15) if r0 == 0xe00 goto pc+2 ; R0=3840
4: (15) if r0 == 0xf00 goto pc+1
4: R0=3840
6: (95) exit
In addition to the new selftests, this change was also verified with
Agni [3]. For the record, the raw SMT is available at [4]. The property
it verifies is that: If a concrete value x is contained in all input
abstract values, after __update_reg_bounds, it will continue to be
contained in all output abstract values.
Link: https://github.com/cilium/cilium/issues/44216 [1]
Link: https://pchaigno.github.io/test-verifier-complexity.html [2]
Link: https://github.com/bpfverif/agni [3]
Link: https://pastebin.com/raw/naCfaqNx [4]
Fixes: 0df1a55afa ("bpf: Warn on internal verifier errors")
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Tested-by: Marco Schirrmeister <mschirrmeister@gmail.com>
Co-developed-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Signed-off-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
Link: https://lore.kernel.org/r/ef254c4f68be19bd393d450188946821c588565d.1772225741.git.paul.chaignon@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This commit introduces tnum_step(), a function that, when given t, and a
number z returns the smallest member of t larger than z. The number z
must be greater or equal to the smallest member of t and less than the
largest member of t.
The first step is to compute j, a number that keeps all of t's known
bits, and matches all unknown bits to z's bits. Since j is a member of
the t, it is already a candidate for result. However, we want our result
to be (minimally) greater than z.
There are only two possible cases:
(1) Case j <= z. In this case, we want to increase the value of j and
make it > z.
(2) Case j > z. In this case, we want to decrease the value of j while
keeping it > z.
(Case 1) j <= z
t = xx11x0x0
z = 10111101 (189)
j = 10111000 (184)
^
k
(Case 1.1) Let's first consider the case where j < z. We will address j
== z later.
Since z > j, there had to be a bit position that was 1 in z and a 0 in
j, beyond which all positions of higher significance are equal in j and
z. Further, this position could not have been unknown in a, because the
unknown positions of a match z. This position had to be a 1 in z and
known 0 in t.
Let k be position of the most significant 1-to-0 flip. In our example, k
= 3 (starting the count at 1 at the least significant bit). Setting (to
1) the unknown bits of t in positions of significance smaller than
k will not produce a result > z. Hence, we must set/unset the unknown
bits at positions of significance higher than k. Specifically, we look
for the next larger combination of 1s and 0s to place in those
positions, relative to the combination that exists in z. We can achieve
this by concatenating bits at unknown positions of t into an integer,
adding 1, and writing the bits of that result back into the
corresponding bit positions previously extracted from z.
>From our example, considering only positions of significance greater
than k:
t = xx..x
z = 10..1
+ 1
-----
11..0
This is the exact combination 1s and 0s we need at the unknown bits of t
in positions of significance greater than k. Further, our result must
only increase the value minimally above z. Hence, unknown bits in
positions of significance smaller than k should remain 0. We finally
have,
result = 11110000 (240)
(Case 1.2) Now consider the case when j = z, for example
t = 1x1x0xxx
z = 10110100 (180)
j = 10110100 (180)
Matching the unknown bits of the t to the bits of z yielded exactly z.
To produce a number greater than z, we must set/unset the unknown bits
in t, and *all* the unknown bits of t candidates for being set/unset. We
can do this similar to Case 1.1, by adding 1 to the bits extracted from
the masked bit positions of z. Essentially, this case is equivalent to
Case 1.1, with k = 0.
t = 1x1x0xxx
z = .0.1.100
+ 1
---------
.0.1.101
This is the exact combination of bits needed in the unknown positions of
t. After recalling the known positions of t, we get
result = 10110101 (181)
(Case 2) j > z
t = x00010x1
z = 10000010 (130)
j = 10001011 (139)
^
k
Since j > z, there had to be a bit position which was 0 in z, and a 1 in
j, beyond which all positions of higher significance are equal in j and
z. This position had to be a 0 in z and known 1 in t. Let k be the
position of the most significant 0-to-1 flip. In our example, k = 4.
Because of the 0-to-1 flip at position k, a member of t can become
greater than z if the bits in positions greater than k are themselves >=
to z. To make that member *minimally* greater than z, the bits in
positions greater than k must be exactly = z. Hence, we simply match all
of t's unknown bits in positions more significant than k to z's bits. In
positions less significant than k, we set all t's unknown bits to 0
to retain minimality.
In our example, in positions of greater significance than k (=4),
t=x000. These positions are matched with z (1000) to produce 1000. In
positions of lower significance than k, t=10x1. All unknown bits are set
to 0 to produce 1001. The final result is:
result = 10001001 (137)
This concludes the computation for a result > z that is a member of t.
The procedure for tnum_step() in this commit implements the idea
described above. As a proof of correctness, we verified the algorithm
against a logical specification of tnum_step. The specification asserts
the following about the inputs t, z and output res that:
1. res is a member of t, and
2. res is strictly greater than z, and
3. there does not exist another value res2 such that
3a. res2 is also a member of t, and
3b. res2 is greater than z
3c. res2 is smaller than res
We checked the implementation against this logical specification using
an SMT solver. The verification formula in SMTLIB format is available
at [1]. The verification returned an "unsat": indicating that no input
assignment exists for which the implementation and the specification
produce different outputs.
In addition, we also automatically generated the logical encoding of the
C implementation using Agni [2] and verified it against the same
specification. This verification also returned an "unsat", confirming
that the implementation is equivalent to the specification. The formula
for this check is also available at [3].
Link: https://pastebin.com/raw/2eRWbiit [1]
Link: https://github.com/bpfverif/agni [2]
Link: https://pastebin.com/raw/EztVbBJ2 [3]
Co-developed-by: Srinivas Narayana <srinivas.narayana@rutgers.edu>
Signed-off-by: Srinivas Narayana <srinivas.narayana@rutgers.edu>
Co-developed-by: Santosh Nagarakatte <santosh.nagarakatte@rutgers.edu>
Signed-off-by: Santosh Nagarakatte <santosh.nagarakatte@rutgers.edu>
Signed-off-by: Harishankar Vishwanathan <harishankar.vishwanathan@gmail.com>
Link: https://lore.kernel.org/r/93fdf71910411c0f19e282ba6d03b4c65f9c5d73.1772225741.git.paul.chaignon@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
On PREEMPT_RT kernels, the per-CPU xdp_dev_bulk_queue (bq) can be
accessed concurrently by multiple preemptible tasks on the same CPU.
The original code assumes bq_enqueue() and __dev_flush() run atomically
with respect to each other on the same CPU, relying on
local_bh_disable() to prevent preemption. However, on PREEMPT_RT,
local_bh_disable() only calls migrate_disable() (when
PREEMPT_RT_NEEDS_BH_LOCK is not set) and does not disable
preemption, which allows CFS scheduling to preempt a task during
bq_xmit_all(), enabling another task on the same CPU to enter
bq_enqueue() and operate on the same per-CPU bq concurrently.
This leads to several races:
1. Double-free / use-after-free on bq->q[]: bq_xmit_all() snapshots
cnt = bq->count, then iterates bq->q[0..cnt-1] to transmit frames.
If preempted after the snapshot, a second task can call bq_enqueue()
-> bq_xmit_all() on the same bq, transmitting (and freeing) the
same frames. When the first task resumes, it operates on stale
pointers in bq->q[], causing use-after-free.
2. bq->count and bq->q[] corruption: concurrent bq_enqueue() modifying
bq->count and bq->q[] while bq_xmit_all() is reading them.
3. dev_rx/xdp_prog teardown race: __dev_flush() clears bq->dev_rx and
bq->xdp_prog after bq_xmit_all(). If preempted between
bq_xmit_all() return and bq->dev_rx = NULL, a preempting
bq_enqueue() sees dev_rx still set (non-NULL), skips adding bq to
the flush_list, and enqueues a frame. When __dev_flush() resumes,
it clears dev_rx and removes bq from the flush_list, orphaning the
newly enqueued frame.
4. __list_del_clearprev() on flush_node: similar to the cpumap race,
both tasks can call __list_del_clearprev() on the same flush_node,
the second dereferences the prev pointer already set to NULL.
The race between task A (__dev_flush -> bq_xmit_all) and task B
(bq_enqueue -> bq_xmit_all) on the same CPU:
Task A (xdp_do_flush) Task B (ndo_xdp_xmit redirect)
---------------------- --------------------------------
__dev_flush(flush_list)
bq_xmit_all(bq)
cnt = bq->count /* e.g. 16 */
/* start iterating bq->q[] */
<-- CFS preempts Task A -->
bq_enqueue(dev, xdpf)
bq->count == DEV_MAP_BULK_SIZE
bq_xmit_all(bq, 0)
cnt = bq->count /* same 16! */
ndo_xdp_xmit(bq->q[])
/* frames freed by driver */
bq->count = 0
<-- Task A resumes -->
ndo_xdp_xmit(bq->q[])
/* use-after-free: frames already freed! */
Fix this by adding a local_lock_t to xdp_dev_bulk_queue and acquiring
it in bq_enqueue() and __dev_flush(). These paths already run under
local_bh_disable(), so use local_lock_nested_bh() which on non-RT is
a pure annotation with no overhead, and on PREEMPT_RT provides a
per-CPU sleeping lock that serializes access to the bq.
Fixes: 3253cb49cb ("softirq: Allow to drop the softirq-BKL lock on PREEMPT_RT")
Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jiayuan Chen <jiayuan.chen@shopee.com>
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Link: https://lore.kernel.org/r/20260225121459.183121-3-jiayuan.chen@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
On PREEMPT_RT kernels, the per-CPU xdp_bulk_queue (bq) can be accessed
concurrently by multiple preemptible tasks on the same CPU.
The original code assumes bq_enqueue() and __cpu_map_flush() run
atomically with respect to each other on the same CPU, relying on
local_bh_disable() to prevent preemption. However, on PREEMPT_RT,
local_bh_disable() only calls migrate_disable() (when
PREEMPT_RT_NEEDS_BH_LOCK is not set) and does not disable
preemption, which allows CFS scheduling to preempt a task during
bq_flush_to_queue(), enabling another task on the same CPU to enter
bq_enqueue() and operate on the same per-CPU bq concurrently.
This leads to several races:
1. Double __list_del_clearprev(): after bq->count is reset in
bq_flush_to_queue(), a preempting task can call bq_enqueue() ->
bq_flush_to_queue() on the same bq when bq->count reaches
CPU_MAP_BULK_SIZE. Both tasks then call __list_del_clearprev()
on the same bq->flush_node, the second call dereferences the
prev pointer that was already set to NULL by the first.
2. bq->count and bq->q[] races: concurrent bq_enqueue() can corrupt
the packet queue while bq_flush_to_queue() is processing it.
The race between task A (__cpu_map_flush -> bq_flush_to_queue) and
task B (bq_enqueue -> bq_flush_to_queue) on the same CPU:
Task A (xdp_do_flush) Task B (cpu_map_enqueue)
---------------------- ------------------------
bq_flush_to_queue(bq)
spin_lock(&q->producer_lock)
/* flush bq->q[] to ptr_ring */
bq->count = 0
spin_unlock(&q->producer_lock)
bq_enqueue(rcpu, xdpf)
<-- CFS preempts Task A --> bq->q[bq->count++] = xdpf
/* ... more enqueues until full ... */
bq_flush_to_queue(bq)
spin_lock(&q->producer_lock)
/* flush to ptr_ring */
spin_unlock(&q->producer_lock)
__list_del_clearprev(flush_node)
/* sets flush_node.prev = NULL */
<-- Task A resumes -->
__list_del_clearprev(flush_node)
flush_node.prev->next = ...
/* prev is NULL -> kernel oops */
Fix this by adding a local_lock_t to xdp_bulk_queue and acquiring it
in bq_enqueue() and __cpu_map_flush(). These paths already run under
local_bh_disable(), so use local_lock_nested_bh() which on non-RT is
a pure annotation with no overhead, and on PREEMPT_RT provides a
per-CPU sleeping lock that serializes access to the bq.
To reproduce, insert an mdelay(100) between bq->count = 0 and
__list_del_clearprev() in bq_flush_to_queue(), then run reproducer
provided by syzkaller.
Fixes: 3253cb49cb ("softirq: Allow to drop the softirq-BKL lock on PREEMPT_RT")
Reported-by: syzbot+2b3391f44313b3983e91@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/69369331.a70a0220.38f243.009d.GAE@google.com/T/
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jiayuan Chen <jiayuan.chen@shopee.com>
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Link: https://lore.kernel.org/r/20260225121459.183121-2-jiayuan.chen@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This assumption will always hold going forward, hence just remove the
various checks and assume it is true with a comment for the uninformed
reader.
Reviewed-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Amery Hung <ameryhung@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260227224806.646888-5-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Currently, when use_kmalloc_nolock is false, the freeing of fields for a
local storage selem is done eagerly before waiting for the RCU or RCU
tasks trace grace period to elapse. This opens up a window where the
program which has access to the selem can recreate the fields after the
freeing of fields is done eagerly, causing memory leaks when the element
is finally freed and returned to the kernel.
Make a few changes to address this. First, delay the freeing of fields
until after the grace periods have expired using a __bpf_selem_free_rcu
wrapper which is eventually invoked after transitioning through the
necessary number of grace period waits. Replace usage of the kfree_rcu
with call_rcu to be able to take a custom callback. Finally, care needs
to be taken to extend the rcu barriers for all cases, and not just when
use_kmalloc_nolock is true, as RCU and RCU tasks trace callbacks can be
in flight for either case and access the smap field, which is used to
obtain the BTF record to walk over special fields in the map value.
While we're at it, drop migrate_disable() from bpf_selem_free_rcu, since
migration should be disabled for RCU callbacks already.
Fixes: 9bac675e63 ("bpf: Postpone bpf_obj_free_fields to the rcu callback")
Reviewed-by: Amery Hung <ameryhung@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260227224806.646888-4-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
BPF hash map may now use the map_check_btf() callback to decide whether
to set a dtor on its bpf_mem_alloc or not. Unlike C++ where members can
opt out of const-ness using mutable, we must lose the const qualifier on
the callback such that we can avoid the ugly cast. Make the change and
adjust all existing users, and lose the comment in hashtab.c.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260227224806.646888-3-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
There is a race window where BPF hash map elements can leak special
fields if the program with access to the map value recreates these
special fields between the check_and_free_fields done on the map value
and its eventual return to the memory allocator.
Several ways were explored prior to this patch, most notably [0] tried
to use a poison value to reject attempts to recreate special fields for
map values that have been logically deleted but still accessible to BPF
programs (either while sitting in the free list or when reused). While
this approach works well for task work, timers, wq, etc., it is harder
to apply the idea to kptrs, which have a similar race and failure mode.
Instead, we change bpf_mem_alloc to allow registering destructor for
allocated elements, such that when they are returned to the allocator,
any special fields created while they were accessible to programs in the
mean time will be freed. If these values get reused, we do not free the
fields again before handing the element back. The special fields thus
may remain initialized while the map value sits in a free list.
When bpf_mem_alloc is retired in the future, a similar concept can be
introduced to kmalloc_nolock-backed kmem_cache, paired with the existing
idea of a constructor.
Note that the destructor registration happens in map_check_btf, after
the BTF record is populated and (at that point) avaiable for inspection
and duplication. Duplication is necessary since the freeing of embedded
bpf_mem_alloc can be decoupled from actual map lifetime due to logic
introduced to reduce the cost of rcu_barrier()s in mem alloc free path in
9f2c6e96c6 ("bpf: Optimize rcu_barrier usage between hash map and bpf_mem_alloc.").
As such, once all callbacks are done, we must also free the duplicated
record. To remove dependency on the bpf_map itself, also stash the key
size of the map to obtain value from htab_elem long after the map is
gone.
[0]: https://lore.kernel.org/bpf/20260216131341.1285427-1-mykyta.yatsenko5@gmail.com
Fixes: 14a324f6a6 ("bpf: Wire up freeing of referenced kptr")
Fixes: 1bfbc267ec ("bpf: Enable bpf_timer and bpf_wq in any context")
Reported-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: syzbot@syzkaller.appspotmail.com
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260227224806.646888-2-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
get_upper_ifindexes() iterates over all upper devices and writes their
indices into an array without checking bounds.
Also the callers assume that the max number of upper devices is
MAX_NEST_DEV and allocate excluded_devices[1+MAX_NEST_DEV] on the stack,
but that assumption is not correct and the number of upper devices could
be larger than MAX_NEST_DEV (e.g., many macvlans), causing a
stack-out-of-bounds write.
Add a max parameter to get_upper_ifindexes() to avoid the issue.
When there are too many upper devices, return -EOVERFLOW and abort the
redirect.
To reproduce, create more than MAX_NEST_DEV(8) macvlans on a device with
an XDP program attached using BPF_F_BROADCAST | BPF_F_EXCLUDE_INGRESS.
Then send a packet to the device to trigger the XDP redirect path.
Reported-by: syzbot+10cc7f13760b31bd2e61@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/698c4ce3.050a0220.340abe.000b.GAE@google.com/T/
Fixes: aeea1b86f9 ("bpf, devmap: Exclude XDP broadcast to master device")
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Kohei Enju <kohei@enjuk.jp>
Link: https://lore.kernel.org/r/20260225053506.4738-1-kohei@enjuk.jp
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Currently, tailcall count is incremented in the interpreter even when
tailcall fails due to non-existent prog. Fix this by holding off on
the tailcall count increment until after NULL check on the prog.
Suggested-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Link: https://lore.kernel.org/r/20260220062959.195101-1-hbathini@linux.ibm.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
For the following scenario:
struct tree_node {
struct bpf_refcount ref;
struct bpf_rb_node node;
struct node_data __kptr * node_data;
u64 key;
};
This means node_data would have the type PTR_TO_BTF_ID | MEM_ALLOC |
NON_OWN_REF | MEM_RCU.
When traversing an rbtree using bpf_rbtree_left/right, if we need to
use bpf_kptr_xchg to read the __kptr pointer, we still need to follow
the remove-read-add sequence.
This patch allows us to use bpf_kptr_xchg to directly read the __kptr
pointer without any prior operations.
Signed-off-by: Kaitao Cheng <chengkaitao@kylinos.cn>
Signed-off-by: Feng Yang <yangfeng@kylinos.cn>
Link: https://lore.kernel.org/r/20260214124042.62229-5-pilgrimtao@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When traversing an rbtree using bpf_rbtree_left/right, if bpf_kptr_xchg
is used to access the __kptr pointer contained in a node, it currently
requires first removing the node with bpf_rbtree_remove and clearing the
NON_OWN_REF flag, then re-adding the node to the original rbtree with
bpf_rbtree_add after usage. This process significantly degrades rbtree
traversal performance. The patch enables accessing __kptr pointers with
the NON_OWN_REF flag set while holding the lock, eliminating the need
for this remove-read-add sequence.
Signed-off-by: Kaitao Cheng <chengkaitao@kylinos.cn>
Signed-off-by: Feng Yang <yangfeng@kylinos.cn>
Link: https://lore.kernel.org/r/20260214124042.62229-3-pilgrimtao@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
For the following scenario:
struct tree_node {
struct bpf_rb_node node;
struct request __kptr *req;
u64 key;
};
struct bpf_rb_root tree_root __contains(tree_node, node);
struct bpf_spin_lock tree_lock;
If we need to traverse all nodes in the rbtree, retrieve the __kptr
pointer from each node, and read kernel data from the referenced
object, using bpf_kptr_xchg appears unavoidable.
This patch skips the BPF verifier checks for bpf_kptr_xchg when
called while holding a lock.
Signed-off-by: Kaitao Cheng <chengkaitao@kylinos.cn>
Link: https://lore.kernel.org/r/20260214124042.62229-2-pilgrimtao@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
