Skip to contentSkip to conversation

bpf: Fix wrong frame passed to check_fastcall_stack_contract() in check_stack_read_fixed_off()

PATCH Superseded Ömer Mete Kaya 2 participants 2 messages Sep 27

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bpf: Fix wrong frame passed to check_fastcall_stack_contract() in check_stack_read_fixed_off()

check_stack_read_fixed_off() calls check_fastcall_stack_contract() with 'state' (the current frame, i.e. the callee) instead of 'reg_state' (the frame that owns the stack being read, i.e. the caller).

In a bpf2bpf call where a callee reads from a PTR_TO_STACK pointing into the caller's frame, 'reg_state' and 'state' are different:
'reg_state' is the caller's frame, while 'state' is the callee's. Passing 'state' causes the fastcall contract to be checked against the callee's subprog instead of the caller's, leaving the caller's fastcall region unprotected and potentially can cause an incorrect fastcall rewrite.

The three sibling functions are all correct:

check_stack_write_fixed_off  -> passes 'state'     (writes to curframe)
check_stack_write_var_off    -> passes 'state'     (writes to curframe)
check_stack_read_var_off     -> passes 'ptr_state' (reads from ptr's frame)

Fix check_stack_read_fixed_off() to match the read pattern and pass 'reg_state' instead of 'state'.

Fixes 5b5f51bff1b6 ("bpf: no_caller_saved_registers attribute for helper calls")
---
 kernel/bpf/verifier.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/bpf/verifier.c +1 -1
index 1cd92336309b..5c523a9bde6b 100644
@@ -4020,7 +4020,7 @@ static int check_stack_read_fixed_off(struct bpf_verifier_env *env,
 	reg = &reg_state->stack[spi].spilled_ptr;
 
 	mark_stack_slot_scratched(env, spi);
-	check_fastcall_stack_contract(env, state, env->insn_idx, off);
+	check_fastcall_stack_contract(env, reg_state, env->insn_idx, off);
 
 	/*
 	 * Refine the in-progress load record's origin to the source stack slot.

--
2.55.0

#1 lore raw
The explanation of why the write siblings are correct may not be accurate. The commit message says they pass 'state' because they "write to curframe", but looking at check_stack_write_fixed_off()…
Quoted 18 lines

commit ff6e19d1f0d8e4001693ac46c23ec7204912e5d2

Author: Ömer Mete Kaya <[email protected]>
Date:   Mon Sep 28 01:23:44 2026 +0300

bpf: Fix wrong frame passed to check_fastcall_stack_contract() in check_stack_read_fixed_off()

check_stack_read_fixed_off() calls check_fastcall_stack_contract() with 'state' (the current frame, i.e. the callee) instead of 'reg_state' (the frame that owns the stack being read, i.e. the caller).

In a bpf2bpf call where a callee reads from a PTR_TO_STACK pointing into the caller's frame, 'reg_state' and 'state' are different:
'reg_state' is the caller's frame, while 'state' is the callee's. Passing 'state' causes the fastcall contract to be checked against the callee's subprog instead of the caller's, leaving the caller's fastcall region unprotected and potentially can cause an incorrect fastcall rewrite.

Quoted 4 lines

The three sibling functions are all correct:
check_stack_write_fixed_off -> passes 'state' (writes to curframe)
check_stack_write_var_off -> passes 'state' (writes to curframe)
check_stack_read_var_off -> passes 'ptr_state' (reads from ptr's frame)

The explanation of why the write siblings are correct may not be accurate. The commit message says they pass 'state' because they "write to curframe", but looking at check_stack_write_fixed_off(), 'state' is actually computed as the frame the destination pointer refers to:

    struct bpf_func_state *state = bpf_func(env, reg);

which is cur->frame[reg->frameno], not necessarily the current frame. The function documents the parameter as "stack frame we're writing to" and compares it against the real current frame separately:

    cur = env->cur_state->frame[env->cur_state->curframe];
    ...
    if (state != cur && reg->type == PTR_TO_STACK) {
        verbose(env, "cannot spill pointers to stack into stack frame of the caller\n");

check_stack_write_var_off() likewise documents 'state' as "func where register points to".

The existing selftest bad_write_in_subprog in verifier_bpf_fastcall.c has a callee write through a caller stack pointer, with the comment "invalidates bpf_fastcall contract for caller". So the write paths already pass the pointer's frame, just like check_stack_read_var_off() passes ptr_state.

Could the explanation be updated to say the write helpers pass 'state' because it's the frame the pointer refers to, not because they write to curframe? The code change itself is correct, but the "(writes to curframe)" wording might mislead future readers about what these functions actually do.

kernel/bpf/verifier.c static int check_stack_read_fixed_off(struct bpf_verifier_env *env, · line 3887

Quoted 13 lines
kernel/bpf/verifier.c +1 -1
index 41b49c56e123..7dd8e32420fd 100644
@@ -3881,7 +3881,7 @@ static int check_stack_read_fixed_off(struct bpf_verifier_env *env,
 	reg = &reg_state->stack[spi].spilled_ptr;
 
 	mark_stack_slot_scratched(env, spi);
-	check_fastcall_stack_contract(env, state, env->insn_idx, off);
+	check_fastcall_stack_contract(env, reg_state, env->insn_idx, off);
 
 	/*
 	 * Refine the in-progress load record's origin to the source stack slot.

---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/36356167987