Skip to contentSkip to conversation

bpf: Fix signedness bug

PATCH In review Ethan Tidmore 4 participants 6 messages Jun 6

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bpf: Fix signedness bug

The function acquire_reference() returns negative error codes and 'id' is an unsigned integer, so the check (id < 0) is always impossible.

Detected by Smatch:
kernel/bpf/verifier.c:13115 check_kfunc_call()
warn: unsigned 'id' is never less than zero.

Fixes 308c7a0ae8859 ("bpf: Refactor object relationship tracking and fix dynptr UAF bug")
---
 kernel/bpf/verifier.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
kernel/bpf/verifier.c +5 -3
index 8ed484cb1a8a..aa8f10fce071 100644
@@ -13111,9 +13111,11 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 		}
 		mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *));
 		if (is_kfunc_acquire(&meta)) {
-			id = acquire_reference(env, insn_idx, 0);
-			if (id < 0)
-				return id;
+			err = acquire_reference(env, insn_idx, 0);
+			if (err < 0)
+				return err;
+			id = err;
+
 			regs[BPF_REG_0].id = id;
 		} else if (is_rbtree_node_type(ptr_type) || is_list_node_type(ptr_type)) {
 			ref_set_non_owning(env, &regs[BPF_REG_0]);
Signature · 3 lines
Thanks,
ET
https://github.com/sponsors/ethantidmore
#1 lore raw
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] The commit message falsely claims that the variable 'id' is an unsigned integer, making the…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The commit message falsely claims that the variable 'id' is an unsigned integer, making the error check (id < 0) impossible. This claim is demonstrably incorrect; 'id' is declared as a signed integer, meaning the patch fixes a non-existent bug. --

commit c4f0a14bac5d07880856111e136127510f48012c
Author: Ethan Tidmore <[email protected]>

bpf: Fix signedness bug

This commit refactors the error handling for acquire_reference() by using a separate err variable, claiming that the id variable is unsigned and cannot be checked for negative error codes.

Quoted 6 lines

The function acquire_reference() returns negative error codes and 'id' is an unsigned integer, so the check (id < 0) is always impossible.

Detected by Smatch:
kernel/bpf/verifier.c:13115 check_kfunc_call()
warn: unsigned 'id' is never less than zero.

[Severity: Medium]
Is this description accurate? Looking at check_kfunc_call(), id appears to be explicitly declared as a signed int, not an unsigned integer.

Since acquire_reference() can return negative error codes like -ENOMEM, wouldn't the original check (id < 0) work correctly?

It looks like the Smatch warning might be a false positive, and the original code was already handling the error properly.

Quoted 7 lines
kernel/bpf/verifier.c +0 -0
index 935595138aa01..be528a0be13cd 100644
@@ -13167,9 +13167,11 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 		}
 		mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *));

kernel/bpf/verifier.c static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, · line 13175

Quoted 10 lines
kernel/bpf/verifier.c +5 -3
 		if (is_kfunc_acquire(&meta)) {
-			id = acquire_reference(env, insn_idx, 0);
-			if (id < 0)
-				return id;
+			err = acquire_reference(env, insn_idx, 0);
+			if (err < 0)
+				return err;
+			id = err;
+
 			regs[BPF_REG_0].id = id;

[Severity: Medium]
Is this change necessary? If id is actually a signed int, the original code would have properly caught the negative error returns from acquire_reference() without needing to introduce the err variable.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#2 lore raw
In reply to Ethan Tidmore
Is this commit message accurate? Looking at check_kfunc_call() in kernel/bpf/verifier.c, the variable 'id' is declared as a signed int:
Quoted 7 lines

commit cc27f12c12ff54b8eace339c3e130317bf205277
Author: Ethan Tidmore <[email protected]>

bpf: Fix signedness bug

The function acquire_reference() returns negative error codes and 'id' is an unsigned integer, so the check (id < 0) is always impossible.

Is this commit message accurate? Looking at check_kfunc_call() in kernel/bpf/verifier.c, the variable 'id' is declared as a signed int:

    const struct btf_param *args;
    u32 i, nargs, ptr_type_id;
    struct btf *desc_btf;
    int id;

Since acquire_reference() can return negative error codes like -ENOMEM, wouldn't the original check (id < 0) work correctly?

The three other call sites of acquire_reference() in the verifier all use the same signed-int pattern without issue:

    verifier.c:mark_stack_slots_dynptr():
        id = acquire_reference(env, insn_idx, dynptr_id);
        if (id < 0)
            return id;

    verifier.c:mark_stack_slots_iter():
        id = acquire_reference(env, insn_idx, ITER_TYPE_TO_ID(iter_type));
        if (id < 0)
            return id;

    verifier.c:check_helper_call():
        id = acquire_reference(env, insn_idx, fn->arg_type[arg_num]);
        if (id < 0)
            return id;

It looks like the Smatch warning might be a false positive, and the original code was already handling the error properly.

Quoted 17 lines

Detected by Smatch:
kernel/bpf/verifier.c:13115 check_kfunc_call()
warn: unsigned 'id' is never less than zero.

Fixes 308c7a0ae8859 ("bpf: Refactor object relationship tracking and fix dynptr UAF bug")
kernel/bpf/verifier.c +0 -1
index 935595138aa0..be528a0be13c 100644
@@ -13167,9 +13167,11 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 		}
 		mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *));
 		if (is_kfunc_acquire(&meta)) {
-			id = acquire_reference(env, insn_idx, 0);

kernel/bpf/verifier.c static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, · line 13177

Quoted 10 lines
kernel/bpf/verifier.c +5 -2
-			if (id < 0)
-				return id;
+			err = acquire_reference(env, insn_idx, 0);
+			if (err < 0)
+				return err;
+			id = err;
+
 			regs[BPF_REG_0].id = id;
 		} else if (is_rbtree_node_type(ptr_type) || is_list_node_type(ptr_type)) {
 			ref_set_non_owning(env, &regs[BPF_REG_0]);

The code change itself is behavior-equivalent since 'err' is also a signed int, so the patch doesn't introduce a runtime regression. But if the premise that 'id' is unsigned is wrong, should the commit message be corrected?

---
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/27050796130

#3 lore raw
Quoted 9 lines · On Fri, Jun 5, 2026 at 8:25 PM <[email protected]> wrote:
Quoted 8 lines

commit cc27f12c12ff54b8eace339c3e130317bf205277
Author: Ethan Tidmore <[email protected]>

bpf: Fix signedness bug

The function acquire_reference() returns negative error codes and 'id' is an unsigned integer, so the check (id < 0) is always impossible.

Quoted 7 lines

Is this commit message accurate? Looking at check_kfunc_call() in kernel/bpf/verifier.c, the variable 'id' is declared as a signed int:

    const struct btf_param *args;
    u32 i, nargs, ptr_type_id;
    struct btf *desc_btf;
    int id;

Hi,

As pointed out by the bot, the fix to the issue has been merged as part of the following patchset.
https://lore.kernel.org/bpf/[email protected]/

Thanks,
Amery

Quoted 63 lines

Since acquire_reference() can return negative error codes like -ENOMEM, wouldn't the original check (id < 0) work correctly?

The three other call sites of acquire_reference() in the verifier all use the same signed-int pattern without issue:

    verifier.c:mark_stack_slots_dynptr():
        id = acquire_reference(env, insn_idx, dynptr_id);
        if (id < 0)
            return id;

    verifier.c:mark_stack_slots_iter():
        id = acquire_reference(env, insn_idx, ITER_TYPE_TO_ID(iter_type));
        if (id < 0)
            return id;

    verifier.c:check_helper_call():
        id = acquire_reference(env, insn_idx, fn->arg_type[arg_num]);
        if (id < 0)
            return id;

It looks like the Smatch warning might be a false positive, and the original code was already handling the error properly.

Quoted 17 lines

Detected by Smatch:
kernel/bpf/verifier.c:13115 check_kfunc_call()
warn: unsigned 'id' is never less than zero.

Fixes 308c7a0ae8859 ("bpf: Refactor object relationship tracking and fix dynptr UAF bug")
kernel/bpf/verifier.c +0 -1
index 935595138aa0..be528a0be13c 100644
@@ -13167,9 +13167,11 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
              }
              mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *));
              if (is_kfunc_acquire(&meta)) {
-                     id = acquire_reference(env, insn_idx, 0);

kernel/bpf/verifier.c static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, · line 13177

Quoted 10 lines
kernel/bpf/verifier.c +5 -2
-                     if (id < 0)
-                             return id;
+                     err = acquire_reference(env, insn_idx, 0);
+                     if (err < 0)
+                             return err;
+                     id = err;
+
                      regs[BPF_REG_0].id = id;
              } else if (is_rbtree_node_type(ptr_type) || is_list_node_type(ptr_type)) {
                      ref_set_non_owning(env, &regs[BPF_REG_0]);

The code change itself is behavior-equivalent since 'err' is also a signed int, so the patch doesn't introduce a runtime regression. But if the premise that 'id' is unsigned is wrong, should the commit message be corrected?

---
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/27050796130

#4 lore raw
In reply to [email protected]
Quoted 8 lines · bot+bpf-ci
Quoted 7 lines

commit cc27f12c12ff54b8eace339c3e130317bf205277
Author: Ethan Tidmore <[email protected]>

bpf: Fix signedness bug

The function acquire_reference() returns negative error codes and 'id' is an unsigned integer, so the check (id < 0) is always impossible.

Quoted 7 lines

Is this commit message accurate? Looking at check_kfunc_call() in kernel/bpf/verifier.c, the variable 'id' is declared as a signed int:

    const struct btf_param *args;
    u32 i, nargs, ptr_type_id;
    struct btf *desc_btf;
    int id;

No, 'id' is assigned as a u32:

	u32 i, nargs, ptr_type_id, id;

--
Thanks,
ET
https://github.com/sponsors/ethantidmore

#5 lore raw
Quoted 13 lines · Ethan Tidmore
Quoted 11 lines · bot+bpf-ci
Quoted 7 lines

commit cc27f12c12ff54b8eace339c3e130317bf205277
Author: Ethan Tidmore <[email protected]>

bpf: Fix signedness bug

The function acquire_reference() returns negative error codes and 'id' is an unsigned integer, so the check (id < 0) is always impossible.

Is this commit message accurate? Looking at check_kfunc_call() in kernel/bpf/verifier.c, the variable 'id' is declared as a signed int:

Quoted 6 lines
Quoted 4 lines
    const struct btf_param *args;
    u32 i, nargs, ptr_type_id;
    struct btf *desc_btf;
    int id;

No, 'id' is assigned as a u32:

You might want to pull bpf-next and check again then.

Quoted 7 lines
        u32 i, nargs, ptr_type_id, id;

--
Thanks,
ET
https://github.com/sponsors/ethantidmore