Skip to contentSkip to conversation

libbpf: Return -EINVAL from btf__align_of() for unknown type ID

PATCH New Abhinav Madake 3 participants 4 messages Sep 19

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

PATCH 1/1 libbpf: Return -EINVAL from btf__align_of() for unknown type ID

btf__align_of() passes the result of btf__type_by_id() straight to btf_kind() without checking for NULL, so an unknown type ID dereferences a NULL pointer instead of returning an error like the other btf__*() lookups do.

BTF parsed from raw data cannot trigger this, since btf_sanity_check() rejects dangling references. BTF built with the btf__add_*() API can:
those only check the ID range, so a forward reference to a type that was never added, followed by btf__align_of() on it (directly, or through the recursive struct/array/typedef cases), crashes.

Check for NULL and return -EINVAL, matching btf__type_by_id().

---
 tools/lib/bpf/btf.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
tools/lib/bpf/btf.c +5 -1
index 908bd3442..d15bc09ac 100644
@@ -986,8 +986,12 @@ done:
 int btf__align_of(const struct btf *btf, __u32 id)
 {
 	const struct btf_type *t = btf__type_by_id(btf, id);
-	__u16 kind = btf_kind(t);
+	__u16 kind;
 
+	if (!t)
+		return libbpf_err(-EINVAL);
+
+	kind = btf_kind(t);
 	switch (kind) {
 	case BTF_KIND_INT:
 	case BTF_KIND_ENUM:
Signature · 1 line
2.47.0
#1 lore raw
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] bpf: btf__align_of() negative error breaks btf_dump.c --

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] bpf: btf__align_of() negative error breaks btf_dump.c
--

commit d6024f50b48c7fc7873ce247e9f420706729d50a
Author: Abhinav Madake <[email protected]>

libbpf: Return -EINVAL from btf__align_of() for unknown type ID

btf__align_of() passes the result of btf__type_by_id() straight to btf_kind() without checking for NULL, so an unknown type ID dereferences a NULL pointer instead of returning an error like the other btf__*() lookups do.

BTF parsed from raw data cannot trigger this, since btf_sanity_check() rejects dangling references. BTF built with the btf__add_*() API can:
those only check the ID range, so a forward reference to a type that was never added, followed by btf__align_of() on it (directly, or through the recursive struct/array/typedef cases), crashes.

Check for NULL and return -EINVAL, matching btf__type_by_id().

tools/lib/bpf/btf.c __s64 btf__resolve_size(const struct btf *btf, __u32 type_id) · line 992

Quoted 13 lines
tools/lib/bpf/btf.c +3 -1
index 908bd344229d5..d15bc09ac53de 100644
@@ -986,8 +986,12 @@ __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
 int btf__align_of(const struct btf *btf, __u32 id)
 {
 	const struct btf_type *t = btf__type_by_id(btf, id);
-	__u16 kind = btf_kind(t);
+	__u16 kind;
 
+	if (!t)
+		return libbpf_err(-EINVAL);

[Severity: Medium]
Will this negative return value break callers in tools/lib/bpf/btf_dump.c that expect 0 on error?

When btf_dump.c processes invalid BTF types, this -EINVAL return bypasses checks for 0, leading to modulo operations with a negative number promoted to a large unsigned integer.

In ptr_is_aligned(), the negative alignment bypasses the check for 0:

tools/lib/bpf/btf_dump.c:ptr_is_aligned() {
	int alignment = btf__align_of(btf, type_id);

	if (alignment == 0)
		return false;

	return ((uintptr_t)data) % alignment == 0;
}

In btf_is_struct_packed(), the error bypasses the false-check (which expects 0), causing an invalid modulo operation with 8 * -22 promoted to unsigned:

tools/lib/bpf/btf_dump.c:btf_is_struct_packed() {
    ...
		align = btf__align_of(btf, m->type);
		bit_sz = btf_member_bitfield_size(t, i);
		if (align && bit_sz == 0 && m->offset % (8 * align) != 0)
			return true;
    ...
}

Should the callers in btf_dump.c be updated to handle negative return values?

Quoted 5 lines
tools/lib/bpf/btf.c +2 -0
+
+	kind = btf_kind(t);
 	switch (kind) {
 	case BTF_KIND_INT:
 	case BTF_KIND_ENUM:
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#2 lore raw

PATCH v2 1/1 libbpf: Check for unknown type ID in btf__align_of()

In reply to Abhinav Madake

btf__align_of() passes the result of btf__type_by_id() to btf_kind() without a NULL check, so an unknown type ID crashes. Parsed BTF cannot hit this since btf_sanity_check() rejects dangling references, but BTF built with btf__add_*() can, as those only check the ID range.

Return 0 with errno set to EINVAL, like the unsupported-kind case.

---
v2: return 0 with errno instead of -EINVAL, callers test for 0.

 tools/lib/bpf/btf.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
tools/lib/bpf/btf.c +5 -1
index 908bd3442..6bfbc7a65 100644
@@ -986,8 +986,12 @@ done:
 int btf__align_of(const struct btf *btf, __u32 id)
 {
 	const struct btf_type *t = btf__type_by_id(btf, id);
-	__u16 kind = btf_kind(t);
+	__u16 kind;
 
+	if (!t)
+		return errno = EINVAL, 0;
+
+	kind = btf_kind(t);
 	switch (kind) {
 	case BTF_KIND_INT:
 	case BTF_KIND_ENUM:
Signature · 1 line
2.47.0
#3 lore raw

Re: [PATCH bpf-next v2] libbpf: Check for unknown type ID in btf__align_of()

Quoted 20 lines · Abhinav Madake

btf__align_of() passes the result of btf__type_by_id() to btf_kind() without a NULL check, so an unknown type ID crashes. Parsed BTF cannot hit this since btf_sanity_check() rejects dangling references, but BTF built with btf__add_*() can, as those only check the ID range.

Return 0 with errno set to EINVAL, like the unsupported-kind case.

---
v2: return 0 with errno instead of -EINVAL, callers test for 0.

 tools/lib/bpf/btf.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
tools/lib/bpf/btf.c +0 -0
index 908bd3442..6bfbc7a65 100644

tools/lib/bpf/btf.c done: · line 992

Quoted 8 lines
tools/lib/bpf/btf.c +3 -1
@@ -986,8 +986,12 @@ done:
 int btf__align_of(const struct btf *btf, __u32 id)
 {
 	const struct btf_type *t = btf__type_by_id(btf, id);
-	__u16 kind = btf_kind(t);
+	__u16 kind;
 
+	if (!t)
+		return errno = EINVAL, 0;

We don't do defensive programming in libbpf.
input is trusted.

pw-bot: cr