Skip to contentSkip to conversation

libbpf: Replace strncpy() with strnlen()+memcpy() in skel_map_create()

PATCH Superseded Kees Cook 3 participants 4 messages Mar 24

Subsystems (2)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Kees Cook

author
#0 lore raw

PATCH v2 1/1 libbpf: Replace strncpy() with strnlen()+memcpy() in skel_map_create()

Replace the deprecated[1] strncpy() with strnlen() on the source followed by memcpy(). Normally strscpy() would be used in this case, but skel_internal.h is shared between kernel and userspace tools, and strscpy() is not available in the userspace build context.

The source map_name is a NUL-terminated C string (the only caller passes a 12 character string literal). The destination attr.map_name is char[BPF_OBJ_NAME_LEN] (16 bytes) in union bpf_attr, passed to the bpf() syscall. The kernel's bpf_obj_name_cpy() requires a NUL terminator within the 16-byte field, rejecting names that use all 16 bytes. Valid names are therefore at most 15 characters.

The attr is pre-zeroed with memset() at the top of the function, so the byte at position 15 is always NUL. The copy is bounded to sizeof(attr.map_name) - 1 (15 bytes) to guarantee NUL-termination is preserved. This is safe because the kernel would reject a 16-byte unterminated name anyway, and the only in-tree caller passes "__loader.map" (12 characters).

While the original strncpy() would have copied a full 16 bytes from an overlong name (producing an unterminated field that the syscall rejects), but this shouldn't be a reachable state. Forcing truncation to 15 bytes, however, seemed to induce a (unrelated?) BPF selftest failure:
https://github.com/kernel-patches/bpf/actions/runs/23472955268/job/68300440546

Allow the literal to exceed 15 characters and exactly reproduce the original strncpy() behavior (potentially lacking a NUL termination).

---
v2: don't force truncation
v1: https://lore.kernel.org/lkml/[email protected]/
Cc: Andrii Nakryiko <[email protected]>
Cc: Eduard Zingerman <[email protected]>
Cc: Alexei Starovoitov <[email protected]>
Cc: Daniel Borkmann <[email protected]>
Cc: Martin KaFai Lau <[email protected]>
Cc: Song Liu <[email protected]>
Cc: Yonghong Song <[email protected]>
Cc: John Fastabend <[email protected]>
Cc: KP Singh <[email protected]>
Cc: Stanislav Fomichev <[email protected]>
Cc: Hao Luo <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: <[email protected]>

---
 tools/lib/bpf/skel_internal.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tools/lib/bpf/skel_internal.h +2 -1
index 6a8f5c7a02eb..8702d6612978 100644
@@ -243,7 +243,8 @@ static inline int skel_map_create(enum bpf_map_type map_type,
 	attr.excl_prog_hash = (unsigned long) excl_prog_hash;
 	attr.excl_prog_hash_size = excl_prog_hash_sz;
 
-	strncpy(attr.map_name, map_name, sizeof(attr.map_name));
+	memcpy(attr.map_name, map_name,
+	       strnlen(map_name, sizeof(attr.map_name)));
 	attr.key_size = key_size;
 	attr.value_size = value_size;
 	attr.max_entries = max_entries;
Signature · 1 line
2.34.1
#1 lore raw
Quoted 12 lines · Kees Cook

Replace the deprecated[1] strncpy() with strnlen() on the source followed by memcpy(). Normally strscpy() would be used in this case, but skel_internal.h is shared between kernel and userspace tools, and strscpy() is not available in the userspace build context.

The source map_name is a NUL-terminated C string (the only caller passes a 12 character string literal). The destination attr.map_name is char[BPF_OBJ_NAME_LEN] (16 bytes) in union bpf_attr, passed to the bpf() syscall. The kernel's bpf_obj_name_cpy() requires a NUL terminator within the 16-byte field, rejecting names that use all 16 bytes. Valid names are therefore at most 15 characters.

Quoted 3 lines

The attr is pre-zeroed with memset() at the top of the function, so the byte at position 15 is always NUL. The copy is bounded to sizeof(attr.map_name) - 1 (15 bytes) to guarantee NUL-termination is

hm, but this version no longer does that, right?

jirka

Quoted 52 lines

preserved. This is safe because the kernel would reject a 16-byte unterminated name anyway, and the only in-tree caller passes "__loader.map" (12 characters).

While the original strncpy() would have copied a full 16 bytes from an overlong name (producing an unterminated field that the syscall rejects), but this shouldn't be a reachable state. Forcing truncation to 15 bytes, however, seemed to induce a (unrelated?) BPF selftest failure:
https://github.com/kernel-patches/bpf/actions/runs/23472955268/job/68300440546

Allow the literal to exceed 15 characters and exactly reproduce the original strncpy() behavior (potentially lacking a NUL termination).

---
v2: don't force truncation
v1: https://lore.kernel.org/lkml/[email protected]/
Cc: Andrii Nakryiko <[email protected]>
Cc: Eduard Zingerman <[email protected]>
Cc: Alexei Starovoitov <[email protected]>
Cc: Daniel Borkmann <[email protected]>
Cc: Martin KaFai Lau <[email protected]>
Cc: Song Liu <[email protected]>
Cc: Yonghong Song <[email protected]>
Cc: John Fastabend <[email protected]>
Cc: KP Singh <[email protected]>
Cc: Stanislav Fomichev <[email protected]>
Cc: Hao Luo <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: <[email protected]>

---
 tools/lib/bpf/skel_internal.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tools/lib/bpf/skel_internal.h +2 -1
index 6a8f5c7a02eb..8702d6612978 100644
@@ -243,7 +243,8 @@ static inline int skel_map_create(enum bpf_map_type map_type,
 	attr.excl_prog_hash = (unsigned long) excl_prog_hash;
 	attr.excl_prog_hash_size = excl_prog_hash_sz;
 
-	strncpy(attr.map_name, map_name, sizeof(attr.map_name));
+	memcpy(attr.map_name, map_name,
+	       strnlen(map_name, sizeof(attr.map_name)));
 	attr.key_size = key_size;
 	attr.value_size = value_size;
 	attr.max_entries = max_entries;
Signature · 1 line
2.34.1
#3 lore raw
In reply to Kees Cook
Quoted 56 lines · Kees Cook

Replace the deprecated[1] strncpy() with strnlen() on the source followed by memcpy(). Normally strscpy() would be used in this case, but skel_internal.h is shared between kernel and userspace tools, and strscpy() is not available in the userspace build context.

The source map_name is a NUL-terminated C string (the only caller passes a 12 character string literal). The destination attr.map_name is char[BPF_OBJ_NAME_LEN] (16 bytes) in union bpf_attr, passed to the bpf() syscall. The kernel's bpf_obj_name_cpy() requires a NUL terminator within the 16-byte field, rejecting names that use all 16 bytes. Valid names are therefore at most 15 characters.

The attr is pre-zeroed with memset() at the top of the function, so the byte at position 15 is always NUL. The copy is bounded to sizeof(attr.map_name) - 1 (15 bytes) to guarantee NUL-termination is preserved. This is safe because the kernel would reject a 16-byte unterminated name anyway, and the only in-tree caller passes "__loader.map" (12 characters).

While the original strncpy() would have copied a full 16 bytes from an overlong name (producing an unterminated field that the syscall rejects), but this shouldn't be a reachable state. Forcing truncation to 15 bytes, however, seemed to induce a (unrelated?) BPF selftest failure:
https://github.com/kernel-patches/bpf/actions/runs/23472955268/job/68300440546

Allow the literal to exceed 15 characters and exactly reproduce the original strncpy() behavior (potentially lacking a NUL termination).

---
v2: don't force truncation
v1: https://lore.kernel.org/lkml/[email protected]/
Cc: Andrii Nakryiko <[email protected]>
Cc: Eduard Zingerman <[email protected]>
Cc: Alexei Starovoitov <[email protected]>
Cc: Daniel Borkmann <[email protected]>
Cc: Martin KaFai Lau <[email protected]>
Cc: Song Liu <[email protected]>
Cc: Yonghong Song <[email protected]>
Cc: John Fastabend <[email protected]>
Cc: KP Singh <[email protected]>
Cc: Stanislav Fomichev <[email protected]>
Cc: Hao Luo <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: <[email protected]>

---
 tools/lib/bpf/skel_internal.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tools/lib/bpf/skel_internal.h +0 -0
index 6a8f5c7a02eb..8702d6612978 100644

tools/lib/bpf/skel_internal.h static inline int skel_map_create(enum bpf_map_type map_type, · line 247

Quoted 6 lines
tools/lib/bpf/skel_internal.h +2 -1
@@ -243,7 +243,8 @@ static inline int skel_map_create(enum bpf_map_type map_type,
        attr.excl_prog_hash = (unsigned long) excl_prog_hash;
        attr.excl_prog_hash_size = excl_prog_hash_sz;
 
-       strncpy(attr.map_name, map_name, sizeof(attr.map_name));
+       memcpy(attr.map_name, map_name,
+              strnlen(map_name, sizeof(attr.map_name)));

wont-fix.

pw-bot: cr

author
#2 lore raw
In reply to Jiri Olsa
Quoted 13 lines · Jiri Olsa
Quoted 12 lines · Kees Cook

Replace the deprecated[1] strncpy() with strnlen() on the source followed by memcpy(). Normally strscpy() would be used in this case, but skel_internal.h is shared between kernel and userspace tools, and strscpy() is not available in the userspace build context.

The source map_name is a NUL-terminated C string (the only caller passes a 12 character string literal). The destination attr.map_name is char[BPF_OBJ_NAME_LEN] (16 bytes) in union bpf_attr, passed to the bpf() syscall. The kernel's bpf_obj_name_cpy() requires a NUL terminator within the 16-byte field, rejecting names that use all 16 bytes. Valid names are therefore at most 15 characters.

Quoted 5 lines
Quoted 3 lines

The attr is pre-zeroed with memset() at the top of the function, so the byte at position 15 is always NUL. The copy is bounded to sizeof(attr.map_name) - 1 (15 bytes) to guarantee NUL-termination is

hm, but this version no longer does that, right?

Arg. Yes, correct. I'll fix the commit log for v3. And yes, the bpf test suite reported that v2 shows no test failures any more.

(Why does this need an unterminated string internally here when the syscall refuses to build one?)

Signature · 1 line
Kees Cook