bpf: Iterate through all PT_NOTE sections when looking for build id
authorJiri Olsa <jolsa@kernel.org>
Wed, 12 Aug 2020 12:31:02 +0000 (14:31 +0200)
committerAlexei Starovoitov <ast@kernel.org>
Thu, 13 Aug 2020 01:14:49 +0000 (18:14 -0700)
Currently when we look for build id within bpf_get_stackid helper
call, we check the first NOTE section and we fail if build id is
not there.

However on some system (Fedora) there can be multiple NOTE sections
in binaries and build id data is not always the first one, like:

  $ readelf -a /usr/bin/ls
  ...
  [ 2] .note.gnu.propert NOTE             0000000000000338  00000338
       0000000000000020  0000000000000000   A       0     0     8358
  [ 3] .note.gnu.build-i NOTE             0000000000000358  00000358
       0000000000000024  0000000000000000   A       0     0     437c
  [ 4] .note.ABI-tag     NOTE             000000000000037c  0000037c
  ...

So the stack_map_get_build_id function will fail on build id retrieval
and fallback to BPF_STACK_BUILD_ID_IP.

This patch is changing the stack_map_get_build_id code to iterate
through all the NOTE sections and try to get build id data from
each of them.

When tracing on sched_switch tracepoint that does bpf_get_stackid
helper call kernel build, I can see about 60% increase of successful
build id retrieval. The rest seems fails on -EFAULT.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200812123102.20032-1-jolsa@kernel.org
kernel/bpf/stackmap.c

index 4fd830a..cfed0ac 100644 (file)
@@ -213,11 +213,13 @@ static int stack_map_get_build_id_32(void *page_addr,
 
        phdr = (Elf32_Phdr *)(page_addr + sizeof(Elf32_Ehdr));
 
 
        phdr = (Elf32_Phdr *)(page_addr + sizeof(Elf32_Ehdr));
 
-       for (i = 0; i < ehdr->e_phnum; ++i)
-               if (phdr[i].p_type == PT_NOTE)
-                       return stack_map_parse_build_id(page_addr, build_id,
-                                       page_addr + phdr[i].p_offset,
-                                       phdr[i].p_filesz);
+       for (i = 0; i < ehdr->e_phnum; ++i) {
+               if (phdr[i].p_type == PT_NOTE &&
+                   !stack_map_parse_build_id(page_addr, build_id,
+                                             page_addr + phdr[i].p_offset,
+                                             phdr[i].p_filesz))
+                       return 0;
+       }
        return -EINVAL;
 }
 
        return -EINVAL;
 }
 
@@ -236,11 +238,13 @@ static int stack_map_get_build_id_64(void *page_addr,
 
        phdr = (Elf64_Phdr *)(page_addr + sizeof(Elf64_Ehdr));
 
 
        phdr = (Elf64_Phdr *)(page_addr + sizeof(Elf64_Ehdr));
 
-       for (i = 0; i < ehdr->e_phnum; ++i)
-               if (phdr[i].p_type == PT_NOTE)
-                       return stack_map_parse_build_id(page_addr, build_id,
-                                       page_addr + phdr[i].p_offset,
-                                       phdr[i].p_filesz);
+       for (i = 0; i < ehdr->e_phnum; ++i) {
+               if (phdr[i].p_type == PT_NOTE &&
+                   !stack_map_parse_build_id(page_addr, build_id,
+                                             page_addr + phdr[i].p_offset,
+                                             phdr[i].p_filesz))
+                       return 0;
+       }
        return -EINVAL;
 }
 
        return -EINVAL;
 }