Skip to contentSkip to conversation

libbpf: Memory error detected by Valgrind

DISCUSSION Alex 2 participants 2 messages Dec 31, 2021 – Jan 6, 2022 BPF

lore
Share & tools
lore

Alex

author
#0 lore raw

With valgrind I discovered the following in v0.6.1:

```
==923672== Memcheck, a memory error detector
==923672== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==923672== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
==923672== Command: ...
==923672==
==923672== Conditional jump or move depends on uninitialised value(s)
==923672==    at 0x46707C: strlen (in ...)
==923672==    by 0x466F4D: strdup (in ...)
==923672==    by 0x41DE90: internal_map_name (libbpf.c:1521)
==923672==    by 0x41DE90: bpf_object__init_internal_map (libbpf.c:1540)
==923672==    by 0x4200B0: bpf_object__init_global_data_maps (libbpf.c:1601)
==923672==    by 0x4266B5: bpf_object__init_maps (libbpf.c:2601)
==923672==    by 0x4266B5: __bpf_object__open.part.0 (libbpf.c:6937)
==923672==    by 0x429609: __bpf_object__open (libbpf.c:6885)
==923672==    by 0x429609: bpf_object__open_mem (libbpf.c:6999)
==923672==    by 0x4315C0: bpf_object__open_skeleton (libbpf.c:11529)
```

I believe map_name in internal_map_name of src/libbpf.c is the culprit:

char map_name[BPF_OBJ_NAME_LEN], *p;

The issue disappears when I change this line to:

char map_name[BPF_OBJ_NAME_LEN] = {}, *p;

Should I submit a patch?

--
Alex

#1 lore raw
Quoted 24 lines · Alex

With valgrind I discovered the following in v0.6.1:

```
==923672== Memcheck, a memory error detector
==923672== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==923672== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
==923672== Command: ...
==923672==
==923672== Conditional jump or move depends on uninitialised value(s)
==923672==    at 0x46707C: strlen (in ...)
==923672==    by 0x466F4D: strdup (in ...)
==923672==    by 0x41DE90: internal_map_name (libbpf.c:1521)
==923672==    by 0x41DE90: bpf_object__init_internal_map (libbpf.c:1540)
==923672==    by 0x4200B0: bpf_object__init_global_data_maps (libbpf.c:1601)
==923672==    by 0x4266B5: bpf_object__init_maps (libbpf.c:2601)
==923672==    by 0x4266B5: __bpf_object__open.part.0 (libbpf.c:6937)
==923672==    by 0x429609: __bpf_object__open (libbpf.c:6885)
==923672==    by 0x429609: bpf_object__open_mem (libbpf.c:6999)
==923672==    by 0x4315C0: bpf_object__open_skeleton (libbpf.c:11529)
```

I believe map_name in internal_map_name of src/libbpf.c is the culprit:

Quoted 7 lines

char map_name[BPF_OBJ_NAME_LEN], *p;

The issue disappears when I change this line to:

char map_name[BPF_OBJ_NAME_LEN] = {}, *p;

Should I submit a patch?

I don't understand where the issue is, tbh. map_name is clearly initialized by snprintf() unconditionally, so by the time we do strdup(map_name) it's a properly initialized string. What am I missing?

Quoted 3 lines

--
Alex