OSDN Git Service

bpf: set inner_map_meta->spin_lock_off correctly
authorYonghong Song <yhs@fb.com>
Wed, 27 Feb 2019 21:22:56 +0000 (13:22 -0800)
committerAlexei Starovoitov <ast@kernel.org>
Thu, 28 Feb 2019 01:03:13 +0000 (17:03 -0800)
commita115d0ed7201a5904c084ae6f07913fe2b9396a6
tree95fefb6e10dc05e5c1114ff4c7e5b4f9ee2e903a
parentd2e614cb0795d935aee879e47aab231247274f13
bpf: set inner_map_meta->spin_lock_off correctly

Commit d83525ca62cf ("bpf: introduce bpf_spin_lock")
introduced bpf_spin_lock and the field spin_lock_off
in kernel internal structure bpf_map has the following
meaning:
  >=0 valid offset, <0 error

For every map created, the kernel will ensure
spin_lock_off has correct value.

Currently, bpf_map->spin_lock_off is not copied
from the inner map to the map_in_map inner_map_meta
during a map_in_map type map creation, so
inner_map_meta->spin_lock_off = 0.
This will give verifier wrong information that
inner_map has bpf_spin_lock and the bpf_spin_lock
is defined at offset 0. An access to offset 0
of a value pointer will trigger the following error:
   bpf_spin_lock cannot be accessed directly by load/store

This patch fixed the issue by copy inner map's spin_lock_off
value to inner_map_meta->spin_lock_off.

Fixes: d83525ca62cf ("bpf: introduce bpf_spin_lock")
Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
kernel/bpf/map_in_map.c