We monitored an unexpected behavoir that bpffs is mounted on a same mount
point lots of times on some of our production envrionments. For example,
$ mount -t bpf
bpffs /sys/fs/bpf bpf rw,relatime 0 0
bpffs /sys/fs/bpf bpf rw,relatime 0 0
bpffs /sys/fs/bpf bpf rw,relatime 0 0
bpffs /sys/fs/bpf bpf rw,relatime 0 0
...
That was casued by a buggy user script which didn't check the mount
point correctly before mounting bpffs. But it also drives us to think more
about if it is okay to allow mounting bpffs on the same mount point
multiple times. After investigation we get the conclusion that it is bad
to allow that behavior, because it can cause unexpected issues, for
example it can break bpftool, which depends on the mount point to get
the pinned files.
Below is the test case wrt bpftool.
First, let's mount bpffs on /var/run/ltcp/bpf multiple times.
$ mount -t bpf
bpffs on /run/ltcp/bpf type bpf (rw,relatime)
bpffs on /run/ltcp/bpf type bpf (rw,relatime)
bpffs on /run/ltcp/bpf type bpf (rw,relatime)
After pinning some bpf progs on this mount point, let's check the pinned
files with bpftool,
$ bpftool prog list -f
87: sock_ops name bpf_sockmap tag a04f5eef06a7f555 gpl
loaded_at 2022-02-23T16:27:38+0800 uid 0
xlated 16B jited 15B memlock 4096B
pinned /run/ltcp/bpf/bpf_sockmap
pinned /run/ltcp/bpf/bpf_sockmap
pinned /run/ltcp/bpf/bpf_sockmap
btf_id 243
89: sk_msg name bpf_redir_proxy tag 57cd311f2e27366b gpl
loaded_at 2022-02-23T16:27:38+0800 uid 0
xlated 16B jited 18B memlock 4096B
pinned /run/ltcp/bpf/bpf_redir_proxy
pinned /run/ltcp/bpf/bpf_redir_proxy
pinned /run/ltcp/bpf/bpf_redir_proxy
btf_id 244
The same pinned file will be showed multiple times.
Finnally after mounting bpffs on the same mount point again, we can't
get the pinnned files via bpftool,
$ bpftool prog list -f
87: sock_ops name bpf_sockmap tag a04f5eef06a7f555 gpl
loaded_at 2022-02-23T16:27:38+0800 uid 0
xlated 16B jited 15B memlock 4096B
btf_id 243
89: sk_msg name bpf_redir_proxy tag 57cd311f2e27366b gpl
loaded_at 2022-02-23T16:27:38+0800 uid 0
xlated 16B jited 18B memlock 4096B
btf_id 244
We should better refuse to mount bpffs on the same mount point. Before
making this change, I also checked why it is allowed in the first place.
The related commits are commit e27f4a942a0e
("bpf: Use mount_nodev not mount_ns to mount the bpf filesystem") and
commit b2197755b263 ("bpf: add support for persistent maps/progs").
Unfortunately they didn't explain why it is allowed. But there should be
no use case which requires to mount bpffs on a same mount point multiple
times, so let's just refuse it.
Signed-off-by: Yafang Shao <laoar.shao@xxxxxxxxx>
Cc: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx>
Cc: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
---
kernel/bpf/inode.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/bpf/inode.c b/kernel/bpf/inode.c
index 4f841e16779e..58374db9376f 100644
--- a/kernel/bpf/inode.c
+++ b/kernel/bpf/inode.c
@@ -763,7 +763,7 @@ static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
static int bpf_get_tree(struct fs_context *fc)
{
- return get_tree_nodev(fc, bpf_fill_super);
+ return get_tree_single(fc, bpf_fill_super);
}
static void bpf_free_fc(struct fs_context *fc)
--
2.17.1
