Hi steve,
Recently I've started running LTP tests with mfsymlinks mount option.
The LTP test "fs_racer" triggers the cifsFileInfo leak when doing
cifs_atomic_open on a symlink file.
After testing and umounting cifs mount, there are some files exist in
/proc/fs/cifs/open_files as,
#cat /proc/fs/cifs/open_files
# Version:1
# Format:
# <tree id> <persistent fid> <flags> <count> <pid> <uid> <filename>
0x2 0x94c08 0x8401 1 1554873 0 11
0x2 0x94b5f 0x8401 1 1554808 0 7
0x2 0x947cf 0x8401 1 1554365 0 0
0x2 0x9470a 0x8401 1 1554339 0 0
and the tcp socket also left,
# netstat -an |grep 445
tcp 0 0 192.168.123.123:38050 192.168.123.124:445
ESTABLISHED
and the cifsd left too.
# ps -ajx |grep cifsd
2 34030 0 0 ? -1 S 0 2:20 [cifsd]
I found the fs_racer cause cifs doing cifs_atomic_open on symlink file,
but the symlink file does not contain a valid i_fop setting, it means,
the cifsFileInfo allocated for the symlink file never be un-referenced.
I'd wanna know, is cifs_atomic_open on symlink file allowed?
If allowed, a cifs_close for symlink file is needed.
Otherwise, cifs_atomic_open needs return error for symlink file.
Any comments are welcome.
Thanks,
Kinglong Mee
