I've been trying to hunt down the bug reported here:
https://bugzilla.linux-nfs.org/show_bug.cgi?id=394
I've not been able to reliably reproduce that, but patches #2 and #3 may
help. The issue I think may be in the management of the sentinel
reference. Responsibility for putting that reference belongs to the
task that clears the HASHED flag, so we need to take care not to put
that ref until we've successfully cleared it.
Hopefully, this will make it a bit more consistent (and close some other
potential races as well).
Jeff Layton (3):
nfsd: nfsd_do_file_acquire should hold rcu_read_lock while getting
refs
nfsd: fix potential race in nfsd_file_close
nfsd: fix nfsd_file_unhash_and_dispose
fs/nfsd/filecache.c | 53 ++++++++++++++++++---------------------------
1 file changed, 21 insertions(+), 32 deletions(-)
--
2.37.3
