diff options
author | Jeff Layton <jlayton@redhat.com> | 2014-01-27 19:46:15 +0100 |
---|---|---|
committer | Trond Myklebust <trond.myklebust@primarydata.com> | 2014-01-27 21:35:56 +0100 |
commit | d529ef83c355f97027ff85298a9709fe06216a66 (patch) | |
tree | b13587538729686d8f366ab734782de0aeba1fa7 /net | |
parent | nfs: handle servers that support only ALLOW ACE type. (diff) | |
download | linux-d529ef83c355f97027ff85298a9709fe06216a66.tar.xz linux-d529ef83c355f97027ff85298a9709fe06216a66.zip |
NFS: fix the handling of NFS_INO_INVALID_DATA flag in nfs_revalidate_mapping
There is a possible race in how the nfs_invalidate_mapping function is
handled. Currently, we go and invalidate the pages in the file and then
clear NFS_INO_INVALID_DATA.
The problem is that it's possible for a stale page to creep into the
mapping after the page was invalidated (i.e., via readahead). If another
writer comes along and sets the flag after that happens but before
invalidate_inode_pages2 returns then we could clear the flag
without the cache having been properly invalidated.
So, we must clear the flag first and then invalidate the pages. Doing
this however, opens another race:
It's possible to have two concurrent read() calls that end up in
nfs_revalidate_mapping at the same time. The first one clears the
NFS_INO_INVALID_DATA flag and then goes to call nfs_invalidate_mapping.
Just before calling that though, the other task races in, checks the
flag and finds it cleared. At that point, it trusts that the mapping is
good and gets the lock on the page, allowing the read() to be satisfied
from the cache even though the data is no longer valid.
These effects are easily manifested by running diotest3 from the LTP
test suite on NFS. That program does a series of DIO writes and buffered
reads. The operations are serialized and page-aligned but the existing
code fails the test since it occasionally allows a read to come out of
the cache incorrectly. While mixing direct and buffered I/O isn't
recommended, I believe it's possible to hit this in other ways that just
use buffered I/O, though that situation is much harder to reproduce.
The problem is that the checking/clearing of that flag and the
invalidation of the mapping really need to be atomic. Fix this by
serializing concurrent invalidations with a bitlock.
At the same time, we also need to allow other places that check
NFS_INO_INVALID_DATA to check whether we might be in the middle of
invalidating the file, so fix up a couple of places that do that
to look for the new NFS_INO_INVALIDATING flag.
Doing this requires us to be careful not to set the bitlock
unnecessarily, so this code only does that if it believes it will
be doing an invalidation.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Diffstat (limited to 'net')
0 files changed, 0 insertions, 0 deletions