linux.git/fs/ceph/dir.c, branch v4.13

ceph: fix race in concurrent readdir

2017-07-17T12:54:59+00:00

For a large directory, program needs to issue multiple readdir
syscalls to get all dentries. When there are multiple programs
read the directory concurrently. Following sequence of events
can happen.

 - program calls readdir with pos = 2. ceph sends readdir request
   to mds. The reply contains N1 entries. ceph adds these N1 entries
   to readdir cache.
 - program calls readdir with pos = N1+2. The readdir is satisfied
   by the readdir cache, N2 entries are returned. (Other program
   calls readdir in the middle, which fills the cache)
 - program calls readdir with pos = N1+N2+2. ceph sends readdir
   request to mds. The reply contains N3 entries and it reaches
   directory end. ceph adds these N3 entries to the readdir cache
   and marks directory complete.

The second readdir call does not update fi->readdir_cache_idx.
ceph add the last N3 entries to wrong places.

Cc: stable@vger.kernel.org # v4.3+
Signed-off-by: "Yan, Zheng" 
Signed-off-by: Ilya Dryomov

ceph: choose readdir frag based on previous readdir reply

2017-05-04T07:19:24+00:00

The dirfragtree is lazily updated, it's not always accurate. Infinite
loops happens in following circumstance.

- client send request to read frag A
- frag A has been fragmented into frag B and C. So mds fills the reply
  with contents of frag B
- client wants to read next frag C. ceph_choose_frag(frag value of C)
  return frag A.

The fix is using previous readdir reply to calculate next readdir frag
when possible.

Signed-off-by: "Yan, Zheng" 
Signed-off-by: Ilya Dryomov

ceph: make seeky readdir more efficient

2017-05-04T07:19:20+00:00

Current cephfs client uses string to indicate start position of
readdir. The string is last entry of previous readdir reply.
This approach does not work for seeky readdir because we can
not easily convert the new postion to a string. For seeky readdir,
mds needs to return dentries from the beginning. Client keeps
retrying if the reply does not contain the dentry it wants.

In current version of ceph, mds sorts CDentry in its cache in
hash order. Client also uses dentry hash to compose dir postion.
For seeky readdir, if client passes the hash part of dir postion
to mds. mds can avoid replying useless dentries.

Signed-off-by: "Yan, Zheng" 
Signed-off-by: Ilya Dryomov

ceph: do a LOOKUP in d_revalidate instead of GETATTR

2017-02-20T11:16:10+00:00

In commit c3f4688a08f (ceph: don't set req->r_locked_dir in
ceph_d_revalidate), we changed the code to do a GETATTR instead of a
LOOKUP as the parent info isn't strictly necessary to revalidate the
dentry. What we missed there though is that in order to update the lease
on the dentry after revalidating it, we _do_ need parent info.

Change ceph_d_revalidate back to doing a LOOKUP instead of a GETATTR so
that we can get the parent info in order to update the lease from
ceph_fill_trace. Note that we set req->r_parent here, but we cannot set
the CEPH_MDS_R_PARENT_LOCKED flag as we can't guarantee that it is.

Signed-off-by: Jeff Layton 
Reviewed-by: Yan, Zheng 
Signed-off-by: Ilya Dryomov

ceph: add a new flag to indicate whether parent is locked

2017-02-20T11:16:08+00:00

struct ceph_mds_request has an r_locked_dir pointer, which is set to
indicate the parent inode and that its i_rwsem is locked.  In some
critical places, we need to be able to indicate the parent inode to the
request handling code, even when its i_rwsem may not be locked.

Most of the code that operates on r_locked_dir doesn't require that the
i_rwsem be locked. We only really need it to handle manipulation of the
dcache. The rest (filling of the inode, updating dentry leases, etc.)
already has its own locking.

Add a new r_req_flags bit that indicates whether the parent is locked
when doing the request, and rename the pointer to "r_parent". For now,
all the places that set r_parent also set this flag, but that will
change in a later patch.

Signed-off-by: Jeff Layton 
Reviewed-by: Yan, Zheng 
Signed-off-by: Ilya Dryomov

ceph: convert bools in ceph_mds_request to a new r_req_flags field

2017-02-20T11:16:08+00:00

Currently, we have a bunch of bool flags in struct ceph_mds_request. We
need more flags though, but each bool takes (at least) a byte. Those
add up over time.

Merge all of the existing bools in this struct into a single unsigned
long, and use the set/test/clear_bit macros to manipulate them. These
are atomic operations, but that is required here to prevent
load/modify/store races. The existing flags are protected by different
locks, so we can't rely on them for that purpose.

Signed-off-by: Jeff Layton 
Reviewed-by: Yan, Zheng 
Signed-off-by: Ilya Dryomov

ceph: cleanup ACCESS_ONCE -> READ_ONCE

2017-02-20T11:16:05+00:00

This removes the uses of ACCESS_ONCE in favor of READ_ONCE

Signed-off-by: Seraphime Kirkovski 
Signed-off-by: Yan, Zheng

ceph: fix endianness of getattr mask in ceph_d_revalidate

2017-01-18T16:58:45+00:00

sparse says:

    fs/ceph/dir.c:1248:50: warning: incorrect type in assignment (different base types)
    fs/ceph/dir.c:1248:50:    expected restricted __le32 [usertype] mask
    fs/ceph/dir.c:1248:50:    got int [signed] [assigned] mask

Fixes: 200fd27c8fa2 ("ceph: use lookup request to revalidate dentry")
Signed-off-by: Jeff Layton 
Reviewed-by: Sage Weil 
Signed-off-by: Ilya Dryomov

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

2016-12-16T18:24:44+00:00

Pull vfs updates from Al Viro:

 - more ->d_init() stuff (work.dcache)

 - pathname resolution cleanups (work.namei)

 - a few missing iov_iter primitives - copy_from_iter_full() and
   friends. Either copy the full requested amount, advance the iterator
   and return true, or fail, return false and do _not_ advance the
   iterator. Quite a few open-coded callers converted (and became more
   readable and harder to fuck up that way) (work.iov_iter)

 - several assorted patches, the big one being logfs removal

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  logfs: remove from tree
  vfs: fix put_compat_statfs64() does not handle errors
  namei: fold should_follow_link() with the step into not-followed link
  namei: pass both WALK_GET and WALK_MORE to should_follow_link()
  namei: invert WALK_PUT logics
  namei: shift interpretation of LOOKUP_FOLLOW inside should_follow_link()
  namei: saner calling conventions for mountpoint_last()
  namei.c: get rid of user_path_parent()
  switch getfrag callbacks to ..._full() primitives
  make skb_add_data,{_nocache}() and skb_copy_to_page_nocache() advance only on success
  [iov_iter] new primitives - copy_from_iter_full() and friends
  don't open-code file_inode()
  ceph: switch to use of ->d_init()
  ceph: unify dentry_operations instances
  lustre: switch to use of ->d_init()

ceph: don't set req->r_locked_dir in ceph_d_revalidate

2016-12-08T13:32:16+00:00

This function sets req->r_locked_dir which is supposed to indicate to
ceph_fill_trace that the parent's i_rwsem is locked for write.
Unfortunately, there is no guarantee that the dir will be locked when
d_revalidate is called, so we really don't want ceph_fill_trace to do
any dcache manipulation from this context. Clear req->r_locked_dir since
it's clearly not safe to do that.

What we really want to know with d_revalidate is whether the dentry
still points to the same inode. ceph_fill_trace installs a pointer to
the inode in req->r_target_inode, so we can just compare that to
d_inode(dentry) to see if it's the same one after the lookup.

Also, since we aren't generally interested in the parent here, we can
switch to using a GETATTR to hint that to the MDS, which also means that
we only need to reserve one cap.

Finally, just remove the d_unhashed check. That's really outside the
purview of a filesystem's d_revalidate. If the thing became unhashed
while we're checking it, then that's up to the VFS to handle anyway.

Fixes: 200fd27c8fa2 ("ceph: use lookup request to revalidate dentry")
Link: http://tracker.ceph.com/issues/18041
Reported-by: Donatas Abraitis 
Signed-off-by: Jeff Layton 
Reviewed-by: "Yan, Zheng" 
Signed-off-by: Ilya Dryomov