glusterfs.git/xlators, branch v3.1.0qa3

fixes for spurious entry self-heal in afr

2010-07-29T06:58:09+00:00

Signed-off-by: Pranith Kumar K 
Signed-off-by: Anand V. Avati 

BUG: 974 (Spurious self-heal?)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=974

glusterd: Bug fixes in add/remove brick.

2010-07-29T05:55:49+00:00

Signed-off-by: Vijay Bellur 
Signed-off-by: Anand V. Avati 

BUG: 1239 ()
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1239

changes to rpc

2010-07-28T12:08:26+00:00

- use mem-pool for requests and saved_frames.
  - preserve the rpc_req structure till rpc invokes program's reply.
    This will enable us to store transport specific data that has to
    last till reply has come (eg., memory regions of chunk lists in
    case of rdma).
  - change signature of rpc_clnt_submit to accept rsphdr_vector and
    rsppayload_vector. The buffers pointed by these vectors will be
    from iobufs and these iobufs are added to an iobref which should
    also be passed as an arguement to rpc_clnt_submit.

Signed-off-by: Raghavendra G 
Signed-off-by: Anand V. Avati 

BUG: 875 (Implement a new protocol to provide proper backward/forward compatibility)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=875

glusterd: Fix a crash when get volume fails

2010-07-28T11:17:28+00:00

Signed-off-by: Vijay Bellur 
Signed-off-by: Anand V. Avati 

BUG: 1239 ()
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1239

fix all clang errors in xlator/protocol

2010-07-28T10:34:58+00:00

Signed-off-by: Amar Tumballi 
Signed-off-by: Anand V. Avati 

BUG: 1092 ()
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1092

removed last few remaining 'ERR_ABORT's from codebase

2010-07-28T10:34:54+00:00

Signed-off-by: Amar Tumballi 
Signed-off-by: Anand V. Avati 

BUG: 966 (NULL check for avoiding NULL dereferencing of pointers..)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=966

some check added to the variables after GF_CALLOC

2010-07-28T10:34:48+00:00

handles some NULL dereference problems
(reported by clang when ran with code where '#define GF_CALLOC NULL').

Signed-off-by: Amar Tumballi 
Signed-off-by: Anand V. Avati 

BUG: 966 (NULL check for avoiding NULL dereferencing of pointers..)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=966

nfs3: NULL fdentry check before removing from fdcache

2010-07-28T10:10:02+00:00

Suppose a file name 1 is created and some data is written to it. After this
another 512 files are newly created and written to. When the the 513th file is
created and an fd_t opened for it, it results in 1's fd_t being replaced in the
fd-lru with 513th file's fd_t. This is the correct behaviour resulting in all
refs getting unref from the fd_t of 1 and the fd and all related state being
freed.

But, in some workloads, some refs are still pending even after the fd_t is
removed from LRU, resulting in the fd still being bound to the inode. In
nfs3svc_remove_cbk, while removing the inode state, we also ensure that any
fd_ts in the cache for this inode are also removed. While removing the fd_t,
this situation where the fd_t has replaced with another, even while a ref
remains on the fd_t, results in a crash in the fdcache_remove path in
nfs3svc_remove_cbk. This happens because the fd_ctx_get results in a NULL value
because the ctx was already deleted when this fd_t was removed from fd-lru
earlier. This patch fixes the crash by introducing a NULL check.

Signed-off-by: Shehjar Tikoo 
Signed-off-by: Anand V. Avati 

BUG: 885 ([NFS Xlator] Crash in nfs3_fdcache_update)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=885

access-control: Fix O_CREATE determination check

2010-07-28T09:24:53+00:00

Signed-off-by: Shehjar Tikoo 
Signed-off-by: Anand V. Avati 

BUG: 1167 (gnfs crash with  fio)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1167

replicate: Set setxattr wait-count to child-count to avoid race

2010-07-28T09:24:49+00:00

A race is seen during a compilebench load against NFS where a NFSv3 MKDIR is
immediately followed by a LOOKUP on the new directory. By the time this lookup
comes in, the setxattr issued by distribute for the mkdir has not returned yet
from all the replicas. Since replicate returns a reply for setxattr after
receiving reply only from one subvol, the MKDIR is unwound from distribute.

When the LOOKUP is received at distribute, the setxattr has still not completed
at the second replica, whereas that is where the lookup is sent by replicate.
Finding that the xattr is missing from the dir, distribute returns ESTALE to
NFS, which return NFS3ERR_STALE to nfs client.

This patch changes replicate setxattr behaviour by forcing replicate to wait for all
children rather than one to call back before unwinding the stack.

Signed-off-by: Shehjar Tikoo 
Signed-off-by: Anand V. Avati 

BUG: 937 (compilebench fails on NFS+dist+repl+all perf xlators)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=937