glusterfs.git/xlators, branch v3.3beta1

nfs: do not free the constant string

2011-07-20T05:30:32+00:00

Change-Id: Iea3838e09efad5cac789f93148fb941d2f7109b1
BUG: 3039
Reviewed-on: http://review.gluster.com/10
Tested-by: Gluster Build System 
Reviewed-by: Anand Avati

glusterd rebalance: use lstat() instead of stat()

2011-07-19T14:59:00+00:00

so the symlinks pointing to directories are not considered
as directories themself and the control flows outside
gluster's scope

Change-Id: Iae910ce6c68886d34ae6e5efe46062481b40cd25
BUG: 3191
Reviewed-on: http://review.gluster.com/26
Tested-by: Gluster Build System 
Reviewed-by: Anand Avati

fuse: fix inode-leaks in fuse_lookup_and_resume codepath.

2011-07-18T06:34:56+00:00

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

BUG: 2525 ([glusterfs-3.1.3qa8]: inode leak in rdma)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2525

io-threads: Initialize mutex and cond

2011-07-17T14:47:26+00:00

Thanks to Emmanuel Dreyfus for sending out the fix.

Signed-off-by: shishir gowda 
Signed-off-by: Anand Avati 

BUG: 3170 (Uninitialized mutex and cond in io-threads)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3170

posix: perform readdir filling in locked region

2011-07-16T19:47:30+00:00

When two application threads share an open dir fd (DIR *) and issue
readdirs, storage/posix will receive separate readdir fops in separate
threads in parallel. This has two-fold issues

1. In the following pair of operations -

   entry = readdir(dir)
       and
   strcpy (gf_dirent->name, entry->d_name)

   @entry is a static buffer in libc which can get reused by another thread
   to get filled with a longer name. This can cause the second operation
   to overflow the buffer as the allocation was for the smaller name.

2. In the following pair of operations -

   seekdir (dir, offset)
      and
   entry = readdir(dir)

   If two threads are executing these sequence in parallel in separate
   threads, then one of them will end up reading wrong/unexpected entries.

It would be sufficient to fix 1. by using readdir_r but that still keeps
the second race open. Hence the patch moves all the set of operations to a
locked region which solves both races.

Signed-off-by: Anand Avati 

BUG: 3171 (Crash in server)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3171

features/locks: Avoid using reqlock to prevent race

2011-07-14T06:50:14+00:00

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

BUG: 2473 (Support for volume and directory level quota)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2473

storage/posix: set all keys present in 'params' dict while creating an entry

2011-07-13T06:16:54+00:00

so, there is no need to do a 'setxattr()' after entry creation, which could now
fail due to ACL on server side.

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

BUG: 3164 (set xattr keys while entry creation)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3164

mgmt/glusterd: write complete string contained in value in to the volinfo file.

2011-07-13T05:09:27+00:00

Currently, we are writing 4096 bytes of value variable to the volinfo file
irrespective of its lenght, thus tampering with the volinfo file structure.

Signed-off-by: Junaid 
Signed-off-by: Anand Avati 

BUG: 3154 (glusterd fails to restart)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3154

cluster/distribute: send proper 'params' dictionary during linkfile creation

2011-07-12T14:12:50+00:00

when the code path enters the 'subvol_filled()' case, local->params is set,
which contains the 'gfid-req' value, but the linkfile creation was not
checking for its existance.

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

BUG: 3159 (mknod (linkfile creation) with no 'gfid-req' key)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3159

server: Reassociating 'old' conn can lead to chaos.

2011-07-12T12:38:10+00:00

Since we moved to a single socket connection b/w client and server,
inherting connection state is unnecessary and can sometimes be
dangerous when clients reconnect even before the server detects
a socket error on the old connection.

Dirty detail: This reassociation results in 'ref count' not decreasing
in tandem with the connection disconnects. This results in a resource
leak.

Signed-off-by: Krishnan Parthasarathi 
Signed-off-by: Anand Avati 

BUG: 3104 (Self-heal does not work in dis-rep!)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3104