We need to flush the msgr workqueue during mon_client shutdown to
ensure that any work affecting our embedded ceph_connection is
finished so that we can be safely destroyed.

Previously, we were flushing the work queue after osd_client
shutdown and before mon_client shutdown to ensure that any osd
connection refs to authorizers are flushed.  Remove the redundant
flush, and document in the comment that the mon_client flush is
needed to cover that case as well.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
(cherry picked from commit f3dea7edd3d449fe7a6d402c1ce56a294b985261)

The bug can cause NULL pointer dereference in write_partial_msg_pages

Signed-off-by: Zheng Yan <zheng.z.yan@intel.com>
Reviewed-by: Alex Elder <elder@inktank.com>
(cherry picked from commit 43643528cce60ca184fe8197efa8e8da7c89a037)

There were a few direct calls to ceph_con_{get,put}() instead of the con
ops from osd_client.c.  This is a bug since those ops aren't defined to
be ceph_con_get/put.

This breaks refcounting on the ceph_osd structs that contain the
ceph_connections, and could lead to all manner of strangeness.

The purpose of the ->get and ->put methods in a ceph connection are
to allow the connection to indicate it has a reference to something
external to the messaging system, *not* to indicate something
external has a reference to the connection.

[elder@inktank.com: added that last sentence]

Signed-off-by: Sage Weil <sage@newdream.net>
Reviewed-by: Alex Elder <elder@inktank.com>
(cherry picked from commit 0d47766f14211a73eaf54cab234db134ece79f49)

In ceph_osdc_release_request(), a reference to the r_reply message
is dropped.  But just after that, that same message is revoked if it
was in use to receive an incoming reply.  Reorder these so we are
sure we hold a reference until we're actually done with the message.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
(cherry picked from commit ab8cb34a4b2f60281a4b18b1f1ad23bc2313d91b)

Pull a 9p bugfix from Eric Van Hensbergen.

* 'v9fs-devel' of git://git.kernel.org/pub/scm/linux/kernel/git/ericvh/v9fs:
  9p: BUG before corrupting memory

Pull NFS client bugfixes from Trond Myklebust:
 "Highlights include:

   - Fix a couple of mount regressions due to the recent cleanups.
   - Fix an Oops in the open recovery code
   - Fix an rpc_pipefs upcall hang that results from some of the net
     namespace work from 3.4.x (stable kernel candidate).
   - Fix a couple of write and o_direct regressions that were found at
     last weeks Bakeathon testing event in Ann Arbor."

* tag 'nfs-for-3.5-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFS: add an endian notation for sparse
  NFSv4.1: integer overflow in decode_cb_sequence_args()
  rpc_pipefs: allow rpc_purge_list to take a NULL waitq pointer
  NFSv4 do not send an empty SETATTR compound
  NFSv2: EOF incorrectly set on short read
  NFS: Use the NFS_DEFAULT_VERSION for v2 and v3 mounts
  NFS: fix directio refcount bug on commit
  NFSv4: Fix unnecessary delegation returns in nfs4_do_open
  NFSv4.1: Convert another trivial printk into a dprintk
  NFS4: Fix open bug when pnfs module blacklisted
  NFS: Remove incorrect BUG_ON in nfs_found_client
  NFS: Map minor mismatch error to protocol not support error.
  NFS: Fix a commit bug
  NFS4: Set parsed mount data version to 4
  NFSv4.1: Ensure we clear session state flags after a session creation
  NFSv4.1: Convert a trivial printk into a dprintk
  NFSv4: Fix up decode_attr_mdsthreshold
  NFSv4: Fix an Oops in the open recovery code
  NFSv4.1: Fix a request leak on the back channel

Bogdan Hamciuc diagnosed and fixed following bug in netpoll_send_udp() :

"skb->len += len;" instead of "skb_put(skb, len);"

Meaning that _if_ a network driver needs to call skb_realloc_headroom(),
only packet headers would be copied, leaving garbage in the payload.

However the skb_realloc_headroom() must be avoided as much as possible
since it requires memory and netpoll tries hard to work even if memory
is exhausted (using a pool of preallocated skbs)

It appears netpoll_send_udp() reserved 16 bytes for the ethernet header,
which happens to work for typicall drivers but not all.

Right thing is to use LL_RESERVED_SPACE(dev)
(And also add dev->needed_tailroom of tailroom)

This patch combines both fixes.

Many thanks to Bogdan for raising this issue.

Reported-by: Bogdan Hamciuc <bogdan.hamciuc@freescale.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Bogdan Hamciuc <bogdan.hamciuc@freescale.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Neil Horman <nhorman@tuxdriver.com>
Reviewed-by: Neil Horman <nhorman@tuxdriver.com>
Reviewed-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In the event that we don't have a dentry for a rpc_pipefs pipe, we still
need to allow the queue_timeout job to clean out the queue. There's just
no waitq to wake up in that event.

Cc: stable@kernel.org
Reported-by: Hans de Bruin <jmdebruin@xmsnet.nl>
Reported-by: Joerg Platte <jplatte@naasa.net>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>