staging: lustre: llite: lookup master inode by ilookup5_nowait
authorwang di <di.wang@intel.com>
Thu, 10 Nov 2016 17:30:43 +0000 (12:30 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 14 Nov 2016 15:13:57 +0000 (16:13 +0100)
commit5ce86eaccfa8859c369514d85387585241c81ca5
treede48455ce275039b28ad9b57b720dcba57de73a0
parent8d2dab899a98d03a4ae3bdf3b73a3827990aa2ad
staging: lustre: llite: lookup master inode by ilookup5_nowait

Do not lookup master inode by ilookup5, instead it should
use ilookup5_nowait, otherwise it will cause dead lock,

1. Client1 send chmod req to the MDT0, then on MDT0, it
   enqueues master and all of its slaves lock, (mdt_attr_set()
   ->mdt_lock_slaves()), after gets master and stripe0 lock,
   it will send the enqueue request(for stripe1) to MDT1, then
   MDT1 finds the lock has been granted to client2. Then MDT1
   sends blocking ast to client2.

2. At the same time, client2 tries to unlink the striped
   dir (rm -rf striped_dir), and during lookup, it will hold
   the master inode of the striped directory, whose inode state
   is NEW, then tries to revalidate all of its slaves,
   (ll_prep_inode()->ll_iget()->ll_read_inode2()->
   ll_update_inode().). And it will be blocked on the server
   side because of 1.

3. Then the client get the blocking_ast request, cancel the
   lock, but being blocked by ilookup5 in ll_md_blocking_ast(),
   because the inode state is still NEW.

Signed-off-by: wang di <di.wang@intel.com>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5344
Reviewed-on: http://review.whamcloud.com/16066
Reviewed-by: John L. Hammond <john.hammond@intel.com>
Reviewed-by: Lai Siyao <lai.siyao@intel.com>
Reviewed-by: James Simmons <uja.ornl@yahoo.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: James Simmons <jsimmons@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/llite/namei.c