staging/lustre/llite: take trunc_sem only at vvp layer
authorPatrick Farrell <paf@cray.com>
Mon, 20 Jun 2016 20:55:35 +0000 (16:55 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 20 Jun 2016 21:28:39 +0000 (14:28 -0700)
The lli_trunc_sem is taken in 'read' mode in both
ll_page_mkwrite and vvp_io_fault_start. This can lead to a
deadlock with another thread which asks for the semaphore
in write mode between thse two read calls.

Since all users of lli_trunc_sem are in the vvp layer, we
can satisfy the requirement to exclude truncate by taking
the semaphore only in vvp_io_fault_start.

Signed-off-by: Patrick Farrell <paf@cray.com>
Reviewed-on: http://review.whamcloud.com/19315
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7981
Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com>
Reviewed-by: Andriy Skulysh <andriy.skulysh@seagate.com>
Signed-off-by: Oleg Drokin <green@linuxhacker.ru>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/llite/llite_mmap.c

index fb1c3b613f59d5718efc16e29f47bf2e730bb913..66ee5db5fce847f1bf0509df4ff9a3adaca886df 100644 (file)
@@ -196,18 +196,11 @@ static int ll_page_mkwrite0(struct vm_area_struct *vma, struct page *vmpage,
 
        set = cfs_block_sigsinv(sigmask(SIGKILL) | sigmask(SIGTERM));
 
-       /* we grab lli_trunc_sem to exclude truncate case.
-        * Otherwise, we could add dirty pages into osc cache
-        * while truncate is on-going.
-        */
        inode = vvp_object_inode(io->ci_obj);
        lli = ll_i2info(inode);
-       down_read(&lli->lli_trunc_sem);
 
        result = cl_io_loop(env, io);
 
-       up_read(&lli->lli_trunc_sem);
-
        cfs_restore_sigs(set);
 
        if (result == 0) {