nfs: Don't busy-wait on SIGKILL in __nfs_iocounter_wait
authorDavid Jeffery <djeffery@redhat.com>
Tue, 5 Aug 2014 15:19:42 +0000 (11:19 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Fri, 22 Aug 2014 22:04:44 +0000 (18:04 -0400)
If a SIGKILL is sent to a task waiting in __nfs_iocounter_wait,
it will busy-wait or soft lockup in its while loop.
nfs_wait_bit_killable won't sleep, and the loop won't exit on
the error return.

Stop the busy-wait by breaking out of the loop when
nfs_wait_bit_killable returns an error.

Signed-off-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
fs/nfs/pagelist.c

index 932c6cc27a7664a29c9bf8d5883990c510e59b02..be7cbce6e4c7a71c0fbd22a03193e20c9fec55ce 100644 (file)
@@ -116,7 +116,7 @@ __nfs_iocounter_wait(struct nfs_io_counter *c)
                if (atomic_read(&c->io_count) == 0)
                        break;
                ret = nfs_wait_bit_killable(&q.key);
-       } while (atomic_read(&c->io_count) != 0);
+       } while (atomic_read(&c->io_count) != 0 && !ret);
        finish_wait(wq, &q.wait);
        return ret;
 }