Imported Upstream version 0.6.4.2

[mirror_zfs-debian.git] / module / zfs / vdev_file.c

diff --git a/module/zfs/vdev_file.c b/module/zfs/vdev_file.c
index 858582acad74b97db7bbabbb43cf312e389d4812..7f43ad8001f4001589dd8bd22e7cb7f35af01bdb 100644 (file)
--- a/module/zfs/vdev_file.c
+++ b/module/zfs/vdev_file.c
@@ -36,6 +36,8 @@
  * Virtual device vector for files.
  */
 
+static taskq_t *vdev_file_taskq;
+
 static void
 vdev_file_hold(vdev_t *vd)
 {
@@ -75,7 +77,7 @@ vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
                goto skip_open;
        }
 
-       vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_PUSHPAGE);
+       vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
 
        /*
         * We always open the files from the root of the global zone, even if
@@ -159,6 +161,17 @@ vdev_file_io_strategy(void *arg)
        zio_interrupt(zio);
 }
 
+static void
+vdev_file_io_fsync(void *arg)
+{
+       zio_t *zio = (zio_t *)arg;
+       vdev_file_t *vf = zio->io_vd->vdev_tsd;
+
+       zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC, kcred, NULL);
+
+       zio_interrupt(zio);
+}
+
 static int
 vdev_file_io_start(zio_t *zio)
 {
@@ -174,6 +187,23 @@ vdev_file_io_start(zio_t *zio)
 
                switch (zio->io_cmd) {
                case DKIOCFLUSHWRITECACHE:
+
+                       if (zfs_nocacheflush)
+                               break;
+
+                       /*
+                        * We cannot safely call vfs_fsync() when PF_FSTRANS
+                        * is set in the current context.  Filesystems like
+                        * XFS include sanity checks to verify it is not
+                        * already set, see xfs_vm_writepage().  Therefore
+                        * the sync must be dispatched to a different context.
+                        */
+                       if (spl_fstrans_check()) {
+                               VERIFY3U(taskq_dispatch(vdev_file_taskq,
+                                   vdev_file_io_fsync, zio, TQ_SLEEP), !=, 0);
+                               return (ZIO_PIPELINE_STOP);
+                       }
+
                        zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
                            kcred, NULL);
                        break;
@@ -184,8 +214,8 @@ vdev_file_io_start(zio_t *zio)
                return (ZIO_PIPELINE_CONTINUE);
        }
 
-       VERIFY3U(taskq_dispatch(system_taskq, vdev_file_io_strategy, zio,
-           TQ_PUSHPAGE), !=, 0);
+       VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
+           TQ_SLEEP), !=, 0);
 
        return (ZIO_PIPELINE_STOP);
 }
@@ -209,6 +239,21 @@ vdev_ops_t vdev_file_ops = {
        B_TRUE                  /* leaf vdev */
 };
 
+void
+vdev_file_init(void)
+{
+       vdev_file_taskq = taskq_create("vdev_file_taskq", 100, minclsyspri,
+           max_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_THREADS_CPU_PCT);
+
+       VERIFY(vdev_file_taskq);
+}
+
+void
+vdev_file_fini(void)
+{
+       taskq_destroy(vdev_file_taskq);
+}
+
 /*
  * From userland we access disks just like files.
  */