d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Wait for all znodes to be released before tearing down the superblock

Browse Source 
By the time we're tearing down our superblock the VFS has started releasing
all our inodes/znodes. Some of this work may have been handed off to our
iput taskq so we need to wait for that work to complete. However the iput
from the taskq can itself result in additional work being added to the
taskq:

dsl_pool_iput_taskq
 iput
  iput_final
   evict
    destroy_inode
     zpl_inode_destroy
      zfs_inode_destroy
       zfs_iput_async(ZTOI(zp->z_xattr_parent))
        taskq_dispatch(dsl_pool_iput_taskq..., iput, ...)

Let's wait until all our znodes have been released.

Signed-off-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #3281
This commit is contained in:
Chris Dunlop 2015-05-02 15:47:06 +10:00 committed by Brian Behlendorf
parent 7a3066ffdd
commit f0da4d1508
1 changed files with 30 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
module/zfs/zfs_vfsops.c
View File

@ -1137,8 +1137,28 @@ zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting)
* drain the iput_taskq to ensure all active references to the
* zfs_sb_t have been handled only then can it be safely destroyed.
*/
if (zsb->z_os)
taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb->z_os)));
if (zsb->z_os) {
/*
* If we're unmounting we have to wait for the list to
* drain completely.
*
* If we're not unmounting there's no guarantee the list
* will drain completely, but iputs run from the taskq
* may add the parents of dir-based xattrs to the taskq
* so we want to wait for these.
*
* We can safely read z_nr_znodes without locking because the
* VFS has already blocked operations which add to the
* z_all_znodes list and thus increment z_nr_znodes.
*/
int round = 0;
while (zsb->z_nr_znodes > 0) {
taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(
zsb->z_os)));
if (++round > 1 && !unmounting)
break;
}
}
rrw_enter(&zsb->z_teardown_lock, RW_WRITER, FTAG);
@ -1182,13 +1202,15 @@ zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting)
*
* Release all holds on dbufs.
*/
mutex_enter(&zsb->z_znodes_lock);
for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
zp = list_next(&zsb->z_all_znodes, zp)) {
if (zp->z_sa_hdl)
zfs_znode_dmu_fini(zp);
if (!unmounting) {
mutex_enter(&zsb->z_znodes_lock);
for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
zp = list_next(&zsb->z_all_znodes, zp)) {
if (zp->z_sa_hdl)
zfs_znode_dmu_fini(zp);
}
mutex_exit(&zsb->z_znodes_lock);
}
mutex_exit(&zsb->z_znodes_lock);
/*
* If we are unmounting, set the unmounted flag and let new VFS ops