2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_inode.h"
37 #include "xfs_dinode.h"
38 #include "xfs_error.h"
39 #include "xfs_mru_cache.h"
40 #include "xfs_filestream.h"
41 #include "xfs_vnodeops.h"
42 #include "xfs_utils.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_inode_item.h"
47 #include <linux/kthread.h>
48 #include <linux/freezer.h>
51 * Sync all the inodes in the given AG according to the
52 * direction given by the flags.
60 xfs_perag_t *pag = &mp->m_perag[ag];
62 uint32_t first_index = 0;
68 xfs_inode_t *ip = NULL;
69 int lock_flags = XFS_ILOCK_SHARED;
72 * use a gang lookup to find the next inode in the tree
73 * as the tree is sparse and a gang lookup walks to find
74 * the number of objects requested.
76 read_lock(&pag->pag_ici_lock);
77 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
78 (void**)&ip, first_index, 1);
81 read_unlock(&pag->pag_ici_lock);
86 * Update the index for the next lookup. Catch overflows
87 * into the next AG range which can occur if we have inodes
88 * in the last block of the AG and we are currently
89 * pointing to the last inode.
91 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
92 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) {
93 read_unlock(&pag->pag_ici_lock);
97 /* nothing to sync during shutdown */
98 if (XFS_FORCED_SHUTDOWN(mp)) {
99 read_unlock(&pag->pag_ici_lock);
104 * If we can't get a reference on the inode, it must be
105 * in reclaim. Leave it for the reclaim code to flush.
109 read_unlock(&pag->pag_ici_lock);
112 read_unlock(&pag->pag_ici_lock);
114 /* avoid new or bad inodes */
115 if (is_bad_inode(inode) ||
116 xfs_iflags_test(ip, XFS_INEW)) {
122 * If we have to flush data or wait for I/O completion
123 * we need to hold the iolock.
125 if (flags & SYNC_DELWRI) {
126 if (VN_DIRTY(inode)) {
127 if (flags & SYNC_TRYLOCK) {
128 if (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
129 lock_flags |= XFS_IOLOCK_SHARED;
131 xfs_ilock(ip, XFS_IOLOCK_SHARED);
132 lock_flags |= XFS_IOLOCK_SHARED;
134 if (lock_flags & XFS_IOLOCK_SHARED) {
135 error = xfs_flush_pages(ip, 0, -1,
136 (flags & SYNC_WAIT) ? 0
141 if (VN_CACHED(inode) && (flags & SYNC_IOWAIT))
144 xfs_ilock(ip, XFS_ILOCK_SHARED);
146 if ((flags & SYNC_ATTR) && !xfs_inode_clean(ip)) {
147 if (flags & SYNC_WAIT) {
149 if (!xfs_inode_clean(ip))
150 error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
153 } else if (xfs_iflock_nowait(ip)) {
154 if (!xfs_inode_clean(ip))
155 error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
160 xfs_iput(ip, lock_flags);
165 * bail out if the filesystem is corrupted.
167 if (error == EFSCORRUPTED)
168 return XFS_ERROR(error);
183 int lflags = XFS_LOG_FORCE;
185 if (mp->m_flags & XFS_MOUNT_RDONLY)
190 if (flags & SYNC_WAIT)
191 lflags |= XFS_LOG_SYNC;
193 for (i = 0; i < mp->m_sb.sb_agcount; i++) {
194 if (!mp->m_perag[i].pag_ici_init)
196 error = xfs_sync_inodes_ag(mp, i, flags);
199 if (error == EFSCORRUPTED)
202 if (flags & SYNC_DELWRI)
203 xfs_log_force(mp, 0, lflags);
205 return XFS_ERROR(last_error);
209 xfs_commit_dummy_trans(
210 struct xfs_mount *mp,
213 struct xfs_inode *ip = mp->m_rootip;
214 struct xfs_trans *tp;
218 * Put a dummy transaction in the log to tell recovery
219 * that all others are OK.
221 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
222 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
224 xfs_trans_cancel(tp, 0);
228 xfs_ilock(ip, XFS_ILOCK_EXCL);
230 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
231 xfs_trans_ihold(tp, ip);
232 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
233 /* XXX(hch): ignoring the error here.. */
234 error = xfs_trans_commit(tp, 0);
236 xfs_iunlock(ip, XFS_ILOCK_EXCL);
238 xfs_log_force(mp, 0, log_flags);
244 struct xfs_mount *mp,
248 struct xfs_buf_log_item *bip;
252 * If this is xfssyncd() then only sync the superblock if we can
253 * lock it without sleeping and it is not pinned.
255 if (flags & SYNC_BDFLUSH) {
256 ASSERT(!(flags & SYNC_WAIT));
258 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
262 bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *);
263 if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp))
266 bp = xfs_getsb(mp, 0);
269 * If the buffer is pinned then push on the log so we won't
270 * get stuck waiting in the write for someone, maybe
271 * ourselves, to flush the log.
273 * Even though we just pushed the log above, we did not have
274 * the superblock buffer locked at that point so it can
275 * become pinned in between there and here.
277 if (XFS_BUF_ISPINNED(bp))
278 xfs_log_force(mp, 0, XFS_LOG_FORCE);
282 if (flags & SYNC_WAIT)
287 return xfs_bwrite(mp, bp);
296 * When remounting a filesystem read-only or freezing the filesystem, we have
297 * two phases to execute. This first phase is syncing the data before we
298 * quiesce the filesystem, and the second is flushing all the inodes out after
299 * we've waited for all the transactions created by the first phase to
300 * complete. The second phase ensures that the inodes are written to their
301 * location on disk rather than just existing in transactions in the log. This
302 * means after a quiesce there is no log replay required to write the inodes to
303 * disk (this is the main difference between a sync and a quiesce).
306 * First stage of freeze - no writers will make progress now we are here,
307 * so we flush delwri and delalloc buffers here, then wait for all I/O to
308 * complete. Data is frozen at that point. Metadata is not frozen,
309 * transactions can still occur here so don't bother flushing the buftarg
310 * because it'll just get dirty again.
314 struct xfs_mount *mp)
318 /* push non-blocking */
319 xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_BDFLUSH);
320 XFS_QM_DQSYNC(mp, SYNC_BDFLUSH);
321 xfs_filestream_flush(mp);
324 xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_WAIT|SYNC_IOWAIT);
325 XFS_QM_DQSYNC(mp, SYNC_WAIT);
327 /* write superblock and hoover up shutdown errors */
328 error = xfs_sync_fsdata(mp, 0);
330 /* flush data-only devices */
331 if (mp->m_rtdev_targp)
332 XFS_bflush(mp->m_rtdev_targp);
339 struct xfs_mount *mp)
341 int count = 0, pincount;
343 xfs_flush_buftarg(mp->m_ddev_targp, 0);
344 xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC);
347 * This loop must run at least twice. The first instance of the loop
348 * will flush most meta data but that will generate more meta data
349 * (typically directory updates). Which then must be flushed and
350 * logged before we can write the unmount record.
353 xfs_sync_inodes(mp, SYNC_ATTR|SYNC_WAIT);
354 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
363 * Second stage of a quiesce. The data is already synced, now we have to take
364 * care of the metadata. New transactions are already blocked, so we need to
365 * wait for any remaining transactions to drain out before proceding.
369 struct xfs_mount *mp)
373 /* wait for all modifications to complete */
374 while (atomic_read(&mp->m_active_trans) > 0)
377 /* flush inodes and push all remaining buffers out to disk */
381 * Just warn here till VFS can correctly support
382 * read-only remount without racing.
384 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
386 /* Push the superblock and write an unmount record */
387 error = xfs_log_sbcount(mp, 1);
389 xfs_fs_cmn_err(CE_WARN, mp,
390 "xfs_attr_quiesce: failed to log sb changes. "
391 "Frozen image may not be consistent.");
392 xfs_log_unmount_write(mp);
393 xfs_unmountfs_writesb(mp);
397 * Enqueue a work item to be picked up by the vfs xfssyncd thread.
398 * Doing this has two advantages:
399 * - It saves on stack space, which is tight in certain situations
400 * - It can be used (with care) as a mechanism to avoid deadlocks.
401 * Flushing while allocating in a full filesystem requires both.
404 xfs_syncd_queue_work(
405 struct xfs_mount *mp,
407 void (*syncer)(struct xfs_mount *, void *),
408 struct completion *completion)
410 struct xfs_sync_work *work;
412 work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP);
413 INIT_LIST_HEAD(&work->w_list);
414 work->w_syncer = syncer;
417 work->w_completion = completion;
418 spin_lock(&mp->m_sync_lock);
419 list_add_tail(&work->w_list, &mp->m_sync_list);
420 spin_unlock(&mp->m_sync_lock);
421 wake_up_process(mp->m_sync_task);
425 * Flush delayed allocate data, attempting to free up reserved space
426 * from existing allocations. At this point a new allocation attempt
427 * has failed with ENOSPC and we are in the process of scratching our
428 * heads, looking about for more room...
431 xfs_flush_inodes_work(
432 struct xfs_mount *mp,
435 struct inode *inode = arg;
436 xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK);
437 xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK | SYNC_IOWAIT);
445 struct inode *inode = VFS_I(ip);
446 DECLARE_COMPLETION_ONSTACK(completion);
449 xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion);
450 wait_for_completion(&completion);
451 xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
455 * Every sync period we need to unpin all items, reclaim inodes, sync
456 * quota and write out the superblock. We might need to cover the log
457 * to indicate it is idle.
461 struct xfs_mount *mp,
466 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
467 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
468 xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC);
469 /* dgc: errors ignored here */
470 error = XFS_QM_DQSYNC(mp, SYNC_BDFLUSH);
471 error = xfs_sync_fsdata(mp, SYNC_BDFLUSH);
472 if (xfs_log_need_covered(mp))
473 error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE);
476 wake_up(&mp->m_wait_single_sync_task);
483 struct xfs_mount *mp = arg;
485 xfs_sync_work_t *work, *n;
489 timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
491 timeleft = schedule_timeout_interruptible(timeleft);
494 if (kthread_should_stop() && list_empty(&mp->m_sync_list))
497 spin_lock(&mp->m_sync_lock);
499 * We can get woken by laptop mode, to do a sync -
500 * that's the (only!) case where the list would be
501 * empty with time remaining.
503 if (!timeleft || list_empty(&mp->m_sync_list)) {
505 timeleft = xfs_syncd_centisecs *
506 msecs_to_jiffies(10);
507 INIT_LIST_HEAD(&mp->m_sync_work.w_list);
508 list_add_tail(&mp->m_sync_work.w_list,
511 list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list)
512 list_move(&work->w_list, &tmp);
513 spin_unlock(&mp->m_sync_lock);
515 list_for_each_entry_safe(work, n, &tmp, w_list) {
516 (*work->w_syncer)(mp, work->w_data);
517 list_del(&work->w_list);
518 if (work == &mp->m_sync_work)
520 if (work->w_completion)
521 complete(work->w_completion);
531 struct xfs_mount *mp)
533 mp->m_sync_work.w_syncer = xfs_sync_worker;
534 mp->m_sync_work.w_mount = mp;
535 mp->m_sync_work.w_completion = NULL;
536 mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd");
537 if (IS_ERR(mp->m_sync_task))
538 return -PTR_ERR(mp->m_sync_task);
544 struct xfs_mount *mp)
546 kthread_stop(mp->m_sync_task);
555 xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino);
557 /* The hash lock here protects a thread in xfs_iget_core from
558 * racing with us on linking the inode back with a vnode.
559 * Once we have the XFS_IRECLAIM flag set it will not touch
562 write_lock(&pag->pag_ici_lock);
563 spin_lock(&ip->i_flags_lock);
564 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
565 !__xfs_iflags_test(ip, XFS_IRECLAIMABLE)) {
566 spin_unlock(&ip->i_flags_lock);
567 write_unlock(&pag->pag_ici_lock);
570 xfs_iunlock(ip, XFS_ILOCK_EXCL);
574 __xfs_iflags_set(ip, XFS_IRECLAIM);
575 spin_unlock(&ip->i_flags_lock);
576 write_unlock(&pag->pag_ici_lock);
577 xfs_put_perag(ip->i_mount, pag);
580 * If the inode is still dirty, then flush it out. If the inode
581 * is not in the AIL, then it will be OK to flush it delwri as
582 * long as xfs_iflush() does not keep any references to the inode.
583 * We leave that decision up to xfs_iflush() since it has the
584 * knowledge of whether it's OK to simply do a delwri flush of
585 * the inode or whether we need to wait until the inode is
586 * pulled from the AIL.
587 * We get the flush lock regardless, though, just to make sure
588 * we don't free it while it is being flushed.
591 xfs_ilock(ip, XFS_ILOCK_EXCL);
596 * In the case of a forced shutdown we rely on xfs_iflush() to
597 * wait for the inode to be unpinned before returning an error.
599 if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) {
600 /* synchronize with xfs_iflush_done */
605 xfs_iunlock(ip, XFS_ILOCK_EXCL);
611 * We set the inode flag atomically with the radix tree tag.
612 * Once we get tag lookups on the radix tree, this inode flag
616 xfs_inode_set_reclaim_tag(
619 xfs_mount_t *mp = ip->i_mount;
620 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
622 read_lock(&pag->pag_ici_lock);
623 spin_lock(&ip->i_flags_lock);
624 radix_tree_tag_set(&pag->pag_ici_root,
625 XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG);
626 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
627 spin_unlock(&ip->i_flags_lock);
628 read_unlock(&pag->pag_ici_lock);
629 xfs_put_perag(mp, pag);
633 __xfs_inode_clear_reclaim_tag(
638 radix_tree_tag_clear(&pag->pag_ici_root,
639 XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG);
643 xfs_inode_clear_reclaim_tag(
646 xfs_mount_t *mp = ip->i_mount;
647 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
649 read_lock(&pag->pag_ici_lock);
650 spin_lock(&ip->i_flags_lock);
651 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
652 spin_unlock(&ip->i_flags_lock);
653 read_unlock(&pag->pag_ici_lock);
654 xfs_put_perag(mp, pag);
659 xfs_reclaim_inodes_ag(
665 xfs_inode_t *ip = NULL;
666 xfs_perag_t *pag = &mp->m_perag[ag];
668 uint32_t first_index;
676 * use a gang lookup to find the next inode in the tree
677 * as the tree is sparse and a gang lookup walks to find
678 * the number of objects requested.
680 read_lock(&pag->pag_ici_lock);
681 nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root,
682 (void**)&ip, first_index, 1,
683 XFS_ICI_RECLAIM_TAG);
686 read_unlock(&pag->pag_ici_lock);
691 * Update the index for the next lookup. Catch overflows
692 * into the next AG range which can occur if we have inodes
693 * in the last block of the AG and we are currently
694 * pointing to the last inode.
696 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
697 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) {
698 read_unlock(&pag->pag_ici_lock);
702 /* ignore if already under reclaim */
703 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
704 read_unlock(&pag->pag_ici_lock);
709 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
710 read_unlock(&pag->pag_ici_lock);
713 if (xfs_ipincount(ip) ||
714 !xfs_iflock_nowait(ip)) {
715 xfs_iunlock(ip, XFS_ILOCK_EXCL);
716 read_unlock(&pag->pag_ici_lock);
720 read_unlock(&pag->pag_ici_lock);
723 * hmmm - this is an inode already in reclaim. Do
724 * we even bother catching it here?
726 if (xfs_reclaim_inode(ip, noblock, mode))
746 for (i = 0; i < mp->m_sb.sb_agcount; i++) {
747 if (!mp->m_perag[i].pag_ici_init)
749 xfs_reclaim_inodes_ag(mp, i, noblock, mode);