ALSA: hda - Fix silent output on ASUS A6Rp
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / xfs / xfs_sync.c
blobf0994aedcd158c2db3d6f9b2bf4d21a4f819bec6
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_dinode.h"
32 #include "xfs_error.h"
33 #include "xfs_filestream.h"
34 #include "xfs_vnodeops.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_quota.h"
37 #include "xfs_trace.h"
38 #include "xfs_fsops.h"
40 #include <linux/kthread.h>
41 #include <linux/freezer.h>
43 struct workqueue_struct *xfs_syncd_wq; /* sync workqueue */
46 * The inode lookup is done in batches to keep the amount of lock traffic and
47 * radix tree lookups to a minimum. The batch size is a trade off between
48 * lookup reduction and stack usage. This is in the reclaim path, so we can't
49 * be too greedy.
51 #define XFS_LOOKUP_BATCH 32
53 STATIC int
54 xfs_inode_ag_walk_grab(
55 struct xfs_inode *ip)
57 struct inode *inode = VFS_I(ip);
59 ASSERT(rcu_read_lock_held());
62 * check for stale RCU freed inode
64 * If the inode has been reallocated, it doesn't matter if it's not in
65 * the AG we are walking - we are walking for writeback, so if it
66 * passes all the "valid inode" checks and is dirty, then we'll write
67 * it back anyway. If it has been reallocated and still being
68 * initialised, the XFS_INEW check below will catch it.
70 spin_lock(&ip->i_flags_lock);
71 if (!ip->i_ino)
72 goto out_unlock_noent;
74 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
75 if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM))
76 goto out_unlock_noent;
77 spin_unlock(&ip->i_flags_lock);
79 /* nothing to sync during shutdown */
80 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
81 return EFSCORRUPTED;
83 /* If we can't grab the inode, it must on it's way to reclaim. */
84 if (!igrab(inode))
85 return ENOENT;
87 if (is_bad_inode(inode)) {
88 IRELE(ip);
89 return ENOENT;
92 /* inode is valid */
93 return 0;
95 out_unlock_noent:
96 spin_unlock(&ip->i_flags_lock);
97 return ENOENT;
100 STATIC int
101 xfs_inode_ag_walk(
102 struct xfs_mount *mp,
103 struct xfs_perag *pag,
104 int (*execute)(struct xfs_inode *ip,
105 struct xfs_perag *pag, int flags),
106 int flags)
108 uint32_t first_index;
109 int last_error = 0;
110 int skipped;
111 int done;
112 int nr_found;
114 restart:
115 done = 0;
116 skipped = 0;
117 first_index = 0;
118 nr_found = 0;
119 do {
120 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
121 int error = 0;
122 int i;
124 rcu_read_lock();
125 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
126 (void **)batch, first_index,
127 XFS_LOOKUP_BATCH);
128 if (!nr_found) {
129 rcu_read_unlock();
130 break;
134 * Grab the inodes before we drop the lock. if we found
135 * nothing, nr == 0 and the loop will be skipped.
137 for (i = 0; i < nr_found; i++) {
138 struct xfs_inode *ip = batch[i];
140 if (done || xfs_inode_ag_walk_grab(ip))
141 batch[i] = NULL;
144 * Update the index for the next lookup. Catch
145 * overflows into the next AG range which can occur if
146 * we have inodes in the last block of the AG and we
147 * are currently pointing to the last inode.
149 * Because we may see inodes that are from the wrong AG
150 * due to RCU freeing and reallocation, only update the
151 * index if it lies in this AG. It was a race that lead
152 * us to see this inode, so another lookup from the
153 * same index will not find it again.
155 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
156 continue;
157 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
158 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
159 done = 1;
162 /* unlock now we've grabbed the inodes. */
163 rcu_read_unlock();
165 for (i = 0; i < nr_found; i++) {
166 if (!batch[i])
167 continue;
168 error = execute(batch[i], pag, flags);
169 IRELE(batch[i]);
170 if (error == EAGAIN) {
171 skipped++;
172 continue;
174 if (error && last_error != EFSCORRUPTED)
175 last_error = error;
178 /* bail out if the filesystem is corrupted. */
179 if (error == EFSCORRUPTED)
180 break;
182 cond_resched();
184 } while (nr_found && !done);
186 if (skipped) {
187 delay(1);
188 goto restart;
190 return last_error;
194 xfs_inode_ag_iterator(
195 struct xfs_mount *mp,
196 int (*execute)(struct xfs_inode *ip,
197 struct xfs_perag *pag, int flags),
198 int flags)
200 struct xfs_perag *pag;
201 int error = 0;
202 int last_error = 0;
203 xfs_agnumber_t ag;
205 ag = 0;
206 while ((pag = xfs_perag_get(mp, ag))) {
207 ag = pag->pag_agno + 1;
208 error = xfs_inode_ag_walk(mp, pag, execute, flags);
209 xfs_perag_put(pag);
210 if (error) {
211 last_error = error;
212 if (error == EFSCORRUPTED)
213 break;
216 return XFS_ERROR(last_error);
219 STATIC int
220 xfs_sync_inode_data(
221 struct xfs_inode *ip,
222 struct xfs_perag *pag,
223 int flags)
225 struct inode *inode = VFS_I(ip);
226 struct address_space *mapping = inode->i_mapping;
227 int error = 0;
229 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
230 return 0;
232 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) {
233 if (flags & SYNC_TRYLOCK)
234 return 0;
235 xfs_ilock(ip, XFS_IOLOCK_SHARED);
238 error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ?
239 0 : XBF_ASYNC, FI_NONE);
240 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
241 return error;
244 STATIC int
245 xfs_sync_inode_attr(
246 struct xfs_inode *ip,
247 struct xfs_perag *pag,
248 int flags)
250 int error = 0;
252 xfs_ilock(ip, XFS_ILOCK_SHARED);
253 if (xfs_inode_clean(ip))
254 goto out_unlock;
255 if (!xfs_iflock_nowait(ip)) {
256 if (!(flags & SYNC_WAIT))
257 goto out_unlock;
258 xfs_iflock(ip);
261 if (xfs_inode_clean(ip)) {
262 xfs_ifunlock(ip);
263 goto out_unlock;
266 error = xfs_iflush(ip, flags);
269 * We don't want to try again on non-blocking flushes that can't run
270 * again immediately. If an inode really must be written, then that's
271 * what the SYNC_WAIT flag is for.
273 if (error == EAGAIN) {
274 ASSERT(!(flags & SYNC_WAIT));
275 error = 0;
278 out_unlock:
279 xfs_iunlock(ip, XFS_ILOCK_SHARED);
280 return error;
284 * Write out pagecache data for the whole filesystem.
286 STATIC int
287 xfs_sync_data(
288 struct xfs_mount *mp,
289 int flags)
291 int error;
293 ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0);
295 error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags);
296 if (error)
297 return XFS_ERROR(error);
299 xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0);
300 return 0;
304 * Write out inode metadata (attributes) for the whole filesystem.
306 STATIC int
307 xfs_sync_attr(
308 struct xfs_mount *mp,
309 int flags)
311 ASSERT((flags & ~SYNC_WAIT) == 0);
313 return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags);
316 STATIC int
317 xfs_sync_fsdata(
318 struct xfs_mount *mp)
320 struct xfs_buf *bp;
321 int error;
324 * If the buffer is pinned then push on the log so we won't get stuck
325 * waiting in the write for someone, maybe ourselves, to flush the log.
327 * Even though we just pushed the log above, we did not have the
328 * superblock buffer locked at that point so it can become pinned in
329 * between there and here.
331 bp = xfs_getsb(mp, 0);
332 if (xfs_buf_ispinned(bp))
333 xfs_log_force(mp, 0);
334 error = xfs_bwrite(bp);
335 xfs_buf_relse(bp);
336 return error;
340 xfs_log_dirty_inode(
341 struct xfs_inode *ip,
342 struct xfs_perag *pag,
343 int flags)
345 struct xfs_mount *mp = ip->i_mount;
346 struct xfs_trans *tp;
347 int error;
349 if (!ip->i_update_core)
350 return 0;
352 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
353 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
354 if (error) {
355 xfs_trans_cancel(tp, 0);
356 return error;
359 xfs_ilock(ip, XFS_ILOCK_EXCL);
360 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
361 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
362 return xfs_trans_commit(tp, 0);
366 * When remounting a filesystem read-only or freezing the filesystem, we have
367 * two phases to execute. This first phase is syncing the data before we
368 * quiesce the filesystem, and the second is flushing all the inodes out after
369 * we've waited for all the transactions created by the first phase to
370 * complete. The second phase ensures that the inodes are written to their
371 * location on disk rather than just existing in transactions in the log. This
372 * means after a quiesce there is no log replay required to write the inodes to
373 * disk (this is the main difference between a sync and a quiesce).
376 * First stage of freeze - no writers will make progress now we are here,
377 * so we flush delwri and delalloc buffers here, then wait for all I/O to
378 * complete. Data is frozen at that point. Metadata is not frozen,
379 * transactions can still occur here so don't bother flushing the buftarg
380 * because it'll just get dirty again.
383 xfs_quiesce_data(
384 struct xfs_mount *mp)
386 int error, error2 = 0;
389 * Log all pending size and timestamp updates. The vfs writeback
390 * code is supposed to do this, but due to its overagressive
391 * livelock detection it will skip inodes where appending writes
392 * were written out in the first non-blocking sync phase if their
393 * completion took long enough that it happened after taking the
394 * timestamp for the cut-off in the blocking phase.
396 xfs_inode_ag_iterator(mp, xfs_log_dirty_inode, 0);
398 xfs_qm_sync(mp, SYNC_TRYLOCK);
399 xfs_qm_sync(mp, SYNC_WAIT);
401 /* force out the newly dirtied log buffers */
402 xfs_log_force(mp, XFS_LOG_SYNC);
404 /* write superblock and hoover up shutdown errors */
405 error = xfs_sync_fsdata(mp);
407 /* make sure all delwri buffers are written out */
408 xfs_flush_buftarg(mp->m_ddev_targp, 1);
410 /* mark the log as covered if needed */
411 if (xfs_log_need_covered(mp))
412 error2 = xfs_fs_log_dummy(mp);
414 /* flush data-only devices */
415 if (mp->m_rtdev_targp)
416 xfs_flush_buftarg(mp->m_rtdev_targp, 1);
418 return error ? error : error2;
421 STATIC void
422 xfs_quiesce_fs(
423 struct xfs_mount *mp)
425 int count = 0, pincount;
427 xfs_reclaim_inodes(mp, 0);
428 xfs_flush_buftarg(mp->m_ddev_targp, 0);
431 * This loop must run at least twice. The first instance of the loop
432 * will flush most meta data but that will generate more meta data
433 * (typically directory updates). Which then must be flushed and
434 * logged before we can write the unmount record. We also so sync
435 * reclaim of inodes to catch any that the above delwri flush skipped.
437 do {
438 xfs_reclaim_inodes(mp, SYNC_WAIT);
439 xfs_sync_attr(mp, SYNC_WAIT);
440 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
441 if (!pincount) {
442 delay(50);
443 count++;
445 } while (count < 2);
449 * Second stage of a quiesce. The data is already synced, now we have to take
450 * care of the metadata. New transactions are already blocked, so we need to
451 * wait for any remaining transactions to drain out before proceeding.
453 void
454 xfs_quiesce_attr(
455 struct xfs_mount *mp)
457 int error = 0;
459 /* wait for all modifications to complete */
460 while (atomic_read(&mp->m_active_trans) > 0)
461 delay(100);
463 /* flush inodes and push all remaining buffers out to disk */
464 xfs_quiesce_fs(mp);
467 * Just warn here till VFS can correctly support
468 * read-only remount without racing.
470 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
472 /* Push the superblock and write an unmount record */
473 error = xfs_log_sbcount(mp);
474 if (error)
475 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
476 "Frozen image may not be consistent.");
477 xfs_log_unmount_write(mp);
478 xfs_unmountfs_writesb(mp);
481 static void
482 xfs_syncd_queue_sync(
483 struct xfs_mount *mp)
485 queue_delayed_work(xfs_syncd_wq, &mp->m_sync_work,
486 msecs_to_jiffies(xfs_syncd_centisecs * 10));
490 * Every sync period we need to unpin all items, reclaim inodes and sync
491 * disk quotas. We might need to cover the log to indicate that the
492 * filesystem is idle and not frozen.
494 STATIC void
495 xfs_sync_worker(
496 struct work_struct *work)
498 struct xfs_mount *mp = container_of(to_delayed_work(work),
499 struct xfs_mount, m_sync_work);
500 int error;
502 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
503 /* dgc: errors ignored here */
504 if (mp->m_super->s_frozen == SB_UNFROZEN &&
505 xfs_log_need_covered(mp))
506 error = xfs_fs_log_dummy(mp);
507 else
508 xfs_log_force(mp, 0);
509 error = xfs_qm_sync(mp, SYNC_TRYLOCK);
511 /* start pushing all the metadata that is currently dirty */
512 xfs_ail_push_all(mp->m_ail);
515 /* queue us up again */
516 xfs_syncd_queue_sync(mp);
520 * Queue a new inode reclaim pass if there are reclaimable inodes and there
521 * isn't a reclaim pass already in progress. By default it runs every 5s based
522 * on the xfs syncd work default of 30s. Perhaps this should have it's own
523 * tunable, but that can be done if this method proves to be ineffective or too
524 * aggressive.
526 static void
527 xfs_syncd_queue_reclaim(
528 struct xfs_mount *mp)
532 * We can have inodes enter reclaim after we've shut down the syncd
533 * workqueue during unmount, so don't allow reclaim work to be queued
534 * during unmount.
536 if (!(mp->m_super->s_flags & MS_ACTIVE))
537 return;
539 rcu_read_lock();
540 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
541 queue_delayed_work(xfs_syncd_wq, &mp->m_reclaim_work,
542 msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
544 rcu_read_unlock();
548 * This is a fast pass over the inode cache to try to get reclaim moving on as
549 * many inodes as possible in a short period of time. It kicks itself every few
550 * seconds, as well as being kicked by the inode cache shrinker when memory
551 * goes low. It scans as quickly as possible avoiding locked inodes or those
552 * already being flushed, and once done schedules a future pass.
554 STATIC void
555 xfs_reclaim_worker(
556 struct work_struct *work)
558 struct xfs_mount *mp = container_of(to_delayed_work(work),
559 struct xfs_mount, m_reclaim_work);
561 xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
562 xfs_syncd_queue_reclaim(mp);
566 * Flush delayed allocate data, attempting to free up reserved space
567 * from existing allocations. At this point a new allocation attempt
568 * has failed with ENOSPC and we are in the process of scratching our
569 * heads, looking about for more room.
571 * Queue a new data flush if there isn't one already in progress and
572 * wait for completion of the flush. This means that we only ever have one
573 * inode flush in progress no matter how many ENOSPC events are occurring and
574 * so will prevent the system from bogging down due to every concurrent
575 * ENOSPC event scanning all the active inodes in the system for writeback.
577 void
578 xfs_flush_inodes(
579 struct xfs_inode *ip)
581 struct xfs_mount *mp = ip->i_mount;
583 queue_work(xfs_syncd_wq, &mp->m_flush_work);
584 flush_work_sync(&mp->m_flush_work);
587 STATIC void
588 xfs_flush_worker(
589 struct work_struct *work)
591 struct xfs_mount *mp = container_of(work,
592 struct xfs_mount, m_flush_work);
594 xfs_sync_data(mp, SYNC_TRYLOCK);
595 xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT);
599 xfs_syncd_init(
600 struct xfs_mount *mp)
602 INIT_WORK(&mp->m_flush_work, xfs_flush_worker);
603 INIT_DELAYED_WORK(&mp->m_sync_work, xfs_sync_worker);
604 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
606 xfs_syncd_queue_sync(mp);
607 xfs_syncd_queue_reclaim(mp);
609 return 0;
612 void
613 xfs_syncd_stop(
614 struct xfs_mount *mp)
616 cancel_delayed_work_sync(&mp->m_sync_work);
617 cancel_delayed_work_sync(&mp->m_reclaim_work);
618 cancel_work_sync(&mp->m_flush_work);
621 void
622 __xfs_inode_set_reclaim_tag(
623 struct xfs_perag *pag,
624 struct xfs_inode *ip)
626 radix_tree_tag_set(&pag->pag_ici_root,
627 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
628 XFS_ICI_RECLAIM_TAG);
630 if (!pag->pag_ici_reclaimable) {
631 /* propagate the reclaim tag up into the perag radix tree */
632 spin_lock(&ip->i_mount->m_perag_lock);
633 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
634 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
635 XFS_ICI_RECLAIM_TAG);
636 spin_unlock(&ip->i_mount->m_perag_lock);
638 /* schedule periodic background inode reclaim */
639 xfs_syncd_queue_reclaim(ip->i_mount);
641 trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno,
642 -1, _RET_IP_);
644 pag->pag_ici_reclaimable++;
648 * We set the inode flag atomically with the radix tree tag.
649 * Once we get tag lookups on the radix tree, this inode flag
650 * can go away.
652 void
653 xfs_inode_set_reclaim_tag(
654 xfs_inode_t *ip)
656 struct xfs_mount *mp = ip->i_mount;
657 struct xfs_perag *pag;
659 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
660 spin_lock(&pag->pag_ici_lock);
661 spin_lock(&ip->i_flags_lock);
662 __xfs_inode_set_reclaim_tag(pag, ip);
663 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
664 spin_unlock(&ip->i_flags_lock);
665 spin_unlock(&pag->pag_ici_lock);
666 xfs_perag_put(pag);
669 STATIC void
670 __xfs_inode_clear_reclaim(
671 xfs_perag_t *pag,
672 xfs_inode_t *ip)
674 pag->pag_ici_reclaimable--;
675 if (!pag->pag_ici_reclaimable) {
676 /* clear the reclaim tag from the perag radix tree */
677 spin_lock(&ip->i_mount->m_perag_lock);
678 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
679 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
680 XFS_ICI_RECLAIM_TAG);
681 spin_unlock(&ip->i_mount->m_perag_lock);
682 trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno,
683 -1, _RET_IP_);
687 void
688 __xfs_inode_clear_reclaim_tag(
689 xfs_mount_t *mp,
690 xfs_perag_t *pag,
691 xfs_inode_t *ip)
693 radix_tree_tag_clear(&pag->pag_ici_root,
694 XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG);
695 __xfs_inode_clear_reclaim(pag, ip);
699 * Grab the inode for reclaim exclusively.
700 * Return 0 if we grabbed it, non-zero otherwise.
702 STATIC int
703 xfs_reclaim_inode_grab(
704 struct xfs_inode *ip,
705 int flags)
707 ASSERT(rcu_read_lock_held());
709 /* quick check for stale RCU freed inode */
710 if (!ip->i_ino)
711 return 1;
714 * do some unlocked checks first to avoid unnecessary lock traffic.
715 * The first is a flush lock check, the second is a already in reclaim
716 * check. Only do these checks if we are not going to block on locks.
718 if ((flags & SYNC_TRYLOCK) &&
719 (!ip->i_flush.done || __xfs_iflags_test(ip, XFS_IRECLAIM))) {
720 return 1;
724 * The radix tree lock here protects a thread in xfs_iget from racing
725 * with us starting reclaim on the inode. Once we have the
726 * XFS_IRECLAIM flag set it will not touch us.
728 * Due to RCU lookup, we may find inodes that have been freed and only
729 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
730 * aren't candidates for reclaim at all, so we must check the
731 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
733 spin_lock(&ip->i_flags_lock);
734 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
735 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
736 /* not a reclaim candidate. */
737 spin_unlock(&ip->i_flags_lock);
738 return 1;
740 __xfs_iflags_set(ip, XFS_IRECLAIM);
741 spin_unlock(&ip->i_flags_lock);
742 return 0;
746 * Inodes in different states need to be treated differently, and the return
747 * value of xfs_iflush is not sufficient to get this right. The following table
748 * lists the inode states and the reclaim actions necessary for non-blocking
749 * reclaim:
752 * inode state iflush ret required action
753 * --------------- ---------- ---------------
754 * bad - reclaim
755 * shutdown EIO unpin and reclaim
756 * clean, unpinned 0 reclaim
757 * stale, unpinned 0 reclaim
758 * clean, pinned(*) 0 requeue
759 * stale, pinned EAGAIN requeue
760 * dirty, delwri ok 0 requeue
761 * dirty, delwri blocked EAGAIN requeue
762 * dirty, sync flush 0 reclaim
764 * (*) dgc: I don't think the clean, pinned state is possible but it gets
765 * handled anyway given the order of checks implemented.
767 * As can be seen from the table, the return value of xfs_iflush() is not
768 * sufficient to correctly decide the reclaim action here. The checks in
769 * xfs_iflush() might look like duplicates, but they are not.
771 * Also, because we get the flush lock first, we know that any inode that has
772 * been flushed delwri has had the flush completed by the time we check that
773 * the inode is clean. The clean inode check needs to be done before flushing
774 * the inode delwri otherwise we would loop forever requeuing clean inodes as
775 * we cannot tell apart a successful delwri flush and a clean inode from the
776 * return value of xfs_iflush().
778 * Note that because the inode is flushed delayed write by background
779 * writeback, the flush lock may already be held here and waiting on it can
780 * result in very long latencies. Hence for sync reclaims, where we wait on the
781 * flush lock, the caller should push out delayed write inodes first before
782 * trying to reclaim them to minimise the amount of time spent waiting. For
783 * background relaim, we just requeue the inode for the next pass.
785 * Hence the order of actions after gaining the locks should be:
786 * bad => reclaim
787 * shutdown => unpin and reclaim
788 * pinned, delwri => requeue
789 * pinned, sync => unpin
790 * stale => reclaim
791 * clean => reclaim
792 * dirty, delwri => flush and requeue
793 * dirty, sync => flush, wait and reclaim
795 STATIC int
796 xfs_reclaim_inode(
797 struct xfs_inode *ip,
798 struct xfs_perag *pag,
799 int sync_mode)
801 int error;
803 restart:
804 error = 0;
805 xfs_ilock(ip, XFS_ILOCK_EXCL);
806 if (!xfs_iflock_nowait(ip)) {
807 if (!(sync_mode & SYNC_WAIT))
808 goto out;
811 * If we only have a single dirty inode in a cluster there is
812 * a fair chance that the AIL push may have pushed it into
813 * the buffer, but xfsbufd won't touch it until 30 seconds
814 * from now, and thus we will lock up here.
816 * Promote the inode buffer to the front of the delwri list
817 * and wake up xfsbufd now.
819 xfs_promote_inode(ip);
820 xfs_iflock(ip);
823 if (is_bad_inode(VFS_I(ip)))
824 goto reclaim;
825 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
826 xfs_iunpin_wait(ip);
827 goto reclaim;
829 if (xfs_ipincount(ip)) {
830 if (!(sync_mode & SYNC_WAIT)) {
831 xfs_ifunlock(ip);
832 goto out;
834 xfs_iunpin_wait(ip);
836 if (xfs_iflags_test(ip, XFS_ISTALE))
837 goto reclaim;
838 if (xfs_inode_clean(ip))
839 goto reclaim;
842 * Now we have an inode that needs flushing.
844 * We do a nonblocking flush here even if we are doing a SYNC_WAIT
845 * reclaim as we can deadlock with inode cluster removal.
846 * xfs_ifree_cluster() can lock the inode buffer before it locks the
847 * ip->i_lock, and we are doing the exact opposite here. As a result,
848 * doing a blocking xfs_itobp() to get the cluster buffer will result
849 * in an ABBA deadlock with xfs_ifree_cluster().
851 * As xfs_ifree_cluser() must gather all inodes that are active in the
852 * cache to mark them stale, if we hit this case we don't actually want
853 * to do IO here - we want the inode marked stale so we can simply
854 * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush,
855 * just unlock the inode, back off and try again. Hopefully the next
856 * pass through will see the stale flag set on the inode.
858 error = xfs_iflush(ip, SYNC_TRYLOCK | sync_mode);
859 if (sync_mode & SYNC_WAIT) {
860 if (error == EAGAIN) {
861 xfs_iunlock(ip, XFS_ILOCK_EXCL);
862 /* backoff longer than in xfs_ifree_cluster */
863 delay(2);
864 goto restart;
866 xfs_iflock(ip);
867 goto reclaim;
871 * When we have to flush an inode but don't have SYNC_WAIT set, we
872 * flush the inode out using a delwri buffer and wait for the next
873 * call into reclaim to find it in a clean state instead of waiting for
874 * it now. We also don't return errors here - if the error is transient
875 * then the next reclaim pass will flush the inode, and if the error
876 * is permanent then the next sync reclaim will reclaim the inode and
877 * pass on the error.
879 if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) {
880 xfs_warn(ip->i_mount,
881 "inode 0x%llx background reclaim flush failed with %d",
882 (long long)ip->i_ino, error);
884 out:
885 xfs_iflags_clear(ip, XFS_IRECLAIM);
886 xfs_iunlock(ip, XFS_ILOCK_EXCL);
888 * We could return EAGAIN here to make reclaim rescan the inode tree in
889 * a short while. However, this just burns CPU time scanning the tree
890 * waiting for IO to complete and xfssyncd never goes back to the idle
891 * state. Instead, return 0 to let the next scheduled background reclaim
892 * attempt to reclaim the inode again.
894 return 0;
896 reclaim:
897 xfs_ifunlock(ip);
898 xfs_iunlock(ip, XFS_ILOCK_EXCL);
900 XFS_STATS_INC(xs_ig_reclaims);
902 * Remove the inode from the per-AG radix tree.
904 * Because radix_tree_delete won't complain even if the item was never
905 * added to the tree assert that it's been there before to catch
906 * problems with the inode life time early on.
908 spin_lock(&pag->pag_ici_lock);
909 if (!radix_tree_delete(&pag->pag_ici_root,
910 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino)))
911 ASSERT(0);
912 __xfs_inode_clear_reclaim(pag, ip);
913 spin_unlock(&pag->pag_ici_lock);
916 * Here we do an (almost) spurious inode lock in order to coordinate
917 * with inode cache radix tree lookups. This is because the lookup
918 * can reference the inodes in the cache without taking references.
920 * We make that OK here by ensuring that we wait until the inode is
921 * unlocked after the lookup before we go ahead and free it. We get
922 * both the ilock and the iolock because the code may need to drop the
923 * ilock one but will still hold the iolock.
925 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
926 xfs_qm_dqdetach(ip);
927 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
929 xfs_inode_free(ip);
930 return error;
935 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
936 * corrupted, we still want to try to reclaim all the inodes. If we don't,
937 * then a shut down during filesystem unmount reclaim walk leak all the
938 * unreclaimed inodes.
941 xfs_reclaim_inodes_ag(
942 struct xfs_mount *mp,
943 int flags,
944 int *nr_to_scan)
946 struct xfs_perag *pag;
947 int error = 0;
948 int last_error = 0;
949 xfs_agnumber_t ag;
950 int trylock = flags & SYNC_TRYLOCK;
951 int skipped;
953 restart:
954 ag = 0;
955 skipped = 0;
956 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
957 unsigned long first_index = 0;
958 int done = 0;
959 int nr_found = 0;
961 ag = pag->pag_agno + 1;
963 if (trylock) {
964 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
965 skipped++;
966 xfs_perag_put(pag);
967 continue;
969 first_index = pag->pag_ici_reclaim_cursor;
970 } else
971 mutex_lock(&pag->pag_ici_reclaim_lock);
973 do {
974 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
975 int i;
977 rcu_read_lock();
978 nr_found = radix_tree_gang_lookup_tag(
979 &pag->pag_ici_root,
980 (void **)batch, first_index,
981 XFS_LOOKUP_BATCH,
982 XFS_ICI_RECLAIM_TAG);
983 if (!nr_found) {
984 done = 1;
985 rcu_read_unlock();
986 break;
990 * Grab the inodes before we drop the lock. if we found
991 * nothing, nr == 0 and the loop will be skipped.
993 for (i = 0; i < nr_found; i++) {
994 struct xfs_inode *ip = batch[i];
996 if (done || xfs_reclaim_inode_grab(ip, flags))
997 batch[i] = NULL;
1000 * Update the index for the next lookup. Catch
1001 * overflows into the next AG range which can
1002 * occur if we have inodes in the last block of
1003 * the AG and we are currently pointing to the
1004 * last inode.
1006 * Because we may see inodes that are from the
1007 * wrong AG due to RCU freeing and
1008 * reallocation, only update the index if it
1009 * lies in this AG. It was a race that lead us
1010 * to see this inode, so another lookup from
1011 * the same index will not find it again.
1013 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1014 pag->pag_agno)
1015 continue;
1016 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1017 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1018 done = 1;
1021 /* unlock now we've grabbed the inodes. */
1022 rcu_read_unlock();
1024 for (i = 0; i < nr_found; i++) {
1025 if (!batch[i])
1026 continue;
1027 error = xfs_reclaim_inode(batch[i], pag, flags);
1028 if (error && last_error != EFSCORRUPTED)
1029 last_error = error;
1032 *nr_to_scan -= XFS_LOOKUP_BATCH;
1034 cond_resched();
1036 } while (nr_found && !done && *nr_to_scan > 0);
1038 if (trylock && !done)
1039 pag->pag_ici_reclaim_cursor = first_index;
1040 else
1041 pag->pag_ici_reclaim_cursor = 0;
1042 mutex_unlock(&pag->pag_ici_reclaim_lock);
1043 xfs_perag_put(pag);
1047 * if we skipped any AG, and we still have scan count remaining, do
1048 * another pass this time using blocking reclaim semantics (i.e
1049 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1050 * ensure that when we get more reclaimers than AGs we block rather
1051 * than spin trying to execute reclaim.
1053 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
1054 trylock = 0;
1055 goto restart;
1057 return XFS_ERROR(last_error);
1061 xfs_reclaim_inodes(
1062 xfs_mount_t *mp,
1063 int mode)
1065 int nr_to_scan = INT_MAX;
1067 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
1071 * Scan a certain number of inodes for reclaim.
1073 * When called we make sure that there is a background (fast) inode reclaim in
1074 * progress, while we will throttle the speed of reclaim via doing synchronous
1075 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1076 * them to be cleaned, which we hope will not be very long due to the
1077 * background walker having already kicked the IO off on those dirty inodes.
1079 void
1080 xfs_reclaim_inodes_nr(
1081 struct xfs_mount *mp,
1082 int nr_to_scan)
1084 /* kick background reclaimer and push the AIL */
1085 xfs_syncd_queue_reclaim(mp);
1086 xfs_ail_push_all(mp->m_ail);
1088 xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
1092 * Return the number of reclaimable inodes in the filesystem for
1093 * the shrinker to determine how much to reclaim.
1096 xfs_reclaim_inodes_count(
1097 struct xfs_mount *mp)
1099 struct xfs_perag *pag;
1100 xfs_agnumber_t ag = 0;
1101 int reclaimable = 0;
1103 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1104 ag = pag->pag_agno + 1;
1105 reclaimable += pag->pag_ici_reclaimable;
1106 xfs_perag_put(pag);
1108 return reclaimable;