]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. | |
3 | * All Rights Reserved. | |
4 | * | |
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. | |
8 | * | |
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. | |
13 | * | |
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 | |
17 | */ | |
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_sb.h" | |
26 | #include "xfs_ag.h" | |
27 | #include "xfs_dir2.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" | |
45 | #include "xfs_rw.h" | |
46 | #include "xfs_quota.h" | |
47 | #include "xfs_trace.h" | |
48 | ||
49 | #include <linux/kthread.h> | |
50 | #include <linux/freezer.h> | |
51 | ||
52 | ||
53 | STATIC xfs_inode_t * | |
54 | xfs_inode_ag_lookup( | |
55 | struct xfs_mount *mp, | |
56 | struct xfs_perag *pag, | |
57 | uint32_t *first_index, | |
58 | int tag) | |
59 | { | |
60 | int nr_found; | |
61 | struct xfs_inode *ip; | |
62 | ||
63 | /* | |
64 | * use a gang lookup to find the next inode in the tree | |
65 | * as the tree is sparse and a gang lookup walks to find | |
66 | * the number of objects requested. | |
67 | */ | |
68 | if (tag == XFS_ICI_NO_TAG) { | |
69 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
70 | (void **)&ip, *first_index, 1); | |
71 | } else { | |
72 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, | |
73 | (void **)&ip, *first_index, 1, tag); | |
74 | } | |
75 | if (!nr_found) | |
76 | return NULL; | |
77 | ||
78 | /* | |
79 | * Update the index for the next lookup. Catch overflows | |
80 | * into the next AG range which can occur if we have inodes | |
81 | * in the last block of the AG and we are currently | |
82 | * pointing to the last inode. | |
83 | */ | |
84 | *first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
85 | if (*first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
86 | return NULL; | |
87 | return ip; | |
88 | } | |
89 | ||
90 | STATIC int | |
91 | xfs_inode_ag_walk( | |
92 | struct xfs_mount *mp, | |
93 | struct xfs_perag *pag, | |
94 | int (*execute)(struct xfs_inode *ip, | |
95 | struct xfs_perag *pag, int flags), | |
96 | int flags, | |
97 | int tag, | |
98 | int exclusive, | |
99 | int *nr_to_scan) | |
100 | { | |
101 | uint32_t first_index; | |
102 | int last_error = 0; | |
103 | int skipped; | |
104 | ||
105 | restart: | |
106 | skipped = 0; | |
107 | first_index = 0; | |
108 | do { | |
109 | int error = 0; | |
110 | xfs_inode_t *ip; | |
111 | ||
112 | if (exclusive) | |
113 | write_lock(&pag->pag_ici_lock); | |
114 | else | |
115 | read_lock(&pag->pag_ici_lock); | |
116 | ip = xfs_inode_ag_lookup(mp, pag, &first_index, tag); | |
117 | if (!ip) { | |
118 | if (exclusive) | |
119 | write_unlock(&pag->pag_ici_lock); | |
120 | else | |
121 | read_unlock(&pag->pag_ici_lock); | |
122 | break; | |
123 | } | |
124 | ||
125 | /* execute releases pag->pag_ici_lock */ | |
126 | error = execute(ip, pag, flags); | |
127 | if (error == EAGAIN) { | |
128 | skipped++; | |
129 | continue; | |
130 | } | |
131 | if (error) | |
132 | last_error = error; | |
133 | ||
134 | /* bail out if the filesystem is corrupted. */ | |
135 | if (error == EFSCORRUPTED) | |
136 | break; | |
137 | ||
138 | } while ((*nr_to_scan)--); | |
139 | ||
140 | if (skipped) { | |
141 | delay(1); | |
142 | goto restart; | |
143 | } | |
144 | return last_error; | |
145 | } | |
146 | ||
147 | int | |
148 | xfs_inode_ag_iterator( | |
149 | struct xfs_mount *mp, | |
150 | int (*execute)(struct xfs_inode *ip, | |
151 | struct xfs_perag *pag, int flags), | |
152 | int flags, | |
153 | int tag, | |
154 | int exclusive, | |
155 | int *nr_to_scan) | |
156 | { | |
157 | int error = 0; | |
158 | int last_error = 0; | |
159 | xfs_agnumber_t ag; | |
160 | int nr; | |
161 | ||
162 | nr = nr_to_scan ? *nr_to_scan : INT_MAX; | |
163 | for (ag = 0; ag < mp->m_sb.sb_agcount; ag++) { | |
164 | struct xfs_perag *pag; | |
165 | ||
166 | pag = xfs_perag_get(mp, ag); | |
167 | error = xfs_inode_ag_walk(mp, pag, execute, flags, tag, | |
168 | exclusive, &nr); | |
169 | xfs_perag_put(pag); | |
170 | if (error) { | |
171 | last_error = error; | |
172 | if (error == EFSCORRUPTED) | |
173 | break; | |
174 | } | |
175 | if (nr <= 0) | |
176 | break; | |
177 | } | |
178 | if (nr_to_scan) | |
179 | *nr_to_scan = nr; | |
180 | return XFS_ERROR(last_error); | |
181 | } | |
182 | ||
183 | /* must be called with pag_ici_lock held and releases it */ | |
184 | int | |
185 | xfs_sync_inode_valid( | |
186 | struct xfs_inode *ip, | |
187 | struct xfs_perag *pag) | |
188 | { | |
189 | struct inode *inode = VFS_I(ip); | |
190 | int error = EFSCORRUPTED; | |
191 | ||
192 | /* nothing to sync during shutdown */ | |
193 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
194 | goto out_unlock; | |
195 | ||
196 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
197 | error = ENOENT; | |
198 | if (xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
199 | goto out_unlock; | |
200 | ||
201 | /* If we can't grab the inode, it must on it's way to reclaim. */ | |
202 | if (!igrab(inode)) | |
203 | goto out_unlock; | |
204 | ||
205 | if (is_bad_inode(inode)) { | |
206 | IRELE(ip); | |
207 | goto out_unlock; | |
208 | } | |
209 | ||
210 | /* inode is valid */ | |
211 | error = 0; | |
212 | out_unlock: | |
213 | read_unlock(&pag->pag_ici_lock); | |
214 | return error; | |
215 | } | |
216 | ||
217 | STATIC int | |
218 | xfs_sync_inode_data( | |
219 | struct xfs_inode *ip, | |
220 | struct xfs_perag *pag, | |
221 | int flags) | |
222 | { | |
223 | struct inode *inode = VFS_I(ip); | |
224 | struct address_space *mapping = inode->i_mapping; | |
225 | int error = 0; | |
226 | ||
227 | error = xfs_sync_inode_valid(ip, pag); | |
228 | if (error) | |
229 | return error; | |
230 | ||
231 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
232 | goto out_wait; | |
233 | ||
234 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
235 | if (flags & SYNC_TRYLOCK) | |
236 | goto out_wait; | |
237 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
238 | } | |
239 | ||
240 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? | |
241 | 0 : XBF_ASYNC, FI_NONE); | |
242 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
243 | ||
244 | out_wait: | |
245 | if (flags & SYNC_WAIT) | |
246 | xfs_ioend_wait(ip); | |
247 | IRELE(ip); | |
248 | return error; | |
249 | } | |
250 | ||
251 | STATIC int | |
252 | xfs_sync_inode_attr( | |
253 | struct xfs_inode *ip, | |
254 | struct xfs_perag *pag, | |
255 | int flags) | |
256 | { | |
257 | int error = 0; | |
258 | ||
259 | error = xfs_sync_inode_valid(ip, pag); | |
260 | if (error) | |
261 | return error; | |
262 | ||
263 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
264 | if (xfs_inode_clean(ip)) | |
265 | goto out_unlock; | |
266 | if (!xfs_iflock_nowait(ip)) { | |
267 | if (!(flags & SYNC_WAIT)) | |
268 | goto out_unlock; | |
269 | xfs_iflock(ip); | |
270 | } | |
271 | ||
272 | if (xfs_inode_clean(ip)) { | |
273 | xfs_ifunlock(ip); | |
274 | goto out_unlock; | |
275 | } | |
276 | ||
277 | error = xfs_iflush(ip, flags); | |
278 | ||
279 | out_unlock: | |
280 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
281 | IRELE(ip); | |
282 | return error; | |
283 | } | |
284 | ||
285 | /* | |
286 | * Write out pagecache data for the whole filesystem. | |
287 | */ | |
288 | int | |
289 | xfs_sync_data( | |
290 | struct xfs_mount *mp, | |
291 | int flags) | |
292 | { | |
293 | int error; | |
294 | ||
295 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); | |
296 | ||
297 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags, | |
298 | XFS_ICI_NO_TAG, 0, NULL); | |
299 | if (error) | |
300 | return XFS_ERROR(error); | |
301 | ||
302 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); | |
303 | return 0; | |
304 | } | |
305 | ||
306 | /* | |
307 | * Write out inode metadata (attributes) for the whole filesystem. | |
308 | */ | |
309 | int | |
310 | xfs_sync_attr( | |
311 | struct xfs_mount *mp, | |
312 | int flags) | |
313 | { | |
314 | ASSERT((flags & ~SYNC_WAIT) == 0); | |
315 | ||
316 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags, | |
317 | XFS_ICI_NO_TAG, 0, NULL); | |
318 | } | |
319 | ||
320 | STATIC int | |
321 | xfs_commit_dummy_trans( | |
322 | struct xfs_mount *mp, | |
323 | uint flags) | |
324 | { | |
325 | struct xfs_inode *ip = mp->m_rootip; | |
326 | struct xfs_trans *tp; | |
327 | int error; | |
328 | ||
329 | /* | |
330 | * Put a dummy transaction in the log to tell recovery | |
331 | * that all others are OK. | |
332 | */ | |
333 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
334 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
335 | if (error) { | |
336 | xfs_trans_cancel(tp, 0); | |
337 | return error; | |
338 | } | |
339 | ||
340 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
341 | ||
342 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
343 | xfs_trans_ihold(tp, ip); | |
344 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
345 | error = xfs_trans_commit(tp, 0); | |
346 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
347 | ||
348 | /* the log force ensures this transaction is pushed to disk */ | |
349 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); | |
350 | return error; | |
351 | } | |
352 | ||
353 | STATIC int | |
354 | xfs_sync_fsdata( | |
355 | struct xfs_mount *mp) | |
356 | { | |
357 | struct xfs_buf *bp; | |
358 | ||
359 | /* | |
360 | * If the buffer is pinned then push on the log so we won't get stuck | |
361 | * waiting in the write for someone, maybe ourselves, to flush the log. | |
362 | * | |
363 | * Even though we just pushed the log above, we did not have the | |
364 | * superblock buffer locked at that point so it can become pinned in | |
365 | * between there and here. | |
366 | */ | |
367 | bp = xfs_getsb(mp, 0); | |
368 | if (XFS_BUF_ISPINNED(bp)) | |
369 | xfs_log_force(mp, 0); | |
370 | ||
371 | return xfs_bwrite(mp, bp); | |
372 | } | |
373 | ||
374 | /* | |
375 | * When remounting a filesystem read-only or freezing the filesystem, we have | |
376 | * two phases to execute. This first phase is syncing the data before we | |
377 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
378 | * we've waited for all the transactions created by the first phase to | |
379 | * complete. The second phase ensures that the inodes are written to their | |
380 | * location on disk rather than just existing in transactions in the log. This | |
381 | * means after a quiesce there is no log replay required to write the inodes to | |
382 | * disk (this is the main difference between a sync and a quiesce). | |
383 | */ | |
384 | /* | |
385 | * First stage of freeze - no writers will make progress now we are here, | |
386 | * so we flush delwri and delalloc buffers here, then wait for all I/O to | |
387 | * complete. Data is frozen at that point. Metadata is not frozen, | |
388 | * transactions can still occur here so don't bother flushing the buftarg | |
389 | * because it'll just get dirty again. | |
390 | */ | |
391 | int | |
392 | xfs_quiesce_data( | |
393 | struct xfs_mount *mp) | |
394 | { | |
395 | int error, error2 = 0; | |
396 | ||
397 | /* push non-blocking */ | |
398 | xfs_sync_data(mp, 0); | |
399 | xfs_qm_sync(mp, SYNC_TRYLOCK); | |
400 | ||
401 | /* push and block till complete */ | |
402 | xfs_sync_data(mp, SYNC_WAIT); | |
403 | xfs_qm_sync(mp, SYNC_WAIT); | |
404 | ||
405 | /* write superblock and hoover up shutdown errors */ | |
406 | error = xfs_sync_fsdata(mp); | |
407 | ||
408 | /* make sure all delwri buffers are written out */ | |
409 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
410 | ||
411 | /* mark the log as covered if needed */ | |
412 | if (xfs_log_need_covered(mp)) | |
413 | error2 = xfs_commit_dummy_trans(mp, SYNC_WAIT); | |
414 | ||
415 | /* flush data-only devices */ | |
416 | if (mp->m_rtdev_targp) | |
417 | XFS_bflush(mp->m_rtdev_targp); | |
418 | ||
419 | return error ? error : error2; | |
420 | } | |
421 | ||
422 | STATIC void | |
423 | xfs_quiesce_fs( | |
424 | struct xfs_mount *mp) | |
425 | { | |
426 | int count = 0, pincount; | |
427 | ||
428 | xfs_reclaim_inodes(mp, 0); | |
429 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
430 | ||
431 | /* | |
432 | * This loop must run at least twice. The first instance of the loop | |
433 | * will flush most meta data but that will generate more meta data | |
434 | * (typically directory updates). Which then must be flushed and | |
435 | * logged before we can write the unmount record. We also so sync | |
436 | * reclaim of inodes to catch any that the above delwri flush skipped. | |
437 | */ | |
438 | do { | |
439 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
440 | xfs_sync_attr(mp, SYNC_WAIT); | |
441 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
442 | if (!pincount) { | |
443 | delay(50); | |
444 | count++; | |
445 | } | |
446 | } while (count < 2); | |
447 | } | |
448 | ||
449 | /* | |
450 | * Second stage of a quiesce. The data is already synced, now we have to take | |
451 | * care of the metadata. New transactions are already blocked, so we need to | |
452 | * wait for any remaining transactions to drain out before proceding. | |
453 | */ | |
454 | void | |
455 | xfs_quiesce_attr( | |
456 | struct xfs_mount *mp) | |
457 | { | |
458 | int error = 0; | |
459 | ||
460 | /* wait for all modifications to complete */ | |
461 | while (atomic_read(&mp->m_active_trans) > 0) | |
462 | delay(100); | |
463 | ||
464 | /* flush inodes and push all remaining buffers out to disk */ | |
465 | xfs_quiesce_fs(mp); | |
466 | ||
467 | /* | |
468 | * Just warn here till VFS can correctly support | |
469 | * read-only remount without racing. | |
470 | */ | |
471 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
472 | ||
473 | /* Push the superblock and write an unmount record */ | |
474 | error = xfs_log_sbcount(mp, 1); | |
475 | if (error) | |
476 | xfs_fs_cmn_err(CE_WARN, mp, | |
477 | "xfs_attr_quiesce: failed to log sb changes. " | |
478 | "Frozen image may not be consistent."); | |
479 | xfs_log_unmount_write(mp); | |
480 | xfs_unmountfs_writesb(mp); | |
481 | } | |
482 | ||
483 | /* | |
484 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
485 | * Doing this has two advantages: | |
486 | * - It saves on stack space, which is tight in certain situations | |
487 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
488 | * Flushing while allocating in a full filesystem requires both. | |
489 | */ | |
490 | STATIC void | |
491 | xfs_syncd_queue_work( | |
492 | struct xfs_mount *mp, | |
493 | void *data, | |
494 | void (*syncer)(struct xfs_mount *, void *), | |
495 | struct completion *completion) | |
496 | { | |
497 | struct xfs_sync_work *work; | |
498 | ||
499 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); | |
500 | INIT_LIST_HEAD(&work->w_list); | |
501 | work->w_syncer = syncer; | |
502 | work->w_data = data; | |
503 | work->w_mount = mp; | |
504 | work->w_completion = completion; | |
505 | spin_lock(&mp->m_sync_lock); | |
506 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
507 | spin_unlock(&mp->m_sync_lock); | |
508 | wake_up_process(mp->m_sync_task); | |
509 | } | |
510 | ||
511 | /* | |
512 | * Flush delayed allocate data, attempting to free up reserved space | |
513 | * from existing allocations. At this point a new allocation attempt | |
514 | * has failed with ENOSPC and we are in the process of scratching our | |
515 | * heads, looking about for more room... | |
516 | */ | |
517 | STATIC void | |
518 | xfs_flush_inodes_work( | |
519 | struct xfs_mount *mp, | |
520 | void *arg) | |
521 | { | |
522 | struct inode *inode = arg; | |
523 | xfs_sync_data(mp, SYNC_TRYLOCK); | |
524 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); | |
525 | iput(inode); | |
526 | } | |
527 | ||
528 | void | |
529 | xfs_flush_inodes( | |
530 | xfs_inode_t *ip) | |
531 | { | |
532 | struct inode *inode = VFS_I(ip); | |
533 | DECLARE_COMPLETION_ONSTACK(completion); | |
534 | ||
535 | igrab(inode); | |
536 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); | |
537 | wait_for_completion(&completion); | |
538 | xfs_log_force(ip->i_mount, XFS_LOG_SYNC); | |
539 | } | |
540 | ||
541 | /* | |
542 | * Every sync period we need to unpin all items, reclaim inodes and sync | |
543 | * disk quotas. We might need to cover the log to indicate that the | |
544 | * filesystem is idle. | |
545 | */ | |
546 | STATIC void | |
547 | xfs_sync_worker( | |
548 | struct xfs_mount *mp, | |
549 | void *unused) | |
550 | { | |
551 | int error; | |
552 | ||
553 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { | |
554 | xfs_log_force(mp, 0); | |
555 | xfs_reclaim_inodes(mp, 0); | |
556 | /* dgc: errors ignored here */ | |
557 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); | |
558 | if (xfs_log_need_covered(mp)) | |
559 | error = xfs_commit_dummy_trans(mp, 0); | |
560 | } | |
561 | mp->m_sync_seq++; | |
562 | wake_up(&mp->m_wait_single_sync_task); | |
563 | } | |
564 | ||
565 | STATIC int | |
566 | xfssyncd( | |
567 | void *arg) | |
568 | { | |
569 | struct xfs_mount *mp = arg; | |
570 | long timeleft; | |
571 | xfs_sync_work_t *work, *n; | |
572 | LIST_HEAD (tmp); | |
573 | ||
574 | set_freezable(); | |
575 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
576 | for (;;) { | |
577 | if (list_empty(&mp->m_sync_list)) | |
578 | timeleft = schedule_timeout_interruptible(timeleft); | |
579 | /* swsusp */ | |
580 | try_to_freeze(); | |
581 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
582 | break; | |
583 | ||
584 | spin_lock(&mp->m_sync_lock); | |
585 | /* | |
586 | * We can get woken by laptop mode, to do a sync - | |
587 | * that's the (only!) case where the list would be | |
588 | * empty with time remaining. | |
589 | */ | |
590 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
591 | if (!timeleft) | |
592 | timeleft = xfs_syncd_centisecs * | |
593 | msecs_to_jiffies(10); | |
594 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
595 | list_add_tail(&mp->m_sync_work.w_list, | |
596 | &mp->m_sync_list); | |
597 | } | |
598 | list_splice_init(&mp->m_sync_list, &tmp); | |
599 | spin_unlock(&mp->m_sync_lock); | |
600 | ||
601 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
602 | (*work->w_syncer)(mp, work->w_data); | |
603 | list_del(&work->w_list); | |
604 | if (work == &mp->m_sync_work) | |
605 | continue; | |
606 | if (work->w_completion) | |
607 | complete(work->w_completion); | |
608 | kmem_free(work); | |
609 | } | |
610 | } | |
611 | ||
612 | return 0; | |
613 | } | |
614 | ||
615 | int | |
616 | xfs_syncd_init( | |
617 | struct xfs_mount *mp) | |
618 | { | |
619 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
620 | mp->m_sync_work.w_mount = mp; | |
621 | mp->m_sync_work.w_completion = NULL; | |
622 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd/%s", mp->m_fsname); | |
623 | if (IS_ERR(mp->m_sync_task)) | |
624 | return -PTR_ERR(mp->m_sync_task); | |
625 | return 0; | |
626 | } | |
627 | ||
628 | void | |
629 | xfs_syncd_stop( | |
630 | struct xfs_mount *mp) | |
631 | { | |
632 | kthread_stop(mp->m_sync_task); | |
633 | } | |
634 | ||
635 | void | |
636 | __xfs_inode_set_reclaim_tag( | |
637 | struct xfs_perag *pag, | |
638 | struct xfs_inode *ip) | |
639 | { | |
640 | radix_tree_tag_set(&pag->pag_ici_root, | |
641 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
642 | XFS_ICI_RECLAIM_TAG); | |
643 | pag->pag_ici_reclaimable++; | |
644 | } | |
645 | ||
646 | /* | |
647 | * We set the inode flag atomically with the radix tree tag. | |
648 | * Once we get tag lookups on the radix tree, this inode flag | |
649 | * can go away. | |
650 | */ | |
651 | void | |
652 | xfs_inode_set_reclaim_tag( | |
653 | xfs_inode_t *ip) | |
654 | { | |
655 | struct xfs_mount *mp = ip->i_mount; | |
656 | struct xfs_perag *pag; | |
657 | ||
658 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
659 | write_lock(&pag->pag_ici_lock); | |
660 | spin_lock(&ip->i_flags_lock); | |
661 | __xfs_inode_set_reclaim_tag(pag, ip); | |
662 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
663 | spin_unlock(&ip->i_flags_lock); | |
664 | write_unlock(&pag->pag_ici_lock); | |
665 | xfs_perag_put(pag); | |
666 | } | |
667 | ||
668 | void | |
669 | __xfs_inode_clear_reclaim_tag( | |
670 | xfs_mount_t *mp, | |
671 | xfs_perag_t *pag, | |
672 | xfs_inode_t *ip) | |
673 | { | |
674 | radix_tree_tag_clear(&pag->pag_ici_root, | |
675 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
676 | pag->pag_ici_reclaimable--; | |
677 | } | |
678 | ||
679 | /* | |
680 | * Inodes in different states need to be treated differently, and the return | |
681 | * value of xfs_iflush is not sufficient to get this right. The following table | |
682 | * lists the inode states and the reclaim actions necessary for non-blocking | |
683 | * reclaim: | |
684 | * | |
685 | * | |
686 | * inode state iflush ret required action | |
687 | * --------------- ---------- --------------- | |
688 | * bad - reclaim | |
689 | * shutdown EIO unpin and reclaim | |
690 | * clean, unpinned 0 reclaim | |
691 | * stale, unpinned 0 reclaim | |
692 | * clean, pinned(*) 0 requeue | |
693 | * stale, pinned EAGAIN requeue | |
694 | * dirty, delwri ok 0 requeue | |
695 | * dirty, delwri blocked EAGAIN requeue | |
696 | * dirty, sync flush 0 reclaim | |
697 | * | |
698 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
699 | * handled anyway given the order of checks implemented. | |
700 | * | |
701 | * As can be seen from the table, the return value of xfs_iflush() is not | |
702 | * sufficient to correctly decide the reclaim action here. The checks in | |
703 | * xfs_iflush() might look like duplicates, but they are not. | |
704 | * | |
705 | * Also, because we get the flush lock first, we know that any inode that has | |
706 | * been flushed delwri has had the flush completed by the time we check that | |
707 | * the inode is clean. The clean inode check needs to be done before flushing | |
708 | * the inode delwri otherwise we would loop forever requeuing clean inodes as | |
709 | * we cannot tell apart a successful delwri flush and a clean inode from the | |
710 | * return value of xfs_iflush(). | |
711 | * | |
712 | * Note that because the inode is flushed delayed write by background | |
713 | * writeback, the flush lock may already be held here and waiting on it can | |
714 | * result in very long latencies. Hence for sync reclaims, where we wait on the | |
715 | * flush lock, the caller should push out delayed write inodes first before | |
716 | * trying to reclaim them to minimise the amount of time spent waiting. For | |
717 | * background relaim, we just requeue the inode for the next pass. | |
718 | * | |
719 | * Hence the order of actions after gaining the locks should be: | |
720 | * bad => reclaim | |
721 | * shutdown => unpin and reclaim | |
722 | * pinned, delwri => requeue | |
723 | * pinned, sync => unpin | |
724 | * stale => reclaim | |
725 | * clean => reclaim | |
726 | * dirty, delwri => flush and requeue | |
727 | * dirty, sync => flush, wait and reclaim | |
728 | */ | |
729 | STATIC int | |
730 | xfs_reclaim_inode( | |
731 | struct xfs_inode *ip, | |
732 | struct xfs_perag *pag, | |
733 | int sync_mode) | |
734 | { | |
735 | int error = 0; | |
736 | ||
737 | /* | |
738 | * The radix tree lock here protects a thread in xfs_iget from racing | |
739 | * with us starting reclaim on the inode. Once we have the | |
740 | * XFS_IRECLAIM flag set it will not touch us. | |
741 | */ | |
742 | spin_lock(&ip->i_flags_lock); | |
743 | ASSERT_ALWAYS(__xfs_iflags_test(ip, XFS_IRECLAIMABLE)); | |
744 | if (__xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
745 | /* ignore as it is already under reclaim */ | |
746 | spin_unlock(&ip->i_flags_lock); | |
747 | write_unlock(&pag->pag_ici_lock); | |
748 | return 0; | |
749 | } | |
750 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
751 | spin_unlock(&ip->i_flags_lock); | |
752 | write_unlock(&pag->pag_ici_lock); | |
753 | ||
754 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
755 | if (!xfs_iflock_nowait(ip)) { | |
756 | if (!(sync_mode & SYNC_WAIT)) | |
757 | goto out; | |
758 | xfs_iflock(ip); | |
759 | } | |
760 | ||
761 | if (is_bad_inode(VFS_I(ip))) | |
762 | goto reclaim; | |
763 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
764 | xfs_iunpin_wait(ip); | |
765 | goto reclaim; | |
766 | } | |
767 | if (xfs_ipincount(ip)) { | |
768 | if (!(sync_mode & SYNC_WAIT)) { | |
769 | xfs_ifunlock(ip); | |
770 | goto out; | |
771 | } | |
772 | xfs_iunpin_wait(ip); | |
773 | } | |
774 | if (xfs_iflags_test(ip, XFS_ISTALE)) | |
775 | goto reclaim; | |
776 | if (xfs_inode_clean(ip)) | |
777 | goto reclaim; | |
778 | ||
779 | /* Now we have an inode that needs flushing */ | |
780 | error = xfs_iflush(ip, sync_mode); | |
781 | if (sync_mode & SYNC_WAIT) { | |
782 | xfs_iflock(ip); | |
783 | goto reclaim; | |
784 | } | |
785 | ||
786 | /* | |
787 | * When we have to flush an inode but don't have SYNC_WAIT set, we | |
788 | * flush the inode out using a delwri buffer and wait for the next | |
789 | * call into reclaim to find it in a clean state instead of waiting for | |
790 | * it now. We also don't return errors here - if the error is transient | |
791 | * then the next reclaim pass will flush the inode, and if the error | |
792 | * is permanent then the next sync reclaim will reclaim the inode and | |
793 | * pass on the error. | |
794 | */ | |
795 | if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
796 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
797 | "inode 0x%llx background reclaim flush failed with %d", | |
798 | (long long)ip->i_ino, error); | |
799 | } | |
800 | out: | |
801 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
802 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
803 | /* | |
804 | * We could return EAGAIN here to make reclaim rescan the inode tree in | |
805 | * a short while. However, this just burns CPU time scanning the tree | |
806 | * waiting for IO to complete and xfssyncd never goes back to the idle | |
807 | * state. Instead, return 0 to let the next scheduled background reclaim | |
808 | * attempt to reclaim the inode again. | |
809 | */ | |
810 | return 0; | |
811 | ||
812 | reclaim: | |
813 | xfs_ifunlock(ip); | |
814 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
815 | xfs_ireclaim(ip); | |
816 | return error; | |
817 | ||
818 | } | |
819 | ||
820 | int | |
821 | xfs_reclaim_inodes( | |
822 | xfs_mount_t *mp, | |
823 | int mode) | |
824 | { | |
825 | return xfs_inode_ag_iterator(mp, xfs_reclaim_inode, mode, | |
826 | XFS_ICI_RECLAIM_TAG, 1, NULL); | |
827 | } | |
828 | ||
829 | /* | |
830 | * Shrinker infrastructure. | |
831 | */ | |
832 | static int | |
833 | xfs_reclaim_inode_shrink( | |
834 | struct shrinker *shrink, | |
835 | int nr_to_scan, | |
836 | gfp_t gfp_mask) | |
837 | { | |
838 | struct xfs_mount *mp; | |
839 | struct xfs_perag *pag; | |
840 | xfs_agnumber_t ag; | |
841 | int reclaimable = 0; | |
842 | ||
843 | mp = container_of(shrink, struct xfs_mount, m_inode_shrink); | |
844 | if (nr_to_scan) { | |
845 | if (!(gfp_mask & __GFP_FS)) | |
846 | return -1; | |
847 | ||
848 | xfs_inode_ag_iterator(mp, xfs_reclaim_inode, 0, | |
849 | XFS_ICI_RECLAIM_TAG, 1, &nr_to_scan); | |
850 | /* if we don't exhaust the scan, don't bother coming back */ | |
851 | if (nr_to_scan > 0) | |
852 | return -1; | |
853 | } | |
854 | ||
855 | for (ag = 0; ag < mp->m_sb.sb_agcount; ag++) { | |
856 | pag = xfs_perag_get(mp, ag); | |
857 | reclaimable += pag->pag_ici_reclaimable; | |
858 | xfs_perag_put(pag); | |
859 | } | |
860 | return reclaimable; | |
861 | } | |
862 | ||
863 | void | |
864 | xfs_inode_shrinker_register( | |
865 | struct xfs_mount *mp) | |
866 | { | |
867 | mp->m_inode_shrink.shrink = xfs_reclaim_inode_shrink; | |
868 | mp->m_inode_shrink.seeks = DEFAULT_SEEKS; | |
869 | register_shrinker(&mp->m_inode_shrink); | |
870 | } | |
871 | ||
872 | void | |
873 | xfs_inode_shrinker_unregister( | |
874 | struct xfs_mount *mp) | |
875 | { | |
876 | unregister_shrinker(&mp->m_inode_shrink); | |
877 | } |