Commit | Line | Data |
---|---|---|
fe4fa4b8 DC |
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" | |
7d095257 | 46 | #include "xfs_quota.h" |
fe4fa4b8 | 47 | |
a167b17e DC |
48 | #include <linux/kthread.h> |
49 | #include <linux/freezer.h> | |
50 | ||
5a34d5cd | 51 | |
75f3cb13 DC |
52 | STATIC xfs_inode_t * |
53 | xfs_inode_ag_lookup( | |
54 | struct xfs_mount *mp, | |
55 | struct xfs_perag *pag, | |
56 | uint32_t *first_index, | |
57 | int tag) | |
58 | { | |
59 | int nr_found; | |
60 | struct xfs_inode *ip; | |
61 | ||
62 | /* | |
63 | * use a gang lookup to find the next inode in the tree | |
64 | * as the tree is sparse and a gang lookup walks to find | |
65 | * the number of objects requested. | |
66 | */ | |
67 | read_lock(&pag->pag_ici_lock); | |
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 | goto unlock; | |
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 | goto unlock; | |
87 | ||
88 | return ip; | |
89 | ||
90 | unlock: | |
91 | read_unlock(&pag->pag_ici_lock); | |
92 | return NULL; | |
93 | } | |
94 | ||
95 | STATIC int | |
96 | xfs_inode_ag_walk( | |
97 | struct xfs_mount *mp, | |
98 | xfs_agnumber_t ag, | |
99 | int (*execute)(struct xfs_inode *ip, | |
100 | struct xfs_perag *pag, int flags), | |
101 | int flags, | |
102 | int tag) | |
103 | { | |
104 | struct xfs_perag *pag = &mp->m_perag[ag]; | |
105 | uint32_t first_index; | |
106 | int last_error = 0; | |
107 | int skipped; | |
108 | ||
109 | restart: | |
110 | skipped = 0; | |
111 | first_index = 0; | |
112 | do { | |
113 | int error = 0; | |
114 | xfs_inode_t *ip; | |
115 | ||
116 | ip = xfs_inode_ag_lookup(mp, pag, &first_index, tag); | |
117 | if (!ip) | |
118 | break; | |
119 | ||
120 | error = execute(ip, pag, flags); | |
121 | if (error == EAGAIN) { | |
122 | skipped++; | |
123 | continue; | |
124 | } | |
125 | if (error) | |
126 | last_error = error; | |
127 | /* | |
128 | * bail out if the filesystem is corrupted. | |
129 | */ | |
130 | if (error == EFSCORRUPTED) | |
131 | break; | |
132 | ||
133 | } while (1); | |
134 | ||
135 | if (skipped) { | |
136 | delay(1); | |
137 | goto restart; | |
138 | } | |
139 | ||
140 | xfs_put_perag(mp, pag); | |
141 | return last_error; | |
142 | } | |
143 | ||
fe588ed3 | 144 | int |
75f3cb13 DC |
145 | xfs_inode_ag_iterator( |
146 | struct xfs_mount *mp, | |
147 | int (*execute)(struct xfs_inode *ip, | |
148 | struct xfs_perag *pag, int flags), | |
149 | int flags, | |
150 | int tag) | |
151 | { | |
152 | int error = 0; | |
153 | int last_error = 0; | |
154 | xfs_agnumber_t ag; | |
155 | ||
156 | for (ag = 0; ag < mp->m_sb.sb_agcount; ag++) { | |
157 | if (!mp->m_perag[ag].pag_ici_init) | |
158 | continue; | |
159 | error = xfs_inode_ag_walk(mp, ag, execute, flags, tag); | |
160 | if (error) { | |
161 | last_error = error; | |
162 | if (error == EFSCORRUPTED) | |
163 | break; | |
164 | } | |
165 | } | |
166 | return XFS_ERROR(last_error); | |
167 | } | |
168 | ||
1da8eeca | 169 | /* must be called with pag_ici_lock held and releases it */ |
fe588ed3 | 170 | int |
1da8eeca DC |
171 | xfs_sync_inode_valid( |
172 | struct xfs_inode *ip, | |
173 | struct xfs_perag *pag) | |
174 | { | |
175 | struct inode *inode = VFS_I(ip); | |
176 | ||
177 | /* nothing to sync during shutdown */ | |
178 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
179 | read_unlock(&pag->pag_ici_lock); | |
180 | return EFSCORRUPTED; | |
181 | } | |
182 | ||
183 | /* | |
184 | * If we can't get a reference on the inode, it must be in reclaim. | |
185 | * Leave it for the reclaim code to flush. Also avoid inodes that | |
186 | * haven't been fully initialised. | |
187 | */ | |
188 | if (!igrab(inode)) { | |
189 | read_unlock(&pag->pag_ici_lock); | |
190 | return ENOENT; | |
191 | } | |
192 | read_unlock(&pag->pag_ici_lock); | |
193 | ||
194 | if (is_bad_inode(inode) || xfs_iflags_test(ip, XFS_INEW)) { | |
195 | IRELE(ip); | |
196 | return ENOENT; | |
197 | } | |
198 | ||
199 | return 0; | |
200 | } | |
201 | ||
5a34d5cd DC |
202 | STATIC int |
203 | xfs_sync_inode_data( | |
204 | struct xfs_inode *ip, | |
75f3cb13 | 205 | struct xfs_perag *pag, |
5a34d5cd DC |
206 | int flags) |
207 | { | |
208 | struct inode *inode = VFS_I(ip); | |
209 | struct address_space *mapping = inode->i_mapping; | |
210 | int error = 0; | |
211 | ||
75f3cb13 DC |
212 | error = xfs_sync_inode_valid(ip, pag); |
213 | if (error) | |
214 | return error; | |
215 | ||
5a34d5cd DC |
216 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
217 | goto out_wait; | |
218 | ||
219 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
220 | if (flags & SYNC_TRYLOCK) | |
221 | goto out_wait; | |
222 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
223 | } | |
224 | ||
225 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? | |
226 | 0 : XFS_B_ASYNC, FI_NONE); | |
227 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
228 | ||
229 | out_wait: | |
230 | if (flags & SYNC_IOWAIT) | |
231 | xfs_ioend_wait(ip); | |
75f3cb13 | 232 | IRELE(ip); |
5a34d5cd DC |
233 | return error; |
234 | } | |
235 | ||
845b6d0c CH |
236 | STATIC int |
237 | xfs_sync_inode_attr( | |
238 | struct xfs_inode *ip, | |
75f3cb13 | 239 | struct xfs_perag *pag, |
845b6d0c CH |
240 | int flags) |
241 | { | |
242 | int error = 0; | |
243 | ||
75f3cb13 DC |
244 | error = xfs_sync_inode_valid(ip, pag); |
245 | if (error) | |
246 | return error; | |
247 | ||
845b6d0c CH |
248 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
249 | if (xfs_inode_clean(ip)) | |
250 | goto out_unlock; | |
251 | if (!xfs_iflock_nowait(ip)) { | |
252 | if (!(flags & SYNC_WAIT)) | |
253 | goto out_unlock; | |
254 | xfs_iflock(ip); | |
255 | } | |
256 | ||
257 | if (xfs_inode_clean(ip)) { | |
258 | xfs_ifunlock(ip); | |
259 | goto out_unlock; | |
260 | } | |
261 | ||
262 | error = xfs_iflush(ip, (flags & SYNC_WAIT) ? | |
263 | XFS_IFLUSH_SYNC : XFS_IFLUSH_DELWRI); | |
264 | ||
265 | out_unlock: | |
266 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
75f3cb13 | 267 | IRELE(ip); |
845b6d0c CH |
268 | return error; |
269 | } | |
270 | ||
683a8970 DC |
271 | int |
272 | xfs_sync_inodes( | |
273 | xfs_mount_t *mp, | |
2030b5ab | 274 | int flags) |
683a8970 | 275 | { |
75f3cb13 | 276 | int error = 0; |
e9f1c6ee | 277 | int lflags = XFS_LOG_FORCE; |
fe4fa4b8 | 278 | |
683a8970 DC |
279 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
280 | return 0; | |
fe4fa4b8 | 281 | |
e9f1c6ee DC |
282 | if (flags & SYNC_WAIT) |
283 | lflags |= XFS_LOG_SYNC; | |
284 | ||
e9f1c6ee | 285 | if (flags & SYNC_DELWRI) |
75f3cb13 | 286 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags, XFS_ICI_NO_TAG); |
e9f1c6ee | 287 | |
75f3cb13 DC |
288 | if (flags & SYNC_ATTR) |
289 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags, XFS_ICI_NO_TAG); | |
290 | ||
291 | if (!error && (flags & SYNC_DELWRI)) | |
292 | xfs_log_force(mp, 0, lflags); | |
293 | return XFS_ERROR(error); | |
fe4fa4b8 DC |
294 | } |
295 | ||
2af75df7 CH |
296 | STATIC int |
297 | xfs_commit_dummy_trans( | |
298 | struct xfs_mount *mp, | |
299 | uint log_flags) | |
300 | { | |
301 | struct xfs_inode *ip = mp->m_rootip; | |
302 | struct xfs_trans *tp; | |
303 | int error; | |
304 | ||
305 | /* | |
306 | * Put a dummy transaction in the log to tell recovery | |
307 | * that all others are OK. | |
308 | */ | |
309 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
310 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
311 | if (error) { | |
312 | xfs_trans_cancel(tp, 0); | |
313 | return error; | |
314 | } | |
315 | ||
316 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
317 | ||
318 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
319 | xfs_trans_ihold(tp, ip); | |
320 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
321 | /* XXX(hch): ignoring the error here.. */ | |
322 | error = xfs_trans_commit(tp, 0); | |
323 | ||
324 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
325 | ||
326 | xfs_log_force(mp, 0, log_flags); | |
327 | return 0; | |
328 | } | |
329 | ||
e9f1c6ee | 330 | int |
2af75df7 CH |
331 | xfs_sync_fsdata( |
332 | struct xfs_mount *mp, | |
333 | int flags) | |
334 | { | |
335 | struct xfs_buf *bp; | |
336 | struct xfs_buf_log_item *bip; | |
337 | int error = 0; | |
338 | ||
339 | /* | |
340 | * If this is xfssyncd() then only sync the superblock if we can | |
341 | * lock it without sleeping and it is not pinned. | |
342 | */ | |
343 | if (flags & SYNC_BDFLUSH) { | |
344 | ASSERT(!(flags & SYNC_WAIT)); | |
345 | ||
346 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
347 | if (!bp) | |
348 | goto out; | |
349 | ||
350 | bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *); | |
351 | if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp)) | |
352 | goto out_brelse; | |
353 | } else { | |
354 | bp = xfs_getsb(mp, 0); | |
355 | ||
356 | /* | |
357 | * If the buffer is pinned then push on the log so we won't | |
358 | * get stuck waiting in the write for someone, maybe | |
359 | * ourselves, to flush the log. | |
360 | * | |
361 | * Even though we just pushed the log above, we did not have | |
362 | * the superblock buffer locked at that point so it can | |
363 | * become pinned in between there and here. | |
364 | */ | |
365 | if (XFS_BUF_ISPINNED(bp)) | |
366 | xfs_log_force(mp, 0, XFS_LOG_FORCE); | |
367 | } | |
368 | ||
369 | ||
370 | if (flags & SYNC_WAIT) | |
371 | XFS_BUF_UNASYNC(bp); | |
372 | else | |
373 | XFS_BUF_ASYNC(bp); | |
374 | ||
375 | return xfs_bwrite(mp, bp); | |
376 | ||
377 | out_brelse: | |
378 | xfs_buf_relse(bp); | |
379 | out: | |
380 | return error; | |
e9f1c6ee DC |
381 | } |
382 | ||
383 | /* | |
a4e4c4f4 DC |
384 | * When remounting a filesystem read-only or freezing the filesystem, we have |
385 | * two phases to execute. This first phase is syncing the data before we | |
386 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
387 | * we've waited for all the transactions created by the first phase to | |
388 | * complete. The second phase ensures that the inodes are written to their | |
389 | * location on disk rather than just existing in transactions in the log. This | |
390 | * means after a quiesce there is no log replay required to write the inodes to | |
391 | * disk (this is the main difference between a sync and a quiesce). | |
392 | */ | |
393 | /* | |
394 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
395 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
396 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
397 | * transactions can still occur here so don't bother flushing the buftarg |
398 | * because it'll just get dirty again. | |
e9f1c6ee DC |
399 | */ |
400 | int | |
401 | xfs_quiesce_data( | |
402 | struct xfs_mount *mp) | |
403 | { | |
404 | int error; | |
405 | ||
406 | /* push non-blocking */ | |
407 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_BDFLUSH); | |
7d095257 | 408 | xfs_qm_sync(mp, SYNC_BDFLUSH); |
e9f1c6ee DC |
409 | xfs_filestream_flush(mp); |
410 | ||
411 | /* push and block */ | |
412 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_WAIT|SYNC_IOWAIT); | |
7d095257 | 413 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 414 | |
a4e4c4f4 | 415 | /* write superblock and hoover up shutdown errors */ |
e9f1c6ee DC |
416 | error = xfs_sync_fsdata(mp, 0); |
417 | ||
a4e4c4f4 | 418 | /* flush data-only devices */ |
e9f1c6ee DC |
419 | if (mp->m_rtdev_targp) |
420 | XFS_bflush(mp->m_rtdev_targp); | |
421 | ||
422 | return error; | |
2af75df7 CH |
423 | } |
424 | ||
76bf105c DC |
425 | STATIC void |
426 | xfs_quiesce_fs( | |
427 | struct xfs_mount *mp) | |
428 | { | |
429 | int count = 0, pincount; | |
430 | ||
431 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
abc10647 | 432 | xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
76bf105c DC |
433 | |
434 | /* | |
435 | * This loop must run at least twice. The first instance of the loop | |
436 | * will flush most meta data but that will generate more meta data | |
437 | * (typically directory updates). Which then must be flushed and | |
438 | * logged before we can write the unmount record. | |
439 | */ | |
440 | do { | |
441 | xfs_sync_inodes(mp, SYNC_ATTR|SYNC_WAIT); | |
442 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
443 | if (!pincount) { | |
444 | delay(50); | |
445 | count++; | |
446 | } | |
447 | } while (count < 2); | |
448 | } | |
449 | ||
450 | /* | |
451 | * Second stage of a quiesce. The data is already synced, now we have to take | |
452 | * care of the metadata. New transactions are already blocked, so we need to | |
453 | * wait for any remaining transactions to drain out before proceding. | |
454 | */ | |
455 | void | |
456 | xfs_quiesce_attr( | |
457 | struct xfs_mount *mp) | |
458 | { | |
459 | int error = 0; | |
460 | ||
461 | /* wait for all modifications to complete */ | |
462 | while (atomic_read(&mp->m_active_trans) > 0) | |
463 | delay(100); | |
464 | ||
465 | /* flush inodes and push all remaining buffers out to disk */ | |
466 | xfs_quiesce_fs(mp); | |
467 | ||
5e106572 FB |
468 | /* |
469 | * Just warn here till VFS can correctly support | |
470 | * read-only remount without racing. | |
471 | */ | |
472 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
473 | |
474 | /* Push the superblock and write an unmount record */ | |
475 | error = xfs_log_sbcount(mp, 1); | |
476 | if (error) | |
477 | xfs_fs_cmn_err(CE_WARN, mp, | |
478 | "xfs_attr_quiesce: failed to log sb changes. " | |
479 | "Frozen image may not be consistent."); | |
480 | xfs_log_unmount_write(mp); | |
481 | xfs_unmountfs_writesb(mp); | |
482 | } | |
483 | ||
a167b17e DC |
484 | /* |
485 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
486 | * Doing this has two advantages: | |
487 | * - It saves on stack space, which is tight in certain situations | |
488 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
489 | * Flushing while allocating in a full filesystem requires both. | |
490 | */ | |
491 | STATIC void | |
492 | xfs_syncd_queue_work( | |
493 | struct xfs_mount *mp, | |
494 | void *data, | |
e43afd72 DC |
495 | void (*syncer)(struct xfs_mount *, void *), |
496 | struct completion *completion) | |
a167b17e | 497 | { |
a8d770d9 | 498 | struct xfs_sync_work *work; |
a167b17e | 499 | |
a8d770d9 | 500 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
a167b17e DC |
501 | INIT_LIST_HEAD(&work->w_list); |
502 | work->w_syncer = syncer; | |
503 | work->w_data = data; | |
504 | work->w_mount = mp; | |
e43afd72 | 505 | work->w_completion = completion; |
a167b17e DC |
506 | spin_lock(&mp->m_sync_lock); |
507 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
508 | spin_unlock(&mp->m_sync_lock); | |
509 | wake_up_process(mp->m_sync_task); | |
510 | } | |
511 | ||
512 | /* | |
513 | * Flush delayed allocate data, attempting to free up reserved space | |
514 | * from existing allocations. At this point a new allocation attempt | |
515 | * has failed with ENOSPC and we are in the process of scratching our | |
516 | * heads, looking about for more room... | |
517 | */ | |
518 | STATIC void | |
a8d770d9 | 519 | xfs_flush_inodes_work( |
a167b17e DC |
520 | struct xfs_mount *mp, |
521 | void *arg) | |
522 | { | |
523 | struct inode *inode = arg; | |
a8d770d9 DC |
524 | xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK); |
525 | xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK | SYNC_IOWAIT); | |
a167b17e DC |
526 | iput(inode); |
527 | } | |
528 | ||
529 | void | |
a8d770d9 | 530 | xfs_flush_inodes( |
a167b17e DC |
531 | xfs_inode_t *ip) |
532 | { | |
533 | struct inode *inode = VFS_I(ip); | |
e43afd72 | 534 | DECLARE_COMPLETION_ONSTACK(completion); |
a167b17e DC |
535 | |
536 | igrab(inode); | |
e43afd72 DC |
537 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); |
538 | wait_for_completion(&completion); | |
a167b17e DC |
539 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); |
540 | } | |
541 | ||
aacaa880 DC |
542 | /* |
543 | * Every sync period we need to unpin all items, reclaim inodes, sync | |
544 | * quota and write out the superblock. We might need to cover the log | |
545 | * to indicate it is idle. | |
546 | */ | |
a167b17e DC |
547 | STATIC void |
548 | xfs_sync_worker( | |
549 | struct xfs_mount *mp, | |
550 | void *unused) | |
551 | { | |
552 | int error; | |
553 | ||
aacaa880 DC |
554 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
555 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
abc10647 | 556 | xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
aacaa880 | 557 | /* dgc: errors ignored here */ |
7d095257 | 558 | error = xfs_qm_sync(mp, SYNC_BDFLUSH); |
aacaa880 DC |
559 | error = xfs_sync_fsdata(mp, SYNC_BDFLUSH); |
560 | if (xfs_log_need_covered(mp)) | |
561 | error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE); | |
562 | } | |
a167b17e DC |
563 | mp->m_sync_seq++; |
564 | wake_up(&mp->m_wait_single_sync_task); | |
565 | } | |
566 | ||
567 | STATIC int | |
568 | xfssyncd( | |
569 | void *arg) | |
570 | { | |
571 | struct xfs_mount *mp = arg; | |
572 | long timeleft; | |
a8d770d9 | 573 | xfs_sync_work_t *work, *n; |
a167b17e DC |
574 | LIST_HEAD (tmp); |
575 | ||
576 | set_freezable(); | |
577 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
578 | for (;;) { | |
579 | timeleft = schedule_timeout_interruptible(timeleft); | |
580 | /* swsusp */ | |
581 | try_to_freeze(); | |
582 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
583 | break; | |
584 | ||
585 | spin_lock(&mp->m_sync_lock); | |
586 | /* | |
587 | * We can get woken by laptop mode, to do a sync - | |
588 | * that's the (only!) case where the list would be | |
589 | * empty with time remaining. | |
590 | */ | |
591 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
592 | if (!timeleft) | |
593 | timeleft = xfs_syncd_centisecs * | |
594 | msecs_to_jiffies(10); | |
595 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
596 | list_add_tail(&mp->m_sync_work.w_list, | |
597 | &mp->m_sync_list); | |
598 | } | |
599 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) | |
600 | list_move(&work->w_list, &tmp); | |
601 | spin_unlock(&mp->m_sync_lock); | |
602 | ||
603 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
604 | (*work->w_syncer)(mp, work->w_data); | |
605 | list_del(&work->w_list); | |
606 | if (work == &mp->m_sync_work) | |
607 | continue; | |
e43afd72 DC |
608 | if (work->w_completion) |
609 | complete(work->w_completion); | |
a167b17e DC |
610 | kmem_free(work); |
611 | } | |
612 | } | |
613 | ||
614 | return 0; | |
615 | } | |
616 | ||
617 | int | |
618 | xfs_syncd_init( | |
619 | struct xfs_mount *mp) | |
620 | { | |
621 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
622 | mp->m_sync_work.w_mount = mp; | |
e43afd72 | 623 | mp->m_sync_work.w_completion = NULL; |
a167b17e DC |
624 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); |
625 | if (IS_ERR(mp->m_sync_task)) | |
626 | return -PTR_ERR(mp->m_sync_task); | |
627 | return 0; | |
628 | } | |
629 | ||
630 | void | |
631 | xfs_syncd_stop( | |
632 | struct xfs_mount *mp) | |
633 | { | |
634 | kthread_stop(mp->m_sync_task); | |
635 | } | |
636 | ||
fce08f2f | 637 | int |
1dc3318a | 638 | xfs_reclaim_inode( |
fce08f2f DC |
639 | xfs_inode_t *ip, |
640 | int locked, | |
641 | int sync_mode) | |
642 | { | |
643 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); | |
644 | ||
645 | /* The hash lock here protects a thread in xfs_iget_core from | |
646 | * racing with us on linking the inode back with a vnode. | |
647 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
648 | * us. | |
649 | */ | |
650 | write_lock(&pag->pag_ici_lock); | |
651 | spin_lock(&ip->i_flags_lock); | |
652 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || | |
653 | !__xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
654 | spin_unlock(&ip->i_flags_lock); | |
655 | write_unlock(&pag->pag_ici_lock); | |
656 | if (locked) { | |
657 | xfs_ifunlock(ip); | |
658 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
659 | } | |
75f3cb13 | 660 | return -EAGAIN; |
fce08f2f DC |
661 | } |
662 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
663 | spin_unlock(&ip->i_flags_lock); | |
664 | write_unlock(&pag->pag_ici_lock); | |
665 | xfs_put_perag(ip->i_mount, pag); | |
666 | ||
667 | /* | |
668 | * If the inode is still dirty, then flush it out. If the inode | |
669 | * is not in the AIL, then it will be OK to flush it delwri as | |
670 | * long as xfs_iflush() does not keep any references to the inode. | |
671 | * We leave that decision up to xfs_iflush() since it has the | |
672 | * knowledge of whether it's OK to simply do a delwri flush of | |
673 | * the inode or whether we need to wait until the inode is | |
674 | * pulled from the AIL. | |
675 | * We get the flush lock regardless, though, just to make sure | |
676 | * we don't free it while it is being flushed. | |
677 | */ | |
678 | if (!locked) { | |
679 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
680 | xfs_iflock(ip); | |
681 | } | |
682 | ||
683 | /* | |
684 | * In the case of a forced shutdown we rely on xfs_iflush() to | |
685 | * wait for the inode to be unpinned before returning an error. | |
686 | */ | |
687 | if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) { | |
688 | /* synchronize with xfs_iflush_done */ | |
689 | xfs_iflock(ip); | |
690 | xfs_ifunlock(ip); | |
691 | } | |
692 | ||
693 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
694 | xfs_ireclaim(ip); | |
695 | return 0; | |
696 | } | |
697 | ||
11654513 DC |
698 | /* |
699 | * We set the inode flag atomically with the radix tree tag. | |
700 | * Once we get tag lookups on the radix tree, this inode flag | |
701 | * can go away. | |
702 | */ | |
396beb85 DC |
703 | void |
704 | xfs_inode_set_reclaim_tag( | |
705 | xfs_inode_t *ip) | |
706 | { | |
707 | xfs_mount_t *mp = ip->i_mount; | |
708 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
709 | ||
710 | read_lock(&pag->pag_ici_lock); | |
711 | spin_lock(&ip->i_flags_lock); | |
712 | radix_tree_tag_set(&pag->pag_ici_root, | |
713 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
11654513 | 714 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
715 | spin_unlock(&ip->i_flags_lock); |
716 | read_unlock(&pag->pag_ici_lock); | |
717 | xfs_put_perag(mp, pag); | |
718 | } | |
719 | ||
720 | void | |
721 | __xfs_inode_clear_reclaim_tag( | |
722 | xfs_mount_t *mp, | |
723 | xfs_perag_t *pag, | |
724 | xfs_inode_t *ip) | |
725 | { | |
726 | radix_tree_tag_clear(&pag->pag_ici_root, | |
727 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
728 | } | |
729 | ||
730 | void | |
731 | xfs_inode_clear_reclaim_tag( | |
732 | xfs_inode_t *ip) | |
733 | { | |
734 | xfs_mount_t *mp = ip->i_mount; | |
735 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
736 | ||
737 | read_lock(&pag->pag_ici_lock); | |
738 | spin_lock(&ip->i_flags_lock); | |
739 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
740 | spin_unlock(&ip->i_flags_lock); | |
741 | read_unlock(&pag->pag_ici_lock); | |
742 | xfs_put_perag(mp, pag); | |
743 | } | |
744 | ||
75f3cb13 DC |
745 | STATIC int |
746 | xfs_reclaim_inode_now( | |
747 | struct xfs_inode *ip, | |
748 | struct xfs_perag *pag, | |
749 | int flags) | |
fce08f2f | 750 | { |
75f3cb13 DC |
751 | /* ignore if already under reclaim */ |
752 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
7a3be02b | 753 | read_unlock(&pag->pag_ici_lock); |
75f3cb13 | 754 | return 0; |
fce08f2f | 755 | } |
75f3cb13 | 756 | read_unlock(&pag->pag_ici_lock); |
7a3be02b | 757 | |
75f3cb13 | 758 | return xfs_reclaim_inode(ip, 0, flags); |
7a3be02b DC |
759 | } |
760 | ||
761 | int | |
762 | xfs_reclaim_inodes( | |
763 | xfs_mount_t *mp, | |
7a3be02b DC |
764 | int mode) |
765 | { | |
75f3cb13 DC |
766 | return xfs_inode_ag_iterator(mp, xfs_reclaim_inode_now, mode, |
767 | XFS_ICI_RECLAIM_TAG); | |
fce08f2f | 768 | } |