]>
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: | |
b0710ccc | 230 | if (flags & SYNC_WAIT) |
5a34d5cd | 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 | ||
075fe102 CH |
271 | /* |
272 | * Write out pagecache data for the whole filesystem. | |
273 | */ | |
683a8970 | 274 | int |
075fe102 CH |
275 | xfs_sync_data( |
276 | struct xfs_mount *mp, | |
277 | int flags) | |
683a8970 | 278 | { |
075fe102 | 279 | int error; |
fe4fa4b8 | 280 | |
b0710ccc | 281 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
fe4fa4b8 | 282 | |
075fe102 CH |
283 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags, |
284 | XFS_ICI_NO_TAG); | |
285 | if (error) | |
286 | return XFS_ERROR(error); | |
e9f1c6ee | 287 | |
075fe102 CH |
288 | xfs_log_force(mp, 0, |
289 | (flags & SYNC_WAIT) ? | |
290 | XFS_LOG_FORCE | XFS_LOG_SYNC : | |
291 | XFS_LOG_FORCE); | |
292 | return 0; | |
293 | } | |
e9f1c6ee | 294 | |
075fe102 CH |
295 | /* |
296 | * Write out inode metadata (attributes) for the whole filesystem. | |
297 | */ | |
298 | int | |
299 | xfs_sync_attr( | |
300 | struct xfs_mount *mp, | |
301 | int flags) | |
302 | { | |
303 | ASSERT((flags & ~SYNC_WAIT) == 0); | |
75f3cb13 | 304 | |
075fe102 CH |
305 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags, |
306 | XFS_ICI_NO_TAG); | |
fe4fa4b8 DC |
307 | } |
308 | ||
2af75df7 CH |
309 | STATIC int |
310 | xfs_commit_dummy_trans( | |
311 | struct xfs_mount *mp, | |
dce5065a | 312 | uint flags) |
2af75df7 CH |
313 | { |
314 | struct xfs_inode *ip = mp->m_rootip; | |
315 | struct xfs_trans *tp; | |
316 | int error; | |
dce5065a DC |
317 | int log_flags = XFS_LOG_FORCE; |
318 | ||
319 | if (flags & SYNC_WAIT) | |
320 | log_flags |= XFS_LOG_SYNC; | |
2af75df7 CH |
321 | |
322 | /* | |
323 | * Put a dummy transaction in the log to tell recovery | |
324 | * that all others are OK. | |
325 | */ | |
326 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
327 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
328 | if (error) { | |
329 | xfs_trans_cancel(tp, 0); | |
330 | return error; | |
331 | } | |
332 | ||
333 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
334 | ||
335 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
336 | xfs_trans_ihold(tp, ip); | |
337 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2af75df7 | 338 | error = xfs_trans_commit(tp, 0); |
2af75df7 CH |
339 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
340 | ||
dce5065a | 341 | /* the log force ensures this transaction is pushed to disk */ |
2af75df7 | 342 | xfs_log_force(mp, 0, log_flags); |
dce5065a | 343 | return error; |
2af75df7 CH |
344 | } |
345 | ||
e9f1c6ee | 346 | int |
2af75df7 CH |
347 | xfs_sync_fsdata( |
348 | struct xfs_mount *mp, | |
349 | int flags) | |
350 | { | |
351 | struct xfs_buf *bp; | |
352 | struct xfs_buf_log_item *bip; | |
353 | int error = 0; | |
354 | ||
355 | /* | |
356 | * If this is xfssyncd() then only sync the superblock if we can | |
357 | * lock it without sleeping and it is not pinned. | |
358 | */ | |
8b5403a6 | 359 | if (flags & SYNC_TRYLOCK) { |
2af75df7 CH |
360 | ASSERT(!(flags & SYNC_WAIT)); |
361 | ||
362 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
363 | if (!bp) | |
364 | goto out; | |
365 | ||
366 | bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *); | |
367 | if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp)) | |
368 | goto out_brelse; | |
369 | } else { | |
370 | bp = xfs_getsb(mp, 0); | |
371 | ||
372 | /* | |
373 | * If the buffer is pinned then push on the log so we won't | |
374 | * get stuck waiting in the write for someone, maybe | |
375 | * ourselves, to flush the log. | |
376 | * | |
377 | * Even though we just pushed the log above, we did not have | |
378 | * the superblock buffer locked at that point so it can | |
379 | * become pinned in between there and here. | |
380 | */ | |
381 | if (XFS_BUF_ISPINNED(bp)) | |
382 | xfs_log_force(mp, 0, XFS_LOG_FORCE); | |
383 | } | |
384 | ||
385 | ||
386 | if (flags & SYNC_WAIT) | |
387 | XFS_BUF_UNASYNC(bp); | |
388 | else | |
389 | XFS_BUF_ASYNC(bp); | |
390 | ||
dce5065a DC |
391 | error = xfs_bwrite(mp, bp); |
392 | if (error) | |
393 | return error; | |
394 | ||
395 | /* | |
396 | * If this is a data integrity sync make sure all pending buffers | |
397 | * are flushed out for the log coverage check below. | |
398 | */ | |
399 | if (flags & SYNC_WAIT) | |
400 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
401 | ||
402 | if (xfs_log_need_covered(mp)) | |
403 | error = xfs_commit_dummy_trans(mp, flags); | |
404 | return error; | |
2af75df7 CH |
405 | |
406 | out_brelse: | |
407 | xfs_buf_relse(bp); | |
408 | out: | |
409 | return error; | |
e9f1c6ee DC |
410 | } |
411 | ||
412 | /* | |
a4e4c4f4 DC |
413 | * When remounting a filesystem read-only or freezing the filesystem, we have |
414 | * two phases to execute. This first phase is syncing the data before we | |
415 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
416 | * we've waited for all the transactions created by the first phase to | |
417 | * complete. The second phase ensures that the inodes are written to their | |
418 | * location on disk rather than just existing in transactions in the log. This | |
419 | * means after a quiesce there is no log replay required to write the inodes to | |
420 | * disk (this is the main difference between a sync and a quiesce). | |
421 | */ | |
422 | /* | |
423 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
424 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
425 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
426 | * transactions can still occur here so don't bother flushing the buftarg |
427 | * because it'll just get dirty again. | |
e9f1c6ee DC |
428 | */ |
429 | int | |
430 | xfs_quiesce_data( | |
431 | struct xfs_mount *mp) | |
432 | { | |
433 | int error; | |
434 | ||
435 | /* push non-blocking */ | |
075fe102 | 436 | xfs_sync_data(mp, 0); |
8b5403a6 | 437 | xfs_qm_sync(mp, SYNC_TRYLOCK); |
e9f1c6ee | 438 | |
c90b07e8 | 439 | /* push and block till complete */ |
b0710ccc | 440 | xfs_sync_data(mp, SYNC_WAIT); |
7d095257 | 441 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 442 | |
c90b07e8 DC |
443 | /* drop inode references pinned by filestreams */ |
444 | xfs_filestream_flush(mp); | |
445 | ||
a4e4c4f4 | 446 | /* write superblock and hoover up shutdown errors */ |
c90b07e8 | 447 | error = xfs_sync_fsdata(mp, SYNC_WAIT); |
e9f1c6ee | 448 | |
a4e4c4f4 | 449 | /* flush data-only devices */ |
e9f1c6ee DC |
450 | if (mp->m_rtdev_targp) |
451 | XFS_bflush(mp->m_rtdev_targp); | |
452 | ||
453 | return error; | |
2af75df7 CH |
454 | } |
455 | ||
76bf105c DC |
456 | STATIC void |
457 | xfs_quiesce_fs( | |
458 | struct xfs_mount *mp) | |
459 | { | |
460 | int count = 0, pincount; | |
461 | ||
462 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
abc10647 | 463 | xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
76bf105c DC |
464 | |
465 | /* | |
466 | * This loop must run at least twice. The first instance of the loop | |
467 | * will flush most meta data but that will generate more meta data | |
468 | * (typically directory updates). Which then must be flushed and | |
469 | * logged before we can write the unmount record. | |
470 | */ | |
471 | do { | |
075fe102 | 472 | xfs_sync_attr(mp, SYNC_WAIT); |
76bf105c DC |
473 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
474 | if (!pincount) { | |
475 | delay(50); | |
476 | count++; | |
477 | } | |
478 | } while (count < 2); | |
479 | } | |
480 | ||
481 | /* | |
482 | * Second stage of a quiesce. The data is already synced, now we have to take | |
483 | * care of the metadata. New transactions are already blocked, so we need to | |
484 | * wait for any remaining transactions to drain out before proceding. | |
485 | */ | |
486 | void | |
487 | xfs_quiesce_attr( | |
488 | struct xfs_mount *mp) | |
489 | { | |
490 | int error = 0; | |
491 | ||
492 | /* wait for all modifications to complete */ | |
493 | while (atomic_read(&mp->m_active_trans) > 0) | |
494 | delay(100); | |
495 | ||
496 | /* flush inodes and push all remaining buffers out to disk */ | |
497 | xfs_quiesce_fs(mp); | |
498 | ||
5e106572 FB |
499 | /* |
500 | * Just warn here till VFS can correctly support | |
501 | * read-only remount without racing. | |
502 | */ | |
503 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
504 | |
505 | /* Push the superblock and write an unmount record */ | |
506 | error = xfs_log_sbcount(mp, 1); | |
507 | if (error) | |
508 | xfs_fs_cmn_err(CE_WARN, mp, | |
509 | "xfs_attr_quiesce: failed to log sb changes. " | |
510 | "Frozen image may not be consistent."); | |
511 | xfs_log_unmount_write(mp); | |
512 | xfs_unmountfs_writesb(mp); | |
513 | } | |
514 | ||
a167b17e DC |
515 | /* |
516 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
517 | * Doing this has two advantages: | |
518 | * - It saves on stack space, which is tight in certain situations | |
519 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
520 | * Flushing while allocating in a full filesystem requires both. | |
521 | */ | |
522 | STATIC void | |
523 | xfs_syncd_queue_work( | |
524 | struct xfs_mount *mp, | |
525 | void *data, | |
e43afd72 DC |
526 | void (*syncer)(struct xfs_mount *, void *), |
527 | struct completion *completion) | |
a167b17e | 528 | { |
a8d770d9 | 529 | struct xfs_sync_work *work; |
a167b17e | 530 | |
a8d770d9 | 531 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
a167b17e DC |
532 | INIT_LIST_HEAD(&work->w_list); |
533 | work->w_syncer = syncer; | |
534 | work->w_data = data; | |
535 | work->w_mount = mp; | |
e43afd72 | 536 | work->w_completion = completion; |
a167b17e DC |
537 | spin_lock(&mp->m_sync_lock); |
538 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
539 | spin_unlock(&mp->m_sync_lock); | |
540 | wake_up_process(mp->m_sync_task); | |
541 | } | |
542 | ||
543 | /* | |
544 | * Flush delayed allocate data, attempting to free up reserved space | |
545 | * from existing allocations. At this point a new allocation attempt | |
546 | * has failed with ENOSPC and we are in the process of scratching our | |
547 | * heads, looking about for more room... | |
548 | */ | |
549 | STATIC void | |
a8d770d9 | 550 | xfs_flush_inodes_work( |
a167b17e DC |
551 | struct xfs_mount *mp, |
552 | void *arg) | |
553 | { | |
554 | struct inode *inode = arg; | |
075fe102 | 555 | xfs_sync_data(mp, SYNC_TRYLOCK); |
b0710ccc | 556 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); |
a167b17e DC |
557 | iput(inode); |
558 | } | |
559 | ||
560 | void | |
a8d770d9 | 561 | xfs_flush_inodes( |
a167b17e DC |
562 | xfs_inode_t *ip) |
563 | { | |
564 | struct inode *inode = VFS_I(ip); | |
e43afd72 | 565 | DECLARE_COMPLETION_ONSTACK(completion); |
a167b17e DC |
566 | |
567 | igrab(inode); | |
e43afd72 DC |
568 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); |
569 | wait_for_completion(&completion); | |
a167b17e DC |
570 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); |
571 | } | |
572 | ||
aacaa880 DC |
573 | /* |
574 | * Every sync period we need to unpin all items, reclaim inodes, sync | |
575 | * quota and write out the superblock. We might need to cover the log | |
576 | * to indicate it is idle. | |
577 | */ | |
a167b17e DC |
578 | STATIC void |
579 | xfs_sync_worker( | |
580 | struct xfs_mount *mp, | |
581 | void *unused) | |
582 | { | |
583 | int error; | |
584 | ||
aacaa880 DC |
585 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
586 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
abc10647 | 587 | xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
aacaa880 | 588 | /* dgc: errors ignored here */ |
8b5403a6 CH |
589 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); |
590 | error = xfs_sync_fsdata(mp, SYNC_TRYLOCK); | |
aacaa880 | 591 | } |
a167b17e DC |
592 | mp->m_sync_seq++; |
593 | wake_up(&mp->m_wait_single_sync_task); | |
594 | } | |
595 | ||
596 | STATIC int | |
597 | xfssyncd( | |
598 | void *arg) | |
599 | { | |
600 | struct xfs_mount *mp = arg; | |
601 | long timeleft; | |
a8d770d9 | 602 | xfs_sync_work_t *work, *n; |
a167b17e DC |
603 | LIST_HEAD (tmp); |
604 | ||
605 | set_freezable(); | |
606 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
607 | for (;;) { | |
608 | timeleft = schedule_timeout_interruptible(timeleft); | |
609 | /* swsusp */ | |
610 | try_to_freeze(); | |
611 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
612 | break; | |
613 | ||
614 | spin_lock(&mp->m_sync_lock); | |
615 | /* | |
616 | * We can get woken by laptop mode, to do a sync - | |
617 | * that's the (only!) case where the list would be | |
618 | * empty with time remaining. | |
619 | */ | |
620 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
621 | if (!timeleft) | |
622 | timeleft = xfs_syncd_centisecs * | |
623 | msecs_to_jiffies(10); | |
624 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
625 | list_add_tail(&mp->m_sync_work.w_list, | |
626 | &mp->m_sync_list); | |
627 | } | |
628 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) | |
629 | list_move(&work->w_list, &tmp); | |
630 | spin_unlock(&mp->m_sync_lock); | |
631 | ||
632 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
633 | (*work->w_syncer)(mp, work->w_data); | |
634 | list_del(&work->w_list); | |
635 | if (work == &mp->m_sync_work) | |
636 | continue; | |
e43afd72 DC |
637 | if (work->w_completion) |
638 | complete(work->w_completion); | |
a167b17e DC |
639 | kmem_free(work); |
640 | } | |
641 | } | |
642 | ||
643 | return 0; | |
644 | } | |
645 | ||
646 | int | |
647 | xfs_syncd_init( | |
648 | struct xfs_mount *mp) | |
649 | { | |
650 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
651 | mp->m_sync_work.w_mount = mp; | |
e43afd72 | 652 | mp->m_sync_work.w_completion = NULL; |
a167b17e DC |
653 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); |
654 | if (IS_ERR(mp->m_sync_task)) | |
655 | return -PTR_ERR(mp->m_sync_task); | |
656 | return 0; | |
657 | } | |
658 | ||
659 | void | |
660 | xfs_syncd_stop( | |
661 | struct xfs_mount *mp) | |
662 | { | |
663 | kthread_stop(mp->m_sync_task); | |
664 | } | |
665 | ||
848ce8f7 | 666 | STATIC int |
1dc3318a | 667 | xfs_reclaim_inode( |
fce08f2f | 668 | xfs_inode_t *ip, |
fce08f2f DC |
669 | int sync_mode) |
670 | { | |
671 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); | |
672 | ||
673 | /* The hash lock here protects a thread in xfs_iget_core from | |
674 | * racing with us on linking the inode back with a vnode. | |
675 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
676 | * us. | |
677 | */ | |
678 | write_lock(&pag->pag_ici_lock); | |
679 | spin_lock(&ip->i_flags_lock); | |
680 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || | |
681 | !__xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
682 | spin_unlock(&ip->i_flags_lock); | |
683 | write_unlock(&pag->pag_ici_lock); | |
75f3cb13 | 684 | return -EAGAIN; |
fce08f2f DC |
685 | } |
686 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
687 | spin_unlock(&ip->i_flags_lock); | |
688 | write_unlock(&pag->pag_ici_lock); | |
689 | xfs_put_perag(ip->i_mount, pag); | |
690 | ||
691 | /* | |
692 | * If the inode is still dirty, then flush it out. If the inode | |
693 | * is not in the AIL, then it will be OK to flush it delwri as | |
694 | * long as xfs_iflush() does not keep any references to the inode. | |
695 | * We leave that decision up to xfs_iflush() since it has the | |
696 | * knowledge of whether it's OK to simply do a delwri flush of | |
697 | * the inode or whether we need to wait until the inode is | |
698 | * pulled from the AIL. | |
699 | * We get the flush lock regardless, though, just to make sure | |
700 | * we don't free it while it is being flushed. | |
701 | */ | |
848ce8f7 CH |
702 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
703 | xfs_iflock(ip); | |
fce08f2f DC |
704 | |
705 | /* | |
706 | * In the case of a forced shutdown we rely on xfs_iflush() to | |
707 | * wait for the inode to be unpinned before returning an error. | |
708 | */ | |
709 | if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) { | |
710 | /* synchronize with xfs_iflush_done */ | |
711 | xfs_iflock(ip); | |
712 | xfs_ifunlock(ip); | |
713 | } | |
714 | ||
715 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
716 | xfs_ireclaim(ip); | |
717 | return 0; | |
718 | } | |
719 | ||
bc990f5c CH |
720 | void |
721 | __xfs_inode_set_reclaim_tag( | |
722 | struct xfs_perag *pag, | |
723 | struct xfs_inode *ip) | |
724 | { | |
725 | radix_tree_tag_set(&pag->pag_ici_root, | |
726 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
727 | XFS_ICI_RECLAIM_TAG); | |
728 | } | |
729 | ||
11654513 DC |
730 | /* |
731 | * We set the inode flag atomically with the radix tree tag. | |
732 | * Once we get tag lookups on the radix tree, this inode flag | |
733 | * can go away. | |
734 | */ | |
396beb85 DC |
735 | void |
736 | xfs_inode_set_reclaim_tag( | |
737 | xfs_inode_t *ip) | |
738 | { | |
739 | xfs_mount_t *mp = ip->i_mount; | |
740 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
741 | ||
742 | read_lock(&pag->pag_ici_lock); | |
743 | spin_lock(&ip->i_flags_lock); | |
bc990f5c | 744 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 745 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
746 | spin_unlock(&ip->i_flags_lock); |
747 | read_unlock(&pag->pag_ici_lock); | |
748 | xfs_put_perag(mp, pag); | |
749 | } | |
750 | ||
751 | void | |
752 | __xfs_inode_clear_reclaim_tag( | |
753 | xfs_mount_t *mp, | |
754 | xfs_perag_t *pag, | |
755 | xfs_inode_t *ip) | |
756 | { | |
757 | radix_tree_tag_clear(&pag->pag_ici_root, | |
758 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
759 | } | |
760 | ||
75f3cb13 DC |
761 | STATIC int |
762 | xfs_reclaim_inode_now( | |
763 | struct xfs_inode *ip, | |
764 | struct xfs_perag *pag, | |
765 | int flags) | |
fce08f2f | 766 | { |
75f3cb13 DC |
767 | /* ignore if already under reclaim */ |
768 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
7a3be02b | 769 | read_unlock(&pag->pag_ici_lock); |
75f3cb13 | 770 | return 0; |
fce08f2f | 771 | } |
75f3cb13 | 772 | read_unlock(&pag->pag_ici_lock); |
7a3be02b | 773 | |
848ce8f7 | 774 | return xfs_reclaim_inode(ip, flags); |
7a3be02b DC |
775 | } |
776 | ||
777 | int | |
778 | xfs_reclaim_inodes( | |
779 | xfs_mount_t *mp, | |
7a3be02b DC |
780 | int mode) |
781 | { | |
75f3cb13 DC |
782 | return xfs_inode_ag_iterator(mp, xfs_reclaim_inode_now, mode, |
783 | XFS_ICI_RECLAIM_TAG); | |
fce08f2f | 784 | } |