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" |
0b1b213f | 47 | #include "xfs_trace.h" |
fe4fa4b8 | 48 | |
a167b17e DC |
49 | #include <linux/kthread.h> |
50 | #include <linux/freezer.h> | |
51 | ||
5a34d5cd | 52 | |
75f3cb13 DC |
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 | */ | |
75f3cb13 DC |
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) | |
c8e20be0 | 76 | return NULL; |
75f3cb13 DC |
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)) | |
c8e20be0 | 86 | return NULL; |
75f3cb13 | 87 | return ip; |
75f3cb13 DC |
88 | } |
89 | ||
90 | STATIC int | |
91 | xfs_inode_ag_walk( | |
92 | struct xfs_mount *mp, | |
5017e97d | 93 | struct xfs_perag *pag, |
75f3cb13 DC |
94 | int (*execute)(struct xfs_inode *ip, |
95 | struct xfs_perag *pag, int flags), | |
96 | int flags, | |
c8e20be0 DC |
97 | int tag, |
98 | int exclusive) | |
75f3cb13 | 99 | { |
75f3cb13 DC |
100 | uint32_t first_index; |
101 | int last_error = 0; | |
102 | int skipped; | |
103 | ||
104 | restart: | |
105 | skipped = 0; | |
106 | first_index = 0; | |
107 | do { | |
108 | int error = 0; | |
109 | xfs_inode_t *ip; | |
110 | ||
c8e20be0 DC |
111 | if (exclusive) |
112 | write_lock(&pag->pag_ici_lock); | |
113 | else | |
114 | read_lock(&pag->pag_ici_lock); | |
75f3cb13 | 115 | ip = xfs_inode_ag_lookup(mp, pag, &first_index, tag); |
c8e20be0 DC |
116 | if (!ip) { |
117 | if (exclusive) | |
118 | write_unlock(&pag->pag_ici_lock); | |
119 | else | |
120 | read_unlock(&pag->pag_ici_lock); | |
75f3cb13 | 121 | break; |
c8e20be0 | 122 | } |
75f3cb13 | 123 | |
c8e20be0 | 124 | /* execute releases pag->pag_ici_lock */ |
75f3cb13 DC |
125 | error = execute(ip, pag, flags); |
126 | if (error == EAGAIN) { | |
127 | skipped++; | |
128 | continue; | |
129 | } | |
130 | if (error) | |
131 | last_error = error; | |
c8e20be0 DC |
132 | |
133 | /* bail out if the filesystem is corrupted. */ | |
75f3cb13 DC |
134 | if (error == EFSCORRUPTED) |
135 | break; | |
136 | ||
137 | } while (1); | |
138 | ||
139 | if (skipped) { | |
140 | delay(1); | |
141 | goto restart; | |
142 | } | |
75f3cb13 DC |
143 | return last_error; |
144 | } | |
145 | ||
fe588ed3 | 146 | int |
75f3cb13 DC |
147 | xfs_inode_ag_iterator( |
148 | struct xfs_mount *mp, | |
149 | int (*execute)(struct xfs_inode *ip, | |
150 | struct xfs_perag *pag, int flags), | |
151 | int flags, | |
c8e20be0 DC |
152 | int tag, |
153 | int exclusive) | |
75f3cb13 DC |
154 | { |
155 | int error = 0; | |
156 | int last_error = 0; | |
157 | xfs_agnumber_t ag; | |
158 | ||
159 | for (ag = 0; ag < mp->m_sb.sb_agcount; ag++) { | |
5017e97d DC |
160 | struct xfs_perag *pag; |
161 | ||
162 | pag = xfs_perag_get(mp, ag); | |
163 | if (!pag->pag_ici_init) { | |
164 | xfs_perag_put(pag); | |
75f3cb13 | 165 | continue; |
5017e97d DC |
166 | } |
167 | error = xfs_inode_ag_walk(mp, pag, execute, flags, tag, | |
c8e20be0 | 168 | exclusive); |
5017e97d | 169 | xfs_perag_put(pag); |
75f3cb13 DC |
170 | if (error) { |
171 | last_error = error; | |
172 | if (error == EFSCORRUPTED) | |
173 | break; | |
174 | } | |
175 | } | |
176 | return XFS_ERROR(last_error); | |
177 | } | |
178 | ||
1da8eeca | 179 | /* must be called with pag_ici_lock held and releases it */ |
fe588ed3 | 180 | int |
1da8eeca DC |
181 | xfs_sync_inode_valid( |
182 | struct xfs_inode *ip, | |
183 | struct xfs_perag *pag) | |
184 | { | |
185 | struct inode *inode = VFS_I(ip); | |
018027be | 186 | int error = EFSCORRUPTED; |
1da8eeca DC |
187 | |
188 | /* nothing to sync during shutdown */ | |
018027be DC |
189 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
190 | goto out_unlock; | |
1da8eeca | 191 | |
018027be DC |
192 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ |
193 | error = ENOENT; | |
194 | if (xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
195 | goto out_unlock; | |
1da8eeca | 196 | |
018027be DC |
197 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
198 | if (!igrab(inode)) | |
199 | goto out_unlock; | |
200 | ||
201 | if (is_bad_inode(inode)) { | |
1da8eeca | 202 | IRELE(ip); |
018027be | 203 | goto out_unlock; |
1da8eeca DC |
204 | } |
205 | ||
018027be DC |
206 | /* inode is valid */ |
207 | error = 0; | |
208 | out_unlock: | |
209 | read_unlock(&pag->pag_ici_lock); | |
210 | return error; | |
1da8eeca DC |
211 | } |
212 | ||
5a34d5cd DC |
213 | STATIC int |
214 | xfs_sync_inode_data( | |
215 | struct xfs_inode *ip, | |
75f3cb13 | 216 | struct xfs_perag *pag, |
5a34d5cd DC |
217 | int flags) |
218 | { | |
219 | struct inode *inode = VFS_I(ip); | |
220 | struct address_space *mapping = inode->i_mapping; | |
221 | int error = 0; | |
222 | ||
75f3cb13 DC |
223 | error = xfs_sync_inode_valid(ip, pag); |
224 | if (error) | |
225 | return error; | |
226 | ||
5a34d5cd DC |
227 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
228 | goto out_wait; | |
229 | ||
230 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
231 | if (flags & SYNC_TRYLOCK) | |
232 | goto out_wait; | |
233 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
234 | } | |
235 | ||
236 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? | |
0cadda1c | 237 | 0 : XBF_ASYNC, FI_NONE); |
5a34d5cd DC |
238 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
239 | ||
240 | out_wait: | |
b0710ccc | 241 | if (flags & SYNC_WAIT) |
5a34d5cd | 242 | xfs_ioend_wait(ip); |
75f3cb13 | 243 | IRELE(ip); |
5a34d5cd DC |
244 | return error; |
245 | } | |
246 | ||
845b6d0c CH |
247 | STATIC int |
248 | xfs_sync_inode_attr( | |
249 | struct xfs_inode *ip, | |
75f3cb13 | 250 | struct xfs_perag *pag, |
845b6d0c CH |
251 | int flags) |
252 | { | |
253 | int error = 0; | |
254 | ||
75f3cb13 DC |
255 | error = xfs_sync_inode_valid(ip, pag); |
256 | if (error) | |
257 | return error; | |
258 | ||
845b6d0c CH |
259 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
260 | if (xfs_inode_clean(ip)) | |
261 | goto out_unlock; | |
262 | if (!xfs_iflock_nowait(ip)) { | |
263 | if (!(flags & SYNC_WAIT)) | |
264 | goto out_unlock; | |
265 | xfs_iflock(ip); | |
266 | } | |
267 | ||
268 | if (xfs_inode_clean(ip)) { | |
269 | xfs_ifunlock(ip); | |
270 | goto out_unlock; | |
271 | } | |
272 | ||
273 | error = xfs_iflush(ip, (flags & SYNC_WAIT) ? | |
274 | XFS_IFLUSH_SYNC : XFS_IFLUSH_DELWRI); | |
275 | ||
276 | out_unlock: | |
277 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
75f3cb13 | 278 | IRELE(ip); |
845b6d0c CH |
279 | return error; |
280 | } | |
281 | ||
075fe102 CH |
282 | /* |
283 | * Write out pagecache data for the whole filesystem. | |
284 | */ | |
683a8970 | 285 | int |
075fe102 CH |
286 | xfs_sync_data( |
287 | struct xfs_mount *mp, | |
288 | int flags) | |
683a8970 | 289 | { |
075fe102 | 290 | int error; |
fe4fa4b8 | 291 | |
b0710ccc | 292 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
fe4fa4b8 | 293 | |
075fe102 | 294 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags, |
c8e20be0 | 295 | XFS_ICI_NO_TAG, 0); |
075fe102 CH |
296 | if (error) |
297 | return XFS_ERROR(error); | |
e9f1c6ee | 298 | |
075fe102 CH |
299 | xfs_log_force(mp, 0, |
300 | (flags & SYNC_WAIT) ? | |
301 | XFS_LOG_FORCE | XFS_LOG_SYNC : | |
302 | XFS_LOG_FORCE); | |
303 | return 0; | |
304 | } | |
e9f1c6ee | 305 | |
075fe102 CH |
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); | |
75f3cb13 | 315 | |
075fe102 | 316 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags, |
c8e20be0 | 317 | XFS_ICI_NO_TAG, 0); |
fe4fa4b8 DC |
318 | } |
319 | ||
2af75df7 CH |
320 | STATIC int |
321 | xfs_commit_dummy_trans( | |
322 | struct xfs_mount *mp, | |
dce5065a | 323 | uint flags) |
2af75df7 CH |
324 | { |
325 | struct xfs_inode *ip = mp->m_rootip; | |
326 | struct xfs_trans *tp; | |
327 | int error; | |
dce5065a DC |
328 | int log_flags = XFS_LOG_FORCE; |
329 | ||
330 | if (flags & SYNC_WAIT) | |
331 | log_flags |= XFS_LOG_SYNC; | |
2af75df7 CH |
332 | |
333 | /* | |
334 | * Put a dummy transaction in the log to tell recovery | |
335 | * that all others are OK. | |
336 | */ | |
337 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
338 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
339 | if (error) { | |
340 | xfs_trans_cancel(tp, 0); | |
341 | return error; | |
342 | } | |
343 | ||
344 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
345 | ||
346 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
347 | xfs_trans_ihold(tp, ip); | |
348 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2af75df7 | 349 | error = xfs_trans_commit(tp, 0); |
2af75df7 CH |
350 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
351 | ||
dce5065a | 352 | /* the log force ensures this transaction is pushed to disk */ |
2af75df7 | 353 | xfs_log_force(mp, 0, log_flags); |
dce5065a | 354 | return error; |
2af75df7 CH |
355 | } |
356 | ||
5d77c0dc | 357 | STATIC int |
2af75df7 CH |
358 | xfs_sync_fsdata( |
359 | struct xfs_mount *mp, | |
360 | int flags) | |
361 | { | |
362 | struct xfs_buf *bp; | |
363 | struct xfs_buf_log_item *bip; | |
364 | int error = 0; | |
365 | ||
366 | /* | |
367 | * If this is xfssyncd() then only sync the superblock if we can | |
368 | * lock it without sleeping and it is not pinned. | |
369 | */ | |
8b5403a6 | 370 | if (flags & SYNC_TRYLOCK) { |
2af75df7 CH |
371 | ASSERT(!(flags & SYNC_WAIT)); |
372 | ||
0cadda1c | 373 | bp = xfs_getsb(mp, XBF_TRYLOCK); |
2af75df7 CH |
374 | if (!bp) |
375 | goto out; | |
376 | ||
377 | bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *); | |
378 | if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp)) | |
379 | goto out_brelse; | |
380 | } else { | |
381 | bp = xfs_getsb(mp, 0); | |
382 | ||
383 | /* | |
384 | * If the buffer is pinned then push on the log so we won't | |
385 | * get stuck waiting in the write for someone, maybe | |
386 | * ourselves, to flush the log. | |
387 | * | |
388 | * Even though we just pushed the log above, we did not have | |
389 | * the superblock buffer locked at that point so it can | |
390 | * become pinned in between there and here. | |
391 | */ | |
392 | if (XFS_BUF_ISPINNED(bp)) | |
393 | xfs_log_force(mp, 0, XFS_LOG_FORCE); | |
394 | } | |
395 | ||
396 | ||
397 | if (flags & SYNC_WAIT) | |
398 | XFS_BUF_UNASYNC(bp); | |
399 | else | |
400 | XFS_BUF_ASYNC(bp); | |
401 | ||
dce5065a DC |
402 | error = xfs_bwrite(mp, bp); |
403 | if (error) | |
404 | return error; | |
405 | ||
406 | /* | |
407 | * If this is a data integrity sync make sure all pending buffers | |
408 | * are flushed out for the log coverage check below. | |
409 | */ | |
410 | if (flags & SYNC_WAIT) | |
411 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
412 | ||
413 | if (xfs_log_need_covered(mp)) | |
414 | error = xfs_commit_dummy_trans(mp, flags); | |
415 | return error; | |
2af75df7 CH |
416 | |
417 | out_brelse: | |
418 | xfs_buf_relse(bp); | |
419 | out: | |
420 | return error; | |
e9f1c6ee DC |
421 | } |
422 | ||
423 | /* | |
a4e4c4f4 DC |
424 | * When remounting a filesystem read-only or freezing the filesystem, we have |
425 | * two phases to execute. This first phase is syncing the data before we | |
426 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
427 | * we've waited for all the transactions created by the first phase to | |
428 | * complete. The second phase ensures that the inodes are written to their | |
429 | * location on disk rather than just existing in transactions in the log. This | |
430 | * means after a quiesce there is no log replay required to write the inodes to | |
431 | * disk (this is the main difference between a sync and a quiesce). | |
432 | */ | |
433 | /* | |
434 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
435 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
436 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
437 | * transactions can still occur here so don't bother flushing the buftarg |
438 | * because it'll just get dirty again. | |
e9f1c6ee DC |
439 | */ |
440 | int | |
441 | xfs_quiesce_data( | |
442 | struct xfs_mount *mp) | |
443 | { | |
444 | int error; | |
445 | ||
446 | /* push non-blocking */ | |
075fe102 | 447 | xfs_sync_data(mp, 0); |
8b5403a6 | 448 | xfs_qm_sync(mp, SYNC_TRYLOCK); |
e9f1c6ee | 449 | |
c90b07e8 | 450 | /* push and block till complete */ |
b0710ccc | 451 | xfs_sync_data(mp, SYNC_WAIT); |
7d095257 | 452 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 453 | |
a4e4c4f4 | 454 | /* write superblock and hoover up shutdown errors */ |
c90b07e8 | 455 | error = xfs_sync_fsdata(mp, SYNC_WAIT); |
e9f1c6ee | 456 | |
a4e4c4f4 | 457 | /* flush data-only devices */ |
e9f1c6ee DC |
458 | if (mp->m_rtdev_targp) |
459 | XFS_bflush(mp->m_rtdev_targp); | |
460 | ||
461 | return error; | |
2af75df7 CH |
462 | } |
463 | ||
76bf105c DC |
464 | STATIC void |
465 | xfs_quiesce_fs( | |
466 | struct xfs_mount *mp) | |
467 | { | |
468 | int count = 0, pincount; | |
469 | ||
470 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
abc10647 | 471 | xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
76bf105c DC |
472 | |
473 | /* | |
474 | * This loop must run at least twice. The first instance of the loop | |
475 | * will flush most meta data but that will generate more meta data | |
476 | * (typically directory updates). Which then must be flushed and | |
477 | * logged before we can write the unmount record. | |
478 | */ | |
479 | do { | |
075fe102 | 480 | xfs_sync_attr(mp, SYNC_WAIT); |
76bf105c DC |
481 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
482 | if (!pincount) { | |
483 | delay(50); | |
484 | count++; | |
485 | } | |
486 | } while (count < 2); | |
487 | } | |
488 | ||
489 | /* | |
490 | * Second stage of a quiesce. The data is already synced, now we have to take | |
491 | * care of the metadata. New transactions are already blocked, so we need to | |
492 | * wait for any remaining transactions to drain out before proceding. | |
493 | */ | |
494 | void | |
495 | xfs_quiesce_attr( | |
496 | struct xfs_mount *mp) | |
497 | { | |
498 | int error = 0; | |
499 | ||
500 | /* wait for all modifications to complete */ | |
501 | while (atomic_read(&mp->m_active_trans) > 0) | |
502 | delay(100); | |
503 | ||
504 | /* flush inodes and push all remaining buffers out to disk */ | |
505 | xfs_quiesce_fs(mp); | |
506 | ||
5e106572 FB |
507 | /* |
508 | * Just warn here till VFS can correctly support | |
509 | * read-only remount without racing. | |
510 | */ | |
511 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
512 | |
513 | /* Push the superblock and write an unmount record */ | |
514 | error = xfs_log_sbcount(mp, 1); | |
515 | if (error) | |
516 | xfs_fs_cmn_err(CE_WARN, mp, | |
517 | "xfs_attr_quiesce: failed to log sb changes. " | |
518 | "Frozen image may not be consistent."); | |
519 | xfs_log_unmount_write(mp); | |
520 | xfs_unmountfs_writesb(mp); | |
521 | } | |
522 | ||
a167b17e DC |
523 | /* |
524 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
525 | * Doing this has two advantages: | |
526 | * - It saves on stack space, which is tight in certain situations | |
527 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
528 | * Flushing while allocating in a full filesystem requires both. | |
529 | */ | |
530 | STATIC void | |
531 | xfs_syncd_queue_work( | |
532 | struct xfs_mount *mp, | |
533 | void *data, | |
e43afd72 DC |
534 | void (*syncer)(struct xfs_mount *, void *), |
535 | struct completion *completion) | |
a167b17e | 536 | { |
a8d770d9 | 537 | struct xfs_sync_work *work; |
a167b17e | 538 | |
a8d770d9 | 539 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
a167b17e DC |
540 | INIT_LIST_HEAD(&work->w_list); |
541 | work->w_syncer = syncer; | |
542 | work->w_data = data; | |
543 | work->w_mount = mp; | |
e43afd72 | 544 | work->w_completion = completion; |
a167b17e DC |
545 | spin_lock(&mp->m_sync_lock); |
546 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
547 | spin_unlock(&mp->m_sync_lock); | |
548 | wake_up_process(mp->m_sync_task); | |
549 | } | |
550 | ||
551 | /* | |
552 | * Flush delayed allocate data, attempting to free up reserved space | |
553 | * from existing allocations. At this point a new allocation attempt | |
554 | * has failed with ENOSPC and we are in the process of scratching our | |
555 | * heads, looking about for more room... | |
556 | */ | |
557 | STATIC void | |
a8d770d9 | 558 | xfs_flush_inodes_work( |
a167b17e DC |
559 | struct xfs_mount *mp, |
560 | void *arg) | |
561 | { | |
562 | struct inode *inode = arg; | |
075fe102 | 563 | xfs_sync_data(mp, SYNC_TRYLOCK); |
b0710ccc | 564 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); |
a167b17e DC |
565 | iput(inode); |
566 | } | |
567 | ||
568 | void | |
a8d770d9 | 569 | xfs_flush_inodes( |
a167b17e DC |
570 | xfs_inode_t *ip) |
571 | { | |
572 | struct inode *inode = VFS_I(ip); | |
e43afd72 | 573 | DECLARE_COMPLETION_ONSTACK(completion); |
a167b17e DC |
574 | |
575 | igrab(inode); | |
e43afd72 DC |
576 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); |
577 | wait_for_completion(&completion); | |
a167b17e DC |
578 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); |
579 | } | |
580 | ||
aacaa880 DC |
581 | /* |
582 | * Every sync period we need to unpin all items, reclaim inodes, sync | |
583 | * quota and write out the superblock. We might need to cover the log | |
584 | * to indicate it is idle. | |
585 | */ | |
a167b17e DC |
586 | STATIC void |
587 | xfs_sync_worker( | |
588 | struct xfs_mount *mp, | |
589 | void *unused) | |
590 | { | |
591 | int error; | |
592 | ||
aacaa880 DC |
593 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
594 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
abc10647 | 595 | xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
aacaa880 | 596 | /* dgc: errors ignored here */ |
8b5403a6 CH |
597 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); |
598 | error = xfs_sync_fsdata(mp, SYNC_TRYLOCK); | |
aacaa880 | 599 | } |
a167b17e DC |
600 | mp->m_sync_seq++; |
601 | wake_up(&mp->m_wait_single_sync_task); | |
602 | } | |
603 | ||
604 | STATIC int | |
605 | xfssyncd( | |
606 | void *arg) | |
607 | { | |
608 | struct xfs_mount *mp = arg; | |
609 | long timeleft; | |
a8d770d9 | 610 | xfs_sync_work_t *work, *n; |
a167b17e DC |
611 | LIST_HEAD (tmp); |
612 | ||
613 | set_freezable(); | |
614 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
615 | for (;;) { | |
616 | timeleft = schedule_timeout_interruptible(timeleft); | |
617 | /* swsusp */ | |
618 | try_to_freeze(); | |
619 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
620 | break; | |
621 | ||
622 | spin_lock(&mp->m_sync_lock); | |
623 | /* | |
624 | * We can get woken by laptop mode, to do a sync - | |
625 | * that's the (only!) case where the list would be | |
626 | * empty with time remaining. | |
627 | */ | |
628 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
629 | if (!timeleft) | |
630 | timeleft = xfs_syncd_centisecs * | |
631 | msecs_to_jiffies(10); | |
632 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
633 | list_add_tail(&mp->m_sync_work.w_list, | |
634 | &mp->m_sync_list); | |
635 | } | |
636 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) | |
637 | list_move(&work->w_list, &tmp); | |
638 | spin_unlock(&mp->m_sync_lock); | |
639 | ||
640 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
641 | (*work->w_syncer)(mp, work->w_data); | |
642 | list_del(&work->w_list); | |
643 | if (work == &mp->m_sync_work) | |
644 | continue; | |
e43afd72 DC |
645 | if (work->w_completion) |
646 | complete(work->w_completion); | |
a167b17e DC |
647 | kmem_free(work); |
648 | } | |
649 | } | |
650 | ||
651 | return 0; | |
652 | } | |
653 | ||
654 | int | |
655 | xfs_syncd_init( | |
656 | struct xfs_mount *mp) | |
657 | { | |
658 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
659 | mp->m_sync_work.w_mount = mp; | |
e43afd72 | 660 | mp->m_sync_work.w_completion = NULL; |
a167b17e DC |
661 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); |
662 | if (IS_ERR(mp->m_sync_task)) | |
663 | return -PTR_ERR(mp->m_sync_task); | |
664 | return 0; | |
665 | } | |
666 | ||
667 | void | |
668 | xfs_syncd_stop( | |
669 | struct xfs_mount *mp) | |
670 | { | |
671 | kthread_stop(mp->m_sync_task); | |
672 | } | |
673 | ||
bc990f5c CH |
674 | void |
675 | __xfs_inode_set_reclaim_tag( | |
676 | struct xfs_perag *pag, | |
677 | struct xfs_inode *ip) | |
678 | { | |
679 | radix_tree_tag_set(&pag->pag_ici_root, | |
680 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
681 | XFS_ICI_RECLAIM_TAG); | |
682 | } | |
683 | ||
11654513 DC |
684 | /* |
685 | * We set the inode flag atomically with the radix tree tag. | |
686 | * Once we get tag lookups on the radix tree, this inode flag | |
687 | * can go away. | |
688 | */ | |
396beb85 DC |
689 | void |
690 | xfs_inode_set_reclaim_tag( | |
691 | xfs_inode_t *ip) | |
692 | { | |
5017e97d DC |
693 | struct xfs_mount *mp = ip->i_mount; |
694 | struct xfs_perag *pag; | |
396beb85 | 695 | |
5017e97d | 696 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
396beb85 DC |
697 | read_lock(&pag->pag_ici_lock); |
698 | spin_lock(&ip->i_flags_lock); | |
bc990f5c | 699 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 700 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
701 | spin_unlock(&ip->i_flags_lock); |
702 | read_unlock(&pag->pag_ici_lock); | |
5017e97d | 703 | xfs_perag_put(pag); |
396beb85 DC |
704 | } |
705 | ||
706 | void | |
707 | __xfs_inode_clear_reclaim_tag( | |
708 | xfs_mount_t *mp, | |
709 | xfs_perag_t *pag, | |
710 | xfs_inode_t *ip) | |
711 | { | |
712 | radix_tree_tag_clear(&pag->pag_ici_root, | |
713 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
714 | } | |
715 | ||
75f3cb13 | 716 | STATIC int |
c8e20be0 | 717 | xfs_reclaim_inode( |
75f3cb13 DC |
718 | struct xfs_inode *ip, |
719 | struct xfs_perag *pag, | |
c8e20be0 | 720 | int sync_mode) |
fce08f2f | 721 | { |
c8e20be0 DC |
722 | /* |
723 | * The radix tree lock here protects a thread in xfs_iget from racing | |
724 | * with us starting reclaim on the inode. Once we have the | |
725 | * XFS_IRECLAIM flag set it will not touch us. | |
726 | */ | |
727 | spin_lock(&ip->i_flags_lock); | |
728 | ASSERT_ALWAYS(__xfs_iflags_test(ip, XFS_IRECLAIMABLE)); | |
729 | if (__xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
730 | /* ignore as it is already under reclaim */ | |
731 | spin_unlock(&ip->i_flags_lock); | |
732 | write_unlock(&pag->pag_ici_lock); | |
75f3cb13 | 733 | return 0; |
fce08f2f | 734 | } |
c8e20be0 DC |
735 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
736 | spin_unlock(&ip->i_flags_lock); | |
737 | write_unlock(&pag->pag_ici_lock); | |
738 | ||
739 | /* | |
740 | * If the inode is still dirty, then flush it out. If the inode | |
741 | * is not in the AIL, then it will be OK to flush it delwri as | |
742 | * long as xfs_iflush() does not keep any references to the inode. | |
743 | * We leave that decision up to xfs_iflush() since it has the | |
744 | * knowledge of whether it's OK to simply do a delwri flush of | |
745 | * the inode or whether we need to wait until the inode is | |
746 | * pulled from the AIL. | |
747 | * We get the flush lock regardless, though, just to make sure | |
748 | * we don't free it while it is being flushed. | |
749 | */ | |
750 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
751 | xfs_iflock(ip); | |
7a3be02b | 752 | |
c8e20be0 DC |
753 | /* |
754 | * In the case of a forced shutdown we rely on xfs_iflush() to | |
755 | * wait for the inode to be unpinned before returning an error. | |
756 | */ | |
757 | if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) { | |
758 | /* synchronize with xfs_iflush_done */ | |
759 | xfs_iflock(ip); | |
760 | xfs_ifunlock(ip); | |
761 | } | |
762 | ||
763 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
764 | xfs_ireclaim(ip); | |
765 | return 0; | |
7a3be02b DC |
766 | } |
767 | ||
768 | int | |
769 | xfs_reclaim_inodes( | |
770 | xfs_mount_t *mp, | |
7a3be02b DC |
771 | int mode) |
772 | { | |
c8e20be0 DC |
773 | return xfs_inode_ag_iterator(mp, xfs_reclaim_inode, mode, |
774 | XFS_ICI_RECLAIM_TAG, 1); | |
fce08f2f | 775 | } |