]>
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" | |
fe4fa4b8 DC |
27 | #include "xfs_mount.h" |
28 | #include "xfs_bmap_btree.h" | |
fe4fa4b8 DC |
29 | #include "xfs_inode.h" |
30 | #include "xfs_dinode.h" | |
31 | #include "xfs_error.h" | |
fe4fa4b8 DC |
32 | #include "xfs_filestream.h" |
33 | #include "xfs_vnodeops.h" | |
fe4fa4b8 | 34 | #include "xfs_inode_item.h" |
7d095257 | 35 | #include "xfs_quota.h" |
0b1b213f | 36 | #include "xfs_trace.h" |
fe4fa4b8 | 37 | |
a167b17e DC |
38 | #include <linux/kthread.h> |
39 | #include <linux/freezer.h> | |
40 | ||
5a34d5cd | 41 | |
75f3cb13 DC |
42 | STATIC xfs_inode_t * |
43 | xfs_inode_ag_lookup( | |
44 | struct xfs_mount *mp, | |
45 | struct xfs_perag *pag, | |
46 | uint32_t *first_index, | |
47 | int tag) | |
48 | { | |
49 | int nr_found; | |
50 | struct xfs_inode *ip; | |
51 | ||
52 | /* | |
53 | * use a gang lookup to find the next inode in the tree | |
54 | * as the tree is sparse and a gang lookup walks to find | |
55 | * the number of objects requested. | |
56 | */ | |
75f3cb13 DC |
57 | if (tag == XFS_ICI_NO_TAG) { |
58 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
59 | (void **)&ip, *first_index, 1); | |
60 | } else { | |
61 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, | |
62 | (void **)&ip, *first_index, 1, tag); | |
63 | } | |
64 | if (!nr_found) | |
c8e20be0 | 65 | return NULL; |
75f3cb13 DC |
66 | |
67 | /* | |
68 | * Update the index for the next lookup. Catch overflows | |
69 | * into the next AG range which can occur if we have inodes | |
70 | * in the last block of the AG and we are currently | |
71 | * pointing to the last inode. | |
72 | */ | |
73 | *first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
74 | if (*first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
c8e20be0 | 75 | return NULL; |
75f3cb13 | 76 | return ip; |
75f3cb13 DC |
77 | } |
78 | ||
79 | STATIC int | |
80 | xfs_inode_ag_walk( | |
81 | struct xfs_mount *mp, | |
5017e97d | 82 | struct xfs_perag *pag, |
75f3cb13 DC |
83 | int (*execute)(struct xfs_inode *ip, |
84 | struct xfs_perag *pag, int flags), | |
85 | int flags, | |
c8e20be0 | 86 | int tag, |
9bf729c0 DC |
87 | int exclusive, |
88 | int *nr_to_scan) | |
75f3cb13 | 89 | { |
75f3cb13 DC |
90 | uint32_t first_index; |
91 | int last_error = 0; | |
92 | int skipped; | |
93 | ||
94 | restart: | |
95 | skipped = 0; | |
96 | first_index = 0; | |
97 | do { | |
98 | int error = 0; | |
99 | xfs_inode_t *ip; | |
100 | ||
c8e20be0 DC |
101 | if (exclusive) |
102 | write_lock(&pag->pag_ici_lock); | |
103 | else | |
104 | read_lock(&pag->pag_ici_lock); | |
75f3cb13 | 105 | ip = xfs_inode_ag_lookup(mp, pag, &first_index, tag); |
c8e20be0 DC |
106 | if (!ip) { |
107 | if (exclusive) | |
108 | write_unlock(&pag->pag_ici_lock); | |
109 | else | |
110 | read_unlock(&pag->pag_ici_lock); | |
75f3cb13 | 111 | break; |
c8e20be0 | 112 | } |
75f3cb13 | 113 | |
c8e20be0 | 114 | /* execute releases pag->pag_ici_lock */ |
75f3cb13 DC |
115 | error = execute(ip, pag, flags); |
116 | if (error == EAGAIN) { | |
117 | skipped++; | |
118 | continue; | |
119 | } | |
120 | if (error) | |
121 | last_error = error; | |
c8e20be0 DC |
122 | |
123 | /* bail out if the filesystem is corrupted. */ | |
75f3cb13 DC |
124 | if (error == EFSCORRUPTED) |
125 | break; | |
126 | ||
9bf729c0 | 127 | } while ((*nr_to_scan)--); |
75f3cb13 DC |
128 | |
129 | if (skipped) { | |
130 | delay(1); | |
131 | goto restart; | |
132 | } | |
75f3cb13 DC |
133 | return last_error; |
134 | } | |
135 | ||
16fd5367 DC |
136 | /* |
137 | * Select the next per-ag structure to iterate during the walk. The reclaim | |
138 | * walk is optimised only to walk AGs with reclaimable inodes in them. | |
139 | */ | |
140 | static struct xfs_perag * | |
141 | xfs_inode_ag_iter_next_pag( | |
142 | struct xfs_mount *mp, | |
143 | xfs_agnumber_t *first, | |
144 | int tag) | |
145 | { | |
146 | struct xfs_perag *pag = NULL; | |
147 | ||
148 | if (tag == XFS_ICI_RECLAIM_TAG) { | |
149 | int found; | |
150 | int ref; | |
151 | ||
152 | spin_lock(&mp->m_perag_lock); | |
153 | found = radix_tree_gang_lookup_tag(&mp->m_perag_tree, | |
154 | (void **)&pag, *first, 1, tag); | |
155 | if (found <= 0) { | |
156 | spin_unlock(&mp->m_perag_lock); | |
157 | return NULL; | |
158 | } | |
159 | *first = pag->pag_agno + 1; | |
160 | /* open coded pag reference increment */ | |
161 | ref = atomic_inc_return(&pag->pag_ref); | |
162 | spin_unlock(&mp->m_perag_lock); | |
163 | trace_xfs_perag_get_reclaim(mp, pag->pag_agno, ref, _RET_IP_); | |
164 | } else { | |
165 | pag = xfs_perag_get(mp, *first); | |
166 | (*first)++; | |
167 | } | |
168 | return pag; | |
169 | } | |
170 | ||
fe588ed3 | 171 | int |
75f3cb13 DC |
172 | xfs_inode_ag_iterator( |
173 | struct xfs_mount *mp, | |
174 | int (*execute)(struct xfs_inode *ip, | |
175 | struct xfs_perag *pag, int flags), | |
176 | int flags, | |
c8e20be0 | 177 | int tag, |
9bf729c0 DC |
178 | int exclusive, |
179 | int *nr_to_scan) | |
75f3cb13 | 180 | { |
16fd5367 | 181 | struct xfs_perag *pag; |
75f3cb13 DC |
182 | int error = 0; |
183 | int last_error = 0; | |
184 | xfs_agnumber_t ag; | |
9bf729c0 | 185 | int nr; |
75f3cb13 | 186 | |
9bf729c0 | 187 | nr = nr_to_scan ? *nr_to_scan : INT_MAX; |
16fd5367 DC |
188 | ag = 0; |
189 | while ((pag = xfs_inode_ag_iter_next_pag(mp, &ag, tag))) { | |
5017e97d | 190 | error = xfs_inode_ag_walk(mp, pag, execute, flags, tag, |
9bf729c0 | 191 | exclusive, &nr); |
5017e97d | 192 | xfs_perag_put(pag); |
75f3cb13 DC |
193 | if (error) { |
194 | last_error = error; | |
195 | if (error == EFSCORRUPTED) | |
196 | break; | |
197 | } | |
9bf729c0 DC |
198 | if (nr <= 0) |
199 | break; | |
75f3cb13 | 200 | } |
9bf729c0 DC |
201 | if (nr_to_scan) |
202 | *nr_to_scan = nr; | |
75f3cb13 DC |
203 | return XFS_ERROR(last_error); |
204 | } | |
205 | ||
1da8eeca | 206 | /* must be called with pag_ici_lock held and releases it */ |
fe588ed3 | 207 | int |
1da8eeca DC |
208 | xfs_sync_inode_valid( |
209 | struct xfs_inode *ip, | |
210 | struct xfs_perag *pag) | |
211 | { | |
212 | struct inode *inode = VFS_I(ip); | |
018027be | 213 | int error = EFSCORRUPTED; |
1da8eeca DC |
214 | |
215 | /* nothing to sync during shutdown */ | |
018027be DC |
216 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
217 | goto out_unlock; | |
1da8eeca | 218 | |
018027be DC |
219 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ |
220 | error = ENOENT; | |
221 | if (xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
222 | goto out_unlock; | |
1da8eeca | 223 | |
018027be DC |
224 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
225 | if (!igrab(inode)) | |
226 | goto out_unlock; | |
227 | ||
228 | if (is_bad_inode(inode)) { | |
1da8eeca | 229 | IRELE(ip); |
018027be | 230 | goto out_unlock; |
1da8eeca DC |
231 | } |
232 | ||
018027be DC |
233 | /* inode is valid */ |
234 | error = 0; | |
235 | out_unlock: | |
236 | read_unlock(&pag->pag_ici_lock); | |
237 | return error; | |
1da8eeca DC |
238 | } |
239 | ||
5a34d5cd DC |
240 | STATIC int |
241 | xfs_sync_inode_data( | |
242 | struct xfs_inode *ip, | |
75f3cb13 | 243 | struct xfs_perag *pag, |
5a34d5cd DC |
244 | int flags) |
245 | { | |
246 | struct inode *inode = VFS_I(ip); | |
247 | struct address_space *mapping = inode->i_mapping; | |
248 | int error = 0; | |
249 | ||
75f3cb13 DC |
250 | error = xfs_sync_inode_valid(ip, pag); |
251 | if (error) | |
252 | return error; | |
253 | ||
5a34d5cd DC |
254 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
255 | goto out_wait; | |
256 | ||
257 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
258 | if (flags & SYNC_TRYLOCK) | |
259 | goto out_wait; | |
260 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
261 | } | |
262 | ||
263 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? | |
0cadda1c | 264 | 0 : XBF_ASYNC, FI_NONE); |
5a34d5cd DC |
265 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
266 | ||
267 | out_wait: | |
b0710ccc | 268 | if (flags & SYNC_WAIT) |
5a34d5cd | 269 | xfs_ioend_wait(ip); |
75f3cb13 | 270 | IRELE(ip); |
5a34d5cd DC |
271 | return error; |
272 | } | |
273 | ||
845b6d0c CH |
274 | STATIC int |
275 | xfs_sync_inode_attr( | |
276 | struct xfs_inode *ip, | |
75f3cb13 | 277 | struct xfs_perag *pag, |
845b6d0c CH |
278 | int flags) |
279 | { | |
280 | int error = 0; | |
281 | ||
75f3cb13 DC |
282 | error = xfs_sync_inode_valid(ip, pag); |
283 | if (error) | |
284 | return error; | |
285 | ||
845b6d0c CH |
286 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
287 | if (xfs_inode_clean(ip)) | |
288 | goto out_unlock; | |
289 | if (!xfs_iflock_nowait(ip)) { | |
290 | if (!(flags & SYNC_WAIT)) | |
291 | goto out_unlock; | |
292 | xfs_iflock(ip); | |
293 | } | |
294 | ||
295 | if (xfs_inode_clean(ip)) { | |
296 | xfs_ifunlock(ip); | |
297 | goto out_unlock; | |
298 | } | |
299 | ||
c854363e | 300 | error = xfs_iflush(ip, flags); |
845b6d0c CH |
301 | |
302 | out_unlock: | |
303 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
75f3cb13 | 304 | IRELE(ip); |
845b6d0c CH |
305 | return error; |
306 | } | |
307 | ||
075fe102 CH |
308 | /* |
309 | * Write out pagecache data for the whole filesystem. | |
310 | */ | |
64c86149 | 311 | STATIC int |
075fe102 CH |
312 | xfs_sync_data( |
313 | struct xfs_mount *mp, | |
314 | int flags) | |
683a8970 | 315 | { |
075fe102 | 316 | int error; |
fe4fa4b8 | 317 | |
b0710ccc | 318 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
fe4fa4b8 | 319 | |
075fe102 | 320 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags, |
9bf729c0 | 321 | XFS_ICI_NO_TAG, 0, NULL); |
075fe102 CH |
322 | if (error) |
323 | return XFS_ERROR(error); | |
e9f1c6ee | 324 | |
a14a348b | 325 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); |
075fe102 CH |
326 | return 0; |
327 | } | |
e9f1c6ee | 328 | |
075fe102 CH |
329 | /* |
330 | * Write out inode metadata (attributes) for the whole filesystem. | |
331 | */ | |
64c86149 | 332 | STATIC int |
075fe102 CH |
333 | xfs_sync_attr( |
334 | struct xfs_mount *mp, | |
335 | int flags) | |
336 | { | |
337 | ASSERT((flags & ~SYNC_WAIT) == 0); | |
75f3cb13 | 338 | |
075fe102 | 339 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags, |
9bf729c0 | 340 | XFS_ICI_NO_TAG, 0, NULL); |
fe4fa4b8 DC |
341 | } |
342 | ||
2af75df7 CH |
343 | STATIC int |
344 | xfs_commit_dummy_trans( | |
345 | struct xfs_mount *mp, | |
dce5065a | 346 | uint flags) |
2af75df7 CH |
347 | { |
348 | struct xfs_inode *ip = mp->m_rootip; | |
349 | struct xfs_trans *tp; | |
350 | int error; | |
351 | ||
352 | /* | |
353 | * Put a dummy transaction in the log to tell recovery | |
354 | * that all others are OK. | |
355 | */ | |
356 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
357 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
358 | if (error) { | |
359 | xfs_trans_cancel(tp, 0); | |
360 | return error; | |
361 | } | |
362 | ||
363 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
364 | ||
898621d5 | 365 | xfs_trans_ijoin(tp, ip); |
2af75df7 | 366 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2af75df7 | 367 | error = xfs_trans_commit(tp, 0); |
2af75df7 CH |
368 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
369 | ||
dce5065a | 370 | /* the log force ensures this transaction is pushed to disk */ |
a14a348b | 371 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); |
dce5065a | 372 | return error; |
2af75df7 CH |
373 | } |
374 | ||
5d77c0dc | 375 | STATIC int |
2af75df7 | 376 | xfs_sync_fsdata( |
df308bcf | 377 | struct xfs_mount *mp) |
2af75df7 CH |
378 | { |
379 | struct xfs_buf *bp; | |
2af75df7 CH |
380 | |
381 | /* | |
df308bcf CH |
382 | * If the buffer is pinned then push on the log so we won't get stuck |
383 | * waiting in the write for someone, maybe ourselves, to flush the log. | |
384 | * | |
385 | * Even though we just pushed the log above, we did not have the | |
386 | * superblock buffer locked at that point so it can become pinned in | |
387 | * between there and here. | |
2af75df7 | 388 | */ |
df308bcf CH |
389 | bp = xfs_getsb(mp, 0); |
390 | if (XFS_BUF_ISPINNED(bp)) | |
391 | xfs_log_force(mp, 0); | |
2af75df7 | 392 | |
df308bcf | 393 | return xfs_bwrite(mp, bp); |
e9f1c6ee DC |
394 | } |
395 | ||
396 | /* | |
a4e4c4f4 DC |
397 | * When remounting a filesystem read-only or freezing the filesystem, we have |
398 | * two phases to execute. This first phase is syncing the data before we | |
399 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
400 | * we've waited for all the transactions created by the first phase to | |
401 | * complete. The second phase ensures that the inodes are written to their | |
402 | * location on disk rather than just existing in transactions in the log. This | |
403 | * means after a quiesce there is no log replay required to write the inodes to | |
404 | * disk (this is the main difference between a sync and a quiesce). | |
405 | */ | |
406 | /* | |
407 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
408 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
409 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
410 | * transactions can still occur here so don't bother flushing the buftarg |
411 | * because it'll just get dirty again. | |
e9f1c6ee DC |
412 | */ |
413 | int | |
414 | xfs_quiesce_data( | |
415 | struct xfs_mount *mp) | |
416 | { | |
df308bcf | 417 | int error, error2 = 0; |
e9f1c6ee DC |
418 | |
419 | /* push non-blocking */ | |
075fe102 | 420 | xfs_sync_data(mp, 0); |
8b5403a6 | 421 | xfs_qm_sync(mp, SYNC_TRYLOCK); |
e9f1c6ee | 422 | |
c90b07e8 | 423 | /* push and block till complete */ |
b0710ccc | 424 | xfs_sync_data(mp, SYNC_WAIT); |
7d095257 | 425 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 426 | |
a4e4c4f4 | 427 | /* write superblock and hoover up shutdown errors */ |
df308bcf CH |
428 | error = xfs_sync_fsdata(mp); |
429 | ||
430 | /* make sure all delwri buffers are written out */ | |
431 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
432 | ||
433 | /* mark the log as covered if needed */ | |
434 | if (xfs_log_need_covered(mp)) | |
435 | error2 = xfs_commit_dummy_trans(mp, SYNC_WAIT); | |
e9f1c6ee | 436 | |
a4e4c4f4 | 437 | /* flush data-only devices */ |
e9f1c6ee DC |
438 | if (mp->m_rtdev_targp) |
439 | XFS_bflush(mp->m_rtdev_targp); | |
440 | ||
df308bcf | 441 | return error ? error : error2; |
2af75df7 CH |
442 | } |
443 | ||
76bf105c DC |
444 | STATIC void |
445 | xfs_quiesce_fs( | |
446 | struct xfs_mount *mp) | |
447 | { | |
448 | int count = 0, pincount; | |
449 | ||
c854363e | 450 | xfs_reclaim_inodes(mp, 0); |
76bf105c | 451 | xfs_flush_buftarg(mp->m_ddev_targp, 0); |
76bf105c DC |
452 | |
453 | /* | |
454 | * This loop must run at least twice. The first instance of the loop | |
455 | * will flush most meta data but that will generate more meta data | |
456 | * (typically directory updates). Which then must be flushed and | |
c854363e DC |
457 | * logged before we can write the unmount record. We also so sync |
458 | * reclaim of inodes to catch any that the above delwri flush skipped. | |
76bf105c DC |
459 | */ |
460 | do { | |
c854363e | 461 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
075fe102 | 462 | xfs_sync_attr(mp, SYNC_WAIT); |
76bf105c DC |
463 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
464 | if (!pincount) { | |
465 | delay(50); | |
466 | count++; | |
467 | } | |
468 | } while (count < 2); | |
469 | } | |
470 | ||
471 | /* | |
472 | * Second stage of a quiesce. The data is already synced, now we have to take | |
473 | * care of the metadata. New transactions are already blocked, so we need to | |
474 | * wait for any remaining transactions to drain out before proceding. | |
475 | */ | |
476 | void | |
477 | xfs_quiesce_attr( | |
478 | struct xfs_mount *mp) | |
479 | { | |
480 | int error = 0; | |
481 | ||
482 | /* wait for all modifications to complete */ | |
483 | while (atomic_read(&mp->m_active_trans) > 0) | |
484 | delay(100); | |
485 | ||
486 | /* flush inodes and push all remaining buffers out to disk */ | |
487 | xfs_quiesce_fs(mp); | |
488 | ||
5e106572 FB |
489 | /* |
490 | * Just warn here till VFS can correctly support | |
491 | * read-only remount without racing. | |
492 | */ | |
493 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
494 | |
495 | /* Push the superblock and write an unmount record */ | |
496 | error = xfs_log_sbcount(mp, 1); | |
497 | if (error) | |
498 | xfs_fs_cmn_err(CE_WARN, mp, | |
499 | "xfs_attr_quiesce: failed to log sb changes. " | |
500 | "Frozen image may not be consistent."); | |
501 | xfs_log_unmount_write(mp); | |
502 | xfs_unmountfs_writesb(mp); | |
503 | } | |
504 | ||
a167b17e DC |
505 | /* |
506 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
507 | * Doing this has two advantages: | |
508 | * - It saves on stack space, which is tight in certain situations | |
509 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
510 | * Flushing while allocating in a full filesystem requires both. | |
511 | */ | |
512 | STATIC void | |
513 | xfs_syncd_queue_work( | |
514 | struct xfs_mount *mp, | |
515 | void *data, | |
e43afd72 DC |
516 | void (*syncer)(struct xfs_mount *, void *), |
517 | struct completion *completion) | |
a167b17e | 518 | { |
a8d770d9 | 519 | struct xfs_sync_work *work; |
a167b17e | 520 | |
a8d770d9 | 521 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
a167b17e DC |
522 | INIT_LIST_HEAD(&work->w_list); |
523 | work->w_syncer = syncer; | |
524 | work->w_data = data; | |
525 | work->w_mount = mp; | |
e43afd72 | 526 | work->w_completion = completion; |
a167b17e DC |
527 | spin_lock(&mp->m_sync_lock); |
528 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
529 | spin_unlock(&mp->m_sync_lock); | |
530 | wake_up_process(mp->m_sync_task); | |
531 | } | |
532 | ||
533 | /* | |
534 | * Flush delayed allocate data, attempting to free up reserved space | |
535 | * from existing allocations. At this point a new allocation attempt | |
536 | * has failed with ENOSPC and we are in the process of scratching our | |
537 | * heads, looking about for more room... | |
538 | */ | |
539 | STATIC void | |
a8d770d9 | 540 | xfs_flush_inodes_work( |
a167b17e DC |
541 | struct xfs_mount *mp, |
542 | void *arg) | |
543 | { | |
544 | struct inode *inode = arg; | |
075fe102 | 545 | xfs_sync_data(mp, SYNC_TRYLOCK); |
b0710ccc | 546 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); |
a167b17e DC |
547 | iput(inode); |
548 | } | |
549 | ||
550 | void | |
a8d770d9 | 551 | xfs_flush_inodes( |
a167b17e DC |
552 | xfs_inode_t *ip) |
553 | { | |
554 | struct inode *inode = VFS_I(ip); | |
e43afd72 | 555 | DECLARE_COMPLETION_ONSTACK(completion); |
a167b17e DC |
556 | |
557 | igrab(inode); | |
e43afd72 DC |
558 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); |
559 | wait_for_completion(&completion); | |
a14a348b | 560 | xfs_log_force(ip->i_mount, XFS_LOG_SYNC); |
a167b17e DC |
561 | } |
562 | ||
aacaa880 | 563 | /* |
df308bcf CH |
564 | * Every sync period we need to unpin all items, reclaim inodes and sync |
565 | * disk quotas. We might need to cover the log to indicate that the | |
566 | * filesystem is idle. | |
aacaa880 | 567 | */ |
a167b17e DC |
568 | STATIC void |
569 | xfs_sync_worker( | |
570 | struct xfs_mount *mp, | |
571 | void *unused) | |
572 | { | |
573 | int error; | |
574 | ||
aacaa880 | 575 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
a14a348b | 576 | xfs_log_force(mp, 0); |
c854363e | 577 | xfs_reclaim_inodes(mp, 0); |
aacaa880 | 578 | /* dgc: errors ignored here */ |
8b5403a6 | 579 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); |
df308bcf CH |
580 | if (xfs_log_need_covered(mp)) |
581 | error = xfs_commit_dummy_trans(mp, 0); | |
aacaa880 | 582 | } |
a167b17e DC |
583 | mp->m_sync_seq++; |
584 | wake_up(&mp->m_wait_single_sync_task); | |
585 | } | |
586 | ||
587 | STATIC int | |
588 | xfssyncd( | |
589 | void *arg) | |
590 | { | |
591 | struct xfs_mount *mp = arg; | |
592 | long timeleft; | |
a8d770d9 | 593 | xfs_sync_work_t *work, *n; |
a167b17e DC |
594 | LIST_HEAD (tmp); |
595 | ||
596 | set_freezable(); | |
597 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
598 | for (;;) { | |
20f6b2c7 DC |
599 | if (list_empty(&mp->m_sync_list)) |
600 | timeleft = schedule_timeout_interruptible(timeleft); | |
a167b17e DC |
601 | /* swsusp */ |
602 | try_to_freeze(); | |
603 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
604 | break; | |
605 | ||
606 | spin_lock(&mp->m_sync_lock); | |
607 | /* | |
608 | * We can get woken by laptop mode, to do a sync - | |
609 | * that's the (only!) case where the list would be | |
610 | * empty with time remaining. | |
611 | */ | |
612 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
613 | if (!timeleft) | |
614 | timeleft = xfs_syncd_centisecs * | |
615 | msecs_to_jiffies(10); | |
616 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
617 | list_add_tail(&mp->m_sync_work.w_list, | |
618 | &mp->m_sync_list); | |
619 | } | |
20f6b2c7 | 620 | list_splice_init(&mp->m_sync_list, &tmp); |
a167b17e DC |
621 | spin_unlock(&mp->m_sync_lock); |
622 | ||
623 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
624 | (*work->w_syncer)(mp, work->w_data); | |
625 | list_del(&work->w_list); | |
626 | if (work == &mp->m_sync_work) | |
627 | continue; | |
e43afd72 DC |
628 | if (work->w_completion) |
629 | complete(work->w_completion); | |
a167b17e DC |
630 | kmem_free(work); |
631 | } | |
632 | } | |
633 | ||
634 | return 0; | |
635 | } | |
636 | ||
637 | int | |
638 | xfs_syncd_init( | |
639 | struct xfs_mount *mp) | |
640 | { | |
641 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
642 | mp->m_sync_work.w_mount = mp; | |
e43afd72 | 643 | mp->m_sync_work.w_completion = NULL; |
e2a07812 | 644 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd/%s", mp->m_fsname); |
a167b17e DC |
645 | if (IS_ERR(mp->m_sync_task)) |
646 | return -PTR_ERR(mp->m_sync_task); | |
647 | return 0; | |
648 | } | |
649 | ||
650 | void | |
651 | xfs_syncd_stop( | |
652 | struct xfs_mount *mp) | |
653 | { | |
654 | kthread_stop(mp->m_sync_task); | |
655 | } | |
656 | ||
bc990f5c CH |
657 | void |
658 | __xfs_inode_set_reclaim_tag( | |
659 | struct xfs_perag *pag, | |
660 | struct xfs_inode *ip) | |
661 | { | |
662 | radix_tree_tag_set(&pag->pag_ici_root, | |
663 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
664 | XFS_ICI_RECLAIM_TAG); | |
16fd5367 DC |
665 | |
666 | if (!pag->pag_ici_reclaimable) { | |
667 | /* propagate the reclaim tag up into the perag radix tree */ | |
668 | spin_lock(&ip->i_mount->m_perag_lock); | |
669 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
670 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
671 | XFS_ICI_RECLAIM_TAG); | |
672 | spin_unlock(&ip->i_mount->m_perag_lock); | |
673 | trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno, | |
674 | -1, _RET_IP_); | |
675 | } | |
9bf729c0 | 676 | pag->pag_ici_reclaimable++; |
bc990f5c CH |
677 | } |
678 | ||
11654513 DC |
679 | /* |
680 | * We set the inode flag atomically with the radix tree tag. | |
681 | * Once we get tag lookups on the radix tree, this inode flag | |
682 | * can go away. | |
683 | */ | |
396beb85 DC |
684 | void |
685 | xfs_inode_set_reclaim_tag( | |
686 | xfs_inode_t *ip) | |
687 | { | |
5017e97d DC |
688 | struct xfs_mount *mp = ip->i_mount; |
689 | struct xfs_perag *pag; | |
396beb85 | 690 | |
5017e97d | 691 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
f1f724e4 | 692 | write_lock(&pag->pag_ici_lock); |
396beb85 | 693 | spin_lock(&ip->i_flags_lock); |
bc990f5c | 694 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 695 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 | 696 | spin_unlock(&ip->i_flags_lock); |
f1f724e4 | 697 | write_unlock(&pag->pag_ici_lock); |
5017e97d | 698 | xfs_perag_put(pag); |
396beb85 DC |
699 | } |
700 | ||
701 | void | |
702 | __xfs_inode_clear_reclaim_tag( | |
703 | xfs_mount_t *mp, | |
704 | xfs_perag_t *pag, | |
705 | xfs_inode_t *ip) | |
706 | { | |
707 | radix_tree_tag_clear(&pag->pag_ici_root, | |
708 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
9bf729c0 | 709 | pag->pag_ici_reclaimable--; |
16fd5367 DC |
710 | if (!pag->pag_ici_reclaimable) { |
711 | /* clear the reclaim tag from the perag radix tree */ | |
712 | spin_lock(&ip->i_mount->m_perag_lock); | |
713 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
714 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
715 | XFS_ICI_RECLAIM_TAG); | |
716 | spin_unlock(&ip->i_mount->m_perag_lock); | |
717 | trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno, | |
718 | -1, _RET_IP_); | |
719 | } | |
396beb85 DC |
720 | } |
721 | ||
777df5af DC |
722 | /* |
723 | * Inodes in different states need to be treated differently, and the return | |
724 | * value of xfs_iflush is not sufficient to get this right. The following table | |
725 | * lists the inode states and the reclaim actions necessary for non-blocking | |
726 | * reclaim: | |
727 | * | |
728 | * | |
729 | * inode state iflush ret required action | |
730 | * --------------- ---------- --------------- | |
731 | * bad - reclaim | |
732 | * shutdown EIO unpin and reclaim | |
733 | * clean, unpinned 0 reclaim | |
734 | * stale, unpinned 0 reclaim | |
c854363e DC |
735 | * clean, pinned(*) 0 requeue |
736 | * stale, pinned EAGAIN requeue | |
737 | * dirty, delwri ok 0 requeue | |
738 | * dirty, delwri blocked EAGAIN requeue | |
739 | * dirty, sync flush 0 reclaim | |
777df5af DC |
740 | * |
741 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
742 | * handled anyway given the order of checks implemented. | |
743 | * | |
c854363e DC |
744 | * As can be seen from the table, the return value of xfs_iflush() is not |
745 | * sufficient to correctly decide the reclaim action here. The checks in | |
746 | * xfs_iflush() might look like duplicates, but they are not. | |
747 | * | |
748 | * Also, because we get the flush lock first, we know that any inode that has | |
749 | * been flushed delwri has had the flush completed by the time we check that | |
750 | * the inode is clean. The clean inode check needs to be done before flushing | |
751 | * the inode delwri otherwise we would loop forever requeuing clean inodes as | |
752 | * we cannot tell apart a successful delwri flush and a clean inode from the | |
753 | * return value of xfs_iflush(). | |
754 | * | |
755 | * Note that because the inode is flushed delayed write by background | |
756 | * writeback, the flush lock may already be held here and waiting on it can | |
757 | * result in very long latencies. Hence for sync reclaims, where we wait on the | |
758 | * flush lock, the caller should push out delayed write inodes first before | |
759 | * trying to reclaim them to minimise the amount of time spent waiting. For | |
760 | * background relaim, we just requeue the inode for the next pass. | |
761 | * | |
777df5af DC |
762 | * Hence the order of actions after gaining the locks should be: |
763 | * bad => reclaim | |
764 | * shutdown => unpin and reclaim | |
c854363e DC |
765 | * pinned, delwri => requeue |
766 | * pinned, sync => unpin | |
777df5af DC |
767 | * stale => reclaim |
768 | * clean => reclaim | |
c854363e DC |
769 | * dirty, delwri => flush and requeue |
770 | * dirty, sync => flush, wait and reclaim | |
777df5af | 771 | */ |
75f3cb13 | 772 | STATIC int |
c8e20be0 | 773 | xfs_reclaim_inode( |
75f3cb13 DC |
774 | struct xfs_inode *ip, |
775 | struct xfs_perag *pag, | |
c8e20be0 | 776 | int sync_mode) |
fce08f2f | 777 | { |
c854363e | 778 | int error = 0; |
777df5af | 779 | |
c8e20be0 DC |
780 | /* |
781 | * The radix tree lock here protects a thread in xfs_iget from racing | |
782 | * with us starting reclaim on the inode. Once we have the | |
783 | * XFS_IRECLAIM flag set it will not touch us. | |
784 | */ | |
785 | spin_lock(&ip->i_flags_lock); | |
786 | ASSERT_ALWAYS(__xfs_iflags_test(ip, XFS_IRECLAIMABLE)); | |
787 | if (__xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
788 | /* ignore as it is already under reclaim */ | |
789 | spin_unlock(&ip->i_flags_lock); | |
790 | write_unlock(&pag->pag_ici_lock); | |
75f3cb13 | 791 | return 0; |
fce08f2f | 792 | } |
c8e20be0 DC |
793 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
794 | spin_unlock(&ip->i_flags_lock); | |
795 | write_unlock(&pag->pag_ici_lock); | |
796 | ||
c8e20be0 | 797 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
798 | if (!xfs_iflock_nowait(ip)) { |
799 | if (!(sync_mode & SYNC_WAIT)) | |
800 | goto out; | |
801 | xfs_iflock(ip); | |
802 | } | |
7a3be02b | 803 | |
777df5af DC |
804 | if (is_bad_inode(VFS_I(ip))) |
805 | goto reclaim; | |
806 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
807 | xfs_iunpin_wait(ip); | |
808 | goto reclaim; | |
809 | } | |
c854363e DC |
810 | if (xfs_ipincount(ip)) { |
811 | if (!(sync_mode & SYNC_WAIT)) { | |
812 | xfs_ifunlock(ip); | |
813 | goto out; | |
814 | } | |
777df5af | 815 | xfs_iunpin_wait(ip); |
c854363e | 816 | } |
777df5af DC |
817 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
818 | goto reclaim; | |
819 | if (xfs_inode_clean(ip)) | |
820 | goto reclaim; | |
821 | ||
822 | /* Now we have an inode that needs flushing */ | |
823 | error = xfs_iflush(ip, sync_mode); | |
c854363e DC |
824 | if (sync_mode & SYNC_WAIT) { |
825 | xfs_iflock(ip); | |
826 | goto reclaim; | |
c8e20be0 DC |
827 | } |
828 | ||
c854363e DC |
829 | /* |
830 | * When we have to flush an inode but don't have SYNC_WAIT set, we | |
831 | * flush the inode out using a delwri buffer and wait for the next | |
832 | * call into reclaim to find it in a clean state instead of waiting for | |
833 | * it now. We also don't return errors here - if the error is transient | |
834 | * then the next reclaim pass will flush the inode, and if the error | |
f1d486a3 | 835 | * is permanent then the next sync reclaim will reclaim the inode and |
c854363e DC |
836 | * pass on the error. |
837 | */ | |
f1d486a3 | 838 | if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
c854363e DC |
839 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, |
840 | "inode 0x%llx background reclaim flush failed with %d", | |
841 | (long long)ip->i_ino, error); | |
842 | } | |
843 | out: | |
844 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
845 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
846 | /* | |
847 | * We could return EAGAIN here to make reclaim rescan the inode tree in | |
848 | * a short while. However, this just burns CPU time scanning the tree | |
849 | * waiting for IO to complete and xfssyncd never goes back to the idle | |
850 | * state. Instead, return 0 to let the next scheduled background reclaim | |
851 | * attempt to reclaim the inode again. | |
852 | */ | |
853 | return 0; | |
854 | ||
777df5af DC |
855 | reclaim: |
856 | xfs_ifunlock(ip); | |
c8e20be0 DC |
857 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
858 | xfs_ireclaim(ip); | |
c854363e DC |
859 | return error; |
860 | ||
7a3be02b DC |
861 | } |
862 | ||
863 | int | |
864 | xfs_reclaim_inodes( | |
865 | xfs_mount_t *mp, | |
7a3be02b DC |
866 | int mode) |
867 | { | |
c8e20be0 | 868 | return xfs_inode_ag_iterator(mp, xfs_reclaim_inode, mode, |
9bf729c0 DC |
869 | XFS_ICI_RECLAIM_TAG, 1, NULL); |
870 | } | |
871 | ||
872 | /* | |
873 | * Shrinker infrastructure. | |
9bf729c0 | 874 | */ |
9bf729c0 DC |
875 | static int |
876 | xfs_reclaim_inode_shrink( | |
7f8275d0 | 877 | struct shrinker *shrink, |
9bf729c0 DC |
878 | int nr_to_scan, |
879 | gfp_t gfp_mask) | |
880 | { | |
881 | struct xfs_mount *mp; | |
882 | struct xfs_perag *pag; | |
883 | xfs_agnumber_t ag; | |
16fd5367 | 884 | int reclaimable; |
9bf729c0 | 885 | |
70e60ce7 | 886 | mp = container_of(shrink, struct xfs_mount, m_inode_shrink); |
9bf729c0 DC |
887 | if (nr_to_scan) { |
888 | if (!(gfp_mask & __GFP_FS)) | |
889 | return -1; | |
890 | ||
70e60ce7 | 891 | xfs_inode_ag_iterator(mp, xfs_reclaim_inode, 0, |
9bf729c0 | 892 | XFS_ICI_RECLAIM_TAG, 1, &nr_to_scan); |
70e60ce7 DC |
893 | /* if we don't exhaust the scan, don't bother coming back */ |
894 | if (nr_to_scan > 0) | |
895 | return -1; | |
896 | } | |
9bf729c0 | 897 | |
16fd5367 DC |
898 | reclaimable = 0; |
899 | ag = 0; | |
900 | while ((pag = xfs_inode_ag_iter_next_pag(mp, &ag, | |
901 | XFS_ICI_RECLAIM_TAG))) { | |
70e60ce7 DC |
902 | reclaimable += pag->pag_ici_reclaimable; |
903 | xfs_perag_put(pag); | |
9bf729c0 | 904 | } |
9bf729c0 DC |
905 | return reclaimable; |
906 | } | |
907 | ||
9bf729c0 DC |
908 | void |
909 | xfs_inode_shrinker_register( | |
910 | struct xfs_mount *mp) | |
911 | { | |
70e60ce7 DC |
912 | mp->m_inode_shrink.shrink = xfs_reclaim_inode_shrink; |
913 | mp->m_inode_shrink.seeks = DEFAULT_SEEKS; | |
914 | register_shrinker(&mp->m_inode_shrink); | |
9bf729c0 DC |
915 | } |
916 | ||
917 | void | |
918 | xfs_inode_shrinker_unregister( | |
919 | struct xfs_mount *mp) | |
920 | { | |
70e60ce7 | 921 | unregister_shrinker(&mp->m_inode_shrink); |
fce08f2f | 922 | } |