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" | |
fd074841 | 25 | #include "xfs_trans_priv.h" |
fe4fa4b8 DC |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
fe4fa4b8 DC |
28 | #include "xfs_mount.h" |
29 | #include "xfs_bmap_btree.h" | |
fe4fa4b8 DC |
30 | #include "xfs_inode.h" |
31 | #include "xfs_dinode.h" | |
32 | #include "xfs_error.h" | |
fe4fa4b8 DC |
33 | #include "xfs_filestream.h" |
34 | #include "xfs_vnodeops.h" | |
fe4fa4b8 | 35 | #include "xfs_inode_item.h" |
7d095257 | 36 | #include "xfs_quota.h" |
0b1b213f | 37 | #include "xfs_trace.h" |
1a387d3b | 38 | #include "xfs_fsops.h" |
fe4fa4b8 | 39 | |
a167b17e DC |
40 | #include <linux/kthread.h> |
41 | #include <linux/freezer.h> | |
42 | ||
c6d09b66 DC |
43 | struct workqueue_struct *xfs_syncd_wq; /* sync workqueue */ |
44 | ||
78ae5256 DC |
45 | /* |
46 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
47 | * radix tree lookups to a minimum. The batch size is a trade off between | |
48 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
49 | * be too greedy. | |
50 | */ | |
51 | #define XFS_LOOKUP_BATCH 32 | |
52 | ||
e13de955 DC |
53 | STATIC int |
54 | xfs_inode_ag_walk_grab( | |
55 | struct xfs_inode *ip) | |
56 | { | |
57 | struct inode *inode = VFS_I(ip); | |
58 | ||
1a3e8f3d DC |
59 | ASSERT(rcu_read_lock_held()); |
60 | ||
61 | /* | |
62 | * check for stale RCU freed inode | |
63 | * | |
64 | * If the inode has been reallocated, it doesn't matter if it's not in | |
65 | * the AG we are walking - we are walking for writeback, so if it | |
66 | * passes all the "valid inode" checks and is dirty, then we'll write | |
67 | * it back anyway. If it has been reallocated and still being | |
68 | * initialised, the XFS_INEW check below will catch it. | |
69 | */ | |
70 | spin_lock(&ip->i_flags_lock); | |
71 | if (!ip->i_ino) | |
72 | goto out_unlock_noent; | |
73 | ||
74 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
75 | if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
76 | goto out_unlock_noent; | |
77 | spin_unlock(&ip->i_flags_lock); | |
78 | ||
e13de955 DC |
79 | /* nothing to sync during shutdown */ |
80 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
81 | return EFSCORRUPTED; | |
82 | ||
e13de955 DC |
83 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
84 | if (!igrab(inode)) | |
85 | return ENOENT; | |
86 | ||
87 | if (is_bad_inode(inode)) { | |
88 | IRELE(ip); | |
89 | return ENOENT; | |
90 | } | |
91 | ||
92 | /* inode is valid */ | |
93 | return 0; | |
1a3e8f3d DC |
94 | |
95 | out_unlock_noent: | |
96 | spin_unlock(&ip->i_flags_lock); | |
97 | return ENOENT; | |
e13de955 DC |
98 | } |
99 | ||
75f3cb13 DC |
100 | STATIC int |
101 | xfs_inode_ag_walk( | |
102 | struct xfs_mount *mp, | |
5017e97d | 103 | struct xfs_perag *pag, |
75f3cb13 DC |
104 | int (*execute)(struct xfs_inode *ip, |
105 | struct xfs_perag *pag, int flags), | |
65d0f205 | 106 | int flags) |
75f3cb13 | 107 | { |
75f3cb13 DC |
108 | uint32_t first_index; |
109 | int last_error = 0; | |
110 | int skipped; | |
65d0f205 | 111 | int done; |
78ae5256 | 112 | int nr_found; |
75f3cb13 DC |
113 | |
114 | restart: | |
65d0f205 | 115 | done = 0; |
75f3cb13 DC |
116 | skipped = 0; |
117 | first_index = 0; | |
78ae5256 | 118 | nr_found = 0; |
75f3cb13 | 119 | do { |
78ae5256 | 120 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
75f3cb13 | 121 | int error = 0; |
78ae5256 | 122 | int i; |
75f3cb13 | 123 | |
1a3e8f3d | 124 | rcu_read_lock(); |
65d0f205 | 125 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, |
78ae5256 DC |
126 | (void **)batch, first_index, |
127 | XFS_LOOKUP_BATCH); | |
65d0f205 | 128 | if (!nr_found) { |
1a3e8f3d | 129 | rcu_read_unlock(); |
75f3cb13 | 130 | break; |
c8e20be0 | 131 | } |
75f3cb13 | 132 | |
65d0f205 | 133 | /* |
78ae5256 DC |
134 | * Grab the inodes before we drop the lock. if we found |
135 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 136 | */ |
78ae5256 DC |
137 | for (i = 0; i < nr_found; i++) { |
138 | struct xfs_inode *ip = batch[i]; | |
139 | ||
140 | if (done || xfs_inode_ag_walk_grab(ip)) | |
141 | batch[i] = NULL; | |
142 | ||
143 | /* | |
1a3e8f3d DC |
144 | * Update the index for the next lookup. Catch |
145 | * overflows into the next AG range which can occur if | |
146 | * we have inodes in the last block of the AG and we | |
147 | * are currently pointing to the last inode. | |
148 | * | |
149 | * Because we may see inodes that are from the wrong AG | |
150 | * due to RCU freeing and reallocation, only update the | |
151 | * index if it lies in this AG. It was a race that lead | |
152 | * us to see this inode, so another lookup from the | |
153 | * same index will not find it again. | |
78ae5256 | 154 | */ |
1a3e8f3d DC |
155 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
156 | continue; | |
78ae5256 DC |
157 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
158 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
159 | done = 1; | |
e13de955 | 160 | } |
78ae5256 DC |
161 | |
162 | /* unlock now we've grabbed the inodes. */ | |
1a3e8f3d | 163 | rcu_read_unlock(); |
e13de955 | 164 | |
78ae5256 DC |
165 | for (i = 0; i < nr_found; i++) { |
166 | if (!batch[i]) | |
167 | continue; | |
168 | error = execute(batch[i], pag, flags); | |
169 | IRELE(batch[i]); | |
170 | if (error == EAGAIN) { | |
171 | skipped++; | |
172 | continue; | |
173 | } | |
174 | if (error && last_error != EFSCORRUPTED) | |
175 | last_error = error; | |
75f3cb13 | 176 | } |
c8e20be0 DC |
177 | |
178 | /* bail out if the filesystem is corrupted. */ | |
75f3cb13 DC |
179 | if (error == EFSCORRUPTED) |
180 | break; | |
181 | ||
78ae5256 | 182 | } while (nr_found && !done); |
75f3cb13 DC |
183 | |
184 | if (skipped) { | |
185 | delay(1); | |
186 | goto restart; | |
187 | } | |
75f3cb13 DC |
188 | return last_error; |
189 | } | |
190 | ||
fe588ed3 | 191 | int |
75f3cb13 DC |
192 | xfs_inode_ag_iterator( |
193 | struct xfs_mount *mp, | |
194 | int (*execute)(struct xfs_inode *ip, | |
195 | struct xfs_perag *pag, int flags), | |
65d0f205 | 196 | int flags) |
75f3cb13 | 197 | { |
16fd5367 | 198 | struct xfs_perag *pag; |
75f3cb13 DC |
199 | int error = 0; |
200 | int last_error = 0; | |
201 | xfs_agnumber_t ag; | |
202 | ||
16fd5367 | 203 | ag = 0; |
65d0f205 DC |
204 | while ((pag = xfs_perag_get(mp, ag))) { |
205 | ag = pag->pag_agno + 1; | |
206 | error = xfs_inode_ag_walk(mp, pag, execute, flags); | |
5017e97d | 207 | xfs_perag_put(pag); |
75f3cb13 DC |
208 | if (error) { |
209 | last_error = error; | |
210 | if (error == EFSCORRUPTED) | |
211 | break; | |
212 | } | |
213 | } | |
214 | return XFS_ERROR(last_error); | |
215 | } | |
216 | ||
5a34d5cd DC |
217 | STATIC int |
218 | xfs_sync_inode_data( | |
219 | struct xfs_inode *ip, | |
75f3cb13 | 220 | struct xfs_perag *pag, |
5a34d5cd DC |
221 | int flags) |
222 | { | |
223 | struct inode *inode = VFS_I(ip); | |
224 | struct address_space *mapping = inode->i_mapping; | |
225 | int error = 0; | |
226 | ||
227 | 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 DC |
242 | xfs_ioend_wait(ip); |
243 | return error; | |
244 | } | |
245 | ||
845b6d0c CH |
246 | STATIC int |
247 | xfs_sync_inode_attr( | |
248 | struct xfs_inode *ip, | |
75f3cb13 | 249 | struct xfs_perag *pag, |
845b6d0c CH |
250 | int flags) |
251 | { | |
252 | int error = 0; | |
253 | ||
254 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
255 | if (xfs_inode_clean(ip)) | |
256 | goto out_unlock; | |
257 | if (!xfs_iflock_nowait(ip)) { | |
258 | if (!(flags & SYNC_WAIT)) | |
259 | goto out_unlock; | |
260 | xfs_iflock(ip); | |
261 | } | |
262 | ||
263 | if (xfs_inode_clean(ip)) { | |
264 | xfs_ifunlock(ip); | |
265 | goto out_unlock; | |
266 | } | |
267 | ||
c854363e | 268 | error = xfs_iflush(ip, flags); |
845b6d0c | 269 | |
ee58abdf DC |
270 | /* |
271 | * We don't want to try again on non-blocking flushes that can't run | |
272 | * again immediately. If an inode really must be written, then that's | |
273 | * what the SYNC_WAIT flag is for. | |
274 | */ | |
275 | if (error == EAGAIN) { | |
276 | ASSERT(!(flags & SYNC_WAIT)); | |
277 | error = 0; | |
278 | } | |
279 | ||
845b6d0c CH |
280 | out_unlock: |
281 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
282 | return error; | |
283 | } | |
284 | ||
075fe102 CH |
285 | /* |
286 | * Write out pagecache data for the whole filesystem. | |
287 | */ | |
64c86149 | 288 | STATIC int |
075fe102 CH |
289 | xfs_sync_data( |
290 | struct xfs_mount *mp, | |
291 | int flags) | |
683a8970 | 292 | { |
075fe102 | 293 | int error; |
fe4fa4b8 | 294 | |
b0710ccc | 295 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
fe4fa4b8 | 296 | |
65d0f205 | 297 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags); |
075fe102 CH |
298 | if (error) |
299 | return XFS_ERROR(error); | |
e9f1c6ee | 300 | |
a14a348b | 301 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); |
075fe102 CH |
302 | return 0; |
303 | } | |
e9f1c6ee | 304 | |
075fe102 CH |
305 | /* |
306 | * Write out inode metadata (attributes) for the whole filesystem. | |
307 | */ | |
64c86149 | 308 | STATIC int |
075fe102 CH |
309 | xfs_sync_attr( |
310 | struct xfs_mount *mp, | |
311 | int flags) | |
312 | { | |
313 | ASSERT((flags & ~SYNC_WAIT) == 0); | |
75f3cb13 | 314 | |
65d0f205 | 315 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags); |
fe4fa4b8 DC |
316 | } |
317 | ||
5d77c0dc | 318 | STATIC int |
2af75df7 | 319 | xfs_sync_fsdata( |
df308bcf | 320 | struct xfs_mount *mp) |
2af75df7 CH |
321 | { |
322 | struct xfs_buf *bp; | |
2af75df7 CH |
323 | |
324 | /* | |
df308bcf CH |
325 | * If the buffer is pinned then push on the log so we won't get stuck |
326 | * waiting in the write for someone, maybe ourselves, to flush the log. | |
327 | * | |
328 | * Even though we just pushed the log above, we did not have the | |
329 | * superblock buffer locked at that point so it can become pinned in | |
330 | * between there and here. | |
2af75df7 | 331 | */ |
df308bcf CH |
332 | bp = xfs_getsb(mp, 0); |
333 | if (XFS_BUF_ISPINNED(bp)) | |
334 | xfs_log_force(mp, 0); | |
2af75df7 | 335 | |
df308bcf | 336 | return xfs_bwrite(mp, bp); |
e9f1c6ee DC |
337 | } |
338 | ||
339 | /* | |
a4e4c4f4 DC |
340 | * When remounting a filesystem read-only or freezing the filesystem, we have |
341 | * two phases to execute. This first phase is syncing the data before we | |
342 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
343 | * we've waited for all the transactions created by the first phase to | |
344 | * complete. The second phase ensures that the inodes are written to their | |
345 | * location on disk rather than just existing in transactions in the log. This | |
346 | * means after a quiesce there is no log replay required to write the inodes to | |
347 | * disk (this is the main difference between a sync and a quiesce). | |
348 | */ | |
349 | /* | |
350 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
351 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
352 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
353 | * transactions can still occur here so don't bother flushing the buftarg |
354 | * because it'll just get dirty again. | |
e9f1c6ee DC |
355 | */ |
356 | int | |
357 | xfs_quiesce_data( | |
358 | struct xfs_mount *mp) | |
359 | { | |
df308bcf | 360 | int error, error2 = 0; |
e9f1c6ee | 361 | |
8b5403a6 | 362 | xfs_qm_sync(mp, SYNC_TRYLOCK); |
7d095257 | 363 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 364 | |
33b8f7c2 CH |
365 | /* force out the newly dirtied log buffers */ |
366 | xfs_log_force(mp, XFS_LOG_SYNC); | |
367 | ||
a4e4c4f4 | 368 | /* write superblock and hoover up shutdown errors */ |
df308bcf CH |
369 | error = xfs_sync_fsdata(mp); |
370 | ||
371 | /* make sure all delwri buffers are written out */ | |
372 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
373 | ||
374 | /* mark the log as covered if needed */ | |
375 | if (xfs_log_need_covered(mp)) | |
c58efdb4 | 376 | error2 = xfs_fs_log_dummy(mp); |
e9f1c6ee | 377 | |
a4e4c4f4 | 378 | /* flush data-only devices */ |
e9f1c6ee DC |
379 | if (mp->m_rtdev_targp) |
380 | XFS_bflush(mp->m_rtdev_targp); | |
381 | ||
df308bcf | 382 | return error ? error : error2; |
2af75df7 CH |
383 | } |
384 | ||
76bf105c DC |
385 | STATIC void |
386 | xfs_quiesce_fs( | |
387 | struct xfs_mount *mp) | |
388 | { | |
389 | int count = 0, pincount; | |
390 | ||
c854363e | 391 | xfs_reclaim_inodes(mp, 0); |
76bf105c | 392 | xfs_flush_buftarg(mp->m_ddev_targp, 0); |
76bf105c DC |
393 | |
394 | /* | |
395 | * This loop must run at least twice. The first instance of the loop | |
396 | * will flush most meta data but that will generate more meta data | |
397 | * (typically directory updates). Which then must be flushed and | |
c854363e DC |
398 | * logged before we can write the unmount record. We also so sync |
399 | * reclaim of inodes to catch any that the above delwri flush skipped. | |
76bf105c DC |
400 | */ |
401 | do { | |
c854363e | 402 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
075fe102 | 403 | xfs_sync_attr(mp, SYNC_WAIT); |
76bf105c DC |
404 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
405 | if (!pincount) { | |
406 | delay(50); | |
407 | count++; | |
408 | } | |
409 | } while (count < 2); | |
410 | } | |
411 | ||
412 | /* | |
413 | * Second stage of a quiesce. The data is already synced, now we have to take | |
414 | * care of the metadata. New transactions are already blocked, so we need to | |
25985edc | 415 | * wait for any remaining transactions to drain out before proceeding. |
76bf105c DC |
416 | */ |
417 | void | |
418 | xfs_quiesce_attr( | |
419 | struct xfs_mount *mp) | |
420 | { | |
421 | int error = 0; | |
422 | ||
423 | /* wait for all modifications to complete */ | |
424 | while (atomic_read(&mp->m_active_trans) > 0) | |
425 | delay(100); | |
426 | ||
427 | /* flush inodes and push all remaining buffers out to disk */ | |
428 | xfs_quiesce_fs(mp); | |
429 | ||
5e106572 FB |
430 | /* |
431 | * Just warn here till VFS can correctly support | |
432 | * read-only remount without racing. | |
433 | */ | |
434 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
435 | |
436 | /* Push the superblock and write an unmount record */ | |
adab0f67 | 437 | error = xfs_log_sbcount(mp); |
76bf105c | 438 | if (error) |
4f10700a | 439 | xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. " |
76bf105c DC |
440 | "Frozen image may not be consistent."); |
441 | xfs_log_unmount_write(mp); | |
442 | xfs_unmountfs_writesb(mp); | |
443 | } | |
444 | ||
c6d09b66 DC |
445 | static void |
446 | xfs_syncd_queue_sync( | |
447 | struct xfs_mount *mp) | |
a167b17e | 448 | { |
c6d09b66 DC |
449 | queue_delayed_work(xfs_syncd_wq, &mp->m_sync_work, |
450 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); | |
a167b17e DC |
451 | } |
452 | ||
aacaa880 | 453 | /* |
df308bcf CH |
454 | * Every sync period we need to unpin all items, reclaim inodes and sync |
455 | * disk quotas. We might need to cover the log to indicate that the | |
1a387d3b | 456 | * filesystem is idle and not frozen. |
aacaa880 | 457 | */ |
a167b17e DC |
458 | STATIC void |
459 | xfs_sync_worker( | |
c6d09b66 | 460 | struct work_struct *work) |
a167b17e | 461 | { |
c6d09b66 DC |
462 | struct xfs_mount *mp = container_of(to_delayed_work(work), |
463 | struct xfs_mount, m_sync_work); | |
a167b17e DC |
464 | int error; |
465 | ||
aacaa880 | 466 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
aacaa880 | 467 | /* dgc: errors ignored here */ |
1a387d3b DC |
468 | if (mp->m_super->s_frozen == SB_UNFROZEN && |
469 | xfs_log_need_covered(mp)) | |
c58efdb4 DC |
470 | error = xfs_fs_log_dummy(mp); |
471 | else | |
472 | xfs_log_force(mp, 0); | |
c58efdb4 | 473 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); |
fd074841 DC |
474 | |
475 | /* start pushing all the metadata that is currently dirty */ | |
476 | xfs_ail_push_all(mp->m_ail); | |
aacaa880 | 477 | } |
c6d09b66 DC |
478 | |
479 | /* queue us up again */ | |
480 | xfs_syncd_queue_sync(mp); | |
a167b17e DC |
481 | } |
482 | ||
a7b339f1 DC |
483 | /* |
484 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
485 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
486 | * on the xfs syncd work default of 30s. Perhaps this should have it's own | |
487 | * tunable, but that can be done if this method proves to be ineffective or too | |
488 | * aggressive. | |
489 | */ | |
490 | static void | |
491 | xfs_syncd_queue_reclaim( | |
492 | struct xfs_mount *mp) | |
a167b17e | 493 | { |
a167b17e | 494 | |
a7b339f1 DC |
495 | /* |
496 | * We can have inodes enter reclaim after we've shut down the syncd | |
497 | * workqueue during unmount, so don't allow reclaim work to be queued | |
498 | * during unmount. | |
499 | */ | |
500 | if (!(mp->m_super->s_flags & MS_ACTIVE)) | |
501 | return; | |
a167b17e | 502 | |
a7b339f1 DC |
503 | rcu_read_lock(); |
504 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
505 | queue_delayed_work(xfs_syncd_wq, &mp->m_reclaim_work, | |
506 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
a167b17e | 507 | } |
a7b339f1 DC |
508 | rcu_read_unlock(); |
509 | } | |
a167b17e | 510 | |
a7b339f1 DC |
511 | /* |
512 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
513 | * many inodes as possible in a short period of time. It kicks itself every few | |
514 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
515 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
516 | * already being flushed, and once done schedules a future pass. | |
517 | */ | |
518 | STATIC void | |
519 | xfs_reclaim_worker( | |
520 | struct work_struct *work) | |
521 | { | |
522 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
523 | struct xfs_mount, m_reclaim_work); | |
524 | ||
525 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
526 | xfs_syncd_queue_reclaim(mp); | |
527 | } | |
528 | ||
89e4cb55 DC |
529 | /* |
530 | * Flush delayed allocate data, attempting to free up reserved space | |
531 | * from existing allocations. At this point a new allocation attempt | |
532 | * has failed with ENOSPC and we are in the process of scratching our | |
533 | * heads, looking about for more room. | |
534 | * | |
535 | * Queue a new data flush if there isn't one already in progress and | |
536 | * wait for completion of the flush. This means that we only ever have one | |
537 | * inode flush in progress no matter how many ENOSPC events are occurring and | |
538 | * so will prevent the system from bogging down due to every concurrent | |
539 | * ENOSPC event scanning all the active inodes in the system for writeback. | |
540 | */ | |
541 | void | |
542 | xfs_flush_inodes( | |
543 | struct xfs_inode *ip) | |
544 | { | |
545 | struct xfs_mount *mp = ip->i_mount; | |
546 | ||
547 | queue_work(xfs_syncd_wq, &mp->m_flush_work); | |
548 | flush_work_sync(&mp->m_flush_work); | |
549 | } | |
550 | ||
551 | STATIC void | |
552 | xfs_flush_worker( | |
553 | struct work_struct *work) | |
554 | { | |
555 | struct xfs_mount *mp = container_of(work, | |
556 | struct xfs_mount, m_flush_work); | |
557 | ||
558 | xfs_sync_data(mp, SYNC_TRYLOCK); | |
559 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); | |
a167b17e DC |
560 | } |
561 | ||
562 | int | |
563 | xfs_syncd_init( | |
564 | struct xfs_mount *mp) | |
565 | { | |
89e4cb55 | 566 | INIT_WORK(&mp->m_flush_work, xfs_flush_worker); |
c6d09b66 | 567 | INIT_DELAYED_WORK(&mp->m_sync_work, xfs_sync_worker); |
a7b339f1 DC |
568 | INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker); |
569 | ||
c6d09b66 | 570 | xfs_syncd_queue_sync(mp); |
a7b339f1 | 571 | xfs_syncd_queue_reclaim(mp); |
c6d09b66 | 572 | |
a167b17e DC |
573 | return 0; |
574 | } | |
575 | ||
576 | void | |
577 | xfs_syncd_stop( | |
578 | struct xfs_mount *mp) | |
579 | { | |
c6d09b66 | 580 | cancel_delayed_work_sync(&mp->m_sync_work); |
a7b339f1 | 581 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
89e4cb55 | 582 | cancel_work_sync(&mp->m_flush_work); |
a167b17e DC |
583 | } |
584 | ||
bc990f5c CH |
585 | void |
586 | __xfs_inode_set_reclaim_tag( | |
587 | struct xfs_perag *pag, | |
588 | struct xfs_inode *ip) | |
589 | { | |
590 | radix_tree_tag_set(&pag->pag_ici_root, | |
591 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
592 | XFS_ICI_RECLAIM_TAG); | |
16fd5367 DC |
593 | |
594 | if (!pag->pag_ici_reclaimable) { | |
595 | /* propagate the reclaim tag up into the perag radix tree */ | |
596 | spin_lock(&ip->i_mount->m_perag_lock); | |
597 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
598 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
599 | XFS_ICI_RECLAIM_TAG); | |
600 | spin_unlock(&ip->i_mount->m_perag_lock); | |
a7b339f1 DC |
601 | |
602 | /* schedule periodic background inode reclaim */ | |
603 | xfs_syncd_queue_reclaim(ip->i_mount); | |
604 | ||
16fd5367 DC |
605 | trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno, |
606 | -1, _RET_IP_); | |
607 | } | |
9bf729c0 | 608 | pag->pag_ici_reclaimable++; |
bc990f5c CH |
609 | } |
610 | ||
11654513 DC |
611 | /* |
612 | * We set the inode flag atomically with the radix tree tag. | |
613 | * Once we get tag lookups on the radix tree, this inode flag | |
614 | * can go away. | |
615 | */ | |
396beb85 DC |
616 | void |
617 | xfs_inode_set_reclaim_tag( | |
618 | xfs_inode_t *ip) | |
619 | { | |
5017e97d DC |
620 | struct xfs_mount *mp = ip->i_mount; |
621 | struct xfs_perag *pag; | |
396beb85 | 622 | |
5017e97d | 623 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1a427ab0 | 624 | spin_lock(&pag->pag_ici_lock); |
396beb85 | 625 | spin_lock(&ip->i_flags_lock); |
bc990f5c | 626 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 627 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 | 628 | spin_unlock(&ip->i_flags_lock); |
1a427ab0 | 629 | spin_unlock(&pag->pag_ici_lock); |
5017e97d | 630 | xfs_perag_put(pag); |
396beb85 DC |
631 | } |
632 | ||
081003ff JW |
633 | STATIC void |
634 | __xfs_inode_clear_reclaim( | |
396beb85 DC |
635 | xfs_perag_t *pag, |
636 | xfs_inode_t *ip) | |
637 | { | |
9bf729c0 | 638 | pag->pag_ici_reclaimable--; |
16fd5367 DC |
639 | if (!pag->pag_ici_reclaimable) { |
640 | /* clear the reclaim tag from the perag radix tree */ | |
641 | spin_lock(&ip->i_mount->m_perag_lock); | |
642 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
643 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
644 | XFS_ICI_RECLAIM_TAG); | |
645 | spin_unlock(&ip->i_mount->m_perag_lock); | |
646 | trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno, | |
647 | -1, _RET_IP_); | |
648 | } | |
396beb85 DC |
649 | } |
650 | ||
081003ff JW |
651 | void |
652 | __xfs_inode_clear_reclaim_tag( | |
653 | xfs_mount_t *mp, | |
654 | xfs_perag_t *pag, | |
655 | xfs_inode_t *ip) | |
656 | { | |
657 | radix_tree_tag_clear(&pag->pag_ici_root, | |
658 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
659 | __xfs_inode_clear_reclaim(pag, ip); | |
660 | } | |
661 | ||
e3a20c0b DC |
662 | /* |
663 | * Grab the inode for reclaim exclusively. | |
664 | * Return 0 if we grabbed it, non-zero otherwise. | |
665 | */ | |
666 | STATIC int | |
667 | xfs_reclaim_inode_grab( | |
668 | struct xfs_inode *ip, | |
669 | int flags) | |
670 | { | |
1a3e8f3d DC |
671 | ASSERT(rcu_read_lock_held()); |
672 | ||
673 | /* quick check for stale RCU freed inode */ | |
674 | if (!ip->i_ino) | |
675 | return 1; | |
e3a20c0b DC |
676 | |
677 | /* | |
1a3e8f3d | 678 | * do some unlocked checks first to avoid unnecessary lock traffic. |
e3a20c0b DC |
679 | * The first is a flush lock check, the second is a already in reclaim |
680 | * check. Only do these checks if we are not going to block on locks. | |
681 | */ | |
682 | if ((flags & SYNC_TRYLOCK) && | |
683 | (!ip->i_flush.done || __xfs_iflags_test(ip, XFS_IRECLAIM))) { | |
684 | return 1; | |
685 | } | |
686 | ||
687 | /* | |
688 | * The radix tree lock here protects a thread in xfs_iget from racing | |
689 | * with us starting reclaim on the inode. Once we have the | |
690 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
691 | * |
692 | * Due to RCU lookup, we may find inodes that have been freed and only | |
693 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
694 | * aren't candidates for reclaim at all, so we must check the | |
695 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
696 | */ |
697 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
698 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
699 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
700 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
701 | spin_unlock(&ip->i_flags_lock); |
702 | return 1; | |
703 | } | |
704 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
705 | spin_unlock(&ip->i_flags_lock); | |
706 | return 0; | |
707 | } | |
708 | ||
777df5af DC |
709 | /* |
710 | * Inodes in different states need to be treated differently, and the return | |
711 | * value of xfs_iflush is not sufficient to get this right. The following table | |
712 | * lists the inode states and the reclaim actions necessary for non-blocking | |
713 | * reclaim: | |
714 | * | |
715 | * | |
716 | * inode state iflush ret required action | |
717 | * --------------- ---------- --------------- | |
718 | * bad - reclaim | |
719 | * shutdown EIO unpin and reclaim | |
720 | * clean, unpinned 0 reclaim | |
721 | * stale, unpinned 0 reclaim | |
c854363e DC |
722 | * clean, pinned(*) 0 requeue |
723 | * stale, pinned EAGAIN requeue | |
724 | * dirty, delwri ok 0 requeue | |
725 | * dirty, delwri blocked EAGAIN requeue | |
726 | * dirty, sync flush 0 reclaim | |
777df5af DC |
727 | * |
728 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
729 | * handled anyway given the order of checks implemented. | |
730 | * | |
c854363e DC |
731 | * As can be seen from the table, the return value of xfs_iflush() is not |
732 | * sufficient to correctly decide the reclaim action here. The checks in | |
733 | * xfs_iflush() might look like duplicates, but they are not. | |
734 | * | |
735 | * Also, because we get the flush lock first, we know that any inode that has | |
736 | * been flushed delwri has had the flush completed by the time we check that | |
737 | * the inode is clean. The clean inode check needs to be done before flushing | |
738 | * the inode delwri otherwise we would loop forever requeuing clean inodes as | |
739 | * we cannot tell apart a successful delwri flush and a clean inode from the | |
740 | * return value of xfs_iflush(). | |
741 | * | |
742 | * Note that because the inode is flushed delayed write by background | |
743 | * writeback, the flush lock may already be held here and waiting on it can | |
744 | * result in very long latencies. Hence for sync reclaims, where we wait on the | |
745 | * flush lock, the caller should push out delayed write inodes first before | |
746 | * trying to reclaim them to minimise the amount of time spent waiting. For | |
747 | * background relaim, we just requeue the inode for the next pass. | |
748 | * | |
777df5af DC |
749 | * Hence the order of actions after gaining the locks should be: |
750 | * bad => reclaim | |
751 | * shutdown => unpin and reclaim | |
c854363e DC |
752 | * pinned, delwri => requeue |
753 | * pinned, sync => unpin | |
777df5af DC |
754 | * stale => reclaim |
755 | * clean => reclaim | |
c854363e DC |
756 | * dirty, delwri => flush and requeue |
757 | * dirty, sync => flush, wait and reclaim | |
777df5af | 758 | */ |
75f3cb13 | 759 | STATIC int |
c8e20be0 | 760 | xfs_reclaim_inode( |
75f3cb13 DC |
761 | struct xfs_inode *ip, |
762 | struct xfs_perag *pag, | |
c8e20be0 | 763 | int sync_mode) |
fce08f2f | 764 | { |
1bfd8d04 | 765 | int error; |
777df5af | 766 | |
1bfd8d04 DC |
767 | restart: |
768 | error = 0; | |
c8e20be0 | 769 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
770 | if (!xfs_iflock_nowait(ip)) { |
771 | if (!(sync_mode & SYNC_WAIT)) | |
772 | goto out; | |
773 | xfs_iflock(ip); | |
774 | } | |
7a3be02b | 775 | |
777df5af DC |
776 | if (is_bad_inode(VFS_I(ip))) |
777 | goto reclaim; | |
778 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
779 | xfs_iunpin_wait(ip); | |
780 | goto reclaim; | |
781 | } | |
c854363e DC |
782 | if (xfs_ipincount(ip)) { |
783 | if (!(sync_mode & SYNC_WAIT)) { | |
784 | xfs_ifunlock(ip); | |
785 | goto out; | |
786 | } | |
777df5af | 787 | xfs_iunpin_wait(ip); |
c854363e | 788 | } |
777df5af DC |
789 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
790 | goto reclaim; | |
791 | if (xfs_inode_clean(ip)) | |
792 | goto reclaim; | |
793 | ||
1bfd8d04 DC |
794 | /* |
795 | * Now we have an inode that needs flushing. | |
796 | * | |
797 | * We do a nonblocking flush here even if we are doing a SYNC_WAIT | |
798 | * reclaim as we can deadlock with inode cluster removal. | |
799 | * xfs_ifree_cluster() can lock the inode buffer before it locks the | |
800 | * ip->i_lock, and we are doing the exact opposite here. As a result, | |
801 | * doing a blocking xfs_itobp() to get the cluster buffer will result | |
802 | * in an ABBA deadlock with xfs_ifree_cluster(). | |
803 | * | |
804 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
805 | * cache to mark them stale, if we hit this case we don't actually want | |
806 | * to do IO here - we want the inode marked stale so we can simply | |
807 | * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush, | |
808 | * just unlock the inode, back off and try again. Hopefully the next | |
809 | * pass through will see the stale flag set on the inode. | |
810 | */ | |
811 | error = xfs_iflush(ip, SYNC_TRYLOCK | sync_mode); | |
c854363e | 812 | if (sync_mode & SYNC_WAIT) { |
1bfd8d04 DC |
813 | if (error == EAGAIN) { |
814 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
815 | /* backoff longer than in xfs_ifree_cluster */ | |
816 | delay(2); | |
817 | goto restart; | |
818 | } | |
c854363e DC |
819 | xfs_iflock(ip); |
820 | goto reclaim; | |
c8e20be0 DC |
821 | } |
822 | ||
c854363e DC |
823 | /* |
824 | * When we have to flush an inode but don't have SYNC_WAIT set, we | |
825 | * flush the inode out using a delwri buffer and wait for the next | |
826 | * call into reclaim to find it in a clean state instead of waiting for | |
827 | * it now. We also don't return errors here - if the error is transient | |
828 | * then the next reclaim pass will flush the inode, and if the error | |
f1d486a3 | 829 | * is permanent then the next sync reclaim will reclaim the inode and |
c854363e DC |
830 | * pass on the error. |
831 | */ | |
f1d486a3 | 832 | if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 833 | xfs_warn(ip->i_mount, |
c854363e DC |
834 | "inode 0x%llx background reclaim flush failed with %d", |
835 | (long long)ip->i_ino, error); | |
836 | } | |
837 | out: | |
838 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
839 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
840 | /* | |
841 | * We could return EAGAIN here to make reclaim rescan the inode tree in | |
842 | * a short while. However, this just burns CPU time scanning the tree | |
843 | * waiting for IO to complete and xfssyncd never goes back to the idle | |
844 | * state. Instead, return 0 to let the next scheduled background reclaim | |
845 | * attempt to reclaim the inode again. | |
846 | */ | |
847 | return 0; | |
848 | ||
777df5af DC |
849 | reclaim: |
850 | xfs_ifunlock(ip); | |
c8e20be0 | 851 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab DC |
852 | |
853 | XFS_STATS_INC(xs_ig_reclaims); | |
854 | /* | |
855 | * Remove the inode from the per-AG radix tree. | |
856 | * | |
857 | * Because radix_tree_delete won't complain even if the item was never | |
858 | * added to the tree assert that it's been there before to catch | |
859 | * problems with the inode life time early on. | |
860 | */ | |
1a427ab0 | 861 | spin_lock(&pag->pag_ici_lock); |
2f11feab DC |
862 | if (!radix_tree_delete(&pag->pag_ici_root, |
863 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino))) | |
864 | ASSERT(0); | |
081003ff | 865 | __xfs_inode_clear_reclaim(pag, ip); |
1a427ab0 | 866 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
867 | |
868 | /* | |
869 | * Here we do an (almost) spurious inode lock in order to coordinate | |
870 | * with inode cache radix tree lookups. This is because the lookup | |
871 | * can reference the inodes in the cache without taking references. | |
872 | * | |
873 | * We make that OK here by ensuring that we wait until the inode is | |
874 | * unlocked after the lookup before we go ahead and free it. We get | |
875 | * both the ilock and the iolock because the code may need to drop the | |
876 | * ilock one but will still hold the iolock. | |
877 | */ | |
878 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
879 | xfs_qm_dqdetach(ip); | |
880 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
881 | ||
882 | xfs_inode_free(ip); | |
c854363e DC |
883 | return error; |
884 | ||
7a3be02b DC |
885 | } |
886 | ||
65d0f205 DC |
887 | /* |
888 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
889 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
890 | * then a shut down during filesystem unmount reclaim walk leak all the | |
891 | * unreclaimed inodes. | |
892 | */ | |
893 | int | |
894 | xfs_reclaim_inodes_ag( | |
895 | struct xfs_mount *mp, | |
896 | int flags, | |
897 | int *nr_to_scan) | |
898 | { | |
899 | struct xfs_perag *pag; | |
900 | int error = 0; | |
901 | int last_error = 0; | |
902 | xfs_agnumber_t ag; | |
69b491c2 DC |
903 | int trylock = flags & SYNC_TRYLOCK; |
904 | int skipped; | |
65d0f205 | 905 | |
69b491c2 | 906 | restart: |
65d0f205 | 907 | ag = 0; |
69b491c2 | 908 | skipped = 0; |
65d0f205 DC |
909 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
910 | unsigned long first_index = 0; | |
911 | int done = 0; | |
e3a20c0b | 912 | int nr_found = 0; |
65d0f205 DC |
913 | |
914 | ag = pag->pag_agno + 1; | |
915 | ||
69b491c2 DC |
916 | if (trylock) { |
917 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
918 | skipped++; | |
f83282a8 | 919 | xfs_perag_put(pag); |
69b491c2 DC |
920 | continue; |
921 | } | |
922 | first_index = pag->pag_ici_reclaim_cursor; | |
923 | } else | |
924 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
925 | ||
65d0f205 | 926 | do { |
e3a20c0b DC |
927 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
928 | int i; | |
65d0f205 | 929 | |
1a3e8f3d | 930 | rcu_read_lock(); |
e3a20c0b DC |
931 | nr_found = radix_tree_gang_lookup_tag( |
932 | &pag->pag_ici_root, | |
933 | (void **)batch, first_index, | |
934 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
935 | XFS_ICI_RECLAIM_TAG); |
936 | if (!nr_found) { | |
b2232219 | 937 | done = 1; |
1a3e8f3d | 938 | rcu_read_unlock(); |
65d0f205 DC |
939 | break; |
940 | } | |
941 | ||
942 | /* | |
e3a20c0b DC |
943 | * Grab the inodes before we drop the lock. if we found |
944 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 945 | */ |
e3a20c0b DC |
946 | for (i = 0; i < nr_found; i++) { |
947 | struct xfs_inode *ip = batch[i]; | |
948 | ||
949 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
950 | batch[i] = NULL; | |
951 | ||
952 | /* | |
953 | * Update the index for the next lookup. Catch | |
954 | * overflows into the next AG range which can | |
955 | * occur if we have inodes in the last block of | |
956 | * the AG and we are currently pointing to the | |
957 | * last inode. | |
1a3e8f3d DC |
958 | * |
959 | * Because we may see inodes that are from the | |
960 | * wrong AG due to RCU freeing and | |
961 | * reallocation, only update the index if it | |
962 | * lies in this AG. It was a race that lead us | |
963 | * to see this inode, so another lookup from | |
964 | * the same index will not find it again. | |
e3a20c0b | 965 | */ |
1a3e8f3d DC |
966 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
967 | pag->pag_agno) | |
968 | continue; | |
e3a20c0b DC |
969 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
970 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
971 | done = 1; | |
972 | } | |
65d0f205 | 973 | |
e3a20c0b | 974 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 975 | rcu_read_unlock(); |
e3a20c0b DC |
976 | |
977 | for (i = 0; i < nr_found; i++) { | |
978 | if (!batch[i]) | |
979 | continue; | |
980 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
981 | if (error && last_error != EFSCORRUPTED) | |
982 | last_error = error; | |
983 | } | |
984 | ||
985 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 986 | |
e3a20c0b | 987 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 988 | |
69b491c2 DC |
989 | if (trylock && !done) |
990 | pag->pag_ici_reclaim_cursor = first_index; | |
991 | else | |
992 | pag->pag_ici_reclaim_cursor = 0; | |
993 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
994 | xfs_perag_put(pag); |
995 | } | |
69b491c2 DC |
996 | |
997 | /* | |
998 | * if we skipped any AG, and we still have scan count remaining, do | |
999 | * another pass this time using blocking reclaim semantics (i.e | |
1000 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1001 | * ensure that when we get more reclaimers than AGs we block rather | |
1002 | * than spin trying to execute reclaim. | |
1003 | */ | |
1004 | if (trylock && skipped && *nr_to_scan > 0) { | |
1005 | trylock = 0; | |
1006 | goto restart; | |
1007 | } | |
65d0f205 DC |
1008 | return XFS_ERROR(last_error); |
1009 | } | |
1010 | ||
7a3be02b DC |
1011 | int |
1012 | xfs_reclaim_inodes( | |
1013 | xfs_mount_t *mp, | |
7a3be02b DC |
1014 | int mode) |
1015 | { | |
65d0f205 DC |
1016 | int nr_to_scan = INT_MAX; |
1017 | ||
1018 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1019 | } |
1020 | ||
1021 | /* | |
a7b339f1 DC |
1022 | * Inode cache shrinker. |
1023 | * | |
1024 | * When called we make sure that there is a background (fast) inode reclaim in | |
1025 | * progress, while we will throttle the speed of reclaim via doiing synchronous | |
1026 | * reclaim of inodes. That means if we come across dirty inodes, we wait for | |
1027 | * them to be cleaned, which we hope will not be very long due to the | |
1028 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1029 | */ |
9bf729c0 DC |
1030 | static int |
1031 | xfs_reclaim_inode_shrink( | |
7f8275d0 | 1032 | struct shrinker *shrink, |
1495f230 | 1033 | struct shrink_control *sc) |
9bf729c0 DC |
1034 | { |
1035 | struct xfs_mount *mp; | |
1036 | struct xfs_perag *pag; | |
1037 | xfs_agnumber_t ag; | |
16fd5367 | 1038 | int reclaimable; |
1495f230 YH |
1039 | int nr_to_scan = sc->nr_to_scan; |
1040 | gfp_t gfp_mask = sc->gfp_mask; | |
9bf729c0 | 1041 | |
70e60ce7 | 1042 | mp = container_of(shrink, struct xfs_mount, m_inode_shrink); |
9bf729c0 | 1043 | if (nr_to_scan) { |
fd074841 | 1044 | /* kick background reclaimer and push the AIL */ |
a7b339f1 | 1045 | xfs_syncd_queue_reclaim(mp); |
fd074841 | 1046 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1047 | |
9bf729c0 DC |
1048 | if (!(gfp_mask & __GFP_FS)) |
1049 | return -1; | |
1050 | ||
a7b339f1 DC |
1051 | xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, |
1052 | &nr_to_scan); | |
65d0f205 | 1053 | /* terminate if we don't exhaust the scan */ |
70e60ce7 DC |
1054 | if (nr_to_scan > 0) |
1055 | return -1; | |
1056 | } | |
9bf729c0 | 1057 | |
16fd5367 DC |
1058 | reclaimable = 0; |
1059 | ag = 0; | |
65d0f205 DC |
1060 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1061 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1062 | reclaimable += pag->pag_ici_reclaimable; |
1063 | xfs_perag_put(pag); | |
9bf729c0 | 1064 | } |
9bf729c0 DC |
1065 | return reclaimable; |
1066 | } | |
1067 | ||
9bf729c0 DC |
1068 | void |
1069 | xfs_inode_shrinker_register( | |
1070 | struct xfs_mount *mp) | |
1071 | { | |
70e60ce7 DC |
1072 | mp->m_inode_shrink.shrink = xfs_reclaim_inode_shrink; |
1073 | mp->m_inode_shrink.seeks = DEFAULT_SEEKS; | |
1074 | register_shrinker(&mp->m_inode_shrink); | |
9bf729c0 DC |
1075 | } |
1076 | ||
1077 | void | |
1078 | xfs_inode_shrinker_unregister( | |
1079 | struct xfs_mount *mp) | |
1080 | { | |
70e60ce7 | 1081 | unregister_shrinker(&mp->m_inode_shrink); |
fce08f2f | 1082 | } |