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 | ||
8daaa831 DC |
182 | cond_resched(); |
183 | ||
78ae5256 | 184 | } while (nr_found && !done); |
75f3cb13 DC |
185 | |
186 | if (skipped) { | |
187 | delay(1); | |
188 | goto restart; | |
189 | } | |
75f3cb13 DC |
190 | return last_error; |
191 | } | |
192 | ||
fe588ed3 | 193 | int |
75f3cb13 DC |
194 | xfs_inode_ag_iterator( |
195 | struct xfs_mount *mp, | |
196 | int (*execute)(struct xfs_inode *ip, | |
197 | struct xfs_perag *pag, int flags), | |
65d0f205 | 198 | int flags) |
75f3cb13 | 199 | { |
16fd5367 | 200 | struct xfs_perag *pag; |
75f3cb13 DC |
201 | int error = 0; |
202 | int last_error = 0; | |
203 | xfs_agnumber_t ag; | |
204 | ||
16fd5367 | 205 | ag = 0; |
65d0f205 DC |
206 | while ((pag = xfs_perag_get(mp, ag))) { |
207 | ag = pag->pag_agno + 1; | |
208 | error = xfs_inode_ag_walk(mp, pag, execute, flags); | |
5017e97d | 209 | xfs_perag_put(pag); |
75f3cb13 DC |
210 | if (error) { |
211 | last_error = error; | |
212 | if (error == EFSCORRUPTED) | |
213 | break; | |
214 | } | |
215 | } | |
216 | return XFS_ERROR(last_error); | |
217 | } | |
218 | ||
5a34d5cd DC |
219 | STATIC int |
220 | xfs_sync_inode_data( | |
221 | struct xfs_inode *ip, | |
75f3cb13 | 222 | struct xfs_perag *pag, |
5a34d5cd DC |
223 | int flags) |
224 | { | |
225 | struct inode *inode = VFS_I(ip); | |
226 | struct address_space *mapping = inode->i_mapping; | |
227 | int error = 0; | |
228 | ||
229 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
230 | goto out_wait; | |
231 | ||
232 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
233 | if (flags & SYNC_TRYLOCK) | |
234 | goto out_wait; | |
235 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
236 | } | |
237 | ||
238 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? | |
0cadda1c | 239 | 0 : XBF_ASYNC, FI_NONE); |
5a34d5cd DC |
240 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
241 | ||
242 | out_wait: | |
b0710ccc | 243 | if (flags & SYNC_WAIT) |
5a34d5cd DC |
244 | xfs_ioend_wait(ip); |
245 | return error; | |
246 | } | |
247 | ||
845b6d0c CH |
248 | STATIC int |
249 | xfs_sync_inode_attr( | |
250 | struct xfs_inode *ip, | |
75f3cb13 | 251 | struct xfs_perag *pag, |
845b6d0c CH |
252 | int flags) |
253 | { | |
254 | int error = 0; | |
255 | ||
256 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
257 | if (xfs_inode_clean(ip)) | |
258 | goto out_unlock; | |
259 | if (!xfs_iflock_nowait(ip)) { | |
260 | if (!(flags & SYNC_WAIT)) | |
261 | goto out_unlock; | |
262 | xfs_iflock(ip); | |
263 | } | |
264 | ||
265 | if (xfs_inode_clean(ip)) { | |
266 | xfs_ifunlock(ip); | |
267 | goto out_unlock; | |
268 | } | |
269 | ||
c854363e | 270 | error = xfs_iflush(ip, flags); |
845b6d0c | 271 | |
ee58abdf DC |
272 | /* |
273 | * We don't want to try again on non-blocking flushes that can't run | |
274 | * again immediately. If an inode really must be written, then that's | |
275 | * what the SYNC_WAIT flag is for. | |
276 | */ | |
277 | if (error == EAGAIN) { | |
278 | ASSERT(!(flags & SYNC_WAIT)); | |
279 | error = 0; | |
280 | } | |
281 | ||
845b6d0c CH |
282 | out_unlock: |
283 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
284 | return error; | |
285 | } | |
286 | ||
075fe102 CH |
287 | /* |
288 | * Write out pagecache data for the whole filesystem. | |
289 | */ | |
64c86149 | 290 | STATIC int |
075fe102 CH |
291 | xfs_sync_data( |
292 | struct xfs_mount *mp, | |
293 | int flags) | |
683a8970 | 294 | { |
075fe102 | 295 | int error; |
fe4fa4b8 | 296 | |
b0710ccc | 297 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
fe4fa4b8 | 298 | |
65d0f205 | 299 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags); |
075fe102 CH |
300 | if (error) |
301 | return XFS_ERROR(error); | |
e9f1c6ee | 302 | |
a14a348b | 303 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); |
075fe102 CH |
304 | return 0; |
305 | } | |
e9f1c6ee | 306 | |
075fe102 CH |
307 | /* |
308 | * Write out inode metadata (attributes) for the whole filesystem. | |
309 | */ | |
64c86149 | 310 | STATIC int |
075fe102 CH |
311 | xfs_sync_attr( |
312 | struct xfs_mount *mp, | |
313 | int flags) | |
314 | { | |
315 | ASSERT((flags & ~SYNC_WAIT) == 0); | |
75f3cb13 | 316 | |
65d0f205 | 317 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags); |
fe4fa4b8 DC |
318 | } |
319 | ||
5d77c0dc | 320 | STATIC int |
2af75df7 | 321 | xfs_sync_fsdata( |
df308bcf | 322 | struct xfs_mount *mp) |
2af75df7 CH |
323 | { |
324 | struct xfs_buf *bp; | |
2af75df7 CH |
325 | |
326 | /* | |
df308bcf CH |
327 | * If the buffer is pinned then push on the log so we won't get stuck |
328 | * waiting in the write for someone, maybe ourselves, to flush the log. | |
329 | * | |
330 | * Even though we just pushed the log above, we did not have the | |
331 | * superblock buffer locked at that point so it can become pinned in | |
332 | * between there and here. | |
2af75df7 | 333 | */ |
df308bcf | 334 | bp = xfs_getsb(mp, 0); |
811e64c7 | 335 | if (xfs_buf_ispinned(bp)) |
df308bcf | 336 | xfs_log_force(mp, 0); |
2af75df7 | 337 | |
df308bcf | 338 | return xfs_bwrite(mp, bp); |
e9f1c6ee DC |
339 | } |
340 | ||
341 | /* | |
a4e4c4f4 DC |
342 | * When remounting a filesystem read-only or freezing the filesystem, we have |
343 | * two phases to execute. This first phase is syncing the data before we | |
344 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
345 | * we've waited for all the transactions created by the first phase to | |
346 | * complete. The second phase ensures that the inodes are written to their | |
347 | * location on disk rather than just existing in transactions in the log. This | |
348 | * means after a quiesce there is no log replay required to write the inodes to | |
349 | * disk (this is the main difference between a sync and a quiesce). | |
350 | */ | |
351 | /* | |
352 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
353 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
354 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
355 | * transactions can still occur here so don't bother flushing the buftarg |
356 | * because it'll just get dirty again. | |
e9f1c6ee DC |
357 | */ |
358 | int | |
359 | xfs_quiesce_data( | |
360 | struct xfs_mount *mp) | |
361 | { | |
df308bcf | 362 | int error, error2 = 0; |
e9f1c6ee | 363 | |
8b5403a6 | 364 | xfs_qm_sync(mp, SYNC_TRYLOCK); |
7d095257 | 365 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 366 | |
33b8f7c2 CH |
367 | /* force out the newly dirtied log buffers */ |
368 | xfs_log_force(mp, XFS_LOG_SYNC); | |
369 | ||
a4e4c4f4 | 370 | /* write superblock and hoover up shutdown errors */ |
df308bcf CH |
371 | error = xfs_sync_fsdata(mp); |
372 | ||
373 | /* make sure all delwri buffers are written out */ | |
374 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
375 | ||
376 | /* mark the log as covered if needed */ | |
377 | if (xfs_log_need_covered(mp)) | |
c58efdb4 | 378 | error2 = xfs_fs_log_dummy(mp); |
e9f1c6ee | 379 | |
a4e4c4f4 | 380 | /* flush data-only devices */ |
e9f1c6ee DC |
381 | if (mp->m_rtdev_targp) |
382 | XFS_bflush(mp->m_rtdev_targp); | |
383 | ||
df308bcf | 384 | return error ? error : error2; |
2af75df7 CH |
385 | } |
386 | ||
76bf105c DC |
387 | STATIC void |
388 | xfs_quiesce_fs( | |
389 | struct xfs_mount *mp) | |
390 | { | |
391 | int count = 0, pincount; | |
392 | ||
c854363e | 393 | xfs_reclaim_inodes(mp, 0); |
76bf105c | 394 | xfs_flush_buftarg(mp->m_ddev_targp, 0); |
76bf105c DC |
395 | |
396 | /* | |
397 | * This loop must run at least twice. The first instance of the loop | |
398 | * will flush most meta data but that will generate more meta data | |
399 | * (typically directory updates). Which then must be flushed and | |
c854363e DC |
400 | * logged before we can write the unmount record. We also so sync |
401 | * reclaim of inodes to catch any that the above delwri flush skipped. | |
76bf105c DC |
402 | */ |
403 | do { | |
c854363e | 404 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
075fe102 | 405 | xfs_sync_attr(mp, SYNC_WAIT); |
76bf105c DC |
406 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
407 | if (!pincount) { | |
408 | delay(50); | |
409 | count++; | |
410 | } | |
411 | } while (count < 2); | |
412 | } | |
413 | ||
414 | /* | |
415 | * Second stage of a quiesce. The data is already synced, now we have to take | |
416 | * care of the metadata. New transactions are already blocked, so we need to | |
25985edc | 417 | * wait for any remaining transactions to drain out before proceeding. |
76bf105c DC |
418 | */ |
419 | void | |
420 | xfs_quiesce_attr( | |
421 | struct xfs_mount *mp) | |
422 | { | |
423 | int error = 0; | |
424 | ||
425 | /* wait for all modifications to complete */ | |
426 | while (atomic_read(&mp->m_active_trans) > 0) | |
427 | delay(100); | |
428 | ||
429 | /* flush inodes and push all remaining buffers out to disk */ | |
430 | xfs_quiesce_fs(mp); | |
431 | ||
5e106572 FB |
432 | /* |
433 | * Just warn here till VFS can correctly support | |
434 | * read-only remount without racing. | |
435 | */ | |
436 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
437 | |
438 | /* Push the superblock and write an unmount record */ | |
adab0f67 | 439 | error = xfs_log_sbcount(mp); |
76bf105c | 440 | if (error) |
4f10700a | 441 | xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. " |
76bf105c DC |
442 | "Frozen image may not be consistent."); |
443 | xfs_log_unmount_write(mp); | |
444 | xfs_unmountfs_writesb(mp); | |
445 | } | |
446 | ||
c6d09b66 DC |
447 | static void |
448 | xfs_syncd_queue_sync( | |
449 | struct xfs_mount *mp) | |
a167b17e | 450 | { |
c6d09b66 DC |
451 | queue_delayed_work(xfs_syncd_wq, &mp->m_sync_work, |
452 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); | |
a167b17e DC |
453 | } |
454 | ||
aacaa880 | 455 | /* |
df308bcf CH |
456 | * Every sync period we need to unpin all items, reclaim inodes and sync |
457 | * disk quotas. We might need to cover the log to indicate that the | |
1a387d3b | 458 | * filesystem is idle and not frozen. |
aacaa880 | 459 | */ |
a167b17e DC |
460 | STATIC void |
461 | xfs_sync_worker( | |
c6d09b66 | 462 | struct work_struct *work) |
a167b17e | 463 | { |
c6d09b66 DC |
464 | struct xfs_mount *mp = container_of(to_delayed_work(work), |
465 | struct xfs_mount, m_sync_work); | |
a167b17e DC |
466 | int error; |
467 | ||
aacaa880 | 468 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
aacaa880 | 469 | /* dgc: errors ignored here */ |
1a387d3b DC |
470 | if (mp->m_super->s_frozen == SB_UNFROZEN && |
471 | xfs_log_need_covered(mp)) | |
c58efdb4 DC |
472 | error = xfs_fs_log_dummy(mp); |
473 | else | |
474 | xfs_log_force(mp, 0); | |
c58efdb4 | 475 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); |
fd074841 DC |
476 | |
477 | /* start pushing all the metadata that is currently dirty */ | |
478 | xfs_ail_push_all(mp->m_ail); | |
aacaa880 | 479 | } |
c6d09b66 DC |
480 | |
481 | /* queue us up again */ | |
482 | xfs_syncd_queue_sync(mp); | |
a167b17e DC |
483 | } |
484 | ||
a7b339f1 DC |
485 | /* |
486 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
487 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
488 | * on the xfs syncd work default of 30s. Perhaps this should have it's own | |
489 | * tunable, but that can be done if this method proves to be ineffective or too | |
490 | * aggressive. | |
491 | */ | |
492 | static void | |
493 | xfs_syncd_queue_reclaim( | |
494 | struct xfs_mount *mp) | |
a167b17e | 495 | { |
a167b17e | 496 | |
a7b339f1 DC |
497 | /* |
498 | * We can have inodes enter reclaim after we've shut down the syncd | |
499 | * workqueue during unmount, so don't allow reclaim work to be queued | |
500 | * during unmount. | |
501 | */ | |
502 | if (!(mp->m_super->s_flags & MS_ACTIVE)) | |
503 | return; | |
a167b17e | 504 | |
a7b339f1 DC |
505 | rcu_read_lock(); |
506 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
507 | queue_delayed_work(xfs_syncd_wq, &mp->m_reclaim_work, | |
508 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
a167b17e | 509 | } |
a7b339f1 DC |
510 | rcu_read_unlock(); |
511 | } | |
a167b17e | 512 | |
a7b339f1 DC |
513 | /* |
514 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
515 | * many inodes as possible in a short period of time. It kicks itself every few | |
516 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
517 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
518 | * already being flushed, and once done schedules a future pass. | |
519 | */ | |
520 | STATIC void | |
521 | xfs_reclaim_worker( | |
522 | struct work_struct *work) | |
523 | { | |
524 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
525 | struct xfs_mount, m_reclaim_work); | |
526 | ||
527 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
528 | xfs_syncd_queue_reclaim(mp); | |
529 | } | |
530 | ||
89e4cb55 DC |
531 | /* |
532 | * Flush delayed allocate data, attempting to free up reserved space | |
533 | * from existing allocations. At this point a new allocation attempt | |
534 | * has failed with ENOSPC and we are in the process of scratching our | |
535 | * heads, looking about for more room. | |
536 | * | |
537 | * Queue a new data flush if there isn't one already in progress and | |
538 | * wait for completion of the flush. This means that we only ever have one | |
539 | * inode flush in progress no matter how many ENOSPC events are occurring and | |
540 | * so will prevent the system from bogging down due to every concurrent | |
541 | * ENOSPC event scanning all the active inodes in the system for writeback. | |
542 | */ | |
543 | void | |
544 | xfs_flush_inodes( | |
545 | struct xfs_inode *ip) | |
546 | { | |
547 | struct xfs_mount *mp = ip->i_mount; | |
548 | ||
549 | queue_work(xfs_syncd_wq, &mp->m_flush_work); | |
550 | flush_work_sync(&mp->m_flush_work); | |
551 | } | |
552 | ||
553 | STATIC void | |
554 | xfs_flush_worker( | |
555 | struct work_struct *work) | |
556 | { | |
557 | struct xfs_mount *mp = container_of(work, | |
558 | struct xfs_mount, m_flush_work); | |
559 | ||
560 | xfs_sync_data(mp, SYNC_TRYLOCK); | |
561 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); | |
a167b17e DC |
562 | } |
563 | ||
564 | int | |
565 | xfs_syncd_init( | |
566 | struct xfs_mount *mp) | |
567 | { | |
89e4cb55 | 568 | INIT_WORK(&mp->m_flush_work, xfs_flush_worker); |
c6d09b66 | 569 | INIT_DELAYED_WORK(&mp->m_sync_work, xfs_sync_worker); |
a7b339f1 DC |
570 | INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker); |
571 | ||
c6d09b66 | 572 | xfs_syncd_queue_sync(mp); |
a7b339f1 | 573 | xfs_syncd_queue_reclaim(mp); |
c6d09b66 | 574 | |
a167b17e DC |
575 | return 0; |
576 | } | |
577 | ||
578 | void | |
579 | xfs_syncd_stop( | |
580 | struct xfs_mount *mp) | |
581 | { | |
c6d09b66 | 582 | cancel_delayed_work_sync(&mp->m_sync_work); |
a7b339f1 | 583 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
89e4cb55 | 584 | cancel_work_sync(&mp->m_flush_work); |
a167b17e DC |
585 | } |
586 | ||
bc990f5c CH |
587 | void |
588 | __xfs_inode_set_reclaim_tag( | |
589 | struct xfs_perag *pag, | |
590 | struct xfs_inode *ip) | |
591 | { | |
592 | radix_tree_tag_set(&pag->pag_ici_root, | |
593 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
594 | XFS_ICI_RECLAIM_TAG); | |
16fd5367 DC |
595 | |
596 | if (!pag->pag_ici_reclaimable) { | |
597 | /* propagate the reclaim tag up into the perag radix tree */ | |
598 | spin_lock(&ip->i_mount->m_perag_lock); | |
599 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
600 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
601 | XFS_ICI_RECLAIM_TAG); | |
602 | spin_unlock(&ip->i_mount->m_perag_lock); | |
a7b339f1 DC |
603 | |
604 | /* schedule periodic background inode reclaim */ | |
605 | xfs_syncd_queue_reclaim(ip->i_mount); | |
606 | ||
16fd5367 DC |
607 | trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno, |
608 | -1, _RET_IP_); | |
609 | } | |
9bf729c0 | 610 | pag->pag_ici_reclaimable++; |
bc990f5c CH |
611 | } |
612 | ||
11654513 DC |
613 | /* |
614 | * We set the inode flag atomically with the radix tree tag. | |
615 | * Once we get tag lookups on the radix tree, this inode flag | |
616 | * can go away. | |
617 | */ | |
396beb85 DC |
618 | void |
619 | xfs_inode_set_reclaim_tag( | |
620 | xfs_inode_t *ip) | |
621 | { | |
5017e97d DC |
622 | struct xfs_mount *mp = ip->i_mount; |
623 | struct xfs_perag *pag; | |
396beb85 | 624 | |
5017e97d | 625 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1a427ab0 | 626 | spin_lock(&pag->pag_ici_lock); |
396beb85 | 627 | spin_lock(&ip->i_flags_lock); |
bc990f5c | 628 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 629 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 | 630 | spin_unlock(&ip->i_flags_lock); |
1a427ab0 | 631 | spin_unlock(&pag->pag_ici_lock); |
5017e97d | 632 | xfs_perag_put(pag); |
396beb85 DC |
633 | } |
634 | ||
081003ff JW |
635 | STATIC void |
636 | __xfs_inode_clear_reclaim( | |
396beb85 DC |
637 | xfs_perag_t *pag, |
638 | xfs_inode_t *ip) | |
639 | { | |
9bf729c0 | 640 | pag->pag_ici_reclaimable--; |
16fd5367 DC |
641 | if (!pag->pag_ici_reclaimable) { |
642 | /* clear the reclaim tag from the perag radix tree */ | |
643 | spin_lock(&ip->i_mount->m_perag_lock); | |
644 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
645 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
646 | XFS_ICI_RECLAIM_TAG); | |
647 | spin_unlock(&ip->i_mount->m_perag_lock); | |
648 | trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno, | |
649 | -1, _RET_IP_); | |
650 | } | |
396beb85 DC |
651 | } |
652 | ||
081003ff JW |
653 | void |
654 | __xfs_inode_clear_reclaim_tag( | |
655 | xfs_mount_t *mp, | |
656 | xfs_perag_t *pag, | |
657 | xfs_inode_t *ip) | |
658 | { | |
659 | radix_tree_tag_clear(&pag->pag_ici_root, | |
660 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
661 | __xfs_inode_clear_reclaim(pag, ip); | |
662 | } | |
663 | ||
e3a20c0b DC |
664 | /* |
665 | * Grab the inode for reclaim exclusively. | |
666 | * Return 0 if we grabbed it, non-zero otherwise. | |
667 | */ | |
668 | STATIC int | |
669 | xfs_reclaim_inode_grab( | |
670 | struct xfs_inode *ip, | |
671 | int flags) | |
672 | { | |
1a3e8f3d DC |
673 | ASSERT(rcu_read_lock_held()); |
674 | ||
675 | /* quick check for stale RCU freed inode */ | |
676 | if (!ip->i_ino) | |
677 | return 1; | |
e3a20c0b DC |
678 | |
679 | /* | |
1a3e8f3d | 680 | * do some unlocked checks first to avoid unnecessary lock traffic. |
e3a20c0b DC |
681 | * The first is a flush lock check, the second is a already in reclaim |
682 | * check. Only do these checks if we are not going to block on locks. | |
683 | */ | |
684 | if ((flags & SYNC_TRYLOCK) && | |
685 | (!ip->i_flush.done || __xfs_iflags_test(ip, XFS_IRECLAIM))) { | |
686 | return 1; | |
687 | } | |
688 | ||
689 | /* | |
690 | * The radix tree lock here protects a thread in xfs_iget from racing | |
691 | * with us starting reclaim on the inode. Once we have the | |
692 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
693 | * |
694 | * Due to RCU lookup, we may find inodes that have been freed and only | |
695 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
696 | * aren't candidates for reclaim at all, so we must check the | |
697 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
698 | */ |
699 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
700 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
701 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
702 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
703 | spin_unlock(&ip->i_flags_lock); |
704 | return 1; | |
705 | } | |
706 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
707 | spin_unlock(&ip->i_flags_lock); | |
708 | return 0; | |
709 | } | |
710 | ||
777df5af DC |
711 | /* |
712 | * Inodes in different states need to be treated differently, and the return | |
713 | * value of xfs_iflush is not sufficient to get this right. The following table | |
714 | * lists the inode states and the reclaim actions necessary for non-blocking | |
715 | * reclaim: | |
716 | * | |
717 | * | |
718 | * inode state iflush ret required action | |
719 | * --------------- ---------- --------------- | |
720 | * bad - reclaim | |
721 | * shutdown EIO unpin and reclaim | |
722 | * clean, unpinned 0 reclaim | |
723 | * stale, unpinned 0 reclaim | |
c854363e DC |
724 | * clean, pinned(*) 0 requeue |
725 | * stale, pinned EAGAIN requeue | |
726 | * dirty, delwri ok 0 requeue | |
727 | * dirty, delwri blocked EAGAIN requeue | |
728 | * dirty, sync flush 0 reclaim | |
777df5af DC |
729 | * |
730 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
731 | * handled anyway given the order of checks implemented. | |
732 | * | |
c854363e DC |
733 | * As can be seen from the table, the return value of xfs_iflush() is not |
734 | * sufficient to correctly decide the reclaim action here. The checks in | |
735 | * xfs_iflush() might look like duplicates, but they are not. | |
736 | * | |
737 | * Also, because we get the flush lock first, we know that any inode that has | |
738 | * been flushed delwri has had the flush completed by the time we check that | |
739 | * the inode is clean. The clean inode check needs to be done before flushing | |
740 | * the inode delwri otherwise we would loop forever requeuing clean inodes as | |
741 | * we cannot tell apart a successful delwri flush and a clean inode from the | |
742 | * return value of xfs_iflush(). | |
743 | * | |
744 | * Note that because the inode is flushed delayed write by background | |
745 | * writeback, the flush lock may already be held here and waiting on it can | |
746 | * result in very long latencies. Hence for sync reclaims, where we wait on the | |
747 | * flush lock, the caller should push out delayed write inodes first before | |
748 | * trying to reclaim them to minimise the amount of time spent waiting. For | |
749 | * background relaim, we just requeue the inode for the next pass. | |
750 | * | |
777df5af DC |
751 | * Hence the order of actions after gaining the locks should be: |
752 | * bad => reclaim | |
753 | * shutdown => unpin and reclaim | |
c854363e DC |
754 | * pinned, delwri => requeue |
755 | * pinned, sync => unpin | |
777df5af DC |
756 | * stale => reclaim |
757 | * clean => reclaim | |
c854363e DC |
758 | * dirty, delwri => flush and requeue |
759 | * dirty, sync => flush, wait and reclaim | |
777df5af | 760 | */ |
75f3cb13 | 761 | STATIC int |
c8e20be0 | 762 | xfs_reclaim_inode( |
75f3cb13 DC |
763 | struct xfs_inode *ip, |
764 | struct xfs_perag *pag, | |
c8e20be0 | 765 | int sync_mode) |
fce08f2f | 766 | { |
1bfd8d04 | 767 | int error; |
777df5af | 768 | |
1bfd8d04 DC |
769 | restart: |
770 | error = 0; | |
c8e20be0 | 771 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
772 | if (!xfs_iflock_nowait(ip)) { |
773 | if (!(sync_mode & SYNC_WAIT)) | |
774 | goto out; | |
775 | xfs_iflock(ip); | |
776 | } | |
7a3be02b | 777 | |
777df5af DC |
778 | if (is_bad_inode(VFS_I(ip))) |
779 | goto reclaim; | |
780 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
781 | xfs_iunpin_wait(ip); | |
782 | goto reclaim; | |
783 | } | |
c854363e DC |
784 | if (xfs_ipincount(ip)) { |
785 | if (!(sync_mode & SYNC_WAIT)) { | |
786 | xfs_ifunlock(ip); | |
787 | goto out; | |
788 | } | |
777df5af | 789 | xfs_iunpin_wait(ip); |
c854363e | 790 | } |
777df5af DC |
791 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
792 | goto reclaim; | |
793 | if (xfs_inode_clean(ip)) | |
794 | goto reclaim; | |
795 | ||
1bfd8d04 DC |
796 | /* |
797 | * Now we have an inode that needs flushing. | |
798 | * | |
799 | * We do a nonblocking flush here even if we are doing a SYNC_WAIT | |
800 | * reclaim as we can deadlock with inode cluster removal. | |
801 | * xfs_ifree_cluster() can lock the inode buffer before it locks the | |
802 | * ip->i_lock, and we are doing the exact opposite here. As a result, | |
803 | * doing a blocking xfs_itobp() to get the cluster buffer will result | |
804 | * in an ABBA deadlock with xfs_ifree_cluster(). | |
805 | * | |
806 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
807 | * cache to mark them stale, if we hit this case we don't actually want | |
808 | * to do IO here - we want the inode marked stale so we can simply | |
809 | * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush, | |
810 | * just unlock the inode, back off and try again. Hopefully the next | |
811 | * pass through will see the stale flag set on the inode. | |
812 | */ | |
813 | error = xfs_iflush(ip, SYNC_TRYLOCK | sync_mode); | |
c854363e | 814 | if (sync_mode & SYNC_WAIT) { |
1bfd8d04 DC |
815 | if (error == EAGAIN) { |
816 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
817 | /* backoff longer than in xfs_ifree_cluster */ | |
818 | delay(2); | |
819 | goto restart; | |
820 | } | |
c854363e DC |
821 | xfs_iflock(ip); |
822 | goto reclaim; | |
c8e20be0 DC |
823 | } |
824 | ||
c854363e DC |
825 | /* |
826 | * When we have to flush an inode but don't have SYNC_WAIT set, we | |
827 | * flush the inode out using a delwri buffer and wait for the next | |
828 | * call into reclaim to find it in a clean state instead of waiting for | |
829 | * it now. We also don't return errors here - if the error is transient | |
830 | * then the next reclaim pass will flush the inode, and if the error | |
f1d486a3 | 831 | * is permanent then the next sync reclaim will reclaim the inode and |
c854363e DC |
832 | * pass on the error. |
833 | */ | |
f1d486a3 | 834 | if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 835 | xfs_warn(ip->i_mount, |
c854363e DC |
836 | "inode 0x%llx background reclaim flush failed with %d", |
837 | (long long)ip->i_ino, error); | |
838 | } | |
839 | out: | |
840 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
841 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
842 | /* | |
843 | * We could return EAGAIN here to make reclaim rescan the inode tree in | |
844 | * a short while. However, this just burns CPU time scanning the tree | |
845 | * waiting for IO to complete and xfssyncd never goes back to the idle | |
846 | * state. Instead, return 0 to let the next scheduled background reclaim | |
847 | * attempt to reclaim the inode again. | |
848 | */ | |
849 | return 0; | |
850 | ||
777df5af DC |
851 | reclaim: |
852 | xfs_ifunlock(ip); | |
c8e20be0 | 853 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab DC |
854 | |
855 | XFS_STATS_INC(xs_ig_reclaims); | |
856 | /* | |
857 | * Remove the inode from the per-AG radix tree. | |
858 | * | |
859 | * Because radix_tree_delete won't complain even if the item was never | |
860 | * added to the tree assert that it's been there before to catch | |
861 | * problems with the inode life time early on. | |
862 | */ | |
1a427ab0 | 863 | spin_lock(&pag->pag_ici_lock); |
2f11feab DC |
864 | if (!radix_tree_delete(&pag->pag_ici_root, |
865 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino))) | |
866 | ASSERT(0); | |
081003ff | 867 | __xfs_inode_clear_reclaim(pag, ip); |
1a427ab0 | 868 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
869 | |
870 | /* | |
871 | * Here we do an (almost) spurious inode lock in order to coordinate | |
872 | * with inode cache radix tree lookups. This is because the lookup | |
873 | * can reference the inodes in the cache without taking references. | |
874 | * | |
875 | * We make that OK here by ensuring that we wait until the inode is | |
876 | * unlocked after the lookup before we go ahead and free it. We get | |
877 | * both the ilock and the iolock because the code may need to drop the | |
878 | * ilock one but will still hold the iolock. | |
879 | */ | |
880 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
881 | xfs_qm_dqdetach(ip); | |
882 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
883 | ||
884 | xfs_inode_free(ip); | |
c854363e DC |
885 | return error; |
886 | ||
7a3be02b DC |
887 | } |
888 | ||
65d0f205 DC |
889 | /* |
890 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
891 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
892 | * then a shut down during filesystem unmount reclaim walk leak all the | |
893 | * unreclaimed inodes. | |
894 | */ | |
895 | int | |
896 | xfs_reclaim_inodes_ag( | |
897 | struct xfs_mount *mp, | |
898 | int flags, | |
899 | int *nr_to_scan) | |
900 | { | |
901 | struct xfs_perag *pag; | |
902 | int error = 0; | |
903 | int last_error = 0; | |
904 | xfs_agnumber_t ag; | |
69b491c2 DC |
905 | int trylock = flags & SYNC_TRYLOCK; |
906 | int skipped; | |
65d0f205 | 907 | |
69b491c2 | 908 | restart: |
65d0f205 | 909 | ag = 0; |
69b491c2 | 910 | skipped = 0; |
65d0f205 DC |
911 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
912 | unsigned long first_index = 0; | |
913 | int done = 0; | |
e3a20c0b | 914 | int nr_found = 0; |
65d0f205 DC |
915 | |
916 | ag = pag->pag_agno + 1; | |
917 | ||
69b491c2 DC |
918 | if (trylock) { |
919 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
920 | skipped++; | |
f83282a8 | 921 | xfs_perag_put(pag); |
69b491c2 DC |
922 | continue; |
923 | } | |
924 | first_index = pag->pag_ici_reclaim_cursor; | |
925 | } else | |
926 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
927 | ||
65d0f205 | 928 | do { |
e3a20c0b DC |
929 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
930 | int i; | |
65d0f205 | 931 | |
1a3e8f3d | 932 | rcu_read_lock(); |
e3a20c0b DC |
933 | nr_found = radix_tree_gang_lookup_tag( |
934 | &pag->pag_ici_root, | |
935 | (void **)batch, first_index, | |
936 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
937 | XFS_ICI_RECLAIM_TAG); |
938 | if (!nr_found) { | |
b2232219 | 939 | done = 1; |
1a3e8f3d | 940 | rcu_read_unlock(); |
65d0f205 DC |
941 | break; |
942 | } | |
943 | ||
944 | /* | |
e3a20c0b DC |
945 | * Grab the inodes before we drop the lock. if we found |
946 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 947 | */ |
e3a20c0b DC |
948 | for (i = 0; i < nr_found; i++) { |
949 | struct xfs_inode *ip = batch[i]; | |
950 | ||
951 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
952 | batch[i] = NULL; | |
953 | ||
954 | /* | |
955 | * Update the index for the next lookup. Catch | |
956 | * overflows into the next AG range which can | |
957 | * occur if we have inodes in the last block of | |
958 | * the AG and we are currently pointing to the | |
959 | * last inode. | |
1a3e8f3d DC |
960 | * |
961 | * Because we may see inodes that are from the | |
962 | * wrong AG due to RCU freeing and | |
963 | * reallocation, only update the index if it | |
964 | * lies in this AG. It was a race that lead us | |
965 | * to see this inode, so another lookup from | |
966 | * the same index will not find it again. | |
e3a20c0b | 967 | */ |
1a3e8f3d DC |
968 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
969 | pag->pag_agno) | |
970 | continue; | |
e3a20c0b DC |
971 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
972 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
973 | done = 1; | |
974 | } | |
65d0f205 | 975 | |
e3a20c0b | 976 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 977 | rcu_read_unlock(); |
e3a20c0b DC |
978 | |
979 | for (i = 0; i < nr_found; i++) { | |
980 | if (!batch[i]) | |
981 | continue; | |
982 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
983 | if (error && last_error != EFSCORRUPTED) | |
984 | last_error = error; | |
985 | } | |
986 | ||
987 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 988 | |
8daaa831 DC |
989 | cond_resched(); |
990 | ||
e3a20c0b | 991 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 992 | |
69b491c2 DC |
993 | if (trylock && !done) |
994 | pag->pag_ici_reclaim_cursor = first_index; | |
995 | else | |
996 | pag->pag_ici_reclaim_cursor = 0; | |
997 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
998 | xfs_perag_put(pag); |
999 | } | |
69b491c2 DC |
1000 | |
1001 | /* | |
1002 | * if we skipped any AG, and we still have scan count remaining, do | |
1003 | * another pass this time using blocking reclaim semantics (i.e | |
1004 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1005 | * ensure that when we get more reclaimers than AGs we block rather | |
1006 | * than spin trying to execute reclaim. | |
1007 | */ | |
8daaa831 | 1008 | if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) { |
69b491c2 DC |
1009 | trylock = 0; |
1010 | goto restart; | |
1011 | } | |
65d0f205 DC |
1012 | return XFS_ERROR(last_error); |
1013 | } | |
1014 | ||
7a3be02b DC |
1015 | int |
1016 | xfs_reclaim_inodes( | |
1017 | xfs_mount_t *mp, | |
7a3be02b DC |
1018 | int mode) |
1019 | { | |
65d0f205 DC |
1020 | int nr_to_scan = INT_MAX; |
1021 | ||
1022 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1023 | } |
1024 | ||
1025 | /* | |
8daaa831 | 1026 | * Scan a certain number of inodes for reclaim. |
a7b339f1 DC |
1027 | * |
1028 | * When called we make sure that there is a background (fast) inode reclaim in | |
8daaa831 | 1029 | * progress, while we will throttle the speed of reclaim via doing synchronous |
a7b339f1 DC |
1030 | * reclaim of inodes. That means if we come across dirty inodes, we wait for |
1031 | * them to be cleaned, which we hope will not be very long due to the | |
1032 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1033 | */ |
8daaa831 DC |
1034 | void |
1035 | xfs_reclaim_inodes_nr( | |
1036 | struct xfs_mount *mp, | |
1037 | int nr_to_scan) | |
9bf729c0 | 1038 | { |
8daaa831 DC |
1039 | /* kick background reclaimer and push the AIL */ |
1040 | xfs_syncd_queue_reclaim(mp); | |
1041 | xfs_ail_push_all(mp->m_ail); | |
a7b339f1 | 1042 | |
8daaa831 DC |
1043 | xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan); |
1044 | } | |
9bf729c0 | 1045 | |
8daaa831 DC |
1046 | /* |
1047 | * Return the number of reclaimable inodes in the filesystem for | |
1048 | * the shrinker to determine how much to reclaim. | |
1049 | */ | |
1050 | int | |
1051 | xfs_reclaim_inodes_count( | |
1052 | struct xfs_mount *mp) | |
1053 | { | |
1054 | struct xfs_perag *pag; | |
1055 | xfs_agnumber_t ag = 0; | |
1056 | int reclaimable = 0; | |
9bf729c0 | 1057 | |
65d0f205 DC |
1058 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1059 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1060 | reclaimable += pag->pag_ici_reclaimable; |
1061 | xfs_perag_put(pag); | |
9bf729c0 | 1062 | } |
9bf729c0 DC |
1063 | return reclaimable; |
1064 | } | |
1065 |