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)) | |
4a06fd26 | 230 | return 0; |
5a34d5cd DC |
231 | |
232 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
233 | if (flags & SYNC_TRYLOCK) | |
4a06fd26 | 234 | return 0; |
5a34d5cd DC |
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 | 240 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
5a34d5cd DC |
241 | return error; |
242 | } | |
243 | ||
845b6d0c CH |
244 | STATIC int |
245 | xfs_sync_inode_attr( | |
246 | struct xfs_inode *ip, | |
75f3cb13 | 247 | struct xfs_perag *pag, |
845b6d0c CH |
248 | int flags) |
249 | { | |
250 | int error = 0; | |
251 | ||
252 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
253 | if (xfs_inode_clean(ip)) | |
254 | goto out_unlock; | |
255 | if (!xfs_iflock_nowait(ip)) { | |
256 | if (!(flags & SYNC_WAIT)) | |
257 | goto out_unlock; | |
258 | xfs_iflock(ip); | |
259 | } | |
260 | ||
261 | if (xfs_inode_clean(ip)) { | |
262 | xfs_ifunlock(ip); | |
263 | goto out_unlock; | |
264 | } | |
265 | ||
c854363e | 266 | error = xfs_iflush(ip, flags); |
845b6d0c | 267 | |
ee58abdf DC |
268 | /* |
269 | * We don't want to try again on non-blocking flushes that can't run | |
270 | * again immediately. If an inode really must be written, then that's | |
271 | * what the SYNC_WAIT flag is for. | |
272 | */ | |
273 | if (error == EAGAIN) { | |
274 | ASSERT(!(flags & SYNC_WAIT)); | |
275 | error = 0; | |
276 | } | |
277 | ||
845b6d0c CH |
278 | out_unlock: |
279 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
280 | return error; | |
281 | } | |
282 | ||
075fe102 CH |
283 | /* |
284 | * Write out pagecache data for the whole filesystem. | |
285 | */ | |
64c86149 | 286 | STATIC int |
075fe102 CH |
287 | xfs_sync_data( |
288 | struct xfs_mount *mp, | |
289 | int flags) | |
683a8970 | 290 | { |
075fe102 | 291 | int error; |
fe4fa4b8 | 292 | |
b0710ccc | 293 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
fe4fa4b8 | 294 | |
65d0f205 | 295 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags); |
075fe102 CH |
296 | if (error) |
297 | return XFS_ERROR(error); | |
e9f1c6ee | 298 | |
a14a348b | 299 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); |
075fe102 CH |
300 | return 0; |
301 | } | |
e9f1c6ee | 302 | |
075fe102 CH |
303 | /* |
304 | * Write out inode metadata (attributes) for the whole filesystem. | |
305 | */ | |
64c86149 | 306 | STATIC int |
075fe102 CH |
307 | xfs_sync_attr( |
308 | struct xfs_mount *mp, | |
309 | int flags) | |
310 | { | |
311 | ASSERT((flags & ~SYNC_WAIT) == 0); | |
75f3cb13 | 312 | |
65d0f205 | 313 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags); |
fe4fa4b8 DC |
314 | } |
315 | ||
5d77c0dc | 316 | STATIC int |
2af75df7 | 317 | xfs_sync_fsdata( |
df308bcf | 318 | struct xfs_mount *mp) |
2af75df7 CH |
319 | { |
320 | struct xfs_buf *bp; | |
c2b006c1 | 321 | int error; |
2af75df7 CH |
322 | |
323 | /* | |
df308bcf CH |
324 | * If the buffer is pinned then push on the log so we won't get stuck |
325 | * waiting in the write for someone, maybe ourselves, to flush the log. | |
326 | * | |
327 | * Even though we just pushed the log above, we did not have the | |
328 | * superblock buffer locked at that point so it can become pinned in | |
329 | * between there and here. | |
2af75df7 | 330 | */ |
df308bcf | 331 | bp = xfs_getsb(mp, 0); |
811e64c7 | 332 | if (xfs_buf_ispinned(bp)) |
df308bcf | 333 | xfs_log_force(mp, 0); |
c2b006c1 CH |
334 | error = xfs_bwrite(bp); |
335 | xfs_buf_relse(bp); | |
336 | return error; | |
e9f1c6ee DC |
337 | } |
338 | ||
be4f1ac8 CH |
339 | int |
340 | xfs_log_dirty_inode( | |
341 | struct xfs_inode *ip, | |
342 | struct xfs_perag *pag, | |
343 | int flags) | |
344 | { | |
345 | struct xfs_mount *mp = ip->i_mount; | |
346 | struct xfs_trans *tp; | |
347 | int error; | |
348 | ||
349 | if (!ip->i_update_core) | |
350 | return 0; | |
351 | ||
352 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
353 | error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0); | |
354 | if (error) { | |
355 | xfs_trans_cancel(tp, 0); | |
356 | return error; | |
357 | } | |
358 | ||
359 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
360 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
361 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
362 | return xfs_trans_commit(tp, 0); | |
363 | } | |
364 | ||
e9f1c6ee | 365 | /* |
a4e4c4f4 DC |
366 | * When remounting a filesystem read-only or freezing the filesystem, we have |
367 | * two phases to execute. This first phase is syncing the data before we | |
368 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
369 | * we've waited for all the transactions created by the first phase to | |
370 | * complete. The second phase ensures that the inodes are written to their | |
371 | * location on disk rather than just existing in transactions in the log. This | |
372 | * means after a quiesce there is no log replay required to write the inodes to | |
373 | * disk (this is the main difference between a sync and a quiesce). | |
374 | */ | |
375 | /* | |
376 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
377 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
378 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
379 | * transactions can still occur here so don't bother flushing the buftarg |
380 | * because it'll just get dirty again. | |
e9f1c6ee DC |
381 | */ |
382 | int | |
383 | xfs_quiesce_data( | |
384 | struct xfs_mount *mp) | |
385 | { | |
df308bcf | 386 | int error, error2 = 0; |
e9f1c6ee | 387 | |
be4f1ac8 CH |
388 | /* |
389 | * Log all pending size and timestamp updates. The vfs writeback | |
390 | * code is supposed to do this, but due to its overagressive | |
391 | * livelock detection it will skip inodes where appending writes | |
392 | * were written out in the first non-blocking sync phase if their | |
393 | * completion took long enough that it happened after taking the | |
394 | * timestamp for the cut-off in the blocking phase. | |
395 | */ | |
396 | xfs_inode_ag_iterator(mp, xfs_log_dirty_inode, 0); | |
397 | ||
34625c66 | 398 | /* force out the log */ |
33b8f7c2 CH |
399 | xfs_log_force(mp, XFS_LOG_SYNC); |
400 | ||
a4e4c4f4 | 401 | /* write superblock and hoover up shutdown errors */ |
df308bcf CH |
402 | error = xfs_sync_fsdata(mp); |
403 | ||
404 | /* make sure all delwri buffers are written out */ | |
405 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
406 | ||
407 | /* mark the log as covered if needed */ | |
408 | if (xfs_log_need_covered(mp)) | |
c58efdb4 | 409 | error2 = xfs_fs_log_dummy(mp); |
e9f1c6ee | 410 | |
a4e4c4f4 | 411 | /* flush data-only devices */ |
e9f1c6ee | 412 | if (mp->m_rtdev_targp) |
a9add83e | 413 | xfs_flush_buftarg(mp->m_rtdev_targp, 1); |
e9f1c6ee | 414 | |
df308bcf | 415 | return error ? error : error2; |
2af75df7 CH |
416 | } |
417 | ||
76bf105c DC |
418 | STATIC void |
419 | xfs_quiesce_fs( | |
420 | struct xfs_mount *mp) | |
421 | { | |
422 | int count = 0, pincount; | |
423 | ||
c854363e | 424 | xfs_reclaim_inodes(mp, 0); |
76bf105c | 425 | xfs_flush_buftarg(mp->m_ddev_targp, 0); |
76bf105c DC |
426 | |
427 | /* | |
428 | * This loop must run at least twice. The first instance of the loop | |
429 | * will flush most meta data but that will generate more meta data | |
430 | * (typically directory updates). Which then must be flushed and | |
c854363e DC |
431 | * logged before we can write the unmount record. We also so sync |
432 | * reclaim of inodes to catch any that the above delwri flush skipped. | |
76bf105c DC |
433 | */ |
434 | do { | |
c854363e | 435 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
075fe102 | 436 | xfs_sync_attr(mp, SYNC_WAIT); |
76bf105c DC |
437 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
438 | if (!pincount) { | |
439 | delay(50); | |
440 | count++; | |
441 | } | |
442 | } while (count < 2); | |
443 | } | |
444 | ||
445 | /* | |
446 | * Second stage of a quiesce. The data is already synced, now we have to take | |
447 | * care of the metadata. New transactions are already blocked, so we need to | |
25985edc | 448 | * wait for any remaining transactions to drain out before proceeding. |
76bf105c DC |
449 | */ |
450 | void | |
451 | xfs_quiesce_attr( | |
452 | struct xfs_mount *mp) | |
453 | { | |
454 | int error = 0; | |
455 | ||
456 | /* wait for all modifications to complete */ | |
457 | while (atomic_read(&mp->m_active_trans) > 0) | |
458 | delay(100); | |
459 | ||
460 | /* flush inodes and push all remaining buffers out to disk */ | |
461 | xfs_quiesce_fs(mp); | |
462 | ||
5e106572 FB |
463 | /* |
464 | * Just warn here till VFS can correctly support | |
465 | * read-only remount without racing. | |
466 | */ | |
467 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
468 | |
469 | /* Push the superblock and write an unmount record */ | |
adab0f67 | 470 | error = xfs_log_sbcount(mp); |
76bf105c | 471 | if (error) |
4f10700a | 472 | xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. " |
76bf105c DC |
473 | "Frozen image may not be consistent."); |
474 | xfs_log_unmount_write(mp); | |
475 | xfs_unmountfs_writesb(mp); | |
476 | } | |
477 | ||
c6d09b66 DC |
478 | static void |
479 | xfs_syncd_queue_sync( | |
480 | struct xfs_mount *mp) | |
a167b17e | 481 | { |
c6d09b66 DC |
482 | queue_delayed_work(xfs_syncd_wq, &mp->m_sync_work, |
483 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); | |
a167b17e DC |
484 | } |
485 | ||
aacaa880 | 486 | /* |
df308bcf CH |
487 | * Every sync period we need to unpin all items, reclaim inodes and sync |
488 | * disk quotas. We might need to cover the log to indicate that the | |
1a387d3b | 489 | * filesystem is idle and not frozen. |
aacaa880 | 490 | */ |
a167b17e DC |
491 | STATIC void |
492 | xfs_sync_worker( | |
c6d09b66 | 493 | struct work_struct *work) |
a167b17e | 494 | { |
c6d09b66 DC |
495 | struct xfs_mount *mp = container_of(to_delayed_work(work), |
496 | struct xfs_mount, m_sync_work); | |
a167b17e DC |
497 | int error; |
498 | ||
aacaa880 | 499 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
aacaa880 | 500 | /* dgc: errors ignored here */ |
1a387d3b DC |
501 | if (mp->m_super->s_frozen == SB_UNFROZEN && |
502 | xfs_log_need_covered(mp)) | |
c58efdb4 DC |
503 | error = xfs_fs_log_dummy(mp); |
504 | else | |
505 | xfs_log_force(mp, 0); | |
fd074841 DC |
506 | |
507 | /* start pushing all the metadata that is currently dirty */ | |
508 | xfs_ail_push_all(mp->m_ail); | |
aacaa880 | 509 | } |
c6d09b66 DC |
510 | |
511 | /* queue us up again */ | |
512 | xfs_syncd_queue_sync(mp); | |
a167b17e DC |
513 | } |
514 | ||
a7b339f1 DC |
515 | /* |
516 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
517 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
518 | * on the xfs syncd work default of 30s. Perhaps this should have it's own | |
519 | * tunable, but that can be done if this method proves to be ineffective or too | |
520 | * aggressive. | |
521 | */ | |
522 | static void | |
523 | xfs_syncd_queue_reclaim( | |
524 | struct xfs_mount *mp) | |
a167b17e | 525 | { |
a167b17e | 526 | |
a7b339f1 DC |
527 | /* |
528 | * We can have inodes enter reclaim after we've shut down the syncd | |
529 | * workqueue during unmount, so don't allow reclaim work to be queued | |
530 | * during unmount. | |
531 | */ | |
532 | if (!(mp->m_super->s_flags & MS_ACTIVE)) | |
533 | return; | |
a167b17e | 534 | |
a7b339f1 DC |
535 | rcu_read_lock(); |
536 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
537 | queue_delayed_work(xfs_syncd_wq, &mp->m_reclaim_work, | |
538 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
a167b17e | 539 | } |
a7b339f1 DC |
540 | rcu_read_unlock(); |
541 | } | |
a167b17e | 542 | |
a7b339f1 DC |
543 | /* |
544 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
545 | * many inodes as possible in a short period of time. It kicks itself every few | |
546 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
547 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
548 | * already being flushed, and once done schedules a future pass. | |
549 | */ | |
550 | STATIC void | |
551 | xfs_reclaim_worker( | |
552 | struct work_struct *work) | |
553 | { | |
554 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
555 | struct xfs_mount, m_reclaim_work); | |
556 | ||
557 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
558 | xfs_syncd_queue_reclaim(mp); | |
559 | } | |
560 | ||
89e4cb55 DC |
561 | /* |
562 | * Flush delayed allocate data, attempting to free up reserved space | |
563 | * from existing allocations. At this point a new allocation attempt | |
564 | * has failed with ENOSPC and we are in the process of scratching our | |
565 | * heads, looking about for more room. | |
566 | * | |
567 | * Queue a new data flush if there isn't one already in progress and | |
568 | * wait for completion of the flush. This means that we only ever have one | |
569 | * inode flush in progress no matter how many ENOSPC events are occurring and | |
570 | * so will prevent the system from bogging down due to every concurrent | |
571 | * ENOSPC event scanning all the active inodes in the system for writeback. | |
572 | */ | |
573 | void | |
574 | xfs_flush_inodes( | |
575 | struct xfs_inode *ip) | |
576 | { | |
577 | struct xfs_mount *mp = ip->i_mount; | |
578 | ||
579 | queue_work(xfs_syncd_wq, &mp->m_flush_work); | |
580 | flush_work_sync(&mp->m_flush_work); | |
581 | } | |
582 | ||
583 | STATIC void | |
584 | xfs_flush_worker( | |
585 | struct work_struct *work) | |
586 | { | |
587 | struct xfs_mount *mp = container_of(work, | |
588 | struct xfs_mount, m_flush_work); | |
589 | ||
590 | xfs_sync_data(mp, SYNC_TRYLOCK); | |
591 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); | |
a167b17e DC |
592 | } |
593 | ||
594 | int | |
595 | xfs_syncd_init( | |
596 | struct xfs_mount *mp) | |
597 | { | |
89e4cb55 | 598 | INIT_WORK(&mp->m_flush_work, xfs_flush_worker); |
c6d09b66 | 599 | INIT_DELAYED_WORK(&mp->m_sync_work, xfs_sync_worker); |
a7b339f1 DC |
600 | INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker); |
601 | ||
c6d09b66 | 602 | xfs_syncd_queue_sync(mp); |
a7b339f1 | 603 | xfs_syncd_queue_reclaim(mp); |
c6d09b66 | 604 | |
a167b17e DC |
605 | return 0; |
606 | } | |
607 | ||
608 | void | |
609 | xfs_syncd_stop( | |
610 | struct xfs_mount *mp) | |
611 | { | |
c6d09b66 | 612 | cancel_delayed_work_sync(&mp->m_sync_work); |
a7b339f1 | 613 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
89e4cb55 | 614 | cancel_work_sync(&mp->m_flush_work); |
a167b17e DC |
615 | } |
616 | ||
bc990f5c CH |
617 | void |
618 | __xfs_inode_set_reclaim_tag( | |
619 | struct xfs_perag *pag, | |
620 | struct xfs_inode *ip) | |
621 | { | |
622 | radix_tree_tag_set(&pag->pag_ici_root, | |
623 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
624 | XFS_ICI_RECLAIM_TAG); | |
16fd5367 DC |
625 | |
626 | if (!pag->pag_ici_reclaimable) { | |
627 | /* propagate the reclaim tag up into the perag radix tree */ | |
628 | spin_lock(&ip->i_mount->m_perag_lock); | |
629 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
630 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
631 | XFS_ICI_RECLAIM_TAG); | |
632 | spin_unlock(&ip->i_mount->m_perag_lock); | |
a7b339f1 DC |
633 | |
634 | /* schedule periodic background inode reclaim */ | |
635 | xfs_syncd_queue_reclaim(ip->i_mount); | |
636 | ||
16fd5367 DC |
637 | trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno, |
638 | -1, _RET_IP_); | |
639 | } | |
9bf729c0 | 640 | pag->pag_ici_reclaimable++; |
bc990f5c CH |
641 | } |
642 | ||
11654513 DC |
643 | /* |
644 | * We set the inode flag atomically with the radix tree tag. | |
645 | * Once we get tag lookups on the radix tree, this inode flag | |
646 | * can go away. | |
647 | */ | |
396beb85 DC |
648 | void |
649 | xfs_inode_set_reclaim_tag( | |
650 | xfs_inode_t *ip) | |
651 | { | |
5017e97d DC |
652 | struct xfs_mount *mp = ip->i_mount; |
653 | struct xfs_perag *pag; | |
396beb85 | 654 | |
5017e97d | 655 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1a427ab0 | 656 | spin_lock(&pag->pag_ici_lock); |
396beb85 | 657 | spin_lock(&ip->i_flags_lock); |
bc990f5c | 658 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 659 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 | 660 | spin_unlock(&ip->i_flags_lock); |
1a427ab0 | 661 | spin_unlock(&pag->pag_ici_lock); |
5017e97d | 662 | xfs_perag_put(pag); |
396beb85 DC |
663 | } |
664 | ||
081003ff JW |
665 | STATIC void |
666 | __xfs_inode_clear_reclaim( | |
396beb85 DC |
667 | xfs_perag_t *pag, |
668 | xfs_inode_t *ip) | |
669 | { | |
9bf729c0 | 670 | pag->pag_ici_reclaimable--; |
16fd5367 DC |
671 | if (!pag->pag_ici_reclaimable) { |
672 | /* clear the reclaim tag from the perag radix tree */ | |
673 | spin_lock(&ip->i_mount->m_perag_lock); | |
674 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
675 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
676 | XFS_ICI_RECLAIM_TAG); | |
677 | spin_unlock(&ip->i_mount->m_perag_lock); | |
678 | trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno, | |
679 | -1, _RET_IP_); | |
680 | } | |
396beb85 DC |
681 | } |
682 | ||
081003ff JW |
683 | void |
684 | __xfs_inode_clear_reclaim_tag( | |
685 | xfs_mount_t *mp, | |
686 | xfs_perag_t *pag, | |
687 | xfs_inode_t *ip) | |
688 | { | |
689 | radix_tree_tag_clear(&pag->pag_ici_root, | |
690 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
691 | __xfs_inode_clear_reclaim(pag, ip); | |
692 | } | |
693 | ||
e3a20c0b DC |
694 | /* |
695 | * Grab the inode for reclaim exclusively. | |
696 | * Return 0 if we grabbed it, non-zero otherwise. | |
697 | */ | |
698 | STATIC int | |
699 | xfs_reclaim_inode_grab( | |
700 | struct xfs_inode *ip, | |
701 | int flags) | |
702 | { | |
1a3e8f3d DC |
703 | ASSERT(rcu_read_lock_held()); |
704 | ||
705 | /* quick check for stale RCU freed inode */ | |
706 | if (!ip->i_ino) | |
707 | return 1; | |
e3a20c0b DC |
708 | |
709 | /* | |
1a3e8f3d | 710 | * do some unlocked checks first to avoid unnecessary lock traffic. |
e3a20c0b DC |
711 | * The first is a flush lock check, the second is a already in reclaim |
712 | * check. Only do these checks if we are not going to block on locks. | |
713 | */ | |
714 | if ((flags & SYNC_TRYLOCK) && | |
715 | (!ip->i_flush.done || __xfs_iflags_test(ip, XFS_IRECLAIM))) { | |
716 | return 1; | |
717 | } | |
718 | ||
719 | /* | |
720 | * The radix tree lock here protects a thread in xfs_iget from racing | |
721 | * with us starting reclaim on the inode. Once we have the | |
722 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
723 | * |
724 | * Due to RCU lookup, we may find inodes that have been freed and only | |
725 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
726 | * aren't candidates for reclaim at all, so we must check the | |
727 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
728 | */ |
729 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
730 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
731 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
732 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
733 | spin_unlock(&ip->i_flags_lock); |
734 | return 1; | |
735 | } | |
736 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
737 | spin_unlock(&ip->i_flags_lock); | |
738 | return 0; | |
739 | } | |
740 | ||
777df5af DC |
741 | /* |
742 | * Inodes in different states need to be treated differently, and the return | |
743 | * value of xfs_iflush is not sufficient to get this right. The following table | |
744 | * lists the inode states and the reclaim actions necessary for non-blocking | |
745 | * reclaim: | |
746 | * | |
747 | * | |
748 | * inode state iflush ret required action | |
749 | * --------------- ---------- --------------- | |
750 | * bad - reclaim | |
751 | * shutdown EIO unpin and reclaim | |
752 | * clean, unpinned 0 reclaim | |
753 | * stale, unpinned 0 reclaim | |
c854363e DC |
754 | * clean, pinned(*) 0 requeue |
755 | * stale, pinned EAGAIN requeue | |
756 | * dirty, delwri ok 0 requeue | |
757 | * dirty, delwri blocked EAGAIN requeue | |
758 | * dirty, sync flush 0 reclaim | |
777df5af DC |
759 | * |
760 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
761 | * handled anyway given the order of checks implemented. | |
762 | * | |
c854363e DC |
763 | * As can be seen from the table, the return value of xfs_iflush() is not |
764 | * sufficient to correctly decide the reclaim action here. The checks in | |
765 | * xfs_iflush() might look like duplicates, but they are not. | |
766 | * | |
767 | * Also, because we get the flush lock first, we know that any inode that has | |
768 | * been flushed delwri has had the flush completed by the time we check that | |
769 | * the inode is clean. The clean inode check needs to be done before flushing | |
770 | * the inode delwri otherwise we would loop forever requeuing clean inodes as | |
771 | * we cannot tell apart a successful delwri flush and a clean inode from the | |
772 | * return value of xfs_iflush(). | |
773 | * | |
774 | * Note that because the inode is flushed delayed write by background | |
775 | * writeback, the flush lock may already be held here and waiting on it can | |
776 | * result in very long latencies. Hence for sync reclaims, where we wait on the | |
777 | * flush lock, the caller should push out delayed write inodes first before | |
778 | * trying to reclaim them to minimise the amount of time spent waiting. For | |
779 | * background relaim, we just requeue the inode for the next pass. | |
780 | * | |
777df5af DC |
781 | * Hence the order of actions after gaining the locks should be: |
782 | * bad => reclaim | |
783 | * shutdown => unpin and reclaim | |
c854363e DC |
784 | * pinned, delwri => requeue |
785 | * pinned, sync => unpin | |
777df5af DC |
786 | * stale => reclaim |
787 | * clean => reclaim | |
c854363e DC |
788 | * dirty, delwri => flush and requeue |
789 | * dirty, sync => flush, wait and reclaim | |
777df5af | 790 | */ |
75f3cb13 | 791 | STATIC int |
c8e20be0 | 792 | xfs_reclaim_inode( |
75f3cb13 DC |
793 | struct xfs_inode *ip, |
794 | struct xfs_perag *pag, | |
c8e20be0 | 795 | int sync_mode) |
fce08f2f | 796 | { |
1bfd8d04 | 797 | int error; |
777df5af | 798 | |
1bfd8d04 DC |
799 | restart: |
800 | error = 0; | |
c8e20be0 | 801 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
802 | if (!xfs_iflock_nowait(ip)) { |
803 | if (!(sync_mode & SYNC_WAIT)) | |
804 | goto out; | |
4dd2cb4a CH |
805 | |
806 | /* | |
807 | * If we only have a single dirty inode in a cluster there is | |
808 | * a fair chance that the AIL push may have pushed it into | |
809 | * the buffer, but xfsbufd won't touch it until 30 seconds | |
810 | * from now, and thus we will lock up here. | |
811 | * | |
812 | * Promote the inode buffer to the front of the delwri list | |
813 | * and wake up xfsbufd now. | |
814 | */ | |
815 | xfs_promote_inode(ip); | |
c854363e DC |
816 | xfs_iflock(ip); |
817 | } | |
7a3be02b | 818 | |
777df5af DC |
819 | if (is_bad_inode(VFS_I(ip))) |
820 | goto reclaim; | |
821 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
822 | xfs_iunpin_wait(ip); | |
823 | goto reclaim; | |
824 | } | |
c854363e DC |
825 | if (xfs_ipincount(ip)) { |
826 | if (!(sync_mode & SYNC_WAIT)) { | |
827 | xfs_ifunlock(ip); | |
828 | goto out; | |
829 | } | |
777df5af | 830 | xfs_iunpin_wait(ip); |
c854363e | 831 | } |
777df5af DC |
832 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
833 | goto reclaim; | |
834 | if (xfs_inode_clean(ip)) | |
835 | goto reclaim; | |
836 | ||
1bfd8d04 DC |
837 | /* |
838 | * Now we have an inode that needs flushing. | |
839 | * | |
840 | * We do a nonblocking flush here even if we are doing a SYNC_WAIT | |
841 | * reclaim as we can deadlock with inode cluster removal. | |
842 | * xfs_ifree_cluster() can lock the inode buffer before it locks the | |
843 | * ip->i_lock, and we are doing the exact opposite here. As a result, | |
844 | * doing a blocking xfs_itobp() to get the cluster buffer will result | |
845 | * in an ABBA deadlock with xfs_ifree_cluster(). | |
846 | * | |
847 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
848 | * cache to mark them stale, if we hit this case we don't actually want | |
849 | * to do IO here - we want the inode marked stale so we can simply | |
850 | * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush, | |
851 | * just unlock the inode, back off and try again. Hopefully the next | |
852 | * pass through will see the stale flag set on the inode. | |
853 | */ | |
854 | error = xfs_iflush(ip, SYNC_TRYLOCK | sync_mode); | |
c854363e | 855 | if (sync_mode & SYNC_WAIT) { |
1bfd8d04 DC |
856 | if (error == EAGAIN) { |
857 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
858 | /* backoff longer than in xfs_ifree_cluster */ | |
859 | delay(2); | |
860 | goto restart; | |
861 | } | |
c854363e DC |
862 | xfs_iflock(ip); |
863 | goto reclaim; | |
c8e20be0 DC |
864 | } |
865 | ||
c854363e DC |
866 | /* |
867 | * When we have to flush an inode but don't have SYNC_WAIT set, we | |
868 | * flush the inode out using a delwri buffer and wait for the next | |
869 | * call into reclaim to find it in a clean state instead of waiting for | |
870 | * it now. We also don't return errors here - if the error is transient | |
871 | * then the next reclaim pass will flush the inode, and if the error | |
f1d486a3 | 872 | * is permanent then the next sync reclaim will reclaim the inode and |
c854363e DC |
873 | * pass on the error. |
874 | */ | |
f1d486a3 | 875 | if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 876 | xfs_warn(ip->i_mount, |
c854363e DC |
877 | "inode 0x%llx background reclaim flush failed with %d", |
878 | (long long)ip->i_ino, error); | |
879 | } | |
880 | out: | |
881 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
882 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
883 | /* | |
884 | * We could return EAGAIN here to make reclaim rescan the inode tree in | |
885 | * a short while. However, this just burns CPU time scanning the tree | |
886 | * waiting for IO to complete and xfssyncd never goes back to the idle | |
887 | * state. Instead, return 0 to let the next scheduled background reclaim | |
888 | * attempt to reclaim the inode again. | |
889 | */ | |
890 | return 0; | |
891 | ||
777df5af DC |
892 | reclaim: |
893 | xfs_ifunlock(ip); | |
c8e20be0 | 894 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab DC |
895 | |
896 | XFS_STATS_INC(xs_ig_reclaims); | |
897 | /* | |
898 | * Remove the inode from the per-AG radix tree. | |
899 | * | |
900 | * Because radix_tree_delete won't complain even if the item was never | |
901 | * added to the tree assert that it's been there before to catch | |
902 | * problems with the inode life time early on. | |
903 | */ | |
1a427ab0 | 904 | spin_lock(&pag->pag_ici_lock); |
2f11feab DC |
905 | if (!radix_tree_delete(&pag->pag_ici_root, |
906 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino))) | |
907 | ASSERT(0); | |
081003ff | 908 | __xfs_inode_clear_reclaim(pag, ip); |
1a427ab0 | 909 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
910 | |
911 | /* | |
912 | * Here we do an (almost) spurious inode lock in order to coordinate | |
913 | * with inode cache radix tree lookups. This is because the lookup | |
914 | * can reference the inodes in the cache without taking references. | |
915 | * | |
916 | * We make that OK here by ensuring that we wait until the inode is | |
917 | * unlocked after the lookup before we go ahead and free it. We get | |
918 | * both the ilock and the iolock because the code may need to drop the | |
919 | * ilock one but will still hold the iolock. | |
920 | */ | |
921 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
922 | xfs_qm_dqdetach(ip); | |
923 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
924 | ||
925 | xfs_inode_free(ip); | |
c854363e DC |
926 | return error; |
927 | ||
7a3be02b DC |
928 | } |
929 | ||
65d0f205 DC |
930 | /* |
931 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
932 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
933 | * then a shut down during filesystem unmount reclaim walk leak all the | |
934 | * unreclaimed inodes. | |
935 | */ | |
936 | int | |
937 | xfs_reclaim_inodes_ag( | |
938 | struct xfs_mount *mp, | |
939 | int flags, | |
940 | int *nr_to_scan) | |
941 | { | |
942 | struct xfs_perag *pag; | |
943 | int error = 0; | |
944 | int last_error = 0; | |
945 | xfs_agnumber_t ag; | |
69b491c2 DC |
946 | int trylock = flags & SYNC_TRYLOCK; |
947 | int skipped; | |
65d0f205 | 948 | |
69b491c2 | 949 | restart: |
65d0f205 | 950 | ag = 0; |
69b491c2 | 951 | skipped = 0; |
65d0f205 DC |
952 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
953 | unsigned long first_index = 0; | |
954 | int done = 0; | |
e3a20c0b | 955 | int nr_found = 0; |
65d0f205 DC |
956 | |
957 | ag = pag->pag_agno + 1; | |
958 | ||
69b491c2 DC |
959 | if (trylock) { |
960 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
961 | skipped++; | |
f83282a8 | 962 | xfs_perag_put(pag); |
69b491c2 DC |
963 | continue; |
964 | } | |
965 | first_index = pag->pag_ici_reclaim_cursor; | |
966 | } else | |
967 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
968 | ||
65d0f205 | 969 | do { |
e3a20c0b DC |
970 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
971 | int i; | |
65d0f205 | 972 | |
1a3e8f3d | 973 | rcu_read_lock(); |
e3a20c0b DC |
974 | nr_found = radix_tree_gang_lookup_tag( |
975 | &pag->pag_ici_root, | |
976 | (void **)batch, first_index, | |
977 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
978 | XFS_ICI_RECLAIM_TAG); |
979 | if (!nr_found) { | |
b2232219 | 980 | done = 1; |
1a3e8f3d | 981 | rcu_read_unlock(); |
65d0f205 DC |
982 | break; |
983 | } | |
984 | ||
985 | /* | |
e3a20c0b DC |
986 | * Grab the inodes before we drop the lock. if we found |
987 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 988 | */ |
e3a20c0b DC |
989 | for (i = 0; i < nr_found; i++) { |
990 | struct xfs_inode *ip = batch[i]; | |
991 | ||
992 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
993 | batch[i] = NULL; | |
994 | ||
995 | /* | |
996 | * Update the index for the next lookup. Catch | |
997 | * overflows into the next AG range which can | |
998 | * occur if we have inodes in the last block of | |
999 | * the AG and we are currently pointing to the | |
1000 | * last inode. | |
1a3e8f3d DC |
1001 | * |
1002 | * Because we may see inodes that are from the | |
1003 | * wrong AG due to RCU freeing and | |
1004 | * reallocation, only update the index if it | |
1005 | * lies in this AG. It was a race that lead us | |
1006 | * to see this inode, so another lookup from | |
1007 | * the same index will not find it again. | |
e3a20c0b | 1008 | */ |
1a3e8f3d DC |
1009 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
1010 | pag->pag_agno) | |
1011 | continue; | |
e3a20c0b DC |
1012 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
1013 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1014 | done = 1; | |
1015 | } | |
65d0f205 | 1016 | |
e3a20c0b | 1017 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 1018 | rcu_read_unlock(); |
e3a20c0b DC |
1019 | |
1020 | for (i = 0; i < nr_found; i++) { | |
1021 | if (!batch[i]) | |
1022 | continue; | |
1023 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
1024 | if (error && last_error != EFSCORRUPTED) | |
1025 | last_error = error; | |
1026 | } | |
1027 | ||
1028 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 1029 | |
8daaa831 DC |
1030 | cond_resched(); |
1031 | ||
e3a20c0b | 1032 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 1033 | |
69b491c2 DC |
1034 | if (trylock && !done) |
1035 | pag->pag_ici_reclaim_cursor = first_index; | |
1036 | else | |
1037 | pag->pag_ici_reclaim_cursor = 0; | |
1038 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
1039 | xfs_perag_put(pag); |
1040 | } | |
69b491c2 DC |
1041 | |
1042 | /* | |
1043 | * if we skipped any AG, and we still have scan count remaining, do | |
1044 | * another pass this time using blocking reclaim semantics (i.e | |
1045 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1046 | * ensure that when we get more reclaimers than AGs we block rather | |
1047 | * than spin trying to execute reclaim. | |
1048 | */ | |
8daaa831 | 1049 | if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) { |
69b491c2 DC |
1050 | trylock = 0; |
1051 | goto restart; | |
1052 | } | |
65d0f205 DC |
1053 | return XFS_ERROR(last_error); |
1054 | } | |
1055 | ||
7a3be02b DC |
1056 | int |
1057 | xfs_reclaim_inodes( | |
1058 | xfs_mount_t *mp, | |
7a3be02b DC |
1059 | int mode) |
1060 | { | |
65d0f205 DC |
1061 | int nr_to_scan = INT_MAX; |
1062 | ||
1063 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1064 | } |
1065 | ||
1066 | /* | |
8daaa831 | 1067 | * Scan a certain number of inodes for reclaim. |
a7b339f1 DC |
1068 | * |
1069 | * When called we make sure that there is a background (fast) inode reclaim in | |
8daaa831 | 1070 | * progress, while we will throttle the speed of reclaim via doing synchronous |
a7b339f1 DC |
1071 | * reclaim of inodes. That means if we come across dirty inodes, we wait for |
1072 | * them to be cleaned, which we hope will not be very long due to the | |
1073 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1074 | */ |
8daaa831 DC |
1075 | void |
1076 | xfs_reclaim_inodes_nr( | |
1077 | struct xfs_mount *mp, | |
1078 | int nr_to_scan) | |
9bf729c0 | 1079 | { |
8daaa831 DC |
1080 | /* kick background reclaimer and push the AIL */ |
1081 | xfs_syncd_queue_reclaim(mp); | |
1082 | xfs_ail_push_all(mp->m_ail); | |
a7b339f1 | 1083 | |
8daaa831 DC |
1084 | xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan); |
1085 | } | |
9bf729c0 | 1086 | |
8daaa831 DC |
1087 | /* |
1088 | * Return the number of reclaimable inodes in the filesystem for | |
1089 | * the shrinker to determine how much to reclaim. | |
1090 | */ | |
1091 | int | |
1092 | xfs_reclaim_inodes_count( | |
1093 | struct xfs_mount *mp) | |
1094 | { | |
1095 | struct xfs_perag *pag; | |
1096 | xfs_agnumber_t ag = 0; | |
1097 | int reclaimable = 0; | |
9bf729c0 | 1098 | |
65d0f205 DC |
1099 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1100 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1101 | reclaimable += pag->pag_ici_reclaimable; |
1102 | xfs_perag_put(pag); | |
9bf729c0 | 1103 | } |
9bf729c0 DC |
1104 | return reclaimable; |
1105 | } | |
1106 |