Commit | Line | Data |
---|---|---|
fe4fa4b8 DC |
1 | /* |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. | |
3 | * All Rights Reserved. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU General Public License as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it would be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
17 | */ | |
18 | #include "xfs.h" | |
19 | #include "xfs_fs.h" | |
20 | #include "xfs_types.h" | |
21 | #include "xfs_bit.h" | |
22 | #include "xfs_log.h" | |
23 | #include "xfs_inum.h" | |
24 | #include "xfs_trans.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
fe4fa4b8 DC |
27 | #include "xfs_mount.h" |
28 | #include "xfs_bmap_btree.h" | |
fe4fa4b8 DC |
29 | #include "xfs_inode.h" |
30 | #include "xfs_dinode.h" | |
31 | #include "xfs_error.h" | |
fe4fa4b8 DC |
32 | #include "xfs_filestream.h" |
33 | #include "xfs_vnodeops.h" | |
fe4fa4b8 | 34 | #include "xfs_inode_item.h" |
7d095257 | 35 | #include "xfs_quota.h" |
0b1b213f | 36 | #include "xfs_trace.h" |
1a387d3b | 37 | #include "xfs_fsops.h" |
fe4fa4b8 | 38 | |
a167b17e DC |
39 | #include <linux/kthread.h> |
40 | #include <linux/freezer.h> | |
41 | ||
5a34d5cd | 42 | |
75f3cb13 DC |
43 | STATIC int |
44 | xfs_inode_ag_walk( | |
45 | struct xfs_mount *mp, | |
5017e97d | 46 | struct xfs_perag *pag, |
75f3cb13 DC |
47 | int (*execute)(struct xfs_inode *ip, |
48 | struct xfs_perag *pag, int flags), | |
65d0f205 | 49 | int flags) |
75f3cb13 | 50 | { |
75f3cb13 DC |
51 | uint32_t first_index; |
52 | int last_error = 0; | |
53 | int skipped; | |
65d0f205 | 54 | int done; |
75f3cb13 DC |
55 | |
56 | restart: | |
65d0f205 | 57 | done = 0; |
75f3cb13 DC |
58 | skipped = 0; |
59 | first_index = 0; | |
60 | do { | |
61 | int error = 0; | |
65d0f205 | 62 | int nr_found; |
75f3cb13 DC |
63 | xfs_inode_t *ip; |
64 | ||
65d0f205 DC |
65 | read_lock(&pag->pag_ici_lock); |
66 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
67 | (void **)&ip, first_index, 1); | |
68 | if (!nr_found) { | |
69 | read_unlock(&pag->pag_ici_lock); | |
75f3cb13 | 70 | break; |
c8e20be0 | 71 | } |
75f3cb13 | 72 | |
65d0f205 DC |
73 | /* |
74 | * Update the index for the next lookup. Catch overflows | |
75 | * into the next AG range which can occur if we have inodes | |
76 | * in the last block of the AG and we are currently | |
77 | * pointing to the last inode. | |
78 | */ | |
79 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
80 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
81 | done = 1; | |
82 | ||
c8e20be0 | 83 | /* execute releases pag->pag_ici_lock */ |
75f3cb13 DC |
84 | error = execute(ip, pag, flags); |
85 | if (error == EAGAIN) { | |
86 | skipped++; | |
87 | continue; | |
88 | } | |
89 | if (error) | |
90 | last_error = error; | |
c8e20be0 DC |
91 | |
92 | /* bail out if the filesystem is corrupted. */ | |
75f3cb13 DC |
93 | if (error == EFSCORRUPTED) |
94 | break; | |
95 | ||
65d0f205 | 96 | } while (!done); |
75f3cb13 DC |
97 | |
98 | if (skipped) { | |
99 | delay(1); | |
100 | goto restart; | |
101 | } | |
75f3cb13 DC |
102 | return last_error; |
103 | } | |
104 | ||
fe588ed3 | 105 | int |
75f3cb13 DC |
106 | xfs_inode_ag_iterator( |
107 | struct xfs_mount *mp, | |
108 | int (*execute)(struct xfs_inode *ip, | |
109 | struct xfs_perag *pag, int flags), | |
65d0f205 | 110 | int flags) |
75f3cb13 | 111 | { |
16fd5367 | 112 | struct xfs_perag *pag; |
75f3cb13 DC |
113 | int error = 0; |
114 | int last_error = 0; | |
115 | xfs_agnumber_t ag; | |
116 | ||
16fd5367 | 117 | ag = 0; |
65d0f205 DC |
118 | while ((pag = xfs_perag_get(mp, ag))) { |
119 | ag = pag->pag_agno + 1; | |
120 | error = xfs_inode_ag_walk(mp, pag, execute, flags); | |
5017e97d | 121 | xfs_perag_put(pag); |
75f3cb13 DC |
122 | if (error) { |
123 | last_error = error; | |
124 | if (error == EFSCORRUPTED) | |
125 | break; | |
126 | } | |
127 | } | |
128 | return XFS_ERROR(last_error); | |
129 | } | |
130 | ||
1da8eeca | 131 | /* must be called with pag_ici_lock held and releases it */ |
fe588ed3 | 132 | int |
1da8eeca DC |
133 | xfs_sync_inode_valid( |
134 | struct xfs_inode *ip, | |
135 | struct xfs_perag *pag) | |
136 | { | |
137 | struct inode *inode = VFS_I(ip); | |
018027be | 138 | int error = EFSCORRUPTED; |
1da8eeca DC |
139 | |
140 | /* nothing to sync during shutdown */ | |
018027be DC |
141 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
142 | goto out_unlock; | |
1da8eeca | 143 | |
018027be DC |
144 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ |
145 | error = ENOENT; | |
146 | if (xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
147 | goto out_unlock; | |
1da8eeca | 148 | |
018027be DC |
149 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
150 | if (!igrab(inode)) | |
151 | goto out_unlock; | |
152 | ||
153 | if (is_bad_inode(inode)) { | |
1da8eeca | 154 | IRELE(ip); |
018027be | 155 | goto out_unlock; |
1da8eeca DC |
156 | } |
157 | ||
018027be DC |
158 | /* inode is valid */ |
159 | error = 0; | |
160 | out_unlock: | |
161 | read_unlock(&pag->pag_ici_lock); | |
162 | return error; | |
1da8eeca DC |
163 | } |
164 | ||
5a34d5cd DC |
165 | STATIC int |
166 | xfs_sync_inode_data( | |
167 | struct xfs_inode *ip, | |
75f3cb13 | 168 | struct xfs_perag *pag, |
5a34d5cd DC |
169 | int flags) |
170 | { | |
171 | struct inode *inode = VFS_I(ip); | |
172 | struct address_space *mapping = inode->i_mapping; | |
173 | int error = 0; | |
174 | ||
75f3cb13 DC |
175 | error = xfs_sync_inode_valid(ip, pag); |
176 | if (error) | |
177 | return error; | |
178 | ||
5a34d5cd DC |
179 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
180 | goto out_wait; | |
181 | ||
182 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
183 | if (flags & SYNC_TRYLOCK) | |
184 | goto out_wait; | |
185 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
186 | } | |
187 | ||
188 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? | |
0cadda1c | 189 | 0 : XBF_ASYNC, FI_NONE); |
5a34d5cd DC |
190 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
191 | ||
192 | out_wait: | |
b0710ccc | 193 | if (flags & SYNC_WAIT) |
5a34d5cd | 194 | xfs_ioend_wait(ip); |
75f3cb13 | 195 | IRELE(ip); |
5a34d5cd DC |
196 | return error; |
197 | } | |
198 | ||
845b6d0c CH |
199 | STATIC int |
200 | xfs_sync_inode_attr( | |
201 | struct xfs_inode *ip, | |
75f3cb13 | 202 | struct xfs_perag *pag, |
845b6d0c CH |
203 | int flags) |
204 | { | |
205 | int error = 0; | |
206 | ||
75f3cb13 DC |
207 | error = xfs_sync_inode_valid(ip, pag); |
208 | if (error) | |
209 | return error; | |
210 | ||
845b6d0c CH |
211 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
212 | if (xfs_inode_clean(ip)) | |
213 | goto out_unlock; | |
214 | if (!xfs_iflock_nowait(ip)) { | |
215 | if (!(flags & SYNC_WAIT)) | |
216 | goto out_unlock; | |
217 | xfs_iflock(ip); | |
218 | } | |
219 | ||
220 | if (xfs_inode_clean(ip)) { | |
221 | xfs_ifunlock(ip); | |
222 | goto out_unlock; | |
223 | } | |
224 | ||
c854363e | 225 | error = xfs_iflush(ip, flags); |
845b6d0c CH |
226 | |
227 | out_unlock: | |
228 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
75f3cb13 | 229 | IRELE(ip); |
845b6d0c CH |
230 | return error; |
231 | } | |
232 | ||
075fe102 CH |
233 | /* |
234 | * Write out pagecache data for the whole filesystem. | |
235 | */ | |
64c86149 | 236 | STATIC int |
075fe102 CH |
237 | xfs_sync_data( |
238 | struct xfs_mount *mp, | |
239 | int flags) | |
683a8970 | 240 | { |
075fe102 | 241 | int error; |
fe4fa4b8 | 242 | |
b0710ccc | 243 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
fe4fa4b8 | 244 | |
65d0f205 | 245 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags); |
075fe102 CH |
246 | if (error) |
247 | return XFS_ERROR(error); | |
e9f1c6ee | 248 | |
a14a348b | 249 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); |
075fe102 CH |
250 | return 0; |
251 | } | |
e9f1c6ee | 252 | |
075fe102 CH |
253 | /* |
254 | * Write out inode metadata (attributes) for the whole filesystem. | |
255 | */ | |
64c86149 | 256 | STATIC int |
075fe102 CH |
257 | xfs_sync_attr( |
258 | struct xfs_mount *mp, | |
259 | int flags) | |
260 | { | |
261 | ASSERT((flags & ~SYNC_WAIT) == 0); | |
75f3cb13 | 262 | |
65d0f205 | 263 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags); |
fe4fa4b8 DC |
264 | } |
265 | ||
5d77c0dc | 266 | STATIC int |
2af75df7 | 267 | xfs_sync_fsdata( |
df308bcf | 268 | struct xfs_mount *mp) |
2af75df7 CH |
269 | { |
270 | struct xfs_buf *bp; | |
2af75df7 CH |
271 | |
272 | /* | |
df308bcf CH |
273 | * If the buffer is pinned then push on the log so we won't get stuck |
274 | * waiting in the write for someone, maybe ourselves, to flush the log. | |
275 | * | |
276 | * Even though we just pushed the log above, we did not have the | |
277 | * superblock buffer locked at that point so it can become pinned in | |
278 | * between there and here. | |
2af75df7 | 279 | */ |
df308bcf CH |
280 | bp = xfs_getsb(mp, 0); |
281 | if (XFS_BUF_ISPINNED(bp)) | |
282 | xfs_log_force(mp, 0); | |
2af75df7 | 283 | |
df308bcf | 284 | return xfs_bwrite(mp, bp); |
e9f1c6ee DC |
285 | } |
286 | ||
287 | /* | |
a4e4c4f4 DC |
288 | * When remounting a filesystem read-only or freezing the filesystem, we have |
289 | * two phases to execute. This first phase is syncing the data before we | |
290 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
291 | * we've waited for all the transactions created by the first phase to | |
292 | * complete. The second phase ensures that the inodes are written to their | |
293 | * location on disk rather than just existing in transactions in the log. This | |
294 | * means after a quiesce there is no log replay required to write the inodes to | |
295 | * disk (this is the main difference between a sync and a quiesce). | |
296 | */ | |
297 | /* | |
298 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
299 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
300 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
301 | * transactions can still occur here so don't bother flushing the buftarg |
302 | * because it'll just get dirty again. | |
e9f1c6ee DC |
303 | */ |
304 | int | |
305 | xfs_quiesce_data( | |
306 | struct xfs_mount *mp) | |
307 | { | |
df308bcf | 308 | int error, error2 = 0; |
e9f1c6ee DC |
309 | |
310 | /* push non-blocking */ | |
075fe102 | 311 | xfs_sync_data(mp, 0); |
8b5403a6 | 312 | xfs_qm_sync(mp, SYNC_TRYLOCK); |
e9f1c6ee | 313 | |
c90b07e8 | 314 | /* push and block till complete */ |
b0710ccc | 315 | xfs_sync_data(mp, SYNC_WAIT); |
7d095257 | 316 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 317 | |
a4e4c4f4 | 318 | /* write superblock and hoover up shutdown errors */ |
df308bcf CH |
319 | error = xfs_sync_fsdata(mp); |
320 | ||
321 | /* make sure all delwri buffers are written out */ | |
322 | xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
323 | ||
324 | /* mark the log as covered if needed */ | |
325 | if (xfs_log_need_covered(mp)) | |
1a387d3b | 326 | error2 = xfs_fs_log_dummy(mp, SYNC_WAIT); |
e9f1c6ee | 327 | |
a4e4c4f4 | 328 | /* flush data-only devices */ |
e9f1c6ee DC |
329 | if (mp->m_rtdev_targp) |
330 | XFS_bflush(mp->m_rtdev_targp); | |
331 | ||
df308bcf | 332 | return error ? error : error2; |
2af75df7 CH |
333 | } |
334 | ||
76bf105c DC |
335 | STATIC void |
336 | xfs_quiesce_fs( | |
337 | struct xfs_mount *mp) | |
338 | { | |
339 | int count = 0, pincount; | |
340 | ||
c854363e | 341 | xfs_reclaim_inodes(mp, 0); |
76bf105c | 342 | xfs_flush_buftarg(mp->m_ddev_targp, 0); |
76bf105c DC |
343 | |
344 | /* | |
345 | * This loop must run at least twice. The first instance of the loop | |
346 | * will flush most meta data but that will generate more meta data | |
347 | * (typically directory updates). Which then must be flushed and | |
c854363e DC |
348 | * logged before we can write the unmount record. We also so sync |
349 | * reclaim of inodes to catch any that the above delwri flush skipped. | |
76bf105c DC |
350 | */ |
351 | do { | |
c854363e | 352 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
075fe102 | 353 | xfs_sync_attr(mp, SYNC_WAIT); |
76bf105c DC |
354 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
355 | if (!pincount) { | |
356 | delay(50); | |
357 | count++; | |
358 | } | |
359 | } while (count < 2); | |
360 | } | |
361 | ||
362 | /* | |
363 | * Second stage of a quiesce. The data is already synced, now we have to take | |
364 | * care of the metadata. New transactions are already blocked, so we need to | |
365 | * wait for any remaining transactions to drain out before proceding. | |
366 | */ | |
367 | void | |
368 | xfs_quiesce_attr( | |
369 | struct xfs_mount *mp) | |
370 | { | |
371 | int error = 0; | |
372 | ||
373 | /* wait for all modifications to complete */ | |
374 | while (atomic_read(&mp->m_active_trans) > 0) | |
375 | delay(100); | |
376 | ||
377 | /* flush inodes and push all remaining buffers out to disk */ | |
378 | xfs_quiesce_fs(mp); | |
379 | ||
5e106572 FB |
380 | /* |
381 | * Just warn here till VFS can correctly support | |
382 | * read-only remount without racing. | |
383 | */ | |
384 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
385 | |
386 | /* Push the superblock and write an unmount record */ | |
387 | error = xfs_log_sbcount(mp, 1); | |
388 | if (error) | |
389 | xfs_fs_cmn_err(CE_WARN, mp, | |
390 | "xfs_attr_quiesce: failed to log sb changes. " | |
391 | "Frozen image may not be consistent."); | |
392 | xfs_log_unmount_write(mp); | |
393 | xfs_unmountfs_writesb(mp); | |
394 | } | |
395 | ||
a167b17e DC |
396 | /* |
397 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
398 | * Doing this has two advantages: | |
399 | * - It saves on stack space, which is tight in certain situations | |
400 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
401 | * Flushing while allocating in a full filesystem requires both. | |
402 | */ | |
403 | STATIC void | |
404 | xfs_syncd_queue_work( | |
405 | struct xfs_mount *mp, | |
406 | void *data, | |
e43afd72 DC |
407 | void (*syncer)(struct xfs_mount *, void *), |
408 | struct completion *completion) | |
a167b17e | 409 | { |
a8d770d9 | 410 | struct xfs_sync_work *work; |
a167b17e | 411 | |
a8d770d9 | 412 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
a167b17e DC |
413 | INIT_LIST_HEAD(&work->w_list); |
414 | work->w_syncer = syncer; | |
415 | work->w_data = data; | |
416 | work->w_mount = mp; | |
e43afd72 | 417 | work->w_completion = completion; |
a167b17e DC |
418 | spin_lock(&mp->m_sync_lock); |
419 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
420 | spin_unlock(&mp->m_sync_lock); | |
421 | wake_up_process(mp->m_sync_task); | |
422 | } | |
423 | ||
424 | /* | |
425 | * Flush delayed allocate data, attempting to free up reserved space | |
426 | * from existing allocations. At this point a new allocation attempt | |
427 | * has failed with ENOSPC and we are in the process of scratching our | |
428 | * heads, looking about for more room... | |
429 | */ | |
430 | STATIC void | |
a8d770d9 | 431 | xfs_flush_inodes_work( |
a167b17e DC |
432 | struct xfs_mount *mp, |
433 | void *arg) | |
434 | { | |
435 | struct inode *inode = arg; | |
075fe102 | 436 | xfs_sync_data(mp, SYNC_TRYLOCK); |
b0710ccc | 437 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); |
a167b17e DC |
438 | iput(inode); |
439 | } | |
440 | ||
441 | void | |
a8d770d9 | 442 | xfs_flush_inodes( |
a167b17e DC |
443 | xfs_inode_t *ip) |
444 | { | |
445 | struct inode *inode = VFS_I(ip); | |
e43afd72 | 446 | DECLARE_COMPLETION_ONSTACK(completion); |
a167b17e DC |
447 | |
448 | igrab(inode); | |
e43afd72 DC |
449 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); |
450 | wait_for_completion(&completion); | |
a14a348b | 451 | xfs_log_force(ip->i_mount, XFS_LOG_SYNC); |
a167b17e DC |
452 | } |
453 | ||
aacaa880 | 454 | /* |
df308bcf CH |
455 | * Every sync period we need to unpin all items, reclaim inodes and sync |
456 | * disk quotas. We might need to cover the log to indicate that the | |
1a387d3b | 457 | * filesystem is idle and not frozen. |
aacaa880 | 458 | */ |
a167b17e DC |
459 | STATIC void |
460 | xfs_sync_worker( | |
461 | struct xfs_mount *mp, | |
462 | void *unused) | |
463 | { | |
464 | int error; | |
465 | ||
aacaa880 | 466 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
a14a348b | 467 | xfs_log_force(mp, 0); |
c854363e | 468 | xfs_reclaim_inodes(mp, 0); |
aacaa880 | 469 | /* dgc: errors ignored here */ |
8b5403a6 | 470 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); |
1a387d3b DC |
471 | if (mp->m_super->s_frozen == SB_UNFROZEN && |
472 | xfs_log_need_covered(mp)) | |
473 | error = xfs_fs_log_dummy(mp, 0); | |
aacaa880 | 474 | } |
a167b17e DC |
475 | mp->m_sync_seq++; |
476 | wake_up(&mp->m_wait_single_sync_task); | |
477 | } | |
478 | ||
479 | STATIC int | |
480 | xfssyncd( | |
481 | void *arg) | |
482 | { | |
483 | struct xfs_mount *mp = arg; | |
484 | long timeleft; | |
a8d770d9 | 485 | xfs_sync_work_t *work, *n; |
a167b17e DC |
486 | LIST_HEAD (tmp); |
487 | ||
488 | set_freezable(); | |
489 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
490 | for (;;) { | |
20f6b2c7 DC |
491 | if (list_empty(&mp->m_sync_list)) |
492 | timeleft = schedule_timeout_interruptible(timeleft); | |
a167b17e DC |
493 | /* swsusp */ |
494 | try_to_freeze(); | |
495 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
496 | break; | |
497 | ||
498 | spin_lock(&mp->m_sync_lock); | |
499 | /* | |
500 | * We can get woken by laptop mode, to do a sync - | |
501 | * that's the (only!) case where the list would be | |
502 | * empty with time remaining. | |
503 | */ | |
504 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
505 | if (!timeleft) | |
506 | timeleft = xfs_syncd_centisecs * | |
507 | msecs_to_jiffies(10); | |
508 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
509 | list_add_tail(&mp->m_sync_work.w_list, | |
510 | &mp->m_sync_list); | |
511 | } | |
20f6b2c7 | 512 | list_splice_init(&mp->m_sync_list, &tmp); |
a167b17e DC |
513 | spin_unlock(&mp->m_sync_lock); |
514 | ||
515 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
516 | (*work->w_syncer)(mp, work->w_data); | |
517 | list_del(&work->w_list); | |
518 | if (work == &mp->m_sync_work) | |
519 | continue; | |
e43afd72 DC |
520 | if (work->w_completion) |
521 | complete(work->w_completion); | |
a167b17e DC |
522 | kmem_free(work); |
523 | } | |
524 | } | |
525 | ||
526 | return 0; | |
527 | } | |
528 | ||
529 | int | |
530 | xfs_syncd_init( | |
531 | struct xfs_mount *mp) | |
532 | { | |
533 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
534 | mp->m_sync_work.w_mount = mp; | |
e43afd72 | 535 | mp->m_sync_work.w_completion = NULL; |
e2a07812 | 536 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd/%s", mp->m_fsname); |
a167b17e DC |
537 | if (IS_ERR(mp->m_sync_task)) |
538 | return -PTR_ERR(mp->m_sync_task); | |
539 | return 0; | |
540 | } | |
541 | ||
542 | void | |
543 | xfs_syncd_stop( | |
544 | struct xfs_mount *mp) | |
545 | { | |
546 | kthread_stop(mp->m_sync_task); | |
547 | } | |
548 | ||
bc990f5c CH |
549 | void |
550 | __xfs_inode_set_reclaim_tag( | |
551 | struct xfs_perag *pag, | |
552 | struct xfs_inode *ip) | |
553 | { | |
554 | radix_tree_tag_set(&pag->pag_ici_root, | |
555 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
556 | XFS_ICI_RECLAIM_TAG); | |
16fd5367 DC |
557 | |
558 | if (!pag->pag_ici_reclaimable) { | |
559 | /* propagate the reclaim tag up into the perag radix tree */ | |
560 | spin_lock(&ip->i_mount->m_perag_lock); | |
561 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
562 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
563 | XFS_ICI_RECLAIM_TAG); | |
564 | spin_unlock(&ip->i_mount->m_perag_lock); | |
565 | trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno, | |
566 | -1, _RET_IP_); | |
567 | } | |
9bf729c0 | 568 | pag->pag_ici_reclaimable++; |
bc990f5c CH |
569 | } |
570 | ||
11654513 DC |
571 | /* |
572 | * We set the inode flag atomically with the radix tree tag. | |
573 | * Once we get tag lookups on the radix tree, this inode flag | |
574 | * can go away. | |
575 | */ | |
396beb85 DC |
576 | void |
577 | xfs_inode_set_reclaim_tag( | |
578 | xfs_inode_t *ip) | |
579 | { | |
5017e97d DC |
580 | struct xfs_mount *mp = ip->i_mount; |
581 | struct xfs_perag *pag; | |
396beb85 | 582 | |
5017e97d | 583 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
f1f724e4 | 584 | write_lock(&pag->pag_ici_lock); |
396beb85 | 585 | spin_lock(&ip->i_flags_lock); |
bc990f5c | 586 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 587 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 | 588 | spin_unlock(&ip->i_flags_lock); |
f1f724e4 | 589 | write_unlock(&pag->pag_ici_lock); |
5017e97d | 590 | xfs_perag_put(pag); |
396beb85 DC |
591 | } |
592 | ||
081003ff JW |
593 | STATIC void |
594 | __xfs_inode_clear_reclaim( | |
396beb85 DC |
595 | xfs_perag_t *pag, |
596 | xfs_inode_t *ip) | |
597 | { | |
9bf729c0 | 598 | pag->pag_ici_reclaimable--; |
16fd5367 DC |
599 | if (!pag->pag_ici_reclaimable) { |
600 | /* clear the reclaim tag from the perag radix tree */ | |
601 | spin_lock(&ip->i_mount->m_perag_lock); | |
602 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
603 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
604 | XFS_ICI_RECLAIM_TAG); | |
605 | spin_unlock(&ip->i_mount->m_perag_lock); | |
606 | trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno, | |
607 | -1, _RET_IP_); | |
608 | } | |
396beb85 DC |
609 | } |
610 | ||
081003ff JW |
611 | void |
612 | __xfs_inode_clear_reclaim_tag( | |
613 | xfs_mount_t *mp, | |
614 | xfs_perag_t *pag, | |
615 | xfs_inode_t *ip) | |
616 | { | |
617 | radix_tree_tag_clear(&pag->pag_ici_root, | |
618 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
619 | __xfs_inode_clear_reclaim(pag, ip); | |
620 | } | |
621 | ||
777df5af DC |
622 | /* |
623 | * Inodes in different states need to be treated differently, and the return | |
624 | * value of xfs_iflush is not sufficient to get this right. The following table | |
625 | * lists the inode states and the reclaim actions necessary for non-blocking | |
626 | * reclaim: | |
627 | * | |
628 | * | |
629 | * inode state iflush ret required action | |
630 | * --------------- ---------- --------------- | |
631 | * bad - reclaim | |
632 | * shutdown EIO unpin and reclaim | |
633 | * clean, unpinned 0 reclaim | |
634 | * stale, unpinned 0 reclaim | |
c854363e DC |
635 | * clean, pinned(*) 0 requeue |
636 | * stale, pinned EAGAIN requeue | |
637 | * dirty, delwri ok 0 requeue | |
638 | * dirty, delwri blocked EAGAIN requeue | |
639 | * dirty, sync flush 0 reclaim | |
777df5af DC |
640 | * |
641 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
642 | * handled anyway given the order of checks implemented. | |
643 | * | |
c854363e DC |
644 | * As can be seen from the table, the return value of xfs_iflush() is not |
645 | * sufficient to correctly decide the reclaim action here. The checks in | |
646 | * xfs_iflush() might look like duplicates, but they are not. | |
647 | * | |
648 | * Also, because we get the flush lock first, we know that any inode that has | |
649 | * been flushed delwri has had the flush completed by the time we check that | |
650 | * the inode is clean. The clean inode check needs to be done before flushing | |
651 | * the inode delwri otherwise we would loop forever requeuing clean inodes as | |
652 | * we cannot tell apart a successful delwri flush and a clean inode from the | |
653 | * return value of xfs_iflush(). | |
654 | * | |
655 | * Note that because the inode is flushed delayed write by background | |
656 | * writeback, the flush lock may already be held here and waiting on it can | |
657 | * result in very long latencies. Hence for sync reclaims, where we wait on the | |
658 | * flush lock, the caller should push out delayed write inodes first before | |
659 | * trying to reclaim them to minimise the amount of time spent waiting. For | |
660 | * background relaim, we just requeue the inode for the next pass. | |
661 | * | |
777df5af DC |
662 | * Hence the order of actions after gaining the locks should be: |
663 | * bad => reclaim | |
664 | * shutdown => unpin and reclaim | |
c854363e DC |
665 | * pinned, delwri => requeue |
666 | * pinned, sync => unpin | |
777df5af DC |
667 | * stale => reclaim |
668 | * clean => reclaim | |
c854363e DC |
669 | * dirty, delwri => flush and requeue |
670 | * dirty, sync => flush, wait and reclaim | |
777df5af | 671 | */ |
75f3cb13 | 672 | STATIC int |
c8e20be0 | 673 | xfs_reclaim_inode( |
75f3cb13 DC |
674 | struct xfs_inode *ip, |
675 | struct xfs_perag *pag, | |
c8e20be0 | 676 | int sync_mode) |
fce08f2f | 677 | { |
c854363e | 678 | int error = 0; |
777df5af | 679 | |
c8e20be0 DC |
680 | /* |
681 | * The radix tree lock here protects a thread in xfs_iget from racing | |
682 | * with us starting reclaim on the inode. Once we have the | |
683 | * XFS_IRECLAIM flag set it will not touch us. | |
684 | */ | |
685 | spin_lock(&ip->i_flags_lock); | |
686 | ASSERT_ALWAYS(__xfs_iflags_test(ip, XFS_IRECLAIMABLE)); | |
687 | if (__xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
688 | /* ignore as it is already under reclaim */ | |
689 | spin_unlock(&ip->i_flags_lock); | |
690 | write_unlock(&pag->pag_ici_lock); | |
75f3cb13 | 691 | return 0; |
fce08f2f | 692 | } |
c8e20be0 DC |
693 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
694 | spin_unlock(&ip->i_flags_lock); | |
695 | write_unlock(&pag->pag_ici_lock); | |
696 | ||
c8e20be0 | 697 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
698 | if (!xfs_iflock_nowait(ip)) { |
699 | if (!(sync_mode & SYNC_WAIT)) | |
700 | goto out; | |
701 | xfs_iflock(ip); | |
702 | } | |
7a3be02b | 703 | |
777df5af DC |
704 | if (is_bad_inode(VFS_I(ip))) |
705 | goto reclaim; | |
706 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
707 | xfs_iunpin_wait(ip); | |
708 | goto reclaim; | |
709 | } | |
c854363e DC |
710 | if (xfs_ipincount(ip)) { |
711 | if (!(sync_mode & SYNC_WAIT)) { | |
712 | xfs_ifunlock(ip); | |
713 | goto out; | |
714 | } | |
777df5af | 715 | xfs_iunpin_wait(ip); |
c854363e | 716 | } |
777df5af DC |
717 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
718 | goto reclaim; | |
719 | if (xfs_inode_clean(ip)) | |
720 | goto reclaim; | |
721 | ||
722 | /* Now we have an inode that needs flushing */ | |
723 | error = xfs_iflush(ip, sync_mode); | |
c854363e DC |
724 | if (sync_mode & SYNC_WAIT) { |
725 | xfs_iflock(ip); | |
726 | goto reclaim; | |
c8e20be0 DC |
727 | } |
728 | ||
c854363e DC |
729 | /* |
730 | * When we have to flush an inode but don't have SYNC_WAIT set, we | |
731 | * flush the inode out using a delwri buffer and wait for the next | |
732 | * call into reclaim to find it in a clean state instead of waiting for | |
733 | * it now. We also don't return errors here - if the error is transient | |
734 | * then the next reclaim pass will flush the inode, and if the error | |
f1d486a3 | 735 | * is permanent then the next sync reclaim will reclaim the inode and |
c854363e DC |
736 | * pass on the error. |
737 | */ | |
f1d486a3 | 738 | if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
c854363e DC |
739 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, |
740 | "inode 0x%llx background reclaim flush failed with %d", | |
741 | (long long)ip->i_ino, error); | |
742 | } | |
743 | out: | |
744 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
745 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
746 | /* | |
747 | * We could return EAGAIN here to make reclaim rescan the inode tree in | |
748 | * a short while. However, this just burns CPU time scanning the tree | |
749 | * waiting for IO to complete and xfssyncd never goes back to the idle | |
750 | * state. Instead, return 0 to let the next scheduled background reclaim | |
751 | * attempt to reclaim the inode again. | |
752 | */ | |
753 | return 0; | |
754 | ||
777df5af DC |
755 | reclaim: |
756 | xfs_ifunlock(ip); | |
c8e20be0 | 757 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab DC |
758 | |
759 | XFS_STATS_INC(xs_ig_reclaims); | |
760 | /* | |
761 | * Remove the inode from the per-AG radix tree. | |
762 | * | |
763 | * Because radix_tree_delete won't complain even if the item was never | |
764 | * added to the tree assert that it's been there before to catch | |
765 | * problems with the inode life time early on. | |
766 | */ | |
767 | write_lock(&pag->pag_ici_lock); | |
768 | if (!radix_tree_delete(&pag->pag_ici_root, | |
769 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino))) | |
770 | ASSERT(0); | |
081003ff | 771 | __xfs_inode_clear_reclaim(pag, ip); |
2f11feab DC |
772 | write_unlock(&pag->pag_ici_lock); |
773 | ||
774 | /* | |
775 | * Here we do an (almost) spurious inode lock in order to coordinate | |
776 | * with inode cache radix tree lookups. This is because the lookup | |
777 | * can reference the inodes in the cache without taking references. | |
778 | * | |
779 | * We make that OK here by ensuring that we wait until the inode is | |
780 | * unlocked after the lookup before we go ahead and free it. We get | |
781 | * both the ilock and the iolock because the code may need to drop the | |
782 | * ilock one but will still hold the iolock. | |
783 | */ | |
784 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
785 | xfs_qm_dqdetach(ip); | |
786 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
787 | ||
788 | xfs_inode_free(ip); | |
c854363e DC |
789 | return error; |
790 | ||
7a3be02b DC |
791 | } |
792 | ||
65d0f205 DC |
793 | /* |
794 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
795 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
796 | * then a shut down during filesystem unmount reclaim walk leak all the | |
797 | * unreclaimed inodes. | |
798 | */ | |
799 | int | |
800 | xfs_reclaim_inodes_ag( | |
801 | struct xfs_mount *mp, | |
802 | int flags, | |
803 | int *nr_to_scan) | |
804 | { | |
805 | struct xfs_perag *pag; | |
806 | int error = 0; | |
807 | int last_error = 0; | |
808 | xfs_agnumber_t ag; | |
809 | ||
810 | ag = 0; | |
811 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { | |
812 | unsigned long first_index = 0; | |
813 | int done = 0; | |
814 | ||
815 | ag = pag->pag_agno + 1; | |
816 | ||
817 | do { | |
818 | struct xfs_inode *ip; | |
819 | int nr_found; | |
820 | ||
821 | write_lock(&pag->pag_ici_lock); | |
822 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, | |
823 | (void **)&ip, first_index, 1, | |
824 | XFS_ICI_RECLAIM_TAG); | |
825 | if (!nr_found) { | |
826 | write_unlock(&pag->pag_ici_lock); | |
827 | break; | |
828 | } | |
829 | ||
830 | /* | |
831 | * Update the index for the next lookup. Catch overflows | |
832 | * into the next AG range which can occur if we have inodes | |
833 | * in the last block of the AG and we are currently | |
834 | * pointing to the last inode. | |
835 | */ | |
836 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
837 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
838 | done = 1; | |
839 | ||
840 | error = xfs_reclaim_inode(ip, pag, flags); | |
841 | if (error && last_error != EFSCORRUPTED) | |
842 | last_error = error; | |
843 | ||
844 | } while (!done && (*nr_to_scan)--); | |
845 | ||
846 | xfs_perag_put(pag); | |
847 | } | |
848 | return XFS_ERROR(last_error); | |
849 | } | |
850 | ||
7a3be02b DC |
851 | int |
852 | xfs_reclaim_inodes( | |
853 | xfs_mount_t *mp, | |
7a3be02b DC |
854 | int mode) |
855 | { | |
65d0f205 DC |
856 | int nr_to_scan = INT_MAX; |
857 | ||
858 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
859 | } |
860 | ||
861 | /* | |
862 | * Shrinker infrastructure. | |
9bf729c0 | 863 | */ |
9bf729c0 DC |
864 | static int |
865 | xfs_reclaim_inode_shrink( | |
7f8275d0 | 866 | struct shrinker *shrink, |
9bf729c0 DC |
867 | int nr_to_scan, |
868 | gfp_t gfp_mask) | |
869 | { | |
870 | struct xfs_mount *mp; | |
871 | struct xfs_perag *pag; | |
872 | xfs_agnumber_t ag; | |
16fd5367 | 873 | int reclaimable; |
9bf729c0 | 874 | |
70e60ce7 | 875 | mp = container_of(shrink, struct xfs_mount, m_inode_shrink); |
9bf729c0 DC |
876 | if (nr_to_scan) { |
877 | if (!(gfp_mask & __GFP_FS)) | |
878 | return -1; | |
879 | ||
65d0f205 DC |
880 | xfs_reclaim_inodes_ag(mp, 0, &nr_to_scan); |
881 | /* terminate if we don't exhaust the scan */ | |
70e60ce7 DC |
882 | if (nr_to_scan > 0) |
883 | return -1; | |
884 | } | |
9bf729c0 | 885 | |
16fd5367 DC |
886 | reclaimable = 0; |
887 | ag = 0; | |
65d0f205 DC |
888 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
889 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
890 | reclaimable += pag->pag_ici_reclaimable; |
891 | xfs_perag_put(pag); | |
9bf729c0 | 892 | } |
9bf729c0 DC |
893 | return reclaimable; |
894 | } | |
895 | ||
9bf729c0 DC |
896 | void |
897 | xfs_inode_shrinker_register( | |
898 | struct xfs_mount *mp) | |
899 | { | |
70e60ce7 DC |
900 | mp->m_inode_shrink.shrink = xfs_reclaim_inode_shrink; |
901 | mp->m_inode_shrink.seeks = DEFAULT_SEEKS; | |
902 | register_shrinker(&mp->m_inode_shrink); | |
9bf729c0 DC |
903 | } |
904 | ||
905 | void | |
906 | xfs_inode_shrinker_unregister( | |
907 | struct xfs_mount *mp) | |
908 | { | |
70e60ce7 | 909 | unregister_shrinker(&mp->m_inode_shrink); |
fce08f2f | 910 | } |