]>
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" | |
6ca1c906 | 20 | #include "xfs_format.h" |
fe4fa4b8 | 21 | #include "xfs_types.h" |
fe4fa4b8 | 22 | #include "xfs_log.h" |
f661f1e0 | 23 | #include "xfs_log_priv.h" |
fe4fa4b8 DC |
24 | #include "xfs_inum.h" |
25 | #include "xfs_trans.h" | |
fd074841 | 26 | #include "xfs_trans_priv.h" |
fe4fa4b8 DC |
27 | #include "xfs_sb.h" |
28 | #include "xfs_ag.h" | |
fe4fa4b8 DC |
29 | #include "xfs_mount.h" |
30 | #include "xfs_bmap_btree.h" | |
fe4fa4b8 DC |
31 | #include "xfs_inode.h" |
32 | #include "xfs_dinode.h" | |
33 | #include "xfs_error.h" | |
fe4fa4b8 | 34 | #include "xfs_filestream.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" |
6d8b79cf | 39 | #include "xfs_icache.h" |
c24b5dfa | 40 | #include "xfs_bmap_util.h" |
fe4fa4b8 | 41 | |
a167b17e DC |
42 | #include <linux/kthread.h> |
43 | #include <linux/freezer.h> | |
44 | ||
33479e05 DC |
45 | STATIC void __xfs_inode_clear_reclaim_tag(struct xfs_mount *mp, |
46 | struct xfs_perag *pag, struct xfs_inode *ip); | |
47 | ||
48 | /* | |
49 | * Allocate and initialise an xfs_inode. | |
50 | */ | |
51 | STATIC struct xfs_inode * | |
52 | xfs_inode_alloc( | |
53 | struct xfs_mount *mp, | |
54 | xfs_ino_t ino) | |
55 | { | |
56 | struct xfs_inode *ip; | |
57 | ||
58 | /* | |
59 | * if this didn't occur in transactions, we could use | |
60 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
61 | * code up to do this anyway. | |
62 | */ | |
63 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
64 | if (!ip) | |
65 | return NULL; | |
66 | if (inode_init_always(mp->m_super, VFS_I(ip))) { | |
67 | kmem_zone_free(xfs_inode_zone, ip); | |
68 | return NULL; | |
69 | } | |
70 | ||
71 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
72 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
73 | ASSERT(!xfs_isiflocked(ip)); | |
74 | ASSERT(ip->i_ino == 0); | |
75 | ||
76 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
77 | ||
78 | /* initialise the xfs inode */ | |
79 | ip->i_ino = ino; | |
80 | ip->i_mount = mp; | |
81 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
82 | ip->i_afp = NULL; | |
83 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
84 | ip->i_flags = 0; | |
85 | ip->i_delayed_blks = 0; | |
86 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
87 | ||
88 | return ip; | |
89 | } | |
90 | ||
91 | STATIC void | |
92 | xfs_inode_free_callback( | |
93 | struct rcu_head *head) | |
94 | { | |
95 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
96 | struct xfs_inode *ip = XFS_I(inode); | |
97 | ||
98 | kmem_zone_free(xfs_inode_zone, ip); | |
99 | } | |
100 | ||
101 | STATIC void | |
102 | xfs_inode_free( | |
103 | struct xfs_inode *ip) | |
104 | { | |
105 | switch (ip->i_d.di_mode & S_IFMT) { | |
106 | case S_IFREG: | |
107 | case S_IFDIR: | |
108 | case S_IFLNK: | |
109 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
110 | break; | |
111 | } | |
112 | ||
113 | if (ip->i_afp) | |
114 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
115 | ||
116 | if (ip->i_itemp) { | |
117 | ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL)); | |
118 | xfs_inode_item_destroy(ip); | |
119 | ip->i_itemp = NULL; | |
120 | } | |
121 | ||
122 | /* asserts to verify all state is correct here */ | |
123 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
124 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
125 | ASSERT(!xfs_isiflocked(ip)); | |
126 | ||
127 | /* | |
128 | * Because we use RCU freeing we need to ensure the inode always | |
129 | * appears to be reclaimed with an invalid inode number when in the | |
130 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
131 | * races. | |
132 | */ | |
133 | spin_lock(&ip->i_flags_lock); | |
134 | ip->i_flags = XFS_IRECLAIM; | |
135 | ip->i_ino = 0; | |
136 | spin_unlock(&ip->i_flags_lock); | |
137 | ||
138 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
139 | } | |
140 | ||
141 | /* | |
142 | * Check the validity of the inode we just found it the cache | |
143 | */ | |
144 | static int | |
145 | xfs_iget_cache_hit( | |
146 | struct xfs_perag *pag, | |
147 | struct xfs_inode *ip, | |
148 | xfs_ino_t ino, | |
149 | int flags, | |
150 | int lock_flags) __releases(RCU) | |
151 | { | |
152 | struct inode *inode = VFS_I(ip); | |
153 | struct xfs_mount *mp = ip->i_mount; | |
154 | int error; | |
155 | ||
156 | /* | |
157 | * check for re-use of an inode within an RCU grace period due to the | |
158 | * radix tree nodes not being updated yet. We monitor for this by | |
159 | * setting the inode number to zero before freeing the inode structure. | |
160 | * If the inode has been reallocated and set up, then the inode number | |
161 | * will not match, so check for that, too. | |
162 | */ | |
163 | spin_lock(&ip->i_flags_lock); | |
164 | if (ip->i_ino != ino) { | |
165 | trace_xfs_iget_skip(ip); | |
166 | XFS_STATS_INC(xs_ig_frecycle); | |
167 | error = EAGAIN; | |
168 | goto out_error; | |
169 | } | |
170 | ||
171 | ||
172 | /* | |
173 | * If we are racing with another cache hit that is currently | |
174 | * instantiating this inode or currently recycling it out of | |
175 | * reclaimabe state, wait for the initialisation to complete | |
176 | * before continuing. | |
177 | * | |
178 | * XXX(hch): eventually we should do something equivalent to | |
179 | * wait_on_inode to wait for these flags to be cleared | |
180 | * instead of polling for it. | |
181 | */ | |
182 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { | |
183 | trace_xfs_iget_skip(ip); | |
184 | XFS_STATS_INC(xs_ig_frecycle); | |
185 | error = EAGAIN; | |
186 | goto out_error; | |
187 | } | |
188 | ||
189 | /* | |
190 | * If lookup is racing with unlink return an error immediately. | |
191 | */ | |
192 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { | |
193 | error = ENOENT; | |
194 | goto out_error; | |
195 | } | |
196 | ||
197 | /* | |
198 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
199 | * Need to carefully get it back into useable state. | |
200 | */ | |
201 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
202 | trace_xfs_iget_reclaim(ip); | |
203 | ||
204 | /* | |
205 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode | |
206 | * from stomping over us while we recycle the inode. We can't | |
207 | * clear the radix tree reclaimable tag yet as it requires | |
208 | * pag_ici_lock to be held exclusive. | |
209 | */ | |
210 | ip->i_flags |= XFS_IRECLAIM; | |
211 | ||
212 | spin_unlock(&ip->i_flags_lock); | |
213 | rcu_read_unlock(); | |
214 | ||
215 | error = -inode_init_always(mp->m_super, inode); | |
216 | if (error) { | |
217 | /* | |
218 | * Re-initializing the inode failed, and we are in deep | |
219 | * trouble. Try to re-add it to the reclaim list. | |
220 | */ | |
221 | rcu_read_lock(); | |
222 | spin_lock(&ip->i_flags_lock); | |
223 | ||
224 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); | |
225 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); | |
226 | trace_xfs_iget_reclaim_fail(ip); | |
227 | goto out_error; | |
228 | } | |
229 | ||
230 | spin_lock(&pag->pag_ici_lock); | |
231 | spin_lock(&ip->i_flags_lock); | |
232 | ||
233 | /* | |
234 | * Clear the per-lifetime state in the inode as we are now | |
235 | * effectively a new inode and need to return to the initial | |
236 | * state before reuse occurs. | |
237 | */ | |
238 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
239 | ip->i_flags |= XFS_INEW; | |
240 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
241 | inode->i_state = I_NEW; | |
242 | ||
243 | ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock)); | |
244 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
245 | ||
246 | spin_unlock(&ip->i_flags_lock); | |
247 | spin_unlock(&pag->pag_ici_lock); | |
248 | } else { | |
249 | /* If the VFS inode is being torn down, pause and try again. */ | |
250 | if (!igrab(inode)) { | |
251 | trace_xfs_iget_skip(ip); | |
252 | error = EAGAIN; | |
253 | goto out_error; | |
254 | } | |
255 | ||
256 | /* We've got a live one. */ | |
257 | spin_unlock(&ip->i_flags_lock); | |
258 | rcu_read_unlock(); | |
259 | trace_xfs_iget_hit(ip); | |
260 | } | |
261 | ||
262 | if (lock_flags != 0) | |
263 | xfs_ilock(ip, lock_flags); | |
264 | ||
265 | xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE); | |
266 | XFS_STATS_INC(xs_ig_found); | |
267 | ||
268 | return 0; | |
269 | ||
270 | out_error: | |
271 | spin_unlock(&ip->i_flags_lock); | |
272 | rcu_read_unlock(); | |
273 | return error; | |
274 | } | |
275 | ||
276 | ||
277 | static int | |
278 | xfs_iget_cache_miss( | |
279 | struct xfs_mount *mp, | |
280 | struct xfs_perag *pag, | |
281 | xfs_trans_t *tp, | |
282 | xfs_ino_t ino, | |
283 | struct xfs_inode **ipp, | |
284 | int flags, | |
285 | int lock_flags) | |
286 | { | |
287 | struct xfs_inode *ip; | |
288 | int error; | |
289 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
290 | int iflags; | |
291 | ||
292 | ip = xfs_inode_alloc(mp, ino); | |
293 | if (!ip) | |
294 | return ENOMEM; | |
295 | ||
296 | error = xfs_iread(mp, tp, ip, flags); | |
297 | if (error) | |
298 | goto out_destroy; | |
299 | ||
300 | trace_xfs_iget_miss(ip); | |
301 | ||
302 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { | |
303 | error = ENOENT; | |
304 | goto out_destroy; | |
305 | } | |
306 | ||
307 | /* | |
308 | * Preload the radix tree so we can insert safely under the | |
309 | * write spinlock. Note that we cannot sleep inside the preload | |
310 | * region. Since we can be called from transaction context, don't | |
311 | * recurse into the file system. | |
312 | */ | |
313 | if (radix_tree_preload(GFP_NOFS)) { | |
314 | error = EAGAIN; | |
315 | goto out_destroy; | |
316 | } | |
317 | ||
318 | /* | |
319 | * Because the inode hasn't been added to the radix-tree yet it can't | |
320 | * be found by another thread, so we can do the non-sleeping lock here. | |
321 | */ | |
322 | if (lock_flags) { | |
323 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
324 | BUG(); | |
325 | } | |
326 | ||
327 | /* | |
328 | * These values must be set before inserting the inode into the radix | |
329 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
330 | * RCU locking mechanism) can find it and that lookup must see that this | |
331 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
332 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
333 | * memory barrier that ensures this detection works correctly at lookup | |
334 | * time. | |
335 | */ | |
336 | iflags = XFS_INEW; | |
337 | if (flags & XFS_IGET_DONTCACHE) | |
338 | iflags |= XFS_IDONTCACHE; | |
113a5683 CS |
339 | ip->i_udquot = NULL; |
340 | ip->i_gdquot = NULL; | |
92f8ff73 | 341 | ip->i_pdquot = NULL; |
33479e05 DC |
342 | xfs_iflags_set(ip, iflags); |
343 | ||
344 | /* insert the new inode */ | |
345 | spin_lock(&pag->pag_ici_lock); | |
346 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
347 | if (unlikely(error)) { | |
348 | WARN_ON(error != -EEXIST); | |
349 | XFS_STATS_INC(xs_ig_dup); | |
350 | error = EAGAIN; | |
351 | goto out_preload_end; | |
352 | } | |
353 | spin_unlock(&pag->pag_ici_lock); | |
354 | radix_tree_preload_end(); | |
355 | ||
356 | *ipp = ip; | |
357 | return 0; | |
358 | ||
359 | out_preload_end: | |
360 | spin_unlock(&pag->pag_ici_lock); | |
361 | radix_tree_preload_end(); | |
362 | if (lock_flags) | |
363 | xfs_iunlock(ip, lock_flags); | |
364 | out_destroy: | |
365 | __destroy_inode(VFS_I(ip)); | |
366 | xfs_inode_free(ip); | |
367 | return error; | |
368 | } | |
369 | ||
370 | /* | |
371 | * Look up an inode by number in the given file system. | |
372 | * The inode is looked up in the cache held in each AG. | |
373 | * If the inode is found in the cache, initialise the vfs inode | |
374 | * if necessary. | |
375 | * | |
376 | * If it is not in core, read it in from the file system's device, | |
377 | * add it to the cache and initialise the vfs inode. | |
378 | * | |
379 | * The inode is locked according to the value of the lock_flags parameter. | |
380 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
381 | * should be taken. | |
382 | * | |
383 | * mp -- the mount point structure for the current file system. It points | |
384 | * to the inode hash table. | |
385 | * tp -- a pointer to the current transaction if there is one. This is | |
386 | * simply passed through to the xfs_iread() call. | |
387 | * ino -- the number of the inode desired. This is the unique identifier | |
388 | * within the file system for the inode being requested. | |
389 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
390 | * for xfs_ilock() for a list of valid values. | |
391 | */ | |
392 | int | |
393 | xfs_iget( | |
394 | xfs_mount_t *mp, | |
395 | xfs_trans_t *tp, | |
396 | xfs_ino_t ino, | |
397 | uint flags, | |
398 | uint lock_flags, | |
399 | xfs_inode_t **ipp) | |
400 | { | |
401 | xfs_inode_t *ip; | |
402 | int error; | |
403 | xfs_perag_t *pag; | |
404 | xfs_agino_t agino; | |
405 | ||
406 | /* | |
407 | * xfs_reclaim_inode() uses the ILOCK to ensure an inode | |
408 | * doesn't get freed while it's being referenced during a | |
409 | * radix tree traversal here. It assumes this function | |
410 | * aqcuires only the ILOCK (and therefore it has no need to | |
411 | * involve the IOLOCK in this synchronization). | |
412 | */ | |
413 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); | |
414 | ||
415 | /* reject inode numbers outside existing AGs */ | |
416 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
417 | return EINVAL; | |
418 | ||
419 | /* get the perag structure and ensure that it's inode capable */ | |
420 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
421 | agino = XFS_INO_TO_AGINO(mp, ino); | |
422 | ||
423 | again: | |
424 | error = 0; | |
425 | rcu_read_lock(); | |
426 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
427 | ||
428 | if (ip) { | |
429 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
430 | if (error) | |
431 | goto out_error_or_again; | |
432 | } else { | |
433 | rcu_read_unlock(); | |
434 | XFS_STATS_INC(xs_ig_missed); | |
435 | ||
436 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
437 | flags, lock_flags); | |
438 | if (error) | |
439 | goto out_error_or_again; | |
440 | } | |
441 | xfs_perag_put(pag); | |
442 | ||
443 | *ipp = ip; | |
444 | ||
445 | /* | |
446 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
447 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
448 | */ | |
449 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) | |
450 | xfs_setup_inode(ip); | |
451 | return 0; | |
452 | ||
453 | out_error_or_again: | |
454 | if (error == EAGAIN) { | |
455 | delay(1); | |
456 | goto again; | |
457 | } | |
458 | xfs_perag_put(pag); | |
459 | return error; | |
460 | } | |
461 | ||
78ae5256 DC |
462 | /* |
463 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
464 | * radix tree lookups to a minimum. The batch size is a trade off between | |
465 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
466 | * be too greedy. | |
467 | */ | |
468 | #define XFS_LOOKUP_BATCH 32 | |
469 | ||
e13de955 DC |
470 | STATIC int |
471 | xfs_inode_ag_walk_grab( | |
472 | struct xfs_inode *ip) | |
473 | { | |
474 | struct inode *inode = VFS_I(ip); | |
475 | ||
1a3e8f3d DC |
476 | ASSERT(rcu_read_lock_held()); |
477 | ||
478 | /* | |
479 | * check for stale RCU freed inode | |
480 | * | |
481 | * If the inode has been reallocated, it doesn't matter if it's not in | |
482 | * the AG we are walking - we are walking for writeback, so if it | |
483 | * passes all the "valid inode" checks and is dirty, then we'll write | |
484 | * it back anyway. If it has been reallocated and still being | |
485 | * initialised, the XFS_INEW check below will catch it. | |
486 | */ | |
487 | spin_lock(&ip->i_flags_lock); | |
488 | if (!ip->i_ino) | |
489 | goto out_unlock_noent; | |
490 | ||
491 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
492 | if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
493 | goto out_unlock_noent; | |
494 | spin_unlock(&ip->i_flags_lock); | |
495 | ||
e13de955 DC |
496 | /* nothing to sync during shutdown */ |
497 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
498 | return EFSCORRUPTED; | |
499 | ||
e13de955 DC |
500 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
501 | if (!igrab(inode)) | |
502 | return ENOENT; | |
503 | ||
504 | if (is_bad_inode(inode)) { | |
505 | IRELE(ip); | |
506 | return ENOENT; | |
507 | } | |
508 | ||
509 | /* inode is valid */ | |
510 | return 0; | |
1a3e8f3d DC |
511 | |
512 | out_unlock_noent: | |
513 | spin_unlock(&ip->i_flags_lock); | |
514 | return ENOENT; | |
e13de955 DC |
515 | } |
516 | ||
75f3cb13 DC |
517 | STATIC int |
518 | xfs_inode_ag_walk( | |
519 | struct xfs_mount *mp, | |
5017e97d | 520 | struct xfs_perag *pag, |
75f3cb13 | 521 | int (*execute)(struct xfs_inode *ip, |
a454f742 BF |
522 | struct xfs_perag *pag, int flags, |
523 | void *args), | |
524 | int flags, | |
525 | void *args, | |
526 | int tag) | |
75f3cb13 | 527 | { |
75f3cb13 DC |
528 | uint32_t first_index; |
529 | int last_error = 0; | |
530 | int skipped; | |
65d0f205 | 531 | int done; |
78ae5256 | 532 | int nr_found; |
75f3cb13 DC |
533 | |
534 | restart: | |
65d0f205 | 535 | done = 0; |
75f3cb13 DC |
536 | skipped = 0; |
537 | first_index = 0; | |
78ae5256 | 538 | nr_found = 0; |
75f3cb13 | 539 | do { |
78ae5256 | 540 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
75f3cb13 | 541 | int error = 0; |
78ae5256 | 542 | int i; |
75f3cb13 | 543 | |
1a3e8f3d | 544 | rcu_read_lock(); |
a454f742 BF |
545 | |
546 | if (tag == -1) | |
547 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
78ae5256 DC |
548 | (void **)batch, first_index, |
549 | XFS_LOOKUP_BATCH); | |
a454f742 BF |
550 | else |
551 | nr_found = radix_tree_gang_lookup_tag( | |
552 | &pag->pag_ici_root, | |
553 | (void **) batch, first_index, | |
554 | XFS_LOOKUP_BATCH, tag); | |
555 | ||
65d0f205 | 556 | if (!nr_found) { |
1a3e8f3d | 557 | rcu_read_unlock(); |
75f3cb13 | 558 | break; |
c8e20be0 | 559 | } |
75f3cb13 | 560 | |
65d0f205 | 561 | /* |
78ae5256 DC |
562 | * Grab the inodes before we drop the lock. if we found |
563 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 564 | */ |
78ae5256 DC |
565 | for (i = 0; i < nr_found; i++) { |
566 | struct xfs_inode *ip = batch[i]; | |
567 | ||
568 | if (done || xfs_inode_ag_walk_grab(ip)) | |
569 | batch[i] = NULL; | |
570 | ||
571 | /* | |
1a3e8f3d DC |
572 | * Update the index for the next lookup. Catch |
573 | * overflows into the next AG range which can occur if | |
574 | * we have inodes in the last block of the AG and we | |
575 | * are currently pointing to the last inode. | |
576 | * | |
577 | * Because we may see inodes that are from the wrong AG | |
578 | * due to RCU freeing and reallocation, only update the | |
579 | * index if it lies in this AG. It was a race that lead | |
580 | * us to see this inode, so another lookup from the | |
581 | * same index will not find it again. | |
78ae5256 | 582 | */ |
1a3e8f3d DC |
583 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
584 | continue; | |
78ae5256 DC |
585 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
586 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
587 | done = 1; | |
e13de955 | 588 | } |
78ae5256 DC |
589 | |
590 | /* unlock now we've grabbed the inodes. */ | |
1a3e8f3d | 591 | rcu_read_unlock(); |
e13de955 | 592 | |
78ae5256 DC |
593 | for (i = 0; i < nr_found; i++) { |
594 | if (!batch[i]) | |
595 | continue; | |
a454f742 | 596 | error = execute(batch[i], pag, flags, args); |
78ae5256 DC |
597 | IRELE(batch[i]); |
598 | if (error == EAGAIN) { | |
599 | skipped++; | |
600 | continue; | |
601 | } | |
602 | if (error && last_error != EFSCORRUPTED) | |
603 | last_error = error; | |
75f3cb13 | 604 | } |
c8e20be0 DC |
605 | |
606 | /* bail out if the filesystem is corrupted. */ | |
75f3cb13 DC |
607 | if (error == EFSCORRUPTED) |
608 | break; | |
609 | ||
8daaa831 DC |
610 | cond_resched(); |
611 | ||
78ae5256 | 612 | } while (nr_found && !done); |
75f3cb13 DC |
613 | |
614 | if (skipped) { | |
615 | delay(1); | |
616 | goto restart; | |
617 | } | |
75f3cb13 DC |
618 | return last_error; |
619 | } | |
620 | ||
579b62fa BF |
621 | /* |
622 | * Background scanning to trim post-EOF preallocated space. This is queued | |
b9fe5052 | 623 | * based on the 'speculative_prealloc_lifetime' tunable (5m by default). |
579b62fa BF |
624 | */ |
625 | STATIC void | |
626 | xfs_queue_eofblocks( | |
627 | struct xfs_mount *mp) | |
628 | { | |
629 | rcu_read_lock(); | |
630 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG)) | |
631 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
632 | &mp->m_eofblocks_work, | |
633 | msecs_to_jiffies(xfs_eofb_secs * 1000)); | |
634 | rcu_read_unlock(); | |
635 | } | |
636 | ||
637 | void | |
638 | xfs_eofblocks_worker( | |
639 | struct work_struct *work) | |
640 | { | |
641 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
642 | struct xfs_mount, m_eofblocks_work); | |
643 | xfs_icache_free_eofblocks(mp, NULL); | |
644 | xfs_queue_eofblocks(mp); | |
645 | } | |
646 | ||
fe588ed3 | 647 | int |
75f3cb13 DC |
648 | xfs_inode_ag_iterator( |
649 | struct xfs_mount *mp, | |
650 | int (*execute)(struct xfs_inode *ip, | |
a454f742 BF |
651 | struct xfs_perag *pag, int flags, |
652 | void *args), | |
653 | int flags, | |
654 | void *args) | |
75f3cb13 | 655 | { |
16fd5367 | 656 | struct xfs_perag *pag; |
75f3cb13 DC |
657 | int error = 0; |
658 | int last_error = 0; | |
659 | xfs_agnumber_t ag; | |
660 | ||
16fd5367 | 661 | ag = 0; |
65d0f205 DC |
662 | while ((pag = xfs_perag_get(mp, ag))) { |
663 | ag = pag->pag_agno + 1; | |
a454f742 BF |
664 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1); |
665 | xfs_perag_put(pag); | |
666 | if (error) { | |
667 | last_error = error; | |
668 | if (error == EFSCORRUPTED) | |
669 | break; | |
670 | } | |
671 | } | |
672 | return XFS_ERROR(last_error); | |
673 | } | |
674 | ||
675 | int | |
676 | xfs_inode_ag_iterator_tag( | |
677 | struct xfs_mount *mp, | |
678 | int (*execute)(struct xfs_inode *ip, | |
679 | struct xfs_perag *pag, int flags, | |
680 | void *args), | |
681 | int flags, | |
682 | void *args, | |
683 | int tag) | |
684 | { | |
685 | struct xfs_perag *pag; | |
686 | int error = 0; | |
687 | int last_error = 0; | |
688 | xfs_agnumber_t ag; | |
689 | ||
690 | ag = 0; | |
691 | while ((pag = xfs_perag_get_tag(mp, ag, tag))) { | |
692 | ag = pag->pag_agno + 1; | |
693 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag); | |
5017e97d | 694 | xfs_perag_put(pag); |
75f3cb13 DC |
695 | if (error) { |
696 | last_error = error; | |
697 | if (error == EFSCORRUPTED) | |
698 | break; | |
699 | } | |
700 | } | |
701 | return XFS_ERROR(last_error); | |
702 | } | |
703 | ||
a7b339f1 DC |
704 | /* |
705 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
706 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
5889608d | 707 | * on the xfs periodic sync default of 30s. Perhaps this should have it's own |
a7b339f1 DC |
708 | * tunable, but that can be done if this method proves to be ineffective or too |
709 | * aggressive. | |
710 | */ | |
711 | static void | |
5889608d | 712 | xfs_reclaim_work_queue( |
a7b339f1 | 713 | struct xfs_mount *mp) |
a167b17e | 714 | { |
a167b17e | 715 | |
a7b339f1 DC |
716 | rcu_read_lock(); |
717 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
5889608d | 718 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, |
a7b339f1 | 719 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); |
a167b17e | 720 | } |
a7b339f1 DC |
721 | rcu_read_unlock(); |
722 | } | |
a167b17e | 723 | |
a7b339f1 DC |
724 | /* |
725 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
726 | * many inodes as possible in a short period of time. It kicks itself every few | |
727 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
728 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
729 | * already being flushed, and once done schedules a future pass. | |
730 | */ | |
33c7a2bc | 731 | void |
a7b339f1 DC |
732 | xfs_reclaim_worker( |
733 | struct work_struct *work) | |
734 | { | |
735 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
736 | struct xfs_mount, m_reclaim_work); | |
737 | ||
738 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
5889608d | 739 | xfs_reclaim_work_queue(mp); |
a7b339f1 DC |
740 | } |
741 | ||
33479e05 | 742 | static void |
bc990f5c CH |
743 | __xfs_inode_set_reclaim_tag( |
744 | struct xfs_perag *pag, | |
745 | struct xfs_inode *ip) | |
746 | { | |
747 | radix_tree_tag_set(&pag->pag_ici_root, | |
748 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
749 | XFS_ICI_RECLAIM_TAG); | |
16fd5367 DC |
750 | |
751 | if (!pag->pag_ici_reclaimable) { | |
752 | /* propagate the reclaim tag up into the perag radix tree */ | |
753 | spin_lock(&ip->i_mount->m_perag_lock); | |
754 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
755 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
756 | XFS_ICI_RECLAIM_TAG); | |
757 | spin_unlock(&ip->i_mount->m_perag_lock); | |
a7b339f1 DC |
758 | |
759 | /* schedule periodic background inode reclaim */ | |
5889608d | 760 | xfs_reclaim_work_queue(ip->i_mount); |
a7b339f1 | 761 | |
16fd5367 DC |
762 | trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno, |
763 | -1, _RET_IP_); | |
764 | } | |
9bf729c0 | 765 | pag->pag_ici_reclaimable++; |
bc990f5c CH |
766 | } |
767 | ||
11654513 DC |
768 | /* |
769 | * We set the inode flag atomically with the radix tree tag. | |
770 | * Once we get tag lookups on the radix tree, this inode flag | |
771 | * can go away. | |
772 | */ | |
396beb85 DC |
773 | void |
774 | xfs_inode_set_reclaim_tag( | |
775 | xfs_inode_t *ip) | |
776 | { | |
5017e97d DC |
777 | struct xfs_mount *mp = ip->i_mount; |
778 | struct xfs_perag *pag; | |
396beb85 | 779 | |
5017e97d | 780 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1a427ab0 | 781 | spin_lock(&pag->pag_ici_lock); |
396beb85 | 782 | spin_lock(&ip->i_flags_lock); |
bc990f5c | 783 | __xfs_inode_set_reclaim_tag(pag, ip); |
11654513 | 784 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 | 785 | spin_unlock(&ip->i_flags_lock); |
1a427ab0 | 786 | spin_unlock(&pag->pag_ici_lock); |
5017e97d | 787 | xfs_perag_put(pag); |
396beb85 DC |
788 | } |
789 | ||
081003ff JW |
790 | STATIC void |
791 | __xfs_inode_clear_reclaim( | |
396beb85 DC |
792 | xfs_perag_t *pag, |
793 | xfs_inode_t *ip) | |
794 | { | |
9bf729c0 | 795 | pag->pag_ici_reclaimable--; |
16fd5367 DC |
796 | if (!pag->pag_ici_reclaimable) { |
797 | /* clear the reclaim tag from the perag radix tree */ | |
798 | spin_lock(&ip->i_mount->m_perag_lock); | |
799 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
800 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
801 | XFS_ICI_RECLAIM_TAG); | |
802 | spin_unlock(&ip->i_mount->m_perag_lock); | |
803 | trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno, | |
804 | -1, _RET_IP_); | |
805 | } | |
396beb85 DC |
806 | } |
807 | ||
33479e05 | 808 | STATIC void |
081003ff JW |
809 | __xfs_inode_clear_reclaim_tag( |
810 | xfs_mount_t *mp, | |
811 | xfs_perag_t *pag, | |
812 | xfs_inode_t *ip) | |
813 | { | |
814 | radix_tree_tag_clear(&pag->pag_ici_root, | |
815 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
816 | __xfs_inode_clear_reclaim(pag, ip); | |
817 | } | |
818 | ||
e3a20c0b DC |
819 | /* |
820 | * Grab the inode for reclaim exclusively. | |
821 | * Return 0 if we grabbed it, non-zero otherwise. | |
822 | */ | |
823 | STATIC int | |
824 | xfs_reclaim_inode_grab( | |
825 | struct xfs_inode *ip, | |
826 | int flags) | |
827 | { | |
1a3e8f3d DC |
828 | ASSERT(rcu_read_lock_held()); |
829 | ||
830 | /* quick check for stale RCU freed inode */ | |
831 | if (!ip->i_ino) | |
832 | return 1; | |
e3a20c0b DC |
833 | |
834 | /* | |
474fce06 CH |
835 | * If we are asked for non-blocking operation, do unlocked checks to |
836 | * see if the inode already is being flushed or in reclaim to avoid | |
837 | * lock traffic. | |
e3a20c0b DC |
838 | */ |
839 | if ((flags & SYNC_TRYLOCK) && | |
474fce06 | 840 | __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM)) |
e3a20c0b | 841 | return 1; |
e3a20c0b DC |
842 | |
843 | /* | |
844 | * The radix tree lock here protects a thread in xfs_iget from racing | |
845 | * with us starting reclaim on the inode. Once we have the | |
846 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
847 | * |
848 | * Due to RCU lookup, we may find inodes that have been freed and only | |
849 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
850 | * aren't candidates for reclaim at all, so we must check the | |
851 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
852 | */ |
853 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
854 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
855 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
856 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
857 | spin_unlock(&ip->i_flags_lock); |
858 | return 1; | |
859 | } | |
860 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
861 | spin_unlock(&ip->i_flags_lock); | |
862 | return 0; | |
863 | } | |
864 | ||
777df5af | 865 | /* |
8a48088f CH |
866 | * Inodes in different states need to be treated differently. The following |
867 | * table lists the inode states and the reclaim actions necessary: | |
777df5af DC |
868 | * |
869 | * inode state iflush ret required action | |
870 | * --------------- ---------- --------------- | |
871 | * bad - reclaim | |
872 | * shutdown EIO unpin and reclaim | |
873 | * clean, unpinned 0 reclaim | |
874 | * stale, unpinned 0 reclaim | |
c854363e DC |
875 | * clean, pinned(*) 0 requeue |
876 | * stale, pinned EAGAIN requeue | |
8a48088f CH |
877 | * dirty, async - requeue |
878 | * dirty, sync 0 reclaim | |
777df5af DC |
879 | * |
880 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
881 | * handled anyway given the order of checks implemented. | |
882 | * | |
c854363e DC |
883 | * Also, because we get the flush lock first, we know that any inode that has |
884 | * been flushed delwri has had the flush completed by the time we check that | |
8a48088f | 885 | * the inode is clean. |
c854363e | 886 | * |
8a48088f CH |
887 | * Note that because the inode is flushed delayed write by AIL pushing, the |
888 | * flush lock may already be held here and waiting on it can result in very | |
889 | * long latencies. Hence for sync reclaims, where we wait on the flush lock, | |
890 | * the caller should push the AIL first before trying to reclaim inodes to | |
891 | * minimise the amount of time spent waiting. For background relaim, we only | |
892 | * bother to reclaim clean inodes anyway. | |
c854363e | 893 | * |
777df5af DC |
894 | * Hence the order of actions after gaining the locks should be: |
895 | * bad => reclaim | |
896 | * shutdown => unpin and reclaim | |
8a48088f | 897 | * pinned, async => requeue |
c854363e | 898 | * pinned, sync => unpin |
777df5af DC |
899 | * stale => reclaim |
900 | * clean => reclaim | |
8a48088f | 901 | * dirty, async => requeue |
c854363e | 902 | * dirty, sync => flush, wait and reclaim |
777df5af | 903 | */ |
75f3cb13 | 904 | STATIC int |
c8e20be0 | 905 | xfs_reclaim_inode( |
75f3cb13 DC |
906 | struct xfs_inode *ip, |
907 | struct xfs_perag *pag, | |
c8e20be0 | 908 | int sync_mode) |
fce08f2f | 909 | { |
4c46819a CH |
910 | struct xfs_buf *bp = NULL; |
911 | int error; | |
777df5af | 912 | |
1bfd8d04 DC |
913 | restart: |
914 | error = 0; | |
c8e20be0 | 915 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
916 | if (!xfs_iflock_nowait(ip)) { |
917 | if (!(sync_mode & SYNC_WAIT)) | |
918 | goto out; | |
919 | xfs_iflock(ip); | |
920 | } | |
7a3be02b | 921 | |
777df5af DC |
922 | if (is_bad_inode(VFS_I(ip))) |
923 | goto reclaim; | |
924 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
925 | xfs_iunpin_wait(ip); | |
04913fdd | 926 | xfs_iflush_abort(ip, false); |
777df5af DC |
927 | goto reclaim; |
928 | } | |
c854363e | 929 | if (xfs_ipincount(ip)) { |
8a48088f CH |
930 | if (!(sync_mode & SYNC_WAIT)) |
931 | goto out_ifunlock; | |
777df5af | 932 | xfs_iunpin_wait(ip); |
c854363e | 933 | } |
777df5af DC |
934 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
935 | goto reclaim; | |
936 | if (xfs_inode_clean(ip)) | |
937 | goto reclaim; | |
938 | ||
8a48088f CH |
939 | /* |
940 | * Never flush out dirty data during non-blocking reclaim, as it would | |
941 | * just contend with AIL pushing trying to do the same job. | |
942 | */ | |
943 | if (!(sync_mode & SYNC_WAIT)) | |
944 | goto out_ifunlock; | |
945 | ||
1bfd8d04 DC |
946 | /* |
947 | * Now we have an inode that needs flushing. | |
948 | * | |
4c46819a | 949 | * Note that xfs_iflush will never block on the inode buffer lock, as |
1bfd8d04 | 950 | * xfs_ifree_cluster() can lock the inode buffer before it locks the |
4c46819a | 951 | * ip->i_lock, and we are doing the exact opposite here. As a result, |
475ee413 CH |
952 | * doing a blocking xfs_imap_to_bp() to get the cluster buffer would |
953 | * result in an ABBA deadlock with xfs_ifree_cluster(). | |
1bfd8d04 DC |
954 | * |
955 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
956 | * cache to mark them stale, if we hit this case we don't actually want | |
957 | * to do IO here - we want the inode marked stale so we can simply | |
4c46819a CH |
958 | * reclaim it. Hence if we get an EAGAIN error here, just unlock the |
959 | * inode, back off and try again. Hopefully the next pass through will | |
960 | * see the stale flag set on the inode. | |
1bfd8d04 | 961 | */ |
4c46819a | 962 | error = xfs_iflush(ip, &bp); |
8a48088f CH |
963 | if (error == EAGAIN) { |
964 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
965 | /* backoff longer than in xfs_ifree_cluster */ | |
966 | delay(2); | |
967 | goto restart; | |
c854363e | 968 | } |
c854363e | 969 | |
4c46819a CH |
970 | if (!error) { |
971 | error = xfs_bwrite(bp); | |
972 | xfs_buf_relse(bp); | |
973 | } | |
974 | ||
975 | xfs_iflock(ip); | |
777df5af DC |
976 | reclaim: |
977 | xfs_ifunlock(ip); | |
c8e20be0 | 978 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab DC |
979 | |
980 | XFS_STATS_INC(xs_ig_reclaims); | |
981 | /* | |
982 | * Remove the inode from the per-AG radix tree. | |
983 | * | |
984 | * Because radix_tree_delete won't complain even if the item was never | |
985 | * added to the tree assert that it's been there before to catch | |
986 | * problems with the inode life time early on. | |
987 | */ | |
1a427ab0 | 988 | spin_lock(&pag->pag_ici_lock); |
2f11feab DC |
989 | if (!radix_tree_delete(&pag->pag_ici_root, |
990 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino))) | |
991 | ASSERT(0); | |
081003ff | 992 | __xfs_inode_clear_reclaim(pag, ip); |
1a427ab0 | 993 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
994 | |
995 | /* | |
996 | * Here we do an (almost) spurious inode lock in order to coordinate | |
997 | * with inode cache radix tree lookups. This is because the lookup | |
998 | * can reference the inodes in the cache without taking references. | |
999 | * | |
1000 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 1001 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 1002 | */ |
ad637a10 | 1003 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1004 | xfs_qm_dqdetach(ip); |
ad637a10 | 1005 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab DC |
1006 | |
1007 | xfs_inode_free(ip); | |
ad637a10 | 1008 | return error; |
8a48088f CH |
1009 | |
1010 | out_ifunlock: | |
1011 | xfs_ifunlock(ip); | |
1012 | out: | |
1013 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
1014 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1015 | /* | |
1016 | * We could return EAGAIN here to make reclaim rescan the inode tree in | |
1017 | * a short while. However, this just burns CPU time scanning the tree | |
5889608d DC |
1018 | * waiting for IO to complete and the reclaim work never goes back to |
1019 | * the idle state. Instead, return 0 to let the next scheduled | |
1020 | * background reclaim attempt to reclaim the inode again. | |
8a48088f CH |
1021 | */ |
1022 | return 0; | |
7a3be02b DC |
1023 | } |
1024 | ||
65d0f205 DC |
1025 | /* |
1026 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
1027 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
1028 | * then a shut down during filesystem unmount reclaim walk leak all the | |
1029 | * unreclaimed inodes. | |
1030 | */ | |
33479e05 | 1031 | STATIC int |
65d0f205 DC |
1032 | xfs_reclaim_inodes_ag( |
1033 | struct xfs_mount *mp, | |
1034 | int flags, | |
1035 | int *nr_to_scan) | |
1036 | { | |
1037 | struct xfs_perag *pag; | |
1038 | int error = 0; | |
1039 | int last_error = 0; | |
1040 | xfs_agnumber_t ag; | |
69b491c2 DC |
1041 | int trylock = flags & SYNC_TRYLOCK; |
1042 | int skipped; | |
65d0f205 | 1043 | |
69b491c2 | 1044 | restart: |
65d0f205 | 1045 | ag = 0; |
69b491c2 | 1046 | skipped = 0; |
65d0f205 DC |
1047 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1048 | unsigned long first_index = 0; | |
1049 | int done = 0; | |
e3a20c0b | 1050 | int nr_found = 0; |
65d0f205 DC |
1051 | |
1052 | ag = pag->pag_agno + 1; | |
1053 | ||
69b491c2 DC |
1054 | if (trylock) { |
1055 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
1056 | skipped++; | |
f83282a8 | 1057 | xfs_perag_put(pag); |
69b491c2 DC |
1058 | continue; |
1059 | } | |
1060 | first_index = pag->pag_ici_reclaim_cursor; | |
1061 | } else | |
1062 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
1063 | ||
65d0f205 | 1064 | do { |
e3a20c0b DC |
1065 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
1066 | int i; | |
65d0f205 | 1067 | |
1a3e8f3d | 1068 | rcu_read_lock(); |
e3a20c0b DC |
1069 | nr_found = radix_tree_gang_lookup_tag( |
1070 | &pag->pag_ici_root, | |
1071 | (void **)batch, first_index, | |
1072 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
1073 | XFS_ICI_RECLAIM_TAG); |
1074 | if (!nr_found) { | |
b2232219 | 1075 | done = 1; |
1a3e8f3d | 1076 | rcu_read_unlock(); |
65d0f205 DC |
1077 | break; |
1078 | } | |
1079 | ||
1080 | /* | |
e3a20c0b DC |
1081 | * Grab the inodes before we drop the lock. if we found |
1082 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 1083 | */ |
e3a20c0b DC |
1084 | for (i = 0; i < nr_found; i++) { |
1085 | struct xfs_inode *ip = batch[i]; | |
1086 | ||
1087 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
1088 | batch[i] = NULL; | |
1089 | ||
1090 | /* | |
1091 | * Update the index for the next lookup. Catch | |
1092 | * overflows into the next AG range which can | |
1093 | * occur if we have inodes in the last block of | |
1094 | * the AG and we are currently pointing to the | |
1095 | * last inode. | |
1a3e8f3d DC |
1096 | * |
1097 | * Because we may see inodes that are from the | |
1098 | * wrong AG due to RCU freeing and | |
1099 | * reallocation, only update the index if it | |
1100 | * lies in this AG. It was a race that lead us | |
1101 | * to see this inode, so another lookup from | |
1102 | * the same index will not find it again. | |
e3a20c0b | 1103 | */ |
1a3e8f3d DC |
1104 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
1105 | pag->pag_agno) | |
1106 | continue; | |
e3a20c0b DC |
1107 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
1108 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1109 | done = 1; | |
1110 | } | |
65d0f205 | 1111 | |
e3a20c0b | 1112 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 1113 | rcu_read_unlock(); |
e3a20c0b DC |
1114 | |
1115 | for (i = 0; i < nr_found; i++) { | |
1116 | if (!batch[i]) | |
1117 | continue; | |
1118 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
1119 | if (error && last_error != EFSCORRUPTED) | |
1120 | last_error = error; | |
1121 | } | |
1122 | ||
1123 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 1124 | |
8daaa831 DC |
1125 | cond_resched(); |
1126 | ||
e3a20c0b | 1127 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 1128 | |
69b491c2 DC |
1129 | if (trylock && !done) |
1130 | pag->pag_ici_reclaim_cursor = first_index; | |
1131 | else | |
1132 | pag->pag_ici_reclaim_cursor = 0; | |
1133 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
1134 | xfs_perag_put(pag); |
1135 | } | |
69b491c2 DC |
1136 | |
1137 | /* | |
1138 | * if we skipped any AG, and we still have scan count remaining, do | |
1139 | * another pass this time using blocking reclaim semantics (i.e | |
1140 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1141 | * ensure that when we get more reclaimers than AGs we block rather | |
1142 | * than spin trying to execute reclaim. | |
1143 | */ | |
8daaa831 | 1144 | if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) { |
69b491c2 DC |
1145 | trylock = 0; |
1146 | goto restart; | |
1147 | } | |
65d0f205 DC |
1148 | return XFS_ERROR(last_error); |
1149 | } | |
1150 | ||
7a3be02b DC |
1151 | int |
1152 | xfs_reclaim_inodes( | |
1153 | xfs_mount_t *mp, | |
7a3be02b DC |
1154 | int mode) |
1155 | { | |
65d0f205 DC |
1156 | int nr_to_scan = INT_MAX; |
1157 | ||
1158 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1159 | } |
1160 | ||
1161 | /* | |
8daaa831 | 1162 | * Scan a certain number of inodes for reclaim. |
a7b339f1 DC |
1163 | * |
1164 | * When called we make sure that there is a background (fast) inode reclaim in | |
8daaa831 | 1165 | * progress, while we will throttle the speed of reclaim via doing synchronous |
a7b339f1 DC |
1166 | * reclaim of inodes. That means if we come across dirty inodes, we wait for |
1167 | * them to be cleaned, which we hope will not be very long due to the | |
1168 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1169 | */ |
8daaa831 DC |
1170 | void |
1171 | xfs_reclaim_inodes_nr( | |
1172 | struct xfs_mount *mp, | |
1173 | int nr_to_scan) | |
9bf729c0 | 1174 | { |
8daaa831 | 1175 | /* kick background reclaimer and push the AIL */ |
5889608d | 1176 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1177 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1178 | |
8daaa831 DC |
1179 | xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan); |
1180 | } | |
9bf729c0 | 1181 | |
8daaa831 DC |
1182 | /* |
1183 | * Return the number of reclaimable inodes in the filesystem for | |
1184 | * the shrinker to determine how much to reclaim. | |
1185 | */ | |
1186 | int | |
1187 | xfs_reclaim_inodes_count( | |
1188 | struct xfs_mount *mp) | |
1189 | { | |
1190 | struct xfs_perag *pag; | |
1191 | xfs_agnumber_t ag = 0; | |
1192 | int reclaimable = 0; | |
9bf729c0 | 1193 | |
65d0f205 DC |
1194 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1195 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1196 | reclaimable += pag->pag_ici_reclaimable; |
1197 | xfs_perag_put(pag); | |
9bf729c0 | 1198 | } |
9bf729c0 DC |
1199 | return reclaimable; |
1200 | } | |
1201 | ||
3e3f9f58 BF |
1202 | STATIC int |
1203 | xfs_inode_match_id( | |
1204 | struct xfs_inode *ip, | |
1205 | struct xfs_eofblocks *eofb) | |
1206 | { | |
b9fe5052 DE |
1207 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && |
1208 | !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
1b556048 | 1209 | return 0; |
3e3f9f58 | 1210 | |
b9fe5052 DE |
1211 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && |
1212 | !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
1b556048 BF |
1213 | return 0; |
1214 | ||
b9fe5052 | 1215 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && |
1b556048 BF |
1216 | xfs_get_projid(ip) != eofb->eof_prid) |
1217 | return 0; | |
1218 | ||
1219 | return 1; | |
3e3f9f58 BF |
1220 | } |
1221 | ||
41176a68 BF |
1222 | STATIC int |
1223 | xfs_inode_free_eofblocks( | |
1224 | struct xfs_inode *ip, | |
1225 | struct xfs_perag *pag, | |
1226 | int flags, | |
1227 | void *args) | |
1228 | { | |
1229 | int ret; | |
3e3f9f58 | 1230 | struct xfs_eofblocks *eofb = args; |
41176a68 BF |
1231 | |
1232 | if (!xfs_can_free_eofblocks(ip, false)) { | |
1233 | /* inode could be preallocated or append-only */ | |
1234 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1235 | xfs_inode_clear_eofblocks_tag(ip); | |
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * If the mapping is dirty the operation can block and wait for some | |
1241 | * time. Unless we are waiting, skip it. | |
1242 | */ | |
1243 | if (!(flags & SYNC_WAIT) && | |
1244 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) | |
1245 | return 0; | |
1246 | ||
00ca79a0 BF |
1247 | if (eofb) { |
1248 | if (!xfs_inode_match_id(ip, eofb)) | |
1249 | return 0; | |
1250 | ||
1251 | /* skip the inode if the file size is too small */ | |
1252 | if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE && | |
1253 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1254 | return 0; | |
1255 | } | |
3e3f9f58 | 1256 | |
41176a68 BF |
1257 | ret = xfs_free_eofblocks(ip->i_mount, ip, true); |
1258 | ||
1259 | /* don't revisit the inode if we're not waiting */ | |
1260 | if (ret == EAGAIN && !(flags & SYNC_WAIT)) | |
1261 | ret = 0; | |
1262 | ||
1263 | return ret; | |
1264 | } | |
1265 | ||
1266 | int | |
1267 | xfs_icache_free_eofblocks( | |
1268 | struct xfs_mount *mp, | |
8ca149de | 1269 | struct xfs_eofblocks *eofb) |
41176a68 | 1270 | { |
8ca149de BF |
1271 | int flags = SYNC_TRYLOCK; |
1272 | ||
1273 | if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC)) | |
1274 | flags = SYNC_WAIT; | |
1275 | ||
41176a68 | 1276 | return xfs_inode_ag_iterator_tag(mp, xfs_inode_free_eofblocks, flags, |
8ca149de | 1277 | eofb, XFS_ICI_EOFBLOCKS_TAG); |
41176a68 BF |
1278 | } |
1279 | ||
27b52867 BF |
1280 | void |
1281 | xfs_inode_set_eofblocks_tag( | |
1282 | xfs_inode_t *ip) | |
1283 | { | |
1284 | struct xfs_mount *mp = ip->i_mount; | |
1285 | struct xfs_perag *pag; | |
1286 | int tagged; | |
1287 | ||
1288 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
1289 | spin_lock(&pag->pag_ici_lock); | |
1290 | trace_xfs_inode_set_eofblocks_tag(ip); | |
1291 | ||
1292 | tagged = radix_tree_tagged(&pag->pag_ici_root, | |
1293 | XFS_ICI_EOFBLOCKS_TAG); | |
1294 | radix_tree_tag_set(&pag->pag_ici_root, | |
1295 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
1296 | XFS_ICI_EOFBLOCKS_TAG); | |
1297 | if (!tagged) { | |
1298 | /* propagate the eofblocks tag up into the perag radix tree */ | |
1299 | spin_lock(&ip->i_mount->m_perag_lock); | |
1300 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
1301 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
1302 | XFS_ICI_EOFBLOCKS_TAG); | |
1303 | spin_unlock(&ip->i_mount->m_perag_lock); | |
579b62fa BF |
1304 | |
1305 | /* kick off background trimming */ | |
1306 | xfs_queue_eofblocks(ip->i_mount); | |
27b52867 BF |
1307 | |
1308 | trace_xfs_perag_set_eofblocks(ip->i_mount, pag->pag_agno, | |
1309 | -1, _RET_IP_); | |
1310 | } | |
1311 | ||
1312 | spin_unlock(&pag->pag_ici_lock); | |
1313 | xfs_perag_put(pag); | |
1314 | } | |
1315 | ||
1316 | void | |
1317 | xfs_inode_clear_eofblocks_tag( | |
1318 | xfs_inode_t *ip) | |
1319 | { | |
1320 | struct xfs_mount *mp = ip->i_mount; | |
1321 | struct xfs_perag *pag; | |
1322 | ||
1323 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
1324 | spin_lock(&pag->pag_ici_lock); | |
1325 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
1326 | ||
1327 | radix_tree_tag_clear(&pag->pag_ici_root, | |
1328 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
1329 | XFS_ICI_EOFBLOCKS_TAG); | |
1330 | if (!radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_EOFBLOCKS_TAG)) { | |
1331 | /* clear the eofblocks tag from the perag radix tree */ | |
1332 | spin_lock(&ip->i_mount->m_perag_lock); | |
1333 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
1334 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
1335 | XFS_ICI_EOFBLOCKS_TAG); | |
1336 | spin_unlock(&ip->i_mount->m_perag_lock); | |
1337 | trace_xfs_perag_clear_eofblocks(ip->i_mount, pag->pag_agno, | |
1338 | -1, _RET_IP_); | |
1339 | } | |
1340 | ||
1341 | spin_unlock(&pag->pag_ici_lock); | |
1342 | xfs_perag_put(pag); | |
1343 | } | |
1344 |