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" |
239880ef DC |
21 | #include "xfs_log_format.h" |
22 | #include "xfs_trans_resv.h" | |
fe4fa4b8 | 23 | #include "xfs_sb.h" |
fe4fa4b8 | 24 | #include "xfs_mount.h" |
fe4fa4b8 | 25 | #include "xfs_inode.h" |
fe4fa4b8 | 26 | #include "xfs_error.h" |
239880ef DC |
27 | #include "xfs_trans.h" |
28 | #include "xfs_trans_priv.h" | |
fe4fa4b8 | 29 | #include "xfs_inode_item.h" |
7d095257 | 30 | #include "xfs_quota.h" |
0b1b213f | 31 | #include "xfs_trace.h" |
6d8b79cf | 32 | #include "xfs_icache.h" |
c24b5dfa | 33 | #include "xfs_bmap_util.h" |
dc06f398 BF |
34 | #include "xfs_dquot_item.h" |
35 | #include "xfs_dquot.h" | |
fe4fa4b8 | 36 | |
a167b17e DC |
37 | #include <linux/kthread.h> |
38 | #include <linux/freezer.h> | |
39 | ||
33479e05 DC |
40 | /* |
41 | * Allocate and initialise an xfs_inode. | |
42 | */ | |
638f4416 | 43 | struct xfs_inode * |
33479e05 DC |
44 | xfs_inode_alloc( |
45 | struct xfs_mount *mp, | |
46 | xfs_ino_t ino) | |
47 | { | |
48 | struct xfs_inode *ip; | |
49 | ||
50 | /* | |
51 | * if this didn't occur in transactions, we could use | |
52 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
53 | * code up to do this anyway. | |
54 | */ | |
55 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
56 | if (!ip) | |
57 | return NULL; | |
58 | if (inode_init_always(mp->m_super, VFS_I(ip))) { | |
59 | kmem_zone_free(xfs_inode_zone, ip); | |
60 | return NULL; | |
61 | } | |
62 | ||
c19b3b05 DC |
63 | /* VFS doesn't initialise i_mode! */ |
64 | VFS_I(ip)->i_mode = 0; | |
65 | ||
ff6d6af2 | 66 | XFS_STATS_INC(mp, vn_active); |
33479e05 DC |
67 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
68 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
69 | ASSERT(!xfs_isiflocked(ip)); | |
70 | ASSERT(ip->i_ino == 0); | |
71 | ||
72 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
73 | ||
74 | /* initialise the xfs inode */ | |
75 | ip->i_ino = ino; | |
76 | ip->i_mount = mp; | |
77 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
78 | ip->i_afp = NULL; | |
79 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
80 | ip->i_flags = 0; | |
81 | ip->i_delayed_blks = 0; | |
f8d55aa0 | 82 | memset(&ip->i_d, 0, sizeof(ip->i_d)); |
33479e05 DC |
83 | |
84 | return ip; | |
85 | } | |
86 | ||
87 | STATIC void | |
88 | xfs_inode_free_callback( | |
89 | struct rcu_head *head) | |
90 | { | |
91 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
92 | struct xfs_inode *ip = XFS_I(inode); | |
93 | ||
c19b3b05 | 94 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
95 | case S_IFREG: |
96 | case S_IFDIR: | |
97 | case S_IFLNK: | |
98 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
99 | break; | |
100 | } | |
101 | ||
102 | if (ip->i_afp) | |
103 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
104 | ||
105 | if (ip->i_itemp) { | |
106 | ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL)); | |
107 | xfs_inode_item_destroy(ip); | |
108 | ip->i_itemp = NULL; | |
109 | } | |
110 | ||
1f2dcfe8 DC |
111 | kmem_zone_free(xfs_inode_zone, ip); |
112 | } | |
113 | ||
8a17d7dd DC |
114 | static void |
115 | __xfs_inode_free( | |
116 | struct xfs_inode *ip) | |
117 | { | |
118 | /* asserts to verify all state is correct here */ | |
119 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
120 | ASSERT(!xfs_isiflocked(ip)); | |
121 | XFS_STATS_DEC(ip->i_mount, vn_active); | |
122 | ||
123 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
124 | } | |
125 | ||
1f2dcfe8 DC |
126 | void |
127 | xfs_inode_free( | |
128 | struct xfs_inode *ip) | |
129 | { | |
33479e05 DC |
130 | /* |
131 | * Because we use RCU freeing we need to ensure the inode always | |
132 | * appears to be reclaimed with an invalid inode number when in the | |
133 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
134 | * races. | |
135 | */ | |
136 | spin_lock(&ip->i_flags_lock); | |
137 | ip->i_flags = XFS_IRECLAIM; | |
138 | ip->i_ino = 0; | |
139 | spin_unlock(&ip->i_flags_lock); | |
140 | ||
8a17d7dd | 141 | __xfs_inode_free(ip); |
33479e05 DC |
142 | } |
143 | ||
ad438c40 DC |
144 | /* |
145 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
146 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
147 | * on the xfs periodic sync default of 30s. Perhaps this should have it's own | |
148 | * tunable, but that can be done if this method proves to be ineffective or too | |
149 | * aggressive. | |
150 | */ | |
151 | static void | |
152 | xfs_reclaim_work_queue( | |
153 | struct xfs_mount *mp) | |
154 | { | |
155 | ||
156 | rcu_read_lock(); | |
157 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
158 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
159 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
160 | } | |
161 | rcu_read_unlock(); | |
162 | } | |
163 | ||
164 | /* | |
165 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
166 | * many inodes as possible in a short period of time. It kicks itself every few | |
167 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
168 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
169 | * already being flushed, and once done schedules a future pass. | |
170 | */ | |
171 | void | |
172 | xfs_reclaim_worker( | |
173 | struct work_struct *work) | |
174 | { | |
175 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
176 | struct xfs_mount, m_reclaim_work); | |
177 | ||
178 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
179 | xfs_reclaim_work_queue(mp); | |
180 | } | |
181 | ||
182 | static void | |
183 | xfs_perag_set_reclaim_tag( | |
184 | struct xfs_perag *pag) | |
185 | { | |
186 | struct xfs_mount *mp = pag->pag_mount; | |
187 | ||
188 | ASSERT(spin_is_locked(&pag->pag_ici_lock)); | |
189 | if (pag->pag_ici_reclaimable++) | |
190 | return; | |
191 | ||
192 | /* propagate the reclaim tag up into the perag radix tree */ | |
193 | spin_lock(&mp->m_perag_lock); | |
194 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, | |
195 | XFS_ICI_RECLAIM_TAG); | |
196 | spin_unlock(&mp->m_perag_lock); | |
197 | ||
198 | /* schedule periodic background inode reclaim */ | |
199 | xfs_reclaim_work_queue(mp); | |
200 | ||
201 | trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
202 | } | |
203 | ||
204 | static void | |
205 | xfs_perag_clear_reclaim_tag( | |
206 | struct xfs_perag *pag) | |
207 | { | |
208 | struct xfs_mount *mp = pag->pag_mount; | |
209 | ||
210 | ASSERT(spin_is_locked(&pag->pag_ici_lock)); | |
211 | if (--pag->pag_ici_reclaimable) | |
212 | return; | |
213 | ||
214 | /* clear the reclaim tag from the perag radix tree */ | |
215 | spin_lock(&mp->m_perag_lock); | |
216 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, | |
217 | XFS_ICI_RECLAIM_TAG); | |
218 | spin_unlock(&mp->m_perag_lock); | |
219 | trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
220 | } | |
221 | ||
222 | ||
223 | /* | |
224 | * We set the inode flag atomically with the radix tree tag. | |
225 | * Once we get tag lookups on the radix tree, this inode flag | |
226 | * can go away. | |
227 | */ | |
228 | void | |
229 | xfs_inode_set_reclaim_tag( | |
230 | struct xfs_inode *ip) | |
231 | { | |
232 | struct xfs_mount *mp = ip->i_mount; | |
233 | struct xfs_perag *pag; | |
234 | ||
235 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
236 | spin_lock(&pag->pag_ici_lock); | |
237 | spin_lock(&ip->i_flags_lock); | |
238 | ||
239 | radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
240 | XFS_ICI_RECLAIM_TAG); | |
241 | xfs_perag_set_reclaim_tag(pag); | |
242 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
243 | ||
244 | spin_unlock(&ip->i_flags_lock); | |
245 | spin_unlock(&pag->pag_ici_lock); | |
246 | xfs_perag_put(pag); | |
247 | } | |
248 | ||
249 | STATIC void | |
250 | xfs_inode_clear_reclaim_tag( | |
251 | struct xfs_perag *pag, | |
252 | xfs_ino_t ino) | |
253 | { | |
254 | radix_tree_tag_clear(&pag->pag_ici_root, | |
255 | XFS_INO_TO_AGINO(pag->pag_mount, ino), | |
256 | XFS_ICI_RECLAIM_TAG); | |
257 | xfs_perag_clear_reclaim_tag(pag); | |
258 | } | |
259 | ||
50997470 DC |
260 | /* |
261 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
262 | * part of the structure. This is made more complex by the fact we store | |
263 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 264 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 265 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 266 | * after reinitialisation even if it fails. |
50997470 DC |
267 | */ |
268 | static int | |
269 | xfs_reinit_inode( | |
270 | struct xfs_mount *mp, | |
271 | struct inode *inode) | |
272 | { | |
273 | int error; | |
54d7b5c1 | 274 | uint32_t nlink = inode->i_nlink; |
9e9a2674 | 275 | uint32_t generation = inode->i_generation; |
83e06f21 | 276 | uint64_t version = inode->i_version; |
c19b3b05 | 277 | umode_t mode = inode->i_mode; |
50997470 DC |
278 | |
279 | error = inode_init_always(mp->m_super, inode); | |
280 | ||
54d7b5c1 | 281 | set_nlink(inode, nlink); |
9e9a2674 | 282 | inode->i_generation = generation; |
83e06f21 | 283 | inode->i_version = version; |
c19b3b05 | 284 | inode->i_mode = mode; |
50997470 DC |
285 | return error; |
286 | } | |
287 | ||
33479e05 DC |
288 | /* |
289 | * Check the validity of the inode we just found it the cache | |
290 | */ | |
291 | static int | |
292 | xfs_iget_cache_hit( | |
293 | struct xfs_perag *pag, | |
294 | struct xfs_inode *ip, | |
295 | xfs_ino_t ino, | |
296 | int flags, | |
297 | int lock_flags) __releases(RCU) | |
298 | { | |
299 | struct inode *inode = VFS_I(ip); | |
300 | struct xfs_mount *mp = ip->i_mount; | |
301 | int error; | |
302 | ||
303 | /* | |
304 | * check for re-use of an inode within an RCU grace period due to the | |
305 | * radix tree nodes not being updated yet. We monitor for this by | |
306 | * setting the inode number to zero before freeing the inode structure. | |
307 | * If the inode has been reallocated and set up, then the inode number | |
308 | * will not match, so check for that, too. | |
309 | */ | |
310 | spin_lock(&ip->i_flags_lock); | |
311 | if (ip->i_ino != ino) { | |
312 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 313 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 314 | error = -EAGAIN; |
33479e05 DC |
315 | goto out_error; |
316 | } | |
317 | ||
318 | ||
319 | /* | |
320 | * If we are racing with another cache hit that is currently | |
321 | * instantiating this inode or currently recycling it out of | |
322 | * reclaimabe state, wait for the initialisation to complete | |
323 | * before continuing. | |
324 | * | |
325 | * XXX(hch): eventually we should do something equivalent to | |
326 | * wait_on_inode to wait for these flags to be cleared | |
327 | * instead of polling for it. | |
328 | */ | |
329 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { | |
330 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 331 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 332 | error = -EAGAIN; |
33479e05 DC |
333 | goto out_error; |
334 | } | |
335 | ||
336 | /* | |
337 | * If lookup is racing with unlink return an error immediately. | |
338 | */ | |
c19b3b05 | 339 | if (VFS_I(ip)->i_mode == 0 && !(flags & XFS_IGET_CREATE)) { |
2451337d | 340 | error = -ENOENT; |
33479e05 DC |
341 | goto out_error; |
342 | } | |
343 | ||
344 | /* | |
345 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
346 | * Need to carefully get it back into useable state. | |
347 | */ | |
348 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
349 | trace_xfs_iget_reclaim(ip); | |
350 | ||
351 | /* | |
352 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode | |
353 | * from stomping over us while we recycle the inode. We can't | |
354 | * clear the radix tree reclaimable tag yet as it requires | |
355 | * pag_ici_lock to be held exclusive. | |
356 | */ | |
357 | ip->i_flags |= XFS_IRECLAIM; | |
358 | ||
359 | spin_unlock(&ip->i_flags_lock); | |
360 | rcu_read_unlock(); | |
361 | ||
50997470 | 362 | error = xfs_reinit_inode(mp, inode); |
33479e05 DC |
363 | if (error) { |
364 | /* | |
365 | * Re-initializing the inode failed, and we are in deep | |
366 | * trouble. Try to re-add it to the reclaim list. | |
367 | */ | |
368 | rcu_read_lock(); | |
369 | spin_lock(&ip->i_flags_lock); | |
370 | ||
371 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); | |
372 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); | |
373 | trace_xfs_iget_reclaim_fail(ip); | |
374 | goto out_error; | |
375 | } | |
376 | ||
377 | spin_lock(&pag->pag_ici_lock); | |
378 | spin_lock(&ip->i_flags_lock); | |
379 | ||
380 | /* | |
381 | * Clear the per-lifetime state in the inode as we are now | |
382 | * effectively a new inode and need to return to the initial | |
383 | * state before reuse occurs. | |
384 | */ | |
385 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
386 | ip->i_flags |= XFS_INEW; | |
545c0889 | 387 | xfs_inode_clear_reclaim_tag(pag, ip->i_ino); |
33479e05 DC |
388 | inode->i_state = I_NEW; |
389 | ||
390 | ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock)); | |
391 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
392 | ||
393 | spin_unlock(&ip->i_flags_lock); | |
394 | spin_unlock(&pag->pag_ici_lock); | |
395 | } else { | |
396 | /* If the VFS inode is being torn down, pause and try again. */ | |
397 | if (!igrab(inode)) { | |
398 | trace_xfs_iget_skip(ip); | |
2451337d | 399 | error = -EAGAIN; |
33479e05 DC |
400 | goto out_error; |
401 | } | |
402 | ||
403 | /* We've got a live one. */ | |
404 | spin_unlock(&ip->i_flags_lock); | |
405 | rcu_read_unlock(); | |
406 | trace_xfs_iget_hit(ip); | |
407 | } | |
408 | ||
409 | if (lock_flags != 0) | |
410 | xfs_ilock(ip, lock_flags); | |
411 | ||
412 | xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE); | |
ff6d6af2 | 413 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
414 | |
415 | return 0; | |
416 | ||
417 | out_error: | |
418 | spin_unlock(&ip->i_flags_lock); | |
419 | rcu_read_unlock(); | |
420 | return error; | |
421 | } | |
422 | ||
423 | ||
424 | static int | |
425 | xfs_iget_cache_miss( | |
426 | struct xfs_mount *mp, | |
427 | struct xfs_perag *pag, | |
428 | xfs_trans_t *tp, | |
429 | xfs_ino_t ino, | |
430 | struct xfs_inode **ipp, | |
431 | int flags, | |
432 | int lock_flags) | |
433 | { | |
434 | struct xfs_inode *ip; | |
435 | int error; | |
436 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
437 | int iflags; | |
438 | ||
439 | ip = xfs_inode_alloc(mp, ino); | |
440 | if (!ip) | |
2451337d | 441 | return -ENOMEM; |
33479e05 DC |
442 | |
443 | error = xfs_iread(mp, tp, ip, flags); | |
444 | if (error) | |
445 | goto out_destroy; | |
446 | ||
447 | trace_xfs_iget_miss(ip); | |
448 | ||
c19b3b05 | 449 | if ((VFS_I(ip)->i_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
2451337d | 450 | error = -ENOENT; |
33479e05 DC |
451 | goto out_destroy; |
452 | } | |
453 | ||
454 | /* | |
455 | * Preload the radix tree so we can insert safely under the | |
456 | * write spinlock. Note that we cannot sleep inside the preload | |
457 | * region. Since we can be called from transaction context, don't | |
458 | * recurse into the file system. | |
459 | */ | |
460 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 461 | error = -EAGAIN; |
33479e05 DC |
462 | goto out_destroy; |
463 | } | |
464 | ||
465 | /* | |
466 | * Because the inode hasn't been added to the radix-tree yet it can't | |
467 | * be found by another thread, so we can do the non-sleeping lock here. | |
468 | */ | |
469 | if (lock_flags) { | |
470 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
471 | BUG(); | |
472 | } | |
473 | ||
474 | /* | |
475 | * These values must be set before inserting the inode into the radix | |
476 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
477 | * RCU locking mechanism) can find it and that lookup must see that this | |
478 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
479 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
480 | * memory barrier that ensures this detection works correctly at lookup | |
481 | * time. | |
482 | */ | |
483 | iflags = XFS_INEW; | |
484 | if (flags & XFS_IGET_DONTCACHE) | |
485 | iflags |= XFS_IDONTCACHE; | |
113a5683 CS |
486 | ip->i_udquot = NULL; |
487 | ip->i_gdquot = NULL; | |
92f8ff73 | 488 | ip->i_pdquot = NULL; |
33479e05 DC |
489 | xfs_iflags_set(ip, iflags); |
490 | ||
491 | /* insert the new inode */ | |
492 | spin_lock(&pag->pag_ici_lock); | |
493 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
494 | if (unlikely(error)) { | |
495 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 496 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 497 | error = -EAGAIN; |
33479e05 DC |
498 | goto out_preload_end; |
499 | } | |
500 | spin_unlock(&pag->pag_ici_lock); | |
501 | radix_tree_preload_end(); | |
502 | ||
503 | *ipp = ip; | |
504 | return 0; | |
505 | ||
506 | out_preload_end: | |
507 | spin_unlock(&pag->pag_ici_lock); | |
508 | radix_tree_preload_end(); | |
509 | if (lock_flags) | |
510 | xfs_iunlock(ip, lock_flags); | |
511 | out_destroy: | |
512 | __destroy_inode(VFS_I(ip)); | |
513 | xfs_inode_free(ip); | |
514 | return error; | |
515 | } | |
516 | ||
517 | /* | |
518 | * Look up an inode by number in the given file system. | |
519 | * The inode is looked up in the cache held in each AG. | |
520 | * If the inode is found in the cache, initialise the vfs inode | |
521 | * if necessary. | |
522 | * | |
523 | * If it is not in core, read it in from the file system's device, | |
524 | * add it to the cache and initialise the vfs inode. | |
525 | * | |
526 | * The inode is locked according to the value of the lock_flags parameter. | |
527 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
528 | * should be taken. | |
529 | * | |
530 | * mp -- the mount point structure for the current file system. It points | |
531 | * to the inode hash table. | |
532 | * tp -- a pointer to the current transaction if there is one. This is | |
533 | * simply passed through to the xfs_iread() call. | |
534 | * ino -- the number of the inode desired. This is the unique identifier | |
535 | * within the file system for the inode being requested. | |
536 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
537 | * for xfs_ilock() for a list of valid values. | |
538 | */ | |
539 | int | |
540 | xfs_iget( | |
541 | xfs_mount_t *mp, | |
542 | xfs_trans_t *tp, | |
543 | xfs_ino_t ino, | |
544 | uint flags, | |
545 | uint lock_flags, | |
546 | xfs_inode_t **ipp) | |
547 | { | |
548 | xfs_inode_t *ip; | |
549 | int error; | |
550 | xfs_perag_t *pag; | |
551 | xfs_agino_t agino; | |
552 | ||
553 | /* | |
554 | * xfs_reclaim_inode() uses the ILOCK to ensure an inode | |
555 | * doesn't get freed while it's being referenced during a | |
556 | * radix tree traversal here. It assumes this function | |
557 | * aqcuires only the ILOCK (and therefore it has no need to | |
558 | * involve the IOLOCK in this synchronization). | |
559 | */ | |
560 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); | |
561 | ||
562 | /* reject inode numbers outside existing AGs */ | |
563 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 564 | return -EINVAL; |
33479e05 | 565 | |
ff6d6af2 | 566 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 567 | |
33479e05 DC |
568 | /* get the perag structure and ensure that it's inode capable */ |
569 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
570 | agino = XFS_INO_TO_AGINO(mp, ino); | |
571 | ||
572 | again: | |
573 | error = 0; | |
574 | rcu_read_lock(); | |
575 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
576 | ||
577 | if (ip) { | |
578 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
579 | if (error) | |
580 | goto out_error_or_again; | |
581 | } else { | |
582 | rcu_read_unlock(); | |
ff6d6af2 | 583 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
584 | |
585 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
586 | flags, lock_flags); | |
587 | if (error) | |
588 | goto out_error_or_again; | |
589 | } | |
590 | xfs_perag_put(pag); | |
591 | ||
592 | *ipp = ip; | |
593 | ||
594 | /* | |
58c90473 | 595 | * If we have a real type for an on-disk inode, we can setup the inode |
33479e05 DC |
596 | * now. If it's a new inode being created, xfs_ialloc will handle it. |
597 | */ | |
c19b3b05 | 598 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 599 | xfs_setup_existing_inode(ip); |
33479e05 DC |
600 | return 0; |
601 | ||
602 | out_error_or_again: | |
2451337d | 603 | if (error == -EAGAIN) { |
33479e05 DC |
604 | delay(1); |
605 | goto again; | |
606 | } | |
607 | xfs_perag_put(pag); | |
608 | return error; | |
609 | } | |
610 | ||
78ae5256 DC |
611 | /* |
612 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
613 | * radix tree lookups to a minimum. The batch size is a trade off between | |
614 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
615 | * be too greedy. | |
616 | */ | |
617 | #define XFS_LOOKUP_BATCH 32 | |
618 | ||
e13de955 DC |
619 | STATIC int |
620 | xfs_inode_ag_walk_grab( | |
621 | struct xfs_inode *ip) | |
622 | { | |
623 | struct inode *inode = VFS_I(ip); | |
624 | ||
1a3e8f3d DC |
625 | ASSERT(rcu_read_lock_held()); |
626 | ||
627 | /* | |
628 | * check for stale RCU freed inode | |
629 | * | |
630 | * If the inode has been reallocated, it doesn't matter if it's not in | |
631 | * the AG we are walking - we are walking for writeback, so if it | |
632 | * passes all the "valid inode" checks and is dirty, then we'll write | |
633 | * it back anyway. If it has been reallocated and still being | |
634 | * initialised, the XFS_INEW check below will catch it. | |
635 | */ | |
636 | spin_lock(&ip->i_flags_lock); | |
637 | if (!ip->i_ino) | |
638 | goto out_unlock_noent; | |
639 | ||
640 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
641 | if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
642 | goto out_unlock_noent; | |
643 | spin_unlock(&ip->i_flags_lock); | |
644 | ||
e13de955 DC |
645 | /* nothing to sync during shutdown */ |
646 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
2451337d | 647 | return -EFSCORRUPTED; |
e13de955 | 648 | |
e13de955 DC |
649 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
650 | if (!igrab(inode)) | |
2451337d | 651 | return -ENOENT; |
e13de955 | 652 | |
e13de955 DC |
653 | /* inode is valid */ |
654 | return 0; | |
1a3e8f3d DC |
655 | |
656 | out_unlock_noent: | |
657 | spin_unlock(&ip->i_flags_lock); | |
2451337d | 658 | return -ENOENT; |
e13de955 DC |
659 | } |
660 | ||
75f3cb13 DC |
661 | STATIC int |
662 | xfs_inode_ag_walk( | |
663 | struct xfs_mount *mp, | |
5017e97d | 664 | struct xfs_perag *pag, |
e0094008 | 665 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
666 | void *args), |
667 | int flags, | |
668 | void *args, | |
669 | int tag) | |
75f3cb13 | 670 | { |
75f3cb13 DC |
671 | uint32_t first_index; |
672 | int last_error = 0; | |
673 | int skipped; | |
65d0f205 | 674 | int done; |
78ae5256 | 675 | int nr_found; |
75f3cb13 DC |
676 | |
677 | restart: | |
65d0f205 | 678 | done = 0; |
75f3cb13 DC |
679 | skipped = 0; |
680 | first_index = 0; | |
78ae5256 | 681 | nr_found = 0; |
75f3cb13 | 682 | do { |
78ae5256 | 683 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
75f3cb13 | 684 | int error = 0; |
78ae5256 | 685 | int i; |
75f3cb13 | 686 | |
1a3e8f3d | 687 | rcu_read_lock(); |
a454f742 BF |
688 | |
689 | if (tag == -1) | |
690 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
78ae5256 DC |
691 | (void **)batch, first_index, |
692 | XFS_LOOKUP_BATCH); | |
a454f742 BF |
693 | else |
694 | nr_found = radix_tree_gang_lookup_tag( | |
695 | &pag->pag_ici_root, | |
696 | (void **) batch, first_index, | |
697 | XFS_LOOKUP_BATCH, tag); | |
698 | ||
65d0f205 | 699 | if (!nr_found) { |
1a3e8f3d | 700 | rcu_read_unlock(); |
75f3cb13 | 701 | break; |
c8e20be0 | 702 | } |
75f3cb13 | 703 | |
65d0f205 | 704 | /* |
78ae5256 DC |
705 | * Grab the inodes before we drop the lock. if we found |
706 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 707 | */ |
78ae5256 DC |
708 | for (i = 0; i < nr_found; i++) { |
709 | struct xfs_inode *ip = batch[i]; | |
710 | ||
711 | if (done || xfs_inode_ag_walk_grab(ip)) | |
712 | batch[i] = NULL; | |
713 | ||
714 | /* | |
1a3e8f3d DC |
715 | * Update the index for the next lookup. Catch |
716 | * overflows into the next AG range which can occur if | |
717 | * we have inodes in the last block of the AG and we | |
718 | * are currently pointing to the last inode. | |
719 | * | |
720 | * Because we may see inodes that are from the wrong AG | |
721 | * due to RCU freeing and reallocation, only update the | |
722 | * index if it lies in this AG. It was a race that lead | |
723 | * us to see this inode, so another lookup from the | |
724 | * same index will not find it again. | |
78ae5256 | 725 | */ |
1a3e8f3d DC |
726 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
727 | continue; | |
78ae5256 DC |
728 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
729 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
730 | done = 1; | |
e13de955 | 731 | } |
78ae5256 DC |
732 | |
733 | /* unlock now we've grabbed the inodes. */ | |
1a3e8f3d | 734 | rcu_read_unlock(); |
e13de955 | 735 | |
78ae5256 DC |
736 | for (i = 0; i < nr_found; i++) { |
737 | if (!batch[i]) | |
738 | continue; | |
e0094008 | 739 | error = execute(batch[i], flags, args); |
78ae5256 | 740 | IRELE(batch[i]); |
2451337d | 741 | if (error == -EAGAIN) { |
78ae5256 DC |
742 | skipped++; |
743 | continue; | |
744 | } | |
2451337d | 745 | if (error && last_error != -EFSCORRUPTED) |
78ae5256 | 746 | last_error = error; |
75f3cb13 | 747 | } |
c8e20be0 DC |
748 | |
749 | /* bail out if the filesystem is corrupted. */ | |
2451337d | 750 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
751 | break; |
752 | ||
8daaa831 DC |
753 | cond_resched(); |
754 | ||
78ae5256 | 755 | } while (nr_found && !done); |
75f3cb13 DC |
756 | |
757 | if (skipped) { | |
758 | delay(1); | |
759 | goto restart; | |
760 | } | |
75f3cb13 DC |
761 | return last_error; |
762 | } | |
763 | ||
579b62fa BF |
764 | /* |
765 | * Background scanning to trim post-EOF preallocated space. This is queued | |
b9fe5052 | 766 | * based on the 'speculative_prealloc_lifetime' tunable (5m by default). |
579b62fa BF |
767 | */ |
768 | STATIC void | |
769 | xfs_queue_eofblocks( | |
770 | struct xfs_mount *mp) | |
771 | { | |
772 | rcu_read_lock(); | |
773 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG)) | |
774 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
775 | &mp->m_eofblocks_work, | |
776 | msecs_to_jiffies(xfs_eofb_secs * 1000)); | |
777 | rcu_read_unlock(); | |
778 | } | |
779 | ||
780 | void | |
781 | xfs_eofblocks_worker( | |
782 | struct work_struct *work) | |
783 | { | |
784 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
785 | struct xfs_mount, m_eofblocks_work); | |
786 | xfs_icache_free_eofblocks(mp, NULL); | |
787 | xfs_queue_eofblocks(mp); | |
788 | } | |
789 | ||
fe588ed3 | 790 | int |
75f3cb13 DC |
791 | xfs_inode_ag_iterator( |
792 | struct xfs_mount *mp, | |
e0094008 | 793 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
794 | void *args), |
795 | int flags, | |
796 | void *args) | |
75f3cb13 | 797 | { |
16fd5367 | 798 | struct xfs_perag *pag; |
75f3cb13 DC |
799 | int error = 0; |
800 | int last_error = 0; | |
801 | xfs_agnumber_t ag; | |
802 | ||
16fd5367 | 803 | ag = 0; |
65d0f205 DC |
804 | while ((pag = xfs_perag_get(mp, ag))) { |
805 | ag = pag->pag_agno + 1; | |
a454f742 BF |
806 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1); |
807 | xfs_perag_put(pag); | |
808 | if (error) { | |
809 | last_error = error; | |
2451337d | 810 | if (error == -EFSCORRUPTED) |
a454f742 BF |
811 | break; |
812 | } | |
813 | } | |
b474c7ae | 814 | return last_error; |
a454f742 BF |
815 | } |
816 | ||
817 | int | |
818 | xfs_inode_ag_iterator_tag( | |
819 | struct xfs_mount *mp, | |
e0094008 | 820 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
821 | void *args), |
822 | int flags, | |
823 | void *args, | |
824 | int tag) | |
825 | { | |
826 | struct xfs_perag *pag; | |
827 | int error = 0; | |
828 | int last_error = 0; | |
829 | xfs_agnumber_t ag; | |
830 | ||
831 | ag = 0; | |
832 | while ((pag = xfs_perag_get_tag(mp, ag, tag))) { | |
833 | ag = pag->pag_agno + 1; | |
834 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag); | |
5017e97d | 835 | xfs_perag_put(pag); |
75f3cb13 DC |
836 | if (error) { |
837 | last_error = error; | |
2451337d | 838 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
839 | break; |
840 | } | |
841 | } | |
b474c7ae | 842 | return last_error; |
75f3cb13 DC |
843 | } |
844 | ||
e3a20c0b DC |
845 | /* |
846 | * Grab the inode for reclaim exclusively. | |
847 | * Return 0 if we grabbed it, non-zero otherwise. | |
848 | */ | |
849 | STATIC int | |
850 | xfs_reclaim_inode_grab( | |
851 | struct xfs_inode *ip, | |
852 | int flags) | |
853 | { | |
1a3e8f3d DC |
854 | ASSERT(rcu_read_lock_held()); |
855 | ||
856 | /* quick check for stale RCU freed inode */ | |
857 | if (!ip->i_ino) | |
858 | return 1; | |
e3a20c0b DC |
859 | |
860 | /* | |
474fce06 CH |
861 | * If we are asked for non-blocking operation, do unlocked checks to |
862 | * see if the inode already is being flushed or in reclaim to avoid | |
863 | * lock traffic. | |
e3a20c0b DC |
864 | */ |
865 | if ((flags & SYNC_TRYLOCK) && | |
474fce06 | 866 | __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM)) |
e3a20c0b | 867 | return 1; |
e3a20c0b DC |
868 | |
869 | /* | |
870 | * The radix tree lock here protects a thread in xfs_iget from racing | |
871 | * with us starting reclaim on the inode. Once we have the | |
872 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
873 | * |
874 | * Due to RCU lookup, we may find inodes that have been freed and only | |
875 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
876 | * aren't candidates for reclaim at all, so we must check the | |
877 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
878 | */ |
879 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
880 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
881 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
882 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
883 | spin_unlock(&ip->i_flags_lock); |
884 | return 1; | |
885 | } | |
886 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
887 | spin_unlock(&ip->i_flags_lock); | |
888 | return 0; | |
889 | } | |
890 | ||
777df5af | 891 | /* |
8a48088f CH |
892 | * Inodes in different states need to be treated differently. The following |
893 | * table lists the inode states and the reclaim actions necessary: | |
777df5af DC |
894 | * |
895 | * inode state iflush ret required action | |
896 | * --------------- ---------- --------------- | |
897 | * bad - reclaim | |
898 | * shutdown EIO unpin and reclaim | |
899 | * clean, unpinned 0 reclaim | |
900 | * stale, unpinned 0 reclaim | |
c854363e DC |
901 | * clean, pinned(*) 0 requeue |
902 | * stale, pinned EAGAIN requeue | |
8a48088f CH |
903 | * dirty, async - requeue |
904 | * dirty, sync 0 reclaim | |
777df5af DC |
905 | * |
906 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
907 | * handled anyway given the order of checks implemented. | |
908 | * | |
c854363e DC |
909 | * Also, because we get the flush lock first, we know that any inode that has |
910 | * been flushed delwri has had the flush completed by the time we check that | |
8a48088f | 911 | * the inode is clean. |
c854363e | 912 | * |
8a48088f CH |
913 | * Note that because the inode is flushed delayed write by AIL pushing, the |
914 | * flush lock may already be held here and waiting on it can result in very | |
915 | * long latencies. Hence for sync reclaims, where we wait on the flush lock, | |
916 | * the caller should push the AIL first before trying to reclaim inodes to | |
917 | * minimise the amount of time spent waiting. For background relaim, we only | |
918 | * bother to reclaim clean inodes anyway. | |
c854363e | 919 | * |
777df5af DC |
920 | * Hence the order of actions after gaining the locks should be: |
921 | * bad => reclaim | |
922 | * shutdown => unpin and reclaim | |
8a48088f | 923 | * pinned, async => requeue |
c854363e | 924 | * pinned, sync => unpin |
777df5af DC |
925 | * stale => reclaim |
926 | * clean => reclaim | |
8a48088f | 927 | * dirty, async => requeue |
c854363e | 928 | * dirty, sync => flush, wait and reclaim |
777df5af | 929 | */ |
75f3cb13 | 930 | STATIC int |
c8e20be0 | 931 | xfs_reclaim_inode( |
75f3cb13 DC |
932 | struct xfs_inode *ip, |
933 | struct xfs_perag *pag, | |
c8e20be0 | 934 | int sync_mode) |
fce08f2f | 935 | { |
4c46819a | 936 | struct xfs_buf *bp = NULL; |
8a17d7dd | 937 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
4c46819a | 938 | int error; |
777df5af | 939 | |
1bfd8d04 DC |
940 | restart: |
941 | error = 0; | |
c8e20be0 | 942 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
943 | if (!xfs_iflock_nowait(ip)) { |
944 | if (!(sync_mode & SYNC_WAIT)) | |
945 | goto out; | |
946 | xfs_iflock(ip); | |
947 | } | |
7a3be02b | 948 | |
777df5af DC |
949 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
950 | xfs_iunpin_wait(ip); | |
04913fdd | 951 | xfs_iflush_abort(ip, false); |
777df5af DC |
952 | goto reclaim; |
953 | } | |
c854363e | 954 | if (xfs_ipincount(ip)) { |
8a48088f CH |
955 | if (!(sync_mode & SYNC_WAIT)) |
956 | goto out_ifunlock; | |
777df5af | 957 | xfs_iunpin_wait(ip); |
c854363e | 958 | } |
777df5af DC |
959 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
960 | goto reclaim; | |
961 | if (xfs_inode_clean(ip)) | |
962 | goto reclaim; | |
963 | ||
8a48088f CH |
964 | /* |
965 | * Never flush out dirty data during non-blocking reclaim, as it would | |
966 | * just contend with AIL pushing trying to do the same job. | |
967 | */ | |
968 | if (!(sync_mode & SYNC_WAIT)) | |
969 | goto out_ifunlock; | |
970 | ||
1bfd8d04 DC |
971 | /* |
972 | * Now we have an inode that needs flushing. | |
973 | * | |
4c46819a | 974 | * Note that xfs_iflush will never block on the inode buffer lock, as |
1bfd8d04 | 975 | * xfs_ifree_cluster() can lock the inode buffer before it locks the |
4c46819a | 976 | * ip->i_lock, and we are doing the exact opposite here. As a result, |
475ee413 CH |
977 | * doing a blocking xfs_imap_to_bp() to get the cluster buffer would |
978 | * result in an ABBA deadlock with xfs_ifree_cluster(). | |
1bfd8d04 DC |
979 | * |
980 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
981 | * cache to mark them stale, if we hit this case we don't actually want | |
982 | * to do IO here - we want the inode marked stale so we can simply | |
4c46819a CH |
983 | * reclaim it. Hence if we get an EAGAIN error here, just unlock the |
984 | * inode, back off and try again. Hopefully the next pass through will | |
985 | * see the stale flag set on the inode. | |
1bfd8d04 | 986 | */ |
4c46819a | 987 | error = xfs_iflush(ip, &bp); |
2451337d | 988 | if (error == -EAGAIN) { |
8a48088f CH |
989 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
990 | /* backoff longer than in xfs_ifree_cluster */ | |
991 | delay(2); | |
992 | goto restart; | |
c854363e | 993 | } |
c854363e | 994 | |
4c46819a CH |
995 | if (!error) { |
996 | error = xfs_bwrite(bp); | |
997 | xfs_buf_relse(bp); | |
998 | } | |
999 | ||
1000 | xfs_iflock(ip); | |
777df5af | 1001 | reclaim: |
8a17d7dd DC |
1002 | /* |
1003 | * Because we use RCU freeing we need to ensure the inode always appears | |
1004 | * to be reclaimed with an invalid inode number when in the free state. | |
1005 | * We do this as early as possible under the ILOCK and flush lock so | |
1006 | * that xfs_iflush_cluster() can be guaranteed to detect races with us | |
1007 | * here. By doing this, we guarantee that once xfs_iflush_cluster has | |
1008 | * locked both the XFS_ILOCK and the flush lock that it will see either | |
1009 | * a valid, flushable inode that will serialise correctly against the | |
1010 | * locks below, or it will see a clean (and invalid) inode that it can | |
1011 | * skip. | |
1012 | */ | |
1013 | spin_lock(&ip->i_flags_lock); | |
1014 | ip->i_flags = XFS_IRECLAIM; | |
1015 | ip->i_ino = 0; | |
1016 | spin_unlock(&ip->i_flags_lock); | |
1017 | ||
777df5af | 1018 | xfs_ifunlock(ip); |
c8e20be0 | 1019 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1020 | |
ff6d6af2 | 1021 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
1022 | /* |
1023 | * Remove the inode from the per-AG radix tree. | |
1024 | * | |
1025 | * Because radix_tree_delete won't complain even if the item was never | |
1026 | * added to the tree assert that it's been there before to catch | |
1027 | * problems with the inode life time early on. | |
1028 | */ | |
1a427ab0 | 1029 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 1030 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 1031 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 1032 | ASSERT(0); |
545c0889 | 1033 | xfs_perag_clear_reclaim_tag(pag); |
1a427ab0 | 1034 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
1035 | |
1036 | /* | |
1037 | * Here we do an (almost) spurious inode lock in order to coordinate | |
1038 | * with inode cache radix tree lookups. This is because the lookup | |
1039 | * can reference the inodes in the cache without taking references. | |
1040 | * | |
1041 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 1042 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 1043 | */ |
ad637a10 | 1044 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1045 | xfs_qm_dqdetach(ip); |
ad637a10 | 1046 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1047 | |
8a17d7dd | 1048 | __xfs_inode_free(ip); |
ad637a10 | 1049 | return error; |
8a48088f CH |
1050 | |
1051 | out_ifunlock: | |
1052 | xfs_ifunlock(ip); | |
1053 | out: | |
1054 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
1055 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1056 | /* | |
2451337d | 1057 | * We could return -EAGAIN here to make reclaim rescan the inode tree in |
8a48088f | 1058 | * a short while. However, this just burns CPU time scanning the tree |
5889608d DC |
1059 | * waiting for IO to complete and the reclaim work never goes back to |
1060 | * the idle state. Instead, return 0 to let the next scheduled | |
1061 | * background reclaim attempt to reclaim the inode again. | |
8a48088f CH |
1062 | */ |
1063 | return 0; | |
7a3be02b DC |
1064 | } |
1065 | ||
65d0f205 DC |
1066 | /* |
1067 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
1068 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
1069 | * then a shut down during filesystem unmount reclaim walk leak all the | |
1070 | * unreclaimed inodes. | |
1071 | */ | |
33479e05 | 1072 | STATIC int |
65d0f205 DC |
1073 | xfs_reclaim_inodes_ag( |
1074 | struct xfs_mount *mp, | |
1075 | int flags, | |
1076 | int *nr_to_scan) | |
1077 | { | |
1078 | struct xfs_perag *pag; | |
1079 | int error = 0; | |
1080 | int last_error = 0; | |
1081 | xfs_agnumber_t ag; | |
69b491c2 DC |
1082 | int trylock = flags & SYNC_TRYLOCK; |
1083 | int skipped; | |
65d0f205 | 1084 | |
69b491c2 | 1085 | restart: |
65d0f205 | 1086 | ag = 0; |
69b491c2 | 1087 | skipped = 0; |
65d0f205 DC |
1088 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1089 | unsigned long first_index = 0; | |
1090 | int done = 0; | |
e3a20c0b | 1091 | int nr_found = 0; |
65d0f205 DC |
1092 | |
1093 | ag = pag->pag_agno + 1; | |
1094 | ||
69b491c2 DC |
1095 | if (trylock) { |
1096 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
1097 | skipped++; | |
f83282a8 | 1098 | xfs_perag_put(pag); |
69b491c2 DC |
1099 | continue; |
1100 | } | |
1101 | first_index = pag->pag_ici_reclaim_cursor; | |
1102 | } else | |
1103 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
1104 | ||
65d0f205 | 1105 | do { |
e3a20c0b DC |
1106 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
1107 | int i; | |
65d0f205 | 1108 | |
1a3e8f3d | 1109 | rcu_read_lock(); |
e3a20c0b DC |
1110 | nr_found = radix_tree_gang_lookup_tag( |
1111 | &pag->pag_ici_root, | |
1112 | (void **)batch, first_index, | |
1113 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
1114 | XFS_ICI_RECLAIM_TAG); |
1115 | if (!nr_found) { | |
b2232219 | 1116 | done = 1; |
1a3e8f3d | 1117 | rcu_read_unlock(); |
65d0f205 DC |
1118 | break; |
1119 | } | |
1120 | ||
1121 | /* | |
e3a20c0b DC |
1122 | * Grab the inodes before we drop the lock. if we found |
1123 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 1124 | */ |
e3a20c0b DC |
1125 | for (i = 0; i < nr_found; i++) { |
1126 | struct xfs_inode *ip = batch[i]; | |
1127 | ||
1128 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
1129 | batch[i] = NULL; | |
1130 | ||
1131 | /* | |
1132 | * Update the index for the next lookup. Catch | |
1133 | * overflows into the next AG range which can | |
1134 | * occur if we have inodes in the last block of | |
1135 | * the AG and we are currently pointing to the | |
1136 | * last inode. | |
1a3e8f3d DC |
1137 | * |
1138 | * Because we may see inodes that are from the | |
1139 | * wrong AG due to RCU freeing and | |
1140 | * reallocation, only update the index if it | |
1141 | * lies in this AG. It was a race that lead us | |
1142 | * to see this inode, so another lookup from | |
1143 | * the same index will not find it again. | |
e3a20c0b | 1144 | */ |
1a3e8f3d DC |
1145 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
1146 | pag->pag_agno) | |
1147 | continue; | |
e3a20c0b DC |
1148 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
1149 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1150 | done = 1; | |
1151 | } | |
65d0f205 | 1152 | |
e3a20c0b | 1153 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 1154 | rcu_read_unlock(); |
e3a20c0b DC |
1155 | |
1156 | for (i = 0; i < nr_found; i++) { | |
1157 | if (!batch[i]) | |
1158 | continue; | |
1159 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
2451337d | 1160 | if (error && last_error != -EFSCORRUPTED) |
e3a20c0b DC |
1161 | last_error = error; |
1162 | } | |
1163 | ||
1164 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 1165 | |
8daaa831 DC |
1166 | cond_resched(); |
1167 | ||
e3a20c0b | 1168 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 1169 | |
69b491c2 DC |
1170 | if (trylock && !done) |
1171 | pag->pag_ici_reclaim_cursor = first_index; | |
1172 | else | |
1173 | pag->pag_ici_reclaim_cursor = 0; | |
1174 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
1175 | xfs_perag_put(pag); |
1176 | } | |
69b491c2 DC |
1177 | |
1178 | /* | |
1179 | * if we skipped any AG, and we still have scan count remaining, do | |
1180 | * another pass this time using blocking reclaim semantics (i.e | |
1181 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1182 | * ensure that when we get more reclaimers than AGs we block rather | |
1183 | * than spin trying to execute reclaim. | |
1184 | */ | |
8daaa831 | 1185 | if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) { |
69b491c2 DC |
1186 | trylock = 0; |
1187 | goto restart; | |
1188 | } | |
b474c7ae | 1189 | return last_error; |
65d0f205 DC |
1190 | } |
1191 | ||
7a3be02b DC |
1192 | int |
1193 | xfs_reclaim_inodes( | |
1194 | xfs_mount_t *mp, | |
7a3be02b DC |
1195 | int mode) |
1196 | { | |
65d0f205 DC |
1197 | int nr_to_scan = INT_MAX; |
1198 | ||
1199 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1200 | } |
1201 | ||
1202 | /* | |
8daaa831 | 1203 | * Scan a certain number of inodes for reclaim. |
a7b339f1 DC |
1204 | * |
1205 | * When called we make sure that there is a background (fast) inode reclaim in | |
8daaa831 | 1206 | * progress, while we will throttle the speed of reclaim via doing synchronous |
a7b339f1 DC |
1207 | * reclaim of inodes. That means if we come across dirty inodes, we wait for |
1208 | * them to be cleaned, which we hope will not be very long due to the | |
1209 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1210 | */ |
0a234c6d | 1211 | long |
8daaa831 DC |
1212 | xfs_reclaim_inodes_nr( |
1213 | struct xfs_mount *mp, | |
1214 | int nr_to_scan) | |
9bf729c0 | 1215 | { |
8daaa831 | 1216 | /* kick background reclaimer and push the AIL */ |
5889608d | 1217 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1218 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1219 | |
0a234c6d | 1220 | return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan); |
8daaa831 | 1221 | } |
9bf729c0 | 1222 | |
8daaa831 DC |
1223 | /* |
1224 | * Return the number of reclaimable inodes in the filesystem for | |
1225 | * the shrinker to determine how much to reclaim. | |
1226 | */ | |
1227 | int | |
1228 | xfs_reclaim_inodes_count( | |
1229 | struct xfs_mount *mp) | |
1230 | { | |
1231 | struct xfs_perag *pag; | |
1232 | xfs_agnumber_t ag = 0; | |
1233 | int reclaimable = 0; | |
9bf729c0 | 1234 | |
65d0f205 DC |
1235 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1236 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1237 | reclaimable += pag->pag_ici_reclaimable; |
1238 | xfs_perag_put(pag); | |
9bf729c0 | 1239 | } |
9bf729c0 DC |
1240 | return reclaimable; |
1241 | } | |
1242 | ||
3e3f9f58 BF |
1243 | STATIC int |
1244 | xfs_inode_match_id( | |
1245 | struct xfs_inode *ip, | |
1246 | struct xfs_eofblocks *eofb) | |
1247 | { | |
b9fe5052 DE |
1248 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && |
1249 | !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
1b556048 | 1250 | return 0; |
3e3f9f58 | 1251 | |
b9fe5052 DE |
1252 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && |
1253 | !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
1b556048 BF |
1254 | return 0; |
1255 | ||
b9fe5052 | 1256 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && |
1b556048 BF |
1257 | xfs_get_projid(ip) != eofb->eof_prid) |
1258 | return 0; | |
1259 | ||
1260 | return 1; | |
3e3f9f58 BF |
1261 | } |
1262 | ||
f4526397 BF |
1263 | /* |
1264 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1265 | * criteria match. This is for global/internal scans only. | |
1266 | */ | |
1267 | STATIC int | |
1268 | xfs_inode_match_id_union( | |
1269 | struct xfs_inode *ip, | |
1270 | struct xfs_eofblocks *eofb) | |
1271 | { | |
1272 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && | |
1273 | uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
1274 | return 1; | |
1275 | ||
1276 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && | |
1277 | gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
1278 | return 1; | |
1279 | ||
1280 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && | |
1281 | xfs_get_projid(ip) == eofb->eof_prid) | |
1282 | return 1; | |
1283 | ||
1284 | return 0; | |
1285 | } | |
1286 | ||
41176a68 BF |
1287 | STATIC int |
1288 | xfs_inode_free_eofblocks( | |
1289 | struct xfs_inode *ip, | |
41176a68 BF |
1290 | int flags, |
1291 | void *args) | |
1292 | { | |
1293 | int ret; | |
3e3f9f58 | 1294 | struct xfs_eofblocks *eofb = args; |
5400da7d | 1295 | bool need_iolock = true; |
f4526397 | 1296 | int match; |
5400da7d BF |
1297 | |
1298 | ASSERT(!eofb || (eofb && eofb->eof_scan_owner != 0)); | |
41176a68 BF |
1299 | |
1300 | if (!xfs_can_free_eofblocks(ip, false)) { | |
1301 | /* inode could be preallocated or append-only */ | |
1302 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1303 | xfs_inode_clear_eofblocks_tag(ip); | |
1304 | return 0; | |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * If the mapping is dirty the operation can block and wait for some | |
1309 | * time. Unless we are waiting, skip it. | |
1310 | */ | |
1311 | if (!(flags & SYNC_WAIT) && | |
1312 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) | |
1313 | return 0; | |
1314 | ||
00ca79a0 | 1315 | if (eofb) { |
f4526397 BF |
1316 | if (eofb->eof_flags & XFS_EOF_FLAGS_UNION) |
1317 | match = xfs_inode_match_id_union(ip, eofb); | |
1318 | else | |
1319 | match = xfs_inode_match_id(ip, eofb); | |
1320 | if (!match) | |
00ca79a0 BF |
1321 | return 0; |
1322 | ||
1323 | /* skip the inode if the file size is too small */ | |
1324 | if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE && | |
1325 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1326 | return 0; | |
5400da7d BF |
1327 | |
1328 | /* | |
1329 | * A scan owner implies we already hold the iolock. Skip it in | |
1330 | * xfs_free_eofblocks() to avoid deadlock. This also eliminates | |
1331 | * the possibility of EAGAIN being returned. | |
1332 | */ | |
1333 | if (eofb->eof_scan_owner == ip->i_ino) | |
1334 | need_iolock = false; | |
00ca79a0 | 1335 | } |
3e3f9f58 | 1336 | |
5400da7d | 1337 | ret = xfs_free_eofblocks(ip->i_mount, ip, need_iolock); |
41176a68 BF |
1338 | |
1339 | /* don't revisit the inode if we're not waiting */ | |
2451337d | 1340 | if (ret == -EAGAIN && !(flags & SYNC_WAIT)) |
41176a68 BF |
1341 | ret = 0; |
1342 | ||
1343 | return ret; | |
1344 | } | |
1345 | ||
1346 | int | |
1347 | xfs_icache_free_eofblocks( | |
1348 | struct xfs_mount *mp, | |
8ca149de | 1349 | struct xfs_eofblocks *eofb) |
41176a68 | 1350 | { |
8ca149de BF |
1351 | int flags = SYNC_TRYLOCK; |
1352 | ||
1353 | if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC)) | |
1354 | flags = SYNC_WAIT; | |
1355 | ||
41176a68 | 1356 | return xfs_inode_ag_iterator_tag(mp, xfs_inode_free_eofblocks, flags, |
8ca149de | 1357 | eofb, XFS_ICI_EOFBLOCKS_TAG); |
41176a68 BF |
1358 | } |
1359 | ||
dc06f398 BF |
1360 | /* |
1361 | * Run eofblocks scans on the quotas applicable to the inode. For inodes with | |
1362 | * multiple quotas, we don't know exactly which quota caused an allocation | |
1363 | * failure. We make a best effort by including each quota under low free space | |
1364 | * conditions (less than 1% free space) in the scan. | |
1365 | */ | |
1366 | int | |
1367 | xfs_inode_free_quota_eofblocks( | |
1368 | struct xfs_inode *ip) | |
1369 | { | |
1370 | int scan = 0; | |
1371 | struct xfs_eofblocks eofb = {0}; | |
1372 | struct xfs_dquot *dq; | |
1373 | ||
1374 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
1375 | ||
1376 | /* | |
1377 | * Set the scan owner to avoid a potential livelock. Otherwise, the scan | |
1378 | * can repeatedly trylock on the inode we're currently processing. We | |
1379 | * run a sync scan to increase effectiveness and use the union filter to | |
1380 | * cover all applicable quotas in a single scan. | |
1381 | */ | |
1382 | eofb.eof_scan_owner = ip->i_ino; | |
1383 | eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC; | |
1384 | ||
1385 | if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) { | |
1386 | dq = xfs_inode_dquot(ip, XFS_DQ_USER); | |
1387 | if (dq && xfs_dquot_lowsp(dq)) { | |
1388 | eofb.eof_uid = VFS_I(ip)->i_uid; | |
1389 | eofb.eof_flags |= XFS_EOF_FLAGS_UID; | |
1390 | scan = 1; | |
1391 | } | |
1392 | } | |
1393 | ||
1394 | if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) { | |
1395 | dq = xfs_inode_dquot(ip, XFS_DQ_GROUP); | |
1396 | if (dq && xfs_dquot_lowsp(dq)) { | |
1397 | eofb.eof_gid = VFS_I(ip)->i_gid; | |
1398 | eofb.eof_flags |= XFS_EOF_FLAGS_GID; | |
1399 | scan = 1; | |
1400 | } | |
1401 | } | |
1402 | ||
1403 | if (scan) | |
1404 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
1405 | ||
1406 | return scan; | |
1407 | } | |
1408 | ||
27b52867 BF |
1409 | void |
1410 | xfs_inode_set_eofblocks_tag( | |
1411 | xfs_inode_t *ip) | |
1412 | { | |
1413 | struct xfs_mount *mp = ip->i_mount; | |
1414 | struct xfs_perag *pag; | |
1415 | int tagged; | |
1416 | ||
1417 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
1418 | spin_lock(&pag->pag_ici_lock); | |
1419 | trace_xfs_inode_set_eofblocks_tag(ip); | |
1420 | ||
1421 | tagged = radix_tree_tagged(&pag->pag_ici_root, | |
1422 | XFS_ICI_EOFBLOCKS_TAG); | |
1423 | radix_tree_tag_set(&pag->pag_ici_root, | |
1424 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
1425 | XFS_ICI_EOFBLOCKS_TAG); | |
1426 | if (!tagged) { | |
1427 | /* propagate the eofblocks tag up into the perag radix tree */ | |
1428 | spin_lock(&ip->i_mount->m_perag_lock); | |
1429 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
1430 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
1431 | XFS_ICI_EOFBLOCKS_TAG); | |
1432 | spin_unlock(&ip->i_mount->m_perag_lock); | |
579b62fa BF |
1433 | |
1434 | /* kick off background trimming */ | |
1435 | xfs_queue_eofblocks(ip->i_mount); | |
27b52867 BF |
1436 | |
1437 | trace_xfs_perag_set_eofblocks(ip->i_mount, pag->pag_agno, | |
1438 | -1, _RET_IP_); | |
1439 | } | |
1440 | ||
1441 | spin_unlock(&pag->pag_ici_lock); | |
1442 | xfs_perag_put(pag); | |
1443 | } | |
1444 | ||
1445 | void | |
1446 | xfs_inode_clear_eofblocks_tag( | |
1447 | xfs_inode_t *ip) | |
1448 | { | |
1449 | struct xfs_mount *mp = ip->i_mount; | |
1450 | struct xfs_perag *pag; | |
1451 | ||
1452 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
1453 | spin_lock(&pag->pag_ici_lock); | |
1454 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
1455 | ||
1456 | radix_tree_tag_clear(&pag->pag_ici_root, | |
1457 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), | |
1458 | XFS_ICI_EOFBLOCKS_TAG); | |
1459 | if (!radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_EOFBLOCKS_TAG)) { | |
1460 | /* clear the eofblocks tag from the perag radix tree */ | |
1461 | spin_lock(&ip->i_mount->m_perag_lock); | |
1462 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
1463 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
1464 | XFS_ICI_EOFBLOCKS_TAG); | |
1465 | spin_unlock(&ip->i_mount->m_perag_lock); | |
1466 | trace_xfs_perag_clear_eofblocks(ip->i_mount, pag->pag_agno, | |
1467 | -1, _RET_IP_); | |
1468 | } | |
1469 | ||
1470 | spin_unlock(&pag->pag_ici_lock); | |
1471 | xfs_perag_put(pag); | |
1472 | } | |
1473 |