]>
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" | |
83104d44 | 36 | #include "xfs_reflink.h" |
fe4fa4b8 | 37 | |
a167b17e DC |
38 | #include <linux/kthread.h> |
39 | #include <linux/freezer.h> | |
f0e28280 | 40 | #include <linux/iversion.h> |
a167b17e | 41 | |
33479e05 DC |
42 | /* |
43 | * Allocate and initialise an xfs_inode. | |
44 | */ | |
638f4416 | 45 | struct xfs_inode * |
33479e05 DC |
46 | xfs_inode_alloc( |
47 | struct xfs_mount *mp, | |
48 | xfs_ino_t ino) | |
49 | { | |
50 | struct xfs_inode *ip; | |
51 | ||
52 | /* | |
53 | * if this didn't occur in transactions, we could use | |
54 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
55 | * code up to do this anyway. | |
56 | */ | |
57 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
58 | if (!ip) | |
59 | return NULL; | |
60 | if (inode_init_always(mp->m_super, VFS_I(ip))) { | |
61 | kmem_zone_free(xfs_inode_zone, ip); | |
62 | return NULL; | |
63 | } | |
64 | ||
c19b3b05 DC |
65 | /* VFS doesn't initialise i_mode! */ |
66 | VFS_I(ip)->i_mode = 0; | |
67 | ||
ff6d6af2 | 68 | XFS_STATS_INC(mp, vn_active); |
33479e05 | 69 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
33479e05 DC |
70 | ASSERT(!xfs_isiflocked(ip)); |
71 | ASSERT(ip->i_ino == 0); | |
72 | ||
33479e05 DC |
73 | /* initialise the xfs inode */ |
74 | ip->i_ino = ino; | |
75 | ip->i_mount = mp; | |
76 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
77 | ip->i_afp = NULL; | |
3993baeb DW |
78 | ip->i_cowfp = NULL; |
79 | ip->i_cnextents = 0; | |
80 | ip->i_cformat = XFS_DINODE_FMT_EXTENTS; | |
33479e05 DC |
81 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); |
82 | ip->i_flags = 0; | |
83 | ip->i_delayed_blks = 0; | |
f8d55aa0 | 84 | memset(&ip->i_d, 0, sizeof(ip->i_d)); |
33479e05 DC |
85 | |
86 | return ip; | |
87 | } | |
88 | ||
89 | STATIC void | |
90 | xfs_inode_free_callback( | |
91 | struct rcu_head *head) | |
92 | { | |
93 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
94 | struct xfs_inode *ip = XFS_I(inode); | |
95 | ||
c19b3b05 | 96 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
97 | case S_IFREG: |
98 | case S_IFDIR: | |
99 | case S_IFLNK: | |
100 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
101 | break; | |
102 | } | |
103 | ||
104 | if (ip->i_afp) | |
105 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
3993baeb DW |
106 | if (ip->i_cowfp) |
107 | xfs_idestroy_fork(ip, XFS_COW_FORK); | |
33479e05 DC |
108 | |
109 | if (ip->i_itemp) { | |
110 | ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL)); | |
111 | xfs_inode_item_destroy(ip); | |
112 | ip->i_itemp = NULL; | |
113 | } | |
114 | ||
1f2dcfe8 DC |
115 | kmem_zone_free(xfs_inode_zone, ip); |
116 | } | |
117 | ||
8a17d7dd DC |
118 | static void |
119 | __xfs_inode_free( | |
120 | struct xfs_inode *ip) | |
121 | { | |
122 | /* asserts to verify all state is correct here */ | |
123 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
8a17d7dd DC |
124 | XFS_STATS_DEC(ip->i_mount, vn_active); |
125 | ||
126 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
127 | } | |
128 | ||
1f2dcfe8 DC |
129 | void |
130 | xfs_inode_free( | |
131 | struct xfs_inode *ip) | |
132 | { | |
98efe8af BF |
133 | ASSERT(!xfs_isiflocked(ip)); |
134 | ||
33479e05 DC |
135 | /* |
136 | * Because we use RCU freeing we need to ensure the inode always | |
137 | * appears to be reclaimed with an invalid inode number when in the | |
138 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
139 | * races. | |
140 | */ | |
141 | spin_lock(&ip->i_flags_lock); | |
142 | ip->i_flags = XFS_IRECLAIM; | |
143 | ip->i_ino = 0; | |
144 | spin_unlock(&ip->i_flags_lock); | |
145 | ||
8a17d7dd | 146 | __xfs_inode_free(ip); |
33479e05 DC |
147 | } |
148 | ||
ad438c40 DC |
149 | /* |
150 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
151 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
152 | * on the xfs periodic sync default of 30s. Perhaps this should have it's own | |
153 | * tunable, but that can be done if this method proves to be ineffective or too | |
154 | * aggressive. | |
155 | */ | |
156 | static void | |
157 | xfs_reclaim_work_queue( | |
158 | struct xfs_mount *mp) | |
159 | { | |
160 | ||
161 | rcu_read_lock(); | |
162 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
163 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
164 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
165 | } | |
166 | rcu_read_unlock(); | |
167 | } | |
168 | ||
169 | /* | |
170 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
171 | * many inodes as possible in a short period of time. It kicks itself every few | |
172 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
173 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
174 | * already being flushed, and once done schedules a future pass. | |
175 | */ | |
176 | void | |
177 | xfs_reclaim_worker( | |
178 | struct work_struct *work) | |
179 | { | |
180 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
181 | struct xfs_mount, m_reclaim_work); | |
182 | ||
183 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
184 | xfs_reclaim_work_queue(mp); | |
185 | } | |
186 | ||
187 | static void | |
188 | xfs_perag_set_reclaim_tag( | |
189 | struct xfs_perag *pag) | |
190 | { | |
191 | struct xfs_mount *mp = pag->pag_mount; | |
192 | ||
95989c46 | 193 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
194 | if (pag->pag_ici_reclaimable++) |
195 | return; | |
196 | ||
197 | /* propagate the reclaim tag up into the perag radix tree */ | |
198 | spin_lock(&mp->m_perag_lock); | |
199 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, | |
200 | XFS_ICI_RECLAIM_TAG); | |
201 | spin_unlock(&mp->m_perag_lock); | |
202 | ||
203 | /* schedule periodic background inode reclaim */ | |
204 | xfs_reclaim_work_queue(mp); | |
205 | ||
206 | trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
207 | } | |
208 | ||
209 | static void | |
210 | xfs_perag_clear_reclaim_tag( | |
211 | struct xfs_perag *pag) | |
212 | { | |
213 | struct xfs_mount *mp = pag->pag_mount; | |
214 | ||
95989c46 | 215 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
216 | if (--pag->pag_ici_reclaimable) |
217 | return; | |
218 | ||
219 | /* clear the reclaim tag from the perag radix tree */ | |
220 | spin_lock(&mp->m_perag_lock); | |
221 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, | |
222 | XFS_ICI_RECLAIM_TAG); | |
223 | spin_unlock(&mp->m_perag_lock); | |
224 | trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
225 | } | |
226 | ||
227 | ||
228 | /* | |
229 | * We set the inode flag atomically with the radix tree tag. | |
230 | * Once we get tag lookups on the radix tree, this inode flag | |
231 | * can go away. | |
232 | */ | |
233 | void | |
234 | xfs_inode_set_reclaim_tag( | |
235 | struct xfs_inode *ip) | |
236 | { | |
237 | struct xfs_mount *mp = ip->i_mount; | |
238 | struct xfs_perag *pag; | |
239 | ||
240 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
241 | spin_lock(&pag->pag_ici_lock); | |
242 | spin_lock(&ip->i_flags_lock); | |
243 | ||
244 | radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
245 | XFS_ICI_RECLAIM_TAG); | |
246 | xfs_perag_set_reclaim_tag(pag); | |
247 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
248 | ||
249 | spin_unlock(&ip->i_flags_lock); | |
250 | spin_unlock(&pag->pag_ici_lock); | |
251 | xfs_perag_put(pag); | |
252 | } | |
253 | ||
254 | STATIC void | |
255 | xfs_inode_clear_reclaim_tag( | |
256 | struct xfs_perag *pag, | |
257 | xfs_ino_t ino) | |
258 | { | |
259 | radix_tree_tag_clear(&pag->pag_ici_root, | |
260 | XFS_INO_TO_AGINO(pag->pag_mount, ino), | |
261 | XFS_ICI_RECLAIM_TAG); | |
262 | xfs_perag_clear_reclaim_tag(pag); | |
263 | } | |
264 | ||
ae2c4ac2 BF |
265 | static void |
266 | xfs_inew_wait( | |
267 | struct xfs_inode *ip) | |
268 | { | |
269 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT); | |
270 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT); | |
271 | ||
272 | do { | |
21417136 | 273 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
ae2c4ac2 BF |
274 | if (!xfs_iflags_test(ip, XFS_INEW)) |
275 | break; | |
276 | schedule(); | |
277 | } while (true); | |
21417136 | 278 | finish_wait(wq, &wait.wq_entry); |
ae2c4ac2 BF |
279 | } |
280 | ||
50997470 DC |
281 | /* |
282 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
283 | * part of the structure. This is made more complex by the fact we store | |
284 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 285 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 286 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 287 | * after reinitialisation even if it fails. |
50997470 DC |
288 | */ |
289 | static int | |
290 | xfs_reinit_inode( | |
291 | struct xfs_mount *mp, | |
292 | struct inode *inode) | |
293 | { | |
294 | int error; | |
54d7b5c1 | 295 | uint32_t nlink = inode->i_nlink; |
9e9a2674 | 296 | uint32_t generation = inode->i_generation; |
f0e28280 | 297 | uint64_t version = inode_peek_iversion(inode); |
c19b3b05 | 298 | umode_t mode = inode->i_mode; |
acd1d715 | 299 | dev_t dev = inode->i_rdev; |
50997470 DC |
300 | |
301 | error = inode_init_always(mp->m_super, inode); | |
302 | ||
54d7b5c1 | 303 | set_nlink(inode, nlink); |
9e9a2674 | 304 | inode->i_generation = generation; |
f0e28280 | 305 | inode_set_iversion_queried(inode, version); |
c19b3b05 | 306 | inode->i_mode = mode; |
acd1d715 | 307 | inode->i_rdev = dev; |
50997470 DC |
308 | return error; |
309 | } | |
310 | ||
33479e05 DC |
311 | /* |
312 | * Check the validity of the inode we just found it the cache | |
313 | */ | |
314 | static int | |
315 | xfs_iget_cache_hit( | |
316 | struct xfs_perag *pag, | |
317 | struct xfs_inode *ip, | |
318 | xfs_ino_t ino, | |
319 | int flags, | |
320 | int lock_flags) __releases(RCU) | |
321 | { | |
322 | struct inode *inode = VFS_I(ip); | |
323 | struct xfs_mount *mp = ip->i_mount; | |
324 | int error; | |
325 | ||
326 | /* | |
327 | * check for re-use of an inode within an RCU grace period due to the | |
328 | * radix tree nodes not being updated yet. We monitor for this by | |
329 | * setting the inode number to zero before freeing the inode structure. | |
330 | * If the inode has been reallocated and set up, then the inode number | |
331 | * will not match, so check for that, too. | |
332 | */ | |
333 | spin_lock(&ip->i_flags_lock); | |
334 | if (ip->i_ino != ino) { | |
335 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 336 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 337 | error = -EAGAIN; |
33479e05 DC |
338 | goto out_error; |
339 | } | |
340 | ||
341 | ||
342 | /* | |
343 | * If we are racing with another cache hit that is currently | |
344 | * instantiating this inode or currently recycling it out of | |
345 | * reclaimabe state, wait for the initialisation to complete | |
346 | * before continuing. | |
347 | * | |
348 | * XXX(hch): eventually we should do something equivalent to | |
349 | * wait_on_inode to wait for these flags to be cleared | |
350 | * instead of polling for it. | |
351 | */ | |
352 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { | |
353 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 354 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 355 | error = -EAGAIN; |
33479e05 DC |
356 | goto out_error; |
357 | } | |
358 | ||
359 | /* | |
360 | * If lookup is racing with unlink return an error immediately. | |
361 | */ | |
c19b3b05 | 362 | if (VFS_I(ip)->i_mode == 0 && !(flags & XFS_IGET_CREATE)) { |
2451337d | 363 | error = -ENOENT; |
33479e05 DC |
364 | goto out_error; |
365 | } | |
366 | ||
367 | /* | |
368 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
369 | * Need to carefully get it back into useable state. | |
370 | */ | |
371 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
372 | trace_xfs_iget_reclaim(ip); | |
373 | ||
378f681c DW |
374 | if (flags & XFS_IGET_INCORE) { |
375 | error = -EAGAIN; | |
376 | goto out_error; | |
377 | } | |
378 | ||
33479e05 DC |
379 | /* |
380 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode | |
381 | * from stomping over us while we recycle the inode. We can't | |
382 | * clear the radix tree reclaimable tag yet as it requires | |
383 | * pag_ici_lock to be held exclusive. | |
384 | */ | |
385 | ip->i_flags |= XFS_IRECLAIM; | |
386 | ||
387 | spin_unlock(&ip->i_flags_lock); | |
388 | rcu_read_unlock(); | |
389 | ||
50997470 | 390 | error = xfs_reinit_inode(mp, inode); |
33479e05 | 391 | if (error) { |
756baca2 | 392 | bool wake; |
33479e05 DC |
393 | /* |
394 | * Re-initializing the inode failed, and we are in deep | |
395 | * trouble. Try to re-add it to the reclaim list. | |
396 | */ | |
397 | rcu_read_lock(); | |
398 | spin_lock(&ip->i_flags_lock); | |
756baca2 | 399 | wake = !!__xfs_iflags_test(ip, XFS_INEW); |
33479e05 | 400 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
756baca2 BF |
401 | if (wake) |
402 | wake_up_bit(&ip->i_flags, __XFS_INEW_BIT); | |
33479e05 DC |
403 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); |
404 | trace_xfs_iget_reclaim_fail(ip); | |
405 | goto out_error; | |
406 | } | |
407 | ||
408 | spin_lock(&pag->pag_ici_lock); | |
409 | spin_lock(&ip->i_flags_lock); | |
410 | ||
411 | /* | |
412 | * Clear the per-lifetime state in the inode as we are now | |
413 | * effectively a new inode and need to return to the initial | |
414 | * state before reuse occurs. | |
415 | */ | |
416 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
417 | ip->i_flags |= XFS_INEW; | |
545c0889 | 418 | xfs_inode_clear_reclaim_tag(pag, ip->i_ino); |
33479e05 DC |
419 | inode->i_state = I_NEW; |
420 | ||
65523218 CH |
421 | ASSERT(!rwsem_is_locked(&inode->i_rwsem)); |
422 | init_rwsem(&inode->i_rwsem); | |
33479e05 DC |
423 | |
424 | spin_unlock(&ip->i_flags_lock); | |
425 | spin_unlock(&pag->pag_ici_lock); | |
426 | } else { | |
427 | /* If the VFS inode is being torn down, pause and try again. */ | |
428 | if (!igrab(inode)) { | |
429 | trace_xfs_iget_skip(ip); | |
2451337d | 430 | error = -EAGAIN; |
33479e05 DC |
431 | goto out_error; |
432 | } | |
433 | ||
434 | /* We've got a live one. */ | |
435 | spin_unlock(&ip->i_flags_lock); | |
436 | rcu_read_unlock(); | |
437 | trace_xfs_iget_hit(ip); | |
438 | } | |
439 | ||
440 | if (lock_flags != 0) | |
441 | xfs_ilock(ip, lock_flags); | |
442 | ||
378f681c DW |
443 | if (!(flags & XFS_IGET_INCORE)) |
444 | xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE); | |
ff6d6af2 | 445 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
446 | |
447 | return 0; | |
448 | ||
449 | out_error: | |
450 | spin_unlock(&ip->i_flags_lock); | |
451 | rcu_read_unlock(); | |
452 | return error; | |
453 | } | |
454 | ||
455 | ||
456 | static int | |
457 | xfs_iget_cache_miss( | |
458 | struct xfs_mount *mp, | |
459 | struct xfs_perag *pag, | |
460 | xfs_trans_t *tp, | |
461 | xfs_ino_t ino, | |
462 | struct xfs_inode **ipp, | |
463 | int flags, | |
464 | int lock_flags) | |
465 | { | |
466 | struct xfs_inode *ip; | |
467 | int error; | |
468 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
469 | int iflags; | |
470 | ||
471 | ip = xfs_inode_alloc(mp, ino); | |
472 | if (!ip) | |
2451337d | 473 | return -ENOMEM; |
33479e05 DC |
474 | |
475 | error = xfs_iread(mp, tp, ip, flags); | |
476 | if (error) | |
477 | goto out_destroy; | |
478 | ||
9cfb9b47 DW |
479 | if (!xfs_inode_verify_forks(ip)) { |
480 | error = -EFSCORRUPTED; | |
481 | goto out_destroy; | |
482 | } | |
483 | ||
33479e05 DC |
484 | trace_xfs_iget_miss(ip); |
485 | ||
c19b3b05 | 486 | if ((VFS_I(ip)->i_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
2451337d | 487 | error = -ENOENT; |
33479e05 DC |
488 | goto out_destroy; |
489 | } | |
490 | ||
491 | /* | |
492 | * Preload the radix tree so we can insert safely under the | |
493 | * write spinlock. Note that we cannot sleep inside the preload | |
494 | * region. Since we can be called from transaction context, don't | |
495 | * recurse into the file system. | |
496 | */ | |
497 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 498 | error = -EAGAIN; |
33479e05 DC |
499 | goto out_destroy; |
500 | } | |
501 | ||
502 | /* | |
503 | * Because the inode hasn't been added to the radix-tree yet it can't | |
504 | * be found by another thread, so we can do the non-sleeping lock here. | |
505 | */ | |
506 | if (lock_flags) { | |
507 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
508 | BUG(); | |
509 | } | |
510 | ||
511 | /* | |
512 | * These values must be set before inserting the inode into the radix | |
513 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
514 | * RCU locking mechanism) can find it and that lookup must see that this | |
515 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
516 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
517 | * memory barrier that ensures this detection works correctly at lookup | |
518 | * time. | |
519 | */ | |
520 | iflags = XFS_INEW; | |
521 | if (flags & XFS_IGET_DONTCACHE) | |
522 | iflags |= XFS_IDONTCACHE; | |
113a5683 CS |
523 | ip->i_udquot = NULL; |
524 | ip->i_gdquot = NULL; | |
92f8ff73 | 525 | ip->i_pdquot = NULL; |
33479e05 DC |
526 | xfs_iflags_set(ip, iflags); |
527 | ||
528 | /* insert the new inode */ | |
529 | spin_lock(&pag->pag_ici_lock); | |
530 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
531 | if (unlikely(error)) { | |
532 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 533 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 534 | error = -EAGAIN; |
33479e05 DC |
535 | goto out_preload_end; |
536 | } | |
537 | spin_unlock(&pag->pag_ici_lock); | |
538 | radix_tree_preload_end(); | |
539 | ||
540 | *ipp = ip; | |
541 | return 0; | |
542 | ||
543 | out_preload_end: | |
544 | spin_unlock(&pag->pag_ici_lock); | |
545 | radix_tree_preload_end(); | |
546 | if (lock_flags) | |
547 | xfs_iunlock(ip, lock_flags); | |
548 | out_destroy: | |
549 | __destroy_inode(VFS_I(ip)); | |
550 | xfs_inode_free(ip); | |
551 | return error; | |
552 | } | |
553 | ||
554 | /* | |
555 | * Look up an inode by number in the given file system. | |
556 | * The inode is looked up in the cache held in each AG. | |
557 | * If the inode is found in the cache, initialise the vfs inode | |
558 | * if necessary. | |
559 | * | |
560 | * If it is not in core, read it in from the file system's device, | |
561 | * add it to the cache and initialise the vfs inode. | |
562 | * | |
563 | * The inode is locked according to the value of the lock_flags parameter. | |
564 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
565 | * should be taken. | |
566 | * | |
567 | * mp -- the mount point structure for the current file system. It points | |
568 | * to the inode hash table. | |
569 | * tp -- a pointer to the current transaction if there is one. This is | |
570 | * simply passed through to the xfs_iread() call. | |
571 | * ino -- the number of the inode desired. This is the unique identifier | |
572 | * within the file system for the inode being requested. | |
573 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
574 | * for xfs_ilock() for a list of valid values. | |
575 | */ | |
576 | int | |
577 | xfs_iget( | |
578 | xfs_mount_t *mp, | |
579 | xfs_trans_t *tp, | |
580 | xfs_ino_t ino, | |
581 | uint flags, | |
582 | uint lock_flags, | |
583 | xfs_inode_t **ipp) | |
584 | { | |
585 | xfs_inode_t *ip; | |
586 | int error; | |
587 | xfs_perag_t *pag; | |
588 | xfs_agino_t agino; | |
589 | ||
590 | /* | |
591 | * xfs_reclaim_inode() uses the ILOCK to ensure an inode | |
592 | * doesn't get freed while it's being referenced during a | |
593 | * radix tree traversal here. It assumes this function | |
594 | * aqcuires only the ILOCK (and therefore it has no need to | |
595 | * involve the IOLOCK in this synchronization). | |
596 | */ | |
597 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); | |
598 | ||
599 | /* reject inode numbers outside existing AGs */ | |
600 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 601 | return -EINVAL; |
33479e05 | 602 | |
ff6d6af2 | 603 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 604 | |
33479e05 DC |
605 | /* get the perag structure and ensure that it's inode capable */ |
606 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
607 | agino = XFS_INO_TO_AGINO(mp, ino); | |
608 | ||
609 | again: | |
610 | error = 0; | |
611 | rcu_read_lock(); | |
612 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
613 | ||
614 | if (ip) { | |
615 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
616 | if (error) | |
617 | goto out_error_or_again; | |
618 | } else { | |
619 | rcu_read_unlock(); | |
378f681c | 620 | if (flags & XFS_IGET_INCORE) { |
ed438b47 | 621 | error = -ENODATA; |
378f681c DW |
622 | goto out_error_or_again; |
623 | } | |
ff6d6af2 | 624 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
625 | |
626 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
627 | flags, lock_flags); | |
628 | if (error) | |
629 | goto out_error_or_again; | |
630 | } | |
631 | xfs_perag_put(pag); | |
632 | ||
633 | *ipp = ip; | |
634 | ||
635 | /* | |
58c90473 | 636 | * If we have a real type for an on-disk inode, we can setup the inode |
33479e05 DC |
637 | * now. If it's a new inode being created, xfs_ialloc will handle it. |
638 | */ | |
c19b3b05 | 639 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 640 | xfs_setup_existing_inode(ip); |
33479e05 DC |
641 | return 0; |
642 | ||
643 | out_error_or_again: | |
378f681c | 644 | if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) { |
33479e05 DC |
645 | delay(1); |
646 | goto again; | |
647 | } | |
648 | xfs_perag_put(pag); | |
649 | return error; | |
650 | } | |
651 | ||
378f681c DW |
652 | /* |
653 | * "Is this a cached inode that's also allocated?" | |
654 | * | |
655 | * Look up an inode by number in the given file system. If the inode is | |
656 | * in cache and isn't in purgatory, return 1 if the inode is allocated | |
657 | * and 0 if it is not. For all other cases (not in cache, being torn | |
658 | * down, etc.), return a negative error code. | |
659 | * | |
660 | * The caller has to prevent inode allocation and freeing activity, | |
661 | * presumably by locking the AGI buffer. This is to ensure that an | |
662 | * inode cannot transition from allocated to freed until the caller is | |
663 | * ready to allow that. If the inode is in an intermediate state (new, | |
664 | * reclaimable, or being reclaimed), -EAGAIN will be returned; if the | |
665 | * inode is not in the cache, -ENOENT will be returned. The caller must | |
666 | * deal with these scenarios appropriately. | |
667 | * | |
668 | * This is a specialized use case for the online scrubber; if you're | |
669 | * reading this, you probably want xfs_iget. | |
670 | */ | |
671 | int | |
672 | xfs_icache_inode_is_allocated( | |
673 | struct xfs_mount *mp, | |
674 | struct xfs_trans *tp, | |
675 | xfs_ino_t ino, | |
676 | bool *inuse) | |
677 | { | |
678 | struct xfs_inode *ip; | |
679 | int error; | |
680 | ||
681 | error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip); | |
682 | if (error) | |
683 | return error; | |
684 | ||
685 | *inuse = !!(VFS_I(ip)->i_mode); | |
686 | IRELE(ip); | |
687 | return 0; | |
688 | } | |
689 | ||
78ae5256 DC |
690 | /* |
691 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
692 | * radix tree lookups to a minimum. The batch size is a trade off between | |
693 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
694 | * be too greedy. | |
695 | */ | |
696 | #define XFS_LOOKUP_BATCH 32 | |
697 | ||
e13de955 DC |
698 | STATIC int |
699 | xfs_inode_ag_walk_grab( | |
ae2c4ac2 BF |
700 | struct xfs_inode *ip, |
701 | int flags) | |
e13de955 DC |
702 | { |
703 | struct inode *inode = VFS_I(ip); | |
ae2c4ac2 | 704 | bool newinos = !!(flags & XFS_AGITER_INEW_WAIT); |
e13de955 | 705 | |
1a3e8f3d DC |
706 | ASSERT(rcu_read_lock_held()); |
707 | ||
708 | /* | |
709 | * check for stale RCU freed inode | |
710 | * | |
711 | * If the inode has been reallocated, it doesn't matter if it's not in | |
712 | * the AG we are walking - we are walking for writeback, so if it | |
713 | * passes all the "valid inode" checks and is dirty, then we'll write | |
714 | * it back anyway. If it has been reallocated and still being | |
715 | * initialised, the XFS_INEW check below will catch it. | |
716 | */ | |
717 | spin_lock(&ip->i_flags_lock); | |
718 | if (!ip->i_ino) | |
719 | goto out_unlock_noent; | |
720 | ||
721 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
ae2c4ac2 BF |
722 | if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) || |
723 | __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
1a3e8f3d DC |
724 | goto out_unlock_noent; |
725 | spin_unlock(&ip->i_flags_lock); | |
726 | ||
e13de955 DC |
727 | /* nothing to sync during shutdown */ |
728 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
2451337d | 729 | return -EFSCORRUPTED; |
e13de955 | 730 | |
e13de955 DC |
731 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
732 | if (!igrab(inode)) | |
2451337d | 733 | return -ENOENT; |
e13de955 | 734 | |
e13de955 DC |
735 | /* inode is valid */ |
736 | return 0; | |
1a3e8f3d DC |
737 | |
738 | out_unlock_noent: | |
739 | spin_unlock(&ip->i_flags_lock); | |
2451337d | 740 | return -ENOENT; |
e13de955 DC |
741 | } |
742 | ||
75f3cb13 DC |
743 | STATIC int |
744 | xfs_inode_ag_walk( | |
745 | struct xfs_mount *mp, | |
5017e97d | 746 | struct xfs_perag *pag, |
e0094008 | 747 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
748 | void *args), |
749 | int flags, | |
750 | void *args, | |
ae2c4ac2 BF |
751 | int tag, |
752 | int iter_flags) | |
75f3cb13 | 753 | { |
75f3cb13 DC |
754 | uint32_t first_index; |
755 | int last_error = 0; | |
756 | int skipped; | |
65d0f205 | 757 | int done; |
78ae5256 | 758 | int nr_found; |
75f3cb13 DC |
759 | |
760 | restart: | |
65d0f205 | 761 | done = 0; |
75f3cb13 DC |
762 | skipped = 0; |
763 | first_index = 0; | |
78ae5256 | 764 | nr_found = 0; |
75f3cb13 | 765 | do { |
78ae5256 | 766 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
75f3cb13 | 767 | int error = 0; |
78ae5256 | 768 | int i; |
75f3cb13 | 769 | |
1a3e8f3d | 770 | rcu_read_lock(); |
a454f742 BF |
771 | |
772 | if (tag == -1) | |
773 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
78ae5256 DC |
774 | (void **)batch, first_index, |
775 | XFS_LOOKUP_BATCH); | |
a454f742 BF |
776 | else |
777 | nr_found = radix_tree_gang_lookup_tag( | |
778 | &pag->pag_ici_root, | |
779 | (void **) batch, first_index, | |
780 | XFS_LOOKUP_BATCH, tag); | |
781 | ||
65d0f205 | 782 | if (!nr_found) { |
1a3e8f3d | 783 | rcu_read_unlock(); |
75f3cb13 | 784 | break; |
c8e20be0 | 785 | } |
75f3cb13 | 786 | |
65d0f205 | 787 | /* |
78ae5256 DC |
788 | * Grab the inodes before we drop the lock. if we found |
789 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 790 | */ |
78ae5256 DC |
791 | for (i = 0; i < nr_found; i++) { |
792 | struct xfs_inode *ip = batch[i]; | |
793 | ||
ae2c4ac2 | 794 | if (done || xfs_inode_ag_walk_grab(ip, iter_flags)) |
78ae5256 DC |
795 | batch[i] = NULL; |
796 | ||
797 | /* | |
1a3e8f3d DC |
798 | * Update the index for the next lookup. Catch |
799 | * overflows into the next AG range which can occur if | |
800 | * we have inodes in the last block of the AG and we | |
801 | * are currently pointing to the last inode. | |
802 | * | |
803 | * Because we may see inodes that are from the wrong AG | |
804 | * due to RCU freeing and reallocation, only update the | |
805 | * index if it lies in this AG. It was a race that lead | |
806 | * us to see this inode, so another lookup from the | |
807 | * same index will not find it again. | |
78ae5256 | 808 | */ |
1a3e8f3d DC |
809 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
810 | continue; | |
78ae5256 DC |
811 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
812 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
813 | done = 1; | |
e13de955 | 814 | } |
78ae5256 DC |
815 | |
816 | /* unlock now we've grabbed the inodes. */ | |
1a3e8f3d | 817 | rcu_read_unlock(); |
e13de955 | 818 | |
78ae5256 DC |
819 | for (i = 0; i < nr_found; i++) { |
820 | if (!batch[i]) | |
821 | continue; | |
ae2c4ac2 BF |
822 | if ((iter_flags & XFS_AGITER_INEW_WAIT) && |
823 | xfs_iflags_test(batch[i], XFS_INEW)) | |
824 | xfs_inew_wait(batch[i]); | |
e0094008 | 825 | error = execute(batch[i], flags, args); |
78ae5256 | 826 | IRELE(batch[i]); |
2451337d | 827 | if (error == -EAGAIN) { |
78ae5256 DC |
828 | skipped++; |
829 | continue; | |
830 | } | |
2451337d | 831 | if (error && last_error != -EFSCORRUPTED) |
78ae5256 | 832 | last_error = error; |
75f3cb13 | 833 | } |
c8e20be0 DC |
834 | |
835 | /* bail out if the filesystem is corrupted. */ | |
2451337d | 836 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
837 | break; |
838 | ||
8daaa831 DC |
839 | cond_resched(); |
840 | ||
78ae5256 | 841 | } while (nr_found && !done); |
75f3cb13 DC |
842 | |
843 | if (skipped) { | |
844 | delay(1); | |
845 | goto restart; | |
846 | } | |
75f3cb13 DC |
847 | return last_error; |
848 | } | |
849 | ||
579b62fa BF |
850 | /* |
851 | * Background scanning to trim post-EOF preallocated space. This is queued | |
b9fe5052 | 852 | * based on the 'speculative_prealloc_lifetime' tunable (5m by default). |
579b62fa | 853 | */ |
fa5a4f57 | 854 | void |
579b62fa BF |
855 | xfs_queue_eofblocks( |
856 | struct xfs_mount *mp) | |
857 | { | |
858 | rcu_read_lock(); | |
859 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG)) | |
860 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
861 | &mp->m_eofblocks_work, | |
862 | msecs_to_jiffies(xfs_eofb_secs * 1000)); | |
863 | rcu_read_unlock(); | |
864 | } | |
865 | ||
866 | void | |
867 | xfs_eofblocks_worker( | |
868 | struct work_struct *work) | |
869 | { | |
870 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
871 | struct xfs_mount, m_eofblocks_work); | |
872 | xfs_icache_free_eofblocks(mp, NULL); | |
873 | xfs_queue_eofblocks(mp); | |
874 | } | |
875 | ||
83104d44 DW |
876 | /* |
877 | * Background scanning to trim preallocated CoW space. This is queued | |
878 | * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default). | |
879 | * (We'll just piggyback on the post-EOF prealloc space workqueue.) | |
880 | */ | |
10ddf64e | 881 | void |
83104d44 DW |
882 | xfs_queue_cowblocks( |
883 | struct xfs_mount *mp) | |
884 | { | |
885 | rcu_read_lock(); | |
886 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG)) | |
887 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
888 | &mp->m_cowblocks_work, | |
889 | msecs_to_jiffies(xfs_cowb_secs * 1000)); | |
890 | rcu_read_unlock(); | |
891 | } | |
892 | ||
893 | void | |
894 | xfs_cowblocks_worker( | |
895 | struct work_struct *work) | |
896 | { | |
897 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
898 | struct xfs_mount, m_cowblocks_work); | |
899 | xfs_icache_free_cowblocks(mp, NULL); | |
900 | xfs_queue_cowblocks(mp); | |
901 | } | |
902 | ||
fe588ed3 | 903 | int |
ae2c4ac2 | 904 | xfs_inode_ag_iterator_flags( |
75f3cb13 | 905 | struct xfs_mount *mp, |
e0094008 | 906 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
907 | void *args), |
908 | int flags, | |
ae2c4ac2 BF |
909 | void *args, |
910 | int iter_flags) | |
75f3cb13 | 911 | { |
16fd5367 | 912 | struct xfs_perag *pag; |
75f3cb13 DC |
913 | int error = 0; |
914 | int last_error = 0; | |
915 | xfs_agnumber_t ag; | |
916 | ||
16fd5367 | 917 | ag = 0; |
65d0f205 DC |
918 | while ((pag = xfs_perag_get(mp, ag))) { |
919 | ag = pag->pag_agno + 1; | |
ae2c4ac2 BF |
920 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1, |
921 | iter_flags); | |
a454f742 BF |
922 | xfs_perag_put(pag); |
923 | if (error) { | |
924 | last_error = error; | |
2451337d | 925 | if (error == -EFSCORRUPTED) |
a454f742 BF |
926 | break; |
927 | } | |
928 | } | |
b474c7ae | 929 | return last_error; |
a454f742 BF |
930 | } |
931 | ||
ae2c4ac2 BF |
932 | int |
933 | xfs_inode_ag_iterator( | |
934 | struct xfs_mount *mp, | |
935 | int (*execute)(struct xfs_inode *ip, int flags, | |
936 | void *args), | |
937 | int flags, | |
938 | void *args) | |
939 | { | |
940 | return xfs_inode_ag_iterator_flags(mp, execute, flags, args, 0); | |
941 | } | |
942 | ||
a454f742 BF |
943 | int |
944 | xfs_inode_ag_iterator_tag( | |
945 | struct xfs_mount *mp, | |
e0094008 | 946 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
947 | void *args), |
948 | int flags, | |
949 | void *args, | |
950 | int tag) | |
951 | { | |
952 | struct xfs_perag *pag; | |
953 | int error = 0; | |
954 | int last_error = 0; | |
955 | xfs_agnumber_t ag; | |
956 | ||
957 | ag = 0; | |
958 | while ((pag = xfs_perag_get_tag(mp, ag, tag))) { | |
959 | ag = pag->pag_agno + 1; | |
ae2c4ac2 BF |
960 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag, |
961 | 0); | |
5017e97d | 962 | xfs_perag_put(pag); |
75f3cb13 DC |
963 | if (error) { |
964 | last_error = error; | |
2451337d | 965 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
966 | break; |
967 | } | |
968 | } | |
b474c7ae | 969 | return last_error; |
75f3cb13 DC |
970 | } |
971 | ||
e3a20c0b DC |
972 | /* |
973 | * Grab the inode for reclaim exclusively. | |
974 | * Return 0 if we grabbed it, non-zero otherwise. | |
975 | */ | |
976 | STATIC int | |
977 | xfs_reclaim_inode_grab( | |
978 | struct xfs_inode *ip, | |
979 | int flags) | |
980 | { | |
1a3e8f3d DC |
981 | ASSERT(rcu_read_lock_held()); |
982 | ||
983 | /* quick check for stale RCU freed inode */ | |
984 | if (!ip->i_ino) | |
985 | return 1; | |
e3a20c0b DC |
986 | |
987 | /* | |
474fce06 CH |
988 | * If we are asked for non-blocking operation, do unlocked checks to |
989 | * see if the inode already is being flushed or in reclaim to avoid | |
990 | * lock traffic. | |
e3a20c0b DC |
991 | */ |
992 | if ((flags & SYNC_TRYLOCK) && | |
474fce06 | 993 | __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM)) |
e3a20c0b | 994 | return 1; |
e3a20c0b DC |
995 | |
996 | /* | |
997 | * The radix tree lock here protects a thread in xfs_iget from racing | |
998 | * with us starting reclaim on the inode. Once we have the | |
999 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
1000 | * |
1001 | * Due to RCU lookup, we may find inodes that have been freed and only | |
1002 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
1003 | * aren't candidates for reclaim at all, so we must check the | |
1004 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
1005 | */ |
1006 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
1007 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
1008 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
1009 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
1010 | spin_unlock(&ip->i_flags_lock); |
1011 | return 1; | |
1012 | } | |
1013 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
1014 | spin_unlock(&ip->i_flags_lock); | |
1015 | return 0; | |
1016 | } | |
1017 | ||
777df5af | 1018 | /* |
8a48088f CH |
1019 | * Inodes in different states need to be treated differently. The following |
1020 | * table lists the inode states and the reclaim actions necessary: | |
777df5af DC |
1021 | * |
1022 | * inode state iflush ret required action | |
1023 | * --------------- ---------- --------------- | |
1024 | * bad - reclaim | |
1025 | * shutdown EIO unpin and reclaim | |
1026 | * clean, unpinned 0 reclaim | |
1027 | * stale, unpinned 0 reclaim | |
c854363e DC |
1028 | * clean, pinned(*) 0 requeue |
1029 | * stale, pinned EAGAIN requeue | |
8a48088f CH |
1030 | * dirty, async - requeue |
1031 | * dirty, sync 0 reclaim | |
777df5af DC |
1032 | * |
1033 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
1034 | * handled anyway given the order of checks implemented. | |
1035 | * | |
c854363e DC |
1036 | * Also, because we get the flush lock first, we know that any inode that has |
1037 | * been flushed delwri has had the flush completed by the time we check that | |
8a48088f | 1038 | * the inode is clean. |
c854363e | 1039 | * |
8a48088f CH |
1040 | * Note that because the inode is flushed delayed write by AIL pushing, the |
1041 | * flush lock may already be held here and waiting on it can result in very | |
1042 | * long latencies. Hence for sync reclaims, where we wait on the flush lock, | |
1043 | * the caller should push the AIL first before trying to reclaim inodes to | |
1044 | * minimise the amount of time spent waiting. For background relaim, we only | |
1045 | * bother to reclaim clean inodes anyway. | |
c854363e | 1046 | * |
777df5af DC |
1047 | * Hence the order of actions after gaining the locks should be: |
1048 | * bad => reclaim | |
1049 | * shutdown => unpin and reclaim | |
8a48088f | 1050 | * pinned, async => requeue |
c854363e | 1051 | * pinned, sync => unpin |
777df5af DC |
1052 | * stale => reclaim |
1053 | * clean => reclaim | |
8a48088f | 1054 | * dirty, async => requeue |
c854363e | 1055 | * dirty, sync => flush, wait and reclaim |
777df5af | 1056 | */ |
75f3cb13 | 1057 | STATIC int |
c8e20be0 | 1058 | xfs_reclaim_inode( |
75f3cb13 DC |
1059 | struct xfs_inode *ip, |
1060 | struct xfs_perag *pag, | |
c8e20be0 | 1061 | int sync_mode) |
fce08f2f | 1062 | { |
4c46819a | 1063 | struct xfs_buf *bp = NULL; |
8a17d7dd | 1064 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
4c46819a | 1065 | int error; |
777df5af | 1066 | |
1bfd8d04 DC |
1067 | restart: |
1068 | error = 0; | |
c8e20be0 | 1069 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
1070 | if (!xfs_iflock_nowait(ip)) { |
1071 | if (!(sync_mode & SYNC_WAIT)) | |
1072 | goto out; | |
1073 | xfs_iflock(ip); | |
1074 | } | |
7a3be02b | 1075 | |
777df5af DC |
1076 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1077 | xfs_iunpin_wait(ip); | |
98efe8af | 1078 | /* xfs_iflush_abort() drops the flush lock */ |
04913fdd | 1079 | xfs_iflush_abort(ip, false); |
777df5af DC |
1080 | goto reclaim; |
1081 | } | |
c854363e | 1082 | if (xfs_ipincount(ip)) { |
8a48088f CH |
1083 | if (!(sync_mode & SYNC_WAIT)) |
1084 | goto out_ifunlock; | |
777df5af | 1085 | xfs_iunpin_wait(ip); |
c854363e | 1086 | } |
98efe8af BF |
1087 | if (xfs_iflags_test(ip, XFS_ISTALE) || xfs_inode_clean(ip)) { |
1088 | xfs_ifunlock(ip); | |
777df5af | 1089 | goto reclaim; |
98efe8af | 1090 | } |
777df5af | 1091 | |
8a48088f CH |
1092 | /* |
1093 | * Never flush out dirty data during non-blocking reclaim, as it would | |
1094 | * just contend with AIL pushing trying to do the same job. | |
1095 | */ | |
1096 | if (!(sync_mode & SYNC_WAIT)) | |
1097 | goto out_ifunlock; | |
1098 | ||
1bfd8d04 DC |
1099 | /* |
1100 | * Now we have an inode that needs flushing. | |
1101 | * | |
4c46819a | 1102 | * Note that xfs_iflush will never block on the inode buffer lock, as |
1bfd8d04 | 1103 | * xfs_ifree_cluster() can lock the inode buffer before it locks the |
4c46819a | 1104 | * ip->i_lock, and we are doing the exact opposite here. As a result, |
475ee413 CH |
1105 | * doing a blocking xfs_imap_to_bp() to get the cluster buffer would |
1106 | * result in an ABBA deadlock with xfs_ifree_cluster(). | |
1bfd8d04 DC |
1107 | * |
1108 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
1109 | * cache to mark them stale, if we hit this case we don't actually want | |
1110 | * to do IO here - we want the inode marked stale so we can simply | |
4c46819a CH |
1111 | * reclaim it. Hence if we get an EAGAIN error here, just unlock the |
1112 | * inode, back off and try again. Hopefully the next pass through will | |
1113 | * see the stale flag set on the inode. | |
1bfd8d04 | 1114 | */ |
4c46819a | 1115 | error = xfs_iflush(ip, &bp); |
2451337d | 1116 | if (error == -EAGAIN) { |
8a48088f CH |
1117 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1118 | /* backoff longer than in xfs_ifree_cluster */ | |
1119 | delay(2); | |
1120 | goto restart; | |
c854363e | 1121 | } |
c854363e | 1122 | |
4c46819a CH |
1123 | if (!error) { |
1124 | error = xfs_bwrite(bp); | |
1125 | xfs_buf_relse(bp); | |
1126 | } | |
1127 | ||
777df5af | 1128 | reclaim: |
98efe8af BF |
1129 | ASSERT(!xfs_isiflocked(ip)); |
1130 | ||
8a17d7dd DC |
1131 | /* |
1132 | * Because we use RCU freeing we need to ensure the inode always appears | |
1133 | * to be reclaimed with an invalid inode number when in the free state. | |
98efe8af | 1134 | * We do this as early as possible under the ILOCK so that |
f2e9ad21 OS |
1135 | * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to |
1136 | * detect races with us here. By doing this, we guarantee that once | |
1137 | * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that | |
1138 | * it will see either a valid inode that will serialise correctly, or it | |
1139 | * will see an invalid inode that it can skip. | |
8a17d7dd DC |
1140 | */ |
1141 | spin_lock(&ip->i_flags_lock); | |
1142 | ip->i_flags = XFS_IRECLAIM; | |
1143 | ip->i_ino = 0; | |
1144 | spin_unlock(&ip->i_flags_lock); | |
1145 | ||
c8e20be0 | 1146 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1147 | |
ff6d6af2 | 1148 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
1149 | /* |
1150 | * Remove the inode from the per-AG radix tree. | |
1151 | * | |
1152 | * Because radix_tree_delete won't complain even if the item was never | |
1153 | * added to the tree assert that it's been there before to catch | |
1154 | * problems with the inode life time early on. | |
1155 | */ | |
1a427ab0 | 1156 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 1157 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 1158 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 1159 | ASSERT(0); |
545c0889 | 1160 | xfs_perag_clear_reclaim_tag(pag); |
1a427ab0 | 1161 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
1162 | |
1163 | /* | |
1164 | * Here we do an (almost) spurious inode lock in order to coordinate | |
1165 | * with inode cache radix tree lookups. This is because the lookup | |
1166 | * can reference the inodes in the cache without taking references. | |
1167 | * | |
1168 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 1169 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 1170 | */ |
ad637a10 | 1171 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1172 | xfs_qm_dqdetach(ip); |
ad637a10 | 1173 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1174 | |
8a17d7dd | 1175 | __xfs_inode_free(ip); |
ad637a10 | 1176 | return error; |
8a48088f CH |
1177 | |
1178 | out_ifunlock: | |
1179 | xfs_ifunlock(ip); | |
1180 | out: | |
1181 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
1182 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1183 | /* | |
2451337d | 1184 | * We could return -EAGAIN here to make reclaim rescan the inode tree in |
8a48088f | 1185 | * a short while. However, this just burns CPU time scanning the tree |
5889608d DC |
1186 | * waiting for IO to complete and the reclaim work never goes back to |
1187 | * the idle state. Instead, return 0 to let the next scheduled | |
1188 | * background reclaim attempt to reclaim the inode again. | |
8a48088f CH |
1189 | */ |
1190 | return 0; | |
7a3be02b DC |
1191 | } |
1192 | ||
65d0f205 DC |
1193 | /* |
1194 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
1195 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
1196 | * then a shut down during filesystem unmount reclaim walk leak all the | |
1197 | * unreclaimed inodes. | |
1198 | */ | |
33479e05 | 1199 | STATIC int |
65d0f205 DC |
1200 | xfs_reclaim_inodes_ag( |
1201 | struct xfs_mount *mp, | |
1202 | int flags, | |
1203 | int *nr_to_scan) | |
1204 | { | |
1205 | struct xfs_perag *pag; | |
1206 | int error = 0; | |
1207 | int last_error = 0; | |
1208 | xfs_agnumber_t ag; | |
69b491c2 DC |
1209 | int trylock = flags & SYNC_TRYLOCK; |
1210 | int skipped; | |
65d0f205 | 1211 | |
69b491c2 | 1212 | restart: |
65d0f205 | 1213 | ag = 0; |
69b491c2 | 1214 | skipped = 0; |
65d0f205 DC |
1215 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1216 | unsigned long first_index = 0; | |
1217 | int done = 0; | |
e3a20c0b | 1218 | int nr_found = 0; |
65d0f205 DC |
1219 | |
1220 | ag = pag->pag_agno + 1; | |
1221 | ||
69b491c2 DC |
1222 | if (trylock) { |
1223 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
1224 | skipped++; | |
f83282a8 | 1225 | xfs_perag_put(pag); |
69b491c2 DC |
1226 | continue; |
1227 | } | |
1228 | first_index = pag->pag_ici_reclaim_cursor; | |
1229 | } else | |
1230 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
1231 | ||
65d0f205 | 1232 | do { |
e3a20c0b DC |
1233 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
1234 | int i; | |
65d0f205 | 1235 | |
1a3e8f3d | 1236 | rcu_read_lock(); |
e3a20c0b DC |
1237 | nr_found = radix_tree_gang_lookup_tag( |
1238 | &pag->pag_ici_root, | |
1239 | (void **)batch, first_index, | |
1240 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
1241 | XFS_ICI_RECLAIM_TAG); |
1242 | if (!nr_found) { | |
b2232219 | 1243 | done = 1; |
1a3e8f3d | 1244 | rcu_read_unlock(); |
65d0f205 DC |
1245 | break; |
1246 | } | |
1247 | ||
1248 | /* | |
e3a20c0b DC |
1249 | * Grab the inodes before we drop the lock. if we found |
1250 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 1251 | */ |
e3a20c0b DC |
1252 | for (i = 0; i < nr_found; i++) { |
1253 | struct xfs_inode *ip = batch[i]; | |
1254 | ||
1255 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
1256 | batch[i] = NULL; | |
1257 | ||
1258 | /* | |
1259 | * Update the index for the next lookup. Catch | |
1260 | * overflows into the next AG range which can | |
1261 | * occur if we have inodes in the last block of | |
1262 | * the AG and we are currently pointing to the | |
1263 | * last inode. | |
1a3e8f3d DC |
1264 | * |
1265 | * Because we may see inodes that are from the | |
1266 | * wrong AG due to RCU freeing and | |
1267 | * reallocation, only update the index if it | |
1268 | * lies in this AG. It was a race that lead us | |
1269 | * to see this inode, so another lookup from | |
1270 | * the same index will not find it again. | |
e3a20c0b | 1271 | */ |
1a3e8f3d DC |
1272 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
1273 | pag->pag_agno) | |
1274 | continue; | |
e3a20c0b DC |
1275 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
1276 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1277 | done = 1; | |
1278 | } | |
65d0f205 | 1279 | |
e3a20c0b | 1280 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 1281 | rcu_read_unlock(); |
e3a20c0b DC |
1282 | |
1283 | for (i = 0; i < nr_found; i++) { | |
1284 | if (!batch[i]) | |
1285 | continue; | |
1286 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
2451337d | 1287 | if (error && last_error != -EFSCORRUPTED) |
e3a20c0b DC |
1288 | last_error = error; |
1289 | } | |
1290 | ||
1291 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 1292 | |
8daaa831 DC |
1293 | cond_resched(); |
1294 | ||
e3a20c0b | 1295 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 1296 | |
69b491c2 DC |
1297 | if (trylock && !done) |
1298 | pag->pag_ici_reclaim_cursor = first_index; | |
1299 | else | |
1300 | pag->pag_ici_reclaim_cursor = 0; | |
1301 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
1302 | xfs_perag_put(pag); |
1303 | } | |
69b491c2 DC |
1304 | |
1305 | /* | |
1306 | * if we skipped any AG, and we still have scan count remaining, do | |
1307 | * another pass this time using blocking reclaim semantics (i.e | |
1308 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1309 | * ensure that when we get more reclaimers than AGs we block rather | |
1310 | * than spin trying to execute reclaim. | |
1311 | */ | |
8daaa831 | 1312 | if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) { |
69b491c2 DC |
1313 | trylock = 0; |
1314 | goto restart; | |
1315 | } | |
b474c7ae | 1316 | return last_error; |
65d0f205 DC |
1317 | } |
1318 | ||
7a3be02b DC |
1319 | int |
1320 | xfs_reclaim_inodes( | |
1321 | xfs_mount_t *mp, | |
7a3be02b DC |
1322 | int mode) |
1323 | { | |
65d0f205 DC |
1324 | int nr_to_scan = INT_MAX; |
1325 | ||
1326 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1327 | } |
1328 | ||
1329 | /* | |
8daaa831 | 1330 | * Scan a certain number of inodes for reclaim. |
a7b339f1 DC |
1331 | * |
1332 | * When called we make sure that there is a background (fast) inode reclaim in | |
8daaa831 | 1333 | * progress, while we will throttle the speed of reclaim via doing synchronous |
a7b339f1 DC |
1334 | * reclaim of inodes. That means if we come across dirty inodes, we wait for |
1335 | * them to be cleaned, which we hope will not be very long due to the | |
1336 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1337 | */ |
0a234c6d | 1338 | long |
8daaa831 DC |
1339 | xfs_reclaim_inodes_nr( |
1340 | struct xfs_mount *mp, | |
1341 | int nr_to_scan) | |
9bf729c0 | 1342 | { |
8daaa831 | 1343 | /* kick background reclaimer and push the AIL */ |
5889608d | 1344 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1345 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1346 | |
0a234c6d | 1347 | return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan); |
8daaa831 | 1348 | } |
9bf729c0 | 1349 | |
8daaa831 DC |
1350 | /* |
1351 | * Return the number of reclaimable inodes in the filesystem for | |
1352 | * the shrinker to determine how much to reclaim. | |
1353 | */ | |
1354 | int | |
1355 | xfs_reclaim_inodes_count( | |
1356 | struct xfs_mount *mp) | |
1357 | { | |
1358 | struct xfs_perag *pag; | |
1359 | xfs_agnumber_t ag = 0; | |
1360 | int reclaimable = 0; | |
9bf729c0 | 1361 | |
65d0f205 DC |
1362 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1363 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1364 | reclaimable += pag->pag_ici_reclaimable; |
1365 | xfs_perag_put(pag); | |
9bf729c0 | 1366 | } |
9bf729c0 DC |
1367 | return reclaimable; |
1368 | } | |
1369 | ||
3e3f9f58 BF |
1370 | STATIC int |
1371 | xfs_inode_match_id( | |
1372 | struct xfs_inode *ip, | |
1373 | struct xfs_eofblocks *eofb) | |
1374 | { | |
b9fe5052 DE |
1375 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && |
1376 | !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
1b556048 | 1377 | return 0; |
3e3f9f58 | 1378 | |
b9fe5052 DE |
1379 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && |
1380 | !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
1b556048 BF |
1381 | return 0; |
1382 | ||
b9fe5052 | 1383 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && |
1b556048 BF |
1384 | xfs_get_projid(ip) != eofb->eof_prid) |
1385 | return 0; | |
1386 | ||
1387 | return 1; | |
3e3f9f58 BF |
1388 | } |
1389 | ||
f4526397 BF |
1390 | /* |
1391 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1392 | * criteria match. This is for global/internal scans only. | |
1393 | */ | |
1394 | STATIC int | |
1395 | xfs_inode_match_id_union( | |
1396 | struct xfs_inode *ip, | |
1397 | struct xfs_eofblocks *eofb) | |
1398 | { | |
1399 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && | |
1400 | uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
1401 | return 1; | |
1402 | ||
1403 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && | |
1404 | gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
1405 | return 1; | |
1406 | ||
1407 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && | |
1408 | xfs_get_projid(ip) == eofb->eof_prid) | |
1409 | return 1; | |
1410 | ||
1411 | return 0; | |
1412 | } | |
1413 | ||
41176a68 BF |
1414 | STATIC int |
1415 | xfs_inode_free_eofblocks( | |
1416 | struct xfs_inode *ip, | |
41176a68 BF |
1417 | int flags, |
1418 | void *args) | |
1419 | { | |
a36b9261 | 1420 | int ret = 0; |
3e3f9f58 | 1421 | struct xfs_eofblocks *eofb = args; |
f4526397 | 1422 | int match; |
5400da7d | 1423 | |
41176a68 BF |
1424 | if (!xfs_can_free_eofblocks(ip, false)) { |
1425 | /* inode could be preallocated or append-only */ | |
1426 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1427 | xfs_inode_clear_eofblocks_tag(ip); | |
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | /* | |
1432 | * If the mapping is dirty the operation can block and wait for some | |
1433 | * time. Unless we are waiting, skip it. | |
1434 | */ | |
1435 | if (!(flags & SYNC_WAIT) && | |
1436 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) | |
1437 | return 0; | |
1438 | ||
00ca79a0 | 1439 | if (eofb) { |
f4526397 BF |
1440 | if (eofb->eof_flags & XFS_EOF_FLAGS_UNION) |
1441 | match = xfs_inode_match_id_union(ip, eofb); | |
1442 | else | |
1443 | match = xfs_inode_match_id(ip, eofb); | |
1444 | if (!match) | |
00ca79a0 BF |
1445 | return 0; |
1446 | ||
1447 | /* skip the inode if the file size is too small */ | |
1448 | if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE && | |
1449 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1450 | return 0; | |
1451 | } | |
3e3f9f58 | 1452 | |
a36b9261 BF |
1453 | /* |
1454 | * If the caller is waiting, return -EAGAIN to keep the background | |
1455 | * scanner moving and revisit the inode in a subsequent pass. | |
1456 | */ | |
c3155097 | 1457 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
a36b9261 BF |
1458 | if (flags & SYNC_WAIT) |
1459 | ret = -EAGAIN; | |
1460 | return ret; | |
1461 | } | |
1462 | ret = xfs_free_eofblocks(ip); | |
c3155097 | 1463 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
41176a68 BF |
1464 | |
1465 | return ret; | |
1466 | } | |
1467 | ||
83104d44 DW |
1468 | static int |
1469 | __xfs_icache_free_eofblocks( | |
41176a68 | 1470 | struct xfs_mount *mp, |
83104d44 DW |
1471 | struct xfs_eofblocks *eofb, |
1472 | int (*execute)(struct xfs_inode *ip, int flags, | |
1473 | void *args), | |
1474 | int tag) | |
41176a68 | 1475 | { |
8ca149de BF |
1476 | int flags = SYNC_TRYLOCK; |
1477 | ||
1478 | if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC)) | |
1479 | flags = SYNC_WAIT; | |
1480 | ||
83104d44 DW |
1481 | return xfs_inode_ag_iterator_tag(mp, execute, flags, |
1482 | eofb, tag); | |
1483 | } | |
1484 | ||
1485 | int | |
1486 | xfs_icache_free_eofblocks( | |
1487 | struct xfs_mount *mp, | |
1488 | struct xfs_eofblocks *eofb) | |
1489 | { | |
1490 | return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_eofblocks, | |
1491 | XFS_ICI_EOFBLOCKS_TAG); | |
41176a68 BF |
1492 | } |
1493 | ||
dc06f398 BF |
1494 | /* |
1495 | * Run eofblocks scans on the quotas applicable to the inode. For inodes with | |
1496 | * multiple quotas, we don't know exactly which quota caused an allocation | |
1497 | * failure. We make a best effort by including each quota under low free space | |
1498 | * conditions (less than 1% free space) in the scan. | |
1499 | */ | |
83104d44 DW |
1500 | static int |
1501 | __xfs_inode_free_quota_eofblocks( | |
1502 | struct xfs_inode *ip, | |
1503 | int (*execute)(struct xfs_mount *mp, | |
1504 | struct xfs_eofblocks *eofb)) | |
dc06f398 BF |
1505 | { |
1506 | int scan = 0; | |
1507 | struct xfs_eofblocks eofb = {0}; | |
1508 | struct xfs_dquot *dq; | |
1509 | ||
dc06f398 | 1510 | /* |
c3155097 | 1511 | * Run a sync scan to increase effectiveness and use the union filter to |
dc06f398 BF |
1512 | * cover all applicable quotas in a single scan. |
1513 | */ | |
dc06f398 BF |
1514 | eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC; |
1515 | ||
1516 | if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) { | |
1517 | dq = xfs_inode_dquot(ip, XFS_DQ_USER); | |
1518 | if (dq && xfs_dquot_lowsp(dq)) { | |
1519 | eofb.eof_uid = VFS_I(ip)->i_uid; | |
1520 | eofb.eof_flags |= XFS_EOF_FLAGS_UID; | |
1521 | scan = 1; | |
1522 | } | |
1523 | } | |
1524 | ||
1525 | if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) { | |
1526 | dq = xfs_inode_dquot(ip, XFS_DQ_GROUP); | |
1527 | if (dq && xfs_dquot_lowsp(dq)) { | |
1528 | eofb.eof_gid = VFS_I(ip)->i_gid; | |
1529 | eofb.eof_flags |= XFS_EOF_FLAGS_GID; | |
1530 | scan = 1; | |
1531 | } | |
1532 | } | |
1533 | ||
1534 | if (scan) | |
83104d44 | 1535 | execute(ip->i_mount, &eofb); |
dc06f398 BF |
1536 | |
1537 | return scan; | |
1538 | } | |
1539 | ||
83104d44 DW |
1540 | int |
1541 | xfs_inode_free_quota_eofblocks( | |
1542 | struct xfs_inode *ip) | |
1543 | { | |
1544 | return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks); | |
1545 | } | |
1546 | ||
91aae6be DW |
1547 | static inline unsigned long |
1548 | xfs_iflag_for_tag( | |
1549 | int tag) | |
1550 | { | |
1551 | switch (tag) { | |
1552 | case XFS_ICI_EOFBLOCKS_TAG: | |
1553 | return XFS_IEOFBLOCKS; | |
1554 | case XFS_ICI_COWBLOCKS_TAG: | |
1555 | return XFS_ICOWBLOCKS; | |
1556 | default: | |
1557 | ASSERT(0); | |
1558 | return 0; | |
1559 | } | |
1560 | } | |
1561 | ||
83104d44 | 1562 | static void |
91aae6be | 1563 | __xfs_inode_set_blocks_tag( |
83104d44 DW |
1564 | xfs_inode_t *ip, |
1565 | void (*execute)(struct xfs_mount *mp), | |
1566 | void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno, | |
1567 | int error, unsigned long caller_ip), | |
1568 | int tag) | |
27b52867 BF |
1569 | { |
1570 | struct xfs_mount *mp = ip->i_mount; | |
1571 | struct xfs_perag *pag; | |
1572 | int tagged; | |
1573 | ||
85a6e764 CH |
1574 | /* |
1575 | * Don't bother locking the AG and looking up in the radix trees | |
1576 | * if we already know that we have the tag set. | |
1577 | */ | |
91aae6be | 1578 | if (ip->i_flags & xfs_iflag_for_tag(tag)) |
85a6e764 CH |
1579 | return; |
1580 | spin_lock(&ip->i_flags_lock); | |
91aae6be | 1581 | ip->i_flags |= xfs_iflag_for_tag(tag); |
85a6e764 CH |
1582 | spin_unlock(&ip->i_flags_lock); |
1583 | ||
27b52867 BF |
1584 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1585 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1586 | |
83104d44 | 1587 | tagged = radix_tree_tagged(&pag->pag_ici_root, tag); |
27b52867 | 1588 | radix_tree_tag_set(&pag->pag_ici_root, |
83104d44 | 1589 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag); |
27b52867 BF |
1590 | if (!tagged) { |
1591 | /* propagate the eofblocks tag up into the perag radix tree */ | |
1592 | spin_lock(&ip->i_mount->m_perag_lock); | |
1593 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
1594 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
83104d44 | 1595 | tag); |
27b52867 | 1596 | spin_unlock(&ip->i_mount->m_perag_lock); |
579b62fa BF |
1597 | |
1598 | /* kick off background trimming */ | |
83104d44 | 1599 | execute(ip->i_mount); |
27b52867 | 1600 | |
83104d44 | 1601 | set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_); |
27b52867 BF |
1602 | } |
1603 | ||
1604 | spin_unlock(&pag->pag_ici_lock); | |
1605 | xfs_perag_put(pag); | |
1606 | } | |
1607 | ||
1608 | void | |
83104d44 | 1609 | xfs_inode_set_eofblocks_tag( |
27b52867 | 1610 | xfs_inode_t *ip) |
83104d44 DW |
1611 | { |
1612 | trace_xfs_inode_set_eofblocks_tag(ip); | |
91aae6be | 1613 | return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks, |
83104d44 DW |
1614 | trace_xfs_perag_set_eofblocks, |
1615 | XFS_ICI_EOFBLOCKS_TAG); | |
1616 | } | |
1617 | ||
1618 | static void | |
91aae6be | 1619 | __xfs_inode_clear_blocks_tag( |
83104d44 DW |
1620 | xfs_inode_t *ip, |
1621 | void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno, | |
1622 | int error, unsigned long caller_ip), | |
1623 | int tag) | |
27b52867 BF |
1624 | { |
1625 | struct xfs_mount *mp = ip->i_mount; | |
1626 | struct xfs_perag *pag; | |
1627 | ||
85a6e764 | 1628 | spin_lock(&ip->i_flags_lock); |
91aae6be | 1629 | ip->i_flags &= ~xfs_iflag_for_tag(tag); |
85a6e764 CH |
1630 | spin_unlock(&ip->i_flags_lock); |
1631 | ||
27b52867 BF |
1632 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1633 | spin_lock(&pag->pag_ici_lock); | |
27b52867 BF |
1634 | |
1635 | radix_tree_tag_clear(&pag->pag_ici_root, | |
83104d44 DW |
1636 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag); |
1637 | if (!radix_tree_tagged(&pag->pag_ici_root, tag)) { | |
27b52867 BF |
1638 | /* clear the eofblocks tag from the perag radix tree */ |
1639 | spin_lock(&ip->i_mount->m_perag_lock); | |
1640 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
1641 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
83104d44 | 1642 | tag); |
27b52867 | 1643 | spin_unlock(&ip->i_mount->m_perag_lock); |
83104d44 | 1644 | clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_); |
27b52867 BF |
1645 | } |
1646 | ||
1647 | spin_unlock(&pag->pag_ici_lock); | |
1648 | xfs_perag_put(pag); | |
1649 | } | |
1650 | ||
83104d44 DW |
1651 | void |
1652 | xfs_inode_clear_eofblocks_tag( | |
1653 | xfs_inode_t *ip) | |
1654 | { | |
1655 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
91aae6be | 1656 | return __xfs_inode_clear_blocks_tag(ip, |
83104d44 DW |
1657 | trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG); |
1658 | } | |
1659 | ||
1660 | /* | |
be78ff0e DW |
1661 | * Set ourselves up to free CoW blocks from this file. If it's already clean |
1662 | * then we can bail out quickly, but otherwise we must back off if the file | |
1663 | * is undergoing some kind of write. | |
83104d44 | 1664 | */ |
be78ff0e DW |
1665 | static bool |
1666 | xfs_prep_free_cowblocks( | |
83104d44 | 1667 | struct xfs_inode *ip, |
be78ff0e | 1668 | struct xfs_ifork *ifp) |
83104d44 | 1669 | { |
39937234 BF |
1670 | /* |
1671 | * Just clear the tag if we have an empty cow fork or none at all. It's | |
1672 | * possible the inode was fully unshared since it was originally tagged. | |
1673 | */ | |
1674 | if (!xfs_is_reflink_inode(ip) || !ifp->if_bytes) { | |
83104d44 DW |
1675 | trace_xfs_inode_free_cowblocks_invalid(ip); |
1676 | xfs_inode_clear_cowblocks_tag(ip); | |
be78ff0e | 1677 | return false; |
83104d44 DW |
1678 | } |
1679 | ||
1680 | /* | |
1681 | * If the mapping is dirty or under writeback we cannot touch the | |
1682 | * CoW fork. Leave it alone if we're in the midst of a directio. | |
1683 | */ | |
a1b7a4de CH |
1684 | if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) || |
1685 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) || | |
83104d44 DW |
1686 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) || |
1687 | atomic_read(&VFS_I(ip)->i_dio_count)) | |
be78ff0e DW |
1688 | return false; |
1689 | ||
1690 | return true; | |
1691 | } | |
1692 | ||
1693 | /* | |
1694 | * Automatic CoW Reservation Freeing | |
1695 | * | |
1696 | * These functions automatically garbage collect leftover CoW reservations | |
1697 | * that were made on behalf of a cowextsize hint when we start to run out | |
1698 | * of quota or when the reservations sit around for too long. If the file | |
1699 | * has dirty pages or is undergoing writeback, its CoW reservations will | |
1700 | * be retained. | |
1701 | * | |
1702 | * The actual garbage collection piggybacks off the same code that runs | |
1703 | * the speculative EOF preallocation garbage collector. | |
1704 | */ | |
1705 | STATIC int | |
1706 | xfs_inode_free_cowblocks( | |
1707 | struct xfs_inode *ip, | |
1708 | int flags, | |
1709 | void *args) | |
1710 | { | |
1711 | struct xfs_eofblocks *eofb = args; | |
1712 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
1713 | int match; | |
1714 | int ret = 0; | |
1715 | ||
1716 | if (!xfs_prep_free_cowblocks(ip, ifp)) | |
83104d44 DW |
1717 | return 0; |
1718 | ||
1719 | if (eofb) { | |
1720 | if (eofb->eof_flags & XFS_EOF_FLAGS_UNION) | |
1721 | match = xfs_inode_match_id_union(ip, eofb); | |
1722 | else | |
1723 | match = xfs_inode_match_id(ip, eofb); | |
1724 | if (!match) | |
1725 | return 0; | |
1726 | ||
1727 | /* skip the inode if the file size is too small */ | |
1728 | if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE && | |
1729 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1730 | return 0; | |
83104d44 DW |
1731 | } |
1732 | ||
1733 | /* Free the CoW blocks */ | |
c3155097 BF |
1734 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
1735 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); | |
83104d44 | 1736 | |
be78ff0e DW |
1737 | /* |
1738 | * Check again, nobody else should be able to dirty blocks or change | |
1739 | * the reflink iflag now that we have the first two locks held. | |
1740 | */ | |
1741 | if (xfs_prep_free_cowblocks(ip, ifp)) | |
1742 | ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false); | |
83104d44 | 1743 | |
c3155097 BF |
1744 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
1745 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
83104d44 DW |
1746 | |
1747 | return ret; | |
1748 | } | |
1749 | ||
1750 | int | |
1751 | xfs_icache_free_cowblocks( | |
1752 | struct xfs_mount *mp, | |
1753 | struct xfs_eofblocks *eofb) | |
1754 | { | |
1755 | return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_cowblocks, | |
1756 | XFS_ICI_COWBLOCKS_TAG); | |
1757 | } | |
1758 | ||
1759 | int | |
1760 | xfs_inode_free_quota_cowblocks( | |
1761 | struct xfs_inode *ip) | |
1762 | { | |
1763 | return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks); | |
1764 | } | |
1765 | ||
1766 | void | |
1767 | xfs_inode_set_cowblocks_tag( | |
1768 | xfs_inode_t *ip) | |
1769 | { | |
7b7381f0 | 1770 | trace_xfs_inode_set_cowblocks_tag(ip); |
91aae6be | 1771 | return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks, |
7b7381f0 | 1772 | trace_xfs_perag_set_cowblocks, |
83104d44 DW |
1773 | XFS_ICI_COWBLOCKS_TAG); |
1774 | } | |
1775 | ||
1776 | void | |
1777 | xfs_inode_clear_cowblocks_tag( | |
1778 | xfs_inode_t *ip) | |
1779 | { | |
7b7381f0 | 1780 | trace_xfs_inode_clear_cowblocks_tag(ip); |
91aae6be | 1781 | return __xfs_inode_clear_blocks_tag(ip, |
7b7381f0 | 1782 | trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG); |
83104d44 | 1783 | } |