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> | |
40 | ||
33479e05 DC |
41 | /* |
42 | * Allocate and initialise an xfs_inode. | |
43 | */ | |
638f4416 | 44 | struct xfs_inode * |
33479e05 DC |
45 | xfs_inode_alloc( |
46 | struct xfs_mount *mp, | |
47 | xfs_ino_t ino) | |
48 | { | |
49 | struct xfs_inode *ip; | |
50 | ||
51 | /* | |
52 | * if this didn't occur in transactions, we could use | |
53 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
54 | * code up to do this anyway. | |
55 | */ | |
56 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
57 | if (!ip) | |
58 | return NULL; | |
59 | if (inode_init_always(mp->m_super, VFS_I(ip))) { | |
60 | kmem_zone_free(xfs_inode_zone, ip); | |
61 | return NULL; | |
62 | } | |
63 | ||
c19b3b05 DC |
64 | /* VFS doesn't initialise i_mode! */ |
65 | VFS_I(ip)->i_mode = 0; | |
66 | ||
ff6d6af2 | 67 | XFS_STATS_INC(mp, vn_active); |
33479e05 | 68 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
33479e05 DC |
69 | ASSERT(!xfs_isiflocked(ip)); |
70 | ASSERT(ip->i_ino == 0); | |
71 | ||
33479e05 DC |
72 | /* initialise the xfs inode */ |
73 | ip->i_ino = ino; | |
74 | ip->i_mount = mp; | |
75 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
76 | ip->i_afp = NULL; | |
3993baeb DW |
77 | ip->i_cowfp = NULL; |
78 | ip->i_cnextents = 0; | |
79 | ip->i_cformat = XFS_DINODE_FMT_EXTENTS; | |
33479e05 DC |
80 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); |
81 | ip->i_flags = 0; | |
82 | ip->i_delayed_blks = 0; | |
f8d55aa0 | 83 | memset(&ip->i_d, 0, sizeof(ip->i_d)); |
33479e05 DC |
84 | |
85 | return ip; | |
86 | } | |
87 | ||
88 | STATIC void | |
89 | xfs_inode_free_callback( | |
90 | struct rcu_head *head) | |
91 | { | |
92 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
93 | struct xfs_inode *ip = XFS_I(inode); | |
94 | ||
c19b3b05 | 95 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
96 | case S_IFREG: |
97 | case S_IFDIR: | |
98 | case S_IFLNK: | |
99 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
100 | break; | |
101 | } | |
102 | ||
103 | if (ip->i_afp) | |
104 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
3993baeb DW |
105 | if (ip->i_cowfp) |
106 | xfs_idestroy_fork(ip, XFS_COW_FORK); | |
33479e05 DC |
107 | |
108 | if (ip->i_itemp) { | |
109 | ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL)); | |
110 | xfs_inode_item_destroy(ip); | |
111 | ip->i_itemp = NULL; | |
112 | } | |
113 | ||
1f2dcfe8 DC |
114 | kmem_zone_free(xfs_inode_zone, ip); |
115 | } | |
116 | ||
8a17d7dd DC |
117 | static void |
118 | __xfs_inode_free( | |
119 | struct xfs_inode *ip) | |
120 | { | |
121 | /* asserts to verify all state is correct here */ | |
122 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
8a17d7dd DC |
123 | XFS_STATS_DEC(ip->i_mount, vn_active); |
124 | ||
125 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
126 | } | |
127 | ||
1f2dcfe8 DC |
128 | void |
129 | xfs_inode_free( | |
130 | struct xfs_inode *ip) | |
131 | { | |
98efe8af BF |
132 | ASSERT(!xfs_isiflocked(ip)); |
133 | ||
33479e05 DC |
134 | /* |
135 | * Because we use RCU freeing we need to ensure the inode always | |
136 | * appears to be reclaimed with an invalid inode number when in the | |
137 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
138 | * races. | |
139 | */ | |
140 | spin_lock(&ip->i_flags_lock); | |
141 | ip->i_flags = XFS_IRECLAIM; | |
142 | ip->i_ino = 0; | |
143 | spin_unlock(&ip->i_flags_lock); | |
144 | ||
8a17d7dd | 145 | __xfs_inode_free(ip); |
33479e05 DC |
146 | } |
147 | ||
ad438c40 DC |
148 | /* |
149 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
150 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
151 | * on the xfs periodic sync default of 30s. Perhaps this should have it's own | |
152 | * tunable, but that can be done if this method proves to be ineffective or too | |
153 | * aggressive. | |
154 | */ | |
155 | static void | |
156 | xfs_reclaim_work_queue( | |
157 | struct xfs_mount *mp) | |
158 | { | |
159 | ||
160 | rcu_read_lock(); | |
161 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
162 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
163 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
164 | } | |
165 | rcu_read_unlock(); | |
166 | } | |
167 | ||
168 | /* | |
169 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
170 | * many inodes as possible in a short period of time. It kicks itself every few | |
171 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
172 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
173 | * already being flushed, and once done schedules a future pass. | |
174 | */ | |
175 | void | |
176 | xfs_reclaim_worker( | |
177 | struct work_struct *work) | |
178 | { | |
179 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
180 | struct xfs_mount, m_reclaim_work); | |
181 | ||
182 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
183 | xfs_reclaim_work_queue(mp); | |
184 | } | |
185 | ||
186 | static void | |
187 | xfs_perag_set_reclaim_tag( | |
188 | struct xfs_perag *pag) | |
189 | { | |
190 | struct xfs_mount *mp = pag->pag_mount; | |
191 | ||
95989c46 | 192 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
193 | if (pag->pag_ici_reclaimable++) |
194 | return; | |
195 | ||
196 | /* propagate the reclaim tag up into the perag radix tree */ | |
197 | spin_lock(&mp->m_perag_lock); | |
198 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, | |
199 | XFS_ICI_RECLAIM_TAG); | |
200 | spin_unlock(&mp->m_perag_lock); | |
201 | ||
202 | /* schedule periodic background inode reclaim */ | |
203 | xfs_reclaim_work_queue(mp); | |
204 | ||
205 | trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
206 | } | |
207 | ||
208 | static void | |
209 | xfs_perag_clear_reclaim_tag( | |
210 | struct xfs_perag *pag) | |
211 | { | |
212 | struct xfs_mount *mp = pag->pag_mount; | |
213 | ||
95989c46 | 214 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
215 | if (--pag->pag_ici_reclaimable) |
216 | return; | |
217 | ||
218 | /* clear the reclaim tag from the perag radix tree */ | |
219 | spin_lock(&mp->m_perag_lock); | |
220 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, | |
221 | XFS_ICI_RECLAIM_TAG); | |
222 | spin_unlock(&mp->m_perag_lock); | |
223 | trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
224 | } | |
225 | ||
226 | ||
227 | /* | |
228 | * We set the inode flag atomically with the radix tree tag. | |
229 | * Once we get tag lookups on the radix tree, this inode flag | |
230 | * can go away. | |
231 | */ | |
232 | void | |
233 | xfs_inode_set_reclaim_tag( | |
234 | struct xfs_inode *ip) | |
235 | { | |
236 | struct xfs_mount *mp = ip->i_mount; | |
237 | struct xfs_perag *pag; | |
238 | ||
239 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
240 | spin_lock(&pag->pag_ici_lock); | |
241 | spin_lock(&ip->i_flags_lock); | |
242 | ||
243 | radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
244 | XFS_ICI_RECLAIM_TAG); | |
245 | xfs_perag_set_reclaim_tag(pag); | |
246 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
247 | ||
248 | spin_unlock(&ip->i_flags_lock); | |
249 | spin_unlock(&pag->pag_ici_lock); | |
250 | xfs_perag_put(pag); | |
251 | } | |
252 | ||
253 | STATIC void | |
254 | xfs_inode_clear_reclaim_tag( | |
255 | struct xfs_perag *pag, | |
256 | xfs_ino_t ino) | |
257 | { | |
258 | radix_tree_tag_clear(&pag->pag_ici_root, | |
259 | XFS_INO_TO_AGINO(pag->pag_mount, ino), | |
260 | XFS_ICI_RECLAIM_TAG); | |
261 | xfs_perag_clear_reclaim_tag(pag); | |
262 | } | |
263 | ||
ae2c4ac2 BF |
264 | static void |
265 | xfs_inew_wait( | |
266 | struct xfs_inode *ip) | |
267 | { | |
268 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT); | |
269 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT); | |
270 | ||
271 | do { | |
21417136 | 272 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
ae2c4ac2 BF |
273 | if (!xfs_iflags_test(ip, XFS_INEW)) |
274 | break; | |
275 | schedule(); | |
276 | } while (true); | |
21417136 | 277 | finish_wait(wq, &wait.wq_entry); |
ae2c4ac2 BF |
278 | } |
279 | ||
50997470 DC |
280 | /* |
281 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
282 | * part of the structure. This is made more complex by the fact we store | |
283 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 284 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 285 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 286 | * after reinitialisation even if it fails. |
50997470 DC |
287 | */ |
288 | static int | |
289 | xfs_reinit_inode( | |
290 | struct xfs_mount *mp, | |
291 | struct inode *inode) | |
292 | { | |
293 | int error; | |
54d7b5c1 | 294 | uint32_t nlink = inode->i_nlink; |
9e9a2674 | 295 | uint32_t generation = inode->i_generation; |
83e06f21 | 296 | uint64_t version = inode->i_version; |
c19b3b05 | 297 | umode_t mode = inode->i_mode; |
50997470 DC |
298 | |
299 | error = inode_init_always(mp->m_super, inode); | |
300 | ||
54d7b5c1 | 301 | set_nlink(inode, nlink); |
9e9a2674 | 302 | inode->i_generation = generation; |
83e06f21 | 303 | inode->i_version = version; |
c19b3b05 | 304 | inode->i_mode = mode; |
50997470 DC |
305 | return error; |
306 | } | |
307 | ||
33479e05 DC |
308 | /* |
309 | * Check the validity of the inode we just found it the cache | |
310 | */ | |
311 | static int | |
312 | xfs_iget_cache_hit( | |
313 | struct xfs_perag *pag, | |
314 | struct xfs_inode *ip, | |
315 | xfs_ino_t ino, | |
316 | int flags, | |
317 | int lock_flags) __releases(RCU) | |
318 | { | |
319 | struct inode *inode = VFS_I(ip); | |
320 | struct xfs_mount *mp = ip->i_mount; | |
321 | int error; | |
322 | ||
323 | /* | |
324 | * check for re-use of an inode within an RCU grace period due to the | |
325 | * radix tree nodes not being updated yet. We monitor for this by | |
326 | * setting the inode number to zero before freeing the inode structure. | |
327 | * If the inode has been reallocated and set up, then the inode number | |
328 | * will not match, so check for that, too. | |
329 | */ | |
330 | spin_lock(&ip->i_flags_lock); | |
331 | if (ip->i_ino != ino) { | |
332 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 333 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 334 | error = -EAGAIN; |
33479e05 DC |
335 | goto out_error; |
336 | } | |
337 | ||
338 | ||
339 | /* | |
340 | * If we are racing with another cache hit that is currently | |
341 | * instantiating this inode or currently recycling it out of | |
342 | * reclaimabe state, wait for the initialisation to complete | |
343 | * before continuing. | |
344 | * | |
345 | * XXX(hch): eventually we should do something equivalent to | |
346 | * wait_on_inode to wait for these flags to be cleared | |
347 | * instead of polling for it. | |
348 | */ | |
349 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { | |
350 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 351 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 352 | error = -EAGAIN; |
33479e05 DC |
353 | goto out_error; |
354 | } | |
355 | ||
356 | /* | |
357 | * If lookup is racing with unlink return an error immediately. | |
358 | */ | |
c19b3b05 | 359 | if (VFS_I(ip)->i_mode == 0 && !(flags & XFS_IGET_CREATE)) { |
2451337d | 360 | error = -ENOENT; |
33479e05 DC |
361 | goto out_error; |
362 | } | |
363 | ||
364 | /* | |
365 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
366 | * Need to carefully get it back into useable state. | |
367 | */ | |
368 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
369 | trace_xfs_iget_reclaim(ip); | |
370 | ||
378f681c DW |
371 | if (flags & XFS_IGET_INCORE) { |
372 | error = -EAGAIN; | |
373 | goto out_error; | |
374 | } | |
375 | ||
33479e05 DC |
376 | /* |
377 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode | |
378 | * from stomping over us while we recycle the inode. We can't | |
379 | * clear the radix tree reclaimable tag yet as it requires | |
380 | * pag_ici_lock to be held exclusive. | |
381 | */ | |
382 | ip->i_flags |= XFS_IRECLAIM; | |
383 | ||
384 | spin_unlock(&ip->i_flags_lock); | |
385 | rcu_read_unlock(); | |
386 | ||
50997470 | 387 | error = xfs_reinit_inode(mp, inode); |
33479e05 | 388 | if (error) { |
756baca2 | 389 | bool wake; |
33479e05 DC |
390 | /* |
391 | * Re-initializing the inode failed, and we are in deep | |
392 | * trouble. Try to re-add it to the reclaim list. | |
393 | */ | |
394 | rcu_read_lock(); | |
395 | spin_lock(&ip->i_flags_lock); | |
756baca2 | 396 | wake = !!__xfs_iflags_test(ip, XFS_INEW); |
33479e05 | 397 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
756baca2 BF |
398 | if (wake) |
399 | wake_up_bit(&ip->i_flags, __XFS_INEW_BIT); | |
33479e05 DC |
400 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); |
401 | trace_xfs_iget_reclaim_fail(ip); | |
402 | goto out_error; | |
403 | } | |
404 | ||
405 | spin_lock(&pag->pag_ici_lock); | |
406 | spin_lock(&ip->i_flags_lock); | |
407 | ||
408 | /* | |
409 | * Clear the per-lifetime state in the inode as we are now | |
410 | * effectively a new inode and need to return to the initial | |
411 | * state before reuse occurs. | |
412 | */ | |
413 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
414 | ip->i_flags |= XFS_INEW; | |
545c0889 | 415 | xfs_inode_clear_reclaim_tag(pag, ip->i_ino); |
33479e05 DC |
416 | inode->i_state = I_NEW; |
417 | ||
65523218 CH |
418 | ASSERT(!rwsem_is_locked(&inode->i_rwsem)); |
419 | init_rwsem(&inode->i_rwsem); | |
33479e05 DC |
420 | |
421 | spin_unlock(&ip->i_flags_lock); | |
422 | spin_unlock(&pag->pag_ici_lock); | |
423 | } else { | |
424 | /* If the VFS inode is being torn down, pause and try again. */ | |
425 | if (!igrab(inode)) { | |
426 | trace_xfs_iget_skip(ip); | |
2451337d | 427 | error = -EAGAIN; |
33479e05 DC |
428 | goto out_error; |
429 | } | |
430 | ||
431 | /* We've got a live one. */ | |
432 | spin_unlock(&ip->i_flags_lock); | |
433 | rcu_read_unlock(); | |
434 | trace_xfs_iget_hit(ip); | |
435 | } | |
436 | ||
437 | if (lock_flags != 0) | |
438 | xfs_ilock(ip, lock_flags); | |
439 | ||
378f681c DW |
440 | if (!(flags & XFS_IGET_INCORE)) |
441 | xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE); | |
ff6d6af2 | 442 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
443 | |
444 | return 0; | |
445 | ||
446 | out_error: | |
447 | spin_unlock(&ip->i_flags_lock); | |
448 | rcu_read_unlock(); | |
449 | return error; | |
450 | } | |
451 | ||
452 | ||
453 | static int | |
454 | xfs_iget_cache_miss( | |
455 | struct xfs_mount *mp, | |
456 | struct xfs_perag *pag, | |
457 | xfs_trans_t *tp, | |
458 | xfs_ino_t ino, | |
459 | struct xfs_inode **ipp, | |
460 | int flags, | |
461 | int lock_flags) | |
462 | { | |
463 | struct xfs_inode *ip; | |
464 | int error; | |
465 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
466 | int iflags; | |
467 | ||
468 | ip = xfs_inode_alloc(mp, ino); | |
469 | if (!ip) | |
2451337d | 470 | return -ENOMEM; |
33479e05 DC |
471 | |
472 | error = xfs_iread(mp, tp, ip, flags); | |
473 | if (error) | |
474 | goto out_destroy; | |
475 | ||
476 | trace_xfs_iget_miss(ip); | |
477 | ||
c19b3b05 | 478 | if ((VFS_I(ip)->i_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
2451337d | 479 | error = -ENOENT; |
33479e05 DC |
480 | goto out_destroy; |
481 | } | |
482 | ||
483 | /* | |
484 | * Preload the radix tree so we can insert safely under the | |
485 | * write spinlock. Note that we cannot sleep inside the preload | |
486 | * region. Since we can be called from transaction context, don't | |
487 | * recurse into the file system. | |
488 | */ | |
489 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 490 | error = -EAGAIN; |
33479e05 DC |
491 | goto out_destroy; |
492 | } | |
493 | ||
494 | /* | |
495 | * Because the inode hasn't been added to the radix-tree yet it can't | |
496 | * be found by another thread, so we can do the non-sleeping lock here. | |
497 | */ | |
498 | if (lock_flags) { | |
499 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
500 | BUG(); | |
501 | } | |
502 | ||
503 | /* | |
504 | * These values must be set before inserting the inode into the radix | |
505 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
506 | * RCU locking mechanism) can find it and that lookup must see that this | |
507 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
508 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
509 | * memory barrier that ensures this detection works correctly at lookup | |
510 | * time. | |
511 | */ | |
512 | iflags = XFS_INEW; | |
513 | if (flags & XFS_IGET_DONTCACHE) | |
514 | iflags |= XFS_IDONTCACHE; | |
113a5683 CS |
515 | ip->i_udquot = NULL; |
516 | ip->i_gdquot = NULL; | |
92f8ff73 | 517 | ip->i_pdquot = NULL; |
33479e05 DC |
518 | xfs_iflags_set(ip, iflags); |
519 | ||
520 | /* insert the new inode */ | |
521 | spin_lock(&pag->pag_ici_lock); | |
522 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
523 | if (unlikely(error)) { | |
524 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 525 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 526 | error = -EAGAIN; |
33479e05 DC |
527 | goto out_preload_end; |
528 | } | |
529 | spin_unlock(&pag->pag_ici_lock); | |
530 | radix_tree_preload_end(); | |
531 | ||
532 | *ipp = ip; | |
533 | return 0; | |
534 | ||
535 | out_preload_end: | |
536 | spin_unlock(&pag->pag_ici_lock); | |
537 | radix_tree_preload_end(); | |
538 | if (lock_flags) | |
539 | xfs_iunlock(ip, lock_flags); | |
540 | out_destroy: | |
541 | __destroy_inode(VFS_I(ip)); | |
542 | xfs_inode_free(ip); | |
543 | return error; | |
544 | } | |
545 | ||
546 | /* | |
547 | * Look up an inode by number in the given file system. | |
548 | * The inode is looked up in the cache held in each AG. | |
549 | * If the inode is found in the cache, initialise the vfs inode | |
550 | * if necessary. | |
551 | * | |
552 | * If it is not in core, read it in from the file system's device, | |
553 | * add it to the cache and initialise the vfs inode. | |
554 | * | |
555 | * The inode is locked according to the value of the lock_flags parameter. | |
556 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
557 | * should be taken. | |
558 | * | |
559 | * mp -- the mount point structure for the current file system. It points | |
560 | * to the inode hash table. | |
561 | * tp -- a pointer to the current transaction if there is one. This is | |
562 | * simply passed through to the xfs_iread() call. | |
563 | * ino -- the number of the inode desired. This is the unique identifier | |
564 | * within the file system for the inode being requested. | |
565 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
566 | * for xfs_ilock() for a list of valid values. | |
567 | */ | |
568 | int | |
569 | xfs_iget( | |
570 | xfs_mount_t *mp, | |
571 | xfs_trans_t *tp, | |
572 | xfs_ino_t ino, | |
573 | uint flags, | |
574 | uint lock_flags, | |
575 | xfs_inode_t **ipp) | |
576 | { | |
577 | xfs_inode_t *ip; | |
578 | int error; | |
579 | xfs_perag_t *pag; | |
580 | xfs_agino_t agino; | |
581 | ||
582 | /* | |
583 | * xfs_reclaim_inode() uses the ILOCK to ensure an inode | |
584 | * doesn't get freed while it's being referenced during a | |
585 | * radix tree traversal here. It assumes this function | |
586 | * aqcuires only the ILOCK (and therefore it has no need to | |
587 | * involve the IOLOCK in this synchronization). | |
588 | */ | |
589 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); | |
590 | ||
591 | /* reject inode numbers outside existing AGs */ | |
592 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 593 | return -EINVAL; |
33479e05 | 594 | |
ff6d6af2 | 595 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 596 | |
33479e05 DC |
597 | /* get the perag structure and ensure that it's inode capable */ |
598 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
599 | agino = XFS_INO_TO_AGINO(mp, ino); | |
600 | ||
601 | again: | |
602 | error = 0; | |
603 | rcu_read_lock(); | |
604 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
605 | ||
606 | if (ip) { | |
607 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
608 | if (error) | |
609 | goto out_error_or_again; | |
610 | } else { | |
611 | rcu_read_unlock(); | |
378f681c | 612 | if (flags & XFS_IGET_INCORE) { |
ed438b47 | 613 | error = -ENODATA; |
378f681c DW |
614 | goto out_error_or_again; |
615 | } | |
ff6d6af2 | 616 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
617 | |
618 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
619 | flags, lock_flags); | |
620 | if (error) | |
621 | goto out_error_or_again; | |
622 | } | |
623 | xfs_perag_put(pag); | |
624 | ||
625 | *ipp = ip; | |
626 | ||
627 | /* | |
58c90473 | 628 | * If we have a real type for an on-disk inode, we can setup the inode |
33479e05 DC |
629 | * now. If it's a new inode being created, xfs_ialloc will handle it. |
630 | */ | |
c19b3b05 | 631 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 632 | xfs_setup_existing_inode(ip); |
33479e05 DC |
633 | return 0; |
634 | ||
635 | out_error_or_again: | |
378f681c | 636 | if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) { |
33479e05 DC |
637 | delay(1); |
638 | goto again; | |
639 | } | |
640 | xfs_perag_put(pag); | |
641 | return error; | |
642 | } | |
643 | ||
378f681c DW |
644 | /* |
645 | * "Is this a cached inode that's also allocated?" | |
646 | * | |
647 | * Look up an inode by number in the given file system. If the inode is | |
648 | * in cache and isn't in purgatory, return 1 if the inode is allocated | |
649 | * and 0 if it is not. For all other cases (not in cache, being torn | |
650 | * down, etc.), return a negative error code. | |
651 | * | |
652 | * The caller has to prevent inode allocation and freeing activity, | |
653 | * presumably by locking the AGI buffer. This is to ensure that an | |
654 | * inode cannot transition from allocated to freed until the caller is | |
655 | * ready to allow that. If the inode is in an intermediate state (new, | |
656 | * reclaimable, or being reclaimed), -EAGAIN will be returned; if the | |
657 | * inode is not in the cache, -ENOENT will be returned. The caller must | |
658 | * deal with these scenarios appropriately. | |
659 | * | |
660 | * This is a specialized use case for the online scrubber; if you're | |
661 | * reading this, you probably want xfs_iget. | |
662 | */ | |
663 | int | |
664 | xfs_icache_inode_is_allocated( | |
665 | struct xfs_mount *mp, | |
666 | struct xfs_trans *tp, | |
667 | xfs_ino_t ino, | |
668 | bool *inuse) | |
669 | { | |
670 | struct xfs_inode *ip; | |
671 | int error; | |
672 | ||
673 | error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip); | |
674 | if (error) | |
675 | return error; | |
676 | ||
677 | *inuse = !!(VFS_I(ip)->i_mode); | |
678 | IRELE(ip); | |
679 | return 0; | |
680 | } | |
681 | ||
78ae5256 DC |
682 | /* |
683 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
684 | * radix tree lookups to a minimum. The batch size is a trade off between | |
685 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
686 | * be too greedy. | |
687 | */ | |
688 | #define XFS_LOOKUP_BATCH 32 | |
689 | ||
e13de955 DC |
690 | STATIC int |
691 | xfs_inode_ag_walk_grab( | |
ae2c4ac2 BF |
692 | struct xfs_inode *ip, |
693 | int flags) | |
e13de955 DC |
694 | { |
695 | struct inode *inode = VFS_I(ip); | |
ae2c4ac2 | 696 | bool newinos = !!(flags & XFS_AGITER_INEW_WAIT); |
e13de955 | 697 | |
1a3e8f3d DC |
698 | ASSERT(rcu_read_lock_held()); |
699 | ||
700 | /* | |
701 | * check for stale RCU freed inode | |
702 | * | |
703 | * If the inode has been reallocated, it doesn't matter if it's not in | |
704 | * the AG we are walking - we are walking for writeback, so if it | |
705 | * passes all the "valid inode" checks and is dirty, then we'll write | |
706 | * it back anyway. If it has been reallocated and still being | |
707 | * initialised, the XFS_INEW check below will catch it. | |
708 | */ | |
709 | spin_lock(&ip->i_flags_lock); | |
710 | if (!ip->i_ino) | |
711 | goto out_unlock_noent; | |
712 | ||
713 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
ae2c4ac2 BF |
714 | if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) || |
715 | __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
1a3e8f3d DC |
716 | goto out_unlock_noent; |
717 | spin_unlock(&ip->i_flags_lock); | |
718 | ||
e13de955 DC |
719 | /* nothing to sync during shutdown */ |
720 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
2451337d | 721 | return -EFSCORRUPTED; |
e13de955 | 722 | |
e13de955 DC |
723 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
724 | if (!igrab(inode)) | |
2451337d | 725 | return -ENOENT; |
e13de955 | 726 | |
e13de955 DC |
727 | /* inode is valid */ |
728 | return 0; | |
1a3e8f3d DC |
729 | |
730 | out_unlock_noent: | |
731 | spin_unlock(&ip->i_flags_lock); | |
2451337d | 732 | return -ENOENT; |
e13de955 DC |
733 | } |
734 | ||
75f3cb13 DC |
735 | STATIC int |
736 | xfs_inode_ag_walk( | |
737 | struct xfs_mount *mp, | |
5017e97d | 738 | struct xfs_perag *pag, |
e0094008 | 739 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
740 | void *args), |
741 | int flags, | |
742 | void *args, | |
ae2c4ac2 BF |
743 | int tag, |
744 | int iter_flags) | |
75f3cb13 | 745 | { |
75f3cb13 DC |
746 | uint32_t first_index; |
747 | int last_error = 0; | |
748 | int skipped; | |
65d0f205 | 749 | int done; |
78ae5256 | 750 | int nr_found; |
75f3cb13 DC |
751 | |
752 | restart: | |
65d0f205 | 753 | done = 0; |
75f3cb13 DC |
754 | skipped = 0; |
755 | first_index = 0; | |
78ae5256 | 756 | nr_found = 0; |
75f3cb13 | 757 | do { |
78ae5256 | 758 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
75f3cb13 | 759 | int error = 0; |
78ae5256 | 760 | int i; |
75f3cb13 | 761 | |
1a3e8f3d | 762 | rcu_read_lock(); |
a454f742 BF |
763 | |
764 | if (tag == -1) | |
765 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
78ae5256 DC |
766 | (void **)batch, first_index, |
767 | XFS_LOOKUP_BATCH); | |
a454f742 BF |
768 | else |
769 | nr_found = radix_tree_gang_lookup_tag( | |
770 | &pag->pag_ici_root, | |
771 | (void **) batch, first_index, | |
772 | XFS_LOOKUP_BATCH, tag); | |
773 | ||
65d0f205 | 774 | if (!nr_found) { |
1a3e8f3d | 775 | rcu_read_unlock(); |
75f3cb13 | 776 | break; |
c8e20be0 | 777 | } |
75f3cb13 | 778 | |
65d0f205 | 779 | /* |
78ae5256 DC |
780 | * Grab the inodes before we drop the lock. if we found |
781 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 782 | */ |
78ae5256 DC |
783 | for (i = 0; i < nr_found; i++) { |
784 | struct xfs_inode *ip = batch[i]; | |
785 | ||
ae2c4ac2 | 786 | if (done || xfs_inode_ag_walk_grab(ip, iter_flags)) |
78ae5256 DC |
787 | batch[i] = NULL; |
788 | ||
789 | /* | |
1a3e8f3d DC |
790 | * Update the index for the next lookup. Catch |
791 | * overflows into the next AG range which can occur if | |
792 | * we have inodes in the last block of the AG and we | |
793 | * are currently pointing to the last inode. | |
794 | * | |
795 | * Because we may see inodes that are from the wrong AG | |
796 | * due to RCU freeing and reallocation, only update the | |
797 | * index if it lies in this AG. It was a race that lead | |
798 | * us to see this inode, so another lookup from the | |
799 | * same index will not find it again. | |
78ae5256 | 800 | */ |
1a3e8f3d DC |
801 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
802 | continue; | |
78ae5256 DC |
803 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
804 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
805 | done = 1; | |
e13de955 | 806 | } |
78ae5256 DC |
807 | |
808 | /* unlock now we've grabbed the inodes. */ | |
1a3e8f3d | 809 | rcu_read_unlock(); |
e13de955 | 810 | |
78ae5256 DC |
811 | for (i = 0; i < nr_found; i++) { |
812 | if (!batch[i]) | |
813 | continue; | |
ae2c4ac2 BF |
814 | if ((iter_flags & XFS_AGITER_INEW_WAIT) && |
815 | xfs_iflags_test(batch[i], XFS_INEW)) | |
816 | xfs_inew_wait(batch[i]); | |
e0094008 | 817 | error = execute(batch[i], flags, args); |
78ae5256 | 818 | IRELE(batch[i]); |
2451337d | 819 | if (error == -EAGAIN) { |
78ae5256 DC |
820 | skipped++; |
821 | continue; | |
822 | } | |
2451337d | 823 | if (error && last_error != -EFSCORRUPTED) |
78ae5256 | 824 | last_error = error; |
75f3cb13 | 825 | } |
c8e20be0 DC |
826 | |
827 | /* bail out if the filesystem is corrupted. */ | |
2451337d | 828 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
829 | break; |
830 | ||
8daaa831 DC |
831 | cond_resched(); |
832 | ||
78ae5256 | 833 | } while (nr_found && !done); |
75f3cb13 DC |
834 | |
835 | if (skipped) { | |
836 | delay(1); | |
837 | goto restart; | |
838 | } | |
75f3cb13 DC |
839 | return last_error; |
840 | } | |
841 | ||
579b62fa BF |
842 | /* |
843 | * Background scanning to trim post-EOF preallocated space. This is queued | |
b9fe5052 | 844 | * based on the 'speculative_prealloc_lifetime' tunable (5m by default). |
579b62fa | 845 | */ |
fa5a4f57 | 846 | void |
579b62fa BF |
847 | xfs_queue_eofblocks( |
848 | struct xfs_mount *mp) | |
849 | { | |
850 | rcu_read_lock(); | |
851 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG)) | |
852 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
853 | &mp->m_eofblocks_work, | |
854 | msecs_to_jiffies(xfs_eofb_secs * 1000)); | |
855 | rcu_read_unlock(); | |
856 | } | |
857 | ||
858 | void | |
859 | xfs_eofblocks_worker( | |
860 | struct work_struct *work) | |
861 | { | |
862 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
863 | struct xfs_mount, m_eofblocks_work); | |
864 | xfs_icache_free_eofblocks(mp, NULL); | |
865 | xfs_queue_eofblocks(mp); | |
866 | } | |
867 | ||
83104d44 DW |
868 | /* |
869 | * Background scanning to trim preallocated CoW space. This is queued | |
870 | * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default). | |
871 | * (We'll just piggyback on the post-EOF prealloc space workqueue.) | |
872 | */ | |
10ddf64e | 873 | void |
83104d44 DW |
874 | xfs_queue_cowblocks( |
875 | struct xfs_mount *mp) | |
876 | { | |
877 | rcu_read_lock(); | |
878 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG)) | |
879 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
880 | &mp->m_cowblocks_work, | |
881 | msecs_to_jiffies(xfs_cowb_secs * 1000)); | |
882 | rcu_read_unlock(); | |
883 | } | |
884 | ||
885 | void | |
886 | xfs_cowblocks_worker( | |
887 | struct work_struct *work) | |
888 | { | |
889 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
890 | struct xfs_mount, m_cowblocks_work); | |
891 | xfs_icache_free_cowblocks(mp, NULL); | |
892 | xfs_queue_cowblocks(mp); | |
893 | } | |
894 | ||
fe588ed3 | 895 | int |
ae2c4ac2 | 896 | xfs_inode_ag_iterator_flags( |
75f3cb13 | 897 | struct xfs_mount *mp, |
e0094008 | 898 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
899 | void *args), |
900 | int flags, | |
ae2c4ac2 BF |
901 | void *args, |
902 | int iter_flags) | |
75f3cb13 | 903 | { |
16fd5367 | 904 | struct xfs_perag *pag; |
75f3cb13 DC |
905 | int error = 0; |
906 | int last_error = 0; | |
907 | xfs_agnumber_t ag; | |
908 | ||
16fd5367 | 909 | ag = 0; |
65d0f205 DC |
910 | while ((pag = xfs_perag_get(mp, ag))) { |
911 | ag = pag->pag_agno + 1; | |
ae2c4ac2 BF |
912 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1, |
913 | iter_flags); | |
a454f742 BF |
914 | xfs_perag_put(pag); |
915 | if (error) { | |
916 | last_error = error; | |
2451337d | 917 | if (error == -EFSCORRUPTED) |
a454f742 BF |
918 | break; |
919 | } | |
920 | } | |
b474c7ae | 921 | return last_error; |
a454f742 BF |
922 | } |
923 | ||
ae2c4ac2 BF |
924 | int |
925 | xfs_inode_ag_iterator( | |
926 | struct xfs_mount *mp, | |
927 | int (*execute)(struct xfs_inode *ip, int flags, | |
928 | void *args), | |
929 | int flags, | |
930 | void *args) | |
931 | { | |
932 | return xfs_inode_ag_iterator_flags(mp, execute, flags, args, 0); | |
933 | } | |
934 | ||
a454f742 BF |
935 | int |
936 | xfs_inode_ag_iterator_tag( | |
937 | struct xfs_mount *mp, | |
e0094008 | 938 | int (*execute)(struct xfs_inode *ip, int flags, |
a454f742 BF |
939 | void *args), |
940 | int flags, | |
941 | void *args, | |
942 | int tag) | |
943 | { | |
944 | struct xfs_perag *pag; | |
945 | int error = 0; | |
946 | int last_error = 0; | |
947 | xfs_agnumber_t ag; | |
948 | ||
949 | ag = 0; | |
950 | while ((pag = xfs_perag_get_tag(mp, ag, tag))) { | |
951 | ag = pag->pag_agno + 1; | |
ae2c4ac2 BF |
952 | error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag, |
953 | 0); | |
5017e97d | 954 | xfs_perag_put(pag); |
75f3cb13 DC |
955 | if (error) { |
956 | last_error = error; | |
2451337d | 957 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
958 | break; |
959 | } | |
960 | } | |
b474c7ae | 961 | return last_error; |
75f3cb13 DC |
962 | } |
963 | ||
e3a20c0b DC |
964 | /* |
965 | * Grab the inode for reclaim exclusively. | |
966 | * Return 0 if we grabbed it, non-zero otherwise. | |
967 | */ | |
968 | STATIC int | |
969 | xfs_reclaim_inode_grab( | |
970 | struct xfs_inode *ip, | |
971 | int flags) | |
972 | { | |
1a3e8f3d DC |
973 | ASSERT(rcu_read_lock_held()); |
974 | ||
975 | /* quick check for stale RCU freed inode */ | |
976 | if (!ip->i_ino) | |
977 | return 1; | |
e3a20c0b DC |
978 | |
979 | /* | |
474fce06 CH |
980 | * If we are asked for non-blocking operation, do unlocked checks to |
981 | * see if the inode already is being flushed or in reclaim to avoid | |
982 | * lock traffic. | |
e3a20c0b DC |
983 | */ |
984 | if ((flags & SYNC_TRYLOCK) && | |
474fce06 | 985 | __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM)) |
e3a20c0b | 986 | return 1; |
e3a20c0b DC |
987 | |
988 | /* | |
989 | * The radix tree lock here protects a thread in xfs_iget from racing | |
990 | * with us starting reclaim on the inode. Once we have the | |
991 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
992 | * |
993 | * Due to RCU lookup, we may find inodes that have been freed and only | |
994 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
995 | * aren't candidates for reclaim at all, so we must check the | |
996 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
997 | */ |
998 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
999 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
1000 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
1001 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
1002 | spin_unlock(&ip->i_flags_lock); |
1003 | return 1; | |
1004 | } | |
1005 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
1006 | spin_unlock(&ip->i_flags_lock); | |
1007 | return 0; | |
1008 | } | |
1009 | ||
777df5af | 1010 | /* |
8a48088f CH |
1011 | * Inodes in different states need to be treated differently. The following |
1012 | * table lists the inode states and the reclaim actions necessary: | |
777df5af DC |
1013 | * |
1014 | * inode state iflush ret required action | |
1015 | * --------------- ---------- --------------- | |
1016 | * bad - reclaim | |
1017 | * shutdown EIO unpin and reclaim | |
1018 | * clean, unpinned 0 reclaim | |
1019 | * stale, unpinned 0 reclaim | |
c854363e DC |
1020 | * clean, pinned(*) 0 requeue |
1021 | * stale, pinned EAGAIN requeue | |
8a48088f CH |
1022 | * dirty, async - requeue |
1023 | * dirty, sync 0 reclaim | |
777df5af DC |
1024 | * |
1025 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
1026 | * handled anyway given the order of checks implemented. | |
1027 | * | |
c854363e DC |
1028 | * Also, because we get the flush lock first, we know that any inode that has |
1029 | * been flushed delwri has had the flush completed by the time we check that | |
8a48088f | 1030 | * the inode is clean. |
c854363e | 1031 | * |
8a48088f CH |
1032 | * Note that because the inode is flushed delayed write by AIL pushing, the |
1033 | * flush lock may already be held here and waiting on it can result in very | |
1034 | * long latencies. Hence for sync reclaims, where we wait on the flush lock, | |
1035 | * the caller should push the AIL first before trying to reclaim inodes to | |
1036 | * minimise the amount of time spent waiting. For background relaim, we only | |
1037 | * bother to reclaim clean inodes anyway. | |
c854363e | 1038 | * |
777df5af DC |
1039 | * Hence the order of actions after gaining the locks should be: |
1040 | * bad => reclaim | |
1041 | * shutdown => unpin and reclaim | |
8a48088f | 1042 | * pinned, async => requeue |
c854363e | 1043 | * pinned, sync => unpin |
777df5af DC |
1044 | * stale => reclaim |
1045 | * clean => reclaim | |
8a48088f | 1046 | * dirty, async => requeue |
c854363e | 1047 | * dirty, sync => flush, wait and reclaim |
777df5af | 1048 | */ |
75f3cb13 | 1049 | STATIC int |
c8e20be0 | 1050 | xfs_reclaim_inode( |
75f3cb13 DC |
1051 | struct xfs_inode *ip, |
1052 | struct xfs_perag *pag, | |
c8e20be0 | 1053 | int sync_mode) |
fce08f2f | 1054 | { |
4c46819a | 1055 | struct xfs_buf *bp = NULL; |
8a17d7dd | 1056 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
4c46819a | 1057 | int error; |
777df5af | 1058 | |
1bfd8d04 DC |
1059 | restart: |
1060 | error = 0; | |
c8e20be0 | 1061 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
1062 | if (!xfs_iflock_nowait(ip)) { |
1063 | if (!(sync_mode & SYNC_WAIT)) | |
1064 | goto out; | |
1065 | xfs_iflock(ip); | |
1066 | } | |
7a3be02b | 1067 | |
777df5af DC |
1068 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1069 | xfs_iunpin_wait(ip); | |
98efe8af | 1070 | /* xfs_iflush_abort() drops the flush lock */ |
04913fdd | 1071 | xfs_iflush_abort(ip, false); |
777df5af DC |
1072 | goto reclaim; |
1073 | } | |
c854363e | 1074 | if (xfs_ipincount(ip)) { |
8a48088f CH |
1075 | if (!(sync_mode & SYNC_WAIT)) |
1076 | goto out_ifunlock; | |
777df5af | 1077 | xfs_iunpin_wait(ip); |
c854363e | 1078 | } |
98efe8af BF |
1079 | if (xfs_iflags_test(ip, XFS_ISTALE) || xfs_inode_clean(ip)) { |
1080 | xfs_ifunlock(ip); | |
777df5af | 1081 | goto reclaim; |
98efe8af | 1082 | } |
777df5af | 1083 | |
8a48088f CH |
1084 | /* |
1085 | * Never flush out dirty data during non-blocking reclaim, as it would | |
1086 | * just contend with AIL pushing trying to do the same job. | |
1087 | */ | |
1088 | if (!(sync_mode & SYNC_WAIT)) | |
1089 | goto out_ifunlock; | |
1090 | ||
1bfd8d04 DC |
1091 | /* |
1092 | * Now we have an inode that needs flushing. | |
1093 | * | |
4c46819a | 1094 | * Note that xfs_iflush will never block on the inode buffer lock, as |
1bfd8d04 | 1095 | * xfs_ifree_cluster() can lock the inode buffer before it locks the |
4c46819a | 1096 | * ip->i_lock, and we are doing the exact opposite here. As a result, |
475ee413 CH |
1097 | * doing a blocking xfs_imap_to_bp() to get the cluster buffer would |
1098 | * result in an ABBA deadlock with xfs_ifree_cluster(). | |
1bfd8d04 DC |
1099 | * |
1100 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
1101 | * cache to mark them stale, if we hit this case we don't actually want | |
1102 | * to do IO here - we want the inode marked stale so we can simply | |
4c46819a CH |
1103 | * reclaim it. Hence if we get an EAGAIN error here, just unlock the |
1104 | * inode, back off and try again. Hopefully the next pass through will | |
1105 | * see the stale flag set on the inode. | |
1bfd8d04 | 1106 | */ |
4c46819a | 1107 | error = xfs_iflush(ip, &bp); |
2451337d | 1108 | if (error == -EAGAIN) { |
8a48088f CH |
1109 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1110 | /* backoff longer than in xfs_ifree_cluster */ | |
1111 | delay(2); | |
1112 | goto restart; | |
c854363e | 1113 | } |
c854363e | 1114 | |
4c46819a CH |
1115 | if (!error) { |
1116 | error = xfs_bwrite(bp); | |
1117 | xfs_buf_relse(bp); | |
1118 | } | |
1119 | ||
777df5af | 1120 | reclaim: |
98efe8af BF |
1121 | ASSERT(!xfs_isiflocked(ip)); |
1122 | ||
8a17d7dd DC |
1123 | /* |
1124 | * Because we use RCU freeing we need to ensure the inode always appears | |
1125 | * to be reclaimed with an invalid inode number when in the free state. | |
98efe8af | 1126 | * We do this as early as possible under the ILOCK so that |
f2e9ad21 OS |
1127 | * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to |
1128 | * detect races with us here. By doing this, we guarantee that once | |
1129 | * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that | |
1130 | * it will see either a valid inode that will serialise correctly, or it | |
1131 | * will see an invalid inode that it can skip. | |
8a17d7dd DC |
1132 | */ |
1133 | spin_lock(&ip->i_flags_lock); | |
1134 | ip->i_flags = XFS_IRECLAIM; | |
1135 | ip->i_ino = 0; | |
1136 | spin_unlock(&ip->i_flags_lock); | |
1137 | ||
c8e20be0 | 1138 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1139 | |
ff6d6af2 | 1140 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
1141 | /* |
1142 | * Remove the inode from the per-AG radix tree. | |
1143 | * | |
1144 | * Because radix_tree_delete won't complain even if the item was never | |
1145 | * added to the tree assert that it's been there before to catch | |
1146 | * problems with the inode life time early on. | |
1147 | */ | |
1a427ab0 | 1148 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 1149 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 1150 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 1151 | ASSERT(0); |
545c0889 | 1152 | xfs_perag_clear_reclaim_tag(pag); |
1a427ab0 | 1153 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
1154 | |
1155 | /* | |
1156 | * Here we do an (almost) spurious inode lock in order to coordinate | |
1157 | * with inode cache radix tree lookups. This is because the lookup | |
1158 | * can reference the inodes in the cache without taking references. | |
1159 | * | |
1160 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 1161 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 1162 | */ |
ad637a10 | 1163 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1164 | xfs_qm_dqdetach(ip); |
ad637a10 | 1165 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1166 | |
8a17d7dd | 1167 | __xfs_inode_free(ip); |
ad637a10 | 1168 | return error; |
8a48088f CH |
1169 | |
1170 | out_ifunlock: | |
1171 | xfs_ifunlock(ip); | |
1172 | out: | |
1173 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
1174 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1175 | /* | |
2451337d | 1176 | * We could return -EAGAIN here to make reclaim rescan the inode tree in |
8a48088f | 1177 | * a short while. However, this just burns CPU time scanning the tree |
5889608d DC |
1178 | * waiting for IO to complete and the reclaim work never goes back to |
1179 | * the idle state. Instead, return 0 to let the next scheduled | |
1180 | * background reclaim attempt to reclaim the inode again. | |
8a48088f CH |
1181 | */ |
1182 | return 0; | |
7a3be02b DC |
1183 | } |
1184 | ||
65d0f205 DC |
1185 | /* |
1186 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
1187 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
1188 | * then a shut down during filesystem unmount reclaim walk leak all the | |
1189 | * unreclaimed inodes. | |
1190 | */ | |
33479e05 | 1191 | STATIC int |
65d0f205 DC |
1192 | xfs_reclaim_inodes_ag( |
1193 | struct xfs_mount *mp, | |
1194 | int flags, | |
1195 | int *nr_to_scan) | |
1196 | { | |
1197 | struct xfs_perag *pag; | |
1198 | int error = 0; | |
1199 | int last_error = 0; | |
1200 | xfs_agnumber_t ag; | |
69b491c2 DC |
1201 | int trylock = flags & SYNC_TRYLOCK; |
1202 | int skipped; | |
65d0f205 | 1203 | |
69b491c2 | 1204 | restart: |
65d0f205 | 1205 | ag = 0; |
69b491c2 | 1206 | skipped = 0; |
65d0f205 DC |
1207 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1208 | unsigned long first_index = 0; | |
1209 | int done = 0; | |
e3a20c0b | 1210 | int nr_found = 0; |
65d0f205 DC |
1211 | |
1212 | ag = pag->pag_agno + 1; | |
1213 | ||
69b491c2 DC |
1214 | if (trylock) { |
1215 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
1216 | skipped++; | |
f83282a8 | 1217 | xfs_perag_put(pag); |
69b491c2 DC |
1218 | continue; |
1219 | } | |
1220 | first_index = pag->pag_ici_reclaim_cursor; | |
1221 | } else | |
1222 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
1223 | ||
65d0f205 | 1224 | do { |
e3a20c0b DC |
1225 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
1226 | int i; | |
65d0f205 | 1227 | |
1a3e8f3d | 1228 | rcu_read_lock(); |
e3a20c0b DC |
1229 | nr_found = radix_tree_gang_lookup_tag( |
1230 | &pag->pag_ici_root, | |
1231 | (void **)batch, first_index, | |
1232 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
1233 | XFS_ICI_RECLAIM_TAG); |
1234 | if (!nr_found) { | |
b2232219 | 1235 | done = 1; |
1a3e8f3d | 1236 | rcu_read_unlock(); |
65d0f205 DC |
1237 | break; |
1238 | } | |
1239 | ||
1240 | /* | |
e3a20c0b DC |
1241 | * Grab the inodes before we drop the lock. if we found |
1242 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 1243 | */ |
e3a20c0b DC |
1244 | for (i = 0; i < nr_found; i++) { |
1245 | struct xfs_inode *ip = batch[i]; | |
1246 | ||
1247 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
1248 | batch[i] = NULL; | |
1249 | ||
1250 | /* | |
1251 | * Update the index for the next lookup. Catch | |
1252 | * overflows into the next AG range which can | |
1253 | * occur if we have inodes in the last block of | |
1254 | * the AG and we are currently pointing to the | |
1255 | * last inode. | |
1a3e8f3d DC |
1256 | * |
1257 | * Because we may see inodes that are from the | |
1258 | * wrong AG due to RCU freeing and | |
1259 | * reallocation, only update the index if it | |
1260 | * lies in this AG. It was a race that lead us | |
1261 | * to see this inode, so another lookup from | |
1262 | * the same index will not find it again. | |
e3a20c0b | 1263 | */ |
1a3e8f3d DC |
1264 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
1265 | pag->pag_agno) | |
1266 | continue; | |
e3a20c0b DC |
1267 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
1268 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1269 | done = 1; | |
1270 | } | |
65d0f205 | 1271 | |
e3a20c0b | 1272 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 1273 | rcu_read_unlock(); |
e3a20c0b DC |
1274 | |
1275 | for (i = 0; i < nr_found; i++) { | |
1276 | if (!batch[i]) | |
1277 | continue; | |
1278 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
2451337d | 1279 | if (error && last_error != -EFSCORRUPTED) |
e3a20c0b DC |
1280 | last_error = error; |
1281 | } | |
1282 | ||
1283 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 1284 | |
8daaa831 DC |
1285 | cond_resched(); |
1286 | ||
e3a20c0b | 1287 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 1288 | |
69b491c2 DC |
1289 | if (trylock && !done) |
1290 | pag->pag_ici_reclaim_cursor = first_index; | |
1291 | else | |
1292 | pag->pag_ici_reclaim_cursor = 0; | |
1293 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
1294 | xfs_perag_put(pag); |
1295 | } | |
69b491c2 DC |
1296 | |
1297 | /* | |
1298 | * if we skipped any AG, and we still have scan count remaining, do | |
1299 | * another pass this time using blocking reclaim semantics (i.e | |
1300 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1301 | * ensure that when we get more reclaimers than AGs we block rather | |
1302 | * than spin trying to execute reclaim. | |
1303 | */ | |
8daaa831 | 1304 | if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) { |
69b491c2 DC |
1305 | trylock = 0; |
1306 | goto restart; | |
1307 | } | |
b474c7ae | 1308 | return last_error; |
65d0f205 DC |
1309 | } |
1310 | ||
7a3be02b DC |
1311 | int |
1312 | xfs_reclaim_inodes( | |
1313 | xfs_mount_t *mp, | |
7a3be02b DC |
1314 | int mode) |
1315 | { | |
65d0f205 DC |
1316 | int nr_to_scan = INT_MAX; |
1317 | ||
1318 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1319 | } |
1320 | ||
1321 | /* | |
8daaa831 | 1322 | * Scan a certain number of inodes for reclaim. |
a7b339f1 DC |
1323 | * |
1324 | * When called we make sure that there is a background (fast) inode reclaim in | |
8daaa831 | 1325 | * progress, while we will throttle the speed of reclaim via doing synchronous |
a7b339f1 DC |
1326 | * reclaim of inodes. That means if we come across dirty inodes, we wait for |
1327 | * them to be cleaned, which we hope will not be very long due to the | |
1328 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1329 | */ |
0a234c6d | 1330 | long |
8daaa831 DC |
1331 | xfs_reclaim_inodes_nr( |
1332 | struct xfs_mount *mp, | |
1333 | int nr_to_scan) | |
9bf729c0 | 1334 | { |
8daaa831 | 1335 | /* kick background reclaimer and push the AIL */ |
5889608d | 1336 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1337 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1338 | |
0a234c6d | 1339 | return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan); |
8daaa831 | 1340 | } |
9bf729c0 | 1341 | |
8daaa831 DC |
1342 | /* |
1343 | * Return the number of reclaimable inodes in the filesystem for | |
1344 | * the shrinker to determine how much to reclaim. | |
1345 | */ | |
1346 | int | |
1347 | xfs_reclaim_inodes_count( | |
1348 | struct xfs_mount *mp) | |
1349 | { | |
1350 | struct xfs_perag *pag; | |
1351 | xfs_agnumber_t ag = 0; | |
1352 | int reclaimable = 0; | |
9bf729c0 | 1353 | |
65d0f205 DC |
1354 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1355 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1356 | reclaimable += pag->pag_ici_reclaimable; |
1357 | xfs_perag_put(pag); | |
9bf729c0 | 1358 | } |
9bf729c0 DC |
1359 | return reclaimable; |
1360 | } | |
1361 | ||
3e3f9f58 BF |
1362 | STATIC int |
1363 | xfs_inode_match_id( | |
1364 | struct xfs_inode *ip, | |
1365 | struct xfs_eofblocks *eofb) | |
1366 | { | |
b9fe5052 DE |
1367 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && |
1368 | !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
1b556048 | 1369 | return 0; |
3e3f9f58 | 1370 | |
b9fe5052 DE |
1371 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && |
1372 | !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
1b556048 BF |
1373 | return 0; |
1374 | ||
b9fe5052 | 1375 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && |
1b556048 BF |
1376 | xfs_get_projid(ip) != eofb->eof_prid) |
1377 | return 0; | |
1378 | ||
1379 | return 1; | |
3e3f9f58 BF |
1380 | } |
1381 | ||
f4526397 BF |
1382 | /* |
1383 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1384 | * criteria match. This is for global/internal scans only. | |
1385 | */ | |
1386 | STATIC int | |
1387 | xfs_inode_match_id_union( | |
1388 | struct xfs_inode *ip, | |
1389 | struct xfs_eofblocks *eofb) | |
1390 | { | |
1391 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && | |
1392 | uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
1393 | return 1; | |
1394 | ||
1395 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && | |
1396 | gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
1397 | return 1; | |
1398 | ||
1399 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && | |
1400 | xfs_get_projid(ip) == eofb->eof_prid) | |
1401 | return 1; | |
1402 | ||
1403 | return 0; | |
1404 | } | |
1405 | ||
41176a68 BF |
1406 | STATIC int |
1407 | xfs_inode_free_eofblocks( | |
1408 | struct xfs_inode *ip, | |
41176a68 BF |
1409 | int flags, |
1410 | void *args) | |
1411 | { | |
a36b9261 | 1412 | int ret = 0; |
3e3f9f58 | 1413 | struct xfs_eofblocks *eofb = args; |
f4526397 | 1414 | int match; |
5400da7d | 1415 | |
41176a68 BF |
1416 | if (!xfs_can_free_eofblocks(ip, false)) { |
1417 | /* inode could be preallocated or append-only */ | |
1418 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1419 | xfs_inode_clear_eofblocks_tag(ip); | |
1420 | return 0; | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * If the mapping is dirty the operation can block and wait for some | |
1425 | * time. Unless we are waiting, skip it. | |
1426 | */ | |
1427 | if (!(flags & SYNC_WAIT) && | |
1428 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) | |
1429 | return 0; | |
1430 | ||
00ca79a0 | 1431 | if (eofb) { |
f4526397 BF |
1432 | if (eofb->eof_flags & XFS_EOF_FLAGS_UNION) |
1433 | match = xfs_inode_match_id_union(ip, eofb); | |
1434 | else | |
1435 | match = xfs_inode_match_id(ip, eofb); | |
1436 | if (!match) | |
00ca79a0 BF |
1437 | return 0; |
1438 | ||
1439 | /* skip the inode if the file size is too small */ | |
1440 | if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE && | |
1441 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1442 | return 0; | |
1443 | } | |
3e3f9f58 | 1444 | |
a36b9261 BF |
1445 | /* |
1446 | * If the caller is waiting, return -EAGAIN to keep the background | |
1447 | * scanner moving and revisit the inode in a subsequent pass. | |
1448 | */ | |
c3155097 | 1449 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
a36b9261 BF |
1450 | if (flags & SYNC_WAIT) |
1451 | ret = -EAGAIN; | |
1452 | return ret; | |
1453 | } | |
1454 | ret = xfs_free_eofblocks(ip); | |
c3155097 | 1455 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
41176a68 BF |
1456 | |
1457 | return ret; | |
1458 | } | |
1459 | ||
83104d44 DW |
1460 | static int |
1461 | __xfs_icache_free_eofblocks( | |
41176a68 | 1462 | struct xfs_mount *mp, |
83104d44 DW |
1463 | struct xfs_eofblocks *eofb, |
1464 | int (*execute)(struct xfs_inode *ip, int flags, | |
1465 | void *args), | |
1466 | int tag) | |
41176a68 | 1467 | { |
8ca149de BF |
1468 | int flags = SYNC_TRYLOCK; |
1469 | ||
1470 | if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC)) | |
1471 | flags = SYNC_WAIT; | |
1472 | ||
83104d44 DW |
1473 | return xfs_inode_ag_iterator_tag(mp, execute, flags, |
1474 | eofb, tag); | |
1475 | } | |
1476 | ||
1477 | int | |
1478 | xfs_icache_free_eofblocks( | |
1479 | struct xfs_mount *mp, | |
1480 | struct xfs_eofblocks *eofb) | |
1481 | { | |
1482 | return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_eofblocks, | |
1483 | XFS_ICI_EOFBLOCKS_TAG); | |
41176a68 BF |
1484 | } |
1485 | ||
dc06f398 BF |
1486 | /* |
1487 | * Run eofblocks scans on the quotas applicable to the inode. For inodes with | |
1488 | * multiple quotas, we don't know exactly which quota caused an allocation | |
1489 | * failure. We make a best effort by including each quota under low free space | |
1490 | * conditions (less than 1% free space) in the scan. | |
1491 | */ | |
83104d44 DW |
1492 | static int |
1493 | __xfs_inode_free_quota_eofblocks( | |
1494 | struct xfs_inode *ip, | |
1495 | int (*execute)(struct xfs_mount *mp, | |
1496 | struct xfs_eofblocks *eofb)) | |
dc06f398 BF |
1497 | { |
1498 | int scan = 0; | |
1499 | struct xfs_eofblocks eofb = {0}; | |
1500 | struct xfs_dquot *dq; | |
1501 | ||
dc06f398 | 1502 | /* |
c3155097 | 1503 | * Run a sync scan to increase effectiveness and use the union filter to |
dc06f398 BF |
1504 | * cover all applicable quotas in a single scan. |
1505 | */ | |
dc06f398 BF |
1506 | eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC; |
1507 | ||
1508 | if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) { | |
1509 | dq = xfs_inode_dquot(ip, XFS_DQ_USER); | |
1510 | if (dq && xfs_dquot_lowsp(dq)) { | |
1511 | eofb.eof_uid = VFS_I(ip)->i_uid; | |
1512 | eofb.eof_flags |= XFS_EOF_FLAGS_UID; | |
1513 | scan = 1; | |
1514 | } | |
1515 | } | |
1516 | ||
1517 | if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) { | |
1518 | dq = xfs_inode_dquot(ip, XFS_DQ_GROUP); | |
1519 | if (dq && xfs_dquot_lowsp(dq)) { | |
1520 | eofb.eof_gid = VFS_I(ip)->i_gid; | |
1521 | eofb.eof_flags |= XFS_EOF_FLAGS_GID; | |
1522 | scan = 1; | |
1523 | } | |
1524 | } | |
1525 | ||
1526 | if (scan) | |
83104d44 | 1527 | execute(ip->i_mount, &eofb); |
dc06f398 BF |
1528 | |
1529 | return scan; | |
1530 | } | |
1531 | ||
83104d44 DW |
1532 | int |
1533 | xfs_inode_free_quota_eofblocks( | |
1534 | struct xfs_inode *ip) | |
1535 | { | |
1536 | return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks); | |
1537 | } | |
1538 | ||
91aae6be DW |
1539 | static inline unsigned long |
1540 | xfs_iflag_for_tag( | |
1541 | int tag) | |
1542 | { | |
1543 | switch (tag) { | |
1544 | case XFS_ICI_EOFBLOCKS_TAG: | |
1545 | return XFS_IEOFBLOCKS; | |
1546 | case XFS_ICI_COWBLOCKS_TAG: | |
1547 | return XFS_ICOWBLOCKS; | |
1548 | default: | |
1549 | ASSERT(0); | |
1550 | return 0; | |
1551 | } | |
1552 | } | |
1553 | ||
83104d44 | 1554 | static void |
91aae6be | 1555 | __xfs_inode_set_blocks_tag( |
83104d44 DW |
1556 | xfs_inode_t *ip, |
1557 | void (*execute)(struct xfs_mount *mp), | |
1558 | void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno, | |
1559 | int error, unsigned long caller_ip), | |
1560 | int tag) | |
27b52867 BF |
1561 | { |
1562 | struct xfs_mount *mp = ip->i_mount; | |
1563 | struct xfs_perag *pag; | |
1564 | int tagged; | |
1565 | ||
85a6e764 CH |
1566 | /* |
1567 | * Don't bother locking the AG and looking up in the radix trees | |
1568 | * if we already know that we have the tag set. | |
1569 | */ | |
91aae6be | 1570 | if (ip->i_flags & xfs_iflag_for_tag(tag)) |
85a6e764 CH |
1571 | return; |
1572 | spin_lock(&ip->i_flags_lock); | |
91aae6be | 1573 | ip->i_flags |= xfs_iflag_for_tag(tag); |
85a6e764 CH |
1574 | spin_unlock(&ip->i_flags_lock); |
1575 | ||
27b52867 BF |
1576 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1577 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1578 | |
83104d44 | 1579 | tagged = radix_tree_tagged(&pag->pag_ici_root, tag); |
27b52867 | 1580 | radix_tree_tag_set(&pag->pag_ici_root, |
83104d44 | 1581 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag); |
27b52867 BF |
1582 | if (!tagged) { |
1583 | /* propagate the eofblocks tag up into the perag radix tree */ | |
1584 | spin_lock(&ip->i_mount->m_perag_lock); | |
1585 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
1586 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
83104d44 | 1587 | tag); |
27b52867 | 1588 | spin_unlock(&ip->i_mount->m_perag_lock); |
579b62fa BF |
1589 | |
1590 | /* kick off background trimming */ | |
83104d44 | 1591 | execute(ip->i_mount); |
27b52867 | 1592 | |
83104d44 | 1593 | set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_); |
27b52867 BF |
1594 | } |
1595 | ||
1596 | spin_unlock(&pag->pag_ici_lock); | |
1597 | xfs_perag_put(pag); | |
1598 | } | |
1599 | ||
1600 | void | |
83104d44 | 1601 | xfs_inode_set_eofblocks_tag( |
27b52867 | 1602 | xfs_inode_t *ip) |
83104d44 DW |
1603 | { |
1604 | trace_xfs_inode_set_eofblocks_tag(ip); | |
91aae6be | 1605 | return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks, |
83104d44 DW |
1606 | trace_xfs_perag_set_eofblocks, |
1607 | XFS_ICI_EOFBLOCKS_TAG); | |
1608 | } | |
1609 | ||
1610 | static void | |
91aae6be | 1611 | __xfs_inode_clear_blocks_tag( |
83104d44 DW |
1612 | xfs_inode_t *ip, |
1613 | void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno, | |
1614 | int error, unsigned long caller_ip), | |
1615 | int tag) | |
27b52867 BF |
1616 | { |
1617 | struct xfs_mount *mp = ip->i_mount; | |
1618 | struct xfs_perag *pag; | |
1619 | ||
85a6e764 | 1620 | spin_lock(&ip->i_flags_lock); |
91aae6be | 1621 | ip->i_flags &= ~xfs_iflag_for_tag(tag); |
85a6e764 CH |
1622 | spin_unlock(&ip->i_flags_lock); |
1623 | ||
27b52867 BF |
1624 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1625 | spin_lock(&pag->pag_ici_lock); | |
27b52867 BF |
1626 | |
1627 | radix_tree_tag_clear(&pag->pag_ici_root, | |
83104d44 DW |
1628 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag); |
1629 | if (!radix_tree_tagged(&pag->pag_ici_root, tag)) { | |
27b52867 BF |
1630 | /* clear the eofblocks tag from the perag radix tree */ |
1631 | spin_lock(&ip->i_mount->m_perag_lock); | |
1632 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
1633 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
83104d44 | 1634 | tag); |
27b52867 | 1635 | spin_unlock(&ip->i_mount->m_perag_lock); |
83104d44 | 1636 | clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_); |
27b52867 BF |
1637 | } |
1638 | ||
1639 | spin_unlock(&pag->pag_ici_lock); | |
1640 | xfs_perag_put(pag); | |
1641 | } | |
1642 | ||
83104d44 DW |
1643 | void |
1644 | xfs_inode_clear_eofblocks_tag( | |
1645 | xfs_inode_t *ip) | |
1646 | { | |
1647 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
91aae6be | 1648 | return __xfs_inode_clear_blocks_tag(ip, |
83104d44 DW |
1649 | trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG); |
1650 | } | |
1651 | ||
1652 | /* | |
1653 | * Automatic CoW Reservation Freeing | |
1654 | * | |
1655 | * These functions automatically garbage collect leftover CoW reservations | |
1656 | * that were made on behalf of a cowextsize hint when we start to run out | |
1657 | * of quota or when the reservations sit around for too long. If the file | |
1658 | * has dirty pages or is undergoing writeback, its CoW reservations will | |
1659 | * be retained. | |
1660 | * | |
1661 | * The actual garbage collection piggybacks off the same code that runs | |
1662 | * the speculative EOF preallocation garbage collector. | |
1663 | */ | |
1664 | STATIC int | |
1665 | xfs_inode_free_cowblocks( | |
1666 | struct xfs_inode *ip, | |
1667 | int flags, | |
1668 | void *args) | |
1669 | { | |
1670 | int ret; | |
1671 | struct xfs_eofblocks *eofb = args; | |
83104d44 | 1672 | int match; |
39937234 | 1673 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
83104d44 | 1674 | |
39937234 BF |
1675 | /* |
1676 | * Just clear the tag if we have an empty cow fork or none at all. It's | |
1677 | * possible the inode was fully unshared since it was originally tagged. | |
1678 | */ | |
1679 | if (!xfs_is_reflink_inode(ip) || !ifp->if_bytes) { | |
83104d44 DW |
1680 | trace_xfs_inode_free_cowblocks_invalid(ip); |
1681 | xfs_inode_clear_cowblocks_tag(ip); | |
1682 | return 0; | |
1683 | } | |
1684 | ||
1685 | /* | |
1686 | * If the mapping is dirty or under writeback we cannot touch the | |
1687 | * CoW fork. Leave it alone if we're in the midst of a directio. | |
1688 | */ | |
a1b7a4de CH |
1689 | if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) || |
1690 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) || | |
83104d44 DW |
1691 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) || |
1692 | atomic_read(&VFS_I(ip)->i_dio_count)) | |
1693 | return 0; | |
1694 | ||
1695 | if (eofb) { | |
1696 | if (eofb->eof_flags & XFS_EOF_FLAGS_UNION) | |
1697 | match = xfs_inode_match_id_union(ip, eofb); | |
1698 | else | |
1699 | match = xfs_inode_match_id(ip, eofb); | |
1700 | if (!match) | |
1701 | return 0; | |
1702 | ||
1703 | /* skip the inode if the file size is too small */ | |
1704 | if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE && | |
1705 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1706 | return 0; | |
83104d44 DW |
1707 | } |
1708 | ||
1709 | /* Free the CoW blocks */ | |
c3155097 BF |
1710 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
1711 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); | |
83104d44 | 1712 | |
3802a345 | 1713 | ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false); |
83104d44 | 1714 | |
c3155097 BF |
1715 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
1716 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
83104d44 DW |
1717 | |
1718 | return ret; | |
1719 | } | |
1720 | ||
1721 | int | |
1722 | xfs_icache_free_cowblocks( | |
1723 | struct xfs_mount *mp, | |
1724 | struct xfs_eofblocks *eofb) | |
1725 | { | |
1726 | return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_cowblocks, | |
1727 | XFS_ICI_COWBLOCKS_TAG); | |
1728 | } | |
1729 | ||
1730 | int | |
1731 | xfs_inode_free_quota_cowblocks( | |
1732 | struct xfs_inode *ip) | |
1733 | { | |
1734 | return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks); | |
1735 | } | |
1736 | ||
1737 | void | |
1738 | xfs_inode_set_cowblocks_tag( | |
1739 | xfs_inode_t *ip) | |
1740 | { | |
7b7381f0 | 1741 | trace_xfs_inode_set_cowblocks_tag(ip); |
91aae6be | 1742 | return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks, |
7b7381f0 | 1743 | trace_xfs_perag_set_cowblocks, |
83104d44 DW |
1744 | XFS_ICI_COWBLOCKS_TAG); |
1745 | } | |
1746 | ||
1747 | void | |
1748 | xfs_inode_clear_cowblocks_tag( | |
1749 | xfs_inode_t *ip) | |
1750 | { | |
7b7381f0 | 1751 | trace_xfs_inode_clear_cowblocks_tag(ip); |
91aae6be | 1752 | return __xfs_inode_clear_blocks_tag(ip, |
7b7381f0 | 1753 | trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG); |
83104d44 | 1754 | } |