Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
fe4fa4b8 DC |
2 | /* |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. | |
4 | * All Rights Reserved. | |
fe4fa4b8 DC |
5 | */ |
6 | #include "xfs.h" | |
7 | #include "xfs_fs.h" | |
5467b34b | 8 | #include "xfs_shared.h" |
6ca1c906 | 9 | #include "xfs_format.h" |
239880ef DC |
10 | #include "xfs_log_format.h" |
11 | #include "xfs_trans_resv.h" | |
fe4fa4b8 | 12 | #include "xfs_sb.h" |
fe4fa4b8 | 13 | #include "xfs_mount.h" |
fe4fa4b8 | 14 | #include "xfs_inode.h" |
239880ef DC |
15 | #include "xfs_trans.h" |
16 | #include "xfs_trans_priv.h" | |
fe4fa4b8 | 17 | #include "xfs_inode_item.h" |
7d095257 | 18 | #include "xfs_quota.h" |
0b1b213f | 19 | #include "xfs_trace.h" |
6d8b79cf | 20 | #include "xfs_icache.h" |
c24b5dfa | 21 | #include "xfs_bmap_util.h" |
dc06f398 BF |
22 | #include "xfs_dquot_item.h" |
23 | #include "xfs_dquot.h" | |
83104d44 | 24 | #include "xfs_reflink.h" |
bb8a66af | 25 | #include "xfs_ialloc.h" |
fe4fa4b8 | 26 | |
f0e28280 | 27 | #include <linux/iversion.h> |
a167b17e | 28 | |
33479e05 DC |
29 | /* |
30 | * Allocate and initialise an xfs_inode. | |
31 | */ | |
638f4416 | 32 | struct xfs_inode * |
33479e05 DC |
33 | xfs_inode_alloc( |
34 | struct xfs_mount *mp, | |
35 | xfs_ino_t ino) | |
36 | { | |
37 | struct xfs_inode *ip; | |
38 | ||
39 | /* | |
40 | * if this didn't occur in transactions, we could use | |
41 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
42 | * code up to do this anyway. | |
43 | */ | |
707e0dda | 44 | ip = kmem_zone_alloc(xfs_inode_zone, 0); |
33479e05 DC |
45 | if (!ip) |
46 | return NULL; | |
47 | if (inode_init_always(mp->m_super, VFS_I(ip))) { | |
377bcd5f | 48 | kmem_cache_free(xfs_inode_zone, ip); |
33479e05 DC |
49 | return NULL; |
50 | } | |
51 | ||
c19b3b05 DC |
52 | /* VFS doesn't initialise i_mode! */ |
53 | VFS_I(ip)->i_mode = 0; | |
54 | ||
ff6d6af2 | 55 | XFS_STATS_INC(mp, vn_active); |
33479e05 | 56 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
33479e05 DC |
57 | ASSERT(!xfs_isiflocked(ip)); |
58 | ASSERT(ip->i_ino == 0); | |
59 | ||
33479e05 DC |
60 | /* initialise the xfs inode */ |
61 | ip->i_ino = ino; | |
62 | ip->i_mount = mp; | |
63 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
64 | ip->i_afp = NULL; | |
3993baeb | 65 | ip->i_cowfp = NULL; |
3ba738df | 66 | memset(&ip->i_df, 0, sizeof(ip->i_df)); |
33479e05 DC |
67 | ip->i_flags = 0; |
68 | ip->i_delayed_blks = 0; | |
f8d55aa0 | 69 | memset(&ip->i_d, 0, sizeof(ip->i_d)); |
6772c1f1 DW |
70 | ip->i_sick = 0; |
71 | ip->i_checked = 0; | |
cb357bf3 DW |
72 | INIT_WORK(&ip->i_ioend_work, xfs_end_io); |
73 | INIT_LIST_HEAD(&ip->i_ioend_list); | |
74 | spin_lock_init(&ip->i_ioend_lock); | |
33479e05 DC |
75 | |
76 | return ip; | |
77 | } | |
78 | ||
79 | STATIC void | |
80 | xfs_inode_free_callback( | |
81 | struct rcu_head *head) | |
82 | { | |
83 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84 | struct xfs_inode *ip = XFS_I(inode); | |
85 | ||
c19b3b05 | 86 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
87 | case S_IFREG: |
88 | case S_IFDIR: | |
89 | case S_IFLNK: | |
ef838512 | 90 | xfs_idestroy_fork(&ip->i_df); |
33479e05 DC |
91 | break; |
92 | } | |
93 | ||
ef838512 CH |
94 | if (ip->i_afp) { |
95 | xfs_idestroy_fork(ip->i_afp); | |
96 | kmem_cache_free(xfs_ifork_zone, ip->i_afp); | |
97 | } | |
98 | if (ip->i_cowfp) { | |
99 | xfs_idestroy_fork(ip->i_cowfp); | |
100 | kmem_cache_free(xfs_ifork_zone, ip->i_cowfp); | |
101 | } | |
33479e05 | 102 | if (ip->i_itemp) { |
22525c17 DC |
103 | ASSERT(!test_bit(XFS_LI_IN_AIL, |
104 | &ip->i_itemp->ili_item.li_flags)); | |
33479e05 DC |
105 | xfs_inode_item_destroy(ip); |
106 | ip->i_itemp = NULL; | |
107 | } | |
108 | ||
377bcd5f | 109 | kmem_cache_free(xfs_inode_zone, ip); |
1f2dcfe8 DC |
110 | } |
111 | ||
8a17d7dd DC |
112 | static void |
113 | __xfs_inode_free( | |
114 | struct xfs_inode *ip) | |
115 | { | |
116 | /* asserts to verify all state is correct here */ | |
117 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
8a17d7dd DC |
118 | XFS_STATS_DEC(ip->i_mount, vn_active); |
119 | ||
120 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
121 | } | |
122 | ||
1f2dcfe8 DC |
123 | void |
124 | xfs_inode_free( | |
125 | struct xfs_inode *ip) | |
126 | { | |
98efe8af BF |
127 | ASSERT(!xfs_isiflocked(ip)); |
128 | ||
33479e05 DC |
129 | /* |
130 | * Because we use RCU freeing we need to ensure the inode always | |
131 | * appears to be reclaimed with an invalid inode number when in the | |
132 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
133 | * races. | |
134 | */ | |
135 | spin_lock(&ip->i_flags_lock); | |
136 | ip->i_flags = XFS_IRECLAIM; | |
137 | ip->i_ino = 0; | |
138 | spin_unlock(&ip->i_flags_lock); | |
139 | ||
8a17d7dd | 140 | __xfs_inode_free(ip); |
33479e05 DC |
141 | } |
142 | ||
ad438c40 DC |
143 | /* |
144 | * Queue a new inode reclaim pass if there are reclaimable inodes and there | |
145 | * isn't a reclaim pass already in progress. By default it runs every 5s based | |
146 | * on the xfs periodic sync default of 30s. Perhaps this should have it's own | |
147 | * tunable, but that can be done if this method proves to be ineffective or too | |
148 | * aggressive. | |
149 | */ | |
150 | static void | |
151 | xfs_reclaim_work_queue( | |
152 | struct xfs_mount *mp) | |
153 | { | |
154 | ||
155 | rcu_read_lock(); | |
156 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
157 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
158 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
159 | } | |
160 | rcu_read_unlock(); | |
161 | } | |
162 | ||
163 | /* | |
164 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
165 | * many inodes as possible in a short period of time. It kicks itself every few | |
166 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
167 | * goes low. It scans as quickly as possible avoiding locked inodes or those | |
168 | * already being flushed, and once done schedules a future pass. | |
169 | */ | |
170 | void | |
171 | xfs_reclaim_worker( | |
172 | struct work_struct *work) | |
173 | { | |
174 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
175 | struct xfs_mount, m_reclaim_work); | |
176 | ||
177 | xfs_reclaim_inodes(mp, SYNC_TRYLOCK); | |
178 | xfs_reclaim_work_queue(mp); | |
179 | } | |
180 | ||
181 | static void | |
182 | xfs_perag_set_reclaim_tag( | |
183 | struct xfs_perag *pag) | |
184 | { | |
185 | struct xfs_mount *mp = pag->pag_mount; | |
186 | ||
95989c46 | 187 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
188 | if (pag->pag_ici_reclaimable++) |
189 | return; | |
190 | ||
191 | /* propagate the reclaim tag up into the perag radix tree */ | |
192 | spin_lock(&mp->m_perag_lock); | |
193 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, | |
194 | XFS_ICI_RECLAIM_TAG); | |
195 | spin_unlock(&mp->m_perag_lock); | |
196 | ||
197 | /* schedule periodic background inode reclaim */ | |
198 | xfs_reclaim_work_queue(mp); | |
199 | ||
200 | trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
201 | } | |
202 | ||
203 | static void | |
204 | xfs_perag_clear_reclaim_tag( | |
205 | struct xfs_perag *pag) | |
206 | { | |
207 | struct xfs_mount *mp = pag->pag_mount; | |
208 | ||
95989c46 | 209 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
210 | if (--pag->pag_ici_reclaimable) |
211 | return; | |
212 | ||
213 | /* clear the reclaim tag from the perag radix tree */ | |
214 | spin_lock(&mp->m_perag_lock); | |
215 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, | |
216 | XFS_ICI_RECLAIM_TAG); | |
217 | spin_unlock(&mp->m_perag_lock); | |
218 | trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
219 | } | |
220 | ||
221 | ||
222 | /* | |
223 | * We set the inode flag atomically with the radix tree tag. | |
224 | * Once we get tag lookups on the radix tree, this inode flag | |
225 | * can go away. | |
226 | */ | |
227 | void | |
228 | xfs_inode_set_reclaim_tag( | |
229 | struct xfs_inode *ip) | |
230 | { | |
231 | struct xfs_mount *mp = ip->i_mount; | |
232 | struct xfs_perag *pag; | |
233 | ||
234 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
235 | spin_lock(&pag->pag_ici_lock); | |
236 | spin_lock(&ip->i_flags_lock); | |
237 | ||
238 | radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
239 | XFS_ICI_RECLAIM_TAG); | |
240 | xfs_perag_set_reclaim_tag(pag); | |
241 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
242 | ||
243 | spin_unlock(&ip->i_flags_lock); | |
244 | spin_unlock(&pag->pag_ici_lock); | |
245 | xfs_perag_put(pag); | |
246 | } | |
247 | ||
248 | STATIC void | |
249 | xfs_inode_clear_reclaim_tag( | |
250 | struct xfs_perag *pag, | |
251 | xfs_ino_t ino) | |
252 | { | |
253 | radix_tree_tag_clear(&pag->pag_ici_root, | |
254 | XFS_INO_TO_AGINO(pag->pag_mount, ino), | |
255 | XFS_ICI_RECLAIM_TAG); | |
256 | xfs_perag_clear_reclaim_tag(pag); | |
257 | } | |
258 | ||
ae2c4ac2 BF |
259 | static void |
260 | xfs_inew_wait( | |
261 | struct xfs_inode *ip) | |
262 | { | |
263 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT); | |
264 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT); | |
265 | ||
266 | do { | |
21417136 | 267 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
ae2c4ac2 BF |
268 | if (!xfs_iflags_test(ip, XFS_INEW)) |
269 | break; | |
270 | schedule(); | |
271 | } while (true); | |
21417136 | 272 | finish_wait(wq, &wait.wq_entry); |
ae2c4ac2 BF |
273 | } |
274 | ||
50997470 DC |
275 | /* |
276 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
277 | * part of the structure. This is made more complex by the fact we store | |
278 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 279 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 280 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 281 | * after reinitialisation even if it fails. |
50997470 DC |
282 | */ |
283 | static int | |
284 | xfs_reinit_inode( | |
285 | struct xfs_mount *mp, | |
286 | struct inode *inode) | |
287 | { | |
288 | int error; | |
54d7b5c1 | 289 | uint32_t nlink = inode->i_nlink; |
9e9a2674 | 290 | uint32_t generation = inode->i_generation; |
f0e28280 | 291 | uint64_t version = inode_peek_iversion(inode); |
c19b3b05 | 292 | umode_t mode = inode->i_mode; |
acd1d715 | 293 | dev_t dev = inode->i_rdev; |
3d8f2821 CH |
294 | kuid_t uid = inode->i_uid; |
295 | kgid_t gid = inode->i_gid; | |
50997470 DC |
296 | |
297 | error = inode_init_always(mp->m_super, inode); | |
298 | ||
54d7b5c1 | 299 | set_nlink(inode, nlink); |
9e9a2674 | 300 | inode->i_generation = generation; |
f0e28280 | 301 | inode_set_iversion_queried(inode, version); |
c19b3b05 | 302 | inode->i_mode = mode; |
acd1d715 | 303 | inode->i_rdev = dev; |
3d8f2821 CH |
304 | inode->i_uid = uid; |
305 | inode->i_gid = gid; | |
50997470 DC |
306 | return error; |
307 | } | |
308 | ||
afca6c5b DC |
309 | /* |
310 | * If we are allocating a new inode, then check what was returned is | |
311 | * actually a free, empty inode. If we are not allocating an inode, | |
312 | * then check we didn't find a free inode. | |
313 | * | |
314 | * Returns: | |
315 | * 0 if the inode free state matches the lookup context | |
316 | * -ENOENT if the inode is free and we are not allocating | |
317 | * -EFSCORRUPTED if there is any state mismatch at all | |
318 | */ | |
319 | static int | |
320 | xfs_iget_check_free_state( | |
321 | struct xfs_inode *ip, | |
322 | int flags) | |
323 | { | |
324 | if (flags & XFS_IGET_CREATE) { | |
325 | /* should be a free inode */ | |
326 | if (VFS_I(ip)->i_mode != 0) { | |
327 | xfs_warn(ip->i_mount, | |
328 | "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)", | |
329 | ip->i_ino, VFS_I(ip)->i_mode); | |
330 | return -EFSCORRUPTED; | |
331 | } | |
332 | ||
333 | if (ip->i_d.di_nblocks != 0) { | |
334 | xfs_warn(ip->i_mount, | |
335 | "Corruption detected! Free inode 0x%llx has blocks allocated!", | |
336 | ip->i_ino); | |
337 | return -EFSCORRUPTED; | |
338 | } | |
339 | return 0; | |
340 | } | |
341 | ||
342 | /* should be an allocated inode */ | |
343 | if (VFS_I(ip)->i_mode == 0) | |
344 | return -ENOENT; | |
345 | ||
346 | return 0; | |
347 | } | |
348 | ||
33479e05 DC |
349 | /* |
350 | * Check the validity of the inode we just found it the cache | |
351 | */ | |
352 | static int | |
353 | xfs_iget_cache_hit( | |
354 | struct xfs_perag *pag, | |
355 | struct xfs_inode *ip, | |
356 | xfs_ino_t ino, | |
357 | int flags, | |
358 | int lock_flags) __releases(RCU) | |
359 | { | |
360 | struct inode *inode = VFS_I(ip); | |
361 | struct xfs_mount *mp = ip->i_mount; | |
362 | int error; | |
363 | ||
364 | /* | |
365 | * check for re-use of an inode within an RCU grace period due to the | |
366 | * radix tree nodes not being updated yet. We monitor for this by | |
367 | * setting the inode number to zero before freeing the inode structure. | |
368 | * If the inode has been reallocated and set up, then the inode number | |
369 | * will not match, so check for that, too. | |
370 | */ | |
371 | spin_lock(&ip->i_flags_lock); | |
372 | if (ip->i_ino != ino) { | |
373 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 374 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 375 | error = -EAGAIN; |
33479e05 DC |
376 | goto out_error; |
377 | } | |
378 | ||
379 | ||
380 | /* | |
381 | * If we are racing with another cache hit that is currently | |
382 | * instantiating this inode or currently recycling it out of | |
383 | * reclaimabe state, wait for the initialisation to complete | |
384 | * before continuing. | |
385 | * | |
386 | * XXX(hch): eventually we should do something equivalent to | |
387 | * wait_on_inode to wait for these flags to be cleared | |
388 | * instead of polling for it. | |
389 | */ | |
390 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { | |
391 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 392 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 393 | error = -EAGAIN; |
33479e05 DC |
394 | goto out_error; |
395 | } | |
396 | ||
397 | /* | |
afca6c5b DC |
398 | * Check the inode free state is valid. This also detects lookup |
399 | * racing with unlinks. | |
33479e05 | 400 | */ |
afca6c5b DC |
401 | error = xfs_iget_check_free_state(ip, flags); |
402 | if (error) | |
33479e05 | 403 | goto out_error; |
33479e05 DC |
404 | |
405 | /* | |
406 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
407 | * Need to carefully get it back into useable state. | |
408 | */ | |
409 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
410 | trace_xfs_iget_reclaim(ip); | |
411 | ||
378f681c DW |
412 | if (flags & XFS_IGET_INCORE) { |
413 | error = -EAGAIN; | |
414 | goto out_error; | |
415 | } | |
416 | ||
33479e05 DC |
417 | /* |
418 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode | |
419 | * from stomping over us while we recycle the inode. We can't | |
420 | * clear the radix tree reclaimable tag yet as it requires | |
421 | * pag_ici_lock to be held exclusive. | |
422 | */ | |
423 | ip->i_flags |= XFS_IRECLAIM; | |
424 | ||
425 | spin_unlock(&ip->i_flags_lock); | |
426 | rcu_read_unlock(); | |
427 | ||
d45344d6 | 428 | ASSERT(!rwsem_is_locked(&inode->i_rwsem)); |
50997470 | 429 | error = xfs_reinit_inode(mp, inode); |
33479e05 | 430 | if (error) { |
756baca2 | 431 | bool wake; |
33479e05 DC |
432 | /* |
433 | * Re-initializing the inode failed, and we are in deep | |
434 | * trouble. Try to re-add it to the reclaim list. | |
435 | */ | |
436 | rcu_read_lock(); | |
437 | spin_lock(&ip->i_flags_lock); | |
756baca2 | 438 | wake = !!__xfs_iflags_test(ip, XFS_INEW); |
33479e05 | 439 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
756baca2 BF |
440 | if (wake) |
441 | wake_up_bit(&ip->i_flags, __XFS_INEW_BIT); | |
33479e05 DC |
442 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); |
443 | trace_xfs_iget_reclaim_fail(ip); | |
444 | goto out_error; | |
445 | } | |
446 | ||
447 | spin_lock(&pag->pag_ici_lock); | |
448 | spin_lock(&ip->i_flags_lock); | |
449 | ||
450 | /* | |
451 | * Clear the per-lifetime state in the inode as we are now | |
452 | * effectively a new inode and need to return to the initial | |
453 | * state before reuse occurs. | |
454 | */ | |
455 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
456 | ip->i_flags |= XFS_INEW; | |
545c0889 | 457 | xfs_inode_clear_reclaim_tag(pag, ip->i_ino); |
33479e05 | 458 | inode->i_state = I_NEW; |
6772c1f1 DW |
459 | ip->i_sick = 0; |
460 | ip->i_checked = 0; | |
33479e05 | 461 | |
33479e05 DC |
462 | spin_unlock(&ip->i_flags_lock); |
463 | spin_unlock(&pag->pag_ici_lock); | |
464 | } else { | |
465 | /* If the VFS inode is being torn down, pause and try again. */ | |
466 | if (!igrab(inode)) { | |
467 | trace_xfs_iget_skip(ip); | |
2451337d | 468 | error = -EAGAIN; |
33479e05 DC |
469 | goto out_error; |
470 | } | |
471 | ||
472 | /* We've got a live one. */ | |
473 | spin_unlock(&ip->i_flags_lock); | |
474 | rcu_read_unlock(); | |
475 | trace_xfs_iget_hit(ip); | |
476 | } | |
477 | ||
478 | if (lock_flags != 0) | |
479 | xfs_ilock(ip, lock_flags); | |
480 | ||
378f681c | 481 | if (!(flags & XFS_IGET_INCORE)) |
dae2f8ed | 482 | xfs_iflags_clear(ip, XFS_ISTALE); |
ff6d6af2 | 483 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
484 | |
485 | return 0; | |
486 | ||
487 | out_error: | |
488 | spin_unlock(&ip->i_flags_lock); | |
489 | rcu_read_unlock(); | |
490 | return error; | |
491 | } | |
492 | ||
493 | ||
494 | static int | |
495 | xfs_iget_cache_miss( | |
496 | struct xfs_mount *mp, | |
497 | struct xfs_perag *pag, | |
498 | xfs_trans_t *tp, | |
499 | xfs_ino_t ino, | |
500 | struct xfs_inode **ipp, | |
501 | int flags, | |
502 | int lock_flags) | |
503 | { | |
504 | struct xfs_inode *ip; | |
505 | int error; | |
506 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
507 | int iflags; | |
508 | ||
509 | ip = xfs_inode_alloc(mp, ino); | |
510 | if (!ip) | |
2451337d | 511 | return -ENOMEM; |
33479e05 | 512 | |
bb8a66af | 513 | error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, flags); |
33479e05 DC |
514 | if (error) |
515 | goto out_destroy; | |
516 | ||
bb8a66af CH |
517 | /* |
518 | * For version 5 superblocks, if we are initialising a new inode and we | |
519 | * are not utilising the XFS_MOUNT_IKEEP inode cluster mode, we can | |
520 | * simply build the new inode core with a random generation number. | |
521 | * | |
522 | * For version 4 (and older) superblocks, log recovery is dependent on | |
523 | * the di_flushiter field being initialised from the current on-disk | |
524 | * value and hence we must also read the inode off disk even when | |
525 | * initializing new inodes. | |
526 | */ | |
527 | if (xfs_sb_version_has_v3inode(&mp->m_sb) && | |
528 | (flags & XFS_IGET_CREATE) && !(mp->m_flags & XFS_MOUNT_IKEEP)) { | |
529 | VFS_I(ip)->i_generation = prandom_u32(); | |
530 | } else { | |
531 | struct xfs_dinode *dip; | |
532 | struct xfs_buf *bp; | |
533 | ||
534 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0); | |
535 | if (error) | |
536 | goto out_destroy; | |
537 | ||
538 | error = xfs_inode_from_disk(ip, dip); | |
539 | if (!error) | |
540 | xfs_buf_set_ref(bp, XFS_INO_REF); | |
541 | xfs_trans_brelse(tp, bp); | |
542 | ||
543 | if (error) | |
544 | goto out_destroy; | |
545 | } | |
546 | ||
33479e05 DC |
547 | trace_xfs_iget_miss(ip); |
548 | ||
ee457001 | 549 | /* |
afca6c5b DC |
550 | * Check the inode free state is valid. This also detects lookup |
551 | * racing with unlinks. | |
ee457001 | 552 | */ |
afca6c5b DC |
553 | error = xfs_iget_check_free_state(ip, flags); |
554 | if (error) | |
33479e05 | 555 | goto out_destroy; |
33479e05 DC |
556 | |
557 | /* | |
558 | * Preload the radix tree so we can insert safely under the | |
559 | * write spinlock. Note that we cannot sleep inside the preload | |
560 | * region. Since we can be called from transaction context, don't | |
561 | * recurse into the file system. | |
562 | */ | |
563 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 564 | error = -EAGAIN; |
33479e05 DC |
565 | goto out_destroy; |
566 | } | |
567 | ||
568 | /* | |
569 | * Because the inode hasn't been added to the radix-tree yet it can't | |
570 | * be found by another thread, so we can do the non-sleeping lock here. | |
571 | */ | |
572 | if (lock_flags) { | |
573 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
574 | BUG(); | |
575 | } | |
576 | ||
577 | /* | |
578 | * These values must be set before inserting the inode into the radix | |
579 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
580 | * RCU locking mechanism) can find it and that lookup must see that this | |
581 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
582 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
583 | * memory barrier that ensures this detection works correctly at lookup | |
584 | * time. | |
585 | */ | |
586 | iflags = XFS_INEW; | |
587 | if (flags & XFS_IGET_DONTCACHE) | |
2c567af4 | 588 | d_mark_dontcache(VFS_I(ip)); |
113a5683 CS |
589 | ip->i_udquot = NULL; |
590 | ip->i_gdquot = NULL; | |
92f8ff73 | 591 | ip->i_pdquot = NULL; |
33479e05 DC |
592 | xfs_iflags_set(ip, iflags); |
593 | ||
594 | /* insert the new inode */ | |
595 | spin_lock(&pag->pag_ici_lock); | |
596 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
597 | if (unlikely(error)) { | |
598 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 599 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 600 | error = -EAGAIN; |
33479e05 DC |
601 | goto out_preload_end; |
602 | } | |
603 | spin_unlock(&pag->pag_ici_lock); | |
604 | radix_tree_preload_end(); | |
605 | ||
606 | *ipp = ip; | |
607 | return 0; | |
608 | ||
609 | out_preload_end: | |
610 | spin_unlock(&pag->pag_ici_lock); | |
611 | radix_tree_preload_end(); | |
612 | if (lock_flags) | |
613 | xfs_iunlock(ip, lock_flags); | |
614 | out_destroy: | |
615 | __destroy_inode(VFS_I(ip)); | |
616 | xfs_inode_free(ip); | |
617 | return error; | |
618 | } | |
619 | ||
620 | /* | |
621 | * Look up an inode by number in the given file system. | |
622 | * The inode is looked up in the cache held in each AG. | |
623 | * If the inode is found in the cache, initialise the vfs inode | |
624 | * if necessary. | |
625 | * | |
626 | * If it is not in core, read it in from the file system's device, | |
627 | * add it to the cache and initialise the vfs inode. | |
628 | * | |
629 | * The inode is locked according to the value of the lock_flags parameter. | |
630 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
631 | * should be taken. | |
632 | * | |
633 | * mp -- the mount point structure for the current file system. It points | |
634 | * to the inode hash table. | |
635 | * tp -- a pointer to the current transaction if there is one. This is | |
636 | * simply passed through to the xfs_iread() call. | |
637 | * ino -- the number of the inode desired. This is the unique identifier | |
638 | * within the file system for the inode being requested. | |
639 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
640 | * for xfs_ilock() for a list of valid values. | |
641 | */ | |
642 | int | |
643 | xfs_iget( | |
644 | xfs_mount_t *mp, | |
645 | xfs_trans_t *tp, | |
646 | xfs_ino_t ino, | |
647 | uint flags, | |
648 | uint lock_flags, | |
649 | xfs_inode_t **ipp) | |
650 | { | |
651 | xfs_inode_t *ip; | |
652 | int error; | |
653 | xfs_perag_t *pag; | |
654 | xfs_agino_t agino; | |
655 | ||
656 | /* | |
657 | * xfs_reclaim_inode() uses the ILOCK to ensure an inode | |
658 | * doesn't get freed while it's being referenced during a | |
659 | * radix tree traversal here. It assumes this function | |
660 | * aqcuires only the ILOCK (and therefore it has no need to | |
661 | * involve the IOLOCK in this synchronization). | |
662 | */ | |
663 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); | |
664 | ||
665 | /* reject inode numbers outside existing AGs */ | |
666 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 667 | return -EINVAL; |
33479e05 | 668 | |
ff6d6af2 | 669 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 670 | |
33479e05 DC |
671 | /* get the perag structure and ensure that it's inode capable */ |
672 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
673 | agino = XFS_INO_TO_AGINO(mp, ino); | |
674 | ||
675 | again: | |
676 | error = 0; | |
677 | rcu_read_lock(); | |
678 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
679 | ||
680 | if (ip) { | |
681 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
682 | if (error) | |
683 | goto out_error_or_again; | |
684 | } else { | |
685 | rcu_read_unlock(); | |
378f681c | 686 | if (flags & XFS_IGET_INCORE) { |
ed438b47 | 687 | error = -ENODATA; |
378f681c DW |
688 | goto out_error_or_again; |
689 | } | |
ff6d6af2 | 690 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
691 | |
692 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
693 | flags, lock_flags); | |
694 | if (error) | |
695 | goto out_error_or_again; | |
696 | } | |
697 | xfs_perag_put(pag); | |
698 | ||
699 | *ipp = ip; | |
700 | ||
701 | /* | |
58c90473 | 702 | * If we have a real type for an on-disk inode, we can setup the inode |
33479e05 DC |
703 | * now. If it's a new inode being created, xfs_ialloc will handle it. |
704 | */ | |
c19b3b05 | 705 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 706 | xfs_setup_existing_inode(ip); |
33479e05 DC |
707 | return 0; |
708 | ||
709 | out_error_or_again: | |
378f681c | 710 | if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) { |
33479e05 DC |
711 | delay(1); |
712 | goto again; | |
713 | } | |
714 | xfs_perag_put(pag); | |
715 | return error; | |
716 | } | |
717 | ||
378f681c DW |
718 | /* |
719 | * "Is this a cached inode that's also allocated?" | |
720 | * | |
721 | * Look up an inode by number in the given file system. If the inode is | |
722 | * in cache and isn't in purgatory, return 1 if the inode is allocated | |
723 | * and 0 if it is not. For all other cases (not in cache, being torn | |
724 | * down, etc.), return a negative error code. | |
725 | * | |
726 | * The caller has to prevent inode allocation and freeing activity, | |
727 | * presumably by locking the AGI buffer. This is to ensure that an | |
728 | * inode cannot transition from allocated to freed until the caller is | |
729 | * ready to allow that. If the inode is in an intermediate state (new, | |
730 | * reclaimable, or being reclaimed), -EAGAIN will be returned; if the | |
731 | * inode is not in the cache, -ENOENT will be returned. The caller must | |
732 | * deal with these scenarios appropriately. | |
733 | * | |
734 | * This is a specialized use case for the online scrubber; if you're | |
735 | * reading this, you probably want xfs_iget. | |
736 | */ | |
737 | int | |
738 | xfs_icache_inode_is_allocated( | |
739 | struct xfs_mount *mp, | |
740 | struct xfs_trans *tp, | |
741 | xfs_ino_t ino, | |
742 | bool *inuse) | |
743 | { | |
744 | struct xfs_inode *ip; | |
745 | int error; | |
746 | ||
747 | error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip); | |
748 | if (error) | |
749 | return error; | |
750 | ||
751 | *inuse = !!(VFS_I(ip)->i_mode); | |
44a8736b | 752 | xfs_irele(ip); |
378f681c DW |
753 | return 0; |
754 | } | |
755 | ||
78ae5256 DC |
756 | /* |
757 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
758 | * radix tree lookups to a minimum. The batch size is a trade off between | |
759 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
760 | * be too greedy. | |
761 | */ | |
762 | #define XFS_LOOKUP_BATCH 32 | |
763 | ||
39b1cfd7 DW |
764 | /* |
765 | * Decide if the given @ip is eligible to be a part of the inode walk, and | |
766 | * grab it if so. Returns true if it's ready to go or false if we should just | |
767 | * ignore it. | |
768 | */ | |
769 | STATIC bool | |
042f65f4 | 770 | xfs_inode_walk_ag_grab( |
ae2c4ac2 BF |
771 | struct xfs_inode *ip, |
772 | int flags) | |
e13de955 DC |
773 | { |
774 | struct inode *inode = VFS_I(ip); | |
042f65f4 | 775 | bool newinos = !!(flags & XFS_INODE_WALK_INEW_WAIT); |
e13de955 | 776 | |
1a3e8f3d DC |
777 | ASSERT(rcu_read_lock_held()); |
778 | ||
779 | /* | |
780 | * check for stale RCU freed inode | |
781 | * | |
782 | * If the inode has been reallocated, it doesn't matter if it's not in | |
783 | * the AG we are walking - we are walking for writeback, so if it | |
784 | * passes all the "valid inode" checks and is dirty, then we'll write | |
785 | * it back anyway. If it has been reallocated and still being | |
786 | * initialised, the XFS_INEW check below will catch it. | |
787 | */ | |
788 | spin_lock(&ip->i_flags_lock); | |
789 | if (!ip->i_ino) | |
790 | goto out_unlock_noent; | |
791 | ||
792 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
ae2c4ac2 BF |
793 | if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) || |
794 | __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
1a3e8f3d DC |
795 | goto out_unlock_noent; |
796 | spin_unlock(&ip->i_flags_lock); | |
797 | ||
e13de955 DC |
798 | /* nothing to sync during shutdown */ |
799 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
39b1cfd7 | 800 | return false; |
e13de955 | 801 | |
e13de955 DC |
802 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
803 | if (!igrab(inode)) | |
39b1cfd7 | 804 | return false; |
e13de955 | 805 | |
e13de955 | 806 | /* inode is valid */ |
39b1cfd7 | 807 | return true; |
1a3e8f3d DC |
808 | |
809 | out_unlock_noent: | |
810 | spin_unlock(&ip->i_flags_lock); | |
39b1cfd7 | 811 | return false; |
e13de955 DC |
812 | } |
813 | ||
5662d38c DW |
814 | /* |
815 | * For a given per-AG structure @pag, grab, @execute, and rele all incore | |
816 | * inodes with the given radix tree @tag. | |
817 | */ | |
75f3cb13 | 818 | STATIC int |
042f65f4 | 819 | xfs_inode_walk_ag( |
5017e97d | 820 | struct xfs_perag *pag, |
964176bd | 821 | int iter_flags, |
390600f8 | 822 | int (*execute)(struct xfs_inode *ip, void *args), |
a454f742 | 823 | void *args, |
964176bd | 824 | int tag) |
75f3cb13 | 825 | { |
964176bd | 826 | struct xfs_mount *mp = pag->pag_mount; |
75f3cb13 DC |
827 | uint32_t first_index; |
828 | int last_error = 0; | |
829 | int skipped; | |
7e88d314 | 830 | bool done; |
78ae5256 | 831 | int nr_found; |
75f3cb13 DC |
832 | |
833 | restart: | |
7e88d314 | 834 | done = false; |
75f3cb13 DC |
835 | skipped = 0; |
836 | first_index = 0; | |
78ae5256 | 837 | nr_found = 0; |
75f3cb13 | 838 | do { |
78ae5256 | 839 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
75f3cb13 | 840 | int error = 0; |
78ae5256 | 841 | int i; |
75f3cb13 | 842 | |
1a3e8f3d | 843 | rcu_read_lock(); |
a454f742 | 844 | |
fc96be95 | 845 | if (tag == XFS_ICI_NO_TAG) |
a454f742 | 846 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, |
78ae5256 DC |
847 | (void **)batch, first_index, |
848 | XFS_LOOKUP_BATCH); | |
a454f742 BF |
849 | else |
850 | nr_found = radix_tree_gang_lookup_tag( | |
851 | &pag->pag_ici_root, | |
852 | (void **) batch, first_index, | |
853 | XFS_LOOKUP_BATCH, tag); | |
854 | ||
65d0f205 | 855 | if (!nr_found) { |
1a3e8f3d | 856 | rcu_read_unlock(); |
75f3cb13 | 857 | break; |
c8e20be0 | 858 | } |
75f3cb13 | 859 | |
65d0f205 | 860 | /* |
78ae5256 DC |
861 | * Grab the inodes before we drop the lock. if we found |
862 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 863 | */ |
78ae5256 DC |
864 | for (i = 0; i < nr_found; i++) { |
865 | struct xfs_inode *ip = batch[i]; | |
866 | ||
042f65f4 | 867 | if (done || !xfs_inode_walk_ag_grab(ip, iter_flags)) |
78ae5256 DC |
868 | batch[i] = NULL; |
869 | ||
870 | /* | |
1a3e8f3d DC |
871 | * Update the index for the next lookup. Catch |
872 | * overflows into the next AG range which can occur if | |
873 | * we have inodes in the last block of the AG and we | |
874 | * are currently pointing to the last inode. | |
875 | * | |
876 | * Because we may see inodes that are from the wrong AG | |
877 | * due to RCU freeing and reallocation, only update the | |
878 | * index if it lies in this AG. It was a race that lead | |
879 | * us to see this inode, so another lookup from the | |
880 | * same index will not find it again. | |
78ae5256 | 881 | */ |
1a3e8f3d DC |
882 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
883 | continue; | |
78ae5256 DC |
884 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
885 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
7e88d314 | 886 | done = true; |
e13de955 | 887 | } |
78ae5256 DC |
888 | |
889 | /* unlock now we've grabbed the inodes. */ | |
1a3e8f3d | 890 | rcu_read_unlock(); |
e13de955 | 891 | |
78ae5256 DC |
892 | for (i = 0; i < nr_found; i++) { |
893 | if (!batch[i]) | |
894 | continue; | |
042f65f4 | 895 | if ((iter_flags & XFS_INODE_WALK_INEW_WAIT) && |
ae2c4ac2 BF |
896 | xfs_iflags_test(batch[i], XFS_INEW)) |
897 | xfs_inew_wait(batch[i]); | |
390600f8 | 898 | error = execute(batch[i], args); |
44a8736b | 899 | xfs_irele(batch[i]); |
2451337d | 900 | if (error == -EAGAIN) { |
78ae5256 DC |
901 | skipped++; |
902 | continue; | |
903 | } | |
2451337d | 904 | if (error && last_error != -EFSCORRUPTED) |
78ae5256 | 905 | last_error = error; |
75f3cb13 | 906 | } |
c8e20be0 DC |
907 | |
908 | /* bail out if the filesystem is corrupted. */ | |
2451337d | 909 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
910 | break; |
911 | ||
8daaa831 DC |
912 | cond_resched(); |
913 | ||
78ae5256 | 914 | } while (nr_found && !done); |
75f3cb13 DC |
915 | |
916 | if (skipped) { | |
917 | delay(1); | |
918 | goto restart; | |
919 | } | |
75f3cb13 DC |
920 | return last_error; |
921 | } | |
922 | ||
5662d38c DW |
923 | /* Fetch the next (possibly tagged) per-AG structure. */ |
924 | static inline struct xfs_perag * | |
925 | xfs_inode_walk_get_perag( | |
926 | struct xfs_mount *mp, | |
927 | xfs_agnumber_t agno, | |
928 | int tag) | |
929 | { | |
930 | if (tag == XFS_ICI_NO_TAG) | |
931 | return xfs_perag_get(mp, agno); | |
932 | return xfs_perag_get_tag(mp, agno, tag); | |
933 | } | |
934 | ||
935 | /* | |
936 | * Call the @execute function on all incore inodes matching the radix tree | |
937 | * @tag. | |
938 | */ | |
939 | int | |
042f65f4 | 940 | xfs_inode_walk( |
5662d38c DW |
941 | struct xfs_mount *mp, |
942 | int iter_flags, | |
943 | int (*execute)(struct xfs_inode *ip, void *args), | |
944 | void *args, | |
945 | int tag) | |
946 | { | |
947 | struct xfs_perag *pag; | |
948 | int error = 0; | |
949 | int last_error = 0; | |
950 | xfs_agnumber_t ag; | |
951 | ||
952 | ag = 0; | |
953 | while ((pag = xfs_inode_walk_get_perag(mp, ag, tag))) { | |
954 | ag = pag->pag_agno + 1; | |
964176bd | 955 | error = xfs_inode_walk_ag(pag, iter_flags, execute, args, tag); |
5662d38c DW |
956 | xfs_perag_put(pag); |
957 | if (error) { | |
958 | last_error = error; | |
959 | if (error == -EFSCORRUPTED) | |
960 | break; | |
961 | } | |
962 | } | |
963 | return last_error; | |
964 | } | |
965 | ||
579b62fa BF |
966 | /* |
967 | * Background scanning to trim post-EOF preallocated space. This is queued | |
b9fe5052 | 968 | * based on the 'speculative_prealloc_lifetime' tunable (5m by default). |
579b62fa | 969 | */ |
fa5a4f57 | 970 | void |
579b62fa BF |
971 | xfs_queue_eofblocks( |
972 | struct xfs_mount *mp) | |
973 | { | |
974 | rcu_read_lock(); | |
975 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG)) | |
976 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
977 | &mp->m_eofblocks_work, | |
978 | msecs_to_jiffies(xfs_eofb_secs * 1000)); | |
979 | rcu_read_unlock(); | |
980 | } | |
981 | ||
982 | void | |
983 | xfs_eofblocks_worker( | |
984 | struct work_struct *work) | |
985 | { | |
986 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
987 | struct xfs_mount, m_eofblocks_work); | |
4b674b9a BF |
988 | |
989 | if (!sb_start_write_trylock(mp->m_super)) | |
990 | return; | |
579b62fa | 991 | xfs_icache_free_eofblocks(mp, NULL); |
4b674b9a BF |
992 | sb_end_write(mp->m_super); |
993 | ||
579b62fa BF |
994 | xfs_queue_eofblocks(mp); |
995 | } | |
996 | ||
83104d44 DW |
997 | /* |
998 | * Background scanning to trim preallocated CoW space. This is queued | |
999 | * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default). | |
1000 | * (We'll just piggyback on the post-EOF prealloc space workqueue.) | |
1001 | */ | |
10ddf64e | 1002 | void |
83104d44 DW |
1003 | xfs_queue_cowblocks( |
1004 | struct xfs_mount *mp) | |
1005 | { | |
1006 | rcu_read_lock(); | |
1007 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG)) | |
1008 | queue_delayed_work(mp->m_eofblocks_workqueue, | |
1009 | &mp->m_cowblocks_work, | |
1010 | msecs_to_jiffies(xfs_cowb_secs * 1000)); | |
1011 | rcu_read_unlock(); | |
1012 | } | |
1013 | ||
1014 | void | |
1015 | xfs_cowblocks_worker( | |
1016 | struct work_struct *work) | |
1017 | { | |
1018 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
1019 | struct xfs_mount, m_cowblocks_work); | |
4b674b9a BF |
1020 | |
1021 | if (!sb_start_write_trylock(mp->m_super)) | |
1022 | return; | |
83104d44 | 1023 | xfs_icache_free_cowblocks(mp, NULL); |
4b674b9a BF |
1024 | sb_end_write(mp->m_super); |
1025 | ||
83104d44 DW |
1026 | xfs_queue_cowblocks(mp); |
1027 | } | |
1028 | ||
e3a20c0b DC |
1029 | /* |
1030 | * Grab the inode for reclaim exclusively. | |
1031 | * Return 0 if we grabbed it, non-zero otherwise. | |
1032 | */ | |
1033 | STATIC int | |
1034 | xfs_reclaim_inode_grab( | |
1035 | struct xfs_inode *ip, | |
1036 | int flags) | |
1037 | { | |
1a3e8f3d DC |
1038 | ASSERT(rcu_read_lock_held()); |
1039 | ||
1040 | /* quick check for stale RCU freed inode */ | |
1041 | if (!ip->i_ino) | |
1042 | return 1; | |
e3a20c0b DC |
1043 | |
1044 | /* | |
474fce06 CH |
1045 | * If we are asked for non-blocking operation, do unlocked checks to |
1046 | * see if the inode already is being flushed or in reclaim to avoid | |
1047 | * lock traffic. | |
e3a20c0b DC |
1048 | */ |
1049 | if ((flags & SYNC_TRYLOCK) && | |
474fce06 | 1050 | __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM)) |
e3a20c0b | 1051 | return 1; |
e3a20c0b DC |
1052 | |
1053 | /* | |
1054 | * The radix tree lock here protects a thread in xfs_iget from racing | |
1055 | * with us starting reclaim on the inode. Once we have the | |
1056 | * XFS_IRECLAIM flag set it will not touch us. | |
1a3e8f3d DC |
1057 | * |
1058 | * Due to RCU lookup, we may find inodes that have been freed and only | |
1059 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that | |
1060 | * aren't candidates for reclaim at all, so we must check the | |
1061 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. | |
e3a20c0b DC |
1062 | */ |
1063 | spin_lock(&ip->i_flags_lock); | |
1a3e8f3d DC |
1064 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
1065 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
1066 | /* not a reclaim candidate. */ | |
e3a20c0b DC |
1067 | spin_unlock(&ip->i_flags_lock); |
1068 | return 1; | |
1069 | } | |
1070 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
1071 | spin_unlock(&ip->i_flags_lock); | |
1072 | return 0; | |
1073 | } | |
1074 | ||
777df5af | 1075 | /* |
8a48088f CH |
1076 | * Inodes in different states need to be treated differently. The following |
1077 | * table lists the inode states and the reclaim actions necessary: | |
777df5af DC |
1078 | * |
1079 | * inode state iflush ret required action | |
1080 | * --------------- ---------- --------------- | |
1081 | * bad - reclaim | |
1082 | * shutdown EIO unpin and reclaim | |
1083 | * clean, unpinned 0 reclaim | |
1084 | * stale, unpinned 0 reclaim | |
c854363e DC |
1085 | * clean, pinned(*) 0 requeue |
1086 | * stale, pinned EAGAIN requeue | |
8a48088f CH |
1087 | * dirty, async - requeue |
1088 | * dirty, sync 0 reclaim | |
777df5af DC |
1089 | * |
1090 | * (*) dgc: I don't think the clean, pinned state is possible but it gets | |
1091 | * handled anyway given the order of checks implemented. | |
1092 | * | |
c854363e DC |
1093 | * Also, because we get the flush lock first, we know that any inode that has |
1094 | * been flushed delwri has had the flush completed by the time we check that | |
8a48088f | 1095 | * the inode is clean. |
c854363e | 1096 | * |
8a48088f CH |
1097 | * Note that because the inode is flushed delayed write by AIL pushing, the |
1098 | * flush lock may already be held here and waiting on it can result in very | |
1099 | * long latencies. Hence for sync reclaims, where we wait on the flush lock, | |
1100 | * the caller should push the AIL first before trying to reclaim inodes to | |
1101 | * minimise the amount of time spent waiting. For background relaim, we only | |
1102 | * bother to reclaim clean inodes anyway. | |
c854363e | 1103 | * |
777df5af DC |
1104 | * Hence the order of actions after gaining the locks should be: |
1105 | * bad => reclaim | |
1106 | * shutdown => unpin and reclaim | |
8a48088f | 1107 | * pinned, async => requeue |
c854363e | 1108 | * pinned, sync => unpin |
777df5af DC |
1109 | * stale => reclaim |
1110 | * clean => reclaim | |
8a48088f | 1111 | * dirty, async => requeue |
c854363e | 1112 | * dirty, sync => flush, wait and reclaim |
777df5af | 1113 | */ |
75f3cb13 | 1114 | STATIC int |
c8e20be0 | 1115 | xfs_reclaim_inode( |
75f3cb13 DC |
1116 | struct xfs_inode *ip, |
1117 | struct xfs_perag *pag, | |
c8e20be0 | 1118 | int sync_mode) |
fce08f2f | 1119 | { |
4c46819a | 1120 | struct xfs_buf *bp = NULL; |
8a17d7dd | 1121 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
4c46819a | 1122 | int error; |
777df5af | 1123 | |
1bfd8d04 DC |
1124 | restart: |
1125 | error = 0; | |
c8e20be0 | 1126 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
c854363e DC |
1127 | if (!xfs_iflock_nowait(ip)) { |
1128 | if (!(sync_mode & SYNC_WAIT)) | |
1129 | goto out; | |
1130 | xfs_iflock(ip); | |
1131 | } | |
7a3be02b | 1132 | |
777df5af DC |
1133 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1134 | xfs_iunpin_wait(ip); | |
98efe8af | 1135 | /* xfs_iflush_abort() drops the flush lock */ |
88fc1879 | 1136 | xfs_iflush_abort(ip); |
777df5af DC |
1137 | goto reclaim; |
1138 | } | |
c854363e | 1139 | if (xfs_ipincount(ip)) { |
8a48088f CH |
1140 | if (!(sync_mode & SYNC_WAIT)) |
1141 | goto out_ifunlock; | |
777df5af | 1142 | xfs_iunpin_wait(ip); |
c854363e | 1143 | } |
96355d5a | 1144 | if (xfs_inode_clean(ip)) { |
98efe8af | 1145 | xfs_ifunlock(ip); |
777df5af | 1146 | goto reclaim; |
98efe8af | 1147 | } |
777df5af | 1148 | |
8a48088f CH |
1149 | /* |
1150 | * Never flush out dirty data during non-blocking reclaim, as it would | |
1151 | * just contend with AIL pushing trying to do the same job. | |
1152 | */ | |
1153 | if (!(sync_mode & SYNC_WAIT)) | |
1154 | goto out_ifunlock; | |
1155 | ||
1bfd8d04 DC |
1156 | /* |
1157 | * Now we have an inode that needs flushing. | |
1158 | * | |
4c46819a | 1159 | * Note that xfs_iflush will never block on the inode buffer lock, as |
1bfd8d04 | 1160 | * xfs_ifree_cluster() can lock the inode buffer before it locks the |
4c46819a | 1161 | * ip->i_lock, and we are doing the exact opposite here. As a result, |
475ee413 CH |
1162 | * doing a blocking xfs_imap_to_bp() to get the cluster buffer would |
1163 | * result in an ABBA deadlock with xfs_ifree_cluster(). | |
1bfd8d04 DC |
1164 | * |
1165 | * As xfs_ifree_cluser() must gather all inodes that are active in the | |
1166 | * cache to mark them stale, if we hit this case we don't actually want | |
1167 | * to do IO here - we want the inode marked stale so we can simply | |
4c46819a CH |
1168 | * reclaim it. Hence if we get an EAGAIN error here, just unlock the |
1169 | * inode, back off and try again. Hopefully the next pass through will | |
1170 | * see the stale flag set on the inode. | |
1bfd8d04 | 1171 | */ |
4c46819a | 1172 | error = xfs_iflush(ip, &bp); |
2451337d | 1173 | if (error == -EAGAIN) { |
8a48088f CH |
1174 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1175 | /* backoff longer than in xfs_ifree_cluster */ | |
1176 | delay(2); | |
1177 | goto restart; | |
c854363e | 1178 | } |
c854363e | 1179 | |
4c46819a CH |
1180 | if (!error) { |
1181 | error = xfs_bwrite(bp); | |
1182 | xfs_buf_relse(bp); | |
1183 | } | |
1184 | ||
777df5af | 1185 | reclaim: |
98efe8af BF |
1186 | ASSERT(!xfs_isiflocked(ip)); |
1187 | ||
8a17d7dd DC |
1188 | /* |
1189 | * Because we use RCU freeing we need to ensure the inode always appears | |
1190 | * to be reclaimed with an invalid inode number when in the free state. | |
98efe8af | 1191 | * We do this as early as possible under the ILOCK so that |
f2e9ad21 OS |
1192 | * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to |
1193 | * detect races with us here. By doing this, we guarantee that once | |
1194 | * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that | |
1195 | * it will see either a valid inode that will serialise correctly, or it | |
1196 | * will see an invalid inode that it can skip. | |
8a17d7dd DC |
1197 | */ |
1198 | spin_lock(&ip->i_flags_lock); | |
1199 | ip->i_flags = XFS_IRECLAIM; | |
1200 | ip->i_ino = 0; | |
1201 | spin_unlock(&ip->i_flags_lock); | |
1202 | ||
c8e20be0 | 1203 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1204 | |
ff6d6af2 | 1205 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
1206 | /* |
1207 | * Remove the inode from the per-AG radix tree. | |
1208 | * | |
1209 | * Because radix_tree_delete won't complain even if the item was never | |
1210 | * added to the tree assert that it's been there before to catch | |
1211 | * problems with the inode life time early on. | |
1212 | */ | |
1a427ab0 | 1213 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 1214 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 1215 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 1216 | ASSERT(0); |
545c0889 | 1217 | xfs_perag_clear_reclaim_tag(pag); |
1a427ab0 | 1218 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
1219 | |
1220 | /* | |
1221 | * Here we do an (almost) spurious inode lock in order to coordinate | |
1222 | * with inode cache radix tree lookups. This is because the lookup | |
1223 | * can reference the inodes in the cache without taking references. | |
1224 | * | |
1225 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 1226 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 1227 | */ |
ad637a10 | 1228 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1229 | xfs_qm_dqdetach(ip); |
ad637a10 | 1230 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
96355d5a | 1231 | ASSERT(xfs_inode_clean(ip)); |
2f11feab | 1232 | |
8a17d7dd | 1233 | __xfs_inode_free(ip); |
ad637a10 | 1234 | return error; |
8a48088f CH |
1235 | |
1236 | out_ifunlock: | |
1237 | xfs_ifunlock(ip); | |
1238 | out: | |
1239 | xfs_iflags_clear(ip, XFS_IRECLAIM); | |
1240 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1241 | /* | |
2451337d | 1242 | * We could return -EAGAIN here to make reclaim rescan the inode tree in |
8a48088f | 1243 | * a short while. However, this just burns CPU time scanning the tree |
5889608d DC |
1244 | * waiting for IO to complete and the reclaim work never goes back to |
1245 | * the idle state. Instead, return 0 to let the next scheduled | |
1246 | * background reclaim attempt to reclaim the inode again. | |
8a48088f CH |
1247 | */ |
1248 | return 0; | |
7a3be02b DC |
1249 | } |
1250 | ||
65d0f205 DC |
1251 | /* |
1252 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
1253 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
1254 | * then a shut down during filesystem unmount reclaim walk leak all the | |
1255 | * unreclaimed inodes. | |
1256 | */ | |
33479e05 | 1257 | STATIC int |
65d0f205 DC |
1258 | xfs_reclaim_inodes_ag( |
1259 | struct xfs_mount *mp, | |
1260 | int flags, | |
1261 | int *nr_to_scan) | |
1262 | { | |
1263 | struct xfs_perag *pag; | |
1264 | int error = 0; | |
1265 | int last_error = 0; | |
1266 | xfs_agnumber_t ag; | |
69b491c2 DC |
1267 | int trylock = flags & SYNC_TRYLOCK; |
1268 | int skipped; | |
65d0f205 | 1269 | |
69b491c2 | 1270 | restart: |
65d0f205 | 1271 | ag = 0; |
69b491c2 | 1272 | skipped = 0; |
65d0f205 DC |
1273 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1274 | unsigned long first_index = 0; | |
1275 | int done = 0; | |
e3a20c0b | 1276 | int nr_found = 0; |
65d0f205 DC |
1277 | |
1278 | ag = pag->pag_agno + 1; | |
1279 | ||
69b491c2 DC |
1280 | if (trylock) { |
1281 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { | |
1282 | skipped++; | |
f83282a8 | 1283 | xfs_perag_put(pag); |
69b491c2 DC |
1284 | continue; |
1285 | } | |
1286 | first_index = pag->pag_ici_reclaim_cursor; | |
1287 | } else | |
1288 | mutex_lock(&pag->pag_ici_reclaim_lock); | |
1289 | ||
65d0f205 | 1290 | do { |
e3a20c0b DC |
1291 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
1292 | int i; | |
65d0f205 | 1293 | |
1a3e8f3d | 1294 | rcu_read_lock(); |
e3a20c0b DC |
1295 | nr_found = radix_tree_gang_lookup_tag( |
1296 | &pag->pag_ici_root, | |
1297 | (void **)batch, first_index, | |
1298 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
1299 | XFS_ICI_RECLAIM_TAG); |
1300 | if (!nr_found) { | |
b2232219 | 1301 | done = 1; |
1a3e8f3d | 1302 | rcu_read_unlock(); |
65d0f205 DC |
1303 | break; |
1304 | } | |
1305 | ||
1306 | /* | |
e3a20c0b DC |
1307 | * Grab the inodes before we drop the lock. if we found |
1308 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 1309 | */ |
e3a20c0b DC |
1310 | for (i = 0; i < nr_found; i++) { |
1311 | struct xfs_inode *ip = batch[i]; | |
1312 | ||
1313 | if (done || xfs_reclaim_inode_grab(ip, flags)) | |
1314 | batch[i] = NULL; | |
1315 | ||
1316 | /* | |
1317 | * Update the index for the next lookup. Catch | |
1318 | * overflows into the next AG range which can | |
1319 | * occur if we have inodes in the last block of | |
1320 | * the AG and we are currently pointing to the | |
1321 | * last inode. | |
1a3e8f3d DC |
1322 | * |
1323 | * Because we may see inodes that are from the | |
1324 | * wrong AG due to RCU freeing and | |
1325 | * reallocation, only update the index if it | |
1326 | * lies in this AG. It was a race that lead us | |
1327 | * to see this inode, so another lookup from | |
1328 | * the same index will not find it again. | |
e3a20c0b | 1329 | */ |
1a3e8f3d DC |
1330 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
1331 | pag->pag_agno) | |
1332 | continue; | |
e3a20c0b DC |
1333 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
1334 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1335 | done = 1; | |
1336 | } | |
65d0f205 | 1337 | |
e3a20c0b | 1338 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 1339 | rcu_read_unlock(); |
e3a20c0b DC |
1340 | |
1341 | for (i = 0; i < nr_found; i++) { | |
1342 | if (!batch[i]) | |
1343 | continue; | |
1344 | error = xfs_reclaim_inode(batch[i], pag, flags); | |
2451337d | 1345 | if (error && last_error != -EFSCORRUPTED) |
e3a20c0b DC |
1346 | last_error = error; |
1347 | } | |
1348 | ||
1349 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
65d0f205 | 1350 | |
8daaa831 DC |
1351 | cond_resched(); |
1352 | ||
e3a20c0b | 1353 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 1354 | |
69b491c2 DC |
1355 | if (trylock && !done) |
1356 | pag->pag_ici_reclaim_cursor = first_index; | |
1357 | else | |
1358 | pag->pag_ici_reclaim_cursor = 0; | |
1359 | mutex_unlock(&pag->pag_ici_reclaim_lock); | |
65d0f205 DC |
1360 | xfs_perag_put(pag); |
1361 | } | |
69b491c2 DC |
1362 | |
1363 | /* | |
1364 | * if we skipped any AG, and we still have scan count remaining, do | |
1365 | * another pass this time using blocking reclaim semantics (i.e | |
1366 | * waiting on the reclaim locks and ignoring the reclaim cursors). This | |
1367 | * ensure that when we get more reclaimers than AGs we block rather | |
1368 | * than spin trying to execute reclaim. | |
1369 | */ | |
8daaa831 | 1370 | if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) { |
69b491c2 DC |
1371 | trylock = 0; |
1372 | goto restart; | |
1373 | } | |
b474c7ae | 1374 | return last_error; |
65d0f205 DC |
1375 | } |
1376 | ||
7a3be02b DC |
1377 | int |
1378 | xfs_reclaim_inodes( | |
1379 | xfs_mount_t *mp, | |
7a3be02b DC |
1380 | int mode) |
1381 | { | |
65d0f205 DC |
1382 | int nr_to_scan = INT_MAX; |
1383 | ||
1384 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); | |
9bf729c0 DC |
1385 | } |
1386 | ||
1387 | /* | |
8daaa831 | 1388 | * Scan a certain number of inodes for reclaim. |
a7b339f1 DC |
1389 | * |
1390 | * When called we make sure that there is a background (fast) inode reclaim in | |
8daaa831 | 1391 | * progress, while we will throttle the speed of reclaim via doing synchronous |
a7b339f1 DC |
1392 | * reclaim of inodes. That means if we come across dirty inodes, we wait for |
1393 | * them to be cleaned, which we hope will not be very long due to the | |
1394 | * background walker having already kicked the IO off on those dirty inodes. | |
9bf729c0 | 1395 | */ |
0a234c6d | 1396 | long |
8daaa831 DC |
1397 | xfs_reclaim_inodes_nr( |
1398 | struct xfs_mount *mp, | |
1399 | int nr_to_scan) | |
9bf729c0 | 1400 | { |
8daaa831 | 1401 | /* kick background reclaimer and push the AIL */ |
5889608d | 1402 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1403 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1404 | |
993f951f | 1405 | return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK, &nr_to_scan); |
8daaa831 | 1406 | } |
9bf729c0 | 1407 | |
8daaa831 DC |
1408 | /* |
1409 | * Return the number of reclaimable inodes in the filesystem for | |
1410 | * the shrinker to determine how much to reclaim. | |
1411 | */ | |
1412 | int | |
1413 | xfs_reclaim_inodes_count( | |
1414 | struct xfs_mount *mp) | |
1415 | { | |
1416 | struct xfs_perag *pag; | |
1417 | xfs_agnumber_t ag = 0; | |
1418 | int reclaimable = 0; | |
9bf729c0 | 1419 | |
65d0f205 DC |
1420 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1421 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1422 | reclaimable += pag->pag_ici_reclaimable; |
1423 | xfs_perag_put(pag); | |
9bf729c0 | 1424 | } |
9bf729c0 DC |
1425 | return reclaimable; |
1426 | } | |
1427 | ||
39b1cfd7 | 1428 | STATIC bool |
3e3f9f58 BF |
1429 | xfs_inode_match_id( |
1430 | struct xfs_inode *ip, | |
1431 | struct xfs_eofblocks *eofb) | |
1432 | { | |
b9fe5052 DE |
1433 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && |
1434 | !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
39b1cfd7 | 1435 | return false; |
3e3f9f58 | 1436 | |
b9fe5052 DE |
1437 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && |
1438 | !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
39b1cfd7 | 1439 | return false; |
1b556048 | 1440 | |
b9fe5052 | 1441 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && |
de7a866f | 1442 | ip->i_d.di_projid != eofb->eof_prid) |
39b1cfd7 | 1443 | return false; |
1b556048 | 1444 | |
39b1cfd7 | 1445 | return true; |
3e3f9f58 BF |
1446 | } |
1447 | ||
f4526397 BF |
1448 | /* |
1449 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1450 | * criteria match. This is for global/internal scans only. | |
1451 | */ | |
39b1cfd7 | 1452 | STATIC bool |
f4526397 BF |
1453 | xfs_inode_match_id_union( |
1454 | struct xfs_inode *ip, | |
1455 | struct xfs_eofblocks *eofb) | |
1456 | { | |
1457 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && | |
1458 | uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
39b1cfd7 | 1459 | return true; |
f4526397 BF |
1460 | |
1461 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && | |
1462 | gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
39b1cfd7 | 1463 | return true; |
f4526397 BF |
1464 | |
1465 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && | |
de7a866f | 1466 | ip->i_d.di_projid == eofb->eof_prid) |
39b1cfd7 | 1467 | return true; |
f4526397 | 1468 | |
39b1cfd7 | 1469 | return false; |
f4526397 BF |
1470 | } |
1471 | ||
a91bf992 DW |
1472 | /* |
1473 | * Is this inode @ip eligible for eof/cow block reclamation, given some | |
1474 | * filtering parameters @eofb? The inode is eligible if @eofb is null or | |
1475 | * if the predicate functions match. | |
1476 | */ | |
1477 | static bool | |
1478 | xfs_inode_matches_eofb( | |
1479 | struct xfs_inode *ip, | |
1480 | struct xfs_eofblocks *eofb) | |
1481 | { | |
39b1cfd7 | 1482 | bool match; |
a91bf992 DW |
1483 | |
1484 | if (!eofb) | |
1485 | return true; | |
1486 | ||
1487 | if (eofb->eof_flags & XFS_EOF_FLAGS_UNION) | |
1488 | match = xfs_inode_match_id_union(ip, eofb); | |
1489 | else | |
1490 | match = xfs_inode_match_id(ip, eofb); | |
1491 | if (!match) | |
1492 | return false; | |
1493 | ||
1494 | /* skip the inode if the file size is too small */ | |
1495 | if ((eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE) && | |
1496 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1497 | return false; | |
1498 | ||
1499 | return true; | |
1500 | } | |
1501 | ||
41176a68 BF |
1502 | STATIC int |
1503 | xfs_inode_free_eofblocks( | |
1504 | struct xfs_inode *ip, | |
41176a68 BF |
1505 | void *args) |
1506 | { | |
390600f8 DW |
1507 | struct xfs_eofblocks *eofb = args; |
1508 | bool wait; | |
390600f8 DW |
1509 | int ret; |
1510 | ||
1511 | wait = eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC); | |
5400da7d | 1512 | |
41176a68 BF |
1513 | if (!xfs_can_free_eofblocks(ip, false)) { |
1514 | /* inode could be preallocated or append-only */ | |
1515 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1516 | xfs_inode_clear_eofblocks_tag(ip); | |
1517 | return 0; | |
1518 | } | |
1519 | ||
1520 | /* | |
1521 | * If the mapping is dirty the operation can block and wait for some | |
1522 | * time. Unless we are waiting, skip it. | |
1523 | */ | |
390600f8 | 1524 | if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) |
41176a68 BF |
1525 | return 0; |
1526 | ||
a91bf992 DW |
1527 | if (!xfs_inode_matches_eofb(ip, eofb)) |
1528 | return 0; | |
3e3f9f58 | 1529 | |
a36b9261 BF |
1530 | /* |
1531 | * If the caller is waiting, return -EAGAIN to keep the background | |
1532 | * scanner moving and revisit the inode in a subsequent pass. | |
1533 | */ | |
c3155097 | 1534 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
390600f8 DW |
1535 | if (wait) |
1536 | return -EAGAIN; | |
1537 | return 0; | |
a36b9261 | 1538 | } |
390600f8 | 1539 | |
a36b9261 | 1540 | ret = xfs_free_eofblocks(ip); |
c3155097 | 1541 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
41176a68 BF |
1542 | |
1543 | return ret; | |
1544 | } | |
1545 | ||
83104d44 DW |
1546 | int |
1547 | xfs_icache_free_eofblocks( | |
1548 | struct xfs_mount *mp, | |
1549 | struct xfs_eofblocks *eofb) | |
1550 | { | |
042f65f4 | 1551 | return xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, eofb, |
83104d44 | 1552 | XFS_ICI_EOFBLOCKS_TAG); |
41176a68 BF |
1553 | } |
1554 | ||
dc06f398 BF |
1555 | /* |
1556 | * Run eofblocks scans on the quotas applicable to the inode. For inodes with | |
1557 | * multiple quotas, we don't know exactly which quota caused an allocation | |
1558 | * failure. We make a best effort by including each quota under low free space | |
1559 | * conditions (less than 1% free space) in the scan. | |
1560 | */ | |
83104d44 DW |
1561 | static int |
1562 | __xfs_inode_free_quota_eofblocks( | |
1563 | struct xfs_inode *ip, | |
1564 | int (*execute)(struct xfs_mount *mp, | |
1565 | struct xfs_eofblocks *eofb)) | |
dc06f398 BF |
1566 | { |
1567 | int scan = 0; | |
1568 | struct xfs_eofblocks eofb = {0}; | |
1569 | struct xfs_dquot *dq; | |
1570 | ||
dc06f398 | 1571 | /* |
c3155097 | 1572 | * Run a sync scan to increase effectiveness and use the union filter to |
dc06f398 BF |
1573 | * cover all applicable quotas in a single scan. |
1574 | */ | |
dc06f398 BF |
1575 | eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC; |
1576 | ||
1577 | if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) { | |
1578 | dq = xfs_inode_dquot(ip, XFS_DQ_USER); | |
1579 | if (dq && xfs_dquot_lowsp(dq)) { | |
1580 | eofb.eof_uid = VFS_I(ip)->i_uid; | |
1581 | eofb.eof_flags |= XFS_EOF_FLAGS_UID; | |
1582 | scan = 1; | |
1583 | } | |
1584 | } | |
1585 | ||
1586 | if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) { | |
1587 | dq = xfs_inode_dquot(ip, XFS_DQ_GROUP); | |
1588 | if (dq && xfs_dquot_lowsp(dq)) { | |
1589 | eofb.eof_gid = VFS_I(ip)->i_gid; | |
1590 | eofb.eof_flags |= XFS_EOF_FLAGS_GID; | |
1591 | scan = 1; | |
1592 | } | |
1593 | } | |
1594 | ||
1595 | if (scan) | |
83104d44 | 1596 | execute(ip->i_mount, &eofb); |
dc06f398 BF |
1597 | |
1598 | return scan; | |
1599 | } | |
1600 | ||
83104d44 DW |
1601 | int |
1602 | xfs_inode_free_quota_eofblocks( | |
1603 | struct xfs_inode *ip) | |
1604 | { | |
1605 | return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks); | |
1606 | } | |
1607 | ||
91aae6be DW |
1608 | static inline unsigned long |
1609 | xfs_iflag_for_tag( | |
1610 | int tag) | |
1611 | { | |
1612 | switch (tag) { | |
1613 | case XFS_ICI_EOFBLOCKS_TAG: | |
1614 | return XFS_IEOFBLOCKS; | |
1615 | case XFS_ICI_COWBLOCKS_TAG: | |
1616 | return XFS_ICOWBLOCKS; | |
1617 | default: | |
1618 | ASSERT(0); | |
1619 | return 0; | |
1620 | } | |
1621 | } | |
1622 | ||
83104d44 | 1623 | static void |
91aae6be | 1624 | __xfs_inode_set_blocks_tag( |
83104d44 DW |
1625 | xfs_inode_t *ip, |
1626 | void (*execute)(struct xfs_mount *mp), | |
1627 | void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno, | |
1628 | int error, unsigned long caller_ip), | |
1629 | int tag) | |
27b52867 BF |
1630 | { |
1631 | struct xfs_mount *mp = ip->i_mount; | |
1632 | struct xfs_perag *pag; | |
1633 | int tagged; | |
1634 | ||
85a6e764 CH |
1635 | /* |
1636 | * Don't bother locking the AG and looking up in the radix trees | |
1637 | * if we already know that we have the tag set. | |
1638 | */ | |
91aae6be | 1639 | if (ip->i_flags & xfs_iflag_for_tag(tag)) |
85a6e764 CH |
1640 | return; |
1641 | spin_lock(&ip->i_flags_lock); | |
91aae6be | 1642 | ip->i_flags |= xfs_iflag_for_tag(tag); |
85a6e764 CH |
1643 | spin_unlock(&ip->i_flags_lock); |
1644 | ||
27b52867 BF |
1645 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1646 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1647 | |
83104d44 | 1648 | tagged = radix_tree_tagged(&pag->pag_ici_root, tag); |
27b52867 | 1649 | radix_tree_tag_set(&pag->pag_ici_root, |
83104d44 | 1650 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag); |
27b52867 BF |
1651 | if (!tagged) { |
1652 | /* propagate the eofblocks tag up into the perag radix tree */ | |
1653 | spin_lock(&ip->i_mount->m_perag_lock); | |
1654 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
1655 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
83104d44 | 1656 | tag); |
27b52867 | 1657 | spin_unlock(&ip->i_mount->m_perag_lock); |
579b62fa BF |
1658 | |
1659 | /* kick off background trimming */ | |
83104d44 | 1660 | execute(ip->i_mount); |
27b52867 | 1661 | |
83104d44 | 1662 | set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_); |
27b52867 BF |
1663 | } |
1664 | ||
1665 | spin_unlock(&pag->pag_ici_lock); | |
1666 | xfs_perag_put(pag); | |
1667 | } | |
1668 | ||
1669 | void | |
83104d44 | 1670 | xfs_inode_set_eofblocks_tag( |
27b52867 | 1671 | xfs_inode_t *ip) |
83104d44 DW |
1672 | { |
1673 | trace_xfs_inode_set_eofblocks_tag(ip); | |
91aae6be | 1674 | return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks, |
83104d44 DW |
1675 | trace_xfs_perag_set_eofblocks, |
1676 | XFS_ICI_EOFBLOCKS_TAG); | |
1677 | } | |
1678 | ||
1679 | static void | |
91aae6be | 1680 | __xfs_inode_clear_blocks_tag( |
83104d44 DW |
1681 | xfs_inode_t *ip, |
1682 | void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno, | |
1683 | int error, unsigned long caller_ip), | |
1684 | int tag) | |
27b52867 BF |
1685 | { |
1686 | struct xfs_mount *mp = ip->i_mount; | |
1687 | struct xfs_perag *pag; | |
1688 | ||
85a6e764 | 1689 | spin_lock(&ip->i_flags_lock); |
91aae6be | 1690 | ip->i_flags &= ~xfs_iflag_for_tag(tag); |
85a6e764 CH |
1691 | spin_unlock(&ip->i_flags_lock); |
1692 | ||
27b52867 BF |
1693 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1694 | spin_lock(&pag->pag_ici_lock); | |
27b52867 BF |
1695 | |
1696 | radix_tree_tag_clear(&pag->pag_ici_root, | |
83104d44 DW |
1697 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag); |
1698 | if (!radix_tree_tagged(&pag->pag_ici_root, tag)) { | |
27b52867 BF |
1699 | /* clear the eofblocks tag from the perag radix tree */ |
1700 | spin_lock(&ip->i_mount->m_perag_lock); | |
1701 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
1702 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
83104d44 | 1703 | tag); |
27b52867 | 1704 | spin_unlock(&ip->i_mount->m_perag_lock); |
83104d44 | 1705 | clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_); |
27b52867 BF |
1706 | } |
1707 | ||
1708 | spin_unlock(&pag->pag_ici_lock); | |
1709 | xfs_perag_put(pag); | |
1710 | } | |
1711 | ||
83104d44 DW |
1712 | void |
1713 | xfs_inode_clear_eofblocks_tag( | |
1714 | xfs_inode_t *ip) | |
1715 | { | |
1716 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
91aae6be | 1717 | return __xfs_inode_clear_blocks_tag(ip, |
83104d44 DW |
1718 | trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG); |
1719 | } | |
1720 | ||
1721 | /* | |
be78ff0e DW |
1722 | * Set ourselves up to free CoW blocks from this file. If it's already clean |
1723 | * then we can bail out quickly, but otherwise we must back off if the file | |
1724 | * is undergoing some kind of write. | |
83104d44 | 1725 | */ |
be78ff0e DW |
1726 | static bool |
1727 | xfs_prep_free_cowblocks( | |
51d62690 | 1728 | struct xfs_inode *ip) |
83104d44 | 1729 | { |
39937234 BF |
1730 | /* |
1731 | * Just clear the tag if we have an empty cow fork or none at all. It's | |
1732 | * possible the inode was fully unshared since it was originally tagged. | |
1733 | */ | |
51d62690 | 1734 | if (!xfs_inode_has_cow_data(ip)) { |
83104d44 DW |
1735 | trace_xfs_inode_free_cowblocks_invalid(ip); |
1736 | xfs_inode_clear_cowblocks_tag(ip); | |
be78ff0e | 1737 | return false; |
83104d44 DW |
1738 | } |
1739 | ||
1740 | /* | |
1741 | * If the mapping is dirty or under writeback we cannot touch the | |
1742 | * CoW fork. Leave it alone if we're in the midst of a directio. | |
1743 | */ | |
a1b7a4de CH |
1744 | if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) || |
1745 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) || | |
83104d44 DW |
1746 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) || |
1747 | atomic_read(&VFS_I(ip)->i_dio_count)) | |
be78ff0e DW |
1748 | return false; |
1749 | ||
1750 | return true; | |
1751 | } | |
1752 | ||
1753 | /* | |
1754 | * Automatic CoW Reservation Freeing | |
1755 | * | |
1756 | * These functions automatically garbage collect leftover CoW reservations | |
1757 | * that were made on behalf of a cowextsize hint when we start to run out | |
1758 | * of quota or when the reservations sit around for too long. If the file | |
1759 | * has dirty pages or is undergoing writeback, its CoW reservations will | |
1760 | * be retained. | |
1761 | * | |
1762 | * The actual garbage collection piggybacks off the same code that runs | |
1763 | * the speculative EOF preallocation garbage collector. | |
1764 | */ | |
1765 | STATIC int | |
1766 | xfs_inode_free_cowblocks( | |
1767 | struct xfs_inode *ip, | |
be78ff0e DW |
1768 | void *args) |
1769 | { | |
1770 | struct xfs_eofblocks *eofb = args; | |
be78ff0e DW |
1771 | int ret = 0; |
1772 | ||
51d62690 | 1773 | if (!xfs_prep_free_cowblocks(ip)) |
83104d44 DW |
1774 | return 0; |
1775 | ||
a91bf992 DW |
1776 | if (!xfs_inode_matches_eofb(ip, eofb)) |
1777 | return 0; | |
83104d44 DW |
1778 | |
1779 | /* Free the CoW blocks */ | |
c3155097 BF |
1780 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
1781 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); | |
83104d44 | 1782 | |
be78ff0e DW |
1783 | /* |
1784 | * Check again, nobody else should be able to dirty blocks or change | |
1785 | * the reflink iflag now that we have the first two locks held. | |
1786 | */ | |
51d62690 | 1787 | if (xfs_prep_free_cowblocks(ip)) |
be78ff0e | 1788 | ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false); |
83104d44 | 1789 | |
c3155097 BF |
1790 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
1791 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
83104d44 DW |
1792 | |
1793 | return ret; | |
1794 | } | |
1795 | ||
1796 | int | |
1797 | xfs_icache_free_cowblocks( | |
1798 | struct xfs_mount *mp, | |
1799 | struct xfs_eofblocks *eofb) | |
1800 | { | |
042f65f4 | 1801 | return xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, eofb, |
83104d44 DW |
1802 | XFS_ICI_COWBLOCKS_TAG); |
1803 | } | |
1804 | ||
1805 | int | |
1806 | xfs_inode_free_quota_cowblocks( | |
1807 | struct xfs_inode *ip) | |
1808 | { | |
1809 | return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks); | |
1810 | } | |
1811 | ||
1812 | void | |
1813 | xfs_inode_set_cowblocks_tag( | |
1814 | xfs_inode_t *ip) | |
1815 | { | |
7b7381f0 | 1816 | trace_xfs_inode_set_cowblocks_tag(ip); |
91aae6be | 1817 | return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks, |
7b7381f0 | 1818 | trace_xfs_perag_set_cowblocks, |
83104d44 DW |
1819 | XFS_ICI_COWBLOCKS_TAG); |
1820 | } | |
1821 | ||
1822 | void | |
1823 | xfs_inode_clear_cowblocks_tag( | |
1824 | xfs_inode_t *ip) | |
1825 | { | |
7b7381f0 | 1826 | trace_xfs_inode_clear_cowblocks_tag(ip); |
91aae6be | 1827 | return __xfs_inode_clear_blocks_tag(ip, |
7b7381f0 | 1828 | trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG); |
83104d44 | 1829 | } |
d6b636eb DW |
1830 | |
1831 | /* Disable post-EOF and CoW block auto-reclamation. */ | |
1832 | void | |
ed30dcbd | 1833 | xfs_stop_block_reaping( |
d6b636eb DW |
1834 | struct xfs_mount *mp) |
1835 | { | |
1836 | cancel_delayed_work_sync(&mp->m_eofblocks_work); | |
1837 | cancel_delayed_work_sync(&mp->m_cowblocks_work); | |
1838 | } | |
1839 | ||
1840 | /* Enable post-EOF and CoW block auto-reclamation. */ | |
1841 | void | |
ed30dcbd | 1842 | xfs_start_block_reaping( |
d6b636eb DW |
1843 | struct xfs_mount *mp) |
1844 | { | |
1845 | xfs_queue_eofblocks(mp); | |
1846 | xfs_queue_cowblocks(mp); | |
1847 | } |