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 | |
1ad2cfe0 DW |
29 | /* |
30 | * Private inode cache walk flags for struct xfs_eofblocks. Must not coincide | |
31 | * with XFS_EOF_FLAGS_*. | |
32 | */ | |
33 | #define XFS_ICWALK_FLAG_DROP_UDQUOT (1U << 31) | |
34 | #define XFS_ICWALK_FLAG_DROP_GDQUOT (1U << 30) | |
35 | #define XFS_ICWALK_FLAG_DROP_PDQUOT (1U << 29) | |
36 | ||
37 | #define XFS_ICWALK_PRIVATE_FLAGS (XFS_ICWALK_FLAG_DROP_UDQUOT | \ | |
38 | XFS_ICWALK_FLAG_DROP_GDQUOT | \ | |
39 | XFS_ICWALK_FLAG_DROP_PDQUOT) | |
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 | /* | |
3050bd0b CM |
52 | * XXX: If this didn't occur in transactions, we could drop GFP_NOFAIL |
53 | * and return NULL here on ENOMEM. | |
33479e05 | 54 | */ |
3050bd0b CM |
55 | ip = kmem_cache_alloc(xfs_inode_zone, GFP_KERNEL | __GFP_NOFAIL); |
56 | ||
33479e05 | 57 | if (inode_init_always(mp->m_super, VFS_I(ip))) { |
377bcd5f | 58 | kmem_cache_free(xfs_inode_zone, ip); |
33479e05 DC |
59 | return NULL; |
60 | } | |
61 | ||
c19b3b05 DC |
62 | /* VFS doesn't initialise i_mode! */ |
63 | VFS_I(ip)->i_mode = 0; | |
64 | ||
ff6d6af2 | 65 | XFS_STATS_INC(mp, vn_active); |
33479e05 | 66 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
33479e05 DC |
67 | ASSERT(ip->i_ino == 0); |
68 | ||
33479e05 DC |
69 | /* initialise the xfs inode */ |
70 | ip->i_ino = ino; | |
71 | ip->i_mount = mp; | |
72 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
73 | ip->i_afp = NULL; | |
3993baeb | 74 | ip->i_cowfp = NULL; |
3ba738df | 75 | memset(&ip->i_df, 0, sizeof(ip->i_df)); |
33479e05 DC |
76 | ip->i_flags = 0; |
77 | ip->i_delayed_blks = 0; | |
3e09ab8f | 78 | ip->i_diflags2 = mp->m_ino_geo.new_diflags2; |
6e73a545 | 79 | ip->i_nblocks = 0; |
7821ea30 | 80 | ip->i_forkoff = 0; |
6772c1f1 DW |
81 | ip->i_sick = 0; |
82 | ip->i_checked = 0; | |
cb357bf3 DW |
83 | INIT_WORK(&ip->i_ioend_work, xfs_end_io); |
84 | INIT_LIST_HEAD(&ip->i_ioend_list); | |
85 | spin_lock_init(&ip->i_ioend_lock); | |
33479e05 DC |
86 | |
87 | return ip; | |
88 | } | |
89 | ||
90 | STATIC void | |
91 | xfs_inode_free_callback( | |
92 | struct rcu_head *head) | |
93 | { | |
94 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
95 | struct xfs_inode *ip = XFS_I(inode); | |
96 | ||
c19b3b05 | 97 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
98 | case S_IFREG: |
99 | case S_IFDIR: | |
100 | case S_IFLNK: | |
ef838512 | 101 | xfs_idestroy_fork(&ip->i_df); |
33479e05 DC |
102 | break; |
103 | } | |
104 | ||
ef838512 CH |
105 | if (ip->i_afp) { |
106 | xfs_idestroy_fork(ip->i_afp); | |
107 | kmem_cache_free(xfs_ifork_zone, ip->i_afp); | |
108 | } | |
109 | if (ip->i_cowfp) { | |
110 | xfs_idestroy_fork(ip->i_cowfp); | |
111 | kmem_cache_free(xfs_ifork_zone, ip->i_cowfp); | |
112 | } | |
33479e05 | 113 | if (ip->i_itemp) { |
22525c17 DC |
114 | ASSERT(!test_bit(XFS_LI_IN_AIL, |
115 | &ip->i_itemp->ili_item.li_flags)); | |
33479e05 DC |
116 | xfs_inode_item_destroy(ip); |
117 | ip->i_itemp = NULL; | |
118 | } | |
119 | ||
377bcd5f | 120 | kmem_cache_free(xfs_inode_zone, ip); |
1f2dcfe8 DC |
121 | } |
122 | ||
8a17d7dd DC |
123 | static void |
124 | __xfs_inode_free( | |
125 | struct xfs_inode *ip) | |
126 | { | |
127 | /* asserts to verify all state is correct here */ | |
128 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
48d55e2a | 129 | ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list)); |
8a17d7dd DC |
130 | XFS_STATS_DEC(ip->i_mount, vn_active); |
131 | ||
132 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
133 | } | |
134 | ||
1f2dcfe8 DC |
135 | void |
136 | xfs_inode_free( | |
137 | struct xfs_inode *ip) | |
138 | { | |
718ecc50 | 139 | ASSERT(!xfs_iflags_test(ip, XFS_IFLUSHING)); |
98efe8af | 140 | |
33479e05 DC |
141 | /* |
142 | * Because we use RCU freeing we need to ensure the inode always | |
143 | * appears to be reclaimed with an invalid inode number when in the | |
144 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
145 | * races. | |
146 | */ | |
147 | spin_lock(&ip->i_flags_lock); | |
148 | ip->i_flags = XFS_IRECLAIM; | |
149 | ip->i_ino = 0; | |
150 | spin_unlock(&ip->i_flags_lock); | |
151 | ||
8a17d7dd | 152 | __xfs_inode_free(ip); |
33479e05 DC |
153 | } |
154 | ||
ad438c40 | 155 | /* |
02511a5a DC |
156 | * Queue background inode reclaim work if there are reclaimable inodes and there |
157 | * isn't reclaim work already scheduled or in progress. | |
ad438c40 DC |
158 | */ |
159 | static void | |
160 | xfs_reclaim_work_queue( | |
161 | struct xfs_mount *mp) | |
162 | { | |
163 | ||
164 | rcu_read_lock(); | |
165 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
166 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
167 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
168 | } | |
169 | rcu_read_unlock(); | |
170 | } | |
171 | ||
ad438c40 DC |
172 | static void |
173 | xfs_perag_set_reclaim_tag( | |
174 | struct xfs_perag *pag) | |
175 | { | |
176 | struct xfs_mount *mp = pag->pag_mount; | |
177 | ||
95989c46 | 178 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
179 | if (pag->pag_ici_reclaimable++) |
180 | return; | |
181 | ||
182 | /* propagate the reclaim tag up into the perag radix tree */ | |
183 | spin_lock(&mp->m_perag_lock); | |
184 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, | |
185 | XFS_ICI_RECLAIM_TAG); | |
186 | spin_unlock(&mp->m_perag_lock); | |
187 | ||
188 | /* schedule periodic background inode reclaim */ | |
189 | xfs_reclaim_work_queue(mp); | |
190 | ||
191 | trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
192 | } | |
193 | ||
194 | static void | |
195 | xfs_perag_clear_reclaim_tag( | |
196 | struct xfs_perag *pag) | |
197 | { | |
198 | struct xfs_mount *mp = pag->pag_mount; | |
199 | ||
95989c46 | 200 | lockdep_assert_held(&pag->pag_ici_lock); |
ad438c40 DC |
201 | if (--pag->pag_ici_reclaimable) |
202 | return; | |
203 | ||
204 | /* clear the reclaim tag from the perag radix tree */ | |
205 | spin_lock(&mp->m_perag_lock); | |
206 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, | |
207 | XFS_ICI_RECLAIM_TAG); | |
208 | spin_unlock(&mp->m_perag_lock); | |
209 | trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_); | |
210 | } | |
211 | ||
212 | ||
213 | /* | |
214 | * We set the inode flag atomically with the radix tree tag. | |
215 | * Once we get tag lookups on the radix tree, this inode flag | |
216 | * can go away. | |
217 | */ | |
218 | void | |
219 | xfs_inode_set_reclaim_tag( | |
220 | struct xfs_inode *ip) | |
221 | { | |
222 | struct xfs_mount *mp = ip->i_mount; | |
223 | struct xfs_perag *pag; | |
224 | ||
225 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); | |
226 | spin_lock(&pag->pag_ici_lock); | |
227 | spin_lock(&ip->i_flags_lock); | |
228 | ||
229 | radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
230 | XFS_ICI_RECLAIM_TAG); | |
231 | xfs_perag_set_reclaim_tag(pag); | |
232 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
233 | ||
234 | spin_unlock(&ip->i_flags_lock); | |
235 | spin_unlock(&pag->pag_ici_lock); | |
236 | xfs_perag_put(pag); | |
237 | } | |
238 | ||
239 | STATIC void | |
240 | xfs_inode_clear_reclaim_tag( | |
241 | struct xfs_perag *pag, | |
242 | xfs_ino_t ino) | |
243 | { | |
244 | radix_tree_tag_clear(&pag->pag_ici_root, | |
245 | XFS_INO_TO_AGINO(pag->pag_mount, ino), | |
246 | XFS_ICI_RECLAIM_TAG); | |
247 | xfs_perag_clear_reclaim_tag(pag); | |
248 | } | |
249 | ||
ae2c4ac2 BF |
250 | static void |
251 | xfs_inew_wait( | |
252 | struct xfs_inode *ip) | |
253 | { | |
254 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT); | |
255 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT); | |
256 | ||
257 | do { | |
21417136 | 258 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
ae2c4ac2 BF |
259 | if (!xfs_iflags_test(ip, XFS_INEW)) |
260 | break; | |
261 | schedule(); | |
262 | } while (true); | |
21417136 | 263 | finish_wait(wq, &wait.wq_entry); |
ae2c4ac2 BF |
264 | } |
265 | ||
50997470 DC |
266 | /* |
267 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
268 | * part of the structure. This is made more complex by the fact we store | |
269 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 270 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 271 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 272 | * after reinitialisation even if it fails. |
50997470 DC |
273 | */ |
274 | static int | |
275 | xfs_reinit_inode( | |
276 | struct xfs_mount *mp, | |
277 | struct inode *inode) | |
278 | { | |
279 | int error; | |
54d7b5c1 | 280 | uint32_t nlink = inode->i_nlink; |
9e9a2674 | 281 | uint32_t generation = inode->i_generation; |
f0e28280 | 282 | uint64_t version = inode_peek_iversion(inode); |
c19b3b05 | 283 | umode_t mode = inode->i_mode; |
acd1d715 | 284 | dev_t dev = inode->i_rdev; |
3d8f2821 CH |
285 | kuid_t uid = inode->i_uid; |
286 | kgid_t gid = inode->i_gid; | |
50997470 DC |
287 | |
288 | error = inode_init_always(mp->m_super, inode); | |
289 | ||
54d7b5c1 | 290 | set_nlink(inode, nlink); |
9e9a2674 | 291 | inode->i_generation = generation; |
f0e28280 | 292 | inode_set_iversion_queried(inode, version); |
c19b3b05 | 293 | inode->i_mode = mode; |
acd1d715 | 294 | inode->i_rdev = dev; |
3d8f2821 CH |
295 | inode->i_uid = uid; |
296 | inode->i_gid = gid; | |
50997470 DC |
297 | return error; |
298 | } | |
299 | ||
afca6c5b DC |
300 | /* |
301 | * If we are allocating a new inode, then check what was returned is | |
302 | * actually a free, empty inode. If we are not allocating an inode, | |
303 | * then check we didn't find a free inode. | |
304 | * | |
305 | * Returns: | |
306 | * 0 if the inode free state matches the lookup context | |
307 | * -ENOENT if the inode is free and we are not allocating | |
308 | * -EFSCORRUPTED if there is any state mismatch at all | |
309 | */ | |
310 | static int | |
311 | xfs_iget_check_free_state( | |
312 | struct xfs_inode *ip, | |
313 | int flags) | |
314 | { | |
315 | if (flags & XFS_IGET_CREATE) { | |
316 | /* should be a free inode */ | |
317 | if (VFS_I(ip)->i_mode != 0) { | |
318 | xfs_warn(ip->i_mount, | |
319 | "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)", | |
320 | ip->i_ino, VFS_I(ip)->i_mode); | |
321 | return -EFSCORRUPTED; | |
322 | } | |
323 | ||
6e73a545 | 324 | if (ip->i_nblocks != 0) { |
afca6c5b DC |
325 | xfs_warn(ip->i_mount, |
326 | "Corruption detected! Free inode 0x%llx has blocks allocated!", | |
327 | ip->i_ino); | |
328 | return -EFSCORRUPTED; | |
329 | } | |
330 | return 0; | |
331 | } | |
332 | ||
333 | /* should be an allocated inode */ | |
334 | if (VFS_I(ip)->i_mode == 0) | |
335 | return -ENOENT; | |
336 | ||
337 | return 0; | |
338 | } | |
339 | ||
33479e05 DC |
340 | /* |
341 | * Check the validity of the inode we just found it the cache | |
342 | */ | |
343 | static int | |
344 | xfs_iget_cache_hit( | |
345 | struct xfs_perag *pag, | |
346 | struct xfs_inode *ip, | |
347 | xfs_ino_t ino, | |
348 | int flags, | |
349 | int lock_flags) __releases(RCU) | |
350 | { | |
351 | struct inode *inode = VFS_I(ip); | |
352 | struct xfs_mount *mp = ip->i_mount; | |
353 | int error; | |
354 | ||
355 | /* | |
356 | * check for re-use of an inode within an RCU grace period due to the | |
357 | * radix tree nodes not being updated yet. We monitor for this by | |
358 | * setting the inode number to zero before freeing the inode structure. | |
359 | * If the inode has been reallocated and set up, then the inode number | |
360 | * will not match, so check for that, too. | |
361 | */ | |
362 | spin_lock(&ip->i_flags_lock); | |
363 | if (ip->i_ino != ino) { | |
364 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 365 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 366 | error = -EAGAIN; |
33479e05 DC |
367 | goto out_error; |
368 | } | |
369 | ||
370 | ||
371 | /* | |
372 | * If we are racing with another cache hit that is currently | |
373 | * instantiating this inode or currently recycling it out of | |
374 | * reclaimabe state, wait for the initialisation to complete | |
375 | * before continuing. | |
376 | * | |
377 | * XXX(hch): eventually we should do something equivalent to | |
378 | * wait_on_inode to wait for these flags to be cleared | |
379 | * instead of polling for it. | |
380 | */ | |
381 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { | |
382 | trace_xfs_iget_skip(ip); | |
ff6d6af2 | 383 | XFS_STATS_INC(mp, xs_ig_frecycle); |
2451337d | 384 | error = -EAGAIN; |
33479e05 DC |
385 | goto out_error; |
386 | } | |
387 | ||
388 | /* | |
afca6c5b DC |
389 | * Check the inode free state is valid. This also detects lookup |
390 | * racing with unlinks. | |
33479e05 | 391 | */ |
afca6c5b DC |
392 | error = xfs_iget_check_free_state(ip, flags); |
393 | if (error) | |
33479e05 | 394 | goto out_error; |
33479e05 DC |
395 | |
396 | /* | |
397 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
398 | * Need to carefully get it back into useable state. | |
399 | */ | |
400 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
401 | trace_xfs_iget_reclaim(ip); | |
402 | ||
378f681c DW |
403 | if (flags & XFS_IGET_INCORE) { |
404 | error = -EAGAIN; | |
405 | goto out_error; | |
406 | } | |
407 | ||
33479e05 DC |
408 | /* |
409 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode | |
410 | * from stomping over us while we recycle the inode. We can't | |
411 | * clear the radix tree reclaimable tag yet as it requires | |
412 | * pag_ici_lock to be held exclusive. | |
413 | */ | |
414 | ip->i_flags |= XFS_IRECLAIM; | |
415 | ||
416 | spin_unlock(&ip->i_flags_lock); | |
417 | rcu_read_unlock(); | |
418 | ||
d45344d6 | 419 | ASSERT(!rwsem_is_locked(&inode->i_rwsem)); |
50997470 | 420 | error = xfs_reinit_inode(mp, inode); |
33479e05 | 421 | if (error) { |
756baca2 | 422 | bool wake; |
33479e05 DC |
423 | /* |
424 | * Re-initializing the inode failed, and we are in deep | |
425 | * trouble. Try to re-add it to the reclaim list. | |
426 | */ | |
427 | rcu_read_lock(); | |
428 | spin_lock(&ip->i_flags_lock); | |
756baca2 | 429 | wake = !!__xfs_iflags_test(ip, XFS_INEW); |
33479e05 | 430 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
756baca2 BF |
431 | if (wake) |
432 | wake_up_bit(&ip->i_flags, __XFS_INEW_BIT); | |
33479e05 DC |
433 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); |
434 | trace_xfs_iget_reclaim_fail(ip); | |
435 | goto out_error; | |
436 | } | |
437 | ||
438 | spin_lock(&pag->pag_ici_lock); | |
439 | spin_lock(&ip->i_flags_lock); | |
440 | ||
441 | /* | |
442 | * Clear the per-lifetime state in the inode as we are now | |
443 | * effectively a new inode and need to return to the initial | |
444 | * state before reuse occurs. | |
445 | */ | |
446 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
447 | ip->i_flags |= XFS_INEW; | |
545c0889 | 448 | xfs_inode_clear_reclaim_tag(pag, ip->i_ino); |
33479e05 | 449 | inode->i_state = I_NEW; |
6772c1f1 DW |
450 | ip->i_sick = 0; |
451 | ip->i_checked = 0; | |
33479e05 | 452 | |
33479e05 DC |
453 | spin_unlock(&ip->i_flags_lock); |
454 | spin_unlock(&pag->pag_ici_lock); | |
455 | } else { | |
456 | /* If the VFS inode is being torn down, pause and try again. */ | |
457 | if (!igrab(inode)) { | |
458 | trace_xfs_iget_skip(ip); | |
2451337d | 459 | error = -EAGAIN; |
33479e05 DC |
460 | goto out_error; |
461 | } | |
462 | ||
463 | /* We've got a live one. */ | |
464 | spin_unlock(&ip->i_flags_lock); | |
465 | rcu_read_unlock(); | |
466 | trace_xfs_iget_hit(ip); | |
467 | } | |
468 | ||
469 | if (lock_flags != 0) | |
470 | xfs_ilock(ip, lock_flags); | |
471 | ||
378f681c | 472 | if (!(flags & XFS_IGET_INCORE)) |
dae2f8ed | 473 | xfs_iflags_clear(ip, XFS_ISTALE); |
ff6d6af2 | 474 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
475 | |
476 | return 0; | |
477 | ||
478 | out_error: | |
479 | spin_unlock(&ip->i_flags_lock); | |
480 | rcu_read_unlock(); | |
481 | return error; | |
482 | } | |
483 | ||
484 | ||
485 | static int | |
486 | xfs_iget_cache_miss( | |
487 | struct xfs_mount *mp, | |
488 | struct xfs_perag *pag, | |
489 | xfs_trans_t *tp, | |
490 | xfs_ino_t ino, | |
491 | struct xfs_inode **ipp, | |
492 | int flags, | |
493 | int lock_flags) | |
494 | { | |
495 | struct xfs_inode *ip; | |
496 | int error; | |
497 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
498 | int iflags; | |
499 | ||
500 | ip = xfs_inode_alloc(mp, ino); | |
501 | if (!ip) | |
2451337d | 502 | return -ENOMEM; |
33479e05 | 503 | |
bb8a66af | 504 | error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, flags); |
33479e05 DC |
505 | if (error) |
506 | goto out_destroy; | |
507 | ||
bb8a66af CH |
508 | /* |
509 | * For version 5 superblocks, if we are initialising a new inode and we | |
510 | * are not utilising the XFS_MOUNT_IKEEP inode cluster mode, we can | |
511 | * simply build the new inode core with a random generation number. | |
512 | * | |
513 | * For version 4 (and older) superblocks, log recovery is dependent on | |
965e0a1a | 514 | * the i_flushiter field being initialised from the current on-disk |
bb8a66af CH |
515 | * value and hence we must also read the inode off disk even when |
516 | * initializing new inodes. | |
517 | */ | |
518 | if (xfs_sb_version_has_v3inode(&mp->m_sb) && | |
519 | (flags & XFS_IGET_CREATE) && !(mp->m_flags & XFS_MOUNT_IKEEP)) { | |
520 | VFS_I(ip)->i_generation = prandom_u32(); | |
521 | } else { | |
bb8a66af CH |
522 | struct xfs_buf *bp; |
523 | ||
af9dcdde | 524 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp); |
bb8a66af CH |
525 | if (error) |
526 | goto out_destroy; | |
527 | ||
af9dcdde CH |
528 | error = xfs_inode_from_disk(ip, |
529 | xfs_buf_offset(bp, ip->i_imap.im_boffset)); | |
bb8a66af CH |
530 | if (!error) |
531 | xfs_buf_set_ref(bp, XFS_INO_REF); | |
532 | xfs_trans_brelse(tp, bp); | |
533 | ||
534 | if (error) | |
535 | goto out_destroy; | |
536 | } | |
537 | ||
33479e05 DC |
538 | trace_xfs_iget_miss(ip); |
539 | ||
ee457001 | 540 | /* |
afca6c5b DC |
541 | * Check the inode free state is valid. This also detects lookup |
542 | * racing with unlinks. | |
ee457001 | 543 | */ |
afca6c5b DC |
544 | error = xfs_iget_check_free_state(ip, flags); |
545 | if (error) | |
33479e05 | 546 | goto out_destroy; |
33479e05 DC |
547 | |
548 | /* | |
549 | * Preload the radix tree so we can insert safely under the | |
550 | * write spinlock. Note that we cannot sleep inside the preload | |
551 | * region. Since we can be called from transaction context, don't | |
552 | * recurse into the file system. | |
553 | */ | |
554 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 555 | error = -EAGAIN; |
33479e05 DC |
556 | goto out_destroy; |
557 | } | |
558 | ||
559 | /* | |
560 | * Because the inode hasn't been added to the radix-tree yet it can't | |
561 | * be found by another thread, so we can do the non-sleeping lock here. | |
562 | */ | |
563 | if (lock_flags) { | |
564 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
565 | BUG(); | |
566 | } | |
567 | ||
568 | /* | |
569 | * These values must be set before inserting the inode into the radix | |
570 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
571 | * RCU locking mechanism) can find it and that lookup must see that this | |
572 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
573 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
574 | * memory barrier that ensures this detection works correctly at lookup | |
575 | * time. | |
576 | */ | |
577 | iflags = XFS_INEW; | |
578 | if (flags & XFS_IGET_DONTCACHE) | |
2c567af4 | 579 | d_mark_dontcache(VFS_I(ip)); |
113a5683 CS |
580 | ip->i_udquot = NULL; |
581 | ip->i_gdquot = NULL; | |
92f8ff73 | 582 | ip->i_pdquot = NULL; |
33479e05 DC |
583 | xfs_iflags_set(ip, iflags); |
584 | ||
585 | /* insert the new inode */ | |
586 | spin_lock(&pag->pag_ici_lock); | |
587 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
588 | if (unlikely(error)) { | |
589 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 590 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 591 | error = -EAGAIN; |
33479e05 DC |
592 | goto out_preload_end; |
593 | } | |
594 | spin_unlock(&pag->pag_ici_lock); | |
595 | radix_tree_preload_end(); | |
596 | ||
597 | *ipp = ip; | |
598 | return 0; | |
599 | ||
600 | out_preload_end: | |
601 | spin_unlock(&pag->pag_ici_lock); | |
602 | radix_tree_preload_end(); | |
603 | if (lock_flags) | |
604 | xfs_iunlock(ip, lock_flags); | |
605 | out_destroy: | |
606 | __destroy_inode(VFS_I(ip)); | |
607 | xfs_inode_free(ip); | |
608 | return error; | |
609 | } | |
610 | ||
611 | /* | |
02511a5a DC |
612 | * Look up an inode by number in the given file system. The inode is looked up |
613 | * in the cache held in each AG. If the inode is found in the cache, initialise | |
614 | * the vfs inode if necessary. | |
33479e05 | 615 | * |
02511a5a DC |
616 | * If it is not in core, read it in from the file system's device, add it to the |
617 | * cache and initialise the vfs inode. | |
33479e05 DC |
618 | * |
619 | * The inode is locked according to the value of the lock_flags parameter. | |
02511a5a DC |
620 | * Inode lookup is only done during metadata operations and not as part of the |
621 | * data IO path. Hence we only allow locking of the XFS_ILOCK during lookup. | |
33479e05 DC |
622 | */ |
623 | int | |
624 | xfs_iget( | |
02511a5a DC |
625 | struct xfs_mount *mp, |
626 | struct xfs_trans *tp, | |
627 | xfs_ino_t ino, | |
628 | uint flags, | |
629 | uint lock_flags, | |
630 | struct xfs_inode **ipp) | |
33479e05 | 631 | { |
02511a5a DC |
632 | struct xfs_inode *ip; |
633 | struct xfs_perag *pag; | |
634 | xfs_agino_t agino; | |
635 | int error; | |
33479e05 | 636 | |
33479e05 DC |
637 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); |
638 | ||
639 | /* reject inode numbers outside existing AGs */ | |
640 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 641 | return -EINVAL; |
33479e05 | 642 | |
ff6d6af2 | 643 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 644 | |
33479e05 DC |
645 | /* get the perag structure and ensure that it's inode capable */ |
646 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
647 | agino = XFS_INO_TO_AGINO(mp, ino); | |
648 | ||
649 | again: | |
650 | error = 0; | |
651 | rcu_read_lock(); | |
652 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
653 | ||
654 | if (ip) { | |
655 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
656 | if (error) | |
657 | goto out_error_or_again; | |
658 | } else { | |
659 | rcu_read_unlock(); | |
378f681c | 660 | if (flags & XFS_IGET_INCORE) { |
ed438b47 | 661 | error = -ENODATA; |
378f681c DW |
662 | goto out_error_or_again; |
663 | } | |
ff6d6af2 | 664 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
665 | |
666 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
667 | flags, lock_flags); | |
668 | if (error) | |
669 | goto out_error_or_again; | |
670 | } | |
671 | xfs_perag_put(pag); | |
672 | ||
673 | *ipp = ip; | |
674 | ||
675 | /* | |
58c90473 | 676 | * If we have a real type for an on-disk inode, we can setup the inode |
33479e05 DC |
677 | * now. If it's a new inode being created, xfs_ialloc will handle it. |
678 | */ | |
c19b3b05 | 679 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 680 | xfs_setup_existing_inode(ip); |
33479e05 DC |
681 | return 0; |
682 | ||
683 | out_error_or_again: | |
378f681c | 684 | if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) { |
33479e05 DC |
685 | delay(1); |
686 | goto again; | |
687 | } | |
688 | xfs_perag_put(pag); | |
689 | return error; | |
690 | } | |
691 | ||
378f681c DW |
692 | /* |
693 | * "Is this a cached inode that's also allocated?" | |
694 | * | |
695 | * Look up an inode by number in the given file system. If the inode is | |
696 | * in cache and isn't in purgatory, return 1 if the inode is allocated | |
697 | * and 0 if it is not. For all other cases (not in cache, being torn | |
698 | * down, etc.), return a negative error code. | |
699 | * | |
700 | * The caller has to prevent inode allocation and freeing activity, | |
701 | * presumably by locking the AGI buffer. This is to ensure that an | |
702 | * inode cannot transition from allocated to freed until the caller is | |
703 | * ready to allow that. If the inode is in an intermediate state (new, | |
704 | * reclaimable, or being reclaimed), -EAGAIN will be returned; if the | |
705 | * inode is not in the cache, -ENOENT will be returned. The caller must | |
706 | * deal with these scenarios appropriately. | |
707 | * | |
708 | * This is a specialized use case for the online scrubber; if you're | |
709 | * reading this, you probably want xfs_iget. | |
710 | */ | |
711 | int | |
712 | xfs_icache_inode_is_allocated( | |
713 | struct xfs_mount *mp, | |
714 | struct xfs_trans *tp, | |
715 | xfs_ino_t ino, | |
716 | bool *inuse) | |
717 | { | |
718 | struct xfs_inode *ip; | |
719 | int error; | |
720 | ||
721 | error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip); | |
722 | if (error) | |
723 | return error; | |
724 | ||
725 | *inuse = !!(VFS_I(ip)->i_mode); | |
44a8736b | 726 | xfs_irele(ip); |
378f681c DW |
727 | return 0; |
728 | } | |
729 | ||
78ae5256 DC |
730 | /* |
731 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
732 | * radix tree lookups to a minimum. The batch size is a trade off between | |
733 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
734 | * be too greedy. | |
735 | */ | |
736 | #define XFS_LOOKUP_BATCH 32 | |
737 | ||
39b1cfd7 DW |
738 | /* |
739 | * Decide if the given @ip is eligible to be a part of the inode walk, and | |
740 | * grab it if so. Returns true if it's ready to go or false if we should just | |
741 | * ignore it. | |
742 | */ | |
743 | STATIC bool | |
042f65f4 | 744 | xfs_inode_walk_ag_grab( |
ae2c4ac2 BF |
745 | struct xfs_inode *ip, |
746 | int flags) | |
e13de955 DC |
747 | { |
748 | struct inode *inode = VFS_I(ip); | |
042f65f4 | 749 | bool newinos = !!(flags & XFS_INODE_WALK_INEW_WAIT); |
e13de955 | 750 | |
1a3e8f3d DC |
751 | ASSERT(rcu_read_lock_held()); |
752 | ||
02511a5a | 753 | /* Check for stale RCU freed inode */ |
1a3e8f3d DC |
754 | spin_lock(&ip->i_flags_lock); |
755 | if (!ip->i_ino) | |
756 | goto out_unlock_noent; | |
757 | ||
758 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ | |
ae2c4ac2 BF |
759 | if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) || |
760 | __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM)) | |
1a3e8f3d DC |
761 | goto out_unlock_noent; |
762 | spin_unlock(&ip->i_flags_lock); | |
763 | ||
e13de955 DC |
764 | /* nothing to sync during shutdown */ |
765 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
39b1cfd7 | 766 | return false; |
e13de955 | 767 | |
e13de955 DC |
768 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
769 | if (!igrab(inode)) | |
39b1cfd7 | 770 | return false; |
e13de955 | 771 | |
e13de955 | 772 | /* inode is valid */ |
39b1cfd7 | 773 | return true; |
1a3e8f3d DC |
774 | |
775 | out_unlock_noent: | |
776 | spin_unlock(&ip->i_flags_lock); | |
39b1cfd7 | 777 | return false; |
e13de955 DC |
778 | } |
779 | ||
5662d38c DW |
780 | /* |
781 | * For a given per-AG structure @pag, grab, @execute, and rele all incore | |
782 | * inodes with the given radix tree @tag. | |
783 | */ | |
75f3cb13 | 784 | STATIC int |
042f65f4 | 785 | xfs_inode_walk_ag( |
5017e97d | 786 | struct xfs_perag *pag, |
964176bd | 787 | int iter_flags, |
390600f8 | 788 | int (*execute)(struct xfs_inode *ip, void *args), |
a454f742 | 789 | void *args, |
964176bd | 790 | int tag) |
75f3cb13 | 791 | { |
964176bd | 792 | struct xfs_mount *mp = pag->pag_mount; |
75f3cb13 DC |
793 | uint32_t first_index; |
794 | int last_error = 0; | |
795 | int skipped; | |
7e88d314 | 796 | bool done; |
78ae5256 | 797 | int nr_found; |
75f3cb13 DC |
798 | |
799 | restart: | |
7e88d314 | 800 | done = false; |
75f3cb13 DC |
801 | skipped = 0; |
802 | first_index = 0; | |
78ae5256 | 803 | nr_found = 0; |
75f3cb13 | 804 | do { |
78ae5256 | 805 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
75f3cb13 | 806 | int error = 0; |
78ae5256 | 807 | int i; |
75f3cb13 | 808 | |
1a3e8f3d | 809 | rcu_read_lock(); |
a454f742 | 810 | |
fc96be95 | 811 | if (tag == XFS_ICI_NO_TAG) |
a454f742 | 812 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, |
78ae5256 DC |
813 | (void **)batch, first_index, |
814 | XFS_LOOKUP_BATCH); | |
a454f742 BF |
815 | else |
816 | nr_found = radix_tree_gang_lookup_tag( | |
817 | &pag->pag_ici_root, | |
818 | (void **) batch, first_index, | |
819 | XFS_LOOKUP_BATCH, tag); | |
820 | ||
65d0f205 | 821 | if (!nr_found) { |
1a3e8f3d | 822 | rcu_read_unlock(); |
75f3cb13 | 823 | break; |
c8e20be0 | 824 | } |
75f3cb13 | 825 | |
65d0f205 | 826 | /* |
78ae5256 DC |
827 | * Grab the inodes before we drop the lock. if we found |
828 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 829 | */ |
78ae5256 DC |
830 | for (i = 0; i < nr_found; i++) { |
831 | struct xfs_inode *ip = batch[i]; | |
832 | ||
042f65f4 | 833 | if (done || !xfs_inode_walk_ag_grab(ip, iter_flags)) |
78ae5256 DC |
834 | batch[i] = NULL; |
835 | ||
836 | /* | |
1a3e8f3d DC |
837 | * Update the index for the next lookup. Catch |
838 | * overflows into the next AG range which can occur if | |
839 | * we have inodes in the last block of the AG and we | |
840 | * are currently pointing to the last inode. | |
841 | * | |
842 | * Because we may see inodes that are from the wrong AG | |
843 | * due to RCU freeing and reallocation, only update the | |
844 | * index if it lies in this AG. It was a race that lead | |
845 | * us to see this inode, so another lookup from the | |
846 | * same index will not find it again. | |
78ae5256 | 847 | */ |
1a3e8f3d DC |
848 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
849 | continue; | |
78ae5256 DC |
850 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
851 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
7e88d314 | 852 | done = true; |
e13de955 | 853 | } |
78ae5256 DC |
854 | |
855 | /* unlock now we've grabbed the inodes. */ | |
1a3e8f3d | 856 | rcu_read_unlock(); |
e13de955 | 857 | |
78ae5256 DC |
858 | for (i = 0; i < nr_found; i++) { |
859 | if (!batch[i]) | |
860 | continue; | |
042f65f4 | 861 | if ((iter_flags & XFS_INODE_WALK_INEW_WAIT) && |
ae2c4ac2 BF |
862 | xfs_iflags_test(batch[i], XFS_INEW)) |
863 | xfs_inew_wait(batch[i]); | |
390600f8 | 864 | error = execute(batch[i], args); |
44a8736b | 865 | xfs_irele(batch[i]); |
2451337d | 866 | if (error == -EAGAIN) { |
78ae5256 DC |
867 | skipped++; |
868 | continue; | |
869 | } | |
2451337d | 870 | if (error && last_error != -EFSCORRUPTED) |
78ae5256 | 871 | last_error = error; |
75f3cb13 | 872 | } |
c8e20be0 DC |
873 | |
874 | /* bail out if the filesystem is corrupted. */ | |
2451337d | 875 | if (error == -EFSCORRUPTED) |
75f3cb13 DC |
876 | break; |
877 | ||
8daaa831 DC |
878 | cond_resched(); |
879 | ||
78ae5256 | 880 | } while (nr_found && !done); |
75f3cb13 DC |
881 | |
882 | if (skipped) { | |
883 | delay(1); | |
884 | goto restart; | |
885 | } | |
75f3cb13 DC |
886 | return last_error; |
887 | } | |
888 | ||
5662d38c DW |
889 | /* Fetch the next (possibly tagged) per-AG structure. */ |
890 | static inline struct xfs_perag * | |
891 | xfs_inode_walk_get_perag( | |
892 | struct xfs_mount *mp, | |
893 | xfs_agnumber_t agno, | |
894 | int tag) | |
895 | { | |
896 | if (tag == XFS_ICI_NO_TAG) | |
897 | return xfs_perag_get(mp, agno); | |
898 | return xfs_perag_get_tag(mp, agno, tag); | |
899 | } | |
900 | ||
901 | /* | |
902 | * Call the @execute function on all incore inodes matching the radix tree | |
903 | * @tag. | |
904 | */ | |
1ad2cfe0 | 905 | static int |
042f65f4 | 906 | xfs_inode_walk( |
5662d38c DW |
907 | struct xfs_mount *mp, |
908 | int iter_flags, | |
909 | int (*execute)(struct xfs_inode *ip, void *args), | |
910 | void *args, | |
911 | int tag) | |
912 | { | |
913 | struct xfs_perag *pag; | |
914 | int error = 0; | |
915 | int last_error = 0; | |
916 | xfs_agnumber_t ag; | |
917 | ||
918 | ag = 0; | |
919 | while ((pag = xfs_inode_walk_get_perag(mp, ag, tag))) { | |
920 | ag = pag->pag_agno + 1; | |
964176bd | 921 | error = xfs_inode_walk_ag(pag, iter_flags, execute, args, tag); |
5662d38c DW |
922 | xfs_perag_put(pag); |
923 | if (error) { | |
924 | last_error = error; | |
925 | if (error == -EFSCORRUPTED) | |
926 | break; | |
927 | } | |
928 | } | |
929 | return last_error; | |
1ad2cfe0 DW |
930 | BUILD_BUG_ON(XFS_ICWALK_PRIVATE_FLAGS & XFS_EOF_FLAGS_VALID); |
931 | } | |
932 | ||
933 | #ifdef CONFIG_XFS_QUOTA | |
934 | /* Drop this inode's dquots. */ | |
935 | static int | |
936 | xfs_dqrele_inode( | |
937 | struct xfs_inode *ip, | |
938 | void *priv) | |
939 | { | |
940 | struct xfs_eofblocks *eofb = priv; | |
941 | ||
942 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
943 | if (eofb->eof_flags & XFS_ICWALK_FLAG_DROP_UDQUOT) { | |
944 | xfs_qm_dqrele(ip->i_udquot); | |
945 | ip->i_udquot = NULL; | |
946 | } | |
947 | if (eofb->eof_flags & XFS_ICWALK_FLAG_DROP_GDQUOT) { | |
948 | xfs_qm_dqrele(ip->i_gdquot); | |
949 | ip->i_gdquot = NULL; | |
950 | } | |
951 | if (eofb->eof_flags & XFS_ICWALK_FLAG_DROP_PDQUOT) { | |
952 | xfs_qm_dqrele(ip->i_pdquot); | |
953 | ip->i_pdquot = NULL; | |
954 | } | |
955 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
956 | return 0; | |
957 | } | |
958 | ||
959 | /* | |
960 | * Detach all dquots from incore inodes if we can. The caller must already | |
961 | * have dropped the relevant XFS_[UGP]QUOTA_ACTIVE flags so that dquots will | |
962 | * not get reattached. | |
963 | */ | |
964 | int | |
965 | xfs_dqrele_all_inodes( | |
966 | struct xfs_mount *mp, | |
967 | unsigned int qflags) | |
968 | { | |
969 | struct xfs_eofblocks eofb = { .eof_flags = 0 }; | |
970 | ||
971 | if (qflags & XFS_UQUOTA_ACCT) | |
972 | eofb.eof_flags |= XFS_ICWALK_FLAG_DROP_UDQUOT; | |
973 | if (qflags & XFS_GQUOTA_ACCT) | |
974 | eofb.eof_flags |= XFS_ICWALK_FLAG_DROP_GDQUOT; | |
975 | if (qflags & XFS_PQUOTA_ACCT) | |
976 | eofb.eof_flags |= XFS_ICWALK_FLAG_DROP_PDQUOT; | |
977 | ||
978 | return xfs_inode_walk(mp, XFS_INODE_WALK_INEW_WAIT, xfs_dqrele_inode, | |
979 | &eofb, XFS_ICI_NO_TAG); | |
5662d38c | 980 | } |
1ad2cfe0 | 981 | #endif /* CONFIG_XFS_QUOTA */ |
5662d38c | 982 | |
e3a20c0b DC |
983 | /* |
984 | * Grab the inode for reclaim exclusively. | |
50718b8d DC |
985 | * |
986 | * We have found this inode via a lookup under RCU, so the inode may have | |
987 | * already been freed, or it may be in the process of being recycled by | |
988 | * xfs_iget(). In both cases, the inode will have XFS_IRECLAIM set. If the inode | |
989 | * has been fully recycled by the time we get the i_flags_lock, XFS_IRECLAIMABLE | |
990 | * will not be set. Hence we need to check for both these flag conditions to | |
991 | * avoid inodes that are no longer reclaim candidates. | |
992 | * | |
993 | * Note: checking for other state flags here, under the i_flags_lock or not, is | |
994 | * racy and should be avoided. Those races should be resolved only after we have | |
995 | * ensured that we are able to reclaim this inode and the world can see that we | |
996 | * are going to reclaim it. | |
997 | * | |
998 | * Return true if we grabbed it, false otherwise. | |
e3a20c0b | 999 | */ |
50718b8d | 1000 | static bool |
e3a20c0b | 1001 | xfs_reclaim_inode_grab( |
50718b8d | 1002 | struct xfs_inode *ip) |
e3a20c0b | 1003 | { |
1a3e8f3d DC |
1004 | ASSERT(rcu_read_lock_held()); |
1005 | ||
e3a20c0b | 1006 | spin_lock(&ip->i_flags_lock); |
1a3e8f3d DC |
1007 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
1008 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
1009 | /* not a reclaim candidate. */ | |
e3a20c0b | 1010 | spin_unlock(&ip->i_flags_lock); |
50718b8d | 1011 | return false; |
e3a20c0b DC |
1012 | } |
1013 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
1014 | spin_unlock(&ip->i_flags_lock); | |
50718b8d | 1015 | return true; |
e3a20c0b DC |
1016 | } |
1017 | ||
777df5af | 1018 | /* |
02511a5a DC |
1019 | * Inode reclaim is non-blocking, so the default action if progress cannot be |
1020 | * made is to "requeue" the inode for reclaim by unlocking it and clearing the | |
1021 | * XFS_IRECLAIM flag. If we are in a shutdown state, we don't care about | |
1022 | * blocking anymore and hence we can wait for the inode to be able to reclaim | |
1023 | * it. | |
777df5af | 1024 | * |
02511a5a DC |
1025 | * We do no IO here - if callers require inodes to be cleaned they must push the |
1026 | * AIL first to trigger writeback of dirty inodes. This enables writeback to be | |
1027 | * done in the background in a non-blocking manner, and enables memory reclaim | |
1028 | * to make progress without blocking. | |
777df5af | 1029 | */ |
4d0bab3a | 1030 | static void |
c8e20be0 | 1031 | xfs_reclaim_inode( |
75f3cb13 | 1032 | struct xfs_inode *ip, |
50718b8d | 1033 | struct xfs_perag *pag) |
fce08f2f | 1034 | { |
8a17d7dd | 1035 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
777df5af | 1036 | |
9552e14d | 1037 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) |
617825fe | 1038 | goto out; |
718ecc50 | 1039 | if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING)) |
9552e14d | 1040 | goto out_iunlock; |
7a3be02b | 1041 | |
777df5af DC |
1042 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1043 | xfs_iunpin_wait(ip); | |
88fc1879 | 1044 | xfs_iflush_abort(ip); |
777df5af DC |
1045 | goto reclaim; |
1046 | } | |
617825fe | 1047 | if (xfs_ipincount(ip)) |
718ecc50 | 1048 | goto out_clear_flush; |
617825fe | 1049 | if (!xfs_inode_clean(ip)) |
718ecc50 | 1050 | goto out_clear_flush; |
8a48088f | 1051 | |
718ecc50 | 1052 | xfs_iflags_clear(ip, XFS_IFLUSHING); |
777df5af | 1053 | reclaim: |
98efe8af | 1054 | |
8a17d7dd DC |
1055 | /* |
1056 | * Because we use RCU freeing we need to ensure the inode always appears | |
1057 | * to be reclaimed with an invalid inode number when in the free state. | |
98efe8af | 1058 | * We do this as early as possible under the ILOCK so that |
f2e9ad21 OS |
1059 | * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to |
1060 | * detect races with us here. By doing this, we guarantee that once | |
1061 | * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that | |
1062 | * it will see either a valid inode that will serialise correctly, or it | |
1063 | * will see an invalid inode that it can skip. | |
8a17d7dd DC |
1064 | */ |
1065 | spin_lock(&ip->i_flags_lock); | |
1066 | ip->i_flags = XFS_IRECLAIM; | |
1067 | ip->i_ino = 0; | |
1068 | spin_unlock(&ip->i_flags_lock); | |
1069 | ||
c8e20be0 | 1070 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 1071 | |
ff6d6af2 | 1072 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
1073 | /* |
1074 | * Remove the inode from the per-AG radix tree. | |
1075 | * | |
1076 | * Because radix_tree_delete won't complain even if the item was never | |
1077 | * added to the tree assert that it's been there before to catch | |
1078 | * problems with the inode life time early on. | |
1079 | */ | |
1a427ab0 | 1080 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 1081 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 1082 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 1083 | ASSERT(0); |
545c0889 | 1084 | xfs_perag_clear_reclaim_tag(pag); |
1a427ab0 | 1085 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
1086 | |
1087 | /* | |
1088 | * Here we do an (almost) spurious inode lock in order to coordinate | |
1089 | * with inode cache radix tree lookups. This is because the lookup | |
1090 | * can reference the inodes in the cache without taking references. | |
1091 | * | |
1092 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 1093 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 1094 | */ |
ad637a10 | 1095 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
3ea06d73 | 1096 | ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot); |
ad637a10 | 1097 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
96355d5a | 1098 | ASSERT(xfs_inode_clean(ip)); |
2f11feab | 1099 | |
8a17d7dd | 1100 | __xfs_inode_free(ip); |
4d0bab3a | 1101 | return; |
8a48088f | 1102 | |
718ecc50 DC |
1103 | out_clear_flush: |
1104 | xfs_iflags_clear(ip, XFS_IFLUSHING); | |
9552e14d | 1105 | out_iunlock: |
8a48088f | 1106 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
9552e14d | 1107 | out: |
617825fe | 1108 | xfs_iflags_clear(ip, XFS_IRECLAIM); |
7a3be02b DC |
1109 | } |
1110 | ||
65d0f205 DC |
1111 | /* |
1112 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is | |
1113 | * corrupted, we still want to try to reclaim all the inodes. If we don't, | |
1114 | * then a shut down during filesystem unmount reclaim walk leak all the | |
1115 | * unreclaimed inodes. | |
617825fe DC |
1116 | * |
1117 | * Returns non-zero if any AGs or inodes were skipped in the reclaim pass | |
1118 | * so that callers that want to block until all dirty inodes are written back | |
1119 | * and reclaimed can sanely loop. | |
65d0f205 | 1120 | */ |
4d0bab3a | 1121 | static void |
65d0f205 DC |
1122 | xfs_reclaim_inodes_ag( |
1123 | struct xfs_mount *mp, | |
65d0f205 DC |
1124 | int *nr_to_scan) |
1125 | { | |
1126 | struct xfs_perag *pag; | |
0e8e2c63 | 1127 | xfs_agnumber_t ag = 0; |
65d0f205 | 1128 | |
65d0f205 DC |
1129 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1130 | unsigned long first_index = 0; | |
1131 | int done = 0; | |
e3a20c0b | 1132 | int nr_found = 0; |
65d0f205 DC |
1133 | |
1134 | ag = pag->pag_agno + 1; | |
1135 | ||
0e8e2c63 | 1136 | first_index = READ_ONCE(pag->pag_ici_reclaim_cursor); |
65d0f205 | 1137 | do { |
e3a20c0b DC |
1138 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
1139 | int i; | |
65d0f205 | 1140 | |
1a3e8f3d | 1141 | rcu_read_lock(); |
e3a20c0b DC |
1142 | nr_found = radix_tree_gang_lookup_tag( |
1143 | &pag->pag_ici_root, | |
1144 | (void **)batch, first_index, | |
1145 | XFS_LOOKUP_BATCH, | |
65d0f205 DC |
1146 | XFS_ICI_RECLAIM_TAG); |
1147 | if (!nr_found) { | |
b2232219 | 1148 | done = 1; |
1a3e8f3d | 1149 | rcu_read_unlock(); |
65d0f205 DC |
1150 | break; |
1151 | } | |
1152 | ||
1153 | /* | |
e3a20c0b DC |
1154 | * Grab the inodes before we drop the lock. if we found |
1155 | * nothing, nr == 0 and the loop will be skipped. | |
65d0f205 | 1156 | */ |
e3a20c0b DC |
1157 | for (i = 0; i < nr_found; i++) { |
1158 | struct xfs_inode *ip = batch[i]; | |
1159 | ||
50718b8d | 1160 | if (done || !xfs_reclaim_inode_grab(ip)) |
e3a20c0b DC |
1161 | batch[i] = NULL; |
1162 | ||
1163 | /* | |
1164 | * Update the index for the next lookup. Catch | |
1165 | * overflows into the next AG range which can | |
1166 | * occur if we have inodes in the last block of | |
1167 | * the AG and we are currently pointing to the | |
1168 | * last inode. | |
1a3e8f3d DC |
1169 | * |
1170 | * Because we may see inodes that are from the | |
1171 | * wrong AG due to RCU freeing and | |
1172 | * reallocation, only update the index if it | |
1173 | * lies in this AG. It was a race that lead us | |
1174 | * to see this inode, so another lookup from | |
1175 | * the same index will not find it again. | |
e3a20c0b | 1176 | */ |
1a3e8f3d DC |
1177 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
1178 | pag->pag_agno) | |
1179 | continue; | |
e3a20c0b DC |
1180 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
1181 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1182 | done = 1; | |
1183 | } | |
65d0f205 | 1184 | |
e3a20c0b | 1185 | /* unlock now we've grabbed the inodes. */ |
1a3e8f3d | 1186 | rcu_read_unlock(); |
e3a20c0b DC |
1187 | |
1188 | for (i = 0; i < nr_found; i++) { | |
4d0bab3a DC |
1189 | if (batch[i]) |
1190 | xfs_reclaim_inode(batch[i], pag); | |
e3a20c0b DC |
1191 | } |
1192 | ||
1193 | *nr_to_scan -= XFS_LOOKUP_BATCH; | |
8daaa831 | 1194 | cond_resched(); |
e3a20c0b | 1195 | } while (nr_found && !done && *nr_to_scan > 0); |
65d0f205 | 1196 | |
0e8e2c63 DC |
1197 | if (done) |
1198 | first_index = 0; | |
1199 | WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index); | |
65d0f205 DC |
1200 | xfs_perag_put(pag); |
1201 | } | |
65d0f205 DC |
1202 | } |
1203 | ||
4d0bab3a | 1204 | void |
7a3be02b | 1205 | xfs_reclaim_inodes( |
4d0bab3a | 1206 | struct xfs_mount *mp) |
7a3be02b | 1207 | { |
65d0f205 DC |
1208 | int nr_to_scan = INT_MAX; |
1209 | ||
4d0bab3a | 1210 | while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { |
617825fe | 1211 | xfs_ail_push_all_sync(mp->m_ail); |
4d0bab3a | 1212 | xfs_reclaim_inodes_ag(mp, &nr_to_scan); |
0f4ec0f1 | 1213 | } |
9bf729c0 DC |
1214 | } |
1215 | ||
1216 | /* | |
02511a5a DC |
1217 | * The shrinker infrastructure determines how many inodes we should scan for |
1218 | * reclaim. We want as many clean inodes ready to reclaim as possible, so we | |
1219 | * push the AIL here. We also want to proactively free up memory if we can to | |
1220 | * minimise the amount of work memory reclaim has to do so we kick the | |
1221 | * background reclaim if it isn't already scheduled. | |
9bf729c0 | 1222 | */ |
0a234c6d | 1223 | long |
8daaa831 DC |
1224 | xfs_reclaim_inodes_nr( |
1225 | struct xfs_mount *mp, | |
1226 | int nr_to_scan) | |
9bf729c0 | 1227 | { |
8daaa831 | 1228 | /* kick background reclaimer and push the AIL */ |
5889608d | 1229 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1230 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1231 | |
50718b8d | 1232 | xfs_reclaim_inodes_ag(mp, &nr_to_scan); |
617825fe | 1233 | return 0; |
8daaa831 | 1234 | } |
9bf729c0 | 1235 | |
8daaa831 DC |
1236 | /* |
1237 | * Return the number of reclaimable inodes in the filesystem for | |
1238 | * the shrinker to determine how much to reclaim. | |
1239 | */ | |
1240 | int | |
1241 | xfs_reclaim_inodes_count( | |
1242 | struct xfs_mount *mp) | |
1243 | { | |
1244 | struct xfs_perag *pag; | |
1245 | xfs_agnumber_t ag = 0; | |
1246 | int reclaimable = 0; | |
9bf729c0 | 1247 | |
65d0f205 DC |
1248 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1249 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1250 | reclaimable += pag->pag_ici_reclaimable; |
1251 | xfs_perag_put(pag); | |
9bf729c0 | 1252 | } |
9bf729c0 DC |
1253 | return reclaimable; |
1254 | } | |
1255 | ||
39b1cfd7 | 1256 | STATIC bool |
3e3f9f58 BF |
1257 | xfs_inode_match_id( |
1258 | struct xfs_inode *ip, | |
1259 | struct xfs_eofblocks *eofb) | |
1260 | { | |
b9fe5052 DE |
1261 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && |
1262 | !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
39b1cfd7 | 1263 | return false; |
3e3f9f58 | 1264 | |
b9fe5052 DE |
1265 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && |
1266 | !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
39b1cfd7 | 1267 | return false; |
1b556048 | 1268 | |
b9fe5052 | 1269 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && |
ceaf603c | 1270 | ip->i_projid != eofb->eof_prid) |
39b1cfd7 | 1271 | return false; |
1b556048 | 1272 | |
39b1cfd7 | 1273 | return true; |
3e3f9f58 BF |
1274 | } |
1275 | ||
f4526397 BF |
1276 | /* |
1277 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1278 | * criteria match. This is for global/internal scans only. | |
1279 | */ | |
39b1cfd7 | 1280 | STATIC bool |
f4526397 BF |
1281 | xfs_inode_match_id_union( |
1282 | struct xfs_inode *ip, | |
1283 | struct xfs_eofblocks *eofb) | |
1284 | { | |
1285 | if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) && | |
1286 | uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid)) | |
39b1cfd7 | 1287 | return true; |
f4526397 BF |
1288 | |
1289 | if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) && | |
1290 | gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid)) | |
39b1cfd7 | 1291 | return true; |
f4526397 BF |
1292 | |
1293 | if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) && | |
ceaf603c | 1294 | ip->i_projid == eofb->eof_prid) |
39b1cfd7 | 1295 | return true; |
f4526397 | 1296 | |
39b1cfd7 | 1297 | return false; |
f4526397 BF |
1298 | } |
1299 | ||
a91bf992 DW |
1300 | /* |
1301 | * Is this inode @ip eligible for eof/cow block reclamation, given some | |
1302 | * filtering parameters @eofb? The inode is eligible if @eofb is null or | |
1303 | * if the predicate functions match. | |
1304 | */ | |
1305 | static bool | |
1306 | xfs_inode_matches_eofb( | |
1307 | struct xfs_inode *ip, | |
1308 | struct xfs_eofblocks *eofb) | |
1309 | { | |
39b1cfd7 | 1310 | bool match; |
a91bf992 DW |
1311 | |
1312 | if (!eofb) | |
1313 | return true; | |
1314 | ||
1315 | if (eofb->eof_flags & XFS_EOF_FLAGS_UNION) | |
1316 | match = xfs_inode_match_id_union(ip, eofb); | |
1317 | else | |
1318 | match = xfs_inode_match_id(ip, eofb); | |
1319 | if (!match) | |
1320 | return false; | |
1321 | ||
1322 | /* skip the inode if the file size is too small */ | |
1323 | if ((eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE) && | |
1324 | XFS_ISIZE(ip) < eofb->eof_min_file_size) | |
1325 | return false; | |
1326 | ||
1327 | return true; | |
1328 | } | |
1329 | ||
4d0bab3a DC |
1330 | /* |
1331 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
1332 | * many inodes as possible in a short period of time. It kicks itself every few | |
1333 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
02511a5a | 1334 | * goes low. |
4d0bab3a DC |
1335 | */ |
1336 | void | |
1337 | xfs_reclaim_worker( | |
1338 | struct work_struct *work) | |
1339 | { | |
1340 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
1341 | struct xfs_mount, m_reclaim_work); | |
1342 | int nr_to_scan = INT_MAX; | |
1343 | ||
1344 | xfs_reclaim_inodes_ag(mp, &nr_to_scan); | |
1345 | xfs_reclaim_work_queue(mp); | |
1346 | } | |
1347 | ||
41176a68 BF |
1348 | STATIC int |
1349 | xfs_inode_free_eofblocks( | |
1350 | struct xfs_inode *ip, | |
0fa4a10a DW |
1351 | void *args, |
1352 | unsigned int *lockflags) | |
41176a68 | 1353 | { |
390600f8 DW |
1354 | struct xfs_eofblocks *eofb = args; |
1355 | bool wait; | |
390600f8 DW |
1356 | |
1357 | wait = eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC); | |
5400da7d | 1358 | |
ce2d3bbe DW |
1359 | if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS)) |
1360 | return 0; | |
1361 | ||
41176a68 BF |
1362 | /* |
1363 | * If the mapping is dirty the operation can block and wait for some | |
1364 | * time. Unless we are waiting, skip it. | |
1365 | */ | |
390600f8 | 1366 | if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) |
41176a68 BF |
1367 | return 0; |
1368 | ||
a91bf992 DW |
1369 | if (!xfs_inode_matches_eofb(ip, eofb)) |
1370 | return 0; | |
3e3f9f58 | 1371 | |
a36b9261 BF |
1372 | /* |
1373 | * If the caller is waiting, return -EAGAIN to keep the background | |
1374 | * scanner moving and revisit the inode in a subsequent pass. | |
1375 | */ | |
c3155097 | 1376 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
390600f8 DW |
1377 | if (wait) |
1378 | return -EAGAIN; | |
1379 | return 0; | |
a36b9261 | 1380 | } |
0fa4a10a | 1381 | *lockflags |= XFS_IOLOCK_EXCL; |
390600f8 | 1382 | |
2b156ff8 DW |
1383 | if (xfs_can_free_eofblocks(ip, false)) |
1384 | return xfs_free_eofblocks(ip); | |
1385 | ||
1386 | /* inode could be preallocated or append-only */ | |
1387 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1388 | xfs_inode_clear_eofblocks_tag(ip); | |
1389 | return 0; | |
41176a68 BF |
1390 | } |
1391 | ||
f9296569 | 1392 | /* |
9669f51d DW |
1393 | * Background scanning to trim preallocated space. This is queued based on the |
1394 | * 'speculative_prealloc_lifetime' tunable (5m by default). | |
f9296569 | 1395 | */ |
9669f51d DW |
1396 | static inline void |
1397 | xfs_blockgc_queue( | |
894ecacf | 1398 | struct xfs_perag *pag) |
f9296569 DW |
1399 | { |
1400 | rcu_read_lock(); | |
894ecacf | 1401 | if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG)) |
3fef46fc | 1402 | queue_delayed_work(pag->pag_mount->m_gc_workqueue, |
894ecacf | 1403 | &pag->pag_blockgc_work, |
9669f51d | 1404 | msecs_to_jiffies(xfs_blockgc_secs * 1000)); |
f9296569 DW |
1405 | rcu_read_unlock(); |
1406 | } | |
1407 | ||
83104d44 | 1408 | static void |
ce2d3bbe DW |
1409 | xfs_blockgc_set_iflag( |
1410 | struct xfs_inode *ip, | |
ce2d3bbe | 1411 | unsigned long iflag) |
27b52867 | 1412 | { |
ce2d3bbe DW |
1413 | struct xfs_mount *mp = ip->i_mount; |
1414 | struct xfs_perag *pag; | |
1415 | int tagged; | |
1416 | ||
1417 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1418 | |
85a6e764 CH |
1419 | /* |
1420 | * Don't bother locking the AG and looking up in the radix trees | |
1421 | * if we already know that we have the tag set. | |
1422 | */ | |
ce2d3bbe | 1423 | if (ip->i_flags & iflag) |
85a6e764 CH |
1424 | return; |
1425 | spin_lock(&ip->i_flags_lock); | |
ce2d3bbe | 1426 | ip->i_flags |= iflag; |
85a6e764 CH |
1427 | spin_unlock(&ip->i_flags_lock); |
1428 | ||
27b52867 BF |
1429 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1430 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1431 | |
ce2d3bbe | 1432 | tagged = radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG); |
27b52867 | 1433 | radix_tree_tag_set(&pag->pag_ici_root, |
ce2d3bbe DW |
1434 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), |
1435 | XFS_ICI_BLOCKGC_TAG); | |
27b52867 | 1436 | if (!tagged) { |
ce2d3bbe | 1437 | /* propagate the blockgc tag up into the perag radix tree */ |
27b52867 BF |
1438 | spin_lock(&ip->i_mount->m_perag_lock); |
1439 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, | |
1440 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
ce2d3bbe | 1441 | XFS_ICI_BLOCKGC_TAG); |
27b52867 | 1442 | spin_unlock(&ip->i_mount->m_perag_lock); |
579b62fa BF |
1443 | |
1444 | /* kick off background trimming */ | |
894ecacf | 1445 | xfs_blockgc_queue(pag); |
27b52867 | 1446 | |
ce2d3bbe DW |
1447 | trace_xfs_perag_set_blockgc(ip->i_mount, pag->pag_agno, -1, |
1448 | _RET_IP_); | |
27b52867 BF |
1449 | } |
1450 | ||
1451 | spin_unlock(&pag->pag_ici_lock); | |
1452 | xfs_perag_put(pag); | |
1453 | } | |
1454 | ||
1455 | void | |
83104d44 | 1456 | xfs_inode_set_eofblocks_tag( |
27b52867 | 1457 | xfs_inode_t *ip) |
83104d44 DW |
1458 | { |
1459 | trace_xfs_inode_set_eofblocks_tag(ip); | |
9669f51d | 1460 | return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1461 | } |
1462 | ||
1463 | static void | |
ce2d3bbe DW |
1464 | xfs_blockgc_clear_iflag( |
1465 | struct xfs_inode *ip, | |
1466 | unsigned long iflag) | |
27b52867 | 1467 | { |
ce2d3bbe DW |
1468 | struct xfs_mount *mp = ip->i_mount; |
1469 | struct xfs_perag *pag; | |
1470 | bool clear_tag; | |
1471 | ||
1472 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1473 | |
85a6e764 | 1474 | spin_lock(&ip->i_flags_lock); |
ce2d3bbe DW |
1475 | ip->i_flags &= ~iflag; |
1476 | clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0; | |
85a6e764 CH |
1477 | spin_unlock(&ip->i_flags_lock); |
1478 | ||
ce2d3bbe DW |
1479 | if (!clear_tag) |
1480 | return; | |
1481 | ||
27b52867 BF |
1482 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1483 | spin_lock(&pag->pag_ici_lock); | |
27b52867 BF |
1484 | |
1485 | radix_tree_tag_clear(&pag->pag_ici_root, | |
ce2d3bbe DW |
1486 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), |
1487 | XFS_ICI_BLOCKGC_TAG); | |
1488 | if (!radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG)) { | |
1489 | /* clear the blockgc tag from the perag radix tree */ | |
27b52867 BF |
1490 | spin_lock(&ip->i_mount->m_perag_lock); |
1491 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, | |
1492 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), | |
ce2d3bbe | 1493 | XFS_ICI_BLOCKGC_TAG); |
27b52867 | 1494 | spin_unlock(&ip->i_mount->m_perag_lock); |
ce2d3bbe DW |
1495 | trace_xfs_perag_clear_blockgc(ip->i_mount, pag->pag_agno, -1, |
1496 | _RET_IP_); | |
27b52867 BF |
1497 | } |
1498 | ||
1499 | spin_unlock(&pag->pag_ici_lock); | |
1500 | xfs_perag_put(pag); | |
1501 | } | |
1502 | ||
83104d44 DW |
1503 | void |
1504 | xfs_inode_clear_eofblocks_tag( | |
1505 | xfs_inode_t *ip) | |
1506 | { | |
1507 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
ce2d3bbe | 1508 | return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1509 | } |
1510 | ||
1511 | /* | |
be78ff0e DW |
1512 | * Set ourselves up to free CoW blocks from this file. If it's already clean |
1513 | * then we can bail out quickly, but otherwise we must back off if the file | |
1514 | * is undergoing some kind of write. | |
83104d44 | 1515 | */ |
be78ff0e DW |
1516 | static bool |
1517 | xfs_prep_free_cowblocks( | |
51d62690 | 1518 | struct xfs_inode *ip) |
83104d44 | 1519 | { |
39937234 BF |
1520 | /* |
1521 | * Just clear the tag if we have an empty cow fork or none at all. It's | |
1522 | * possible the inode was fully unshared since it was originally tagged. | |
1523 | */ | |
51d62690 | 1524 | if (!xfs_inode_has_cow_data(ip)) { |
83104d44 DW |
1525 | trace_xfs_inode_free_cowblocks_invalid(ip); |
1526 | xfs_inode_clear_cowblocks_tag(ip); | |
be78ff0e | 1527 | return false; |
83104d44 DW |
1528 | } |
1529 | ||
1530 | /* | |
1531 | * If the mapping is dirty or under writeback we cannot touch the | |
1532 | * CoW fork. Leave it alone if we're in the midst of a directio. | |
1533 | */ | |
a1b7a4de CH |
1534 | if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) || |
1535 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) || | |
83104d44 DW |
1536 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) || |
1537 | atomic_read(&VFS_I(ip)->i_dio_count)) | |
be78ff0e DW |
1538 | return false; |
1539 | ||
1540 | return true; | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * Automatic CoW Reservation Freeing | |
1545 | * | |
1546 | * These functions automatically garbage collect leftover CoW reservations | |
1547 | * that were made on behalf of a cowextsize hint when we start to run out | |
1548 | * of quota or when the reservations sit around for too long. If the file | |
1549 | * has dirty pages or is undergoing writeback, its CoW reservations will | |
1550 | * be retained. | |
1551 | * | |
1552 | * The actual garbage collection piggybacks off the same code that runs | |
1553 | * the speculative EOF preallocation garbage collector. | |
1554 | */ | |
1555 | STATIC int | |
1556 | xfs_inode_free_cowblocks( | |
1557 | struct xfs_inode *ip, | |
0fa4a10a DW |
1558 | void *args, |
1559 | unsigned int *lockflags) | |
be78ff0e DW |
1560 | { |
1561 | struct xfs_eofblocks *eofb = args; | |
f41a0716 | 1562 | bool wait; |
be78ff0e DW |
1563 | int ret = 0; |
1564 | ||
f41a0716 DW |
1565 | wait = eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC); |
1566 | ||
ce2d3bbe DW |
1567 | if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS)) |
1568 | return 0; | |
1569 | ||
51d62690 | 1570 | if (!xfs_prep_free_cowblocks(ip)) |
83104d44 DW |
1571 | return 0; |
1572 | ||
a91bf992 DW |
1573 | if (!xfs_inode_matches_eofb(ip, eofb)) |
1574 | return 0; | |
83104d44 | 1575 | |
f41a0716 DW |
1576 | /* |
1577 | * If the caller is waiting, return -EAGAIN to keep the background | |
1578 | * scanner moving and revisit the inode in a subsequent pass. | |
1579 | */ | |
0fa4a10a DW |
1580 | if (!(*lockflags & XFS_IOLOCK_EXCL) && |
1581 | !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
f41a0716 DW |
1582 | if (wait) |
1583 | return -EAGAIN; | |
1584 | return 0; | |
1585 | } | |
0fa4a10a DW |
1586 | *lockflags |= XFS_IOLOCK_EXCL; |
1587 | ||
f41a0716 DW |
1588 | if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) { |
1589 | if (wait) | |
0fa4a10a DW |
1590 | return -EAGAIN; |
1591 | return 0; | |
f41a0716 | 1592 | } |
0fa4a10a | 1593 | *lockflags |= XFS_MMAPLOCK_EXCL; |
83104d44 | 1594 | |
be78ff0e DW |
1595 | /* |
1596 | * Check again, nobody else should be able to dirty blocks or change | |
1597 | * the reflink iflag now that we have the first two locks held. | |
1598 | */ | |
51d62690 | 1599 | if (xfs_prep_free_cowblocks(ip)) |
be78ff0e | 1600 | ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false); |
83104d44 DW |
1601 | return ret; |
1602 | } | |
1603 | ||
83104d44 DW |
1604 | void |
1605 | xfs_inode_set_cowblocks_tag( | |
1606 | xfs_inode_t *ip) | |
1607 | { | |
7b7381f0 | 1608 | trace_xfs_inode_set_cowblocks_tag(ip); |
9669f51d | 1609 | return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 DW |
1610 | } |
1611 | ||
1612 | void | |
1613 | xfs_inode_clear_cowblocks_tag( | |
1614 | xfs_inode_t *ip) | |
1615 | { | |
7b7381f0 | 1616 | trace_xfs_inode_clear_cowblocks_tag(ip); |
ce2d3bbe | 1617 | return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 | 1618 | } |
d6b636eb | 1619 | |
894ecacf DW |
1620 | #define for_each_perag_tag(mp, next_agno, pag, tag) \ |
1621 | for ((next_agno) = 0, (pag) = xfs_perag_get_tag((mp), 0, (tag)); \ | |
1622 | (pag) != NULL; \ | |
1623 | (next_agno) = (pag)->pag_agno + 1, \ | |
1624 | xfs_perag_put(pag), \ | |
1625 | (pag) = xfs_perag_get_tag((mp), (next_agno), (tag))) | |
1626 | ||
1627 | ||
d6b636eb DW |
1628 | /* Disable post-EOF and CoW block auto-reclamation. */ |
1629 | void | |
c9a6526f | 1630 | xfs_blockgc_stop( |
d6b636eb DW |
1631 | struct xfs_mount *mp) |
1632 | { | |
894ecacf DW |
1633 | struct xfs_perag *pag; |
1634 | xfs_agnumber_t agno; | |
1635 | ||
1636 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1637 | cancel_delayed_work_sync(&pag->pag_blockgc_work); | |
d6b636eb DW |
1638 | } |
1639 | ||
1640 | /* Enable post-EOF and CoW block auto-reclamation. */ | |
1641 | void | |
c9a6526f | 1642 | xfs_blockgc_start( |
d6b636eb DW |
1643 | struct xfs_mount *mp) |
1644 | { | |
894ecacf DW |
1645 | struct xfs_perag *pag; |
1646 | xfs_agnumber_t agno; | |
1647 | ||
1648 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1649 | xfs_blockgc_queue(pag); | |
d6b636eb | 1650 | } |
3d4feec0 | 1651 | |
41956753 DW |
1652 | /* Scan one incore inode for block preallocations that we can remove. */ |
1653 | static int | |
1654 | xfs_blockgc_scan_inode( | |
1655 | struct xfs_inode *ip, | |
1656 | void *args) | |
85c5b270 | 1657 | { |
0fa4a10a | 1658 | unsigned int lockflags = 0; |
85c5b270 DW |
1659 | int error; |
1660 | ||
0fa4a10a | 1661 | error = xfs_inode_free_eofblocks(ip, args, &lockflags); |
85c5b270 | 1662 | if (error) |
0fa4a10a | 1663 | goto unlock; |
85c5b270 | 1664 | |
0fa4a10a DW |
1665 | error = xfs_inode_free_cowblocks(ip, args, &lockflags); |
1666 | unlock: | |
1667 | if (lockflags) | |
1668 | xfs_iunlock(ip, lockflags); | |
1669 | return error; | |
85c5b270 DW |
1670 | } |
1671 | ||
9669f51d DW |
1672 | /* Background worker that trims preallocated space. */ |
1673 | void | |
1674 | xfs_blockgc_worker( | |
1675 | struct work_struct *work) | |
1676 | { | |
894ecacf DW |
1677 | struct xfs_perag *pag = container_of(to_delayed_work(work), |
1678 | struct xfs_perag, pag_blockgc_work); | |
1679 | struct xfs_mount *mp = pag->pag_mount; | |
9669f51d DW |
1680 | int error; |
1681 | ||
1682 | if (!sb_start_write_trylock(mp->m_super)) | |
1683 | return; | |
894ecacf | 1684 | error = xfs_inode_walk_ag(pag, 0, xfs_blockgc_scan_inode, NULL, |
41956753 | 1685 | XFS_ICI_BLOCKGC_TAG); |
9669f51d | 1686 | if (error) |
894ecacf DW |
1687 | xfs_info(mp, "AG %u preallocation gc worker failed, err=%d", |
1688 | pag->pag_agno, error); | |
9669f51d | 1689 | sb_end_write(mp->m_super); |
894ecacf | 1690 | xfs_blockgc_queue(pag); |
9669f51d DW |
1691 | } |
1692 | ||
85c5b270 DW |
1693 | /* |
1694 | * Try to free space in the filesystem by purging eofblocks and cowblocks. | |
1695 | */ | |
1696 | int | |
1697 | xfs_blockgc_free_space( | |
1698 | struct xfs_mount *mp, | |
1699 | struct xfs_eofblocks *eofb) | |
1700 | { | |
1701 | trace_xfs_blockgc_free_space(mp, eofb, _RET_IP_); | |
1702 | ||
41956753 DW |
1703 | return xfs_inode_walk(mp, 0, xfs_blockgc_scan_inode, eofb, |
1704 | XFS_ICI_BLOCKGC_TAG); | |
85c5b270 DW |
1705 | } |
1706 | ||
3d4feec0 | 1707 | /* |
c237dd7c DW |
1708 | * Run cow/eofblocks scans on the supplied dquots. We don't know exactly which |
1709 | * quota caused an allocation failure, so we make a best effort by including | |
1710 | * each quota under low free space conditions (less than 1% free space) in the | |
1711 | * scan. | |
111068f8 DW |
1712 | * |
1713 | * Callers must not hold any inode's ILOCK. If requesting a synchronous scan | |
1714 | * (XFS_EOF_FLAGS_SYNC), the caller also must not hold any inode's IOLOCK or | |
1715 | * MMAPLOCK. | |
3d4feec0 | 1716 | */ |
111068f8 | 1717 | int |
c237dd7c DW |
1718 | xfs_blockgc_free_dquots( |
1719 | struct xfs_mount *mp, | |
1720 | struct xfs_dquot *udqp, | |
1721 | struct xfs_dquot *gdqp, | |
1722 | struct xfs_dquot *pdqp, | |
111068f8 | 1723 | unsigned int eof_flags) |
3d4feec0 DW |
1724 | { |
1725 | struct xfs_eofblocks eofb = {0}; | |
3d4feec0 DW |
1726 | bool do_work = false; |
1727 | ||
c237dd7c DW |
1728 | if (!udqp && !gdqp && !pdqp) |
1729 | return 0; | |
1730 | ||
3d4feec0 | 1731 | /* |
111068f8 DW |
1732 | * Run a scan to free blocks using the union filter to cover all |
1733 | * applicable quotas in a single scan. | |
3d4feec0 | 1734 | */ |
111068f8 | 1735 | eofb.eof_flags = XFS_EOF_FLAGS_UNION | eof_flags; |
3d4feec0 | 1736 | |
c237dd7c DW |
1737 | if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) { |
1738 | eofb.eof_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id); | |
1739 | eofb.eof_flags |= XFS_EOF_FLAGS_UID; | |
1740 | do_work = true; | |
3d4feec0 DW |
1741 | } |
1742 | ||
c237dd7c DW |
1743 | if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) { |
1744 | eofb.eof_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id); | |
1745 | eofb.eof_flags |= XFS_EOF_FLAGS_GID; | |
1746 | do_work = true; | |
3d4feec0 DW |
1747 | } |
1748 | ||
c237dd7c DW |
1749 | if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) { |
1750 | eofb.eof_prid = pdqp->q_id; | |
1751 | eofb.eof_flags |= XFS_EOF_FLAGS_PRID; | |
1752 | do_work = true; | |
3d4feec0 DW |
1753 | } |
1754 | ||
1755 | if (!do_work) | |
111068f8 | 1756 | return 0; |
3d4feec0 | 1757 | |
85c5b270 | 1758 | return xfs_blockgc_free_space(mp, &eofb); |
c237dd7c DW |
1759 | } |
1760 | ||
1761 | /* Run cow/eofblocks scans on the quotas attached to the inode. */ | |
1762 | int | |
1763 | xfs_blockgc_free_quota( | |
1764 | struct xfs_inode *ip, | |
1765 | unsigned int eof_flags) | |
1766 | { | |
1767 | return xfs_blockgc_free_dquots(ip->i_mount, | |
1768 | xfs_inode_dquot(ip, XFS_DQTYPE_USER), | |
1769 | xfs_inode_dquot(ip, XFS_DQTYPE_GROUP), | |
1770 | xfs_inode_dquot(ip, XFS_DQTYPE_PROJ), eof_flags); | |
3d4feec0 | 1771 | } |