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_mount.h" |
fe4fa4b8 | 13 | #include "xfs_inode.h" |
239880ef DC |
14 | #include "xfs_trans.h" |
15 | #include "xfs_trans_priv.h" | |
fe4fa4b8 | 16 | #include "xfs_inode_item.h" |
7d095257 | 17 | #include "xfs_quota.h" |
0b1b213f | 18 | #include "xfs_trace.h" |
6d8b79cf | 19 | #include "xfs_icache.h" |
c24b5dfa | 20 | #include "xfs_bmap_util.h" |
dc06f398 BF |
21 | #include "xfs_dquot_item.h" |
22 | #include "xfs_dquot.h" | |
83104d44 | 23 | #include "xfs_reflink.h" |
bb8a66af | 24 | #include "xfs_ialloc.h" |
9bbafc71 | 25 | #include "xfs_ag.h" |
01728b44 | 26 | #include "xfs_log_priv.h" |
fe4fa4b8 | 27 | |
f0e28280 | 28 | #include <linux/iversion.h> |
a167b17e | 29 | |
c809d7e9 DW |
30 | /* Radix tree tags for incore inode tree. */ |
31 | ||
32 | /* inode is to be reclaimed */ | |
33 | #define XFS_ICI_RECLAIM_TAG 0 | |
34 | /* Inode has speculative preallocations (posteof or cow) to clean. */ | |
35 | #define XFS_ICI_BLOCKGC_TAG 1 | |
36 | ||
37 | /* | |
38 | * The goal for walking incore inodes. These can correspond with incore inode | |
39 | * radix tree tags when convenient. Avoid existing XFS_IWALK namespace. | |
40 | */ | |
41 | enum xfs_icwalk_goal { | |
c809d7e9 DW |
42 | /* Goals directly associated with tagged inodes. */ |
43 | XFS_ICWALK_BLOCKGC = XFS_ICI_BLOCKGC_TAG, | |
f1bc5c56 | 44 | XFS_ICWALK_RECLAIM = XFS_ICI_RECLAIM_TAG, |
c809d7e9 DW |
45 | }; |
46 | ||
7fdff526 | 47 | static int xfs_icwalk(struct xfs_mount *mp, |
b26b2bf1 | 48 | enum xfs_icwalk_goal goal, struct xfs_icwalk *icw); |
7fdff526 | 49 | static int xfs_icwalk_ag(struct xfs_perag *pag, |
b26b2bf1 | 50 | enum xfs_icwalk_goal goal, struct xfs_icwalk *icw); |
df600197 | 51 | |
1ad2cfe0 | 52 | /* |
b26b2bf1 DW |
53 | * Private inode cache walk flags for struct xfs_icwalk. Must not |
54 | * coincide with XFS_ICWALK_FLAGS_VALID. | |
1ad2cfe0 | 55 | */ |
1ad2cfe0 | 56 | |
f1bc5c56 DW |
57 | /* Stop scanning after icw_scan_limit inodes. */ |
58 | #define XFS_ICWALK_FLAG_SCAN_LIMIT (1U << 28) | |
59 | ||
9492750a | 60 | #define XFS_ICWALK_FLAG_RECLAIM_SICK (1U << 27) |
2d53f66b | 61 | #define XFS_ICWALK_FLAG_UNION (1U << 26) /* union filter algorithm */ |
9492750a | 62 | |
777eb1fa | 63 | #define XFS_ICWALK_PRIVATE_FLAGS (XFS_ICWALK_FLAG_SCAN_LIMIT | \ |
2d53f66b DW |
64 | XFS_ICWALK_FLAG_RECLAIM_SICK | \ |
65 | XFS_ICWALK_FLAG_UNION) | |
1ad2cfe0 | 66 | |
33479e05 DC |
67 | /* |
68 | * Allocate and initialise an xfs_inode. | |
69 | */ | |
638f4416 | 70 | struct xfs_inode * |
33479e05 DC |
71 | xfs_inode_alloc( |
72 | struct xfs_mount *mp, | |
73 | xfs_ino_t ino) | |
74 | { | |
75 | struct xfs_inode *ip; | |
76 | ||
77 | /* | |
3050bd0b CM |
78 | * XXX: If this didn't occur in transactions, we could drop GFP_NOFAIL |
79 | * and return NULL here on ENOMEM. | |
33479e05 | 80 | */ |
fd60b288 | 81 | ip = alloc_inode_sb(mp->m_super, xfs_inode_cache, GFP_KERNEL | __GFP_NOFAIL); |
3050bd0b | 82 | |
33479e05 | 83 | if (inode_init_always(mp->m_super, VFS_I(ip))) { |
182696fb | 84 | kmem_cache_free(xfs_inode_cache, ip); |
33479e05 DC |
85 | return NULL; |
86 | } | |
87 | ||
f38a032b | 88 | /* VFS doesn't initialise i_mode or i_state! */ |
c19b3b05 | 89 | VFS_I(ip)->i_mode = 0; |
f38a032b | 90 | VFS_I(ip)->i_state = 0; |
67958013 | 91 | mapping_set_large_folios(VFS_I(ip)->i_mapping); |
c19b3b05 | 92 | |
ff6d6af2 | 93 | XFS_STATS_INC(mp, vn_active); |
33479e05 | 94 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
33479e05 DC |
95 | ASSERT(ip->i_ino == 0); |
96 | ||
33479e05 DC |
97 | /* initialise the xfs inode */ |
98 | ip->i_ino = ino; | |
99 | ip->i_mount = mp; | |
100 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
101 | ip->i_afp = NULL; | |
3993baeb | 102 | ip->i_cowfp = NULL; |
3ba738df | 103 | memset(&ip->i_df, 0, sizeof(ip->i_df)); |
33479e05 DC |
104 | ip->i_flags = 0; |
105 | ip->i_delayed_blks = 0; | |
3e09ab8f | 106 | ip->i_diflags2 = mp->m_ino_geo.new_diflags2; |
6e73a545 | 107 | ip->i_nblocks = 0; |
7821ea30 | 108 | ip->i_forkoff = 0; |
6772c1f1 DW |
109 | ip->i_sick = 0; |
110 | ip->i_checked = 0; | |
cb357bf3 DW |
111 | INIT_WORK(&ip->i_ioend_work, xfs_end_io); |
112 | INIT_LIST_HEAD(&ip->i_ioend_list); | |
113 | spin_lock_init(&ip->i_ioend_lock); | |
33479e05 DC |
114 | |
115 | return ip; | |
116 | } | |
117 | ||
118 | STATIC void | |
119 | xfs_inode_free_callback( | |
120 | struct rcu_head *head) | |
121 | { | |
122 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
123 | struct xfs_inode *ip = XFS_I(inode); | |
124 | ||
c19b3b05 | 125 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
126 | case S_IFREG: |
127 | case S_IFDIR: | |
128 | case S_IFLNK: | |
ef838512 | 129 | xfs_idestroy_fork(&ip->i_df); |
33479e05 DC |
130 | break; |
131 | } | |
132 | ||
ef838512 CH |
133 | if (ip->i_afp) { |
134 | xfs_idestroy_fork(ip->i_afp); | |
182696fb | 135 | kmem_cache_free(xfs_ifork_cache, ip->i_afp); |
ef838512 CH |
136 | } |
137 | if (ip->i_cowfp) { | |
138 | xfs_idestroy_fork(ip->i_cowfp); | |
182696fb | 139 | kmem_cache_free(xfs_ifork_cache, ip->i_cowfp); |
ef838512 | 140 | } |
33479e05 | 141 | if (ip->i_itemp) { |
22525c17 DC |
142 | ASSERT(!test_bit(XFS_LI_IN_AIL, |
143 | &ip->i_itemp->ili_item.li_flags)); | |
33479e05 DC |
144 | xfs_inode_item_destroy(ip); |
145 | ip->i_itemp = NULL; | |
146 | } | |
147 | ||
182696fb | 148 | kmem_cache_free(xfs_inode_cache, ip); |
1f2dcfe8 DC |
149 | } |
150 | ||
8a17d7dd DC |
151 | static void |
152 | __xfs_inode_free( | |
153 | struct xfs_inode *ip) | |
154 | { | |
155 | /* asserts to verify all state is correct here */ | |
156 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
48d55e2a | 157 | ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list)); |
8a17d7dd DC |
158 | XFS_STATS_DEC(ip->i_mount, vn_active); |
159 | ||
160 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
161 | } | |
162 | ||
1f2dcfe8 DC |
163 | void |
164 | xfs_inode_free( | |
165 | struct xfs_inode *ip) | |
166 | { | |
718ecc50 | 167 | ASSERT(!xfs_iflags_test(ip, XFS_IFLUSHING)); |
98efe8af | 168 | |
33479e05 DC |
169 | /* |
170 | * Because we use RCU freeing we need to ensure the inode always | |
171 | * appears to be reclaimed with an invalid inode number when in the | |
172 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
173 | * races. | |
174 | */ | |
175 | spin_lock(&ip->i_flags_lock); | |
176 | ip->i_flags = XFS_IRECLAIM; | |
177 | ip->i_ino = 0; | |
178 | spin_unlock(&ip->i_flags_lock); | |
179 | ||
8a17d7dd | 180 | __xfs_inode_free(ip); |
33479e05 DC |
181 | } |
182 | ||
ad438c40 | 183 | /* |
02511a5a DC |
184 | * Queue background inode reclaim work if there are reclaimable inodes and there |
185 | * isn't reclaim work already scheduled or in progress. | |
ad438c40 DC |
186 | */ |
187 | static void | |
188 | xfs_reclaim_work_queue( | |
189 | struct xfs_mount *mp) | |
190 | { | |
191 | ||
192 | rcu_read_lock(); | |
193 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
194 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
195 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
196 | } | |
197 | rcu_read_unlock(); | |
198 | } | |
199 | ||
c076ae7a DW |
200 | /* |
201 | * Background scanning to trim preallocated space. This is queued based on the | |
202 | * 'speculative_prealloc_lifetime' tunable (5m by default). | |
203 | */ | |
204 | static inline void | |
205 | xfs_blockgc_queue( | |
ad438c40 | 206 | struct xfs_perag *pag) |
c076ae7a | 207 | { |
6f649091 DW |
208 | struct xfs_mount *mp = pag->pag_mount; |
209 | ||
210 | if (!xfs_is_blockgc_enabled(mp)) | |
211 | return; | |
212 | ||
c076ae7a DW |
213 | rcu_read_lock(); |
214 | if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG)) | |
ab23a776 | 215 | queue_delayed_work(pag->pag_mount->m_blockgc_wq, |
c076ae7a DW |
216 | &pag->pag_blockgc_work, |
217 | msecs_to_jiffies(xfs_blockgc_secs * 1000)); | |
218 | rcu_read_unlock(); | |
219 | } | |
220 | ||
221 | /* Set a tag on both the AG incore inode tree and the AG radix tree. */ | |
222 | static void | |
223 | xfs_perag_set_inode_tag( | |
224 | struct xfs_perag *pag, | |
225 | xfs_agino_t agino, | |
226 | unsigned int tag) | |
ad438c40 DC |
227 | { |
228 | struct xfs_mount *mp = pag->pag_mount; | |
c076ae7a | 229 | bool was_tagged; |
ad438c40 | 230 | |
95989c46 | 231 | lockdep_assert_held(&pag->pag_ici_lock); |
c076ae7a DW |
232 | |
233 | was_tagged = radix_tree_tagged(&pag->pag_ici_root, tag); | |
234 | radix_tree_tag_set(&pag->pag_ici_root, agino, tag); | |
235 | ||
236 | if (tag == XFS_ICI_RECLAIM_TAG) | |
237 | pag->pag_ici_reclaimable++; | |
238 | ||
239 | if (was_tagged) | |
ad438c40 DC |
240 | return; |
241 | ||
c076ae7a | 242 | /* propagate the tag up into the perag radix tree */ |
ad438c40 | 243 | spin_lock(&mp->m_perag_lock); |
c076ae7a | 244 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, tag); |
ad438c40 DC |
245 | spin_unlock(&mp->m_perag_lock); |
246 | ||
c076ae7a DW |
247 | /* start background work */ |
248 | switch (tag) { | |
249 | case XFS_ICI_RECLAIM_TAG: | |
250 | xfs_reclaim_work_queue(mp); | |
251 | break; | |
252 | case XFS_ICI_BLOCKGC_TAG: | |
253 | xfs_blockgc_queue(pag); | |
254 | break; | |
255 | } | |
ad438c40 | 256 | |
c076ae7a | 257 | trace_xfs_perag_set_inode_tag(mp, pag->pag_agno, tag, _RET_IP_); |
ad438c40 DC |
258 | } |
259 | ||
c076ae7a | 260 | /* Clear a tag on both the AG incore inode tree and the AG radix tree. */ |
ad438c40 | 261 | static void |
c076ae7a DW |
262 | xfs_perag_clear_inode_tag( |
263 | struct xfs_perag *pag, | |
264 | xfs_agino_t agino, | |
265 | unsigned int tag) | |
ad438c40 DC |
266 | { |
267 | struct xfs_mount *mp = pag->pag_mount; | |
268 | ||
95989c46 | 269 | lockdep_assert_held(&pag->pag_ici_lock); |
c076ae7a DW |
270 | |
271 | /* | |
272 | * Reclaim can signal (with a null agino) that it cleared its own tag | |
273 | * by removing the inode from the radix tree. | |
274 | */ | |
275 | if (agino != NULLAGINO) | |
276 | radix_tree_tag_clear(&pag->pag_ici_root, agino, tag); | |
277 | else | |
278 | ASSERT(tag == XFS_ICI_RECLAIM_TAG); | |
279 | ||
280 | if (tag == XFS_ICI_RECLAIM_TAG) | |
281 | pag->pag_ici_reclaimable--; | |
282 | ||
283 | if (radix_tree_tagged(&pag->pag_ici_root, tag)) | |
ad438c40 DC |
284 | return; |
285 | ||
c076ae7a | 286 | /* clear the tag from the perag radix tree */ |
ad438c40 | 287 | spin_lock(&mp->m_perag_lock); |
c076ae7a | 288 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, tag); |
ad438c40 | 289 | spin_unlock(&mp->m_perag_lock); |
ad438c40 | 290 | |
c076ae7a DW |
291 | trace_xfs_perag_clear_inode_tag(mp, pag->pag_agno, tag, _RET_IP_); |
292 | } | |
ad438c40 | 293 | |
50997470 DC |
294 | /* |
295 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
296 | * part of the structure. This is made more complex by the fact we store | |
297 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 298 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 299 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 300 | * after reinitialisation even if it fails. |
50997470 DC |
301 | */ |
302 | static int | |
303 | xfs_reinit_inode( | |
304 | struct xfs_mount *mp, | |
305 | struct inode *inode) | |
306 | { | |
ff7bebeb DW |
307 | int error; |
308 | uint32_t nlink = inode->i_nlink; | |
309 | uint32_t generation = inode->i_generation; | |
310 | uint64_t version = inode_peek_iversion(inode); | |
311 | umode_t mode = inode->i_mode; | |
312 | dev_t dev = inode->i_rdev; | |
313 | kuid_t uid = inode->i_uid; | |
314 | kgid_t gid = inode->i_gid; | |
50997470 DC |
315 | |
316 | error = inode_init_always(mp->m_super, inode); | |
317 | ||
54d7b5c1 | 318 | set_nlink(inode, nlink); |
9e9a2674 | 319 | inode->i_generation = generation; |
f0e28280 | 320 | inode_set_iversion_queried(inode, version); |
c19b3b05 | 321 | inode->i_mode = mode; |
acd1d715 | 322 | inode->i_rdev = dev; |
3d8f2821 CH |
323 | inode->i_uid = uid; |
324 | inode->i_gid = gid; | |
67958013 | 325 | mapping_set_large_folios(inode->i_mapping); |
50997470 DC |
326 | return error; |
327 | } | |
328 | ||
ff7bebeb DW |
329 | /* |
330 | * Carefully nudge an inode whose VFS state has been torn down back into a | |
331 | * usable state. Drops the i_flags_lock and the rcu read lock. | |
332 | */ | |
333 | static int | |
334 | xfs_iget_recycle( | |
335 | struct xfs_perag *pag, | |
336 | struct xfs_inode *ip) __releases(&ip->i_flags_lock) | |
337 | { | |
338 | struct xfs_mount *mp = ip->i_mount; | |
339 | struct inode *inode = VFS_I(ip); | |
340 | int error; | |
341 | ||
342 | trace_xfs_iget_recycle(ip); | |
343 | ||
344 | /* | |
345 | * We need to make it look like the inode is being reclaimed to prevent | |
346 | * the actual reclaim workers from stomping over us while we recycle | |
347 | * the inode. We can't clear the radix tree tag yet as it requires | |
348 | * pag_ici_lock to be held exclusive. | |
349 | */ | |
350 | ip->i_flags |= XFS_IRECLAIM; | |
351 | ||
352 | spin_unlock(&ip->i_flags_lock); | |
353 | rcu_read_unlock(); | |
354 | ||
355 | ASSERT(!rwsem_is_locked(&inode->i_rwsem)); | |
356 | error = xfs_reinit_inode(mp, inode); | |
357 | if (error) { | |
ff7bebeb DW |
358 | /* |
359 | * Re-initializing the inode failed, and we are in deep | |
360 | * trouble. Try to re-add it to the reclaim list. | |
361 | */ | |
362 | rcu_read_lock(); | |
363 | spin_lock(&ip->i_flags_lock); | |
ff7bebeb | 364 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
ff7bebeb DW |
365 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); |
366 | spin_unlock(&ip->i_flags_lock); | |
367 | rcu_read_unlock(); | |
368 | ||
369 | trace_xfs_iget_recycle_fail(ip); | |
370 | return error; | |
371 | } | |
372 | ||
373 | spin_lock(&pag->pag_ici_lock); | |
374 | spin_lock(&ip->i_flags_lock); | |
375 | ||
376 | /* | |
377 | * Clear the per-lifetime state in the inode as we are now effectively | |
378 | * a new inode and need to return to the initial state before reuse | |
379 | * occurs. | |
380 | */ | |
381 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
382 | ip->i_flags |= XFS_INEW; | |
383 | xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
384 | XFS_ICI_RECLAIM_TAG); | |
385 | inode->i_state = I_NEW; | |
386 | spin_unlock(&ip->i_flags_lock); | |
387 | spin_unlock(&pag->pag_ici_lock); | |
388 | ||
389 | return 0; | |
390 | } | |
391 | ||
afca6c5b DC |
392 | /* |
393 | * If we are allocating a new inode, then check what was returned is | |
394 | * actually a free, empty inode. If we are not allocating an inode, | |
395 | * then check we didn't find a free inode. | |
396 | * | |
397 | * Returns: | |
398 | * 0 if the inode free state matches the lookup context | |
399 | * -ENOENT if the inode is free and we are not allocating | |
400 | * -EFSCORRUPTED if there is any state mismatch at all | |
401 | */ | |
402 | static int | |
403 | xfs_iget_check_free_state( | |
404 | struct xfs_inode *ip, | |
405 | int flags) | |
406 | { | |
407 | if (flags & XFS_IGET_CREATE) { | |
408 | /* should be a free inode */ | |
409 | if (VFS_I(ip)->i_mode != 0) { | |
410 | xfs_warn(ip->i_mount, | |
411 | "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)", | |
412 | ip->i_ino, VFS_I(ip)->i_mode); | |
413 | return -EFSCORRUPTED; | |
414 | } | |
415 | ||
6e73a545 | 416 | if (ip->i_nblocks != 0) { |
afca6c5b DC |
417 | xfs_warn(ip->i_mount, |
418 | "Corruption detected! Free inode 0x%llx has blocks allocated!", | |
419 | ip->i_ino); | |
420 | return -EFSCORRUPTED; | |
421 | } | |
422 | return 0; | |
423 | } | |
424 | ||
425 | /* should be an allocated inode */ | |
426 | if (VFS_I(ip)->i_mode == 0) | |
427 | return -ENOENT; | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
ab23a776 DC |
432 | /* Make all pending inactivation work start immediately. */ |
433 | static void | |
434 | xfs_inodegc_queue_all( | |
435 | struct xfs_mount *mp) | |
436 | { | |
437 | struct xfs_inodegc *gc; | |
438 | int cpu; | |
439 | ||
440 | for_each_online_cpu(cpu) { | |
441 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
442 | if (!llist_empty(&gc->list)) | |
7cf2b0f9 | 443 | mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0); |
ab23a776 DC |
444 | } |
445 | } | |
446 | ||
33479e05 DC |
447 | /* |
448 | * Check the validity of the inode we just found it the cache | |
449 | */ | |
450 | static int | |
451 | xfs_iget_cache_hit( | |
452 | struct xfs_perag *pag, | |
453 | struct xfs_inode *ip, | |
454 | xfs_ino_t ino, | |
455 | int flags, | |
456 | int lock_flags) __releases(RCU) | |
457 | { | |
458 | struct inode *inode = VFS_I(ip); | |
459 | struct xfs_mount *mp = ip->i_mount; | |
460 | int error; | |
461 | ||
462 | /* | |
463 | * check for re-use of an inode within an RCU grace period due to the | |
464 | * radix tree nodes not being updated yet. We monitor for this by | |
465 | * setting the inode number to zero before freeing the inode structure. | |
466 | * If the inode has been reallocated and set up, then the inode number | |
467 | * will not match, so check for that, too. | |
468 | */ | |
469 | spin_lock(&ip->i_flags_lock); | |
77b4d286 DW |
470 | if (ip->i_ino != ino) |
471 | goto out_skip; | |
33479e05 DC |
472 | |
473 | /* | |
474 | * If we are racing with another cache hit that is currently | |
475 | * instantiating this inode or currently recycling it out of | |
ff7bebeb | 476 | * reclaimable state, wait for the initialisation to complete |
33479e05 DC |
477 | * before continuing. |
478 | * | |
ab23a776 DC |
479 | * If we're racing with the inactivation worker we also want to wait. |
480 | * If we're creating a new file, it's possible that the worker | |
481 | * previously marked the inode as free on disk but hasn't finished | |
482 | * updating the incore state yet. The AGI buffer will be dirty and | |
483 | * locked to the icreate transaction, so a synchronous push of the | |
484 | * inodegc workers would result in deadlock. For a regular iget, the | |
485 | * worker is running already, so we might as well wait. | |
486 | * | |
33479e05 DC |
487 | * XXX(hch): eventually we should do something equivalent to |
488 | * wait_on_inode to wait for these flags to be cleared | |
489 | * instead of polling for it. | |
490 | */ | |
ab23a776 | 491 | if (ip->i_flags & (XFS_INEW | XFS_IRECLAIM | XFS_INACTIVATING)) |
77b4d286 | 492 | goto out_skip; |
33479e05 | 493 | |
ab23a776 DC |
494 | if (ip->i_flags & XFS_NEED_INACTIVE) { |
495 | /* Unlinked inodes cannot be re-grabbed. */ | |
496 | if (VFS_I(ip)->i_nlink == 0) { | |
497 | error = -ENOENT; | |
498 | goto out_error; | |
499 | } | |
500 | goto out_inodegc_flush; | |
501 | } | |
502 | ||
33479e05 | 503 | /* |
afca6c5b DC |
504 | * Check the inode free state is valid. This also detects lookup |
505 | * racing with unlinks. | |
33479e05 | 506 | */ |
afca6c5b DC |
507 | error = xfs_iget_check_free_state(ip, flags); |
508 | if (error) | |
33479e05 | 509 | goto out_error; |
33479e05 | 510 | |
77b4d286 DW |
511 | /* Skip inodes that have no vfs state. */ |
512 | if ((flags & XFS_IGET_INCORE) && | |
513 | (ip->i_flags & XFS_IRECLAIMABLE)) | |
514 | goto out_skip; | |
378f681c | 515 | |
77b4d286 DW |
516 | /* The inode fits the selection criteria; process it. */ |
517 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
ff7bebeb DW |
518 | /* Drops i_flags_lock and RCU read lock. */ |
519 | error = xfs_iget_recycle(pag, ip); | |
520 | if (error) | |
521 | return error; | |
33479e05 DC |
522 | } else { |
523 | /* If the VFS inode is being torn down, pause and try again. */ | |
77b4d286 DW |
524 | if (!igrab(inode)) |
525 | goto out_skip; | |
33479e05 DC |
526 | |
527 | /* We've got a live one. */ | |
528 | spin_unlock(&ip->i_flags_lock); | |
529 | rcu_read_unlock(); | |
530 | trace_xfs_iget_hit(ip); | |
531 | } | |
532 | ||
533 | if (lock_flags != 0) | |
534 | xfs_ilock(ip, lock_flags); | |
535 | ||
378f681c | 536 | if (!(flags & XFS_IGET_INCORE)) |
dae2f8ed | 537 | xfs_iflags_clear(ip, XFS_ISTALE); |
ff6d6af2 | 538 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
539 | |
540 | return 0; | |
541 | ||
77b4d286 DW |
542 | out_skip: |
543 | trace_xfs_iget_skip(ip); | |
544 | XFS_STATS_INC(mp, xs_ig_frecycle); | |
545 | error = -EAGAIN; | |
33479e05 DC |
546 | out_error: |
547 | spin_unlock(&ip->i_flags_lock); | |
548 | rcu_read_unlock(); | |
549 | return error; | |
ab23a776 DC |
550 | |
551 | out_inodegc_flush: | |
552 | spin_unlock(&ip->i_flags_lock); | |
553 | rcu_read_unlock(); | |
554 | /* | |
555 | * Do not wait for the workers, because the caller could hold an AGI | |
556 | * buffer lock. We're just going to sleep in a loop anyway. | |
557 | */ | |
558 | if (xfs_is_inodegc_enabled(mp)) | |
559 | xfs_inodegc_queue_all(mp); | |
560 | return -EAGAIN; | |
33479e05 DC |
561 | } |
562 | ||
33479e05 DC |
563 | static int |
564 | xfs_iget_cache_miss( | |
565 | struct xfs_mount *mp, | |
566 | struct xfs_perag *pag, | |
567 | xfs_trans_t *tp, | |
568 | xfs_ino_t ino, | |
569 | struct xfs_inode **ipp, | |
570 | int flags, | |
571 | int lock_flags) | |
572 | { | |
573 | struct xfs_inode *ip; | |
574 | int error; | |
575 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
576 | int iflags; | |
577 | ||
578 | ip = xfs_inode_alloc(mp, ino); | |
579 | if (!ip) | |
2451337d | 580 | return -ENOMEM; |
33479e05 | 581 | |
bb8a66af | 582 | error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, flags); |
33479e05 DC |
583 | if (error) |
584 | goto out_destroy; | |
585 | ||
bb8a66af CH |
586 | /* |
587 | * For version 5 superblocks, if we are initialising a new inode and we | |
0560f31a | 588 | * are not utilising the XFS_FEAT_IKEEP inode cluster mode, we can |
bb8a66af CH |
589 | * simply build the new inode core with a random generation number. |
590 | * | |
591 | * For version 4 (and older) superblocks, log recovery is dependent on | |
965e0a1a | 592 | * the i_flushiter field being initialised from the current on-disk |
bb8a66af CH |
593 | * value and hence we must also read the inode off disk even when |
594 | * initializing new inodes. | |
595 | */ | |
38c26bfd | 596 | if (xfs_has_v3inodes(mp) && |
0560f31a | 597 | (flags & XFS_IGET_CREATE) && !xfs_has_ikeep(mp)) { |
bb8a66af CH |
598 | VFS_I(ip)->i_generation = prandom_u32(); |
599 | } else { | |
bb8a66af CH |
600 | struct xfs_buf *bp; |
601 | ||
af9dcdde | 602 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp); |
bb8a66af CH |
603 | if (error) |
604 | goto out_destroy; | |
605 | ||
af9dcdde CH |
606 | error = xfs_inode_from_disk(ip, |
607 | xfs_buf_offset(bp, ip->i_imap.im_boffset)); | |
bb8a66af CH |
608 | if (!error) |
609 | xfs_buf_set_ref(bp, XFS_INO_REF); | |
610 | xfs_trans_brelse(tp, bp); | |
611 | ||
612 | if (error) | |
613 | goto out_destroy; | |
614 | } | |
615 | ||
33479e05 DC |
616 | trace_xfs_iget_miss(ip); |
617 | ||
ee457001 | 618 | /* |
afca6c5b DC |
619 | * Check the inode free state is valid. This also detects lookup |
620 | * racing with unlinks. | |
ee457001 | 621 | */ |
afca6c5b DC |
622 | error = xfs_iget_check_free_state(ip, flags); |
623 | if (error) | |
33479e05 | 624 | goto out_destroy; |
33479e05 DC |
625 | |
626 | /* | |
627 | * Preload the radix tree so we can insert safely under the | |
628 | * write spinlock. Note that we cannot sleep inside the preload | |
629 | * region. Since we can be called from transaction context, don't | |
630 | * recurse into the file system. | |
631 | */ | |
632 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 633 | error = -EAGAIN; |
33479e05 DC |
634 | goto out_destroy; |
635 | } | |
636 | ||
637 | /* | |
638 | * Because the inode hasn't been added to the radix-tree yet it can't | |
639 | * be found by another thread, so we can do the non-sleeping lock here. | |
640 | */ | |
641 | if (lock_flags) { | |
642 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
643 | BUG(); | |
644 | } | |
645 | ||
646 | /* | |
647 | * These values must be set before inserting the inode into the radix | |
648 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
649 | * RCU locking mechanism) can find it and that lookup must see that this | |
650 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
651 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
652 | * memory barrier that ensures this detection works correctly at lookup | |
653 | * time. | |
654 | */ | |
655 | iflags = XFS_INEW; | |
656 | if (flags & XFS_IGET_DONTCACHE) | |
2c567af4 | 657 | d_mark_dontcache(VFS_I(ip)); |
113a5683 CS |
658 | ip->i_udquot = NULL; |
659 | ip->i_gdquot = NULL; | |
92f8ff73 | 660 | ip->i_pdquot = NULL; |
33479e05 DC |
661 | xfs_iflags_set(ip, iflags); |
662 | ||
663 | /* insert the new inode */ | |
664 | spin_lock(&pag->pag_ici_lock); | |
665 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
666 | if (unlikely(error)) { | |
667 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 668 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 669 | error = -EAGAIN; |
33479e05 DC |
670 | goto out_preload_end; |
671 | } | |
672 | spin_unlock(&pag->pag_ici_lock); | |
673 | radix_tree_preload_end(); | |
674 | ||
675 | *ipp = ip; | |
676 | return 0; | |
677 | ||
678 | out_preload_end: | |
679 | spin_unlock(&pag->pag_ici_lock); | |
680 | radix_tree_preload_end(); | |
681 | if (lock_flags) | |
682 | xfs_iunlock(ip, lock_flags); | |
683 | out_destroy: | |
684 | __destroy_inode(VFS_I(ip)); | |
685 | xfs_inode_free(ip); | |
686 | return error; | |
687 | } | |
688 | ||
689 | /* | |
02511a5a DC |
690 | * Look up an inode by number in the given file system. The inode is looked up |
691 | * in the cache held in each AG. If the inode is found in the cache, initialise | |
692 | * the vfs inode if necessary. | |
33479e05 | 693 | * |
02511a5a DC |
694 | * If it is not in core, read it in from the file system's device, add it to the |
695 | * cache and initialise the vfs inode. | |
33479e05 DC |
696 | * |
697 | * The inode is locked according to the value of the lock_flags parameter. | |
02511a5a DC |
698 | * Inode lookup is only done during metadata operations and not as part of the |
699 | * data IO path. Hence we only allow locking of the XFS_ILOCK during lookup. | |
33479e05 DC |
700 | */ |
701 | int | |
702 | xfs_iget( | |
02511a5a DC |
703 | struct xfs_mount *mp, |
704 | struct xfs_trans *tp, | |
705 | xfs_ino_t ino, | |
706 | uint flags, | |
707 | uint lock_flags, | |
708 | struct xfs_inode **ipp) | |
33479e05 | 709 | { |
02511a5a DC |
710 | struct xfs_inode *ip; |
711 | struct xfs_perag *pag; | |
712 | xfs_agino_t agino; | |
713 | int error; | |
33479e05 | 714 | |
33479e05 DC |
715 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); |
716 | ||
717 | /* reject inode numbers outside existing AGs */ | |
718 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 719 | return -EINVAL; |
33479e05 | 720 | |
ff6d6af2 | 721 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 722 | |
33479e05 DC |
723 | /* get the perag structure and ensure that it's inode capable */ |
724 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
725 | agino = XFS_INO_TO_AGINO(mp, ino); | |
726 | ||
727 | again: | |
728 | error = 0; | |
729 | rcu_read_lock(); | |
730 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
731 | ||
732 | if (ip) { | |
733 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
734 | if (error) | |
735 | goto out_error_or_again; | |
736 | } else { | |
737 | rcu_read_unlock(); | |
378f681c | 738 | if (flags & XFS_IGET_INCORE) { |
ed438b47 | 739 | error = -ENODATA; |
378f681c DW |
740 | goto out_error_or_again; |
741 | } | |
ff6d6af2 | 742 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
743 | |
744 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
745 | flags, lock_flags); | |
746 | if (error) | |
747 | goto out_error_or_again; | |
748 | } | |
749 | xfs_perag_put(pag); | |
750 | ||
751 | *ipp = ip; | |
752 | ||
753 | /* | |
58c90473 | 754 | * If we have a real type for an on-disk inode, we can setup the inode |
132c460e YX |
755 | * now. If it's a new inode being created, xfs_init_new_inode will |
756 | * handle it. | |
33479e05 | 757 | */ |
c19b3b05 | 758 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 759 | xfs_setup_existing_inode(ip); |
33479e05 DC |
760 | return 0; |
761 | ||
762 | out_error_or_again: | |
378f681c | 763 | if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) { |
33479e05 DC |
764 | delay(1); |
765 | goto again; | |
766 | } | |
767 | xfs_perag_put(pag); | |
768 | return error; | |
769 | } | |
770 | ||
378f681c DW |
771 | /* |
772 | * "Is this a cached inode that's also allocated?" | |
773 | * | |
774 | * Look up an inode by number in the given file system. If the inode is | |
775 | * in cache and isn't in purgatory, return 1 if the inode is allocated | |
776 | * and 0 if it is not. For all other cases (not in cache, being torn | |
777 | * down, etc.), return a negative error code. | |
778 | * | |
779 | * The caller has to prevent inode allocation and freeing activity, | |
780 | * presumably by locking the AGI buffer. This is to ensure that an | |
781 | * inode cannot transition from allocated to freed until the caller is | |
782 | * ready to allow that. If the inode is in an intermediate state (new, | |
783 | * reclaimable, or being reclaimed), -EAGAIN will be returned; if the | |
784 | * inode is not in the cache, -ENOENT will be returned. The caller must | |
785 | * deal with these scenarios appropriately. | |
786 | * | |
787 | * This is a specialized use case for the online scrubber; if you're | |
788 | * reading this, you probably want xfs_iget. | |
789 | */ | |
790 | int | |
791 | xfs_icache_inode_is_allocated( | |
792 | struct xfs_mount *mp, | |
793 | struct xfs_trans *tp, | |
794 | xfs_ino_t ino, | |
795 | bool *inuse) | |
796 | { | |
797 | struct xfs_inode *ip; | |
798 | int error; | |
799 | ||
800 | error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip); | |
801 | if (error) | |
802 | return error; | |
803 | ||
804 | *inuse = !!(VFS_I(ip)->i_mode); | |
44a8736b | 805 | xfs_irele(ip); |
378f681c DW |
806 | return 0; |
807 | } | |
808 | ||
e3a20c0b DC |
809 | /* |
810 | * Grab the inode for reclaim exclusively. | |
50718b8d DC |
811 | * |
812 | * We have found this inode via a lookup under RCU, so the inode may have | |
813 | * already been freed, or it may be in the process of being recycled by | |
814 | * xfs_iget(). In both cases, the inode will have XFS_IRECLAIM set. If the inode | |
815 | * has been fully recycled by the time we get the i_flags_lock, XFS_IRECLAIMABLE | |
816 | * will not be set. Hence we need to check for both these flag conditions to | |
817 | * avoid inodes that are no longer reclaim candidates. | |
818 | * | |
819 | * Note: checking for other state flags here, under the i_flags_lock or not, is | |
820 | * racy and should be avoided. Those races should be resolved only after we have | |
821 | * ensured that we are able to reclaim this inode and the world can see that we | |
822 | * are going to reclaim it. | |
823 | * | |
824 | * Return true if we grabbed it, false otherwise. | |
e3a20c0b | 825 | */ |
50718b8d | 826 | static bool |
f1bc5c56 | 827 | xfs_reclaim_igrab( |
9492750a | 828 | struct xfs_inode *ip, |
b26b2bf1 | 829 | struct xfs_icwalk *icw) |
e3a20c0b | 830 | { |
1a3e8f3d DC |
831 | ASSERT(rcu_read_lock_held()); |
832 | ||
e3a20c0b | 833 | spin_lock(&ip->i_flags_lock); |
1a3e8f3d DC |
834 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
835 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
836 | /* not a reclaim candidate. */ | |
e3a20c0b | 837 | spin_unlock(&ip->i_flags_lock); |
50718b8d | 838 | return false; |
e3a20c0b | 839 | } |
9492750a DW |
840 | |
841 | /* Don't reclaim a sick inode unless the caller asked for it. */ | |
842 | if (ip->i_sick && | |
b26b2bf1 | 843 | (!icw || !(icw->icw_flags & XFS_ICWALK_FLAG_RECLAIM_SICK))) { |
9492750a DW |
844 | spin_unlock(&ip->i_flags_lock); |
845 | return false; | |
846 | } | |
847 | ||
e3a20c0b DC |
848 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
849 | spin_unlock(&ip->i_flags_lock); | |
50718b8d | 850 | return true; |
e3a20c0b DC |
851 | } |
852 | ||
777df5af | 853 | /* |
02511a5a DC |
854 | * Inode reclaim is non-blocking, so the default action if progress cannot be |
855 | * made is to "requeue" the inode for reclaim by unlocking it and clearing the | |
856 | * XFS_IRECLAIM flag. If we are in a shutdown state, we don't care about | |
857 | * blocking anymore and hence we can wait for the inode to be able to reclaim | |
858 | * it. | |
777df5af | 859 | * |
02511a5a DC |
860 | * We do no IO here - if callers require inodes to be cleaned they must push the |
861 | * AIL first to trigger writeback of dirty inodes. This enables writeback to be | |
862 | * done in the background in a non-blocking manner, and enables memory reclaim | |
863 | * to make progress without blocking. | |
777df5af | 864 | */ |
4d0bab3a | 865 | static void |
c8e20be0 | 866 | xfs_reclaim_inode( |
75f3cb13 | 867 | struct xfs_inode *ip, |
50718b8d | 868 | struct xfs_perag *pag) |
fce08f2f | 869 | { |
8a17d7dd | 870 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
777df5af | 871 | |
9552e14d | 872 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) |
617825fe | 873 | goto out; |
718ecc50 | 874 | if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING)) |
9552e14d | 875 | goto out_iunlock; |
7a3be02b | 876 | |
01728b44 DC |
877 | /* |
878 | * Check for log shutdown because aborting the inode can move the log | |
879 | * tail and corrupt in memory state. This is fine if the log is shut | |
880 | * down, but if the log is still active and only the mount is shut down | |
881 | * then the in-memory log tail movement caused by the abort can be | |
882 | * incorrectly propagated to disk. | |
883 | */ | |
884 | if (xlog_is_shutdown(ip->i_mount->m_log)) { | |
777df5af | 885 | xfs_iunpin_wait(ip); |
d2d7c047 | 886 | xfs_iflush_shutdown_abort(ip); |
777df5af DC |
887 | goto reclaim; |
888 | } | |
617825fe | 889 | if (xfs_ipincount(ip)) |
718ecc50 | 890 | goto out_clear_flush; |
617825fe | 891 | if (!xfs_inode_clean(ip)) |
718ecc50 | 892 | goto out_clear_flush; |
8a48088f | 893 | |
718ecc50 | 894 | xfs_iflags_clear(ip, XFS_IFLUSHING); |
777df5af | 895 | reclaim: |
ab23a776 | 896 | trace_xfs_inode_reclaiming(ip); |
98efe8af | 897 | |
8a17d7dd DC |
898 | /* |
899 | * Because we use RCU freeing we need to ensure the inode always appears | |
900 | * to be reclaimed with an invalid inode number when in the free state. | |
98efe8af | 901 | * We do this as early as possible under the ILOCK so that |
f2e9ad21 OS |
902 | * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to |
903 | * detect races with us here. By doing this, we guarantee that once | |
904 | * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that | |
905 | * it will see either a valid inode that will serialise correctly, or it | |
906 | * will see an invalid inode that it can skip. | |
8a17d7dd DC |
907 | */ |
908 | spin_lock(&ip->i_flags_lock); | |
909 | ip->i_flags = XFS_IRECLAIM; | |
910 | ip->i_ino = 0; | |
255794c7 DW |
911 | ip->i_sick = 0; |
912 | ip->i_checked = 0; | |
8a17d7dd DC |
913 | spin_unlock(&ip->i_flags_lock); |
914 | ||
c8e20be0 | 915 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 916 | |
ff6d6af2 | 917 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
918 | /* |
919 | * Remove the inode from the per-AG radix tree. | |
920 | * | |
921 | * Because radix_tree_delete won't complain even if the item was never | |
922 | * added to the tree assert that it's been there before to catch | |
923 | * problems with the inode life time early on. | |
924 | */ | |
1a427ab0 | 925 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 926 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 927 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 928 | ASSERT(0); |
c076ae7a | 929 | xfs_perag_clear_inode_tag(pag, NULLAGINO, XFS_ICI_RECLAIM_TAG); |
1a427ab0 | 930 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
931 | |
932 | /* | |
933 | * Here we do an (almost) spurious inode lock in order to coordinate | |
934 | * with inode cache radix tree lookups. This is because the lookup | |
935 | * can reference the inodes in the cache without taking references. | |
936 | * | |
937 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 938 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 939 | */ |
ad637a10 | 940 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
3ea06d73 | 941 | ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot); |
ad637a10 | 942 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
96355d5a | 943 | ASSERT(xfs_inode_clean(ip)); |
2f11feab | 944 | |
8a17d7dd | 945 | __xfs_inode_free(ip); |
4d0bab3a | 946 | return; |
8a48088f | 947 | |
718ecc50 DC |
948 | out_clear_flush: |
949 | xfs_iflags_clear(ip, XFS_IFLUSHING); | |
9552e14d | 950 | out_iunlock: |
8a48088f | 951 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
9552e14d | 952 | out: |
617825fe | 953 | xfs_iflags_clear(ip, XFS_IRECLAIM); |
7a3be02b DC |
954 | } |
955 | ||
9492750a DW |
956 | /* Reclaim sick inodes if we're unmounting or the fs went down. */ |
957 | static inline bool | |
958 | xfs_want_reclaim_sick( | |
959 | struct xfs_mount *mp) | |
960 | { | |
2e973b2c | 961 | return xfs_is_unmounting(mp) || xfs_has_norecovery(mp) || |
75c8c50f | 962 | xfs_is_shutdown(mp); |
9492750a DW |
963 | } |
964 | ||
4d0bab3a | 965 | void |
7a3be02b | 966 | xfs_reclaim_inodes( |
4d0bab3a | 967 | struct xfs_mount *mp) |
7a3be02b | 968 | { |
b26b2bf1 DW |
969 | struct xfs_icwalk icw = { |
970 | .icw_flags = 0, | |
9492750a DW |
971 | }; |
972 | ||
973 | if (xfs_want_reclaim_sick(mp)) | |
b26b2bf1 | 974 | icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK; |
9492750a | 975 | |
4d0bab3a | 976 | while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { |
617825fe | 977 | xfs_ail_push_all_sync(mp->m_ail); |
b26b2bf1 | 978 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw); |
0f4ec0f1 | 979 | } |
9bf729c0 DC |
980 | } |
981 | ||
982 | /* | |
02511a5a DC |
983 | * The shrinker infrastructure determines how many inodes we should scan for |
984 | * reclaim. We want as many clean inodes ready to reclaim as possible, so we | |
985 | * push the AIL here. We also want to proactively free up memory if we can to | |
986 | * minimise the amount of work memory reclaim has to do so we kick the | |
987 | * background reclaim if it isn't already scheduled. | |
9bf729c0 | 988 | */ |
0a234c6d | 989 | long |
8daaa831 DC |
990 | xfs_reclaim_inodes_nr( |
991 | struct xfs_mount *mp, | |
10be350b | 992 | unsigned long nr_to_scan) |
9bf729c0 | 993 | { |
b26b2bf1 DW |
994 | struct xfs_icwalk icw = { |
995 | .icw_flags = XFS_ICWALK_FLAG_SCAN_LIMIT, | |
10be350b | 996 | .icw_scan_limit = min_t(unsigned long, LONG_MAX, nr_to_scan), |
f1bc5c56 DW |
997 | }; |
998 | ||
9492750a | 999 | if (xfs_want_reclaim_sick(mp)) |
b26b2bf1 | 1000 | icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK; |
9492750a | 1001 | |
8daaa831 | 1002 | /* kick background reclaimer and push the AIL */ |
5889608d | 1003 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1004 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1005 | |
b26b2bf1 | 1006 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw); |
617825fe | 1007 | return 0; |
8daaa831 | 1008 | } |
9bf729c0 | 1009 | |
8daaa831 DC |
1010 | /* |
1011 | * Return the number of reclaimable inodes in the filesystem for | |
1012 | * the shrinker to determine how much to reclaim. | |
1013 | */ | |
10be350b | 1014 | long |
8daaa831 DC |
1015 | xfs_reclaim_inodes_count( |
1016 | struct xfs_mount *mp) | |
1017 | { | |
1018 | struct xfs_perag *pag; | |
1019 | xfs_agnumber_t ag = 0; | |
10be350b | 1020 | long reclaimable = 0; |
9bf729c0 | 1021 | |
65d0f205 DC |
1022 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1023 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1024 | reclaimable += pag->pag_ici_reclaimable; |
1025 | xfs_perag_put(pag); | |
9bf729c0 | 1026 | } |
9bf729c0 DC |
1027 | return reclaimable; |
1028 | } | |
1029 | ||
39b1cfd7 | 1030 | STATIC bool |
b26b2bf1 | 1031 | xfs_icwalk_match_id( |
3e3f9f58 | 1032 | struct xfs_inode *ip, |
b26b2bf1 | 1033 | struct xfs_icwalk *icw) |
3e3f9f58 | 1034 | { |
b26b2bf1 DW |
1035 | if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) && |
1036 | !uid_eq(VFS_I(ip)->i_uid, icw->icw_uid)) | |
39b1cfd7 | 1037 | return false; |
3e3f9f58 | 1038 | |
b26b2bf1 DW |
1039 | if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) && |
1040 | !gid_eq(VFS_I(ip)->i_gid, icw->icw_gid)) | |
39b1cfd7 | 1041 | return false; |
1b556048 | 1042 | |
b26b2bf1 DW |
1043 | if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) && |
1044 | ip->i_projid != icw->icw_prid) | |
39b1cfd7 | 1045 | return false; |
1b556048 | 1046 | |
39b1cfd7 | 1047 | return true; |
3e3f9f58 BF |
1048 | } |
1049 | ||
f4526397 BF |
1050 | /* |
1051 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1052 | * criteria match. This is for global/internal scans only. | |
1053 | */ | |
39b1cfd7 | 1054 | STATIC bool |
b26b2bf1 | 1055 | xfs_icwalk_match_id_union( |
f4526397 | 1056 | struct xfs_inode *ip, |
b26b2bf1 | 1057 | struct xfs_icwalk *icw) |
f4526397 | 1058 | { |
b26b2bf1 DW |
1059 | if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) && |
1060 | uid_eq(VFS_I(ip)->i_uid, icw->icw_uid)) | |
39b1cfd7 | 1061 | return true; |
f4526397 | 1062 | |
b26b2bf1 DW |
1063 | if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) && |
1064 | gid_eq(VFS_I(ip)->i_gid, icw->icw_gid)) | |
39b1cfd7 | 1065 | return true; |
f4526397 | 1066 | |
b26b2bf1 DW |
1067 | if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) && |
1068 | ip->i_projid == icw->icw_prid) | |
39b1cfd7 | 1069 | return true; |
f4526397 | 1070 | |
39b1cfd7 | 1071 | return false; |
f4526397 BF |
1072 | } |
1073 | ||
a91bf992 DW |
1074 | /* |
1075 | * Is this inode @ip eligible for eof/cow block reclamation, given some | |
b26b2bf1 | 1076 | * filtering parameters @icw? The inode is eligible if @icw is null or |
a91bf992 DW |
1077 | * if the predicate functions match. |
1078 | */ | |
1079 | static bool | |
b26b2bf1 | 1080 | xfs_icwalk_match( |
a91bf992 | 1081 | struct xfs_inode *ip, |
b26b2bf1 | 1082 | struct xfs_icwalk *icw) |
a91bf992 | 1083 | { |
39b1cfd7 | 1084 | bool match; |
a91bf992 | 1085 | |
b26b2bf1 | 1086 | if (!icw) |
a91bf992 DW |
1087 | return true; |
1088 | ||
b26b2bf1 DW |
1089 | if (icw->icw_flags & XFS_ICWALK_FLAG_UNION) |
1090 | match = xfs_icwalk_match_id_union(ip, icw); | |
a91bf992 | 1091 | else |
b26b2bf1 | 1092 | match = xfs_icwalk_match_id(ip, icw); |
a91bf992 DW |
1093 | if (!match) |
1094 | return false; | |
1095 | ||
1096 | /* skip the inode if the file size is too small */ | |
b26b2bf1 DW |
1097 | if ((icw->icw_flags & XFS_ICWALK_FLAG_MINFILESIZE) && |
1098 | XFS_ISIZE(ip) < icw->icw_min_file_size) | |
a91bf992 DW |
1099 | return false; |
1100 | ||
1101 | return true; | |
1102 | } | |
1103 | ||
4d0bab3a DC |
1104 | /* |
1105 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
1106 | * many inodes as possible in a short period of time. It kicks itself every few | |
1107 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
02511a5a | 1108 | * goes low. |
4d0bab3a DC |
1109 | */ |
1110 | void | |
1111 | xfs_reclaim_worker( | |
1112 | struct work_struct *work) | |
1113 | { | |
1114 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
1115 | struct xfs_mount, m_reclaim_work); | |
4d0bab3a | 1116 | |
f1bc5c56 | 1117 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, NULL); |
4d0bab3a DC |
1118 | xfs_reclaim_work_queue(mp); |
1119 | } | |
1120 | ||
41176a68 BF |
1121 | STATIC int |
1122 | xfs_inode_free_eofblocks( | |
1123 | struct xfs_inode *ip, | |
b26b2bf1 | 1124 | struct xfs_icwalk *icw, |
0fa4a10a | 1125 | unsigned int *lockflags) |
41176a68 | 1126 | { |
390600f8 | 1127 | bool wait; |
390600f8 | 1128 | |
b26b2bf1 | 1129 | wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC); |
5400da7d | 1130 | |
ce2d3bbe DW |
1131 | if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS)) |
1132 | return 0; | |
1133 | ||
41176a68 BF |
1134 | /* |
1135 | * If the mapping is dirty the operation can block and wait for some | |
1136 | * time. Unless we are waiting, skip it. | |
1137 | */ | |
390600f8 | 1138 | if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) |
41176a68 BF |
1139 | return 0; |
1140 | ||
b26b2bf1 | 1141 | if (!xfs_icwalk_match(ip, icw)) |
a91bf992 | 1142 | return 0; |
3e3f9f58 | 1143 | |
a36b9261 BF |
1144 | /* |
1145 | * If the caller is waiting, return -EAGAIN to keep the background | |
1146 | * scanner moving and revisit the inode in a subsequent pass. | |
1147 | */ | |
c3155097 | 1148 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
390600f8 DW |
1149 | if (wait) |
1150 | return -EAGAIN; | |
1151 | return 0; | |
a36b9261 | 1152 | } |
0fa4a10a | 1153 | *lockflags |= XFS_IOLOCK_EXCL; |
390600f8 | 1154 | |
2b156ff8 DW |
1155 | if (xfs_can_free_eofblocks(ip, false)) |
1156 | return xfs_free_eofblocks(ip); | |
1157 | ||
1158 | /* inode could be preallocated or append-only */ | |
1159 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1160 | xfs_inode_clear_eofblocks_tag(ip); | |
1161 | return 0; | |
41176a68 BF |
1162 | } |
1163 | ||
83104d44 | 1164 | static void |
ce2d3bbe DW |
1165 | xfs_blockgc_set_iflag( |
1166 | struct xfs_inode *ip, | |
ce2d3bbe | 1167 | unsigned long iflag) |
27b52867 | 1168 | { |
ce2d3bbe DW |
1169 | struct xfs_mount *mp = ip->i_mount; |
1170 | struct xfs_perag *pag; | |
ce2d3bbe DW |
1171 | |
1172 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1173 | |
85a6e764 CH |
1174 | /* |
1175 | * Don't bother locking the AG and looking up in the radix trees | |
1176 | * if we already know that we have the tag set. | |
1177 | */ | |
ce2d3bbe | 1178 | if (ip->i_flags & iflag) |
85a6e764 CH |
1179 | return; |
1180 | spin_lock(&ip->i_flags_lock); | |
ce2d3bbe | 1181 | ip->i_flags |= iflag; |
85a6e764 CH |
1182 | spin_unlock(&ip->i_flags_lock); |
1183 | ||
27b52867 BF |
1184 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1185 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1186 | |
c076ae7a DW |
1187 | xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1188 | XFS_ICI_BLOCKGC_TAG); | |
27b52867 BF |
1189 | |
1190 | spin_unlock(&pag->pag_ici_lock); | |
1191 | xfs_perag_put(pag); | |
1192 | } | |
1193 | ||
1194 | void | |
83104d44 | 1195 | xfs_inode_set_eofblocks_tag( |
27b52867 | 1196 | xfs_inode_t *ip) |
83104d44 DW |
1197 | { |
1198 | trace_xfs_inode_set_eofblocks_tag(ip); | |
9669f51d | 1199 | return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1200 | } |
1201 | ||
1202 | static void | |
ce2d3bbe DW |
1203 | xfs_blockgc_clear_iflag( |
1204 | struct xfs_inode *ip, | |
1205 | unsigned long iflag) | |
27b52867 | 1206 | { |
ce2d3bbe DW |
1207 | struct xfs_mount *mp = ip->i_mount; |
1208 | struct xfs_perag *pag; | |
1209 | bool clear_tag; | |
1210 | ||
1211 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1212 | |
85a6e764 | 1213 | spin_lock(&ip->i_flags_lock); |
ce2d3bbe DW |
1214 | ip->i_flags &= ~iflag; |
1215 | clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0; | |
85a6e764 CH |
1216 | spin_unlock(&ip->i_flags_lock); |
1217 | ||
ce2d3bbe DW |
1218 | if (!clear_tag) |
1219 | return; | |
1220 | ||
27b52867 BF |
1221 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1222 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1223 | |
c076ae7a DW |
1224 | xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1225 | XFS_ICI_BLOCKGC_TAG); | |
27b52867 BF |
1226 | |
1227 | spin_unlock(&pag->pag_ici_lock); | |
1228 | xfs_perag_put(pag); | |
1229 | } | |
1230 | ||
83104d44 DW |
1231 | void |
1232 | xfs_inode_clear_eofblocks_tag( | |
1233 | xfs_inode_t *ip) | |
1234 | { | |
1235 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
ce2d3bbe | 1236 | return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1237 | } |
1238 | ||
1239 | /* | |
be78ff0e DW |
1240 | * Set ourselves up to free CoW blocks from this file. If it's already clean |
1241 | * then we can bail out quickly, but otherwise we must back off if the file | |
1242 | * is undergoing some kind of write. | |
83104d44 | 1243 | */ |
be78ff0e DW |
1244 | static bool |
1245 | xfs_prep_free_cowblocks( | |
51d62690 | 1246 | struct xfs_inode *ip) |
83104d44 | 1247 | { |
39937234 BF |
1248 | /* |
1249 | * Just clear the tag if we have an empty cow fork or none at all. It's | |
1250 | * possible the inode was fully unshared since it was originally tagged. | |
1251 | */ | |
51d62690 | 1252 | if (!xfs_inode_has_cow_data(ip)) { |
83104d44 DW |
1253 | trace_xfs_inode_free_cowblocks_invalid(ip); |
1254 | xfs_inode_clear_cowblocks_tag(ip); | |
be78ff0e | 1255 | return false; |
83104d44 DW |
1256 | } |
1257 | ||
1258 | /* | |
1259 | * If the mapping is dirty or under writeback we cannot touch the | |
1260 | * CoW fork. Leave it alone if we're in the midst of a directio. | |
1261 | */ | |
a1b7a4de CH |
1262 | if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) || |
1263 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) || | |
83104d44 DW |
1264 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) || |
1265 | atomic_read(&VFS_I(ip)->i_dio_count)) | |
be78ff0e DW |
1266 | return false; |
1267 | ||
1268 | return true; | |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * Automatic CoW Reservation Freeing | |
1273 | * | |
1274 | * These functions automatically garbage collect leftover CoW reservations | |
1275 | * that were made on behalf of a cowextsize hint when we start to run out | |
1276 | * of quota or when the reservations sit around for too long. If the file | |
1277 | * has dirty pages or is undergoing writeback, its CoW reservations will | |
1278 | * be retained. | |
1279 | * | |
1280 | * The actual garbage collection piggybacks off the same code that runs | |
1281 | * the speculative EOF preallocation garbage collector. | |
1282 | */ | |
1283 | STATIC int | |
1284 | xfs_inode_free_cowblocks( | |
1285 | struct xfs_inode *ip, | |
b26b2bf1 | 1286 | struct xfs_icwalk *icw, |
0fa4a10a | 1287 | unsigned int *lockflags) |
be78ff0e | 1288 | { |
f41a0716 | 1289 | bool wait; |
be78ff0e DW |
1290 | int ret = 0; |
1291 | ||
b26b2bf1 | 1292 | wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC); |
f41a0716 | 1293 | |
ce2d3bbe DW |
1294 | if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS)) |
1295 | return 0; | |
1296 | ||
51d62690 | 1297 | if (!xfs_prep_free_cowblocks(ip)) |
83104d44 DW |
1298 | return 0; |
1299 | ||
b26b2bf1 | 1300 | if (!xfs_icwalk_match(ip, icw)) |
a91bf992 | 1301 | return 0; |
83104d44 | 1302 | |
f41a0716 DW |
1303 | /* |
1304 | * If the caller is waiting, return -EAGAIN to keep the background | |
1305 | * scanner moving and revisit the inode in a subsequent pass. | |
1306 | */ | |
0fa4a10a DW |
1307 | if (!(*lockflags & XFS_IOLOCK_EXCL) && |
1308 | !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
f41a0716 DW |
1309 | if (wait) |
1310 | return -EAGAIN; | |
1311 | return 0; | |
1312 | } | |
0fa4a10a DW |
1313 | *lockflags |= XFS_IOLOCK_EXCL; |
1314 | ||
f41a0716 DW |
1315 | if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) { |
1316 | if (wait) | |
0fa4a10a DW |
1317 | return -EAGAIN; |
1318 | return 0; | |
f41a0716 | 1319 | } |
0fa4a10a | 1320 | *lockflags |= XFS_MMAPLOCK_EXCL; |
83104d44 | 1321 | |
be78ff0e DW |
1322 | /* |
1323 | * Check again, nobody else should be able to dirty blocks or change | |
1324 | * the reflink iflag now that we have the first two locks held. | |
1325 | */ | |
51d62690 | 1326 | if (xfs_prep_free_cowblocks(ip)) |
be78ff0e | 1327 | ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false); |
83104d44 DW |
1328 | return ret; |
1329 | } | |
1330 | ||
83104d44 DW |
1331 | void |
1332 | xfs_inode_set_cowblocks_tag( | |
1333 | xfs_inode_t *ip) | |
1334 | { | |
7b7381f0 | 1335 | trace_xfs_inode_set_cowblocks_tag(ip); |
9669f51d | 1336 | return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 DW |
1337 | } |
1338 | ||
1339 | void | |
1340 | xfs_inode_clear_cowblocks_tag( | |
1341 | xfs_inode_t *ip) | |
1342 | { | |
7b7381f0 | 1343 | trace_xfs_inode_clear_cowblocks_tag(ip); |
ce2d3bbe | 1344 | return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 | 1345 | } |
d6b636eb DW |
1346 | |
1347 | /* Disable post-EOF and CoW block auto-reclamation. */ | |
1348 | void | |
c9a6526f | 1349 | xfs_blockgc_stop( |
d6b636eb DW |
1350 | struct xfs_mount *mp) |
1351 | { | |
894ecacf DW |
1352 | struct xfs_perag *pag; |
1353 | xfs_agnumber_t agno; | |
1354 | ||
6f649091 DW |
1355 | if (!xfs_clear_blockgc_enabled(mp)) |
1356 | return; | |
1357 | ||
1358 | for_each_perag(mp, agno, pag) | |
894ecacf | 1359 | cancel_delayed_work_sync(&pag->pag_blockgc_work); |
6f649091 | 1360 | trace_xfs_blockgc_stop(mp, __return_address); |
d6b636eb DW |
1361 | } |
1362 | ||
1363 | /* Enable post-EOF and CoW block auto-reclamation. */ | |
1364 | void | |
c9a6526f | 1365 | xfs_blockgc_start( |
d6b636eb DW |
1366 | struct xfs_mount *mp) |
1367 | { | |
894ecacf DW |
1368 | struct xfs_perag *pag; |
1369 | xfs_agnumber_t agno; | |
1370 | ||
6f649091 DW |
1371 | if (xfs_set_blockgc_enabled(mp)) |
1372 | return; | |
1373 | ||
1374 | trace_xfs_blockgc_start(mp, __return_address); | |
894ecacf DW |
1375 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) |
1376 | xfs_blockgc_queue(pag); | |
d6b636eb | 1377 | } |
3d4feec0 | 1378 | |
d20d5edc DW |
1379 | /* Don't try to run block gc on an inode that's in any of these states. */ |
1380 | #define XFS_BLOCKGC_NOGRAB_IFLAGS (XFS_INEW | \ | |
ab23a776 DC |
1381 | XFS_NEED_INACTIVE | \ |
1382 | XFS_INACTIVATING | \ | |
d20d5edc DW |
1383 | XFS_IRECLAIMABLE | \ |
1384 | XFS_IRECLAIM) | |
df600197 | 1385 | /* |
b9baaef4 DW |
1386 | * Decide if the given @ip is eligible for garbage collection of speculative |
1387 | * preallocations, and grab it if so. Returns true if it's ready to go or | |
1388 | * false if we should just ignore it. | |
df600197 DW |
1389 | */ |
1390 | static bool | |
b9baaef4 | 1391 | xfs_blockgc_igrab( |
7fdff526 | 1392 | struct xfs_inode *ip) |
df600197 DW |
1393 | { |
1394 | struct inode *inode = VFS_I(ip); | |
df600197 DW |
1395 | |
1396 | ASSERT(rcu_read_lock_held()); | |
1397 | ||
1398 | /* Check for stale RCU freed inode */ | |
1399 | spin_lock(&ip->i_flags_lock); | |
1400 | if (!ip->i_ino) | |
1401 | goto out_unlock_noent; | |
1402 | ||
d20d5edc | 1403 | if (ip->i_flags & XFS_BLOCKGC_NOGRAB_IFLAGS) |
df600197 DW |
1404 | goto out_unlock_noent; |
1405 | spin_unlock(&ip->i_flags_lock); | |
1406 | ||
1407 | /* nothing to sync during shutdown */ | |
75c8c50f | 1408 | if (xfs_is_shutdown(ip->i_mount)) |
df600197 DW |
1409 | return false; |
1410 | ||
1411 | /* If we can't grab the inode, it must on it's way to reclaim. */ | |
1412 | if (!igrab(inode)) | |
1413 | return false; | |
1414 | ||
1415 | /* inode is valid */ | |
1416 | return true; | |
1417 | ||
1418 | out_unlock_noent: | |
1419 | spin_unlock(&ip->i_flags_lock); | |
1420 | return false; | |
1421 | } | |
1422 | ||
41956753 DW |
1423 | /* Scan one incore inode for block preallocations that we can remove. */ |
1424 | static int | |
1425 | xfs_blockgc_scan_inode( | |
1426 | struct xfs_inode *ip, | |
b26b2bf1 | 1427 | struct xfs_icwalk *icw) |
85c5b270 | 1428 | { |
0fa4a10a | 1429 | unsigned int lockflags = 0; |
85c5b270 DW |
1430 | int error; |
1431 | ||
b26b2bf1 | 1432 | error = xfs_inode_free_eofblocks(ip, icw, &lockflags); |
85c5b270 | 1433 | if (error) |
0fa4a10a | 1434 | goto unlock; |
85c5b270 | 1435 | |
b26b2bf1 | 1436 | error = xfs_inode_free_cowblocks(ip, icw, &lockflags); |
0fa4a10a DW |
1437 | unlock: |
1438 | if (lockflags) | |
1439 | xfs_iunlock(ip, lockflags); | |
594ab00b | 1440 | xfs_irele(ip); |
0fa4a10a | 1441 | return error; |
85c5b270 DW |
1442 | } |
1443 | ||
9669f51d DW |
1444 | /* Background worker that trims preallocated space. */ |
1445 | void | |
1446 | xfs_blockgc_worker( | |
1447 | struct work_struct *work) | |
1448 | { | |
894ecacf DW |
1449 | struct xfs_perag *pag = container_of(to_delayed_work(work), |
1450 | struct xfs_perag, pag_blockgc_work); | |
1451 | struct xfs_mount *mp = pag->pag_mount; | |
9669f51d DW |
1452 | int error; |
1453 | ||
6f649091 DW |
1454 | trace_xfs_blockgc_worker(mp, __return_address); |
1455 | ||
f427cf5c | 1456 | error = xfs_icwalk_ag(pag, XFS_ICWALK_BLOCKGC, NULL); |
9669f51d | 1457 | if (error) |
894ecacf DW |
1458 | xfs_info(mp, "AG %u preallocation gc worker failed, err=%d", |
1459 | pag->pag_agno, error); | |
894ecacf | 1460 | xfs_blockgc_queue(pag); |
9669f51d DW |
1461 | } |
1462 | ||
85c5b270 | 1463 | /* |
2eb66502 DW |
1464 | * Try to free space in the filesystem by purging inactive inodes, eofblocks |
1465 | * and cowblocks. | |
85c5b270 DW |
1466 | */ |
1467 | int | |
1468 | xfs_blockgc_free_space( | |
1469 | struct xfs_mount *mp, | |
b26b2bf1 | 1470 | struct xfs_icwalk *icw) |
85c5b270 | 1471 | { |
2eb66502 DW |
1472 | int error; |
1473 | ||
b26b2bf1 | 1474 | trace_xfs_blockgc_free_space(mp, icw, _RET_IP_); |
85c5b270 | 1475 | |
2eb66502 DW |
1476 | error = xfs_icwalk(mp, XFS_ICWALK_BLOCKGC, icw); |
1477 | if (error) | |
1478 | return error; | |
1479 | ||
1480 | xfs_inodegc_flush(mp); | |
1481 | return 0; | |
85c5b270 DW |
1482 | } |
1483 | ||
e8d04c2a DW |
1484 | /* |
1485 | * Reclaim all the free space that we can by scheduling the background blockgc | |
1486 | * and inodegc workers immediately and waiting for them all to clear. | |
1487 | */ | |
1488 | void | |
1489 | xfs_blockgc_flush_all( | |
1490 | struct xfs_mount *mp) | |
1491 | { | |
1492 | struct xfs_perag *pag; | |
1493 | xfs_agnumber_t agno; | |
1494 | ||
1495 | trace_xfs_blockgc_flush_all(mp, __return_address); | |
1496 | ||
1497 | /* | |
1498 | * For each blockgc worker, move its queue time up to now. If it | |
1499 | * wasn't queued, it will not be requeued. Then flush whatever's | |
1500 | * left. | |
1501 | */ | |
1502 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1503 | mod_delayed_work(pag->pag_mount->m_blockgc_wq, | |
1504 | &pag->pag_blockgc_work, 0); | |
1505 | ||
1506 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1507 | flush_delayed_work(&pag->pag_blockgc_work); | |
1508 | ||
1509 | xfs_inodegc_flush(mp); | |
1510 | } | |
1511 | ||
3d4feec0 | 1512 | /* |
c237dd7c DW |
1513 | * Run cow/eofblocks scans on the supplied dquots. We don't know exactly which |
1514 | * quota caused an allocation failure, so we make a best effort by including | |
1515 | * each quota under low free space conditions (less than 1% free space) in the | |
1516 | * scan. | |
111068f8 DW |
1517 | * |
1518 | * Callers must not hold any inode's ILOCK. If requesting a synchronous scan | |
2d53f66b | 1519 | * (XFS_ICWALK_FLAG_SYNC), the caller also must not hold any inode's IOLOCK or |
111068f8 | 1520 | * MMAPLOCK. |
3d4feec0 | 1521 | */ |
111068f8 | 1522 | int |
c237dd7c DW |
1523 | xfs_blockgc_free_dquots( |
1524 | struct xfs_mount *mp, | |
1525 | struct xfs_dquot *udqp, | |
1526 | struct xfs_dquot *gdqp, | |
1527 | struct xfs_dquot *pdqp, | |
2d53f66b | 1528 | unsigned int iwalk_flags) |
3d4feec0 | 1529 | { |
b26b2bf1 | 1530 | struct xfs_icwalk icw = {0}; |
3d4feec0 DW |
1531 | bool do_work = false; |
1532 | ||
c237dd7c DW |
1533 | if (!udqp && !gdqp && !pdqp) |
1534 | return 0; | |
1535 | ||
3d4feec0 | 1536 | /* |
111068f8 DW |
1537 | * Run a scan to free blocks using the union filter to cover all |
1538 | * applicable quotas in a single scan. | |
3d4feec0 | 1539 | */ |
b26b2bf1 | 1540 | icw.icw_flags = XFS_ICWALK_FLAG_UNION | iwalk_flags; |
3d4feec0 | 1541 | |
c237dd7c | 1542 | if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) { |
b26b2bf1 DW |
1543 | icw.icw_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id); |
1544 | icw.icw_flags |= XFS_ICWALK_FLAG_UID; | |
c237dd7c | 1545 | do_work = true; |
3d4feec0 DW |
1546 | } |
1547 | ||
c237dd7c | 1548 | if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) { |
b26b2bf1 DW |
1549 | icw.icw_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id); |
1550 | icw.icw_flags |= XFS_ICWALK_FLAG_GID; | |
c237dd7c | 1551 | do_work = true; |
3d4feec0 DW |
1552 | } |
1553 | ||
c237dd7c | 1554 | if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) { |
b26b2bf1 DW |
1555 | icw.icw_prid = pdqp->q_id; |
1556 | icw.icw_flags |= XFS_ICWALK_FLAG_PRID; | |
c237dd7c | 1557 | do_work = true; |
3d4feec0 DW |
1558 | } |
1559 | ||
1560 | if (!do_work) | |
111068f8 | 1561 | return 0; |
3d4feec0 | 1562 | |
b26b2bf1 | 1563 | return xfs_blockgc_free_space(mp, &icw); |
c237dd7c DW |
1564 | } |
1565 | ||
1566 | /* Run cow/eofblocks scans on the quotas attached to the inode. */ | |
1567 | int | |
1568 | xfs_blockgc_free_quota( | |
1569 | struct xfs_inode *ip, | |
2d53f66b | 1570 | unsigned int iwalk_flags) |
c237dd7c DW |
1571 | { |
1572 | return xfs_blockgc_free_dquots(ip->i_mount, | |
1573 | xfs_inode_dquot(ip, XFS_DQTYPE_USER), | |
1574 | xfs_inode_dquot(ip, XFS_DQTYPE_GROUP), | |
2d53f66b | 1575 | xfs_inode_dquot(ip, XFS_DQTYPE_PROJ), iwalk_flags); |
3d4feec0 | 1576 | } |
df600197 DW |
1577 | |
1578 | /* XFS Inode Cache Walking Code */ | |
1579 | ||
f1bc5c56 DW |
1580 | /* |
1581 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
1582 | * radix tree lookups to a minimum. The batch size is a trade off between | |
1583 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
1584 | * be too greedy. | |
1585 | */ | |
1586 | #define XFS_LOOKUP_BATCH 32 | |
1587 | ||
1588 | ||
b9baaef4 DW |
1589 | /* |
1590 | * Decide if we want to grab this inode in anticipation of doing work towards | |
594ab00b | 1591 | * the goal. |
b9baaef4 DW |
1592 | */ |
1593 | static inline bool | |
1594 | xfs_icwalk_igrab( | |
1595 | enum xfs_icwalk_goal goal, | |
9492750a | 1596 | struct xfs_inode *ip, |
b26b2bf1 | 1597 | struct xfs_icwalk *icw) |
b9baaef4 DW |
1598 | { |
1599 | switch (goal) { | |
b9baaef4 | 1600 | case XFS_ICWALK_BLOCKGC: |
7fdff526 | 1601 | return xfs_blockgc_igrab(ip); |
f1bc5c56 | 1602 | case XFS_ICWALK_RECLAIM: |
b26b2bf1 | 1603 | return xfs_reclaim_igrab(ip, icw); |
b9baaef4 DW |
1604 | default: |
1605 | return false; | |
1606 | } | |
1607 | } | |
1608 | ||
594ab00b DW |
1609 | /* |
1610 | * Process an inode. Each processing function must handle any state changes | |
1611 | * made by the icwalk igrab function. Return -EAGAIN to skip an inode. | |
1612 | */ | |
f427cf5c DW |
1613 | static inline int |
1614 | xfs_icwalk_process_inode( | |
1615 | enum xfs_icwalk_goal goal, | |
1616 | struct xfs_inode *ip, | |
f1bc5c56 | 1617 | struct xfs_perag *pag, |
b26b2bf1 | 1618 | struct xfs_icwalk *icw) |
f427cf5c | 1619 | { |
594ab00b | 1620 | int error = 0; |
f427cf5c DW |
1621 | |
1622 | switch (goal) { | |
f427cf5c | 1623 | case XFS_ICWALK_BLOCKGC: |
b26b2bf1 | 1624 | error = xfs_blockgc_scan_inode(ip, icw); |
f427cf5c | 1625 | break; |
f1bc5c56 DW |
1626 | case XFS_ICWALK_RECLAIM: |
1627 | xfs_reclaim_inode(ip, pag); | |
1628 | break; | |
f427cf5c | 1629 | } |
f427cf5c DW |
1630 | return error; |
1631 | } | |
1632 | ||
df600197 | 1633 | /* |
f427cf5c DW |
1634 | * For a given per-AG structure @pag and a goal, grab qualifying inodes and |
1635 | * process them in some manner. | |
df600197 DW |
1636 | */ |
1637 | static int | |
c1115c0c | 1638 | xfs_icwalk_ag( |
df600197 | 1639 | struct xfs_perag *pag, |
f427cf5c | 1640 | enum xfs_icwalk_goal goal, |
b26b2bf1 | 1641 | struct xfs_icwalk *icw) |
df600197 DW |
1642 | { |
1643 | struct xfs_mount *mp = pag->pag_mount; | |
1644 | uint32_t first_index; | |
1645 | int last_error = 0; | |
1646 | int skipped; | |
1647 | bool done; | |
1648 | int nr_found; | |
1649 | ||
1650 | restart: | |
1651 | done = false; | |
1652 | skipped = 0; | |
f1bc5c56 DW |
1653 | if (goal == XFS_ICWALK_RECLAIM) |
1654 | first_index = READ_ONCE(pag->pag_ici_reclaim_cursor); | |
1655 | else | |
1656 | first_index = 0; | |
df600197 DW |
1657 | nr_found = 0; |
1658 | do { | |
1659 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; | |
1660 | int error = 0; | |
1661 | int i; | |
1662 | ||
1663 | rcu_read_lock(); | |
1664 | ||
a437b9b4 CH |
1665 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, |
1666 | (void **) batch, first_index, | |
1667 | XFS_LOOKUP_BATCH, goal); | |
df600197 | 1668 | if (!nr_found) { |
f1bc5c56 | 1669 | done = true; |
df600197 DW |
1670 | rcu_read_unlock(); |
1671 | break; | |
1672 | } | |
1673 | ||
1674 | /* | |
1675 | * Grab the inodes before we drop the lock. if we found | |
1676 | * nothing, nr == 0 and the loop will be skipped. | |
1677 | */ | |
1678 | for (i = 0; i < nr_found; i++) { | |
1679 | struct xfs_inode *ip = batch[i]; | |
1680 | ||
b26b2bf1 | 1681 | if (done || !xfs_icwalk_igrab(goal, ip, icw)) |
df600197 DW |
1682 | batch[i] = NULL; |
1683 | ||
1684 | /* | |
1685 | * Update the index for the next lookup. Catch | |
1686 | * overflows into the next AG range which can occur if | |
1687 | * we have inodes in the last block of the AG and we | |
1688 | * are currently pointing to the last inode. | |
1689 | * | |
1690 | * Because we may see inodes that are from the wrong AG | |
1691 | * due to RCU freeing and reallocation, only update the | |
1692 | * index if it lies in this AG. It was a race that lead | |
1693 | * us to see this inode, so another lookup from the | |
1694 | * same index will not find it again. | |
1695 | */ | |
1696 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) | |
1697 | continue; | |
1698 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
1699 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1700 | done = true; | |
1701 | } | |
1702 | ||
1703 | /* unlock now we've grabbed the inodes. */ | |
1704 | rcu_read_unlock(); | |
1705 | ||
1706 | for (i = 0; i < nr_found; i++) { | |
1707 | if (!batch[i]) | |
1708 | continue; | |
f1bc5c56 | 1709 | error = xfs_icwalk_process_inode(goal, batch[i], pag, |
b26b2bf1 | 1710 | icw); |
df600197 DW |
1711 | if (error == -EAGAIN) { |
1712 | skipped++; | |
1713 | continue; | |
1714 | } | |
1715 | if (error && last_error != -EFSCORRUPTED) | |
1716 | last_error = error; | |
1717 | } | |
1718 | ||
1719 | /* bail out if the filesystem is corrupted. */ | |
1720 | if (error == -EFSCORRUPTED) | |
1721 | break; | |
1722 | ||
1723 | cond_resched(); | |
1724 | ||
b26b2bf1 DW |
1725 | if (icw && (icw->icw_flags & XFS_ICWALK_FLAG_SCAN_LIMIT)) { |
1726 | icw->icw_scan_limit -= XFS_LOOKUP_BATCH; | |
1727 | if (icw->icw_scan_limit <= 0) | |
f1bc5c56 DW |
1728 | break; |
1729 | } | |
df600197 DW |
1730 | } while (nr_found && !done); |
1731 | ||
f1bc5c56 DW |
1732 | if (goal == XFS_ICWALK_RECLAIM) { |
1733 | if (done) | |
1734 | first_index = 0; | |
1735 | WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index); | |
1736 | } | |
1737 | ||
df600197 DW |
1738 | if (skipped) { |
1739 | delay(1); | |
1740 | goto restart; | |
1741 | } | |
1742 | return last_error; | |
1743 | } | |
1744 | ||
f427cf5c | 1745 | /* Walk all incore inodes to achieve a given goal. */ |
df600197 | 1746 | static int |
c1115c0c | 1747 | xfs_icwalk( |
df600197 | 1748 | struct xfs_mount *mp, |
f427cf5c | 1749 | enum xfs_icwalk_goal goal, |
b26b2bf1 | 1750 | struct xfs_icwalk *icw) |
df600197 DW |
1751 | { |
1752 | struct xfs_perag *pag; | |
1753 | int error = 0; | |
1754 | int last_error = 0; | |
a437b9b4 | 1755 | xfs_agnumber_t agno; |
df600197 | 1756 | |
a437b9b4 | 1757 | for_each_perag_tag(mp, agno, pag, goal) { |
b26b2bf1 | 1758 | error = xfs_icwalk_ag(pag, goal, icw); |
df600197 DW |
1759 | if (error) { |
1760 | last_error = error; | |
a437b9b4 CH |
1761 | if (error == -EFSCORRUPTED) { |
1762 | xfs_perag_put(pag); | |
df600197 | 1763 | break; |
a437b9b4 | 1764 | } |
df600197 DW |
1765 | } |
1766 | } | |
1767 | return last_error; | |
2d53f66b | 1768 | BUILD_BUG_ON(XFS_ICWALK_PRIVATE_FLAGS & XFS_ICWALK_FLAGS_VALID); |
df600197 | 1769 | } |
c6c2066d DW |
1770 | |
1771 | #ifdef DEBUG | |
1772 | static void | |
1773 | xfs_check_delalloc( | |
1774 | struct xfs_inode *ip, | |
1775 | int whichfork) | |
1776 | { | |
1777 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
1778 | struct xfs_bmbt_irec got; | |
1779 | struct xfs_iext_cursor icur; | |
1780 | ||
1781 | if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got)) | |
1782 | return; | |
1783 | do { | |
1784 | if (isnullstartblock(got.br_startblock)) { | |
1785 | xfs_warn(ip->i_mount, | |
1786 | "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]", | |
1787 | ip->i_ino, | |
1788 | whichfork == XFS_DATA_FORK ? "data" : "cow", | |
1789 | got.br_startoff, got.br_blockcount); | |
1790 | } | |
1791 | } while (xfs_iext_next_extent(ifp, &icur, &got)); | |
1792 | } | |
1793 | #else | |
1794 | #define xfs_check_delalloc(ip, whichfork) do { } while (0) | |
1795 | #endif | |
1796 | ||
ab23a776 DC |
1797 | /* Schedule the inode for reclaim. */ |
1798 | static void | |
1799 | xfs_inodegc_set_reclaimable( | |
c6c2066d DW |
1800 | struct xfs_inode *ip) |
1801 | { | |
1802 | struct xfs_mount *mp = ip->i_mount; | |
1803 | struct xfs_perag *pag; | |
c6c2066d | 1804 | |
75c8c50f | 1805 | if (!xfs_is_shutdown(mp) && ip->i_delayed_blks) { |
c6c2066d DW |
1806 | xfs_check_delalloc(ip, XFS_DATA_FORK); |
1807 | xfs_check_delalloc(ip, XFS_COW_FORK); | |
1808 | ASSERT(0); | |
1809 | } | |
1810 | ||
c6c2066d DW |
1811 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1812 | spin_lock(&pag->pag_ici_lock); | |
1813 | spin_lock(&ip->i_flags_lock); | |
1814 | ||
ab23a776 DC |
1815 | trace_xfs_inode_set_reclaimable(ip); |
1816 | ip->i_flags &= ~(XFS_NEED_INACTIVE | XFS_INACTIVATING); | |
1817 | ip->i_flags |= XFS_IRECLAIMABLE; | |
c6c2066d DW |
1818 | xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1819 | XFS_ICI_RECLAIM_TAG); | |
c6c2066d DW |
1820 | |
1821 | spin_unlock(&ip->i_flags_lock); | |
1822 | spin_unlock(&pag->pag_ici_lock); | |
1823 | xfs_perag_put(pag); | |
1824 | } | |
ab23a776 DC |
1825 | |
1826 | /* | |
1827 | * Free all speculative preallocations and possibly even the inode itself. | |
1828 | * This is the last chance to make changes to an otherwise unreferenced file | |
1829 | * before incore reclamation happens. | |
1830 | */ | |
1831 | static void | |
1832 | xfs_inodegc_inactivate( | |
1833 | struct xfs_inode *ip) | |
1834 | { | |
1835 | trace_xfs_inode_inactivating(ip); | |
1836 | xfs_inactive(ip); | |
1837 | xfs_inodegc_set_reclaimable(ip); | |
1838 | } | |
1839 | ||
1840 | void | |
1841 | xfs_inodegc_worker( | |
1842 | struct work_struct *work) | |
1843 | { | |
7cf2b0f9 DC |
1844 | struct xfs_inodegc *gc = container_of(to_delayed_work(work), |
1845 | struct xfs_inodegc, work); | |
ab23a776 DC |
1846 | struct llist_node *node = llist_del_all(&gc->list); |
1847 | struct xfs_inode *ip, *n; | |
1848 | ||
1849 | WRITE_ONCE(gc->items, 0); | |
1850 | ||
1851 | if (!node) | |
1852 | return; | |
1853 | ||
1854 | ip = llist_entry(node, struct xfs_inode, i_gclist); | |
40b1de00 | 1855 | trace_xfs_inodegc_worker(ip->i_mount, READ_ONCE(gc->shrinker_hits)); |
ab23a776 | 1856 | |
40b1de00 | 1857 | WRITE_ONCE(gc->shrinker_hits, 0); |
ab23a776 DC |
1858 | llist_for_each_entry_safe(ip, n, node, i_gclist) { |
1859 | xfs_iflags_set(ip, XFS_INACTIVATING); | |
1860 | xfs_inodegc_inactivate(ip); | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | /* | |
5e672cd6 DC |
1865 | * Expedite all pending inodegc work to run immediately. This does not wait for |
1866 | * completion of the work. | |
ab23a776 DC |
1867 | */ |
1868 | void | |
5e672cd6 | 1869 | xfs_inodegc_push( |
ab23a776 DC |
1870 | struct xfs_mount *mp) |
1871 | { | |
ab23a776 DC |
1872 | if (!xfs_is_inodegc_enabled(mp)) |
1873 | return; | |
5e672cd6 DC |
1874 | trace_xfs_inodegc_push(mp, __return_address); |
1875 | xfs_inodegc_queue_all(mp); | |
1876 | } | |
ab23a776 | 1877 | |
5e672cd6 DC |
1878 | /* |
1879 | * Force all currently queued inode inactivation work to run immediately and | |
1880 | * wait for the work to finish. | |
1881 | */ | |
1882 | void | |
1883 | xfs_inodegc_flush( | |
1884 | struct xfs_mount *mp) | |
1885 | { | |
1886 | xfs_inodegc_push(mp); | |
ab23a776 | 1887 | trace_xfs_inodegc_flush(mp, __return_address); |
6191cf3a | 1888 | flush_workqueue(mp->m_inodegc_wq); |
ab23a776 DC |
1889 | } |
1890 | ||
1891 | /* | |
1892 | * Flush all the pending work and then disable the inode inactivation background | |
1893 | * workers and wait for them to stop. | |
1894 | */ | |
1895 | void | |
1896 | xfs_inodegc_stop( | |
1897 | struct xfs_mount *mp) | |
1898 | { | |
ab23a776 DC |
1899 | if (!xfs_clear_inodegc_enabled(mp)) |
1900 | return; | |
1901 | ||
1902 | xfs_inodegc_queue_all(mp); | |
6191cf3a | 1903 | drain_workqueue(mp->m_inodegc_wq); |
ab23a776 | 1904 | |
ab23a776 DC |
1905 | trace_xfs_inodegc_stop(mp, __return_address); |
1906 | } | |
1907 | ||
1908 | /* | |
1909 | * Enable the inode inactivation background workers and schedule deferred inode | |
1910 | * inactivation work if there is any. | |
1911 | */ | |
1912 | void | |
1913 | xfs_inodegc_start( | |
1914 | struct xfs_mount *mp) | |
1915 | { | |
1916 | if (xfs_set_inodegc_enabled(mp)) | |
1917 | return; | |
1918 | ||
1919 | trace_xfs_inodegc_start(mp, __return_address); | |
1920 | xfs_inodegc_queue_all(mp); | |
1921 | } | |
1922 | ||
65f03d86 DW |
1923 | #ifdef CONFIG_XFS_RT |
1924 | static inline bool | |
1925 | xfs_inodegc_want_queue_rt_file( | |
1926 | struct xfs_inode *ip) | |
1927 | { | |
1928 | struct xfs_mount *mp = ip->i_mount; | |
65f03d86 DW |
1929 | |
1930 | if (!XFS_IS_REALTIME_INODE(ip)) | |
1931 | return false; | |
1932 | ||
2229276c DW |
1933 | if (__percpu_counter_compare(&mp->m_frextents, |
1934 | mp->m_low_rtexts[XFS_LOWSP_5_PCNT], | |
1935 | XFS_FDBLOCKS_BATCH) < 0) | |
1936 | return true; | |
1937 | ||
1938 | return false; | |
65f03d86 DW |
1939 | } |
1940 | #else | |
1941 | # define xfs_inodegc_want_queue_rt_file(ip) (false) | |
1942 | #endif /* CONFIG_XFS_RT */ | |
1943 | ||
ab23a776 DC |
1944 | /* |
1945 | * Schedule the inactivation worker when: | |
1946 | * | |
1947 | * - We've accumulated more than one inode cluster buffer's worth of inodes. | |
7d6f07d2 | 1948 | * - There is less than 5% free space left. |
108523b8 | 1949 | * - Any of the quotas for this inode are near an enforcement limit. |
ab23a776 DC |
1950 | */ |
1951 | static inline bool | |
1952 | xfs_inodegc_want_queue_work( | |
1953 | struct xfs_inode *ip, | |
1954 | unsigned int items) | |
1955 | { | |
1956 | struct xfs_mount *mp = ip->i_mount; | |
1957 | ||
1958 | if (items > mp->m_ino_geo.inodes_per_cluster) | |
1959 | return true; | |
1960 | ||
7d6f07d2 DW |
1961 | if (__percpu_counter_compare(&mp->m_fdblocks, |
1962 | mp->m_low_space[XFS_LOWSP_5_PCNT], | |
1963 | XFS_FDBLOCKS_BATCH) < 0) | |
1964 | return true; | |
1965 | ||
65f03d86 DW |
1966 | if (xfs_inodegc_want_queue_rt_file(ip)) |
1967 | return true; | |
1968 | ||
108523b8 DW |
1969 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_USER)) |
1970 | return true; | |
1971 | ||
1972 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_GROUP)) | |
1973 | return true; | |
1974 | ||
1975 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_PROJ)) | |
1976 | return true; | |
1977 | ||
ab23a776 DC |
1978 | return false; |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * Upper bound on the number of inodes in each AG that can be queued for | |
1983 | * inactivation at any given time, to avoid monopolizing the workqueue. | |
1984 | */ | |
1985 | #define XFS_INODEGC_MAX_BACKLOG (4 * XFS_INODES_PER_CHUNK) | |
1986 | ||
1987 | /* | |
1988 | * Make the frontend wait for inactivations when: | |
1989 | * | |
40b1de00 | 1990 | * - Memory shrinkers queued the inactivation worker and it hasn't finished. |
ab23a776 DC |
1991 | * - The queue depth exceeds the maximum allowable percpu backlog. |
1992 | * | |
1993 | * Note: If the current thread is running a transaction, we don't ever want to | |
1994 | * wait for other transactions because that could introduce a deadlock. | |
1995 | */ | |
1996 | static inline bool | |
1997 | xfs_inodegc_want_flush_work( | |
1998 | struct xfs_inode *ip, | |
40b1de00 DW |
1999 | unsigned int items, |
2000 | unsigned int shrinker_hits) | |
ab23a776 DC |
2001 | { |
2002 | if (current->journal_info) | |
2003 | return false; | |
2004 | ||
40b1de00 DW |
2005 | if (shrinker_hits > 0) |
2006 | return true; | |
2007 | ||
ab23a776 DC |
2008 | if (items > XFS_INODEGC_MAX_BACKLOG) |
2009 | return true; | |
2010 | ||
2011 | return false; | |
2012 | } | |
2013 | ||
2014 | /* | |
2015 | * Queue a background inactivation worker if there are inodes that need to be | |
2016 | * inactivated and higher level xfs code hasn't disabled the background | |
2017 | * workers. | |
2018 | */ | |
2019 | static void | |
2020 | xfs_inodegc_queue( | |
2021 | struct xfs_inode *ip) | |
2022 | { | |
2023 | struct xfs_mount *mp = ip->i_mount; | |
2024 | struct xfs_inodegc *gc; | |
2025 | int items; | |
40b1de00 | 2026 | unsigned int shrinker_hits; |
7cf2b0f9 | 2027 | unsigned long queue_delay = 1; |
ab23a776 DC |
2028 | |
2029 | trace_xfs_inode_set_need_inactive(ip); | |
2030 | spin_lock(&ip->i_flags_lock); | |
2031 | ip->i_flags |= XFS_NEED_INACTIVE; | |
2032 | spin_unlock(&ip->i_flags_lock); | |
2033 | ||
2034 | gc = get_cpu_ptr(mp->m_inodegc); | |
2035 | llist_add(&ip->i_gclist, &gc->list); | |
2036 | items = READ_ONCE(gc->items); | |
2037 | WRITE_ONCE(gc->items, items + 1); | |
40b1de00 | 2038 | shrinker_hits = READ_ONCE(gc->shrinker_hits); |
ab23a776 | 2039 | |
7cf2b0f9 DC |
2040 | /* |
2041 | * We queue the work while holding the current CPU so that the work | |
2042 | * is scheduled to run on this CPU. | |
2043 | */ | |
2044 | if (!xfs_is_inodegc_enabled(mp)) { | |
2045 | put_cpu_ptr(gc); | |
ab23a776 | 2046 | return; |
ab23a776 DC |
2047 | } |
2048 | ||
7cf2b0f9 DC |
2049 | if (xfs_inodegc_want_queue_work(ip, items)) |
2050 | queue_delay = 0; | |
2051 | ||
2052 | trace_xfs_inodegc_queue(mp, __return_address); | |
2053 | mod_delayed_work(mp->m_inodegc_wq, &gc->work, queue_delay); | |
2054 | put_cpu_ptr(gc); | |
2055 | ||
40b1de00 | 2056 | if (xfs_inodegc_want_flush_work(ip, items, shrinker_hits)) { |
ab23a776 | 2057 | trace_xfs_inodegc_throttle(mp, __return_address); |
7cf2b0f9 | 2058 | flush_delayed_work(&gc->work); |
ab23a776 DC |
2059 | } |
2060 | } | |
2061 | ||
2062 | /* | |
2063 | * Fold the dead CPU inodegc queue into the current CPUs queue. | |
2064 | */ | |
2065 | void | |
2066 | xfs_inodegc_cpu_dead( | |
2067 | struct xfs_mount *mp, | |
2068 | unsigned int dead_cpu) | |
2069 | { | |
2070 | struct xfs_inodegc *dead_gc, *gc; | |
2071 | struct llist_node *first, *last; | |
2072 | unsigned int count = 0; | |
2073 | ||
2074 | dead_gc = per_cpu_ptr(mp->m_inodegc, dead_cpu); | |
7cf2b0f9 | 2075 | cancel_delayed_work_sync(&dead_gc->work); |
ab23a776 DC |
2076 | |
2077 | if (llist_empty(&dead_gc->list)) | |
2078 | return; | |
2079 | ||
2080 | first = dead_gc->list.first; | |
2081 | last = first; | |
2082 | while (last->next) { | |
2083 | last = last->next; | |
2084 | count++; | |
2085 | } | |
2086 | dead_gc->list.first = NULL; | |
2087 | dead_gc->items = 0; | |
2088 | ||
2089 | /* Add pending work to current CPU */ | |
2090 | gc = get_cpu_ptr(mp->m_inodegc); | |
2091 | llist_add_batch(first, last, &gc->list); | |
2092 | count += READ_ONCE(gc->items); | |
2093 | WRITE_ONCE(gc->items, count); | |
ab23a776 DC |
2094 | |
2095 | if (xfs_is_inodegc_enabled(mp)) { | |
2096 | trace_xfs_inodegc_queue(mp, __return_address); | |
7cf2b0f9 | 2097 | mod_delayed_work(mp->m_inodegc_wq, &gc->work, 0); |
ab23a776 | 2098 | } |
7cf2b0f9 | 2099 | put_cpu_ptr(gc); |
ab23a776 DC |
2100 | } |
2101 | ||
2102 | /* | |
2103 | * We set the inode flag atomically with the radix tree tag. Once we get tag | |
2104 | * lookups on the radix tree, this inode flag can go away. | |
2105 | * | |
2106 | * We always use background reclaim here because even if the inode is clean, it | |
2107 | * still may be under IO and hence we have wait for IO completion to occur | |
2108 | * before we can reclaim the inode. The background reclaim path handles this | |
2109 | * more efficiently than we can here, so simply let background reclaim tear down | |
2110 | * all inodes. | |
2111 | */ | |
2112 | void | |
2113 | xfs_inode_mark_reclaimable( | |
2114 | struct xfs_inode *ip) | |
2115 | { | |
2116 | struct xfs_mount *mp = ip->i_mount; | |
2117 | bool need_inactive; | |
2118 | ||
2119 | XFS_STATS_INC(mp, vn_reclaim); | |
2120 | ||
2121 | /* | |
2122 | * We should never get here with any of the reclaim flags already set. | |
2123 | */ | |
2124 | ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_ALL_IRECLAIM_FLAGS)); | |
2125 | ||
2126 | need_inactive = xfs_inode_needs_inactive(ip); | |
2127 | if (need_inactive) { | |
2128 | xfs_inodegc_queue(ip); | |
2129 | return; | |
2130 | } | |
2131 | ||
2132 | /* Going straight to reclaim, so drop the dquots. */ | |
2133 | xfs_qm_dqdetach(ip); | |
2134 | xfs_inodegc_set_reclaimable(ip); | |
2135 | } | |
40b1de00 DW |
2136 | |
2137 | /* | |
2138 | * Register a phony shrinker so that we can run background inodegc sooner when | |
2139 | * there's memory pressure. Inactivation does not itself free any memory but | |
2140 | * it does make inodes reclaimable, which eventually frees memory. | |
2141 | * | |
2142 | * The count function, seek value, and batch value are crafted to trigger the | |
2143 | * scan function during the second round of scanning. Hopefully this means | |
2144 | * that we reclaimed enough memory that initiating metadata transactions won't | |
2145 | * make things worse. | |
2146 | */ | |
2147 | #define XFS_INODEGC_SHRINKER_COUNT (1UL << DEF_PRIORITY) | |
2148 | #define XFS_INODEGC_SHRINKER_BATCH ((XFS_INODEGC_SHRINKER_COUNT / 2) + 1) | |
2149 | ||
2150 | static unsigned long | |
2151 | xfs_inodegc_shrinker_count( | |
2152 | struct shrinker *shrink, | |
2153 | struct shrink_control *sc) | |
2154 | { | |
2155 | struct xfs_mount *mp = container_of(shrink, struct xfs_mount, | |
2156 | m_inodegc_shrinker); | |
2157 | struct xfs_inodegc *gc; | |
2158 | int cpu; | |
2159 | ||
2160 | if (!xfs_is_inodegc_enabled(mp)) | |
2161 | return 0; | |
2162 | ||
2163 | for_each_online_cpu(cpu) { | |
2164 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
2165 | if (!llist_empty(&gc->list)) | |
2166 | return XFS_INODEGC_SHRINKER_COUNT; | |
2167 | } | |
2168 | ||
2169 | return 0; | |
2170 | } | |
2171 | ||
2172 | static unsigned long | |
2173 | xfs_inodegc_shrinker_scan( | |
2174 | struct shrinker *shrink, | |
2175 | struct shrink_control *sc) | |
2176 | { | |
2177 | struct xfs_mount *mp = container_of(shrink, struct xfs_mount, | |
2178 | m_inodegc_shrinker); | |
2179 | struct xfs_inodegc *gc; | |
2180 | int cpu; | |
2181 | bool no_items = true; | |
2182 | ||
2183 | if (!xfs_is_inodegc_enabled(mp)) | |
2184 | return SHRINK_STOP; | |
2185 | ||
2186 | trace_xfs_inodegc_shrinker_scan(mp, sc, __return_address); | |
2187 | ||
2188 | for_each_online_cpu(cpu) { | |
2189 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
2190 | if (!llist_empty(&gc->list)) { | |
2191 | unsigned int h = READ_ONCE(gc->shrinker_hits); | |
2192 | ||
2193 | WRITE_ONCE(gc->shrinker_hits, h + 1); | |
7cf2b0f9 | 2194 | mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0); |
40b1de00 DW |
2195 | no_items = false; |
2196 | } | |
2197 | } | |
2198 | ||
2199 | /* | |
2200 | * If there are no inodes to inactivate, we don't want the shrinker | |
2201 | * to think there's deferred work to call us back about. | |
2202 | */ | |
2203 | if (no_items) | |
2204 | return LONG_MAX; | |
2205 | ||
2206 | return SHRINK_STOP; | |
2207 | } | |
2208 | ||
2209 | /* Register a shrinker so we can accelerate inodegc and throttle queuing. */ | |
2210 | int | |
2211 | xfs_inodegc_register_shrinker( | |
2212 | struct xfs_mount *mp) | |
2213 | { | |
2214 | struct shrinker *shrink = &mp->m_inodegc_shrinker; | |
2215 | ||
2216 | shrink->count_objects = xfs_inodegc_shrinker_count; | |
2217 | shrink->scan_objects = xfs_inodegc_shrinker_scan; | |
2218 | shrink->seeks = 0; | |
2219 | shrink->flags = SHRINKER_NONSLAB; | |
2220 | shrink->batch = XFS_INODEGC_SHRINKER_BATCH; | |
2221 | ||
2222 | return register_shrinker(shrink); | |
2223 | } |