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)); | |
3993baeb | 101 | ip->i_cowfp = NULL; |
2ed5b09b DW |
102 | memset(&ip->i_af, 0, sizeof(ip->i_af)); |
103 | ip->i_af.if_format = XFS_DINODE_FMT_EXTENTS; | |
3ba738df | 104 | memset(&ip->i_df, 0, sizeof(ip->i_df)); |
2ed5b09b | 105 | ip->i_df.if_present = 1; |
33479e05 DC |
106 | ip->i_flags = 0; |
107 | ip->i_delayed_blks = 0; | |
3e09ab8f | 108 | ip->i_diflags2 = mp->m_ino_geo.new_diflags2; |
6e73a545 | 109 | ip->i_nblocks = 0; |
7821ea30 | 110 | ip->i_forkoff = 0; |
6772c1f1 DW |
111 | ip->i_sick = 0; |
112 | ip->i_checked = 0; | |
cb357bf3 DW |
113 | INIT_WORK(&ip->i_ioend_work, xfs_end_io); |
114 | INIT_LIST_HEAD(&ip->i_ioend_list); | |
115 | spin_lock_init(&ip->i_ioend_lock); | |
33479e05 DC |
116 | |
117 | return ip; | |
118 | } | |
119 | ||
120 | STATIC void | |
121 | xfs_inode_free_callback( | |
122 | struct rcu_head *head) | |
123 | { | |
124 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
125 | struct xfs_inode *ip = XFS_I(inode); | |
126 | ||
c19b3b05 | 127 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
128 | case S_IFREG: |
129 | case S_IFDIR: | |
130 | case S_IFLNK: | |
ef838512 | 131 | xfs_idestroy_fork(&ip->i_df); |
33479e05 DC |
132 | break; |
133 | } | |
134 | ||
2ed5b09b DW |
135 | if (ip->i_af.if_present) { |
136 | xfs_idestroy_fork(&ip->i_af); | |
137 | xfs_ifork_zap_attr(ip); | |
ef838512 CH |
138 | } |
139 | if (ip->i_cowfp) { | |
140 | xfs_idestroy_fork(ip->i_cowfp); | |
182696fb | 141 | kmem_cache_free(xfs_ifork_cache, ip->i_cowfp); |
ef838512 | 142 | } |
33479e05 | 143 | if (ip->i_itemp) { |
22525c17 DC |
144 | ASSERT(!test_bit(XFS_LI_IN_AIL, |
145 | &ip->i_itemp->ili_item.li_flags)); | |
33479e05 DC |
146 | xfs_inode_item_destroy(ip); |
147 | ip->i_itemp = NULL; | |
148 | } | |
149 | ||
182696fb | 150 | kmem_cache_free(xfs_inode_cache, ip); |
1f2dcfe8 DC |
151 | } |
152 | ||
8a17d7dd DC |
153 | static void |
154 | __xfs_inode_free( | |
155 | struct xfs_inode *ip) | |
156 | { | |
157 | /* asserts to verify all state is correct here */ | |
158 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
48d55e2a | 159 | ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list)); |
8a17d7dd DC |
160 | XFS_STATS_DEC(ip->i_mount, vn_active); |
161 | ||
162 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
163 | } | |
164 | ||
1f2dcfe8 DC |
165 | void |
166 | xfs_inode_free( | |
167 | struct xfs_inode *ip) | |
168 | { | |
718ecc50 | 169 | ASSERT(!xfs_iflags_test(ip, XFS_IFLUSHING)); |
98efe8af | 170 | |
33479e05 DC |
171 | /* |
172 | * Because we use RCU freeing we need to ensure the inode always | |
173 | * appears to be reclaimed with an invalid inode number when in the | |
174 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
175 | * races. | |
176 | */ | |
177 | spin_lock(&ip->i_flags_lock); | |
178 | ip->i_flags = XFS_IRECLAIM; | |
179 | ip->i_ino = 0; | |
180 | spin_unlock(&ip->i_flags_lock); | |
181 | ||
8a17d7dd | 182 | __xfs_inode_free(ip); |
33479e05 DC |
183 | } |
184 | ||
ad438c40 | 185 | /* |
02511a5a DC |
186 | * Queue background inode reclaim work if there are reclaimable inodes and there |
187 | * isn't reclaim work already scheduled or in progress. | |
ad438c40 DC |
188 | */ |
189 | static void | |
190 | xfs_reclaim_work_queue( | |
191 | struct xfs_mount *mp) | |
192 | { | |
193 | ||
194 | rcu_read_lock(); | |
195 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
196 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
197 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
198 | } | |
199 | rcu_read_unlock(); | |
200 | } | |
201 | ||
c076ae7a DW |
202 | /* |
203 | * Background scanning to trim preallocated space. This is queued based on the | |
204 | * 'speculative_prealloc_lifetime' tunable (5m by default). | |
205 | */ | |
206 | static inline void | |
207 | xfs_blockgc_queue( | |
ad438c40 | 208 | struct xfs_perag *pag) |
c076ae7a | 209 | { |
6f649091 DW |
210 | struct xfs_mount *mp = pag->pag_mount; |
211 | ||
212 | if (!xfs_is_blockgc_enabled(mp)) | |
213 | return; | |
214 | ||
c076ae7a DW |
215 | rcu_read_lock(); |
216 | if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG)) | |
ab23a776 | 217 | queue_delayed_work(pag->pag_mount->m_blockgc_wq, |
c076ae7a DW |
218 | &pag->pag_blockgc_work, |
219 | msecs_to_jiffies(xfs_blockgc_secs * 1000)); | |
220 | rcu_read_unlock(); | |
221 | } | |
222 | ||
223 | /* Set a tag on both the AG incore inode tree and the AG radix tree. */ | |
224 | static void | |
225 | xfs_perag_set_inode_tag( | |
226 | struct xfs_perag *pag, | |
227 | xfs_agino_t agino, | |
228 | unsigned int tag) | |
ad438c40 DC |
229 | { |
230 | struct xfs_mount *mp = pag->pag_mount; | |
c076ae7a | 231 | bool was_tagged; |
ad438c40 | 232 | |
95989c46 | 233 | lockdep_assert_held(&pag->pag_ici_lock); |
c076ae7a DW |
234 | |
235 | was_tagged = radix_tree_tagged(&pag->pag_ici_root, tag); | |
236 | radix_tree_tag_set(&pag->pag_ici_root, agino, tag); | |
237 | ||
238 | if (tag == XFS_ICI_RECLAIM_TAG) | |
239 | pag->pag_ici_reclaimable++; | |
240 | ||
241 | if (was_tagged) | |
ad438c40 DC |
242 | return; |
243 | ||
c076ae7a | 244 | /* propagate the tag up into the perag radix tree */ |
ad438c40 | 245 | spin_lock(&mp->m_perag_lock); |
c076ae7a | 246 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, tag); |
ad438c40 DC |
247 | spin_unlock(&mp->m_perag_lock); |
248 | ||
c076ae7a DW |
249 | /* start background work */ |
250 | switch (tag) { | |
251 | case XFS_ICI_RECLAIM_TAG: | |
252 | xfs_reclaim_work_queue(mp); | |
253 | break; | |
254 | case XFS_ICI_BLOCKGC_TAG: | |
255 | xfs_blockgc_queue(pag); | |
256 | break; | |
257 | } | |
ad438c40 | 258 | |
c076ae7a | 259 | trace_xfs_perag_set_inode_tag(mp, pag->pag_agno, tag, _RET_IP_); |
ad438c40 DC |
260 | } |
261 | ||
c076ae7a | 262 | /* Clear a tag on both the AG incore inode tree and the AG radix tree. */ |
ad438c40 | 263 | static void |
c076ae7a DW |
264 | xfs_perag_clear_inode_tag( |
265 | struct xfs_perag *pag, | |
266 | xfs_agino_t agino, | |
267 | unsigned int tag) | |
ad438c40 DC |
268 | { |
269 | struct xfs_mount *mp = pag->pag_mount; | |
270 | ||
95989c46 | 271 | lockdep_assert_held(&pag->pag_ici_lock); |
c076ae7a DW |
272 | |
273 | /* | |
274 | * Reclaim can signal (with a null agino) that it cleared its own tag | |
275 | * by removing the inode from the radix tree. | |
276 | */ | |
277 | if (agino != NULLAGINO) | |
278 | radix_tree_tag_clear(&pag->pag_ici_root, agino, tag); | |
279 | else | |
280 | ASSERT(tag == XFS_ICI_RECLAIM_TAG); | |
281 | ||
282 | if (tag == XFS_ICI_RECLAIM_TAG) | |
283 | pag->pag_ici_reclaimable--; | |
284 | ||
285 | if (radix_tree_tagged(&pag->pag_ici_root, tag)) | |
ad438c40 DC |
286 | return; |
287 | ||
c076ae7a | 288 | /* clear the tag from the perag radix tree */ |
ad438c40 | 289 | spin_lock(&mp->m_perag_lock); |
c076ae7a | 290 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, tag); |
ad438c40 | 291 | spin_unlock(&mp->m_perag_lock); |
ad438c40 | 292 | |
c076ae7a DW |
293 | trace_xfs_perag_clear_inode_tag(mp, pag->pag_agno, tag, _RET_IP_); |
294 | } | |
ad438c40 | 295 | |
50997470 DC |
296 | /* |
297 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
298 | * part of the structure. This is made more complex by the fact we store | |
299 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 300 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 301 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 302 | * after reinitialisation even if it fails. |
50997470 DC |
303 | */ |
304 | static int | |
305 | xfs_reinit_inode( | |
306 | struct xfs_mount *mp, | |
307 | struct inode *inode) | |
308 | { | |
ff7bebeb DW |
309 | int error; |
310 | uint32_t nlink = inode->i_nlink; | |
311 | uint32_t generation = inode->i_generation; | |
312 | uint64_t version = inode_peek_iversion(inode); | |
313 | umode_t mode = inode->i_mode; | |
314 | dev_t dev = inode->i_rdev; | |
315 | kuid_t uid = inode->i_uid; | |
316 | kgid_t gid = inode->i_gid; | |
50997470 DC |
317 | |
318 | error = inode_init_always(mp->m_super, inode); | |
319 | ||
54d7b5c1 | 320 | set_nlink(inode, nlink); |
9e9a2674 | 321 | inode->i_generation = generation; |
f0e28280 | 322 | inode_set_iversion_queried(inode, version); |
c19b3b05 | 323 | inode->i_mode = mode; |
acd1d715 | 324 | inode->i_rdev = dev; |
3d8f2821 CH |
325 | inode->i_uid = uid; |
326 | inode->i_gid = gid; | |
67958013 | 327 | mapping_set_large_folios(inode->i_mapping); |
50997470 DC |
328 | return error; |
329 | } | |
330 | ||
ff7bebeb DW |
331 | /* |
332 | * Carefully nudge an inode whose VFS state has been torn down back into a | |
333 | * usable state. Drops the i_flags_lock and the rcu read lock. | |
334 | */ | |
335 | static int | |
336 | xfs_iget_recycle( | |
337 | struct xfs_perag *pag, | |
338 | struct xfs_inode *ip) __releases(&ip->i_flags_lock) | |
339 | { | |
340 | struct xfs_mount *mp = ip->i_mount; | |
341 | struct inode *inode = VFS_I(ip); | |
342 | int error; | |
343 | ||
344 | trace_xfs_iget_recycle(ip); | |
345 | ||
346 | /* | |
347 | * We need to make it look like the inode is being reclaimed to prevent | |
348 | * the actual reclaim workers from stomping over us while we recycle | |
349 | * the inode. We can't clear the radix tree tag yet as it requires | |
350 | * pag_ici_lock to be held exclusive. | |
351 | */ | |
352 | ip->i_flags |= XFS_IRECLAIM; | |
353 | ||
354 | spin_unlock(&ip->i_flags_lock); | |
355 | rcu_read_unlock(); | |
356 | ||
357 | ASSERT(!rwsem_is_locked(&inode->i_rwsem)); | |
358 | error = xfs_reinit_inode(mp, inode); | |
359 | if (error) { | |
ff7bebeb DW |
360 | /* |
361 | * Re-initializing the inode failed, and we are in deep | |
362 | * trouble. Try to re-add it to the reclaim list. | |
363 | */ | |
364 | rcu_read_lock(); | |
365 | spin_lock(&ip->i_flags_lock); | |
ff7bebeb | 366 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
ff7bebeb DW |
367 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); |
368 | spin_unlock(&ip->i_flags_lock); | |
369 | rcu_read_unlock(); | |
370 | ||
371 | trace_xfs_iget_recycle_fail(ip); | |
372 | return error; | |
373 | } | |
374 | ||
375 | spin_lock(&pag->pag_ici_lock); | |
376 | spin_lock(&ip->i_flags_lock); | |
377 | ||
378 | /* | |
379 | * Clear the per-lifetime state in the inode as we are now effectively | |
380 | * a new inode and need to return to the initial state before reuse | |
381 | * occurs. | |
382 | */ | |
383 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
384 | ip->i_flags |= XFS_INEW; | |
385 | xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
386 | XFS_ICI_RECLAIM_TAG); | |
387 | inode->i_state = I_NEW; | |
388 | spin_unlock(&ip->i_flags_lock); | |
389 | spin_unlock(&pag->pag_ici_lock); | |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
afca6c5b DC |
394 | /* |
395 | * If we are allocating a new inode, then check what was returned is | |
396 | * actually a free, empty inode. If we are not allocating an inode, | |
397 | * then check we didn't find a free inode. | |
398 | * | |
399 | * Returns: | |
400 | * 0 if the inode free state matches the lookup context | |
401 | * -ENOENT if the inode is free and we are not allocating | |
402 | * -EFSCORRUPTED if there is any state mismatch at all | |
403 | */ | |
404 | static int | |
405 | xfs_iget_check_free_state( | |
406 | struct xfs_inode *ip, | |
407 | int flags) | |
408 | { | |
409 | if (flags & XFS_IGET_CREATE) { | |
410 | /* should be a free inode */ | |
411 | if (VFS_I(ip)->i_mode != 0) { | |
412 | xfs_warn(ip->i_mount, | |
413 | "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)", | |
414 | ip->i_ino, VFS_I(ip)->i_mode); | |
415 | return -EFSCORRUPTED; | |
416 | } | |
417 | ||
6e73a545 | 418 | if (ip->i_nblocks != 0) { |
afca6c5b DC |
419 | xfs_warn(ip->i_mount, |
420 | "Corruption detected! Free inode 0x%llx has blocks allocated!", | |
421 | ip->i_ino); | |
422 | return -EFSCORRUPTED; | |
423 | } | |
424 | return 0; | |
425 | } | |
426 | ||
427 | /* should be an allocated inode */ | |
428 | if (VFS_I(ip)->i_mode == 0) | |
429 | return -ENOENT; | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
ab23a776 DC |
434 | /* Make all pending inactivation work start immediately. */ |
435 | static void | |
436 | xfs_inodegc_queue_all( | |
437 | struct xfs_mount *mp) | |
438 | { | |
439 | struct xfs_inodegc *gc; | |
440 | int cpu; | |
441 | ||
442 | for_each_online_cpu(cpu) { | |
443 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
444 | if (!llist_empty(&gc->list)) | |
7cf2b0f9 | 445 | mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0); |
ab23a776 DC |
446 | } |
447 | } | |
448 | ||
33479e05 DC |
449 | /* |
450 | * Check the validity of the inode we just found it the cache | |
451 | */ | |
452 | static int | |
453 | xfs_iget_cache_hit( | |
454 | struct xfs_perag *pag, | |
455 | struct xfs_inode *ip, | |
456 | xfs_ino_t ino, | |
457 | int flags, | |
458 | int lock_flags) __releases(RCU) | |
459 | { | |
460 | struct inode *inode = VFS_I(ip); | |
461 | struct xfs_mount *mp = ip->i_mount; | |
462 | int error; | |
463 | ||
464 | /* | |
465 | * check for re-use of an inode within an RCU grace period due to the | |
466 | * radix tree nodes not being updated yet. We monitor for this by | |
467 | * setting the inode number to zero before freeing the inode structure. | |
468 | * If the inode has been reallocated and set up, then the inode number | |
469 | * will not match, so check for that, too. | |
470 | */ | |
471 | spin_lock(&ip->i_flags_lock); | |
77b4d286 DW |
472 | if (ip->i_ino != ino) |
473 | goto out_skip; | |
33479e05 DC |
474 | |
475 | /* | |
476 | * If we are racing with another cache hit that is currently | |
477 | * instantiating this inode or currently recycling it out of | |
ff7bebeb | 478 | * reclaimable state, wait for the initialisation to complete |
33479e05 DC |
479 | * before continuing. |
480 | * | |
ab23a776 DC |
481 | * If we're racing with the inactivation worker we also want to wait. |
482 | * If we're creating a new file, it's possible that the worker | |
483 | * previously marked the inode as free on disk but hasn't finished | |
484 | * updating the incore state yet. The AGI buffer will be dirty and | |
485 | * locked to the icreate transaction, so a synchronous push of the | |
486 | * inodegc workers would result in deadlock. For a regular iget, the | |
487 | * worker is running already, so we might as well wait. | |
488 | * | |
33479e05 DC |
489 | * XXX(hch): eventually we should do something equivalent to |
490 | * wait_on_inode to wait for these flags to be cleared | |
491 | * instead of polling for it. | |
492 | */ | |
ab23a776 | 493 | if (ip->i_flags & (XFS_INEW | XFS_IRECLAIM | XFS_INACTIVATING)) |
77b4d286 | 494 | goto out_skip; |
33479e05 | 495 | |
ab23a776 DC |
496 | if (ip->i_flags & XFS_NEED_INACTIVE) { |
497 | /* Unlinked inodes cannot be re-grabbed. */ | |
498 | if (VFS_I(ip)->i_nlink == 0) { | |
499 | error = -ENOENT; | |
500 | goto out_error; | |
501 | } | |
502 | goto out_inodegc_flush; | |
503 | } | |
504 | ||
33479e05 | 505 | /* |
afca6c5b DC |
506 | * Check the inode free state is valid. This also detects lookup |
507 | * racing with unlinks. | |
33479e05 | 508 | */ |
afca6c5b DC |
509 | error = xfs_iget_check_free_state(ip, flags); |
510 | if (error) | |
33479e05 | 511 | goto out_error; |
33479e05 | 512 | |
77b4d286 DW |
513 | /* Skip inodes that have no vfs state. */ |
514 | if ((flags & XFS_IGET_INCORE) && | |
515 | (ip->i_flags & XFS_IRECLAIMABLE)) | |
516 | goto out_skip; | |
378f681c | 517 | |
77b4d286 DW |
518 | /* The inode fits the selection criteria; process it. */ |
519 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
ff7bebeb DW |
520 | /* Drops i_flags_lock and RCU read lock. */ |
521 | error = xfs_iget_recycle(pag, ip); | |
522 | if (error) | |
523 | return error; | |
33479e05 DC |
524 | } else { |
525 | /* If the VFS inode is being torn down, pause and try again. */ | |
77b4d286 DW |
526 | if (!igrab(inode)) |
527 | goto out_skip; | |
33479e05 DC |
528 | |
529 | /* We've got a live one. */ | |
530 | spin_unlock(&ip->i_flags_lock); | |
531 | rcu_read_unlock(); | |
532 | trace_xfs_iget_hit(ip); | |
533 | } | |
534 | ||
535 | if (lock_flags != 0) | |
536 | xfs_ilock(ip, lock_flags); | |
537 | ||
378f681c | 538 | if (!(flags & XFS_IGET_INCORE)) |
dae2f8ed | 539 | xfs_iflags_clear(ip, XFS_ISTALE); |
ff6d6af2 | 540 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
541 | |
542 | return 0; | |
543 | ||
77b4d286 DW |
544 | out_skip: |
545 | trace_xfs_iget_skip(ip); | |
546 | XFS_STATS_INC(mp, xs_ig_frecycle); | |
547 | error = -EAGAIN; | |
33479e05 DC |
548 | out_error: |
549 | spin_unlock(&ip->i_flags_lock); | |
550 | rcu_read_unlock(); | |
551 | return error; | |
ab23a776 DC |
552 | |
553 | out_inodegc_flush: | |
554 | spin_unlock(&ip->i_flags_lock); | |
555 | rcu_read_unlock(); | |
556 | /* | |
557 | * Do not wait for the workers, because the caller could hold an AGI | |
558 | * buffer lock. We're just going to sleep in a loop anyway. | |
559 | */ | |
560 | if (xfs_is_inodegc_enabled(mp)) | |
561 | xfs_inodegc_queue_all(mp); | |
562 | return -EAGAIN; | |
33479e05 DC |
563 | } |
564 | ||
33479e05 DC |
565 | static int |
566 | xfs_iget_cache_miss( | |
567 | struct xfs_mount *mp, | |
568 | struct xfs_perag *pag, | |
569 | xfs_trans_t *tp, | |
570 | xfs_ino_t ino, | |
571 | struct xfs_inode **ipp, | |
572 | int flags, | |
573 | int lock_flags) | |
574 | { | |
575 | struct xfs_inode *ip; | |
576 | int error; | |
577 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
578 | int iflags; | |
579 | ||
580 | ip = xfs_inode_alloc(mp, ino); | |
581 | if (!ip) | |
2451337d | 582 | return -ENOMEM; |
33479e05 | 583 | |
bb8a66af | 584 | error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, flags); |
33479e05 DC |
585 | if (error) |
586 | goto out_destroy; | |
587 | ||
bb8a66af CH |
588 | /* |
589 | * For version 5 superblocks, if we are initialising a new inode and we | |
0560f31a | 590 | * are not utilising the XFS_FEAT_IKEEP inode cluster mode, we can |
bb8a66af CH |
591 | * simply build the new inode core with a random generation number. |
592 | * | |
593 | * For version 4 (and older) superblocks, log recovery is dependent on | |
965e0a1a | 594 | * the i_flushiter field being initialised from the current on-disk |
bb8a66af CH |
595 | * value and hence we must also read the inode off disk even when |
596 | * initializing new inodes. | |
597 | */ | |
38c26bfd | 598 | if (xfs_has_v3inodes(mp) && |
0560f31a | 599 | (flags & XFS_IGET_CREATE) && !xfs_has_ikeep(mp)) { |
bb8a66af CH |
600 | VFS_I(ip)->i_generation = prandom_u32(); |
601 | } else { | |
bb8a66af CH |
602 | struct xfs_buf *bp; |
603 | ||
af9dcdde | 604 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp); |
bb8a66af CH |
605 | if (error) |
606 | goto out_destroy; | |
607 | ||
af9dcdde CH |
608 | error = xfs_inode_from_disk(ip, |
609 | xfs_buf_offset(bp, ip->i_imap.im_boffset)); | |
bb8a66af CH |
610 | if (!error) |
611 | xfs_buf_set_ref(bp, XFS_INO_REF); | |
612 | xfs_trans_brelse(tp, bp); | |
613 | ||
614 | if (error) | |
615 | goto out_destroy; | |
616 | } | |
617 | ||
33479e05 DC |
618 | trace_xfs_iget_miss(ip); |
619 | ||
ee457001 | 620 | /* |
afca6c5b DC |
621 | * Check the inode free state is valid. This also detects lookup |
622 | * racing with unlinks. | |
ee457001 | 623 | */ |
afca6c5b DC |
624 | error = xfs_iget_check_free_state(ip, flags); |
625 | if (error) | |
33479e05 | 626 | goto out_destroy; |
33479e05 DC |
627 | |
628 | /* | |
629 | * Preload the radix tree so we can insert safely under the | |
630 | * write spinlock. Note that we cannot sleep inside the preload | |
631 | * region. Since we can be called from transaction context, don't | |
632 | * recurse into the file system. | |
633 | */ | |
634 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 635 | error = -EAGAIN; |
33479e05 DC |
636 | goto out_destroy; |
637 | } | |
638 | ||
639 | /* | |
640 | * Because the inode hasn't been added to the radix-tree yet it can't | |
641 | * be found by another thread, so we can do the non-sleeping lock here. | |
642 | */ | |
643 | if (lock_flags) { | |
644 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
645 | BUG(); | |
646 | } | |
647 | ||
648 | /* | |
649 | * These values must be set before inserting the inode into the radix | |
650 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
651 | * RCU locking mechanism) can find it and that lookup must see that this | |
652 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
653 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
654 | * memory barrier that ensures this detection works correctly at lookup | |
655 | * time. | |
656 | */ | |
657 | iflags = XFS_INEW; | |
658 | if (flags & XFS_IGET_DONTCACHE) | |
2c567af4 | 659 | d_mark_dontcache(VFS_I(ip)); |
113a5683 CS |
660 | ip->i_udquot = NULL; |
661 | ip->i_gdquot = NULL; | |
92f8ff73 | 662 | ip->i_pdquot = NULL; |
33479e05 DC |
663 | xfs_iflags_set(ip, iflags); |
664 | ||
665 | /* insert the new inode */ | |
666 | spin_lock(&pag->pag_ici_lock); | |
667 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
668 | if (unlikely(error)) { | |
669 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 670 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 671 | error = -EAGAIN; |
33479e05 DC |
672 | goto out_preload_end; |
673 | } | |
674 | spin_unlock(&pag->pag_ici_lock); | |
675 | radix_tree_preload_end(); | |
676 | ||
677 | *ipp = ip; | |
678 | return 0; | |
679 | ||
680 | out_preload_end: | |
681 | spin_unlock(&pag->pag_ici_lock); | |
682 | radix_tree_preload_end(); | |
683 | if (lock_flags) | |
684 | xfs_iunlock(ip, lock_flags); | |
685 | out_destroy: | |
686 | __destroy_inode(VFS_I(ip)); | |
687 | xfs_inode_free(ip); | |
688 | return error; | |
689 | } | |
690 | ||
691 | /* | |
02511a5a DC |
692 | * Look up an inode by number in the given file system. The inode is looked up |
693 | * in the cache held in each AG. If the inode is found in the cache, initialise | |
694 | * the vfs inode if necessary. | |
33479e05 | 695 | * |
02511a5a DC |
696 | * If it is not in core, read it in from the file system's device, add it to the |
697 | * cache and initialise the vfs inode. | |
33479e05 DC |
698 | * |
699 | * The inode is locked according to the value of the lock_flags parameter. | |
02511a5a DC |
700 | * Inode lookup is only done during metadata operations and not as part of the |
701 | * data IO path. Hence we only allow locking of the XFS_ILOCK during lookup. | |
33479e05 DC |
702 | */ |
703 | int | |
704 | xfs_iget( | |
02511a5a DC |
705 | struct xfs_mount *mp, |
706 | struct xfs_trans *tp, | |
707 | xfs_ino_t ino, | |
708 | uint flags, | |
709 | uint lock_flags, | |
710 | struct xfs_inode **ipp) | |
33479e05 | 711 | { |
02511a5a DC |
712 | struct xfs_inode *ip; |
713 | struct xfs_perag *pag; | |
714 | xfs_agino_t agino; | |
715 | int error; | |
33479e05 | 716 | |
33479e05 DC |
717 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); |
718 | ||
719 | /* reject inode numbers outside existing AGs */ | |
720 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 721 | return -EINVAL; |
33479e05 | 722 | |
ff6d6af2 | 723 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 724 | |
33479e05 DC |
725 | /* get the perag structure and ensure that it's inode capable */ |
726 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
727 | agino = XFS_INO_TO_AGINO(mp, ino); | |
728 | ||
729 | again: | |
730 | error = 0; | |
731 | rcu_read_lock(); | |
732 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
733 | ||
734 | if (ip) { | |
735 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
736 | if (error) | |
737 | goto out_error_or_again; | |
738 | } else { | |
739 | rcu_read_unlock(); | |
378f681c | 740 | if (flags & XFS_IGET_INCORE) { |
ed438b47 | 741 | error = -ENODATA; |
378f681c DW |
742 | goto out_error_or_again; |
743 | } | |
ff6d6af2 | 744 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
745 | |
746 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
747 | flags, lock_flags); | |
748 | if (error) | |
749 | goto out_error_or_again; | |
750 | } | |
751 | xfs_perag_put(pag); | |
752 | ||
753 | *ipp = ip; | |
754 | ||
755 | /* | |
58c90473 | 756 | * If we have a real type for an on-disk inode, we can setup the inode |
132c460e YX |
757 | * now. If it's a new inode being created, xfs_init_new_inode will |
758 | * handle it. | |
33479e05 | 759 | */ |
c19b3b05 | 760 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 761 | xfs_setup_existing_inode(ip); |
33479e05 DC |
762 | return 0; |
763 | ||
764 | out_error_or_again: | |
378f681c | 765 | if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) { |
33479e05 DC |
766 | delay(1); |
767 | goto again; | |
768 | } | |
769 | xfs_perag_put(pag); | |
770 | return error; | |
771 | } | |
772 | ||
378f681c DW |
773 | /* |
774 | * "Is this a cached inode that's also allocated?" | |
775 | * | |
776 | * Look up an inode by number in the given file system. If the inode is | |
777 | * in cache and isn't in purgatory, return 1 if the inode is allocated | |
778 | * and 0 if it is not. For all other cases (not in cache, being torn | |
779 | * down, etc.), return a negative error code. | |
780 | * | |
781 | * The caller has to prevent inode allocation and freeing activity, | |
782 | * presumably by locking the AGI buffer. This is to ensure that an | |
783 | * inode cannot transition from allocated to freed until the caller is | |
784 | * ready to allow that. If the inode is in an intermediate state (new, | |
785 | * reclaimable, or being reclaimed), -EAGAIN will be returned; if the | |
786 | * inode is not in the cache, -ENOENT will be returned. The caller must | |
787 | * deal with these scenarios appropriately. | |
788 | * | |
789 | * This is a specialized use case for the online scrubber; if you're | |
790 | * reading this, you probably want xfs_iget. | |
791 | */ | |
792 | int | |
793 | xfs_icache_inode_is_allocated( | |
794 | struct xfs_mount *mp, | |
795 | struct xfs_trans *tp, | |
796 | xfs_ino_t ino, | |
797 | bool *inuse) | |
798 | { | |
799 | struct xfs_inode *ip; | |
800 | int error; | |
801 | ||
802 | error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip); | |
803 | if (error) | |
804 | return error; | |
805 | ||
806 | *inuse = !!(VFS_I(ip)->i_mode); | |
44a8736b | 807 | xfs_irele(ip); |
378f681c DW |
808 | return 0; |
809 | } | |
810 | ||
e3a20c0b DC |
811 | /* |
812 | * Grab the inode for reclaim exclusively. | |
50718b8d DC |
813 | * |
814 | * We have found this inode via a lookup under RCU, so the inode may have | |
815 | * already been freed, or it may be in the process of being recycled by | |
816 | * xfs_iget(). In both cases, the inode will have XFS_IRECLAIM set. If the inode | |
817 | * has been fully recycled by the time we get the i_flags_lock, XFS_IRECLAIMABLE | |
818 | * will not be set. Hence we need to check for both these flag conditions to | |
819 | * avoid inodes that are no longer reclaim candidates. | |
820 | * | |
821 | * Note: checking for other state flags here, under the i_flags_lock or not, is | |
822 | * racy and should be avoided. Those races should be resolved only after we have | |
823 | * ensured that we are able to reclaim this inode and the world can see that we | |
824 | * are going to reclaim it. | |
825 | * | |
826 | * Return true if we grabbed it, false otherwise. | |
e3a20c0b | 827 | */ |
50718b8d | 828 | static bool |
f1bc5c56 | 829 | xfs_reclaim_igrab( |
9492750a | 830 | struct xfs_inode *ip, |
b26b2bf1 | 831 | struct xfs_icwalk *icw) |
e3a20c0b | 832 | { |
1a3e8f3d DC |
833 | ASSERT(rcu_read_lock_held()); |
834 | ||
e3a20c0b | 835 | spin_lock(&ip->i_flags_lock); |
1a3e8f3d DC |
836 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
837 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
838 | /* not a reclaim candidate. */ | |
e3a20c0b | 839 | spin_unlock(&ip->i_flags_lock); |
50718b8d | 840 | return false; |
e3a20c0b | 841 | } |
9492750a DW |
842 | |
843 | /* Don't reclaim a sick inode unless the caller asked for it. */ | |
844 | if (ip->i_sick && | |
b26b2bf1 | 845 | (!icw || !(icw->icw_flags & XFS_ICWALK_FLAG_RECLAIM_SICK))) { |
9492750a DW |
846 | spin_unlock(&ip->i_flags_lock); |
847 | return false; | |
848 | } | |
849 | ||
e3a20c0b DC |
850 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
851 | spin_unlock(&ip->i_flags_lock); | |
50718b8d | 852 | return true; |
e3a20c0b DC |
853 | } |
854 | ||
777df5af | 855 | /* |
02511a5a DC |
856 | * Inode reclaim is non-blocking, so the default action if progress cannot be |
857 | * made is to "requeue" the inode for reclaim by unlocking it and clearing the | |
858 | * XFS_IRECLAIM flag. If we are in a shutdown state, we don't care about | |
859 | * blocking anymore and hence we can wait for the inode to be able to reclaim | |
860 | * it. | |
777df5af | 861 | * |
02511a5a DC |
862 | * We do no IO here - if callers require inodes to be cleaned they must push the |
863 | * AIL first to trigger writeback of dirty inodes. This enables writeback to be | |
864 | * done in the background in a non-blocking manner, and enables memory reclaim | |
865 | * to make progress without blocking. | |
777df5af | 866 | */ |
4d0bab3a | 867 | static void |
c8e20be0 | 868 | xfs_reclaim_inode( |
75f3cb13 | 869 | struct xfs_inode *ip, |
50718b8d | 870 | struct xfs_perag *pag) |
fce08f2f | 871 | { |
8a17d7dd | 872 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
777df5af | 873 | |
9552e14d | 874 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) |
617825fe | 875 | goto out; |
718ecc50 | 876 | if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING)) |
9552e14d | 877 | goto out_iunlock; |
7a3be02b | 878 | |
01728b44 DC |
879 | /* |
880 | * Check for log shutdown because aborting the inode can move the log | |
881 | * tail and corrupt in memory state. This is fine if the log is shut | |
882 | * down, but if the log is still active and only the mount is shut down | |
883 | * then the in-memory log tail movement caused by the abort can be | |
884 | * incorrectly propagated to disk. | |
885 | */ | |
886 | if (xlog_is_shutdown(ip->i_mount->m_log)) { | |
777df5af | 887 | xfs_iunpin_wait(ip); |
d2d7c047 | 888 | xfs_iflush_shutdown_abort(ip); |
777df5af DC |
889 | goto reclaim; |
890 | } | |
617825fe | 891 | if (xfs_ipincount(ip)) |
718ecc50 | 892 | goto out_clear_flush; |
617825fe | 893 | if (!xfs_inode_clean(ip)) |
718ecc50 | 894 | goto out_clear_flush; |
8a48088f | 895 | |
718ecc50 | 896 | xfs_iflags_clear(ip, XFS_IFLUSHING); |
777df5af | 897 | reclaim: |
ab23a776 | 898 | trace_xfs_inode_reclaiming(ip); |
98efe8af | 899 | |
8a17d7dd DC |
900 | /* |
901 | * Because we use RCU freeing we need to ensure the inode always appears | |
902 | * to be reclaimed with an invalid inode number when in the free state. | |
98efe8af | 903 | * We do this as early as possible under the ILOCK so that |
f2e9ad21 OS |
904 | * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to |
905 | * detect races with us here. By doing this, we guarantee that once | |
906 | * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that | |
907 | * it will see either a valid inode that will serialise correctly, or it | |
908 | * will see an invalid inode that it can skip. | |
8a17d7dd DC |
909 | */ |
910 | spin_lock(&ip->i_flags_lock); | |
911 | ip->i_flags = XFS_IRECLAIM; | |
912 | ip->i_ino = 0; | |
255794c7 DW |
913 | ip->i_sick = 0; |
914 | ip->i_checked = 0; | |
8a17d7dd DC |
915 | spin_unlock(&ip->i_flags_lock); |
916 | ||
c8e20be0 | 917 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 918 | |
ff6d6af2 | 919 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
920 | /* |
921 | * Remove the inode from the per-AG radix tree. | |
922 | * | |
923 | * Because radix_tree_delete won't complain even if the item was never | |
924 | * added to the tree assert that it's been there before to catch | |
925 | * problems with the inode life time early on. | |
926 | */ | |
1a427ab0 | 927 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 928 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 929 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 930 | ASSERT(0); |
c076ae7a | 931 | xfs_perag_clear_inode_tag(pag, NULLAGINO, XFS_ICI_RECLAIM_TAG); |
1a427ab0 | 932 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
933 | |
934 | /* | |
935 | * Here we do an (almost) spurious inode lock in order to coordinate | |
936 | * with inode cache radix tree lookups. This is because the lookup | |
937 | * can reference the inodes in the cache without taking references. | |
938 | * | |
939 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 940 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 941 | */ |
ad637a10 | 942 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
3ea06d73 | 943 | ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot); |
ad637a10 | 944 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
96355d5a | 945 | ASSERT(xfs_inode_clean(ip)); |
2f11feab | 946 | |
8a17d7dd | 947 | __xfs_inode_free(ip); |
4d0bab3a | 948 | return; |
8a48088f | 949 | |
718ecc50 DC |
950 | out_clear_flush: |
951 | xfs_iflags_clear(ip, XFS_IFLUSHING); | |
9552e14d | 952 | out_iunlock: |
8a48088f | 953 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
9552e14d | 954 | out: |
617825fe | 955 | xfs_iflags_clear(ip, XFS_IRECLAIM); |
7a3be02b DC |
956 | } |
957 | ||
9492750a DW |
958 | /* Reclaim sick inodes if we're unmounting or the fs went down. */ |
959 | static inline bool | |
960 | xfs_want_reclaim_sick( | |
961 | struct xfs_mount *mp) | |
962 | { | |
2e973b2c | 963 | return xfs_is_unmounting(mp) || xfs_has_norecovery(mp) || |
75c8c50f | 964 | xfs_is_shutdown(mp); |
9492750a DW |
965 | } |
966 | ||
4d0bab3a | 967 | void |
7a3be02b | 968 | xfs_reclaim_inodes( |
4d0bab3a | 969 | struct xfs_mount *mp) |
7a3be02b | 970 | { |
b26b2bf1 DW |
971 | struct xfs_icwalk icw = { |
972 | .icw_flags = 0, | |
9492750a DW |
973 | }; |
974 | ||
975 | if (xfs_want_reclaim_sick(mp)) | |
b26b2bf1 | 976 | icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK; |
9492750a | 977 | |
4d0bab3a | 978 | while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { |
617825fe | 979 | xfs_ail_push_all_sync(mp->m_ail); |
b26b2bf1 | 980 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw); |
0f4ec0f1 | 981 | } |
9bf729c0 DC |
982 | } |
983 | ||
984 | /* | |
02511a5a DC |
985 | * The shrinker infrastructure determines how many inodes we should scan for |
986 | * reclaim. We want as many clean inodes ready to reclaim as possible, so we | |
987 | * push the AIL here. We also want to proactively free up memory if we can to | |
988 | * minimise the amount of work memory reclaim has to do so we kick the | |
989 | * background reclaim if it isn't already scheduled. | |
9bf729c0 | 990 | */ |
0a234c6d | 991 | long |
8daaa831 DC |
992 | xfs_reclaim_inodes_nr( |
993 | struct xfs_mount *mp, | |
10be350b | 994 | unsigned long nr_to_scan) |
9bf729c0 | 995 | { |
b26b2bf1 DW |
996 | struct xfs_icwalk icw = { |
997 | .icw_flags = XFS_ICWALK_FLAG_SCAN_LIMIT, | |
10be350b | 998 | .icw_scan_limit = min_t(unsigned long, LONG_MAX, nr_to_scan), |
f1bc5c56 DW |
999 | }; |
1000 | ||
9492750a | 1001 | if (xfs_want_reclaim_sick(mp)) |
b26b2bf1 | 1002 | icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK; |
9492750a | 1003 | |
8daaa831 | 1004 | /* kick background reclaimer and push the AIL */ |
5889608d | 1005 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1006 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1007 | |
b26b2bf1 | 1008 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw); |
617825fe | 1009 | return 0; |
8daaa831 | 1010 | } |
9bf729c0 | 1011 | |
8daaa831 DC |
1012 | /* |
1013 | * Return the number of reclaimable inodes in the filesystem for | |
1014 | * the shrinker to determine how much to reclaim. | |
1015 | */ | |
10be350b | 1016 | long |
8daaa831 DC |
1017 | xfs_reclaim_inodes_count( |
1018 | struct xfs_mount *mp) | |
1019 | { | |
1020 | struct xfs_perag *pag; | |
1021 | xfs_agnumber_t ag = 0; | |
10be350b | 1022 | long reclaimable = 0; |
9bf729c0 | 1023 | |
65d0f205 DC |
1024 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1025 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1026 | reclaimable += pag->pag_ici_reclaimable; |
1027 | xfs_perag_put(pag); | |
9bf729c0 | 1028 | } |
9bf729c0 DC |
1029 | return reclaimable; |
1030 | } | |
1031 | ||
39b1cfd7 | 1032 | STATIC bool |
b26b2bf1 | 1033 | xfs_icwalk_match_id( |
3e3f9f58 | 1034 | struct xfs_inode *ip, |
b26b2bf1 | 1035 | struct xfs_icwalk *icw) |
3e3f9f58 | 1036 | { |
b26b2bf1 DW |
1037 | if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) && |
1038 | !uid_eq(VFS_I(ip)->i_uid, icw->icw_uid)) | |
39b1cfd7 | 1039 | return false; |
3e3f9f58 | 1040 | |
b26b2bf1 DW |
1041 | if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) && |
1042 | !gid_eq(VFS_I(ip)->i_gid, icw->icw_gid)) | |
39b1cfd7 | 1043 | return false; |
1b556048 | 1044 | |
b26b2bf1 DW |
1045 | if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) && |
1046 | ip->i_projid != icw->icw_prid) | |
39b1cfd7 | 1047 | return false; |
1b556048 | 1048 | |
39b1cfd7 | 1049 | return true; |
3e3f9f58 BF |
1050 | } |
1051 | ||
f4526397 BF |
1052 | /* |
1053 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1054 | * criteria match. This is for global/internal scans only. | |
1055 | */ | |
39b1cfd7 | 1056 | STATIC bool |
b26b2bf1 | 1057 | xfs_icwalk_match_id_union( |
f4526397 | 1058 | struct xfs_inode *ip, |
b26b2bf1 | 1059 | struct xfs_icwalk *icw) |
f4526397 | 1060 | { |
b26b2bf1 DW |
1061 | if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) && |
1062 | uid_eq(VFS_I(ip)->i_uid, icw->icw_uid)) | |
39b1cfd7 | 1063 | return true; |
f4526397 | 1064 | |
b26b2bf1 DW |
1065 | if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) && |
1066 | gid_eq(VFS_I(ip)->i_gid, icw->icw_gid)) | |
39b1cfd7 | 1067 | return true; |
f4526397 | 1068 | |
b26b2bf1 DW |
1069 | if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) && |
1070 | ip->i_projid == icw->icw_prid) | |
39b1cfd7 | 1071 | return true; |
f4526397 | 1072 | |
39b1cfd7 | 1073 | return false; |
f4526397 BF |
1074 | } |
1075 | ||
a91bf992 DW |
1076 | /* |
1077 | * Is this inode @ip eligible for eof/cow block reclamation, given some | |
b26b2bf1 | 1078 | * filtering parameters @icw? The inode is eligible if @icw is null or |
a91bf992 DW |
1079 | * if the predicate functions match. |
1080 | */ | |
1081 | static bool | |
b26b2bf1 | 1082 | xfs_icwalk_match( |
a91bf992 | 1083 | struct xfs_inode *ip, |
b26b2bf1 | 1084 | struct xfs_icwalk *icw) |
a91bf992 | 1085 | { |
39b1cfd7 | 1086 | bool match; |
a91bf992 | 1087 | |
b26b2bf1 | 1088 | if (!icw) |
a91bf992 DW |
1089 | return true; |
1090 | ||
b26b2bf1 DW |
1091 | if (icw->icw_flags & XFS_ICWALK_FLAG_UNION) |
1092 | match = xfs_icwalk_match_id_union(ip, icw); | |
a91bf992 | 1093 | else |
b26b2bf1 | 1094 | match = xfs_icwalk_match_id(ip, icw); |
a91bf992 DW |
1095 | if (!match) |
1096 | return false; | |
1097 | ||
1098 | /* skip the inode if the file size is too small */ | |
b26b2bf1 DW |
1099 | if ((icw->icw_flags & XFS_ICWALK_FLAG_MINFILESIZE) && |
1100 | XFS_ISIZE(ip) < icw->icw_min_file_size) | |
a91bf992 DW |
1101 | return false; |
1102 | ||
1103 | return true; | |
1104 | } | |
1105 | ||
4d0bab3a DC |
1106 | /* |
1107 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
1108 | * many inodes as possible in a short period of time. It kicks itself every few | |
1109 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
02511a5a | 1110 | * goes low. |
4d0bab3a DC |
1111 | */ |
1112 | void | |
1113 | xfs_reclaim_worker( | |
1114 | struct work_struct *work) | |
1115 | { | |
1116 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
1117 | struct xfs_mount, m_reclaim_work); | |
4d0bab3a | 1118 | |
f1bc5c56 | 1119 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, NULL); |
4d0bab3a DC |
1120 | xfs_reclaim_work_queue(mp); |
1121 | } | |
1122 | ||
41176a68 BF |
1123 | STATIC int |
1124 | xfs_inode_free_eofblocks( | |
1125 | struct xfs_inode *ip, | |
b26b2bf1 | 1126 | struct xfs_icwalk *icw, |
0fa4a10a | 1127 | unsigned int *lockflags) |
41176a68 | 1128 | { |
390600f8 | 1129 | bool wait; |
390600f8 | 1130 | |
b26b2bf1 | 1131 | wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC); |
5400da7d | 1132 | |
ce2d3bbe DW |
1133 | if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS)) |
1134 | return 0; | |
1135 | ||
41176a68 BF |
1136 | /* |
1137 | * If the mapping is dirty the operation can block and wait for some | |
1138 | * time. Unless we are waiting, skip it. | |
1139 | */ | |
390600f8 | 1140 | if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) |
41176a68 BF |
1141 | return 0; |
1142 | ||
b26b2bf1 | 1143 | if (!xfs_icwalk_match(ip, icw)) |
a91bf992 | 1144 | return 0; |
3e3f9f58 | 1145 | |
a36b9261 BF |
1146 | /* |
1147 | * If the caller is waiting, return -EAGAIN to keep the background | |
1148 | * scanner moving and revisit the inode in a subsequent pass. | |
1149 | */ | |
c3155097 | 1150 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
390600f8 DW |
1151 | if (wait) |
1152 | return -EAGAIN; | |
1153 | return 0; | |
a36b9261 | 1154 | } |
0fa4a10a | 1155 | *lockflags |= XFS_IOLOCK_EXCL; |
390600f8 | 1156 | |
2b156ff8 DW |
1157 | if (xfs_can_free_eofblocks(ip, false)) |
1158 | return xfs_free_eofblocks(ip); | |
1159 | ||
1160 | /* inode could be preallocated or append-only */ | |
1161 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1162 | xfs_inode_clear_eofblocks_tag(ip); | |
1163 | return 0; | |
41176a68 BF |
1164 | } |
1165 | ||
83104d44 | 1166 | static void |
ce2d3bbe DW |
1167 | xfs_blockgc_set_iflag( |
1168 | struct xfs_inode *ip, | |
ce2d3bbe | 1169 | unsigned long iflag) |
27b52867 | 1170 | { |
ce2d3bbe DW |
1171 | struct xfs_mount *mp = ip->i_mount; |
1172 | struct xfs_perag *pag; | |
ce2d3bbe DW |
1173 | |
1174 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1175 | |
85a6e764 CH |
1176 | /* |
1177 | * Don't bother locking the AG and looking up in the radix trees | |
1178 | * if we already know that we have the tag set. | |
1179 | */ | |
ce2d3bbe | 1180 | if (ip->i_flags & iflag) |
85a6e764 CH |
1181 | return; |
1182 | spin_lock(&ip->i_flags_lock); | |
ce2d3bbe | 1183 | ip->i_flags |= iflag; |
85a6e764 CH |
1184 | spin_unlock(&ip->i_flags_lock); |
1185 | ||
27b52867 BF |
1186 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1187 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1188 | |
c076ae7a DW |
1189 | xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1190 | XFS_ICI_BLOCKGC_TAG); | |
27b52867 BF |
1191 | |
1192 | spin_unlock(&pag->pag_ici_lock); | |
1193 | xfs_perag_put(pag); | |
1194 | } | |
1195 | ||
1196 | void | |
83104d44 | 1197 | xfs_inode_set_eofblocks_tag( |
27b52867 | 1198 | xfs_inode_t *ip) |
83104d44 DW |
1199 | { |
1200 | trace_xfs_inode_set_eofblocks_tag(ip); | |
9669f51d | 1201 | return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1202 | } |
1203 | ||
1204 | static void | |
ce2d3bbe DW |
1205 | xfs_blockgc_clear_iflag( |
1206 | struct xfs_inode *ip, | |
1207 | unsigned long iflag) | |
27b52867 | 1208 | { |
ce2d3bbe DW |
1209 | struct xfs_mount *mp = ip->i_mount; |
1210 | struct xfs_perag *pag; | |
1211 | bool clear_tag; | |
1212 | ||
1213 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1214 | |
85a6e764 | 1215 | spin_lock(&ip->i_flags_lock); |
ce2d3bbe DW |
1216 | ip->i_flags &= ~iflag; |
1217 | clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0; | |
85a6e764 CH |
1218 | spin_unlock(&ip->i_flags_lock); |
1219 | ||
ce2d3bbe DW |
1220 | if (!clear_tag) |
1221 | return; | |
1222 | ||
27b52867 BF |
1223 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1224 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1225 | |
c076ae7a DW |
1226 | xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1227 | XFS_ICI_BLOCKGC_TAG); | |
27b52867 BF |
1228 | |
1229 | spin_unlock(&pag->pag_ici_lock); | |
1230 | xfs_perag_put(pag); | |
1231 | } | |
1232 | ||
83104d44 DW |
1233 | void |
1234 | xfs_inode_clear_eofblocks_tag( | |
1235 | xfs_inode_t *ip) | |
1236 | { | |
1237 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
ce2d3bbe | 1238 | return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1239 | } |
1240 | ||
1241 | /* | |
be78ff0e DW |
1242 | * Set ourselves up to free CoW blocks from this file. If it's already clean |
1243 | * then we can bail out quickly, but otherwise we must back off if the file | |
1244 | * is undergoing some kind of write. | |
83104d44 | 1245 | */ |
be78ff0e DW |
1246 | static bool |
1247 | xfs_prep_free_cowblocks( | |
51d62690 | 1248 | struct xfs_inode *ip) |
83104d44 | 1249 | { |
39937234 BF |
1250 | /* |
1251 | * Just clear the tag if we have an empty cow fork or none at all. It's | |
1252 | * possible the inode was fully unshared since it was originally tagged. | |
1253 | */ | |
51d62690 | 1254 | if (!xfs_inode_has_cow_data(ip)) { |
83104d44 DW |
1255 | trace_xfs_inode_free_cowblocks_invalid(ip); |
1256 | xfs_inode_clear_cowblocks_tag(ip); | |
be78ff0e | 1257 | return false; |
83104d44 DW |
1258 | } |
1259 | ||
1260 | /* | |
1261 | * If the mapping is dirty or under writeback we cannot touch the | |
1262 | * CoW fork. Leave it alone if we're in the midst of a directio. | |
1263 | */ | |
a1b7a4de CH |
1264 | if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) || |
1265 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) || | |
83104d44 DW |
1266 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) || |
1267 | atomic_read(&VFS_I(ip)->i_dio_count)) | |
be78ff0e DW |
1268 | return false; |
1269 | ||
1270 | return true; | |
1271 | } | |
1272 | ||
1273 | /* | |
1274 | * Automatic CoW Reservation Freeing | |
1275 | * | |
1276 | * These functions automatically garbage collect leftover CoW reservations | |
1277 | * that were made on behalf of a cowextsize hint when we start to run out | |
1278 | * of quota or when the reservations sit around for too long. If the file | |
1279 | * has dirty pages or is undergoing writeback, its CoW reservations will | |
1280 | * be retained. | |
1281 | * | |
1282 | * The actual garbage collection piggybacks off the same code that runs | |
1283 | * the speculative EOF preallocation garbage collector. | |
1284 | */ | |
1285 | STATIC int | |
1286 | xfs_inode_free_cowblocks( | |
1287 | struct xfs_inode *ip, | |
b26b2bf1 | 1288 | struct xfs_icwalk *icw, |
0fa4a10a | 1289 | unsigned int *lockflags) |
be78ff0e | 1290 | { |
f41a0716 | 1291 | bool wait; |
be78ff0e DW |
1292 | int ret = 0; |
1293 | ||
b26b2bf1 | 1294 | wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC); |
f41a0716 | 1295 | |
ce2d3bbe DW |
1296 | if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS)) |
1297 | return 0; | |
1298 | ||
51d62690 | 1299 | if (!xfs_prep_free_cowblocks(ip)) |
83104d44 DW |
1300 | return 0; |
1301 | ||
b26b2bf1 | 1302 | if (!xfs_icwalk_match(ip, icw)) |
a91bf992 | 1303 | return 0; |
83104d44 | 1304 | |
f41a0716 DW |
1305 | /* |
1306 | * If the caller is waiting, return -EAGAIN to keep the background | |
1307 | * scanner moving and revisit the inode in a subsequent pass. | |
1308 | */ | |
0fa4a10a DW |
1309 | if (!(*lockflags & XFS_IOLOCK_EXCL) && |
1310 | !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
f41a0716 DW |
1311 | if (wait) |
1312 | return -EAGAIN; | |
1313 | return 0; | |
1314 | } | |
0fa4a10a DW |
1315 | *lockflags |= XFS_IOLOCK_EXCL; |
1316 | ||
f41a0716 DW |
1317 | if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) { |
1318 | if (wait) | |
0fa4a10a DW |
1319 | return -EAGAIN; |
1320 | return 0; | |
f41a0716 | 1321 | } |
0fa4a10a | 1322 | *lockflags |= XFS_MMAPLOCK_EXCL; |
83104d44 | 1323 | |
be78ff0e DW |
1324 | /* |
1325 | * Check again, nobody else should be able to dirty blocks or change | |
1326 | * the reflink iflag now that we have the first two locks held. | |
1327 | */ | |
51d62690 | 1328 | if (xfs_prep_free_cowblocks(ip)) |
be78ff0e | 1329 | ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false); |
83104d44 DW |
1330 | return ret; |
1331 | } | |
1332 | ||
83104d44 DW |
1333 | void |
1334 | xfs_inode_set_cowblocks_tag( | |
1335 | xfs_inode_t *ip) | |
1336 | { | |
7b7381f0 | 1337 | trace_xfs_inode_set_cowblocks_tag(ip); |
9669f51d | 1338 | return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 DW |
1339 | } |
1340 | ||
1341 | void | |
1342 | xfs_inode_clear_cowblocks_tag( | |
1343 | xfs_inode_t *ip) | |
1344 | { | |
7b7381f0 | 1345 | trace_xfs_inode_clear_cowblocks_tag(ip); |
ce2d3bbe | 1346 | return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 | 1347 | } |
d6b636eb DW |
1348 | |
1349 | /* Disable post-EOF and CoW block auto-reclamation. */ | |
1350 | void | |
c9a6526f | 1351 | xfs_blockgc_stop( |
d6b636eb DW |
1352 | struct xfs_mount *mp) |
1353 | { | |
894ecacf DW |
1354 | struct xfs_perag *pag; |
1355 | xfs_agnumber_t agno; | |
1356 | ||
6f649091 DW |
1357 | if (!xfs_clear_blockgc_enabled(mp)) |
1358 | return; | |
1359 | ||
1360 | for_each_perag(mp, agno, pag) | |
894ecacf | 1361 | cancel_delayed_work_sync(&pag->pag_blockgc_work); |
6f649091 | 1362 | trace_xfs_blockgc_stop(mp, __return_address); |
d6b636eb DW |
1363 | } |
1364 | ||
1365 | /* Enable post-EOF and CoW block auto-reclamation. */ | |
1366 | void | |
c9a6526f | 1367 | xfs_blockgc_start( |
d6b636eb DW |
1368 | struct xfs_mount *mp) |
1369 | { | |
894ecacf DW |
1370 | struct xfs_perag *pag; |
1371 | xfs_agnumber_t agno; | |
1372 | ||
6f649091 DW |
1373 | if (xfs_set_blockgc_enabled(mp)) |
1374 | return; | |
1375 | ||
1376 | trace_xfs_blockgc_start(mp, __return_address); | |
894ecacf DW |
1377 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) |
1378 | xfs_blockgc_queue(pag); | |
d6b636eb | 1379 | } |
3d4feec0 | 1380 | |
d20d5edc DW |
1381 | /* Don't try to run block gc on an inode that's in any of these states. */ |
1382 | #define XFS_BLOCKGC_NOGRAB_IFLAGS (XFS_INEW | \ | |
ab23a776 DC |
1383 | XFS_NEED_INACTIVE | \ |
1384 | XFS_INACTIVATING | \ | |
d20d5edc DW |
1385 | XFS_IRECLAIMABLE | \ |
1386 | XFS_IRECLAIM) | |
df600197 | 1387 | /* |
b9baaef4 DW |
1388 | * Decide if the given @ip is eligible for garbage collection of speculative |
1389 | * preallocations, and grab it if so. Returns true if it's ready to go or | |
1390 | * false if we should just ignore it. | |
df600197 DW |
1391 | */ |
1392 | static bool | |
b9baaef4 | 1393 | xfs_blockgc_igrab( |
7fdff526 | 1394 | struct xfs_inode *ip) |
df600197 DW |
1395 | { |
1396 | struct inode *inode = VFS_I(ip); | |
df600197 DW |
1397 | |
1398 | ASSERT(rcu_read_lock_held()); | |
1399 | ||
1400 | /* Check for stale RCU freed inode */ | |
1401 | spin_lock(&ip->i_flags_lock); | |
1402 | if (!ip->i_ino) | |
1403 | goto out_unlock_noent; | |
1404 | ||
d20d5edc | 1405 | if (ip->i_flags & XFS_BLOCKGC_NOGRAB_IFLAGS) |
df600197 DW |
1406 | goto out_unlock_noent; |
1407 | spin_unlock(&ip->i_flags_lock); | |
1408 | ||
1409 | /* nothing to sync during shutdown */ | |
75c8c50f | 1410 | if (xfs_is_shutdown(ip->i_mount)) |
df600197 DW |
1411 | return false; |
1412 | ||
1413 | /* If we can't grab the inode, it must on it's way to reclaim. */ | |
1414 | if (!igrab(inode)) | |
1415 | return false; | |
1416 | ||
1417 | /* inode is valid */ | |
1418 | return true; | |
1419 | ||
1420 | out_unlock_noent: | |
1421 | spin_unlock(&ip->i_flags_lock); | |
1422 | return false; | |
1423 | } | |
1424 | ||
41956753 DW |
1425 | /* Scan one incore inode for block preallocations that we can remove. */ |
1426 | static int | |
1427 | xfs_blockgc_scan_inode( | |
1428 | struct xfs_inode *ip, | |
b26b2bf1 | 1429 | struct xfs_icwalk *icw) |
85c5b270 | 1430 | { |
0fa4a10a | 1431 | unsigned int lockflags = 0; |
85c5b270 DW |
1432 | int error; |
1433 | ||
b26b2bf1 | 1434 | error = xfs_inode_free_eofblocks(ip, icw, &lockflags); |
85c5b270 | 1435 | if (error) |
0fa4a10a | 1436 | goto unlock; |
85c5b270 | 1437 | |
b26b2bf1 | 1438 | error = xfs_inode_free_cowblocks(ip, icw, &lockflags); |
0fa4a10a DW |
1439 | unlock: |
1440 | if (lockflags) | |
1441 | xfs_iunlock(ip, lockflags); | |
594ab00b | 1442 | xfs_irele(ip); |
0fa4a10a | 1443 | return error; |
85c5b270 DW |
1444 | } |
1445 | ||
9669f51d DW |
1446 | /* Background worker that trims preallocated space. */ |
1447 | void | |
1448 | xfs_blockgc_worker( | |
1449 | struct work_struct *work) | |
1450 | { | |
894ecacf DW |
1451 | struct xfs_perag *pag = container_of(to_delayed_work(work), |
1452 | struct xfs_perag, pag_blockgc_work); | |
1453 | struct xfs_mount *mp = pag->pag_mount; | |
9669f51d DW |
1454 | int error; |
1455 | ||
6f649091 DW |
1456 | trace_xfs_blockgc_worker(mp, __return_address); |
1457 | ||
f427cf5c | 1458 | error = xfs_icwalk_ag(pag, XFS_ICWALK_BLOCKGC, NULL); |
9669f51d | 1459 | if (error) |
894ecacf DW |
1460 | xfs_info(mp, "AG %u preallocation gc worker failed, err=%d", |
1461 | pag->pag_agno, error); | |
894ecacf | 1462 | xfs_blockgc_queue(pag); |
9669f51d DW |
1463 | } |
1464 | ||
85c5b270 | 1465 | /* |
2eb66502 DW |
1466 | * Try to free space in the filesystem by purging inactive inodes, eofblocks |
1467 | * and cowblocks. | |
85c5b270 DW |
1468 | */ |
1469 | int | |
1470 | xfs_blockgc_free_space( | |
1471 | struct xfs_mount *mp, | |
b26b2bf1 | 1472 | struct xfs_icwalk *icw) |
85c5b270 | 1473 | { |
2eb66502 DW |
1474 | int error; |
1475 | ||
b26b2bf1 | 1476 | trace_xfs_blockgc_free_space(mp, icw, _RET_IP_); |
85c5b270 | 1477 | |
2eb66502 DW |
1478 | error = xfs_icwalk(mp, XFS_ICWALK_BLOCKGC, icw); |
1479 | if (error) | |
1480 | return error; | |
1481 | ||
1482 | xfs_inodegc_flush(mp); | |
1483 | return 0; | |
85c5b270 DW |
1484 | } |
1485 | ||
e8d04c2a DW |
1486 | /* |
1487 | * Reclaim all the free space that we can by scheduling the background blockgc | |
1488 | * and inodegc workers immediately and waiting for them all to clear. | |
1489 | */ | |
1490 | void | |
1491 | xfs_blockgc_flush_all( | |
1492 | struct xfs_mount *mp) | |
1493 | { | |
1494 | struct xfs_perag *pag; | |
1495 | xfs_agnumber_t agno; | |
1496 | ||
1497 | trace_xfs_blockgc_flush_all(mp, __return_address); | |
1498 | ||
1499 | /* | |
1500 | * For each blockgc worker, move its queue time up to now. If it | |
1501 | * wasn't queued, it will not be requeued. Then flush whatever's | |
1502 | * left. | |
1503 | */ | |
1504 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1505 | mod_delayed_work(pag->pag_mount->m_blockgc_wq, | |
1506 | &pag->pag_blockgc_work, 0); | |
1507 | ||
1508 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1509 | flush_delayed_work(&pag->pag_blockgc_work); | |
1510 | ||
1511 | xfs_inodegc_flush(mp); | |
1512 | } | |
1513 | ||
3d4feec0 | 1514 | /* |
c237dd7c DW |
1515 | * Run cow/eofblocks scans on the supplied dquots. We don't know exactly which |
1516 | * quota caused an allocation failure, so we make a best effort by including | |
1517 | * each quota under low free space conditions (less than 1% free space) in the | |
1518 | * scan. | |
111068f8 DW |
1519 | * |
1520 | * Callers must not hold any inode's ILOCK. If requesting a synchronous scan | |
2d53f66b | 1521 | * (XFS_ICWALK_FLAG_SYNC), the caller also must not hold any inode's IOLOCK or |
111068f8 | 1522 | * MMAPLOCK. |
3d4feec0 | 1523 | */ |
111068f8 | 1524 | int |
c237dd7c DW |
1525 | xfs_blockgc_free_dquots( |
1526 | struct xfs_mount *mp, | |
1527 | struct xfs_dquot *udqp, | |
1528 | struct xfs_dquot *gdqp, | |
1529 | struct xfs_dquot *pdqp, | |
2d53f66b | 1530 | unsigned int iwalk_flags) |
3d4feec0 | 1531 | { |
b26b2bf1 | 1532 | struct xfs_icwalk icw = {0}; |
3d4feec0 DW |
1533 | bool do_work = false; |
1534 | ||
c237dd7c DW |
1535 | if (!udqp && !gdqp && !pdqp) |
1536 | return 0; | |
1537 | ||
3d4feec0 | 1538 | /* |
111068f8 DW |
1539 | * Run a scan to free blocks using the union filter to cover all |
1540 | * applicable quotas in a single scan. | |
3d4feec0 | 1541 | */ |
b26b2bf1 | 1542 | icw.icw_flags = XFS_ICWALK_FLAG_UNION | iwalk_flags; |
3d4feec0 | 1543 | |
c237dd7c | 1544 | if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) { |
b26b2bf1 DW |
1545 | icw.icw_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id); |
1546 | icw.icw_flags |= XFS_ICWALK_FLAG_UID; | |
c237dd7c | 1547 | do_work = true; |
3d4feec0 DW |
1548 | } |
1549 | ||
c237dd7c | 1550 | if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) { |
b26b2bf1 DW |
1551 | icw.icw_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id); |
1552 | icw.icw_flags |= XFS_ICWALK_FLAG_GID; | |
c237dd7c | 1553 | do_work = true; |
3d4feec0 DW |
1554 | } |
1555 | ||
c237dd7c | 1556 | if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) { |
b26b2bf1 DW |
1557 | icw.icw_prid = pdqp->q_id; |
1558 | icw.icw_flags |= XFS_ICWALK_FLAG_PRID; | |
c237dd7c | 1559 | do_work = true; |
3d4feec0 DW |
1560 | } |
1561 | ||
1562 | if (!do_work) | |
111068f8 | 1563 | return 0; |
3d4feec0 | 1564 | |
b26b2bf1 | 1565 | return xfs_blockgc_free_space(mp, &icw); |
c237dd7c DW |
1566 | } |
1567 | ||
1568 | /* Run cow/eofblocks scans on the quotas attached to the inode. */ | |
1569 | int | |
1570 | xfs_blockgc_free_quota( | |
1571 | struct xfs_inode *ip, | |
2d53f66b | 1572 | unsigned int iwalk_flags) |
c237dd7c DW |
1573 | { |
1574 | return xfs_blockgc_free_dquots(ip->i_mount, | |
1575 | xfs_inode_dquot(ip, XFS_DQTYPE_USER), | |
1576 | xfs_inode_dquot(ip, XFS_DQTYPE_GROUP), | |
2d53f66b | 1577 | xfs_inode_dquot(ip, XFS_DQTYPE_PROJ), iwalk_flags); |
3d4feec0 | 1578 | } |
df600197 DW |
1579 | |
1580 | /* XFS Inode Cache Walking Code */ | |
1581 | ||
f1bc5c56 DW |
1582 | /* |
1583 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
1584 | * radix tree lookups to a minimum. The batch size is a trade off between | |
1585 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
1586 | * be too greedy. | |
1587 | */ | |
1588 | #define XFS_LOOKUP_BATCH 32 | |
1589 | ||
1590 | ||
b9baaef4 DW |
1591 | /* |
1592 | * Decide if we want to grab this inode in anticipation of doing work towards | |
594ab00b | 1593 | * the goal. |
b9baaef4 DW |
1594 | */ |
1595 | static inline bool | |
1596 | xfs_icwalk_igrab( | |
1597 | enum xfs_icwalk_goal goal, | |
9492750a | 1598 | struct xfs_inode *ip, |
b26b2bf1 | 1599 | struct xfs_icwalk *icw) |
b9baaef4 DW |
1600 | { |
1601 | switch (goal) { | |
b9baaef4 | 1602 | case XFS_ICWALK_BLOCKGC: |
7fdff526 | 1603 | return xfs_blockgc_igrab(ip); |
f1bc5c56 | 1604 | case XFS_ICWALK_RECLAIM: |
b26b2bf1 | 1605 | return xfs_reclaim_igrab(ip, icw); |
b9baaef4 DW |
1606 | default: |
1607 | return false; | |
1608 | } | |
1609 | } | |
1610 | ||
594ab00b DW |
1611 | /* |
1612 | * Process an inode. Each processing function must handle any state changes | |
1613 | * made by the icwalk igrab function. Return -EAGAIN to skip an inode. | |
1614 | */ | |
f427cf5c DW |
1615 | static inline int |
1616 | xfs_icwalk_process_inode( | |
1617 | enum xfs_icwalk_goal goal, | |
1618 | struct xfs_inode *ip, | |
f1bc5c56 | 1619 | struct xfs_perag *pag, |
b26b2bf1 | 1620 | struct xfs_icwalk *icw) |
f427cf5c | 1621 | { |
594ab00b | 1622 | int error = 0; |
f427cf5c DW |
1623 | |
1624 | switch (goal) { | |
f427cf5c | 1625 | case XFS_ICWALK_BLOCKGC: |
b26b2bf1 | 1626 | error = xfs_blockgc_scan_inode(ip, icw); |
f427cf5c | 1627 | break; |
f1bc5c56 DW |
1628 | case XFS_ICWALK_RECLAIM: |
1629 | xfs_reclaim_inode(ip, pag); | |
1630 | break; | |
f427cf5c | 1631 | } |
f427cf5c DW |
1632 | return error; |
1633 | } | |
1634 | ||
df600197 | 1635 | /* |
f427cf5c DW |
1636 | * For a given per-AG structure @pag and a goal, grab qualifying inodes and |
1637 | * process them in some manner. | |
df600197 DW |
1638 | */ |
1639 | static int | |
c1115c0c | 1640 | xfs_icwalk_ag( |
df600197 | 1641 | struct xfs_perag *pag, |
f427cf5c | 1642 | enum xfs_icwalk_goal goal, |
b26b2bf1 | 1643 | struct xfs_icwalk *icw) |
df600197 DW |
1644 | { |
1645 | struct xfs_mount *mp = pag->pag_mount; | |
1646 | uint32_t first_index; | |
1647 | int last_error = 0; | |
1648 | int skipped; | |
1649 | bool done; | |
1650 | int nr_found; | |
1651 | ||
1652 | restart: | |
1653 | done = false; | |
1654 | skipped = 0; | |
f1bc5c56 DW |
1655 | if (goal == XFS_ICWALK_RECLAIM) |
1656 | first_index = READ_ONCE(pag->pag_ici_reclaim_cursor); | |
1657 | else | |
1658 | first_index = 0; | |
df600197 DW |
1659 | nr_found = 0; |
1660 | do { | |
1661 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; | |
1662 | int error = 0; | |
1663 | int i; | |
1664 | ||
1665 | rcu_read_lock(); | |
1666 | ||
a437b9b4 CH |
1667 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, |
1668 | (void **) batch, first_index, | |
1669 | XFS_LOOKUP_BATCH, goal); | |
df600197 | 1670 | if (!nr_found) { |
f1bc5c56 | 1671 | done = true; |
df600197 DW |
1672 | rcu_read_unlock(); |
1673 | break; | |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * Grab the inodes before we drop the lock. if we found | |
1678 | * nothing, nr == 0 and the loop will be skipped. | |
1679 | */ | |
1680 | for (i = 0; i < nr_found; i++) { | |
1681 | struct xfs_inode *ip = batch[i]; | |
1682 | ||
b26b2bf1 | 1683 | if (done || !xfs_icwalk_igrab(goal, ip, icw)) |
df600197 DW |
1684 | batch[i] = NULL; |
1685 | ||
1686 | /* | |
1687 | * Update the index for the next lookup. Catch | |
1688 | * overflows into the next AG range which can occur if | |
1689 | * we have inodes in the last block of the AG and we | |
1690 | * are currently pointing to the last inode. | |
1691 | * | |
1692 | * Because we may see inodes that are from the wrong AG | |
1693 | * due to RCU freeing and reallocation, only update the | |
1694 | * index if it lies in this AG. It was a race that lead | |
1695 | * us to see this inode, so another lookup from the | |
1696 | * same index will not find it again. | |
1697 | */ | |
1698 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) | |
1699 | continue; | |
1700 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
1701 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1702 | done = true; | |
1703 | } | |
1704 | ||
1705 | /* unlock now we've grabbed the inodes. */ | |
1706 | rcu_read_unlock(); | |
1707 | ||
1708 | for (i = 0; i < nr_found; i++) { | |
1709 | if (!batch[i]) | |
1710 | continue; | |
f1bc5c56 | 1711 | error = xfs_icwalk_process_inode(goal, batch[i], pag, |
b26b2bf1 | 1712 | icw); |
df600197 DW |
1713 | if (error == -EAGAIN) { |
1714 | skipped++; | |
1715 | continue; | |
1716 | } | |
1717 | if (error && last_error != -EFSCORRUPTED) | |
1718 | last_error = error; | |
1719 | } | |
1720 | ||
1721 | /* bail out if the filesystem is corrupted. */ | |
1722 | if (error == -EFSCORRUPTED) | |
1723 | break; | |
1724 | ||
1725 | cond_resched(); | |
1726 | ||
b26b2bf1 DW |
1727 | if (icw && (icw->icw_flags & XFS_ICWALK_FLAG_SCAN_LIMIT)) { |
1728 | icw->icw_scan_limit -= XFS_LOOKUP_BATCH; | |
1729 | if (icw->icw_scan_limit <= 0) | |
f1bc5c56 DW |
1730 | break; |
1731 | } | |
df600197 DW |
1732 | } while (nr_found && !done); |
1733 | ||
f1bc5c56 DW |
1734 | if (goal == XFS_ICWALK_RECLAIM) { |
1735 | if (done) | |
1736 | first_index = 0; | |
1737 | WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index); | |
1738 | } | |
1739 | ||
df600197 DW |
1740 | if (skipped) { |
1741 | delay(1); | |
1742 | goto restart; | |
1743 | } | |
1744 | return last_error; | |
1745 | } | |
1746 | ||
f427cf5c | 1747 | /* Walk all incore inodes to achieve a given goal. */ |
df600197 | 1748 | static int |
c1115c0c | 1749 | xfs_icwalk( |
df600197 | 1750 | struct xfs_mount *mp, |
f427cf5c | 1751 | enum xfs_icwalk_goal goal, |
b26b2bf1 | 1752 | struct xfs_icwalk *icw) |
df600197 DW |
1753 | { |
1754 | struct xfs_perag *pag; | |
1755 | int error = 0; | |
1756 | int last_error = 0; | |
a437b9b4 | 1757 | xfs_agnumber_t agno; |
df600197 | 1758 | |
a437b9b4 | 1759 | for_each_perag_tag(mp, agno, pag, goal) { |
b26b2bf1 | 1760 | error = xfs_icwalk_ag(pag, goal, icw); |
df600197 DW |
1761 | if (error) { |
1762 | last_error = error; | |
a437b9b4 CH |
1763 | if (error == -EFSCORRUPTED) { |
1764 | xfs_perag_put(pag); | |
df600197 | 1765 | break; |
a437b9b4 | 1766 | } |
df600197 DW |
1767 | } |
1768 | } | |
1769 | return last_error; | |
2d53f66b | 1770 | BUILD_BUG_ON(XFS_ICWALK_PRIVATE_FLAGS & XFS_ICWALK_FLAGS_VALID); |
df600197 | 1771 | } |
c6c2066d DW |
1772 | |
1773 | #ifdef DEBUG | |
1774 | static void | |
1775 | xfs_check_delalloc( | |
1776 | struct xfs_inode *ip, | |
1777 | int whichfork) | |
1778 | { | |
732436ef | 1779 | struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork); |
c6c2066d DW |
1780 | struct xfs_bmbt_irec got; |
1781 | struct xfs_iext_cursor icur; | |
1782 | ||
1783 | if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got)) | |
1784 | return; | |
1785 | do { | |
1786 | if (isnullstartblock(got.br_startblock)) { | |
1787 | xfs_warn(ip->i_mount, | |
1788 | "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]", | |
1789 | ip->i_ino, | |
1790 | whichfork == XFS_DATA_FORK ? "data" : "cow", | |
1791 | got.br_startoff, got.br_blockcount); | |
1792 | } | |
1793 | } while (xfs_iext_next_extent(ifp, &icur, &got)); | |
1794 | } | |
1795 | #else | |
1796 | #define xfs_check_delalloc(ip, whichfork) do { } while (0) | |
1797 | #endif | |
1798 | ||
ab23a776 DC |
1799 | /* Schedule the inode for reclaim. */ |
1800 | static void | |
1801 | xfs_inodegc_set_reclaimable( | |
c6c2066d DW |
1802 | struct xfs_inode *ip) |
1803 | { | |
1804 | struct xfs_mount *mp = ip->i_mount; | |
1805 | struct xfs_perag *pag; | |
c6c2066d | 1806 | |
75c8c50f | 1807 | if (!xfs_is_shutdown(mp) && ip->i_delayed_blks) { |
c6c2066d DW |
1808 | xfs_check_delalloc(ip, XFS_DATA_FORK); |
1809 | xfs_check_delalloc(ip, XFS_COW_FORK); | |
1810 | ASSERT(0); | |
1811 | } | |
1812 | ||
c6c2066d DW |
1813 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1814 | spin_lock(&pag->pag_ici_lock); | |
1815 | spin_lock(&ip->i_flags_lock); | |
1816 | ||
ab23a776 DC |
1817 | trace_xfs_inode_set_reclaimable(ip); |
1818 | ip->i_flags &= ~(XFS_NEED_INACTIVE | XFS_INACTIVATING); | |
1819 | ip->i_flags |= XFS_IRECLAIMABLE; | |
c6c2066d DW |
1820 | xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1821 | XFS_ICI_RECLAIM_TAG); | |
c6c2066d DW |
1822 | |
1823 | spin_unlock(&ip->i_flags_lock); | |
1824 | spin_unlock(&pag->pag_ici_lock); | |
1825 | xfs_perag_put(pag); | |
1826 | } | |
ab23a776 DC |
1827 | |
1828 | /* | |
1829 | * Free all speculative preallocations and possibly even the inode itself. | |
1830 | * This is the last chance to make changes to an otherwise unreferenced file | |
1831 | * before incore reclamation happens. | |
1832 | */ | |
1833 | static void | |
1834 | xfs_inodegc_inactivate( | |
1835 | struct xfs_inode *ip) | |
1836 | { | |
1837 | trace_xfs_inode_inactivating(ip); | |
1838 | xfs_inactive(ip); | |
1839 | xfs_inodegc_set_reclaimable(ip); | |
1840 | } | |
1841 | ||
1842 | void | |
1843 | xfs_inodegc_worker( | |
1844 | struct work_struct *work) | |
1845 | { | |
7cf2b0f9 DC |
1846 | struct xfs_inodegc *gc = container_of(to_delayed_work(work), |
1847 | struct xfs_inodegc, work); | |
ab23a776 DC |
1848 | struct llist_node *node = llist_del_all(&gc->list); |
1849 | struct xfs_inode *ip, *n; | |
1850 | ||
1851 | WRITE_ONCE(gc->items, 0); | |
1852 | ||
1853 | if (!node) | |
1854 | return; | |
1855 | ||
1856 | ip = llist_entry(node, struct xfs_inode, i_gclist); | |
40b1de00 | 1857 | trace_xfs_inodegc_worker(ip->i_mount, READ_ONCE(gc->shrinker_hits)); |
ab23a776 | 1858 | |
40b1de00 | 1859 | WRITE_ONCE(gc->shrinker_hits, 0); |
ab23a776 DC |
1860 | llist_for_each_entry_safe(ip, n, node, i_gclist) { |
1861 | xfs_iflags_set(ip, XFS_INACTIVATING); | |
1862 | xfs_inodegc_inactivate(ip); | |
1863 | } | |
1864 | } | |
1865 | ||
1866 | /* | |
5e672cd6 DC |
1867 | * Expedite all pending inodegc work to run immediately. This does not wait for |
1868 | * completion of the work. | |
ab23a776 DC |
1869 | */ |
1870 | void | |
5e672cd6 | 1871 | xfs_inodegc_push( |
ab23a776 DC |
1872 | struct xfs_mount *mp) |
1873 | { | |
ab23a776 DC |
1874 | if (!xfs_is_inodegc_enabled(mp)) |
1875 | return; | |
5e672cd6 DC |
1876 | trace_xfs_inodegc_push(mp, __return_address); |
1877 | xfs_inodegc_queue_all(mp); | |
1878 | } | |
ab23a776 | 1879 | |
5e672cd6 DC |
1880 | /* |
1881 | * Force all currently queued inode inactivation work to run immediately and | |
1882 | * wait for the work to finish. | |
1883 | */ | |
1884 | void | |
1885 | xfs_inodegc_flush( | |
1886 | struct xfs_mount *mp) | |
1887 | { | |
1888 | xfs_inodegc_push(mp); | |
ab23a776 | 1889 | trace_xfs_inodegc_flush(mp, __return_address); |
6191cf3a | 1890 | flush_workqueue(mp->m_inodegc_wq); |
ab23a776 DC |
1891 | } |
1892 | ||
1893 | /* | |
1894 | * Flush all the pending work and then disable the inode inactivation background | |
1895 | * workers and wait for them to stop. | |
1896 | */ | |
1897 | void | |
1898 | xfs_inodegc_stop( | |
1899 | struct xfs_mount *mp) | |
1900 | { | |
ab23a776 DC |
1901 | if (!xfs_clear_inodegc_enabled(mp)) |
1902 | return; | |
1903 | ||
1904 | xfs_inodegc_queue_all(mp); | |
6191cf3a | 1905 | drain_workqueue(mp->m_inodegc_wq); |
ab23a776 | 1906 | |
ab23a776 DC |
1907 | trace_xfs_inodegc_stop(mp, __return_address); |
1908 | } | |
1909 | ||
1910 | /* | |
1911 | * Enable the inode inactivation background workers and schedule deferred inode | |
1912 | * inactivation work if there is any. | |
1913 | */ | |
1914 | void | |
1915 | xfs_inodegc_start( | |
1916 | struct xfs_mount *mp) | |
1917 | { | |
1918 | if (xfs_set_inodegc_enabled(mp)) | |
1919 | return; | |
1920 | ||
1921 | trace_xfs_inodegc_start(mp, __return_address); | |
1922 | xfs_inodegc_queue_all(mp); | |
1923 | } | |
1924 | ||
65f03d86 DW |
1925 | #ifdef CONFIG_XFS_RT |
1926 | static inline bool | |
1927 | xfs_inodegc_want_queue_rt_file( | |
1928 | struct xfs_inode *ip) | |
1929 | { | |
1930 | struct xfs_mount *mp = ip->i_mount; | |
65f03d86 DW |
1931 | |
1932 | if (!XFS_IS_REALTIME_INODE(ip)) | |
1933 | return false; | |
1934 | ||
2229276c DW |
1935 | if (__percpu_counter_compare(&mp->m_frextents, |
1936 | mp->m_low_rtexts[XFS_LOWSP_5_PCNT], | |
1937 | XFS_FDBLOCKS_BATCH) < 0) | |
1938 | return true; | |
1939 | ||
1940 | return false; | |
65f03d86 DW |
1941 | } |
1942 | #else | |
1943 | # define xfs_inodegc_want_queue_rt_file(ip) (false) | |
1944 | #endif /* CONFIG_XFS_RT */ | |
1945 | ||
ab23a776 DC |
1946 | /* |
1947 | * Schedule the inactivation worker when: | |
1948 | * | |
1949 | * - We've accumulated more than one inode cluster buffer's worth of inodes. | |
7d6f07d2 | 1950 | * - There is less than 5% free space left. |
108523b8 | 1951 | * - Any of the quotas for this inode are near an enforcement limit. |
ab23a776 DC |
1952 | */ |
1953 | static inline bool | |
1954 | xfs_inodegc_want_queue_work( | |
1955 | struct xfs_inode *ip, | |
1956 | unsigned int items) | |
1957 | { | |
1958 | struct xfs_mount *mp = ip->i_mount; | |
1959 | ||
1960 | if (items > mp->m_ino_geo.inodes_per_cluster) | |
1961 | return true; | |
1962 | ||
7d6f07d2 DW |
1963 | if (__percpu_counter_compare(&mp->m_fdblocks, |
1964 | mp->m_low_space[XFS_LOWSP_5_PCNT], | |
1965 | XFS_FDBLOCKS_BATCH) < 0) | |
1966 | return true; | |
1967 | ||
65f03d86 DW |
1968 | if (xfs_inodegc_want_queue_rt_file(ip)) |
1969 | return true; | |
1970 | ||
108523b8 DW |
1971 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_USER)) |
1972 | return true; | |
1973 | ||
1974 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_GROUP)) | |
1975 | return true; | |
1976 | ||
1977 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_PROJ)) | |
1978 | return true; | |
1979 | ||
ab23a776 DC |
1980 | return false; |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * Upper bound on the number of inodes in each AG that can be queued for | |
1985 | * inactivation at any given time, to avoid monopolizing the workqueue. | |
1986 | */ | |
1987 | #define XFS_INODEGC_MAX_BACKLOG (4 * XFS_INODES_PER_CHUNK) | |
1988 | ||
1989 | /* | |
1990 | * Make the frontend wait for inactivations when: | |
1991 | * | |
40b1de00 | 1992 | * - Memory shrinkers queued the inactivation worker and it hasn't finished. |
ab23a776 DC |
1993 | * - The queue depth exceeds the maximum allowable percpu backlog. |
1994 | * | |
1995 | * Note: If the current thread is running a transaction, we don't ever want to | |
1996 | * wait for other transactions because that could introduce a deadlock. | |
1997 | */ | |
1998 | static inline bool | |
1999 | xfs_inodegc_want_flush_work( | |
2000 | struct xfs_inode *ip, | |
40b1de00 DW |
2001 | unsigned int items, |
2002 | unsigned int shrinker_hits) | |
ab23a776 DC |
2003 | { |
2004 | if (current->journal_info) | |
2005 | return false; | |
2006 | ||
40b1de00 DW |
2007 | if (shrinker_hits > 0) |
2008 | return true; | |
2009 | ||
ab23a776 DC |
2010 | if (items > XFS_INODEGC_MAX_BACKLOG) |
2011 | return true; | |
2012 | ||
2013 | return false; | |
2014 | } | |
2015 | ||
2016 | /* | |
2017 | * Queue a background inactivation worker if there are inodes that need to be | |
2018 | * inactivated and higher level xfs code hasn't disabled the background | |
2019 | * workers. | |
2020 | */ | |
2021 | static void | |
2022 | xfs_inodegc_queue( | |
2023 | struct xfs_inode *ip) | |
2024 | { | |
2025 | struct xfs_mount *mp = ip->i_mount; | |
2026 | struct xfs_inodegc *gc; | |
2027 | int items; | |
40b1de00 | 2028 | unsigned int shrinker_hits; |
7cf2b0f9 | 2029 | unsigned long queue_delay = 1; |
ab23a776 DC |
2030 | |
2031 | trace_xfs_inode_set_need_inactive(ip); | |
2032 | spin_lock(&ip->i_flags_lock); | |
2033 | ip->i_flags |= XFS_NEED_INACTIVE; | |
2034 | spin_unlock(&ip->i_flags_lock); | |
2035 | ||
2036 | gc = get_cpu_ptr(mp->m_inodegc); | |
2037 | llist_add(&ip->i_gclist, &gc->list); | |
2038 | items = READ_ONCE(gc->items); | |
2039 | WRITE_ONCE(gc->items, items + 1); | |
40b1de00 | 2040 | shrinker_hits = READ_ONCE(gc->shrinker_hits); |
ab23a776 | 2041 | |
7cf2b0f9 DC |
2042 | /* |
2043 | * We queue the work while holding the current CPU so that the work | |
2044 | * is scheduled to run on this CPU. | |
2045 | */ | |
2046 | if (!xfs_is_inodegc_enabled(mp)) { | |
2047 | put_cpu_ptr(gc); | |
ab23a776 | 2048 | return; |
ab23a776 DC |
2049 | } |
2050 | ||
7cf2b0f9 DC |
2051 | if (xfs_inodegc_want_queue_work(ip, items)) |
2052 | queue_delay = 0; | |
2053 | ||
2054 | trace_xfs_inodegc_queue(mp, __return_address); | |
2055 | mod_delayed_work(mp->m_inodegc_wq, &gc->work, queue_delay); | |
2056 | put_cpu_ptr(gc); | |
2057 | ||
40b1de00 | 2058 | if (xfs_inodegc_want_flush_work(ip, items, shrinker_hits)) { |
ab23a776 | 2059 | trace_xfs_inodegc_throttle(mp, __return_address); |
7cf2b0f9 | 2060 | flush_delayed_work(&gc->work); |
ab23a776 DC |
2061 | } |
2062 | } | |
2063 | ||
2064 | /* | |
2065 | * Fold the dead CPU inodegc queue into the current CPUs queue. | |
2066 | */ | |
2067 | void | |
2068 | xfs_inodegc_cpu_dead( | |
2069 | struct xfs_mount *mp, | |
2070 | unsigned int dead_cpu) | |
2071 | { | |
2072 | struct xfs_inodegc *dead_gc, *gc; | |
2073 | struct llist_node *first, *last; | |
2074 | unsigned int count = 0; | |
2075 | ||
2076 | dead_gc = per_cpu_ptr(mp->m_inodegc, dead_cpu); | |
7cf2b0f9 | 2077 | cancel_delayed_work_sync(&dead_gc->work); |
ab23a776 DC |
2078 | |
2079 | if (llist_empty(&dead_gc->list)) | |
2080 | return; | |
2081 | ||
2082 | first = dead_gc->list.first; | |
2083 | last = first; | |
2084 | while (last->next) { | |
2085 | last = last->next; | |
2086 | count++; | |
2087 | } | |
2088 | dead_gc->list.first = NULL; | |
2089 | dead_gc->items = 0; | |
2090 | ||
2091 | /* Add pending work to current CPU */ | |
2092 | gc = get_cpu_ptr(mp->m_inodegc); | |
2093 | llist_add_batch(first, last, &gc->list); | |
2094 | count += READ_ONCE(gc->items); | |
2095 | WRITE_ONCE(gc->items, count); | |
ab23a776 DC |
2096 | |
2097 | if (xfs_is_inodegc_enabled(mp)) { | |
2098 | trace_xfs_inodegc_queue(mp, __return_address); | |
7cf2b0f9 | 2099 | mod_delayed_work(mp->m_inodegc_wq, &gc->work, 0); |
ab23a776 | 2100 | } |
7cf2b0f9 | 2101 | put_cpu_ptr(gc); |
ab23a776 DC |
2102 | } |
2103 | ||
2104 | /* | |
2105 | * We set the inode flag atomically with the radix tree tag. Once we get tag | |
2106 | * lookups on the radix tree, this inode flag can go away. | |
2107 | * | |
2108 | * We always use background reclaim here because even if the inode is clean, it | |
2109 | * still may be under IO and hence we have wait for IO completion to occur | |
2110 | * before we can reclaim the inode. The background reclaim path handles this | |
2111 | * more efficiently than we can here, so simply let background reclaim tear down | |
2112 | * all inodes. | |
2113 | */ | |
2114 | void | |
2115 | xfs_inode_mark_reclaimable( | |
2116 | struct xfs_inode *ip) | |
2117 | { | |
2118 | struct xfs_mount *mp = ip->i_mount; | |
2119 | bool need_inactive; | |
2120 | ||
2121 | XFS_STATS_INC(mp, vn_reclaim); | |
2122 | ||
2123 | /* | |
2124 | * We should never get here with any of the reclaim flags already set. | |
2125 | */ | |
2126 | ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_ALL_IRECLAIM_FLAGS)); | |
2127 | ||
2128 | need_inactive = xfs_inode_needs_inactive(ip); | |
2129 | if (need_inactive) { | |
2130 | xfs_inodegc_queue(ip); | |
2131 | return; | |
2132 | } | |
2133 | ||
2134 | /* Going straight to reclaim, so drop the dquots. */ | |
2135 | xfs_qm_dqdetach(ip); | |
2136 | xfs_inodegc_set_reclaimable(ip); | |
2137 | } | |
40b1de00 DW |
2138 | |
2139 | /* | |
2140 | * Register a phony shrinker so that we can run background inodegc sooner when | |
2141 | * there's memory pressure. Inactivation does not itself free any memory but | |
2142 | * it does make inodes reclaimable, which eventually frees memory. | |
2143 | * | |
2144 | * The count function, seek value, and batch value are crafted to trigger the | |
2145 | * scan function during the second round of scanning. Hopefully this means | |
2146 | * that we reclaimed enough memory that initiating metadata transactions won't | |
2147 | * make things worse. | |
2148 | */ | |
2149 | #define XFS_INODEGC_SHRINKER_COUNT (1UL << DEF_PRIORITY) | |
2150 | #define XFS_INODEGC_SHRINKER_BATCH ((XFS_INODEGC_SHRINKER_COUNT / 2) + 1) | |
2151 | ||
2152 | static unsigned long | |
2153 | xfs_inodegc_shrinker_count( | |
2154 | struct shrinker *shrink, | |
2155 | struct shrink_control *sc) | |
2156 | { | |
2157 | struct xfs_mount *mp = container_of(shrink, struct xfs_mount, | |
2158 | m_inodegc_shrinker); | |
2159 | struct xfs_inodegc *gc; | |
2160 | int cpu; | |
2161 | ||
2162 | if (!xfs_is_inodegc_enabled(mp)) | |
2163 | return 0; | |
2164 | ||
2165 | for_each_online_cpu(cpu) { | |
2166 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
2167 | if (!llist_empty(&gc->list)) | |
2168 | return XFS_INODEGC_SHRINKER_COUNT; | |
2169 | } | |
2170 | ||
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | static unsigned long | |
2175 | xfs_inodegc_shrinker_scan( | |
2176 | struct shrinker *shrink, | |
2177 | struct shrink_control *sc) | |
2178 | { | |
2179 | struct xfs_mount *mp = container_of(shrink, struct xfs_mount, | |
2180 | m_inodegc_shrinker); | |
2181 | struct xfs_inodegc *gc; | |
2182 | int cpu; | |
2183 | bool no_items = true; | |
2184 | ||
2185 | if (!xfs_is_inodegc_enabled(mp)) | |
2186 | return SHRINK_STOP; | |
2187 | ||
2188 | trace_xfs_inodegc_shrinker_scan(mp, sc, __return_address); | |
2189 | ||
2190 | for_each_online_cpu(cpu) { | |
2191 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
2192 | if (!llist_empty(&gc->list)) { | |
2193 | unsigned int h = READ_ONCE(gc->shrinker_hits); | |
2194 | ||
2195 | WRITE_ONCE(gc->shrinker_hits, h + 1); | |
7cf2b0f9 | 2196 | mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0); |
40b1de00 DW |
2197 | no_items = false; |
2198 | } | |
2199 | } | |
2200 | ||
2201 | /* | |
2202 | * If there are no inodes to inactivate, we don't want the shrinker | |
2203 | * to think there's deferred work to call us back about. | |
2204 | */ | |
2205 | if (no_items) | |
2206 | return LONG_MAX; | |
2207 | ||
2208 | return SHRINK_STOP; | |
2209 | } | |
2210 | ||
2211 | /* Register a shrinker so we can accelerate inodegc and throttle queuing. */ | |
2212 | int | |
2213 | xfs_inodegc_register_shrinker( | |
2214 | struct xfs_mount *mp) | |
2215 | { | |
2216 | struct shrinker *shrink = &mp->m_inodegc_shrinker; | |
2217 | ||
2218 | shrink->count_objects = xfs_inodegc_shrinker_count; | |
2219 | shrink->scan_objects = xfs_inodegc_shrinker_scan; | |
2220 | shrink->seeks = 0; | |
2221 | shrink->flags = SHRINKER_NONSLAB; | |
2222 | shrink->batch = XFS_INODEGC_SHRINKER_BATCH; | |
2223 | ||
2224 | return register_shrinker(shrink); | |
2225 | } |