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)); |
33479e05 DC |
105 | ip->i_flags = 0; |
106 | ip->i_delayed_blks = 0; | |
3e09ab8f | 107 | ip->i_diflags2 = mp->m_ino_geo.new_diflags2; |
6e73a545 | 108 | ip->i_nblocks = 0; |
7821ea30 | 109 | ip->i_forkoff = 0; |
6772c1f1 DW |
110 | ip->i_sick = 0; |
111 | ip->i_checked = 0; | |
cb357bf3 DW |
112 | INIT_WORK(&ip->i_ioend_work, xfs_end_io); |
113 | INIT_LIST_HEAD(&ip->i_ioend_list); | |
114 | spin_lock_init(&ip->i_ioend_lock); | |
2fd26cc0 DC |
115 | ip->i_next_unlinked = NULLAGINO; |
116 | ip->i_prev_unlinked = NULLAGINO; | |
33479e05 DC |
117 | |
118 | return ip; | |
119 | } | |
120 | ||
121 | STATIC void | |
122 | xfs_inode_free_callback( | |
123 | struct rcu_head *head) | |
124 | { | |
125 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
126 | struct xfs_inode *ip = XFS_I(inode); | |
127 | ||
c19b3b05 | 128 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
33479e05 DC |
129 | case S_IFREG: |
130 | case S_IFDIR: | |
131 | case S_IFLNK: | |
ef838512 | 132 | xfs_idestroy_fork(&ip->i_df); |
33479e05 DC |
133 | break; |
134 | } | |
135 | ||
e45d7cb2 DW |
136 | xfs_ifork_zap_attr(ip); |
137 | ||
ef838512 CH |
138 | if (ip->i_cowfp) { |
139 | xfs_idestroy_fork(ip->i_cowfp); | |
182696fb | 140 | kmem_cache_free(xfs_ifork_cache, ip->i_cowfp); |
ef838512 | 141 | } |
33479e05 | 142 | if (ip->i_itemp) { |
22525c17 DC |
143 | ASSERT(!test_bit(XFS_LI_IN_AIL, |
144 | &ip->i_itemp->ili_item.li_flags)); | |
33479e05 DC |
145 | xfs_inode_item_destroy(ip); |
146 | ip->i_itemp = NULL; | |
147 | } | |
148 | ||
182696fb | 149 | kmem_cache_free(xfs_inode_cache, ip); |
1f2dcfe8 DC |
150 | } |
151 | ||
8a17d7dd DC |
152 | static void |
153 | __xfs_inode_free( | |
154 | struct xfs_inode *ip) | |
155 | { | |
156 | /* asserts to verify all state is correct here */ | |
157 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
48d55e2a | 158 | ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list)); |
8a17d7dd DC |
159 | XFS_STATS_DEC(ip->i_mount, vn_active); |
160 | ||
161 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
162 | } | |
163 | ||
1f2dcfe8 DC |
164 | void |
165 | xfs_inode_free( | |
166 | struct xfs_inode *ip) | |
167 | { | |
718ecc50 | 168 | ASSERT(!xfs_iflags_test(ip, XFS_IFLUSHING)); |
98efe8af | 169 | |
33479e05 DC |
170 | /* |
171 | * Because we use RCU freeing we need to ensure the inode always | |
172 | * appears to be reclaimed with an invalid inode number when in the | |
173 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
174 | * races. | |
175 | */ | |
176 | spin_lock(&ip->i_flags_lock); | |
177 | ip->i_flags = XFS_IRECLAIM; | |
178 | ip->i_ino = 0; | |
179 | spin_unlock(&ip->i_flags_lock); | |
180 | ||
8a17d7dd | 181 | __xfs_inode_free(ip); |
33479e05 DC |
182 | } |
183 | ||
ad438c40 | 184 | /* |
02511a5a DC |
185 | * Queue background inode reclaim work if there are reclaimable inodes and there |
186 | * isn't reclaim work already scheduled or in progress. | |
ad438c40 DC |
187 | */ |
188 | static void | |
189 | xfs_reclaim_work_queue( | |
190 | struct xfs_mount *mp) | |
191 | { | |
192 | ||
193 | rcu_read_lock(); | |
194 | if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { | |
195 | queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work, | |
196 | msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10)); | |
197 | } | |
198 | rcu_read_unlock(); | |
199 | } | |
200 | ||
c076ae7a DW |
201 | /* |
202 | * Background scanning to trim preallocated space. This is queued based on the | |
203 | * 'speculative_prealloc_lifetime' tunable (5m by default). | |
204 | */ | |
205 | static inline void | |
206 | xfs_blockgc_queue( | |
ad438c40 | 207 | struct xfs_perag *pag) |
c076ae7a | 208 | { |
6f649091 DW |
209 | struct xfs_mount *mp = pag->pag_mount; |
210 | ||
211 | if (!xfs_is_blockgc_enabled(mp)) | |
212 | return; | |
213 | ||
c076ae7a DW |
214 | rcu_read_lock(); |
215 | if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG)) | |
ab23a776 | 216 | queue_delayed_work(pag->pag_mount->m_blockgc_wq, |
c076ae7a DW |
217 | &pag->pag_blockgc_work, |
218 | msecs_to_jiffies(xfs_blockgc_secs * 1000)); | |
219 | rcu_read_unlock(); | |
220 | } | |
221 | ||
222 | /* Set a tag on both the AG incore inode tree and the AG radix tree. */ | |
223 | static void | |
224 | xfs_perag_set_inode_tag( | |
225 | struct xfs_perag *pag, | |
226 | xfs_agino_t agino, | |
227 | unsigned int tag) | |
ad438c40 DC |
228 | { |
229 | struct xfs_mount *mp = pag->pag_mount; | |
c076ae7a | 230 | bool was_tagged; |
ad438c40 | 231 | |
95989c46 | 232 | lockdep_assert_held(&pag->pag_ici_lock); |
c076ae7a DW |
233 | |
234 | was_tagged = radix_tree_tagged(&pag->pag_ici_root, tag); | |
235 | radix_tree_tag_set(&pag->pag_ici_root, agino, tag); | |
236 | ||
237 | if (tag == XFS_ICI_RECLAIM_TAG) | |
238 | pag->pag_ici_reclaimable++; | |
239 | ||
240 | if (was_tagged) | |
ad438c40 DC |
241 | return; |
242 | ||
c076ae7a | 243 | /* propagate the tag up into the perag radix tree */ |
ad438c40 | 244 | spin_lock(&mp->m_perag_lock); |
c076ae7a | 245 | radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, tag); |
ad438c40 DC |
246 | spin_unlock(&mp->m_perag_lock); |
247 | ||
c076ae7a DW |
248 | /* start background work */ |
249 | switch (tag) { | |
250 | case XFS_ICI_RECLAIM_TAG: | |
251 | xfs_reclaim_work_queue(mp); | |
252 | break; | |
253 | case XFS_ICI_BLOCKGC_TAG: | |
254 | xfs_blockgc_queue(pag); | |
255 | break; | |
256 | } | |
ad438c40 | 257 | |
368e2d09 | 258 | trace_xfs_perag_set_inode_tag(pag, _RET_IP_); |
ad438c40 DC |
259 | } |
260 | ||
c076ae7a | 261 | /* Clear a tag on both the AG incore inode tree and the AG radix tree. */ |
ad438c40 | 262 | static void |
c076ae7a DW |
263 | xfs_perag_clear_inode_tag( |
264 | struct xfs_perag *pag, | |
265 | xfs_agino_t agino, | |
266 | unsigned int tag) | |
ad438c40 DC |
267 | { |
268 | struct xfs_mount *mp = pag->pag_mount; | |
269 | ||
95989c46 | 270 | lockdep_assert_held(&pag->pag_ici_lock); |
c076ae7a DW |
271 | |
272 | /* | |
273 | * Reclaim can signal (with a null agino) that it cleared its own tag | |
274 | * by removing the inode from the radix tree. | |
275 | */ | |
276 | if (agino != NULLAGINO) | |
277 | radix_tree_tag_clear(&pag->pag_ici_root, agino, tag); | |
278 | else | |
279 | ASSERT(tag == XFS_ICI_RECLAIM_TAG); | |
280 | ||
281 | if (tag == XFS_ICI_RECLAIM_TAG) | |
282 | pag->pag_ici_reclaimable--; | |
283 | ||
284 | if (radix_tree_tagged(&pag->pag_ici_root, tag)) | |
ad438c40 DC |
285 | return; |
286 | ||
c076ae7a | 287 | /* clear the tag from the perag radix tree */ |
ad438c40 | 288 | spin_lock(&mp->m_perag_lock); |
c076ae7a | 289 | radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, tag); |
ad438c40 | 290 | spin_unlock(&mp->m_perag_lock); |
ad438c40 | 291 | |
368e2d09 | 292 | trace_xfs_perag_clear_inode_tag(pag, _RET_IP_); |
c076ae7a | 293 | } |
ad438c40 | 294 | |
50997470 DC |
295 | /* |
296 | * When we recycle a reclaimable inode, we need to re-initialise the VFS inode | |
297 | * part of the structure. This is made more complex by the fact we store | |
298 | * information about the on-disk values in the VFS inode and so we can't just | |
83e06f21 | 299 | * overwrite the values unconditionally. Hence we save the parameters we |
50997470 | 300 | * need to retain across reinitialisation, and rewrite them into the VFS inode |
83e06f21 | 301 | * after reinitialisation even if it fails. |
50997470 DC |
302 | */ |
303 | static int | |
304 | xfs_reinit_inode( | |
305 | struct xfs_mount *mp, | |
306 | struct inode *inode) | |
307 | { | |
ff7bebeb DW |
308 | int error; |
309 | uint32_t nlink = inode->i_nlink; | |
310 | uint32_t generation = inode->i_generation; | |
311 | uint64_t version = inode_peek_iversion(inode); | |
312 | umode_t mode = inode->i_mode; | |
313 | dev_t dev = inode->i_rdev; | |
314 | kuid_t uid = inode->i_uid; | |
315 | kgid_t gid = inode->i_gid; | |
50997470 DC |
316 | |
317 | error = inode_init_always(mp->m_super, inode); | |
318 | ||
54d7b5c1 | 319 | set_nlink(inode, nlink); |
9e9a2674 | 320 | inode->i_generation = generation; |
f0e28280 | 321 | inode_set_iversion_queried(inode, version); |
c19b3b05 | 322 | inode->i_mode = mode; |
acd1d715 | 323 | inode->i_rdev = dev; |
3d8f2821 CH |
324 | inode->i_uid = uid; |
325 | inode->i_gid = gid; | |
67958013 | 326 | mapping_set_large_folios(inode->i_mapping); |
50997470 DC |
327 | return error; |
328 | } | |
329 | ||
ff7bebeb DW |
330 | /* |
331 | * Carefully nudge an inode whose VFS state has been torn down back into a | |
332 | * usable state. Drops the i_flags_lock and the rcu read lock. | |
333 | */ | |
334 | static int | |
335 | xfs_iget_recycle( | |
336 | struct xfs_perag *pag, | |
337 | struct xfs_inode *ip) __releases(&ip->i_flags_lock) | |
338 | { | |
339 | struct xfs_mount *mp = ip->i_mount; | |
340 | struct inode *inode = VFS_I(ip); | |
341 | int error; | |
342 | ||
343 | trace_xfs_iget_recycle(ip); | |
344 | ||
28b4b059 LL |
345 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) |
346 | return -EAGAIN; | |
347 | ||
ff7bebeb DW |
348 | /* |
349 | * We need to make it look like the inode is being reclaimed to prevent | |
350 | * the actual reclaim workers from stomping over us while we recycle | |
351 | * the inode. We can't clear the radix tree tag yet as it requires | |
352 | * pag_ici_lock to be held exclusive. | |
353 | */ | |
354 | ip->i_flags |= XFS_IRECLAIM; | |
355 | ||
356 | spin_unlock(&ip->i_flags_lock); | |
357 | rcu_read_unlock(); | |
358 | ||
359 | ASSERT(!rwsem_is_locked(&inode->i_rwsem)); | |
360 | error = xfs_reinit_inode(mp, inode); | |
28b4b059 | 361 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
ff7bebeb | 362 | if (error) { |
ff7bebeb DW |
363 | /* |
364 | * Re-initializing the inode failed, and we are in deep | |
365 | * trouble. Try to re-add it to the reclaim list. | |
366 | */ | |
367 | rcu_read_lock(); | |
368 | spin_lock(&ip->i_flags_lock); | |
ff7bebeb | 369 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
ff7bebeb DW |
370 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); |
371 | spin_unlock(&ip->i_flags_lock); | |
372 | rcu_read_unlock(); | |
373 | ||
374 | trace_xfs_iget_recycle_fail(ip); | |
375 | return error; | |
376 | } | |
377 | ||
378 | spin_lock(&pag->pag_ici_lock); | |
379 | spin_lock(&ip->i_flags_lock); | |
380 | ||
381 | /* | |
382 | * Clear the per-lifetime state in the inode as we are now effectively | |
383 | * a new inode and need to return to the initial state before reuse | |
384 | * occurs. | |
385 | */ | |
386 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
387 | ip->i_flags |= XFS_INEW; | |
388 | xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), | |
389 | XFS_ICI_RECLAIM_TAG); | |
390 | inode->i_state = I_NEW; | |
391 | spin_unlock(&ip->i_flags_lock); | |
392 | spin_unlock(&pag->pag_ici_lock); | |
393 | ||
394 | return 0; | |
395 | } | |
396 | ||
afca6c5b DC |
397 | /* |
398 | * If we are allocating a new inode, then check what was returned is | |
399 | * actually a free, empty inode. If we are not allocating an inode, | |
400 | * then check we didn't find a free inode. | |
401 | * | |
402 | * Returns: | |
403 | * 0 if the inode free state matches the lookup context | |
404 | * -ENOENT if the inode is free and we are not allocating | |
405 | * -EFSCORRUPTED if there is any state mismatch at all | |
406 | */ | |
407 | static int | |
408 | xfs_iget_check_free_state( | |
409 | struct xfs_inode *ip, | |
410 | int flags) | |
411 | { | |
412 | if (flags & XFS_IGET_CREATE) { | |
413 | /* should be a free inode */ | |
414 | if (VFS_I(ip)->i_mode != 0) { | |
415 | xfs_warn(ip->i_mount, | |
416 | "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)", | |
417 | ip->i_ino, VFS_I(ip)->i_mode); | |
418 | return -EFSCORRUPTED; | |
419 | } | |
420 | ||
6e73a545 | 421 | if (ip->i_nblocks != 0) { |
afca6c5b DC |
422 | xfs_warn(ip->i_mount, |
423 | "Corruption detected! Free inode 0x%llx has blocks allocated!", | |
424 | ip->i_ino); | |
425 | return -EFSCORRUPTED; | |
426 | } | |
427 | return 0; | |
428 | } | |
429 | ||
430 | /* should be an allocated inode */ | |
431 | if (VFS_I(ip)->i_mode == 0) | |
432 | return -ENOENT; | |
433 | ||
434 | return 0; | |
435 | } | |
436 | ||
ab23a776 | 437 | /* Make all pending inactivation work start immediately. */ |
2254a739 | 438 | static bool |
ab23a776 DC |
439 | xfs_inodegc_queue_all( |
440 | struct xfs_mount *mp) | |
441 | { | |
442 | struct xfs_inodegc *gc; | |
443 | int cpu; | |
2254a739 | 444 | bool ret = false; |
ab23a776 DC |
445 | |
446 | for_each_online_cpu(cpu) { | |
447 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
2254a739 | 448 | if (!llist_empty(&gc->list)) { |
7cf2b0f9 | 449 | mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0); |
2254a739 DW |
450 | ret = true; |
451 | } | |
ab23a776 | 452 | } |
2254a739 DW |
453 | |
454 | return ret; | |
ab23a776 DC |
455 | } |
456 | ||
33479e05 DC |
457 | /* |
458 | * Check the validity of the inode we just found it the cache | |
459 | */ | |
460 | static int | |
461 | xfs_iget_cache_hit( | |
462 | struct xfs_perag *pag, | |
463 | struct xfs_inode *ip, | |
464 | xfs_ino_t ino, | |
465 | int flags, | |
466 | int lock_flags) __releases(RCU) | |
467 | { | |
468 | struct inode *inode = VFS_I(ip); | |
469 | struct xfs_mount *mp = ip->i_mount; | |
470 | int error; | |
471 | ||
472 | /* | |
473 | * check for re-use of an inode within an RCU grace period due to the | |
474 | * radix tree nodes not being updated yet. We monitor for this by | |
475 | * setting the inode number to zero before freeing the inode structure. | |
476 | * If the inode has been reallocated and set up, then the inode number | |
477 | * will not match, so check for that, too. | |
478 | */ | |
479 | spin_lock(&ip->i_flags_lock); | |
77b4d286 DW |
480 | if (ip->i_ino != ino) |
481 | goto out_skip; | |
33479e05 DC |
482 | |
483 | /* | |
484 | * If we are racing with another cache hit that is currently | |
485 | * instantiating this inode or currently recycling it out of | |
ff7bebeb | 486 | * reclaimable state, wait for the initialisation to complete |
33479e05 DC |
487 | * before continuing. |
488 | * | |
ab23a776 DC |
489 | * If we're racing with the inactivation worker we also want to wait. |
490 | * If we're creating a new file, it's possible that the worker | |
491 | * previously marked the inode as free on disk but hasn't finished | |
492 | * updating the incore state yet. The AGI buffer will be dirty and | |
493 | * locked to the icreate transaction, so a synchronous push of the | |
494 | * inodegc workers would result in deadlock. For a regular iget, the | |
495 | * worker is running already, so we might as well wait. | |
496 | * | |
33479e05 DC |
497 | * XXX(hch): eventually we should do something equivalent to |
498 | * wait_on_inode to wait for these flags to be cleared | |
499 | * instead of polling for it. | |
500 | */ | |
ab23a776 | 501 | if (ip->i_flags & (XFS_INEW | XFS_IRECLAIM | XFS_INACTIVATING)) |
77b4d286 | 502 | goto out_skip; |
33479e05 | 503 | |
ab23a776 DC |
504 | if (ip->i_flags & XFS_NEED_INACTIVE) { |
505 | /* Unlinked inodes cannot be re-grabbed. */ | |
506 | if (VFS_I(ip)->i_nlink == 0) { | |
507 | error = -ENOENT; | |
508 | goto out_error; | |
509 | } | |
510 | goto out_inodegc_flush; | |
511 | } | |
512 | ||
33479e05 | 513 | /* |
afca6c5b DC |
514 | * Check the inode free state is valid. This also detects lookup |
515 | * racing with unlinks. | |
33479e05 | 516 | */ |
afca6c5b DC |
517 | error = xfs_iget_check_free_state(ip, flags); |
518 | if (error) | |
33479e05 | 519 | goto out_error; |
33479e05 | 520 | |
77b4d286 DW |
521 | /* Skip inodes that have no vfs state. */ |
522 | if ((flags & XFS_IGET_INCORE) && | |
523 | (ip->i_flags & XFS_IRECLAIMABLE)) | |
524 | goto out_skip; | |
378f681c | 525 | |
77b4d286 DW |
526 | /* The inode fits the selection criteria; process it. */ |
527 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
ff7bebeb DW |
528 | /* Drops i_flags_lock and RCU read lock. */ |
529 | error = xfs_iget_recycle(pag, ip); | |
28b4b059 LL |
530 | if (error == -EAGAIN) |
531 | goto out_skip; | |
ff7bebeb DW |
532 | if (error) |
533 | return error; | |
33479e05 DC |
534 | } else { |
535 | /* If the VFS inode is being torn down, pause and try again. */ | |
77b4d286 DW |
536 | if (!igrab(inode)) |
537 | goto out_skip; | |
33479e05 DC |
538 | |
539 | /* We've got a live one. */ | |
540 | spin_unlock(&ip->i_flags_lock); | |
541 | rcu_read_unlock(); | |
542 | trace_xfs_iget_hit(ip); | |
543 | } | |
544 | ||
545 | if (lock_flags != 0) | |
546 | xfs_ilock(ip, lock_flags); | |
547 | ||
378f681c | 548 | if (!(flags & XFS_IGET_INCORE)) |
dae2f8ed | 549 | xfs_iflags_clear(ip, XFS_ISTALE); |
ff6d6af2 | 550 | XFS_STATS_INC(mp, xs_ig_found); |
33479e05 DC |
551 | |
552 | return 0; | |
553 | ||
77b4d286 DW |
554 | out_skip: |
555 | trace_xfs_iget_skip(ip); | |
556 | XFS_STATS_INC(mp, xs_ig_frecycle); | |
557 | error = -EAGAIN; | |
33479e05 DC |
558 | out_error: |
559 | spin_unlock(&ip->i_flags_lock); | |
560 | rcu_read_unlock(); | |
561 | return error; | |
ab23a776 DC |
562 | |
563 | out_inodegc_flush: | |
564 | spin_unlock(&ip->i_flags_lock); | |
565 | rcu_read_unlock(); | |
566 | /* | |
567 | * Do not wait for the workers, because the caller could hold an AGI | |
568 | * buffer lock. We're just going to sleep in a loop anyway. | |
569 | */ | |
570 | if (xfs_is_inodegc_enabled(mp)) | |
571 | xfs_inodegc_queue_all(mp); | |
572 | return -EAGAIN; | |
33479e05 DC |
573 | } |
574 | ||
33479e05 DC |
575 | static int |
576 | xfs_iget_cache_miss( | |
577 | struct xfs_mount *mp, | |
578 | struct xfs_perag *pag, | |
579 | xfs_trans_t *tp, | |
580 | xfs_ino_t ino, | |
581 | struct xfs_inode **ipp, | |
582 | int flags, | |
583 | int lock_flags) | |
584 | { | |
585 | struct xfs_inode *ip; | |
586 | int error; | |
587 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
588 | int iflags; | |
589 | ||
590 | ip = xfs_inode_alloc(mp, ino); | |
591 | if (!ip) | |
2451337d | 592 | return -ENOMEM; |
33479e05 | 593 | |
498f0adb | 594 | error = xfs_imap(pag, tp, ip->i_ino, &ip->i_imap, flags); |
33479e05 DC |
595 | if (error) |
596 | goto out_destroy; | |
597 | ||
bb8a66af CH |
598 | /* |
599 | * For version 5 superblocks, if we are initialising a new inode and we | |
0560f31a | 600 | * are not utilising the XFS_FEAT_IKEEP inode cluster mode, we can |
bb8a66af CH |
601 | * simply build the new inode core with a random generation number. |
602 | * | |
603 | * For version 4 (and older) superblocks, log recovery is dependent on | |
965e0a1a | 604 | * the i_flushiter field being initialised from the current on-disk |
bb8a66af CH |
605 | * value and hence we must also read the inode off disk even when |
606 | * initializing new inodes. | |
607 | */ | |
38c26bfd | 608 | if (xfs_has_v3inodes(mp) && |
0560f31a | 609 | (flags & XFS_IGET_CREATE) && !xfs_has_ikeep(mp)) { |
a251c17a | 610 | VFS_I(ip)->i_generation = get_random_u32(); |
bb8a66af | 611 | } else { |
bb8a66af CH |
612 | struct xfs_buf *bp; |
613 | ||
af9dcdde | 614 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp); |
bb8a66af CH |
615 | if (error) |
616 | goto out_destroy; | |
617 | ||
af9dcdde CH |
618 | error = xfs_inode_from_disk(ip, |
619 | xfs_buf_offset(bp, ip->i_imap.im_boffset)); | |
bb8a66af CH |
620 | if (!error) |
621 | xfs_buf_set_ref(bp, XFS_INO_REF); | |
622 | xfs_trans_brelse(tp, bp); | |
623 | ||
624 | if (error) | |
625 | goto out_destroy; | |
626 | } | |
627 | ||
33479e05 DC |
628 | trace_xfs_iget_miss(ip); |
629 | ||
ee457001 | 630 | /* |
afca6c5b DC |
631 | * Check the inode free state is valid. This also detects lookup |
632 | * racing with unlinks. | |
ee457001 | 633 | */ |
afca6c5b DC |
634 | error = xfs_iget_check_free_state(ip, flags); |
635 | if (error) | |
33479e05 | 636 | goto out_destroy; |
33479e05 DC |
637 | |
638 | /* | |
639 | * Preload the radix tree so we can insert safely under the | |
640 | * write spinlock. Note that we cannot sleep inside the preload | |
641 | * region. Since we can be called from transaction context, don't | |
642 | * recurse into the file system. | |
643 | */ | |
644 | if (radix_tree_preload(GFP_NOFS)) { | |
2451337d | 645 | error = -EAGAIN; |
33479e05 DC |
646 | goto out_destroy; |
647 | } | |
648 | ||
649 | /* | |
650 | * Because the inode hasn't been added to the radix-tree yet it can't | |
651 | * be found by another thread, so we can do the non-sleeping lock here. | |
652 | */ | |
653 | if (lock_flags) { | |
654 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
655 | BUG(); | |
656 | } | |
657 | ||
658 | /* | |
659 | * These values must be set before inserting the inode into the radix | |
660 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
661 | * RCU locking mechanism) can find it and that lookup must see that this | |
662 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
663 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
664 | * memory barrier that ensures this detection works correctly at lookup | |
665 | * time. | |
666 | */ | |
667 | iflags = XFS_INEW; | |
668 | if (flags & XFS_IGET_DONTCACHE) | |
2c567af4 | 669 | d_mark_dontcache(VFS_I(ip)); |
113a5683 CS |
670 | ip->i_udquot = NULL; |
671 | ip->i_gdquot = NULL; | |
92f8ff73 | 672 | ip->i_pdquot = NULL; |
33479e05 DC |
673 | xfs_iflags_set(ip, iflags); |
674 | ||
675 | /* insert the new inode */ | |
676 | spin_lock(&pag->pag_ici_lock); | |
677 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
678 | if (unlikely(error)) { | |
679 | WARN_ON(error != -EEXIST); | |
ff6d6af2 | 680 | XFS_STATS_INC(mp, xs_ig_dup); |
2451337d | 681 | error = -EAGAIN; |
33479e05 DC |
682 | goto out_preload_end; |
683 | } | |
684 | spin_unlock(&pag->pag_ici_lock); | |
685 | radix_tree_preload_end(); | |
686 | ||
687 | *ipp = ip; | |
688 | return 0; | |
689 | ||
690 | out_preload_end: | |
691 | spin_unlock(&pag->pag_ici_lock); | |
692 | radix_tree_preload_end(); | |
693 | if (lock_flags) | |
694 | xfs_iunlock(ip, lock_flags); | |
695 | out_destroy: | |
696 | __destroy_inode(VFS_I(ip)); | |
697 | xfs_inode_free(ip); | |
698 | return error; | |
699 | } | |
700 | ||
701 | /* | |
02511a5a DC |
702 | * Look up an inode by number in the given file system. The inode is looked up |
703 | * in the cache held in each AG. If the inode is found in the cache, initialise | |
704 | * the vfs inode if necessary. | |
33479e05 | 705 | * |
02511a5a DC |
706 | * If it is not in core, read it in from the file system's device, add it to the |
707 | * cache and initialise the vfs inode. | |
33479e05 DC |
708 | * |
709 | * The inode is locked according to the value of the lock_flags parameter. | |
02511a5a DC |
710 | * Inode lookup is only done during metadata operations and not as part of the |
711 | * data IO path. Hence we only allow locking of the XFS_ILOCK during lookup. | |
33479e05 DC |
712 | */ |
713 | int | |
714 | xfs_iget( | |
02511a5a DC |
715 | struct xfs_mount *mp, |
716 | struct xfs_trans *tp, | |
717 | xfs_ino_t ino, | |
718 | uint flags, | |
719 | uint lock_flags, | |
720 | struct xfs_inode **ipp) | |
33479e05 | 721 | { |
02511a5a DC |
722 | struct xfs_inode *ip; |
723 | struct xfs_perag *pag; | |
724 | xfs_agino_t agino; | |
725 | int error; | |
33479e05 | 726 | |
33479e05 DC |
727 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); |
728 | ||
729 | /* reject inode numbers outside existing AGs */ | |
730 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) | |
2451337d | 731 | return -EINVAL; |
33479e05 | 732 | |
ff6d6af2 | 733 | XFS_STATS_INC(mp, xs_ig_attempts); |
8774cf8b | 734 | |
33479e05 DC |
735 | /* get the perag structure and ensure that it's inode capable */ |
736 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); | |
737 | agino = XFS_INO_TO_AGINO(mp, ino); | |
738 | ||
739 | again: | |
740 | error = 0; | |
741 | rcu_read_lock(); | |
742 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
743 | ||
744 | if (ip) { | |
745 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); | |
746 | if (error) | |
747 | goto out_error_or_again; | |
748 | } else { | |
749 | rcu_read_unlock(); | |
378f681c | 750 | if (flags & XFS_IGET_INCORE) { |
ed438b47 | 751 | error = -ENODATA; |
378f681c DW |
752 | goto out_error_or_again; |
753 | } | |
ff6d6af2 | 754 | XFS_STATS_INC(mp, xs_ig_missed); |
33479e05 DC |
755 | |
756 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, | |
757 | flags, lock_flags); | |
758 | if (error) | |
759 | goto out_error_or_again; | |
760 | } | |
761 | xfs_perag_put(pag); | |
762 | ||
763 | *ipp = ip; | |
764 | ||
765 | /* | |
58c90473 | 766 | * If we have a real type for an on-disk inode, we can setup the inode |
132c460e YX |
767 | * now. If it's a new inode being created, xfs_init_new_inode will |
768 | * handle it. | |
33479e05 | 769 | */ |
c19b3b05 | 770 | if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0) |
58c90473 | 771 | xfs_setup_existing_inode(ip); |
33479e05 DC |
772 | return 0; |
773 | ||
774 | out_error_or_again: | |
302436c2 DW |
775 | if (!(flags & (XFS_IGET_INCORE | XFS_IGET_NORETRY)) && |
776 | error == -EAGAIN) { | |
33479e05 DC |
777 | delay(1); |
778 | goto again; | |
779 | } | |
780 | xfs_perag_put(pag); | |
781 | return error; | |
782 | } | |
783 | ||
378f681c DW |
784 | /* |
785 | * "Is this a cached inode that's also allocated?" | |
786 | * | |
787 | * Look up an inode by number in the given file system. If the inode is | |
788 | * in cache and isn't in purgatory, return 1 if the inode is allocated | |
789 | * and 0 if it is not. For all other cases (not in cache, being torn | |
790 | * down, etc.), return a negative error code. | |
791 | * | |
792 | * The caller has to prevent inode allocation and freeing activity, | |
793 | * presumably by locking the AGI buffer. This is to ensure that an | |
794 | * inode cannot transition from allocated to freed until the caller is | |
795 | * ready to allow that. If the inode is in an intermediate state (new, | |
796 | * reclaimable, or being reclaimed), -EAGAIN will be returned; if the | |
797 | * inode is not in the cache, -ENOENT will be returned. The caller must | |
798 | * deal with these scenarios appropriately. | |
799 | * | |
800 | * This is a specialized use case for the online scrubber; if you're | |
801 | * reading this, you probably want xfs_iget. | |
802 | */ | |
803 | int | |
804 | xfs_icache_inode_is_allocated( | |
805 | struct xfs_mount *mp, | |
806 | struct xfs_trans *tp, | |
807 | xfs_ino_t ino, | |
808 | bool *inuse) | |
809 | { | |
810 | struct xfs_inode *ip; | |
811 | int error; | |
812 | ||
813 | error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip); | |
814 | if (error) | |
815 | return error; | |
816 | ||
817 | *inuse = !!(VFS_I(ip)->i_mode); | |
44a8736b | 818 | xfs_irele(ip); |
378f681c DW |
819 | return 0; |
820 | } | |
821 | ||
e3a20c0b DC |
822 | /* |
823 | * Grab the inode for reclaim exclusively. | |
50718b8d DC |
824 | * |
825 | * We have found this inode via a lookup under RCU, so the inode may have | |
826 | * already been freed, or it may be in the process of being recycled by | |
827 | * xfs_iget(). In both cases, the inode will have XFS_IRECLAIM set. If the inode | |
828 | * has been fully recycled by the time we get the i_flags_lock, XFS_IRECLAIMABLE | |
829 | * will not be set. Hence we need to check for both these flag conditions to | |
830 | * avoid inodes that are no longer reclaim candidates. | |
831 | * | |
832 | * Note: checking for other state flags here, under the i_flags_lock or not, is | |
833 | * racy and should be avoided. Those races should be resolved only after we have | |
834 | * ensured that we are able to reclaim this inode and the world can see that we | |
835 | * are going to reclaim it. | |
836 | * | |
837 | * Return true if we grabbed it, false otherwise. | |
e3a20c0b | 838 | */ |
50718b8d | 839 | static bool |
f1bc5c56 | 840 | xfs_reclaim_igrab( |
9492750a | 841 | struct xfs_inode *ip, |
b26b2bf1 | 842 | struct xfs_icwalk *icw) |
e3a20c0b | 843 | { |
1a3e8f3d DC |
844 | ASSERT(rcu_read_lock_held()); |
845 | ||
e3a20c0b | 846 | spin_lock(&ip->i_flags_lock); |
1a3e8f3d DC |
847 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
848 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
849 | /* not a reclaim candidate. */ | |
e3a20c0b | 850 | spin_unlock(&ip->i_flags_lock); |
50718b8d | 851 | return false; |
e3a20c0b | 852 | } |
9492750a DW |
853 | |
854 | /* Don't reclaim a sick inode unless the caller asked for it. */ | |
855 | if (ip->i_sick && | |
b26b2bf1 | 856 | (!icw || !(icw->icw_flags & XFS_ICWALK_FLAG_RECLAIM_SICK))) { |
9492750a DW |
857 | spin_unlock(&ip->i_flags_lock); |
858 | return false; | |
859 | } | |
860 | ||
e3a20c0b DC |
861 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
862 | spin_unlock(&ip->i_flags_lock); | |
50718b8d | 863 | return true; |
e3a20c0b DC |
864 | } |
865 | ||
777df5af | 866 | /* |
02511a5a DC |
867 | * Inode reclaim is non-blocking, so the default action if progress cannot be |
868 | * made is to "requeue" the inode for reclaim by unlocking it and clearing the | |
869 | * XFS_IRECLAIM flag. If we are in a shutdown state, we don't care about | |
870 | * blocking anymore and hence we can wait for the inode to be able to reclaim | |
871 | * it. | |
777df5af | 872 | * |
02511a5a DC |
873 | * We do no IO here - if callers require inodes to be cleaned they must push the |
874 | * AIL first to trigger writeback of dirty inodes. This enables writeback to be | |
875 | * done in the background in a non-blocking manner, and enables memory reclaim | |
876 | * to make progress without blocking. | |
777df5af | 877 | */ |
4d0bab3a | 878 | static void |
c8e20be0 | 879 | xfs_reclaim_inode( |
75f3cb13 | 880 | struct xfs_inode *ip, |
50718b8d | 881 | struct xfs_perag *pag) |
fce08f2f | 882 | { |
8a17d7dd | 883 | xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */ |
777df5af | 884 | |
9552e14d | 885 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) |
617825fe | 886 | goto out; |
718ecc50 | 887 | if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING)) |
9552e14d | 888 | goto out_iunlock; |
7a3be02b | 889 | |
01728b44 DC |
890 | /* |
891 | * Check for log shutdown because aborting the inode can move the log | |
892 | * tail and corrupt in memory state. This is fine if the log is shut | |
893 | * down, but if the log is still active and only the mount is shut down | |
894 | * then the in-memory log tail movement caused by the abort can be | |
895 | * incorrectly propagated to disk. | |
896 | */ | |
897 | if (xlog_is_shutdown(ip->i_mount->m_log)) { | |
777df5af | 898 | xfs_iunpin_wait(ip); |
d2d7c047 | 899 | xfs_iflush_shutdown_abort(ip); |
777df5af DC |
900 | goto reclaim; |
901 | } | |
617825fe | 902 | if (xfs_ipincount(ip)) |
718ecc50 | 903 | goto out_clear_flush; |
617825fe | 904 | if (!xfs_inode_clean(ip)) |
718ecc50 | 905 | goto out_clear_flush; |
8a48088f | 906 | |
718ecc50 | 907 | xfs_iflags_clear(ip, XFS_IFLUSHING); |
777df5af | 908 | reclaim: |
ab23a776 | 909 | trace_xfs_inode_reclaiming(ip); |
98efe8af | 910 | |
8a17d7dd DC |
911 | /* |
912 | * Because we use RCU freeing we need to ensure the inode always appears | |
913 | * to be reclaimed with an invalid inode number when in the free state. | |
98efe8af | 914 | * We do this as early as possible under the ILOCK so that |
f2e9ad21 OS |
915 | * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to |
916 | * detect races with us here. By doing this, we guarantee that once | |
917 | * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that | |
918 | * it will see either a valid inode that will serialise correctly, or it | |
919 | * will see an invalid inode that it can skip. | |
8a17d7dd DC |
920 | */ |
921 | spin_lock(&ip->i_flags_lock); | |
922 | ip->i_flags = XFS_IRECLAIM; | |
923 | ip->i_ino = 0; | |
255794c7 DW |
924 | ip->i_sick = 0; |
925 | ip->i_checked = 0; | |
8a17d7dd DC |
926 | spin_unlock(&ip->i_flags_lock); |
927 | ||
fad743d7 | 928 | ASSERT(!ip->i_itemp || ip->i_itemp->ili_item.li_buf == NULL); |
c8e20be0 | 929 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2f11feab | 930 | |
ff6d6af2 | 931 | XFS_STATS_INC(ip->i_mount, xs_ig_reclaims); |
2f11feab DC |
932 | /* |
933 | * Remove the inode from the per-AG radix tree. | |
934 | * | |
935 | * Because radix_tree_delete won't complain even if the item was never | |
936 | * added to the tree assert that it's been there before to catch | |
937 | * problems with the inode life time early on. | |
938 | */ | |
1a427ab0 | 939 | spin_lock(&pag->pag_ici_lock); |
2f11feab | 940 | if (!radix_tree_delete(&pag->pag_ici_root, |
8a17d7dd | 941 | XFS_INO_TO_AGINO(ip->i_mount, ino))) |
2f11feab | 942 | ASSERT(0); |
c076ae7a | 943 | xfs_perag_clear_inode_tag(pag, NULLAGINO, XFS_ICI_RECLAIM_TAG); |
1a427ab0 | 944 | spin_unlock(&pag->pag_ici_lock); |
2f11feab DC |
945 | |
946 | /* | |
947 | * Here we do an (almost) spurious inode lock in order to coordinate | |
948 | * with inode cache radix tree lookups. This is because the lookup | |
949 | * can reference the inodes in the cache without taking references. | |
950 | * | |
951 | * We make that OK here by ensuring that we wait until the inode is | |
ad637a10 | 952 | * unlocked after the lookup before we go ahead and free it. |
2f11feab | 953 | */ |
ad637a10 | 954 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
3ea06d73 | 955 | ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot); |
ad637a10 | 956 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
96355d5a | 957 | ASSERT(xfs_inode_clean(ip)); |
2f11feab | 958 | |
8a17d7dd | 959 | __xfs_inode_free(ip); |
4d0bab3a | 960 | return; |
8a48088f | 961 | |
718ecc50 DC |
962 | out_clear_flush: |
963 | xfs_iflags_clear(ip, XFS_IFLUSHING); | |
9552e14d | 964 | out_iunlock: |
8a48088f | 965 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
9552e14d | 966 | out: |
617825fe | 967 | xfs_iflags_clear(ip, XFS_IRECLAIM); |
7a3be02b DC |
968 | } |
969 | ||
9492750a DW |
970 | /* Reclaim sick inodes if we're unmounting or the fs went down. */ |
971 | static inline bool | |
972 | xfs_want_reclaim_sick( | |
973 | struct xfs_mount *mp) | |
974 | { | |
2e973b2c | 975 | return xfs_is_unmounting(mp) || xfs_has_norecovery(mp) || |
75c8c50f | 976 | xfs_is_shutdown(mp); |
9492750a DW |
977 | } |
978 | ||
4d0bab3a | 979 | void |
7a3be02b | 980 | xfs_reclaim_inodes( |
4d0bab3a | 981 | struct xfs_mount *mp) |
7a3be02b | 982 | { |
b26b2bf1 DW |
983 | struct xfs_icwalk icw = { |
984 | .icw_flags = 0, | |
9492750a DW |
985 | }; |
986 | ||
987 | if (xfs_want_reclaim_sick(mp)) | |
b26b2bf1 | 988 | icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK; |
9492750a | 989 | |
4d0bab3a | 990 | while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) { |
617825fe | 991 | xfs_ail_push_all_sync(mp->m_ail); |
b26b2bf1 | 992 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw); |
0f4ec0f1 | 993 | } |
9bf729c0 DC |
994 | } |
995 | ||
996 | /* | |
02511a5a DC |
997 | * The shrinker infrastructure determines how many inodes we should scan for |
998 | * reclaim. We want as many clean inodes ready to reclaim as possible, so we | |
999 | * push the AIL here. We also want to proactively free up memory if we can to | |
1000 | * minimise the amount of work memory reclaim has to do so we kick the | |
1001 | * background reclaim if it isn't already scheduled. | |
9bf729c0 | 1002 | */ |
0a234c6d | 1003 | long |
8daaa831 DC |
1004 | xfs_reclaim_inodes_nr( |
1005 | struct xfs_mount *mp, | |
10be350b | 1006 | unsigned long nr_to_scan) |
9bf729c0 | 1007 | { |
b26b2bf1 DW |
1008 | struct xfs_icwalk icw = { |
1009 | .icw_flags = XFS_ICWALK_FLAG_SCAN_LIMIT, | |
10be350b | 1010 | .icw_scan_limit = min_t(unsigned long, LONG_MAX, nr_to_scan), |
f1bc5c56 DW |
1011 | }; |
1012 | ||
9492750a | 1013 | if (xfs_want_reclaim_sick(mp)) |
b26b2bf1 | 1014 | icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK; |
9492750a | 1015 | |
8daaa831 | 1016 | /* kick background reclaimer and push the AIL */ |
5889608d | 1017 | xfs_reclaim_work_queue(mp); |
8daaa831 | 1018 | xfs_ail_push_all(mp->m_ail); |
a7b339f1 | 1019 | |
b26b2bf1 | 1020 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw); |
617825fe | 1021 | return 0; |
8daaa831 | 1022 | } |
9bf729c0 | 1023 | |
8daaa831 DC |
1024 | /* |
1025 | * Return the number of reclaimable inodes in the filesystem for | |
1026 | * the shrinker to determine how much to reclaim. | |
1027 | */ | |
10be350b | 1028 | long |
8daaa831 DC |
1029 | xfs_reclaim_inodes_count( |
1030 | struct xfs_mount *mp) | |
1031 | { | |
1032 | struct xfs_perag *pag; | |
1033 | xfs_agnumber_t ag = 0; | |
10be350b | 1034 | long reclaimable = 0; |
9bf729c0 | 1035 | |
65d0f205 DC |
1036 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
1037 | ag = pag->pag_agno + 1; | |
70e60ce7 DC |
1038 | reclaimable += pag->pag_ici_reclaimable; |
1039 | xfs_perag_put(pag); | |
9bf729c0 | 1040 | } |
9bf729c0 DC |
1041 | return reclaimable; |
1042 | } | |
1043 | ||
39b1cfd7 | 1044 | STATIC bool |
b26b2bf1 | 1045 | xfs_icwalk_match_id( |
3e3f9f58 | 1046 | struct xfs_inode *ip, |
b26b2bf1 | 1047 | struct xfs_icwalk *icw) |
3e3f9f58 | 1048 | { |
b26b2bf1 DW |
1049 | if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) && |
1050 | !uid_eq(VFS_I(ip)->i_uid, icw->icw_uid)) | |
39b1cfd7 | 1051 | return false; |
3e3f9f58 | 1052 | |
b26b2bf1 DW |
1053 | if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) && |
1054 | !gid_eq(VFS_I(ip)->i_gid, icw->icw_gid)) | |
39b1cfd7 | 1055 | return false; |
1b556048 | 1056 | |
b26b2bf1 DW |
1057 | if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) && |
1058 | ip->i_projid != icw->icw_prid) | |
39b1cfd7 | 1059 | return false; |
1b556048 | 1060 | |
39b1cfd7 | 1061 | return true; |
3e3f9f58 BF |
1062 | } |
1063 | ||
f4526397 BF |
1064 | /* |
1065 | * A union-based inode filtering algorithm. Process the inode if any of the | |
1066 | * criteria match. This is for global/internal scans only. | |
1067 | */ | |
39b1cfd7 | 1068 | STATIC bool |
b26b2bf1 | 1069 | xfs_icwalk_match_id_union( |
f4526397 | 1070 | struct xfs_inode *ip, |
b26b2bf1 | 1071 | struct xfs_icwalk *icw) |
f4526397 | 1072 | { |
b26b2bf1 DW |
1073 | if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) && |
1074 | uid_eq(VFS_I(ip)->i_uid, icw->icw_uid)) | |
39b1cfd7 | 1075 | return true; |
f4526397 | 1076 | |
b26b2bf1 DW |
1077 | if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) && |
1078 | gid_eq(VFS_I(ip)->i_gid, icw->icw_gid)) | |
39b1cfd7 | 1079 | return true; |
f4526397 | 1080 | |
b26b2bf1 DW |
1081 | if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) && |
1082 | ip->i_projid == icw->icw_prid) | |
39b1cfd7 | 1083 | return true; |
f4526397 | 1084 | |
39b1cfd7 | 1085 | return false; |
f4526397 BF |
1086 | } |
1087 | ||
a91bf992 DW |
1088 | /* |
1089 | * Is this inode @ip eligible for eof/cow block reclamation, given some | |
b26b2bf1 | 1090 | * filtering parameters @icw? The inode is eligible if @icw is null or |
a91bf992 DW |
1091 | * if the predicate functions match. |
1092 | */ | |
1093 | static bool | |
b26b2bf1 | 1094 | xfs_icwalk_match( |
a91bf992 | 1095 | struct xfs_inode *ip, |
b26b2bf1 | 1096 | struct xfs_icwalk *icw) |
a91bf992 | 1097 | { |
39b1cfd7 | 1098 | bool match; |
a91bf992 | 1099 | |
b26b2bf1 | 1100 | if (!icw) |
a91bf992 DW |
1101 | return true; |
1102 | ||
b26b2bf1 DW |
1103 | if (icw->icw_flags & XFS_ICWALK_FLAG_UNION) |
1104 | match = xfs_icwalk_match_id_union(ip, icw); | |
a91bf992 | 1105 | else |
b26b2bf1 | 1106 | match = xfs_icwalk_match_id(ip, icw); |
a91bf992 DW |
1107 | if (!match) |
1108 | return false; | |
1109 | ||
1110 | /* skip the inode if the file size is too small */ | |
b26b2bf1 DW |
1111 | if ((icw->icw_flags & XFS_ICWALK_FLAG_MINFILESIZE) && |
1112 | XFS_ISIZE(ip) < icw->icw_min_file_size) | |
a91bf992 DW |
1113 | return false; |
1114 | ||
1115 | return true; | |
1116 | } | |
1117 | ||
4d0bab3a DC |
1118 | /* |
1119 | * This is a fast pass over the inode cache to try to get reclaim moving on as | |
1120 | * many inodes as possible in a short period of time. It kicks itself every few | |
1121 | * seconds, as well as being kicked by the inode cache shrinker when memory | |
02511a5a | 1122 | * goes low. |
4d0bab3a DC |
1123 | */ |
1124 | void | |
1125 | xfs_reclaim_worker( | |
1126 | struct work_struct *work) | |
1127 | { | |
1128 | struct xfs_mount *mp = container_of(to_delayed_work(work), | |
1129 | struct xfs_mount, m_reclaim_work); | |
4d0bab3a | 1130 | |
f1bc5c56 | 1131 | xfs_icwalk(mp, XFS_ICWALK_RECLAIM, NULL); |
4d0bab3a DC |
1132 | xfs_reclaim_work_queue(mp); |
1133 | } | |
1134 | ||
41176a68 BF |
1135 | STATIC int |
1136 | xfs_inode_free_eofblocks( | |
1137 | struct xfs_inode *ip, | |
b26b2bf1 | 1138 | struct xfs_icwalk *icw, |
0fa4a10a | 1139 | unsigned int *lockflags) |
41176a68 | 1140 | { |
390600f8 | 1141 | bool wait; |
390600f8 | 1142 | |
b26b2bf1 | 1143 | wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC); |
5400da7d | 1144 | |
ce2d3bbe DW |
1145 | if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS)) |
1146 | return 0; | |
1147 | ||
41176a68 BF |
1148 | /* |
1149 | * If the mapping is dirty the operation can block and wait for some | |
1150 | * time. Unless we are waiting, skip it. | |
1151 | */ | |
390600f8 | 1152 | if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY)) |
41176a68 BF |
1153 | return 0; |
1154 | ||
b26b2bf1 | 1155 | if (!xfs_icwalk_match(ip, icw)) |
a91bf992 | 1156 | return 0; |
3e3f9f58 | 1157 | |
a36b9261 BF |
1158 | /* |
1159 | * If the caller is waiting, return -EAGAIN to keep the background | |
1160 | * scanner moving and revisit the inode in a subsequent pass. | |
1161 | */ | |
c3155097 | 1162 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
390600f8 DW |
1163 | if (wait) |
1164 | return -EAGAIN; | |
1165 | return 0; | |
a36b9261 | 1166 | } |
0fa4a10a | 1167 | *lockflags |= XFS_IOLOCK_EXCL; |
390600f8 | 1168 | |
2b156ff8 DW |
1169 | if (xfs_can_free_eofblocks(ip, false)) |
1170 | return xfs_free_eofblocks(ip); | |
1171 | ||
1172 | /* inode could be preallocated or append-only */ | |
1173 | trace_xfs_inode_free_eofblocks_invalid(ip); | |
1174 | xfs_inode_clear_eofblocks_tag(ip); | |
1175 | return 0; | |
41176a68 BF |
1176 | } |
1177 | ||
83104d44 | 1178 | static void |
ce2d3bbe DW |
1179 | xfs_blockgc_set_iflag( |
1180 | struct xfs_inode *ip, | |
ce2d3bbe | 1181 | unsigned long iflag) |
27b52867 | 1182 | { |
ce2d3bbe DW |
1183 | struct xfs_mount *mp = ip->i_mount; |
1184 | struct xfs_perag *pag; | |
ce2d3bbe DW |
1185 | |
1186 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1187 | |
85a6e764 CH |
1188 | /* |
1189 | * Don't bother locking the AG and looking up in the radix trees | |
1190 | * if we already know that we have the tag set. | |
1191 | */ | |
ce2d3bbe | 1192 | if (ip->i_flags & iflag) |
85a6e764 CH |
1193 | return; |
1194 | spin_lock(&ip->i_flags_lock); | |
ce2d3bbe | 1195 | ip->i_flags |= iflag; |
85a6e764 CH |
1196 | spin_unlock(&ip->i_flags_lock); |
1197 | ||
27b52867 BF |
1198 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1199 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1200 | |
c076ae7a DW |
1201 | xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1202 | XFS_ICI_BLOCKGC_TAG); | |
27b52867 BF |
1203 | |
1204 | spin_unlock(&pag->pag_ici_lock); | |
1205 | xfs_perag_put(pag); | |
1206 | } | |
1207 | ||
1208 | void | |
83104d44 | 1209 | xfs_inode_set_eofblocks_tag( |
27b52867 | 1210 | xfs_inode_t *ip) |
83104d44 DW |
1211 | { |
1212 | trace_xfs_inode_set_eofblocks_tag(ip); | |
9669f51d | 1213 | return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1214 | } |
1215 | ||
1216 | static void | |
ce2d3bbe DW |
1217 | xfs_blockgc_clear_iflag( |
1218 | struct xfs_inode *ip, | |
1219 | unsigned long iflag) | |
27b52867 | 1220 | { |
ce2d3bbe DW |
1221 | struct xfs_mount *mp = ip->i_mount; |
1222 | struct xfs_perag *pag; | |
1223 | bool clear_tag; | |
1224 | ||
1225 | ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0); | |
27b52867 | 1226 | |
85a6e764 | 1227 | spin_lock(&ip->i_flags_lock); |
ce2d3bbe DW |
1228 | ip->i_flags &= ~iflag; |
1229 | clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0; | |
85a6e764 CH |
1230 | spin_unlock(&ip->i_flags_lock); |
1231 | ||
ce2d3bbe DW |
1232 | if (!clear_tag) |
1233 | return; | |
1234 | ||
27b52867 BF |
1235 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1236 | spin_lock(&pag->pag_ici_lock); | |
27b52867 | 1237 | |
c076ae7a DW |
1238 | xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1239 | XFS_ICI_BLOCKGC_TAG); | |
27b52867 BF |
1240 | |
1241 | spin_unlock(&pag->pag_ici_lock); | |
1242 | xfs_perag_put(pag); | |
1243 | } | |
1244 | ||
83104d44 DW |
1245 | void |
1246 | xfs_inode_clear_eofblocks_tag( | |
1247 | xfs_inode_t *ip) | |
1248 | { | |
1249 | trace_xfs_inode_clear_eofblocks_tag(ip); | |
ce2d3bbe | 1250 | return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS); |
83104d44 DW |
1251 | } |
1252 | ||
1253 | /* | |
be78ff0e DW |
1254 | * Set ourselves up to free CoW blocks from this file. If it's already clean |
1255 | * then we can bail out quickly, but otherwise we must back off if the file | |
1256 | * is undergoing some kind of write. | |
83104d44 | 1257 | */ |
be78ff0e DW |
1258 | static bool |
1259 | xfs_prep_free_cowblocks( | |
51d62690 | 1260 | struct xfs_inode *ip) |
83104d44 | 1261 | { |
39937234 BF |
1262 | /* |
1263 | * Just clear the tag if we have an empty cow fork or none at all. It's | |
1264 | * possible the inode was fully unshared since it was originally tagged. | |
1265 | */ | |
51d62690 | 1266 | if (!xfs_inode_has_cow_data(ip)) { |
83104d44 DW |
1267 | trace_xfs_inode_free_cowblocks_invalid(ip); |
1268 | xfs_inode_clear_cowblocks_tag(ip); | |
be78ff0e | 1269 | return false; |
83104d44 DW |
1270 | } |
1271 | ||
1272 | /* | |
1273 | * If the mapping is dirty or under writeback we cannot touch the | |
1274 | * CoW fork. Leave it alone if we're in the midst of a directio. | |
1275 | */ | |
a1b7a4de CH |
1276 | if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) || |
1277 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) || | |
83104d44 DW |
1278 | mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) || |
1279 | atomic_read(&VFS_I(ip)->i_dio_count)) | |
be78ff0e DW |
1280 | return false; |
1281 | ||
1282 | return true; | |
1283 | } | |
1284 | ||
1285 | /* | |
1286 | * Automatic CoW Reservation Freeing | |
1287 | * | |
1288 | * These functions automatically garbage collect leftover CoW reservations | |
1289 | * that were made on behalf of a cowextsize hint when we start to run out | |
1290 | * of quota or when the reservations sit around for too long. If the file | |
1291 | * has dirty pages or is undergoing writeback, its CoW reservations will | |
1292 | * be retained. | |
1293 | * | |
1294 | * The actual garbage collection piggybacks off the same code that runs | |
1295 | * the speculative EOF preallocation garbage collector. | |
1296 | */ | |
1297 | STATIC int | |
1298 | xfs_inode_free_cowblocks( | |
1299 | struct xfs_inode *ip, | |
b26b2bf1 | 1300 | struct xfs_icwalk *icw, |
0fa4a10a | 1301 | unsigned int *lockflags) |
be78ff0e | 1302 | { |
f41a0716 | 1303 | bool wait; |
be78ff0e DW |
1304 | int ret = 0; |
1305 | ||
b26b2bf1 | 1306 | wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC); |
f41a0716 | 1307 | |
ce2d3bbe DW |
1308 | if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS)) |
1309 | return 0; | |
1310 | ||
51d62690 | 1311 | if (!xfs_prep_free_cowblocks(ip)) |
83104d44 DW |
1312 | return 0; |
1313 | ||
b26b2bf1 | 1314 | if (!xfs_icwalk_match(ip, icw)) |
a91bf992 | 1315 | return 0; |
83104d44 | 1316 | |
f41a0716 DW |
1317 | /* |
1318 | * If the caller is waiting, return -EAGAIN to keep the background | |
1319 | * scanner moving and revisit the inode in a subsequent pass. | |
1320 | */ | |
0fa4a10a DW |
1321 | if (!(*lockflags & XFS_IOLOCK_EXCL) && |
1322 | !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
f41a0716 DW |
1323 | if (wait) |
1324 | return -EAGAIN; | |
1325 | return 0; | |
1326 | } | |
0fa4a10a DW |
1327 | *lockflags |= XFS_IOLOCK_EXCL; |
1328 | ||
f41a0716 DW |
1329 | if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) { |
1330 | if (wait) | |
0fa4a10a DW |
1331 | return -EAGAIN; |
1332 | return 0; | |
f41a0716 | 1333 | } |
0fa4a10a | 1334 | *lockflags |= XFS_MMAPLOCK_EXCL; |
83104d44 | 1335 | |
be78ff0e DW |
1336 | /* |
1337 | * Check again, nobody else should be able to dirty blocks or change | |
1338 | * the reflink iflag now that we have the first two locks held. | |
1339 | */ | |
51d62690 | 1340 | if (xfs_prep_free_cowblocks(ip)) |
be78ff0e | 1341 | ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false); |
83104d44 DW |
1342 | return ret; |
1343 | } | |
1344 | ||
83104d44 DW |
1345 | void |
1346 | xfs_inode_set_cowblocks_tag( | |
1347 | xfs_inode_t *ip) | |
1348 | { | |
7b7381f0 | 1349 | trace_xfs_inode_set_cowblocks_tag(ip); |
9669f51d | 1350 | return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 DW |
1351 | } |
1352 | ||
1353 | void | |
1354 | xfs_inode_clear_cowblocks_tag( | |
1355 | xfs_inode_t *ip) | |
1356 | { | |
7b7381f0 | 1357 | trace_xfs_inode_clear_cowblocks_tag(ip); |
ce2d3bbe | 1358 | return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS); |
83104d44 | 1359 | } |
d6b636eb DW |
1360 | |
1361 | /* Disable post-EOF and CoW block auto-reclamation. */ | |
1362 | void | |
c9a6526f | 1363 | xfs_blockgc_stop( |
d6b636eb DW |
1364 | struct xfs_mount *mp) |
1365 | { | |
894ecacf DW |
1366 | struct xfs_perag *pag; |
1367 | xfs_agnumber_t agno; | |
1368 | ||
6f649091 DW |
1369 | if (!xfs_clear_blockgc_enabled(mp)) |
1370 | return; | |
1371 | ||
1372 | for_each_perag(mp, agno, pag) | |
894ecacf | 1373 | cancel_delayed_work_sync(&pag->pag_blockgc_work); |
6f649091 | 1374 | trace_xfs_blockgc_stop(mp, __return_address); |
d6b636eb DW |
1375 | } |
1376 | ||
1377 | /* Enable post-EOF and CoW block auto-reclamation. */ | |
1378 | void | |
c9a6526f | 1379 | xfs_blockgc_start( |
d6b636eb DW |
1380 | struct xfs_mount *mp) |
1381 | { | |
894ecacf DW |
1382 | struct xfs_perag *pag; |
1383 | xfs_agnumber_t agno; | |
1384 | ||
6f649091 DW |
1385 | if (xfs_set_blockgc_enabled(mp)) |
1386 | return; | |
1387 | ||
1388 | trace_xfs_blockgc_start(mp, __return_address); | |
894ecacf DW |
1389 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) |
1390 | xfs_blockgc_queue(pag); | |
d6b636eb | 1391 | } |
3d4feec0 | 1392 | |
d20d5edc DW |
1393 | /* Don't try to run block gc on an inode that's in any of these states. */ |
1394 | #define XFS_BLOCKGC_NOGRAB_IFLAGS (XFS_INEW | \ | |
ab23a776 DC |
1395 | XFS_NEED_INACTIVE | \ |
1396 | XFS_INACTIVATING | \ | |
d20d5edc DW |
1397 | XFS_IRECLAIMABLE | \ |
1398 | XFS_IRECLAIM) | |
df600197 | 1399 | /* |
b9baaef4 DW |
1400 | * Decide if the given @ip is eligible for garbage collection of speculative |
1401 | * preallocations, and grab it if so. Returns true if it's ready to go or | |
1402 | * false if we should just ignore it. | |
df600197 DW |
1403 | */ |
1404 | static bool | |
b9baaef4 | 1405 | xfs_blockgc_igrab( |
7fdff526 | 1406 | struct xfs_inode *ip) |
df600197 DW |
1407 | { |
1408 | struct inode *inode = VFS_I(ip); | |
df600197 DW |
1409 | |
1410 | ASSERT(rcu_read_lock_held()); | |
1411 | ||
1412 | /* Check for stale RCU freed inode */ | |
1413 | spin_lock(&ip->i_flags_lock); | |
1414 | if (!ip->i_ino) | |
1415 | goto out_unlock_noent; | |
1416 | ||
d20d5edc | 1417 | if (ip->i_flags & XFS_BLOCKGC_NOGRAB_IFLAGS) |
df600197 DW |
1418 | goto out_unlock_noent; |
1419 | spin_unlock(&ip->i_flags_lock); | |
1420 | ||
1421 | /* nothing to sync during shutdown */ | |
75c8c50f | 1422 | if (xfs_is_shutdown(ip->i_mount)) |
df600197 DW |
1423 | return false; |
1424 | ||
1425 | /* If we can't grab the inode, it must on it's way to reclaim. */ | |
1426 | if (!igrab(inode)) | |
1427 | return false; | |
1428 | ||
1429 | /* inode is valid */ | |
1430 | return true; | |
1431 | ||
1432 | out_unlock_noent: | |
1433 | spin_unlock(&ip->i_flags_lock); | |
1434 | return false; | |
1435 | } | |
1436 | ||
41956753 DW |
1437 | /* Scan one incore inode for block preallocations that we can remove. */ |
1438 | static int | |
1439 | xfs_blockgc_scan_inode( | |
1440 | struct xfs_inode *ip, | |
b26b2bf1 | 1441 | struct xfs_icwalk *icw) |
85c5b270 | 1442 | { |
0fa4a10a | 1443 | unsigned int lockflags = 0; |
85c5b270 DW |
1444 | int error; |
1445 | ||
b26b2bf1 | 1446 | error = xfs_inode_free_eofblocks(ip, icw, &lockflags); |
85c5b270 | 1447 | if (error) |
0fa4a10a | 1448 | goto unlock; |
85c5b270 | 1449 | |
b26b2bf1 | 1450 | error = xfs_inode_free_cowblocks(ip, icw, &lockflags); |
0fa4a10a DW |
1451 | unlock: |
1452 | if (lockflags) | |
1453 | xfs_iunlock(ip, lockflags); | |
594ab00b | 1454 | xfs_irele(ip); |
0fa4a10a | 1455 | return error; |
85c5b270 DW |
1456 | } |
1457 | ||
9669f51d DW |
1458 | /* Background worker that trims preallocated space. */ |
1459 | void | |
1460 | xfs_blockgc_worker( | |
1461 | struct work_struct *work) | |
1462 | { | |
894ecacf DW |
1463 | struct xfs_perag *pag = container_of(to_delayed_work(work), |
1464 | struct xfs_perag, pag_blockgc_work); | |
1465 | struct xfs_mount *mp = pag->pag_mount; | |
9669f51d DW |
1466 | int error; |
1467 | ||
6f649091 DW |
1468 | trace_xfs_blockgc_worker(mp, __return_address); |
1469 | ||
f427cf5c | 1470 | error = xfs_icwalk_ag(pag, XFS_ICWALK_BLOCKGC, NULL); |
9669f51d | 1471 | if (error) |
894ecacf DW |
1472 | xfs_info(mp, "AG %u preallocation gc worker failed, err=%d", |
1473 | pag->pag_agno, error); | |
894ecacf | 1474 | xfs_blockgc_queue(pag); |
9669f51d DW |
1475 | } |
1476 | ||
85c5b270 | 1477 | /* |
2eb66502 DW |
1478 | * Try to free space in the filesystem by purging inactive inodes, eofblocks |
1479 | * and cowblocks. | |
85c5b270 DW |
1480 | */ |
1481 | int | |
1482 | xfs_blockgc_free_space( | |
1483 | struct xfs_mount *mp, | |
b26b2bf1 | 1484 | struct xfs_icwalk *icw) |
85c5b270 | 1485 | { |
2eb66502 DW |
1486 | int error; |
1487 | ||
b26b2bf1 | 1488 | trace_xfs_blockgc_free_space(mp, icw, _RET_IP_); |
85c5b270 | 1489 | |
2eb66502 DW |
1490 | error = xfs_icwalk(mp, XFS_ICWALK_BLOCKGC, icw); |
1491 | if (error) | |
1492 | return error; | |
1493 | ||
1494 | xfs_inodegc_flush(mp); | |
1495 | return 0; | |
85c5b270 DW |
1496 | } |
1497 | ||
e8d04c2a DW |
1498 | /* |
1499 | * Reclaim all the free space that we can by scheduling the background blockgc | |
1500 | * and inodegc workers immediately and waiting for them all to clear. | |
1501 | */ | |
1502 | void | |
1503 | xfs_blockgc_flush_all( | |
1504 | struct xfs_mount *mp) | |
1505 | { | |
1506 | struct xfs_perag *pag; | |
1507 | xfs_agnumber_t agno; | |
1508 | ||
1509 | trace_xfs_blockgc_flush_all(mp, __return_address); | |
1510 | ||
1511 | /* | |
1512 | * For each blockgc worker, move its queue time up to now. If it | |
1513 | * wasn't queued, it will not be requeued. Then flush whatever's | |
1514 | * left. | |
1515 | */ | |
1516 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1517 | mod_delayed_work(pag->pag_mount->m_blockgc_wq, | |
1518 | &pag->pag_blockgc_work, 0); | |
1519 | ||
1520 | for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG) | |
1521 | flush_delayed_work(&pag->pag_blockgc_work); | |
1522 | ||
1523 | xfs_inodegc_flush(mp); | |
1524 | } | |
1525 | ||
3d4feec0 | 1526 | /* |
c237dd7c DW |
1527 | * Run cow/eofblocks scans on the supplied dquots. We don't know exactly which |
1528 | * quota caused an allocation failure, so we make a best effort by including | |
1529 | * each quota under low free space conditions (less than 1% free space) in the | |
1530 | * scan. | |
111068f8 DW |
1531 | * |
1532 | * Callers must not hold any inode's ILOCK. If requesting a synchronous scan | |
2d53f66b | 1533 | * (XFS_ICWALK_FLAG_SYNC), the caller also must not hold any inode's IOLOCK or |
111068f8 | 1534 | * MMAPLOCK. |
3d4feec0 | 1535 | */ |
111068f8 | 1536 | int |
c237dd7c DW |
1537 | xfs_blockgc_free_dquots( |
1538 | struct xfs_mount *mp, | |
1539 | struct xfs_dquot *udqp, | |
1540 | struct xfs_dquot *gdqp, | |
1541 | struct xfs_dquot *pdqp, | |
2d53f66b | 1542 | unsigned int iwalk_flags) |
3d4feec0 | 1543 | { |
b26b2bf1 | 1544 | struct xfs_icwalk icw = {0}; |
3d4feec0 DW |
1545 | bool do_work = false; |
1546 | ||
c237dd7c DW |
1547 | if (!udqp && !gdqp && !pdqp) |
1548 | return 0; | |
1549 | ||
3d4feec0 | 1550 | /* |
111068f8 DW |
1551 | * Run a scan to free blocks using the union filter to cover all |
1552 | * applicable quotas in a single scan. | |
3d4feec0 | 1553 | */ |
b26b2bf1 | 1554 | icw.icw_flags = XFS_ICWALK_FLAG_UNION | iwalk_flags; |
3d4feec0 | 1555 | |
c237dd7c | 1556 | if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) { |
b26b2bf1 DW |
1557 | icw.icw_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id); |
1558 | icw.icw_flags |= XFS_ICWALK_FLAG_UID; | |
c237dd7c | 1559 | do_work = true; |
3d4feec0 DW |
1560 | } |
1561 | ||
c237dd7c | 1562 | if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) { |
b26b2bf1 DW |
1563 | icw.icw_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id); |
1564 | icw.icw_flags |= XFS_ICWALK_FLAG_GID; | |
c237dd7c | 1565 | do_work = true; |
3d4feec0 DW |
1566 | } |
1567 | ||
c237dd7c | 1568 | if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) { |
b26b2bf1 DW |
1569 | icw.icw_prid = pdqp->q_id; |
1570 | icw.icw_flags |= XFS_ICWALK_FLAG_PRID; | |
c237dd7c | 1571 | do_work = true; |
3d4feec0 DW |
1572 | } |
1573 | ||
1574 | if (!do_work) | |
111068f8 | 1575 | return 0; |
3d4feec0 | 1576 | |
b26b2bf1 | 1577 | return xfs_blockgc_free_space(mp, &icw); |
c237dd7c DW |
1578 | } |
1579 | ||
1580 | /* Run cow/eofblocks scans on the quotas attached to the inode. */ | |
1581 | int | |
1582 | xfs_blockgc_free_quota( | |
1583 | struct xfs_inode *ip, | |
2d53f66b | 1584 | unsigned int iwalk_flags) |
c237dd7c DW |
1585 | { |
1586 | return xfs_blockgc_free_dquots(ip->i_mount, | |
1587 | xfs_inode_dquot(ip, XFS_DQTYPE_USER), | |
1588 | xfs_inode_dquot(ip, XFS_DQTYPE_GROUP), | |
2d53f66b | 1589 | xfs_inode_dquot(ip, XFS_DQTYPE_PROJ), iwalk_flags); |
3d4feec0 | 1590 | } |
df600197 DW |
1591 | |
1592 | /* XFS Inode Cache Walking Code */ | |
1593 | ||
f1bc5c56 DW |
1594 | /* |
1595 | * The inode lookup is done in batches to keep the amount of lock traffic and | |
1596 | * radix tree lookups to a minimum. The batch size is a trade off between | |
1597 | * lookup reduction and stack usage. This is in the reclaim path, so we can't | |
1598 | * be too greedy. | |
1599 | */ | |
1600 | #define XFS_LOOKUP_BATCH 32 | |
1601 | ||
1602 | ||
b9baaef4 DW |
1603 | /* |
1604 | * Decide if we want to grab this inode in anticipation of doing work towards | |
594ab00b | 1605 | * the goal. |
b9baaef4 DW |
1606 | */ |
1607 | static inline bool | |
1608 | xfs_icwalk_igrab( | |
1609 | enum xfs_icwalk_goal goal, | |
9492750a | 1610 | struct xfs_inode *ip, |
b26b2bf1 | 1611 | struct xfs_icwalk *icw) |
b9baaef4 DW |
1612 | { |
1613 | switch (goal) { | |
b9baaef4 | 1614 | case XFS_ICWALK_BLOCKGC: |
7fdff526 | 1615 | return xfs_blockgc_igrab(ip); |
f1bc5c56 | 1616 | case XFS_ICWALK_RECLAIM: |
b26b2bf1 | 1617 | return xfs_reclaim_igrab(ip, icw); |
b9baaef4 DW |
1618 | default: |
1619 | return false; | |
1620 | } | |
1621 | } | |
1622 | ||
594ab00b DW |
1623 | /* |
1624 | * Process an inode. Each processing function must handle any state changes | |
1625 | * made by the icwalk igrab function. Return -EAGAIN to skip an inode. | |
1626 | */ | |
f427cf5c DW |
1627 | static inline int |
1628 | xfs_icwalk_process_inode( | |
1629 | enum xfs_icwalk_goal goal, | |
1630 | struct xfs_inode *ip, | |
f1bc5c56 | 1631 | struct xfs_perag *pag, |
b26b2bf1 | 1632 | struct xfs_icwalk *icw) |
f427cf5c | 1633 | { |
594ab00b | 1634 | int error = 0; |
f427cf5c DW |
1635 | |
1636 | switch (goal) { | |
f427cf5c | 1637 | case XFS_ICWALK_BLOCKGC: |
b26b2bf1 | 1638 | error = xfs_blockgc_scan_inode(ip, icw); |
f427cf5c | 1639 | break; |
f1bc5c56 DW |
1640 | case XFS_ICWALK_RECLAIM: |
1641 | xfs_reclaim_inode(ip, pag); | |
1642 | break; | |
f427cf5c | 1643 | } |
f427cf5c DW |
1644 | return error; |
1645 | } | |
1646 | ||
df600197 | 1647 | /* |
f427cf5c DW |
1648 | * For a given per-AG structure @pag and a goal, grab qualifying inodes and |
1649 | * process them in some manner. | |
df600197 DW |
1650 | */ |
1651 | static int | |
c1115c0c | 1652 | xfs_icwalk_ag( |
df600197 | 1653 | struct xfs_perag *pag, |
f427cf5c | 1654 | enum xfs_icwalk_goal goal, |
b26b2bf1 | 1655 | struct xfs_icwalk *icw) |
df600197 DW |
1656 | { |
1657 | struct xfs_mount *mp = pag->pag_mount; | |
1658 | uint32_t first_index; | |
1659 | int last_error = 0; | |
1660 | int skipped; | |
1661 | bool done; | |
1662 | int nr_found; | |
1663 | ||
1664 | restart: | |
1665 | done = false; | |
1666 | skipped = 0; | |
f1bc5c56 DW |
1667 | if (goal == XFS_ICWALK_RECLAIM) |
1668 | first_index = READ_ONCE(pag->pag_ici_reclaim_cursor); | |
1669 | else | |
1670 | first_index = 0; | |
df600197 DW |
1671 | nr_found = 0; |
1672 | do { | |
1673 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; | |
1674 | int error = 0; | |
1675 | int i; | |
1676 | ||
1677 | rcu_read_lock(); | |
1678 | ||
a437b9b4 CH |
1679 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, |
1680 | (void **) batch, first_index, | |
1681 | XFS_LOOKUP_BATCH, goal); | |
df600197 | 1682 | if (!nr_found) { |
f1bc5c56 | 1683 | done = true; |
df600197 DW |
1684 | rcu_read_unlock(); |
1685 | break; | |
1686 | } | |
1687 | ||
1688 | /* | |
1689 | * Grab the inodes before we drop the lock. if we found | |
1690 | * nothing, nr == 0 and the loop will be skipped. | |
1691 | */ | |
1692 | for (i = 0; i < nr_found; i++) { | |
1693 | struct xfs_inode *ip = batch[i]; | |
1694 | ||
b26b2bf1 | 1695 | if (done || !xfs_icwalk_igrab(goal, ip, icw)) |
df600197 DW |
1696 | batch[i] = NULL; |
1697 | ||
1698 | /* | |
1699 | * Update the index for the next lookup. Catch | |
1700 | * overflows into the next AG range which can occur if | |
1701 | * we have inodes in the last block of the AG and we | |
1702 | * are currently pointing to the last inode. | |
1703 | * | |
1704 | * Because we may see inodes that are from the wrong AG | |
1705 | * due to RCU freeing and reallocation, only update the | |
1706 | * index if it lies in this AG. It was a race that lead | |
1707 | * us to see this inode, so another lookup from the | |
1708 | * same index will not find it again. | |
1709 | */ | |
1710 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) | |
1711 | continue; | |
1712 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
1713 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) | |
1714 | done = true; | |
1715 | } | |
1716 | ||
1717 | /* unlock now we've grabbed the inodes. */ | |
1718 | rcu_read_unlock(); | |
1719 | ||
1720 | for (i = 0; i < nr_found; i++) { | |
1721 | if (!batch[i]) | |
1722 | continue; | |
f1bc5c56 | 1723 | error = xfs_icwalk_process_inode(goal, batch[i], pag, |
b26b2bf1 | 1724 | icw); |
df600197 DW |
1725 | if (error == -EAGAIN) { |
1726 | skipped++; | |
1727 | continue; | |
1728 | } | |
1729 | if (error && last_error != -EFSCORRUPTED) | |
1730 | last_error = error; | |
1731 | } | |
1732 | ||
1733 | /* bail out if the filesystem is corrupted. */ | |
1734 | if (error == -EFSCORRUPTED) | |
1735 | break; | |
1736 | ||
1737 | cond_resched(); | |
1738 | ||
b26b2bf1 DW |
1739 | if (icw && (icw->icw_flags & XFS_ICWALK_FLAG_SCAN_LIMIT)) { |
1740 | icw->icw_scan_limit -= XFS_LOOKUP_BATCH; | |
1741 | if (icw->icw_scan_limit <= 0) | |
f1bc5c56 DW |
1742 | break; |
1743 | } | |
df600197 DW |
1744 | } while (nr_found && !done); |
1745 | ||
f1bc5c56 DW |
1746 | if (goal == XFS_ICWALK_RECLAIM) { |
1747 | if (done) | |
1748 | first_index = 0; | |
1749 | WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index); | |
1750 | } | |
1751 | ||
df600197 DW |
1752 | if (skipped) { |
1753 | delay(1); | |
1754 | goto restart; | |
1755 | } | |
1756 | return last_error; | |
1757 | } | |
1758 | ||
f427cf5c | 1759 | /* Walk all incore inodes to achieve a given goal. */ |
df600197 | 1760 | static int |
c1115c0c | 1761 | xfs_icwalk( |
df600197 | 1762 | struct xfs_mount *mp, |
f427cf5c | 1763 | enum xfs_icwalk_goal goal, |
b26b2bf1 | 1764 | struct xfs_icwalk *icw) |
df600197 DW |
1765 | { |
1766 | struct xfs_perag *pag; | |
1767 | int error = 0; | |
1768 | int last_error = 0; | |
a437b9b4 | 1769 | xfs_agnumber_t agno; |
df600197 | 1770 | |
a437b9b4 | 1771 | for_each_perag_tag(mp, agno, pag, goal) { |
b26b2bf1 | 1772 | error = xfs_icwalk_ag(pag, goal, icw); |
df600197 DW |
1773 | if (error) { |
1774 | last_error = error; | |
a437b9b4 | 1775 | if (error == -EFSCORRUPTED) { |
c4d5660a | 1776 | xfs_perag_rele(pag); |
df600197 | 1777 | break; |
a437b9b4 | 1778 | } |
df600197 DW |
1779 | } |
1780 | } | |
1781 | return last_error; | |
2d53f66b | 1782 | BUILD_BUG_ON(XFS_ICWALK_PRIVATE_FLAGS & XFS_ICWALK_FLAGS_VALID); |
df600197 | 1783 | } |
c6c2066d DW |
1784 | |
1785 | #ifdef DEBUG | |
1786 | static void | |
1787 | xfs_check_delalloc( | |
1788 | struct xfs_inode *ip, | |
1789 | int whichfork) | |
1790 | { | |
732436ef | 1791 | struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork); |
c6c2066d DW |
1792 | struct xfs_bmbt_irec got; |
1793 | struct xfs_iext_cursor icur; | |
1794 | ||
1795 | if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got)) | |
1796 | return; | |
1797 | do { | |
1798 | if (isnullstartblock(got.br_startblock)) { | |
1799 | xfs_warn(ip->i_mount, | |
1800 | "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]", | |
1801 | ip->i_ino, | |
1802 | whichfork == XFS_DATA_FORK ? "data" : "cow", | |
1803 | got.br_startoff, got.br_blockcount); | |
1804 | } | |
1805 | } while (xfs_iext_next_extent(ifp, &icur, &got)); | |
1806 | } | |
1807 | #else | |
1808 | #define xfs_check_delalloc(ip, whichfork) do { } while (0) | |
1809 | #endif | |
1810 | ||
ab23a776 DC |
1811 | /* Schedule the inode for reclaim. */ |
1812 | static void | |
1813 | xfs_inodegc_set_reclaimable( | |
c6c2066d DW |
1814 | struct xfs_inode *ip) |
1815 | { | |
1816 | struct xfs_mount *mp = ip->i_mount; | |
1817 | struct xfs_perag *pag; | |
c6c2066d | 1818 | |
75c8c50f | 1819 | if (!xfs_is_shutdown(mp) && ip->i_delayed_blks) { |
c6c2066d DW |
1820 | xfs_check_delalloc(ip, XFS_DATA_FORK); |
1821 | xfs_check_delalloc(ip, XFS_COW_FORK); | |
1822 | ASSERT(0); | |
1823 | } | |
1824 | ||
c6c2066d DW |
1825 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1826 | spin_lock(&pag->pag_ici_lock); | |
1827 | spin_lock(&ip->i_flags_lock); | |
1828 | ||
ab23a776 DC |
1829 | trace_xfs_inode_set_reclaimable(ip); |
1830 | ip->i_flags &= ~(XFS_NEED_INACTIVE | XFS_INACTIVATING); | |
1831 | ip->i_flags |= XFS_IRECLAIMABLE; | |
c6c2066d DW |
1832 | xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino), |
1833 | XFS_ICI_RECLAIM_TAG); | |
c6c2066d DW |
1834 | |
1835 | spin_unlock(&ip->i_flags_lock); | |
1836 | spin_unlock(&pag->pag_ici_lock); | |
1837 | xfs_perag_put(pag); | |
1838 | } | |
ab23a776 DC |
1839 | |
1840 | /* | |
1841 | * Free all speculative preallocations and possibly even the inode itself. | |
1842 | * This is the last chance to make changes to an otherwise unreferenced file | |
1843 | * before incore reclamation happens. | |
1844 | */ | |
1845 | static void | |
1846 | xfs_inodegc_inactivate( | |
1847 | struct xfs_inode *ip) | |
1848 | { | |
1849 | trace_xfs_inode_inactivating(ip); | |
1850 | xfs_inactive(ip); | |
1851 | xfs_inodegc_set_reclaimable(ip); | |
1852 | } | |
1853 | ||
1854 | void | |
1855 | xfs_inodegc_worker( | |
1856 | struct work_struct *work) | |
1857 | { | |
7cf2b0f9 DC |
1858 | struct xfs_inodegc *gc = container_of(to_delayed_work(work), |
1859 | struct xfs_inodegc, work); | |
ab23a776 DC |
1860 | struct llist_node *node = llist_del_all(&gc->list); |
1861 | struct xfs_inode *ip, *n; | |
4da11251 | 1862 | unsigned int nofs_flag; |
ab23a776 | 1863 | |
b37c4c83 DW |
1864 | ASSERT(gc->cpu == smp_processor_id()); |
1865 | ||
ab23a776 DC |
1866 | WRITE_ONCE(gc->items, 0); |
1867 | ||
1868 | if (!node) | |
1869 | return; | |
1870 | ||
4da11251 WG |
1871 | /* |
1872 | * We can allocate memory here while doing writeback on behalf of | |
1873 | * memory reclaim. To avoid memory allocation deadlocks set the | |
1874 | * task-wide nofs context for the following operations. | |
1875 | */ | |
1876 | nofs_flag = memalloc_nofs_save(); | |
1877 | ||
ab23a776 | 1878 | ip = llist_entry(node, struct xfs_inode, i_gclist); |
40b1de00 | 1879 | trace_xfs_inodegc_worker(ip->i_mount, READ_ONCE(gc->shrinker_hits)); |
ab23a776 | 1880 | |
40b1de00 | 1881 | WRITE_ONCE(gc->shrinker_hits, 0); |
ab23a776 DC |
1882 | llist_for_each_entry_safe(ip, n, node, i_gclist) { |
1883 | xfs_iflags_set(ip, XFS_INACTIVATING); | |
1884 | xfs_inodegc_inactivate(ip); | |
1885 | } | |
4da11251 WG |
1886 | |
1887 | memalloc_nofs_restore(nofs_flag); | |
ab23a776 DC |
1888 | } |
1889 | ||
1890 | /* | |
5e672cd6 DC |
1891 | * Expedite all pending inodegc work to run immediately. This does not wait for |
1892 | * completion of the work. | |
ab23a776 DC |
1893 | */ |
1894 | void | |
5e672cd6 | 1895 | xfs_inodegc_push( |
ab23a776 DC |
1896 | struct xfs_mount *mp) |
1897 | { | |
ab23a776 DC |
1898 | if (!xfs_is_inodegc_enabled(mp)) |
1899 | return; | |
5e672cd6 DC |
1900 | trace_xfs_inodegc_push(mp, __return_address); |
1901 | xfs_inodegc_queue_all(mp); | |
1902 | } | |
ab23a776 | 1903 | |
5e672cd6 DC |
1904 | /* |
1905 | * Force all currently queued inode inactivation work to run immediately and | |
1906 | * wait for the work to finish. | |
1907 | */ | |
1908 | void | |
1909 | xfs_inodegc_flush( | |
1910 | struct xfs_mount *mp) | |
1911 | { | |
1912 | xfs_inodegc_push(mp); | |
ab23a776 | 1913 | trace_xfs_inodegc_flush(mp, __return_address); |
6191cf3a | 1914 | flush_workqueue(mp->m_inodegc_wq); |
ab23a776 DC |
1915 | } |
1916 | ||
1917 | /* | |
1918 | * Flush all the pending work and then disable the inode inactivation background | |
2254a739 DW |
1919 | * workers and wait for them to stop. Caller must hold sb->s_umount to |
1920 | * coordinate changes in the inodegc_enabled state. | |
ab23a776 DC |
1921 | */ |
1922 | void | |
1923 | xfs_inodegc_stop( | |
1924 | struct xfs_mount *mp) | |
1925 | { | |
2254a739 DW |
1926 | bool rerun; |
1927 | ||
ab23a776 DC |
1928 | if (!xfs_clear_inodegc_enabled(mp)) |
1929 | return; | |
1930 | ||
2254a739 DW |
1931 | /* |
1932 | * Drain all pending inodegc work, including inodes that could be | |
1933 | * queued by racing xfs_inodegc_queue or xfs_inodegc_shrinker_scan | |
1934 | * threads that sample the inodegc state just prior to us clearing it. | |
1935 | * The inodegc flag state prevents new threads from queuing more | |
1936 | * inodes, so we queue pending work items and flush the workqueue until | |
1937 | * all inodegc lists are empty. IOWs, we cannot use drain_workqueue | |
1938 | * here because it does not allow other unserialized mechanisms to | |
1939 | * reschedule inodegc work while this draining is in progress. | |
1940 | */ | |
ab23a776 | 1941 | xfs_inodegc_queue_all(mp); |
2254a739 DW |
1942 | do { |
1943 | flush_workqueue(mp->m_inodegc_wq); | |
1944 | rerun = xfs_inodegc_queue_all(mp); | |
1945 | } while (rerun); | |
ab23a776 | 1946 | |
ab23a776 DC |
1947 | trace_xfs_inodegc_stop(mp, __return_address); |
1948 | } | |
1949 | ||
1950 | /* | |
1951 | * Enable the inode inactivation background workers and schedule deferred inode | |
2254a739 DW |
1952 | * inactivation work if there is any. Caller must hold sb->s_umount to |
1953 | * coordinate changes in the inodegc_enabled state. | |
ab23a776 DC |
1954 | */ |
1955 | void | |
1956 | xfs_inodegc_start( | |
1957 | struct xfs_mount *mp) | |
1958 | { | |
1959 | if (xfs_set_inodegc_enabled(mp)) | |
1960 | return; | |
1961 | ||
1962 | trace_xfs_inodegc_start(mp, __return_address); | |
1963 | xfs_inodegc_queue_all(mp); | |
1964 | } | |
1965 | ||
65f03d86 DW |
1966 | #ifdef CONFIG_XFS_RT |
1967 | static inline bool | |
1968 | xfs_inodegc_want_queue_rt_file( | |
1969 | struct xfs_inode *ip) | |
1970 | { | |
1971 | struct xfs_mount *mp = ip->i_mount; | |
65f03d86 DW |
1972 | |
1973 | if (!XFS_IS_REALTIME_INODE(ip)) | |
1974 | return false; | |
1975 | ||
2229276c DW |
1976 | if (__percpu_counter_compare(&mp->m_frextents, |
1977 | mp->m_low_rtexts[XFS_LOWSP_5_PCNT], | |
1978 | XFS_FDBLOCKS_BATCH) < 0) | |
1979 | return true; | |
1980 | ||
1981 | return false; | |
65f03d86 DW |
1982 | } |
1983 | #else | |
1984 | # define xfs_inodegc_want_queue_rt_file(ip) (false) | |
1985 | #endif /* CONFIG_XFS_RT */ | |
1986 | ||
ab23a776 DC |
1987 | /* |
1988 | * Schedule the inactivation worker when: | |
1989 | * | |
1990 | * - We've accumulated more than one inode cluster buffer's worth of inodes. | |
7d6f07d2 | 1991 | * - There is less than 5% free space left. |
108523b8 | 1992 | * - Any of the quotas for this inode are near an enforcement limit. |
ab23a776 DC |
1993 | */ |
1994 | static inline bool | |
1995 | xfs_inodegc_want_queue_work( | |
1996 | struct xfs_inode *ip, | |
1997 | unsigned int items) | |
1998 | { | |
1999 | struct xfs_mount *mp = ip->i_mount; | |
2000 | ||
2001 | if (items > mp->m_ino_geo.inodes_per_cluster) | |
2002 | return true; | |
2003 | ||
7d6f07d2 DW |
2004 | if (__percpu_counter_compare(&mp->m_fdblocks, |
2005 | mp->m_low_space[XFS_LOWSP_5_PCNT], | |
2006 | XFS_FDBLOCKS_BATCH) < 0) | |
2007 | return true; | |
2008 | ||
65f03d86 DW |
2009 | if (xfs_inodegc_want_queue_rt_file(ip)) |
2010 | return true; | |
2011 | ||
108523b8 DW |
2012 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_USER)) |
2013 | return true; | |
2014 | ||
2015 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_GROUP)) | |
2016 | return true; | |
2017 | ||
2018 | if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_PROJ)) | |
2019 | return true; | |
2020 | ||
ab23a776 DC |
2021 | return false; |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * Upper bound on the number of inodes in each AG that can be queued for | |
2026 | * inactivation at any given time, to avoid monopolizing the workqueue. | |
2027 | */ | |
2028 | #define XFS_INODEGC_MAX_BACKLOG (4 * XFS_INODES_PER_CHUNK) | |
2029 | ||
2030 | /* | |
2031 | * Make the frontend wait for inactivations when: | |
2032 | * | |
40b1de00 | 2033 | * - Memory shrinkers queued the inactivation worker and it hasn't finished. |
ab23a776 DC |
2034 | * - The queue depth exceeds the maximum allowable percpu backlog. |
2035 | * | |
2036 | * Note: If the current thread is running a transaction, we don't ever want to | |
2037 | * wait for other transactions because that could introduce a deadlock. | |
2038 | */ | |
2039 | static inline bool | |
2040 | xfs_inodegc_want_flush_work( | |
2041 | struct xfs_inode *ip, | |
40b1de00 DW |
2042 | unsigned int items, |
2043 | unsigned int shrinker_hits) | |
ab23a776 DC |
2044 | { |
2045 | if (current->journal_info) | |
2046 | return false; | |
2047 | ||
40b1de00 DW |
2048 | if (shrinker_hits > 0) |
2049 | return true; | |
2050 | ||
ab23a776 DC |
2051 | if (items > XFS_INODEGC_MAX_BACKLOG) |
2052 | return true; | |
2053 | ||
2054 | return false; | |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * Queue a background inactivation worker if there are inodes that need to be | |
2059 | * inactivated and higher level xfs code hasn't disabled the background | |
2060 | * workers. | |
2061 | */ | |
2062 | static void | |
2063 | xfs_inodegc_queue( | |
2064 | struct xfs_inode *ip) | |
2065 | { | |
2066 | struct xfs_mount *mp = ip->i_mount; | |
2067 | struct xfs_inodegc *gc; | |
2068 | int items; | |
40b1de00 | 2069 | unsigned int shrinker_hits; |
7cf2b0f9 | 2070 | unsigned long queue_delay = 1; |
ab23a776 DC |
2071 | |
2072 | trace_xfs_inode_set_need_inactive(ip); | |
2073 | spin_lock(&ip->i_flags_lock); | |
2074 | ip->i_flags |= XFS_NEED_INACTIVE; | |
2075 | spin_unlock(&ip->i_flags_lock); | |
2076 | ||
2077 | gc = get_cpu_ptr(mp->m_inodegc); | |
2078 | llist_add(&ip->i_gclist, &gc->list); | |
2079 | items = READ_ONCE(gc->items); | |
2080 | WRITE_ONCE(gc->items, items + 1); | |
40b1de00 | 2081 | shrinker_hits = READ_ONCE(gc->shrinker_hits); |
ab23a776 | 2082 | |
7cf2b0f9 DC |
2083 | /* |
2084 | * We queue the work while holding the current CPU so that the work | |
2085 | * is scheduled to run on this CPU. | |
2086 | */ | |
2087 | if (!xfs_is_inodegc_enabled(mp)) { | |
2088 | put_cpu_ptr(gc); | |
ab23a776 | 2089 | return; |
ab23a776 DC |
2090 | } |
2091 | ||
7cf2b0f9 DC |
2092 | if (xfs_inodegc_want_queue_work(ip, items)) |
2093 | queue_delay = 0; | |
2094 | ||
2095 | trace_xfs_inodegc_queue(mp, __return_address); | |
03e0add8 DW |
2096 | mod_delayed_work_on(current_cpu(), mp->m_inodegc_wq, &gc->work, |
2097 | queue_delay); | |
7cf2b0f9 DC |
2098 | put_cpu_ptr(gc); |
2099 | ||
40b1de00 | 2100 | if (xfs_inodegc_want_flush_work(ip, items, shrinker_hits)) { |
ab23a776 | 2101 | trace_xfs_inodegc_throttle(mp, __return_address); |
7cf2b0f9 | 2102 | flush_delayed_work(&gc->work); |
ab23a776 DC |
2103 | } |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * Fold the dead CPU inodegc queue into the current CPUs queue. | |
2108 | */ | |
2109 | void | |
2110 | xfs_inodegc_cpu_dead( | |
2111 | struct xfs_mount *mp, | |
2112 | unsigned int dead_cpu) | |
2113 | { | |
2114 | struct xfs_inodegc *dead_gc, *gc; | |
2115 | struct llist_node *first, *last; | |
2116 | unsigned int count = 0; | |
2117 | ||
2118 | dead_gc = per_cpu_ptr(mp->m_inodegc, dead_cpu); | |
7cf2b0f9 | 2119 | cancel_delayed_work_sync(&dead_gc->work); |
ab23a776 DC |
2120 | |
2121 | if (llist_empty(&dead_gc->list)) | |
2122 | return; | |
2123 | ||
2124 | first = dead_gc->list.first; | |
2125 | last = first; | |
2126 | while (last->next) { | |
2127 | last = last->next; | |
2128 | count++; | |
2129 | } | |
2130 | dead_gc->list.first = NULL; | |
2131 | dead_gc->items = 0; | |
2132 | ||
2133 | /* Add pending work to current CPU */ | |
2134 | gc = get_cpu_ptr(mp->m_inodegc); | |
2135 | llist_add_batch(first, last, &gc->list); | |
2136 | count += READ_ONCE(gc->items); | |
2137 | WRITE_ONCE(gc->items, count); | |
ab23a776 DC |
2138 | |
2139 | if (xfs_is_inodegc_enabled(mp)) { | |
2140 | trace_xfs_inodegc_queue(mp, __return_address); | |
03e0add8 DW |
2141 | mod_delayed_work_on(current_cpu(), mp->m_inodegc_wq, &gc->work, |
2142 | 0); | |
ab23a776 | 2143 | } |
7cf2b0f9 | 2144 | put_cpu_ptr(gc); |
ab23a776 DC |
2145 | } |
2146 | ||
2147 | /* | |
2148 | * We set the inode flag atomically with the radix tree tag. Once we get tag | |
2149 | * lookups on the radix tree, this inode flag can go away. | |
2150 | * | |
2151 | * We always use background reclaim here because even if the inode is clean, it | |
2152 | * still may be under IO and hence we have wait for IO completion to occur | |
2153 | * before we can reclaim the inode. The background reclaim path handles this | |
2154 | * more efficiently than we can here, so simply let background reclaim tear down | |
2155 | * all inodes. | |
2156 | */ | |
2157 | void | |
2158 | xfs_inode_mark_reclaimable( | |
2159 | struct xfs_inode *ip) | |
2160 | { | |
2161 | struct xfs_mount *mp = ip->i_mount; | |
2162 | bool need_inactive; | |
2163 | ||
2164 | XFS_STATS_INC(mp, vn_reclaim); | |
2165 | ||
2166 | /* | |
2167 | * We should never get here with any of the reclaim flags already set. | |
2168 | */ | |
2169 | ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_ALL_IRECLAIM_FLAGS)); | |
2170 | ||
2171 | need_inactive = xfs_inode_needs_inactive(ip); | |
2172 | if (need_inactive) { | |
2173 | xfs_inodegc_queue(ip); | |
2174 | return; | |
2175 | } | |
2176 | ||
2177 | /* Going straight to reclaim, so drop the dquots. */ | |
2178 | xfs_qm_dqdetach(ip); | |
2179 | xfs_inodegc_set_reclaimable(ip); | |
2180 | } | |
40b1de00 DW |
2181 | |
2182 | /* | |
2183 | * Register a phony shrinker so that we can run background inodegc sooner when | |
2184 | * there's memory pressure. Inactivation does not itself free any memory but | |
2185 | * it does make inodes reclaimable, which eventually frees memory. | |
2186 | * | |
2187 | * The count function, seek value, and batch value are crafted to trigger the | |
2188 | * scan function during the second round of scanning. Hopefully this means | |
2189 | * that we reclaimed enough memory that initiating metadata transactions won't | |
2190 | * make things worse. | |
2191 | */ | |
2192 | #define XFS_INODEGC_SHRINKER_COUNT (1UL << DEF_PRIORITY) | |
2193 | #define XFS_INODEGC_SHRINKER_BATCH ((XFS_INODEGC_SHRINKER_COUNT / 2) + 1) | |
2194 | ||
2195 | static unsigned long | |
2196 | xfs_inodegc_shrinker_count( | |
2197 | struct shrinker *shrink, | |
2198 | struct shrink_control *sc) | |
2199 | { | |
2200 | struct xfs_mount *mp = container_of(shrink, struct xfs_mount, | |
2201 | m_inodegc_shrinker); | |
2202 | struct xfs_inodegc *gc; | |
2203 | int cpu; | |
2204 | ||
2205 | if (!xfs_is_inodegc_enabled(mp)) | |
2206 | return 0; | |
2207 | ||
2208 | for_each_online_cpu(cpu) { | |
2209 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
2210 | if (!llist_empty(&gc->list)) | |
2211 | return XFS_INODEGC_SHRINKER_COUNT; | |
2212 | } | |
2213 | ||
2214 | return 0; | |
2215 | } | |
2216 | ||
2217 | static unsigned long | |
2218 | xfs_inodegc_shrinker_scan( | |
2219 | struct shrinker *shrink, | |
2220 | struct shrink_control *sc) | |
2221 | { | |
2222 | struct xfs_mount *mp = container_of(shrink, struct xfs_mount, | |
2223 | m_inodegc_shrinker); | |
2224 | struct xfs_inodegc *gc; | |
2225 | int cpu; | |
2226 | bool no_items = true; | |
2227 | ||
2228 | if (!xfs_is_inodegc_enabled(mp)) | |
2229 | return SHRINK_STOP; | |
2230 | ||
2231 | trace_xfs_inodegc_shrinker_scan(mp, sc, __return_address); | |
2232 | ||
2233 | for_each_online_cpu(cpu) { | |
2234 | gc = per_cpu_ptr(mp->m_inodegc, cpu); | |
2235 | if (!llist_empty(&gc->list)) { | |
2236 | unsigned int h = READ_ONCE(gc->shrinker_hits); | |
2237 | ||
2238 | WRITE_ONCE(gc->shrinker_hits, h + 1); | |
7cf2b0f9 | 2239 | mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0); |
40b1de00 DW |
2240 | no_items = false; |
2241 | } | |
2242 | } | |
2243 | ||
2244 | /* | |
2245 | * If there are no inodes to inactivate, we don't want the shrinker | |
2246 | * to think there's deferred work to call us back about. | |
2247 | */ | |
2248 | if (no_items) | |
2249 | return LONG_MAX; | |
2250 | ||
2251 | return SHRINK_STOP; | |
2252 | } | |
2253 | ||
2254 | /* Register a shrinker so we can accelerate inodegc and throttle queuing. */ | |
2255 | int | |
2256 | xfs_inodegc_register_shrinker( | |
2257 | struct xfs_mount *mp) | |
2258 | { | |
2259 | struct shrinker *shrink = &mp->m_inodegc_shrinker; | |
2260 | ||
2261 | shrink->count_objects = xfs_inodegc_shrinker_count; | |
2262 | shrink->scan_objects = xfs_inodegc_shrinker_scan; | |
2263 | shrink->seeks = 0; | |
2264 | shrink->flags = SHRINKER_NONSLAB; | |
2265 | shrink->batch = XFS_INODEGC_SHRINKER_BATCH; | |
2266 | ||
e33c267a | 2267 | return register_shrinker(shrink, "xfs-inodegc:%s", mp->m_super->s_id); |
40b1de00 | 2268 | } |