Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 : * All Rights Reserved.
5 : */
6 : #include "xfs.h"
7 : #include "xfs_fs.h"
8 : #include "xfs_shared.h"
9 : #include "xfs_format.h"
10 : #include "xfs_log_format.h"
11 : #include "xfs_trans_resv.h"
12 : #include "xfs_mount.h"
13 : #include "xfs_inode.h"
14 : #include "xfs_trans.h"
15 : #include "xfs_trans_priv.h"
16 : #include "xfs_inode_item.h"
17 : #include "xfs_quota.h"
18 : #include "xfs_trace.h"
19 : #include "xfs_icache.h"
20 : #include "xfs_bmap_util.h"
21 : #include "xfs_dquot_item.h"
22 : #include "xfs_dquot.h"
23 : #include "xfs_reflink.h"
24 : #include "xfs_ialloc.h"
25 : #include "xfs_ag.h"
26 : #include "xfs_log_priv.h"
27 : #include "xfs_health.h"
28 : #include "xfs_da_format.h"
29 : #include "xfs_dir2.h"
30 : #include "xfs_imeta.h"
31 :
32 : #include <linux/iversion.h>
33 :
34 : /* Radix tree tags for incore inode tree. */
35 :
36 : /* inode is to be reclaimed */
37 : #define XFS_ICI_RECLAIM_TAG 0
38 : /* Inode has speculative preallocations (posteof or cow) to clean. */
39 : #define XFS_ICI_BLOCKGC_TAG 1
40 :
41 : /*
42 : * The goal for walking incore inodes. These can correspond with incore inode
43 : * radix tree tags when convenient. Avoid existing XFS_IWALK namespace.
44 : */
45 : enum xfs_icwalk_goal {
46 : /* Goals directly associated with tagged inodes. */
47 : XFS_ICWALK_BLOCKGC = XFS_ICI_BLOCKGC_TAG,
48 : XFS_ICWALK_RECLAIM = XFS_ICI_RECLAIM_TAG,
49 : };
50 :
51 : static int xfs_icwalk(struct xfs_mount *mp,
52 : enum xfs_icwalk_goal goal, struct xfs_icwalk *icw);
53 : static int xfs_icwalk_ag(struct xfs_perag *pag,
54 : enum xfs_icwalk_goal goal, struct xfs_icwalk *icw);
55 :
56 : /*
57 : * Private inode cache walk flags for struct xfs_icwalk. Must not
58 : * coincide with XFS_ICWALK_FLAGS_VALID.
59 : */
60 :
61 : /* Stop scanning after icw_scan_limit inodes. */
62 : #define XFS_ICWALK_FLAG_SCAN_LIMIT (1U << 28)
63 :
64 : #define XFS_ICWALK_FLAG_RECLAIM_SICK (1U << 27)
65 : #define XFS_ICWALK_FLAG_UNION (1U << 26) /* union filter algorithm */
66 :
67 : #define XFS_ICWALK_PRIVATE_FLAGS (XFS_ICWALK_FLAG_SCAN_LIMIT | \
68 : XFS_ICWALK_FLAG_RECLAIM_SICK | \
69 : XFS_ICWALK_FLAG_UNION)
70 :
71 : /*
72 : * Allocate and initialise an xfs_inode.
73 : */
74 : struct xfs_inode *
75 509835937 : xfs_inode_alloc(
76 : struct xfs_mount *mp,
77 : xfs_ino_t ino)
78 : {
79 509835937 : struct xfs_inode *ip;
80 :
81 : /*
82 : * XXX: If this didn't occur in transactions, we could drop GFP_NOFAIL
83 : * and return NULL here on ENOMEM.
84 : */
85 509835937 : ip = alloc_inode_sb(mp->m_super, xfs_inode_cache, GFP_KERNEL | __GFP_NOFAIL);
86 :
87 509864658 : if (inode_init_always(mp->m_super, VFS_I(ip))) {
88 0 : kmem_cache_free(xfs_inode_cache, ip);
89 0 : return NULL;
90 : }
91 :
92 : /* VFS doesn't initialise i_mode or i_state! */
93 509856215 : VFS_I(ip)->i_mode = 0;
94 509856215 : VFS_I(ip)->i_state = 0;
95 509856215 : mapping_set_large_folios(VFS_I(ip)->i_mapping);
96 :
97 509790763 : XFS_STATS_INC(mp, vn_active);
98 509790763 : ASSERT(atomic_read(&ip->i_pincount) == 0);
99 509790763 : ASSERT(ip->i_ino == 0);
100 :
101 : /* initialise the xfs inode */
102 509790763 : ip->i_ino = ino;
103 509790763 : ip->i_mount = mp;
104 509790763 : memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
105 509790763 : ip->i_cowfp = NULL;
106 509790763 : memset(&ip->i_af, 0, sizeof(ip->i_af));
107 509790763 : ip->i_af.if_format = XFS_DINODE_FMT_EXTENTS;
108 509790763 : memset(&ip->i_df, 0, sizeof(ip->i_df));
109 509790763 : ip->i_flags = 0;
110 509790763 : ip->i_delayed_blks = 0;
111 509790763 : ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
112 509790763 : ip->i_nblocks = 0;
113 509790763 : ip->i_forkoff = 0;
114 509790763 : ip->i_sick = 0;
115 509790763 : ip->i_checked = 0;
116 509790763 : INIT_WORK(&ip->i_ioend_work, xfs_end_io);
117 509790763 : INIT_LIST_HEAD(&ip->i_ioend_list);
118 509790763 : spin_lock_init(&ip->i_ioend_lock);
119 509788447 : ip->i_next_unlinked = NULLAGINO;
120 509788447 : ip->i_prev_unlinked = 0;
121 :
122 509788447 : return ip;
123 : }
124 :
125 : STATIC void
126 506958733 : xfs_inode_free_callback(
127 : struct rcu_head *head)
128 : {
129 506958733 : struct inode *inode = container_of(head, struct inode, i_rcu);
130 506958733 : struct xfs_inode *ip = XFS_I(inode);
131 :
132 506958733 : switch (VFS_I(ip)->i_mode & S_IFMT) {
133 290042439 : case S_IFREG:
134 : case S_IFDIR:
135 : case S_IFLNK:
136 290042439 : xfs_idestroy_fork(&ip->i_df);
137 290042439 : break;
138 : }
139 :
140 507222445 : xfs_ifork_zap_attr(ip);
141 :
142 508751246 : if (ip->i_cowfp) {
143 102860662 : xfs_idestroy_fork(ip->i_cowfp);
144 102844614 : kmem_cache_free(xfs_ifork_cache, ip->i_cowfp);
145 : }
146 508308771 : if (ip->i_itemp) {
147 54855272 : ASSERT(!test_bit(XFS_LI_IN_AIL,
148 : &ip->i_itemp->ili_item.li_flags));
149 54855272 : xfs_inode_item_destroy(ip);
150 55058729 : ip->i_itemp = NULL;
151 : }
152 :
153 508512228 : kmem_cache_free(xfs_inode_cache, ip);
154 508242907 : }
155 :
156 : static void
157 509885902 : __xfs_inode_free(
158 : struct xfs_inode *ip)
159 : {
160 : /* asserts to verify all state is correct here */
161 509885902 : ASSERT(atomic_read(&ip->i_pincount) == 0);
162 509885902 : ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list));
163 509885902 : XFS_STATS_DEC(ip->i_mount, vn_active);
164 :
165 509885902 : call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
166 509885574 : }
167 :
168 : void
169 793169 : xfs_inode_free(
170 : struct xfs_inode *ip)
171 : {
172 1586341 : ASSERT(!xfs_iflags_test(ip, XFS_IFLUSHING));
173 :
174 : /*
175 : * Because we use RCU freeing we need to ensure the inode always
176 : * appears to be reclaimed with an invalid inode number when in the
177 : * free state. The ip->i_flags_lock provides the barrier against lookup
178 : * races.
179 : */
180 793172 : spin_lock(&ip->i_flags_lock);
181 793174 : ip->i_flags = XFS_IRECLAIM;
182 793174 : ip->i_ino = 0;
183 793174 : spin_unlock(&ip->i_flags_lock);
184 :
185 793174 : __xfs_inode_free(ip);
186 793169 : }
187 :
188 : /*
189 : * Queue background inode reclaim work if there are reclaimable inodes and there
190 : * isn't reclaim work already scheduled or in progress.
191 : */
192 : static void
193 5626337 : xfs_reclaim_work_queue(
194 : struct xfs_mount *mp)
195 : {
196 :
197 5626337 : rcu_read_lock();
198 5626261 : if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
199 5611473 : queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
200 5611473 : msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
201 : }
202 5626020 : rcu_read_unlock();
203 5625780 : }
204 :
205 : /*
206 : * Background scanning to trim preallocated space. This is queued based on the
207 : * 'speculative_prealloc_lifetime' tunable (5m by default).
208 : */
209 : static inline void
210 954285 : xfs_blockgc_queue(
211 : struct xfs_perag *pag)
212 : {
213 954285 : struct xfs_mount *mp = pag->pag_mount;
214 :
215 1908570 : if (!xfs_is_blockgc_enabled(mp))
216 : return;
217 :
218 954285 : rcu_read_lock();
219 954281 : if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG))
220 938516 : queue_delayed_work(pag->pag_mount->m_blockgc_wq,
221 : &pag->pag_blockgc_work,
222 938516 : msecs_to_jiffies(xfs_blockgc_secs * 1000));
223 954227 : rcu_read_unlock();
224 : }
225 :
226 : /* Set a tag on both the AG incore inode tree and the AG radix tree. */
227 : static void
228 1078608068 : xfs_perag_set_inode_tag(
229 : struct xfs_perag *pag,
230 : xfs_agino_t agino,
231 : unsigned int tag)
232 : {
233 1078608068 : struct xfs_mount *mp = pag->pag_mount;
234 1078608068 : bool was_tagged;
235 :
236 1078608068 : lockdep_assert_held(&pag->pag_ici_lock);
237 :
238 1078608068 : was_tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
239 1078609130 : radix_tree_tag_set(&pag->pag_ici_root, agino, tag);
240 :
241 1078576732 : if (tag == XFS_ICI_RECLAIM_TAG)
242 1075661946 : pag->pag_ici_reclaimable++;
243 :
244 1078576732 : if (was_tagged)
245 : return;
246 :
247 : /* propagate the tag up into the perag radix tree */
248 5366186 : spin_lock(&mp->m_perag_lock);
249 5367426 : radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, tag);
250 5367426 : spin_unlock(&mp->m_perag_lock);
251 :
252 : /* start background work */
253 5366622 : switch (tag) {
254 5066950 : case XFS_ICI_RECLAIM_TAG:
255 5066950 : xfs_reclaim_work_queue(mp);
256 5066950 : break;
257 299672 : case XFS_ICI_BLOCKGC_TAG:
258 299672 : xfs_blockgc_queue(pag);
259 299672 : break;
260 : }
261 :
262 5366645 : trace_xfs_perag_set_inode_tag(pag, _RET_IP_);
263 : }
264 :
265 : /* Clear a tag on both the AG incore inode tree and the AG radix tree. */
266 : static void
267 1089346061 : xfs_perag_clear_inode_tag(
268 : struct xfs_perag *pag,
269 : xfs_agino_t agino,
270 : unsigned int tag)
271 : {
272 1089346061 : struct xfs_mount *mp = pag->pag_mount;
273 :
274 1089346061 : lockdep_assert_held(&pag->pag_ici_lock);
275 :
276 : /*
277 : * Reclaim can signal (with a null agino) that it cleared its own tag
278 : * by removing the inode from the radix tree.
279 : */
280 1089346061 : if (agino != NULLAGINO)
281 580253540 : radix_tree_tag_clear(&pag->pag_ici_root, agino, tag);
282 : else
283 509092521 : ASSERT(tag == XFS_ICI_RECLAIM_TAG);
284 :
285 1089348366 : if (tag == XFS_ICI_RECLAIM_TAG)
286 1075633386 : pag->pag_ici_reclaimable--;
287 :
288 1089348366 : if (radix_tree_tagged(&pag->pag_ici_root, tag))
289 : return;
290 :
291 : /* clear the tag from the perag radix tree */
292 9772930 : spin_lock(&mp->m_perag_lock);
293 9773393 : radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, tag);
294 9773393 : spin_unlock(&mp->m_perag_lock);
295 :
296 9773371 : trace_xfs_perag_clear_inode_tag(pag, _RET_IP_);
297 : }
298 :
299 : /*
300 : * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
301 : * part of the structure. This is made more complex by the fact we store
302 : * information about the on-disk values in the VFS inode and so we can't just
303 : * overwrite the values unconditionally. Hence we save the parameters we
304 : * need to retain across reinitialisation, and rewrite them into the VFS inode
305 : * after reinitialisation even if it fails.
306 : */
307 : static int
308 566001901 : xfs_reinit_inode(
309 : struct xfs_mount *mp,
310 : struct inode *inode)
311 : {
312 566001901 : int error;
313 566001901 : uint32_t nlink = inode->i_nlink;
314 566001901 : uint32_t generation = inode->i_generation;
315 566001901 : uint64_t version = inode_peek_iversion(inode);
316 566001901 : umode_t mode = inode->i_mode;
317 566001901 : dev_t dev = inode->i_rdev;
318 566001901 : kuid_t uid = inode->i_uid;
319 566001901 : kgid_t gid = inode->i_gid;
320 :
321 566001901 : error = inode_init_always(mp->m_super, inode);
322 :
323 566575224 : set_nlink(inode, nlink);
324 566593138 : inode->i_generation = generation;
325 566593138 : inode_set_iversion_queried(inode, version);
326 566593138 : inode->i_mode = mode;
327 566593138 : inode->i_rdev = dev;
328 566593138 : inode->i_uid = uid;
329 566593138 : inode->i_gid = gid;
330 566593138 : mapping_set_large_folios(inode->i_mapping);
331 566556003 : return error;
332 : }
333 :
334 : /*
335 : * Carefully nudge an inode whose VFS state has been torn down back into a
336 : * usable state. Drops the i_flags_lock and the rcu read lock.
337 : */
338 : static int
339 566590485 : xfs_iget_recycle(
340 : struct xfs_perag *pag,
341 : struct xfs_inode *ip) __releases(&ip->i_flags_lock)
342 : {
343 566590485 : struct xfs_mount *mp = ip->i_mount;
344 566590485 : struct inode *inode = VFS_I(ip);
345 566590485 : int error;
346 :
347 566590485 : trace_xfs_iget_recycle(ip);
348 :
349 566590541 : if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
350 : return -EAGAIN;
351 :
352 : /*
353 : * We need to make it look like the inode is being reclaimed to prevent
354 : * the actual reclaim workers from stomping over us while we recycle
355 : * the inode. We can't clear the radix tree tag yet as it requires
356 : * pag_ici_lock to be held exclusive.
357 : */
358 566602494 : ip->i_flags |= XFS_IRECLAIM;
359 :
360 566602494 : spin_unlock(&ip->i_flags_lock);
361 566600144 : rcu_read_unlock();
362 :
363 566492145 : ASSERT(!rwsem_is_locked(&inode->i_rwsem));
364 566492145 : error = xfs_reinit_inode(mp, inode);
365 566530236 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
366 566591598 : if (error) {
367 : /*
368 : * Re-initializing the inode failed, and we are in deep
369 : * trouble. Try to re-add it to the reclaim list.
370 : */
371 0 : rcu_read_lock();
372 0 : spin_lock(&ip->i_flags_lock);
373 0 : ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
374 0 : ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
375 0 : spin_unlock(&ip->i_flags_lock);
376 0 : rcu_read_unlock();
377 :
378 0 : trace_xfs_iget_recycle_fail(ip);
379 0 : return error;
380 : }
381 :
382 566591598 : spin_lock(&pag->pag_ici_lock);
383 566621407 : spin_lock(&ip->i_flags_lock);
384 :
385 : /*
386 : * Clear the per-lifetime state in the inode as we are now effectively
387 : * a new inode and need to return to the initial state before reuse
388 : * occurs.
389 : */
390 566622036 : ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
391 566622036 : ip->i_flags |= XFS_INEW;
392 566622036 : xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
393 : XFS_ICI_RECLAIM_TAG);
394 566617642 : inode->i_state = I_NEW;
395 566617642 : spin_unlock(&ip->i_flags_lock);
396 566620484 : spin_unlock(&pag->pag_ici_lock);
397 :
398 566620484 : return 0;
399 : }
400 :
401 : /*
402 : * If we are allocating a new inode, then check what was returned is
403 : * actually a free, empty inode. If we are not allocating an inode,
404 : * then check we didn't find a free inode.
405 : *
406 : * Returns:
407 : * 0 if the inode free state matches the lookup context
408 : * -ENOENT if the inode is free and we are not allocating
409 : * -EFSCORRUPTED if there is any state mismatch at all
410 : */
411 : static int
412 75937587329 : xfs_iget_check_free_state(
413 : struct xfs_inode *ip,
414 : int flags)
415 : {
416 75937587329 : if (flags & XFS_IGET_CREATE) {
417 : /* should be a free inode */
418 64359267 : if (VFS_I(ip)->i_mode != 0) {
419 0 : xfs_warn(ip->i_mount,
420 : "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
421 : ip->i_ino, VFS_I(ip)->i_mode);
422 0 : xfs_agno_mark_sick(ip->i_mount,
423 0 : XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
424 : XFS_SICK_AG_INOBT);
425 0 : return -EFSCORRUPTED;
426 : }
427 :
428 64359267 : if (ip->i_nblocks != 0) {
429 0 : xfs_warn(ip->i_mount,
430 : "Corruption detected! Free inode 0x%llx has blocks allocated!",
431 : ip->i_ino);
432 0 : xfs_agno_mark_sick(ip->i_mount,
433 0 : XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
434 : XFS_SICK_AG_INOBT);
435 0 : return -EFSCORRUPTED;
436 : }
437 : return 0;
438 : }
439 :
440 : /* should be an allocated inode */
441 75873228062 : if (VFS_I(ip)->i_mode == 0)
442 2166487 : return -ENOENT;
443 :
444 : return 0;
445 : }
446 :
447 : /* Make all pending inactivation work start immediately. */
448 : static bool
449 14414572 : xfs_inodegc_queue_all(
450 : struct xfs_mount *mp)
451 : {
452 14414572 : struct xfs_inodegc *gc;
453 14414572 : int cpu;
454 14414572 : bool ret = false;
455 :
456 43222279 : for_each_online_cpu(cpu) {
457 28807286 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
458 28807286 : if (!llist_empty(&gc->list)) {
459 877940 : mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0);
460 877940 : ret = true;
461 : }
462 : }
463 :
464 14403065 : return ret;
465 : }
466 :
467 : /* Wait for all queued work and collect errors */
468 : static int
469 5055050 : xfs_inodegc_wait_all(
470 : struct xfs_mount *mp)
471 : {
472 5055050 : int cpu;
473 5055050 : int error = 0;
474 :
475 5055050 : flush_workqueue(mp->m_inodegc_wq);
476 20221793 : for_each_online_cpu(cpu) {
477 10110717 : struct xfs_inodegc *gc;
478 :
479 10110717 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
480 10110717 : if (gc->error && !error)
481 1707 : error = gc->error;
482 10110717 : gc->error = 0;
483 : }
484 :
485 5056039 : return error;
486 : }
487 :
488 : /*
489 : * Check the validity of the inode we just found it the cache
490 : */
491 : static int
492 75249629947 : xfs_iget_cache_hit(
493 : struct xfs_perag *pag,
494 : struct xfs_inode *ip,
495 : xfs_ino_t ino,
496 : int flags,
497 : int lock_flags) __releases(RCU)
498 : {
499 75249629947 : struct inode *inode = VFS_I(ip);
500 75249629947 : struct xfs_mount *mp = ip->i_mount;
501 75249629947 : int error;
502 :
503 : /*
504 : * check for re-use of an inode within an RCU grace period due to the
505 : * radix tree nodes not being updated yet. We monitor for this by
506 : * setting the inode number to zero before freeing the inode structure.
507 : * If the inode has been reallocated and set up, then the inode number
508 : * will not match, so check for that, too.
509 : */
510 75249629947 : spin_lock(&ip->i_flags_lock);
511 75504923126 : if (ip->i_ino != ino)
512 483 : goto out_skip;
513 :
514 : /*
515 : * If we are racing with another cache hit that is currently
516 : * instantiating this inode or currently recycling it out of
517 : * reclaimable state, wait for the initialisation to complete
518 : * before continuing.
519 : *
520 : * If we're racing with the inactivation worker we also want to wait.
521 : * If we're creating a new file, it's possible that the worker
522 : * previously marked the inode as free on disk but hasn't finished
523 : * updating the incore state yet. The AGI buffer will be dirty and
524 : * locked to the icreate transaction, so a synchronous push of the
525 : * inodegc workers would result in deadlock. For a regular iget, the
526 : * worker is running already, so we might as well wait.
527 : *
528 : * XXX(hch): eventually we should do something equivalent to
529 : * wait_on_inode to wait for these flags to be cleared
530 : * instead of polling for it.
531 : */
532 75504922643 : if (ip->i_flags & (XFS_INEW | XFS_IRECLAIM | XFS_INACTIVATING))
533 246532 : goto out_skip;
534 :
535 75504676111 : if (ip->i_flags & XFS_NEED_INACTIVE) {
536 : /* Unlinked inodes cannot be re-grabbed. */
537 1223008 : if (VFS_I(ip)->i_nlink == 0) {
538 1181081 : error = -ENOENT;
539 1181081 : goto out_error;
540 : }
541 41927 : goto out_inodegc_flush;
542 : }
543 :
544 : /*
545 : * Check the inode free state is valid. This also detects lookup
546 : * racing with unlinks.
547 : */
548 75503453103 : error = xfs_iget_check_free_state(ip, flags);
549 75443297507 : if (error)
550 2166481 : goto out_error;
551 :
552 : /* Skip inodes that have no vfs state. */
553 75441131026 : if ((flags & XFS_IGET_INCORE) &&
554 0 : (ip->i_flags & XFS_IRECLAIMABLE))
555 0 : goto out_skip;
556 :
557 : /* The inode fits the selection criteria; process it. */
558 75441131026 : if (ip->i_flags & XFS_IRECLAIMABLE) {
559 : /* Drops i_flags_lock and RCU read lock. */
560 566591273 : error = xfs_iget_recycle(pag, ip);
561 566596195 : if (error == -EAGAIN)
562 3 : goto out_skip;
563 566596192 : if (error)
564 : return error;
565 : } else {
566 : /* If the VFS inode is being torn down, pause and try again. */
567 74874539753 : if (!igrab(inode))
568 167523 : goto out_skip;
569 :
570 : /* We've got a live one. */
571 74935722757 : spin_unlock(&ip->i_flags_lock);
572 75003332008 : rcu_read_unlock();
573 74895947582 : trace_xfs_iget_hit(ip);
574 : }
575 :
576 75348648738 : if (lock_flags != 0)
577 68626264449 : xfs_ilock(ip, lock_flags);
578 :
579 75311719751 : if (!(flags & XFS_IGET_INCORE))
580 75434709205 : xfs_iflags_clear(ip, XFS_ISTALE);
581 75458257560 : XFS_STATS_INC(mp, xs_ig_found);
582 :
583 75458257560 : return 0;
584 :
585 414541 : out_skip:
586 414541 : trace_xfs_iget_skip(ip);
587 414541 : XFS_STATS_INC(mp, xs_ig_frecycle);
588 414541 : error = -EAGAIN;
589 3762103 : out_error:
590 3762103 : spin_unlock(&ip->i_flags_lock);
591 3762097 : rcu_read_unlock();
592 3762097 : return error;
593 :
594 : out_inodegc_flush:
595 41927 : spin_unlock(&ip->i_flags_lock);
596 41927 : rcu_read_unlock();
597 : /*
598 : * Do not wait for the workers, because the caller could hold an AGI
599 : * buffer lock. We're just going to sleep in a loop anyway.
600 : */
601 83854 : if (xfs_is_inodegc_enabled(mp))
602 41908 : xfs_inodegc_queue_all(mp);
603 : return -EAGAIN;
604 : }
605 :
606 : static int
607 509843350 : xfs_iget_cache_miss(
608 : struct xfs_mount *mp,
609 : struct xfs_perag *pag,
610 : xfs_trans_t *tp,
611 : xfs_ino_t ino,
612 : struct xfs_inode **ipp,
613 : int flags,
614 : int lock_flags)
615 : {
616 509843350 : struct xfs_inode *ip;
617 509843350 : int error;
618 509843350 : xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
619 509843350 : int iflags;
620 :
621 509843350 : ip = xfs_inode_alloc(mp, ino);
622 509844046 : if (!ip)
623 : return -ENOMEM;
624 :
625 509844046 : error = xfs_imap(pag, tp, ip->i_ino, &ip->i_imap, flags);
626 509858883 : if (error)
627 67993 : goto out_destroy;
628 :
629 : /*
630 : * For version 5 superblocks, if we are initialising a new inode and we
631 : * are not utilising the XFS_FEAT_IKEEP inode cluster mode, we can
632 : * simply build the new inode core with a random generation number.
633 : *
634 : * For version 4 (and older) superblocks, log recovery is dependent on
635 : * the i_flushiter field being initialised from the current on-disk
636 : * value and hence we must also read the inode off disk even when
637 : * initializing new inodes.
638 : */
639 509790890 : if (xfs_has_v3inodes(mp) &&
640 509790472 : (flags & XFS_IGET_CREATE) && !xfs_has_ikeep(mp)) {
641 30450933 : VFS_I(ip)->i_generation = get_random_u32();
642 : } else {
643 479339957 : struct xfs_buf *bp;
644 :
645 479339957 : error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp);
646 479354723 : if (error)
647 4413 : goto out_destroy;
648 :
649 479355787 : error = xfs_inode_from_disk(ip,
650 479350512 : xfs_buf_offset(bp, ip->i_imap.im_boffset));
651 479345275 : if (!error)
652 479345073 : xfs_buf_set_ref(bp, XFS_INO_REF);
653 : else
654 202 : xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
655 479349280 : xfs_trans_brelse(tp, bp);
656 :
657 479354234 : if (error)
658 202 : goto out_destroy;
659 : }
660 :
661 509803928 : trace_xfs_iget_miss(ip);
662 :
663 : /*
664 : * Check the inode free state is valid. This also detects lookup
665 : * racing with unlinks.
666 : */
667 509787069 : error = xfs_iget_check_free_state(ip, flags);
668 509801609 : if (error)
669 0 : goto out_destroy;
670 :
671 : /*
672 : * Preload the radix tree so we can insert safely under the
673 : * write spinlock. Note that we cannot sleep inside the preload
674 : * region. Since we can be called from transaction context, don't
675 : * recurse into the file system.
676 : */
677 509801609 : if (radix_tree_preload(GFP_NOFS)) {
678 0 : error = -EAGAIN;
679 0 : goto out_destroy;
680 : }
681 :
682 : /*
683 : * Because the inode hasn't been added to the radix-tree yet it can't
684 : * be found by another thread, so we can do the non-sleeping lock here.
685 : */
686 509805825 : if (lock_flags) {
687 496642449 : if (!xfs_ilock_nowait(ip, lock_flags))
688 0 : BUG();
689 : }
690 :
691 : /*
692 : * These values must be set before inserting the inode into the radix
693 : * tree as the moment it is inserted a concurrent lookup (allowed by the
694 : * RCU locking mechanism) can find it and that lookup must see that this
695 : * is an inode currently under construction (i.e. that XFS_INEW is set).
696 : * The ip->i_flags_lock that protects the XFS_INEW flag forms the
697 : * memory barrier that ensures this detection works correctly at lookup
698 : * time.
699 : */
700 509807843 : iflags = XFS_INEW;
701 509807843 : if (flags & XFS_IGET_DONTCACHE)
702 466227194 : d_mark_dontcache(VFS_I(ip));
703 509808448 : ip->i_udquot = NULL;
704 509808448 : ip->i_gdquot = NULL;
705 509808448 : ip->i_pdquot = NULL;
706 509808448 : xfs_iflags_set(ip, iflags);
707 :
708 : /* insert the new inode */
709 509809483 : spin_lock(&pag->pag_ici_lock);
710 509810350 : error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
711 509808763 : if (unlikely(error)) {
712 720768 : WARN_ON(error != -EEXIST);
713 720768 : XFS_STATS_INC(mp, xs_ig_dup);
714 720768 : error = -EAGAIN;
715 720768 : goto out_preload_end;
716 : }
717 509087995 : spin_unlock(&pag->pag_ici_lock);
718 509082296 : radix_tree_preload_end();
719 :
720 509086839 : *ipp = ip;
721 509086839 : return 0;
722 :
723 : out_preload_end:
724 720768 : spin_unlock(&pag->pag_ici_lock);
725 720767 : radix_tree_preload_end();
726 720764 : if (lock_flags)
727 720108 : xfs_iunlock(ip, lock_flags);
728 656 : out_destroy:
729 793174 : __destroy_inode(VFS_I(ip));
730 793172 : xfs_inode_free(ip);
731 793172 : return error;
732 : }
733 :
734 : /*
735 : * Look up an inode by number in the given file system. The inode is looked up
736 : * in the cache held in each AG. If the inode is found in the cache, initialise
737 : * the vfs inode if necessary.
738 : *
739 : * If it is not in core, read it in from the file system's device, add it to the
740 : * cache and initialise the vfs inode.
741 : *
742 : * The inode is locked according to the value of the lock_flags parameter.
743 : * Inode lookup is only done during metadata operations and not as part of the
744 : * data IO path. Hence we only allow locking of the XFS_ILOCK during lookup.
745 : */
746 : int
747 75508942380 : xfs_iget(
748 : struct xfs_mount *mp,
749 : struct xfs_trans *tp,
750 : xfs_ino_t ino,
751 : uint flags,
752 : uint lock_flags,
753 : struct xfs_inode **ipp)
754 : {
755 75508942380 : struct xfs_inode *ip;
756 75508942380 : struct xfs_perag *pag;
757 75508942380 : xfs_agino_t agino;
758 75508942380 : int error;
759 :
760 75508942380 : ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
761 :
762 : /* reject inode numbers outside existing AGs */
763 75508942380 : if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
764 : return -EINVAL;
765 :
766 75508109737 : XFS_STATS_INC(mp, xs_ig_attempts);
767 :
768 : /* get the perag structure and ensure that it's inode capable */
769 75508109737 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
770 75736324075 : agino = XFS_INO_TO_AGINO(mp, ino);
771 :
772 75737480838 : again:
773 75737480838 : error = 0;
774 75737480838 : rcu_read_lock();
775 75997237068 : ip = radix_tree_lookup(&pag->pag_ici_root, agino);
776 :
777 75687303508 : if (ip) {
778 75177463330 : error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
779 75311430637 : if (error)
780 3804011 : goto out_error_or_again;
781 : } else {
782 509840178 : rcu_read_unlock();
783 509852798 : if (flags & XFS_IGET_INCORE) {
784 0 : error = -ENODATA;
785 0 : goto out_error_or_again;
786 : }
787 509852798 : XFS_STATS_INC(mp, xs_ig_missed);
788 :
789 509852798 : error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
790 : flags, lock_flags);
791 509875297 : if (error)
792 793162 : goto out_error_or_again;
793 : }
794 75816708761 : xfs_perag_put(pag);
795 :
796 76002559562 : *ipp = ip;
797 :
798 : /*
799 : * If we have a real type for an on-disk inode, we can setup the inode
800 : * now. If it's a new inode being created, xfs_init_new_inode will
801 : * handle it.
802 : */
803 >15191*10^7 : if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
804 1009657665 : xfs_setup_existing_inode(ip);
805 : return 0;
806 :
807 4597173 : out_error_or_again:
808 4597173 : if (!(flags & (XFS_IGET_INCORE | XFS_IGET_NORETRY)) &&
809 : error == -EAGAIN) {
810 1157066 : delay(1);
811 1156763 : goto again;
812 : }
813 3440107 : xfs_perag_put(pag);
814 3440107 : return error;
815 : }
816 :
817 : /* Get a metadata inode. The ftype must match exactly. */
818 : int
819 385920 : xfs_imeta_iget(
820 : struct xfs_mount *mp,
821 : xfs_ino_t ino,
822 : unsigned char ftype,
823 : struct xfs_inode **ipp)
824 : {
825 385920 : struct xfs_inode *ip;
826 385920 : int error;
827 :
828 385920 : ASSERT(ftype != XFS_DIR3_FT_UNKNOWN);
829 :
830 385920 : error = xfs_iget(mp, NULL, ino, XFS_IGET_UNTRUSTED, 0, &ip);
831 385920 : if (error == -EFSCORRUPTED)
832 4 : goto whine;
833 385916 : if (error)
834 : return error;
835 :
836 385916 : if (VFS_I(ip)->i_nlink == 0)
837 0 : goto bad_rele;
838 385916 : if (xfs_mode_to_ftype(VFS_I(ip)->i_mode) != ftype)
839 0 : goto bad_rele;
840 385916 : if (xfs_has_metadir(mp) && !xfs_is_metadir_inode(ip))
841 0 : goto bad_rele;
842 :
843 385916 : *ipp = ip;
844 385916 : return 0;
845 0 : bad_rele:
846 0 : xfs_irele(ip);
847 4 : whine:
848 4 : xfs_err(mp, "metadata inode 0x%llx is corrupt", ino);
849 4 : xfs_fs_mark_sick(mp, XFS_SICK_FS_METADIR);
850 4 : return -EFSCORRUPTED;
851 : }
852 :
853 : /*
854 : * Grab the inode for reclaim exclusively.
855 : *
856 : * We have found this inode via a lookup under RCU, so the inode may have
857 : * already been freed, or it may be in the process of being recycled by
858 : * xfs_iget(). In both cases, the inode will have XFS_IRECLAIM set. If the inode
859 : * has been fully recycled by the time we get the i_flags_lock, XFS_IRECLAIMABLE
860 : * will not be set. Hence we need to check for both these flag conditions to
861 : * avoid inodes that are no longer reclaim candidates.
862 : *
863 : * Note: checking for other state flags here, under the i_flags_lock or not, is
864 : * racy and should be avoided. Those races should be resolved only after we have
865 : * ensured that we are able to reclaim this inode and the world can see that we
866 : * are going to reclaim it.
867 : *
868 : * Return true if we grabbed it, false otherwise.
869 : */
870 : static bool
871 529344291 : xfs_reclaim_igrab(
872 : struct xfs_inode *ip,
873 : struct xfs_icwalk *icw)
874 : {
875 529344291 : ASSERT(rcu_read_lock_held());
876 :
877 529344291 : spin_lock(&ip->i_flags_lock);
878 529353772 : if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
879 : __xfs_iflags_test(ip, XFS_IRECLAIM)) {
880 : /* not a reclaim candidate. */
881 80174 : spin_unlock(&ip->i_flags_lock);
882 80174 : return false;
883 : }
884 :
885 : /* Don't reclaim a sick inode unless the caller asked for it. */
886 529273598 : if (ip->i_sick &&
887 32 : (!icw || !(icw->icw_flags & XFS_ICWALK_FLAG_RECLAIM_SICK))) {
888 0 : spin_unlock(&ip->i_flags_lock);
889 0 : return false;
890 : }
891 :
892 529273598 : __xfs_iflags_set(ip, XFS_IRECLAIM);
893 529273598 : spin_unlock(&ip->i_flags_lock);
894 529273598 : return true;
895 : }
896 :
897 : /*
898 : * Inode reclaim is non-blocking, so the default action if progress cannot be
899 : * made is to "requeue" the inode for reclaim by unlocking it and clearing the
900 : * XFS_IRECLAIM flag. If we are in a shutdown state, we don't care about
901 : * blocking anymore and hence we can wait for the inode to be able to reclaim
902 : * it.
903 : *
904 : * We do no IO here - if callers require inodes to be cleaned they must push the
905 : * AIL first to trigger writeback of dirty inodes. This enables writeback to be
906 : * done in the background in a non-blocking manner, and enables memory reclaim
907 : * to make progress without blocking.
908 : */
909 : static void
910 529278523 : xfs_reclaim_inode(
911 : struct xfs_inode *ip,
912 : struct xfs_perag *pag)
913 : {
914 529278523 : xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
915 :
916 529278523 : if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
917 5 : goto out;
918 529279984 : if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING))
919 12446997 : goto out_iunlock;
920 :
921 : /*
922 : * Check for log shutdown because aborting the inode can move the log
923 : * tail and corrupt in memory state. This is fine if the log is shut
924 : * down, but if the log is still active and only the mount is shut down
925 : * then the in-memory log tail movement caused by the abort can be
926 : * incorrectly propagated to disk.
927 : */
928 1033659418 : if (xlog_is_shutdown(ip->i_mount->m_log)) {
929 305761830 : xfs_iunpin_wait(ip);
930 305761830 : xfs_iflush_shutdown_abort(ip);
931 305761830 : goto reclaim;
932 : }
933 211067879 : if (xfs_ipincount(ip))
934 1531340 : goto out_clear_flush;
935 209536539 : if (!xfs_inode_clean(ip))
936 6212954 : goto out_clear_flush;
937 :
938 203323585 : xfs_iflags_clear(ip, XFS_IFLUSHING);
939 509091974 : reclaim:
940 509091974 : trace_xfs_inode_reclaiming(ip);
941 :
942 : /*
943 : * Because we use RCU freeing we need to ensure the inode always appears
944 : * to be reclaimed with an invalid inode number when in the free state.
945 : * We do this as early as possible under the ILOCK so that
946 : * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
947 : * detect races with us here. By doing this, we guarantee that once
948 : * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
949 : * it will see either a valid inode that will serialise correctly, or it
950 : * will see an invalid inode that it can skip.
951 : */
952 509091941 : spin_lock(&ip->i_flags_lock);
953 509092515 : ip->i_flags = XFS_IRECLAIM;
954 509092515 : ip->i_ino = 0;
955 509092515 : ip->i_sick = 0;
956 509092515 : ip->i_checked = 0;
957 509092515 : spin_unlock(&ip->i_flags_lock);
958 :
959 509092394 : ASSERT(!ip->i_itemp || ip->i_itemp->ili_item.li_buf == NULL);
960 509092394 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
961 :
962 509079841 : XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
963 : /*
964 : * Remove the inode from the per-AG radix tree.
965 : *
966 : * Because radix_tree_delete won't complain even if the item was never
967 : * added to the tree assert that it's been there before to catch
968 : * problems with the inode life time early on.
969 : */
970 509079841 : spin_lock(&pag->pag_ici_lock);
971 1018185750 : if (!xfs_is_shutdown(pag->pag_mount)) {
972 : /* had better not be on any unlinked list! */
973 203330964 : ASSERT(!xfs_inode_on_unlinked_list(ip));
974 203330964 : if (xfs_inode_on_unlinked_list(ip))
975 0 : xfs_emerg(pag->pag_mount, "IUNLINK ino 0x%llx nlink %u mode 0o%o prevun 0x%x nextun 0x%x", ino, VFS_I(ip)->i_nlink, VFS_I(ip)->i_mode, ip->i_prev_unlinked, ip->i_next_unlinked);
976 : }
977 509092662 : if (!radix_tree_delete(&pag->pag_ici_root,
978 509092875 : XFS_INO_TO_AGINO(ip->i_mount, ino)))
979 0 : ASSERT(0);
980 509092662 : xfs_perag_clear_inode_tag(pag, NULLAGINO, XFS_ICI_RECLAIM_TAG);
981 509092755 : spin_unlock(&pag->pag_ici_lock);
982 :
983 : /*
984 : * Here we do an (almost) spurious inode lock in order to coordinate
985 : * with inode cache radix tree lookups. This is because the lookup
986 : * can reference the inodes in the cache without taking references.
987 : *
988 : * We make that OK here by ensuring that we wait until the inode is
989 : * unlocked after the lookup before we go ahead and free it.
990 : */
991 509092898 : xfs_ilock(ip, XFS_ILOCK_EXCL);
992 509092838 : ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot);
993 509092838 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
994 564257913 : ASSERT(xfs_inode_clean(ip));
995 :
996 509092901 : __xfs_inode_free(ip);
997 509092901 : return;
998 :
999 7744294 : out_clear_flush:
1000 7744294 : xfs_iflags_clear(ip, XFS_IFLUSHING);
1001 20191305 : out_iunlock:
1002 20191305 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
1003 20191193 : out:
1004 20191193 : xfs_iflags_clear(ip, XFS_IRECLAIM);
1005 : }
1006 :
1007 : /* Reclaim sick inodes if we're unmounting or the fs went down. */
1008 : static inline bool
1009 545189 : xfs_want_reclaim_sick(
1010 : struct xfs_mount *mp)
1011 : {
1012 1611236 : return xfs_is_unmounting(mp) || xfs_has_norecovery(mp) ||
1013 : xfs_is_shutdown(mp);
1014 : }
1015 :
1016 : void
1017 24331 : xfs_reclaim_inodes(
1018 : struct xfs_mount *mp)
1019 : {
1020 24331 : struct xfs_icwalk icw = {
1021 : .icw_flags = 0,
1022 : };
1023 :
1024 24331 : if (xfs_want_reclaim_sick(mp))
1025 24331 : icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1026 :
1027 64692 : while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
1028 40361 : xfs_ail_push_all_sync(mp->m_ail);
1029 40361 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1030 : }
1031 24331 : }
1032 :
1033 : /*
1034 : * The shrinker infrastructure determines how many inodes we should scan for
1035 : * reclaim. We want as many clean inodes ready to reclaim as possible, so we
1036 : * push the AIL here. We also want to proactively free up memory if we can to
1037 : * minimise the amount of work memory reclaim has to do so we kick the
1038 : * background reclaim if it isn't already scheduled.
1039 : */
1040 : long
1041 520858 : xfs_reclaim_inodes_nr(
1042 : struct xfs_mount *mp,
1043 : unsigned long nr_to_scan)
1044 : {
1045 520858 : struct xfs_icwalk icw = {
1046 : .icw_flags = XFS_ICWALK_FLAG_SCAN_LIMIT,
1047 520858 : .icw_scan_limit = min_t(unsigned long, LONG_MAX, nr_to_scan),
1048 : };
1049 :
1050 520858 : if (xfs_want_reclaim_sick(mp))
1051 2 : icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1052 :
1053 : /* kick background reclaimer and push the AIL */
1054 520858 : xfs_reclaim_work_queue(mp);
1055 520858 : xfs_ail_push_all(mp->m_ail);
1056 :
1057 520858 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1058 520858 : return 0;
1059 : }
1060 :
1061 : /*
1062 : * Return the number of reclaimable inodes in the filesystem for
1063 : * the shrinker to determine how much to reclaim.
1064 : */
1065 : long
1066 1035540 : xfs_reclaim_inodes_count(
1067 : struct xfs_mount *mp)
1068 : {
1069 1035540 : struct xfs_perag *pag;
1070 1035540 : xfs_agnumber_t ag = 0;
1071 1035540 : long reclaimable = 0;
1072 :
1073 3474045 : while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1074 2438504 : ag = pag->pag_agno + 1;
1075 2438504 : reclaimable += pag->pag_ici_reclaimable;
1076 2438504 : xfs_perag_put(pag);
1077 : }
1078 1035541 : return reclaimable;
1079 : }
1080 :
1081 : STATIC bool
1082 110851 : xfs_icwalk_match_id(
1083 : struct xfs_inode *ip,
1084 : struct xfs_icwalk *icw)
1085 : {
1086 110851 : if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1087 : !uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1088 : return false;
1089 :
1090 110851 : if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) &&
1091 : !gid_eq(VFS_I(ip)->i_gid, icw->icw_gid))
1092 : return false;
1093 :
1094 110851 : if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) &&
1095 0 : ip->i_projid != icw->icw_prid)
1096 0 : return false;
1097 :
1098 : return true;
1099 : }
1100 :
1101 : /*
1102 : * A union-based inode filtering algorithm. Process the inode if any of the
1103 : * criteria match. This is for global/internal scans only.
1104 : */
1105 : STATIC bool
1106 1508 : xfs_icwalk_match_id_union(
1107 : struct xfs_inode *ip,
1108 : struct xfs_icwalk *icw)
1109 : {
1110 1508 : if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1111 : uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1112 : return true;
1113 :
1114 513 : if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) &&
1115 : gid_eq(VFS_I(ip)->i_gid, icw->icw_gid))
1116 : return true;
1117 :
1118 0 : if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) &&
1119 0 : ip->i_projid == icw->icw_prid)
1120 0 : return true;
1121 :
1122 : return false;
1123 : }
1124 :
1125 : /*
1126 : * Is this inode @ip eligible for eof/cow block reclamation, given some
1127 : * filtering parameters @icw? The inode is eligible if @icw is null or
1128 : * if the predicate functions match.
1129 : */
1130 : static bool
1131 814159 : xfs_icwalk_match(
1132 : struct xfs_inode *ip,
1133 : struct xfs_icwalk *icw)
1134 : {
1135 814159 : bool match;
1136 :
1137 814159 : if (!icw)
1138 : return true;
1139 :
1140 112377 : if (icw->icw_flags & XFS_ICWALK_FLAG_UNION)
1141 1508 : match = xfs_icwalk_match_id_union(ip, icw);
1142 : else
1143 110869 : match = xfs_icwalk_match_id(ip, icw);
1144 112377 : if (!match)
1145 : return false;
1146 :
1147 : /* skip the inode if the file size is too small */
1148 112388 : if ((icw->icw_flags & XFS_ICWALK_FLAG_MINFILESIZE) &&
1149 0 : XFS_ISIZE(ip) < icw->icw_min_file_size)
1150 0 : return false;
1151 :
1152 : return true;
1153 : }
1154 :
1155 : /*
1156 : * This is a fast pass over the inode cache to try to get reclaim moving on as
1157 : * many inodes as possible in a short period of time. It kicks itself every few
1158 : * seconds, as well as being kicked by the inode cache shrinker when memory
1159 : * goes low.
1160 : */
1161 : void
1162 37888 : xfs_reclaim_worker(
1163 : struct work_struct *work)
1164 : {
1165 37888 : struct xfs_mount *mp = container_of(to_delayed_work(work),
1166 : struct xfs_mount, m_reclaim_work);
1167 :
1168 37888 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, NULL);
1169 37888 : xfs_reclaim_work_queue(mp);
1170 37888 : }
1171 :
1172 : STATIC int
1173 1381457 : xfs_inode_free_eofblocks(
1174 : struct xfs_inode *ip,
1175 : struct xfs_icwalk *icw,
1176 : unsigned int *lockflags)
1177 : {
1178 1381457 : bool wait;
1179 :
1180 1381457 : wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1181 :
1182 2762761 : if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS))
1183 : return 0;
1184 :
1185 : /*
1186 : * If the mapping is dirty the operation can block and wait for some
1187 : * time. Unless we are waiting, skip it.
1188 : */
1189 1205517 : if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1190 : return 0;
1191 :
1192 663077 : if (!xfs_icwalk_match(ip, icw))
1193 : return 0;
1194 :
1195 : /*
1196 : * If the caller is waiting, return -EAGAIN to keep the background
1197 : * scanner moving and revisit the inode in a subsequent pass.
1198 : */
1199 663077 : if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1200 106413 : if (wait)
1201 : return -EAGAIN;
1202 33910 : return 0;
1203 : }
1204 556840 : *lockflags |= XFS_IOLOCK_EXCL;
1205 :
1206 556840 : if (xfs_can_free_eofblocks(ip, false))
1207 186403 : return xfs_free_eofblocks(ip);
1208 :
1209 : /* inode could be preallocated or append-only */
1210 370437 : trace_xfs_inode_free_eofblocks_invalid(ip);
1211 370432 : xfs_inode_clear_eofblocks_tag(ip);
1212 370432 : return 0;
1213 : }
1214 :
1215 : static void
1216 4197407 : xfs_blockgc_set_iflag(
1217 : struct xfs_inode *ip,
1218 : unsigned long iflag)
1219 : {
1220 4197407 : struct xfs_mount *mp = ip->i_mount;
1221 4197407 : struct xfs_perag *pag;
1222 :
1223 4197407 : ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
1224 :
1225 : /*
1226 : * Don't bother locking the AG and looking up in the radix trees
1227 : * if we already know that we have the tag set.
1228 : */
1229 4197407 : if (ip->i_flags & iflag)
1230 : return;
1231 2919447 : spin_lock(&ip->i_flags_lock);
1232 2919434 : ip->i_flags |= iflag;
1233 2919434 : spin_unlock(&ip->i_flags_lock);
1234 :
1235 2919424 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1236 2919421 : spin_lock(&pag->pag_ici_lock);
1237 :
1238 2919438 : xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1239 : XFS_ICI_BLOCKGC_TAG);
1240 :
1241 2919443 : spin_unlock(&pag->pag_ici_lock);
1242 2919445 : xfs_perag_put(pag);
1243 : }
1244 :
1245 : void
1246 3409799 : xfs_inode_set_eofblocks_tag(
1247 : xfs_inode_t *ip)
1248 : {
1249 3409799 : trace_xfs_inode_set_eofblocks_tag(ip);
1250 3409769 : return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS);
1251 : }
1252 :
1253 : static void
1254 15481560 : xfs_blockgc_clear_iflag(
1255 : struct xfs_inode *ip,
1256 : unsigned long iflag)
1257 : {
1258 15481560 : struct xfs_mount *mp = ip->i_mount;
1259 15481560 : struct xfs_perag *pag;
1260 15481560 : bool clear_tag;
1261 :
1262 15481560 : ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
1263 :
1264 15481560 : spin_lock(&ip->i_flags_lock);
1265 15481784 : ip->i_flags &= ~iflag;
1266 15481784 : clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0;
1267 15481784 : spin_unlock(&ip->i_flags_lock);
1268 :
1269 15480746 : if (!clear_tag)
1270 : return;
1271 :
1272 13634851 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1273 13634791 : spin_lock(&pag->pag_ici_lock);
1274 :
1275 13634986 : xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1276 : XFS_ICI_BLOCKGC_TAG);
1277 :
1278 13635776 : spin_unlock(&pag->pag_ici_lock);
1279 13635560 : xfs_perag_put(pag);
1280 : }
1281 :
1282 : void
1283 9610828 : xfs_inode_clear_eofblocks_tag(
1284 : xfs_inode_t *ip)
1285 : {
1286 9610828 : trace_xfs_inode_clear_eofblocks_tag(ip);
1287 9610902 : return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS);
1288 : }
1289 :
1290 : /*
1291 : * Set ourselves up to free CoW blocks from this file. If it's already clean
1292 : * then we can bail out quickly, but otherwise we must back off if the file
1293 : * is undergoing some kind of write.
1294 : */
1295 : static bool
1296 328142 : xfs_prep_free_cowblocks(
1297 : struct xfs_inode *ip)
1298 : {
1299 : /*
1300 : * Just clear the tag if we have an empty cow fork or none at all. It's
1301 : * possible the inode was fully unshared since it was originally tagged.
1302 : */
1303 656284 : if (!xfs_inode_has_cow_data(ip)) {
1304 20028 : trace_xfs_inode_free_cowblocks_invalid(ip);
1305 20028 : xfs_inode_clear_cowblocks_tag(ip);
1306 20028 : return false;
1307 : }
1308 :
1309 : /*
1310 : * If the mapping is dirty or under writeback we cannot touch the
1311 : * CoW fork. Leave it alone if we're in the midst of a directio.
1312 : */
1313 308114 : if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1314 303093 : mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1315 301323 : mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1316 : atomic_read(&VFS_I(ip)->i_dio_count))
1317 6794 : return false;
1318 :
1319 : return true;
1320 : }
1321 :
1322 : /*
1323 : * Automatic CoW Reservation Freeing
1324 : *
1325 : * These functions automatically garbage collect leftover CoW reservations
1326 : * that were made on behalf of a cowextsize hint when we start to run out
1327 : * of quota or when the reservations sit around for too long. If the file
1328 : * has dirty pages or is undergoing writeback, its CoW reservations will
1329 : * be retained.
1330 : *
1331 : * The actual garbage collection piggybacks off the same code that runs
1332 : * the speculative EOF preallocation garbage collector.
1333 : */
1334 : STATIC int
1335 1308735 : xfs_inode_free_cowblocks(
1336 : struct xfs_inode *ip,
1337 : struct xfs_icwalk *icw,
1338 : unsigned int *lockflags)
1339 : {
1340 1308735 : bool wait;
1341 1308735 : int ret = 0;
1342 :
1343 1308735 : wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1344 :
1345 2617767 : if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS))
1346 : return 0;
1347 :
1348 177947 : if (!xfs_prep_free_cowblocks(ip))
1349 : return 0;
1350 :
1351 151111 : if (!xfs_icwalk_match(ip, icw))
1352 : return 0;
1353 :
1354 : /*
1355 : * If the caller is waiting, return -EAGAIN to keep the background
1356 : * scanner moving and revisit the inode in a subsequent pass.
1357 : */
1358 300224 : if (!(*lockflags & XFS_IOLOCK_EXCL) &&
1359 149084 : !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1360 901 : if (wait)
1361 : return -EAGAIN;
1362 406 : return 0;
1363 : }
1364 150239 : *lockflags |= XFS_IOLOCK_EXCL;
1365 :
1366 150239 : if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) {
1367 0 : if (wait)
1368 : return -EAGAIN;
1369 0 : return 0;
1370 : }
1371 150222 : *lockflags |= XFS_MMAPLOCK_EXCL;
1372 :
1373 : /*
1374 : * Check again, nobody else should be able to dirty blocks or change
1375 : * the reflink iflag now that we have the first two locks held.
1376 : */
1377 150222 : if (xfs_prep_free_cowblocks(ip))
1378 150222 : ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
1379 : return ret;
1380 : }
1381 :
1382 : void
1383 787606 : xfs_inode_set_cowblocks_tag(
1384 : xfs_inode_t *ip)
1385 : {
1386 787606 : trace_xfs_inode_set_cowblocks_tag(ip);
1387 787606 : return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS);
1388 : }
1389 :
1390 : void
1391 5870686 : xfs_inode_clear_cowblocks_tag(
1392 : xfs_inode_t *ip)
1393 : {
1394 5870686 : trace_xfs_inode_clear_cowblocks_tag(ip);
1395 5870702 : return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS);
1396 : }
1397 :
1398 : /* Disable post-EOF and CoW block auto-reclamation. */
1399 : void
1400 72995 : xfs_blockgc_stop(
1401 : struct xfs_mount *mp)
1402 : {
1403 72995 : struct xfs_perag *pag;
1404 72995 : xfs_agnumber_t agno;
1405 :
1406 72995 : if (!xfs_clear_blockgc_enabled(mp))
1407 14 : return;
1408 :
1409 402101 : for_each_perag(mp, agno, pag)
1410 329120 : cancel_delayed_work_sync(&pag->pag_blockgc_work);
1411 72981 : trace_xfs_blockgc_stop(mp, __return_address);
1412 : }
1413 :
1414 : /* Enable post-EOF and CoW block auto-reclamation. */
1415 : void
1416 73013 : xfs_blockgc_start(
1417 : struct xfs_mount *mp)
1418 : {
1419 73013 : struct xfs_perag *pag;
1420 73013 : xfs_agnumber_t agno;
1421 :
1422 73013 : if (xfs_set_blockgc_enabled(mp))
1423 : return;
1424 :
1425 73009 : trace_xfs_blockgc_start(mp, __return_address);
1426 241127 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1427 168118 : xfs_blockgc_queue(pag);
1428 : }
1429 :
1430 : /* Don't try to run block gc on an inode that's in any of these states. */
1431 : #define XFS_BLOCKGC_NOGRAB_IFLAGS (XFS_INEW | \
1432 : XFS_NEED_INACTIVE | \
1433 : XFS_INACTIVATING | \
1434 : XFS_IRECLAIMABLE | \
1435 : XFS_IRECLAIM)
1436 : /*
1437 : * Decide if the given @ip is eligible for garbage collection of speculative
1438 : * preallocations, and grab it if so. Returns true if it's ready to go or
1439 : * false if we should just ignore it.
1440 : */
1441 : static bool
1442 1513800 : xfs_blockgc_igrab(
1443 : struct xfs_inode *ip)
1444 : {
1445 1513800 : struct inode *inode = VFS_I(ip);
1446 :
1447 1513800 : ASSERT(rcu_read_lock_held());
1448 :
1449 : /* Check for stale RCU freed inode */
1450 1513800 : spin_lock(&ip->i_flags_lock);
1451 1513973 : if (!ip->i_ino)
1452 0 : goto out_unlock_noent;
1453 :
1454 1513973 : if (ip->i_flags & XFS_BLOCKGC_NOGRAB_IFLAGS)
1455 132749 : goto out_unlock_noent;
1456 1381224 : spin_unlock(&ip->i_flags_lock);
1457 :
1458 : /* nothing to sync during shutdown */
1459 2762438 : if (xfs_is_shutdown(ip->i_mount))
1460 : return false;
1461 :
1462 : /* If we can't grab the inode, it must on it's way to reclaim. */
1463 1381198 : if (!igrab(inode))
1464 272 : return false;
1465 :
1466 : /* inode is valid */
1467 : return true;
1468 :
1469 132749 : out_unlock_noent:
1470 132749 : spin_unlock(&ip->i_flags_lock);
1471 132749 : return false;
1472 : }
1473 :
1474 : /* Scan one incore inode for block preallocations that we can remove. */
1475 : static int
1476 1381458 : xfs_blockgc_scan_inode(
1477 : struct xfs_inode *ip,
1478 : struct xfs_icwalk *icw)
1479 : {
1480 1381458 : unsigned int lockflags = 0;
1481 1381458 : int error;
1482 :
1483 1381458 : error = xfs_inode_free_eofblocks(ip, icw, &lockflags);
1484 1381283 : if (error)
1485 72482 : goto unlock;
1486 :
1487 1308801 : error = xfs_inode_free_cowblocks(ip, icw, &lockflags);
1488 1381458 : unlock:
1489 1381458 : if (lockflags)
1490 705079 : xfs_iunlock(ip, lockflags);
1491 1381395 : xfs_irele(ip);
1492 1381467 : return error;
1493 : }
1494 :
1495 : /* Background worker that trims preallocated space. */
1496 : void
1497 486516 : xfs_blockgc_worker(
1498 : struct work_struct *work)
1499 : {
1500 486516 : struct xfs_perag *pag = container_of(to_delayed_work(work),
1501 : struct xfs_perag, pag_blockgc_work);
1502 486516 : struct xfs_mount *mp = pag->pag_mount;
1503 486516 : int error;
1504 :
1505 486516 : trace_xfs_blockgc_worker(mp, __return_address);
1506 :
1507 486520 : error = xfs_icwalk_ag(pag, XFS_ICWALK_BLOCKGC, NULL);
1508 486486 : if (error)
1509 0 : xfs_info(mp, "AG %u preallocation gc worker failed, err=%d",
1510 : pag->pag_agno, error);
1511 486486 : xfs_blockgc_queue(pag);
1512 486423 : }
1513 :
1514 : /*
1515 : * Try to free space in the filesystem by purging inactive inodes, eofblocks
1516 : * and cowblocks.
1517 : */
1518 : int
1519 436571 : xfs_blockgc_free_space(
1520 : struct xfs_mount *mp,
1521 : struct xfs_icwalk *icw)
1522 : {
1523 436571 : int error;
1524 :
1525 436571 : trace_xfs_blockgc_free_space(mp, icw, _RET_IP_);
1526 :
1527 436482 : error = xfs_icwalk(mp, XFS_ICWALK_BLOCKGC, icw);
1528 436432 : if (error)
1529 : return error;
1530 :
1531 436394 : return xfs_inodegc_flush(mp);
1532 : }
1533 :
1534 : /*
1535 : * Reclaim all the free space that we can by scheduling the background blockgc
1536 : * and inodegc workers immediately and waiting for them all to clear.
1537 : */
1538 : int
1539 4302302 : xfs_blockgc_flush_all(
1540 : struct xfs_mount *mp)
1541 : {
1542 4302302 : struct xfs_perag *pag;
1543 4302302 : xfs_agnumber_t agno;
1544 :
1545 4302302 : trace_xfs_blockgc_flush_all(mp, __return_address);
1546 :
1547 : /*
1548 : * For each blockgc worker, move its queue time up to now. If it
1549 : * wasn't queued, it will not be requeued. Then flush whatever's
1550 : * left.
1551 : */
1552 4739391 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1553 437065 : mod_delayed_work(pag->pag_mount->m_blockgc_wq,
1554 : &pag->pag_blockgc_work, 0);
1555 :
1556 4734903 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1557 432784 : flush_delayed_work(&pag->pag_blockgc_work);
1558 :
1559 4301975 : return xfs_inodegc_flush(mp);
1560 : }
1561 :
1562 : /*
1563 : * Run cow/eofblocks scans on the supplied dquots. We don't know exactly which
1564 : * quota caused an allocation failure, so we make a best effort by including
1565 : * each quota under low free space conditions (less than 1% free space) in the
1566 : * scan.
1567 : *
1568 : * Callers must not hold any inode's ILOCK. If requesting a synchronous scan
1569 : * (XFS_ICWALK_FLAG_SYNC), the caller also must not hold any inode's IOLOCK or
1570 : * MMAPLOCK.
1571 : */
1572 : int
1573 6832 : xfs_blockgc_free_dquots(
1574 : struct xfs_mount *mp,
1575 : struct xfs_dquot *udqp,
1576 : struct xfs_dquot *gdqp,
1577 : struct xfs_dquot *pdqp,
1578 : unsigned int iwalk_flags)
1579 : {
1580 6832 : struct xfs_icwalk icw = {0};
1581 6832 : bool do_work = false;
1582 :
1583 6832 : if (!udqp && !gdqp && !pdqp)
1584 : return 0;
1585 :
1586 : /*
1587 : * Run a scan to free blocks using the union filter to cover all
1588 : * applicable quotas in a single scan.
1589 : */
1590 6832 : icw.icw_flags = XFS_ICWALK_FLAG_UNION | iwalk_flags;
1591 :
1592 6832 : if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) {
1593 4657 : icw.icw_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id);
1594 4657 : icw.icw_flags |= XFS_ICWALK_FLAG_UID;
1595 4657 : do_work = true;
1596 : }
1597 :
1598 6832 : if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) {
1599 6588 : icw.icw_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id);
1600 6588 : icw.icw_flags |= XFS_ICWALK_FLAG_GID;
1601 6588 : do_work = true;
1602 : }
1603 :
1604 6832 : if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) {
1605 6612 : icw.icw_prid = pdqp->q_id;
1606 6612 : icw.icw_flags |= XFS_ICWALK_FLAG_PRID;
1607 6612 : do_work = true;
1608 : }
1609 :
1610 6832 : if (!do_work)
1611 : return 0;
1612 :
1613 6694 : return xfs_blockgc_free_space(mp, &icw);
1614 : }
1615 :
1616 : /* Run cow/eofblocks scans on the quotas attached to the inode. */
1617 : int
1618 4514 : xfs_blockgc_free_quota(
1619 : struct xfs_inode *ip,
1620 : unsigned int iwalk_flags)
1621 : {
1622 4514 : return xfs_blockgc_free_dquots(ip->i_mount,
1623 : xfs_inode_dquot(ip, XFS_DQTYPE_USER),
1624 : xfs_inode_dquot(ip, XFS_DQTYPE_GROUP),
1625 : xfs_inode_dquot(ip, XFS_DQTYPE_PROJ), iwalk_flags);
1626 : }
1627 :
1628 : /* XFS Inode Cache Walking Code */
1629 :
1630 : /*
1631 : * The inode lookup is done in batches to keep the amount of lock traffic and
1632 : * radix tree lookups to a minimum. The batch size is a trade off between
1633 : * lookup reduction and stack usage. This is in the reclaim path, so we can't
1634 : * be too greedy.
1635 : */
1636 : #define XFS_LOOKUP_BATCH 32
1637 :
1638 :
1639 : /*
1640 : * Decide if we want to grab this inode in anticipation of doing work towards
1641 : * the goal.
1642 : */
1643 : static inline bool
1644 530857520 : xfs_icwalk_igrab(
1645 : enum xfs_icwalk_goal goal,
1646 : struct xfs_inode *ip,
1647 : struct xfs_icwalk *icw)
1648 : {
1649 530857520 : switch (goal) {
1650 1513795 : case XFS_ICWALK_BLOCKGC:
1651 1513795 : return xfs_blockgc_igrab(ip);
1652 529343725 : case XFS_ICWALK_RECLAIM:
1653 529343725 : return xfs_reclaim_igrab(ip, icw);
1654 : default:
1655 : return false;
1656 : }
1657 : }
1658 :
1659 : /*
1660 : * Process an inode. Each processing function must handle any state changes
1661 : * made by the icwalk igrab function. Return -EAGAIN to skip an inode.
1662 : */
1663 : static inline int
1664 530659790 : xfs_icwalk_process_inode(
1665 : enum xfs_icwalk_goal goal,
1666 : struct xfs_inode *ip,
1667 : struct xfs_perag *pag,
1668 : struct xfs_icwalk *icw)
1669 : {
1670 530659790 : int error = 0;
1671 :
1672 530659790 : switch (goal) {
1673 1381512 : case XFS_ICWALK_BLOCKGC:
1674 1381512 : error = xfs_blockgc_scan_inode(ip, icw);
1675 1381512 : break;
1676 529278278 : case XFS_ICWALK_RECLAIM:
1677 529278278 : xfs_reclaim_inode(ip, pag);
1678 529278278 : break;
1679 : }
1680 530661593 : return error;
1681 : }
1682 :
1683 : /*
1684 : * For a given per-AG structure @pag and a goal, grab qualifying inodes and
1685 : * process them in some manner.
1686 : */
1687 : static int
1688 2680392 : xfs_icwalk_ag(
1689 : struct xfs_perag *pag,
1690 : enum xfs_icwalk_goal goal,
1691 : struct xfs_icwalk *icw)
1692 : {
1693 2680392 : struct xfs_mount *mp = pag->pag_mount;
1694 2680392 : uint32_t first_index;
1695 2680392 : int last_error = 0;
1696 2746684 : int skipped;
1697 2746684 : bool done;
1698 2746684 : int nr_found;
1699 :
1700 2746684 : restart:
1701 2746684 : done = false;
1702 2746684 : skipped = 0;
1703 2746684 : if (goal == XFS_ICWALK_RECLAIM)
1704 2110836 : first_index = READ_ONCE(pag->pag_ici_reclaim_cursor);
1705 : else
1706 : first_index = 0;
1707 : nr_found = 0;
1708 18692964 : do {
1709 18692964 : struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1710 18692964 : int error = 0;
1711 18692964 : int i;
1712 :
1713 18692964 : rcu_read_lock();
1714 :
1715 18692267 : nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root,
1716 : (void **) batch, first_index,
1717 : XFS_LOOKUP_BATCH, goal);
1718 18692683 : if (!nr_found) {
1719 1154065 : done = true;
1720 1154065 : rcu_read_unlock();
1721 2746744 : break;
1722 : }
1723 :
1724 : /*
1725 : * Grab the inodes before we drop the lock. if we found
1726 : * nothing, nr == 0 and the loop will be skipped.
1727 : */
1728 548396028 : for (i = 0; i < nr_found; i++) {
1729 530857051 : struct xfs_inode *ip = batch[i];
1730 :
1731 530857051 : if (done || !xfs_icwalk_igrab(goal, ip, icw))
1732 213004 : batch[i] = NULL;
1733 :
1734 : /*
1735 : * Update the index for the next lookup. Catch
1736 : * overflows into the next AG range which can occur if
1737 : * we have inodes in the last block of the AG and we
1738 : * are currently pointing to the last inode.
1739 : *
1740 : * Because we may see inodes that are from the wrong AG
1741 : * due to RCU freeing and reallocation, only update the
1742 : * index if it lies in this AG. It was a race that lead
1743 : * us to see this inode, so another lookup from the
1744 : * same index will not find it again.
1745 : */
1746 530857410 : if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
1747 4746 : continue;
1748 530852664 : first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1749 530852664 : if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1750 0 : done = true;
1751 : }
1752 :
1753 : /* unlock now we've grabbed the inodes. */
1754 17538977 : rcu_read_unlock();
1755 :
1756 565951823 : for (i = 0; i < nr_found; i++) {
1757 530873923 : if (!batch[i])
1758 213201 : continue;
1759 530660722 : error = xfs_icwalk_process_inode(goal, batch[i], pag,
1760 : icw);
1761 530660679 : if (error == -EAGAIN) {
1762 72968 : skipped++;
1763 72968 : continue;
1764 : }
1765 530587711 : if (error && last_error != -EFSCORRUPTED)
1766 0 : last_error = error;
1767 : }
1768 :
1769 : /* bail out if the filesystem is corrupted. */
1770 17538923 : if (error == -EFSCORRUPTED)
1771 : break;
1772 :
1773 17538923 : cond_resched();
1774 :
1775 17538992 : if (icw && (icw->icw_flags & XFS_ICWALK_FLAG_SCAN_LIMIT)) {
1776 5596250 : icw->icw_scan_limit -= XFS_LOOKUP_BATCH;
1777 5596250 : if (icw->icw_scan_limit <= 0)
1778 : break;
1779 : }
1780 15946280 : } while (nr_found && !done);
1781 :
1782 2746744 : if (goal == XFS_ICWALK_RECLAIM) {
1783 2110836 : if (done)
1784 518124 : first_index = 0;
1785 2110836 : WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index);
1786 : }
1787 :
1788 2746744 : if (skipped) {
1789 66366 : delay(1);
1790 66292 : goto restart;
1791 : }
1792 2680378 : return last_error;
1793 : }
1794 :
1795 : /* Walk all incore inodes to achieve a given goal. */
1796 : static int
1797 1035585 : xfs_icwalk(
1798 : struct xfs_mount *mp,
1799 : enum xfs_icwalk_goal goal,
1800 : struct xfs_icwalk *icw)
1801 : {
1802 1035585 : struct xfs_perag *pag;
1803 1035585 : int error = 0;
1804 1035585 : int last_error = 0;
1805 1035585 : xfs_agnumber_t agno;
1806 :
1807 3229513 : for_each_perag_tag(mp, agno, pag, goal) {
1808 2193956 : error = xfs_icwalk_ag(pag, goal, icw);
1809 2193928 : if (error) {
1810 0 : last_error = error;
1811 0 : if (error == -EFSCORRUPTED) {
1812 0 : xfs_perag_rele(pag);
1813 0 : break;
1814 : }
1815 : }
1816 : }
1817 1035583 : return last_error;
1818 : BUILD_BUG_ON(XFS_ICWALK_PRIVATE_FLAGS & XFS_ICWALK_FLAGS_VALID);
1819 : }
1820 :
1821 : #ifdef DEBUG
1822 : static void
1823 0 : xfs_check_delalloc(
1824 : struct xfs_inode *ip,
1825 : int whichfork)
1826 : {
1827 0 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1828 0 : struct xfs_bmbt_irec got;
1829 0 : struct xfs_iext_cursor icur;
1830 :
1831 0 : if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
1832 0 : return;
1833 0 : do {
1834 0 : if (isnullstartblock(got.br_startblock)) {
1835 0 : xfs_warn(ip->i_mount,
1836 : "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
1837 : ip->i_ino,
1838 : whichfork == XFS_DATA_FORK ? "data" : "cow",
1839 : got.br_startoff, got.br_blockcount);
1840 : }
1841 0 : } while (xfs_iext_next_extent(ifp, &icur, &got));
1842 : }
1843 : #else
1844 : #define xfs_check_delalloc(ip, whichfork) do { } while (0)
1845 : #endif
1846 :
1847 : /* Schedule the inode for reclaim. */
1848 : static void
1849 1075637417 : xfs_inodegc_set_reclaimable(
1850 : struct xfs_inode *ip)
1851 : {
1852 1075637417 : struct xfs_mount *mp = ip->i_mount;
1853 1075637417 : struct xfs_perag *pag;
1854 :
1855 2151274834 : if (!xfs_is_shutdown(mp) && ip->i_delayed_blks) {
1856 0 : xfs_check_delalloc(ip, XFS_DATA_FORK);
1857 0 : xfs_check_delalloc(ip, XFS_COW_FORK);
1858 0 : ASSERT(0);
1859 : }
1860 :
1861 1075637417 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1862 1075597272 : spin_lock(&pag->pag_ici_lock);
1863 1075706332 : spin_lock(&ip->i_flags_lock);
1864 :
1865 2151422762 : if (!xfs_is_shutdown(pag->pag_mount)) {
1866 : /* had better not be on any unlinked list! */
1867 770862489 : ASSERT(!xfs_inode_on_unlinked_list(ip));
1868 770862489 : if (xfs_inode_on_unlinked_list(ip))
1869 1 : xfs_emerg(pag->pag_mount, "IUNLINK mark reclaim ino 0x%llx nlink %u mode 0o%o prevun 0x%x nextun 0x%x", ip->i_ino, VFS_I(ip)->i_nlink, VFS_I(ip)->i_mode, ip->i_prev_unlinked, ip->i_next_unlinked);
1870 : }
1871 :
1872 1075711381 : trace_xfs_inode_set_reclaimable(ip);
1873 1075687775 : ip->i_flags &= ~(XFS_NEED_INACTIVE | XFS_INACTIVATING);
1874 1075687775 : ip->i_flags |= XFS_IRECLAIMABLE;
1875 1075687775 : xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1876 : XFS_ICI_RECLAIM_TAG);
1877 :
1878 1075657875 : spin_unlock(&ip->i_flags_lock);
1879 1075690283 : spin_unlock(&pag->pag_ici_lock);
1880 1075693517 : xfs_perag_put(pag);
1881 1075648687 : }
1882 :
1883 : /*
1884 : * Free all speculative preallocations and possibly even the inode itself.
1885 : * This is the last chance to make changes to an otherwise unreferenced file
1886 : * before incore reclamation happens.
1887 : */
1888 : static int
1889 38271508 : xfs_inodegc_inactivate(
1890 : struct xfs_inode *ip)
1891 : {
1892 38271508 : int error;
1893 :
1894 38271508 : trace_xfs_inode_inactivating(ip);
1895 38262732 : error = xfs_inactive(ip);
1896 38275198 : xfs_inodegc_set_reclaimable(ip);
1897 38270182 : return error;
1898 :
1899 : }
1900 :
1901 : void
1902 4613622 : xfs_inodegc_worker(
1903 : struct work_struct *work)
1904 : {
1905 4613622 : struct xfs_inodegc *gc = container_of(to_delayed_work(work),
1906 : struct xfs_inodegc, work);
1907 4613622 : struct llist_node *node = llist_del_all(&gc->list);
1908 4612385 : struct xfs_inode *ip, *n;
1909 4612385 : unsigned int nofs_flag;
1910 :
1911 4612385 : ASSERT(gc->cpu == smp_processor_id());
1912 :
1913 4612240 : WRITE_ONCE(gc->items, 0);
1914 :
1915 4612240 : if (!node)
1916 : return;
1917 :
1918 : /*
1919 : * We can allocate memory here while doing writeback on behalf of
1920 : * memory reclaim. To avoid memory allocation deadlocks set the
1921 : * task-wide nofs context for the following operations.
1922 : */
1923 4609079 : nofs_flag = memalloc_nofs_save();
1924 :
1925 4609079 : ip = llist_entry(node, struct xfs_inode, i_gclist);
1926 4609079 : trace_xfs_inodegc_worker(ip->i_mount, READ_ONCE(gc->shrinker_hits));
1927 :
1928 4601039 : WRITE_ONCE(gc->shrinker_hits, 0);
1929 42871055 : llist_for_each_entry_safe(ip, n, node, i_gclist) {
1930 38258045 : int error;
1931 :
1932 38258045 : xfs_iflags_set(ip, XFS_INACTIVATING);
1933 38265501 : error = xfs_inodegc_inactivate(ip);
1934 38270016 : if (error && !gc->error)
1935 1757 : gc->error = error;
1936 : }
1937 :
1938 4613010 : memalloc_nofs_restore(nofs_flag);
1939 : }
1940 :
1941 : /*
1942 : * Expedite all pending inodegc work to run immediately. This does not wait for
1943 : * completion of the work.
1944 : */
1945 : void
1946 14184053 : xfs_inodegc_push(
1947 : struct xfs_mount *mp)
1948 : {
1949 28368106 : if (!xfs_is_inodegc_enabled(mp))
1950 : return;
1951 14164346 : trace_xfs_inodegc_push(mp, __return_address);
1952 14151600 : xfs_inodegc_queue_all(mp);
1953 : }
1954 :
1955 : /*
1956 : * Force all currently queued inode inactivation work to run immediately and
1957 : * wait for the work to finish.
1958 : */
1959 : int
1960 5055021 : xfs_inodegc_flush(
1961 : struct xfs_mount *mp)
1962 : {
1963 5055021 : xfs_inodegc_push(mp);
1964 5054949 : trace_xfs_inodegc_flush(mp, __return_address);
1965 5054962 : return xfs_inodegc_wait_all(mp);
1966 : }
1967 :
1968 : /*
1969 : * Flush all the pending work and then disable the inode inactivation background
1970 : * workers and wait for them to stop. Caller must hold sb->s_umount to
1971 : * coordinate changes in the inodegc_enabled state.
1972 : */
1973 : void
1974 73029 : xfs_inodegc_stop(
1975 : struct xfs_mount *mp)
1976 : {
1977 73029 : bool rerun;
1978 :
1979 73029 : if (!xfs_clear_inodegc_enabled(mp))
1980 : return;
1981 :
1982 : /*
1983 : * Drain all pending inodegc work, including inodes that could be
1984 : * queued by racing xfs_inodegc_queue or xfs_inodegc_shrinker_scan
1985 : * threads that sample the inodegc state just prior to us clearing it.
1986 : * The inodegc flag state prevents new threads from queuing more
1987 : * inodes, so we queue pending work items and flush the workqueue until
1988 : * all inodegc lists are empty. IOWs, we cannot use drain_workqueue
1989 : * here because it does not allow other unserialized mechanisms to
1990 : * reschedule inodegc work while this draining is in progress.
1991 : */
1992 73015 : xfs_inodegc_queue_all(mp);
1993 73015 : do {
1994 73015 : flush_workqueue(mp->m_inodegc_wq);
1995 73015 : rerun = xfs_inodegc_queue_all(mp);
1996 73015 : } while (rerun);
1997 :
1998 73015 : trace_xfs_inodegc_stop(mp, __return_address);
1999 : }
2000 :
2001 : /*
2002 : * Enable the inode inactivation background workers and schedule deferred inode
2003 : * inactivation work if there is any. Caller must hold sb->s_umount to
2004 : * coordinate changes in the inodegc_enabled state.
2005 : */
2006 : void
2007 73013 : xfs_inodegc_start(
2008 : struct xfs_mount *mp)
2009 : {
2010 73013 : if (xfs_set_inodegc_enabled(mp))
2011 : return;
2012 :
2013 73009 : trace_xfs_inodegc_start(mp, __return_address);
2014 73009 : xfs_inodegc_queue_all(mp);
2015 : }
2016 :
2017 : #ifdef CONFIG_XFS_RT
2018 : static inline bool
2019 32071125 : xfs_inodegc_want_queue_rt_file(
2020 : struct xfs_inode *ip)
2021 : {
2022 32071125 : struct xfs_mount *mp = ip->i_mount;
2023 :
2024 32071125 : if (!XFS_IS_REALTIME_INODE(ip))
2025 : return false;
2026 :
2027 1842611 : if (__percpu_counter_compare(&mp->m_frextents,
2028 1842614 : mp->m_low_rtexts[XFS_LOWSP_5_PCNT],
2029 : XFS_FDBLOCKS_BATCH) < 0)
2030 0 : return true;
2031 :
2032 : return false;
2033 : }
2034 : #else
2035 : # define xfs_inodegc_want_queue_rt_file(ip) (false)
2036 : #endif /* CONFIG_XFS_RT */
2037 :
2038 : /*
2039 : * Schedule the inactivation worker when:
2040 : *
2041 : * - We've accumulated more than one inode cluster buffer's worth of inodes.
2042 : * - There is less than 5% free space left.
2043 : * - Any of the quotas for this inode are near an enforcement limit.
2044 : */
2045 : static inline bool
2046 38274917 : xfs_inodegc_want_queue_work(
2047 : struct xfs_inode *ip,
2048 : unsigned int items)
2049 : {
2050 38274917 : struct xfs_mount *mp = ip->i_mount;
2051 :
2052 38274917 : if (items > mp->m_ino_geo.inodes_per_cluster)
2053 : return true;
2054 :
2055 32209619 : if (__percpu_counter_compare(&mp->m_fdblocks,
2056 32209768 : mp->m_low_space[XFS_LOWSP_5_PCNT],
2057 : XFS_FDBLOCKS_BATCH) < 0)
2058 : return true;
2059 :
2060 32071348 : if (xfs_inodegc_want_queue_rt_file(ip))
2061 : return true;
2062 :
2063 32071546 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_USER))
2064 : return true;
2065 :
2066 32071083 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_GROUP))
2067 : return true;
2068 :
2069 32071157 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_PROJ))
2070 38 : return true;
2071 :
2072 : return false;
2073 : }
2074 :
2075 : /*
2076 : * Upper bound on the number of inodes in each AG that can be queued for
2077 : * inactivation at any given time, to avoid monopolizing the workqueue.
2078 : */
2079 : #define XFS_INODEGC_MAX_BACKLOG (4 * XFS_INODES_PER_CHUNK)
2080 :
2081 : /*
2082 : * Make the frontend wait for inactivations when:
2083 : *
2084 : * - Memory shrinkers queued the inactivation worker and it hasn't finished.
2085 : * - The queue depth exceeds the maximum allowable percpu backlog.
2086 : *
2087 : * Note: If the current thread is running a transaction, we don't ever want to
2088 : * wait for other transactions because that could introduce a deadlock.
2089 : */
2090 : static inline bool
2091 : xfs_inodegc_want_flush_work(
2092 : struct xfs_inode *ip,
2093 : unsigned int items,
2094 : unsigned int shrinker_hits)
2095 : {
2096 38273992 : if (current->journal_info)
2097 : return false;
2098 :
2099 38236780 : if (shrinker_hits > 0)
2100 : return true;
2101 :
2102 38236776 : if (items > XFS_INODEGC_MAX_BACKLOG)
2103 : return true;
2104 :
2105 : return false;
2106 : }
2107 :
2108 : /*
2109 : * Queue a background inactivation worker if there are inodes that need to be
2110 : * inactivated and higher level xfs code hasn't disabled the background
2111 : * workers.
2112 : */
2113 : static void
2114 38273139 : xfs_inodegc_queue(
2115 : struct xfs_inode *ip)
2116 : {
2117 38273139 : struct xfs_mount *mp = ip->i_mount;
2118 38273139 : struct xfs_inodegc *gc;
2119 38273139 : int items;
2120 38273139 : unsigned int shrinker_hits;
2121 38273139 : unsigned long queue_delay = 1;
2122 :
2123 38273139 : trace_xfs_inode_set_need_inactive(ip);
2124 38274517 : spin_lock(&ip->i_flags_lock);
2125 38274796 : ip->i_flags |= XFS_NEED_INACTIVE;
2126 38274796 : spin_unlock(&ip->i_flags_lock);
2127 :
2128 38274711 : gc = get_cpu_ptr(mp->m_inodegc);
2129 38274945 : llist_add(&ip->i_gclist, &gc->list);
2130 38274931 : items = READ_ONCE(gc->items);
2131 38274931 : WRITE_ONCE(gc->items, items + 1);
2132 38274931 : shrinker_hits = READ_ONCE(gc->shrinker_hits);
2133 :
2134 : /*
2135 : * We queue the work while holding the current CPU so that the work
2136 : * is scheduled to run on this CPU.
2137 : */
2138 76549862 : if (!xfs_is_inodegc_enabled(mp)) {
2139 0 : put_cpu_ptr(gc);
2140 0 : return;
2141 : }
2142 :
2143 38274931 : if (xfs_inodegc_want_queue_work(ip, items))
2144 6203697 : queue_delay = 0;
2145 :
2146 38274610 : trace_xfs_inodegc_queue(mp, __return_address);
2147 38274401 : mod_delayed_work_on(current_cpu(), mp->m_inodegc_wq, &gc->work,
2148 : queue_delay);
2149 38273376 : put_cpu_ptr(gc);
2150 :
2151 38273992 : if (xfs_inodegc_want_flush_work(ip, items, shrinker_hits)) {
2152 967864 : trace_xfs_inodegc_throttle(mp, __return_address);
2153 967850 : flush_delayed_work(&gc->work);
2154 : }
2155 : }
2156 :
2157 : /*
2158 : * Fold the dead CPU inodegc queue into the current CPUs queue.
2159 : */
2160 : void
2161 9 : xfs_inodegc_cpu_dead(
2162 : struct xfs_mount *mp,
2163 : unsigned int dead_cpu)
2164 : {
2165 9 : struct xfs_inodegc *dead_gc, *gc;
2166 9 : struct llist_node *first, *last;
2167 9 : unsigned int count = 0;
2168 :
2169 9 : dead_gc = per_cpu_ptr(mp->m_inodegc, dead_cpu);
2170 9 : cancel_delayed_work_sync(&dead_gc->work);
2171 :
2172 9 : if (llist_empty(&dead_gc->list))
2173 : return;
2174 :
2175 0 : first = dead_gc->list.first;
2176 0 : last = first;
2177 0 : while (last->next) {
2178 0 : last = last->next;
2179 0 : count++;
2180 : }
2181 0 : dead_gc->list.first = NULL;
2182 0 : dead_gc->items = 0;
2183 :
2184 : /* Add pending work to current CPU */
2185 0 : gc = get_cpu_ptr(mp->m_inodegc);
2186 0 : llist_add_batch(first, last, &gc->list);
2187 0 : count += READ_ONCE(gc->items);
2188 0 : WRITE_ONCE(gc->items, count);
2189 :
2190 0 : if (xfs_is_inodegc_enabled(mp)) {
2191 0 : trace_xfs_inodegc_queue(mp, __return_address);
2192 0 : mod_delayed_work_on(current_cpu(), mp->m_inodegc_wq, &gc->work,
2193 : 0);
2194 : }
2195 0 : put_cpu_ptr(gc);
2196 : }
2197 :
2198 : /*
2199 : * We set the inode flag atomically with the radix tree tag. Once we get tag
2200 : * lookups on the radix tree, this inode flag can go away.
2201 : *
2202 : * We always use background reclaim here because even if the inode is clean, it
2203 : * still may be under IO and hence we have wait for IO completion to occur
2204 : * before we can reclaim the inode. The background reclaim path handles this
2205 : * more efficiently than we can here, so simply let background reclaim tear down
2206 : * all inodes.
2207 : */
2208 : void
2209 1075513810 : xfs_inode_mark_reclaimable(
2210 : struct xfs_inode *ip)
2211 : {
2212 1075513810 : struct xfs_mount *mp = ip->i_mount;
2213 1075513810 : bool need_inactive;
2214 :
2215 1075513810 : XFS_STATS_INC(mp, vn_reclaim);
2216 :
2217 : /*
2218 : * We should never get here with any of the reclaim flags already set.
2219 : */
2220 2151218122 : ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_ALL_IRECLAIM_FLAGS));
2221 :
2222 1075704312 : need_inactive = xfs_inode_needs_inactive(ip);
2223 1075618950 : if (need_inactive) {
2224 38272841 : xfs_inodegc_queue(ip);
2225 38272841 : return;
2226 : }
2227 :
2228 : /* Going straight to reclaim, so drop the dquots. */
2229 1037346109 : xfs_qm_dqdetach(ip);
2230 1037335845 : xfs_inodegc_set_reclaimable(ip);
2231 : }
2232 :
2233 : /*
2234 : * Register a phony shrinker so that we can run background inodegc sooner when
2235 : * there's memory pressure. Inactivation does not itself free any memory but
2236 : * it does make inodes reclaimable, which eventually frees memory.
2237 : *
2238 : * The count function, seek value, and batch value are crafted to trigger the
2239 : * scan function during the second round of scanning. Hopefully this means
2240 : * that we reclaimed enough memory that initiating metadata transactions won't
2241 : * make things worse.
2242 : */
2243 : #define XFS_INODEGC_SHRINKER_COUNT (1UL << DEF_PRIORITY)
2244 : #define XFS_INODEGC_SHRINKER_BATCH ((XFS_INODEGC_SHRINKER_COUNT / 2) + 1)
2245 :
2246 : static unsigned long
2247 6547 : xfs_inodegc_shrinker_count(
2248 : struct shrinker *shrink,
2249 : struct shrink_control *sc)
2250 : {
2251 6547 : struct xfs_mount *mp = container_of(shrink, struct xfs_mount,
2252 : m_inodegc_shrinker);
2253 6547 : struct xfs_inodegc *gc;
2254 6547 : int cpu;
2255 :
2256 13094 : if (!xfs_is_inodegc_enabled(mp))
2257 : return 0;
2258 :
2259 19322 : for_each_online_cpu(cpu) {
2260 12966 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
2261 12966 : if (!llist_empty(&gc->list))
2262 : return XFS_INODEGC_SHRINKER_COUNT;
2263 : }
2264 :
2265 : return 0;
2266 : }
2267 :
2268 : static unsigned long
2269 119 : xfs_inodegc_shrinker_scan(
2270 : struct shrinker *shrink,
2271 : struct shrink_control *sc)
2272 : {
2273 119 : struct xfs_mount *mp = container_of(shrink, struct xfs_mount,
2274 : m_inodegc_shrinker);
2275 119 : struct xfs_inodegc *gc;
2276 119 : int cpu;
2277 119 : bool no_items = true;
2278 :
2279 238 : if (!xfs_is_inodegc_enabled(mp))
2280 : return SHRINK_STOP;
2281 :
2282 119 : trace_xfs_inodegc_shrinker_scan(mp, sc, __return_address);
2283 :
2284 476 : for_each_online_cpu(cpu) {
2285 238 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
2286 238 : if (!llist_empty(&gc->list)) {
2287 122 : unsigned int h = READ_ONCE(gc->shrinker_hits);
2288 :
2289 122 : WRITE_ONCE(gc->shrinker_hits, h + 1);
2290 122 : mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0);
2291 122 : no_items = false;
2292 : }
2293 : }
2294 :
2295 : /*
2296 : * If there are no inodes to inactivate, we don't want the shrinker
2297 : * to think there's deferred work to call us back about.
2298 : */
2299 119 : if (no_items)
2300 0 : return LONG_MAX;
2301 :
2302 : return SHRINK_STOP;
2303 : }
2304 :
2305 : /* Register a shrinker so we can accelerate inodegc and throttle queuing. */
2306 : int
2307 24333 : xfs_inodegc_register_shrinker(
2308 : struct xfs_mount *mp)
2309 : {
2310 24333 : struct shrinker *shrink = &mp->m_inodegc_shrinker;
2311 :
2312 24333 : shrink->count_objects = xfs_inodegc_shrinker_count;
2313 24333 : shrink->scan_objects = xfs_inodegc_shrinker_scan;
2314 24333 : shrink->seeks = 0;
2315 24333 : shrink->flags = SHRINKER_NONSLAB;
2316 24333 : shrink->batch = XFS_INODEGC_SHRINKER_BATCH;
2317 :
2318 24333 : return register_shrinker(shrink, "xfs-inodegc:%s", mp->m_super->s_id);
2319 : }
|
