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 444998433 : xfs_inode_alloc(
76 : struct xfs_mount *mp,
77 : xfs_ino_t ino)
78 : {
79 444998433 : 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 444998433 : ip = alloc_inode_sb(mp->m_super, xfs_inode_cache, GFP_KERNEL | __GFP_NOFAIL);
86 :
87 445710232 : 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 445620923 : VFS_I(ip)->i_mode = 0;
94 445620923 : VFS_I(ip)->i_state = 0;
95 445620923 : mapping_set_large_folios(VFS_I(ip)->i_mapping);
96 :
97 445638967 : XFS_STATS_INC(mp, vn_active);
98 445322733 : ASSERT(atomic_read(&ip->i_pincount) == 0);
99 445322733 : ASSERT(ip->i_ino == 0);
100 :
101 : /* initialise the xfs inode */
102 445322733 : ip->i_ino = ino;
103 445322733 : ip->i_mount = mp;
104 445322733 : memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
105 445322733 : ip->i_cowfp = NULL;
106 445322733 : memset(&ip->i_af, 0, sizeof(ip->i_af));
107 445322733 : ip->i_af.if_format = XFS_DINODE_FMT_EXTENTS;
108 445322733 : memset(&ip->i_df, 0, sizeof(ip->i_df));
109 445322733 : ip->i_flags = 0;
110 445322733 : ip->i_delayed_blks = 0;
111 445322733 : ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
112 445322733 : ip->i_nblocks = 0;
113 445322733 : ip->i_forkoff = 0;
114 445322733 : ip->i_sick = 0;
115 445322733 : ip->i_checked = 0;
116 445322733 : INIT_WORK(&ip->i_ioend_work, xfs_end_io);
117 445322733 : INIT_LIST_HEAD(&ip->i_ioend_list);
118 445322733 : spin_lock_init(&ip->i_ioend_lock);
119 445686716 : ip->i_next_unlinked = NULLAGINO;
120 445686716 : ip->i_prev_unlinked = 0;
121 :
122 445686716 : return ip;
123 : }
124 :
125 : STATIC void
126 444709486 : xfs_inode_free_callback(
127 : struct rcu_head *head)
128 : {
129 444709486 : struct inode *inode = container_of(head, struct inode, i_rcu);
130 444709486 : struct xfs_inode *ip = XFS_I(inode);
131 :
132 444709486 : switch (VFS_I(ip)->i_mode & S_IFMT) {
133 287198690 : case S_IFREG:
134 : case S_IFDIR:
135 : case S_IFLNK:
136 287198690 : xfs_idestroy_fork(&ip->i_df);
137 287198690 : break;
138 : }
139 :
140 444726567 : xfs_ifork_zap_attr(ip);
141 :
142 445240087 : if (ip->i_cowfp) {
143 96025217 : xfs_idestroy_fork(ip->i_cowfp);
144 96011894 : kmem_cache_free(xfs_ifork_cache, ip->i_cowfp);
145 : }
146 445092811 : if (ip->i_itemp) {
147 83950298 : ASSERT(!test_bit(XFS_LI_IN_AIL,
148 : &ip->i_itemp->ili_item.li_flags));
149 83950298 : xfs_inode_item_destroy(ip);
150 83986216 : ip->i_itemp = NULL;
151 : }
152 :
153 445128729 : kmem_cache_free(xfs_inode_cache, ip);
154 444948469 : }
155 :
156 : static void
157 445921946 : __xfs_inode_free(
158 : struct xfs_inode *ip)
159 : {
160 : /* asserts to verify all state is correct here */
161 445921946 : ASSERT(atomic_read(&ip->i_pincount) == 0);
162 445921946 : ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list));
163 445921946 : XFS_STATS_DEC(ip->i_mount, vn_active);
164 :
165 445921941 : call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
166 445922057 : }
167 :
168 : void
169 816345 : xfs_inode_free(
170 : struct xfs_inode *ip)
171 : {
172 1632691 : 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 816346 : spin_lock(&ip->i_flags_lock);
181 816347 : ip->i_flags = XFS_IRECLAIM;
182 816347 : ip->i_ino = 0;
183 816347 : spin_unlock(&ip->i_flags_lock);
184 :
185 816347 : __xfs_inode_free(ip);
186 816344 : }
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 9297240 : xfs_reclaim_work_queue(
194 : struct xfs_mount *mp)
195 : {
196 :
197 9297240 : rcu_read_lock();
198 9295309 : if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
199 9271618 : queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
200 9273397 : msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
201 : }
202 9298380 : rcu_read_unlock();
203 9295682 : }
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 5823881 : xfs_blockgc_queue(
211 : struct xfs_perag *pag)
212 : {
213 5823881 : struct xfs_mount *mp = pag->pag_mount;
214 :
215 11647762 : if (!xfs_is_blockgc_enabled(mp))
216 : return;
217 :
218 5823565 : rcu_read_lock();
219 5823503 : if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG))
220 5748337 : queue_delayed_work(pag->pag_mount->m_blockgc_wq,
221 : &pag->pag_blockgc_work,
222 5748298 : msecs_to_jiffies(xfs_blockgc_secs * 1000));
223 5823671 : 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 886078071 : xfs_perag_set_inode_tag(
229 : struct xfs_perag *pag,
230 : xfs_agino_t agino,
231 : unsigned int tag)
232 : {
233 886078071 : struct xfs_mount *mp = pag->pag_mount;
234 886078071 : bool was_tagged;
235 :
236 886078071 : lockdep_assert_held(&pag->pag_ici_lock);
237 :
238 886078071 : was_tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
239 886056534 : radix_tree_tag_set(&pag->pag_ici_root, agino, tag);
240 :
241 886116844 : if (tag == XFS_ICI_RECLAIM_TAG)
242 879623973 : pag->pag_ici_reclaimable++;
243 :
244 886116844 : if (was_tagged)
245 : return;
246 :
247 : /* propagate the tag up into the perag radix tree */
248 9816852 : spin_lock(&mp->m_perag_lock);
249 9830850 : radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, tag);
250 9830850 : spin_unlock(&mp->m_perag_lock);
251 :
252 : /* start background work */
253 9830832 : switch (tag) {
254 9182324 : case XFS_ICI_RECLAIM_TAG:
255 9182324 : xfs_reclaim_work_queue(mp);
256 9182324 : break;
257 648508 : case XFS_ICI_BLOCKGC_TAG:
258 648508 : xfs_blockgc_queue(pag);
259 648508 : break;
260 : }
261 :
262 9814714 : 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 904934145 : xfs_perag_clear_inode_tag(
268 : struct xfs_perag *pag,
269 : xfs_agino_t agino,
270 : unsigned int tag)
271 : {
272 904934145 : struct xfs_mount *mp = pag->pag_mount;
273 :
274 904934145 : 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 904934145 : if (agino != NULLAGINO)
281 459828436 : radix_tree_tag_clear(&pag->pag_ici_root, agino, tag);
282 : else
283 445105709 : ASSERT(tag == XFS_ICI_RECLAIM_TAG);
284 :
285 904926282 : if (tag == XFS_ICI_RECLAIM_TAG)
286 879740589 : pag->pag_ici_reclaimable--;
287 :
288 904926282 : if (radix_tree_tagged(&pag->pag_ici_root, tag))
289 : return;
290 :
291 : /* clear the tag from the perag radix tree */
292 23002925 : spin_lock(&mp->m_perag_lock);
293 23115517 : radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, tag);
294 23115517 : spin_unlock(&mp->m_perag_lock);
295 :
296 23115242 : 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 434660510 : xfs_reinit_inode(
309 : struct xfs_mount *mp,
310 : struct inode *inode)
311 : {
312 434660510 : int error;
313 434660510 : uint32_t nlink = inode->i_nlink;
314 434660510 : uint32_t generation = inode->i_generation;
315 434660510 : uint64_t version = inode_peek_iversion(inode);
316 434660510 : umode_t mode = inode->i_mode;
317 434660510 : dev_t dev = inode->i_rdev;
318 434660510 : kuid_t uid = inode->i_uid;
319 434660510 : kgid_t gid = inode->i_gid;
320 :
321 434660510 : error = inode_init_always(mp->m_super, inode);
322 :
323 434645066 : set_nlink(inode, nlink);
324 434627604 : inode->i_generation = generation;
325 434627604 : inode_set_iversion_queried(inode, version);
326 434627604 : inode->i_mode = mode;
327 434627604 : inode->i_rdev = dev;
328 434627604 : inode->i_uid = uid;
329 434627604 : inode->i_gid = gid;
330 434627604 : mapping_set_large_folios(inode->i_mapping);
331 434629126 : 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 434637674 : xfs_iget_recycle(
340 : struct xfs_perag *pag,
341 : struct xfs_inode *ip) __releases(&ip->i_flags_lock)
342 : {
343 434637674 : struct xfs_mount *mp = ip->i_mount;
344 434637674 : struct inode *inode = VFS_I(ip);
345 434637674 : int error;
346 :
347 434637674 : trace_xfs_iget_recycle(ip);
348 :
349 434629245 : 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 434638504 : ip->i_flags |= XFS_IRECLAIM;
359 :
360 434638504 : spin_unlock(&ip->i_flags_lock);
361 434666683 : rcu_read_unlock();
362 :
363 434663512 : ASSERT(!rwsem_is_locked(&inode->i_rwsem));
364 434663512 : error = xfs_reinit_inode(mp, inode);
365 434630666 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
366 434625597 : 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 434625597 : spin_lock(&pag->pag_ici_lock);
383 434673936 : 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 434681908 : ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
391 434681908 : ip->i_flags |= XFS_INEW;
392 434681908 : xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
393 : XFS_ICI_RECLAIM_TAG);
394 434627787 : inode->i_state = I_NEW;
395 434627787 : spin_unlock(&ip->i_flags_lock);
396 434662040 : spin_unlock(&pag->pag_ici_lock);
397 :
398 434662040 : 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 >14093*10^7 : xfs_iget_check_free_state(
413 : struct xfs_inode *ip,
414 : int flags)
415 : {
416 >14093*10^7 : if (flags & XFS_IGET_CREATE) {
417 : /* should be a free inode */
418 123720568 : 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 123720568 : 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 >14081*10^7 : if (VFS_I(ip)->i_mode == 0)
442 2563920 : return -ENOENT;
443 :
444 : return 0;
445 : }
446 :
447 : /* Make all pending inactivation work start immediately. */
448 : static bool
449 47051756 : xfs_inodegc_queue_all(
450 : struct xfs_mount *mp)
451 : {
452 47051756 : struct xfs_inodegc *gc;
453 47051756 : int cpu;
454 47051756 : bool ret = false;
455 :
456 235033760 : for_each_online_cpu(cpu) {
457 187931501 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
458 187970961 : if (!llist_empty(&gc->list)) {
459 4008420 : mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0);
460 4008420 : ret = true;
461 : }
462 : }
463 :
464 47032650 : return ret;
465 : }
466 :
467 : /* Wait for all queued work and collect errors */
468 : static int
469 16950883 : xfs_inodegc_wait_all(
470 : struct xfs_mount *mp)
471 : {
472 16950883 : int cpu;
473 16950883 : int error = 0;
474 :
475 16950883 : flush_workqueue(mp->m_inodegc_wq);
476 101776772 : for_each_online_cpu(cpu) {
477 67858395 : struct xfs_inodegc *gc;
478 :
479 67858395 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
480 67858773 : if (gc->error && !error)
481 2106 : error = gc->error;
482 67858773 : gc->error = 0;
483 : }
484 :
485 16966262 : return error;
486 : }
487 :
488 : /*
489 : * Check the validity of the inode we just found it the cache
490 : */
491 : static int
492 >13975*10^7 : 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 >13975*10^7 : struct inode *inode = VFS_I(ip);
500 >13975*10^7 : struct xfs_mount *mp = ip->i_mount;
501 >13975*10^7 : 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 >13975*10^7 : spin_lock(&ip->i_flags_lock);
511 >14113*10^7 : if (ip->i_ino != ino)
512 10 : 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 >14113*10^7 : if (ip->i_flags & (XFS_INEW | XFS_IRECLAIM | XFS_INACTIVATING))
533 493465 : goto out_skip;
534 :
535 >14113*10^7 : if (ip->i_flags & XFS_NEED_INACTIVE) {
536 : /* Unlinked inodes cannot be re-grabbed. */
537 5038589 : if (VFS_I(ip)->i_nlink == 0) {
538 5016905 : error = -ENOENT;
539 5016905 : goto out_error;
540 : }
541 21684 : 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 >14113*10^7 : error = xfs_iget_check_free_state(ip, flags);
549 >14003*10^7 : if (error)
550 2563911 : goto out_error;
551 :
552 : /* Skip inodes that have no vfs state. */
553 >14003*10^7 : 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 >14003*10^7 : if (ip->i_flags & XFS_IRECLAIMABLE) {
559 : /* Drops i_flags_lock and RCU read lock. */
560 434641832 : error = xfs_iget_recycle(pag, ip);
561 434667835 : if (error == -EAGAIN)
562 4 : goto out_skip;
563 434667831 : if (error)
564 : return error;
565 : } else {
566 : /* If the VFS inode is being torn down, pause and try again. */
567 >13959*10^7 : if (!igrab(inode))
568 167759 : goto out_skip;
569 :
570 : /* We've got a live one. */
571 >14026*10^7 : spin_unlock(&ip->i_flags_lock);
572 >14045*10^7 : rcu_read_unlock();
573 >14044*10^7 : trace_xfs_iget_hit(ip);
574 : }
575 :
576 >13961*10^7 : if (lock_flags != 0)
577 81301242018 : xfs_ilock(ip, lock_flags);
578 :
579 >14041*10^7 : if (!(flags & XFS_IGET_INCORE))
580 >14040*10^7 : xfs_iflags_clear(ip, XFS_ISTALE);
581 >14106*10^7 : XFS_STATS_INC(mp, xs_ig_found);
582 :
583 >14053*10^7 : return 0;
584 :
585 661238 : out_skip:
586 661238 : trace_xfs_iget_skip(ip);
587 661133 : XFS_STATS_INC(mp, xs_ig_frecycle);
588 661158 : error = -EAGAIN;
589 8241974 : out_error:
590 8241974 : spin_unlock(&ip->i_flags_lock);
591 8242034 : rcu_read_unlock();
592 8242034 : return error;
593 :
594 : out_inodegc_flush:
595 21684 : spin_unlock(&ip->i_flags_lock);
596 21684 : 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 43364 : if (xfs_is_inodegc_enabled(mp))
602 21681 : xfs_inodegc_queue_all(mp);
603 : return -EAGAIN;
604 : }
605 :
606 : static int
607 445077311 : 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 445077311 : struct xfs_inode *ip;
617 445077311 : int error;
618 445077311 : xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
619 445077311 : int iflags;
620 :
621 445077311 : ip = xfs_inode_alloc(mp, ino);
622 445508052 : if (!ip)
623 : return -ENOMEM;
624 :
625 445508052 : error = xfs_imap(pag, tp, ip->i_ino, &ip->i_imap, flags);
626 445512442 : if (error)
627 114707 : 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 445397735 : if (xfs_has_v3inodes(mp) &&
640 445395462 : (flags & XFS_IGET_CREATE) && !xfs_has_ikeep(mp)) {
641 69036331 : VFS_I(ip)->i_generation = get_random_u32();
642 : } else {
643 376361404 : struct xfs_buf *bp;
644 :
645 376361404 : error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp);
646 376243895 : if (error)
647 6433 : goto out_destroy;
648 :
649 376225925 : error = xfs_inode_from_disk(ip,
650 376238566 : xfs_buf_offset(bp, ip->i_imap.im_boffset));
651 376219736 : if (!error)
652 376218632 : xfs_buf_set_ref(bp, XFS_INO_REF);
653 : else
654 1104 : xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
655 376225861 : xfs_trans_brelse(tp, bp);
656 :
657 376248386 : if (error)
658 1104 : goto out_destroy;
659 : }
660 :
661 445354252 : 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 445116378 : error = xfs_iget_check_free_state(ip, flags);
668 445030023 : 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 445030023 : 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 445259179 : if (lock_flags) {
687 419191816 : 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 445290059 : iflags = XFS_INEW;
701 445290059 : if (flags & XFS_IGET_DONTCACHE)
702 350317232 : d_mark_dontcache(VFS_I(ip));
703 445292888 : ip->i_udquot = NULL;
704 445292888 : ip->i_gdquot = NULL;
705 445292888 : ip->i_pdquot = NULL;
706 445292888 : xfs_iflags_set(ip, iflags);
707 :
708 : /* insert the new inode */
709 445588571 : spin_lock(&pag->pag_ici_lock);
710 445632400 : error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
711 445506420 : if (unlikely(error)) {
712 695207 : WARN_ON(error != -EEXIST);
713 695207 : XFS_STATS_INC(mp, xs_ig_dup);
714 695207 : error = -EAGAIN;
715 695207 : goto out_preload_end;
716 : }
717 444811213 : spin_unlock(&pag->pag_ici_lock);
718 444918808 : radix_tree_preload_end();
719 :
720 444300664 : *ipp = ip;
721 444300664 : return 0;
722 :
723 : out_preload_end:
724 695207 : spin_unlock(&pag->pag_ici_lock);
725 695207 : radix_tree_preload_end();
726 695207 : if (lock_flags)
727 689937 : xfs_iunlock(ip, lock_flags);
728 5270 : out_destroy:
729 816345 : __destroy_inode(VFS_I(ip));
730 816344 : xfs_inode_free(ip);
731 816344 : 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 >14131*10^7 : 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 >14131*10^7 : struct xfs_inode *ip;
756 >14131*10^7 : struct xfs_perag *pag;
757 >14131*10^7 : xfs_agino_t agino;
758 >14131*10^7 : int error;
759 :
760 >14131*10^7 : ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
761 :
762 : /* reject inode numbers outside existing AGs */
763 >14131*10^7 : if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
764 1603482 : return -EINVAL;
765 :
766 >14131*10^7 : XFS_STATS_INC(mp, xs_ig_attempts);
767 :
768 : /* get the perag structure and ensure that it's inode capable */
769 >13925*10^7 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
770 >14121*10^7 : agino = XFS_INO_TO_AGINO(mp, ino);
771 :
772 >14121*10^7 : again:
773 >14121*10^7 : error = 0;
774 >14121*10^7 : rcu_read_lock();
775 >14091*10^7 : ip = radix_tree_lookup(&pag->pag_ici_root, agino);
776 :
777 >14085*10^7 : if (ip) {
778 >14041*10^7 : error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
779 >13983*10^7 : if (error)
780 8263559 : goto out_error_or_again;
781 : } else {
782 445429412 : rcu_read_unlock();
783 445442774 : if (flags & XFS_IGET_INCORE) {
784 0 : error = -ENODATA;
785 0 : goto out_error_or_again;
786 : }
787 445442774 : XFS_STATS_INC(mp, xs_ig_missed);
788 :
789 445064706 : error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
790 : flags, lock_flags);
791 445059621 : if (error)
792 816343 : goto out_error_or_again;
793 : }
794 >14027*10^7 : xfs_perag_put(pag);
795 :
796 >14133*10^7 : *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 >28259*10^7 : if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
804 755186052 : xfs_setup_existing_inode(ip);
805 : return 0;
806 :
807 9079902 : out_error_or_again:
808 9079902 : if (!(flags & (XFS_IGET_INCORE | XFS_IGET_NORETRY)) &&
809 : error == -EAGAIN) {
810 1272362 : delay(1);
811 1265566 : goto again;
812 : }
813 7807540 : xfs_perag_put(pag);
814 7807540 : return error;
815 : }
816 :
817 : /* Get a metadata inode. The ftype must match exactly. */
818 : int
819 1578493 : xfs_imeta_iget(
820 : struct xfs_mount *mp,
821 : xfs_ino_t ino,
822 : unsigned char ftype,
823 : struct xfs_inode **ipp)
824 : {
825 1578493 : struct xfs_inode *ip;
826 1578493 : int error;
827 :
828 1578493 : ASSERT(ftype != XFS_DIR3_FT_UNKNOWN);
829 :
830 1578493 : error = xfs_iget(mp, NULL, ino, XFS_IGET_UNTRUSTED, 0, &ip);
831 1578493 : if (error == -EFSCORRUPTED)
832 21 : goto whine;
833 1578472 : if (error)
834 : return error;
835 :
836 1578472 : if (VFS_I(ip)->i_nlink == 0)
837 0 : goto bad_rele;
838 1578472 : if (xfs_mode_to_ftype(VFS_I(ip)->i_mode) != ftype)
839 0 : goto bad_rele;
840 1578472 : if (xfs_has_metadir(mp) && !xfs_is_metadir_inode(ip))
841 4 : goto bad_rele;
842 :
843 1578468 : *ipp = ip;
844 1578468 : return 0;
845 4 : bad_rele:
846 4 : xfs_irele(ip);
847 25 : whine:
848 25 : xfs_err(mp, "metadata inode 0x%llx is corrupt", ino);
849 25 : xfs_fs_mark_sick(mp, XFS_SICK_FS_METADIR);
850 25 : 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 474328903 : xfs_reclaim_igrab(
872 : struct xfs_inode *ip,
873 : struct xfs_icwalk *icw)
874 : {
875 474328903 : ASSERT(rcu_read_lock_held());
876 :
877 474328903 : spin_lock(&ip->i_flags_lock);
878 474329080 : if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
879 : __xfs_iflags_test(ip, XFS_IRECLAIM)) {
880 : /* not a reclaim candidate. */
881 2281 : spin_unlock(&ip->i_flags_lock);
882 2281 : return false;
883 : }
884 :
885 : /* Don't reclaim a sick inode unless the caller asked for it. */
886 474326799 : if (ip->i_sick &&
887 27908 : (!icw || !(icw->icw_flags & XFS_ICWALK_FLAG_RECLAIM_SICK))) {
888 0 : spin_unlock(&ip->i_flags_lock);
889 0 : return false;
890 : }
891 :
892 474326799 : __xfs_iflags_set(ip, XFS_IRECLAIM);
893 474326799 : spin_unlock(&ip->i_flags_lock);
894 474326799 : 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 474326811 : xfs_reclaim_inode(
911 : struct xfs_inode *ip,
912 : struct xfs_perag *pag)
913 : {
914 474326811 : xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
915 :
916 474326811 : if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
917 7107 : goto out;
918 474319583 : if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING))
919 8731308 : 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 931176724 : if (xlog_is_shutdown(ip->i_mount->m_log)) {
929 325411208 : xfs_iunpin_wait(ip);
930 325411207 : xfs_iflush_shutdown_abort(ip);
931 325411208 : goto reclaim;
932 : }
933 140177154 : if (xfs_ipincount(ip))
934 7355521 : goto out_clear_flush;
935 132821633 : if (!xfs_inode_clean(ip))
936 13127200 : goto out_clear_flush;
937 :
938 119694433 : xfs_iflags_clear(ip, XFS_IFLUSHING);
939 445105660 : reclaim:
940 445105660 : 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 445105627 : spin_lock(&ip->i_flags_lock);
953 445105693 : ip->i_flags = XFS_IRECLAIM;
954 445105693 : ip->i_ino = 0;
955 445105693 : ip->i_sick = 0;
956 445105693 : ip->i_checked = 0;
957 445105693 : spin_unlock(&ip->i_flags_lock);
958 :
959 445105720 : ASSERT(!ip->i_itemp || ip->i_itemp->ili_item.li_buf == NULL);
960 445105720 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
961 :
962 445105676 : 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 445105675 : spin_lock(&pag->pag_ici_lock);
971 890211468 : if (!xfs_is_shutdown(pag->pag_mount)) {
972 : /* had better not be on any unlinked list! */
973 119694527 : ASSERT(!xfs_inode_on_unlinked_list(ip));
974 119694527 : 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 445105714 : if (!radix_tree_delete(&pag->pag_ici_root,
978 445105734 : XFS_INO_TO_AGINO(ip->i_mount, ino)))
979 0 : ASSERT(0);
980 445105714 : xfs_perag_clear_inode_tag(pag, NULLAGINO, XFS_ICI_RECLAIM_TAG);
981 445105623 : 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 445105684 : xfs_ilock(ip, XFS_ILOCK_EXCL);
992 445105708 : ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot);
993 445105708 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
994 529262387 : ASSERT(xfs_inode_clean(ip));
995 :
996 445105653 : __xfs_inode_free(ip);
997 445105653 : return;
998 :
999 20482721 : out_clear_flush:
1000 20482721 : xfs_iflags_clear(ip, XFS_IFLUSHING);
1001 29214029 : out_iunlock:
1002 29214029 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
1003 29221135 : out:
1004 29221135 : 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 106910 : xfs_want_reclaim_sick(
1010 : struct xfs_mount *mp)
1011 : {
1012 253905 : return xfs_is_unmounting(mp) || xfs_has_norecovery(mp) ||
1013 : xfs_is_shutdown(mp);
1014 : }
1015 :
1016 : void
1017 66825 : xfs_reclaim_inodes(
1018 : struct xfs_mount *mp)
1019 : {
1020 66825 : struct xfs_icwalk icw = {
1021 : .icw_flags = 0,
1022 : };
1023 :
1024 66825 : if (xfs_want_reclaim_sick(mp))
1025 66825 : icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1026 :
1027 397328 : while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
1028 330503 : xfs_ail_push_all_sync(mp->m_ail);
1029 330503 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1030 : }
1031 66825 : }
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 40085 : xfs_reclaim_inodes_nr(
1042 : struct xfs_mount *mp,
1043 : unsigned long nr_to_scan)
1044 : {
1045 40085 : struct xfs_icwalk icw = {
1046 : .icw_flags = XFS_ICWALK_FLAG_SCAN_LIMIT,
1047 40085 : .icw_scan_limit = min_t(unsigned long, LONG_MAX, nr_to_scan),
1048 : };
1049 :
1050 40085 : if (xfs_want_reclaim_sick(mp))
1051 9 : icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1052 :
1053 : /* kick background reclaimer and push the AIL */
1054 40085 : xfs_reclaim_work_queue(mp);
1055 40085 : xfs_ail_push_all(mp->m_ail);
1056 :
1057 40085 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1058 40085 : 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 386465 : xfs_reclaim_inodes_count(
1067 : struct xfs_mount *mp)
1068 : {
1069 386465 : struct xfs_perag *pag;
1070 386465 : xfs_agnumber_t ag = 0;
1071 386465 : long reclaimable = 0;
1072 :
1073 807449 : while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1074 420984 : ag = pag->pag_agno + 1;
1075 420984 : reclaimable += pag->pag_ici_reclaimable;
1076 420984 : xfs_perag_put(pag);
1077 : }
1078 386465 : return reclaimable;
1079 : }
1080 :
1081 : STATIC bool
1082 1908811 : xfs_icwalk_match_id(
1083 : struct xfs_inode *ip,
1084 : struct xfs_icwalk *icw)
1085 : {
1086 1908811 : if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1087 : !uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1088 : return false;
1089 :
1090 1908811 : if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) &&
1091 : !gid_eq(VFS_I(ip)->i_gid, icw->icw_gid))
1092 : return false;
1093 :
1094 1908811 : 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 3613 : xfs_icwalk_match_id_union(
1107 : struct xfs_inode *ip,
1108 : struct xfs_icwalk *icw)
1109 : {
1110 3613 : if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1111 : uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1112 : return true;
1113 :
1114 804 : 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 4401598 : xfs_icwalk_match(
1132 : struct xfs_inode *ip,
1133 : struct xfs_icwalk *icw)
1134 : {
1135 4401598 : bool match;
1136 :
1137 4401598 : if (!icw)
1138 : return true;
1139 :
1140 1901001 : if (icw->icw_flags & XFS_ICWALK_FLAG_UNION)
1141 3612 : match = xfs_icwalk_match_id_union(ip, icw);
1142 : else
1143 1897389 : match = xfs_icwalk_match_id(ip, icw);
1144 1901001 : if (!match)
1145 : return false;
1146 :
1147 : /* skip the inode if the file size is too small */
1148 1896569 : 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 83906 : xfs_reclaim_worker(
1163 : struct work_struct *work)
1164 : {
1165 83906 : struct xfs_mount *mp = container_of(to_delayed_work(work),
1166 : struct xfs_mount, m_reclaim_work);
1167 :
1168 83906 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, NULL);
1169 83906 : xfs_reclaim_work_queue(mp);
1170 83906 : }
1171 :
1172 : STATIC int
1173 35882413 : xfs_inode_free_eofblocks(
1174 : struct xfs_inode *ip,
1175 : struct xfs_icwalk *icw,
1176 : unsigned int *lockflags)
1177 : {
1178 35882413 : bool wait;
1179 :
1180 35882413 : wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1181 :
1182 72075148 : 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 7619008 : if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1190 : return 0;
1191 :
1192 2821475 : 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 2821475 : if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1200 1405016 : if (wait)
1201 : return -EAGAIN;
1202 190954 : return 0;
1203 : }
1204 1445445 : *lockflags |= XFS_IOLOCK_EXCL;
1205 :
1206 1445445 : if (xfs_can_free_eofblocks(ip, false))
1207 593709 : return xfs_free_eofblocks(ip);
1208 :
1209 : /* inode could be preallocated or append-only */
1210 851631 : trace_xfs_inode_free_eofblocks_invalid(ip);
1211 851582 : xfs_inode_clear_eofblocks_tag(ip);
1212 851582 : return 0;
1213 : }
1214 :
1215 : static void
1216 13934396 : xfs_blockgc_set_iflag(
1217 : struct xfs_inode *ip,
1218 : unsigned long iflag)
1219 : {
1220 13934396 : struct xfs_mount *mp = ip->i_mount;
1221 13934396 : struct xfs_perag *pag;
1222 :
1223 13934396 : 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 13934396 : if (ip->i_flags & iflag)
1230 : return;
1231 6455625 : spin_lock(&ip->i_flags_lock);
1232 6466738 : ip->i_flags |= iflag;
1233 6466738 : spin_unlock(&ip->i_flags_lock);
1234 :
1235 6465248 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1236 6475478 : spin_lock(&pag->pag_ici_lock);
1237 :
1238 6475161 : xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1239 : XFS_ICI_BLOCKGC_TAG);
1240 :
1241 6463299 : spin_unlock(&pag->pag_ici_lock);
1242 6461622 : xfs_perag_put(pag);
1243 : }
1244 :
1245 : void
1246 6415246 : xfs_inode_set_eofblocks_tag(
1247 : xfs_inode_t *ip)
1248 : {
1249 6415246 : trace_xfs_inode_set_eofblocks_tag(ip);
1250 6414973 : return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS);
1251 : }
1252 :
1253 : static void
1254 32750332 : xfs_blockgc_clear_iflag(
1255 : struct xfs_inode *ip,
1256 : unsigned long iflag)
1257 : {
1258 32750332 : struct xfs_mount *mp = ip->i_mount;
1259 32750332 : struct xfs_perag *pag;
1260 32750332 : bool clear_tag;
1261 :
1262 32750332 : ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
1263 :
1264 32750332 : spin_lock(&ip->i_flags_lock);
1265 32832685 : ip->i_flags &= ~iflag;
1266 32832685 : clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0;
1267 32832685 : spin_unlock(&ip->i_flags_lock);
1268 :
1269 32823605 : if (!clear_tag)
1270 : return;
1271 :
1272 25224724 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1273 25243794 : spin_lock(&pag->pag_ici_lock);
1274 :
1275 25244528 : xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1276 : XFS_ICI_BLOCKGC_TAG);
1277 :
1278 25224399 : spin_unlock(&pag->pag_ici_lock);
1279 25253547 : xfs_perag_put(pag);
1280 : }
1281 :
1282 : void
1283 22135972 : xfs_inode_clear_eofblocks_tag(
1284 : xfs_inode_t *ip)
1285 : {
1286 22135972 : trace_xfs_inode_clear_eofblocks_tag(ip);
1287 22078193 : 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 29500476 : 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 59000952 : if (!xfs_inode_has_cow_data(ip)) {
1304 282923 : trace_xfs_inode_free_cowblocks_invalid(ip);
1305 282493 : xfs_inode_clear_cowblocks_tag(ip);
1306 282493 : 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 32379226 : if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1314 6322948 : mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1315 5646893 : mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1316 : atomic_read(&VFS_I(ip)->i_dio_count))
1317 26735202 : 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 34845248 : xfs_inode_free_cowblocks(
1336 : struct xfs_inode *ip,
1337 : struct xfs_icwalk *icw,
1338 : unsigned int *lockflags)
1339 : {
1340 34845248 : bool wait;
1341 34845248 : int ret = 0;
1342 :
1343 34845248 : wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1344 :
1345 69950333 : if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS))
1346 : return 0;
1347 :
1348 28643059 : if (!xfs_prep_free_cowblocks(ip))
1349 : return 0;
1350 :
1351 1556144 : 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 3058908 : if (!(*lockflags & XFS_IOLOCK_EXCL) &&
1359 1501612 : !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1360 630801 : if (wait)
1361 : return -EAGAIN;
1362 111793 : return 0;
1363 : }
1364 926495 : *lockflags |= XFS_IOLOCK_EXCL;
1365 :
1366 926495 : if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) {
1367 39 : if (wait)
1368 : return -EAGAIN;
1369 37 : return 0;
1370 : }
1371 926648 : *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 926648 : if (xfs_prep_free_cowblocks(ip))
1378 926589 : ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
1379 : return ret;
1380 : }
1381 :
1382 : void
1383 7523392 : xfs_inode_set_cowblocks_tag(
1384 : xfs_inode_t *ip)
1385 : {
1386 7523392 : trace_xfs_inode_set_cowblocks_tag(ip);
1387 7518101 : return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS);
1388 : }
1389 :
1390 : void
1391 10654684 : xfs_inode_clear_cowblocks_tag(
1392 : xfs_inode_t *ip)
1393 : {
1394 10654684 : trace_xfs_inode_clear_cowblocks_tag(ip);
1395 10652670 : return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS);
1396 : }
1397 :
1398 : /* Disable post-EOF and CoW block auto-reclamation. */
1399 : void
1400 136724 : xfs_blockgc_stop(
1401 : struct xfs_mount *mp)
1402 : {
1403 136724 : struct xfs_perag *pag;
1404 136724 : xfs_agnumber_t agno;
1405 :
1406 136724 : if (!xfs_clear_blockgc_enabled(mp))
1407 71 : return;
1408 :
1409 977439 : for_each_perag(mp, agno, pag)
1410 840786 : cancel_delayed_work_sync(&pag->pag_blockgc_work);
1411 136653 : trace_xfs_blockgc_stop(mp, __return_address);
1412 : }
1413 :
1414 : /* Enable post-EOF and CoW block auto-reclamation. */
1415 : void
1416 136842 : xfs_blockgc_start(
1417 : struct xfs_mount *mp)
1418 : {
1419 136842 : struct xfs_perag *pag;
1420 136842 : xfs_agnumber_t agno;
1421 :
1422 136842 : if (xfs_set_blockgc_enabled(mp))
1423 : return;
1424 :
1425 136820 : trace_xfs_blockgc_start(mp, __return_address);
1426 351662 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1427 214842 : 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 36461558 : xfs_blockgc_igrab(
1443 : struct xfs_inode *ip)
1444 : {
1445 36461558 : struct inode *inode = VFS_I(ip);
1446 :
1447 36461558 : ASSERT(rcu_read_lock_held());
1448 :
1449 : /* Check for stale RCU freed inode */
1450 36461558 : spin_lock(&ip->i_flags_lock);
1451 36733512 : if (!ip->i_ino)
1452 0 : goto out_unlock_noent;
1453 :
1454 36733512 : if (ip->i_flags & XFS_BLOCKGC_NOGRAB_IFLAGS)
1455 520830 : goto out_unlock_noent;
1456 36212682 : spin_unlock(&ip->i_flags_lock);
1457 :
1458 : /* nothing to sync during shutdown */
1459 72232802 : 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 36099090 : if (!igrab(inode))
1464 3972 : return false;
1465 :
1466 : /* inode is valid */
1467 : return true;
1468 :
1469 520830 : out_unlock_noent:
1470 520830 : spin_unlock(&ip->i_flags_lock);
1471 520830 : return false;
1472 : }
1473 :
1474 : /* Scan one incore inode for block preallocations that we can remove. */
1475 : static int
1476 35914745 : xfs_blockgc_scan_inode(
1477 : struct xfs_inode *ip,
1478 : struct xfs_icwalk *icw)
1479 : {
1480 35914745 : unsigned int lockflags = 0;
1481 35914745 : int error;
1482 :
1483 35914745 : error = xfs_inode_free_eofblocks(ip, icw, &lockflags);
1484 36138999 : if (error)
1485 1212531 : goto unlock;
1486 :
1487 34926468 : error = xfs_inode_free_cowblocks(ip, icw, &lockflags);
1488 36111592 : unlock:
1489 36111592 : if (lockflags)
1490 2317196 : xfs_iunlock(ip, lockflags);
1491 36110941 : xfs_irele(ip);
1492 36135193 : return error;
1493 : }
1494 :
1495 : /* Background worker that trims preallocated space. */
1496 : void
1497 4961915 : xfs_blockgc_worker(
1498 : struct work_struct *work)
1499 : {
1500 4961915 : struct xfs_perag *pag = container_of(to_delayed_work(work),
1501 : struct xfs_perag, pag_blockgc_work);
1502 4961915 : struct xfs_mount *mp = pag->pag_mount;
1503 4961915 : int error;
1504 :
1505 4961915 : trace_xfs_blockgc_worker(mp, __return_address);
1506 :
1507 4961910 : error = xfs_icwalk_ag(pag, XFS_ICWALK_BLOCKGC, NULL);
1508 4960802 : if (error)
1509 0 : xfs_info(mp, "AG %u preallocation gc worker failed, err=%d",
1510 : pag->pag_agno, error);
1511 4960802 : xfs_blockgc_queue(pag);
1512 4960008 : }
1513 :
1514 : /*
1515 : * Try to free space in the filesystem by purging inactive inodes, eofblocks
1516 : * and cowblocks.
1517 : */
1518 : int
1519 1687319 : xfs_blockgc_free_space(
1520 : struct xfs_mount *mp,
1521 : struct xfs_icwalk *icw)
1522 : {
1523 1687319 : int error;
1524 :
1525 1687319 : trace_xfs_blockgc_free_space(mp, icw, _RET_IP_);
1526 :
1527 1686803 : error = xfs_icwalk(mp, XFS_ICWALK_BLOCKGC, icw);
1528 1686168 : if (error)
1529 : return error;
1530 :
1531 1686138 : 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 12654503 : xfs_blockgc_flush_all(
1540 : struct xfs_mount *mp)
1541 : {
1542 12654503 : struct xfs_perag *pag;
1543 12654503 : xfs_agnumber_t agno;
1544 :
1545 12654503 : 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 17062176 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1553 4413977 : mod_delayed_work(pag->pag_mount->m_blockgc_wq,
1554 : &pag->pag_blockgc_work, 0);
1555 :
1556 17028634 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1557 4375215 : flush_delayed_work(&pag->pag_blockgc_work);
1558 :
1559 12656644 : 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 36002 : 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 36002 : struct xfs_icwalk icw = {0};
1581 36002 : bool do_work = false;
1582 :
1583 36002 : 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 36002 : icw.icw_flags = XFS_ICWALK_FLAG_UNION | iwalk_flags;
1591 :
1592 36002 : if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) {
1593 21263 : icw.icw_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id);
1594 21263 : icw.icw_flags |= XFS_ICWALK_FLAG_UID;
1595 21263 : do_work = true;
1596 : }
1597 :
1598 36002 : if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) {
1599 34589 : icw.icw_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id);
1600 34589 : icw.icw_flags |= XFS_ICWALK_FLAG_GID;
1601 34589 : do_work = true;
1602 : }
1603 :
1604 36002 : if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) {
1605 30638 : icw.icw_prid = pdqp->q_id;
1606 30638 : icw.icw_flags |= XFS_ICWALK_FLAG_PRID;
1607 30638 : do_work = true;
1608 : }
1609 :
1610 36002 : if (!do_work)
1611 : return 0;
1612 :
1613 35173 : return xfs_blockgc_free_space(mp, &icw);
1614 : }
1615 :
1616 : /* Run cow/eofblocks scans on the quotas attached to the inode. */
1617 : int
1618 16887 : xfs_blockgc_free_quota(
1619 : struct xfs_inode *ip,
1620 : unsigned int iwalk_flags)
1621 : {
1622 16887 : 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 510818591 : xfs_icwalk_igrab(
1645 : enum xfs_icwalk_goal goal,
1646 : struct xfs_inode *ip,
1647 : struct xfs_icwalk *icw)
1648 : {
1649 510818591 : switch (goal) {
1650 36489670 : case XFS_ICWALK_BLOCKGC:
1651 36489670 : return xfs_blockgc_igrab(ip);
1652 474328921 : case XFS_ICWALK_RECLAIM:
1653 474328921 : 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 510290671 : 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 510290671 : int error = 0;
1671 :
1672 510290671 : switch (goal) {
1673 35963855 : case XFS_ICWALK_BLOCKGC:
1674 35963855 : error = xfs_blockgc_scan_inode(ip, icw);
1675 35963855 : break;
1676 474326816 : case XFS_ICWALK_RECLAIM:
1677 474326816 : xfs_reclaim_inode(ip, pag);
1678 474326816 : break;
1679 : }
1680 510416254 : 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 6327553 : xfs_icwalk_ag(
1689 : struct xfs_perag *pag,
1690 : enum xfs_icwalk_goal goal,
1691 : struct xfs_icwalk *icw)
1692 : {
1693 6327553 : struct xfs_mount *mp = pag->pag_mount;
1694 6327553 : uint32_t first_index;
1695 6327553 : int last_error = 0;
1696 7489800 : int skipped;
1697 7489800 : bool done;
1698 7489800 : int nr_found;
1699 :
1700 7489800 : restart:
1701 7489800 : done = false;
1702 7489800 : skipped = 0;
1703 7489800 : if (goal == XFS_ICWALK_RECLAIM)
1704 680479 : first_index = READ_ONCE(pag->pag_ici_reclaim_cursor);
1705 : else
1706 : first_index = 0;
1707 : nr_found = 0;
1708 29818088 : do {
1709 29818088 : struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1710 29818088 : int error = 0;
1711 29818088 : int i;
1712 :
1713 29818088 : rcu_read_lock();
1714 :
1715 29799120 : nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root,
1716 : (void **) batch, first_index,
1717 : XFS_LOOKUP_BATCH, goal);
1718 29789713 : if (!nr_found) {
1719 7355134 : done = true;
1720 7355134 : rcu_read_unlock();
1721 7477586 : 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 533349024 : for (i = 0; i < nr_found; i++) {
1729 510905967 : struct xfs_inode *ip = batch[i];
1730 :
1731 510863314 : if (done || !xfs_icwalk_igrab(goal, ip, icw))
1732 520262 : 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 510914445 : if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
1747 114 : continue;
1748 510914331 : first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1749 510914331 : if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1750 3 : done = true;
1751 : }
1752 :
1753 : /* unlock now we've grabbed the inodes. */
1754 22443057 : rcu_read_unlock();
1755 :
1756 555833986 : for (i = 0; i < nr_found; i++) {
1757 510962316 : if (!batch[i])
1758 530822 : continue;
1759 510370300 : error = xfs_icwalk_process_inode(goal, batch[i], pag,
1760 : icw);
1761 510418450 : if (error == -EAGAIN) {
1762 1718080 : skipped++;
1763 1718080 : continue;
1764 : }
1765 508700370 : if (error && last_error != -EFSCORRUPTED)
1766 0 : last_error = error;
1767 : }
1768 :
1769 : /* bail out if the filesystem is corrupted. */
1770 22428613 : if (error == -EFSCORRUPTED)
1771 : break;
1772 :
1773 22428613 : cond_resched();
1774 :
1775 22436472 : if (icw && (icw->icw_flags & XFS_ICWALK_FLAG_SCAN_LIMIT)) {
1776 373125 : icw->icw_scan_limit -= XFS_LOOKUP_BATCH;
1777 373125 : if (icw->icw_scan_limit <= 0)
1778 : break;
1779 : }
1780 22328291 : } while (nr_found && !done);
1781 :
1782 7477589 : if (goal == XFS_ICWALK_RECLAIM) {
1783 680479 : if (done)
1784 572298 : first_index = 0;
1785 680479 : WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index);
1786 : }
1787 :
1788 7477589 : if (skipped) {
1789 1152024 : delay(1);
1790 1162247 : goto restart;
1791 : }
1792 6325565 : return last_error;
1793 : }
1794 :
1795 : /* Walk all incore inodes to achieve a given goal. */
1796 : static int
1797 2141338 : xfs_icwalk(
1798 : struct xfs_mount *mp,
1799 : enum xfs_icwalk_goal goal,
1800 : struct xfs_icwalk *icw)
1801 : {
1802 2141338 : struct xfs_perag *pag;
1803 2141338 : int error = 0;
1804 2141338 : int last_error = 0;
1805 2141338 : xfs_agnumber_t agno;
1806 :
1807 3506793 : for_each_perag_tag(mp, agno, pag, goal) {
1808 1366249 : error = xfs_icwalk_ag(pag, goal, icw);
1809 1365455 : 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 2141158 : 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 879573715 : xfs_inodegc_set_reclaimable(
1850 : struct xfs_inode *ip)
1851 : {
1852 879573715 : struct xfs_mount *mp = ip->i_mount;
1853 879573715 : struct xfs_perag *pag;
1854 :
1855 1759147430 : 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 879573715 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1862 879712170 : spin_lock(&pag->pag_ici_lock);
1863 879755263 : spin_lock(&ip->i_flags_lock);
1864 :
1865 1759529114 : if (!xfs_is_shutdown(pag->pag_mount)) {
1866 : /* had better not be on any unlinked list! */
1867 556108633 : ASSERT(!xfs_inode_on_unlinked_list(ip));
1868 556108633 : 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 879764557 : trace_xfs_inode_set_reclaimable(ip);
1873 879579790 : ip->i_flags &= ~(XFS_NEED_INACTIVE | XFS_INACTIVATING);
1874 879579790 : ip->i_flags |= XFS_IRECLAIMABLE;
1875 879579790 : xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1876 : XFS_ICI_RECLAIM_TAG);
1877 :
1878 879605713 : spin_unlock(&ip->i_flags_lock);
1879 879716519 : spin_unlock(&pag->pag_ici_lock);
1880 879714549 : xfs_perag_put(pag);
1881 879704888 : }
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 75824412 : xfs_inodegc_inactivate(
1890 : struct xfs_inode *ip)
1891 : {
1892 75824412 : int error;
1893 :
1894 75824412 : trace_xfs_inode_inactivating(ip);
1895 75696781 : error = xfs_inactive(ip);
1896 75975146 : xfs_inodegc_set_reclaimable(ip);
1897 75927580 : return error;
1898 :
1899 : }
1900 :
1901 : void
1902 11963352 : xfs_inodegc_worker(
1903 : struct work_struct *work)
1904 : {
1905 11963352 : struct xfs_inodegc *gc = container_of(to_delayed_work(work),
1906 : struct xfs_inodegc, work);
1907 11963352 : struct llist_node *node = llist_del_all(&gc->list);
1908 11998999 : struct xfs_inode *ip, *n;
1909 11998999 : unsigned int nofs_flag;
1910 :
1911 11998999 : ASSERT(gc->cpu == smp_processor_id());
1912 :
1913 11993306 : WRITE_ONCE(gc->items, 0);
1914 :
1915 11993306 : 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 11989132 : nofs_flag = memalloc_nofs_save();
1924 :
1925 11989132 : ip = llist_entry(node, struct xfs_inode, i_gclist);
1926 11989132 : trace_xfs_inodegc_worker(ip->i_mount, READ_ONCE(gc->shrinker_hits));
1927 :
1928 11979014 : WRITE_ONCE(gc->shrinker_hits, 0);
1929 87887070 : llist_for_each_entry_safe(ip, n, node, i_gclist) {
1930 75870078 : int error;
1931 :
1932 75870078 : xfs_iflags_set(ip, XFS_INACTIVATING);
1933 75817003 : error = xfs_inodegc_inactivate(ip);
1934 75908056 : if (error && !gc->error)
1935 2271 : gc->error = error;
1936 : }
1937 :
1938 12016992 : 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 46682112 : xfs_inodegc_push(
1947 : struct xfs_mount *mp)
1948 : {
1949 93364224 : if (!xfs_is_inodegc_enabled(mp))
1950 : return;
1951 46612121 : trace_xfs_inodegc_push(mp, __return_address);
1952 46582088 : 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 16946721 : xfs_inodegc_flush(
1961 : struct xfs_mount *mp)
1962 : {
1963 16946721 : xfs_inodegc_push(mp);
1964 16955376 : trace_xfs_inodegc_flush(mp, __return_address);
1965 16950526 : 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 136902 : xfs_inodegc_stop(
1975 : struct xfs_mount *mp)
1976 : {
1977 136902 : bool rerun;
1978 :
1979 136902 : 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 136831 : xfs_inodegc_queue_all(mp);
1993 136831 : do {
1994 136831 : flush_workqueue(mp->m_inodegc_wq);
1995 136831 : rerun = xfs_inodegc_queue_all(mp);
1996 136831 : } while (rerun);
1997 :
1998 136831 : 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 136842 : xfs_inodegc_start(
2008 : struct xfs_mount *mp)
2009 : {
2010 136842 : if (xfs_set_inodegc_enabled(mp))
2011 : return;
2012 :
2013 136820 : trace_xfs_inodegc_start(mp, __return_address);
2014 136820 : xfs_inodegc_queue_all(mp);
2015 : }
2016 :
2017 : #ifdef CONFIG_XFS_RT
2018 : static inline bool
2019 59856979 : xfs_inodegc_want_queue_rt_file(
2020 : struct xfs_inode *ip)
2021 : {
2022 59856979 : struct xfs_mount *mp = ip->i_mount;
2023 :
2024 59856979 : if (!XFS_IS_REALTIME_INODE(ip))
2025 : return false;
2026 :
2027 8040559 : if (__percpu_counter_compare(&mp->m_frextents,
2028 8040630 : mp->m_low_rtexts[XFS_LOWSP_5_PCNT],
2029 : XFS_FDBLOCKS_BATCH) < 0)
2030 196295 : 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 75956492 : xfs_inodegc_want_queue_work(
2047 : struct xfs_inode *ip,
2048 : unsigned int items)
2049 : {
2050 75956492 : struct xfs_mount *mp = ip->i_mount;
2051 :
2052 75956492 : if (items > mp->m_ino_geo.inodes_per_cluster)
2053 : return true;
2054 :
2055 60407718 : if (__percpu_counter_compare(&mp->m_fdblocks,
2056 60414857 : mp->m_low_space[XFS_LOWSP_5_PCNT],
2057 : XFS_FDBLOCKS_BATCH) < 0)
2058 : return true;
2059 :
2060 59859943 : if (xfs_inodegc_want_queue_rt_file(ip))
2061 : return true;
2062 :
2063 59656671 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_USER))
2064 : return true;
2065 :
2066 59632973 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_GROUP))
2067 : return true;
2068 :
2069 59650832 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_PROJ))
2070 213 : 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 75911120 : if (current->journal_info)
2097 : return false;
2098 :
2099 75833624 : if (shrinker_hits > 0)
2100 : return true;
2101 :
2102 75833622 : 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 75928439 : xfs_inodegc_queue(
2115 : struct xfs_inode *ip)
2116 : {
2117 75928439 : struct xfs_mount *mp = ip->i_mount;
2118 75928439 : struct xfs_inodegc *gc;
2119 75928439 : int items;
2120 75928439 : unsigned int shrinker_hits;
2121 75928439 : unsigned long queue_delay = 1;
2122 :
2123 75928439 : trace_xfs_inode_set_need_inactive(ip);
2124 75902193 : spin_lock(&ip->i_flags_lock);
2125 75975304 : ip->i_flags |= XFS_NEED_INACTIVE;
2126 75975304 : spin_unlock(&ip->i_flags_lock);
2127 :
2128 75986529 : gc = get_cpu_ptr(mp->m_inodegc);
2129 75967415 : llist_add(&ip->i_gclist, &gc->list);
2130 75969675 : items = READ_ONCE(gc->items);
2131 75969675 : WRITE_ONCE(gc->items, items + 1);
2132 75969675 : 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 151939350 : if (!xfs_is_inodegc_enabled(mp)) {
2139 0 : put_cpu_ptr(gc);
2140 0 : return;
2141 : }
2142 :
2143 75969675 : if (xfs_inodegc_want_queue_work(ip, items))
2144 16291802 : queue_delay = 0;
2145 :
2146 75936138 : trace_xfs_inodegc_queue(mp, __return_address);
2147 75924090 : mod_delayed_work_on(current_cpu(), mp->m_inodegc_wq, &gc->work,
2148 : queue_delay);
2149 75872587 : put_cpu_ptr(gc);
2150 :
2151 75911120 : if (xfs_inodegc_want_flush_work(ip, items, shrinker_hits)) {
2152 5749308 : trace_xfs_inodegc_throttle(mp, __return_address);
2153 5744478 : 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 158 : xfs_inodegc_cpu_dead(
2162 : struct xfs_mount *mp,
2163 : unsigned int dead_cpu)
2164 : {
2165 158 : struct xfs_inodegc *dead_gc, *gc;
2166 158 : struct llist_node *first, *last;
2167 158 : unsigned int count = 0;
2168 :
2169 158 : dead_gc = per_cpu_ptr(mp->m_inodegc, dead_cpu);
2170 158 : cancel_delayed_work_sync(&dead_gc->work);
2171 :
2172 158 : 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 879675307 : xfs_inode_mark_reclaimable(
2210 : struct xfs_inode *ip)
2211 : {
2212 879675307 : struct xfs_mount *mp = ip->i_mount;
2213 879675307 : bool need_inactive;
2214 :
2215 879675307 : XFS_STATS_INC(mp, vn_reclaim);
2216 :
2217 : /*
2218 : * We should never get here with any of the reclaim flags already set.
2219 : */
2220 1759370332 : ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_ALL_IRECLAIM_FLAGS));
2221 :
2222 879744819 : need_inactive = xfs_inode_needs_inactive(ip);
2223 879498492 : if (need_inactive) {
2224 75899130 : xfs_inodegc_queue(ip);
2225 75899130 : return;
2226 : }
2227 :
2228 : /* Going straight to reclaim, so drop the dquots. */
2229 803599362 : xfs_qm_dqdetach(ip);
2230 803603829 : 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 7597 : xfs_inodegc_shrinker_count(
2248 : struct shrinker *shrink,
2249 : struct shrink_control *sc)
2250 : {
2251 7597 : struct xfs_mount *mp = container_of(shrink, struct xfs_mount,
2252 : m_inodegc_shrinker);
2253 7597 : struct xfs_inodegc *gc;
2254 7597 : int cpu;
2255 :
2256 15194 : if (!xfs_is_inodegc_enabled(mp))
2257 : return 0;
2258 :
2259 37824 : for_each_online_cpu(cpu) {
2260 30288 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
2261 30288 : if (!llist_empty(&gc->list))
2262 : return XFS_INODEGC_SHRINKER_COUNT;
2263 : }
2264 :
2265 : return 0;
2266 : }
2267 :
2268 : static unsigned long
2269 32 : xfs_inodegc_shrinker_scan(
2270 : struct shrinker *shrink,
2271 : struct shrink_control *sc)
2272 : {
2273 32 : struct xfs_mount *mp = container_of(shrink, struct xfs_mount,
2274 : m_inodegc_shrinker);
2275 32 : struct xfs_inodegc *gc;
2276 32 : int cpu;
2277 32 : bool no_items = true;
2278 :
2279 64 : if (!xfs_is_inodegc_enabled(mp))
2280 : return SHRINK_STOP;
2281 :
2282 32 : trace_xfs_inodegc_shrinker_scan(mp, sc, __return_address);
2283 :
2284 192 : for_each_online_cpu(cpu) {
2285 128 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
2286 128 : if (!llist_empty(&gc->list)) {
2287 32 : unsigned int h = READ_ONCE(gc->shrinker_hits);
2288 :
2289 32 : WRITE_ONCE(gc->shrinker_hits, h + 1);
2290 32 : mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0);
2291 32 : 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 32 : 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 66856 : xfs_inodegc_register_shrinker(
2308 : struct xfs_mount *mp)
2309 : {
2310 66856 : struct shrinker *shrink = &mp->m_inodegc_shrinker;
2311 :
2312 66856 : shrink->count_objects = xfs_inodegc_shrinker_count;
2313 66856 : shrink->scan_objects = xfs_inodegc_shrinker_scan;
2314 66856 : shrink->seeks = 0;
2315 66856 : shrink->flags = SHRINKER_NONSLAB;
2316 66856 : shrink->batch = XFS_INODEGC_SHRINKER_BATCH;
2317 :
2318 66856 : return register_shrinker(shrink, "xfs-inodegc:%s", mp->m_super->s_id);
2319 : }
|
