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 :
28 : #include <linux/iversion.h>
29 :
30 : /* Radix tree tags for incore inode tree. */
31 :
32 : /* inode is to be reclaimed */
33 : #define XFS_ICI_RECLAIM_TAG 0
34 : /* Inode has speculative preallocations (posteof or cow) to clean. */
35 : #define XFS_ICI_BLOCKGC_TAG 1
36 :
37 : /*
38 : * The goal for walking incore inodes. These can correspond with incore inode
39 : * radix tree tags when convenient. Avoid existing XFS_IWALK namespace.
40 : */
41 : enum xfs_icwalk_goal {
42 : /* Goals directly associated with tagged inodes. */
43 : XFS_ICWALK_BLOCKGC = XFS_ICI_BLOCKGC_TAG,
44 : XFS_ICWALK_RECLAIM = XFS_ICI_RECLAIM_TAG,
45 : };
46 :
47 : static int xfs_icwalk(struct xfs_mount *mp,
48 : enum xfs_icwalk_goal goal, struct xfs_icwalk *icw);
49 : static int xfs_icwalk_ag(struct xfs_perag *pag,
50 : enum xfs_icwalk_goal goal, struct xfs_icwalk *icw);
51 :
52 : /*
53 : * Private inode cache walk flags for struct xfs_icwalk. Must not
54 : * coincide with XFS_ICWALK_FLAGS_VALID.
55 : */
56 :
57 : /* Stop scanning after icw_scan_limit inodes. */
58 : #define XFS_ICWALK_FLAG_SCAN_LIMIT (1U << 28)
59 :
60 : #define XFS_ICWALK_FLAG_RECLAIM_SICK (1U << 27)
61 : #define XFS_ICWALK_FLAG_UNION (1U << 26) /* union filter algorithm */
62 :
63 : #define XFS_ICWALK_PRIVATE_FLAGS (XFS_ICWALK_FLAG_SCAN_LIMIT | \
64 : XFS_ICWALK_FLAG_RECLAIM_SICK | \
65 : XFS_ICWALK_FLAG_UNION)
66 :
67 : /*
68 : * Allocate and initialise an xfs_inode.
69 : */
70 : struct xfs_inode *
71 440000898 : xfs_inode_alloc(
72 : struct xfs_mount *mp,
73 : xfs_ino_t ino)
74 : {
75 440000898 : struct xfs_inode *ip;
76 :
77 : /*
78 : * XXX: If this didn't occur in transactions, we could drop GFP_NOFAIL
79 : * and return NULL here on ENOMEM.
80 : */
81 440000898 : ip = alloc_inode_sb(mp->m_super, xfs_inode_cache, GFP_KERNEL | __GFP_NOFAIL);
82 :
83 440350090 : if (inode_init_always(mp->m_super, VFS_I(ip))) {
84 0 : kmem_cache_free(xfs_inode_cache, ip);
85 0 : return NULL;
86 : }
87 :
88 : /* VFS doesn't initialise i_mode or i_state! */
89 440371359 : VFS_I(ip)->i_mode = 0;
90 440371359 : VFS_I(ip)->i_state = 0;
91 440371359 : mapping_set_large_folios(VFS_I(ip)->i_mapping);
92 :
93 440375746 : XFS_STATS_INC(mp, vn_active);
94 439983615 : ASSERT(atomic_read(&ip->i_pincount) == 0);
95 439983615 : ASSERT(ip->i_ino == 0);
96 :
97 : /* initialise the xfs inode */
98 439983615 : ip->i_ino = ino;
99 439983615 : ip->i_mount = mp;
100 439983615 : memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
101 439983615 : ip->i_cowfp = NULL;
102 439983615 : memset(&ip->i_af, 0, sizeof(ip->i_af));
103 439983615 : ip->i_af.if_format = XFS_DINODE_FMT_EXTENTS;
104 439983615 : memset(&ip->i_df, 0, sizeof(ip->i_df));
105 439983615 : ip->i_flags = 0;
106 439983615 : ip->i_delayed_blks = 0;
107 439983615 : ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
108 439983615 : ip->i_nblocks = 0;
109 439983615 : ip->i_forkoff = 0;
110 439983615 : ip->i_sick = 0;
111 439983615 : ip->i_checked = 0;
112 439983615 : INIT_WORK(&ip->i_ioend_work, xfs_end_io);
113 439983615 : INIT_LIST_HEAD(&ip->i_ioend_list);
114 439983615 : spin_lock_init(&ip->i_ioend_lock);
115 440325282 : ip->i_next_unlinked = NULLAGINO;
116 440325282 : ip->i_prev_unlinked = NULLAGINO;
117 :
118 440325282 : return ip;
119 : }
120 :
121 : STATIC void
122 439514671 : xfs_inode_free_callback(
123 : struct rcu_head *head)
124 : {
125 439514671 : struct inode *inode = container_of(head, struct inode, i_rcu);
126 439514671 : struct xfs_inode *ip = XFS_I(inode);
127 :
128 439514671 : switch (VFS_I(ip)->i_mode & S_IFMT) {
129 282098078 : case S_IFREG:
130 : case S_IFDIR:
131 : case S_IFLNK:
132 282098078 : xfs_idestroy_fork(&ip->i_df);
133 282098078 : break;
134 : }
135 :
136 439525497 : xfs_ifork_zap_attr(ip);
137 :
138 439883299 : if (ip->i_cowfp) {
139 97073260 : xfs_idestroy_fork(ip->i_cowfp);
140 97069355 : kmem_cache_free(xfs_ifork_cache, ip->i_cowfp);
141 : }
142 439776026 : if (ip->i_itemp) {
143 57732866 : ASSERT(!test_bit(XFS_LI_IN_AIL,
144 : &ip->i_itemp->ili_item.li_flags));
145 57732866 : xfs_inode_item_destroy(ip);
146 57773738 : ip->i_itemp = NULL;
147 : }
148 :
149 439816898 : kmem_cache_free(xfs_inode_cache, ip);
150 439785164 : }
151 :
152 : static void
153 440580202 : __xfs_inode_free(
154 : struct xfs_inode *ip)
155 : {
156 : /* asserts to verify all state is correct here */
157 440580202 : ASSERT(atomic_read(&ip->i_pincount) == 0);
158 440580202 : ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list));
159 440580202 : XFS_STATS_DEC(ip->i_mount, vn_active);
160 :
161 440580234 : call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
162 440580569 : }
163 :
164 : void
165 704913 : xfs_inode_free(
166 : struct xfs_inode *ip)
167 : {
168 1409824 : ASSERT(!xfs_iflags_test(ip, XFS_IFLUSHING));
169 :
170 : /*
171 : * Because we use RCU freeing we need to ensure the inode always
172 : * appears to be reclaimed with an invalid inode number when in the
173 : * free state. The ip->i_flags_lock provides the barrier against lookup
174 : * races.
175 : */
176 704911 : spin_lock(&ip->i_flags_lock);
177 704912 : ip->i_flags = XFS_IRECLAIM;
178 704912 : ip->i_ino = 0;
179 704912 : spin_unlock(&ip->i_flags_lock);
180 :
181 704912 : __xfs_inode_free(ip);
182 704910 : }
183 :
184 : /*
185 : * Queue background inode reclaim work if there are reclaimable inodes and there
186 : * isn't reclaim work already scheduled or in progress.
187 : */
188 : static void
189 3931602 : xfs_reclaim_work_queue(
190 : struct xfs_mount *mp)
191 : {
192 :
193 3931602 : rcu_read_lock();
194 3931094 : if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
195 3922059 : queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
196 3922912 : msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
197 : }
198 3931773 : rcu_read_unlock();
199 3929744 : }
200 :
201 : /*
202 : * Background scanning to trim preallocated space. This is queued based on the
203 : * 'speculative_prealloc_lifetime' tunable (5m by default).
204 : */
205 : static inline void
206 3163122 : xfs_blockgc_queue(
207 : struct xfs_perag *pag)
208 : {
209 3163122 : struct xfs_mount *mp = pag->pag_mount;
210 :
211 6326244 : if (!xfs_is_blockgc_enabled(mp))
212 : return;
213 :
214 3163006 : rcu_read_lock();
215 3162920 : if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG))
216 3120323 : queue_delayed_work(pag->pag_mount->m_blockgc_wq,
217 : &pag->pag_blockgc_work,
218 3120338 : msecs_to_jiffies(xfs_blockgc_secs * 1000));
219 3162968 : rcu_read_unlock();
220 : }
221 :
222 : /* Set a tag on both the AG incore inode tree and the AG radix tree. */
223 : static void
224 1091685323 : xfs_perag_set_inode_tag(
225 : struct xfs_perag *pag,
226 : xfs_agino_t agino,
227 : unsigned int tag)
228 : {
229 1091685323 : struct xfs_mount *mp = pag->pag_mount;
230 1091685323 : bool was_tagged;
231 :
232 1091685323 : lockdep_assert_held(&pag->pag_ici_lock);
233 :
234 1091685323 : was_tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
235 1091765141 : radix_tree_tag_set(&pag->pag_ici_root, agino, tag);
236 :
237 1092001645 : if (tag == XFS_ICI_RECLAIM_TAG)
238 1086995575 : pag->pag_ici_reclaimable++;
239 :
240 1092001645 : if (was_tagged)
241 : return;
242 :
243 : /* propagate the tag up into the perag radix tree */
244 4282022 : spin_lock(&mp->m_perag_lock);
245 4289417 : radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, tag);
246 4289417 : spin_unlock(&mp->m_perag_lock);
247 :
248 : /* start background work */
249 4289407 : switch (tag) {
250 3855408 : case XFS_ICI_RECLAIM_TAG:
251 3855408 : xfs_reclaim_work_queue(mp);
252 3855408 : break;
253 433999 : case XFS_ICI_BLOCKGC_TAG:
254 433999 : xfs_blockgc_queue(pag);
255 433999 : break;
256 : }
257 :
258 4282064 : trace_xfs_perag_set_inode_tag(pag, _RET_IP_);
259 : }
260 :
261 : /* Clear a tag on both the AG incore inode tree and the AG radix tree. */
262 : static void
263 1106448209 : xfs_perag_clear_inode_tag(
264 : struct xfs_perag *pag,
265 : xfs_agino_t agino,
266 : unsigned int tag)
267 : {
268 1106448209 : struct xfs_mount *mp = pag->pag_mount;
269 :
270 1106448209 : lockdep_assert_held(&pag->pag_ici_lock);
271 :
272 : /*
273 : * Reclaim can signal (with a null agino) that it cleared its own tag
274 : * by removing the inode from the radix tree.
275 : */
276 1106448209 : if (agino != NULLAGINO)
277 666572524 : radix_tree_tag_clear(&pag->pag_ici_root, agino, tag);
278 : else
279 439875685 : ASSERT(tag == XFS_ICI_RECLAIM_TAG);
280 :
281 1106388970 : if (tag == XFS_ICI_RECLAIM_TAG)
282 1087385245 : pag->pag_ici_reclaimable--;
283 :
284 1106388970 : if (radix_tree_tagged(&pag->pag_ici_root, tag))
285 : return;
286 :
287 : /* clear the tag from the perag radix tree */
288 15684093 : spin_lock(&mp->m_perag_lock);
289 15751171 : radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, tag);
290 15751171 : spin_unlock(&mp->m_perag_lock);
291 :
292 15750442 : trace_xfs_perag_clear_inode_tag(pag, _RET_IP_);
293 : }
294 :
295 : /*
296 : * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
297 : * part of the structure. This is made more complex by the fact we store
298 : * information about the on-disk values in the VFS inode and so we can't just
299 : * overwrite the values unconditionally. Hence we save the parameters we
300 : * need to retain across reinitialisation, and rewrite them into the VFS inode
301 : * after reinitialisation even if it fails.
302 : */
303 : static int
304 647263808 : xfs_reinit_inode(
305 : struct xfs_mount *mp,
306 : struct inode *inode)
307 : {
308 647263808 : int error;
309 647263808 : uint32_t nlink = inode->i_nlink;
310 647263808 : uint32_t generation = inode->i_generation;
311 647263808 : uint64_t version = inode_peek_iversion(inode);
312 647263808 : umode_t mode = inode->i_mode;
313 647263808 : dev_t dev = inode->i_rdev;
314 647263808 : kuid_t uid = inode->i_uid;
315 647263808 : kgid_t gid = inode->i_gid;
316 :
317 647263808 : error = inode_init_always(mp->m_super, inode);
318 :
319 647132291 : set_nlink(inode, nlink);
320 647055063 : inode->i_generation = generation;
321 647055063 : inode_set_iversion_queried(inode, version);
322 647055063 : inode->i_mode = mode;
323 647055063 : inode->i_rdev = dev;
324 647055063 : inode->i_uid = uid;
325 647055063 : inode->i_gid = gid;
326 647055063 : mapping_set_large_folios(inode->i_mapping);
327 647037720 : return error;
328 : }
329 :
330 : /*
331 : * Carefully nudge an inode whose VFS state has been torn down back into a
332 : * usable state. Drops the i_flags_lock and the rcu read lock.
333 : */
334 : static int
335 646234668 : xfs_iget_recycle(
336 : struct xfs_perag *pag,
337 : struct xfs_inode *ip) __releases(&ip->i_flags_lock)
338 : {
339 646234668 : struct xfs_mount *mp = ip->i_mount;
340 646234668 : struct inode *inode = VFS_I(ip);
341 646234668 : int error;
342 :
343 646234668 : trace_xfs_iget_recycle(ip);
344 :
345 646179328 : if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
346 : return -EAGAIN;
347 :
348 : /*
349 : * We need to make it look like the inode is being reclaimed to prevent
350 : * the actual reclaim workers from stomping over us while we recycle
351 : * the inode. We can't clear the radix tree tag yet as it requires
352 : * pag_ici_lock to be held exclusive.
353 : */
354 646510033 : ip->i_flags |= XFS_IRECLAIM;
355 :
356 646510033 : spin_unlock(&ip->i_flags_lock);
357 647336062 : rcu_read_unlock();
358 :
359 647170089 : ASSERT(!rwsem_is_locked(&inode->i_rwsem));
360 647170089 : error = xfs_reinit_inode(mp, inode);
361 647044853 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
362 646732441 : if (error) {
363 : /*
364 : * Re-initializing the inode failed, and we are in deep
365 : * trouble. Try to re-add it to the reclaim list.
366 : */
367 0 : rcu_read_lock();
368 0 : spin_lock(&ip->i_flags_lock);
369 0 : ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
370 0 : ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
371 0 : spin_unlock(&ip->i_flags_lock);
372 0 : rcu_read_unlock();
373 :
374 0 : trace_xfs_iget_recycle_fail(ip);
375 0 : return error;
376 : }
377 :
378 646732441 : spin_lock(&pag->pag_ici_lock);
379 647472105 : spin_lock(&ip->i_flags_lock);
380 :
381 : /*
382 : * Clear the per-lifetime state in the inode as we are now effectively
383 : * a new inode and need to return to the initial state before reuse
384 : * occurs.
385 : */
386 647586623 : ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
387 647586623 : ip->i_flags |= XFS_INEW;
388 647586623 : xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
389 : XFS_ICI_RECLAIM_TAG);
390 647242642 : inode->i_state = I_NEW;
391 1294485284 : spin_unlock(&ip->i_flags_lock);
392 647626233 : spin_unlock(&pag->pag_ici_lock);
393 :
394 647626233 : return 0;
395 : }
396 :
397 : /*
398 : * If we are allocating a new inode, then check what was returned is
399 : * actually a free, empty inode. If we are not allocating an inode,
400 : * then check we didn't find a free inode.
401 : *
402 : * Returns:
403 : * 0 if the inode free state matches the lookup context
404 : * -ENOENT if the inode is free and we are not allocating
405 : * -EFSCORRUPTED if there is any state mismatch at all
406 : */
407 : static int
408 57475212577 : xfs_iget_check_free_state(
409 : struct xfs_inode *ip,
410 : int flags)
411 : {
412 57475212577 : if (flags & XFS_IGET_CREATE) {
413 : /* should be a free inode */
414 84539389 : if (VFS_I(ip)->i_mode != 0) {
415 0 : xfs_warn(ip->i_mount,
416 : "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
417 : ip->i_ino, VFS_I(ip)->i_mode);
418 0 : return -EFSCORRUPTED;
419 : }
420 :
421 84539389 : if (ip->i_nblocks != 0) {
422 0 : xfs_warn(ip->i_mount,
423 : "Corruption detected! Free inode 0x%llx has blocks allocated!",
424 : ip->i_ino);
425 0 : return -EFSCORRUPTED;
426 : }
427 : return 0;
428 : }
429 :
430 : /* should be an allocated inode */
431 57390673188 : if (VFS_I(ip)->i_mode == 0)
432 5272298 : return -ENOENT;
433 :
434 : return 0;
435 : }
436 :
437 : /* Make all pending inactivation work start immediately. */
438 : static bool
439 36573816 : xfs_inodegc_queue_all(
440 : struct xfs_mount *mp)
441 : {
442 36573816 : struct xfs_inodegc *gc;
443 36573816 : int cpu;
444 36573816 : bool ret = false;
445 :
446 183046942 : for_each_online_cpu(cpu) {
447 146247766 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
448 146420033 : if (!llist_empty(&gc->list)) {
449 1593648 : mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0);
450 1593648 : ret = true;
451 : }
452 : }
453 :
454 36792502 : return ret;
455 : }
456 :
457 : /* Wait for all queued work and collect errors */
458 : static int
459 14385999 : xfs_inodegc_wait_all(
460 : struct xfs_mount *mp)
461 : {
462 14385999 : int cpu;
463 14385999 : int error = 0;
464 :
465 14385999 : flush_workqueue(mp->m_inodegc_wq);
466 87161387 : for_each_online_cpu(cpu) {
467 58218711 : struct xfs_inodegc *gc;
468 :
469 58218711 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
470 58218806 : if (gc->error && !error)
471 2260 : error = gc->error;
472 58218806 : gc->error = 0;
473 : }
474 :
475 14555902 : return error;
476 : }
477 :
478 : /*
479 : * Check the validity of the inode we just found it the cache
480 : */
481 : static int
482 56867034145 : xfs_iget_cache_hit(
483 : struct xfs_perag *pag,
484 : struct xfs_inode *ip,
485 : xfs_ino_t ino,
486 : int flags,
487 : int lock_flags) __releases(RCU)
488 : {
489 56867034145 : struct inode *inode = VFS_I(ip);
490 56867034145 : struct xfs_mount *mp = ip->i_mount;
491 56867034145 : int error;
492 :
493 : /*
494 : * check for re-use of an inode within an RCU grace period due to the
495 : * radix tree nodes not being updated yet. We monitor for this by
496 : * setting the inode number to zero before freeing the inode structure.
497 : * If the inode has been reallocated and set up, then the inode number
498 : * will not match, so check for that, too.
499 : */
500 56867034145 : spin_lock(&ip->i_flags_lock);
501 57896732367 : if (ip->i_ino != ino)
502 22 : goto out_skip;
503 :
504 : /*
505 : * If we are racing with another cache hit that is currently
506 : * instantiating this inode or currently recycling it out of
507 : * reclaimable state, wait for the initialisation to complete
508 : * before continuing.
509 : *
510 : * If we're racing with the inactivation worker we also want to wait.
511 : * If we're creating a new file, it's possible that the worker
512 : * previously marked the inode as free on disk but hasn't finished
513 : * updating the incore state yet. The AGI buffer will be dirty and
514 : * locked to the icreate transaction, so a synchronous push of the
515 : * inodegc workers would result in deadlock. For a regular iget, the
516 : * worker is running already, so we might as well wait.
517 : *
518 : * XXX(hch): eventually we should do something equivalent to
519 : * wait_on_inode to wait for these flags to be cleared
520 : * instead of polling for it.
521 : */
522 57896732345 : if (ip->i_flags & (XFS_INEW | XFS_IRECLAIM | XFS_INACTIVATING))
523 295833 : goto out_skip;
524 :
525 57896436512 : if (ip->i_flags & XFS_NEED_INACTIVE) {
526 : /* Unlinked inodes cannot be re-grabbed. */
527 2856477 : if (VFS_I(ip)->i_nlink == 0) {
528 2833530 : error = -ENOENT;
529 2833530 : goto out_error;
530 : }
531 22947 : goto out_inodegc_flush;
532 : }
533 :
534 : /*
535 : * Check the inode free state is valid. This also detects lookup
536 : * racing with unlinks.
537 : */
538 57893580035 : error = xfs_iget_check_free_state(ip, flags);
539 56632720115 : if (error)
540 5272308 : goto out_error;
541 :
542 : /* Skip inodes that have no vfs state. */
543 56627447807 : if ((flags & XFS_IGET_INCORE) &&
544 2174606004 : (ip->i_flags & XFS_IRECLAIMABLE))
545 91620 : goto out_skip;
546 :
547 : /* The inode fits the selection criteria; process it. */
548 56627356187 : if (ip->i_flags & XFS_IRECLAIMABLE) {
549 : /* Drops i_flags_lock and RCU read lock. */
550 646877741 : error = xfs_iget_recycle(pag, ip);
551 647552802 : if (error == -EAGAIN)
552 3 : goto out_skip;
553 647552799 : if (error)
554 : return error;
555 : } else {
556 : /* If the VFS inode is being torn down, pause and try again. */
557 55980478446 : if (!igrab(inode))
558 274002 : goto out_skip;
559 :
560 : /* We've got a live one. */
561 56557780567 : spin_unlock(&ip->i_flags_lock);
562 56962432408 : rcu_read_unlock();
563 56956404452 : trace_xfs_iget_hit(ip);
564 : }
565 :
566 56557118906 : if (lock_flags != 0)
567 51918372692 : xfs_ilock(ip, lock_flags);
568 :
569 57098688360 : if (!(flags & XFS_IGET_INCORE))
570 54949339390 : xfs_iflags_clear(ip, XFS_ISTALE);
571 57231245964 : XFS_STATS_INC(mp, xs_ig_found);
572 :
573 56582939682 : return 0;
574 :
575 661480 : out_skip:
576 661480 : trace_xfs_iget_skip(ip);
577 661187 : XFS_STATS_INC(mp, xs_ig_frecycle);
578 661131 : error = -EAGAIN;
579 8766969 : out_error:
580 8766969 : spin_unlock(&ip->i_flags_lock);
581 8767575 : rcu_read_unlock();
582 8767575 : return error;
583 :
584 : out_inodegc_flush:
585 22947 : spin_unlock(&ip->i_flags_lock);
586 22946 : rcu_read_unlock();
587 : /*
588 : * Do not wait for the workers, because the caller could hold an AGI
589 : * buffer lock. We're just going to sleep in a loop anyway.
590 : */
591 45892 : if (xfs_is_inodegc_enabled(mp))
592 22942 : xfs_inodegc_queue_all(mp);
593 : return -EAGAIN;
594 : }
595 :
596 : static int
597 440042033 : xfs_iget_cache_miss(
598 : struct xfs_mount *mp,
599 : struct xfs_perag *pag,
600 : xfs_trans_t *tp,
601 : xfs_ino_t ino,
602 : struct xfs_inode **ipp,
603 : int flags,
604 : int lock_flags)
605 : {
606 440042033 : struct xfs_inode *ip;
607 440042033 : int error;
608 440042033 : xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
609 440042033 : int iflags;
610 :
611 440042033 : ip = xfs_inode_alloc(mp, ino);
612 439992057 : if (!ip)
613 : return -ENOMEM;
614 :
615 439992057 : error = xfs_imap(pag, tp, ip->i_ino, &ip->i_imap, flags);
616 440009329 : if (error)
617 69189 : goto out_destroy;
618 :
619 : /*
620 : * For version 5 superblocks, if we are initialising a new inode and we
621 : * are not utilising the XFS_FEAT_IKEEP inode cluster mode, we can
622 : * simply build the new inode core with a random generation number.
623 : *
624 : * For version 4 (and older) superblocks, log recovery is dependent on
625 : * the i_flushiter field being initialised from the current on-disk
626 : * value and hence we must also read the inode off disk even when
627 : * initializing new inodes.
628 : */
629 439940140 : if (xfs_has_v3inodes(mp) &&
630 439938042 : (flags & XFS_IGET_CREATE) && !xfs_has_ikeep(mp)) {
631 47694716 : VFS_I(ip)->i_generation = get_random_u32();
632 : } else {
633 392245424 : struct xfs_buf *bp;
634 :
635 392245424 : error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp);
636 392479599 : if (error)
637 7340 : goto out_destroy;
638 :
639 392452600 : error = xfs_inode_from_disk(ip,
640 392473364 : xfs_buf_offset(bp, ip->i_imap.im_boffset));
641 392470386 : if (!error)
642 392469239 : xfs_buf_set_ref(bp, XFS_INO_REF);
643 392468031 : xfs_trans_brelse(tp, bp);
644 :
645 392506997 : if (error)
646 1105 : goto out_destroy;
647 : }
648 :
649 440053999 : trace_xfs_iget_miss(ip);
650 :
651 : /*
652 : * Check the inode free state is valid. This also detects lookup
653 : * racing with unlinks.
654 : */
655 439761049 : error = xfs_iget_check_free_state(ip, flags);
656 439797657 : if (error)
657 0 : goto out_destroy;
658 :
659 : /*
660 : * Preload the radix tree so we can insert safely under the
661 : * write spinlock. Note that we cannot sleep inside the preload
662 : * region. Since we can be called from transaction context, don't
663 : * recurse into the file system.
664 : */
665 439797657 : if (radix_tree_preload(GFP_NOFS)) {
666 0 : error = -EAGAIN;
667 0 : goto out_destroy;
668 : }
669 :
670 : /*
671 : * Because the inode hasn't been added to the radix-tree yet it can't
672 : * be found by another thread, so we can do the non-sleeping lock here.
673 : */
674 439920349 : if (lock_flags) {
675 420766308 : if (!xfs_ilock_nowait(ip, lock_flags))
676 0 : BUG();
677 : }
678 :
679 : /*
680 : * These values must be set before inserting the inode into the radix
681 : * tree as the moment it is inserted a concurrent lookup (allowed by the
682 : * RCU locking mechanism) can find it and that lookup must see that this
683 : * is an inode currently under construction (i.e. that XFS_INEW is set).
684 : * The ip->i_flags_lock that protects the XFS_INEW flag forms the
685 : * memory barrier that ensures this detection works correctly at lookup
686 : * time.
687 : */
688 440086118 : iflags = XFS_INEW;
689 440086118 : if (flags & XFS_IGET_DONTCACHE)
690 373392656 : d_mark_dontcache(VFS_I(ip));
691 440099841 : ip->i_udquot = NULL;
692 440099841 : ip->i_gdquot = NULL;
693 440099841 : ip->i_pdquot = NULL;
694 440099841 : xfs_iflags_set(ip, iflags);
695 :
696 : /* insert the new inode */
697 440280062 : spin_lock(&pag->pag_ici_lock);
698 440387201 : error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
699 439983340 : if (unlikely(error)) {
700 628384 : WARN_ON(error != -EEXIST);
701 628384 : XFS_STATS_INC(mp, xs_ig_dup);
702 628384 : error = -EAGAIN;
703 628384 : goto out_preload_end;
704 : }
705 439354956 : spin_unlock(&pag->pag_ici_lock);
706 439533426 : radix_tree_preload_end();
707 :
708 439240768 : *ipp = ip;
709 439240768 : return 0;
710 :
711 : out_preload_end:
712 628384 : spin_unlock(&pag->pag_ici_lock);
713 628384 : radix_tree_preload_end();
714 628384 : if (lock_flags)
715 624204 : xfs_iunlock(ip, lock_flags);
716 4180 : out_destroy:
717 704913 : __destroy_inode(VFS_I(ip));
718 704913 : xfs_inode_free(ip);
719 704913 : return error;
720 : }
721 :
722 : /*
723 : * Look up an inode by number in the given file system. The inode is looked up
724 : * in the cache held in each AG. If the inode is found in the cache, initialise
725 : * the vfs inode if necessary.
726 : *
727 : * If it is not in core, read it in from the file system's device, add it to the
728 : * cache and initialise the vfs inode.
729 : *
730 : * The inode is locked according to the value of the lock_flags parameter.
731 : * Inode lookup is only done during metadata operations and not as part of the
732 : * data IO path. Hence we only allow locking of the XFS_ILOCK during lookup.
733 : */
734 : int
735 57388208611 : xfs_iget(
736 : struct xfs_mount *mp,
737 : struct xfs_trans *tp,
738 : xfs_ino_t ino,
739 : uint flags,
740 : uint lock_flags,
741 : struct xfs_inode **ipp)
742 : {
743 57388208611 : struct xfs_inode *ip;
744 57388208611 : struct xfs_perag *pag;
745 57388208611 : xfs_agino_t agino;
746 57388208611 : int error;
747 :
748 57388208611 : ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
749 :
750 : /* reject inode numbers outside existing AGs */
751 57388208611 : if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
752 678300 : return -EINVAL;
753 :
754 57387530311 : XFS_STATS_INC(mp, xs_ig_attempts);
755 :
756 : /* get the perag structure and ensure that it's inode capable */
757 56545365969 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
758 57465699680 : agino = XFS_INO_TO_AGINO(mp, ino);
759 :
760 57466902101 : again:
761 57466902101 : error = 0;
762 57466902101 : rcu_read_lock();
763 57466850878 : ip = radix_tree_lookup(&pag->pag_ici_root, agino);
764 :
765 57890359641 : if (ip) {
766 57391499309 : error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
767 56390322019 : if (error)
768 8790187 : goto out_error_or_again;
769 : } else {
770 498860332 : rcu_read_unlock();
771 498871335 : if (flags & XFS_IGET_INCORE) {
772 58677694 : error = -ENODATA;
773 58677694 : goto out_error_or_again;
774 : }
775 440193641 : XFS_STATS_INC(mp, xs_ig_missed);
776 :
777 439972431 : error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
778 : flags, lock_flags);
779 439965060 : if (error)
780 704910 : goto out_error_or_again;
781 : }
782 56820791982 : xfs_perag_put(pag);
783 :
784 58009233559 : *ipp = ip;
785 :
786 : /*
787 : * If we have a real type for an on-disk inode, we can setup the inode
788 : * now. If it's a new inode being created, xfs_init_new_inode will
789 : * handle it.
790 : */
791 >11631*10^7 : if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
792 1001988506 : xfs_setup_existing_inode(ip);
793 : return 0;
794 :
795 68172791 : out_error_or_again:
796 68172791 : if (!(flags & (XFS_IGET_INCORE | XFS_IGET_NORETRY)) &&
797 : error == -EAGAIN) {
798 1218010 : delay(1);
799 1202421 : goto again;
800 : }
801 66954781 : xfs_perag_put(pag);
802 66954781 : return error;
803 : }
804 :
805 : /*
806 : * "Is this a cached inode that's also allocated?"
807 : *
808 : * Look up an inode by number in the given file system. If the inode is
809 : * in cache and isn't in purgatory, return 1 if the inode is allocated
810 : * and 0 if it is not. For all other cases (not in cache, being torn
811 : * down, etc.), return a negative error code.
812 : *
813 : * The caller has to prevent inode allocation and freeing activity,
814 : * presumably by locking the AGI buffer. This is to ensure that an
815 : * inode cannot transition from allocated to freed until the caller is
816 : * ready to allow that. If the inode is in an intermediate state (new,
817 : * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
818 : * inode is not in the cache, -ENOENT will be returned. The caller must
819 : * deal with these scenarios appropriately.
820 : *
821 : * This is a specialized use case for the online scrubber; if you're
822 : * reading this, you probably want xfs_iget.
823 : */
824 : int
825 2234123251 : xfs_icache_inode_is_allocated(
826 : struct xfs_mount *mp,
827 : struct xfs_trans *tp,
828 : xfs_ino_t ino,
829 : bool *inuse)
830 : {
831 2234123251 : struct xfs_inode *ip;
832 2234123251 : int error;
833 :
834 2234123251 : error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip);
835 2237612374 : if (error)
836 : return error;
837 :
838 2177978466 : *inuse = !!(VFS_I(ip)->i_mode);
839 2177978466 : xfs_irele(ip);
840 2177978466 : return 0;
841 : }
842 :
843 : /*
844 : * Grab the inode for reclaim exclusively.
845 : *
846 : * We have found this inode via a lookup under RCU, so the inode may have
847 : * already been freed, or it may be in the process of being recycled by
848 : * xfs_iget(). In both cases, the inode will have XFS_IRECLAIM set. If the inode
849 : * has been fully recycled by the time we get the i_flags_lock, XFS_IRECLAIMABLE
850 : * will not be set. Hence we need to check for both these flag conditions to
851 : * avoid inodes that are no longer reclaim candidates.
852 : *
853 : * Note: checking for other state flags here, under the i_flags_lock or not, is
854 : * racy and should be avoided. Those races should be resolved only after we have
855 : * ensured that we are able to reclaim this inode and the world can see that we
856 : * are going to reclaim it.
857 : *
858 : * Return true if we grabbed it, false otherwise.
859 : */
860 : static bool
861 456697242 : xfs_reclaim_igrab(
862 : struct xfs_inode *ip,
863 : struct xfs_icwalk *icw)
864 : {
865 456697242 : ASSERT(rcu_read_lock_held());
866 :
867 456697242 : spin_lock(&ip->i_flags_lock);
868 456698260 : if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
869 : __xfs_iflags_test(ip, XFS_IRECLAIM)) {
870 : /* not a reclaim candidate. */
871 11973 : spin_unlock(&ip->i_flags_lock);
872 11973 : return false;
873 : }
874 :
875 : /* Don't reclaim a sick inode unless the caller asked for it. */
876 456686287 : if (ip->i_sick &&
877 33 : (!icw || !(icw->icw_flags & XFS_ICWALK_FLAG_RECLAIM_SICK))) {
878 0 : spin_unlock(&ip->i_flags_lock);
879 0 : return false;
880 : }
881 :
882 456686287 : __xfs_iflags_set(ip, XFS_IRECLAIM);
883 456686287 : spin_unlock(&ip->i_flags_lock);
884 456686287 : return true;
885 : }
886 :
887 : /*
888 : * Inode reclaim is non-blocking, so the default action if progress cannot be
889 : * made is to "requeue" the inode for reclaim by unlocking it and clearing the
890 : * XFS_IRECLAIM flag. If we are in a shutdown state, we don't care about
891 : * blocking anymore and hence we can wait for the inode to be able to reclaim
892 : * it.
893 : *
894 : * We do no IO here - if callers require inodes to be cleaned they must push the
895 : * AIL first to trigger writeback of dirty inodes. This enables writeback to be
896 : * done in the background in a non-blocking manner, and enables memory reclaim
897 : * to make progress without blocking.
898 : */
899 : static void
900 456686034 : xfs_reclaim_inode(
901 : struct xfs_inode *ip,
902 : struct xfs_perag *pag)
903 : {
904 456686034 : xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
905 :
906 456686034 : if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
907 10950 : goto out;
908 456674299 : if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING))
909 8309420 : goto out_iunlock;
910 :
911 : /*
912 : * Check for log shutdown because aborting the inode can move the log
913 : * tail and corrupt in memory state. This is fine if the log is shut
914 : * down, but if the log is still active and only the mount is shut down
915 : * then the in-memory log tail movement caused by the abort can be
916 : * incorrectly propagated to disk.
917 : */
918 896731848 : if (xlog_is_shutdown(ip->i_mount->m_log)) {
919 339864664 : xfs_iunpin_wait(ip);
920 339864664 : xfs_iflush_shutdown_abort(ip);
921 339864664 : goto reclaim;
922 : }
923 108501260 : if (xfs_ipincount(ip))
924 3099242 : goto out_clear_flush;
925 105402018 : if (!xfs_inode_clean(ip))
926 5391056 : goto out_clear_flush;
927 :
928 100010962 : xfs_iflags_clear(ip, XFS_IFLUSHING);
929 439875700 : reclaim:
930 439875700 : trace_xfs_inode_reclaiming(ip);
931 :
932 : /*
933 : * Because we use RCU freeing we need to ensure the inode always appears
934 : * to be reclaimed with an invalid inode number when in the free state.
935 : * We do this as early as possible under the ILOCK so that
936 : * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
937 : * detect races with us here. By doing this, we guarantee that once
938 : * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
939 : * it will see either a valid inode that will serialise correctly, or it
940 : * will see an invalid inode that it can skip.
941 : */
942 439875548 : spin_lock(&ip->i_flags_lock);
943 439875791 : ip->i_flags = XFS_IRECLAIM;
944 439875791 : ip->i_ino = 0;
945 439875791 : ip->i_sick = 0;
946 439875791 : ip->i_checked = 0;
947 439875791 : spin_unlock(&ip->i_flags_lock);
948 :
949 439875767 : ASSERT(!ip->i_itemp || ip->i_itemp->ili_item.li_buf == NULL);
950 439875767 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
951 :
952 439875424 : XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
953 : /*
954 : * Remove the inode from the per-AG radix tree.
955 : *
956 : * Because radix_tree_delete won't complain even if the item was never
957 : * added to the tree assert that it's been there before to catch
958 : * problems with the inode life time early on.
959 : */
960 439875452 : spin_lock(&pag->pag_ici_lock);
961 439875651 : if (!radix_tree_delete(&pag->pag_ici_root,
962 439875821 : XFS_INO_TO_AGINO(ip->i_mount, ino)))
963 0 : ASSERT(0);
964 439875651 : xfs_perag_clear_inode_tag(pag, NULLAGINO, XFS_ICI_RECLAIM_TAG);
965 439875410 : spin_unlock(&pag->pag_ici_lock);
966 :
967 : /*
968 : * Here we do an (almost) spurious inode lock in order to coordinate
969 : * with inode cache radix tree lookups. This is because the lookup
970 : * can reference the inodes in the cache without taking references.
971 : *
972 : * We make that OK here by ensuring that we wait until the inode is
973 : * unlocked after the lookup before we go ahead and free it.
974 : */
975 439875800 : xfs_ilock(ip, XFS_ILOCK_EXCL);
976 439875726 : ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot);
977 439875726 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
978 497681792 : ASSERT(xfs_inode_clean(ip));
979 :
980 439875535 : __xfs_inode_free(ip);
981 439875535 : return;
982 :
983 8490298 : out_clear_flush:
984 8490298 : xfs_iflags_clear(ip, XFS_IFLUSHING);
985 16799718 : out_iunlock:
986 16799718 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
987 16810665 : out:
988 16810665 : xfs_iflags_clear(ip, XFS_IRECLAIM);
989 : }
990 :
991 : /* Reclaim sick inodes if we're unmounting or the fs went down. */
992 : static inline bool
993 107119 : xfs_want_reclaim_sick(
994 : struct xfs_mount *mp)
995 : {
996 262123 : return xfs_is_unmounting(mp) || xfs_has_norecovery(mp) ||
997 : xfs_is_shutdown(mp);
998 : }
999 :
1000 : void
1001 59234 : xfs_reclaim_inodes(
1002 : struct xfs_mount *mp)
1003 : {
1004 59234 : struct xfs_icwalk icw = {
1005 : .icw_flags = 0,
1006 : };
1007 :
1008 59234 : if (xfs_want_reclaim_sick(mp))
1009 59234 : icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1010 :
1011 594400 : while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
1012 535166 : xfs_ail_push_all_sync(mp->m_ail);
1013 535166 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1014 : }
1015 59234 : }
1016 :
1017 : /*
1018 : * The shrinker infrastructure determines how many inodes we should scan for
1019 : * reclaim. We want as many clean inodes ready to reclaim as possible, so we
1020 : * push the AIL here. We also want to proactively free up memory if we can to
1021 : * minimise the amount of work memory reclaim has to do so we kick the
1022 : * background reclaim if it isn't already scheduled.
1023 : */
1024 : long
1025 47885 : xfs_reclaim_inodes_nr(
1026 : struct xfs_mount *mp,
1027 : unsigned long nr_to_scan)
1028 : {
1029 47885 : struct xfs_icwalk icw = {
1030 : .icw_flags = XFS_ICWALK_FLAG_SCAN_LIMIT,
1031 47885 : .icw_scan_limit = min_t(unsigned long, LONG_MAX, nr_to_scan),
1032 : };
1033 :
1034 47885 : if (xfs_want_reclaim_sick(mp))
1035 0 : icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1036 :
1037 : /* kick background reclaimer and push the AIL */
1038 47885 : xfs_reclaim_work_queue(mp);
1039 47885 : xfs_ail_push_all(mp->m_ail);
1040 :
1041 47885 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1042 47885 : return 0;
1043 : }
1044 :
1045 : /*
1046 : * Return the number of reclaimable inodes in the filesystem for
1047 : * the shrinker to determine how much to reclaim.
1048 : */
1049 : long
1050 257854 : xfs_reclaim_inodes_count(
1051 : struct xfs_mount *mp)
1052 : {
1053 257854 : struct xfs_perag *pag;
1054 257854 : xfs_agnumber_t ag = 0;
1055 257854 : long reclaimable = 0;
1056 :
1057 611921 : while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1058 354067 : ag = pag->pag_agno + 1;
1059 354067 : reclaimable += pag->pag_ici_reclaimable;
1060 354067 : xfs_perag_put(pag);
1061 : }
1062 257855 : return reclaimable;
1063 : }
1064 :
1065 : STATIC bool
1066 2879084 : xfs_icwalk_match_id(
1067 : struct xfs_inode *ip,
1068 : struct xfs_icwalk *icw)
1069 : {
1070 2879084 : if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1071 : !uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1072 : return false;
1073 :
1074 2879084 : if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) &&
1075 : !gid_eq(VFS_I(ip)->i_gid, icw->icw_gid))
1076 : return false;
1077 :
1078 2879084 : if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) &&
1079 0 : ip->i_projid != icw->icw_prid)
1080 0 : return false;
1081 :
1082 : return true;
1083 : }
1084 :
1085 : /*
1086 : * A union-based inode filtering algorithm. Process the inode if any of the
1087 : * criteria match. This is for global/internal scans only.
1088 : */
1089 : STATIC bool
1090 3228 : xfs_icwalk_match_id_union(
1091 : struct xfs_inode *ip,
1092 : struct xfs_icwalk *icw)
1093 : {
1094 3228 : if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1095 : uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1096 : return true;
1097 :
1098 148 : if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) &&
1099 : gid_eq(VFS_I(ip)->i_gid, icw->icw_gid))
1100 : return true;
1101 :
1102 0 : if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) &&
1103 0 : ip->i_projid == icw->icw_prid)
1104 0 : return true;
1105 :
1106 : return false;
1107 : }
1108 :
1109 : /*
1110 : * Is this inode @ip eligible for eof/cow block reclamation, given some
1111 : * filtering parameters @icw? The inode is eligible if @icw is null or
1112 : * if the predicate functions match.
1113 : */
1114 : static bool
1115 4338782 : xfs_icwalk_match(
1116 : struct xfs_inode *ip,
1117 : struct xfs_icwalk *icw)
1118 : {
1119 4338782 : bool match;
1120 :
1121 4338782 : if (!icw)
1122 : return true;
1123 :
1124 2863667 : if (icw->icw_flags & XFS_ICWALK_FLAG_UNION)
1125 3225 : match = xfs_icwalk_match_id_union(ip, icw);
1126 : else
1127 2860442 : match = xfs_icwalk_match_id(ip, icw);
1128 2863667 : if (!match)
1129 : return false;
1130 :
1131 : /* skip the inode if the file size is too small */
1132 2858091 : if ((icw->icw_flags & XFS_ICWALK_FLAG_MINFILESIZE) &&
1133 0 : XFS_ISIZE(ip) < icw->icw_min_file_size)
1134 0 : return false;
1135 :
1136 : return true;
1137 : }
1138 :
1139 : /*
1140 : * This is a fast pass over the inode cache to try to get reclaim moving on as
1141 : * many inodes as possible in a short period of time. It kicks itself every few
1142 : * seconds, as well as being kicked by the inode cache shrinker when memory
1143 : * goes low.
1144 : */
1145 : void
1146 32096 : xfs_reclaim_worker(
1147 : struct work_struct *work)
1148 : {
1149 32096 : struct xfs_mount *mp = container_of(to_delayed_work(work),
1150 : struct xfs_mount, m_reclaim_work);
1151 :
1152 32096 : xfs_icwalk(mp, XFS_ICWALK_RECLAIM, NULL);
1153 32096 : xfs_reclaim_work_queue(mp);
1154 32096 : }
1155 :
1156 : STATIC int
1157 27721429 : xfs_inode_free_eofblocks(
1158 : struct xfs_inode *ip,
1159 : struct xfs_icwalk *icw,
1160 : unsigned int *lockflags)
1161 : {
1162 27721429 : bool wait;
1163 :
1164 27721429 : wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1165 :
1166 55752533 : if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS))
1167 : return 0;
1168 :
1169 : /*
1170 : * If the mapping is dirty the operation can block and wait for some
1171 : * time. Unless we are waiting, skip it.
1172 : */
1173 4695425 : if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1174 : return 0;
1175 :
1176 2707720 : if (!xfs_icwalk_match(ip, icw))
1177 : return 0;
1178 :
1179 : /*
1180 : * If the caller is waiting, return -EAGAIN to keep the background
1181 : * scanner moving and revisit the inode in a subsequent pass.
1182 : */
1183 2707720 : if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1184 1777070 : if (wait)
1185 : return -EAGAIN;
1186 99592 : return 0;
1187 : }
1188 957502 : *lockflags |= XFS_IOLOCK_EXCL;
1189 :
1190 957502 : if (xfs_can_free_eofblocks(ip, false))
1191 390626 : return xfs_free_eofblocks(ip);
1192 :
1193 : /* inode could be preallocated or append-only */
1194 566851 : trace_xfs_inode_free_eofblocks_invalid(ip);
1195 566839 : xfs_inode_clear_eofblocks_tag(ip);
1196 566839 : return 0;
1197 : }
1198 :
1199 : static void
1200 8938691 : xfs_blockgc_set_iflag(
1201 : struct xfs_inode *ip,
1202 : unsigned long iflag)
1203 : {
1204 8938691 : struct xfs_mount *mp = ip->i_mount;
1205 8938691 : struct xfs_perag *pag;
1206 :
1207 8938691 : ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
1208 :
1209 : /*
1210 : * Don't bother locking the AG and looking up in the radix trees
1211 : * if we already know that we have the tag set.
1212 : */
1213 8938691 : if (ip->i_flags & iflag)
1214 : return;
1215 4877912 : spin_lock(&ip->i_flags_lock);
1216 4882995 : ip->i_flags |= iflag;
1217 4882995 : spin_unlock(&ip->i_flags_lock);
1218 :
1219 4882554 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1220 4888354 : spin_lock(&pag->pag_ici_lock);
1221 :
1222 4887794 : xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1223 : XFS_ICI_BLOCKGC_TAG);
1224 :
1225 4881738 : spin_unlock(&pag->pag_ici_lock);
1226 4879879 : xfs_perag_put(pag);
1227 : }
1228 :
1229 : void
1230 4803750 : xfs_inode_set_eofblocks_tag(
1231 : xfs_inode_t *ip)
1232 : {
1233 4803750 : trace_xfs_inode_set_eofblocks_tag(ip);
1234 4803504 : return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS);
1235 : }
1236 :
1237 : static void
1238 24970123 : xfs_blockgc_clear_iflag(
1239 : struct xfs_inode *ip,
1240 : unsigned long iflag)
1241 : {
1242 24970123 : struct xfs_mount *mp = ip->i_mount;
1243 24970123 : struct xfs_perag *pag;
1244 24970123 : bool clear_tag;
1245 :
1246 24970123 : ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
1247 :
1248 24970123 : spin_lock(&ip->i_flags_lock);
1249 25024437 : ip->i_flags &= ~iflag;
1250 25024437 : clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0;
1251 25024437 : spin_unlock(&ip->i_flags_lock);
1252 :
1253 24997565 : if (!clear_tag)
1254 : return;
1255 :
1256 19047261 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1257 19086664 : spin_lock(&pag->pag_ici_lock);
1258 :
1259 19087857 : xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1260 : XFS_ICI_BLOCKGC_TAG);
1261 :
1262 19084015 : spin_unlock(&pag->pag_ici_lock);
1263 19097963 : xfs_perag_put(pag);
1264 : }
1265 :
1266 : void
1267 19330444 : xfs_inode_clear_eofblocks_tag(
1268 : xfs_inode_t *ip)
1269 : {
1270 19330444 : trace_xfs_inode_clear_eofblocks_tag(ip);
1271 19275966 : return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS);
1272 : }
1273 :
1274 : /*
1275 : * Set ourselves up to free CoW blocks from this file. If it's already clean
1276 : * then we can bail out quickly, but otherwise we must back off if the file
1277 : * is undergoing some kind of write.
1278 : */
1279 : static bool
1280 23856404 : xfs_prep_free_cowblocks(
1281 : struct xfs_inode *ip)
1282 : {
1283 : /*
1284 : * Just clear the tag if we have an empty cow fork or none at all. It's
1285 : * possible the inode was fully unshared since it was originally tagged.
1286 : */
1287 47712808 : if (!xfs_inode_has_cow_data(ip)) {
1288 302242 : trace_xfs_inode_free_cowblocks_invalid(ip);
1289 301840 : xfs_inode_clear_cowblocks_tag(ip);
1290 301840 : return false;
1291 : }
1292 :
1293 : /*
1294 : * If the mapping is dirty or under writeback we cannot touch the
1295 : * CoW fork. Leave it alone if we're in the midst of a directio.
1296 : */
1297 26501882 : if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1298 5894494 : mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1299 5066199 : mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1300 : atomic_read(&VFS_I(ip)->i_dio_count))
1301 21437709 : return false;
1302 :
1303 : return true;
1304 : }
1305 :
1306 : /*
1307 : * Automatic CoW Reservation Freeing
1308 : *
1309 : * These functions automatically garbage collect leftover CoW reservations
1310 : * that were made on behalf of a cowextsize hint when we start to run out
1311 : * of quota or when the reservations sit around for too long. If the file
1312 : * has dirty pages or is undergoing writeback, its CoW reservations will
1313 : * be retained.
1314 : *
1315 : * The actual garbage collection piggybacks off the same code that runs
1316 : * the speculative EOF preallocation garbage collector.
1317 : */
1318 : STATIC int
1319 26191497 : xfs_inode_free_cowblocks(
1320 : struct xfs_inode *ip,
1321 : struct xfs_icwalk *icw,
1322 : unsigned int *lockflags)
1323 : {
1324 26191497 : bool wait;
1325 26191497 : int ret = 0;
1326 :
1327 26191497 : wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1328 :
1329 52756080 : if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS))
1330 : return 0;
1331 :
1332 23458410 : if (!xfs_prep_free_cowblocks(ip))
1333 : return 0;
1334 :
1335 1611442 : if (!xfs_icwalk_match(ip, icw))
1336 : return 0;
1337 :
1338 : /*
1339 : * If the caller is waiting, return -EAGAIN to keep the background
1340 : * scanner moving and revisit the inode in a subsequent pass.
1341 : */
1342 3196923 : if (!(*lockflags & XFS_IOLOCK_EXCL) &&
1343 1575510 : !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1344 1114623 : if (wait)
1345 : return -EAGAIN;
1346 82419 : return 0;
1347 : }
1348 506790 : *lockflags |= XFS_IOLOCK_EXCL;
1349 :
1350 506790 : if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) {
1351 100 : if (wait)
1352 : return -EAGAIN;
1353 63 : return 0;
1354 : }
1355 505879 : *lockflags |= XFS_MMAPLOCK_EXCL;
1356 :
1357 : /*
1358 : * Check again, nobody else should be able to dirty blocks or change
1359 : * the reflink iflag now that we have the first two locks held.
1360 : */
1361 505879 : if (xfs_prep_free_cowblocks(ip))
1362 505868 : ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
1363 : return ret;
1364 : }
1365 :
1366 : void
1367 4136628 : xfs_inode_set_cowblocks_tag(
1368 : xfs_inode_t *ip)
1369 : {
1370 4136628 : trace_xfs_inode_set_cowblocks_tag(ip);
1371 4132429 : return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS);
1372 : }
1373 :
1374 : void
1375 5677874 : xfs_inode_clear_cowblocks_tag(
1376 : xfs_inode_t *ip)
1377 : {
1378 5677874 : trace_xfs_inode_clear_cowblocks_tag(ip);
1379 5675409 : return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS);
1380 : }
1381 :
1382 : /* Disable post-EOF and CoW block auto-reclamation. */
1383 : void
1384 122114 : xfs_blockgc_stop(
1385 : struct xfs_mount *mp)
1386 : {
1387 122114 : struct xfs_perag *pag;
1388 122114 : xfs_agnumber_t agno;
1389 :
1390 122114 : if (!xfs_clear_blockgc_enabled(mp))
1391 72 : return;
1392 :
1393 838609 : for_each_perag(mp, agno, pag)
1394 716567 : cancel_delayed_work_sync(&pag->pag_blockgc_work);
1395 122042 : trace_xfs_blockgc_stop(mp, __return_address);
1396 : }
1397 :
1398 : /* Enable post-EOF and CoW block auto-reclamation. */
1399 : void
1400 122205 : xfs_blockgc_start(
1401 : struct xfs_mount *mp)
1402 : {
1403 122205 : struct xfs_perag *pag;
1404 122205 : xfs_agnumber_t agno;
1405 :
1406 122205 : if (xfs_set_blockgc_enabled(mp))
1407 : return;
1408 :
1409 122183 : trace_xfs_blockgc_start(mp, __return_address);
1410 197783 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1411 75600 : xfs_blockgc_queue(pag);
1412 : }
1413 :
1414 : /* Don't try to run block gc on an inode that's in any of these states. */
1415 : #define XFS_BLOCKGC_NOGRAB_IFLAGS (XFS_INEW | \
1416 : XFS_NEED_INACTIVE | \
1417 : XFS_INACTIVATING | \
1418 : XFS_IRECLAIMABLE | \
1419 : XFS_IRECLAIM)
1420 : /*
1421 : * Decide if the given @ip is eligible for garbage collection of speculative
1422 : * preallocations, and grab it if so. Returns true if it's ready to go or
1423 : * false if we should just ignore it.
1424 : */
1425 : static bool
1426 28018071 : xfs_blockgc_igrab(
1427 : struct xfs_inode *ip)
1428 : {
1429 28018071 : struct inode *inode = VFS_I(ip);
1430 :
1431 28018071 : ASSERT(rcu_read_lock_held());
1432 :
1433 : /* Check for stale RCU freed inode */
1434 28018071 : spin_lock(&ip->i_flags_lock);
1435 28468038 : if (!ip->i_ino)
1436 0 : goto out_unlock_noent;
1437 :
1438 28468038 : if (ip->i_flags & XFS_BLOCKGC_NOGRAB_IFLAGS)
1439 296331 : goto out_unlock_noent;
1440 28171707 : spin_unlock(&ip->i_flags_lock);
1441 :
1442 : /* nothing to sync during shutdown */
1443 56136218 : if (xfs_is_shutdown(ip->i_mount))
1444 : return false;
1445 :
1446 : /* If we can't grab the inode, it must on it's way to reclaim. */
1447 28044380 : if (!igrab(inode))
1448 1947 : return false;
1449 :
1450 : /* inode is valid */
1451 : return true;
1452 :
1453 296331 : out_unlock_noent:
1454 296331 : spin_unlock(&ip->i_flags_lock);
1455 296331 : return false;
1456 : }
1457 :
1458 : /* Scan one incore inode for block preallocations that we can remove. */
1459 : static int
1460 27773972 : xfs_blockgc_scan_inode(
1461 : struct xfs_inode *ip,
1462 : struct xfs_icwalk *icw)
1463 : {
1464 27773972 : unsigned int lockflags = 0;
1465 27773972 : int error;
1466 :
1467 27773972 : error = xfs_inode_free_eofblocks(ip, icw, &lockflags);
1468 27942353 : if (error)
1469 1671953 : goto unlock;
1470 :
1471 26270400 : error = xfs_inode_free_cowblocks(ip, icw, &lockflags);
1472 28039151 : unlock:
1473 28039151 : if (lockflags)
1474 1427583 : xfs_iunlock(ip, lockflags);
1475 28039101 : xfs_irele(ip);
1476 28006302 : return error;
1477 : }
1478 :
1479 : /* Background worker that trims preallocated space. */
1480 : void
1481 2654568 : xfs_blockgc_worker(
1482 : struct work_struct *work)
1483 : {
1484 2654568 : struct xfs_perag *pag = container_of(to_delayed_work(work),
1485 : struct xfs_perag, pag_blockgc_work);
1486 2654568 : struct xfs_mount *mp = pag->pag_mount;
1487 2654568 : int error;
1488 :
1489 2654568 : trace_xfs_blockgc_worker(mp, __return_address);
1490 :
1491 2654564 : error = xfs_icwalk_ag(pag, XFS_ICWALK_BLOCKGC, NULL);
1492 2653592 : if (error)
1493 0 : xfs_info(mp, "AG %u preallocation gc worker failed, err=%d",
1494 : pag->pag_agno, error);
1495 2653592 : xfs_blockgc_queue(pag);
1496 2653278 : }
1497 :
1498 : /*
1499 : * Try to free space in the filesystem by purging inactive inodes, eofblocks
1500 : * and cowblocks.
1501 : */
1502 : int
1503 1169123 : xfs_blockgc_free_space(
1504 : struct xfs_mount *mp,
1505 : struct xfs_icwalk *icw)
1506 : {
1507 1169123 : int error;
1508 :
1509 1169123 : trace_xfs_blockgc_free_space(mp, icw, _RET_IP_);
1510 :
1511 1168671 : error = xfs_icwalk(mp, XFS_ICWALK_BLOCKGC, icw);
1512 1168057 : if (error)
1513 : return error;
1514 :
1515 1168027 : return xfs_inodegc_flush(mp);
1516 : }
1517 :
1518 : /*
1519 : * Reclaim all the free space that we can by scheduling the background blockgc
1520 : * and inodegc workers immediately and waiting for them all to clear.
1521 : */
1522 : int
1523 12207189 : xfs_blockgc_flush_all(
1524 : struct xfs_mount *mp)
1525 : {
1526 12207189 : struct xfs_perag *pag;
1527 12207189 : xfs_agnumber_t agno;
1528 :
1529 12207189 : trace_xfs_blockgc_flush_all(mp, __return_address);
1530 :
1531 : /*
1532 : * For each blockgc worker, move its queue time up to now. If it
1533 : * wasn't queued, it will not be requeued. Then flush whatever's
1534 : * left.
1535 : */
1536 14190919 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1537 2271623 : mod_delayed_work(pag->pag_mount->m_blockgc_wq,
1538 : &pag->pag_blockgc_work, 0);
1539 :
1540 14354913 : for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1541 2257244 : flush_delayed_work(&pag->pag_blockgc_work);
1542 :
1543 12205171 : return xfs_inodegc_flush(mp);
1544 : }
1545 :
1546 : /*
1547 : * Run cow/eofblocks scans on the supplied dquots. We don't know exactly which
1548 : * quota caused an allocation failure, so we make a best effort by including
1549 : * each quota under low free space conditions (less than 1% free space) in the
1550 : * scan.
1551 : *
1552 : * Callers must not hold any inode's ILOCK. If requesting a synchronous scan
1553 : * (XFS_ICWALK_FLAG_SYNC), the caller also must not hold any inode's IOLOCK or
1554 : * MMAPLOCK.
1555 : */
1556 : int
1557 32989 : xfs_blockgc_free_dquots(
1558 : struct xfs_mount *mp,
1559 : struct xfs_dquot *udqp,
1560 : struct xfs_dquot *gdqp,
1561 : struct xfs_dquot *pdqp,
1562 : unsigned int iwalk_flags)
1563 : {
1564 32989 : struct xfs_icwalk icw = {0};
1565 32989 : bool do_work = false;
1566 :
1567 32989 : if (!udqp && !gdqp && !pdqp)
1568 : return 0;
1569 :
1570 : /*
1571 : * Run a scan to free blocks using the union filter to cover all
1572 : * applicable quotas in a single scan.
1573 : */
1574 32989 : icw.icw_flags = XFS_ICWALK_FLAG_UNION | iwalk_flags;
1575 :
1576 32989 : if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) {
1577 32013 : icw.icw_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id);
1578 32010 : icw.icw_flags |= XFS_ICWALK_FLAG_UID;
1579 32010 : do_work = true;
1580 : }
1581 :
1582 32986 : if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) {
1583 32119 : icw.icw_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id);
1584 32119 : icw.icw_flags |= XFS_ICWALK_FLAG_GID;
1585 32119 : do_work = true;
1586 : }
1587 :
1588 32986 : if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) {
1589 26854 : icw.icw_prid = pdqp->q_id;
1590 26854 : icw.icw_flags |= XFS_ICWALK_FLAG_PRID;
1591 26854 : do_work = true;
1592 : }
1593 :
1594 32986 : if (!do_work)
1595 : return 0;
1596 :
1597 32571 : return xfs_blockgc_free_space(mp, &icw);
1598 : }
1599 :
1600 : /* Run cow/eofblocks scans on the quotas attached to the inode. */
1601 : int
1602 22037 : xfs_blockgc_free_quota(
1603 : struct xfs_inode *ip,
1604 : unsigned int iwalk_flags)
1605 : {
1606 22037 : return xfs_blockgc_free_dquots(ip->i_mount,
1607 : xfs_inode_dquot(ip, XFS_DQTYPE_USER),
1608 : xfs_inode_dquot(ip, XFS_DQTYPE_GROUP),
1609 : xfs_inode_dquot(ip, XFS_DQTYPE_PROJ), iwalk_flags);
1610 : }
1611 :
1612 : /* XFS Inode Cache Walking Code */
1613 :
1614 : /*
1615 : * The inode lookup is done in batches to keep the amount of lock traffic and
1616 : * radix tree lookups to a minimum. The batch size is a trade off between
1617 : * lookup reduction and stack usage. This is in the reclaim path, so we can't
1618 : * be too greedy.
1619 : */
1620 : #define XFS_LOOKUP_BATCH 32
1621 :
1622 :
1623 : /*
1624 : * Decide if we want to grab this inode in anticipation of doing work towards
1625 : * the goal.
1626 : */
1627 : static inline bool
1628 484758407 : xfs_icwalk_igrab(
1629 : enum xfs_icwalk_goal goal,
1630 : struct xfs_inode *ip,
1631 : struct xfs_icwalk *icw)
1632 : {
1633 484758407 : switch (goal) {
1634 28061045 : case XFS_ICWALK_BLOCKGC:
1635 28061045 : return xfs_blockgc_igrab(ip);
1636 456697362 : case XFS_ICWALK_RECLAIM:
1637 456697362 : return xfs_reclaim_igrab(ip, icw);
1638 : default:
1639 : return false;
1640 : }
1641 : }
1642 :
1643 : /*
1644 : * Process an inode. Each processing function must handle any state changes
1645 : * made by the icwalk igrab function. Return -EAGAIN to skip an inode.
1646 : */
1647 : static inline int
1648 484523841 : xfs_icwalk_process_inode(
1649 : enum xfs_icwalk_goal goal,
1650 : struct xfs_inode *ip,
1651 : struct xfs_perag *pag,
1652 : struct xfs_icwalk *icw)
1653 : {
1654 484523841 : int error = 0;
1655 :
1656 484523841 : switch (goal) {
1657 27837759 : case XFS_ICWALK_BLOCKGC:
1658 27837759 : error = xfs_blockgc_scan_inode(ip, icw);
1659 27837759 : break;
1660 456686082 : case XFS_ICWALK_RECLAIM:
1661 456686082 : xfs_reclaim_inode(ip, pag);
1662 456686082 : break;
1663 : }
1664 484644598 : return error;
1665 : }
1666 :
1667 : /*
1668 : * For a given per-AG structure @pag and a goal, grab qualifying inodes and
1669 : * process them in some manner.
1670 : */
1671 : static int
1672 3932680 : xfs_icwalk_ag(
1673 : struct xfs_perag *pag,
1674 : enum xfs_icwalk_goal goal,
1675 : struct xfs_icwalk *icw)
1676 : {
1677 3932680 : struct xfs_mount *mp = pag->pag_mount;
1678 3932680 : uint32_t first_index;
1679 3932680 : int last_error = 0;
1680 5464347 : int skipped;
1681 5464347 : bool done;
1682 5464347 : int nr_found;
1683 :
1684 5464347 : restart:
1685 5464347 : done = false;
1686 5464347 : skipped = 0;
1687 5464347 : if (goal == XFS_ICWALK_RECLAIM)
1688 804981 : first_index = READ_ONCE(pag->pag_ici_reclaim_cursor);
1689 : else
1690 : first_index = 0;
1691 : nr_found = 0;
1692 25210037 : do {
1693 25210037 : struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1694 25210037 : int error = 0;
1695 25210037 : int i;
1696 :
1697 25210037 : rcu_read_lock();
1698 :
1699 25157603 : nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root,
1700 : (void **) batch, first_index,
1701 : XFS_LOOKUP_BATCH, goal);
1702 25187941 : if (!nr_found) {
1703 5312867 : done = true;
1704 5312867 : rcu_read_unlock();
1705 5463983 : break;
1706 : }
1707 :
1708 : /*
1709 : * Grab the inodes before we drop the lock. if we found
1710 : * nothing, nr == 0 and the loop will be skipped.
1711 : */
1712 504779604 : for (i = 0; i < nr_found; i++) {
1713 484880627 : struct xfs_inode *ip = batch[i];
1714 :
1715 484836433 : if (done || !xfs_icwalk_igrab(goal, ip, icw))
1716 321453 : batch[i] = NULL;
1717 :
1718 : /*
1719 : * Update the index for the next lookup. Catch
1720 : * overflows into the next AG range which can occur if
1721 : * we have inodes in the last block of the AG and we
1722 : * are currently pointing to the last inode.
1723 : *
1724 : * Because we may see inodes that are from the wrong AG
1725 : * due to RCU freeing and reallocation, only update the
1726 : * index if it lies in this AG. It was a race that lead
1727 : * us to see this inode, so another lookup from the
1728 : * same index will not find it again.
1729 : */
1730 484904530 : if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
1731 440 : continue;
1732 484904090 : first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1733 484904090 : if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1734 32 : done = true;
1735 : }
1736 :
1737 : /* unlock now we've grabbed the inodes. */
1738 19898977 : rcu_read_unlock();
1739 :
1740 524800463 : for (i = 0; i < nr_found; i++) {
1741 485023271 : if (!batch[i])
1742 308793 : continue;
1743 484624031 : error = xfs_icwalk_process_inode(goal, batch[i], pag,
1744 : icw);
1745 484695187 : if (error == -EAGAIN) {
1746 2675062 : skipped++;
1747 2675062 : continue;
1748 : }
1749 482020125 : if (error && last_error != -EFSCORRUPTED)
1750 0 : last_error = error;
1751 : }
1752 :
1753 : /* bail out if the filesystem is corrupted. */
1754 19878215 : if (error == -EFSCORRUPTED)
1755 : break;
1756 :
1757 19878215 : cond_resched();
1758 :
1759 19881112 : if (icw && (icw->icw_flags & XFS_ICWALK_FLAG_SCAN_LIMIT)) {
1760 419399 : icw->icw_scan_limit -= XFS_LOOKUP_BATCH;
1761 419399 : if (icw->icw_scan_limit <= 0)
1762 : break;
1763 : }
1764 19745722 : } while (nr_found && !done);
1765 :
1766 5464015 : if (goal == XFS_ICWALK_RECLAIM) {
1767 804981 : if (done)
1768 669591 : first_index = 0;
1769 804981 : WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index);
1770 : }
1771 :
1772 5464015 : if (skipped) {
1773 1532633 : delay(1);
1774 1531667 : goto restart;
1775 : }
1776 3931382 : return last_error;
1777 : }
1778 :
1779 : /* Walk all incore inodes to achieve a given goal. */
1780 : static int
1781 1783860 : xfs_icwalk(
1782 : struct xfs_mount *mp,
1783 : enum xfs_icwalk_goal goal,
1784 : struct xfs_icwalk *icw)
1785 : {
1786 1783860 : struct xfs_perag *pag;
1787 1783860 : int error = 0;
1788 1783860 : int last_error = 0;
1789 1783860 : xfs_agnumber_t agno;
1790 :
1791 3061977 : for_each_perag_tag(mp, agno, pag, goal) {
1792 1278560 : error = xfs_icwalk_ag(pag, goal, icw);
1793 1278117 : if (error) {
1794 0 : last_error = error;
1795 0 : if (error == -EFSCORRUPTED) {
1796 0 : xfs_perag_rele(pag);
1797 0 : break;
1798 : }
1799 : }
1800 : }
1801 1783688 : return last_error;
1802 : BUILD_BUG_ON(XFS_ICWALK_PRIVATE_FLAGS & XFS_ICWALK_FLAGS_VALID);
1803 : }
1804 :
1805 : #ifdef DEBUG
1806 : static void
1807 0 : xfs_check_delalloc(
1808 : struct xfs_inode *ip,
1809 : int whichfork)
1810 : {
1811 0 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1812 0 : struct xfs_bmbt_irec got;
1813 0 : struct xfs_iext_cursor icur;
1814 :
1815 0 : if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
1816 0 : return;
1817 0 : do {
1818 0 : if (isnullstartblock(got.br_startblock)) {
1819 0 : xfs_warn(ip->i_mount,
1820 : "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
1821 : ip->i_ino,
1822 : whichfork == XFS_DATA_FORK ? "data" : "cow",
1823 : got.br_startoff, got.br_blockcount);
1824 : }
1825 0 : } while (xfs_iext_next_extent(ifp, &icur, &got));
1826 : }
1827 : #else
1828 : #define xfs_check_delalloc(ip, whichfork) do { } while (0)
1829 : #endif
1830 :
1831 : /* Schedule the inode for reclaim. */
1832 : static void
1833 1086144762 : xfs_inodegc_set_reclaimable(
1834 : struct xfs_inode *ip)
1835 : {
1836 1086144762 : struct xfs_mount *mp = ip->i_mount;
1837 1086144762 : struct xfs_perag *pag;
1838 :
1839 2172289524 : if (!xfs_is_shutdown(mp) && ip->i_delayed_blks) {
1840 0 : xfs_check_delalloc(ip, XFS_DATA_FORK);
1841 0 : xfs_check_delalloc(ip, XFS_COW_FORK);
1842 0 : ASSERT(0);
1843 : }
1844 :
1845 1086144762 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1846 1087086795 : spin_lock(&pag->pag_ici_lock);
1847 1087395271 : spin_lock(&ip->i_flags_lock);
1848 :
1849 1087476632 : trace_xfs_inode_set_reclaimable(ip);
1850 1086768609 : ip->i_flags &= ~(XFS_NEED_INACTIVE | XFS_INACTIVATING);
1851 1086768609 : ip->i_flags |= XFS_IRECLAIMABLE;
1852 1086768609 : xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1853 : XFS_ICI_RECLAIM_TAG);
1854 :
1855 1086972967 : spin_unlock(&ip->i_flags_lock);
1856 1087409367 : spin_unlock(&pag->pag_ici_lock);
1857 1087358703 : xfs_perag_put(pag);
1858 1087181650 : }
1859 :
1860 : /*
1861 : * Free all speculative preallocations and possibly even the inode itself.
1862 : * This is the last chance to make changes to an otherwise unreferenced file
1863 : * before incore reclamation happens.
1864 : */
1865 : static int
1866 58052332 : xfs_inodegc_inactivate(
1867 : struct xfs_inode *ip)
1868 : {
1869 58052332 : int error;
1870 :
1871 58052332 : trace_xfs_inode_inactivating(ip);
1872 57986876 : error = xfs_inactive(ip);
1873 58077096 : xfs_inodegc_set_reclaimable(ip);
1874 58100246 : return error;
1875 :
1876 : }
1877 :
1878 : void
1879 5435694 : xfs_inodegc_worker(
1880 : struct work_struct *work)
1881 : {
1882 5435694 : struct xfs_inodegc *gc = container_of(to_delayed_work(work),
1883 : struct xfs_inodegc, work);
1884 5435694 : struct llist_node *node = llist_del_all(&gc->list);
1885 5444655 : struct xfs_inode *ip, *n;
1886 5444655 : unsigned int nofs_flag;
1887 :
1888 5444655 : ASSERT(gc->cpu == smp_processor_id());
1889 :
1890 5442631 : WRITE_ONCE(gc->items, 0);
1891 :
1892 5442631 : if (!node)
1893 : return;
1894 :
1895 : /*
1896 : * We can allocate memory here while doing writeback on behalf of
1897 : * memory reclaim. To avoid memory allocation deadlocks set the
1898 : * task-wide nofs context for the following operations.
1899 : */
1900 5440719 : nofs_flag = memalloc_nofs_save();
1901 :
1902 5440719 : ip = llist_entry(node, struct xfs_inode, i_gclist);
1903 5440719 : trace_xfs_inodegc_worker(ip->i_mount, READ_ONCE(gc->shrinker_hits));
1904 :
1905 5438479 : WRITE_ONCE(gc->shrinker_hits, 0);
1906 63532842 : llist_for_each_entry_safe(ip, n, node, i_gclist) {
1907 58082863 : int error;
1908 :
1909 58082863 : xfs_iflags_set(ip, XFS_INACTIVATING);
1910 58050169 : error = xfs_inodegc_inactivate(ip);
1911 58094363 : if (error && !gc->error)
1912 2467 : gc->error = error;
1913 : }
1914 :
1915 5449979 : memalloc_nofs_restore(nofs_flag);
1916 : }
1917 :
1918 : /*
1919 : * Expedite all pending inodegc work to run immediately. This does not wait for
1920 : * completion of the work.
1921 : */
1922 : void
1923 36288197 : xfs_inodegc_push(
1924 : struct xfs_mount *mp)
1925 : {
1926 72576394 : if (!xfs_is_inodegc_enabled(mp))
1927 : return;
1928 36052566 : trace_xfs_inodegc_push(mp, __return_address);
1929 36040884 : xfs_inodegc_queue_all(mp);
1930 : }
1931 :
1932 : /*
1933 : * Force all currently queued inode inactivation work to run immediately and
1934 : * wait for the work to finish.
1935 : */
1936 : int
1937 14314238 : xfs_inodegc_flush(
1938 : struct xfs_mount *mp)
1939 : {
1940 14314238 : xfs_inodegc_push(mp);
1941 14427754 : trace_xfs_inodegc_flush(mp, __return_address);
1942 14396712 : return xfs_inodegc_wait_all(mp);
1943 : }
1944 :
1945 : /*
1946 : * Flush all the pending work and then disable the inode inactivation background
1947 : * workers and wait for them to stop. Caller must hold sb->s_umount to
1948 : * coordinate changes in the inodegc_enabled state.
1949 : */
1950 : void
1951 122242 : xfs_inodegc_stop(
1952 : struct xfs_mount *mp)
1953 : {
1954 122242 : bool rerun;
1955 :
1956 122242 : if (!xfs_clear_inodegc_enabled(mp))
1957 : return;
1958 :
1959 : /*
1960 : * Drain all pending inodegc work, including inodes that could be
1961 : * queued by racing xfs_inodegc_queue or xfs_inodegc_shrinker_scan
1962 : * threads that sample the inodegc state just prior to us clearing it.
1963 : * The inodegc flag state prevents new threads from queuing more
1964 : * inodes, so we queue pending work items and flush the workqueue until
1965 : * all inodegc lists are empty. IOWs, we cannot use drain_workqueue
1966 : * here because it does not allow other unserialized mechanisms to
1967 : * reschedule inodegc work while this draining is in progress.
1968 : */
1969 122170 : xfs_inodegc_queue_all(mp);
1970 122170 : do {
1971 122170 : flush_workqueue(mp->m_inodegc_wq);
1972 122170 : rerun = xfs_inodegc_queue_all(mp);
1973 122170 : } while (rerun);
1974 :
1975 122170 : trace_xfs_inodegc_stop(mp, __return_address);
1976 : }
1977 :
1978 : /*
1979 : * Enable the inode inactivation background workers and schedule deferred inode
1980 : * inactivation work if there is any. Caller must hold sb->s_umount to
1981 : * coordinate changes in the inodegc_enabled state.
1982 : */
1983 : void
1984 122205 : xfs_inodegc_start(
1985 : struct xfs_mount *mp)
1986 : {
1987 122205 : if (xfs_set_inodegc_enabled(mp))
1988 : return;
1989 :
1990 122183 : trace_xfs_inodegc_start(mp, __return_address);
1991 122183 : xfs_inodegc_queue_all(mp);
1992 : }
1993 :
1994 : #ifdef CONFIG_XFS_RT
1995 : static inline bool
1996 42332368 : xfs_inodegc_want_queue_rt_file(
1997 : struct xfs_inode *ip)
1998 : {
1999 42332368 : struct xfs_mount *mp = ip->i_mount;
2000 :
2001 42332368 : if (!XFS_IS_REALTIME_INODE(ip))
2002 : return false;
2003 :
2004 6794669 : if (__percpu_counter_compare(&mp->m_frextents,
2005 6794731 : mp->m_low_rtexts[XFS_LOWSP_5_PCNT],
2006 : XFS_FDBLOCKS_BATCH) < 0)
2007 196412 : return true;
2008 :
2009 : return false;
2010 : }
2011 : #else
2012 : # define xfs_inodegc_want_queue_rt_file(ip) (false)
2013 : #endif /* CONFIG_XFS_RT */
2014 :
2015 : /*
2016 : * Schedule the inactivation worker when:
2017 : *
2018 : * - We've accumulated more than one inode cluster buffer's worth of inodes.
2019 : * - There is less than 5% free space left.
2020 : * - Any of the quotas for this inode are near an enforcement limit.
2021 : */
2022 : static inline bool
2023 58073506 : xfs_inodegc_want_queue_work(
2024 : struct xfs_inode *ip,
2025 : unsigned int items)
2026 : {
2027 58073506 : struct xfs_mount *mp = ip->i_mount;
2028 :
2029 58073506 : if (items > mp->m_ino_geo.inodes_per_cluster)
2030 : return true;
2031 :
2032 42667012 : if (__percpu_counter_compare(&mp->m_fdblocks,
2033 42675311 : mp->m_low_space[XFS_LOWSP_5_PCNT],
2034 : XFS_FDBLOCKS_BATCH) < 0)
2035 : return true;
2036 :
2037 42333079 : if (xfs_inodegc_want_queue_rt_file(ip))
2038 : return true;
2039 :
2040 42130051 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_USER))
2041 : return true;
2042 :
2043 42135282 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_GROUP))
2044 : return true;
2045 :
2046 42135459 : if (xfs_inode_near_dquot_enforcement(ip, XFS_DQTYPE_PROJ))
2047 114 : return true;
2048 :
2049 : return false;
2050 : }
2051 :
2052 : /*
2053 : * Upper bound on the number of inodes in each AG that can be queued for
2054 : * inactivation at any given time, to avoid monopolizing the workqueue.
2055 : */
2056 : #define XFS_INODEGC_MAX_BACKLOG (4 * XFS_INODES_PER_CHUNK)
2057 :
2058 : /*
2059 : * Make the frontend wait for inactivations when:
2060 : *
2061 : * - Memory shrinkers queued the inactivation worker and it hasn't finished.
2062 : * - The queue depth exceeds the maximum allowable percpu backlog.
2063 : *
2064 : * Note: If the current thread is running a transaction, we don't ever want to
2065 : * wait for other transactions because that could introduce a deadlock.
2066 : */
2067 : static inline bool
2068 : xfs_inodegc_want_flush_work(
2069 : struct xfs_inode *ip,
2070 : unsigned int items,
2071 : unsigned int shrinker_hits)
2072 : {
2073 58070013 : if (current->journal_info)
2074 : return false;
2075 :
2076 58008238 : if (shrinker_hits > 0)
2077 : return true;
2078 :
2079 58008238 : if (items > XFS_INODEGC_MAX_BACKLOG)
2080 : return true;
2081 :
2082 : return false;
2083 : }
2084 :
2085 : /*
2086 : * Queue a background inactivation worker if there are inodes that need to be
2087 : * inactivated and higher level xfs code hasn't disabled the background
2088 : * workers.
2089 : */
2090 : static void
2091 58049066 : xfs_inodegc_queue(
2092 : struct xfs_inode *ip)
2093 : {
2094 58049066 : struct xfs_mount *mp = ip->i_mount;
2095 58049066 : struct xfs_inodegc *gc;
2096 58049066 : int items;
2097 58049066 : unsigned int shrinker_hits;
2098 58049066 : unsigned long queue_delay = 1;
2099 :
2100 58049066 : trace_xfs_inode_set_need_inactive(ip);
2101 58029056 : spin_lock(&ip->i_flags_lock);
2102 58111945 : ip->i_flags |= XFS_NEED_INACTIVE;
2103 58111945 : spin_unlock(&ip->i_flags_lock);
2104 :
2105 58119529 : gc = get_cpu_ptr(mp->m_inodegc);
2106 58103406 : llist_add(&ip->i_gclist, &gc->list);
2107 58090904 : items = READ_ONCE(gc->items);
2108 58090904 : WRITE_ONCE(gc->items, items + 1);
2109 58090904 : shrinker_hits = READ_ONCE(gc->shrinker_hits);
2110 :
2111 : /*
2112 : * We queue the work while holding the current CPU so that the work
2113 : * is scheduled to run on this CPU.
2114 : */
2115 116181808 : if (!xfs_is_inodegc_enabled(mp)) {
2116 0 : put_cpu_ptr(gc);
2117 0 : return;
2118 : }
2119 :
2120 58090904 : if (xfs_inodegc_want_queue_work(ip, items))
2121 15931018 : queue_delay = 0;
2122 :
2123 58064694 : trace_xfs_inodegc_queue(mp, __return_address);
2124 58056167 : mod_delayed_work_on(current_cpu(), mp->m_inodegc_wq, &gc->work,
2125 : queue_delay);
2126 58033399 : put_cpu_ptr(gc);
2127 :
2128 58070013 : if (xfs_inodegc_want_flush_work(ip, items, shrinker_hits)) {
2129 5638780 : trace_xfs_inodegc_throttle(mp, __return_address);
2130 5631216 : flush_delayed_work(&gc->work);
2131 : }
2132 : }
2133 :
2134 : /*
2135 : * Fold the dead CPU inodegc queue into the current CPUs queue.
2136 : */
2137 : void
2138 140 : xfs_inodegc_cpu_dead(
2139 : struct xfs_mount *mp,
2140 : unsigned int dead_cpu)
2141 : {
2142 140 : struct xfs_inodegc *dead_gc, *gc;
2143 140 : struct llist_node *first, *last;
2144 140 : unsigned int count = 0;
2145 :
2146 140 : dead_gc = per_cpu_ptr(mp->m_inodegc, dead_cpu);
2147 140 : cancel_delayed_work_sync(&dead_gc->work);
2148 :
2149 140 : if (llist_empty(&dead_gc->list))
2150 : return;
2151 :
2152 0 : first = dead_gc->list.first;
2153 0 : last = first;
2154 0 : while (last->next) {
2155 0 : last = last->next;
2156 0 : count++;
2157 : }
2158 0 : dead_gc->list.first = NULL;
2159 0 : dead_gc->items = 0;
2160 :
2161 : /* Add pending work to current CPU */
2162 0 : gc = get_cpu_ptr(mp->m_inodegc);
2163 0 : llist_add_batch(first, last, &gc->list);
2164 0 : count += READ_ONCE(gc->items);
2165 0 : WRITE_ONCE(gc->items, count);
2166 :
2167 0 : if (xfs_is_inodegc_enabled(mp)) {
2168 0 : trace_xfs_inodegc_queue(mp, __return_address);
2169 0 : mod_delayed_work_on(current_cpu(), mp->m_inodegc_wq, &gc->work,
2170 : 0);
2171 : }
2172 0 : put_cpu_ptr(gc);
2173 : }
2174 :
2175 : /*
2176 : * We set the inode flag atomically with the radix tree tag. Once we get tag
2177 : * lookups on the radix tree, this inode flag can go away.
2178 : *
2179 : * We always use background reclaim here because even if the inode is clean, it
2180 : * still may be under IO and hence we have wait for IO completion to occur
2181 : * before we can reclaim the inode. The background reclaim path handles this
2182 : * more efficiently than we can here, so simply let background reclaim tear down
2183 : * all inodes.
2184 : */
2185 : void
2186 1087148287 : xfs_inode_mark_reclaimable(
2187 : struct xfs_inode *ip)
2188 : {
2189 1087148287 : struct xfs_mount *mp = ip->i_mount;
2190 1087148287 : bool need_inactive;
2191 :
2192 1087148287 : XFS_STATS_INC(mp, vn_reclaim);
2193 :
2194 : /*
2195 : * We should never get here with any of the reclaim flags already set.
2196 : */
2197 2173649712 : ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_ALL_IRECLAIM_FLAGS));
2198 :
2199 1087427296 : need_inactive = xfs_inode_needs_inactive(ip);
2200 1086235480 : if (need_inactive) {
2201 58032369 : xfs_inodegc_queue(ip);
2202 58032369 : return;
2203 : }
2204 :
2205 : /* Going straight to reclaim, so drop the dquots. */
2206 1028203111 : xfs_qm_dqdetach(ip);
2207 1028202646 : xfs_inodegc_set_reclaimable(ip);
2208 : }
2209 :
2210 : /*
2211 : * Register a phony shrinker so that we can run background inodegc sooner when
2212 : * there's memory pressure. Inactivation does not itself free any memory but
2213 : * it does make inodes reclaimable, which eventually frees memory.
2214 : *
2215 : * The count function, seek value, and batch value are crafted to trigger the
2216 : * scan function during the second round of scanning. Hopefully this means
2217 : * that we reclaimed enough memory that initiating metadata transactions won't
2218 : * make things worse.
2219 : */
2220 : #define XFS_INODEGC_SHRINKER_COUNT (1UL << DEF_PRIORITY)
2221 : #define XFS_INODEGC_SHRINKER_BATCH ((XFS_INODEGC_SHRINKER_COUNT / 2) + 1)
2222 :
2223 : static unsigned long
2224 7034 : xfs_inodegc_shrinker_count(
2225 : struct shrinker *shrink,
2226 : struct shrink_control *sc)
2227 : {
2228 7034 : struct xfs_mount *mp = container_of(shrink, struct xfs_mount,
2229 : m_inodegc_shrinker);
2230 7034 : struct xfs_inodegc *gc;
2231 7034 : int cpu;
2232 :
2233 14068 : if (!xfs_is_inodegc_enabled(mp))
2234 : return 0;
2235 :
2236 35014 : for_each_online_cpu(cpu) {
2237 28038 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
2238 28038 : if (!llist_empty(&gc->list))
2239 : return XFS_INODEGC_SHRINKER_COUNT;
2240 : }
2241 :
2242 : return 0;
2243 : }
2244 :
2245 : static unsigned long
2246 22 : xfs_inodegc_shrinker_scan(
2247 : struct shrinker *shrink,
2248 : struct shrink_control *sc)
2249 : {
2250 22 : struct xfs_mount *mp = container_of(shrink, struct xfs_mount,
2251 : m_inodegc_shrinker);
2252 22 : struct xfs_inodegc *gc;
2253 22 : int cpu;
2254 22 : bool no_items = true;
2255 :
2256 44 : if (!xfs_is_inodegc_enabled(mp))
2257 : return SHRINK_STOP;
2258 :
2259 22 : trace_xfs_inodegc_shrinker_scan(mp, sc, __return_address);
2260 :
2261 132 : for_each_online_cpu(cpu) {
2262 88 : gc = per_cpu_ptr(mp->m_inodegc, cpu);
2263 88 : if (!llist_empty(&gc->list)) {
2264 29 : unsigned int h = READ_ONCE(gc->shrinker_hits);
2265 :
2266 29 : WRITE_ONCE(gc->shrinker_hits, h + 1);
2267 29 : mod_delayed_work_on(cpu, mp->m_inodegc_wq, &gc->work, 0);
2268 29 : no_items = false;
2269 : }
2270 : }
2271 :
2272 : /*
2273 : * If there are no inodes to inactivate, we don't want the shrinker
2274 : * to think there's deferred work to call us back about.
2275 : */
2276 22 : if (no_items)
2277 0 : return LONG_MAX;
2278 :
2279 : return SHRINK_STOP;
2280 : }
2281 :
2282 : /* Register a shrinker so we can accelerate inodegc and throttle queuing. */
2283 : int
2284 59289 : xfs_inodegc_register_shrinker(
2285 : struct xfs_mount *mp)
2286 : {
2287 59289 : struct shrinker *shrink = &mp->m_inodegc_shrinker;
2288 :
2289 59289 : shrink->count_objects = xfs_inodegc_shrinker_count;
2290 59289 : shrink->scan_objects = xfs_inodegc_shrinker_scan;
2291 59289 : shrink->seeks = 0;
2292 59289 : shrink->flags = SHRINKER_NONSLAB;
2293 59289 : shrink->batch = XFS_INODEGC_SHRINKER_BATCH;
2294 :
2295 59289 : return register_shrinker(shrink, "xfs-inodegc:%s", mp->m_super->s_id);
2296 : }
|
