blob: 990210fcb9c326f923c0c25a86557a0aba3bb40f [file] [log] [blame]
David Chinnerfe4fa4b2008-10-30 17:06:08 +11001/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
Dave Chinner6ca1c902013-08-12 20:49:26 +100020#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110023#include "xfs_sb.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110024#include "xfs_mount.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110025#include "xfs_inode.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110026#include "xfs_error.h"
Dave Chinner239880e2013-10-23 10:50:10 +110027#include "xfs_trans.h"
28#include "xfs_trans_priv.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110029#include "xfs_inode_item.h"
Christoph Hellwig7d095252009-06-08 15:33:32 +020030#include "xfs_quota.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000031#include "xfs_trace.h"
Dave Chinner6d8b79c2012-10-08 21:56:09 +110032#include "xfs_icache.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100033#include "xfs_bmap_util.h"
Brian Fosterdc06f3982014-07-24 19:49:28 +100034#include "xfs_dquot_item.h"
35#include "xfs_dquot.h"
Darrick J. Wong83104d42016-10-03 09:11:46 -070036#include "xfs_reflink.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110037
David Chinnera167b172008-10-30 17:06:18 +110038#include <linux/kthread.h>
39#include <linux/freezer.h>
40
Dave Chinner33479e02012-10-08 21:56:11 +110041/*
42 * Allocate and initialise an xfs_inode.
43 */
Dave Chinner638f44162013-08-30 10:23:45 +100044struct xfs_inode *
Dave Chinner33479e02012-10-08 21:56:11 +110045xfs_inode_alloc(
46 struct xfs_mount *mp,
47 xfs_ino_t ino)
48{
49 struct xfs_inode *ip;
50
51 /*
52 * if this didn't occur in transactions, we could use
53 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
54 * code up to do this anyway.
55 */
56 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
57 if (!ip)
58 return NULL;
59 if (inode_init_always(mp->m_super, VFS_I(ip))) {
60 kmem_zone_free(xfs_inode_zone, ip);
61 return NULL;
62 }
63
Dave Chinnerc19b3b052016-02-09 16:54:58 +110064 /* VFS doesn't initialise i_mode! */
65 VFS_I(ip)->i_mode = 0;
66
Bill O'Donnellff6d6af2015-10-12 18:21:22 +110067 XFS_STATS_INC(mp, vn_active);
Dave Chinner33479e02012-10-08 21:56:11 +110068 ASSERT(atomic_read(&ip->i_pincount) == 0);
Dave Chinner33479e02012-10-08 21:56:11 +110069 ASSERT(!xfs_isiflocked(ip));
70 ASSERT(ip->i_ino == 0);
71
Dave Chinner33479e02012-10-08 21:56:11 +110072 /* initialise the xfs inode */
73 ip->i_ino = ino;
74 ip->i_mount = mp;
75 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
76 ip->i_afp = NULL;
Darrick J. Wong3993bae2016-10-03 09:11:32 -070077 ip->i_cowfp = NULL;
78 ip->i_cnextents = 0;
79 ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
Dave Chinner33479e02012-10-08 21:56:11 +110080 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
81 ip->i_flags = 0;
82 ip->i_delayed_blks = 0;
Dave Chinnerf8d55aa0522016-02-09 16:54:58 +110083 memset(&ip->i_d, 0, sizeof(ip->i_d));
Dave Chinner33479e02012-10-08 21:56:11 +110084
85 return ip;
86}
87
88STATIC void
89xfs_inode_free_callback(
90 struct rcu_head *head)
91{
92 struct inode *inode = container_of(head, struct inode, i_rcu);
93 struct xfs_inode *ip = XFS_I(inode);
94
Dave Chinnerc19b3b052016-02-09 16:54:58 +110095 switch (VFS_I(ip)->i_mode & S_IFMT) {
Dave Chinner33479e02012-10-08 21:56:11 +110096 case S_IFREG:
97 case S_IFDIR:
98 case S_IFLNK:
99 xfs_idestroy_fork(ip, XFS_DATA_FORK);
100 break;
101 }
102
103 if (ip->i_afp)
104 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
Darrick J. Wong3993bae2016-10-03 09:11:32 -0700105 if (ip->i_cowfp)
106 xfs_idestroy_fork(ip, XFS_COW_FORK);
Dave Chinner33479e02012-10-08 21:56:11 +1100107
108 if (ip->i_itemp) {
109 ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL));
110 xfs_inode_item_destroy(ip);
111 ip->i_itemp = NULL;
112 }
113
Dave Chinner1f2dcfe2016-05-18 14:01:53 +1000114 kmem_zone_free(xfs_inode_zone, ip);
115}
116
Dave Chinner8a17d7d2016-05-18 14:09:12 +1000117static void
118__xfs_inode_free(
119 struct xfs_inode *ip)
120{
121 /* asserts to verify all state is correct here */
122 ASSERT(atomic_read(&ip->i_pincount) == 0);
Dave Chinner8a17d7d2016-05-18 14:09:12 +1000123 XFS_STATS_DEC(ip->i_mount, vn_active);
124
125 call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
126}
127
Dave Chinner1f2dcfe2016-05-18 14:01:53 +1000128void
129xfs_inode_free(
130 struct xfs_inode *ip)
131{
Brian Foster98efe8a2016-11-10 08:23:22 +1100132 ASSERT(!xfs_isiflocked(ip));
133
Dave Chinner33479e02012-10-08 21:56:11 +1100134 /*
135 * Because we use RCU freeing we need to ensure the inode always
136 * appears to be reclaimed with an invalid inode number when in the
137 * free state. The ip->i_flags_lock provides the barrier against lookup
138 * races.
139 */
140 spin_lock(&ip->i_flags_lock);
141 ip->i_flags = XFS_IRECLAIM;
142 ip->i_ino = 0;
143 spin_unlock(&ip->i_flags_lock);
144
Dave Chinner8a17d7d2016-05-18 14:09:12 +1000145 __xfs_inode_free(ip);
Dave Chinner33479e02012-10-08 21:56:11 +1100146}
147
148/*
Dave Chinnerad438c42016-05-18 14:20:08 +1000149 * Queue a new inode reclaim pass if there are reclaimable inodes and there
150 * isn't a reclaim pass already in progress. By default it runs every 5s based
151 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
152 * tunable, but that can be done if this method proves to be ineffective or too
153 * aggressive.
154 */
155static void
156xfs_reclaim_work_queue(
157 struct xfs_mount *mp)
158{
159
160 rcu_read_lock();
161 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
162 queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
163 msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
164 }
165 rcu_read_unlock();
166}
167
168/*
169 * This is a fast pass over the inode cache to try to get reclaim moving on as
170 * many inodes as possible in a short period of time. It kicks itself every few
171 * seconds, as well as being kicked by the inode cache shrinker when memory
172 * goes low. It scans as quickly as possible avoiding locked inodes or those
173 * already being flushed, and once done schedules a future pass.
174 */
175void
176xfs_reclaim_worker(
177 struct work_struct *work)
178{
179 struct xfs_mount *mp = container_of(to_delayed_work(work),
180 struct xfs_mount, m_reclaim_work);
181
182 xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
183 xfs_reclaim_work_queue(mp);
184}
185
186static void
187xfs_perag_set_reclaim_tag(
188 struct xfs_perag *pag)
189{
190 struct xfs_mount *mp = pag->pag_mount;
191
Brian Foster95989c42017-06-08 08:23:07 -0700192 lockdep_assert_held(&pag->pag_ici_lock);
Dave Chinnerad438c42016-05-18 14:20:08 +1000193 if (pag->pag_ici_reclaimable++)
194 return;
195
196 /* propagate the reclaim tag up into the perag radix tree */
197 spin_lock(&mp->m_perag_lock);
198 radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno,
199 XFS_ICI_RECLAIM_TAG);
200 spin_unlock(&mp->m_perag_lock);
201
202 /* schedule periodic background inode reclaim */
203 xfs_reclaim_work_queue(mp);
204
205 trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
206}
207
208static void
209xfs_perag_clear_reclaim_tag(
210 struct xfs_perag *pag)
211{
212 struct xfs_mount *mp = pag->pag_mount;
213
Brian Foster95989c42017-06-08 08:23:07 -0700214 lockdep_assert_held(&pag->pag_ici_lock);
Dave Chinnerad438c42016-05-18 14:20:08 +1000215 if (--pag->pag_ici_reclaimable)
216 return;
217
218 /* clear the reclaim tag from the perag radix tree */
219 spin_lock(&mp->m_perag_lock);
220 radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno,
221 XFS_ICI_RECLAIM_TAG);
222 spin_unlock(&mp->m_perag_lock);
223 trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
224}
225
226
227/*
228 * We set the inode flag atomically with the radix tree tag.
229 * Once we get tag lookups on the radix tree, this inode flag
230 * can go away.
231 */
232void
233xfs_inode_set_reclaim_tag(
234 struct xfs_inode *ip)
235{
236 struct xfs_mount *mp = ip->i_mount;
237 struct xfs_perag *pag;
238
239 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
240 spin_lock(&pag->pag_ici_lock);
241 spin_lock(&ip->i_flags_lock);
242
243 radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino),
244 XFS_ICI_RECLAIM_TAG);
245 xfs_perag_set_reclaim_tag(pag);
246 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
247
248 spin_unlock(&ip->i_flags_lock);
249 spin_unlock(&pag->pag_ici_lock);
250 xfs_perag_put(pag);
251}
252
253STATIC void
254xfs_inode_clear_reclaim_tag(
255 struct xfs_perag *pag,
256 xfs_ino_t ino)
257{
258 radix_tree_tag_clear(&pag->pag_ici_root,
259 XFS_INO_TO_AGINO(pag->pag_mount, ino),
260 XFS_ICI_RECLAIM_TAG);
261 xfs_perag_clear_reclaim_tag(pag);
262}
263
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700264static void
265xfs_inew_wait(
266 struct xfs_inode *ip)
267{
268 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT);
269 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT);
270
271 do {
272 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
273 if (!xfs_iflags_test(ip, XFS_INEW))
274 break;
275 schedule();
276 } while (true);
277 finish_wait(wq, &wait.wait);
278}
279
Dave Chinnerad438c42016-05-18 14:20:08 +1000280/*
Dave Chinner50997472016-02-09 16:54:58 +1100281 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
282 * part of the structure. This is made more complex by the fact we store
283 * information about the on-disk values in the VFS inode and so we can't just
Dave Chinner83e06f22016-02-09 16:54:58 +1100284 * overwrite the values unconditionally. Hence we save the parameters we
Dave Chinner50997472016-02-09 16:54:58 +1100285 * need to retain across reinitialisation, and rewrite them into the VFS inode
Dave Chinner83e06f22016-02-09 16:54:58 +1100286 * after reinitialisation even if it fails.
Dave Chinner50997472016-02-09 16:54:58 +1100287 */
288static int
289xfs_reinit_inode(
290 struct xfs_mount *mp,
291 struct inode *inode)
292{
293 int error;
Dave Chinner54d7b5c2016-02-09 16:54:58 +1100294 uint32_t nlink = inode->i_nlink;
Dave Chinner9e9a2672016-02-09 16:54:58 +1100295 uint32_t generation = inode->i_generation;
Dave Chinner83e06f22016-02-09 16:54:58 +1100296 uint64_t version = inode->i_version;
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100297 umode_t mode = inode->i_mode;
Dave Chinner50997472016-02-09 16:54:58 +1100298
299 error = inode_init_always(mp->m_super, inode);
300
Dave Chinner54d7b5c2016-02-09 16:54:58 +1100301 set_nlink(inode, nlink);
Dave Chinner9e9a2672016-02-09 16:54:58 +1100302 inode->i_generation = generation;
Dave Chinner83e06f22016-02-09 16:54:58 +1100303 inode->i_version = version;
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100304 inode->i_mode = mode;
Dave Chinner50997472016-02-09 16:54:58 +1100305 return error;
306}
307
308/*
Dave Chinner33479e02012-10-08 21:56:11 +1100309 * Check the validity of the inode we just found it the cache
310 */
311static int
312xfs_iget_cache_hit(
313 struct xfs_perag *pag,
314 struct xfs_inode *ip,
315 xfs_ino_t ino,
316 int flags,
317 int lock_flags) __releases(RCU)
318{
319 struct inode *inode = VFS_I(ip);
320 struct xfs_mount *mp = ip->i_mount;
321 int error;
322
323 /*
324 * check for re-use of an inode within an RCU grace period due to the
325 * radix tree nodes not being updated yet. We monitor for this by
326 * setting the inode number to zero before freeing the inode structure.
327 * If the inode has been reallocated and set up, then the inode number
328 * will not match, so check for that, too.
329 */
330 spin_lock(&ip->i_flags_lock);
331 if (ip->i_ino != ino) {
332 trace_xfs_iget_skip(ip);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100333 XFS_STATS_INC(mp, xs_ig_frecycle);
Dave Chinner24513372014-06-25 14:58:08 +1000334 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100335 goto out_error;
336 }
337
338
339 /*
340 * If we are racing with another cache hit that is currently
341 * instantiating this inode or currently recycling it out of
342 * reclaimabe state, wait for the initialisation to complete
343 * before continuing.
344 *
345 * XXX(hch): eventually we should do something equivalent to
346 * wait_on_inode to wait for these flags to be cleared
347 * instead of polling for it.
348 */
349 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
350 trace_xfs_iget_skip(ip);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100351 XFS_STATS_INC(mp, xs_ig_frecycle);
Dave Chinner24513372014-06-25 14:58:08 +1000352 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100353 goto out_error;
354 }
355
356 /*
357 * If lookup is racing with unlink return an error immediately.
358 */
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100359 if (VFS_I(ip)->i_mode == 0 && !(flags & XFS_IGET_CREATE)) {
Dave Chinner24513372014-06-25 14:58:08 +1000360 error = -ENOENT;
Dave Chinner33479e02012-10-08 21:56:11 +1100361 goto out_error;
362 }
363
364 /*
365 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
366 * Need to carefully get it back into useable state.
367 */
368 if (ip->i_flags & XFS_IRECLAIMABLE) {
369 trace_xfs_iget_reclaim(ip);
370
371 /*
372 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
373 * from stomping over us while we recycle the inode. We can't
374 * clear the radix tree reclaimable tag yet as it requires
375 * pag_ici_lock to be held exclusive.
376 */
377 ip->i_flags |= XFS_IRECLAIM;
378
379 spin_unlock(&ip->i_flags_lock);
380 rcu_read_unlock();
381
Dave Chinner50997472016-02-09 16:54:58 +1100382 error = xfs_reinit_inode(mp, inode);
Dave Chinner33479e02012-10-08 21:56:11 +1100383 if (error) {
Brian Foster756baca2017-04-26 08:30:39 -0700384 bool wake;
Dave Chinner33479e02012-10-08 21:56:11 +1100385 /*
386 * Re-initializing the inode failed, and we are in deep
387 * trouble. Try to re-add it to the reclaim list.
388 */
389 rcu_read_lock();
390 spin_lock(&ip->i_flags_lock);
Brian Foster756baca2017-04-26 08:30:39 -0700391 wake = !!__xfs_iflags_test(ip, XFS_INEW);
Dave Chinner33479e02012-10-08 21:56:11 +1100392 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
Brian Foster756baca2017-04-26 08:30:39 -0700393 if (wake)
394 wake_up_bit(&ip->i_flags, __XFS_INEW_BIT);
Dave Chinner33479e02012-10-08 21:56:11 +1100395 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
396 trace_xfs_iget_reclaim_fail(ip);
397 goto out_error;
398 }
399
400 spin_lock(&pag->pag_ici_lock);
401 spin_lock(&ip->i_flags_lock);
402
403 /*
404 * Clear the per-lifetime state in the inode as we are now
405 * effectively a new inode and need to return to the initial
406 * state before reuse occurs.
407 */
408 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
409 ip->i_flags |= XFS_INEW;
Dave Chinner545c0882016-05-18 14:11:41 +1000410 xfs_inode_clear_reclaim_tag(pag, ip->i_ino);
Dave Chinner33479e02012-10-08 21:56:11 +1100411 inode->i_state = I_NEW;
412
Christoph Hellwig65523212016-11-30 14:33:25 +1100413 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
414 init_rwsem(&inode->i_rwsem);
Dave Chinner33479e02012-10-08 21:56:11 +1100415
416 spin_unlock(&ip->i_flags_lock);
417 spin_unlock(&pag->pag_ici_lock);
418 } else {
419 /* If the VFS inode is being torn down, pause and try again. */
420 if (!igrab(inode)) {
421 trace_xfs_iget_skip(ip);
Dave Chinner24513372014-06-25 14:58:08 +1000422 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100423 goto out_error;
424 }
425
426 /* We've got a live one. */
427 spin_unlock(&ip->i_flags_lock);
428 rcu_read_unlock();
429 trace_xfs_iget_hit(ip);
430 }
431
432 if (lock_flags != 0)
433 xfs_ilock(ip, lock_flags);
434
435 xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100436 XFS_STATS_INC(mp, xs_ig_found);
Dave Chinner33479e02012-10-08 21:56:11 +1100437
438 return 0;
439
440out_error:
441 spin_unlock(&ip->i_flags_lock);
442 rcu_read_unlock();
443 return error;
444}
445
446
447static int
448xfs_iget_cache_miss(
449 struct xfs_mount *mp,
450 struct xfs_perag *pag,
451 xfs_trans_t *tp,
452 xfs_ino_t ino,
453 struct xfs_inode **ipp,
454 int flags,
455 int lock_flags)
456{
457 struct xfs_inode *ip;
458 int error;
459 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
460 int iflags;
461
462 ip = xfs_inode_alloc(mp, ino);
463 if (!ip)
Dave Chinner24513372014-06-25 14:58:08 +1000464 return -ENOMEM;
Dave Chinner33479e02012-10-08 21:56:11 +1100465
466 error = xfs_iread(mp, tp, ip, flags);
467 if (error)
468 goto out_destroy;
469
470 trace_xfs_iget_miss(ip);
471
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100472 if ((VFS_I(ip)->i_mode == 0) && !(flags & XFS_IGET_CREATE)) {
Dave Chinner24513372014-06-25 14:58:08 +1000473 error = -ENOENT;
Dave Chinner33479e02012-10-08 21:56:11 +1100474 goto out_destroy;
475 }
476
477 /*
478 * Preload the radix tree so we can insert safely under the
479 * write spinlock. Note that we cannot sleep inside the preload
480 * region. Since we can be called from transaction context, don't
481 * recurse into the file system.
482 */
483 if (radix_tree_preload(GFP_NOFS)) {
Dave Chinner24513372014-06-25 14:58:08 +1000484 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100485 goto out_destroy;
486 }
487
488 /*
489 * Because the inode hasn't been added to the radix-tree yet it can't
490 * be found by another thread, so we can do the non-sleeping lock here.
491 */
492 if (lock_flags) {
493 if (!xfs_ilock_nowait(ip, lock_flags))
494 BUG();
495 }
496
497 /*
498 * These values must be set before inserting the inode into the radix
499 * tree as the moment it is inserted a concurrent lookup (allowed by the
500 * RCU locking mechanism) can find it and that lookup must see that this
501 * is an inode currently under construction (i.e. that XFS_INEW is set).
502 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
503 * memory barrier that ensures this detection works correctly at lookup
504 * time.
505 */
506 iflags = XFS_INEW;
507 if (flags & XFS_IGET_DONTCACHE)
508 iflags |= XFS_IDONTCACHE;
Chandra Seetharaman113a5682013-06-27 17:25:07 -0500509 ip->i_udquot = NULL;
510 ip->i_gdquot = NULL;
Chandra Seetharaman92f8ff72013-07-11 00:00:40 -0500511 ip->i_pdquot = NULL;
Dave Chinner33479e02012-10-08 21:56:11 +1100512 xfs_iflags_set(ip, iflags);
513
514 /* insert the new inode */
515 spin_lock(&pag->pag_ici_lock);
516 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
517 if (unlikely(error)) {
518 WARN_ON(error != -EEXIST);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100519 XFS_STATS_INC(mp, xs_ig_dup);
Dave Chinner24513372014-06-25 14:58:08 +1000520 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100521 goto out_preload_end;
522 }
523 spin_unlock(&pag->pag_ici_lock);
524 radix_tree_preload_end();
525
526 *ipp = ip;
527 return 0;
528
529out_preload_end:
530 spin_unlock(&pag->pag_ici_lock);
531 radix_tree_preload_end();
532 if (lock_flags)
533 xfs_iunlock(ip, lock_flags);
534out_destroy:
535 __destroy_inode(VFS_I(ip));
536 xfs_inode_free(ip);
537 return error;
538}
539
540/*
541 * Look up an inode by number in the given file system.
542 * The inode is looked up in the cache held in each AG.
543 * If the inode is found in the cache, initialise the vfs inode
544 * if necessary.
545 *
546 * If it is not in core, read it in from the file system's device,
547 * add it to the cache and initialise the vfs inode.
548 *
549 * The inode is locked according to the value of the lock_flags parameter.
550 * This flag parameter indicates how and if the inode's IO lock and inode lock
551 * should be taken.
552 *
553 * mp -- the mount point structure for the current file system. It points
554 * to the inode hash table.
555 * tp -- a pointer to the current transaction if there is one. This is
556 * simply passed through to the xfs_iread() call.
557 * ino -- the number of the inode desired. This is the unique identifier
558 * within the file system for the inode being requested.
559 * lock_flags -- flags indicating how to lock the inode. See the comment
560 * for xfs_ilock() for a list of valid values.
561 */
562int
563xfs_iget(
564 xfs_mount_t *mp,
565 xfs_trans_t *tp,
566 xfs_ino_t ino,
567 uint flags,
568 uint lock_flags,
569 xfs_inode_t **ipp)
570{
571 xfs_inode_t *ip;
572 int error;
573 xfs_perag_t *pag;
574 xfs_agino_t agino;
575
576 /*
577 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
578 * doesn't get freed while it's being referenced during a
579 * radix tree traversal here. It assumes this function
580 * aqcuires only the ILOCK (and therefore it has no need to
581 * involve the IOLOCK in this synchronization).
582 */
583 ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
584
585 /* reject inode numbers outside existing AGs */
586 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
Dave Chinner24513372014-06-25 14:58:08 +1000587 return -EINVAL;
Dave Chinner33479e02012-10-08 21:56:11 +1100588
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100589 XFS_STATS_INC(mp, xs_ig_attempts);
Lucas Stach8774cf82015-08-28 14:50:56 +1000590
Dave Chinner33479e02012-10-08 21:56:11 +1100591 /* get the perag structure and ensure that it's inode capable */
592 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
593 agino = XFS_INO_TO_AGINO(mp, ino);
594
595again:
596 error = 0;
597 rcu_read_lock();
598 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
599
600 if (ip) {
601 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
602 if (error)
603 goto out_error_or_again;
604 } else {
605 rcu_read_unlock();
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100606 XFS_STATS_INC(mp, xs_ig_missed);
Dave Chinner33479e02012-10-08 21:56:11 +1100607
608 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
609 flags, lock_flags);
610 if (error)
611 goto out_error_or_again;
612 }
613 xfs_perag_put(pag);
614
615 *ipp = ip;
616
617 /*
Dave Chinner58c90472015-02-23 22:38:08 +1100618 * If we have a real type for an on-disk inode, we can setup the inode
Dave Chinner33479e02012-10-08 21:56:11 +1100619 * now. If it's a new inode being created, xfs_ialloc will handle it.
620 */
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100621 if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
Dave Chinner58c90472015-02-23 22:38:08 +1100622 xfs_setup_existing_inode(ip);
Dave Chinner33479e02012-10-08 21:56:11 +1100623 return 0;
624
625out_error_or_again:
Dave Chinner24513372014-06-25 14:58:08 +1000626 if (error == -EAGAIN) {
Dave Chinner33479e02012-10-08 21:56:11 +1100627 delay(1);
628 goto again;
629 }
630 xfs_perag_put(pag);
631 return error;
632}
633
Dave Chinner78ae5252010-09-28 12:28:19 +1000634/*
635 * The inode lookup is done in batches to keep the amount of lock traffic and
636 * radix tree lookups to a minimum. The batch size is a trade off between
637 * lookup reduction and stack usage. This is in the reclaim path, so we can't
638 * be too greedy.
639 */
640#define XFS_LOOKUP_BATCH 32
641
Dave Chinnere13de952010-09-28 12:28:06 +1000642STATIC int
643xfs_inode_ag_walk_grab(
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700644 struct xfs_inode *ip,
645 int flags)
Dave Chinnere13de952010-09-28 12:28:06 +1000646{
647 struct inode *inode = VFS_I(ip);
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700648 bool newinos = !!(flags & XFS_AGITER_INEW_WAIT);
Dave Chinnere13de952010-09-28 12:28:06 +1000649
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100650 ASSERT(rcu_read_lock_held());
651
652 /*
653 * check for stale RCU freed inode
654 *
655 * If the inode has been reallocated, it doesn't matter if it's not in
656 * the AG we are walking - we are walking for writeback, so if it
657 * passes all the "valid inode" checks and is dirty, then we'll write
658 * it back anyway. If it has been reallocated and still being
659 * initialised, the XFS_INEW check below will catch it.
660 */
661 spin_lock(&ip->i_flags_lock);
662 if (!ip->i_ino)
663 goto out_unlock_noent;
664
665 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700666 if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) ||
667 __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM))
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100668 goto out_unlock_noent;
669 spin_unlock(&ip->i_flags_lock);
670
Dave Chinnere13de952010-09-28 12:28:06 +1000671 /* nothing to sync during shutdown */
672 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Dave Chinner24513372014-06-25 14:58:08 +1000673 return -EFSCORRUPTED;
Dave Chinnere13de952010-09-28 12:28:06 +1000674
Dave Chinnere13de952010-09-28 12:28:06 +1000675 /* If we can't grab the inode, it must on it's way to reclaim. */
676 if (!igrab(inode))
Dave Chinner24513372014-06-25 14:58:08 +1000677 return -ENOENT;
Dave Chinnere13de952010-09-28 12:28:06 +1000678
Dave Chinnere13de952010-09-28 12:28:06 +1000679 /* inode is valid */
680 return 0;
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100681
682out_unlock_noent:
683 spin_unlock(&ip->i_flags_lock);
Dave Chinner24513372014-06-25 14:58:08 +1000684 return -ENOENT;
Dave Chinnere13de952010-09-28 12:28:06 +1000685}
686
Dave Chinner75f3cb12009-06-08 15:35:14 +0200687STATIC int
688xfs_inode_ag_walk(
689 struct xfs_mount *mp,
Dave Chinner5017e972010-01-11 11:47:40 +0000690 struct xfs_perag *pag,
Eric Sandeene0094002014-04-14 19:04:19 +1000691 int (*execute)(struct xfs_inode *ip, int flags,
Brian Fostera454f742012-11-06 09:50:39 -0500692 void *args),
693 int flags,
694 void *args,
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700695 int tag,
696 int iter_flags)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200697{
Dave Chinner75f3cb12009-06-08 15:35:14 +0200698 uint32_t first_index;
699 int last_error = 0;
700 int skipped;
Dave Chinner65d0f202010-09-24 18:40:15 +1000701 int done;
Dave Chinner78ae5252010-09-28 12:28:19 +1000702 int nr_found;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200703
704restart:
Dave Chinner65d0f202010-09-24 18:40:15 +1000705 done = 0;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200706 skipped = 0;
707 first_index = 0;
Dave Chinner78ae5252010-09-28 12:28:19 +1000708 nr_found = 0;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200709 do {
Dave Chinner78ae5252010-09-28 12:28:19 +1000710 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
Dave Chinner75f3cb12009-06-08 15:35:14 +0200711 int error = 0;
Dave Chinner78ae5252010-09-28 12:28:19 +1000712 int i;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200713
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100714 rcu_read_lock();
Brian Fostera454f742012-11-06 09:50:39 -0500715
716 if (tag == -1)
717 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
Dave Chinner78ae5252010-09-28 12:28:19 +1000718 (void **)batch, first_index,
719 XFS_LOOKUP_BATCH);
Brian Fostera454f742012-11-06 09:50:39 -0500720 else
721 nr_found = radix_tree_gang_lookup_tag(
722 &pag->pag_ici_root,
723 (void **) batch, first_index,
724 XFS_LOOKUP_BATCH, tag);
725
Dave Chinner65d0f202010-09-24 18:40:15 +1000726 if (!nr_found) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100727 rcu_read_unlock();
Dave Chinner75f3cb12009-06-08 15:35:14 +0200728 break;
Dave Chinnerc8e20be2010-01-10 23:51:45 +0000729 }
Dave Chinner75f3cb12009-06-08 15:35:14 +0200730
Dave Chinner65d0f202010-09-24 18:40:15 +1000731 /*
Dave Chinner78ae5252010-09-28 12:28:19 +1000732 * Grab the inodes before we drop the lock. if we found
733 * nothing, nr == 0 and the loop will be skipped.
Dave Chinner65d0f202010-09-24 18:40:15 +1000734 */
Dave Chinner78ae5252010-09-28 12:28:19 +1000735 for (i = 0; i < nr_found; i++) {
736 struct xfs_inode *ip = batch[i];
Dave Chinner65d0f202010-09-24 18:40:15 +1000737
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700738 if (done || xfs_inode_ag_walk_grab(ip, iter_flags))
Dave Chinner78ae5252010-09-28 12:28:19 +1000739 batch[i] = NULL;
740
741 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100742 * Update the index for the next lookup. Catch
743 * overflows into the next AG range which can occur if
744 * we have inodes in the last block of the AG and we
745 * are currently pointing to the last inode.
746 *
747 * Because we may see inodes that are from the wrong AG
748 * due to RCU freeing and reallocation, only update the
749 * index if it lies in this AG. It was a race that lead
750 * us to see this inode, so another lookup from the
751 * same index will not find it again.
Dave Chinner78ae5252010-09-28 12:28:19 +1000752 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100753 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
754 continue;
Dave Chinner78ae5252010-09-28 12:28:19 +1000755 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
756 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
757 done = 1;
Dave Chinnere13de952010-09-28 12:28:06 +1000758 }
Dave Chinner78ae5252010-09-28 12:28:19 +1000759
760 /* unlock now we've grabbed the inodes. */
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100761 rcu_read_unlock();
Dave Chinnere13de952010-09-28 12:28:06 +1000762
Dave Chinner78ae5252010-09-28 12:28:19 +1000763 for (i = 0; i < nr_found; i++) {
764 if (!batch[i])
765 continue;
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700766 if ((iter_flags & XFS_AGITER_INEW_WAIT) &&
767 xfs_iflags_test(batch[i], XFS_INEW))
768 xfs_inew_wait(batch[i]);
Eric Sandeene0094002014-04-14 19:04:19 +1000769 error = execute(batch[i], flags, args);
Dave Chinner78ae5252010-09-28 12:28:19 +1000770 IRELE(batch[i]);
Dave Chinner24513372014-06-25 14:58:08 +1000771 if (error == -EAGAIN) {
Dave Chinner78ae5252010-09-28 12:28:19 +1000772 skipped++;
773 continue;
774 }
Dave Chinner24513372014-06-25 14:58:08 +1000775 if (error && last_error != -EFSCORRUPTED)
Dave Chinner78ae5252010-09-28 12:28:19 +1000776 last_error = error;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200777 }
Dave Chinnerc8e20be2010-01-10 23:51:45 +0000778
779 /* bail out if the filesystem is corrupted. */
Dave Chinner24513372014-06-25 14:58:08 +1000780 if (error == -EFSCORRUPTED)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200781 break;
782
Dave Chinner8daaa832011-07-08 14:14:46 +1000783 cond_resched();
784
Dave Chinner78ae5252010-09-28 12:28:19 +1000785 } while (nr_found && !done);
Dave Chinner75f3cb12009-06-08 15:35:14 +0200786
787 if (skipped) {
788 delay(1);
789 goto restart;
790 }
Dave Chinner75f3cb12009-06-08 15:35:14 +0200791 return last_error;
792}
793
Brian Foster579b62f2012-11-06 09:50:47 -0500794/*
795 * Background scanning to trim post-EOF preallocated space. This is queued
Dwight Engenb9fe5052013-08-15 14:08:02 -0400796 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
Brian Foster579b62f2012-11-06 09:50:47 -0500797 */
Brian Fosterfa5a4f52016-06-21 11:53:28 +1000798void
Brian Foster579b62f2012-11-06 09:50:47 -0500799xfs_queue_eofblocks(
800 struct xfs_mount *mp)
801{
802 rcu_read_lock();
803 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
804 queue_delayed_work(mp->m_eofblocks_workqueue,
805 &mp->m_eofblocks_work,
806 msecs_to_jiffies(xfs_eofb_secs * 1000));
807 rcu_read_unlock();
808}
809
810void
811xfs_eofblocks_worker(
812 struct work_struct *work)
813{
814 struct xfs_mount *mp = container_of(to_delayed_work(work),
815 struct xfs_mount, m_eofblocks_work);
816 xfs_icache_free_eofblocks(mp, NULL);
817 xfs_queue_eofblocks(mp);
818}
819
Darrick J. Wong83104d42016-10-03 09:11:46 -0700820/*
821 * Background scanning to trim preallocated CoW space. This is queued
822 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
823 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
824 */
825STATIC void
826xfs_queue_cowblocks(
827 struct xfs_mount *mp)
828{
829 rcu_read_lock();
830 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
831 queue_delayed_work(mp->m_eofblocks_workqueue,
832 &mp->m_cowblocks_work,
833 msecs_to_jiffies(xfs_cowb_secs * 1000));
834 rcu_read_unlock();
835}
836
837void
838xfs_cowblocks_worker(
839 struct work_struct *work)
840{
841 struct xfs_mount *mp = container_of(to_delayed_work(work),
842 struct xfs_mount, m_cowblocks_work);
843 xfs_icache_free_cowblocks(mp, NULL);
844 xfs_queue_cowblocks(mp);
845}
846
Christoph Hellwigfe588ed2009-06-08 15:35:27 +0200847int
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700848xfs_inode_ag_iterator_flags(
Dave Chinner75f3cb12009-06-08 15:35:14 +0200849 struct xfs_mount *mp,
Eric Sandeene0094002014-04-14 19:04:19 +1000850 int (*execute)(struct xfs_inode *ip, int flags,
Brian Fostera454f742012-11-06 09:50:39 -0500851 void *args),
852 int flags,
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700853 void *args,
854 int iter_flags)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200855{
Dave Chinner16fd5362010-07-20 09:43:39 +1000856 struct xfs_perag *pag;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200857 int error = 0;
858 int last_error = 0;
859 xfs_agnumber_t ag;
860
Dave Chinner16fd5362010-07-20 09:43:39 +1000861 ag = 0;
Dave Chinner65d0f202010-09-24 18:40:15 +1000862 while ((pag = xfs_perag_get(mp, ag))) {
863 ag = pag->pag_agno + 1;
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700864 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1,
865 iter_flags);
Brian Fostera454f742012-11-06 09:50:39 -0500866 xfs_perag_put(pag);
867 if (error) {
868 last_error = error;
Dave Chinner24513372014-06-25 14:58:08 +1000869 if (error == -EFSCORRUPTED)
Brian Fostera454f742012-11-06 09:50:39 -0500870 break;
871 }
872 }
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000873 return last_error;
Brian Fostera454f742012-11-06 09:50:39 -0500874}
875
876int
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700877xfs_inode_ag_iterator(
878 struct xfs_mount *mp,
879 int (*execute)(struct xfs_inode *ip, int flags,
880 void *args),
881 int flags,
882 void *args)
883{
884 return xfs_inode_ag_iterator_flags(mp, execute, flags, args, 0);
885}
886
887int
Brian Fostera454f742012-11-06 09:50:39 -0500888xfs_inode_ag_iterator_tag(
889 struct xfs_mount *mp,
Eric Sandeene0094002014-04-14 19:04:19 +1000890 int (*execute)(struct xfs_inode *ip, int flags,
Brian Fostera454f742012-11-06 09:50:39 -0500891 void *args),
892 int flags,
893 void *args,
894 int tag)
895{
896 struct xfs_perag *pag;
897 int error = 0;
898 int last_error = 0;
899 xfs_agnumber_t ag;
900
901 ag = 0;
902 while ((pag = xfs_perag_get_tag(mp, ag, tag))) {
903 ag = pag->pag_agno + 1;
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700904 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag,
905 0);
Dave Chinner5017e972010-01-11 11:47:40 +0000906 xfs_perag_put(pag);
Dave Chinner75f3cb12009-06-08 15:35:14 +0200907 if (error) {
908 last_error = error;
Dave Chinner24513372014-06-25 14:58:08 +1000909 if (error == -EFSCORRUPTED)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200910 break;
911 }
912 }
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000913 return last_error;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200914}
915
David Chinner76bf1052008-10-30 17:16:21 +1100916/*
Dave Chinnere3a20c02010-09-24 19:51:50 +1000917 * Grab the inode for reclaim exclusively.
918 * Return 0 if we grabbed it, non-zero otherwise.
919 */
920STATIC int
921xfs_reclaim_inode_grab(
922 struct xfs_inode *ip,
923 int flags)
924{
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100925 ASSERT(rcu_read_lock_held());
926
927 /* quick check for stale RCU freed inode */
928 if (!ip->i_ino)
929 return 1;
Dave Chinnere3a20c02010-09-24 19:51:50 +1000930
931 /*
Christoph Hellwig474fce02011-12-18 20:00:09 +0000932 * If we are asked for non-blocking operation, do unlocked checks to
933 * see if the inode already is being flushed or in reclaim to avoid
934 * lock traffic.
Dave Chinnere3a20c02010-09-24 19:51:50 +1000935 */
936 if ((flags & SYNC_TRYLOCK) &&
Christoph Hellwig474fce02011-12-18 20:00:09 +0000937 __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM))
Dave Chinnere3a20c02010-09-24 19:51:50 +1000938 return 1;
Dave Chinnere3a20c02010-09-24 19:51:50 +1000939
940 /*
941 * The radix tree lock here protects a thread in xfs_iget from racing
942 * with us starting reclaim on the inode. Once we have the
943 * XFS_IRECLAIM flag set it will not touch us.
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100944 *
945 * Due to RCU lookup, we may find inodes that have been freed and only
946 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
947 * aren't candidates for reclaim at all, so we must check the
948 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
Dave Chinnere3a20c02010-09-24 19:51:50 +1000949 */
950 spin_lock(&ip->i_flags_lock);
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100951 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
952 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
953 /* not a reclaim candidate. */
Dave Chinnere3a20c02010-09-24 19:51:50 +1000954 spin_unlock(&ip->i_flags_lock);
955 return 1;
956 }
957 __xfs_iflags_set(ip, XFS_IRECLAIM);
958 spin_unlock(&ip->i_flags_lock);
959 return 0;
960}
961
962/*
Christoph Hellwig8a480882012-04-23 15:58:35 +1000963 * Inodes in different states need to be treated differently. The following
964 * table lists the inode states and the reclaim actions necessary:
Dave Chinner777df5a2010-02-06 12:37:26 +1100965 *
966 * inode state iflush ret required action
967 * --------------- ---------- ---------------
968 * bad - reclaim
969 * shutdown EIO unpin and reclaim
970 * clean, unpinned 0 reclaim
971 * stale, unpinned 0 reclaim
Dave Chinnerc8543632010-02-06 12:39:36 +1100972 * clean, pinned(*) 0 requeue
973 * stale, pinned EAGAIN requeue
Christoph Hellwig8a480882012-04-23 15:58:35 +1000974 * dirty, async - requeue
975 * dirty, sync 0 reclaim
Dave Chinner777df5a2010-02-06 12:37:26 +1100976 *
977 * (*) dgc: I don't think the clean, pinned state is possible but it gets
978 * handled anyway given the order of checks implemented.
979 *
Dave Chinnerc8543632010-02-06 12:39:36 +1100980 * Also, because we get the flush lock first, we know that any inode that has
981 * been flushed delwri has had the flush completed by the time we check that
Christoph Hellwig8a480882012-04-23 15:58:35 +1000982 * the inode is clean.
Dave Chinnerc8543632010-02-06 12:39:36 +1100983 *
Christoph Hellwig8a480882012-04-23 15:58:35 +1000984 * Note that because the inode is flushed delayed write by AIL pushing, the
985 * flush lock may already be held here and waiting on it can result in very
986 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
987 * the caller should push the AIL first before trying to reclaim inodes to
988 * minimise the amount of time spent waiting. For background relaim, we only
989 * bother to reclaim clean inodes anyway.
Dave Chinnerc8543632010-02-06 12:39:36 +1100990 *
Dave Chinner777df5a2010-02-06 12:37:26 +1100991 * Hence the order of actions after gaining the locks should be:
992 * bad => reclaim
993 * shutdown => unpin and reclaim
Christoph Hellwig8a480882012-04-23 15:58:35 +1000994 * pinned, async => requeue
Dave Chinnerc8543632010-02-06 12:39:36 +1100995 * pinned, sync => unpin
Dave Chinner777df5a2010-02-06 12:37:26 +1100996 * stale => reclaim
997 * clean => reclaim
Christoph Hellwig8a480882012-04-23 15:58:35 +1000998 * dirty, async => requeue
Dave Chinnerc8543632010-02-06 12:39:36 +1100999 * dirty, sync => flush, wait and reclaim
Dave Chinner777df5a2010-02-06 12:37:26 +11001000 */
Dave Chinner75f3cb12009-06-08 15:35:14 +02001001STATIC int
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001002xfs_reclaim_inode(
Dave Chinner75f3cb12009-06-08 15:35:14 +02001003 struct xfs_inode *ip,
1004 struct xfs_perag *pag,
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001005 int sync_mode)
David Chinner7a3be022008-10-30 17:37:37 +11001006{
Christoph Hellwig4c468192012-04-23 15:58:36 +10001007 struct xfs_buf *bp = NULL;
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001008 xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
Christoph Hellwig4c468192012-04-23 15:58:36 +10001009 int error;
Dave Chinner777df5a2010-02-06 12:37:26 +11001010
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001011restart:
1012 error = 0;
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001013 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinnerc8543632010-02-06 12:39:36 +11001014 if (!xfs_iflock_nowait(ip)) {
1015 if (!(sync_mode & SYNC_WAIT))
1016 goto out;
1017 xfs_iflock(ip);
1018 }
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001019
Dave Chinner777df5a2010-02-06 12:37:26 +11001020 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1021 xfs_iunpin_wait(ip);
Brian Foster98efe8a2016-11-10 08:23:22 +11001022 /* xfs_iflush_abort() drops the flush lock */
Dave Chinner04913fd2012-04-23 15:58:41 +10001023 xfs_iflush_abort(ip, false);
Dave Chinner777df5a2010-02-06 12:37:26 +11001024 goto reclaim;
1025 }
Dave Chinnerc8543632010-02-06 12:39:36 +11001026 if (xfs_ipincount(ip)) {
Christoph Hellwig8a480882012-04-23 15:58:35 +10001027 if (!(sync_mode & SYNC_WAIT))
1028 goto out_ifunlock;
Dave Chinner777df5a2010-02-06 12:37:26 +11001029 xfs_iunpin_wait(ip);
Dave Chinnerc8543632010-02-06 12:39:36 +11001030 }
Brian Foster98efe8a2016-11-10 08:23:22 +11001031 if (xfs_iflags_test(ip, XFS_ISTALE) || xfs_inode_clean(ip)) {
1032 xfs_ifunlock(ip);
Dave Chinner777df5a2010-02-06 12:37:26 +11001033 goto reclaim;
Brian Foster98efe8a2016-11-10 08:23:22 +11001034 }
Dave Chinner777df5a2010-02-06 12:37:26 +11001035
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001036 /*
Christoph Hellwig8a480882012-04-23 15:58:35 +10001037 * Never flush out dirty data during non-blocking reclaim, as it would
1038 * just contend with AIL pushing trying to do the same job.
1039 */
1040 if (!(sync_mode & SYNC_WAIT))
1041 goto out_ifunlock;
1042
1043 /*
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001044 * Now we have an inode that needs flushing.
1045 *
Christoph Hellwig4c468192012-04-23 15:58:36 +10001046 * Note that xfs_iflush will never block on the inode buffer lock, as
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001047 * xfs_ifree_cluster() can lock the inode buffer before it locks the
Christoph Hellwig4c468192012-04-23 15:58:36 +10001048 * ip->i_lock, and we are doing the exact opposite here. As a result,
Christoph Hellwig475ee412012-07-03 12:21:22 -04001049 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1050 * result in an ABBA deadlock with xfs_ifree_cluster().
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001051 *
1052 * As xfs_ifree_cluser() must gather all inodes that are active in the
1053 * cache to mark them stale, if we hit this case we don't actually want
1054 * to do IO here - we want the inode marked stale so we can simply
Christoph Hellwig4c468192012-04-23 15:58:36 +10001055 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1056 * inode, back off and try again. Hopefully the next pass through will
1057 * see the stale flag set on the inode.
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001058 */
Christoph Hellwig4c468192012-04-23 15:58:36 +10001059 error = xfs_iflush(ip, &bp);
Dave Chinner24513372014-06-25 14:58:08 +10001060 if (error == -EAGAIN) {
Christoph Hellwig8a480882012-04-23 15:58:35 +10001061 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1062 /* backoff longer than in xfs_ifree_cluster */
1063 delay(2);
1064 goto restart;
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001065 }
Dave Chinnerc8543632010-02-06 12:39:36 +11001066
Christoph Hellwig4c468192012-04-23 15:58:36 +10001067 if (!error) {
1068 error = xfs_bwrite(bp);
1069 xfs_buf_relse(bp);
1070 }
1071
Dave Chinner777df5a2010-02-06 12:37:26 +11001072reclaim:
Brian Foster98efe8a2016-11-10 08:23:22 +11001073 ASSERT(!xfs_isiflocked(ip));
1074
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001075 /*
1076 * Because we use RCU freeing we need to ensure the inode always appears
1077 * to be reclaimed with an invalid inode number when in the free state.
Brian Foster98efe8a2016-11-10 08:23:22 +11001078 * We do this as early as possible under the ILOCK so that
1079 * xfs_iflush_cluster() can be guaranteed to detect races with us here.
1080 * By doing this, we guarantee that once xfs_iflush_cluster has locked
1081 * XFS_ILOCK that it will see either a valid, flushable inode that will
1082 * serialise correctly, or it will see a clean (and invalid) inode that
1083 * it can skip.
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001084 */
1085 spin_lock(&ip->i_flags_lock);
1086 ip->i_flags = XFS_IRECLAIM;
1087 ip->i_ino = 0;
1088 spin_unlock(&ip->i_flags_lock);
1089
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001090 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001091
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11001092 XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001093 /*
1094 * Remove the inode from the per-AG radix tree.
1095 *
1096 * Because radix_tree_delete won't complain even if the item was never
1097 * added to the tree assert that it's been there before to catch
1098 * problems with the inode life time early on.
1099 */
Dave Chinner1a427ab2010-12-16 17:08:41 +11001100 spin_lock(&pag->pag_ici_lock);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001101 if (!radix_tree_delete(&pag->pag_ici_root,
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001102 XFS_INO_TO_AGINO(ip->i_mount, ino)))
Dave Chinner2f11fea2010-07-20 17:53:25 +10001103 ASSERT(0);
Dave Chinner545c0882016-05-18 14:11:41 +10001104 xfs_perag_clear_reclaim_tag(pag);
Dave Chinner1a427ab2010-12-16 17:08:41 +11001105 spin_unlock(&pag->pag_ici_lock);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001106
1107 /*
1108 * Here we do an (almost) spurious inode lock in order to coordinate
1109 * with inode cache radix tree lookups. This is because the lookup
1110 * can reference the inodes in the cache without taking references.
1111 *
1112 * We make that OK here by ensuring that we wait until the inode is
Alex Elderad637a12012-02-16 22:01:00 +00001113 * unlocked after the lookup before we go ahead and free it.
Dave Chinner2f11fea2010-07-20 17:53:25 +10001114 */
Alex Elderad637a12012-02-16 22:01:00 +00001115 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001116 xfs_qm_dqdetach(ip);
Alex Elderad637a12012-02-16 22:01:00 +00001117 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001118
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001119 __xfs_inode_free(ip);
Alex Elderad637a12012-02-16 22:01:00 +00001120 return error;
Christoph Hellwig8a480882012-04-23 15:58:35 +10001121
1122out_ifunlock:
1123 xfs_ifunlock(ip);
1124out:
1125 xfs_iflags_clear(ip, XFS_IRECLAIM);
1126 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1127 /*
Dave Chinner24513372014-06-25 14:58:08 +10001128 * We could return -EAGAIN here to make reclaim rescan the inode tree in
Christoph Hellwig8a480882012-04-23 15:58:35 +10001129 * a short while. However, this just burns CPU time scanning the tree
Dave Chinner58896082012-10-08 21:56:05 +11001130 * waiting for IO to complete and the reclaim work never goes back to
1131 * the idle state. Instead, return 0 to let the next scheduled
1132 * background reclaim attempt to reclaim the inode again.
Christoph Hellwig8a480882012-04-23 15:58:35 +10001133 */
1134 return 0;
David Chinner7a3be022008-10-30 17:37:37 +11001135}
1136
Dave Chinner65d0f202010-09-24 18:40:15 +10001137/*
1138 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1139 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1140 * then a shut down during filesystem unmount reclaim walk leak all the
1141 * unreclaimed inodes.
1142 */
Dave Chinner33479e02012-10-08 21:56:11 +11001143STATIC int
Dave Chinner65d0f202010-09-24 18:40:15 +10001144xfs_reclaim_inodes_ag(
1145 struct xfs_mount *mp,
1146 int flags,
1147 int *nr_to_scan)
1148{
1149 struct xfs_perag *pag;
1150 int error = 0;
1151 int last_error = 0;
1152 xfs_agnumber_t ag;
Dave Chinner69b491c2010-09-27 11:09:51 +10001153 int trylock = flags & SYNC_TRYLOCK;
1154 int skipped;
Dave Chinner65d0f202010-09-24 18:40:15 +10001155
Dave Chinner69b491c2010-09-27 11:09:51 +10001156restart:
Dave Chinner65d0f202010-09-24 18:40:15 +10001157 ag = 0;
Dave Chinner69b491c2010-09-27 11:09:51 +10001158 skipped = 0;
Dave Chinner65d0f202010-09-24 18:40:15 +10001159 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1160 unsigned long first_index = 0;
1161 int done = 0;
Dave Chinnere3a20c02010-09-24 19:51:50 +10001162 int nr_found = 0;
Dave Chinner65d0f202010-09-24 18:40:15 +10001163
1164 ag = pag->pag_agno + 1;
1165
Dave Chinner69b491c2010-09-27 11:09:51 +10001166 if (trylock) {
1167 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
1168 skipped++;
Dave Chinnerf83282a2010-11-08 08:55:04 +00001169 xfs_perag_put(pag);
Dave Chinner69b491c2010-09-27 11:09:51 +10001170 continue;
1171 }
1172 first_index = pag->pag_ici_reclaim_cursor;
1173 } else
1174 mutex_lock(&pag->pag_ici_reclaim_lock);
1175
Dave Chinner65d0f202010-09-24 18:40:15 +10001176 do {
Dave Chinnere3a20c02010-09-24 19:51:50 +10001177 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1178 int i;
Dave Chinner65d0f202010-09-24 18:40:15 +10001179
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001180 rcu_read_lock();
Dave Chinnere3a20c02010-09-24 19:51:50 +10001181 nr_found = radix_tree_gang_lookup_tag(
1182 &pag->pag_ici_root,
1183 (void **)batch, first_index,
1184 XFS_LOOKUP_BATCH,
Dave Chinner65d0f202010-09-24 18:40:15 +10001185 XFS_ICI_RECLAIM_TAG);
1186 if (!nr_found) {
Dave Chinnerb2232212011-05-06 02:54:04 +00001187 done = 1;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001188 rcu_read_unlock();
Dave Chinner65d0f202010-09-24 18:40:15 +10001189 break;
1190 }
1191
1192 /*
Dave Chinnere3a20c02010-09-24 19:51:50 +10001193 * Grab the inodes before we drop the lock. if we found
1194 * nothing, nr == 0 and the loop will be skipped.
Dave Chinner65d0f202010-09-24 18:40:15 +10001195 */
Dave Chinnere3a20c02010-09-24 19:51:50 +10001196 for (i = 0; i < nr_found; i++) {
1197 struct xfs_inode *ip = batch[i];
Dave Chinner65d0f202010-09-24 18:40:15 +10001198
Dave Chinnere3a20c02010-09-24 19:51:50 +10001199 if (done || xfs_reclaim_inode_grab(ip, flags))
1200 batch[i] = NULL;
Dave Chinner65d0f202010-09-24 18:40:15 +10001201
Dave Chinnere3a20c02010-09-24 19:51:50 +10001202 /*
1203 * Update the index for the next lookup. Catch
1204 * overflows into the next AG range which can
1205 * occur if we have inodes in the last block of
1206 * the AG and we are currently pointing to the
1207 * last inode.
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001208 *
1209 * Because we may see inodes that are from the
1210 * wrong AG due to RCU freeing and
1211 * reallocation, only update the index if it
1212 * lies in this AG. It was a race that lead us
1213 * to see this inode, so another lookup from
1214 * the same index will not find it again.
Dave Chinnere3a20c02010-09-24 19:51:50 +10001215 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001216 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1217 pag->pag_agno)
1218 continue;
Dave Chinnere3a20c02010-09-24 19:51:50 +10001219 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1220 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1221 done = 1;
1222 }
1223
1224 /* unlock now we've grabbed the inodes. */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001225 rcu_read_unlock();
Dave Chinnere3a20c02010-09-24 19:51:50 +10001226
1227 for (i = 0; i < nr_found; i++) {
1228 if (!batch[i])
1229 continue;
1230 error = xfs_reclaim_inode(batch[i], pag, flags);
Dave Chinner24513372014-06-25 14:58:08 +10001231 if (error && last_error != -EFSCORRUPTED)
Dave Chinnere3a20c02010-09-24 19:51:50 +10001232 last_error = error;
1233 }
1234
1235 *nr_to_scan -= XFS_LOOKUP_BATCH;
1236
Dave Chinner8daaa832011-07-08 14:14:46 +10001237 cond_resched();
1238
Dave Chinnere3a20c02010-09-24 19:51:50 +10001239 } while (nr_found && !done && *nr_to_scan > 0);
Dave Chinner65d0f202010-09-24 18:40:15 +10001240
Dave Chinner69b491c2010-09-27 11:09:51 +10001241 if (trylock && !done)
1242 pag->pag_ici_reclaim_cursor = first_index;
1243 else
1244 pag->pag_ici_reclaim_cursor = 0;
1245 mutex_unlock(&pag->pag_ici_reclaim_lock);
Dave Chinner65d0f202010-09-24 18:40:15 +10001246 xfs_perag_put(pag);
1247 }
Dave Chinner69b491c2010-09-27 11:09:51 +10001248
1249 /*
1250 * if we skipped any AG, and we still have scan count remaining, do
1251 * another pass this time using blocking reclaim semantics (i.e
1252 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1253 * ensure that when we get more reclaimers than AGs we block rather
1254 * than spin trying to execute reclaim.
1255 */
Dave Chinner8daaa832011-07-08 14:14:46 +10001256 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
Dave Chinner69b491c2010-09-27 11:09:51 +10001257 trylock = 0;
1258 goto restart;
1259 }
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001260 return last_error;
Dave Chinner65d0f202010-09-24 18:40:15 +10001261}
1262
David Chinnerfce08f22008-10-30 17:37:03 +11001263int
David Chinner1dc33182008-10-30 17:37:15 +11001264xfs_reclaim_inodes(
David Chinnerfce08f22008-10-30 17:37:03 +11001265 xfs_mount_t *mp,
David Chinnerfce08f22008-10-30 17:37:03 +11001266 int mode)
1267{
Dave Chinner65d0f202010-09-24 18:40:15 +10001268 int nr_to_scan = INT_MAX;
1269
1270 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
Dave Chinner9bf729c2010-04-29 09:55:50 +10001271}
1272
1273/*
Dave Chinner8daaa832011-07-08 14:14:46 +10001274 * Scan a certain number of inodes for reclaim.
Dave Chinnera7b339f2011-04-08 12:45:07 +10001275 *
1276 * When called we make sure that there is a background (fast) inode reclaim in
Dave Chinner8daaa832011-07-08 14:14:46 +10001277 * progress, while we will throttle the speed of reclaim via doing synchronous
Dave Chinnera7b339f2011-04-08 12:45:07 +10001278 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1279 * them to be cleaned, which we hope will not be very long due to the
1280 * background walker having already kicked the IO off on those dirty inodes.
Dave Chinner9bf729c2010-04-29 09:55:50 +10001281 */
Dave Chinner0a234c62013-08-28 10:17:57 +10001282long
Dave Chinner8daaa832011-07-08 14:14:46 +10001283xfs_reclaim_inodes_nr(
1284 struct xfs_mount *mp,
1285 int nr_to_scan)
Dave Chinner9bf729c2010-04-29 09:55:50 +10001286{
Dave Chinner8daaa832011-07-08 14:14:46 +10001287 /* kick background reclaimer and push the AIL */
Dave Chinner58896082012-10-08 21:56:05 +11001288 xfs_reclaim_work_queue(mp);
Dave Chinner8daaa832011-07-08 14:14:46 +10001289 xfs_ail_push_all(mp->m_ail);
Dave Chinner9bf729c2010-04-29 09:55:50 +10001290
Dave Chinner0a234c62013-08-28 10:17:57 +10001291 return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
Dave Chinner8daaa832011-07-08 14:14:46 +10001292}
Dave Chinnera7b339f2011-04-08 12:45:07 +10001293
Dave Chinner8daaa832011-07-08 14:14:46 +10001294/*
1295 * Return the number of reclaimable inodes in the filesystem for
1296 * the shrinker to determine how much to reclaim.
1297 */
1298int
1299xfs_reclaim_inodes_count(
1300 struct xfs_mount *mp)
1301{
1302 struct xfs_perag *pag;
1303 xfs_agnumber_t ag = 0;
1304 int reclaimable = 0;
Dave Chinner9bf729c2010-04-29 09:55:50 +10001305
Dave Chinner65d0f202010-09-24 18:40:15 +10001306 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1307 ag = pag->pag_agno + 1;
Dave Chinner70e60ce2010-07-20 08:07:02 +10001308 reclaimable += pag->pag_ici_reclaimable;
1309 xfs_perag_put(pag);
Dave Chinner9bf729c2010-04-29 09:55:50 +10001310 }
Dave Chinner9bf729c2010-04-29 09:55:50 +10001311 return reclaimable;
1312}
1313
Brian Foster41176a62012-11-06 09:50:42 -05001314STATIC int
Brian Foster3e3f9f52012-11-07 12:21:13 -05001315xfs_inode_match_id(
1316 struct xfs_inode *ip,
1317 struct xfs_eofblocks *eofb)
1318{
Dwight Engenb9fe5052013-08-15 14:08:02 -04001319 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1320 !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
Brian Foster1b556042012-11-06 09:50:45 -05001321 return 0;
Brian Foster3e3f9f52012-11-07 12:21:13 -05001322
Dwight Engenb9fe5052013-08-15 14:08:02 -04001323 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1324 !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
Brian Foster1b556042012-11-06 09:50:45 -05001325 return 0;
1326
Dwight Engenb9fe5052013-08-15 14:08:02 -04001327 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
Brian Foster1b556042012-11-06 09:50:45 -05001328 xfs_get_projid(ip) != eofb->eof_prid)
1329 return 0;
1330
1331 return 1;
Brian Foster3e3f9f52012-11-07 12:21:13 -05001332}
1333
Brian Fosterf4526392014-07-24 19:44:28 +10001334/*
1335 * A union-based inode filtering algorithm. Process the inode if any of the
1336 * criteria match. This is for global/internal scans only.
1337 */
1338STATIC int
1339xfs_inode_match_id_union(
1340 struct xfs_inode *ip,
1341 struct xfs_eofblocks *eofb)
1342{
1343 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1344 uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1345 return 1;
1346
1347 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1348 gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1349 return 1;
1350
1351 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1352 xfs_get_projid(ip) == eofb->eof_prid)
1353 return 1;
1354
1355 return 0;
1356}
1357
Brian Foster3e3f9f52012-11-07 12:21:13 -05001358STATIC int
Brian Foster41176a62012-11-06 09:50:42 -05001359xfs_inode_free_eofblocks(
1360 struct xfs_inode *ip,
Brian Foster41176a62012-11-06 09:50:42 -05001361 int flags,
1362 void *args)
1363{
Brian Fostera36b9262017-01-27 23:22:55 -08001364 int ret = 0;
Brian Foster3e3f9f52012-11-07 12:21:13 -05001365 struct xfs_eofblocks *eofb = args;
Brian Fosterf4526392014-07-24 19:44:28 +10001366 int match;
Brian Foster5400da72014-07-24 19:40:22 +10001367
Brian Foster41176a62012-11-06 09:50:42 -05001368 if (!xfs_can_free_eofblocks(ip, false)) {
1369 /* inode could be preallocated or append-only */
1370 trace_xfs_inode_free_eofblocks_invalid(ip);
1371 xfs_inode_clear_eofblocks_tag(ip);
1372 return 0;
1373 }
1374
1375 /*
1376 * If the mapping is dirty the operation can block and wait for some
1377 * time. Unless we are waiting, skip it.
1378 */
1379 if (!(flags & SYNC_WAIT) &&
1380 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1381 return 0;
1382
Brian Foster00ca79a2012-11-07 12:21:14 -05001383 if (eofb) {
Brian Fosterf4526392014-07-24 19:44:28 +10001384 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1385 match = xfs_inode_match_id_union(ip, eofb);
1386 else
1387 match = xfs_inode_match_id(ip, eofb);
1388 if (!match)
Brian Foster00ca79a2012-11-07 12:21:14 -05001389 return 0;
1390
1391 /* skip the inode if the file size is too small */
1392 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1393 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1394 return 0;
1395 }
Brian Foster3e3f9f52012-11-07 12:21:13 -05001396
Brian Fostera36b9262017-01-27 23:22:55 -08001397 /*
1398 * If the caller is waiting, return -EAGAIN to keep the background
1399 * scanner moving and revisit the inode in a subsequent pass.
1400 */
Brian Fosterc3155092017-01-27 23:22:56 -08001401 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
Brian Fostera36b9262017-01-27 23:22:55 -08001402 if (flags & SYNC_WAIT)
1403 ret = -EAGAIN;
1404 return ret;
1405 }
1406 ret = xfs_free_eofblocks(ip);
Brian Fosterc3155092017-01-27 23:22:56 -08001407 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Brian Foster41176a62012-11-06 09:50:42 -05001408
1409 return ret;
1410}
1411
Darrick J. Wong83104d42016-10-03 09:11:46 -07001412static int
1413__xfs_icache_free_eofblocks(
Brian Foster41176a62012-11-06 09:50:42 -05001414 struct xfs_mount *mp,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001415 struct xfs_eofblocks *eofb,
1416 int (*execute)(struct xfs_inode *ip, int flags,
1417 void *args),
1418 int tag)
Brian Foster41176a62012-11-06 09:50:42 -05001419{
Brian Foster8ca149d2012-11-07 12:21:12 -05001420 int flags = SYNC_TRYLOCK;
1421
1422 if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC))
1423 flags = SYNC_WAIT;
1424
Darrick J. Wong83104d42016-10-03 09:11:46 -07001425 return xfs_inode_ag_iterator_tag(mp, execute, flags,
1426 eofb, tag);
1427}
1428
1429int
1430xfs_icache_free_eofblocks(
1431 struct xfs_mount *mp,
1432 struct xfs_eofblocks *eofb)
1433{
1434 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_eofblocks,
1435 XFS_ICI_EOFBLOCKS_TAG);
Brian Foster41176a62012-11-06 09:50:42 -05001436}
1437
Brian Fosterdc06f3982014-07-24 19:49:28 +10001438/*
1439 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1440 * multiple quotas, we don't know exactly which quota caused an allocation
1441 * failure. We make a best effort by including each quota under low free space
1442 * conditions (less than 1% free space) in the scan.
1443 */
Darrick J. Wong83104d42016-10-03 09:11:46 -07001444static int
1445__xfs_inode_free_quota_eofblocks(
1446 struct xfs_inode *ip,
1447 int (*execute)(struct xfs_mount *mp,
1448 struct xfs_eofblocks *eofb))
Brian Fosterdc06f3982014-07-24 19:49:28 +10001449{
1450 int scan = 0;
1451 struct xfs_eofblocks eofb = {0};
1452 struct xfs_dquot *dq;
1453
Brian Fosterdc06f3982014-07-24 19:49:28 +10001454 /*
Brian Fosterc3155092017-01-27 23:22:56 -08001455 * Run a sync scan to increase effectiveness and use the union filter to
Brian Fosterdc06f3982014-07-24 19:49:28 +10001456 * cover all applicable quotas in a single scan.
1457 */
Brian Fosterdc06f3982014-07-24 19:49:28 +10001458 eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC;
1459
1460 if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) {
1461 dq = xfs_inode_dquot(ip, XFS_DQ_USER);
1462 if (dq && xfs_dquot_lowsp(dq)) {
1463 eofb.eof_uid = VFS_I(ip)->i_uid;
1464 eofb.eof_flags |= XFS_EOF_FLAGS_UID;
1465 scan = 1;
1466 }
1467 }
1468
1469 if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) {
1470 dq = xfs_inode_dquot(ip, XFS_DQ_GROUP);
1471 if (dq && xfs_dquot_lowsp(dq)) {
1472 eofb.eof_gid = VFS_I(ip)->i_gid;
1473 eofb.eof_flags |= XFS_EOF_FLAGS_GID;
1474 scan = 1;
1475 }
1476 }
1477
1478 if (scan)
Darrick J. Wong83104d42016-10-03 09:11:46 -07001479 execute(ip->i_mount, &eofb);
Brian Fosterdc06f3982014-07-24 19:49:28 +10001480
1481 return scan;
1482}
1483
Darrick J. Wong83104d42016-10-03 09:11:46 -07001484int
1485xfs_inode_free_quota_eofblocks(
1486 struct xfs_inode *ip)
1487{
1488 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks);
1489}
1490
1491static void
1492__xfs_inode_set_eofblocks_tag(
1493 xfs_inode_t *ip,
1494 void (*execute)(struct xfs_mount *mp),
1495 void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1496 int error, unsigned long caller_ip),
1497 int tag)
Brian Foster27b52862012-11-06 09:50:38 -05001498{
1499 struct xfs_mount *mp = ip->i_mount;
1500 struct xfs_perag *pag;
1501 int tagged;
1502
Christoph Hellwig85a6e762016-09-19 11:09:48 +10001503 /*
1504 * Don't bother locking the AG and looking up in the radix trees
1505 * if we already know that we have the tag set.
1506 */
1507 if (ip->i_flags & XFS_IEOFBLOCKS)
1508 return;
1509 spin_lock(&ip->i_flags_lock);
1510 ip->i_flags |= XFS_IEOFBLOCKS;
1511 spin_unlock(&ip->i_flags_lock);
1512
Brian Foster27b52862012-11-06 09:50:38 -05001513 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1514 spin_lock(&pag->pag_ici_lock);
Brian Foster27b52862012-11-06 09:50:38 -05001515
Darrick J. Wong83104d42016-10-03 09:11:46 -07001516 tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
Brian Foster27b52862012-11-06 09:50:38 -05001517 radix_tree_tag_set(&pag->pag_ici_root,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001518 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
Brian Foster27b52862012-11-06 09:50:38 -05001519 if (!tagged) {
1520 /* propagate the eofblocks tag up into the perag radix tree */
1521 spin_lock(&ip->i_mount->m_perag_lock);
1522 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
1523 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
Darrick J. Wong83104d42016-10-03 09:11:46 -07001524 tag);
Brian Foster27b52862012-11-06 09:50:38 -05001525 spin_unlock(&ip->i_mount->m_perag_lock);
1526
Brian Foster579b62f2012-11-06 09:50:47 -05001527 /* kick off background trimming */
Darrick J. Wong83104d42016-10-03 09:11:46 -07001528 execute(ip->i_mount);
Brian Foster579b62f2012-11-06 09:50:47 -05001529
Darrick J. Wong83104d42016-10-03 09:11:46 -07001530 set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
Brian Foster27b52862012-11-06 09:50:38 -05001531 }
1532
1533 spin_unlock(&pag->pag_ici_lock);
1534 xfs_perag_put(pag);
1535}
1536
1537void
Darrick J. Wong83104d42016-10-03 09:11:46 -07001538xfs_inode_set_eofblocks_tag(
Brian Foster27b52862012-11-06 09:50:38 -05001539 xfs_inode_t *ip)
1540{
Darrick J. Wong83104d42016-10-03 09:11:46 -07001541 trace_xfs_inode_set_eofblocks_tag(ip);
1542 return __xfs_inode_set_eofblocks_tag(ip, xfs_queue_eofblocks,
1543 trace_xfs_perag_set_eofblocks,
1544 XFS_ICI_EOFBLOCKS_TAG);
1545}
1546
1547static void
1548__xfs_inode_clear_eofblocks_tag(
1549 xfs_inode_t *ip,
1550 void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1551 int error, unsigned long caller_ip),
1552 int tag)
1553{
Brian Foster27b52862012-11-06 09:50:38 -05001554 struct xfs_mount *mp = ip->i_mount;
1555 struct xfs_perag *pag;
1556
Christoph Hellwig85a6e762016-09-19 11:09:48 +10001557 spin_lock(&ip->i_flags_lock);
1558 ip->i_flags &= ~XFS_IEOFBLOCKS;
1559 spin_unlock(&ip->i_flags_lock);
1560
Brian Foster27b52862012-11-06 09:50:38 -05001561 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1562 spin_lock(&pag->pag_ici_lock);
Brian Foster27b52862012-11-06 09:50:38 -05001563
1564 radix_tree_tag_clear(&pag->pag_ici_root,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001565 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1566 if (!radix_tree_tagged(&pag->pag_ici_root, tag)) {
Brian Foster27b52862012-11-06 09:50:38 -05001567 /* clear the eofblocks tag from the perag radix tree */
1568 spin_lock(&ip->i_mount->m_perag_lock);
1569 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
1570 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
Darrick J. Wong83104d42016-10-03 09:11:46 -07001571 tag);
Brian Foster27b52862012-11-06 09:50:38 -05001572 spin_unlock(&ip->i_mount->m_perag_lock);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001573 clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
Brian Foster27b52862012-11-06 09:50:38 -05001574 }
1575
1576 spin_unlock(&pag->pag_ici_lock);
1577 xfs_perag_put(pag);
1578}
1579
Darrick J. Wong83104d42016-10-03 09:11:46 -07001580void
1581xfs_inode_clear_eofblocks_tag(
1582 xfs_inode_t *ip)
1583{
1584 trace_xfs_inode_clear_eofblocks_tag(ip);
1585 return __xfs_inode_clear_eofblocks_tag(ip,
1586 trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG);
1587}
1588
1589/*
1590 * Automatic CoW Reservation Freeing
1591 *
1592 * These functions automatically garbage collect leftover CoW reservations
1593 * that were made on behalf of a cowextsize hint when we start to run out
1594 * of quota or when the reservations sit around for too long. If the file
1595 * has dirty pages or is undergoing writeback, its CoW reservations will
1596 * be retained.
1597 *
1598 * The actual garbage collection piggybacks off the same code that runs
1599 * the speculative EOF preallocation garbage collector.
1600 */
1601STATIC int
1602xfs_inode_free_cowblocks(
1603 struct xfs_inode *ip,
1604 int flags,
1605 void *args)
1606{
1607 int ret;
1608 struct xfs_eofblocks *eofb = args;
Darrick J. Wong83104d42016-10-03 09:11:46 -07001609 int match;
Brian Foster39937232016-11-08 12:53:33 +11001610 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001611
Brian Foster39937232016-11-08 12:53:33 +11001612 /*
1613 * Just clear the tag if we have an empty cow fork or none at all. It's
1614 * possible the inode was fully unshared since it was originally tagged.
1615 */
1616 if (!xfs_is_reflink_inode(ip) || !ifp->if_bytes) {
Darrick J. Wong83104d42016-10-03 09:11:46 -07001617 trace_xfs_inode_free_cowblocks_invalid(ip);
1618 xfs_inode_clear_cowblocks_tag(ip);
1619 return 0;
1620 }
1621
1622 /*
1623 * If the mapping is dirty or under writeback we cannot touch the
1624 * CoW fork. Leave it alone if we're in the midst of a directio.
1625 */
Christoph Hellwiga1b7a4d2017-01-03 18:39:33 -08001626 if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1627 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
Darrick J. Wong83104d42016-10-03 09:11:46 -07001628 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1629 atomic_read(&VFS_I(ip)->i_dio_count))
1630 return 0;
1631
1632 if (eofb) {
1633 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1634 match = xfs_inode_match_id_union(ip, eofb);
1635 else
1636 match = xfs_inode_match_id(ip, eofb);
1637 if (!match)
1638 return 0;
1639
1640 /* skip the inode if the file size is too small */
1641 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1642 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1643 return 0;
Darrick J. Wong83104d42016-10-03 09:11:46 -07001644 }
1645
1646 /* Free the CoW blocks */
Brian Fosterc3155092017-01-27 23:22:56 -08001647 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1648 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001649
Christoph Hellwig3802a342017-03-07 16:45:58 -08001650 ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001651
Brian Fosterc3155092017-01-27 23:22:56 -08001652 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1653 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001654
1655 return ret;
1656}
1657
1658int
1659xfs_icache_free_cowblocks(
1660 struct xfs_mount *mp,
1661 struct xfs_eofblocks *eofb)
1662{
1663 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_cowblocks,
1664 XFS_ICI_COWBLOCKS_TAG);
1665}
1666
1667int
1668xfs_inode_free_quota_cowblocks(
1669 struct xfs_inode *ip)
1670{
1671 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks);
1672}
1673
1674void
1675xfs_inode_set_cowblocks_tag(
1676 xfs_inode_t *ip)
1677{
Brian Foster7b7381f2016-10-24 14:21:00 +11001678 trace_xfs_inode_set_cowblocks_tag(ip);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001679 return __xfs_inode_set_eofblocks_tag(ip, xfs_queue_cowblocks,
Brian Foster7b7381f2016-10-24 14:21:00 +11001680 trace_xfs_perag_set_cowblocks,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001681 XFS_ICI_COWBLOCKS_TAG);
1682}
1683
1684void
1685xfs_inode_clear_cowblocks_tag(
1686 xfs_inode_t *ip)
1687{
Brian Foster7b7381f2016-10-24 14:21:00 +11001688 trace_xfs_inode_clear_cowblocks_tag(ip);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001689 return __xfs_inode_clear_eofblocks_tag(ip,
Brian Foster7b7381f2016-10-24 14:21:00 +11001690 trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001691}