blob: 245483cc282ba0032f5f0a16649d5b7dbd45cffb [file] [log] [blame]
Dave Chinner0b61f8a2018-06-05 19:42:14 -07001// SPDX-License-Identifier: GPL-2.0
David Chinnerfe4fa4b2008-10-30 17:06:08 +11002/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
David Chinnerfe4fa4b2008-10-30 17:06:08 +11005 */
6#include "xfs.h"
7#include "xfs_fs.h"
Dave Chinner6ca1c902013-08-12 20:49:26 +10008#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +11009#include "xfs_log_format.h"
10#include "xfs_trans_resv.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110011#include "xfs_sb.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110012#include "xfs_mount.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110013#include "xfs_inode.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110014#include "xfs_error.h"
Dave Chinner239880e2013-10-23 10:50:10 +110015#include "xfs_trans.h"
16#include "xfs_trans_priv.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110017#include "xfs_inode_item.h"
Christoph Hellwig7d095252009-06-08 15:33:32 +020018#include "xfs_quota.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000019#include "xfs_trace.h"
Dave Chinner6d8b79c2012-10-08 21:56:09 +110020#include "xfs_icache.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100021#include "xfs_bmap_util.h"
Brian Fosterdc06f3982014-07-24 19:49:28 +100022#include "xfs_dquot_item.h"
23#include "xfs_dquot.h"
Darrick J. Wong83104d42016-10-03 09:11:46 -070024#include "xfs_reflink.h"
David Chinnerfe4fa4b2008-10-30 17:06:08 +110025
David Chinnera167b172008-10-30 17:06:18 +110026#include <linux/kthread.h>
27#include <linux/freezer.h>
Jeff Laytonf0e28282017-12-11 06:35:19 -050028#include <linux/iversion.h>
David Chinnera167b172008-10-30 17:06:18 +110029
Dave Chinner33479e02012-10-08 21:56:11 +110030/*
31 * Allocate and initialise an xfs_inode.
32 */
Dave Chinner638f44162013-08-30 10:23:45 +100033struct xfs_inode *
Dave Chinner33479e02012-10-08 21:56:11 +110034xfs_inode_alloc(
35 struct xfs_mount *mp,
36 xfs_ino_t ino)
37{
38 struct xfs_inode *ip;
39
40 /*
41 * if this didn't occur in transactions, we could use
42 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
43 * code up to do this anyway.
44 */
45 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
46 if (!ip)
47 return NULL;
48 if (inode_init_always(mp->m_super, VFS_I(ip))) {
49 kmem_zone_free(xfs_inode_zone, ip);
50 return NULL;
51 }
52
Dave Chinnerc19b3b052016-02-09 16:54:58 +110053 /* VFS doesn't initialise i_mode! */
54 VFS_I(ip)->i_mode = 0;
55
Bill O'Donnellff6d6af2015-10-12 18:21:22 +110056 XFS_STATS_INC(mp, vn_active);
Dave Chinner33479e02012-10-08 21:56:11 +110057 ASSERT(atomic_read(&ip->i_pincount) == 0);
Dave Chinner33479e02012-10-08 21:56:11 +110058 ASSERT(!xfs_isiflocked(ip));
59 ASSERT(ip->i_ino == 0);
60
Dave Chinner33479e02012-10-08 21:56:11 +110061 /* initialise the xfs inode */
62 ip->i_ino = ino;
63 ip->i_mount = mp;
64 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
65 ip->i_afp = NULL;
Darrick J. Wong3993bae2016-10-03 09:11:32 -070066 ip->i_cowfp = NULL;
67 ip->i_cnextents = 0;
68 ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
Christoph Hellwig3ba738d2018-07-17 16:51:50 -070069 memset(&ip->i_df, 0, sizeof(ip->i_df));
Dave Chinner33479e02012-10-08 21:56:11 +110070 ip->i_flags = 0;
71 ip->i_delayed_blks = 0;
Dave Chinnerf8d55aa0522016-02-09 16:54:58 +110072 memset(&ip->i_d, 0, sizeof(ip->i_d));
Dave Chinner33479e02012-10-08 21:56:11 +110073
74 return ip;
75}
76
77STATIC void
78xfs_inode_free_callback(
79 struct rcu_head *head)
80{
81 struct inode *inode = container_of(head, struct inode, i_rcu);
82 struct xfs_inode *ip = XFS_I(inode);
83
Dave Chinnerc19b3b052016-02-09 16:54:58 +110084 switch (VFS_I(ip)->i_mode & S_IFMT) {
Dave Chinner33479e02012-10-08 21:56:11 +110085 case S_IFREG:
86 case S_IFDIR:
87 case S_IFLNK:
88 xfs_idestroy_fork(ip, XFS_DATA_FORK);
89 break;
90 }
91
92 if (ip->i_afp)
93 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
Darrick J. Wong3993bae2016-10-03 09:11:32 -070094 if (ip->i_cowfp)
95 xfs_idestroy_fork(ip, XFS_COW_FORK);
Dave Chinner33479e02012-10-08 21:56:11 +110096
97 if (ip->i_itemp) {
Dave Chinner22525c12018-05-09 07:47:34 -070098 ASSERT(!test_bit(XFS_LI_IN_AIL,
99 &ip->i_itemp->ili_item.li_flags));
Dave Chinner33479e02012-10-08 21:56:11 +1100100 xfs_inode_item_destroy(ip);
101 ip->i_itemp = NULL;
102 }
103
Dave Chinner1f2dcfe2016-05-18 14:01:53 +1000104 kmem_zone_free(xfs_inode_zone, ip);
105}
106
Dave Chinner8a17d7d2016-05-18 14:09:12 +1000107static void
108__xfs_inode_free(
109 struct xfs_inode *ip)
110{
111 /* asserts to verify all state is correct here */
112 ASSERT(atomic_read(&ip->i_pincount) == 0);
Dave Chinner8a17d7d2016-05-18 14:09:12 +1000113 XFS_STATS_DEC(ip->i_mount, vn_active);
114
115 call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
116}
117
Dave Chinner1f2dcfe2016-05-18 14:01:53 +1000118void
119xfs_inode_free(
120 struct xfs_inode *ip)
121{
Brian Foster98efe8a2016-11-10 08:23:22 +1100122 ASSERT(!xfs_isiflocked(ip));
123
Dave Chinner33479e02012-10-08 21:56:11 +1100124 /*
125 * Because we use RCU freeing we need to ensure the inode always
126 * appears to be reclaimed with an invalid inode number when in the
127 * free state. The ip->i_flags_lock provides the barrier against lookup
128 * races.
129 */
130 spin_lock(&ip->i_flags_lock);
131 ip->i_flags = XFS_IRECLAIM;
132 ip->i_ino = 0;
133 spin_unlock(&ip->i_flags_lock);
134
Dave Chinner8a17d7d2016-05-18 14:09:12 +1000135 __xfs_inode_free(ip);
Dave Chinner33479e02012-10-08 21:56:11 +1100136}
137
138/*
Dave Chinnerad438c42016-05-18 14:20:08 +1000139 * Queue a new inode reclaim pass if there are reclaimable inodes and there
140 * isn't a reclaim pass already in progress. By default it runs every 5s based
141 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
142 * tunable, but that can be done if this method proves to be ineffective or too
143 * aggressive.
144 */
145static void
146xfs_reclaim_work_queue(
147 struct xfs_mount *mp)
148{
149
150 rcu_read_lock();
151 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
152 queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
153 msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
154 }
155 rcu_read_unlock();
156}
157
158/*
159 * This is a fast pass over the inode cache to try to get reclaim moving on as
160 * many inodes as possible in a short period of time. It kicks itself every few
161 * seconds, as well as being kicked by the inode cache shrinker when memory
162 * goes low. It scans as quickly as possible avoiding locked inodes or those
163 * already being flushed, and once done schedules a future pass.
164 */
165void
166xfs_reclaim_worker(
167 struct work_struct *work)
168{
169 struct xfs_mount *mp = container_of(to_delayed_work(work),
170 struct xfs_mount, m_reclaim_work);
171
172 xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
173 xfs_reclaim_work_queue(mp);
174}
175
176static void
177xfs_perag_set_reclaim_tag(
178 struct xfs_perag *pag)
179{
180 struct xfs_mount *mp = pag->pag_mount;
181
Brian Foster95989c42017-06-08 08:23:07 -0700182 lockdep_assert_held(&pag->pag_ici_lock);
Dave Chinnerad438c42016-05-18 14:20:08 +1000183 if (pag->pag_ici_reclaimable++)
184 return;
185
186 /* propagate the reclaim tag up into the perag radix tree */
187 spin_lock(&mp->m_perag_lock);
188 radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno,
189 XFS_ICI_RECLAIM_TAG);
190 spin_unlock(&mp->m_perag_lock);
191
192 /* schedule periodic background inode reclaim */
193 xfs_reclaim_work_queue(mp);
194
195 trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
196}
197
198static void
199xfs_perag_clear_reclaim_tag(
200 struct xfs_perag *pag)
201{
202 struct xfs_mount *mp = pag->pag_mount;
203
Brian Foster95989c42017-06-08 08:23:07 -0700204 lockdep_assert_held(&pag->pag_ici_lock);
Dave Chinnerad438c42016-05-18 14:20:08 +1000205 if (--pag->pag_ici_reclaimable)
206 return;
207
208 /* clear the reclaim tag from the perag radix tree */
209 spin_lock(&mp->m_perag_lock);
210 radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno,
211 XFS_ICI_RECLAIM_TAG);
212 spin_unlock(&mp->m_perag_lock);
213 trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
214}
215
216
217/*
218 * We set the inode flag atomically with the radix tree tag.
219 * Once we get tag lookups on the radix tree, this inode flag
220 * can go away.
221 */
222void
223xfs_inode_set_reclaim_tag(
224 struct xfs_inode *ip)
225{
226 struct xfs_mount *mp = ip->i_mount;
227 struct xfs_perag *pag;
228
229 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
230 spin_lock(&pag->pag_ici_lock);
231 spin_lock(&ip->i_flags_lock);
232
233 radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino),
234 XFS_ICI_RECLAIM_TAG);
235 xfs_perag_set_reclaim_tag(pag);
236 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
237
238 spin_unlock(&ip->i_flags_lock);
239 spin_unlock(&pag->pag_ici_lock);
240 xfs_perag_put(pag);
241}
242
243STATIC void
244xfs_inode_clear_reclaim_tag(
245 struct xfs_perag *pag,
246 xfs_ino_t ino)
247{
248 radix_tree_tag_clear(&pag->pag_ici_root,
249 XFS_INO_TO_AGINO(pag->pag_mount, ino),
250 XFS_ICI_RECLAIM_TAG);
251 xfs_perag_clear_reclaim_tag(pag);
252}
253
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700254static void
255xfs_inew_wait(
256 struct xfs_inode *ip)
257{
258 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT);
259 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT);
260
261 do {
Ingo Molnar21417132017-03-05 11:25:39 +0100262 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700263 if (!xfs_iflags_test(ip, XFS_INEW))
264 break;
265 schedule();
266 } while (true);
Ingo Molnar21417132017-03-05 11:25:39 +0100267 finish_wait(wq, &wait.wq_entry);
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700268}
269
Dave Chinnerad438c42016-05-18 14:20:08 +1000270/*
Dave Chinner50997472016-02-09 16:54:58 +1100271 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
272 * part of the structure. This is made more complex by the fact we store
273 * information about the on-disk values in the VFS inode and so we can't just
Dave Chinner83e06f22016-02-09 16:54:58 +1100274 * overwrite the values unconditionally. Hence we save the parameters we
Dave Chinner50997472016-02-09 16:54:58 +1100275 * need to retain across reinitialisation, and rewrite them into the VFS inode
Dave Chinner83e06f22016-02-09 16:54:58 +1100276 * after reinitialisation even if it fails.
Dave Chinner50997472016-02-09 16:54:58 +1100277 */
278static int
279xfs_reinit_inode(
280 struct xfs_mount *mp,
281 struct inode *inode)
282{
283 int error;
Dave Chinner54d7b5c2016-02-09 16:54:58 +1100284 uint32_t nlink = inode->i_nlink;
Dave Chinner9e9a2672016-02-09 16:54:58 +1100285 uint32_t generation = inode->i_generation;
Jeff Laytonf0e28282017-12-11 06:35:19 -0500286 uint64_t version = inode_peek_iversion(inode);
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100287 umode_t mode = inode->i_mode;
Amir Goldsteinacd1d712018-01-26 11:24:40 -0800288 dev_t dev = inode->i_rdev;
Dave Chinner50997472016-02-09 16:54:58 +1100289
290 error = inode_init_always(mp->m_super, inode);
291
Dave Chinner54d7b5c2016-02-09 16:54:58 +1100292 set_nlink(inode, nlink);
Dave Chinner9e9a2672016-02-09 16:54:58 +1100293 inode->i_generation = generation;
Jeff Laytonf0e28282017-12-11 06:35:19 -0500294 inode_set_iversion_queried(inode, version);
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100295 inode->i_mode = mode;
Amir Goldsteinacd1d712018-01-26 11:24:40 -0800296 inode->i_rdev = dev;
Dave Chinner50997472016-02-09 16:54:58 +1100297 return error;
298}
299
300/*
Dave Chinnerafca6c52018-04-17 17:17:34 -0700301 * If we are allocating a new inode, then check what was returned is
302 * actually a free, empty inode. If we are not allocating an inode,
303 * then check we didn't find a free inode.
304 *
305 * Returns:
306 * 0 if the inode free state matches the lookup context
307 * -ENOENT if the inode is free and we are not allocating
308 * -EFSCORRUPTED if there is any state mismatch at all
309 */
310static int
311xfs_iget_check_free_state(
312 struct xfs_inode *ip,
313 int flags)
314{
315 if (flags & XFS_IGET_CREATE) {
316 /* should be a free inode */
317 if (VFS_I(ip)->i_mode != 0) {
318 xfs_warn(ip->i_mount,
319"Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
320 ip->i_ino, VFS_I(ip)->i_mode);
321 return -EFSCORRUPTED;
322 }
323
324 if (ip->i_d.di_nblocks != 0) {
325 xfs_warn(ip->i_mount,
326"Corruption detected! Free inode 0x%llx has blocks allocated!",
327 ip->i_ino);
328 return -EFSCORRUPTED;
329 }
330 return 0;
331 }
332
333 /* should be an allocated inode */
334 if (VFS_I(ip)->i_mode == 0)
335 return -ENOENT;
336
337 return 0;
338}
339
340/*
Dave Chinner33479e02012-10-08 21:56:11 +1100341 * Check the validity of the inode we just found it the cache
342 */
343static int
344xfs_iget_cache_hit(
345 struct xfs_perag *pag,
346 struct xfs_inode *ip,
347 xfs_ino_t ino,
348 int flags,
349 int lock_flags) __releases(RCU)
350{
351 struct inode *inode = VFS_I(ip);
352 struct xfs_mount *mp = ip->i_mount;
353 int error;
354
355 /*
356 * check for re-use of an inode within an RCU grace period due to the
357 * radix tree nodes not being updated yet. We monitor for this by
358 * setting the inode number to zero before freeing the inode structure.
359 * If the inode has been reallocated and set up, then the inode number
360 * will not match, so check for that, too.
361 */
362 spin_lock(&ip->i_flags_lock);
363 if (ip->i_ino != ino) {
364 trace_xfs_iget_skip(ip);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100365 XFS_STATS_INC(mp, xs_ig_frecycle);
Dave Chinner24513372014-06-25 14:58:08 +1000366 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100367 goto out_error;
368 }
369
370
371 /*
372 * If we are racing with another cache hit that is currently
373 * instantiating this inode or currently recycling it out of
374 * reclaimabe state, wait for the initialisation to complete
375 * before continuing.
376 *
377 * XXX(hch): eventually we should do something equivalent to
378 * wait_on_inode to wait for these flags to be cleared
379 * instead of polling for it.
380 */
381 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
382 trace_xfs_iget_skip(ip);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100383 XFS_STATS_INC(mp, xs_ig_frecycle);
Dave Chinner24513372014-06-25 14:58:08 +1000384 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100385 goto out_error;
386 }
387
388 /*
Dave Chinnerafca6c52018-04-17 17:17:34 -0700389 * Check the inode free state is valid. This also detects lookup
390 * racing with unlinks.
Dave Chinner33479e02012-10-08 21:56:11 +1100391 */
Dave Chinnerafca6c52018-04-17 17:17:34 -0700392 error = xfs_iget_check_free_state(ip, flags);
393 if (error)
Dave Chinner33479e02012-10-08 21:56:11 +1100394 goto out_error;
Dave Chinner33479e02012-10-08 21:56:11 +1100395
396 /*
397 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
398 * Need to carefully get it back into useable state.
399 */
400 if (ip->i_flags & XFS_IRECLAIMABLE) {
401 trace_xfs_iget_reclaim(ip);
402
Darrick J. Wong378f6812017-06-19 08:58:56 -0700403 if (flags & XFS_IGET_INCORE) {
404 error = -EAGAIN;
405 goto out_error;
406 }
407
Dave Chinner33479e02012-10-08 21:56:11 +1100408 /*
409 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
410 * from stomping over us while we recycle the inode. We can't
411 * clear the radix tree reclaimable tag yet as it requires
412 * pag_ici_lock to be held exclusive.
413 */
414 ip->i_flags |= XFS_IRECLAIM;
415
416 spin_unlock(&ip->i_flags_lock);
417 rcu_read_unlock();
418
Dave Chinner50997472016-02-09 16:54:58 +1100419 error = xfs_reinit_inode(mp, inode);
Dave Chinner33479e02012-10-08 21:56:11 +1100420 if (error) {
Brian Foster756baca2017-04-26 08:30:39 -0700421 bool wake;
Dave Chinner33479e02012-10-08 21:56:11 +1100422 /*
423 * Re-initializing the inode failed, and we are in deep
424 * trouble. Try to re-add it to the reclaim list.
425 */
426 rcu_read_lock();
427 spin_lock(&ip->i_flags_lock);
Brian Foster756baca2017-04-26 08:30:39 -0700428 wake = !!__xfs_iflags_test(ip, XFS_INEW);
Dave Chinner33479e02012-10-08 21:56:11 +1100429 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
Brian Foster756baca2017-04-26 08:30:39 -0700430 if (wake)
431 wake_up_bit(&ip->i_flags, __XFS_INEW_BIT);
Dave Chinner33479e02012-10-08 21:56:11 +1100432 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
433 trace_xfs_iget_reclaim_fail(ip);
434 goto out_error;
435 }
436
437 spin_lock(&pag->pag_ici_lock);
438 spin_lock(&ip->i_flags_lock);
439
440 /*
441 * Clear the per-lifetime state in the inode as we are now
442 * effectively a new inode and need to return to the initial
443 * state before reuse occurs.
444 */
445 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
446 ip->i_flags |= XFS_INEW;
Dave Chinner545c0882016-05-18 14:11:41 +1000447 xfs_inode_clear_reclaim_tag(pag, ip->i_ino);
Dave Chinner33479e02012-10-08 21:56:11 +1100448 inode->i_state = I_NEW;
449
Christoph Hellwig65523212016-11-30 14:33:25 +1100450 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
451 init_rwsem(&inode->i_rwsem);
Dave Chinner33479e02012-10-08 21:56:11 +1100452
453 spin_unlock(&ip->i_flags_lock);
454 spin_unlock(&pag->pag_ici_lock);
455 } else {
456 /* If the VFS inode is being torn down, pause and try again. */
457 if (!igrab(inode)) {
458 trace_xfs_iget_skip(ip);
Dave Chinner24513372014-06-25 14:58:08 +1000459 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100460 goto out_error;
461 }
462
463 /* We've got a live one. */
464 spin_unlock(&ip->i_flags_lock);
465 rcu_read_unlock();
466 trace_xfs_iget_hit(ip);
467 }
468
469 if (lock_flags != 0)
470 xfs_ilock(ip, lock_flags);
471
Darrick J. Wong378f6812017-06-19 08:58:56 -0700472 if (!(flags & XFS_IGET_INCORE))
473 xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100474 XFS_STATS_INC(mp, xs_ig_found);
Dave Chinner33479e02012-10-08 21:56:11 +1100475
476 return 0;
477
478out_error:
479 spin_unlock(&ip->i_flags_lock);
480 rcu_read_unlock();
481 return error;
482}
483
484
485static int
486xfs_iget_cache_miss(
487 struct xfs_mount *mp,
488 struct xfs_perag *pag,
489 xfs_trans_t *tp,
490 xfs_ino_t ino,
491 struct xfs_inode **ipp,
492 int flags,
493 int lock_flags)
494{
495 struct xfs_inode *ip;
496 int error;
497 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
498 int iflags;
499
500 ip = xfs_inode_alloc(mp, ino);
501 if (!ip)
Dave Chinner24513372014-06-25 14:58:08 +1000502 return -ENOMEM;
Dave Chinner33479e02012-10-08 21:56:11 +1100503
504 error = xfs_iread(mp, tp, ip, flags);
505 if (error)
506 goto out_destroy;
507
Darrick J. Wong9cfb9b42018-01-08 10:51:06 -0800508 if (!xfs_inode_verify_forks(ip)) {
509 error = -EFSCORRUPTED;
510 goto out_destroy;
511 }
512
Dave Chinner33479e02012-10-08 21:56:11 +1100513 trace_xfs_iget_miss(ip);
514
Dave Chinneree457002018-03-23 10:22:53 -0700515
516 /*
Dave Chinnerafca6c52018-04-17 17:17:34 -0700517 * Check the inode free state is valid. This also detects lookup
518 * racing with unlinks.
Dave Chinneree457002018-03-23 10:22:53 -0700519 */
Dave Chinnerafca6c52018-04-17 17:17:34 -0700520 error = xfs_iget_check_free_state(ip, flags);
521 if (error)
Dave Chinner33479e02012-10-08 21:56:11 +1100522 goto out_destroy;
Dave Chinner33479e02012-10-08 21:56:11 +1100523
524 /*
525 * Preload the radix tree so we can insert safely under the
526 * write spinlock. Note that we cannot sleep inside the preload
527 * region. Since we can be called from transaction context, don't
528 * recurse into the file system.
529 */
530 if (radix_tree_preload(GFP_NOFS)) {
Dave Chinner24513372014-06-25 14:58:08 +1000531 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100532 goto out_destroy;
533 }
534
535 /*
536 * Because the inode hasn't been added to the radix-tree yet it can't
537 * be found by another thread, so we can do the non-sleeping lock here.
538 */
539 if (lock_flags) {
540 if (!xfs_ilock_nowait(ip, lock_flags))
541 BUG();
542 }
543
544 /*
545 * These values must be set before inserting the inode into the radix
546 * tree as the moment it is inserted a concurrent lookup (allowed by the
547 * RCU locking mechanism) can find it and that lookup must see that this
548 * is an inode currently under construction (i.e. that XFS_INEW is set).
549 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
550 * memory barrier that ensures this detection works correctly at lookup
551 * time.
552 */
553 iflags = XFS_INEW;
554 if (flags & XFS_IGET_DONTCACHE)
555 iflags |= XFS_IDONTCACHE;
Chandra Seetharaman113a5682013-06-27 17:25:07 -0500556 ip->i_udquot = NULL;
557 ip->i_gdquot = NULL;
Chandra Seetharaman92f8ff72013-07-11 00:00:40 -0500558 ip->i_pdquot = NULL;
Dave Chinner33479e02012-10-08 21:56:11 +1100559 xfs_iflags_set(ip, iflags);
560
561 /* insert the new inode */
562 spin_lock(&pag->pag_ici_lock);
563 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
564 if (unlikely(error)) {
565 WARN_ON(error != -EEXIST);
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100566 XFS_STATS_INC(mp, xs_ig_dup);
Dave Chinner24513372014-06-25 14:58:08 +1000567 error = -EAGAIN;
Dave Chinner33479e02012-10-08 21:56:11 +1100568 goto out_preload_end;
569 }
570 spin_unlock(&pag->pag_ici_lock);
571 radix_tree_preload_end();
572
573 *ipp = ip;
574 return 0;
575
576out_preload_end:
577 spin_unlock(&pag->pag_ici_lock);
578 radix_tree_preload_end();
579 if (lock_flags)
580 xfs_iunlock(ip, lock_flags);
581out_destroy:
582 __destroy_inode(VFS_I(ip));
583 xfs_inode_free(ip);
584 return error;
585}
586
587/*
588 * Look up an inode by number in the given file system.
589 * The inode is looked up in the cache held in each AG.
590 * If the inode is found in the cache, initialise the vfs inode
591 * if necessary.
592 *
593 * If it is not in core, read it in from the file system's device,
594 * add it to the cache and initialise the vfs inode.
595 *
596 * The inode is locked according to the value of the lock_flags parameter.
597 * This flag parameter indicates how and if the inode's IO lock and inode lock
598 * should be taken.
599 *
600 * mp -- the mount point structure for the current file system. It points
601 * to the inode hash table.
602 * tp -- a pointer to the current transaction if there is one. This is
603 * simply passed through to the xfs_iread() call.
604 * ino -- the number of the inode desired. This is the unique identifier
605 * within the file system for the inode being requested.
606 * lock_flags -- flags indicating how to lock the inode. See the comment
607 * for xfs_ilock() for a list of valid values.
608 */
609int
610xfs_iget(
611 xfs_mount_t *mp,
612 xfs_trans_t *tp,
613 xfs_ino_t ino,
614 uint flags,
615 uint lock_flags,
616 xfs_inode_t **ipp)
617{
618 xfs_inode_t *ip;
619 int error;
620 xfs_perag_t *pag;
621 xfs_agino_t agino;
622
623 /*
624 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
625 * doesn't get freed while it's being referenced during a
626 * radix tree traversal here. It assumes this function
627 * aqcuires only the ILOCK (and therefore it has no need to
628 * involve the IOLOCK in this synchronization).
629 */
630 ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
631
632 /* reject inode numbers outside existing AGs */
633 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
Dave Chinner24513372014-06-25 14:58:08 +1000634 return -EINVAL;
Dave Chinner33479e02012-10-08 21:56:11 +1100635
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100636 XFS_STATS_INC(mp, xs_ig_attempts);
Lucas Stach8774cf82015-08-28 14:50:56 +1000637
Dave Chinner33479e02012-10-08 21:56:11 +1100638 /* get the perag structure and ensure that it's inode capable */
639 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
640 agino = XFS_INO_TO_AGINO(mp, ino);
641
642again:
643 error = 0;
644 rcu_read_lock();
645 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
646
647 if (ip) {
648 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
649 if (error)
650 goto out_error_or_again;
651 } else {
652 rcu_read_unlock();
Darrick J. Wong378f6812017-06-19 08:58:56 -0700653 if (flags & XFS_IGET_INCORE) {
Darrick J. Wonged438b42017-10-17 21:37:32 -0700654 error = -ENODATA;
Darrick J. Wong378f6812017-06-19 08:58:56 -0700655 goto out_error_or_again;
656 }
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100657 XFS_STATS_INC(mp, xs_ig_missed);
Dave Chinner33479e02012-10-08 21:56:11 +1100658
659 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
660 flags, lock_flags);
661 if (error)
662 goto out_error_or_again;
663 }
664 xfs_perag_put(pag);
665
666 *ipp = ip;
667
668 /*
Dave Chinner58c90472015-02-23 22:38:08 +1100669 * If we have a real type for an on-disk inode, we can setup the inode
Dave Chinner33479e02012-10-08 21:56:11 +1100670 * now. If it's a new inode being created, xfs_ialloc will handle it.
671 */
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100672 if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
Dave Chinner58c90472015-02-23 22:38:08 +1100673 xfs_setup_existing_inode(ip);
Dave Chinner33479e02012-10-08 21:56:11 +1100674 return 0;
675
676out_error_or_again:
Darrick J. Wong378f6812017-06-19 08:58:56 -0700677 if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) {
Dave Chinner33479e02012-10-08 21:56:11 +1100678 delay(1);
679 goto again;
680 }
681 xfs_perag_put(pag);
682 return error;
683}
684
Dave Chinner78ae5252010-09-28 12:28:19 +1000685/*
Darrick J. Wong378f6812017-06-19 08:58:56 -0700686 * "Is this a cached inode that's also allocated?"
687 *
688 * Look up an inode by number in the given file system. If the inode is
689 * in cache and isn't in purgatory, return 1 if the inode is allocated
690 * and 0 if it is not. For all other cases (not in cache, being torn
691 * down, etc.), return a negative error code.
692 *
693 * The caller has to prevent inode allocation and freeing activity,
694 * presumably by locking the AGI buffer. This is to ensure that an
695 * inode cannot transition from allocated to freed until the caller is
696 * ready to allow that. If the inode is in an intermediate state (new,
697 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
698 * inode is not in the cache, -ENOENT will be returned. The caller must
699 * deal with these scenarios appropriately.
700 *
701 * This is a specialized use case for the online scrubber; if you're
702 * reading this, you probably want xfs_iget.
703 */
704int
705xfs_icache_inode_is_allocated(
706 struct xfs_mount *mp,
707 struct xfs_trans *tp,
708 xfs_ino_t ino,
709 bool *inuse)
710{
711 struct xfs_inode *ip;
712 int error;
713
714 error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip);
715 if (error)
716 return error;
717
718 *inuse = !!(VFS_I(ip)->i_mode);
Darrick J. Wong44a87362018-07-25 12:52:32 -0700719 xfs_irele(ip);
Darrick J. Wong378f6812017-06-19 08:58:56 -0700720 return 0;
721}
722
723/*
Dave Chinner78ae5252010-09-28 12:28:19 +1000724 * The inode lookup is done in batches to keep the amount of lock traffic and
725 * radix tree lookups to a minimum. The batch size is a trade off between
726 * lookup reduction and stack usage. This is in the reclaim path, so we can't
727 * be too greedy.
728 */
729#define XFS_LOOKUP_BATCH 32
730
Dave Chinnere13de952010-09-28 12:28:06 +1000731STATIC int
732xfs_inode_ag_walk_grab(
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700733 struct xfs_inode *ip,
734 int flags)
Dave Chinnere13de952010-09-28 12:28:06 +1000735{
736 struct inode *inode = VFS_I(ip);
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700737 bool newinos = !!(flags & XFS_AGITER_INEW_WAIT);
Dave Chinnere13de952010-09-28 12:28:06 +1000738
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100739 ASSERT(rcu_read_lock_held());
740
741 /*
742 * check for stale RCU freed inode
743 *
744 * If the inode has been reallocated, it doesn't matter if it's not in
745 * the AG we are walking - we are walking for writeback, so if it
746 * passes all the "valid inode" checks and is dirty, then we'll write
747 * it back anyway. If it has been reallocated and still being
748 * initialised, the XFS_INEW check below will catch it.
749 */
750 spin_lock(&ip->i_flags_lock);
751 if (!ip->i_ino)
752 goto out_unlock_noent;
753
754 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700755 if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) ||
756 __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM))
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100757 goto out_unlock_noent;
758 spin_unlock(&ip->i_flags_lock);
759
Dave Chinnere13de952010-09-28 12:28:06 +1000760 /* nothing to sync during shutdown */
761 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Dave Chinner24513372014-06-25 14:58:08 +1000762 return -EFSCORRUPTED;
Dave Chinnere13de952010-09-28 12:28:06 +1000763
Dave Chinnere13de952010-09-28 12:28:06 +1000764 /* If we can't grab the inode, it must on it's way to reclaim. */
765 if (!igrab(inode))
Dave Chinner24513372014-06-25 14:58:08 +1000766 return -ENOENT;
Dave Chinnere13de952010-09-28 12:28:06 +1000767
Dave Chinnere13de952010-09-28 12:28:06 +1000768 /* inode is valid */
769 return 0;
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100770
771out_unlock_noent:
772 spin_unlock(&ip->i_flags_lock);
Dave Chinner24513372014-06-25 14:58:08 +1000773 return -ENOENT;
Dave Chinnere13de952010-09-28 12:28:06 +1000774}
775
Dave Chinner75f3cb12009-06-08 15:35:14 +0200776STATIC int
777xfs_inode_ag_walk(
778 struct xfs_mount *mp,
Dave Chinner5017e972010-01-11 11:47:40 +0000779 struct xfs_perag *pag,
Eric Sandeene0094002014-04-14 19:04:19 +1000780 int (*execute)(struct xfs_inode *ip, int flags,
Brian Fostera454f742012-11-06 09:50:39 -0500781 void *args),
782 int flags,
783 void *args,
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700784 int tag,
785 int iter_flags)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200786{
Dave Chinner75f3cb12009-06-08 15:35:14 +0200787 uint32_t first_index;
788 int last_error = 0;
789 int skipped;
Dave Chinner65d0f202010-09-24 18:40:15 +1000790 int done;
Dave Chinner78ae5252010-09-28 12:28:19 +1000791 int nr_found;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200792
793restart:
Dave Chinner65d0f202010-09-24 18:40:15 +1000794 done = 0;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200795 skipped = 0;
796 first_index = 0;
Dave Chinner78ae5252010-09-28 12:28:19 +1000797 nr_found = 0;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200798 do {
Dave Chinner78ae5252010-09-28 12:28:19 +1000799 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
Dave Chinner75f3cb12009-06-08 15:35:14 +0200800 int error = 0;
Dave Chinner78ae5252010-09-28 12:28:19 +1000801 int i;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200802
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100803 rcu_read_lock();
Brian Fostera454f742012-11-06 09:50:39 -0500804
805 if (tag == -1)
806 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
Dave Chinner78ae5252010-09-28 12:28:19 +1000807 (void **)batch, first_index,
808 XFS_LOOKUP_BATCH);
Brian Fostera454f742012-11-06 09:50:39 -0500809 else
810 nr_found = radix_tree_gang_lookup_tag(
811 &pag->pag_ici_root,
812 (void **) batch, first_index,
813 XFS_LOOKUP_BATCH, tag);
814
Dave Chinner65d0f202010-09-24 18:40:15 +1000815 if (!nr_found) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100816 rcu_read_unlock();
Dave Chinner75f3cb12009-06-08 15:35:14 +0200817 break;
Dave Chinnerc8e20be2010-01-10 23:51:45 +0000818 }
Dave Chinner75f3cb12009-06-08 15:35:14 +0200819
Dave Chinner65d0f202010-09-24 18:40:15 +1000820 /*
Dave Chinner78ae5252010-09-28 12:28:19 +1000821 * Grab the inodes before we drop the lock. if we found
822 * nothing, nr == 0 and the loop will be skipped.
Dave Chinner65d0f202010-09-24 18:40:15 +1000823 */
Dave Chinner78ae5252010-09-28 12:28:19 +1000824 for (i = 0; i < nr_found; i++) {
825 struct xfs_inode *ip = batch[i];
Dave Chinner65d0f202010-09-24 18:40:15 +1000826
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700827 if (done || xfs_inode_ag_walk_grab(ip, iter_flags))
Dave Chinner78ae5252010-09-28 12:28:19 +1000828 batch[i] = NULL;
829
830 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100831 * Update the index for the next lookup. Catch
832 * overflows into the next AG range which can occur if
833 * we have inodes in the last block of the AG and we
834 * are currently pointing to the last inode.
835 *
836 * Because we may see inodes that are from the wrong AG
837 * due to RCU freeing and reallocation, only update the
838 * index if it lies in this AG. It was a race that lead
839 * us to see this inode, so another lookup from the
840 * same index will not find it again.
Dave Chinner78ae5252010-09-28 12:28:19 +1000841 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100842 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
843 continue;
Dave Chinner78ae5252010-09-28 12:28:19 +1000844 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
845 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
846 done = 1;
Dave Chinnere13de952010-09-28 12:28:06 +1000847 }
Dave Chinner78ae5252010-09-28 12:28:19 +1000848
849 /* unlock now we've grabbed the inodes. */
Dave Chinner1a3e8f32010-12-17 17:29:43 +1100850 rcu_read_unlock();
Dave Chinnere13de952010-09-28 12:28:06 +1000851
Dave Chinner78ae5252010-09-28 12:28:19 +1000852 for (i = 0; i < nr_found; i++) {
853 if (!batch[i])
854 continue;
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700855 if ((iter_flags & XFS_AGITER_INEW_WAIT) &&
856 xfs_iflags_test(batch[i], XFS_INEW))
857 xfs_inew_wait(batch[i]);
Eric Sandeene0094002014-04-14 19:04:19 +1000858 error = execute(batch[i], flags, args);
Darrick J. Wong44a87362018-07-25 12:52:32 -0700859 xfs_irele(batch[i]);
Dave Chinner24513372014-06-25 14:58:08 +1000860 if (error == -EAGAIN) {
Dave Chinner78ae5252010-09-28 12:28:19 +1000861 skipped++;
862 continue;
863 }
Dave Chinner24513372014-06-25 14:58:08 +1000864 if (error && last_error != -EFSCORRUPTED)
Dave Chinner78ae5252010-09-28 12:28:19 +1000865 last_error = error;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200866 }
Dave Chinnerc8e20be2010-01-10 23:51:45 +0000867
868 /* bail out if the filesystem is corrupted. */
Dave Chinner24513372014-06-25 14:58:08 +1000869 if (error == -EFSCORRUPTED)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200870 break;
871
Dave Chinner8daaa832011-07-08 14:14:46 +1000872 cond_resched();
873
Dave Chinner78ae5252010-09-28 12:28:19 +1000874 } while (nr_found && !done);
Dave Chinner75f3cb12009-06-08 15:35:14 +0200875
876 if (skipped) {
877 delay(1);
878 goto restart;
879 }
Dave Chinner75f3cb12009-06-08 15:35:14 +0200880 return last_error;
881}
882
Brian Foster579b62f2012-11-06 09:50:47 -0500883/*
884 * Background scanning to trim post-EOF preallocated space. This is queued
Dwight Engenb9fe5052013-08-15 14:08:02 -0400885 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
Brian Foster579b62f2012-11-06 09:50:47 -0500886 */
Brian Fosterfa5a4f52016-06-21 11:53:28 +1000887void
Brian Foster579b62f2012-11-06 09:50:47 -0500888xfs_queue_eofblocks(
889 struct xfs_mount *mp)
890{
891 rcu_read_lock();
892 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
893 queue_delayed_work(mp->m_eofblocks_workqueue,
894 &mp->m_eofblocks_work,
895 msecs_to_jiffies(xfs_eofb_secs * 1000));
896 rcu_read_unlock();
897}
898
899void
900xfs_eofblocks_worker(
901 struct work_struct *work)
902{
903 struct xfs_mount *mp = container_of(to_delayed_work(work),
904 struct xfs_mount, m_eofblocks_work);
905 xfs_icache_free_eofblocks(mp, NULL);
906 xfs_queue_eofblocks(mp);
907}
908
Darrick J. Wong83104d42016-10-03 09:11:46 -0700909/*
910 * Background scanning to trim preallocated CoW space. This is queued
911 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
912 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
913 */
Darrick J. Wong10ddf642017-12-14 15:46:05 -0800914void
Darrick J. Wong83104d42016-10-03 09:11:46 -0700915xfs_queue_cowblocks(
916 struct xfs_mount *mp)
917{
918 rcu_read_lock();
919 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
920 queue_delayed_work(mp->m_eofblocks_workqueue,
921 &mp->m_cowblocks_work,
922 msecs_to_jiffies(xfs_cowb_secs * 1000));
923 rcu_read_unlock();
924}
925
926void
927xfs_cowblocks_worker(
928 struct work_struct *work)
929{
930 struct xfs_mount *mp = container_of(to_delayed_work(work),
931 struct xfs_mount, m_cowblocks_work);
932 xfs_icache_free_cowblocks(mp, NULL);
933 xfs_queue_cowblocks(mp);
934}
935
Christoph Hellwigfe588ed2009-06-08 15:35:27 +0200936int
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700937xfs_inode_ag_iterator_flags(
Dave Chinner75f3cb12009-06-08 15:35:14 +0200938 struct xfs_mount *mp,
Eric Sandeene0094002014-04-14 19:04:19 +1000939 int (*execute)(struct xfs_inode *ip, int flags,
Brian Fostera454f742012-11-06 09:50:39 -0500940 void *args),
941 int flags,
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700942 void *args,
943 int iter_flags)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200944{
Dave Chinner16fd5362010-07-20 09:43:39 +1000945 struct xfs_perag *pag;
Dave Chinner75f3cb12009-06-08 15:35:14 +0200946 int error = 0;
947 int last_error = 0;
948 xfs_agnumber_t ag;
949
Dave Chinner16fd5362010-07-20 09:43:39 +1000950 ag = 0;
Dave Chinner65d0f202010-09-24 18:40:15 +1000951 while ((pag = xfs_perag_get(mp, ag))) {
952 ag = pag->pag_agno + 1;
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700953 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1,
954 iter_flags);
Brian Fostera454f742012-11-06 09:50:39 -0500955 xfs_perag_put(pag);
956 if (error) {
957 last_error = error;
Dave Chinner24513372014-06-25 14:58:08 +1000958 if (error == -EFSCORRUPTED)
Brian Fostera454f742012-11-06 09:50:39 -0500959 break;
960 }
961 }
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000962 return last_error;
Brian Fostera454f742012-11-06 09:50:39 -0500963}
964
965int
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700966xfs_inode_ag_iterator(
967 struct xfs_mount *mp,
968 int (*execute)(struct xfs_inode *ip, int flags,
969 void *args),
970 int flags,
971 void *args)
972{
973 return xfs_inode_ag_iterator_flags(mp, execute, flags, args, 0);
974}
975
976int
Brian Fostera454f742012-11-06 09:50:39 -0500977xfs_inode_ag_iterator_tag(
978 struct xfs_mount *mp,
Eric Sandeene0094002014-04-14 19:04:19 +1000979 int (*execute)(struct xfs_inode *ip, int flags,
Brian Fostera454f742012-11-06 09:50:39 -0500980 void *args),
981 int flags,
982 void *args,
983 int tag)
984{
985 struct xfs_perag *pag;
986 int error = 0;
987 int last_error = 0;
988 xfs_agnumber_t ag;
989
990 ag = 0;
991 while ((pag = xfs_perag_get_tag(mp, ag, tag))) {
992 ag = pag->pag_agno + 1;
Brian Fosterae2c4ac2017-04-26 08:30:39 -0700993 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag,
994 0);
Dave Chinner5017e972010-01-11 11:47:40 +0000995 xfs_perag_put(pag);
Dave Chinner75f3cb12009-06-08 15:35:14 +0200996 if (error) {
997 last_error = error;
Dave Chinner24513372014-06-25 14:58:08 +1000998 if (error == -EFSCORRUPTED)
Dave Chinner75f3cb12009-06-08 15:35:14 +0200999 break;
1000 }
1001 }
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001002 return last_error;
Dave Chinner75f3cb12009-06-08 15:35:14 +02001003}
1004
David Chinner76bf1052008-10-30 17:16:21 +11001005/*
Dave Chinnere3a20c02010-09-24 19:51:50 +10001006 * Grab the inode for reclaim exclusively.
1007 * Return 0 if we grabbed it, non-zero otherwise.
1008 */
1009STATIC int
1010xfs_reclaim_inode_grab(
1011 struct xfs_inode *ip,
1012 int flags)
1013{
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001014 ASSERT(rcu_read_lock_held());
1015
1016 /* quick check for stale RCU freed inode */
1017 if (!ip->i_ino)
1018 return 1;
Dave Chinnere3a20c02010-09-24 19:51:50 +10001019
1020 /*
Christoph Hellwig474fce02011-12-18 20:00:09 +00001021 * If we are asked for non-blocking operation, do unlocked checks to
1022 * see if the inode already is being flushed or in reclaim to avoid
1023 * lock traffic.
Dave Chinnere3a20c02010-09-24 19:51:50 +10001024 */
1025 if ((flags & SYNC_TRYLOCK) &&
Christoph Hellwig474fce02011-12-18 20:00:09 +00001026 __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM))
Dave Chinnere3a20c02010-09-24 19:51:50 +10001027 return 1;
Dave Chinnere3a20c02010-09-24 19:51:50 +10001028
1029 /*
1030 * The radix tree lock here protects a thread in xfs_iget from racing
1031 * with us starting reclaim on the inode. Once we have the
1032 * XFS_IRECLAIM flag set it will not touch us.
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001033 *
1034 * Due to RCU lookup, we may find inodes that have been freed and only
1035 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
1036 * aren't candidates for reclaim at all, so we must check the
1037 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
Dave Chinnere3a20c02010-09-24 19:51:50 +10001038 */
1039 spin_lock(&ip->i_flags_lock);
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001040 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
1041 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
1042 /* not a reclaim candidate. */
Dave Chinnere3a20c02010-09-24 19:51:50 +10001043 spin_unlock(&ip->i_flags_lock);
1044 return 1;
1045 }
1046 __xfs_iflags_set(ip, XFS_IRECLAIM);
1047 spin_unlock(&ip->i_flags_lock);
1048 return 0;
1049}
1050
1051/*
Christoph Hellwig8a480882012-04-23 15:58:35 +10001052 * Inodes in different states need to be treated differently. The following
1053 * table lists the inode states and the reclaim actions necessary:
Dave Chinner777df5a2010-02-06 12:37:26 +11001054 *
1055 * inode state iflush ret required action
1056 * --------------- ---------- ---------------
1057 * bad - reclaim
1058 * shutdown EIO unpin and reclaim
1059 * clean, unpinned 0 reclaim
1060 * stale, unpinned 0 reclaim
Dave Chinnerc8543632010-02-06 12:39:36 +11001061 * clean, pinned(*) 0 requeue
1062 * stale, pinned EAGAIN requeue
Christoph Hellwig8a480882012-04-23 15:58:35 +10001063 * dirty, async - requeue
1064 * dirty, sync 0 reclaim
Dave Chinner777df5a2010-02-06 12:37:26 +11001065 *
1066 * (*) dgc: I don't think the clean, pinned state is possible but it gets
1067 * handled anyway given the order of checks implemented.
1068 *
Dave Chinnerc8543632010-02-06 12:39:36 +11001069 * Also, because we get the flush lock first, we know that any inode that has
1070 * been flushed delwri has had the flush completed by the time we check that
Christoph Hellwig8a480882012-04-23 15:58:35 +10001071 * the inode is clean.
Dave Chinnerc8543632010-02-06 12:39:36 +11001072 *
Christoph Hellwig8a480882012-04-23 15:58:35 +10001073 * Note that because the inode is flushed delayed write by AIL pushing, the
1074 * flush lock may already be held here and waiting on it can result in very
1075 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
1076 * the caller should push the AIL first before trying to reclaim inodes to
1077 * minimise the amount of time spent waiting. For background relaim, we only
1078 * bother to reclaim clean inodes anyway.
Dave Chinnerc8543632010-02-06 12:39:36 +11001079 *
Dave Chinner777df5a2010-02-06 12:37:26 +11001080 * Hence the order of actions after gaining the locks should be:
1081 * bad => reclaim
1082 * shutdown => unpin and reclaim
Christoph Hellwig8a480882012-04-23 15:58:35 +10001083 * pinned, async => requeue
Dave Chinnerc8543632010-02-06 12:39:36 +11001084 * pinned, sync => unpin
Dave Chinner777df5a2010-02-06 12:37:26 +11001085 * stale => reclaim
1086 * clean => reclaim
Christoph Hellwig8a480882012-04-23 15:58:35 +10001087 * dirty, async => requeue
Dave Chinnerc8543632010-02-06 12:39:36 +11001088 * dirty, sync => flush, wait and reclaim
Dave Chinner777df5a2010-02-06 12:37:26 +11001089 */
Dave Chinner75f3cb12009-06-08 15:35:14 +02001090STATIC int
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001091xfs_reclaim_inode(
Dave Chinner75f3cb12009-06-08 15:35:14 +02001092 struct xfs_inode *ip,
1093 struct xfs_perag *pag,
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001094 int sync_mode)
David Chinner7a3be022008-10-30 17:37:37 +11001095{
Christoph Hellwig4c468192012-04-23 15:58:36 +10001096 struct xfs_buf *bp = NULL;
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001097 xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
Christoph Hellwig4c468192012-04-23 15:58:36 +10001098 int error;
Dave Chinner777df5a2010-02-06 12:37:26 +11001099
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001100restart:
1101 error = 0;
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001102 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinnerc8543632010-02-06 12:39:36 +11001103 if (!xfs_iflock_nowait(ip)) {
1104 if (!(sync_mode & SYNC_WAIT))
1105 goto out;
1106 xfs_iflock(ip);
1107 }
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001108
Dave Chinner777df5a2010-02-06 12:37:26 +11001109 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1110 xfs_iunpin_wait(ip);
Brian Foster98efe8a2016-11-10 08:23:22 +11001111 /* xfs_iflush_abort() drops the flush lock */
Dave Chinner04913fd2012-04-23 15:58:41 +10001112 xfs_iflush_abort(ip, false);
Dave Chinner777df5a2010-02-06 12:37:26 +11001113 goto reclaim;
1114 }
Dave Chinnerc8543632010-02-06 12:39:36 +11001115 if (xfs_ipincount(ip)) {
Christoph Hellwig8a480882012-04-23 15:58:35 +10001116 if (!(sync_mode & SYNC_WAIT))
1117 goto out_ifunlock;
Dave Chinner777df5a2010-02-06 12:37:26 +11001118 xfs_iunpin_wait(ip);
Dave Chinnerc8543632010-02-06 12:39:36 +11001119 }
Brian Foster98efe8a2016-11-10 08:23:22 +11001120 if (xfs_iflags_test(ip, XFS_ISTALE) || xfs_inode_clean(ip)) {
1121 xfs_ifunlock(ip);
Dave Chinner777df5a2010-02-06 12:37:26 +11001122 goto reclaim;
Brian Foster98efe8a2016-11-10 08:23:22 +11001123 }
Dave Chinner777df5a2010-02-06 12:37:26 +11001124
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001125 /*
Christoph Hellwig8a480882012-04-23 15:58:35 +10001126 * Never flush out dirty data during non-blocking reclaim, as it would
1127 * just contend with AIL pushing trying to do the same job.
1128 */
1129 if (!(sync_mode & SYNC_WAIT))
1130 goto out_ifunlock;
1131
1132 /*
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001133 * Now we have an inode that needs flushing.
1134 *
Christoph Hellwig4c468192012-04-23 15:58:36 +10001135 * Note that xfs_iflush will never block on the inode buffer lock, as
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001136 * xfs_ifree_cluster() can lock the inode buffer before it locks the
Christoph Hellwig4c468192012-04-23 15:58:36 +10001137 * ip->i_lock, and we are doing the exact opposite here. As a result,
Christoph Hellwig475ee412012-07-03 12:21:22 -04001138 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1139 * result in an ABBA deadlock with xfs_ifree_cluster().
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001140 *
1141 * As xfs_ifree_cluser() must gather all inodes that are active in the
1142 * cache to mark them stale, if we hit this case we don't actually want
1143 * to do IO here - we want the inode marked stale so we can simply
Christoph Hellwig4c468192012-04-23 15:58:36 +10001144 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1145 * inode, back off and try again. Hopefully the next pass through will
1146 * see the stale flag set on the inode.
Dave Chinner1bfd8d02011-03-26 09:13:55 +11001147 */
Christoph Hellwig4c468192012-04-23 15:58:36 +10001148 error = xfs_iflush(ip, &bp);
Dave Chinner24513372014-06-25 14:58:08 +10001149 if (error == -EAGAIN) {
Christoph Hellwig8a480882012-04-23 15:58:35 +10001150 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1151 /* backoff longer than in xfs_ifree_cluster */
1152 delay(2);
1153 goto restart;
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001154 }
Dave Chinnerc8543632010-02-06 12:39:36 +11001155
Christoph Hellwig4c468192012-04-23 15:58:36 +10001156 if (!error) {
1157 error = xfs_bwrite(bp);
1158 xfs_buf_relse(bp);
1159 }
1160
Dave Chinner777df5a2010-02-06 12:37:26 +11001161reclaim:
Brian Foster98efe8a2016-11-10 08:23:22 +11001162 ASSERT(!xfs_isiflocked(ip));
1163
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001164 /*
1165 * Because we use RCU freeing we need to ensure the inode always appears
1166 * to be reclaimed with an invalid inode number when in the free state.
Brian Foster98efe8a2016-11-10 08:23:22 +11001167 * We do this as early as possible under the ILOCK so that
Omar Sandovalf2e9ad22017-08-25 10:05:26 -07001168 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
1169 * detect races with us here. By doing this, we guarantee that once
1170 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
1171 * it will see either a valid inode that will serialise correctly, or it
1172 * will see an invalid inode that it can skip.
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001173 */
1174 spin_lock(&ip->i_flags_lock);
1175 ip->i_flags = XFS_IRECLAIM;
1176 ip->i_ino = 0;
1177 spin_unlock(&ip->i_flags_lock);
1178
Dave Chinnerc8e20be2010-01-10 23:51:45 +00001179 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001180
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11001181 XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001182 /*
1183 * Remove the inode from the per-AG radix tree.
1184 *
1185 * Because radix_tree_delete won't complain even if the item was never
1186 * added to the tree assert that it's been there before to catch
1187 * problems with the inode life time early on.
1188 */
Dave Chinner1a427ab2010-12-16 17:08:41 +11001189 spin_lock(&pag->pag_ici_lock);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001190 if (!radix_tree_delete(&pag->pag_ici_root,
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001191 XFS_INO_TO_AGINO(ip->i_mount, ino)))
Dave Chinner2f11fea2010-07-20 17:53:25 +10001192 ASSERT(0);
Dave Chinner545c0882016-05-18 14:11:41 +10001193 xfs_perag_clear_reclaim_tag(pag);
Dave Chinner1a427ab2010-12-16 17:08:41 +11001194 spin_unlock(&pag->pag_ici_lock);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001195
1196 /*
1197 * Here we do an (almost) spurious inode lock in order to coordinate
1198 * with inode cache radix tree lookups. This is because the lookup
1199 * can reference the inodes in the cache without taking references.
1200 *
1201 * We make that OK here by ensuring that we wait until the inode is
Alex Elderad637a12012-02-16 22:01:00 +00001202 * unlocked after the lookup before we go ahead and free it.
Dave Chinner2f11fea2010-07-20 17:53:25 +10001203 */
Alex Elderad637a12012-02-16 22:01:00 +00001204 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001205 xfs_qm_dqdetach(ip);
Alex Elderad637a12012-02-16 22:01:00 +00001206 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner2f11fea2010-07-20 17:53:25 +10001207
Dave Chinner8a17d7d2016-05-18 14:09:12 +10001208 __xfs_inode_free(ip);
Alex Elderad637a12012-02-16 22:01:00 +00001209 return error;
Christoph Hellwig8a480882012-04-23 15:58:35 +10001210
1211out_ifunlock:
1212 xfs_ifunlock(ip);
1213out:
1214 xfs_iflags_clear(ip, XFS_IRECLAIM);
1215 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1216 /*
Dave Chinner24513372014-06-25 14:58:08 +10001217 * We could return -EAGAIN here to make reclaim rescan the inode tree in
Christoph Hellwig8a480882012-04-23 15:58:35 +10001218 * a short while. However, this just burns CPU time scanning the tree
Dave Chinner58896082012-10-08 21:56:05 +11001219 * waiting for IO to complete and the reclaim work never goes back to
1220 * the idle state. Instead, return 0 to let the next scheduled
1221 * background reclaim attempt to reclaim the inode again.
Christoph Hellwig8a480882012-04-23 15:58:35 +10001222 */
1223 return 0;
David Chinner7a3be022008-10-30 17:37:37 +11001224}
1225
Dave Chinner65d0f202010-09-24 18:40:15 +10001226/*
1227 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1228 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1229 * then a shut down during filesystem unmount reclaim walk leak all the
1230 * unreclaimed inodes.
1231 */
Dave Chinner33479e02012-10-08 21:56:11 +11001232STATIC int
Dave Chinner65d0f202010-09-24 18:40:15 +10001233xfs_reclaim_inodes_ag(
1234 struct xfs_mount *mp,
1235 int flags,
1236 int *nr_to_scan)
1237{
1238 struct xfs_perag *pag;
1239 int error = 0;
1240 int last_error = 0;
1241 xfs_agnumber_t ag;
Dave Chinner69b491c2010-09-27 11:09:51 +10001242 int trylock = flags & SYNC_TRYLOCK;
1243 int skipped;
Dave Chinner65d0f202010-09-24 18:40:15 +10001244
Dave Chinner69b491c2010-09-27 11:09:51 +10001245restart:
Dave Chinner65d0f202010-09-24 18:40:15 +10001246 ag = 0;
Dave Chinner69b491c2010-09-27 11:09:51 +10001247 skipped = 0;
Dave Chinner65d0f202010-09-24 18:40:15 +10001248 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1249 unsigned long first_index = 0;
1250 int done = 0;
Dave Chinnere3a20c02010-09-24 19:51:50 +10001251 int nr_found = 0;
Dave Chinner65d0f202010-09-24 18:40:15 +10001252
1253 ag = pag->pag_agno + 1;
1254
Dave Chinner69b491c2010-09-27 11:09:51 +10001255 if (trylock) {
1256 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
1257 skipped++;
Dave Chinnerf83282a2010-11-08 08:55:04 +00001258 xfs_perag_put(pag);
Dave Chinner69b491c2010-09-27 11:09:51 +10001259 continue;
1260 }
1261 first_index = pag->pag_ici_reclaim_cursor;
1262 } else
1263 mutex_lock(&pag->pag_ici_reclaim_lock);
1264
Dave Chinner65d0f202010-09-24 18:40:15 +10001265 do {
Dave Chinnere3a20c02010-09-24 19:51:50 +10001266 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1267 int i;
Dave Chinner65d0f202010-09-24 18:40:15 +10001268
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001269 rcu_read_lock();
Dave Chinnere3a20c02010-09-24 19:51:50 +10001270 nr_found = radix_tree_gang_lookup_tag(
1271 &pag->pag_ici_root,
1272 (void **)batch, first_index,
1273 XFS_LOOKUP_BATCH,
Dave Chinner65d0f202010-09-24 18:40:15 +10001274 XFS_ICI_RECLAIM_TAG);
1275 if (!nr_found) {
Dave Chinnerb2232212011-05-06 02:54:04 +00001276 done = 1;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001277 rcu_read_unlock();
Dave Chinner65d0f202010-09-24 18:40:15 +10001278 break;
1279 }
1280
1281 /*
Dave Chinnere3a20c02010-09-24 19:51:50 +10001282 * Grab the inodes before we drop the lock. if we found
1283 * nothing, nr == 0 and the loop will be skipped.
Dave Chinner65d0f202010-09-24 18:40:15 +10001284 */
Dave Chinnere3a20c02010-09-24 19:51:50 +10001285 for (i = 0; i < nr_found; i++) {
1286 struct xfs_inode *ip = batch[i];
Dave Chinner65d0f202010-09-24 18:40:15 +10001287
Dave Chinnere3a20c02010-09-24 19:51:50 +10001288 if (done || xfs_reclaim_inode_grab(ip, flags))
1289 batch[i] = NULL;
Dave Chinner65d0f202010-09-24 18:40:15 +10001290
Dave Chinnere3a20c02010-09-24 19:51:50 +10001291 /*
1292 * Update the index for the next lookup. Catch
1293 * overflows into the next AG range which can
1294 * occur if we have inodes in the last block of
1295 * the AG and we are currently pointing to the
1296 * last inode.
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001297 *
1298 * Because we may see inodes that are from the
1299 * wrong AG due to RCU freeing and
1300 * reallocation, only update the index if it
1301 * lies in this AG. It was a race that lead us
1302 * to see this inode, so another lookup from
1303 * the same index will not find it again.
Dave Chinnere3a20c02010-09-24 19:51:50 +10001304 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001305 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1306 pag->pag_agno)
1307 continue;
Dave Chinnere3a20c02010-09-24 19:51:50 +10001308 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1309 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1310 done = 1;
1311 }
1312
1313 /* unlock now we've grabbed the inodes. */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001314 rcu_read_unlock();
Dave Chinnere3a20c02010-09-24 19:51:50 +10001315
1316 for (i = 0; i < nr_found; i++) {
1317 if (!batch[i])
1318 continue;
1319 error = xfs_reclaim_inode(batch[i], pag, flags);
Dave Chinner24513372014-06-25 14:58:08 +10001320 if (error && last_error != -EFSCORRUPTED)
Dave Chinnere3a20c02010-09-24 19:51:50 +10001321 last_error = error;
1322 }
1323
1324 *nr_to_scan -= XFS_LOOKUP_BATCH;
1325
Dave Chinner8daaa832011-07-08 14:14:46 +10001326 cond_resched();
1327
Dave Chinnere3a20c02010-09-24 19:51:50 +10001328 } while (nr_found && !done && *nr_to_scan > 0);
Dave Chinner65d0f202010-09-24 18:40:15 +10001329
Dave Chinner69b491c2010-09-27 11:09:51 +10001330 if (trylock && !done)
1331 pag->pag_ici_reclaim_cursor = first_index;
1332 else
1333 pag->pag_ici_reclaim_cursor = 0;
1334 mutex_unlock(&pag->pag_ici_reclaim_lock);
Dave Chinner65d0f202010-09-24 18:40:15 +10001335 xfs_perag_put(pag);
1336 }
Dave Chinner69b491c2010-09-27 11:09:51 +10001337
1338 /*
1339 * if we skipped any AG, and we still have scan count remaining, do
1340 * another pass this time using blocking reclaim semantics (i.e
1341 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1342 * ensure that when we get more reclaimers than AGs we block rather
1343 * than spin trying to execute reclaim.
1344 */
Dave Chinner8daaa832011-07-08 14:14:46 +10001345 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
Dave Chinner69b491c2010-09-27 11:09:51 +10001346 trylock = 0;
1347 goto restart;
1348 }
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001349 return last_error;
Dave Chinner65d0f202010-09-24 18:40:15 +10001350}
1351
David Chinnerfce08f22008-10-30 17:37:03 +11001352int
David Chinner1dc33182008-10-30 17:37:15 +11001353xfs_reclaim_inodes(
David Chinnerfce08f22008-10-30 17:37:03 +11001354 xfs_mount_t *mp,
David Chinnerfce08f22008-10-30 17:37:03 +11001355 int mode)
1356{
Dave Chinner65d0f202010-09-24 18:40:15 +10001357 int nr_to_scan = INT_MAX;
1358
1359 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
Dave Chinner9bf729c2010-04-29 09:55:50 +10001360}
1361
1362/*
Dave Chinner8daaa832011-07-08 14:14:46 +10001363 * Scan a certain number of inodes for reclaim.
Dave Chinnera7b339f2011-04-08 12:45:07 +10001364 *
1365 * When called we make sure that there is a background (fast) inode reclaim in
Dave Chinner8daaa832011-07-08 14:14:46 +10001366 * progress, while we will throttle the speed of reclaim via doing synchronous
Dave Chinnera7b339f2011-04-08 12:45:07 +10001367 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1368 * them to be cleaned, which we hope will not be very long due to the
1369 * background walker having already kicked the IO off on those dirty inodes.
Dave Chinner9bf729c2010-04-29 09:55:50 +10001370 */
Dave Chinner0a234c62013-08-28 10:17:57 +10001371long
Dave Chinner8daaa832011-07-08 14:14:46 +10001372xfs_reclaim_inodes_nr(
1373 struct xfs_mount *mp,
1374 int nr_to_scan)
Dave Chinner9bf729c2010-04-29 09:55:50 +10001375{
Dave Chinner8daaa832011-07-08 14:14:46 +10001376 /* kick background reclaimer and push the AIL */
Dave Chinner58896082012-10-08 21:56:05 +11001377 xfs_reclaim_work_queue(mp);
Dave Chinner8daaa832011-07-08 14:14:46 +10001378 xfs_ail_push_all(mp->m_ail);
Dave Chinner9bf729c2010-04-29 09:55:50 +10001379
Dave Chinner0a234c62013-08-28 10:17:57 +10001380 return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
Dave Chinner8daaa832011-07-08 14:14:46 +10001381}
Dave Chinnera7b339f2011-04-08 12:45:07 +10001382
Dave Chinner8daaa832011-07-08 14:14:46 +10001383/*
1384 * Return the number of reclaimable inodes in the filesystem for
1385 * the shrinker to determine how much to reclaim.
1386 */
1387int
1388xfs_reclaim_inodes_count(
1389 struct xfs_mount *mp)
1390{
1391 struct xfs_perag *pag;
1392 xfs_agnumber_t ag = 0;
1393 int reclaimable = 0;
Dave Chinner9bf729c2010-04-29 09:55:50 +10001394
Dave Chinner65d0f202010-09-24 18:40:15 +10001395 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1396 ag = pag->pag_agno + 1;
Dave Chinner70e60ce2010-07-20 08:07:02 +10001397 reclaimable += pag->pag_ici_reclaimable;
1398 xfs_perag_put(pag);
Dave Chinner9bf729c2010-04-29 09:55:50 +10001399 }
Dave Chinner9bf729c2010-04-29 09:55:50 +10001400 return reclaimable;
1401}
1402
Brian Foster41176a62012-11-06 09:50:42 -05001403STATIC int
Brian Foster3e3f9f52012-11-07 12:21:13 -05001404xfs_inode_match_id(
1405 struct xfs_inode *ip,
1406 struct xfs_eofblocks *eofb)
1407{
Dwight Engenb9fe5052013-08-15 14:08:02 -04001408 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1409 !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
Brian Foster1b556042012-11-06 09:50:45 -05001410 return 0;
Brian Foster3e3f9f52012-11-07 12:21:13 -05001411
Dwight Engenb9fe5052013-08-15 14:08:02 -04001412 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1413 !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
Brian Foster1b556042012-11-06 09:50:45 -05001414 return 0;
1415
Dwight Engenb9fe5052013-08-15 14:08:02 -04001416 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
Brian Foster1b556042012-11-06 09:50:45 -05001417 xfs_get_projid(ip) != eofb->eof_prid)
1418 return 0;
1419
1420 return 1;
Brian Foster3e3f9f52012-11-07 12:21:13 -05001421}
1422
Brian Fosterf4526392014-07-24 19:44:28 +10001423/*
1424 * A union-based inode filtering algorithm. Process the inode if any of the
1425 * criteria match. This is for global/internal scans only.
1426 */
1427STATIC int
1428xfs_inode_match_id_union(
1429 struct xfs_inode *ip,
1430 struct xfs_eofblocks *eofb)
1431{
1432 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1433 uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1434 return 1;
1435
1436 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1437 gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1438 return 1;
1439
1440 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1441 xfs_get_projid(ip) == eofb->eof_prid)
1442 return 1;
1443
1444 return 0;
1445}
1446
Brian Foster3e3f9f52012-11-07 12:21:13 -05001447STATIC int
Brian Foster41176a62012-11-06 09:50:42 -05001448xfs_inode_free_eofblocks(
1449 struct xfs_inode *ip,
Brian Foster41176a62012-11-06 09:50:42 -05001450 int flags,
1451 void *args)
1452{
Brian Fostera36b9262017-01-27 23:22:55 -08001453 int ret = 0;
Brian Foster3e3f9f52012-11-07 12:21:13 -05001454 struct xfs_eofblocks *eofb = args;
Brian Fosterf4526392014-07-24 19:44:28 +10001455 int match;
Brian Foster5400da72014-07-24 19:40:22 +10001456
Brian Foster41176a62012-11-06 09:50:42 -05001457 if (!xfs_can_free_eofblocks(ip, false)) {
1458 /* inode could be preallocated or append-only */
1459 trace_xfs_inode_free_eofblocks_invalid(ip);
1460 xfs_inode_clear_eofblocks_tag(ip);
1461 return 0;
1462 }
1463
1464 /*
1465 * If the mapping is dirty the operation can block and wait for some
1466 * time. Unless we are waiting, skip it.
1467 */
1468 if (!(flags & SYNC_WAIT) &&
1469 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1470 return 0;
1471
Brian Foster00ca79a2012-11-07 12:21:14 -05001472 if (eofb) {
Brian Fosterf4526392014-07-24 19:44:28 +10001473 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1474 match = xfs_inode_match_id_union(ip, eofb);
1475 else
1476 match = xfs_inode_match_id(ip, eofb);
1477 if (!match)
Brian Foster00ca79a2012-11-07 12:21:14 -05001478 return 0;
1479
1480 /* skip the inode if the file size is too small */
1481 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1482 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1483 return 0;
1484 }
Brian Foster3e3f9f52012-11-07 12:21:13 -05001485
Brian Fostera36b9262017-01-27 23:22:55 -08001486 /*
1487 * If the caller is waiting, return -EAGAIN to keep the background
1488 * scanner moving and revisit the inode in a subsequent pass.
1489 */
Brian Fosterc3155092017-01-27 23:22:56 -08001490 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
Brian Fostera36b9262017-01-27 23:22:55 -08001491 if (flags & SYNC_WAIT)
1492 ret = -EAGAIN;
1493 return ret;
1494 }
1495 ret = xfs_free_eofblocks(ip);
Brian Fosterc3155092017-01-27 23:22:56 -08001496 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Brian Foster41176a62012-11-06 09:50:42 -05001497
1498 return ret;
1499}
1500
Darrick J. Wong83104d42016-10-03 09:11:46 -07001501static int
1502__xfs_icache_free_eofblocks(
Brian Foster41176a62012-11-06 09:50:42 -05001503 struct xfs_mount *mp,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001504 struct xfs_eofblocks *eofb,
1505 int (*execute)(struct xfs_inode *ip, int flags,
1506 void *args),
1507 int tag)
Brian Foster41176a62012-11-06 09:50:42 -05001508{
Brian Foster8ca149d2012-11-07 12:21:12 -05001509 int flags = SYNC_TRYLOCK;
1510
1511 if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC))
1512 flags = SYNC_WAIT;
1513
Darrick J. Wong83104d42016-10-03 09:11:46 -07001514 return xfs_inode_ag_iterator_tag(mp, execute, flags,
1515 eofb, tag);
1516}
1517
1518int
1519xfs_icache_free_eofblocks(
1520 struct xfs_mount *mp,
1521 struct xfs_eofblocks *eofb)
1522{
1523 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_eofblocks,
1524 XFS_ICI_EOFBLOCKS_TAG);
Brian Foster41176a62012-11-06 09:50:42 -05001525}
1526
Brian Fosterdc06f3982014-07-24 19:49:28 +10001527/*
1528 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1529 * multiple quotas, we don't know exactly which quota caused an allocation
1530 * failure. We make a best effort by including each quota under low free space
1531 * conditions (less than 1% free space) in the scan.
1532 */
Darrick J. Wong83104d42016-10-03 09:11:46 -07001533static int
1534__xfs_inode_free_quota_eofblocks(
1535 struct xfs_inode *ip,
1536 int (*execute)(struct xfs_mount *mp,
1537 struct xfs_eofblocks *eofb))
Brian Fosterdc06f3982014-07-24 19:49:28 +10001538{
1539 int scan = 0;
1540 struct xfs_eofblocks eofb = {0};
1541 struct xfs_dquot *dq;
1542
Brian Fosterdc06f3982014-07-24 19:49:28 +10001543 /*
Brian Fosterc3155092017-01-27 23:22:56 -08001544 * Run a sync scan to increase effectiveness and use the union filter to
Brian Fosterdc06f3982014-07-24 19:49:28 +10001545 * cover all applicable quotas in a single scan.
1546 */
Brian Fosterdc06f3982014-07-24 19:49:28 +10001547 eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC;
1548
1549 if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) {
1550 dq = xfs_inode_dquot(ip, XFS_DQ_USER);
1551 if (dq && xfs_dquot_lowsp(dq)) {
1552 eofb.eof_uid = VFS_I(ip)->i_uid;
1553 eofb.eof_flags |= XFS_EOF_FLAGS_UID;
1554 scan = 1;
1555 }
1556 }
1557
1558 if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) {
1559 dq = xfs_inode_dquot(ip, XFS_DQ_GROUP);
1560 if (dq && xfs_dquot_lowsp(dq)) {
1561 eofb.eof_gid = VFS_I(ip)->i_gid;
1562 eofb.eof_flags |= XFS_EOF_FLAGS_GID;
1563 scan = 1;
1564 }
1565 }
1566
1567 if (scan)
Darrick J. Wong83104d42016-10-03 09:11:46 -07001568 execute(ip->i_mount, &eofb);
Brian Fosterdc06f3982014-07-24 19:49:28 +10001569
1570 return scan;
1571}
1572
Darrick J. Wong83104d42016-10-03 09:11:46 -07001573int
1574xfs_inode_free_quota_eofblocks(
1575 struct xfs_inode *ip)
1576{
1577 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks);
1578}
1579
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001580static inline unsigned long
1581xfs_iflag_for_tag(
1582 int tag)
1583{
1584 switch (tag) {
1585 case XFS_ICI_EOFBLOCKS_TAG:
1586 return XFS_IEOFBLOCKS;
1587 case XFS_ICI_COWBLOCKS_TAG:
1588 return XFS_ICOWBLOCKS;
1589 default:
1590 ASSERT(0);
1591 return 0;
1592 }
1593}
1594
Darrick J. Wong83104d42016-10-03 09:11:46 -07001595static void
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001596__xfs_inode_set_blocks_tag(
Darrick J. Wong83104d42016-10-03 09:11:46 -07001597 xfs_inode_t *ip,
1598 void (*execute)(struct xfs_mount *mp),
1599 void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1600 int error, unsigned long caller_ip),
1601 int tag)
Brian Foster27b52862012-11-06 09:50:38 -05001602{
1603 struct xfs_mount *mp = ip->i_mount;
1604 struct xfs_perag *pag;
1605 int tagged;
1606
Christoph Hellwig85a6e762016-09-19 11:09:48 +10001607 /*
1608 * Don't bother locking the AG and looking up in the radix trees
1609 * if we already know that we have the tag set.
1610 */
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001611 if (ip->i_flags & xfs_iflag_for_tag(tag))
Christoph Hellwig85a6e762016-09-19 11:09:48 +10001612 return;
1613 spin_lock(&ip->i_flags_lock);
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001614 ip->i_flags |= xfs_iflag_for_tag(tag);
Christoph Hellwig85a6e762016-09-19 11:09:48 +10001615 spin_unlock(&ip->i_flags_lock);
1616
Brian Foster27b52862012-11-06 09:50:38 -05001617 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1618 spin_lock(&pag->pag_ici_lock);
Brian Foster27b52862012-11-06 09:50:38 -05001619
Darrick J. Wong83104d42016-10-03 09:11:46 -07001620 tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
Brian Foster27b52862012-11-06 09:50:38 -05001621 radix_tree_tag_set(&pag->pag_ici_root,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001622 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
Brian Foster27b52862012-11-06 09:50:38 -05001623 if (!tagged) {
1624 /* propagate the eofblocks tag up into the perag radix tree */
1625 spin_lock(&ip->i_mount->m_perag_lock);
1626 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
1627 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
Darrick J. Wong83104d42016-10-03 09:11:46 -07001628 tag);
Brian Foster27b52862012-11-06 09:50:38 -05001629 spin_unlock(&ip->i_mount->m_perag_lock);
1630
Brian Foster579b62f2012-11-06 09:50:47 -05001631 /* kick off background trimming */
Darrick J. Wong83104d42016-10-03 09:11:46 -07001632 execute(ip->i_mount);
Brian Foster579b62f2012-11-06 09:50:47 -05001633
Darrick J. Wong83104d42016-10-03 09:11:46 -07001634 set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
Brian Foster27b52862012-11-06 09:50:38 -05001635 }
1636
1637 spin_unlock(&pag->pag_ici_lock);
1638 xfs_perag_put(pag);
1639}
1640
1641void
Darrick J. Wong83104d42016-10-03 09:11:46 -07001642xfs_inode_set_eofblocks_tag(
Brian Foster27b52862012-11-06 09:50:38 -05001643 xfs_inode_t *ip)
1644{
Darrick J. Wong83104d42016-10-03 09:11:46 -07001645 trace_xfs_inode_set_eofblocks_tag(ip);
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001646 return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001647 trace_xfs_perag_set_eofblocks,
1648 XFS_ICI_EOFBLOCKS_TAG);
1649}
1650
1651static void
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001652__xfs_inode_clear_blocks_tag(
Darrick J. Wong83104d42016-10-03 09:11:46 -07001653 xfs_inode_t *ip,
1654 void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1655 int error, unsigned long caller_ip),
1656 int tag)
1657{
Brian Foster27b52862012-11-06 09:50:38 -05001658 struct xfs_mount *mp = ip->i_mount;
1659 struct xfs_perag *pag;
1660
Christoph Hellwig85a6e762016-09-19 11:09:48 +10001661 spin_lock(&ip->i_flags_lock);
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001662 ip->i_flags &= ~xfs_iflag_for_tag(tag);
Christoph Hellwig85a6e762016-09-19 11:09:48 +10001663 spin_unlock(&ip->i_flags_lock);
1664
Brian Foster27b52862012-11-06 09:50:38 -05001665 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1666 spin_lock(&pag->pag_ici_lock);
Brian Foster27b52862012-11-06 09:50:38 -05001667
1668 radix_tree_tag_clear(&pag->pag_ici_root,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001669 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1670 if (!radix_tree_tagged(&pag->pag_ici_root, tag)) {
Brian Foster27b52862012-11-06 09:50:38 -05001671 /* clear the eofblocks tag from the perag radix tree */
1672 spin_lock(&ip->i_mount->m_perag_lock);
1673 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
1674 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
Darrick J. Wong83104d42016-10-03 09:11:46 -07001675 tag);
Brian Foster27b52862012-11-06 09:50:38 -05001676 spin_unlock(&ip->i_mount->m_perag_lock);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001677 clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
Brian Foster27b52862012-11-06 09:50:38 -05001678 }
1679
1680 spin_unlock(&pag->pag_ici_lock);
1681 xfs_perag_put(pag);
1682}
1683
Darrick J. Wong83104d42016-10-03 09:11:46 -07001684void
1685xfs_inode_clear_eofblocks_tag(
1686 xfs_inode_t *ip)
1687{
1688 trace_xfs_inode_clear_eofblocks_tag(ip);
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001689 return __xfs_inode_clear_blocks_tag(ip,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001690 trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG);
1691}
1692
1693/*
Darrick J. Wongbe78ff02018-01-16 19:03:59 -08001694 * Set ourselves up to free CoW blocks from this file. If it's already clean
1695 * then we can bail out quickly, but otherwise we must back off if the file
1696 * is undergoing some kind of write.
1697 */
1698static bool
1699xfs_prep_free_cowblocks(
Christoph Hellwig51d62692018-07-17 16:51:51 -07001700 struct xfs_inode *ip)
Darrick J. Wongbe78ff02018-01-16 19:03:59 -08001701{
1702 /*
1703 * Just clear the tag if we have an empty cow fork or none at all. It's
1704 * possible the inode was fully unshared since it was originally tagged.
1705 */
Christoph Hellwig51d62692018-07-17 16:51:51 -07001706 if (!xfs_inode_has_cow_data(ip)) {
Darrick J. Wongbe78ff02018-01-16 19:03:59 -08001707 trace_xfs_inode_free_cowblocks_invalid(ip);
1708 xfs_inode_clear_cowblocks_tag(ip);
1709 return false;
1710 }
1711
1712 /*
1713 * If the mapping is dirty or under writeback we cannot touch the
1714 * CoW fork. Leave it alone if we're in the midst of a directio.
1715 */
1716 if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1717 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1718 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1719 atomic_read(&VFS_I(ip)->i_dio_count))
1720 return false;
1721
1722 return true;
1723}
1724
1725/*
Darrick J. Wong83104d42016-10-03 09:11:46 -07001726 * Automatic CoW Reservation Freeing
1727 *
1728 * These functions automatically garbage collect leftover CoW reservations
1729 * that were made on behalf of a cowextsize hint when we start to run out
1730 * of quota or when the reservations sit around for too long. If the file
1731 * has dirty pages or is undergoing writeback, its CoW reservations will
1732 * be retained.
1733 *
1734 * The actual garbage collection piggybacks off the same code that runs
1735 * the speculative EOF preallocation garbage collector.
1736 */
1737STATIC int
1738xfs_inode_free_cowblocks(
1739 struct xfs_inode *ip,
1740 int flags,
1741 void *args)
1742{
Darrick J. Wongbe78ff02018-01-16 19:03:59 -08001743 struct xfs_eofblocks *eofb = args;
Darrick J. Wongbe78ff02018-01-16 19:03:59 -08001744 int match;
1745 int ret = 0;
Darrick J. Wong83104d42016-10-03 09:11:46 -07001746
Christoph Hellwig51d62692018-07-17 16:51:51 -07001747 if (!xfs_prep_free_cowblocks(ip))
Darrick J. Wong83104d42016-10-03 09:11:46 -07001748 return 0;
1749
1750 if (eofb) {
1751 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1752 match = xfs_inode_match_id_union(ip, eofb);
1753 else
1754 match = xfs_inode_match_id(ip, eofb);
1755 if (!match)
1756 return 0;
1757
1758 /* skip the inode if the file size is too small */
1759 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1760 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1761 return 0;
Darrick J. Wong83104d42016-10-03 09:11:46 -07001762 }
1763
1764 /* Free the CoW blocks */
Brian Fosterc3155092017-01-27 23:22:56 -08001765 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1766 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001767
Darrick J. Wongbe78ff02018-01-16 19:03:59 -08001768 /*
1769 * Check again, nobody else should be able to dirty blocks or change
1770 * the reflink iflag now that we have the first two locks held.
1771 */
Christoph Hellwig51d62692018-07-17 16:51:51 -07001772 if (xfs_prep_free_cowblocks(ip))
Darrick J. Wongbe78ff02018-01-16 19:03:59 -08001773 ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001774
Brian Fosterc3155092017-01-27 23:22:56 -08001775 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1776 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001777
1778 return ret;
1779}
1780
1781int
1782xfs_icache_free_cowblocks(
1783 struct xfs_mount *mp,
1784 struct xfs_eofblocks *eofb)
1785{
1786 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_cowblocks,
1787 XFS_ICI_COWBLOCKS_TAG);
1788}
1789
1790int
1791xfs_inode_free_quota_cowblocks(
1792 struct xfs_inode *ip)
1793{
1794 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks);
1795}
1796
1797void
1798xfs_inode_set_cowblocks_tag(
1799 xfs_inode_t *ip)
1800{
Brian Foster7b7381f2016-10-24 14:21:00 +11001801 trace_xfs_inode_set_cowblocks_tag(ip);
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001802 return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks,
Brian Foster7b7381f2016-10-24 14:21:00 +11001803 trace_xfs_perag_set_cowblocks,
Darrick J. Wong83104d42016-10-03 09:11:46 -07001804 XFS_ICI_COWBLOCKS_TAG);
1805}
1806
1807void
1808xfs_inode_clear_cowblocks_tag(
1809 xfs_inode_t *ip)
1810{
Brian Foster7b7381f2016-10-24 14:21:00 +11001811 trace_xfs_inode_clear_cowblocks_tag(ip);
Darrick J. Wong91aae6b2017-12-14 15:42:22 -08001812 return __xfs_inode_clear_blocks_tag(ip,
Brian Foster7b7381f2016-10-24 14:21:00 +11001813 trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG);
Darrick J. Wong83104d42016-10-03 09:11:46 -07001814}
Darrick J. Wongd6b636e2018-05-09 10:03:56 -07001815
1816/* Disable post-EOF and CoW block auto-reclamation. */
1817void
1818xfs_icache_disable_reclaim(
1819 struct xfs_mount *mp)
1820{
1821 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1822 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1823}
1824
1825/* Enable post-EOF and CoW block auto-reclamation. */
1826void
1827xfs_icache_enable_reclaim(
1828 struct xfs_mount *mp)
1829{
1830 xfs_queue_eofblocks(mp);
1831 xfs_queue_cowblocks(mp);
1832}