2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_btree.h"
34 #include "xfs_ialloc.h"
35 #include "xfs_quota.h"
36 #include "xfs_utils.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_inode_item.h"
40 #include "xfs_trace.h"
44 * Define xfs inode iolock lockdep classes. We need to ensure that all active
45 * inodes are considered the same for lockdep purposes, including inodes that
46 * are recycled through the XFS_IRECLAIMABLE state. This is the the only way to
47 * guarantee the locks are considered the same when there are multiple lock
48 * initialisation siteѕ. Also, define a reclaimable inode class so it is
49 * obvious in lockdep reports which class the report is against.
51 static struct lock_class_key xfs_iolock_active;
52 struct lock_class_key xfs_iolock_reclaimable;
55 * Allocate and initialise an xfs_inode.
57 STATIC struct xfs_inode *
65 * if this didn't occur in transactions, we could use
66 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
67 * code up to do this anyway.
69 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
72 if (inode_init_always(mp->m_super, VFS_I(ip))) {
73 kmem_zone_free(xfs_inode_zone, ip);
77 ASSERT(atomic_read(&ip->i_pincount) == 0);
78 ASSERT(!spin_is_locked(&ip->i_flags_lock));
79 ASSERT(!xfs_isiflocked(ip));
80 ASSERT(ip->i_ino == 0);
82 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
83 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
84 &xfs_iolock_active, "xfs_iolock_active");
86 /* initialise the xfs inode */
89 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
91 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
93 ip->i_delayed_blks = 0;
94 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
100 xfs_inode_free_callback(
101 struct rcu_head *head)
103 struct inode *inode = container_of(head, struct inode, i_rcu);
104 struct xfs_inode *ip = XFS_I(inode);
106 kmem_zone_free(xfs_inode_zone, ip);
111 struct xfs_inode *ip)
113 switch (ip->i_d.di_mode & S_IFMT) {
117 xfs_idestroy_fork(ip, XFS_DATA_FORK);
122 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
125 ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL));
126 xfs_inode_item_destroy(ip);
130 /* asserts to verify all state is correct here */
131 ASSERT(atomic_read(&ip->i_pincount) == 0);
132 ASSERT(!spin_is_locked(&ip->i_flags_lock));
133 ASSERT(!xfs_isiflocked(ip));
136 * Because we use RCU freeing we need to ensure the inode always
137 * appears to be reclaimed with an invalid inode number when in the
138 * free state. The ip->i_flags_lock provides the barrier against lookup
141 spin_lock(&ip->i_flags_lock);
142 ip->i_flags = XFS_IRECLAIM;
144 spin_unlock(&ip->i_flags_lock);
146 call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
150 * Check the validity of the inode we just found it the cache
154 struct xfs_perag *pag,
155 struct xfs_inode *ip,
158 int lock_flags) __releases(RCU)
160 struct inode *inode = VFS_I(ip);
161 struct xfs_mount *mp = ip->i_mount;
165 * check for re-use of an inode within an RCU grace period due to the
166 * radix tree nodes not being updated yet. We monitor for this by
167 * setting the inode number to zero before freeing the inode structure.
168 * If the inode has been reallocated and set up, then the inode number
169 * will not match, so check for that, too.
171 spin_lock(&ip->i_flags_lock);
172 if (ip->i_ino != ino) {
173 trace_xfs_iget_skip(ip);
174 XFS_STATS_INC(xs_ig_frecycle);
181 * If we are racing with another cache hit that is currently
182 * instantiating this inode or currently recycling it out of
183 * reclaimabe state, wait for the initialisation to complete
186 * XXX(hch): eventually we should do something equivalent to
187 * wait_on_inode to wait for these flags to be cleared
188 * instead of polling for it.
190 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
191 trace_xfs_iget_skip(ip);
192 XFS_STATS_INC(xs_ig_frecycle);
198 * If lookup is racing with unlink return an error immediately.
200 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
206 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
207 * Need to carefully get it back into useable state.
209 if (ip->i_flags & XFS_IRECLAIMABLE) {
210 trace_xfs_iget_reclaim(ip);
213 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
214 * from stomping over us while we recycle the inode. We can't
215 * clear the radix tree reclaimable tag yet as it requires
216 * pag_ici_lock to be held exclusive.
218 ip->i_flags |= XFS_IRECLAIM;
220 spin_unlock(&ip->i_flags_lock);
223 error = -inode_init_always(mp->m_super, inode);
226 * Re-initializing the inode failed, and we are in deep
227 * trouble. Try to re-add it to the reclaim list.
230 spin_lock(&ip->i_flags_lock);
232 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
233 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
234 trace_xfs_iget_reclaim_fail(ip);
238 spin_lock(&pag->pag_ici_lock);
239 spin_lock(&ip->i_flags_lock);
242 * Clear the per-lifetime state in the inode as we are now
243 * effectively a new inode and need to return to the initial
244 * state before reuse occurs.
246 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
247 ip->i_flags |= XFS_INEW;
248 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
249 inode->i_state = I_NEW;
251 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
252 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
253 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
254 &xfs_iolock_active, "xfs_iolock_active");
256 spin_unlock(&ip->i_flags_lock);
257 spin_unlock(&pag->pag_ici_lock);
259 /* If the VFS inode is being torn down, pause and try again. */
261 trace_xfs_iget_skip(ip);
266 /* We've got a live one. */
267 spin_unlock(&ip->i_flags_lock);
269 trace_xfs_iget_hit(ip);
273 xfs_ilock(ip, lock_flags);
275 xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE);
276 XFS_STATS_INC(xs_ig_found);
281 spin_unlock(&ip->i_flags_lock);
289 struct xfs_mount *mp,
290 struct xfs_perag *pag,
293 struct xfs_inode **ipp,
297 struct xfs_inode *ip;
299 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
302 ip = xfs_inode_alloc(mp, ino);
306 error = xfs_iread(mp, tp, ip, flags);
310 trace_xfs_iget_miss(ip);
312 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
318 * Preload the radix tree so we can insert safely under the
319 * write spinlock. Note that we cannot sleep inside the preload
320 * region. Since we can be called from transaction context, don't
321 * recurse into the file system.
323 if (radix_tree_preload(GFP_NOFS)) {
329 * Because the inode hasn't been added to the radix-tree yet it can't
330 * be found by another thread, so we can do the non-sleeping lock here.
333 if (!xfs_ilock_nowait(ip, lock_flags))
338 * These values must be set before inserting the inode into the radix
339 * tree as the moment it is inserted a concurrent lookup (allowed by the
340 * RCU locking mechanism) can find it and that lookup must see that this
341 * is an inode currently under construction (i.e. that XFS_INEW is set).
342 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
343 * memory barrier that ensures this detection works correctly at lookup
347 if (flags & XFS_IGET_DONTCACHE)
348 iflags |= XFS_IDONTCACHE;
349 ip->i_udquot = ip->i_gdquot = NULL;
350 xfs_iflags_set(ip, iflags);
352 /* insert the new inode */
353 spin_lock(&pag->pag_ici_lock);
354 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
355 if (unlikely(error)) {
356 WARN_ON(error != -EEXIST);
357 XFS_STATS_INC(xs_ig_dup);
359 goto out_preload_end;
361 spin_unlock(&pag->pag_ici_lock);
362 radix_tree_preload_end();
368 spin_unlock(&pag->pag_ici_lock);
369 radix_tree_preload_end();
371 xfs_iunlock(ip, lock_flags);
373 __destroy_inode(VFS_I(ip));
379 * Look up an inode by number in the given file system.
380 * The inode is looked up in the cache held in each AG.
381 * If the inode is found in the cache, initialise the vfs inode
384 * If it is not in core, read it in from the file system's device,
385 * add it to the cache and initialise the vfs inode.
387 * The inode is locked according to the value of the lock_flags parameter.
388 * This flag parameter indicates how and if the inode's IO lock and inode lock
391 * mp -- the mount point structure for the current file system. It points
392 * to the inode hash table.
393 * tp -- a pointer to the current transaction if there is one. This is
394 * simply passed through to the xfs_iread() call.
395 * ino -- the number of the inode desired. This is the unique identifier
396 * within the file system for the inode being requested.
397 * lock_flags -- flags indicating how to lock the inode. See the comment
398 * for xfs_ilock() for a list of valid values.
415 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
416 * doesn't get freed while it's being referenced during a
417 * radix tree traversal here. It assumes this function
418 * aqcuires only the ILOCK (and therefore it has no need to
419 * involve the IOLOCK in this synchronization).
421 ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
423 /* reject inode numbers outside existing AGs */
424 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
427 /* get the perag structure and ensure that it's inode capable */
428 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
429 agino = XFS_INO_TO_AGINO(mp, ino);
434 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
437 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
439 goto out_error_or_again;
442 XFS_STATS_INC(xs_ig_missed);
444 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
447 goto out_error_or_again;
454 * If we have a real type for an on-disk inode, we can set ops(&unlock)
455 * now. If it's a new inode being created, xfs_ialloc will handle it.
457 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
462 if (error == EAGAIN) {
471 * This is a wrapper routine around the xfs_ilock() routine
472 * used to centralize some grungy code. It is used in places
473 * that wish to lock the inode solely for reading the extents.
474 * The reason these places can't just call xfs_ilock(SHARED)
475 * is that the inode lock also guards to bringing in of the
476 * extents from disk for a file in b-tree format. If the inode
477 * is in b-tree format, then we need to lock the inode exclusively
478 * until the extents are read in. Locking it exclusively all
479 * the time would limit our parallelism unnecessarily, though.
480 * What we do instead is check to see if the extents have been
481 * read in yet, and only lock the inode exclusively if they
484 * The function returns a value which should be given to the
485 * corresponding xfs_iunlock_map_shared(). This value is
486 * the mode in which the lock was actually taken.
489 xfs_ilock_map_shared(
494 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
495 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
496 lock_mode = XFS_ILOCK_EXCL;
498 lock_mode = XFS_ILOCK_SHARED;
501 xfs_ilock(ip, lock_mode);
507 * This is simply the unlock routine to go with xfs_ilock_map_shared().
508 * All it does is call xfs_iunlock() with the given lock_mode.
511 xfs_iunlock_map_shared(
513 unsigned int lock_mode)
515 xfs_iunlock(ip, lock_mode);
519 * The xfs inode contains 2 locks: a multi-reader lock called the
520 * i_iolock and a multi-reader lock called the i_lock. This routine
521 * allows either or both of the locks to be obtained.
523 * The 2 locks should always be ordered so that the IO lock is
524 * obtained first in order to prevent deadlock.
526 * ip -- the inode being locked
527 * lock_flags -- this parameter indicates the inode's locks
528 * to be locked. It can be:
533 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
534 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
535 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
536 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
544 * You can't set both SHARED and EXCL for the same lock,
545 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
546 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
548 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
549 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
550 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
551 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
552 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
554 if (lock_flags & XFS_IOLOCK_EXCL)
555 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
556 else if (lock_flags & XFS_IOLOCK_SHARED)
557 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
559 if (lock_flags & XFS_ILOCK_EXCL)
560 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
561 else if (lock_flags & XFS_ILOCK_SHARED)
562 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
564 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
568 * This is just like xfs_ilock(), except that the caller
569 * is guaranteed not to sleep. It returns 1 if it gets
570 * the requested locks and 0 otherwise. If the IO lock is
571 * obtained but the inode lock cannot be, then the IO lock
572 * is dropped before returning.
574 * ip -- the inode being locked
575 * lock_flags -- this parameter indicates the inode's locks to be
576 * to be locked. See the comment for xfs_ilock() for a list
585 * You can't set both SHARED and EXCL for the same lock,
586 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
587 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
589 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
590 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
591 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
592 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
593 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
595 if (lock_flags & XFS_IOLOCK_EXCL) {
596 if (!mrtryupdate(&ip->i_iolock))
598 } else if (lock_flags & XFS_IOLOCK_SHARED) {
599 if (!mrtryaccess(&ip->i_iolock))
602 if (lock_flags & XFS_ILOCK_EXCL) {
603 if (!mrtryupdate(&ip->i_lock))
604 goto out_undo_iolock;
605 } else if (lock_flags & XFS_ILOCK_SHARED) {
606 if (!mrtryaccess(&ip->i_lock))
607 goto out_undo_iolock;
609 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
613 if (lock_flags & XFS_IOLOCK_EXCL)
614 mrunlock_excl(&ip->i_iolock);
615 else if (lock_flags & XFS_IOLOCK_SHARED)
616 mrunlock_shared(&ip->i_iolock);
622 * xfs_iunlock() is used to drop the inode locks acquired with
623 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
624 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
625 * that we know which locks to drop.
627 * ip -- the inode being unlocked
628 * lock_flags -- this parameter indicates the inode's locks to be
629 * to be unlocked. See the comment for xfs_ilock() for a list
630 * of valid values for this parameter.
639 * You can't set both SHARED and EXCL for the same lock,
640 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
641 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
643 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
644 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
645 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
646 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
647 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
648 ASSERT(lock_flags != 0);
650 if (lock_flags & XFS_IOLOCK_EXCL)
651 mrunlock_excl(&ip->i_iolock);
652 else if (lock_flags & XFS_IOLOCK_SHARED)
653 mrunlock_shared(&ip->i_iolock);
655 if (lock_flags & XFS_ILOCK_EXCL)
656 mrunlock_excl(&ip->i_lock);
657 else if (lock_flags & XFS_ILOCK_SHARED)
658 mrunlock_shared(&ip->i_lock);
660 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
664 * give up write locks. the i/o lock cannot be held nested
665 * if it is being demoted.
672 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
673 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
675 if (lock_flags & XFS_ILOCK_EXCL)
676 mrdemote(&ip->i_lock);
677 if (lock_flags & XFS_IOLOCK_EXCL)
678 mrdemote(&ip->i_iolock);
680 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
689 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
690 if (!(lock_flags & XFS_ILOCK_SHARED))
691 return !!ip->i_lock.mr_writer;
692 return rwsem_is_locked(&ip->i_lock.mr_lock);
695 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
696 if (!(lock_flags & XFS_IOLOCK_SHARED))
697 return !!ip->i_iolock.mr_writer;
698 return rwsem_is_locked(&ip->i_iolock.mr_lock);
708 struct xfs_inode *ip)
710 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
711 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
714 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
715 if (xfs_isiflocked(ip))
717 } while (!xfs_iflock_nowait(ip));
719 finish_wait(wq, &wait.wait);