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"
25 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_quota.h"
37 #include "xfs_utils.h"
38 #include "xfs_trans_priv.h"
39 #include "xfs_inode_item.h"
41 #include "xfs_trace.h"
45 * Define xfs inode iolock lockdep classes. We need to ensure that all active
46 * inodes are considered the same for lockdep purposes, including inodes that
47 * are recycled through the XFS_IRECLAIMABLE state. This is the the only way to
48 * guarantee the locks are considered the same when there are multiple lock
49 * initialisation siteѕ. Also, define a reclaimable inode class so it is
50 * obvious in lockdep reports which class the report is against.
52 static struct lock_class_key xfs_iolock_active;
53 struct lock_class_key xfs_iolock_reclaimable;
56 * Allocate and initialise an xfs_inode.
58 STATIC struct xfs_inode *
66 * if this didn't occur in transactions, we could use
67 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
68 * code up to do this anyway.
70 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
73 if (inode_init_always(mp->m_super, VFS_I(ip))) {
74 kmem_zone_free(xfs_inode_zone, ip);
78 ASSERT(atomic_read(&ip->i_pincount) == 0);
79 ASSERT(!spin_is_locked(&ip->i_flags_lock));
80 ASSERT(completion_done(&ip->i_flush));
81 ASSERT(ip->i_ino == 0);
83 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
84 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
85 &xfs_iolock_active, "xfs_iolock_active");
87 /* initialise the xfs inode */
90 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
92 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
94 ip->i_update_core = 0;
95 ip->i_delayed_blks = 0;
96 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
104 xfs_inode_free_callback(
105 struct rcu_head *head)
107 struct inode *inode = container_of(head, struct inode, i_rcu);
108 struct xfs_inode *ip = XFS_I(inode);
110 INIT_LIST_HEAD(&inode->i_dentry);
111 kmem_zone_free(xfs_inode_zone, ip);
116 struct xfs_inode *ip)
118 switch (ip->i_d.di_mode & S_IFMT) {
122 xfs_idestroy_fork(ip, XFS_DATA_FORK);
127 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
131 * Only if we are shutting down the fs will we see an
132 * inode still in the AIL. If it is there, we should remove
133 * it to prevent a use-after-free from occurring.
135 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
136 struct xfs_ail *ailp = lip->li_ailp;
138 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
139 XFS_FORCED_SHUTDOWN(ip->i_mount));
140 if (lip->li_flags & XFS_LI_IN_AIL) {
141 spin_lock(&ailp->xa_lock);
142 if (lip->li_flags & XFS_LI_IN_AIL)
143 xfs_trans_ail_delete(ailp, lip);
145 spin_unlock(&ailp->xa_lock);
147 xfs_inode_item_destroy(ip);
151 /* asserts to verify all state is correct here */
152 ASSERT(atomic_read(&ip->i_pincount) == 0);
153 ASSERT(!spin_is_locked(&ip->i_flags_lock));
154 ASSERT(completion_done(&ip->i_flush));
157 * Because we use RCU freeing we need to ensure the inode always
158 * appears to be reclaimed with an invalid inode number when in the
159 * free state. The ip->i_flags_lock provides the barrier against lookup
162 spin_lock(&ip->i_flags_lock);
163 ip->i_flags = XFS_IRECLAIM;
165 spin_unlock(&ip->i_flags_lock);
167 call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
171 * Check the validity of the inode we just found it the cache
175 struct xfs_perag *pag,
176 struct xfs_inode *ip,
179 int lock_flags) __releases(RCU)
181 struct inode *inode = VFS_I(ip);
182 struct xfs_mount *mp = ip->i_mount;
186 * check for re-use of an inode within an RCU grace period due to the
187 * radix tree nodes not being updated yet. We monitor for this by
188 * setting the inode number to zero before freeing the inode structure.
189 * If the inode has been reallocated and set up, then the inode number
190 * will not match, so check for that, too.
192 spin_lock(&ip->i_flags_lock);
193 if (ip->i_ino != ino) {
194 trace_xfs_iget_skip(ip);
195 XFS_STATS_INC(xs_ig_frecycle);
202 * If we are racing with another cache hit that is currently
203 * instantiating this inode or currently recycling it out of
204 * reclaimabe state, wait for the initialisation to complete
207 * XXX(hch): eventually we should do something equivalent to
208 * wait_on_inode to wait for these flags to be cleared
209 * instead of polling for it.
211 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
212 trace_xfs_iget_skip(ip);
213 XFS_STATS_INC(xs_ig_frecycle);
219 * If lookup is racing with unlink return an error immediately.
221 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
227 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
228 * Need to carefully get it back into useable state.
230 if (ip->i_flags & XFS_IRECLAIMABLE) {
231 trace_xfs_iget_reclaim(ip);
234 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
235 * from stomping over us while we recycle the inode. We can't
236 * clear the radix tree reclaimable tag yet as it requires
237 * pag_ici_lock to be held exclusive.
239 ip->i_flags |= XFS_IRECLAIM;
241 spin_unlock(&ip->i_flags_lock);
244 error = -inode_init_always(mp->m_super, inode);
247 * Re-initializing the inode failed, and we are in deep
248 * trouble. Try to re-add it to the reclaim list.
251 spin_lock(&ip->i_flags_lock);
253 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
254 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
255 trace_xfs_iget_reclaim_fail(ip);
259 spin_lock(&pag->pag_ici_lock);
260 spin_lock(&ip->i_flags_lock);
263 * Clear the per-lifetime state in the inode as we are now
264 * effectively a new inode and need to return to the initial
265 * state before reuse occurs.
267 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
268 ip->i_flags |= XFS_INEW;
269 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
270 inode->i_state = I_NEW;
272 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
273 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
274 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
275 &xfs_iolock_active, "xfs_iolock_active");
277 spin_unlock(&ip->i_flags_lock);
278 spin_unlock(&pag->pag_ici_lock);
280 /* If the VFS inode is being torn down, pause and try again. */
282 trace_xfs_iget_skip(ip);
287 /* We've got a live one. */
288 spin_unlock(&ip->i_flags_lock);
290 trace_xfs_iget_hit(ip);
294 xfs_ilock(ip, lock_flags);
296 xfs_iflags_clear(ip, XFS_ISTALE);
297 XFS_STATS_INC(xs_ig_found);
302 spin_unlock(&ip->i_flags_lock);
310 struct xfs_mount *mp,
311 struct xfs_perag *pag,
314 struct xfs_inode **ipp,
318 struct xfs_inode *ip;
320 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
322 ip = xfs_inode_alloc(mp, ino);
326 error = xfs_iread(mp, tp, ip, flags);
330 trace_xfs_iget_miss(ip);
332 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
338 * Preload the radix tree so we can insert safely under the
339 * write spinlock. Note that we cannot sleep inside the preload
342 if (radix_tree_preload(GFP_KERNEL)) {
348 * Because the inode hasn't been added to the radix-tree yet it can't
349 * be found by another thread, so we can do the non-sleeping lock here.
352 if (!xfs_ilock_nowait(ip, lock_flags))
357 * These values must be set before inserting the inode into the radix
358 * tree as the moment it is inserted a concurrent lookup (allowed by the
359 * RCU locking mechanism) can find it and that lookup must see that this
360 * is an inode currently under construction (i.e. that XFS_INEW is set).
361 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
362 * memory barrier that ensures this detection works correctly at lookup
365 ip->i_udquot = ip->i_gdquot = NULL;
366 xfs_iflags_set(ip, XFS_INEW);
368 /* insert the new inode */
369 spin_lock(&pag->pag_ici_lock);
370 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
371 if (unlikely(error)) {
372 WARN_ON(error != -EEXIST);
373 XFS_STATS_INC(xs_ig_dup);
375 goto out_preload_end;
377 spin_unlock(&pag->pag_ici_lock);
378 radix_tree_preload_end();
384 spin_unlock(&pag->pag_ici_lock);
385 radix_tree_preload_end();
387 xfs_iunlock(ip, lock_flags);
389 __destroy_inode(VFS_I(ip));
395 * Look up an inode by number in the given file system.
396 * The inode is looked up in the cache held in each AG.
397 * If the inode is found in the cache, initialise the vfs inode
400 * If it is not in core, read it in from the file system's device,
401 * add it to the cache and initialise the vfs inode.
403 * The inode is locked according to the value of the lock_flags parameter.
404 * This flag parameter indicates how and if the inode's IO lock and inode lock
407 * mp -- the mount point structure for the current file system. It points
408 * to the inode hash table.
409 * tp -- a pointer to the current transaction if there is one. This is
410 * simply passed through to the xfs_iread() call.
411 * ino -- the number of the inode desired. This is the unique identifier
412 * within the file system for the inode being requested.
413 * lock_flags -- flags indicating how to lock the inode. See the comment
414 * for xfs_ilock() for a list of valid values.
430 /* reject inode numbers outside existing AGs */
431 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
434 /* get the perag structure and ensure that it's inode capable */
435 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
436 agino = XFS_INO_TO_AGINO(mp, ino);
441 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
444 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
446 goto out_error_or_again;
449 XFS_STATS_INC(xs_ig_missed);
451 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
454 goto out_error_or_again;
460 ASSERT(ip->i_df.if_ext_max ==
461 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
463 * If we have a real type for an on-disk inode, we can set ops(&unlock)
464 * now. If it's a new inode being created, xfs_ialloc will handle it.
466 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
471 if (error == EAGAIN) {
480 * This is a wrapper routine around the xfs_ilock() routine
481 * used to centralize some grungy code. It is used in places
482 * that wish to lock the inode solely for reading the extents.
483 * The reason these places can't just call xfs_ilock(SHARED)
484 * is that the inode lock also guards to bringing in of the
485 * extents from disk for a file in b-tree format. If the inode
486 * is in b-tree format, then we need to lock the inode exclusively
487 * until the extents are read in. Locking it exclusively all
488 * the time would limit our parallelism unnecessarily, though.
489 * What we do instead is check to see if the extents have been
490 * read in yet, and only lock the inode exclusively if they
493 * The function returns a value which should be given to the
494 * corresponding xfs_iunlock_map_shared(). This value is
495 * the mode in which the lock was actually taken.
498 xfs_ilock_map_shared(
503 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
504 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
505 lock_mode = XFS_ILOCK_EXCL;
507 lock_mode = XFS_ILOCK_SHARED;
510 xfs_ilock(ip, lock_mode);
516 * This is simply the unlock routine to go with xfs_ilock_map_shared().
517 * All it does is call xfs_iunlock() with the given lock_mode.
520 xfs_iunlock_map_shared(
522 unsigned int lock_mode)
524 xfs_iunlock(ip, lock_mode);
528 * The xfs inode contains 2 locks: a multi-reader lock called the
529 * i_iolock and a multi-reader lock called the i_lock. This routine
530 * allows either or both of the locks to be obtained.
532 * The 2 locks should always be ordered so that the IO lock is
533 * obtained first in order to prevent deadlock.
535 * ip -- the inode being locked
536 * lock_flags -- this parameter indicates the inode's locks
537 * to be locked. It can be:
542 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
543 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
544 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
545 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
553 * You can't set both SHARED and EXCL for the same lock,
554 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
555 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
557 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
558 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
559 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
560 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
561 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
563 if (lock_flags & XFS_IOLOCK_EXCL)
564 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
565 else if (lock_flags & XFS_IOLOCK_SHARED)
566 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
568 if (lock_flags & XFS_ILOCK_EXCL)
569 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
570 else if (lock_flags & XFS_ILOCK_SHARED)
571 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
573 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
577 * This is just like xfs_ilock(), except that the caller
578 * is guaranteed not to sleep. It returns 1 if it gets
579 * the requested locks and 0 otherwise. If the IO lock is
580 * obtained but the inode lock cannot be, then the IO lock
581 * is dropped before returning.
583 * ip -- the inode being locked
584 * lock_flags -- this parameter indicates the inode's locks to be
585 * to be locked. See the comment for xfs_ilock() for a list
594 * You can't set both SHARED and EXCL for the same lock,
595 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
596 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
598 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
599 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
600 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
601 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
602 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
604 if (lock_flags & XFS_IOLOCK_EXCL) {
605 if (!mrtryupdate(&ip->i_iolock))
607 } else if (lock_flags & XFS_IOLOCK_SHARED) {
608 if (!mrtryaccess(&ip->i_iolock))
611 if (lock_flags & XFS_ILOCK_EXCL) {
612 if (!mrtryupdate(&ip->i_lock))
613 goto out_undo_iolock;
614 } else if (lock_flags & XFS_ILOCK_SHARED) {
615 if (!mrtryaccess(&ip->i_lock))
616 goto out_undo_iolock;
618 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
622 if (lock_flags & XFS_IOLOCK_EXCL)
623 mrunlock_excl(&ip->i_iolock);
624 else if (lock_flags & XFS_IOLOCK_SHARED)
625 mrunlock_shared(&ip->i_iolock);
631 * xfs_iunlock() is used to drop the inode locks acquired with
632 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
633 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
634 * that we know which locks to drop.
636 * ip -- the inode being unlocked
637 * lock_flags -- this parameter indicates the inode's locks to be
638 * to be unlocked. See the comment for xfs_ilock() for a list
639 * of valid values for this parameter.
648 * You can't set both SHARED and EXCL for the same lock,
649 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
650 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
652 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
653 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
654 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
655 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
656 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
657 XFS_LOCK_DEP_MASK)) == 0);
658 ASSERT(lock_flags != 0);
660 if (lock_flags & XFS_IOLOCK_EXCL)
661 mrunlock_excl(&ip->i_iolock);
662 else if (lock_flags & XFS_IOLOCK_SHARED)
663 mrunlock_shared(&ip->i_iolock);
665 if (lock_flags & XFS_ILOCK_EXCL)
666 mrunlock_excl(&ip->i_lock);
667 else if (lock_flags & XFS_ILOCK_SHARED)
668 mrunlock_shared(&ip->i_lock);
670 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
671 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
673 * Let the AIL know that this item has been unlocked in case
674 * it is in the AIL and anyone is waiting on it. Don't do
675 * this if the caller has asked us not to.
677 xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp,
678 (xfs_log_item_t*)(ip->i_itemp));
680 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
684 * give up write locks. the i/o lock cannot be held nested
685 * if it is being demoted.
692 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
693 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
695 if (lock_flags & XFS_ILOCK_EXCL)
696 mrdemote(&ip->i_lock);
697 if (lock_flags & XFS_IOLOCK_EXCL)
698 mrdemote(&ip->i_iolock);
700 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
709 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
710 if (!(lock_flags & XFS_ILOCK_SHARED))
711 return !!ip->i_lock.mr_writer;
712 return rwsem_is_locked(&ip->i_lock.mr_lock);
715 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
716 if (!(lock_flags & XFS_IOLOCK_SHARED))
717 return !!ip->i_iolock.mr_writer;
718 return rwsem_is_locked(&ip->i_iolock.mr_lock);