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_iocount
) == 0);
79 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
80 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
81 ASSERT(completion_done(&ip
->i_flush
));
82 ASSERT(ip
->i_ino
== 0);
84 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
85 lockdep_set_class_and_name(&ip
->i_iolock
.mr_lock
,
86 &xfs_iolock_active
, "xfs_iolock_active");
88 /* initialise the xfs inode */
91 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
93 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
95 ip
->i_update_core
= 0;
96 ip
->i_delayed_blks
= 0;
97 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
105 xfs_inode_free_callback(
106 struct rcu_head
*head
)
108 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
109 struct xfs_inode
*ip
= XFS_I(inode
);
111 INIT_LIST_HEAD(&inode
->i_dentry
);
112 kmem_zone_free(xfs_inode_zone
, ip
);
117 struct xfs_inode
*ip
)
119 switch (ip
->i_d
.di_mode
& S_IFMT
) {
123 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
128 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
132 * Only if we are shutting down the fs will we see an
133 * inode still in the AIL. If it is there, we should remove
134 * it to prevent a use-after-free from occurring.
136 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
137 struct xfs_ail
*ailp
= lip
->li_ailp
;
139 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
140 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
141 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
142 spin_lock(&ailp
->xa_lock
);
143 if (lip
->li_flags
& XFS_LI_IN_AIL
)
144 xfs_trans_ail_delete(ailp
, lip
);
146 spin_unlock(&ailp
->xa_lock
);
148 xfs_inode_item_destroy(ip
);
152 /* asserts to verify all state is correct here */
153 ASSERT(atomic_read(&ip
->i_iocount
) == 0);
154 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
155 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
156 ASSERT(completion_done(&ip
->i_flush
));
159 * Because we use RCU freeing we need to ensure the inode always
160 * appears to be reclaimed with an invalid inode number when in the
161 * free state. The ip->i_flags_lock provides the barrier against lookup
164 spin_lock(&ip
->i_flags_lock
);
165 ip
->i_flags
= XFS_IRECLAIM
;
167 spin_unlock(&ip
->i_flags_lock
);
169 call_rcu(&VFS_I(ip
)->i_rcu
, xfs_inode_free_callback
);
173 * Check the validity of the inode we just found it the cache
177 struct xfs_perag
*pag
,
178 struct xfs_inode
*ip
,
181 int lock_flags
) __releases(RCU
)
183 struct inode
*inode
= VFS_I(ip
);
184 struct xfs_mount
*mp
= ip
->i_mount
;
188 * check for re-use of an inode within an RCU grace period due to the
189 * radix tree nodes not being updated yet. We monitor for this by
190 * setting the inode number to zero before freeing the inode structure.
191 * If the inode has been reallocated and set up, then the inode number
192 * will not match, so check for that, too.
194 spin_lock(&ip
->i_flags_lock
);
195 if (ip
->i_ino
!= ino
) {
196 trace_xfs_iget_skip(ip
);
197 XFS_STATS_INC(xs_ig_frecycle
);
204 * If we are racing with another cache hit that is currently
205 * instantiating this inode or currently recycling it out of
206 * reclaimabe state, wait for the initialisation to complete
209 * XXX(hch): eventually we should do something equivalent to
210 * wait_on_inode to wait for these flags to be cleared
211 * instead of polling for it.
213 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
214 trace_xfs_iget_skip(ip
);
215 XFS_STATS_INC(xs_ig_frecycle
);
221 * If lookup is racing with unlink return an error immediately.
223 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
229 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
230 * Need to carefully get it back into useable state.
232 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
233 trace_xfs_iget_reclaim(ip
);
236 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
237 * from stomping over us while we recycle the inode. We can't
238 * clear the radix tree reclaimable tag yet as it requires
239 * pag_ici_lock to be held exclusive.
241 ip
->i_flags
|= XFS_IRECLAIM
;
243 spin_unlock(&ip
->i_flags_lock
);
246 error
= -inode_init_always(mp
->m_super
, inode
);
249 * Re-initializing the inode failed, and we are in deep
250 * trouble. Try to re-add it to the reclaim list.
253 spin_lock(&ip
->i_flags_lock
);
255 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
256 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
257 trace_xfs_iget_reclaim_fail(ip
);
261 spin_lock(&pag
->pag_ici_lock
);
262 spin_lock(&ip
->i_flags_lock
);
265 * Clear the per-lifetime state in the inode as we are now
266 * effectively a new inode and need to return to the initial
267 * state before reuse occurs.
269 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
270 ip
->i_flags
|= XFS_INEW
;
271 __xfs_inode_clear_reclaim_tag(mp
, pag
, ip
);
272 inode
->i_state
= I_NEW
;
274 ASSERT(!rwsem_is_locked(&ip
->i_iolock
.mr_lock
));
275 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
276 lockdep_set_class_and_name(&ip
->i_iolock
.mr_lock
,
277 &xfs_iolock_active
, "xfs_iolock_active");
279 spin_unlock(&ip
->i_flags_lock
);
280 spin_unlock(&pag
->pag_ici_lock
);
282 /* If the VFS inode is being torn down, pause and try again. */
284 trace_xfs_iget_skip(ip
);
289 /* We've got a live one. */
290 spin_unlock(&ip
->i_flags_lock
);
292 trace_xfs_iget_hit(ip
);
296 xfs_ilock(ip
, lock_flags
);
298 xfs_iflags_clear(ip
, XFS_ISTALE
);
299 XFS_STATS_INC(xs_ig_found
);
304 spin_unlock(&ip
->i_flags_lock
);
312 struct xfs_mount
*mp
,
313 struct xfs_perag
*pag
,
316 struct xfs_inode
**ipp
,
320 struct xfs_inode
*ip
;
322 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
324 ip
= xfs_inode_alloc(mp
, ino
);
328 error
= xfs_iread(mp
, tp
, ip
, flags
);
332 trace_xfs_iget_miss(ip
);
334 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
340 * Preload the radix tree so we can insert safely under the
341 * write spinlock. Note that we cannot sleep inside the preload
344 if (radix_tree_preload(GFP_KERNEL
)) {
350 * Because the inode hasn't been added to the radix-tree yet it can't
351 * be found by another thread, so we can do the non-sleeping lock here.
354 if (!xfs_ilock_nowait(ip
, lock_flags
))
358 spin_lock(&pag
->pag_ici_lock
);
360 /* insert the new inode */
361 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
362 if (unlikely(error
)) {
363 WARN_ON(error
!= -EEXIST
);
364 XFS_STATS_INC(xs_ig_dup
);
366 goto out_preload_end
;
369 /* These values _must_ be set before releasing the radix tree lock! */
370 ip
->i_udquot
= ip
->i_gdquot
= NULL
;
371 xfs_iflags_set(ip
, XFS_INEW
);
373 spin_unlock(&pag
->pag_ici_lock
);
374 radix_tree_preload_end();
380 spin_unlock(&pag
->pag_ici_lock
);
381 radix_tree_preload_end();
383 xfs_iunlock(ip
, lock_flags
);
385 __destroy_inode(VFS_I(ip
));
391 * Look up an inode by number in the given file system.
392 * The inode is looked up in the cache held in each AG.
393 * If the inode is found in the cache, initialise the vfs inode
396 * If it is not in core, read it in from the file system's device,
397 * add it to the cache and initialise the vfs inode.
399 * The inode is locked according to the value of the lock_flags parameter.
400 * This flag parameter indicates how and if the inode's IO lock and inode lock
403 * mp -- the mount point structure for the current file system. It points
404 * to the inode hash table.
405 * tp -- a pointer to the current transaction if there is one. This is
406 * simply passed through to the xfs_iread() call.
407 * ino -- the number of the inode desired. This is the unique identifier
408 * within the file system for the inode being requested.
409 * lock_flags -- flags indicating how to lock the inode. See the comment
410 * for xfs_ilock() for a list of valid values.
426 /* reject inode numbers outside existing AGs */
427 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
430 /* get the perag structure and ensure that it's inode capable */
431 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
432 agino
= XFS_INO_TO_AGINO(mp
, ino
);
437 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
440 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
442 goto out_error_or_again
;
445 XFS_STATS_INC(xs_ig_missed
);
447 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
450 goto out_error_or_again
;
456 ASSERT(ip
->i_df
.if_ext_max
==
457 XFS_IFORK_DSIZE(ip
) / sizeof(xfs_bmbt_rec_t
));
459 * If we have a real type for an on-disk inode, we can set ops(&unlock)
460 * now. If it's a new inode being created, xfs_ialloc will handle it.
462 if (xfs_iflags_test(ip
, XFS_INEW
) && ip
->i_d
.di_mode
!= 0)
467 if (error
== EAGAIN
) {
476 * This is a wrapper routine around the xfs_ilock() routine
477 * used to centralize some grungy code. It is used in places
478 * that wish to lock the inode solely for reading the extents.
479 * The reason these places can't just call xfs_ilock(SHARED)
480 * is that the inode lock also guards to bringing in of the
481 * extents from disk for a file in b-tree format. If the inode
482 * is in b-tree format, then we need to lock the inode exclusively
483 * until the extents are read in. Locking it exclusively all
484 * the time would limit our parallelism unnecessarily, though.
485 * What we do instead is check to see if the extents have been
486 * read in yet, and only lock the inode exclusively if they
489 * The function returns a value which should be given to the
490 * corresponding xfs_iunlock_map_shared(). This value is
491 * the mode in which the lock was actually taken.
494 xfs_ilock_map_shared(
499 if ((ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) &&
500 ((ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)) {
501 lock_mode
= XFS_ILOCK_EXCL
;
503 lock_mode
= XFS_ILOCK_SHARED
;
506 xfs_ilock(ip
, lock_mode
);
512 * This is simply the unlock routine to go with xfs_ilock_map_shared().
513 * All it does is call xfs_iunlock() with the given lock_mode.
516 xfs_iunlock_map_shared(
518 unsigned int lock_mode
)
520 xfs_iunlock(ip
, lock_mode
);
524 * The xfs inode contains 2 locks: a multi-reader lock called the
525 * i_iolock and a multi-reader lock called the i_lock. This routine
526 * allows either or both of the locks to be obtained.
528 * The 2 locks should always be ordered so that the IO lock is
529 * obtained first in order to prevent deadlock.
531 * ip -- the inode being locked
532 * lock_flags -- this parameter indicates the inode's locks
533 * to be locked. It can be:
538 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
539 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
540 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
541 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
549 * You can't set both SHARED and EXCL for the same lock,
550 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
551 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
553 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
554 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
555 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
556 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
557 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
559 if (lock_flags
& XFS_IOLOCK_EXCL
)
560 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
561 else if (lock_flags
& XFS_IOLOCK_SHARED
)
562 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
564 if (lock_flags
& XFS_ILOCK_EXCL
)
565 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
566 else if (lock_flags
& XFS_ILOCK_SHARED
)
567 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
569 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
573 * This is just like xfs_ilock(), except that the caller
574 * is guaranteed not to sleep. It returns 1 if it gets
575 * the requested locks and 0 otherwise. If the IO lock is
576 * obtained but the inode lock cannot be, then the IO lock
577 * is dropped before returning.
579 * ip -- the inode being locked
580 * lock_flags -- this parameter indicates the inode's locks to be
581 * to be locked. See the comment for xfs_ilock() for a list
590 * You can't set both SHARED and EXCL for the same lock,
591 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
592 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
594 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
595 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
596 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
597 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
598 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
600 if (lock_flags
& XFS_IOLOCK_EXCL
) {
601 if (!mrtryupdate(&ip
->i_iolock
))
603 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
604 if (!mrtryaccess(&ip
->i_iolock
))
607 if (lock_flags
& XFS_ILOCK_EXCL
) {
608 if (!mrtryupdate(&ip
->i_lock
))
609 goto out_undo_iolock
;
610 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
611 if (!mrtryaccess(&ip
->i_lock
))
612 goto out_undo_iolock
;
614 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
618 if (lock_flags
& XFS_IOLOCK_EXCL
)
619 mrunlock_excl(&ip
->i_iolock
);
620 else if (lock_flags
& XFS_IOLOCK_SHARED
)
621 mrunlock_shared(&ip
->i_iolock
);
627 * xfs_iunlock() is used to drop the inode locks acquired with
628 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
629 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
630 * that we know which locks to drop.
632 * ip -- the inode being unlocked
633 * lock_flags -- this parameter indicates the inode's locks to be
634 * to be unlocked. See the comment for xfs_ilock() for a list
635 * of valid values for this parameter.
644 * You can't set both SHARED and EXCL for the same lock,
645 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
646 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
648 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
649 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
650 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
651 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
652 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_IUNLOCK_NONOTIFY
|
653 XFS_LOCK_DEP_MASK
)) == 0);
654 ASSERT(lock_flags
!= 0);
656 if (lock_flags
& XFS_IOLOCK_EXCL
)
657 mrunlock_excl(&ip
->i_iolock
);
658 else if (lock_flags
& XFS_IOLOCK_SHARED
)
659 mrunlock_shared(&ip
->i_iolock
);
661 if (lock_flags
& XFS_ILOCK_EXCL
)
662 mrunlock_excl(&ip
->i_lock
);
663 else if (lock_flags
& XFS_ILOCK_SHARED
)
664 mrunlock_shared(&ip
->i_lock
);
666 if ((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) &&
667 !(lock_flags
& XFS_IUNLOCK_NONOTIFY
) && ip
->i_itemp
) {
669 * Let the AIL know that this item has been unlocked in case
670 * it is in the AIL and anyone is waiting on it. Don't do
671 * this if the caller has asked us not to.
673 xfs_trans_unlocked_item(ip
->i_itemp
->ili_item
.li_ailp
,
674 (xfs_log_item_t
*)(ip
->i_itemp
));
676 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
680 * give up write locks. the i/o lock cannot be held nested
681 * if it is being demoted.
688 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
689 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
691 if (lock_flags
& XFS_ILOCK_EXCL
)
692 mrdemote(&ip
->i_lock
);
693 if (lock_flags
& XFS_IOLOCK_EXCL
)
694 mrdemote(&ip
->i_iolock
);
696 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
705 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
706 if (!(lock_flags
& XFS_ILOCK_SHARED
))
707 return !!ip
->i_lock
.mr_writer
;
708 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
711 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
712 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
713 return !!ip
->i_iolock
.mr_writer
;
714 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);