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_btree_trace.h"
42 #include "xfs_trace.h"
46 * Allocate and initialise an xfs_inode.
48 STATIC
struct xfs_inode
*
56 * if this didn't occur in transactions, we could use
57 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
58 * code up to do this anyway.
60 ip
= kmem_zone_alloc(xfs_inode_zone
, KM_SLEEP
);
63 if (inode_init_always(mp
->m_super
, VFS_I(ip
))) {
64 kmem_zone_free(xfs_inode_zone
, ip
);
68 ASSERT(atomic_read(&ip
->i_iocount
) == 0);
69 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
70 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
71 ASSERT(completion_done(&ip
->i_flush
));
73 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
75 /* initialise the xfs inode */
78 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
80 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
82 ip
->i_update_core
= 0;
83 ip
->i_delayed_blks
= 0;
84 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
88 /* prevent anyone from using this yet */
89 VFS_I(ip
)->i_state
= I_NEW
;
95 xfs_inode_free_callback(
96 struct rcu_head
*head
)
98 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
99 struct xfs_inode
*ip
= XFS_I(inode
);
101 INIT_LIST_HEAD(&inode
->i_dentry
);
102 kmem_zone_free(xfs_inode_zone
, ip
);
107 struct xfs_inode
*ip
)
109 switch (ip
->i_d
.di_mode
& S_IFMT
) {
113 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
118 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
122 * Only if we are shutting down the fs will we see an
123 * inode still in the AIL. If it is there, we should remove
124 * it to prevent a use-after-free from occurring.
126 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
127 struct xfs_ail
*ailp
= lip
->li_ailp
;
129 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
130 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
131 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
132 spin_lock(&ailp
->xa_lock
);
133 if (lip
->li_flags
& XFS_LI_IN_AIL
)
134 xfs_trans_ail_delete(ailp
, lip
);
136 spin_unlock(&ailp
->xa_lock
);
138 xfs_inode_item_destroy(ip
);
142 /* asserts to verify all state is correct here */
143 ASSERT(atomic_read(&ip
->i_iocount
) == 0);
144 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
145 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
146 ASSERT(completion_done(&ip
->i_flush
));
148 call_rcu(&ip
->i_vnode
.i_rcu
, xfs_inode_free_callback
);
152 * Check the validity of the inode we just found it the cache
156 struct xfs_perag
*pag
,
157 struct xfs_inode
*ip
,
159 int lock_flags
) __releases(pag
->pag_ici_lock
)
161 struct inode
*inode
= VFS_I(ip
);
162 struct xfs_mount
*mp
= ip
->i_mount
;
165 spin_lock(&ip
->i_flags_lock
);
168 * If we are racing with another cache hit that is currently
169 * instantiating this inode or currently recycling it out of
170 * reclaimabe state, wait for the initialisation to complete
173 * XXX(hch): eventually we should do something equivalent to
174 * wait_on_inode to wait for these flags to be cleared
175 * instead of polling for it.
177 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
178 trace_xfs_iget_skip(ip
);
179 XFS_STATS_INC(xs_ig_frecycle
);
185 * If lookup is racing with unlink return an error immediately.
187 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
193 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
194 * Need to carefully get it back into useable state.
196 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
197 trace_xfs_iget_reclaim(ip
);
200 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
201 * from stomping over us while we recycle the inode. We can't
202 * clear the radix tree reclaimable tag yet as it requires
203 * pag_ici_lock to be held exclusive.
205 ip
->i_flags
|= XFS_IRECLAIM
;
207 spin_unlock(&ip
->i_flags_lock
);
208 read_unlock(&pag
->pag_ici_lock
);
210 error
= -inode_init_always(mp
->m_super
, inode
);
213 * Re-initializing the inode failed, and we are in deep
214 * trouble. Try to re-add it to the reclaim list.
216 read_lock(&pag
->pag_ici_lock
);
217 spin_lock(&ip
->i_flags_lock
);
219 ip
->i_flags
&= ~XFS_INEW
;
220 ip
->i_flags
|= XFS_IRECLAIMABLE
;
221 __xfs_inode_set_reclaim_tag(pag
, ip
);
222 trace_xfs_iget_reclaim_fail(ip
);
226 write_lock(&pag
->pag_ici_lock
);
227 spin_lock(&ip
->i_flags_lock
);
228 ip
->i_flags
&= ~(XFS_IRECLAIMABLE
| XFS_IRECLAIM
);
229 ip
->i_flags
|= XFS_INEW
;
230 __xfs_inode_clear_reclaim_tag(mp
, pag
, ip
);
231 inode
->i_state
= I_NEW
;
232 spin_unlock(&ip
->i_flags_lock
);
233 write_unlock(&pag
->pag_ici_lock
);
235 /* If the VFS inode is being torn down, pause and try again. */
237 trace_xfs_iget_skip(ip
);
242 /* We've got a live one. */
243 spin_unlock(&ip
->i_flags_lock
);
244 read_unlock(&pag
->pag_ici_lock
);
245 trace_xfs_iget_hit(ip
);
249 xfs_ilock(ip
, lock_flags
);
251 xfs_iflags_clear(ip
, XFS_ISTALE
);
252 XFS_STATS_INC(xs_ig_found
);
257 spin_unlock(&ip
->i_flags_lock
);
258 read_unlock(&pag
->pag_ici_lock
);
265 struct xfs_mount
*mp
,
266 struct xfs_perag
*pag
,
269 struct xfs_inode
**ipp
,
273 struct xfs_inode
*ip
;
275 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
277 ip
= xfs_inode_alloc(mp
, ino
);
281 error
= xfs_iread(mp
, tp
, ip
, flags
);
285 trace_xfs_iget_miss(ip
);
287 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
293 * Preload the radix tree so we can insert safely under the
294 * write spinlock. Note that we cannot sleep inside the preload
297 if (radix_tree_preload(GFP_KERNEL
)) {
303 * Because the inode hasn't been added to the radix-tree yet it can't
304 * be found by another thread, so we can do the non-sleeping lock here.
307 if (!xfs_ilock_nowait(ip
, lock_flags
))
311 write_lock(&pag
->pag_ici_lock
);
313 /* insert the new inode */
314 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
315 if (unlikely(error
)) {
316 WARN_ON(error
!= -EEXIST
);
317 XFS_STATS_INC(xs_ig_dup
);
319 goto out_preload_end
;
322 /* These values _must_ be set before releasing the radix tree lock! */
323 ip
->i_udquot
= ip
->i_gdquot
= NULL
;
324 xfs_iflags_set(ip
, XFS_INEW
);
326 write_unlock(&pag
->pag_ici_lock
);
327 radix_tree_preload_end();
333 write_unlock(&pag
->pag_ici_lock
);
334 radix_tree_preload_end();
336 xfs_iunlock(ip
, lock_flags
);
338 __destroy_inode(VFS_I(ip
));
344 * Look up an inode by number in the given file system.
345 * The inode is looked up in the cache held in each AG.
346 * If the inode is found in the cache, initialise the vfs inode
349 * If it is not in core, read it in from the file system's device,
350 * add it to the cache and initialise the vfs inode.
352 * The inode is locked according to the value of the lock_flags parameter.
353 * This flag parameter indicates how and if the inode's IO lock and inode lock
356 * mp -- the mount point structure for the current file system. It points
357 * to the inode hash table.
358 * tp -- a pointer to the current transaction if there is one. This is
359 * simply passed through to the xfs_iread() call.
360 * ino -- the number of the inode desired. This is the unique identifier
361 * within the file system for the inode being requested.
362 * lock_flags -- flags indicating how to lock the inode. See the comment
363 * for xfs_ilock() for a list of valid values.
379 /* reject inode numbers outside existing AGs */
380 if (XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
383 /* get the perag structure and ensure that it's inode capable */
384 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
385 agino
= XFS_INO_TO_AGINO(mp
, ino
);
389 read_lock(&pag
->pag_ici_lock
);
390 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
393 error
= xfs_iget_cache_hit(pag
, ip
, flags
, lock_flags
);
395 goto out_error_or_again
;
397 read_unlock(&pag
->pag_ici_lock
);
398 XFS_STATS_INC(xs_ig_missed
);
400 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
403 goto out_error_or_again
;
409 ASSERT(ip
->i_df
.if_ext_max
==
410 XFS_IFORK_DSIZE(ip
) / sizeof(xfs_bmbt_rec_t
));
412 * If we have a real type for an on-disk inode, we can set ops(&unlock)
413 * now. If it's a new inode being created, xfs_ialloc will handle it.
415 if (xfs_iflags_test(ip
, XFS_INEW
) && ip
->i_d
.di_mode
!= 0)
420 if (error
== EAGAIN
) {
429 * This is a wrapper routine around the xfs_ilock() routine
430 * used to centralize some grungy code. It is used in places
431 * that wish to lock the inode solely for reading the extents.
432 * The reason these places can't just call xfs_ilock(SHARED)
433 * is that the inode lock also guards to bringing in of the
434 * extents from disk for a file in b-tree format. If the inode
435 * is in b-tree format, then we need to lock the inode exclusively
436 * until the extents are read in. Locking it exclusively all
437 * the time would limit our parallelism unnecessarily, though.
438 * What we do instead is check to see if the extents have been
439 * read in yet, and only lock the inode exclusively if they
442 * The function returns a value which should be given to the
443 * corresponding xfs_iunlock_map_shared(). This value is
444 * the mode in which the lock was actually taken.
447 xfs_ilock_map_shared(
452 if ((ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) &&
453 ((ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)) {
454 lock_mode
= XFS_ILOCK_EXCL
;
456 lock_mode
= XFS_ILOCK_SHARED
;
459 xfs_ilock(ip
, lock_mode
);
465 * This is simply the unlock routine to go with xfs_ilock_map_shared().
466 * All it does is call xfs_iunlock() with the given lock_mode.
469 xfs_iunlock_map_shared(
471 unsigned int lock_mode
)
473 xfs_iunlock(ip
, lock_mode
);
477 * The xfs inode contains 2 locks: a multi-reader lock called the
478 * i_iolock and a multi-reader lock called the i_lock. This routine
479 * allows either or both of the locks to be obtained.
481 * The 2 locks should always be ordered so that the IO lock is
482 * obtained first in order to prevent deadlock.
484 * ip -- the inode being locked
485 * lock_flags -- this parameter indicates the inode's locks
486 * to be locked. It can be:
491 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
492 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
493 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
494 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
502 * You can't set both SHARED and EXCL for the same lock,
503 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
504 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
506 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
507 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
508 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
509 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
510 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
512 if (lock_flags
& XFS_IOLOCK_EXCL
)
513 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
514 else if (lock_flags
& XFS_IOLOCK_SHARED
)
515 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
517 if (lock_flags
& XFS_ILOCK_EXCL
)
518 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
519 else if (lock_flags
& XFS_ILOCK_SHARED
)
520 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
522 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
526 * This is just like xfs_ilock(), except that the caller
527 * is guaranteed not to sleep. It returns 1 if it gets
528 * the requested locks and 0 otherwise. If the IO lock is
529 * obtained but the inode lock cannot be, then the IO lock
530 * is dropped before returning.
532 * ip -- the inode being locked
533 * lock_flags -- this parameter indicates the inode's locks to be
534 * to be locked. See the comment for xfs_ilock() for a list
543 * You can't set both SHARED and EXCL for the same lock,
544 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
545 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
547 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
548 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
549 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
550 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
551 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
553 if (lock_flags
& XFS_IOLOCK_EXCL
) {
554 if (!mrtryupdate(&ip
->i_iolock
))
556 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
557 if (!mrtryaccess(&ip
->i_iolock
))
560 if (lock_flags
& XFS_ILOCK_EXCL
) {
561 if (!mrtryupdate(&ip
->i_lock
))
562 goto out_undo_iolock
;
563 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
564 if (!mrtryaccess(&ip
->i_lock
))
565 goto out_undo_iolock
;
567 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
571 if (lock_flags
& XFS_IOLOCK_EXCL
)
572 mrunlock_excl(&ip
->i_iolock
);
573 else if (lock_flags
& XFS_IOLOCK_SHARED
)
574 mrunlock_shared(&ip
->i_iolock
);
580 * xfs_iunlock() is used to drop the inode locks acquired with
581 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
582 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
583 * that we know which locks to drop.
585 * ip -- the inode being unlocked
586 * lock_flags -- this parameter indicates the inode's locks to be
587 * to be unlocked. See the comment for xfs_ilock() for a list
588 * of valid values for this parameter.
597 * You can't set both SHARED and EXCL for the same lock,
598 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
599 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
601 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
602 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
603 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
604 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
605 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_IUNLOCK_NONOTIFY
|
606 XFS_LOCK_DEP_MASK
)) == 0);
607 ASSERT(lock_flags
!= 0);
609 if (lock_flags
& XFS_IOLOCK_EXCL
)
610 mrunlock_excl(&ip
->i_iolock
);
611 else if (lock_flags
& XFS_IOLOCK_SHARED
)
612 mrunlock_shared(&ip
->i_iolock
);
614 if (lock_flags
& XFS_ILOCK_EXCL
)
615 mrunlock_excl(&ip
->i_lock
);
616 else if (lock_flags
& XFS_ILOCK_SHARED
)
617 mrunlock_shared(&ip
->i_lock
);
619 if ((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) &&
620 !(lock_flags
& XFS_IUNLOCK_NONOTIFY
) && ip
->i_itemp
) {
622 * Let the AIL know that this item has been unlocked in case
623 * it is in the AIL and anyone is waiting on it. Don't do
624 * this if the caller has asked us not to.
626 xfs_trans_unlocked_item(ip
->i_itemp
->ili_item
.li_ailp
,
627 (xfs_log_item_t
*)(ip
->i_itemp
));
629 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
633 * give up write locks. the i/o lock cannot be held nested
634 * if it is being demoted.
641 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
642 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
644 if (lock_flags
& XFS_ILOCK_EXCL
)
645 mrdemote(&ip
->i_lock
);
646 if (lock_flags
& XFS_IOLOCK_EXCL
)
647 mrdemote(&ip
->i_iolock
);
649 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
658 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
659 if (!(lock_flags
& XFS_ILOCK_SHARED
))
660 return !!ip
->i_lock
.mr_writer
;
661 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
664 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
665 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
666 return !!ip
->i_iolock
.mr_writer
;
667 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);