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
;
98 switch (ip
->i_d
.di_mode
& S_IFMT
) {
102 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
107 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
111 * Only if we are shutting down the fs will we see an
112 * inode still in the AIL. If it is there, we should remove
113 * it to prevent a use-after-free from occurring.
115 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
116 struct xfs_ail
*ailp
= lip
->li_ailp
;
118 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
119 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
120 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
121 spin_lock(&ailp
->xa_lock
);
122 if (lip
->li_flags
& XFS_LI_IN_AIL
)
123 xfs_trans_ail_delete(ailp
, lip
);
125 spin_unlock(&ailp
->xa_lock
);
127 xfs_inode_item_destroy(ip
);
131 /* asserts to verify all state is correct here */
132 ASSERT(atomic_read(&ip
->i_iocount
) == 0);
133 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
134 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
135 ASSERT(completion_done(&ip
->i_flush
));
137 kmem_zone_free(xfs_inode_zone
, ip
);
141 * Check the validity of the inode we just found it the cache
145 struct xfs_perag
*pag
,
146 struct xfs_inode
*ip
,
148 int lock_flags
) __releases(pag
->pag_ici_lock
)
150 struct inode
*inode
= VFS_I(ip
);
151 struct xfs_mount
*mp
= ip
->i_mount
;
154 spin_lock(&ip
->i_flags_lock
);
157 * If we are racing with another cache hit that is currently
158 * instantiating this inode or currently recycling it out of
159 * reclaimabe state, wait for the initialisation to complete
162 * XXX(hch): eventually we should do something equivalent to
163 * wait_on_inode to wait for these flags to be cleared
164 * instead of polling for it.
166 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
167 trace_xfs_iget_skip(ip
);
168 XFS_STATS_INC(xs_ig_frecycle
);
174 * If lookup is racing with unlink return an error immediately.
176 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
182 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
183 * Need to carefully get it back into useable state.
185 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
186 trace_xfs_iget_reclaim(ip
);
189 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
190 * from stomping over us while we recycle the inode. We can't
191 * clear the radix tree reclaimable tag yet as it requires
192 * pag_ici_lock to be held exclusive.
194 ip
->i_flags
|= XFS_IRECLAIM
;
196 spin_unlock(&ip
->i_flags_lock
);
197 read_unlock(&pag
->pag_ici_lock
);
199 error
= -inode_init_always(mp
->m_super
, inode
);
202 * Re-initializing the inode failed, and we are in deep
203 * trouble. Try to re-add it to the reclaim list.
205 read_lock(&pag
->pag_ici_lock
);
206 spin_lock(&ip
->i_flags_lock
);
208 ip
->i_flags
&= ~XFS_INEW
;
209 ip
->i_flags
|= XFS_IRECLAIMABLE
;
210 __xfs_inode_set_reclaim_tag(pag
, ip
);
211 trace_xfs_iget_reclaim_fail(ip
);
215 write_lock(&pag
->pag_ici_lock
);
216 spin_lock(&ip
->i_flags_lock
);
217 ip
->i_flags
&= ~(XFS_IRECLAIMABLE
| XFS_IRECLAIM
);
218 ip
->i_flags
|= XFS_INEW
;
219 __xfs_inode_clear_reclaim_tag(mp
, pag
, ip
);
220 inode
->i_state
= I_NEW
;
221 spin_unlock(&ip
->i_flags_lock
);
222 write_unlock(&pag
->pag_ici_lock
);
224 /* If the VFS inode is being torn down, pause and try again. */
226 trace_xfs_iget_skip(ip
);
231 /* We've got a live one. */
232 spin_unlock(&ip
->i_flags_lock
);
233 read_unlock(&pag
->pag_ici_lock
);
234 trace_xfs_iget_hit(ip
);
238 xfs_ilock(ip
, lock_flags
);
240 xfs_iflags_clear(ip
, XFS_ISTALE
);
241 XFS_STATS_INC(xs_ig_found
);
246 spin_unlock(&ip
->i_flags_lock
);
247 read_unlock(&pag
->pag_ici_lock
);
254 struct xfs_mount
*mp
,
255 struct xfs_perag
*pag
,
258 struct xfs_inode
**ipp
,
262 struct xfs_inode
*ip
;
264 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
266 ip
= xfs_inode_alloc(mp
, ino
);
270 error
= xfs_iread(mp
, tp
, ip
, flags
);
274 trace_xfs_iget_miss(ip
);
276 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
282 * Preload the radix tree so we can insert safely under the
283 * write spinlock. Note that we cannot sleep inside the preload
286 if (radix_tree_preload(GFP_KERNEL
)) {
292 * Because the inode hasn't been added to the radix-tree yet it can't
293 * be found by another thread, so we can do the non-sleeping lock here.
296 if (!xfs_ilock_nowait(ip
, lock_flags
))
300 write_lock(&pag
->pag_ici_lock
);
302 /* insert the new inode */
303 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
304 if (unlikely(error
)) {
305 WARN_ON(error
!= -EEXIST
);
306 XFS_STATS_INC(xs_ig_dup
);
308 goto out_preload_end
;
311 /* These values _must_ be set before releasing the radix tree lock! */
312 ip
->i_udquot
= ip
->i_gdquot
= NULL
;
313 xfs_iflags_set(ip
, XFS_INEW
);
315 write_unlock(&pag
->pag_ici_lock
);
316 radix_tree_preload_end();
322 write_unlock(&pag
->pag_ici_lock
);
323 radix_tree_preload_end();
325 xfs_iunlock(ip
, lock_flags
);
327 __destroy_inode(VFS_I(ip
));
333 * Look up an inode by number in the given file system.
334 * The inode is looked up in the cache held in each AG.
335 * If the inode is found in the cache, initialise the vfs inode
338 * If it is not in core, read it in from the file system's device,
339 * add it to the cache and initialise the vfs inode.
341 * The inode is locked according to the value of the lock_flags parameter.
342 * This flag parameter indicates how and if the inode's IO lock and inode lock
345 * mp -- the mount point structure for the current file system. It points
346 * to the inode hash table.
347 * tp -- a pointer to the current transaction if there is one. This is
348 * simply passed through to the xfs_iread() call.
349 * ino -- the number of the inode desired. This is the unique identifier
350 * within the file system for the inode being requested.
351 * lock_flags -- flags indicating how to lock the inode. See the comment
352 * for xfs_ilock() for a list of valid values.
368 /* reject inode numbers outside existing AGs */
369 if (XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
372 /* get the perag structure and ensure that it's inode capable */
373 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
374 agino
= XFS_INO_TO_AGINO(mp
, ino
);
378 read_lock(&pag
->pag_ici_lock
);
379 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
382 error
= xfs_iget_cache_hit(pag
, ip
, flags
, lock_flags
);
384 goto out_error_or_again
;
386 read_unlock(&pag
->pag_ici_lock
);
387 XFS_STATS_INC(xs_ig_missed
);
389 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
392 goto out_error_or_again
;
398 ASSERT(ip
->i_df
.if_ext_max
==
399 XFS_IFORK_DSIZE(ip
) / sizeof(xfs_bmbt_rec_t
));
401 * If we have a real type for an on-disk inode, we can set ops(&unlock)
402 * now. If it's a new inode being created, xfs_ialloc will handle it.
404 if (xfs_iflags_test(ip
, XFS_INEW
) && ip
->i_d
.di_mode
!= 0)
409 if (error
== EAGAIN
) {
418 * This is a wrapper routine around the xfs_ilock() routine
419 * used to centralize some grungy code. It is used in places
420 * that wish to lock the inode solely for reading the extents.
421 * The reason these places can't just call xfs_ilock(SHARED)
422 * is that the inode lock also guards to bringing in of the
423 * extents from disk for a file in b-tree format. If the inode
424 * is in b-tree format, then we need to lock the inode exclusively
425 * until the extents are read in. Locking it exclusively all
426 * the time would limit our parallelism unnecessarily, though.
427 * What we do instead is check to see if the extents have been
428 * read in yet, and only lock the inode exclusively if they
431 * The function returns a value which should be given to the
432 * corresponding xfs_iunlock_map_shared(). This value is
433 * the mode in which the lock was actually taken.
436 xfs_ilock_map_shared(
441 if ((ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) &&
442 ((ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)) {
443 lock_mode
= XFS_ILOCK_EXCL
;
445 lock_mode
= XFS_ILOCK_SHARED
;
448 xfs_ilock(ip
, lock_mode
);
454 * This is simply the unlock routine to go with xfs_ilock_map_shared().
455 * All it does is call xfs_iunlock() with the given lock_mode.
458 xfs_iunlock_map_shared(
460 unsigned int lock_mode
)
462 xfs_iunlock(ip
, lock_mode
);
466 * The xfs inode contains 2 locks: a multi-reader lock called the
467 * i_iolock and a multi-reader lock called the i_lock. This routine
468 * allows either or both of the locks to be obtained.
470 * The 2 locks should always be ordered so that the IO lock is
471 * obtained first in order to prevent deadlock.
473 * ip -- the inode being locked
474 * lock_flags -- this parameter indicates the inode's locks
475 * to be locked. It can be:
480 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
481 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
482 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
483 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
491 * You can't set both SHARED and EXCL for the same lock,
492 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
493 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
495 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
496 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
497 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
498 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
499 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
501 if (lock_flags
& XFS_IOLOCK_EXCL
)
502 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
503 else if (lock_flags
& XFS_IOLOCK_SHARED
)
504 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
506 if (lock_flags
& XFS_ILOCK_EXCL
)
507 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
508 else if (lock_flags
& XFS_ILOCK_SHARED
)
509 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
511 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
515 * This is just like xfs_ilock(), except that the caller
516 * is guaranteed not to sleep. It returns 1 if it gets
517 * the requested locks and 0 otherwise. If the IO lock is
518 * obtained but the inode lock cannot be, then the IO lock
519 * is dropped before returning.
521 * ip -- the inode being locked
522 * lock_flags -- this parameter indicates the inode's locks to be
523 * to be locked. See the comment for xfs_ilock() for a list
532 * You can't set both SHARED and EXCL for the same lock,
533 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
534 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
536 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
537 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
538 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
539 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
540 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
542 if (lock_flags
& XFS_IOLOCK_EXCL
) {
543 if (!mrtryupdate(&ip
->i_iolock
))
545 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
546 if (!mrtryaccess(&ip
->i_iolock
))
549 if (lock_flags
& XFS_ILOCK_EXCL
) {
550 if (!mrtryupdate(&ip
->i_lock
))
551 goto out_undo_iolock
;
552 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
553 if (!mrtryaccess(&ip
->i_lock
))
554 goto out_undo_iolock
;
556 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
560 if (lock_flags
& XFS_IOLOCK_EXCL
)
561 mrunlock_excl(&ip
->i_iolock
);
562 else if (lock_flags
& XFS_IOLOCK_SHARED
)
563 mrunlock_shared(&ip
->i_iolock
);
569 * xfs_iunlock() is used to drop the inode locks acquired with
570 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
571 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
572 * that we know which locks to drop.
574 * ip -- the inode being unlocked
575 * lock_flags -- this parameter indicates the inode's locks to be
576 * to be unlocked. See the comment for xfs_ilock() for a list
577 * of valid values for this parameter.
586 * You can't set both SHARED and EXCL for the same lock,
587 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
588 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
590 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
591 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
592 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
593 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
594 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_IUNLOCK_NONOTIFY
|
595 XFS_LOCK_DEP_MASK
)) == 0);
596 ASSERT(lock_flags
!= 0);
598 if (lock_flags
& XFS_IOLOCK_EXCL
)
599 mrunlock_excl(&ip
->i_iolock
);
600 else if (lock_flags
& XFS_IOLOCK_SHARED
)
601 mrunlock_shared(&ip
->i_iolock
);
603 if (lock_flags
& XFS_ILOCK_EXCL
)
604 mrunlock_excl(&ip
->i_lock
);
605 else if (lock_flags
& XFS_ILOCK_SHARED
)
606 mrunlock_shared(&ip
->i_lock
);
608 if ((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) &&
609 !(lock_flags
& XFS_IUNLOCK_NONOTIFY
) && ip
->i_itemp
) {
611 * Let the AIL know that this item has been unlocked in case
612 * it is in the AIL and anyone is waiting on it. Don't do
613 * this if the caller has asked us not to.
615 xfs_trans_unlocked_item(ip
->i_itemp
->ili_item
.li_ailp
,
616 (xfs_log_item_t
*)(ip
->i_itemp
));
618 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
622 * give up write locks. the i/o lock cannot be held nested
623 * if it is being demoted.
630 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
631 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
633 if (lock_flags
& XFS_ILOCK_EXCL
)
634 mrdemote(&ip
->i_lock
);
635 if (lock_flags
& XFS_IOLOCK_EXCL
)
636 mrdemote(&ip
->i_iolock
);
638 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
647 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
648 if (!(lock_flags
& XFS_ILOCK_SHARED
))
649 return !!ip
->i_lock
.mr_writer
;
650 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
653 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
654 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
655 return !!ip
->i_iolock
.mr_writer
;
656 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);