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"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_quota.h"
40 #include "xfs_utils.h"
43 * Look up an inode by number in the given file system.
44 * The inode is looked up in the cache held in each AG.
45 * If the inode is found in the cache, attach it to the provided
48 * If it is not in core, read it in from the file system's device,
49 * add it to the cache and attach the provided vnode.
51 * The inode is locked according to the value of the lock_flags parameter.
52 * This flag parameter indicates how and if the inode's IO lock and inode lock
55 * mp -- the mount point structure for the current file system. It points
56 * to the inode hash table.
57 * tp -- a pointer to the current transaction if there is one. This is
58 * simply passed through to the xfs_iread() call.
59 * ino -- the number of the inode desired. This is the unique identifier
60 * within the file system for the inode being requested.
61 * lock_flags -- flags indicating how to lock the inode. See the comment
62 * for xfs_ilock() for a list of valid values.
63 * bno -- the block number starting the buffer containing the inode,
64 * if known (as by bulkstat), else 0.
77 struct inode
*old_inode
;
81 unsigned long first_index
, mask
;
85 /* the radix tree exists only in inode capable AGs */
86 if (XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_maxagi
)
89 /* get the perag structure and ensure that it's inode capable */
90 pag
= xfs_get_perag(mp
, ino
);
91 if (!pag
->pagi_inodeok
)
93 ASSERT(pag
->pag_ici_init
);
94 agino
= XFS_INO_TO_AGINO(mp
, ino
);
97 read_lock(&pag
->pag_ici_lock
);
98 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
102 * If INEW is set this inode is being set up
103 * we need to pause and try again.
105 if (xfs_iflags_test(ip
, XFS_INEW
)) {
106 read_unlock(&pag
->pag_ici_lock
);
108 XFS_STATS_INC(xs_ig_frecycle
);
113 old_inode
= ip
->i_vnode
;
114 if (old_inode
== NULL
) {
116 * If IRECLAIM is set this inode is
117 * on its way out of the system,
118 * we need to pause and try again.
120 if (xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
121 read_unlock(&pag
->pag_ici_lock
);
123 XFS_STATS_INC(xs_ig_frecycle
);
127 ASSERT(xfs_iflags_test(ip
, XFS_IRECLAIMABLE
));
130 * If lookup is racing with unlink, then we
131 * should return an error immediately so we
132 * don't remove it from the reclaim list and
133 * potentially leak the inode.
135 if ((ip
->i_d
.di_mode
== 0) &&
136 !(flags
& XFS_IGET_CREATE
)) {
137 read_unlock(&pag
->pag_ici_lock
);
138 xfs_put_perag(mp
, pag
);
142 xfs_itrace_exit_tag(ip
, "xfs_iget.alloc");
144 XFS_STATS_INC(xs_ig_found
);
145 xfs_iflags_clear(ip
, XFS_IRECLAIMABLE
);
146 read_unlock(&pag
->pag_ici_lock
);
149 list_del_init(&ip
->i_reclaim
);
150 XFS_MOUNT_IUNLOCK(mp
);
154 } else if (inode
!= old_inode
) {
155 /* The inode is being torn down, pause and
158 if (old_inode
->i_state
& (I_FREEING
| I_CLEAR
)) {
159 read_unlock(&pag
->pag_ici_lock
);
161 XFS_STATS_INC(xs_ig_frecycle
);
165 /* Chances are the other vnode (the one in the inode) is being torn
166 * down right now, and we landed on top of it. Question is, what do
167 * we do? Unhook the old inode and hook up the new one?
170 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
177 read_unlock(&pag
->pag_ici_lock
);
178 XFS_STATS_INC(xs_ig_found
);
181 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
182 xfs_put_perag(mp
, pag
);
187 xfs_ilock(ip
, lock_flags
);
189 xfs_iflags_clear(ip
, XFS_ISTALE
);
190 xfs_itrace_exit_tag(ip
, "xfs_iget.found");
197 read_unlock(&pag
->pag_ici_lock
);
198 XFS_STATS_INC(xs_ig_missed
);
201 * Read the disk inode attributes into a new inode structure and get
202 * a new vnode for it. This should also initialize i_ino and i_mount.
204 error
= xfs_iread(mp
, tp
, ino
, &ip
, bno
,
205 (flags
& XFS_IGET_BULKSTAT
) ? XFS_IMAP_BULKSTAT
: 0);
207 xfs_put_perag(mp
, pag
);
211 xfs_itrace_exit_tag(ip
, "xfs_iget.alloc");
214 mrlock_init(&ip
->i_lock
, MRLOCK_ALLOW_EQUAL_PRI
|MRLOCK_BARRIER
,
215 "xfsino", ip
->i_ino
);
216 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
217 init_waitqueue_head(&ip
->i_ipin_wait
);
218 atomic_set(&ip
->i_pincount
, 0);
219 initnsema(&ip
->i_flock
, 1, "xfsfino");
222 xfs_ilock(ip
, lock_flags
);
224 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
226 xfs_put_perag(mp
, pag
);
231 * Preload the radix tree so we can insert safely under the
234 if (radix_tree_preload(GFP_KERNEL
)) {
239 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
240 first_index
= agino
& mask
;
241 write_lock(&pag
->pag_ici_lock
);
243 * insert the new inode
245 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
246 if (unlikely(error
)) {
247 BUG_ON(error
!= -EEXIST
);
248 write_unlock(&pag
->pag_ici_lock
);
249 radix_tree_preload_end();
251 XFS_STATS_INC(xs_ig_dup
);
256 * These values _must_ be set before releasing the radix tree lock!
258 ip
->i_udquot
= ip
->i_gdquot
= NULL
;
259 xfs_iflags_set(ip
, XFS_INEW
);
261 write_unlock(&pag
->pag_ici_lock
);
262 radix_tree_preload_end();
265 * Link ip to its mount and thread it on the mount's inode list.
268 if ((iq
= mp
->m_inodes
)) {
269 ASSERT(iq
->i_mprev
->i_mnext
== iq
);
270 ip
->i_mprev
= iq
->i_mprev
;
271 iq
->i_mprev
->i_mnext
= ip
;
280 XFS_MOUNT_IUNLOCK(mp
);
281 xfs_put_perag(mp
, pag
);
284 ASSERT(ip
->i_df
.if_ext_max
==
285 XFS_IFORK_DSIZE(ip
) / sizeof(xfs_bmbt_rec_t
));
287 xfs_iflags_set(ip
, XFS_IMODIFIED
);
291 * Set up the Linux with the Linux inode.
294 inode
->i_private
= ip
;
297 * If we have a real type for an on-disk inode, we can set ops(&unlock)
298 * now. If it's a new inode being created, xfs_ialloc will handle it.
300 if (ip
->i_d
.di_mode
!= 0)
307 * The 'normal' internal xfs_iget, if needed it will
308 * 'allocate', or 'get', the vnode.
324 XFS_STATS_INC(xs_ig_attempts
);
327 inode
= iget_locked(mp
->m_super
, ino
);
329 /* If we got no inode we are out of memory */
332 if (inode
->i_state
& I_NEW
) {
333 XFS_STATS_INC(vn_active
);
334 XFS_STATS_INC(vn_alloc
);
336 error
= xfs_iget_core(inode
, mp
, tp
, ino
, flags
,
337 lock_flags
, ipp
, bno
);
339 make_bad_inode(inode
);
340 if (inode
->i_state
& I_NEW
)
341 unlock_new_inode(inode
);
348 * If the inode is not fully constructed due to
349 * filehandle mismatches wait for the inode to go
350 * away and try again.
352 * iget_locked will call __wait_on_freeing_inode
353 * to wait for the inode to go away.
355 if (is_bad_inode(inode
)) {
369 xfs_ilock(ip
, lock_flags
);
370 XFS_STATS_INC(xs_ig_found
);
376 * Look for the inode corresponding to the given ino in the hash table.
377 * If it is there and its i_transp pointer matches tp, return it.
378 * Otherwise, return NULL.
381 xfs_inode_incore(xfs_mount_t
*mp
,
388 pag
= xfs_get_perag(mp
, ino
);
389 read_lock(&pag
->pag_ici_lock
);
390 ip
= radix_tree_lookup(&pag
->pag_ici_root
, XFS_INO_TO_AGINO(mp
, ino
));
391 read_unlock(&pag
->pag_ici_lock
);
392 xfs_put_perag(mp
, pag
);
394 /* the returned inode must match the transaction */
395 if (ip
&& (ip
->i_transp
!= tp
))
401 * Decrement reference count of an inode structure and unlock it.
403 * ip -- the inode being released
404 * lock_flags -- this parameter indicates the inode's locks to be
405 * to be released. See the comment on xfs_iunlock() for a list
409 xfs_iput(xfs_inode_t
*ip
,
412 xfs_itrace_entry(ip
);
413 xfs_iunlock(ip
, lock_flags
);
418 * Special iput for brand-new inodes that are still locked
425 struct inode
*inode
= VFS_I(ip
);
427 xfs_itrace_entry(ip
);
429 if ((ip
->i_d
.di_mode
== 0)) {
430 ASSERT(!xfs_iflags_test(ip
, XFS_IRECLAIMABLE
));
431 make_bad_inode(inode
);
433 if (inode
->i_state
& I_NEW
)
434 unlock_new_inode(inode
);
436 xfs_iunlock(ip
, lock_flags
);
442 * This routine embodies the part of the reclaim code that pulls
443 * the inode from the inode hash table and the mount structure's
445 * This should only be called from xfs_reclaim().
448 xfs_ireclaim(xfs_inode_t
*ip
)
451 * Remove from old hash list and mount list.
453 XFS_STATS_INC(xs_ig_reclaims
);
458 * Here we do a spurious inode lock in order to coordinate with
459 * xfs_sync(). This is because xfs_sync() references the inodes
460 * in the mount list without taking references on the corresponding
461 * vnodes. We make that OK here by ensuring that we wait until
462 * the inode is unlocked in xfs_sync() before we go ahead and
463 * free it. We get both the regular lock and the io lock because
464 * the xfs_sync() code may need to drop the regular one but will
465 * still hold the io lock.
467 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
470 * Release dquots (and their references) if any. An inode may escape
471 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
473 XFS_QM_DQDETACH(ip
->i_mount
, ip
);
476 * Pull our behavior descriptor from the vnode chain.
479 ip
->i_vnode
->i_private
= NULL
;
484 * Free all memory associated with the inode.
486 xfs_iunlock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
491 * This routine removes an about-to-be-destroyed inode from
492 * all of the lists in which it is located with the exception
493 * of the behavior chain.
499 xfs_mount_t
*mp
= ip
->i_mount
;
500 xfs_perag_t
*pag
= xfs_get_perag(mp
, ip
->i_ino
);
503 write_lock(&pag
->pag_ici_lock
);
504 radix_tree_delete(&pag
->pag_ici_root
, XFS_INO_TO_AGINO(mp
, ip
->i_ino
));
505 write_unlock(&pag
->pag_ici_lock
);
506 xfs_put_perag(mp
, pag
);
509 * Remove from mount's inode list.
512 ASSERT((ip
->i_mnext
!= NULL
) && (ip
->i_mprev
!= NULL
));
514 iq
->i_mprev
= ip
->i_mprev
;
515 ip
->i_mprev
->i_mnext
= iq
;
518 * Fix up the head pointer if it points to the inode being deleted.
520 if (mp
->m_inodes
== ip
) {
528 /* Deal with the deleted inodes list */
529 list_del_init(&ip
->i_reclaim
);
532 XFS_MOUNT_IUNLOCK(mp
);
536 * This is a wrapper routine around the xfs_ilock() routine
537 * used to centralize some grungy code. It is used in places
538 * that wish to lock the inode solely for reading the extents.
539 * The reason these places can't just call xfs_ilock(SHARED)
540 * is that the inode lock also guards to bringing in of the
541 * extents from disk for a file in b-tree format. If the inode
542 * is in b-tree format, then we need to lock the inode exclusively
543 * until the extents are read in. Locking it exclusively all
544 * the time would limit our parallelism unnecessarily, though.
545 * What we do instead is check to see if the extents have been
546 * read in yet, and only lock the inode exclusively if they
549 * The function returns a value which should be given to the
550 * corresponding xfs_iunlock_map_shared(). This value is
551 * the mode in which the lock was actually taken.
554 xfs_ilock_map_shared(
559 if ((ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) &&
560 ((ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)) {
561 lock_mode
= XFS_ILOCK_EXCL
;
563 lock_mode
= XFS_ILOCK_SHARED
;
566 xfs_ilock(ip
, lock_mode
);
572 * This is simply the unlock routine to go with xfs_ilock_map_shared().
573 * All it does is call xfs_iunlock() with the given lock_mode.
576 xfs_iunlock_map_shared(
578 unsigned int lock_mode
)
580 xfs_iunlock(ip
, lock_mode
);
584 * The xfs inode contains 2 locks: a multi-reader lock called the
585 * i_iolock and a multi-reader lock called the i_lock. This routine
586 * allows either or both of the locks to be obtained.
588 * The 2 locks should always be ordered so that the IO lock is
589 * obtained first in order to prevent deadlock.
591 * ip -- the inode being locked
592 * lock_flags -- this parameter indicates the inode's locks
593 * to be locked. It can be:
598 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
599 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
600 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
601 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
609 * You can't set both SHARED and EXCL for the same lock,
610 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
611 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
613 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
614 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
615 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
616 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
617 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
619 if (lock_flags
& XFS_IOLOCK_EXCL
)
620 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
621 else if (lock_flags
& XFS_IOLOCK_SHARED
)
622 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
624 if (lock_flags
& XFS_ILOCK_EXCL
)
625 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
626 else if (lock_flags
& XFS_ILOCK_SHARED
)
627 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
629 xfs_ilock_trace(ip
, 1, lock_flags
, (inst_t
*)__return_address
);
633 * This is just like xfs_ilock(), except that the caller
634 * is guaranteed not to sleep. It returns 1 if it gets
635 * the requested locks and 0 otherwise. If the IO lock is
636 * obtained but the inode lock cannot be, then the IO lock
637 * is dropped before returning.
639 * ip -- the inode being locked
640 * lock_flags -- this parameter indicates the inode's locks to be
641 * to be locked. See the comment for xfs_ilock() for a list
650 * You can't set both SHARED and EXCL for the same lock,
651 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
652 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
654 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
655 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
656 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
657 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
658 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
660 if (lock_flags
& XFS_IOLOCK_EXCL
) {
661 if (!mrtryupdate(&ip
->i_iolock
))
663 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
664 if (!mrtryaccess(&ip
->i_iolock
))
667 if (lock_flags
& XFS_ILOCK_EXCL
) {
668 if (!mrtryupdate(&ip
->i_lock
))
669 goto out_undo_iolock
;
670 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
671 if (!mrtryaccess(&ip
->i_lock
))
672 goto out_undo_iolock
;
674 xfs_ilock_trace(ip
, 2, lock_flags
, (inst_t
*)__return_address
);
678 if (lock_flags
& XFS_IOLOCK_EXCL
)
679 mrunlock_excl(&ip
->i_iolock
);
680 else if (lock_flags
& XFS_IOLOCK_SHARED
)
681 mrunlock_shared(&ip
->i_iolock
);
687 * xfs_iunlock() is used to drop the inode locks acquired with
688 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
689 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
690 * that we know which locks to drop.
692 * ip -- the inode being unlocked
693 * lock_flags -- this parameter indicates the inode's locks to be
694 * to be unlocked. See the comment for xfs_ilock() for a list
695 * of valid values for this parameter.
704 * You can't set both SHARED and EXCL for the same lock,
705 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
706 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
708 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
709 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
710 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
711 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
712 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_IUNLOCK_NONOTIFY
|
713 XFS_LOCK_DEP_MASK
)) == 0);
714 ASSERT(lock_flags
!= 0);
716 if (lock_flags
& XFS_IOLOCK_EXCL
)
717 mrunlock_excl(&ip
->i_iolock
);
718 else if (lock_flags
& XFS_IOLOCK_SHARED
)
719 mrunlock_shared(&ip
->i_iolock
);
721 if (lock_flags
& XFS_ILOCK_EXCL
)
722 mrunlock_excl(&ip
->i_lock
);
723 else if (lock_flags
& XFS_ILOCK_SHARED
)
724 mrunlock_shared(&ip
->i_lock
);
726 if ((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) &&
727 !(lock_flags
& XFS_IUNLOCK_NONOTIFY
) && ip
->i_itemp
) {
729 * Let the AIL know that this item has been unlocked in case
730 * it is in the AIL and anyone is waiting on it. Don't do
731 * this if the caller has asked us not to.
733 xfs_trans_unlocked_item(ip
->i_mount
,
734 (xfs_log_item_t
*)(ip
->i_itemp
));
736 xfs_ilock_trace(ip
, 3, lock_flags
, (inst_t
*)__return_address
);
740 * give up write locks. the i/o lock cannot be held nested
741 * if it is being demoted.
748 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
749 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
751 if (lock_flags
& XFS_ILOCK_EXCL
)
752 mrdemote(&ip
->i_lock
);
753 if (lock_flags
& XFS_IOLOCK_EXCL
)
754 mrdemote(&ip
->i_iolock
);
759 * Debug-only routine, without additional rw_semaphore APIs, we can
760 * now only answer requests regarding whether we hold the lock for write
761 * (reader state is outside our visibility, we only track writer state).
763 * Note: this means !xfs_isilocked would give false positives, so don't do that.
770 if ((lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) ==
772 if (!ip
->i_lock
.mr_writer
)
776 if ((lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) ==
778 if (!ip
->i_iolock
.mr_writer
)
787 * The following three routines simply manage the i_flock
788 * semaphore embedded in the inode. This semaphore synchronizes
789 * processes attempting to flush the in-core inode back to disk.
792 xfs_iflock(xfs_inode_t
*ip
)
794 psema(&(ip
->i_flock
), PINOD
|PLTWAIT
);
798 xfs_iflock_nowait(xfs_inode_t
*ip
)
800 return (cpsema(&(ip
->i_flock
)));
804 xfs_ifunlock(xfs_inode_t
*ip
)
806 ASSERT(issemalocked(&(ip
->i_flock
)));
807 vsema(&(ip
->i_flock
));