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 xfs_icluster_t *icl, *new_icl = NULL;
82 unsigned long first_index, mask;
86 /* the radix tree exists only in inode capable AGs */
87 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
90 /* get the perag structure and ensure that it's inode capable */
91 pag = xfs_get_perag(mp, ino);
92 if (!pag->pagi_inodeok)
94 ASSERT(pag->pag_ici_init);
95 agino = XFS_INO_TO_AGINO(mp, ino);
98 read_lock(&pag->pag_ici_lock);
99 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
103 * If INEW is set this inode is being set up
104 * we need to pause and try again.
106 if (xfs_iflags_test(ip, XFS_INEW)) {
107 read_unlock(&pag->pag_ici_lock);
109 XFS_STATS_INC(xs_ig_frecycle);
114 old_inode = ip->i_vnode;
115 if (old_inode == NULL) {
117 * If IRECLAIM is set this inode is
118 * on its way out of the system,
119 * we need to pause and try again.
121 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
122 read_unlock(&pag->pag_ici_lock);
124 XFS_STATS_INC(xs_ig_frecycle);
128 ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
131 * If lookup is racing with unlink, then we
132 * should return an error immediately so we
133 * don't remove it from the reclaim list and
134 * potentially leak the inode.
136 if ((ip->i_d.di_mode == 0) &&
137 !(flags & XFS_IGET_CREATE)) {
138 read_unlock(&pag->pag_ici_lock);
139 xfs_put_perag(mp, pag);
144 * There may be transactions sitting in the
145 * incore log buffers or being flushed to disk
146 * at this time. We can't clear the
147 * XFS_IRECLAIMABLE flag until these
148 * transactions have hit the disk, otherwise we
149 * will void the guarantee the flag provides
152 if (xfs_ipincount(ip)) {
153 read_unlock(&pag->pag_ici_lock);
155 XFS_LOG_FORCE|XFS_LOG_SYNC);
156 XFS_STATS_INC(xs_ig_frecycle);
160 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
162 XFS_STATS_INC(xs_ig_found);
164 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
165 read_unlock(&pag->pag_ici_lock);
168 list_del_init(&ip->i_reclaim);
169 XFS_MOUNT_IUNLOCK(mp);
173 } else if (inode != old_inode) {
174 /* The inode is being torn down, pause and
177 if (old_inode->i_state & (I_FREEING | I_CLEAR)) {
178 read_unlock(&pag->pag_ici_lock);
180 XFS_STATS_INC(xs_ig_frecycle);
184 /* Chances are the other vnode (the one in the inode) is being torn
185 * down right now, and we landed on top of it. Question is, what do
186 * we do? Unhook the old inode and hook up the new one?
189 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
196 read_unlock(&pag->pag_ici_lock);
197 XFS_STATS_INC(xs_ig_found);
200 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
201 xfs_put_perag(mp, pag);
206 xfs_ilock(ip, lock_flags);
208 xfs_iflags_clear(ip, XFS_ISTALE);
209 xfs_itrace_exit_tag(ip, "xfs_iget.found");
216 read_unlock(&pag->pag_ici_lock);
217 XFS_STATS_INC(xs_ig_missed);
220 * Read the disk inode attributes into a new inode structure and get
221 * a new vnode for it. This should also initialize i_ino and i_mount.
223 error = xfs_iread(mp, tp, ino, &ip, bno,
224 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
226 xfs_put_perag(mp, pag);
230 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
233 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
234 "xfsino", ip->i_ino);
235 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
236 init_waitqueue_head(&ip->i_ipin_wait);
237 atomic_set(&ip->i_pincount, 0);
238 initnsema(&ip->i_flock, 1, "xfsfino");
241 xfs_ilock(ip, lock_flags);
243 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
245 xfs_put_perag(mp, pag);
250 * This is a bit messy - we preallocate everything we _might_
251 * need before we pick up the ici lock. That way we don't have to
252 * juggle locks and go all the way back to the start.
254 new_icl = kmem_zone_alloc(xfs_icluster_zone, KM_SLEEP);
255 if (radix_tree_preload(GFP_KERNEL)) {
259 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
260 first_index = agino & mask;
261 write_lock(&pag->pag_ici_lock);
264 * Find the cluster if it exists
267 if (radix_tree_gang_lookup(&pag->pag_ici_root, (void**)&iq,
269 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) == first_index)
274 * insert the new inode
276 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
277 if (unlikely(error)) {
278 BUG_ON(error != -EEXIST);
279 write_unlock(&pag->pag_ici_lock);
280 radix_tree_preload_end();
282 XFS_STATS_INC(xs_ig_dup);
287 * These values _must_ be set before releasing ihlock!
289 ip->i_udquot = ip->i_gdquot = NULL;
290 xfs_iflags_set(ip, XFS_INEW);
292 ASSERT(ip->i_cluster == NULL);
295 spin_lock_init(&new_icl->icl_lock);
296 INIT_HLIST_HEAD(&new_icl->icl_inodes);
300 ASSERT(!hlist_empty(&icl->icl_inodes));
302 spin_lock(&icl->icl_lock);
303 hlist_add_head(&ip->i_cnode, &icl->icl_inodes);
305 spin_unlock(&icl->icl_lock);
307 write_unlock(&pag->pag_ici_lock);
308 radix_tree_preload_end();
310 kmem_zone_free(xfs_icluster_zone, new_icl);
313 * Link ip to its mount and thread it on the mount's inode list.
316 if ((iq = mp->m_inodes)) {
317 ASSERT(iq->i_mprev->i_mnext == iq);
318 ip->i_mprev = iq->i_mprev;
319 iq->i_mprev->i_mnext = ip;
328 XFS_MOUNT_IUNLOCK(mp);
329 xfs_put_perag(mp, pag);
332 ASSERT(ip->i_df.if_ext_max ==
333 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
335 xfs_iflags_set(ip, XFS_IMODIFIED);
339 * If we have a real type for an on-disk inode, we can set ops(&unlock)
340 * now. If it's a new inode being created, xfs_ialloc will handle it.
342 xfs_initialize_vnode(mp, inode, ip);
348 * The 'normal' internal xfs_iget, if needed it will
349 * 'allocate', or 'get', the vnode.
365 XFS_STATS_INC(xs_ig_attempts);
368 inode = iget_locked(mp->m_super, ino);
370 /* If we got no inode we are out of memory */
373 if (inode->i_state & I_NEW) {
374 XFS_STATS_INC(vn_active);
375 XFS_STATS_INC(vn_alloc);
377 error = xfs_iget_core(inode, mp, tp, ino, flags,
378 lock_flags, ipp, bno);
380 make_bad_inode(inode);
381 if (inode->i_state & I_NEW)
382 unlock_new_inode(inode);
389 * If the inode is not fully constructed due to
390 * filehandle mismatches wait for the inode to go
391 * away and try again.
393 * iget_locked will call __wait_on_freeing_inode
394 * to wait for the inode to go away.
396 if (is_bad_inode(inode)) {
410 xfs_ilock(ip, lock_flags);
411 XFS_STATS_INC(xs_ig_found);
417 * Look for the inode corresponding to the given ino in the hash table.
418 * If it is there and its i_transp pointer matches tp, return it.
419 * Otherwise, return NULL.
422 xfs_inode_incore(xfs_mount_t *mp,
429 pag = xfs_get_perag(mp, ino);
430 read_lock(&pag->pag_ici_lock);
431 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
432 read_unlock(&pag->pag_ici_lock);
433 xfs_put_perag(mp, pag);
435 /* the returned inode must match the transaction */
436 if (ip && (ip->i_transp != tp))
442 * Decrement reference count of an inode structure and unlock it.
444 * ip -- the inode being released
445 * lock_flags -- this parameter indicates the inode's locks to be
446 * to be released. See the comment on xfs_iunlock() for a list
450 xfs_iput(xfs_inode_t *ip,
453 xfs_itrace_entry(ip);
454 xfs_iunlock(ip, lock_flags);
459 * Special iput for brand-new inodes that are still locked
462 xfs_iput_new(xfs_inode_t *ip,
465 struct inode *inode = ip->i_vnode;
467 xfs_itrace_entry(ip);
469 if ((ip->i_d.di_mode == 0)) {
470 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
471 make_bad_inode(inode);
473 if (inode->i_state & I_NEW)
474 unlock_new_inode(inode);
476 xfs_iunlock(ip, lock_flags);
482 * This routine embodies the part of the reclaim code that pulls
483 * the inode from the inode hash table and the mount structure's
485 * This should only be called from xfs_reclaim().
488 xfs_ireclaim(xfs_inode_t *ip)
491 * Remove from old hash list and mount list.
493 XFS_STATS_INC(xs_ig_reclaims);
498 * Here we do a spurious inode lock in order to coordinate with
499 * xfs_sync(). This is because xfs_sync() references the inodes
500 * in the mount list without taking references on the corresponding
501 * vnodes. We make that OK here by ensuring that we wait until
502 * the inode is unlocked in xfs_sync() before we go ahead and
503 * free it. We get both the regular lock and the io lock because
504 * the xfs_sync() code may need to drop the regular one but will
505 * still hold the io lock.
507 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
510 * Release dquots (and their references) if any. An inode may escape
511 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
513 XFS_QM_DQDETACH(ip->i_mount, ip);
516 * Pull our behavior descriptor from the vnode chain.
519 ip->i_vnode->i_private = NULL;
524 * Free all memory associated with the inode.
526 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
531 * This routine removes an about-to-be-destroyed inode from
532 * all of the lists in which it is located with the exception
533 * of the behavior chain.
539 xfs_mount_t *mp = ip->i_mount;
540 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
543 write_lock(&pag->pag_ici_lock);
544 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
545 write_unlock(&pag->pag_ici_lock);
546 xfs_put_perag(mp, pag);
549 * Remove from cluster list
552 spin_lock(&ip->i_cluster->icl_lock);
553 hlist_del(&ip->i_cnode);
554 spin_unlock(&ip->i_cluster->icl_lock);
556 /* was last inode in cluster? */
557 if (hlist_empty(&ip->i_cluster->icl_inodes))
558 kmem_zone_free(xfs_icluster_zone, ip->i_cluster);
561 * Remove from mount's inode list.
564 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
566 iq->i_mprev = ip->i_mprev;
567 ip->i_mprev->i_mnext = iq;
570 * Fix up the head pointer if it points to the inode being deleted.
572 if (mp->m_inodes == ip) {
580 /* Deal with the deleted inodes list */
581 list_del_init(&ip->i_reclaim);
584 XFS_MOUNT_IUNLOCK(mp);
588 * This is a wrapper routine around the xfs_ilock() routine
589 * used to centralize some grungy code. It is used in places
590 * that wish to lock the inode solely for reading the extents.
591 * The reason these places can't just call xfs_ilock(SHARED)
592 * is that the inode lock also guards to bringing in of the
593 * extents from disk for a file in b-tree format. If the inode
594 * is in b-tree format, then we need to lock the inode exclusively
595 * until the extents are read in. Locking it exclusively all
596 * the time would limit our parallelism unnecessarily, though.
597 * What we do instead is check to see if the extents have been
598 * read in yet, and only lock the inode exclusively if they
601 * The function returns a value which should be given to the
602 * corresponding xfs_iunlock_map_shared(). This value is
603 * the mode in which the lock was actually taken.
606 xfs_ilock_map_shared(
611 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
612 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
613 lock_mode = XFS_ILOCK_EXCL;
615 lock_mode = XFS_ILOCK_SHARED;
618 xfs_ilock(ip, lock_mode);
624 * This is simply the unlock routine to go with xfs_ilock_map_shared().
625 * All it does is call xfs_iunlock() with the given lock_mode.
628 xfs_iunlock_map_shared(
630 unsigned int lock_mode)
632 xfs_iunlock(ip, lock_mode);
636 * The xfs inode contains 2 locks: a multi-reader lock called the
637 * i_iolock and a multi-reader lock called the i_lock. This routine
638 * allows either or both of the locks to be obtained.
640 * The 2 locks should always be ordered so that the IO lock is
641 * obtained first in order to prevent deadlock.
643 * ip -- the inode being locked
644 * lock_flags -- this parameter indicates the inode's locks
645 * to be locked. It can be:
650 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
651 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
652 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
653 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
656 xfs_ilock(xfs_inode_t *ip,
660 * You can't set both SHARED and EXCL for the same lock,
661 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
662 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
664 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
665 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
666 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
667 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
668 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
670 if (lock_flags & XFS_IOLOCK_EXCL) {
671 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
672 } else if (lock_flags & XFS_IOLOCK_SHARED) {
673 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
675 if (lock_flags & XFS_ILOCK_EXCL) {
676 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
677 } else if (lock_flags & XFS_ILOCK_SHARED) {
678 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
680 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
684 * This is just like xfs_ilock(), except that the caller
685 * is guaranteed not to sleep. It returns 1 if it gets
686 * the requested locks and 0 otherwise. If the IO lock is
687 * obtained but the inode lock cannot be, then the IO lock
688 * is dropped before returning.
690 * ip -- the inode being locked
691 * lock_flags -- this parameter indicates the inode's locks to be
692 * to be locked. See the comment for xfs_ilock() for a list
697 xfs_ilock_nowait(xfs_inode_t *ip,
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_LOCK_DEP_MASK)) == 0);
715 if (lock_flags & XFS_IOLOCK_EXCL) {
716 iolocked = mrtryupdate(&ip->i_iolock);
720 } else if (lock_flags & XFS_IOLOCK_SHARED) {
721 iolocked = mrtryaccess(&ip->i_iolock);
726 if (lock_flags & XFS_ILOCK_EXCL) {
727 ilocked = mrtryupdate(&ip->i_lock);
730 mrunlock(&ip->i_iolock);
734 } else if (lock_flags & XFS_ILOCK_SHARED) {
735 ilocked = mrtryaccess(&ip->i_lock);
738 mrunlock(&ip->i_iolock);
743 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
748 * xfs_iunlock() is used to drop the inode locks acquired with
749 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
750 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
751 * that we know which locks to drop.
753 * ip -- the inode being unlocked
754 * lock_flags -- this parameter indicates the inode's locks to be
755 * to be unlocked. See the comment for xfs_ilock() for a list
756 * of valid values for this parameter.
760 xfs_iunlock(xfs_inode_t *ip,
764 * You can't set both SHARED and EXCL for the same lock,
765 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
766 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
768 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
769 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
770 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
771 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
772 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
773 XFS_LOCK_DEP_MASK)) == 0);
774 ASSERT(lock_flags != 0);
776 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
777 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
778 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
779 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
780 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
781 mrunlock(&ip->i_iolock);
784 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
785 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
786 (ismrlocked(&ip->i_lock, MR_ACCESS)));
787 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
788 (ismrlocked(&ip->i_lock, MR_UPDATE)));
789 mrunlock(&ip->i_lock);
792 * Let the AIL know that this item has been unlocked in case
793 * it is in the AIL and anyone is waiting on it. Don't do
794 * this if the caller has asked us not to.
796 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
797 ip->i_itemp != NULL) {
798 xfs_trans_unlocked_item(ip->i_mount,
799 (xfs_log_item_t*)(ip->i_itemp));
802 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
806 * give up write locks. the i/o lock cannot be held nested
807 * if it is being demoted.
810 xfs_ilock_demote(xfs_inode_t *ip,
813 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
814 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
816 if (lock_flags & XFS_ILOCK_EXCL) {
817 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
818 mrdemote(&ip->i_lock);
820 if (lock_flags & XFS_IOLOCK_EXCL) {
821 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
822 mrdemote(&ip->i_iolock);
827 * The following three routines simply manage the i_flock
828 * semaphore embedded in the inode. This semaphore synchronizes
829 * processes attempting to flush the in-core inode back to disk.
832 xfs_iflock(xfs_inode_t *ip)
834 psema(&(ip->i_flock), PINOD|PLTWAIT);
838 xfs_iflock_nowait(xfs_inode_t *ip)
840 return (cpsema(&(ip->i_flock)));
844 xfs_ifunlock(xfs_inode_t *ip)
846 ASSERT(issemalocked(&(ip->i_flock)));
847 vsema(&(ip->i_flock));