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);
143 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
145 XFS_STATS_INC(xs_ig_found);
146 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
147 read_unlock(&pag->pag_ici_lock);
150 list_del_init(&ip->i_reclaim);
151 XFS_MOUNT_IUNLOCK(mp);
155 } else if (inode != old_inode) {
156 /* The inode is being torn down, pause and
159 if (old_inode->i_state & (I_FREEING | I_CLEAR)) {
160 read_unlock(&pag->pag_ici_lock);
162 XFS_STATS_INC(xs_ig_frecycle);
166 /* Chances are the other vnode (the one in the inode) is being torn
167 * down right now, and we landed on top of it. Question is, what do
168 * we do? Unhook the old inode and hook up the new one?
171 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
178 read_unlock(&pag->pag_ici_lock);
179 XFS_STATS_INC(xs_ig_found);
182 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
183 xfs_put_perag(mp, pag);
188 xfs_ilock(ip, lock_flags);
190 xfs_iflags_clear(ip, XFS_ISTALE);
191 xfs_itrace_exit_tag(ip, "xfs_iget.found");
198 read_unlock(&pag->pag_ici_lock);
199 XFS_STATS_INC(xs_ig_missed);
202 * Read the disk inode attributes into a new inode structure and get
203 * a new vnode for it. This should also initialize i_ino and i_mount.
205 error = xfs_iread(mp, tp, ino, &ip, bno,
206 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
208 xfs_put_perag(mp, pag);
212 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
215 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
216 "xfsino", ip->i_ino);
217 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
218 init_waitqueue_head(&ip->i_ipin_wait);
219 atomic_set(&ip->i_pincount, 0);
220 initnsema(&ip->i_flock, 1, "xfsfino");
223 xfs_ilock(ip, lock_flags);
225 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
227 xfs_put_perag(mp, pag);
232 * This is a bit messy - we preallocate everything we _might_
233 * need before we pick up the ici lock. That way we don't have to
234 * juggle locks and go all the way back to the start.
236 new_icl = kmem_zone_alloc(xfs_icluster_zone, KM_SLEEP);
237 if (radix_tree_preload(GFP_KERNEL)) {
241 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
242 first_index = agino & mask;
243 write_lock(&pag->pag_ici_lock);
246 * Find the cluster if it exists
249 if (radix_tree_gang_lookup(&pag->pag_ici_root, (void**)&iq,
251 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) == first_index)
256 * insert the new inode
258 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
259 if (unlikely(error)) {
260 BUG_ON(error != -EEXIST);
261 write_unlock(&pag->pag_ici_lock);
262 radix_tree_preload_end();
264 XFS_STATS_INC(xs_ig_dup);
269 * These values _must_ be set before releasing ihlock!
271 ip->i_udquot = ip->i_gdquot = NULL;
272 xfs_iflags_set(ip, XFS_INEW);
274 ASSERT(ip->i_cluster == NULL);
277 spin_lock_init(&new_icl->icl_lock);
278 INIT_HLIST_HEAD(&new_icl->icl_inodes);
282 ASSERT(!hlist_empty(&icl->icl_inodes));
284 spin_lock(&icl->icl_lock);
285 hlist_add_head(&ip->i_cnode, &icl->icl_inodes);
287 spin_unlock(&icl->icl_lock);
289 write_unlock(&pag->pag_ici_lock);
290 radix_tree_preload_end();
292 kmem_zone_free(xfs_icluster_zone, new_icl);
295 * Link ip to its mount and thread it on the mount's inode list.
298 if ((iq = mp->m_inodes)) {
299 ASSERT(iq->i_mprev->i_mnext == iq);
300 ip->i_mprev = iq->i_mprev;
301 iq->i_mprev->i_mnext = ip;
310 XFS_MOUNT_IUNLOCK(mp);
311 xfs_put_perag(mp, pag);
314 ASSERT(ip->i_df.if_ext_max ==
315 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
317 xfs_iflags_set(ip, XFS_IMODIFIED);
321 * If we have a real type for an on-disk inode, we can set ops(&unlock)
322 * now. If it's a new inode being created, xfs_ialloc will handle it.
324 xfs_initialize_vnode(mp, inode, ip);
330 * The 'normal' internal xfs_iget, if needed it will
331 * 'allocate', or 'get', the vnode.
347 XFS_STATS_INC(xs_ig_attempts);
350 inode = iget_locked(mp->m_super, ino);
352 /* If we got no inode we are out of memory */
355 if (inode->i_state & I_NEW) {
356 XFS_STATS_INC(vn_active);
357 XFS_STATS_INC(vn_alloc);
359 error = xfs_iget_core(inode, mp, tp, ino, flags,
360 lock_flags, ipp, bno);
362 make_bad_inode(inode);
363 if (inode->i_state & I_NEW)
364 unlock_new_inode(inode);
371 * If the inode is not fully constructed due to
372 * filehandle mismatches wait for the inode to go
373 * away and try again.
375 * iget_locked will call __wait_on_freeing_inode
376 * to wait for the inode to go away.
378 if (is_bad_inode(inode)) {
392 xfs_ilock(ip, lock_flags);
393 XFS_STATS_INC(xs_ig_found);
399 * Look for the inode corresponding to the given ino in the hash table.
400 * If it is there and its i_transp pointer matches tp, return it.
401 * Otherwise, return NULL.
404 xfs_inode_incore(xfs_mount_t *mp,
411 pag = xfs_get_perag(mp, ino);
412 read_lock(&pag->pag_ici_lock);
413 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
414 read_unlock(&pag->pag_ici_lock);
415 xfs_put_perag(mp, pag);
417 /* the returned inode must match the transaction */
418 if (ip && (ip->i_transp != tp))
424 * Decrement reference count of an inode structure and unlock it.
426 * ip -- the inode being released
427 * lock_flags -- this parameter indicates the inode's locks to be
428 * to be released. See the comment on xfs_iunlock() for a list
432 xfs_iput(xfs_inode_t *ip,
435 xfs_itrace_entry(ip);
436 xfs_iunlock(ip, lock_flags);
441 * Special iput for brand-new inodes that are still locked
444 xfs_iput_new(xfs_inode_t *ip,
447 struct inode *inode = ip->i_vnode;
449 xfs_itrace_entry(ip);
451 if ((ip->i_d.di_mode == 0)) {
452 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
453 make_bad_inode(inode);
455 if (inode->i_state & I_NEW)
456 unlock_new_inode(inode);
458 xfs_iunlock(ip, lock_flags);
464 * This routine embodies the part of the reclaim code that pulls
465 * the inode from the inode hash table and the mount structure's
467 * This should only be called from xfs_reclaim().
470 xfs_ireclaim(xfs_inode_t *ip)
473 * Remove from old hash list and mount list.
475 XFS_STATS_INC(xs_ig_reclaims);
480 * Here we do a spurious inode lock in order to coordinate with
481 * xfs_sync(). This is because xfs_sync() references the inodes
482 * in the mount list without taking references on the corresponding
483 * vnodes. We make that OK here by ensuring that we wait until
484 * the inode is unlocked in xfs_sync() before we go ahead and
485 * free it. We get both the regular lock and the io lock because
486 * the xfs_sync() code may need to drop the regular one but will
487 * still hold the io lock.
489 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
492 * Release dquots (and their references) if any. An inode may escape
493 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
495 XFS_QM_DQDETACH(ip->i_mount, ip);
498 * Pull our behavior descriptor from the vnode chain.
501 ip->i_vnode->i_private = NULL;
506 * Free all memory associated with the inode.
508 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
513 * This routine removes an about-to-be-destroyed inode from
514 * all of the lists in which it is located with the exception
515 * of the behavior chain.
521 xfs_mount_t *mp = ip->i_mount;
522 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
525 write_lock(&pag->pag_ici_lock);
526 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
527 write_unlock(&pag->pag_ici_lock);
528 xfs_put_perag(mp, pag);
531 * Remove from cluster list
534 spin_lock(&ip->i_cluster->icl_lock);
535 hlist_del(&ip->i_cnode);
536 spin_unlock(&ip->i_cluster->icl_lock);
538 /* was last inode in cluster? */
539 if (hlist_empty(&ip->i_cluster->icl_inodes))
540 kmem_zone_free(xfs_icluster_zone, ip->i_cluster);
543 * Remove from mount's inode list.
546 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
548 iq->i_mprev = ip->i_mprev;
549 ip->i_mprev->i_mnext = iq;
552 * Fix up the head pointer if it points to the inode being deleted.
554 if (mp->m_inodes == ip) {
562 /* Deal with the deleted inodes list */
563 list_del_init(&ip->i_reclaim);
566 XFS_MOUNT_IUNLOCK(mp);
570 * This is a wrapper routine around the xfs_ilock() routine
571 * used to centralize some grungy code. It is used in places
572 * that wish to lock the inode solely for reading the extents.
573 * The reason these places can't just call xfs_ilock(SHARED)
574 * is that the inode lock also guards to bringing in of the
575 * extents from disk for a file in b-tree format. If the inode
576 * is in b-tree format, then we need to lock the inode exclusively
577 * until the extents are read in. Locking it exclusively all
578 * the time would limit our parallelism unnecessarily, though.
579 * What we do instead is check to see if the extents have been
580 * read in yet, and only lock the inode exclusively if they
583 * The function returns a value which should be given to the
584 * corresponding xfs_iunlock_map_shared(). This value is
585 * the mode in which the lock was actually taken.
588 xfs_ilock_map_shared(
593 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
594 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
595 lock_mode = XFS_ILOCK_EXCL;
597 lock_mode = XFS_ILOCK_SHARED;
600 xfs_ilock(ip, lock_mode);
606 * This is simply the unlock routine to go with xfs_ilock_map_shared().
607 * All it does is call xfs_iunlock() with the given lock_mode.
610 xfs_iunlock_map_shared(
612 unsigned int lock_mode)
614 xfs_iunlock(ip, lock_mode);
618 * The xfs inode contains 2 locks: a multi-reader lock called the
619 * i_iolock and a multi-reader lock called the i_lock. This routine
620 * allows either or both of the locks to be obtained.
622 * The 2 locks should always be ordered so that the IO lock is
623 * obtained first in order to prevent deadlock.
625 * ip -- the inode being locked
626 * lock_flags -- this parameter indicates the inode's locks
627 * to be locked. It can be:
632 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
633 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
634 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
635 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
638 xfs_ilock(xfs_inode_t *ip,
642 * You can't set both SHARED and EXCL for the same lock,
643 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
644 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
646 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
647 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
648 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
649 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
650 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
652 if (lock_flags & XFS_IOLOCK_EXCL) {
653 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
654 } else if (lock_flags & XFS_IOLOCK_SHARED) {
655 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
657 if (lock_flags & XFS_ILOCK_EXCL) {
658 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
659 } else if (lock_flags & XFS_ILOCK_SHARED) {
660 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
662 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
666 * This is just like xfs_ilock(), except that the caller
667 * is guaranteed not to sleep. It returns 1 if it gets
668 * the requested locks and 0 otherwise. If the IO lock is
669 * obtained but the inode lock cannot be, then the IO lock
670 * is dropped before returning.
672 * ip -- the inode being locked
673 * lock_flags -- this parameter indicates the inode's locks to be
674 * to be locked. See the comment for xfs_ilock() for a list
679 xfs_ilock_nowait(xfs_inode_t *ip,
686 * You can't set both SHARED and EXCL for the same lock,
687 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
688 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
690 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
691 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
692 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
693 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
694 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
697 if (lock_flags & XFS_IOLOCK_EXCL) {
698 iolocked = mrtryupdate(&ip->i_iolock);
702 } else if (lock_flags & XFS_IOLOCK_SHARED) {
703 iolocked = mrtryaccess(&ip->i_iolock);
708 if (lock_flags & XFS_ILOCK_EXCL) {
709 ilocked = mrtryupdate(&ip->i_lock);
712 mrunlock(&ip->i_iolock);
716 } else if (lock_flags & XFS_ILOCK_SHARED) {
717 ilocked = mrtryaccess(&ip->i_lock);
720 mrunlock(&ip->i_iolock);
725 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
730 * xfs_iunlock() is used to drop the inode locks acquired with
731 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
732 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
733 * that we know which locks to drop.
735 * ip -- the inode being unlocked
736 * lock_flags -- this parameter indicates the inode's locks to be
737 * to be unlocked. See the comment for xfs_ilock() for a list
738 * of valid values for this parameter.
742 xfs_iunlock(xfs_inode_t *ip,
746 * You can't set both SHARED and EXCL for the same lock,
747 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
748 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
750 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
751 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
752 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
753 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
754 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
755 XFS_LOCK_DEP_MASK)) == 0);
756 ASSERT(lock_flags != 0);
758 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
759 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
760 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
761 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
762 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
763 mrunlock(&ip->i_iolock);
766 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
767 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
768 (ismrlocked(&ip->i_lock, MR_ACCESS)));
769 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
770 (ismrlocked(&ip->i_lock, MR_UPDATE)));
771 mrunlock(&ip->i_lock);
774 * Let the AIL know that this item has been unlocked in case
775 * it is in the AIL and anyone is waiting on it. Don't do
776 * this if the caller has asked us not to.
778 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
779 ip->i_itemp != NULL) {
780 xfs_trans_unlocked_item(ip->i_mount,
781 (xfs_log_item_t*)(ip->i_itemp));
784 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
788 * give up write locks. the i/o lock cannot be held nested
789 * if it is being demoted.
792 xfs_ilock_demote(xfs_inode_t *ip,
795 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
796 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
798 if (lock_flags & XFS_ILOCK_EXCL) {
799 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
800 mrdemote(&ip->i_lock);
802 if (lock_flags & XFS_IOLOCK_EXCL) {
803 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
804 mrdemote(&ip->i_iolock);
809 * The following three routines simply manage the i_flock
810 * semaphore embedded in the inode. This semaphore synchronizes
811 * processes attempting to flush the in-core inode back to disk.
814 xfs_iflock(xfs_inode_t *ip)
816 psema(&(ip->i_flock), PINOD|PLTWAIT);
820 xfs_iflock_nowait(xfs_inode_t *ip)
822 return (cpsema(&(ip->i_flock)));
826 xfs_ifunlock(xfs_inode_t *ip)
828 ASSERT(issemalocked(&(ip->i_flock)));
829 vsema(&(ip->i_flock));