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;
80 unsigned long first_index, mask;
84 /* the radix tree exists only in inode capable AGs */
85 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
88 /* get the perag structure and ensure that it's inode capable */
89 pag = xfs_get_perag(mp, ino);
90 if (!pag->pagi_inodeok)
92 ASSERT(pag->pag_ici_init);
93 agino = XFS_INO_TO_AGINO(mp, ino);
96 read_lock(&pag->pag_ici_lock);
97 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
101 * If INEW is set this inode is being set up
102 * we need to pause and try again.
104 if (xfs_iflags_test(ip, XFS_INEW)) {
105 read_unlock(&pag->pag_ici_lock);
107 XFS_STATS_INC(xs_ig_frecycle);
112 old_inode = ip->i_vnode;
113 if (old_inode == NULL) {
115 * If IRECLAIM is set this inode is
116 * on its way out of the system,
117 * we need to pause and try again.
119 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
120 read_unlock(&pag->pag_ici_lock);
122 XFS_STATS_INC(xs_ig_frecycle);
126 ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
129 * If lookup is racing with unlink, then we
130 * should return an error immediately so we
131 * don't remove it from the reclaim list and
132 * potentially leak the inode.
134 if ((ip->i_d.di_mode == 0) &&
135 !(flags & XFS_IGET_CREATE)) {
136 read_unlock(&pag->pag_ici_lock);
137 xfs_put_perag(mp, pag);
141 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
143 XFS_STATS_INC(xs_ig_found);
144 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
145 read_unlock(&pag->pag_ici_lock);
148 list_del_init(&ip->i_reclaim);
149 XFS_MOUNT_IUNLOCK(mp);
153 } else if (inode != old_inode) {
154 /* The inode is being torn down, pause and
157 if (old_inode->i_state & (I_FREEING | I_CLEAR)) {
158 read_unlock(&pag->pag_ici_lock);
160 XFS_STATS_INC(xs_ig_frecycle);
164 /* Chances are the other vnode (the one in the inode) is being torn
165 * down right now, and we landed on top of it. Question is, what do
166 * we do? Unhook the old inode and hook up the new one?
169 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
176 read_unlock(&pag->pag_ici_lock);
177 XFS_STATS_INC(xs_ig_found);
180 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
181 xfs_put_perag(mp, pag);
186 xfs_ilock(ip, lock_flags);
188 xfs_iflags_clear(ip, XFS_ISTALE);
189 xfs_itrace_exit_tag(ip, "xfs_iget.found");
196 read_unlock(&pag->pag_ici_lock);
197 XFS_STATS_INC(xs_ig_missed);
200 * Read the disk inode attributes into a new inode structure and get
201 * a new vnode for it. This should also initialize i_ino and i_mount.
203 error = xfs_iread(mp, tp, ino, &ip, bno,
204 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
206 xfs_put_perag(mp, pag);
210 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
212 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
214 xfs_put_perag(mp, pag);
219 * Preload the radix tree so we can insert safely under the
222 if (radix_tree_preload(GFP_KERNEL)) {
229 xfs_ilock(ip, lock_flags);
231 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
232 first_index = agino & mask;
233 write_lock(&pag->pag_ici_lock);
235 * insert the new inode
237 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
238 if (unlikely(error)) {
239 BUG_ON(error != -EEXIST);
240 write_unlock(&pag->pag_ici_lock);
241 radix_tree_preload_end();
243 xfs_iunlock(ip, lock_flags);
245 XFS_STATS_INC(xs_ig_dup);
250 * These values _must_ be set before releasing the radix tree lock!
252 ip->i_udquot = ip->i_gdquot = NULL;
253 xfs_iflags_set(ip, XFS_INEW);
255 write_unlock(&pag->pag_ici_lock);
256 radix_tree_preload_end();
257 xfs_put_perag(mp, pag);
260 ASSERT(ip->i_df.if_ext_max ==
261 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
263 xfs_iflags_set(ip, XFS_IMODIFIED);
267 * Set up the Linux with the Linux inode.
270 inode->i_private = ip;
273 * If we have a real type for an on-disk inode, we can set ops(&unlock)
274 * now. If it's a new inode being created, xfs_ialloc will handle it.
276 if (ip->i_d.di_mode != 0)
283 * The 'normal' internal xfs_iget, if needed it will
284 * 'allocate', or 'get', the vnode.
300 XFS_STATS_INC(xs_ig_attempts);
303 inode = iget_locked(mp->m_super, ino);
305 /* If we got no inode we are out of memory */
308 if (inode->i_state & I_NEW) {
309 XFS_STATS_INC(vn_active);
310 XFS_STATS_INC(vn_alloc);
312 error = xfs_iget_core(inode, mp, tp, ino, flags,
313 lock_flags, ipp, bno);
315 make_bad_inode(inode);
316 if (inode->i_state & I_NEW)
317 unlock_new_inode(inode);
324 * If the inode is not fully constructed due to
325 * filehandle mismatches wait for the inode to go
326 * away and try again.
328 * iget_locked will call __wait_on_freeing_inode
329 * to wait for the inode to go away.
331 if (is_bad_inode(inode)) {
345 xfs_ilock(ip, lock_flags);
346 XFS_STATS_INC(xs_ig_found);
352 * Look for the inode corresponding to the given ino in the hash table.
353 * If it is there and its i_transp pointer matches tp, return it.
354 * Otherwise, return NULL.
357 xfs_inode_incore(xfs_mount_t *mp,
364 pag = xfs_get_perag(mp, ino);
365 read_lock(&pag->pag_ici_lock);
366 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
367 read_unlock(&pag->pag_ici_lock);
368 xfs_put_perag(mp, pag);
370 /* the returned inode must match the transaction */
371 if (ip && (ip->i_transp != tp))
377 * Decrement reference count of an inode structure and unlock it.
379 * ip -- the inode being released
380 * lock_flags -- this parameter indicates the inode's locks to be
381 * to be released. See the comment on xfs_iunlock() for a list
385 xfs_iput(xfs_inode_t *ip,
388 xfs_itrace_entry(ip);
389 xfs_iunlock(ip, lock_flags);
394 * Special iput for brand-new inodes that are still locked
401 struct inode *inode = VFS_I(ip);
403 xfs_itrace_entry(ip);
405 if ((ip->i_d.di_mode == 0)) {
406 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
407 make_bad_inode(inode);
409 if (inode->i_state & I_NEW)
410 unlock_new_inode(inode);
412 xfs_iunlock(ip, lock_flags);
418 * This routine embodies the part of the reclaim code that pulls
419 * the inode from the inode hash table and the mount structure's
421 * This should only be called from xfs_reclaim().
424 xfs_ireclaim(xfs_inode_t *ip)
427 * Remove from old hash list and mount list.
429 XFS_STATS_INC(xs_ig_reclaims);
434 * Here we do a spurious inode lock in order to coordinate with
435 * xfs_sync(). This is because xfs_sync() references the inodes
436 * in the mount list without taking references on the corresponding
437 * vnodes. We make that OK here by ensuring that we wait until
438 * the inode is unlocked in xfs_sync() before we go ahead and
439 * free it. We get both the regular lock and the io lock because
440 * the xfs_sync() code may need to drop the regular one but will
441 * still hold the io lock.
443 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
446 * Release dquots (and their references) if any. An inode may escape
447 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
449 XFS_QM_DQDETACH(ip->i_mount, ip);
452 * Pull our behavior descriptor from the vnode chain.
455 ip->i_vnode->i_private = NULL;
460 * Free all memory associated with the inode.
462 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
467 * This routine removes an about-to-be-destroyed inode from
468 * all of the lists in which it is located with the exception
469 * of the behavior chain.
475 xfs_mount_t *mp = ip->i_mount;
476 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
478 write_lock(&pag->pag_ici_lock);
479 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
480 write_unlock(&pag->pag_ici_lock);
481 xfs_put_perag(mp, pag);
483 /* Deal with the deleted inodes list */
485 list_del_init(&ip->i_reclaim);
487 XFS_MOUNT_IUNLOCK(mp);
491 * This is a wrapper routine around the xfs_ilock() routine
492 * used to centralize some grungy code. It is used in places
493 * that wish to lock the inode solely for reading the extents.
494 * The reason these places can't just call xfs_ilock(SHARED)
495 * is that the inode lock also guards to bringing in of the
496 * extents from disk for a file in b-tree format. If the inode
497 * is in b-tree format, then we need to lock the inode exclusively
498 * until the extents are read in. Locking it exclusively all
499 * the time would limit our parallelism unnecessarily, though.
500 * What we do instead is check to see if the extents have been
501 * read in yet, and only lock the inode exclusively if they
504 * The function returns a value which should be given to the
505 * corresponding xfs_iunlock_map_shared(). This value is
506 * the mode in which the lock was actually taken.
509 xfs_ilock_map_shared(
514 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
515 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
516 lock_mode = XFS_ILOCK_EXCL;
518 lock_mode = XFS_ILOCK_SHARED;
521 xfs_ilock(ip, lock_mode);
527 * This is simply the unlock routine to go with xfs_ilock_map_shared().
528 * All it does is call xfs_iunlock() with the given lock_mode.
531 xfs_iunlock_map_shared(
533 unsigned int lock_mode)
535 xfs_iunlock(ip, lock_mode);
539 * The xfs inode contains 2 locks: a multi-reader lock called the
540 * i_iolock and a multi-reader lock called the i_lock. This routine
541 * allows either or both of the locks to be obtained.
543 * The 2 locks should always be ordered so that the IO lock is
544 * obtained first in order to prevent deadlock.
546 * ip -- the inode being locked
547 * lock_flags -- this parameter indicates the inode's locks
548 * to be locked. It can be:
553 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
554 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
555 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
556 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
564 * You can't set both SHARED and EXCL for the same lock,
565 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
566 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
568 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
569 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
570 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
571 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
572 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
574 if (lock_flags & XFS_IOLOCK_EXCL)
575 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
576 else if (lock_flags & XFS_IOLOCK_SHARED)
577 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
579 if (lock_flags & XFS_ILOCK_EXCL)
580 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
581 else if (lock_flags & XFS_ILOCK_SHARED)
582 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
584 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
588 * This is just like xfs_ilock(), except that the caller
589 * is guaranteed not to sleep. It returns 1 if it gets
590 * the requested locks and 0 otherwise. If the IO lock is
591 * obtained but the inode lock cannot be, then the IO lock
592 * is dropped before returning.
594 * ip -- the inode being locked
595 * lock_flags -- this parameter indicates the inode's locks to be
596 * to be locked. See the comment for xfs_ilock() for a list
605 * You can't set both SHARED and EXCL for the same lock,
606 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
607 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
609 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
610 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
611 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
612 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
613 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
615 if (lock_flags & XFS_IOLOCK_EXCL) {
616 if (!mrtryupdate(&ip->i_iolock))
618 } else if (lock_flags & XFS_IOLOCK_SHARED) {
619 if (!mrtryaccess(&ip->i_iolock))
622 if (lock_flags & XFS_ILOCK_EXCL) {
623 if (!mrtryupdate(&ip->i_lock))
624 goto out_undo_iolock;
625 } else if (lock_flags & XFS_ILOCK_SHARED) {
626 if (!mrtryaccess(&ip->i_lock))
627 goto out_undo_iolock;
629 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
633 if (lock_flags & XFS_IOLOCK_EXCL)
634 mrunlock_excl(&ip->i_iolock);
635 else if (lock_flags & XFS_IOLOCK_SHARED)
636 mrunlock_shared(&ip->i_iolock);
642 * xfs_iunlock() is used to drop the inode locks acquired with
643 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
644 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
645 * that we know which locks to drop.
647 * ip -- the inode being unlocked
648 * lock_flags -- this parameter indicates the inode's locks to be
649 * to be unlocked. See the comment for xfs_ilock() for a list
650 * of valid values for this parameter.
659 * You can't set both SHARED and EXCL for the same lock,
660 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
661 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
663 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
664 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
665 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
666 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
667 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
668 XFS_LOCK_DEP_MASK)) == 0);
669 ASSERT(lock_flags != 0);
671 if (lock_flags & XFS_IOLOCK_EXCL)
672 mrunlock_excl(&ip->i_iolock);
673 else if (lock_flags & XFS_IOLOCK_SHARED)
674 mrunlock_shared(&ip->i_iolock);
676 if (lock_flags & XFS_ILOCK_EXCL)
677 mrunlock_excl(&ip->i_lock);
678 else if (lock_flags & XFS_ILOCK_SHARED)
679 mrunlock_shared(&ip->i_lock);
681 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
682 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
684 * Let the AIL know that this item has been unlocked in case
685 * it is in the AIL and anyone is waiting on it. Don't do
686 * this if the caller has asked us not to.
688 xfs_trans_unlocked_item(ip->i_mount,
689 (xfs_log_item_t*)(ip->i_itemp));
691 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
695 * give up write locks. the i/o lock cannot be held nested
696 * if it is being demoted.
703 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
704 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
706 if (lock_flags & XFS_ILOCK_EXCL)
707 mrdemote(&ip->i_lock);
708 if (lock_flags & XFS_IOLOCK_EXCL)
709 mrdemote(&ip->i_iolock);
714 * Debug-only routine, without additional rw_semaphore APIs, we can
715 * now only answer requests regarding whether we hold the lock for write
716 * (reader state is outside our visibility, we only track writer state).
718 * Note: this means !xfs_isilocked would give false positives, so don't do that.
725 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
727 if (!ip->i_lock.mr_writer)
731 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
733 if (!ip->i_iolock.mr_writer)