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"
41 #include "xfs_trans_priv.h"
42 #include "xfs_inode_item.h"
44 #include "xfs_btree_trace.h"
45 #include "xfs_dir2_trace.h"
49 * Allocate and initialise an xfs_inode.
51 STATIC struct xfs_inode *
59 * if this didn't occur in transactions, we could use
60 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
61 * code up to do this anyway.
63 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
67 ASSERT(atomic_read(&ip->i_iocount) == 0);
68 ASSERT(atomic_read(&ip->i_pincount) == 0);
69 ASSERT(!spin_is_locked(&ip->i_flags_lock));
70 ASSERT(completion_done(&ip->i_flush));
73 * initialise the VFS inode here to get failures
74 * out of the way early.
76 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
77 kmem_zone_free(xfs_inode_zone, ip);
81 /* initialise the xfs inode */
84 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
86 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
88 ip->i_update_core = 0;
89 ip->i_update_size = 0;
90 ip->i_delayed_blks = 0;
91 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
96 * Initialize inode's trace buffers.
98 #ifdef XFS_INODE_TRACE
99 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
101 #ifdef XFS_BMAP_TRACE
102 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
104 #ifdef XFS_BTREE_TRACE
105 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
108 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
110 #ifdef XFS_ILOCK_TRACE
111 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
113 #ifdef XFS_DIR2_TRACE
114 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
121 * Check the validity of the inode we just found it the cache
125 struct xfs_perag *pag,
126 struct xfs_inode *ip,
128 int lock_flags) __releases(pag->pag_ici_lock)
130 struct xfs_mount *mp = ip->i_mount;
134 * If INEW is set this inode is being set up
135 * If IRECLAIM is set this inode is being torn down
136 * Pause and try again.
138 if (xfs_iflags_test(ip, (XFS_INEW|XFS_IRECLAIM))) {
139 XFS_STATS_INC(xs_ig_frecycle);
143 /* If IRECLAIMABLE is set, we've torn down the vfs inode part */
144 if (xfs_iflags_test(ip, XFS_IRECLAIMABLE)) {
147 * If lookup is racing with unlink, then we should return an
148 * error immediately so we don't remove it from the reclaim
149 * list and potentially leak the inode.
151 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
156 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
159 * We need to re-initialise the VFS inode as it has been
160 * 'freed' by the VFS. Do this here so we can deal with
161 * errors cleanly, then tag it so it can be set up correctly
164 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
170 * We must set the XFS_INEW flag before clearing the
171 * XFS_IRECLAIMABLE flag so that if a racing lookup does
172 * not find the XFS_IRECLAIMABLE above but has the igrab()
173 * below succeed we can safely check XFS_INEW to detect
174 * that this inode is still being initialised.
176 xfs_iflags_set(ip, XFS_INEW);
177 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
179 /* clear the radix tree reclaim flag as well. */
180 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
181 } else if (!igrab(VFS_I(ip))) {
182 /* If the VFS inode is being torn down, pause and try again. */
183 XFS_STATS_INC(xs_ig_frecycle);
185 } else if (xfs_iflags_test(ip, XFS_INEW)) {
187 * We are racing with another cache hit that is
188 * currently recycling this inode out of the XFS_IRECLAIMABLE
189 * state. Wait for the initialisation to complete before
192 wait_on_inode(VFS_I(ip));
195 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
201 /* We've got a live one. */
202 read_unlock(&pag->pag_ici_lock);
205 xfs_ilock(ip, lock_flags);
207 xfs_iflags_clear(ip, XFS_ISTALE);
208 xfs_itrace_exit_tag(ip, "xfs_iget.found");
209 XFS_STATS_INC(xs_ig_found);
213 read_unlock(&pag->pag_ici_lock);
220 struct xfs_mount *mp,
221 struct xfs_perag *pag,
224 struct xfs_inode **ipp,
227 int lock_flags) __releases(pag->pag_ici_lock)
229 struct xfs_inode *ip;
231 unsigned long first_index, mask;
232 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
234 ip = xfs_inode_alloc(mp, ino);
238 error = xfs_iread(mp, tp, ip, bno, flags);
242 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
244 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
250 xfs_ilock(ip, lock_flags);
253 * Preload the radix tree so we can insert safely under the
254 * write spinlock. Note that we cannot sleep inside the preload
257 if (radix_tree_preload(GFP_KERNEL)) {
262 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
263 first_index = agino & mask;
264 write_lock(&pag->pag_ici_lock);
266 /* insert the new inode */
267 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
268 if (unlikely(error)) {
269 WARN_ON(error != -EEXIST);
270 XFS_STATS_INC(xs_ig_dup);
272 goto out_preload_end;
275 /* These values _must_ be set before releasing the radix tree lock! */
276 ip->i_udquot = ip->i_gdquot = NULL;
277 xfs_iflags_set(ip, XFS_INEW);
279 write_unlock(&pag->pag_ici_lock);
280 radix_tree_preload_end();
285 write_unlock(&pag->pag_ici_lock);
286 radix_tree_preload_end();
289 xfs_iunlock(ip, lock_flags);
291 xfs_destroy_inode(ip);
296 * Look up an inode by number in the given file system.
297 * The inode is looked up in the cache held in each AG.
298 * If the inode is found in the cache, initialise the vfs inode
301 * If it is not in core, read it in from the file system's device,
302 * add it to the cache and initialise the vfs inode.
304 * The inode is locked according to the value of the lock_flags parameter.
305 * This flag parameter indicates how and if the inode's IO lock and inode lock
308 * mp -- the mount point structure for the current file system. It points
309 * to the inode hash table.
310 * tp -- a pointer to the current transaction if there is one. This is
311 * simply passed through to the xfs_iread() call.
312 * ino -- the number of the inode desired. This is the unique identifier
313 * within the file system for the inode being requested.
314 * lock_flags -- flags indicating how to lock the inode. See the comment
315 * for xfs_ilock() for a list of valid values.
316 * bno -- the block number starting the buffer containing the inode,
317 * if known (as by bulkstat), else 0.
334 /* the radix tree exists only in inode capable AGs */
335 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
338 /* get the perag structure and ensure that it's inode capable */
339 pag = xfs_get_perag(mp, ino);
340 if (!pag->pagi_inodeok)
342 ASSERT(pag->pag_ici_init);
343 agino = XFS_INO_TO_AGINO(mp, ino);
347 read_lock(&pag->pag_ici_lock);
348 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
351 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
353 goto out_error_or_again;
355 read_unlock(&pag->pag_ici_lock);
356 XFS_STATS_INC(xs_ig_missed);
358 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno,
361 goto out_error_or_again;
363 xfs_put_perag(mp, pag);
365 xfs_iflags_set(ip, XFS_IMODIFIED);
368 ASSERT(ip->i_df.if_ext_max ==
369 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
371 * If we have a real type for an on-disk inode, we can set ops(&unlock)
372 * now. If it's a new inode being created, xfs_ialloc will handle it.
374 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
379 if (error == EAGAIN) {
383 xfs_put_perag(mp, pag);
389 * Look for the inode corresponding to the given ino in the hash table.
390 * If it is there and its i_transp pointer matches tp, return it.
391 * Otherwise, return NULL.
394 xfs_inode_incore(xfs_mount_t *mp,
401 pag = xfs_get_perag(mp, ino);
402 read_lock(&pag->pag_ici_lock);
403 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
404 read_unlock(&pag->pag_ici_lock);
405 xfs_put_perag(mp, pag);
407 /* the returned inode must match the transaction */
408 if (ip && (ip->i_transp != tp))
414 * Decrement reference count of an inode structure and unlock it.
416 * ip -- the inode being released
417 * lock_flags -- this parameter indicates the inode's locks to be
418 * to be released. See the comment on xfs_iunlock() for a list
422 xfs_iput(xfs_inode_t *ip,
425 xfs_itrace_entry(ip);
426 xfs_iunlock(ip, lock_flags);
431 * Special iput for brand-new inodes that are still locked
438 struct inode *inode = VFS_I(ip);
440 xfs_itrace_entry(ip);
442 if ((ip->i_d.di_mode == 0)) {
443 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
444 make_bad_inode(inode);
446 if (inode->i_state & I_NEW)
447 unlock_new_inode(inode);
449 xfs_iunlock(ip, lock_flags);
455 * This routine embodies the part of the reclaim code that pulls
456 * the inode from the inode hash table and the mount structure's
458 * This should only be called from xfs_reclaim().
461 xfs_ireclaim(xfs_inode_t *ip)
464 * Remove from old hash list and mount list.
466 XFS_STATS_INC(xs_ig_reclaims);
471 * Here we do a spurious inode lock in order to coordinate with inode
472 * cache radix tree lookups. This is because the lookup can reference
473 * the inodes in the cache without taking references. We make that OK
474 * here by ensuring that we wait until the inode is unlocked after the
475 * lookup before we go ahead and free it. We get both the ilock and
476 * the iolock because the code may need to drop the ilock one but will
477 * still hold the iolock.
479 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
482 * Release dquots (and their references) if any. An inode may escape
483 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
485 XFS_QM_DQDETACH(ip->i_mount, ip);
488 * Free all memory associated with the inode.
490 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
495 * This routine removes an about-to-be-destroyed inode from
496 * all of the lists in which it is located with the exception
497 * of the behavior chain.
503 xfs_mount_t *mp = ip->i_mount;
504 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
506 write_lock(&pag->pag_ici_lock);
507 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
508 write_unlock(&pag->pag_ici_lock);
509 xfs_put_perag(mp, pag);
515 * This is a wrapper routine around the xfs_ilock() routine
516 * used to centralize some grungy code. It is used in places
517 * that wish to lock the inode solely for reading the extents.
518 * The reason these places can't just call xfs_ilock(SHARED)
519 * is that the inode lock also guards to bringing in of the
520 * extents from disk for a file in b-tree format. If the inode
521 * is in b-tree format, then we need to lock the inode exclusively
522 * until the extents are read in. Locking it exclusively all
523 * the time would limit our parallelism unnecessarily, though.
524 * What we do instead is check to see if the extents have been
525 * read in yet, and only lock the inode exclusively if they
528 * The function returns a value which should be given to the
529 * corresponding xfs_iunlock_map_shared(). This value is
530 * the mode in which the lock was actually taken.
533 xfs_ilock_map_shared(
538 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
539 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
540 lock_mode = XFS_ILOCK_EXCL;
542 lock_mode = XFS_ILOCK_SHARED;
545 xfs_ilock(ip, lock_mode);
551 * This is simply the unlock routine to go with xfs_ilock_map_shared().
552 * All it does is call xfs_iunlock() with the given lock_mode.
555 xfs_iunlock_map_shared(
557 unsigned int lock_mode)
559 xfs_iunlock(ip, lock_mode);
563 * The xfs inode contains 2 locks: a multi-reader lock called the
564 * i_iolock and a multi-reader lock called the i_lock. This routine
565 * allows either or both of the locks to be obtained.
567 * The 2 locks should always be ordered so that the IO lock is
568 * obtained first in order to prevent deadlock.
570 * ip -- the inode being locked
571 * lock_flags -- this parameter indicates the inode's locks
572 * to be locked. It can be:
577 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
578 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
579 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
580 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
588 * You can't set both SHARED and EXCL for the same lock,
589 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
590 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
592 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
593 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
594 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
595 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
596 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
598 if (lock_flags & XFS_IOLOCK_EXCL)
599 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
600 else if (lock_flags & XFS_IOLOCK_SHARED)
601 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
603 if (lock_flags & XFS_ILOCK_EXCL)
604 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
605 else if (lock_flags & XFS_ILOCK_SHARED)
606 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
608 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
612 * This is just like xfs_ilock(), except that the caller
613 * is guaranteed not to sleep. It returns 1 if it gets
614 * the requested locks and 0 otherwise. If the IO lock is
615 * obtained but the inode lock cannot be, then the IO lock
616 * is dropped before returning.
618 * ip -- the inode being locked
619 * lock_flags -- this parameter indicates the inode's locks to be
620 * to be locked. See the comment for xfs_ilock() for a list
629 * You can't set both SHARED and EXCL for the same lock,
630 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
631 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
633 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
634 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
635 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
636 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
637 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
639 if (lock_flags & XFS_IOLOCK_EXCL) {
640 if (!mrtryupdate(&ip->i_iolock))
642 } else if (lock_flags & XFS_IOLOCK_SHARED) {
643 if (!mrtryaccess(&ip->i_iolock))
646 if (lock_flags & XFS_ILOCK_EXCL) {
647 if (!mrtryupdate(&ip->i_lock))
648 goto out_undo_iolock;
649 } else if (lock_flags & XFS_ILOCK_SHARED) {
650 if (!mrtryaccess(&ip->i_lock))
651 goto out_undo_iolock;
653 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
657 if (lock_flags & XFS_IOLOCK_EXCL)
658 mrunlock_excl(&ip->i_iolock);
659 else if (lock_flags & XFS_IOLOCK_SHARED)
660 mrunlock_shared(&ip->i_iolock);
666 * xfs_iunlock() is used to drop the inode locks acquired with
667 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
668 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
669 * that we know which locks to drop.
671 * ip -- the inode being unlocked
672 * lock_flags -- this parameter indicates the inode's locks to be
673 * to be unlocked. See the comment for xfs_ilock() for a list
674 * of valid values for this parameter.
683 * You can't set both SHARED and EXCL for the same lock,
684 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
685 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
687 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
688 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
689 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
690 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
691 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
692 XFS_LOCK_DEP_MASK)) == 0);
693 ASSERT(lock_flags != 0);
695 if (lock_flags & XFS_IOLOCK_EXCL)
696 mrunlock_excl(&ip->i_iolock);
697 else if (lock_flags & XFS_IOLOCK_SHARED)
698 mrunlock_shared(&ip->i_iolock);
700 if (lock_flags & XFS_ILOCK_EXCL)
701 mrunlock_excl(&ip->i_lock);
702 else if (lock_flags & XFS_ILOCK_SHARED)
703 mrunlock_shared(&ip->i_lock);
705 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
706 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
708 * Let the AIL know that this item has been unlocked in case
709 * it is in the AIL and anyone is waiting on it. Don't do
710 * this if the caller has asked us not to.
712 xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp,
713 (xfs_log_item_t*)(ip->i_itemp));
715 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
719 * give up write locks. the i/o lock cannot be held nested
720 * if it is being demoted.
727 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
728 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
730 if (lock_flags & XFS_ILOCK_EXCL)
731 mrdemote(&ip->i_lock);
732 if (lock_flags & XFS_IOLOCK_EXCL)
733 mrdemote(&ip->i_iolock);
738 * Debug-only routine, without additional rw_semaphore APIs, we can
739 * now only answer requests regarding whether we hold the lock for write
740 * (reader state is outside our visibility, we only track writer state).
742 * Note: this means !xfs_isilocked would give false positives, so don't do that.
749 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
751 if (!ip->i_lock.mr_writer)
755 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
757 if (!ip->i_iolock.mr_writer)