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"
25 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_quota.h"
37 #include "xfs_utils.h"
38 #include "xfs_trans_priv.h"
39 #include "xfs_inode_item.h"
41 #include "xfs_btree_trace.h"
42 #include "xfs_trace.h"
46 * Allocate and initialise an xfs_inode.
48 STATIC struct xfs_inode *
56 * if this didn't occur in transactions, we could use
57 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
58 * code up to do this anyway.
60 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
63 if (inode_init_always(mp->m_super, VFS_I(ip))) {
64 kmem_zone_free(xfs_inode_zone, ip);
68 ASSERT(atomic_read(&ip->i_iocount) == 0);
69 ASSERT(atomic_read(&ip->i_pincount) == 0);
70 ASSERT(!spin_is_locked(&ip->i_flags_lock));
71 ASSERT(completion_done(&ip->i_flush));
73 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
75 /* initialise the xfs inode */
78 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
80 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
82 ip->i_update_core = 0;
83 ip->i_delayed_blks = 0;
84 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
88 /* prevent anyone from using this yet */
89 VFS_I(ip)->i_state = I_NEW;
98 switch (ip->i_d.di_mode & S_IFMT) {
102 xfs_idestroy_fork(ip, XFS_DATA_FORK);
107 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
111 * Only if we are shutting down the fs will we see an
112 * inode still in the AIL. If it is there, we should remove
113 * it to prevent a use-after-free from occurring.
115 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
116 struct xfs_ail *ailp = lip->li_ailp;
118 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
119 XFS_FORCED_SHUTDOWN(ip->i_mount));
120 if (lip->li_flags & XFS_LI_IN_AIL) {
121 spin_lock(&ailp->xa_lock);
122 if (lip->li_flags & XFS_LI_IN_AIL)
123 xfs_trans_ail_delete(ailp, lip);
125 spin_unlock(&ailp->xa_lock);
127 xfs_inode_item_destroy(ip);
131 /* asserts to verify all state is correct here */
132 ASSERT(atomic_read(&ip->i_iocount) == 0);
133 ASSERT(atomic_read(&ip->i_pincount) == 0);
134 ASSERT(!spin_is_locked(&ip->i_flags_lock));
135 ASSERT(completion_done(&ip->i_flush));
137 kmem_zone_free(xfs_inode_zone, ip);
141 * Check the validity of the inode we just found it the cache
145 struct xfs_perag *pag,
146 struct xfs_inode *ip,
148 int lock_flags) __releases(pag->pag_ici_lock)
150 struct inode *inode = VFS_I(ip);
151 struct xfs_mount *mp = ip->i_mount;
154 spin_lock(&ip->i_flags_lock);
157 * If we are racing with another cache hit that is currently
158 * instantiating this inode or currently recycling it out of
159 * reclaimabe state, wait for the initialisation to complete
162 * XXX(hch): eventually we should do something equivalent to
163 * wait_on_inode to wait for these flags to be cleared
164 * instead of polling for it.
166 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
167 trace_xfs_iget_skip(ip);
168 XFS_STATS_INC(xs_ig_frecycle);
174 * If lookup is racing with unlink return an error immediately.
176 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
182 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
183 * Need to carefully get it back into useable state.
185 if (ip->i_flags & XFS_IRECLAIMABLE) {
186 trace_xfs_iget_reclaim(ip);
189 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
190 * from stomping over us while we recycle the inode. We can't
191 * clear the radix tree reclaimable tag yet as it requires
192 * pag_ici_lock to be held exclusive.
194 ip->i_flags |= XFS_IRECLAIM;
196 spin_unlock(&ip->i_flags_lock);
197 read_unlock(&pag->pag_ici_lock);
199 error = -inode_init_always(mp->m_super, inode);
202 * Re-initializing the inode failed, and we are in deep
203 * trouble. Try to re-add it to the reclaim list.
205 read_lock(&pag->pag_ici_lock);
206 spin_lock(&ip->i_flags_lock);
208 ip->i_flags &= ~XFS_INEW;
209 ip->i_flags |= XFS_IRECLAIMABLE;
210 __xfs_inode_set_reclaim_tag(pag, ip);
211 trace_xfs_iget_reclaim_fail(ip);
215 write_lock(&pag->pag_ici_lock);
216 spin_lock(&ip->i_flags_lock);
217 ip->i_flags &= ~(XFS_IRECLAIMABLE | XFS_IRECLAIM);
218 ip->i_flags |= XFS_INEW;
219 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
220 inode->i_state = I_NEW;
221 spin_unlock(&ip->i_flags_lock);
222 write_unlock(&pag->pag_ici_lock);
224 /* If the VFS inode is being torn down, pause and try again. */
226 trace_xfs_iget_skip(ip);
231 /* We've got a live one. */
232 spin_unlock(&ip->i_flags_lock);
233 read_unlock(&pag->pag_ici_lock);
234 trace_xfs_iget_hit(ip);
238 xfs_ilock(ip, lock_flags);
240 xfs_iflags_clear(ip, XFS_ISTALE);
241 XFS_STATS_INC(xs_ig_found);
246 spin_unlock(&ip->i_flags_lock);
247 read_unlock(&pag->pag_ici_lock);
254 struct xfs_mount *mp,
255 struct xfs_perag *pag,
258 struct xfs_inode **ipp,
262 struct xfs_inode *ip;
264 unsigned long first_index, mask;
265 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
267 ip = xfs_inode_alloc(mp, ino);
271 error = xfs_iread(mp, tp, ip, flags);
275 trace_xfs_iget_miss(ip);
277 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
283 * Preload the radix tree so we can insert safely under the
284 * write spinlock. Note that we cannot sleep inside the preload
287 if (radix_tree_preload(GFP_KERNEL)) {
293 * Because the inode hasn't been added to the radix-tree yet it can't
294 * be found by another thread, so we can do the non-sleeping lock here.
297 if (!xfs_ilock_nowait(ip, lock_flags))
301 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
302 first_index = agino & mask;
303 write_lock(&pag->pag_ici_lock);
305 /* insert the new inode */
306 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
307 if (unlikely(error)) {
308 WARN_ON(error != -EEXIST);
309 XFS_STATS_INC(xs_ig_dup);
311 goto out_preload_end;
314 /* These values _must_ be set before releasing the radix tree lock! */
315 ip->i_udquot = ip->i_gdquot = NULL;
316 xfs_iflags_set(ip, XFS_INEW);
318 write_unlock(&pag->pag_ici_lock);
319 radix_tree_preload_end();
325 write_unlock(&pag->pag_ici_lock);
326 radix_tree_preload_end();
328 xfs_iunlock(ip, lock_flags);
330 __destroy_inode(VFS_I(ip));
336 * Look up an inode by number in the given file system.
337 * The inode is looked up in the cache held in each AG.
338 * If the inode is found in the cache, initialise the vfs inode
341 * If it is not in core, read it in from the file system's device,
342 * add it to the cache and initialise the vfs inode.
344 * The inode is locked according to the value of the lock_flags parameter.
345 * This flag parameter indicates how and if the inode's IO lock and inode lock
348 * mp -- the mount point structure for the current file system. It points
349 * to the inode hash table.
350 * tp -- a pointer to the current transaction if there is one. This is
351 * simply passed through to the xfs_iread() call.
352 * ino -- the number of the inode desired. This is the unique identifier
353 * within the file system for the inode being requested.
354 * lock_flags -- flags indicating how to lock the inode. See the comment
355 * for xfs_ilock() for a list of valid values.
371 /* the radix tree exists only in inode capable AGs */
372 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
375 /* get the perag structure and ensure that it's inode capable */
376 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
377 agino = XFS_INO_TO_AGINO(mp, ino);
381 read_lock(&pag->pag_ici_lock);
382 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
385 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
387 goto out_error_or_again;
389 read_unlock(&pag->pag_ici_lock);
390 XFS_STATS_INC(xs_ig_missed);
392 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
395 goto out_error_or_again;
401 ASSERT(ip->i_df.if_ext_max ==
402 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
404 * If we have a real type for an on-disk inode, we can set ops(&unlock)
405 * now. If it's a new inode being created, xfs_ialloc will handle it.
407 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
412 if (error == EAGAIN) {
421 * Decrement reference count of an inode structure and unlock it.
423 * ip -- the inode being released
424 * lock_flags -- this parameter indicates the inode's locks to be
425 * to be released. See the comment on xfs_iunlock() for a list
429 xfs_iput(xfs_inode_t *ip,
432 xfs_itrace_entry(ip);
433 xfs_iunlock(ip, lock_flags);
438 * Special iput for brand-new inodes that are still locked
445 struct inode *inode = VFS_I(ip);
447 xfs_itrace_entry(ip);
449 if ((ip->i_d.di_mode == 0)) {
450 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
451 make_bad_inode(inode);
453 if (inode->i_state & I_NEW)
454 unlock_new_inode(inode);
456 xfs_iunlock(ip, lock_flags);
461 * This is called free all the memory associated with an inode.
462 * It must free the inode itself and any buffers allocated for
463 * if_extents/if_data and if_broot. It must also free the lock
464 * associated with the inode.
466 * Note: because we don't initialise everything on reallocation out
467 * of the zone, we must ensure we nullify everything correctly before
468 * freeing the structure.
472 struct xfs_inode *ip)
474 struct xfs_mount *mp = ip->i_mount;
475 struct xfs_perag *pag;
476 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
478 XFS_STATS_INC(xs_ig_reclaims);
481 * Remove the inode from the per-AG radix tree.
483 * Because radix_tree_delete won't complain even if the item was never
484 * added to the tree assert that it's been there before to catch
485 * problems with the inode life time early on.
487 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
488 write_lock(&pag->pag_ici_lock);
489 if (!radix_tree_delete(&pag->pag_ici_root, agino))
491 write_unlock(&pag->pag_ici_lock);
495 * Here we do an (almost) spurious inode lock in order to coordinate
496 * with inode cache radix tree lookups. This is because the lookup
497 * can reference the inodes in the cache without taking references.
499 * We make that OK here by ensuring that we wait until the inode is
500 * unlocked after the lookup before we go ahead and free it. We get
501 * both the ilock and the iolock because the code may need to drop the
502 * ilock one but will still hold the iolock.
504 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
506 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
512 * This is a wrapper routine around the xfs_ilock() routine
513 * used to centralize some grungy code. It is used in places
514 * that wish to lock the inode solely for reading the extents.
515 * The reason these places can't just call xfs_ilock(SHARED)
516 * is that the inode lock also guards to bringing in of the
517 * extents from disk for a file in b-tree format. If the inode
518 * is in b-tree format, then we need to lock the inode exclusively
519 * until the extents are read in. Locking it exclusively all
520 * the time would limit our parallelism unnecessarily, though.
521 * What we do instead is check to see if the extents have been
522 * read in yet, and only lock the inode exclusively if they
525 * The function returns a value which should be given to the
526 * corresponding xfs_iunlock_map_shared(). This value is
527 * the mode in which the lock was actually taken.
530 xfs_ilock_map_shared(
535 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
536 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
537 lock_mode = XFS_ILOCK_EXCL;
539 lock_mode = XFS_ILOCK_SHARED;
542 xfs_ilock(ip, lock_mode);
548 * This is simply the unlock routine to go with xfs_ilock_map_shared().
549 * All it does is call xfs_iunlock() with the given lock_mode.
552 xfs_iunlock_map_shared(
554 unsigned int lock_mode)
556 xfs_iunlock(ip, lock_mode);
560 * The xfs inode contains 2 locks: a multi-reader lock called the
561 * i_iolock and a multi-reader lock called the i_lock. This routine
562 * allows either or both of the locks to be obtained.
564 * The 2 locks should always be ordered so that the IO lock is
565 * obtained first in order to prevent deadlock.
567 * ip -- the inode being locked
568 * lock_flags -- this parameter indicates the inode's locks
569 * to be locked. It can be:
574 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
575 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
576 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
577 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
585 * You can't set both SHARED and EXCL for the same lock,
586 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
587 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
589 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
590 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
591 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
592 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
593 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
595 if (lock_flags & XFS_IOLOCK_EXCL)
596 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
597 else if (lock_flags & XFS_IOLOCK_SHARED)
598 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
600 if (lock_flags & XFS_ILOCK_EXCL)
601 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
602 else if (lock_flags & XFS_ILOCK_SHARED)
603 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
605 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
609 * This is just like xfs_ilock(), except that the caller
610 * is guaranteed not to sleep. It returns 1 if it gets
611 * the requested locks and 0 otherwise. If the IO lock is
612 * obtained but the inode lock cannot be, then the IO lock
613 * is dropped before returning.
615 * ip -- the inode being locked
616 * lock_flags -- this parameter indicates the inode's locks to be
617 * to be locked. See the comment for xfs_ilock() for a list
626 * You can't set both SHARED and EXCL for the same lock,
627 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
628 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
630 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
631 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
632 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
633 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
634 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
636 if (lock_flags & XFS_IOLOCK_EXCL) {
637 if (!mrtryupdate(&ip->i_iolock))
639 } else if (lock_flags & XFS_IOLOCK_SHARED) {
640 if (!mrtryaccess(&ip->i_iolock))
643 if (lock_flags & XFS_ILOCK_EXCL) {
644 if (!mrtryupdate(&ip->i_lock))
645 goto out_undo_iolock;
646 } else if (lock_flags & XFS_ILOCK_SHARED) {
647 if (!mrtryaccess(&ip->i_lock))
648 goto out_undo_iolock;
650 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
654 if (lock_flags & XFS_IOLOCK_EXCL)
655 mrunlock_excl(&ip->i_iolock);
656 else if (lock_flags & XFS_IOLOCK_SHARED)
657 mrunlock_shared(&ip->i_iolock);
663 * xfs_iunlock() is used to drop the inode locks acquired with
664 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
665 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
666 * that we know which locks to drop.
668 * ip -- the inode being unlocked
669 * lock_flags -- this parameter indicates the inode's locks to be
670 * to be unlocked. See the comment for xfs_ilock() for a list
671 * of valid values for this parameter.
680 * You can't set both SHARED and EXCL for the same lock,
681 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
682 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
684 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
685 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
686 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
687 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
688 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
689 XFS_LOCK_DEP_MASK)) == 0);
690 ASSERT(lock_flags != 0);
692 if (lock_flags & XFS_IOLOCK_EXCL)
693 mrunlock_excl(&ip->i_iolock);
694 else if (lock_flags & XFS_IOLOCK_SHARED)
695 mrunlock_shared(&ip->i_iolock);
697 if (lock_flags & XFS_ILOCK_EXCL)
698 mrunlock_excl(&ip->i_lock);
699 else if (lock_flags & XFS_ILOCK_SHARED)
700 mrunlock_shared(&ip->i_lock);
702 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
703 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
705 * Let the AIL know that this item has been unlocked in case
706 * it is in the AIL and anyone is waiting on it. Don't do
707 * this if the caller has asked us not to.
709 xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp,
710 (xfs_log_item_t*)(ip->i_itemp));
712 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
716 * give up write locks. the i/o lock cannot be held nested
717 * if it is being demoted.
724 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
725 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
727 if (lock_flags & XFS_ILOCK_EXCL)
728 mrdemote(&ip->i_lock);
729 if (lock_flags & XFS_IOLOCK_EXCL)
730 mrdemote(&ip->i_iolock);
732 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
741 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
742 if (!(lock_flags & XFS_ILOCK_SHARED))
743 return !!ip->i_lock.mr_writer;
744 return rwsem_is_locked(&ip->i_lock.mr_lock);
747 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
748 if (!(lock_flags & XFS_IOLOCK_SHARED))
749 return !!ip->i_iolock.mr_writer;
750 return rwsem_is_locked(&ip->i_iolock.mr_lock);