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 * Check the validity of the inode we just found it the cache
47 struct xfs_perag *pag,
50 int lock_flags) __releases(pag->pag_ici_lock)
52 struct xfs_mount *mp = ip->i_mount;
56 * If INEW is set this inode is being set up
57 * If IRECLAIM is set this inode is being torn down
58 * Pause and try again.
60 if (xfs_iflags_test(ip, (XFS_INEW|XFS_IRECLAIM))) {
62 XFS_STATS_INC(xs_ig_frecycle);
66 /* If IRECLAIMABLE is set, we've torn down the vfs inode part */
67 if (xfs_iflags_test(ip, XFS_IRECLAIMABLE)) {
70 * If lookup is racing with unlink, then we should return an
71 * error immediately so we don't remove it from the reclaim
72 * list and potentially leak the inode.
75 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
80 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
83 * We need to re-initialise the VFS inode as it has been
84 * 'freed' by the VFS. Do this here so we can deal with
85 * errors cleanly, then tag it so it can be set up correctly
88 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
92 xfs_iflags_set(ip, XFS_INEW);
93 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
95 /* clear the radix tree reclaim flag as well. */
96 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
97 read_unlock(&pag->pag_ici_lock);
98 } else if (!igrab(VFS_I(ip))) {
99 /* If the VFS inode is being torn down, pause and try again. */
101 XFS_STATS_INC(xs_ig_frecycle);
104 /* we've got a live one */
105 read_unlock(&pag->pag_ici_lock);
108 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
114 xfs_ilock(ip, lock_flags);
116 xfs_iflags_clear(ip, XFS_ISTALE);
117 xfs_itrace_exit_tag(ip, "xfs_iget.found");
118 XFS_STATS_INC(xs_ig_found);
122 read_unlock(&pag->pag_ici_lock);
130 struct xfs_mount *mp,
131 struct xfs_perag *pag,
134 struct xfs_inode **ipp,
137 int lock_flags) __releases(pag->pag_ici_lock)
139 struct xfs_inode *ip;
141 unsigned long first_index, mask;
142 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
145 * Read the disk inode attributes into a new inode structure and get
146 * a new vnode for it. This should also initialize i_ino and i_mount.
148 error = xfs_iread(mp, tp, ino, &ip, bno,
149 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
153 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
155 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
161 * Preload the radix tree so we can insert safely under the
164 if (radix_tree_preload(GFP_KERNEL)) {
170 xfs_ilock(ip, lock_flags);
172 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
173 first_index = agino & mask;
174 write_lock(&pag->pag_ici_lock);
176 /* insert the new inode */
177 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
178 if (unlikely(error)) {
179 WARN_ON(error != -EEXIST);
180 XFS_STATS_INC(xs_ig_dup);
185 /* These values _must_ be set before releasing the radix tree lock! */
186 ip->i_udquot = ip->i_gdquot = NULL;
187 xfs_iflags_set(ip, XFS_INEW);
189 write_unlock(&pag->pag_ici_lock);
190 radix_tree_preload_end();
195 write_unlock(&pag->pag_ici_lock);
196 radix_tree_preload_end();
203 * Look up an inode by number in the given file system.
204 * The inode is looked up in the cache held in each AG.
205 * If the inode is found in the cache, initialise the vfs inode
208 * If it is not in core, read it in from the file system's device,
209 * add it to the cache and initialise the vfs inode.
211 * The inode is locked according to the value of the lock_flags parameter.
212 * This flag parameter indicates how and if the inode's IO lock and inode lock
215 * mp -- the mount point structure for the current file system. It points
216 * to the inode hash table.
217 * tp -- a pointer to the current transaction if there is one. This is
218 * simply passed through to the xfs_iread() call.
219 * ino -- the number of the inode desired. This is the unique identifier
220 * within the file system for the inode being requested.
221 * lock_flags -- flags indicating how to lock the inode. See the comment
222 * for xfs_ilock() for a list of valid values.
223 * bno -- the block number starting the buffer containing the inode,
224 * if known (as by bulkstat), else 0.
241 /* the radix tree exists only in inode capable AGs */
242 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
245 /* get the perag structure and ensure that it's inode capable */
246 pag = xfs_get_perag(mp, ino);
247 if (!pag->pagi_inodeok)
249 ASSERT(pag->pag_ici_init);
250 agino = XFS_INO_TO_AGINO(mp, ino);
254 read_lock(&pag->pag_ici_lock);
255 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
258 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
260 goto out_error_or_again;
262 read_unlock(&pag->pag_ici_lock);
263 XFS_STATS_INC(xs_ig_missed);
265 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno,
268 goto out_error_or_again;
270 xfs_put_perag(mp, pag);
272 xfs_iflags_set(ip, XFS_IMODIFIED);
275 ASSERT(ip->i_df.if_ext_max ==
276 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
278 * If we have a real type for an on-disk inode, we can set ops(&unlock)
279 * now. If it's a new inode being created, xfs_ialloc will handle it.
281 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
286 if (error == EAGAIN) {
290 xfs_put_perag(mp, pag);
296 * Look for the inode corresponding to the given ino in the hash table.
297 * If it is there and its i_transp pointer matches tp, return it.
298 * Otherwise, return NULL.
301 xfs_inode_incore(xfs_mount_t *mp,
308 pag = xfs_get_perag(mp, ino);
309 read_lock(&pag->pag_ici_lock);
310 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
311 read_unlock(&pag->pag_ici_lock);
312 xfs_put_perag(mp, pag);
314 /* the returned inode must match the transaction */
315 if (ip && (ip->i_transp != tp))
321 * Decrement reference count of an inode structure and unlock it.
323 * ip -- the inode being released
324 * lock_flags -- this parameter indicates the inode's locks to be
325 * to be released. See the comment on xfs_iunlock() for a list
329 xfs_iput(xfs_inode_t *ip,
332 xfs_itrace_entry(ip);
333 xfs_iunlock(ip, lock_flags);
338 * Special iput for brand-new inodes that are still locked
345 struct inode *inode = VFS_I(ip);
347 xfs_itrace_entry(ip);
349 if ((ip->i_d.di_mode == 0)) {
350 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
351 make_bad_inode(inode);
353 if (inode->i_state & I_NEW)
354 unlock_new_inode(inode);
356 xfs_iunlock(ip, lock_flags);
362 * This routine embodies the part of the reclaim code that pulls
363 * the inode from the inode hash table and the mount structure's
365 * This should only be called from xfs_reclaim().
368 xfs_ireclaim(xfs_inode_t *ip)
371 * Remove from old hash list and mount list.
373 XFS_STATS_INC(xs_ig_reclaims);
378 * Here we do a spurious inode lock in order to coordinate with inode
379 * cache radix tree lookups. This is because the lookup can reference
380 * the inodes in the cache without taking references. We make that OK
381 * here by ensuring that we wait until the inode is unlocked after the
382 * lookup before we go ahead and free it. We get both the ilock and
383 * the iolock because the code may need to drop the ilock one but will
384 * still hold the iolock.
386 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
389 * Release dquots (and their references) if any. An inode may escape
390 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
392 XFS_QM_DQDETACH(ip->i_mount, ip);
395 * Free all memory associated with the inode.
397 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
402 * This routine removes an about-to-be-destroyed inode from
403 * all of the lists in which it is located with the exception
404 * of the behavior chain.
410 xfs_mount_t *mp = ip->i_mount;
411 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
413 write_lock(&pag->pag_ici_lock);
414 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
415 write_unlock(&pag->pag_ici_lock);
416 xfs_put_perag(mp, pag);
422 * This is a wrapper routine around the xfs_ilock() routine
423 * used to centralize some grungy code. It is used in places
424 * that wish to lock the inode solely for reading the extents.
425 * The reason these places can't just call xfs_ilock(SHARED)
426 * is that the inode lock also guards to bringing in of the
427 * extents from disk for a file in b-tree format. If the inode
428 * is in b-tree format, then we need to lock the inode exclusively
429 * until the extents are read in. Locking it exclusively all
430 * the time would limit our parallelism unnecessarily, though.
431 * What we do instead is check to see if the extents have been
432 * read in yet, and only lock the inode exclusively if they
435 * The function returns a value which should be given to the
436 * corresponding xfs_iunlock_map_shared(). This value is
437 * the mode in which the lock was actually taken.
440 xfs_ilock_map_shared(
445 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
446 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
447 lock_mode = XFS_ILOCK_EXCL;
449 lock_mode = XFS_ILOCK_SHARED;
452 xfs_ilock(ip, lock_mode);
458 * This is simply the unlock routine to go with xfs_ilock_map_shared().
459 * All it does is call xfs_iunlock() with the given lock_mode.
462 xfs_iunlock_map_shared(
464 unsigned int lock_mode)
466 xfs_iunlock(ip, lock_mode);
470 * The xfs inode contains 2 locks: a multi-reader lock called the
471 * i_iolock and a multi-reader lock called the i_lock. This routine
472 * allows either or both of the locks to be obtained.
474 * The 2 locks should always be ordered so that the IO lock is
475 * obtained first in order to prevent deadlock.
477 * ip -- the inode being locked
478 * lock_flags -- this parameter indicates the inode's locks
479 * to be locked. It can be:
484 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
485 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
486 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
487 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
495 * You can't set both SHARED and EXCL for the same lock,
496 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
497 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
499 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
500 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
501 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
502 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
503 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
505 if (lock_flags & XFS_IOLOCK_EXCL)
506 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
507 else if (lock_flags & XFS_IOLOCK_SHARED)
508 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
510 if (lock_flags & XFS_ILOCK_EXCL)
511 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
512 else if (lock_flags & XFS_ILOCK_SHARED)
513 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
515 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
519 * This is just like xfs_ilock(), except that the caller
520 * is guaranteed not to sleep. It returns 1 if it gets
521 * the requested locks and 0 otherwise. If the IO lock is
522 * obtained but the inode lock cannot be, then the IO lock
523 * is dropped before returning.
525 * ip -- the inode being locked
526 * lock_flags -- this parameter indicates the inode's locks to be
527 * to be locked. See the comment for xfs_ilock() for a list
536 * You can't set both SHARED and EXCL for the same lock,
537 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
538 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
540 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
541 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
542 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
543 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
544 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
546 if (lock_flags & XFS_IOLOCK_EXCL) {
547 if (!mrtryupdate(&ip->i_iolock))
549 } else if (lock_flags & XFS_IOLOCK_SHARED) {
550 if (!mrtryaccess(&ip->i_iolock))
553 if (lock_flags & XFS_ILOCK_EXCL) {
554 if (!mrtryupdate(&ip->i_lock))
555 goto out_undo_iolock;
556 } else if (lock_flags & XFS_ILOCK_SHARED) {
557 if (!mrtryaccess(&ip->i_lock))
558 goto out_undo_iolock;
560 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
564 if (lock_flags & XFS_IOLOCK_EXCL)
565 mrunlock_excl(&ip->i_iolock);
566 else if (lock_flags & XFS_IOLOCK_SHARED)
567 mrunlock_shared(&ip->i_iolock);
573 * xfs_iunlock() is used to drop the inode locks acquired with
574 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
575 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
576 * that we know which locks to drop.
578 * ip -- the inode being unlocked
579 * lock_flags -- this parameter indicates the inode's locks to be
580 * to be unlocked. See the comment for xfs_ilock() for a list
581 * of valid values for this parameter.
590 * You can't set both SHARED and EXCL for the same lock,
591 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
592 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
594 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
595 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
596 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
597 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
598 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
599 XFS_LOCK_DEP_MASK)) == 0);
600 ASSERT(lock_flags != 0);
602 if (lock_flags & XFS_IOLOCK_EXCL)
603 mrunlock_excl(&ip->i_iolock);
604 else if (lock_flags & XFS_IOLOCK_SHARED)
605 mrunlock_shared(&ip->i_iolock);
607 if (lock_flags & XFS_ILOCK_EXCL)
608 mrunlock_excl(&ip->i_lock);
609 else if (lock_flags & XFS_ILOCK_SHARED)
610 mrunlock_shared(&ip->i_lock);
612 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
613 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
615 * Let the AIL know that this item has been unlocked in case
616 * it is in the AIL and anyone is waiting on it. Don't do
617 * this if the caller has asked us not to.
619 xfs_trans_unlocked_item(ip->i_mount,
620 (xfs_log_item_t*)(ip->i_itemp));
622 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
626 * give up write locks. the i/o lock cannot be held nested
627 * if it is being demoted.
634 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
635 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
637 if (lock_flags & XFS_ILOCK_EXCL)
638 mrdemote(&ip->i_lock);
639 if (lock_flags & XFS_IOLOCK_EXCL)
640 mrdemote(&ip->i_iolock);
645 * Debug-only routine, without additional rw_semaphore APIs, we can
646 * now only answer requests regarding whether we hold the lock for write
647 * (reader state is outside our visibility, we only track writer state).
649 * Note: this means !xfs_isilocked would give false positives, so don't do that.
656 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
658 if (!ip->i_lock.mr_writer)
662 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
664 if (!ip->i_iolock.mr_writer)