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);
94 read_unlock(&pag->pag_ici_lock);
97 list_del_init(&ip->i_reclaim);
98 XFS_MOUNT_IUNLOCK(mp);
99 } else if (!igrab(VFS_I(ip))) {
100 /* If the VFS inode is being torn down, pause and try again. */
102 XFS_STATS_INC(xs_ig_frecycle);
105 /* we've got a live one */
106 read_unlock(&pag->pag_ici_lock);
109 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
115 xfs_ilock(ip, lock_flags);
117 xfs_iflags_clear(ip, XFS_ISTALE);
118 xfs_itrace_exit_tag(ip, "xfs_iget.found");
119 XFS_STATS_INC(xs_ig_found);
123 read_unlock(&pag->pag_ici_lock);
131 struct xfs_mount *mp,
132 struct xfs_perag *pag,
135 struct xfs_inode **ipp,
138 int lock_flags) __releases(pag->pag_ici_lock)
140 struct xfs_inode *ip;
142 unsigned long first_index, mask;
143 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
146 * Read the disk inode attributes into a new inode structure and get
147 * a new vnode for it. This should also initialize i_ino and i_mount.
149 error = xfs_iread(mp, tp, ino, &ip, bno,
150 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
154 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
156 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
162 * Preload the radix tree so we can insert safely under the
165 if (radix_tree_preload(GFP_KERNEL)) {
171 xfs_ilock(ip, lock_flags);
173 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
174 first_index = agino & mask;
175 write_lock(&pag->pag_ici_lock);
177 /* insert the new inode */
178 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
179 if (unlikely(error)) {
180 WARN_ON(error != -EEXIST);
181 XFS_STATS_INC(xs_ig_dup);
186 /* These values _must_ be set before releasing the radix tree lock! */
187 ip->i_udquot = ip->i_gdquot = NULL;
188 xfs_iflags_set(ip, XFS_INEW);
190 write_unlock(&pag->pag_ici_lock);
191 radix_tree_preload_end();
196 write_unlock(&pag->pag_ici_lock);
197 radix_tree_preload_end();
204 * Look up an inode by number in the given file system.
205 * The inode is looked up in the cache held in each AG.
206 * If the inode is found in the cache, initialise the vfs inode
209 * If it is not in core, read it in from the file system's device,
210 * add it to the cache and initialise the vfs inode.
212 * The inode is locked according to the value of the lock_flags parameter.
213 * This flag parameter indicates how and if the inode's IO lock and inode lock
216 * mp -- the mount point structure for the current file system. It points
217 * to the inode hash table.
218 * tp -- a pointer to the current transaction if there is one. This is
219 * simply passed through to the xfs_iread() call.
220 * ino -- the number of the inode desired. This is the unique identifier
221 * within the file system for the inode being requested.
222 * lock_flags -- flags indicating how to lock the inode. See the comment
223 * for xfs_ilock() for a list of valid values.
224 * bno -- the block number starting the buffer containing the inode,
225 * if known (as by bulkstat), else 0.
242 /* the radix tree exists only in inode capable AGs */
243 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
246 /* get the perag structure and ensure that it's inode capable */
247 pag = xfs_get_perag(mp, ino);
248 if (!pag->pagi_inodeok)
250 ASSERT(pag->pag_ici_init);
251 agino = XFS_INO_TO_AGINO(mp, ino);
255 read_lock(&pag->pag_ici_lock);
256 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
259 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
261 goto out_error_or_again;
263 read_unlock(&pag->pag_ici_lock);
264 XFS_STATS_INC(xs_ig_missed);
266 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno,
269 goto out_error_or_again;
271 xfs_put_perag(mp, pag);
273 xfs_iflags_set(ip, XFS_IMODIFIED);
276 ASSERT(ip->i_df.if_ext_max ==
277 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
279 * If we have a real type for an on-disk inode, we can set ops(&unlock)
280 * now. If it's a new inode being created, xfs_ialloc will handle it.
282 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
287 if (error == EAGAIN) {
291 xfs_put_perag(mp, pag);
297 * Look for the inode corresponding to the given ino in the hash table.
298 * If it is there and its i_transp pointer matches tp, return it.
299 * Otherwise, return NULL.
302 xfs_inode_incore(xfs_mount_t *mp,
309 pag = xfs_get_perag(mp, ino);
310 read_lock(&pag->pag_ici_lock);
311 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
312 read_unlock(&pag->pag_ici_lock);
313 xfs_put_perag(mp, pag);
315 /* the returned inode must match the transaction */
316 if (ip && (ip->i_transp != tp))
322 * Decrement reference count of an inode structure and unlock it.
324 * ip -- the inode being released
325 * lock_flags -- this parameter indicates the inode's locks to be
326 * to be released. See the comment on xfs_iunlock() for a list
330 xfs_iput(xfs_inode_t *ip,
333 xfs_itrace_entry(ip);
334 xfs_iunlock(ip, lock_flags);
339 * Special iput for brand-new inodes that are still locked
346 struct inode *inode = VFS_I(ip);
348 xfs_itrace_entry(ip);
350 if ((ip->i_d.di_mode == 0)) {
351 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
352 make_bad_inode(inode);
354 if (inode->i_state & I_NEW)
355 unlock_new_inode(inode);
357 xfs_iunlock(ip, lock_flags);
363 * This routine embodies the part of the reclaim code that pulls
364 * the inode from the inode hash table and the mount structure's
366 * This should only be called from xfs_reclaim().
369 xfs_ireclaim(xfs_inode_t *ip)
372 * Remove from old hash list and mount list.
374 XFS_STATS_INC(xs_ig_reclaims);
379 * Here we do a spurious inode lock in order to coordinate with inode
380 * cache radix tree lookups. This is because the lookup can reference
381 * the inodes in the cache without taking references. We make that OK
382 * here by ensuring that we wait until the inode is unlocked after the
383 * lookup before we go ahead and free it. We get both the ilock and
384 * the iolock because the code may need to drop the ilock one but will
385 * still hold the iolock.
387 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
390 * Release dquots (and their references) if any. An inode may escape
391 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
393 XFS_QM_DQDETACH(ip->i_mount, ip);
396 * Free all memory associated with the inode.
398 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
403 * This routine removes an about-to-be-destroyed inode from
404 * all of the lists in which it is located with the exception
405 * of the behavior chain.
411 xfs_mount_t *mp = ip->i_mount;
412 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
414 write_lock(&pag->pag_ici_lock);
415 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
416 write_unlock(&pag->pag_ici_lock);
417 xfs_put_perag(mp, pag);
419 /* Deal with the deleted inodes list */
421 list_del_init(&ip->i_reclaim);
423 XFS_MOUNT_IUNLOCK(mp);
427 * This is a wrapper routine around the xfs_ilock() routine
428 * used to centralize some grungy code. It is used in places
429 * that wish to lock the inode solely for reading the extents.
430 * The reason these places can't just call xfs_ilock(SHARED)
431 * is that the inode lock also guards to bringing in of the
432 * extents from disk for a file in b-tree format. If the inode
433 * is in b-tree format, then we need to lock the inode exclusively
434 * until the extents are read in. Locking it exclusively all
435 * the time would limit our parallelism unnecessarily, though.
436 * What we do instead is check to see if the extents have been
437 * read in yet, and only lock the inode exclusively if they
440 * The function returns a value which should be given to the
441 * corresponding xfs_iunlock_map_shared(). This value is
442 * the mode in which the lock was actually taken.
445 xfs_ilock_map_shared(
450 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
451 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
452 lock_mode = XFS_ILOCK_EXCL;
454 lock_mode = XFS_ILOCK_SHARED;
457 xfs_ilock(ip, lock_mode);
463 * This is simply the unlock routine to go with xfs_ilock_map_shared().
464 * All it does is call xfs_iunlock() with the given lock_mode.
467 xfs_iunlock_map_shared(
469 unsigned int lock_mode)
471 xfs_iunlock(ip, lock_mode);
475 * The xfs inode contains 2 locks: a multi-reader lock called the
476 * i_iolock and a multi-reader lock called the i_lock. This routine
477 * allows either or both of the locks to be obtained.
479 * The 2 locks should always be ordered so that the IO lock is
480 * obtained first in order to prevent deadlock.
482 * ip -- the inode being locked
483 * lock_flags -- this parameter indicates the inode's locks
484 * to be locked. It can be:
489 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
490 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
491 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
492 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
500 * You can't set both SHARED and EXCL for the same lock,
501 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
502 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
504 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
505 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
506 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
507 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
508 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
510 if (lock_flags & XFS_IOLOCK_EXCL)
511 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
512 else if (lock_flags & XFS_IOLOCK_SHARED)
513 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
515 if (lock_flags & XFS_ILOCK_EXCL)
516 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
517 else if (lock_flags & XFS_ILOCK_SHARED)
518 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
520 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
524 * This is just like xfs_ilock(), except that the caller
525 * is guaranteed not to sleep. It returns 1 if it gets
526 * the requested locks and 0 otherwise. If the IO lock is
527 * obtained but the inode lock cannot be, then the IO lock
528 * is dropped before returning.
530 * ip -- the inode being locked
531 * lock_flags -- this parameter indicates the inode's locks to be
532 * to be locked. See the comment for xfs_ilock() for a list
541 * You can't set both SHARED and EXCL for the same lock,
542 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
543 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
545 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
546 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
547 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
548 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
549 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
551 if (lock_flags & XFS_IOLOCK_EXCL) {
552 if (!mrtryupdate(&ip->i_iolock))
554 } else if (lock_flags & XFS_IOLOCK_SHARED) {
555 if (!mrtryaccess(&ip->i_iolock))
558 if (lock_flags & XFS_ILOCK_EXCL) {
559 if (!mrtryupdate(&ip->i_lock))
560 goto out_undo_iolock;
561 } else if (lock_flags & XFS_ILOCK_SHARED) {
562 if (!mrtryaccess(&ip->i_lock))
563 goto out_undo_iolock;
565 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
569 if (lock_flags & XFS_IOLOCK_EXCL)
570 mrunlock_excl(&ip->i_iolock);
571 else if (lock_flags & XFS_IOLOCK_SHARED)
572 mrunlock_shared(&ip->i_iolock);
578 * xfs_iunlock() is used to drop the inode locks acquired with
579 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
580 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
581 * that we know which locks to drop.
583 * ip -- the inode being unlocked
584 * lock_flags -- this parameter indicates the inode's locks to be
585 * to be unlocked. See the comment for xfs_ilock() for a list
586 * of valid values for this parameter.
595 * You can't set both SHARED and EXCL for the same lock,
596 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
597 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
599 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
600 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
601 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
602 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
603 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
604 XFS_LOCK_DEP_MASK)) == 0);
605 ASSERT(lock_flags != 0);
607 if (lock_flags & XFS_IOLOCK_EXCL)
608 mrunlock_excl(&ip->i_iolock);
609 else if (lock_flags & XFS_IOLOCK_SHARED)
610 mrunlock_shared(&ip->i_iolock);
612 if (lock_flags & XFS_ILOCK_EXCL)
613 mrunlock_excl(&ip->i_lock);
614 else if (lock_flags & XFS_ILOCK_SHARED)
615 mrunlock_shared(&ip->i_lock);
617 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
618 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
620 * Let the AIL know that this item has been unlocked in case
621 * it is in the AIL and anyone is waiting on it. Don't do
622 * this if the caller has asked us not to.
624 xfs_trans_unlocked_item(ip->i_mount,
625 (xfs_log_item_t*)(ip->i_itemp));
627 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
631 * give up write locks. the i/o lock cannot be held nested
632 * if it is being demoted.
639 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
640 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
642 if (lock_flags & XFS_ILOCK_EXCL)
643 mrdemote(&ip->i_lock);
644 if (lock_flags & XFS_IOLOCK_EXCL)
645 mrdemote(&ip->i_iolock);
650 * Debug-only routine, without additional rw_semaphore APIs, we can
651 * now only answer requests regarding whether we hold the lock for write
652 * (reader state is outside our visibility, we only track writer state).
654 * Note: this means !xfs_isilocked would give false positives, so don't do that.
661 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
663 if (!ip->i_lock.mr_writer)
667 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
669 if (!ip->i_iolock.mr_writer)