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);
100 list_del_init(&ip->i_reclaim);
101 XFS_MOUNT_IUNLOCK(mp);
102 } else if (!igrab(VFS_I(ip))) {
103 /* If the VFS inode is being torn down, pause and try again. */
105 XFS_STATS_INC(xs_ig_frecycle);
108 /* we've got a live one */
109 read_unlock(&pag->pag_ici_lock);
112 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
118 xfs_ilock(ip, lock_flags);
120 xfs_iflags_clear(ip, XFS_ISTALE);
121 xfs_itrace_exit_tag(ip, "xfs_iget.found");
122 XFS_STATS_INC(xs_ig_found);
126 read_unlock(&pag->pag_ici_lock);
134 struct xfs_mount *mp,
135 struct xfs_perag *pag,
138 struct xfs_inode **ipp,
141 int lock_flags) __releases(pag->pag_ici_lock)
143 struct xfs_inode *ip;
145 unsigned long first_index, mask;
146 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
149 * Read the disk inode attributes into a new inode structure and get
150 * a new vnode for it. This should also initialize i_ino and i_mount.
152 error = xfs_iread(mp, tp, ino, &ip, bno,
153 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
157 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
159 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
165 * Preload the radix tree so we can insert safely under the
168 if (radix_tree_preload(GFP_KERNEL)) {
174 xfs_ilock(ip, lock_flags);
176 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
177 first_index = agino & mask;
178 write_lock(&pag->pag_ici_lock);
180 /* insert the new inode */
181 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
182 if (unlikely(error)) {
183 WARN_ON(error != -EEXIST);
184 XFS_STATS_INC(xs_ig_dup);
189 /* These values _must_ be set before releasing the radix tree lock! */
190 ip->i_udquot = ip->i_gdquot = NULL;
191 xfs_iflags_set(ip, XFS_INEW);
193 write_unlock(&pag->pag_ici_lock);
194 radix_tree_preload_end();
199 write_unlock(&pag->pag_ici_lock);
200 radix_tree_preload_end();
207 * Look up an inode by number in the given file system.
208 * The inode is looked up in the cache held in each AG.
209 * If the inode is found in the cache, initialise the vfs inode
212 * If it is not in core, read it in from the file system's device,
213 * add it to the cache and initialise the vfs inode.
215 * The inode is locked according to the value of the lock_flags parameter.
216 * This flag parameter indicates how and if the inode's IO lock and inode lock
219 * mp -- the mount point structure for the current file system. It points
220 * to the inode hash table.
221 * tp -- a pointer to the current transaction if there is one. This is
222 * simply passed through to the xfs_iread() call.
223 * ino -- the number of the inode desired. This is the unique identifier
224 * within the file system for the inode being requested.
225 * lock_flags -- flags indicating how to lock the inode. See the comment
226 * for xfs_ilock() for a list of valid values.
227 * bno -- the block number starting the buffer containing the inode,
228 * if known (as by bulkstat), else 0.
245 /* the radix tree exists only in inode capable AGs */
246 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
249 /* get the perag structure and ensure that it's inode capable */
250 pag = xfs_get_perag(mp, ino);
251 if (!pag->pagi_inodeok)
253 ASSERT(pag->pag_ici_init);
254 agino = XFS_INO_TO_AGINO(mp, ino);
258 read_lock(&pag->pag_ici_lock);
259 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
262 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
264 goto out_error_or_again;
266 read_unlock(&pag->pag_ici_lock);
267 XFS_STATS_INC(xs_ig_missed);
269 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno,
272 goto out_error_or_again;
274 xfs_put_perag(mp, pag);
276 xfs_iflags_set(ip, XFS_IMODIFIED);
279 ASSERT(ip->i_df.if_ext_max ==
280 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
282 * If we have a real type for an on-disk inode, we can set ops(&unlock)
283 * now. If it's a new inode being created, xfs_ialloc will handle it.
285 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
290 if (error == EAGAIN) {
294 xfs_put_perag(mp, pag);
300 * Look for the inode corresponding to the given ino in the hash table.
301 * If it is there and its i_transp pointer matches tp, return it.
302 * Otherwise, return NULL.
305 xfs_inode_incore(xfs_mount_t *mp,
312 pag = xfs_get_perag(mp, ino);
313 read_lock(&pag->pag_ici_lock);
314 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
315 read_unlock(&pag->pag_ici_lock);
316 xfs_put_perag(mp, pag);
318 /* the returned inode must match the transaction */
319 if (ip && (ip->i_transp != tp))
325 * Decrement reference count of an inode structure and unlock it.
327 * ip -- the inode being released
328 * lock_flags -- this parameter indicates the inode's locks to be
329 * to be released. See the comment on xfs_iunlock() for a list
333 xfs_iput(xfs_inode_t *ip,
336 xfs_itrace_entry(ip);
337 xfs_iunlock(ip, lock_flags);
342 * Special iput for brand-new inodes that are still locked
349 struct inode *inode = VFS_I(ip);
351 xfs_itrace_entry(ip);
353 if ((ip->i_d.di_mode == 0)) {
354 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
355 make_bad_inode(inode);
357 if (inode->i_state & I_NEW)
358 unlock_new_inode(inode);
360 xfs_iunlock(ip, lock_flags);
366 * This routine embodies the part of the reclaim code that pulls
367 * the inode from the inode hash table and the mount structure's
369 * This should only be called from xfs_reclaim().
372 xfs_ireclaim(xfs_inode_t *ip)
375 * Remove from old hash list and mount list.
377 XFS_STATS_INC(xs_ig_reclaims);
382 * Here we do a spurious inode lock in order to coordinate with inode
383 * cache radix tree lookups. This is because the lookup can reference
384 * the inodes in the cache without taking references. We make that OK
385 * here by ensuring that we wait until the inode is unlocked after the
386 * lookup before we go ahead and free it. We get both the ilock and
387 * the iolock because the code may need to drop the ilock one but will
388 * still hold the iolock.
390 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
393 * Release dquots (and their references) if any. An inode may escape
394 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
396 XFS_QM_DQDETACH(ip->i_mount, ip);
399 * Free all memory associated with the inode.
401 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
406 * This routine removes an about-to-be-destroyed inode from
407 * all of the lists in which it is located with the exception
408 * of the behavior chain.
414 xfs_mount_t *mp = ip->i_mount;
415 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
417 write_lock(&pag->pag_ici_lock);
418 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
419 write_unlock(&pag->pag_ici_lock);
420 xfs_put_perag(mp, pag);
422 /* Deal with the deleted inodes list */
424 list_del_init(&ip->i_reclaim);
426 XFS_MOUNT_IUNLOCK(mp);
430 * This is a wrapper routine around the xfs_ilock() routine
431 * used to centralize some grungy code. It is used in places
432 * that wish to lock the inode solely for reading the extents.
433 * The reason these places can't just call xfs_ilock(SHARED)
434 * is that the inode lock also guards to bringing in of the
435 * extents from disk for a file in b-tree format. If the inode
436 * is in b-tree format, then we need to lock the inode exclusively
437 * until the extents are read in. Locking it exclusively all
438 * the time would limit our parallelism unnecessarily, though.
439 * What we do instead is check to see if the extents have been
440 * read in yet, and only lock the inode exclusively if they
443 * The function returns a value which should be given to the
444 * corresponding xfs_iunlock_map_shared(). This value is
445 * the mode in which the lock was actually taken.
448 xfs_ilock_map_shared(
453 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
454 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
455 lock_mode = XFS_ILOCK_EXCL;
457 lock_mode = XFS_ILOCK_SHARED;
460 xfs_ilock(ip, lock_mode);
466 * This is simply the unlock routine to go with xfs_ilock_map_shared().
467 * All it does is call xfs_iunlock() with the given lock_mode.
470 xfs_iunlock_map_shared(
472 unsigned int lock_mode)
474 xfs_iunlock(ip, lock_mode);
478 * The xfs inode contains 2 locks: a multi-reader lock called the
479 * i_iolock and a multi-reader lock called the i_lock. This routine
480 * allows either or both of the locks to be obtained.
482 * The 2 locks should always be ordered so that the IO lock is
483 * obtained first in order to prevent deadlock.
485 * ip -- the inode being locked
486 * lock_flags -- this parameter indicates the inode's locks
487 * to be locked. It can be:
492 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
493 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
494 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
495 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
503 * You can't set both SHARED and EXCL for the same lock,
504 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
505 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
507 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
508 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
509 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
510 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
511 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
513 if (lock_flags & XFS_IOLOCK_EXCL)
514 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
515 else if (lock_flags & XFS_IOLOCK_SHARED)
516 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
518 if (lock_flags & XFS_ILOCK_EXCL)
519 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
520 else if (lock_flags & XFS_ILOCK_SHARED)
521 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
523 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
527 * This is just like xfs_ilock(), except that the caller
528 * is guaranteed not to sleep. It returns 1 if it gets
529 * the requested locks and 0 otherwise. If the IO lock is
530 * obtained but the inode lock cannot be, then the IO lock
531 * is dropped before returning.
533 * ip -- the inode being locked
534 * lock_flags -- this parameter indicates the inode's locks to be
535 * to be locked. See the comment for xfs_ilock() for a list
544 * You can't set both SHARED and EXCL for the same lock,
545 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
546 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
548 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
549 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
550 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
551 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
552 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
554 if (lock_flags & XFS_IOLOCK_EXCL) {
555 if (!mrtryupdate(&ip->i_iolock))
557 } else if (lock_flags & XFS_IOLOCK_SHARED) {
558 if (!mrtryaccess(&ip->i_iolock))
561 if (lock_flags & XFS_ILOCK_EXCL) {
562 if (!mrtryupdate(&ip->i_lock))
563 goto out_undo_iolock;
564 } else if (lock_flags & XFS_ILOCK_SHARED) {
565 if (!mrtryaccess(&ip->i_lock))
566 goto out_undo_iolock;
568 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
572 if (lock_flags & XFS_IOLOCK_EXCL)
573 mrunlock_excl(&ip->i_iolock);
574 else if (lock_flags & XFS_IOLOCK_SHARED)
575 mrunlock_shared(&ip->i_iolock);
581 * xfs_iunlock() is used to drop the inode locks acquired with
582 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
583 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
584 * that we know which locks to drop.
586 * ip -- the inode being unlocked
587 * lock_flags -- this parameter indicates the inode's locks to be
588 * to be unlocked. See the comment for xfs_ilock() for a list
589 * of valid values for this parameter.
598 * You can't set both SHARED and EXCL for the same lock,
599 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
600 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
602 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
603 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
604 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
605 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
606 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
607 XFS_LOCK_DEP_MASK)) == 0);
608 ASSERT(lock_flags != 0);
610 if (lock_flags & XFS_IOLOCK_EXCL)
611 mrunlock_excl(&ip->i_iolock);
612 else if (lock_flags & XFS_IOLOCK_SHARED)
613 mrunlock_shared(&ip->i_iolock);
615 if (lock_flags & XFS_ILOCK_EXCL)
616 mrunlock_excl(&ip->i_lock);
617 else if (lock_flags & XFS_ILOCK_SHARED)
618 mrunlock_shared(&ip->i_lock);
620 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
621 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
623 * Let the AIL know that this item has been unlocked in case
624 * it is in the AIL and anyone is waiting on it. Don't do
625 * this if the caller has asked us not to.
627 xfs_trans_unlocked_item(ip->i_mount,
628 (xfs_log_item_t*)(ip->i_itemp));
630 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
634 * give up write locks. the i/o lock cannot be held nested
635 * if it is being demoted.
642 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
643 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
645 if (lock_flags & XFS_ILOCK_EXCL)
646 mrdemote(&ip->i_lock);
647 if (lock_flags & XFS_IOLOCK_EXCL)
648 mrdemote(&ip->i_iolock);
653 * Debug-only routine, without additional rw_semaphore APIs, we can
654 * now only answer requests regarding whether we hold the lock for write
655 * (reader state is outside our visibility, we only track writer state).
657 * Note: this means !xfs_isilocked would give false positives, so don't do that.
664 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
666 if (!ip->i_lock.mr_writer)
670 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
672 if (!ip->i_iolock.mr_writer)