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"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_quota.h"
42 #include "xfs_utils.h"
45 * Initialize the inode hash table for the newly mounted file system.
46 * Choose an initial table size based on user specified value, else
47 * use a simple algorithm using the maximum number of inodes as an
48 * indicator for table size, and clamp it between one and some large
52 xfs_ihash_init(xfs_mount_t *mp)
55 uint i, flags = KM_SLEEP | KM_MAYFAIL;
58 icount = mp->m_maxicount ? mp->m_maxicount :
59 (mp->m_sb.sb_dblocks << mp->m_sb.sb_inopblog);
60 mp->m_ihsize = 1 << max_t(uint, 8,
61 (xfs_highbit64(icount) + 1) / 2);
62 mp->m_ihsize = min_t(uint, mp->m_ihsize,
63 (64 * NBPP) / sizeof(xfs_ihash_t));
66 while (!(mp->m_ihash = (xfs_ihash_t *)kmem_zalloc(mp->m_ihsize *
67 sizeof(xfs_ihash_t), flags))) {
68 if ((mp->m_ihsize >>= 1) <= NBPP)
71 for (i = 0; i < mp->m_ihsize; i++) {
72 rwlock_init(&(mp->m_ihash[i].ih_lock));
77 * Free up structures allocated by xfs_ihash_init, at unmount time.
80 xfs_ihash_free(xfs_mount_t *mp)
82 kmem_free(mp->m_ihash, mp->m_ihsize*sizeof(xfs_ihash_t));
87 * Initialize the inode cluster hash table for the newly mounted file system.
88 * Its size is derived from the ihash table size.
91 xfs_chash_init(xfs_mount_t *mp)
95 mp->m_chsize = max_t(uint, 1, mp->m_ihsize /
96 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog));
97 mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize);
98 mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize
99 * sizeof(xfs_chash_t),
101 for (i = 0; i < mp->m_chsize; i++) {
102 spinlock_init(&mp->m_chash[i].ch_lock,"xfshash");
107 * Free up structures allocated by xfs_chash_init, at unmount time.
110 xfs_chash_free(xfs_mount_t *mp)
114 for (i = 0; i < mp->m_chsize; i++) {
115 spinlock_destroy(&mp->m_chash[i].ch_lock);
118 kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t));
123 * Try to move an inode to the front of its hash list if possible
124 * (and if its not there already). Called right after obtaining
125 * the list version number and then dropping the read_lock on the
126 * hash list in question (which is done right after looking up the
127 * inode in question...).
137 if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) {
138 if (likely(version == ih->ih_version)) {
139 /* remove from list */
140 if ((iq = ip->i_next)) {
141 iq->i_prevp = ip->i_prevp;
145 /* insert at list head */
147 iq->i_prevp = &ip->i_next;
149 ip->i_prevp = &ih->ih_next;
152 write_unlock(&ih->ih_lock);
157 * Look up an inode by number in the given file system.
158 * The inode is looked up in the hash table for the file system
159 * represented by the mount point parameter mp. Each bucket of
160 * the hash table is guarded by an individual semaphore.
162 * If the inode is found in the hash table, its corresponding vnode
163 * is obtained with a call to vn_get(). This call takes care of
164 * coordination with the reclamation of the inode and vnode. Note
165 * that the vmap structure is filled in while holding the hash lock.
166 * This gives us the state of the inode/vnode when we found it and
167 * is used for coordination in vn_get().
169 * If it is not in core, read it in from the file system's device and
170 * add the inode into the hash table.
172 * The inode is locked according to the value of the lock_flags parameter.
173 * This flag parameter indicates how and if the inode's IO lock and inode lock
176 * mp -- the mount point structure for the current file system. It points
177 * to the inode hash table.
178 * tp -- a pointer to the current transaction if there is one. This is
179 * simply passed through to the xfs_iread() call.
180 * ino -- the number of the inode desired. This is the unique identifier
181 * within the file system for the inode being requested.
182 * lock_flags -- flags indicating how to lock the inode. See the comment
183 * for xfs_ilock() for a list of valid values.
184 * bno -- the block number starting the buffer containing the inode,
185 * if known (as by bulkstat), else 0.
206 xfs_chashlist_t *chl, *chlnew;
210 ih = XFS_IHASH(mp, ino);
213 read_lock(&ih->ih_lock);
215 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
216 if (ip->i_ino == ino) {
218 * If INEW is set this inode is being set up
219 * we need to pause and try again.
221 if (ip->i_flags & XFS_INEW) {
222 read_unlock(&ih->ih_lock);
224 XFS_STATS_INC(xs_ig_frecycle);
229 inode_vp = XFS_ITOV_NULL(ip);
230 if (inode_vp == NULL) {
232 * If IRECLAIM is set this inode is
233 * on its way out of the system,
234 * we need to pause and try again.
236 if (ip->i_flags & XFS_IRECLAIM) {
237 read_unlock(&ih->ih_lock);
239 XFS_STATS_INC(xs_ig_frecycle);
244 vn_trace_exit(vp, "xfs_iget.alloc",
245 (inst_t *)__return_address);
247 XFS_STATS_INC(xs_ig_found);
249 ip->i_flags &= ~XFS_IRECLAIMABLE;
250 version = ih->ih_version;
251 read_unlock(&ih->ih_lock);
252 xfs_ihash_promote(ih, ip, version);
255 list_del_init(&ip->i_reclaim);
256 XFS_MOUNT_IUNLOCK(mp);
260 } else if (vp != inode_vp) {
261 struct inode *inode = LINVFS_GET_IP(inode_vp);
263 /* The inode is being torn down, pause and
266 if (inode->i_state & (I_FREEING | I_CLEAR)) {
267 read_unlock(&ih->ih_lock);
269 XFS_STATS_INC(xs_ig_frecycle);
273 /* Chances are the other vnode (the one in the inode) is being torn
274 * down right now, and we landed on top of it. Question is, what do
275 * we do? Unhook the old inode and hook up the new one?
278 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
283 * Inode cache hit: if ip is not at the front of
284 * its hash chain, move it there now.
285 * Do this with the lock held for update, but
286 * do statistics after releasing the lock.
288 version = ih->ih_version;
289 read_unlock(&ih->ih_lock);
290 xfs_ihash_promote(ih, ip, version);
291 XFS_STATS_INC(xs_ig_found);
294 if (ip->i_d.di_mode == 0) {
295 if (!(flags & IGET_CREATE))
297 xfs_iocore_inode_reinit(ip);
301 xfs_ilock(ip, lock_flags);
303 ip->i_flags &= ~XFS_ISTALE;
305 vn_trace_exit(vp, "xfs_iget.found",
306 (inst_t *)__return_address);
312 * Inode cache miss: save the hash chain version stamp and unlock
313 * the chain, so we don't deadlock in vn_alloc.
315 XFS_STATS_INC(xs_ig_missed);
317 version = ih->ih_version;
319 read_unlock(&ih->ih_lock);
322 * Read the disk inode attributes into a new inode structure and get
323 * a new vnode for it. This should also initialize i_ino and i_mount.
325 error = xfs_iread(mp, tp, ino, &ip, bno);
330 vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address);
332 xfs_inode_lock_init(ip, vp);
333 xfs_iocore_inode_init(ip);
335 if (lock_flags != 0) {
336 xfs_ilock(ip, lock_flags);
339 if ((ip->i_d.di_mode == 0) && !(flags & IGET_CREATE)) {
345 * Put ip on its hash chain, unless someone else hashed a duplicate
346 * after we released the hash lock.
348 write_lock(&ih->ih_lock);
350 if (ih->ih_version != version) {
351 for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) {
352 if (iq->i_ino == ino) {
353 write_unlock(&ih->ih_lock);
356 XFS_STATS_INC(xs_ig_dup);
363 * These values _must_ be set before releasing ihlock!
366 if ((iq = ih->ih_next)) {
367 iq->i_prevp = &ip->i_next;
370 ip->i_prevp = &ih->ih_next;
372 ip->i_udquot = ip->i_gdquot = NULL;
374 ip->i_flags |= XFS_INEW;
376 write_unlock(&ih->ih_lock);
379 * put ip on its cluster's hash chain
381 ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL &&
382 ip->i_cnext == NULL);
385 ch = XFS_CHASH(mp, ip->i_blkno);
387 s = mutex_spinlock(&ch->ch_lock);
388 for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
389 if (chl->chl_blkno == ip->i_blkno) {
391 /* insert this inode into the doubly-linked list
392 * where chl points */
393 if ((iq = chl->chl_ip)) {
394 ip->i_cprev = iq->i_cprev;
395 iq->i_cprev->i_cnext = ip;
408 /* no hash list found for this block; add a new hash list */
410 if (chlnew == NULL) {
411 mutex_spinunlock(&ch->ch_lock, s);
412 ASSERT(xfs_chashlist_zone != NULL);
413 chlnew = (xfs_chashlist_t *)
414 kmem_zone_alloc(xfs_chashlist_zone,
416 ASSERT(chlnew != NULL);
421 ip->i_chash = chlnew;
423 chlnew->chl_blkno = ip->i_blkno;
424 chlnew->chl_next = ch->ch_list;
425 ch->ch_list = chlnew;
429 if (chlnew != NULL) {
430 kmem_zone_free(xfs_chashlist_zone, chlnew);
434 mutex_spinunlock(&ch->ch_lock, s);
438 * Link ip to its mount and thread it on the mount's inode list.
441 if ((iq = mp->m_inodes)) {
442 ASSERT(iq->i_mprev->i_mnext == iq);
443 ip->i_mprev = iq->i_mprev;
444 iq->i_mprev->i_mnext = ip;
453 XFS_MOUNT_IUNLOCK(mp);
456 ASSERT(ip->i_df.if_ext_max ==
457 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
459 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
460 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
465 * If we have a real type for an on-disk inode, we can set ops(&unlock)
466 * now. If it's a new inode being created, xfs_ialloc will handle it.
468 VFS_INIT_VNODE(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1);
475 * The 'normal' internal xfs_iget, if needed it will
476 * 'allocate', or 'get', the vnode.
492 XFS_STATS_INC(xs_ig_attempts);
495 if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) {
499 vp = LINVFS_GET_VP(inode);
500 if (inode->i_state & I_NEW) {
501 vn_initialize(inode);
502 error = xfs_iget_core(vp, mp, tp, ino, flags,
503 lock_flags, ipp, bno);
506 if (inode->i_state & I_NEW)
507 unlock_new_inode(inode);
512 * If the inode is not fully constructed due to
513 * filehandle mistmatches wait for the inode to go
514 * away and try again.
516 * iget_locked will call __wait_on_freeing_inode
517 * to wait for the inode to go away.
519 if (is_bad_inode(inode) ||
520 ((bdp = vn_bhv_lookup(VN_BHV_HEAD(vp),
521 &xfs_vnodeops)) == NULL)) {
527 ip = XFS_BHVTOI(bdp);
529 xfs_ilock(ip, lock_flags);
530 XFS_STATS_INC(xs_ig_found);
535 error = ENOMEM; /* If we got no inode we are out of memory */
541 * Do the setup for the various locks within the incore inode.
548 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
549 "xfsino", (long)vp->v_number);
550 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", vp->v_number);
551 init_waitqueue_head(&ip->i_ipin_wait);
552 atomic_set(&ip->i_pincount, 0);
553 init_sema(&ip->i_flock, 1, "xfsfino", vp->v_number);
557 * Look for the inode corresponding to the given ino in the hash table.
558 * If it is there and its i_transp pointer matches tp, return it.
559 * Otherwise, return NULL.
562 xfs_inode_incore(xfs_mount_t *mp,
570 ih = XFS_IHASH(mp, ino);
571 read_lock(&ih->ih_lock);
572 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
573 if (ip->i_ino == ino) {
575 * If we find it and tp matches, return it.
576 * Also move it to the front of the hash list
577 * if we find it and it is not already there.
578 * Otherwise break from the loop and return
581 if (ip->i_transp == tp) {
582 version = ih->ih_version;
583 read_unlock(&ih->ih_lock);
584 xfs_ihash_promote(ih, ip, version);
590 read_unlock(&ih->ih_lock);
595 * Decrement reference count of an inode structure and unlock it.
597 * ip -- the inode being released
598 * lock_flags -- this parameter indicates the inode's locks to be
599 * to be released. See the comment on xfs_iunlock() for a list
603 xfs_iput(xfs_inode_t *ip,
606 vnode_t *vp = XFS_ITOV(ip);
608 vn_trace_entry(vp, "xfs_iput", (inst_t *)__return_address);
610 xfs_iunlock(ip, lock_flags);
616 * Special iput for brand-new inodes that are still locked
619 xfs_iput_new(xfs_inode_t *ip,
622 vnode_t *vp = XFS_ITOV(ip);
623 struct inode *inode = LINVFS_GET_IP(vp);
625 vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address);
627 if ((ip->i_d.di_mode == 0)) {
628 ASSERT(!(ip->i_flags & XFS_IRECLAIMABLE));
631 if (inode->i_state & I_NEW)
632 unlock_new_inode(inode);
634 xfs_iunlock(ip, lock_flags);
640 * This routine embodies the part of the reclaim code that pulls
641 * the inode from the inode hash table and the mount structure's
643 * This should only be called from xfs_reclaim().
646 xfs_ireclaim(xfs_inode_t *ip)
651 * Remove from old hash list and mount list.
653 XFS_STATS_INC(xs_ig_reclaims);
658 * Here we do a spurious inode lock in order to coordinate with
659 * xfs_sync(). This is because xfs_sync() references the inodes
660 * in the mount list without taking references on the corresponding
661 * vnodes. We make that OK here by ensuring that we wait until
662 * the inode is unlocked in xfs_sync() before we go ahead and
663 * free it. We get both the regular lock and the io lock because
664 * the xfs_sync() code may need to drop the regular one but will
665 * still hold the io lock.
667 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
670 * Release dquots (and their references) if any. An inode may escape
671 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
673 XFS_QM_DQDETACH(ip->i_mount, ip);
676 * Pull our behavior descriptor from the vnode chain.
678 vp = XFS_ITOV_NULL(ip);
680 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
684 * Free all memory associated with the inode.
690 * This routine removes an about-to-be-destroyed inode from
691 * all of the lists in which it is located with the exception
692 * of the behavior chain.
702 xfs_chashlist_t *chl, *chm;
706 write_lock(&ih->ih_lock);
707 if ((iq = ip->i_next)) {
708 iq->i_prevp = ip->i_prevp;
712 write_unlock(&ih->ih_lock);
715 * Remove from cluster hash list
716 * 1) delete the chashlist if this is the last inode on the chashlist
717 * 2) unchain from list of inodes
718 * 3) point chashlist->chl_ip to 'chl_next' if to this inode.
721 ch = XFS_CHASH(mp, ip->i_blkno);
722 s = mutex_spinlock(&ch->ch_lock);
724 if (ip->i_cnext == ip) {
725 /* Last inode on chashlist */
726 ASSERT(ip->i_cnext == ip && ip->i_cprev == ip);
727 ASSERT(ip->i_chash != NULL);
729 for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
730 if (chl->chl_blkno == ip->i_blkno) {
732 /* first item on the list */
733 ch->ch_list = chl->chl_next;
735 chm->chl_next = chl->chl_next;
737 kmem_zone_free(xfs_chashlist_zone, chl);
740 ASSERT(chl->chl_ip != ip);
744 ASSERT_ALWAYS(chl != NULL);
746 /* delete one inode from a non-empty list */
748 iq->i_cprev = ip->i_cprev;
749 ip->i_cprev->i_cnext = iq;
750 if (ip->i_chash->chl_ip == ip) {
751 ip->i_chash->chl_ip = iq;
753 ip->i_chash = __return_address;
754 ip->i_cprev = __return_address;
755 ip->i_cnext = __return_address;
757 mutex_spinunlock(&ch->ch_lock, s);
760 * Remove from mount's inode list.
763 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
765 iq->i_mprev = ip->i_mprev;
766 ip->i_mprev->i_mnext = iq;
769 * Fix up the head pointer if it points to the inode being deleted.
771 if (mp->m_inodes == ip) {
779 /* Deal with the deleted inodes list */
780 list_del_init(&ip->i_reclaim);
783 XFS_MOUNT_IUNLOCK(mp);
787 * This is a wrapper routine around the xfs_ilock() routine
788 * used to centralize some grungy code. It is used in places
789 * that wish to lock the inode solely for reading the extents.
790 * The reason these places can't just call xfs_ilock(SHARED)
791 * is that the inode lock also guards to bringing in of the
792 * extents from disk for a file in b-tree format. If the inode
793 * is in b-tree format, then we need to lock the inode exclusively
794 * until the extents are read in. Locking it exclusively all
795 * the time would limit our parallelism unnecessarily, though.
796 * What we do instead is check to see if the extents have been
797 * read in yet, and only lock the inode exclusively if they
800 * The function returns a value which should be given to the
801 * corresponding xfs_iunlock_map_shared(). This value is
802 * the mode in which the lock was actually taken.
805 xfs_ilock_map_shared(
810 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
811 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
812 lock_mode = XFS_ILOCK_EXCL;
814 lock_mode = XFS_ILOCK_SHARED;
817 xfs_ilock(ip, lock_mode);
823 * This is simply the unlock routine to go with xfs_ilock_map_shared().
824 * All it does is call xfs_iunlock() with the given lock_mode.
827 xfs_iunlock_map_shared(
829 unsigned int lock_mode)
831 xfs_iunlock(ip, lock_mode);
835 * The xfs inode contains 2 locks: a multi-reader lock called the
836 * i_iolock and a multi-reader lock called the i_lock. This routine
837 * allows either or both of the locks to be obtained.
839 * The 2 locks should always be ordered so that the IO lock is
840 * obtained first in order to prevent deadlock.
842 * ip -- the inode being locked
843 * lock_flags -- this parameter indicates the inode's locks
844 * to be locked. It can be:
849 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
850 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
851 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
852 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
855 xfs_ilock(xfs_inode_t *ip,
859 * You can't set both SHARED and EXCL for the same lock,
860 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
861 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
863 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
864 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
865 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
866 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
867 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
869 if (lock_flags & XFS_IOLOCK_EXCL) {
870 mrupdate(&ip->i_iolock);
871 } else if (lock_flags & XFS_IOLOCK_SHARED) {
872 mraccess(&ip->i_iolock);
874 if (lock_flags & XFS_ILOCK_EXCL) {
875 mrupdate(&ip->i_lock);
876 } else if (lock_flags & XFS_ILOCK_SHARED) {
877 mraccess(&ip->i_lock);
879 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
883 * This is just like xfs_ilock(), except that the caller
884 * is guaranteed not to sleep. It returns 1 if it gets
885 * the requested locks and 0 otherwise. If the IO lock is
886 * obtained but the inode lock cannot be, then the IO lock
887 * is dropped before returning.
889 * ip -- the inode being locked
890 * lock_flags -- this parameter indicates the inode's locks to be
891 * to be locked. See the comment for xfs_ilock() for a list
896 xfs_ilock_nowait(xfs_inode_t *ip,
903 * You can't set both SHARED and EXCL for the same lock,
904 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
905 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
907 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
908 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
909 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
910 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
911 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
914 if (lock_flags & XFS_IOLOCK_EXCL) {
915 iolocked = mrtryupdate(&ip->i_iolock);
919 } else if (lock_flags & XFS_IOLOCK_SHARED) {
920 iolocked = mrtryaccess(&ip->i_iolock);
925 if (lock_flags & XFS_ILOCK_EXCL) {
926 ilocked = mrtryupdate(&ip->i_lock);
929 mrunlock(&ip->i_iolock);
933 } else if (lock_flags & XFS_ILOCK_SHARED) {
934 ilocked = mrtryaccess(&ip->i_lock);
937 mrunlock(&ip->i_iolock);
942 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
947 * xfs_iunlock() is used to drop the inode locks acquired with
948 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
949 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
950 * that we know which locks to drop.
952 * ip -- the inode being unlocked
953 * lock_flags -- this parameter indicates the inode's locks to be
954 * to be unlocked. See the comment for xfs_ilock() for a list
955 * of valid values for this parameter.
959 xfs_iunlock(xfs_inode_t *ip,
963 * You can't set both SHARED and EXCL for the same lock,
964 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
965 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
967 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
968 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
969 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
970 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
971 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY)) == 0);
972 ASSERT(lock_flags != 0);
974 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
975 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
976 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
977 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
978 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
979 mrunlock(&ip->i_iolock);
982 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
983 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
984 (ismrlocked(&ip->i_lock, MR_ACCESS)));
985 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
986 (ismrlocked(&ip->i_lock, MR_UPDATE)));
987 mrunlock(&ip->i_lock);
990 * Let the AIL know that this item has been unlocked in case
991 * it is in the AIL and anyone is waiting on it. Don't do
992 * this if the caller has asked us not to.
994 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
995 ip->i_itemp != NULL) {
996 xfs_trans_unlocked_item(ip->i_mount,
997 (xfs_log_item_t*)(ip->i_itemp));
1000 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
1004 * give up write locks. the i/o lock cannot be held nested
1005 * if it is being demoted.
1008 xfs_ilock_demote(xfs_inode_t *ip,
1011 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
1012 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
1014 if (lock_flags & XFS_ILOCK_EXCL) {
1015 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
1016 mrdemote(&ip->i_lock);
1018 if (lock_flags & XFS_IOLOCK_EXCL) {
1019 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1020 mrdemote(&ip->i_iolock);
1025 * The following three routines simply manage the i_flock
1026 * semaphore embedded in the inode. This semaphore synchronizes
1027 * processes attempting to flush the in-core inode back to disk.
1030 xfs_iflock(xfs_inode_t *ip)
1032 psema(&(ip->i_flock), PINOD|PLTWAIT);
1036 xfs_iflock_nowait(xfs_inode_t *ip)
1038 return (cpsema(&(ip->i_flock)));
1042 xfs_ifunlock(xfs_inode_t *ip)
1044 ASSERT(valusema(&(ip->i_flock)) <= 0);
1045 vsema(&(ip->i_flock));