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 * Initialize the inode hash table for the newly mounted file system.
44 * Choose an initial table size based on user specified value, else
45 * use a simple algorithm using the maximum number of inodes as an
46 * indicator for table size, and clamp it between one and some large
50 xfs_ihash_init(xfs_mount_t *mp)
53 uint i, flags = KM_SLEEP | KM_MAYFAIL;
56 icount = mp->m_maxicount ? mp->m_maxicount :
57 (mp->m_sb.sb_dblocks << mp->m_sb.sb_inopblog);
58 mp->m_ihsize = 1 << max_t(uint, 8,
59 (xfs_highbit64(icount) + 1) / 2);
60 mp->m_ihsize = min_t(uint, mp->m_ihsize,
61 (64 * NBPP) / sizeof(xfs_ihash_t));
64 while (!(mp->m_ihash = (xfs_ihash_t *)kmem_zalloc(mp->m_ihsize *
65 sizeof(xfs_ihash_t), flags))) {
66 if ((mp->m_ihsize >>= 1) <= NBPP)
69 for (i = 0; i < mp->m_ihsize; i++) {
70 rwlock_init(&(mp->m_ihash[i].ih_lock));
75 * Free up structures allocated by xfs_ihash_init, at unmount time.
78 xfs_ihash_free(xfs_mount_t *mp)
80 kmem_free(mp->m_ihash, mp->m_ihsize*sizeof(xfs_ihash_t));
85 * Initialize the inode cluster hash table for the newly mounted file system.
86 * Its size is derived from the ihash table size.
89 xfs_chash_init(xfs_mount_t *mp)
93 mp->m_chsize = max_t(uint, 1, mp->m_ihsize /
94 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog));
95 mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize);
96 mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize
97 * sizeof(xfs_chash_t),
99 for (i = 0; i < mp->m_chsize; i++) {
100 spinlock_init(&mp->m_chash[i].ch_lock,"xfshash");
105 * Free up structures allocated by xfs_chash_init, at unmount time.
108 xfs_chash_free(xfs_mount_t *mp)
112 for (i = 0; i < mp->m_chsize; i++) {
113 spinlock_destroy(&mp->m_chash[i].ch_lock);
116 kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t));
121 * Try to move an inode to the front of its hash list if possible
122 * (and if its not there already). Called right after obtaining
123 * the list version number and then dropping the read_lock on the
124 * hash list in question (which is done right after looking up the
125 * inode in question...).
135 if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) {
136 if (likely(version == ih->ih_version)) {
137 /* remove from list */
138 if ((iq = ip->i_next)) {
139 iq->i_prevp = ip->i_prevp;
143 /* insert at list head */
145 iq->i_prevp = &ip->i_next;
147 ip->i_prevp = &ih->ih_next;
150 write_unlock(&ih->ih_lock);
155 * Look up an inode by number in the given file system.
156 * The inode is looked up in the hash table for the file system
157 * represented by the mount point parameter mp. Each bucket of
158 * the hash table is guarded by an individual semaphore.
160 * If the inode is found in the hash table, its corresponding vnode
161 * is obtained with a call to vn_get(). This call takes care of
162 * coordination with the reclamation of the inode and vnode. Note
163 * that the vmap structure is filled in while holding the hash lock.
164 * This gives us the state of the inode/vnode when we found it and
165 * is used for coordination in vn_get().
167 * If it is not in core, read it in from the file system's device and
168 * add the inode into the hash table.
170 * The inode is locked according to the value of the lock_flags parameter.
171 * This flag parameter indicates how and if the inode's IO lock and inode lock
174 * mp -- the mount point structure for the current file system. It points
175 * to the inode hash table.
176 * tp -- a pointer to the current transaction if there is one. This is
177 * simply passed through to the xfs_iread() call.
178 * ino -- the number of the inode desired. This is the unique identifier
179 * within the file system for the inode being requested.
180 * lock_flags -- flags indicating how to lock the inode. See the comment
181 * for xfs_ilock() for a list of valid values.
182 * bno -- the block number starting the buffer containing the inode,
183 * if known (as by bulkstat), else 0.
199 bhv_vnode_t *inode_vp;
204 xfs_chashlist_t *chl, *chlnew;
208 ih = XFS_IHASH(mp, ino);
211 read_lock(&ih->ih_lock);
213 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
214 if (ip->i_ino == ino) {
216 * If INEW is set this inode is being set up
217 * we need to pause and try again.
219 if (ip->i_flags & XFS_INEW) {
220 read_unlock(&ih->ih_lock);
222 XFS_STATS_INC(xs_ig_frecycle);
227 inode_vp = XFS_ITOV_NULL(ip);
228 if (inode_vp == NULL) {
230 * If IRECLAIM is set this inode is
231 * on its way out of the system,
232 * we need to pause and try again.
234 if (ip->i_flags & XFS_IRECLAIM) {
235 read_unlock(&ih->ih_lock);
237 XFS_STATS_INC(xs_ig_frecycle);
242 vn_trace_exit(vp, "xfs_iget.alloc",
243 (inst_t *)__return_address);
245 XFS_STATS_INC(xs_ig_found);
247 ip->i_flags &= ~XFS_IRECLAIMABLE;
248 version = ih->ih_version;
249 read_unlock(&ih->ih_lock);
250 xfs_ihash_promote(ih, ip, version);
253 list_del_init(&ip->i_reclaim);
254 XFS_MOUNT_IUNLOCK(mp);
258 } else if (vp != inode_vp) {
259 struct inode *inode = vn_to_inode(inode_vp);
261 /* The inode is being torn down, pause and
264 if (inode->i_state & (I_FREEING | I_CLEAR)) {
265 read_unlock(&ih->ih_lock);
267 XFS_STATS_INC(xs_ig_frecycle);
271 /* Chances are the other vnode (the one in the inode) is being torn
272 * down right now, and we landed on top of it. Question is, what do
273 * we do? Unhook the old inode and hook up the new one?
276 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
281 * Inode cache hit: if ip is not at the front of
282 * its hash chain, move it there now.
283 * Do this with the lock held for update, but
284 * do statistics after releasing the lock.
286 version = ih->ih_version;
287 read_unlock(&ih->ih_lock);
288 xfs_ihash_promote(ih, ip, version);
289 XFS_STATS_INC(xs_ig_found);
292 if (ip->i_d.di_mode == 0) {
293 if (!(flags & IGET_CREATE))
295 xfs_iocore_inode_reinit(ip);
299 xfs_ilock(ip, lock_flags);
301 ip->i_flags &= ~XFS_ISTALE;
303 vn_trace_exit(vp, "xfs_iget.found",
304 (inst_t *)__return_address);
310 * Inode cache miss: save the hash chain version stamp and unlock
311 * the chain, so we don't deadlock in vn_alloc.
313 XFS_STATS_INC(xs_ig_missed);
315 version = ih->ih_version;
317 read_unlock(&ih->ih_lock);
320 * Read the disk inode attributes into a new inode structure and get
321 * a new vnode for it. This should also initialize i_ino and i_mount.
323 error = xfs_iread(mp, tp, ino, &ip, bno);
328 vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address);
330 xfs_inode_lock_init(ip, vp);
331 xfs_iocore_inode_init(ip);
333 if (lock_flags != 0) {
334 xfs_ilock(ip, lock_flags);
337 if ((ip->i_d.di_mode == 0) && !(flags & IGET_CREATE)) {
343 * Put ip on its hash chain, unless someone else hashed a duplicate
344 * after we released the hash lock.
346 write_lock(&ih->ih_lock);
348 if (ih->ih_version != version) {
349 for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) {
350 if (iq->i_ino == ino) {
351 write_unlock(&ih->ih_lock);
354 XFS_STATS_INC(xs_ig_dup);
361 * These values _must_ be set before releasing ihlock!
364 if ((iq = ih->ih_next)) {
365 iq->i_prevp = &ip->i_next;
368 ip->i_prevp = &ih->ih_next;
370 ip->i_udquot = ip->i_gdquot = NULL;
372 ip->i_flags |= XFS_INEW;
374 write_unlock(&ih->ih_lock);
377 * put ip on its cluster's hash chain
379 ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL &&
380 ip->i_cnext == NULL);
383 ch = XFS_CHASH(mp, ip->i_blkno);
385 s = mutex_spinlock(&ch->ch_lock);
386 for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
387 if (chl->chl_blkno == ip->i_blkno) {
389 /* insert this inode into the doubly-linked list
390 * where chl points */
391 if ((iq = chl->chl_ip)) {
392 ip->i_cprev = iq->i_cprev;
393 iq->i_cprev->i_cnext = ip;
406 /* no hash list found for this block; add a new hash list */
408 if (chlnew == NULL) {
409 mutex_spinunlock(&ch->ch_lock, s);
410 ASSERT(xfs_chashlist_zone != NULL);
411 chlnew = (xfs_chashlist_t *)
412 kmem_zone_alloc(xfs_chashlist_zone,
414 ASSERT(chlnew != NULL);
419 ip->i_chash = chlnew;
421 chlnew->chl_blkno = ip->i_blkno;
423 ch->ch_list->chl_prev = chlnew;
424 chlnew->chl_next = ch->ch_list;
425 chlnew->chl_prev = NULL;
426 ch->ch_list = chlnew;
430 if (chlnew != NULL) {
431 kmem_zone_free(xfs_chashlist_zone, chlnew);
435 mutex_spinunlock(&ch->ch_lock, s);
439 * Link ip to its mount and thread it on the mount's inode list.
442 if ((iq = mp->m_inodes)) {
443 ASSERT(iq->i_mprev->i_mnext == iq);
444 ip->i_mprev = iq->i_mprev;
445 iq->i_mprev->i_mnext = ip;
454 XFS_MOUNT_IUNLOCK(mp);
457 ASSERT(ip->i_df.if_ext_max ==
458 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
460 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
461 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
466 * If we have a real type for an on-disk inode, we can set ops(&unlock)
467 * now. If it's a new inode being created, xfs_ialloc will handle it.
469 bhv_vfs_init_vnode(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1);
476 * The 'normal' internal xfs_iget, if needed it will
477 * 'allocate', or 'get', the vnode.
490 bhv_vnode_t *vp = NULL;
493 XFS_STATS_INC(xs_ig_attempts);
496 if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) {
499 vp = vn_from_inode(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 mismatches 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 ((ip = xfs_vtoi(vp)) == NULL)) {
527 xfs_ilock(ip, lock_flags);
528 XFS_STATS_INC(xs_ig_found);
533 error = ENOMEM; /* If we got no inode we are out of memory */
539 * Do the setup for the various locks within the incore inode.
546 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
547 "xfsino", (long)vp->v_number);
548 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", vp->v_number);
549 init_waitqueue_head(&ip->i_ipin_wait);
550 atomic_set(&ip->i_pincount, 0);
551 init_sema(&ip->i_flock, 1, "xfsfino", vp->v_number);
555 * Look for the inode corresponding to the given ino in the hash table.
556 * If it is there and its i_transp pointer matches tp, return it.
557 * Otherwise, return NULL.
560 xfs_inode_incore(xfs_mount_t *mp,
568 ih = XFS_IHASH(mp, ino);
569 read_lock(&ih->ih_lock);
570 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
571 if (ip->i_ino == ino) {
573 * If we find it and tp matches, return it.
574 * Also move it to the front of the hash list
575 * if we find it and it is not already there.
576 * Otherwise break from the loop and return
579 if (ip->i_transp == tp) {
580 version = ih->ih_version;
581 read_unlock(&ih->ih_lock);
582 xfs_ihash_promote(ih, ip, version);
588 read_unlock(&ih->ih_lock);
593 * Decrement reference count of an inode structure and unlock it.
595 * ip -- the inode being released
596 * lock_flags -- this parameter indicates the inode's locks to be
597 * to be released. See the comment on xfs_iunlock() for a list
601 xfs_iput(xfs_inode_t *ip,
604 bhv_vnode_t *vp = XFS_ITOV(ip);
606 vn_trace_entry(vp, "xfs_iput", (inst_t *)__return_address);
607 xfs_iunlock(ip, lock_flags);
612 * Special iput for brand-new inodes that are still locked
615 xfs_iput_new(xfs_inode_t *ip,
618 bhv_vnode_t *vp = XFS_ITOV(ip);
619 struct inode *inode = vn_to_inode(vp);
621 vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address);
623 if ((ip->i_d.di_mode == 0)) {
624 ASSERT(!(ip->i_flags & XFS_IRECLAIMABLE));
627 if (inode->i_state & I_NEW)
628 unlock_new_inode(inode);
630 xfs_iunlock(ip, lock_flags);
636 * This routine embodies the part of the reclaim code that pulls
637 * the inode from the inode hash table and the mount structure's
639 * This should only be called from xfs_reclaim().
642 xfs_ireclaim(xfs_inode_t *ip)
647 * Remove from old hash list and mount list.
649 XFS_STATS_INC(xs_ig_reclaims);
654 * Here we do a spurious inode lock in order to coordinate with
655 * xfs_sync(). This is because xfs_sync() references the inodes
656 * in the mount list without taking references on the corresponding
657 * vnodes. We make that OK here by ensuring that we wait until
658 * the inode is unlocked in xfs_sync() before we go ahead and
659 * free it. We get both the regular lock and the io lock because
660 * the xfs_sync() code may need to drop the regular one but will
661 * still hold the io lock.
663 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
666 * Release dquots (and their references) if any. An inode may escape
667 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
669 XFS_QM_DQDETACH(ip->i_mount, ip);
672 * Pull our behavior descriptor from the vnode chain.
674 vp = XFS_ITOV_NULL(ip);
676 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
680 * Free all memory associated with the inode.
686 * This routine removes an about-to-be-destroyed inode from
687 * all of the lists in which it is located with the exception
688 * of the behavior chain.
698 xfs_chashlist_t *chl, *chm;
702 write_lock(&ih->ih_lock);
703 if ((iq = ip->i_next)) {
704 iq->i_prevp = ip->i_prevp;
708 write_unlock(&ih->ih_lock);
711 * Remove from cluster hash list
712 * 1) delete the chashlist if this is the last inode on the chashlist
713 * 2) unchain from list of inodes
714 * 3) point chashlist->chl_ip to 'chl_next' if to this inode.
717 ch = XFS_CHASH(mp, ip->i_blkno);
718 s = mutex_spinlock(&ch->ch_lock);
720 if (ip->i_cnext == ip) {
721 /* Last inode on chashlist */
722 ASSERT(ip->i_cnext == ip && ip->i_cprev == ip);
723 ASSERT(ip->i_chash != NULL);
727 chl->chl_prev->chl_next = chl->chl_next;
729 ch->ch_list = chl->chl_next;
731 chl->chl_next->chl_prev = chl->chl_prev;
732 kmem_zone_free(xfs_chashlist_zone, chl);
734 /* delete one inode from a non-empty list */
736 iq->i_cprev = ip->i_cprev;
737 ip->i_cprev->i_cnext = iq;
738 if (ip->i_chash->chl_ip == ip) {
739 ip->i_chash->chl_ip = iq;
741 ip->i_chash = __return_address;
742 ip->i_cprev = __return_address;
743 ip->i_cnext = __return_address;
745 mutex_spinunlock(&ch->ch_lock, s);
748 * Remove from mount's inode list.
751 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
753 iq->i_mprev = ip->i_mprev;
754 ip->i_mprev->i_mnext = iq;
757 * Fix up the head pointer if it points to the inode being deleted.
759 if (mp->m_inodes == ip) {
767 /* Deal with the deleted inodes list */
768 list_del_init(&ip->i_reclaim);
771 XFS_MOUNT_IUNLOCK(mp);
775 * This is a wrapper routine around the xfs_ilock() routine
776 * used to centralize some grungy code. It is used in places
777 * that wish to lock the inode solely for reading the extents.
778 * The reason these places can't just call xfs_ilock(SHARED)
779 * is that the inode lock also guards to bringing in of the
780 * extents from disk for a file in b-tree format. If the inode
781 * is in b-tree format, then we need to lock the inode exclusively
782 * until the extents are read in. Locking it exclusively all
783 * the time would limit our parallelism unnecessarily, though.
784 * What we do instead is check to see if the extents have been
785 * read in yet, and only lock the inode exclusively if they
788 * The function returns a value which should be given to the
789 * corresponding xfs_iunlock_map_shared(). This value is
790 * the mode in which the lock was actually taken.
793 xfs_ilock_map_shared(
798 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
799 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
800 lock_mode = XFS_ILOCK_EXCL;
802 lock_mode = XFS_ILOCK_SHARED;
805 xfs_ilock(ip, lock_mode);
811 * This is simply the unlock routine to go with xfs_ilock_map_shared().
812 * All it does is call xfs_iunlock() with the given lock_mode.
815 xfs_iunlock_map_shared(
817 unsigned int lock_mode)
819 xfs_iunlock(ip, lock_mode);
823 * The xfs inode contains 2 locks: a multi-reader lock called the
824 * i_iolock and a multi-reader lock called the i_lock. This routine
825 * allows either or both of the locks to be obtained.
827 * The 2 locks should always be ordered so that the IO lock is
828 * obtained first in order to prevent deadlock.
830 * ip -- the inode being locked
831 * lock_flags -- this parameter indicates the inode's locks
832 * to be locked. It can be:
837 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
838 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
839 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
840 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
843 xfs_ilock(xfs_inode_t *ip,
847 * You can't set both SHARED and EXCL for the same lock,
848 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
849 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
851 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
852 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
853 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
854 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
855 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
857 if (lock_flags & XFS_IOLOCK_EXCL) {
858 mrupdate(&ip->i_iolock);
859 } else if (lock_flags & XFS_IOLOCK_SHARED) {
860 mraccess(&ip->i_iolock);
862 if (lock_flags & XFS_ILOCK_EXCL) {
863 mrupdate(&ip->i_lock);
864 } else if (lock_flags & XFS_ILOCK_SHARED) {
865 mraccess(&ip->i_lock);
867 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
871 * This is just like xfs_ilock(), except that the caller
872 * is guaranteed not to sleep. It returns 1 if it gets
873 * the requested locks and 0 otherwise. If the IO lock is
874 * obtained but the inode lock cannot be, then the IO lock
875 * is dropped before returning.
877 * ip -- the inode being locked
878 * lock_flags -- this parameter indicates the inode's locks to be
879 * to be locked. See the comment for xfs_ilock() for a list
884 xfs_ilock_nowait(xfs_inode_t *ip,
891 * You can't set both SHARED and EXCL for the same lock,
892 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
893 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
895 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
896 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
897 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
898 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
899 ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0);
902 if (lock_flags & XFS_IOLOCK_EXCL) {
903 iolocked = mrtryupdate(&ip->i_iolock);
907 } else if (lock_flags & XFS_IOLOCK_SHARED) {
908 iolocked = mrtryaccess(&ip->i_iolock);
913 if (lock_flags & XFS_ILOCK_EXCL) {
914 ilocked = mrtryupdate(&ip->i_lock);
917 mrunlock(&ip->i_iolock);
921 } else if (lock_flags & XFS_ILOCK_SHARED) {
922 ilocked = mrtryaccess(&ip->i_lock);
925 mrunlock(&ip->i_iolock);
930 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
935 * xfs_iunlock() is used to drop the inode locks acquired with
936 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
937 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
938 * that we know which locks to drop.
940 * ip -- the inode being unlocked
941 * lock_flags -- this parameter indicates the inode's locks to be
942 * to be unlocked. See the comment for xfs_ilock() for a list
943 * of valid values for this parameter.
947 xfs_iunlock(xfs_inode_t *ip,
951 * You can't set both SHARED and EXCL for the same lock,
952 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
953 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
955 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
956 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
957 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
958 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
959 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY)) == 0);
960 ASSERT(lock_flags != 0);
962 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
963 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
964 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
965 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
966 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
967 mrunlock(&ip->i_iolock);
970 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
971 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
972 (ismrlocked(&ip->i_lock, MR_ACCESS)));
973 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
974 (ismrlocked(&ip->i_lock, MR_UPDATE)));
975 mrunlock(&ip->i_lock);
978 * Let the AIL know that this item has been unlocked in case
979 * it is in the AIL and anyone is waiting on it. Don't do
980 * this if the caller has asked us not to.
982 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
983 ip->i_itemp != NULL) {
984 xfs_trans_unlocked_item(ip->i_mount,
985 (xfs_log_item_t*)(ip->i_itemp));
988 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
992 * give up write locks. the i/o lock cannot be held nested
993 * if it is being demoted.
996 xfs_ilock_demote(xfs_inode_t *ip,
999 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
1000 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
1002 if (lock_flags & XFS_ILOCK_EXCL) {
1003 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
1004 mrdemote(&ip->i_lock);
1006 if (lock_flags & XFS_IOLOCK_EXCL) {
1007 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1008 mrdemote(&ip->i_iolock);
1013 * The following three routines simply manage the i_flock
1014 * semaphore embedded in the inode. This semaphore synchronizes
1015 * processes attempting to flush the in-core inode back to disk.
1018 xfs_iflock(xfs_inode_t *ip)
1020 psema(&(ip->i_flock), PINOD|PLTWAIT);
1024 xfs_iflock_nowait(xfs_inode_t *ip)
1026 return (cpsema(&(ip->i_flock)));
1030 xfs_ifunlock(xfs_inode_t *ip)
1032 ASSERT(issemalocked(&(ip->i_flock)));
1033 vsema(&(ip->i_flock));