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)
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 mp->m_ihash = kmem_zalloc_greedy(&mp->m_ihsize,
65 NBPC * sizeof(xfs_ihash_t),
66 mp->m_ihsize * sizeof(xfs_ihash_t),
67 KM_SLEEP | KM_MAYFAIL | KM_LARGE);
68 mp->m_ihsize /= sizeof(xfs_ihash_t);
69 for (i = 0; i < mp->m_ihsize; i++)
70 rwlock_init(&(mp->m_ihash[i].ih_lock));
74 * Free up structures allocated by xfs_ihash_init, at unmount time.
77 xfs_ihash_free(xfs_mount_t *mp)
79 kmem_free(mp->m_ihash, mp->m_ihsize * sizeof(xfs_ihash_t));
84 * Initialize the inode cluster hash table for the newly mounted file system.
85 * Its size is derived from the ihash table size.
88 xfs_chash_init(xfs_mount_t *mp)
92 mp->m_chsize = max_t(uint, 1, mp->m_ihsize /
93 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog));
94 mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize);
95 mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize
96 * sizeof(xfs_chash_t),
98 for (i = 0; i < mp->m_chsize; i++) {
99 spinlock_init(&mp->m_chash[i].ch_lock,"xfshash");
104 * Free up structures allocated by xfs_chash_init, at unmount time.
107 xfs_chash_free(xfs_mount_t *mp)
111 for (i = 0; i < mp->m_chsize; i++) {
112 spinlock_destroy(&mp->m_chash[i].ch_lock);
115 kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t));
120 * Try to move an inode to the front of its hash list if possible
121 * (and if its not there already). Called right after obtaining
122 * the list version number and then dropping the read_lock on the
123 * hash list in question (which is done right after looking up the
124 * inode in question...).
134 if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) {
135 if (likely(version == ih->ih_version)) {
136 /* remove from list */
137 if ((iq = ip->i_next)) {
138 iq->i_prevp = ip->i_prevp;
142 /* insert at list head */
144 iq->i_prevp = &ip->i_next;
146 ip->i_prevp = &ih->ih_next;
149 write_unlock(&ih->ih_lock);
154 * Look up an inode by number in the given file system.
155 * The inode is looked up in the hash table for the file system
156 * represented by the mount point parameter mp. Each bucket of
157 * the hash table is guarded by an individual semaphore.
159 * If the inode is found in the hash table, its corresponding vnode
160 * is obtained with a call to vn_get(). This call takes care of
161 * coordination with the reclamation of the inode and vnode. Note
162 * that the vmap structure is filled in while holding the hash lock.
163 * This gives us the state of the inode/vnode when we found it and
164 * is used for coordination in vn_get().
166 * If it is not in core, read it in from the file system's device and
167 * add the inode into the hash table.
169 * The inode is locked according to the value of the lock_flags parameter.
170 * This flag parameter indicates how and if the inode's IO lock and inode lock
173 * mp -- the mount point structure for the current file system. It points
174 * to the inode hash table.
175 * tp -- a pointer to the current transaction if there is one. This is
176 * simply passed through to the xfs_iread() call.
177 * ino -- the number of the inode desired. This is the unique identifier
178 * within the file system for the inode being requested.
179 * lock_flags -- flags indicating how to lock the inode. See the comment
180 * for xfs_ilock() for a list of valid values.
181 * bno -- the block number starting the buffer containing the inode,
182 * if known (as by bulkstat), else 0.
198 bhv_vnode_t *inode_vp;
203 xfs_chashlist_t *chl, *chlnew;
207 ih = XFS_IHASH(mp, ino);
210 read_lock(&ih->ih_lock);
212 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
213 if (ip->i_ino == ino) {
215 * If INEW is set this inode is being set up
216 * we need to pause and try again.
218 if (xfs_iflags_test(ip, XFS_INEW)) {
219 read_unlock(&ih->ih_lock);
221 XFS_STATS_INC(xs_ig_frecycle);
226 inode_vp = XFS_ITOV_NULL(ip);
227 if (inode_vp == NULL) {
229 * If IRECLAIM is set this inode is
230 * on its way out of the system,
231 * we need to pause and try again.
233 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
234 read_unlock(&ih->ih_lock);
236 XFS_STATS_INC(xs_ig_frecycle);
240 ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
243 * If lookup is racing with unlink, then we
244 * should return an error immediately so we
245 * don't remove it from the reclaim list and
246 * potentially leak the inode.
248 if ((ip->i_d.di_mode == 0) &&
249 !(flags & XFS_IGET_CREATE)) {
250 read_unlock(&ih->ih_lock);
255 * There may be transactions sitting in the
256 * incore log buffers or being flushed to disk
257 * at this time. We can't clear the
258 * XFS_IRECLAIMABLE flag until these
259 * transactions have hit the disk, otherwise we
260 * will void the guarantee the flag provides
263 if (xfs_ipincount(ip)) {
264 read_unlock(&ih->ih_lock);
266 XFS_LOG_FORCE|XFS_LOG_SYNC);
267 XFS_STATS_INC(xs_ig_frecycle);
271 vn_trace_exit(vp, "xfs_iget.alloc",
272 (inst_t *)__return_address);
274 XFS_STATS_INC(xs_ig_found);
276 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
277 version = ih->ih_version;
278 read_unlock(&ih->ih_lock);
279 xfs_ihash_promote(ih, ip, version);
282 list_del_init(&ip->i_reclaim);
283 XFS_MOUNT_IUNLOCK(mp);
287 } else if (vp != inode_vp) {
288 struct inode *inode = vn_to_inode(inode_vp);
290 /* The inode is being torn down, pause and
293 if (inode->i_state & (I_FREEING | I_CLEAR)) {
294 read_unlock(&ih->ih_lock);
296 XFS_STATS_INC(xs_ig_frecycle);
300 /* Chances are the other vnode (the one in the inode) is being torn
301 * down right now, and we landed on top of it. Question is, what do
302 * we do? Unhook the old inode and hook up the new one?
305 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
310 * Inode cache hit: if ip is not at the front of
311 * its hash chain, move it there now.
312 * Do this with the lock held for update, but
313 * do statistics after releasing the lock.
315 version = ih->ih_version;
316 read_unlock(&ih->ih_lock);
317 xfs_ihash_promote(ih, ip, version);
318 XFS_STATS_INC(xs_ig_found);
321 if (ip->i_d.di_mode == 0) {
322 if (!(flags & XFS_IGET_CREATE))
324 xfs_iocore_inode_reinit(ip);
328 xfs_ilock(ip, lock_flags);
330 xfs_iflags_clear(ip, XFS_ISTALE);
331 vn_trace_exit(vp, "xfs_iget.found",
332 (inst_t *)__return_address);
338 * Inode cache miss: save the hash chain version stamp and unlock
339 * the chain, so we don't deadlock in vn_alloc.
341 XFS_STATS_INC(xs_ig_missed);
343 version = ih->ih_version;
345 read_unlock(&ih->ih_lock);
348 * Read the disk inode attributes into a new inode structure and get
349 * a new vnode for it. This should also initialize i_ino and i_mount.
351 error = xfs_iread(mp, tp, ino, &ip, bno,
352 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
356 vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address);
358 xfs_inode_lock_init(ip, vp);
359 xfs_iocore_inode_init(ip);
362 xfs_ilock(ip, lock_flags);
364 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
370 * Put ip on its hash chain, unless someone else hashed a duplicate
371 * after we released the hash lock.
373 write_lock(&ih->ih_lock);
375 if (ih->ih_version != version) {
376 for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) {
377 if (iq->i_ino == ino) {
378 write_unlock(&ih->ih_lock);
381 XFS_STATS_INC(xs_ig_dup);
388 * These values _must_ be set before releasing ihlock!
391 if ((iq = ih->ih_next)) {
392 iq->i_prevp = &ip->i_next;
395 ip->i_prevp = &ih->ih_next;
397 ip->i_udquot = ip->i_gdquot = NULL;
399 xfs_iflags_set(ip, XFS_INEW);
400 write_unlock(&ih->ih_lock);
403 * put ip on its cluster's hash chain
405 ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL &&
406 ip->i_cnext == NULL);
409 ch = XFS_CHASH(mp, ip->i_blkno);
411 s = mutex_spinlock(&ch->ch_lock);
412 for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
413 if (chl->chl_blkno == ip->i_blkno) {
415 /* insert this inode into the doubly-linked list
416 * where chl points */
417 if ((iq = chl->chl_ip)) {
418 ip->i_cprev = iq->i_cprev;
419 iq->i_cprev->i_cnext = ip;
432 /* no hash list found for this block; add a new hash list */
434 if (chlnew == NULL) {
435 mutex_spinunlock(&ch->ch_lock, s);
436 ASSERT(xfs_chashlist_zone != NULL);
437 chlnew = (xfs_chashlist_t *)
438 kmem_zone_alloc(xfs_chashlist_zone,
440 ASSERT(chlnew != NULL);
445 ip->i_chash = chlnew;
447 chlnew->chl_blkno = ip->i_blkno;
449 ch->ch_list->chl_prev = chlnew;
450 chlnew->chl_next = ch->ch_list;
451 chlnew->chl_prev = NULL;
452 ch->ch_list = chlnew;
456 if (chlnew != NULL) {
457 kmem_zone_free(xfs_chashlist_zone, chlnew);
461 mutex_spinunlock(&ch->ch_lock, s);
465 * Link ip to its mount and thread it on the mount's inode list.
468 if ((iq = mp->m_inodes)) {
469 ASSERT(iq->i_mprev->i_mnext == iq);
470 ip->i_mprev = iq->i_mprev;
471 iq->i_mprev->i_mnext = ip;
480 XFS_MOUNT_IUNLOCK(mp);
483 ASSERT(ip->i_df.if_ext_max ==
484 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
486 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
487 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
492 * If we have a real type for an on-disk inode, we can set ops(&unlock)
493 * now. If it's a new inode being created, xfs_ialloc will handle it.
495 bhv_vfs_init_vnode(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1);
502 * The 'normal' internal xfs_iget, if needed it will
503 * 'allocate', or 'get', the vnode.
516 bhv_vnode_t *vp = NULL;
519 XFS_STATS_INC(xs_ig_attempts);
522 if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) {
525 vp = vn_from_inode(inode);
526 if (inode->i_state & I_NEW) {
527 vn_initialize(inode);
528 error = xfs_iget_core(vp, mp, tp, ino, flags,
529 lock_flags, ipp, bno);
532 if (inode->i_state & I_NEW)
533 unlock_new_inode(inode);
538 * If the inode is not fully constructed due to
539 * filehandle mismatches wait for the inode to go
540 * away and try again.
542 * iget_locked will call __wait_on_freeing_inode
543 * to wait for the inode to go away.
545 if (is_bad_inode(inode) ||
546 ((ip = xfs_vtoi(vp)) == NULL)) {
553 xfs_ilock(ip, lock_flags);
554 XFS_STATS_INC(xs_ig_found);
559 error = ENOMEM; /* If we got no inode we are out of memory */
565 * Do the setup for the various locks within the incore inode.
572 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
573 "xfsino", (long)vp->v_number);
574 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", vp->v_number);
575 init_waitqueue_head(&ip->i_ipin_wait);
576 atomic_set(&ip->i_pincount, 0);
577 initnsema(&ip->i_flock, 1, "xfsfino");
581 * Look for the inode corresponding to the given ino in the hash table.
582 * If it is there and its i_transp pointer matches tp, return it.
583 * Otherwise, return NULL.
586 xfs_inode_incore(xfs_mount_t *mp,
594 ih = XFS_IHASH(mp, ino);
595 read_lock(&ih->ih_lock);
596 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
597 if (ip->i_ino == ino) {
599 * If we find it and tp matches, return it.
600 * Also move it to the front of the hash list
601 * if we find it and it is not already there.
602 * Otherwise break from the loop and return
605 if (ip->i_transp == tp) {
606 version = ih->ih_version;
607 read_unlock(&ih->ih_lock);
608 xfs_ihash_promote(ih, ip, version);
614 read_unlock(&ih->ih_lock);
619 * Decrement reference count of an inode structure and unlock it.
621 * ip -- the inode being released
622 * lock_flags -- this parameter indicates the inode's locks to be
623 * to be released. See the comment on xfs_iunlock() for a list
627 xfs_iput(xfs_inode_t *ip,
630 bhv_vnode_t *vp = XFS_ITOV(ip);
632 vn_trace_entry(vp, "xfs_iput", (inst_t *)__return_address);
633 xfs_iunlock(ip, lock_flags);
638 * Special iput for brand-new inodes that are still locked
641 xfs_iput_new(xfs_inode_t *ip,
644 bhv_vnode_t *vp = XFS_ITOV(ip);
645 struct inode *inode = vn_to_inode(vp);
647 vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address);
649 if ((ip->i_d.di_mode == 0)) {
650 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
653 if (inode->i_state & I_NEW)
654 unlock_new_inode(inode);
656 xfs_iunlock(ip, lock_flags);
662 * This routine embodies the part of the reclaim code that pulls
663 * the inode from the inode hash table and the mount structure's
665 * This should only be called from xfs_reclaim().
668 xfs_ireclaim(xfs_inode_t *ip)
673 * Remove from old hash list and mount list.
675 XFS_STATS_INC(xs_ig_reclaims);
680 * Here we do a spurious inode lock in order to coordinate with
681 * xfs_sync(). This is because xfs_sync() references the inodes
682 * in the mount list without taking references on the corresponding
683 * vnodes. We make that OK here by ensuring that we wait until
684 * the inode is unlocked in xfs_sync() before we go ahead and
685 * free it. We get both the regular lock and the io lock because
686 * the xfs_sync() code may need to drop the regular one but will
687 * still hold the io lock.
689 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
692 * Release dquots (and their references) if any. An inode may escape
693 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
695 XFS_QM_DQDETACH(ip->i_mount, ip);
698 * Pull our behavior descriptor from the vnode chain.
700 vp = XFS_ITOV_NULL(ip);
702 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
706 * Free all memory associated with the inode.
708 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
713 * This routine removes an about-to-be-destroyed inode from
714 * all of the lists in which it is located with the exception
715 * of the behavior chain.
725 xfs_chashlist_t *chl, *chm;
729 write_lock(&ih->ih_lock);
730 if ((iq = ip->i_next)) {
731 iq->i_prevp = ip->i_prevp;
735 write_unlock(&ih->ih_lock);
738 * Remove from cluster hash list
739 * 1) delete the chashlist if this is the last inode on the chashlist
740 * 2) unchain from list of inodes
741 * 3) point chashlist->chl_ip to 'chl_next' if to this inode.
744 ch = XFS_CHASH(mp, ip->i_blkno);
745 s = mutex_spinlock(&ch->ch_lock);
747 if (ip->i_cnext == ip) {
748 /* Last inode on chashlist */
749 ASSERT(ip->i_cnext == ip && ip->i_cprev == ip);
750 ASSERT(ip->i_chash != NULL);
754 chl->chl_prev->chl_next = chl->chl_next;
756 ch->ch_list = chl->chl_next;
758 chl->chl_next->chl_prev = chl->chl_prev;
759 kmem_zone_free(xfs_chashlist_zone, chl);
761 /* delete one inode from a non-empty list */
763 iq->i_cprev = ip->i_cprev;
764 ip->i_cprev->i_cnext = iq;
765 if (ip->i_chash->chl_ip == ip) {
766 ip->i_chash->chl_ip = iq;
768 ip->i_chash = __return_address;
769 ip->i_cprev = __return_address;
770 ip->i_cnext = __return_address;
772 mutex_spinunlock(&ch->ch_lock, s);
775 * Remove from mount's inode list.
778 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
780 iq->i_mprev = ip->i_mprev;
781 ip->i_mprev->i_mnext = iq;
784 * Fix up the head pointer if it points to the inode being deleted.
786 if (mp->m_inodes == ip) {
794 /* Deal with the deleted inodes list */
795 list_del_init(&ip->i_reclaim);
798 XFS_MOUNT_IUNLOCK(mp);
802 * This is a wrapper routine around the xfs_ilock() routine
803 * used to centralize some grungy code. It is used in places
804 * that wish to lock the inode solely for reading the extents.
805 * The reason these places can't just call xfs_ilock(SHARED)
806 * is that the inode lock also guards to bringing in of the
807 * extents from disk for a file in b-tree format. If the inode
808 * is in b-tree format, then we need to lock the inode exclusively
809 * until the extents are read in. Locking it exclusively all
810 * the time would limit our parallelism unnecessarily, though.
811 * What we do instead is check to see if the extents have been
812 * read in yet, and only lock the inode exclusively if they
815 * The function returns a value which should be given to the
816 * corresponding xfs_iunlock_map_shared(). This value is
817 * the mode in which the lock was actually taken.
820 xfs_ilock_map_shared(
825 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
826 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
827 lock_mode = XFS_ILOCK_EXCL;
829 lock_mode = XFS_ILOCK_SHARED;
832 xfs_ilock(ip, lock_mode);
838 * This is simply the unlock routine to go with xfs_ilock_map_shared().
839 * All it does is call xfs_iunlock() with the given lock_mode.
842 xfs_iunlock_map_shared(
844 unsigned int lock_mode)
846 xfs_iunlock(ip, lock_mode);
850 * The xfs inode contains 2 locks: a multi-reader lock called the
851 * i_iolock and a multi-reader lock called the i_lock. This routine
852 * allows either or both of the locks to be obtained.
854 * The 2 locks should always be ordered so that the IO lock is
855 * obtained first in order to prevent deadlock.
857 * ip -- the inode being locked
858 * lock_flags -- this parameter indicates the inode's locks
859 * to be locked. It can be:
864 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
865 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
866 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
867 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
870 xfs_ilock(xfs_inode_t *ip,
874 * You can't set both SHARED and EXCL for the same lock,
875 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
876 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
878 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
879 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
880 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
881 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
882 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
884 if (lock_flags & XFS_IOLOCK_EXCL) {
885 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
886 } else if (lock_flags & XFS_IOLOCK_SHARED) {
887 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
889 if (lock_flags & XFS_ILOCK_EXCL) {
890 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
891 } else if (lock_flags & XFS_ILOCK_SHARED) {
892 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
894 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
898 * This is just like xfs_ilock(), except that the caller
899 * is guaranteed not to sleep. It returns 1 if it gets
900 * the requested locks and 0 otherwise. If the IO lock is
901 * obtained but the inode lock cannot be, then the IO lock
902 * is dropped before returning.
904 * ip -- the inode being locked
905 * lock_flags -- this parameter indicates the inode's locks to be
906 * to be locked. See the comment for xfs_ilock() for a list
911 xfs_ilock_nowait(xfs_inode_t *ip,
918 * You can't set both SHARED and EXCL for the same lock,
919 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
920 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
922 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
923 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
924 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
925 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
926 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
929 if (lock_flags & XFS_IOLOCK_EXCL) {
930 iolocked = mrtryupdate(&ip->i_iolock);
934 } else if (lock_flags & XFS_IOLOCK_SHARED) {
935 iolocked = mrtryaccess(&ip->i_iolock);
940 if (lock_flags & XFS_ILOCK_EXCL) {
941 ilocked = mrtryupdate(&ip->i_lock);
944 mrunlock(&ip->i_iolock);
948 } else if (lock_flags & XFS_ILOCK_SHARED) {
949 ilocked = mrtryaccess(&ip->i_lock);
952 mrunlock(&ip->i_iolock);
957 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
962 * xfs_iunlock() is used to drop the inode locks acquired with
963 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
964 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
965 * that we know which locks to drop.
967 * ip -- the inode being unlocked
968 * lock_flags -- this parameter indicates the inode's locks to be
969 * to be unlocked. See the comment for xfs_ilock() for a list
970 * of valid values for this parameter.
974 xfs_iunlock(xfs_inode_t *ip,
978 * You can't set both SHARED and EXCL for the same lock,
979 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
980 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
982 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
983 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
984 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
985 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
986 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
987 XFS_LOCK_DEP_MASK)) == 0);
988 ASSERT(lock_flags != 0);
990 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
991 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
992 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
993 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
994 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
995 mrunlock(&ip->i_iolock);
998 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
999 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
1000 (ismrlocked(&ip->i_lock, MR_ACCESS)));
1001 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
1002 (ismrlocked(&ip->i_lock, MR_UPDATE)));
1003 mrunlock(&ip->i_lock);
1006 * Let the AIL know that this item has been unlocked in case
1007 * it is in the AIL and anyone is waiting on it. Don't do
1008 * this if the caller has asked us not to.
1010 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
1011 ip->i_itemp != NULL) {
1012 xfs_trans_unlocked_item(ip->i_mount,
1013 (xfs_log_item_t*)(ip->i_itemp));
1016 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
1020 * give up write locks. the i/o lock cannot be held nested
1021 * if it is being demoted.
1024 xfs_ilock_demote(xfs_inode_t *ip,
1027 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
1028 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
1030 if (lock_flags & XFS_ILOCK_EXCL) {
1031 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
1032 mrdemote(&ip->i_lock);
1034 if (lock_flags & XFS_IOLOCK_EXCL) {
1035 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1036 mrdemote(&ip->i_iolock);
1041 * The following three routines simply manage the i_flock
1042 * semaphore embedded in the inode. This semaphore synchronizes
1043 * processes attempting to flush the in-core inode back to disk.
1046 xfs_iflock(xfs_inode_t *ip)
1048 psema(&(ip->i_flock), PINOD|PLTWAIT);
1052 xfs_iflock_nowait(xfs_inode_t *ip)
1054 return (cpsema(&(ip->i_flock)));
1058 xfs_ifunlock(xfs_inode_t *ip)
1060 ASSERT(issemalocked(&(ip->i_flock)));
1061 vsema(&(ip->i_flock));