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 (ip
->i_flags
& 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 (ip
->i_flags
& XFS_IRECLAIM
) {
234 read_unlock(&ih
->ih_lock
);
236 XFS_STATS_INC(xs_ig_frecycle
);
241 vn_trace_exit(vp
, "xfs_iget.alloc",
242 (inst_t
*)__return_address
);
244 XFS_STATS_INC(xs_ig_found
);
246 ip
->i_flags
&= ~XFS_IRECLAIMABLE
;
247 version
= ih
->ih_version
;
248 read_unlock(&ih
->ih_lock
);
249 xfs_ihash_promote(ih
, ip
, version
);
252 list_del_init(&ip
->i_reclaim
);
253 XFS_MOUNT_IUNLOCK(mp
);
257 } else if (vp
!= inode_vp
) {
258 struct inode
*inode
= vn_to_inode(inode_vp
);
260 /* The inode is being torn down, pause and
263 if (inode
->i_state
& (I_FREEING
| I_CLEAR
)) {
264 read_unlock(&ih
->ih_lock
);
266 XFS_STATS_INC(xs_ig_frecycle
);
270 /* Chances are the other vnode (the one in the inode) is being torn
271 * down right now, and we landed on top of it. Question is, what do
272 * we do? Unhook the old inode and hook up the new one?
275 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
280 * Inode cache hit: if ip is not at the front of
281 * its hash chain, move it there now.
282 * Do this with the lock held for update, but
283 * do statistics after releasing the lock.
285 version
= ih
->ih_version
;
286 read_unlock(&ih
->ih_lock
);
287 xfs_ihash_promote(ih
, ip
, version
);
288 XFS_STATS_INC(xs_ig_found
);
291 if (ip
->i_d
.di_mode
== 0) {
292 if (!(flags
& XFS_IGET_CREATE
))
294 xfs_iocore_inode_reinit(ip
);
298 xfs_ilock(ip
, lock_flags
);
300 ip
->i_flags
&= ~XFS_ISTALE
;
302 vn_trace_exit(vp
, "xfs_iget.found",
303 (inst_t
*)__return_address
);
309 * Inode cache miss: save the hash chain version stamp and unlock
310 * the chain, so we don't deadlock in vn_alloc.
312 XFS_STATS_INC(xs_ig_missed
);
314 version
= ih
->ih_version
;
316 read_unlock(&ih
->ih_lock
);
319 * Read the disk inode attributes into a new inode structure and get
320 * a new vnode for it. This should also initialize i_ino and i_mount.
322 error
= xfs_iread(mp
, tp
, ino
, &ip
, bno
,
323 (flags
& XFS_IGET_BULKSTAT
) ? XFS_IMAP_BULKSTAT
: 0);
327 vn_trace_exit(vp
, "xfs_iget.alloc", (inst_t
*)__return_address
);
329 xfs_inode_lock_init(ip
, vp
);
330 xfs_iocore_inode_init(ip
);
333 xfs_ilock(ip
, lock_flags
);
335 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
341 * Put ip on its hash chain, unless someone else hashed a duplicate
342 * after we released the hash lock.
344 write_lock(&ih
->ih_lock
);
346 if (ih
->ih_version
!= version
) {
347 for (iq
= ih
->ih_next
; iq
!= NULL
; iq
= iq
->i_next
) {
348 if (iq
->i_ino
== ino
) {
349 write_unlock(&ih
->ih_lock
);
352 XFS_STATS_INC(xs_ig_dup
);
359 * These values _must_ be set before releasing ihlock!
362 if ((iq
= ih
->ih_next
)) {
363 iq
->i_prevp
= &ip
->i_next
;
366 ip
->i_prevp
= &ih
->ih_next
;
368 ip
->i_udquot
= ip
->i_gdquot
= NULL
;
370 ip
->i_flags
|= XFS_INEW
;
372 write_unlock(&ih
->ih_lock
);
375 * put ip on its cluster's hash chain
377 ASSERT(ip
->i_chash
== NULL
&& ip
->i_cprev
== NULL
&&
378 ip
->i_cnext
== NULL
);
381 ch
= XFS_CHASH(mp
, ip
->i_blkno
);
383 s
= mutex_spinlock(&ch
->ch_lock
);
384 for (chl
= ch
->ch_list
; chl
!= NULL
; chl
= chl
->chl_next
) {
385 if (chl
->chl_blkno
== ip
->i_blkno
) {
387 /* insert this inode into the doubly-linked list
388 * where chl points */
389 if ((iq
= chl
->chl_ip
)) {
390 ip
->i_cprev
= iq
->i_cprev
;
391 iq
->i_cprev
->i_cnext
= ip
;
404 /* no hash list found for this block; add a new hash list */
406 if (chlnew
== NULL
) {
407 mutex_spinunlock(&ch
->ch_lock
, s
);
408 ASSERT(xfs_chashlist_zone
!= NULL
);
409 chlnew
= (xfs_chashlist_t
*)
410 kmem_zone_alloc(xfs_chashlist_zone
,
412 ASSERT(chlnew
!= NULL
);
417 ip
->i_chash
= chlnew
;
419 chlnew
->chl_blkno
= ip
->i_blkno
;
421 ch
->ch_list
->chl_prev
= chlnew
;
422 chlnew
->chl_next
= ch
->ch_list
;
423 chlnew
->chl_prev
= NULL
;
424 ch
->ch_list
= chlnew
;
428 if (chlnew
!= NULL
) {
429 kmem_zone_free(xfs_chashlist_zone
, chlnew
);
433 mutex_spinunlock(&ch
->ch_lock
, s
);
437 * Link ip to its mount and thread it on the mount's inode list.
440 if ((iq
= mp
->m_inodes
)) {
441 ASSERT(iq
->i_mprev
->i_mnext
== iq
);
442 ip
->i_mprev
= iq
->i_mprev
;
443 iq
->i_mprev
->i_mnext
= ip
;
452 XFS_MOUNT_IUNLOCK(mp
);
455 ASSERT(ip
->i_df
.if_ext_max
==
456 XFS_IFORK_DSIZE(ip
) / sizeof(xfs_bmbt_rec_t
));
458 ASSERT(((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) != 0) ==
459 ((ip
->i_iocore
.io_flags
& XFS_IOCORE_RT
) != 0));
464 * If we have a real type for an on-disk inode, we can set ops(&unlock)
465 * now. If it's a new inode being created, xfs_ialloc will handle it.
467 bhv_vfs_init_vnode(XFS_MTOVFS(mp
), vp
, XFS_ITOBHV(ip
), 1);
474 * The 'normal' internal xfs_iget, if needed it will
475 * 'allocate', or 'get', the vnode.
488 bhv_vnode_t
*vp
= NULL
;
491 XFS_STATS_INC(xs_ig_attempts
);
494 if ((inode
= iget_locked(XFS_MTOVFS(mp
)->vfs_super
, ino
))) {
497 vp
= vn_from_inode(inode
);
498 if (inode
->i_state
& I_NEW
) {
499 vn_initialize(inode
);
500 error
= xfs_iget_core(vp
, mp
, tp
, ino
, flags
,
501 lock_flags
, ipp
, bno
);
504 if (inode
->i_state
& I_NEW
)
505 unlock_new_inode(inode
);
510 * If the inode is not fully constructed due to
511 * filehandle mismatches wait for the inode to go
512 * away and try again.
514 * iget_locked will call __wait_on_freeing_inode
515 * to wait for the inode to go away.
517 if (is_bad_inode(inode
) ||
518 ((ip
= xfs_vtoi(vp
)) == NULL
)) {
525 xfs_ilock(ip
, lock_flags
);
526 XFS_STATS_INC(xs_ig_found
);
531 error
= ENOMEM
; /* If we got no inode we are out of memory */
537 * Do the setup for the various locks within the incore inode.
544 mrlock_init(&ip
->i_lock
, MRLOCK_ALLOW_EQUAL_PRI
|MRLOCK_BARRIER
,
545 "xfsino", (long)vp
->v_number
);
546 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", vp
->v_number
);
547 init_waitqueue_head(&ip
->i_ipin_wait
);
548 atomic_set(&ip
->i_pincount
, 0);
549 initnsema(&ip
->i_flock
, 1, "xfsfino");
553 * Look for the inode corresponding to the given ino in the hash table.
554 * If it is there and its i_transp pointer matches tp, return it.
555 * Otherwise, return NULL.
558 xfs_inode_incore(xfs_mount_t
*mp
,
566 ih
= XFS_IHASH(mp
, ino
);
567 read_lock(&ih
->ih_lock
);
568 for (ip
= ih
->ih_next
; ip
!= NULL
; ip
= ip
->i_next
) {
569 if (ip
->i_ino
== ino
) {
571 * If we find it and tp matches, return it.
572 * Also move it to the front of the hash list
573 * if we find it and it is not already there.
574 * Otherwise break from the loop and return
577 if (ip
->i_transp
== tp
) {
578 version
= ih
->ih_version
;
579 read_unlock(&ih
->ih_lock
);
580 xfs_ihash_promote(ih
, ip
, version
);
586 read_unlock(&ih
->ih_lock
);
591 * Decrement reference count of an inode structure and unlock it.
593 * ip -- the inode being released
594 * lock_flags -- this parameter indicates the inode's locks to be
595 * to be released. See the comment on xfs_iunlock() for a list
599 xfs_iput(xfs_inode_t
*ip
,
602 bhv_vnode_t
*vp
= XFS_ITOV(ip
);
604 vn_trace_entry(vp
, "xfs_iput", (inst_t
*)__return_address
);
605 xfs_iunlock(ip
, lock_flags
);
610 * Special iput for brand-new inodes that are still locked
613 xfs_iput_new(xfs_inode_t
*ip
,
616 bhv_vnode_t
*vp
= XFS_ITOV(ip
);
617 struct inode
*inode
= vn_to_inode(vp
);
619 vn_trace_entry(vp
, "xfs_iput_new", (inst_t
*)__return_address
);
621 if ((ip
->i_d
.di_mode
== 0)) {
622 ASSERT(!(ip
->i_flags
& XFS_IRECLAIMABLE
));
625 if (inode
->i_state
& I_NEW
)
626 unlock_new_inode(inode
);
628 xfs_iunlock(ip
, lock_flags
);
634 * This routine embodies the part of the reclaim code that pulls
635 * the inode from the inode hash table and the mount structure's
637 * This should only be called from xfs_reclaim().
640 xfs_ireclaim(xfs_inode_t
*ip
)
645 * Remove from old hash list and mount list.
647 XFS_STATS_INC(xs_ig_reclaims
);
652 * Here we do a spurious inode lock in order to coordinate with
653 * xfs_sync(). This is because xfs_sync() references the inodes
654 * in the mount list without taking references on the corresponding
655 * vnodes. We make that OK here by ensuring that we wait until
656 * the inode is unlocked in xfs_sync() before we go ahead and
657 * free it. We get both the regular lock and the io lock because
658 * the xfs_sync() code may need to drop the regular one but will
659 * still hold the io lock.
661 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
664 * Release dquots (and their references) if any. An inode may escape
665 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
667 XFS_QM_DQDETACH(ip
->i_mount
, ip
);
670 * Pull our behavior descriptor from the vnode chain.
672 vp
= XFS_ITOV_NULL(ip
);
674 vn_bhv_remove(VN_BHV_HEAD(vp
), XFS_ITOBHV(ip
));
678 * Free all memory associated with the inode.
684 * This routine removes an about-to-be-destroyed inode from
685 * all of the lists in which it is located with the exception
686 * of the behavior chain.
696 xfs_chashlist_t
*chl
, *chm
;
700 write_lock(&ih
->ih_lock
);
701 if ((iq
= ip
->i_next
)) {
702 iq
->i_prevp
= ip
->i_prevp
;
706 write_unlock(&ih
->ih_lock
);
709 * Remove from cluster hash list
710 * 1) delete the chashlist if this is the last inode on the chashlist
711 * 2) unchain from list of inodes
712 * 3) point chashlist->chl_ip to 'chl_next' if to this inode.
715 ch
= XFS_CHASH(mp
, ip
->i_blkno
);
716 s
= mutex_spinlock(&ch
->ch_lock
);
718 if (ip
->i_cnext
== ip
) {
719 /* Last inode on chashlist */
720 ASSERT(ip
->i_cnext
== ip
&& ip
->i_cprev
== ip
);
721 ASSERT(ip
->i_chash
!= NULL
);
725 chl
->chl_prev
->chl_next
= chl
->chl_next
;
727 ch
->ch_list
= chl
->chl_next
;
729 chl
->chl_next
->chl_prev
= chl
->chl_prev
;
730 kmem_zone_free(xfs_chashlist_zone
, chl
);
732 /* delete one inode from a non-empty list */
734 iq
->i_cprev
= ip
->i_cprev
;
735 ip
->i_cprev
->i_cnext
= iq
;
736 if (ip
->i_chash
->chl_ip
== ip
) {
737 ip
->i_chash
->chl_ip
= iq
;
739 ip
->i_chash
= __return_address
;
740 ip
->i_cprev
= __return_address
;
741 ip
->i_cnext
= __return_address
;
743 mutex_spinunlock(&ch
->ch_lock
, s
);
746 * Remove from mount's inode list.
749 ASSERT((ip
->i_mnext
!= NULL
) && (ip
->i_mprev
!= NULL
));
751 iq
->i_mprev
= ip
->i_mprev
;
752 ip
->i_mprev
->i_mnext
= iq
;
755 * Fix up the head pointer if it points to the inode being deleted.
757 if (mp
->m_inodes
== ip
) {
765 /* Deal with the deleted inodes list */
766 list_del_init(&ip
->i_reclaim
);
769 XFS_MOUNT_IUNLOCK(mp
);
773 * This is a wrapper routine around the xfs_ilock() routine
774 * used to centralize some grungy code. It is used in places
775 * that wish to lock the inode solely for reading the extents.
776 * The reason these places can't just call xfs_ilock(SHARED)
777 * is that the inode lock also guards to bringing in of the
778 * extents from disk for a file in b-tree format. If the inode
779 * is in b-tree format, then we need to lock the inode exclusively
780 * until the extents are read in. Locking it exclusively all
781 * the time would limit our parallelism unnecessarily, though.
782 * What we do instead is check to see if the extents have been
783 * read in yet, and only lock the inode exclusively if they
786 * The function returns a value which should be given to the
787 * corresponding xfs_iunlock_map_shared(). This value is
788 * the mode in which the lock was actually taken.
791 xfs_ilock_map_shared(
796 if ((ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) &&
797 ((ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)) {
798 lock_mode
= XFS_ILOCK_EXCL
;
800 lock_mode
= XFS_ILOCK_SHARED
;
803 xfs_ilock(ip
, lock_mode
);
809 * This is simply the unlock routine to go with xfs_ilock_map_shared().
810 * All it does is call xfs_iunlock() with the given lock_mode.
813 xfs_iunlock_map_shared(
815 unsigned int lock_mode
)
817 xfs_iunlock(ip
, lock_mode
);
821 * The xfs inode contains 2 locks: a multi-reader lock called the
822 * i_iolock and a multi-reader lock called the i_lock. This routine
823 * allows either or both of the locks to be obtained.
825 * The 2 locks should always be ordered so that the IO lock is
826 * obtained first in order to prevent deadlock.
828 * ip -- the inode being locked
829 * lock_flags -- this parameter indicates the inode's locks
830 * to be locked. It can be:
835 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
836 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
837 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
838 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
841 xfs_ilock(xfs_inode_t
*ip
,
845 * You can't set both SHARED and EXCL for the same lock,
846 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
847 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
849 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
850 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
851 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
852 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
853 ASSERT((lock_flags
& ~XFS_LOCK_MASK
) == 0);
855 if (lock_flags
& XFS_IOLOCK_EXCL
) {
856 mrupdate(&ip
->i_iolock
);
857 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
858 mraccess(&ip
->i_iolock
);
860 if (lock_flags
& XFS_ILOCK_EXCL
) {
861 mrupdate(&ip
->i_lock
);
862 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
863 mraccess(&ip
->i_lock
);
865 xfs_ilock_trace(ip
, 1, lock_flags
, (inst_t
*)__return_address
);
869 * This is just like xfs_ilock(), except that the caller
870 * is guaranteed not to sleep. It returns 1 if it gets
871 * the requested locks and 0 otherwise. If the IO lock is
872 * obtained but the inode lock cannot be, then the IO lock
873 * is dropped before returning.
875 * ip -- the inode being locked
876 * lock_flags -- this parameter indicates the inode's locks to be
877 * to be locked. See the comment for xfs_ilock() for a list
882 xfs_ilock_nowait(xfs_inode_t
*ip
,
889 * You can't set both SHARED and EXCL for the same lock,
890 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
891 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
893 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
894 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
895 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
896 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
897 ASSERT((lock_flags
& ~XFS_LOCK_MASK
) == 0);
900 if (lock_flags
& XFS_IOLOCK_EXCL
) {
901 iolocked
= mrtryupdate(&ip
->i_iolock
);
905 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
906 iolocked
= mrtryaccess(&ip
->i_iolock
);
911 if (lock_flags
& XFS_ILOCK_EXCL
) {
912 ilocked
= mrtryupdate(&ip
->i_lock
);
915 mrunlock(&ip
->i_iolock
);
919 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
920 ilocked
= mrtryaccess(&ip
->i_lock
);
923 mrunlock(&ip
->i_iolock
);
928 xfs_ilock_trace(ip
, 2, lock_flags
, (inst_t
*)__return_address
);
933 * xfs_iunlock() is used to drop the inode locks acquired with
934 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
935 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
936 * that we know which locks to drop.
938 * ip -- the inode being unlocked
939 * lock_flags -- this parameter indicates the inode's locks to be
940 * to be unlocked. See the comment for xfs_ilock() for a list
941 * of valid values for this parameter.
945 xfs_iunlock(xfs_inode_t
*ip
,
949 * You can't set both SHARED and EXCL for the same lock,
950 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
951 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
953 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
954 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
955 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
956 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
957 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_IUNLOCK_NONOTIFY
)) == 0);
958 ASSERT(lock_flags
!= 0);
960 if (lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) {
961 ASSERT(!(lock_flags
& XFS_IOLOCK_SHARED
) ||
962 (ismrlocked(&ip
->i_iolock
, MR_ACCESS
)));
963 ASSERT(!(lock_flags
& XFS_IOLOCK_EXCL
) ||
964 (ismrlocked(&ip
->i_iolock
, MR_UPDATE
)));
965 mrunlock(&ip
->i_iolock
);
968 if (lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) {
969 ASSERT(!(lock_flags
& XFS_ILOCK_SHARED
) ||
970 (ismrlocked(&ip
->i_lock
, MR_ACCESS
)));
971 ASSERT(!(lock_flags
& XFS_ILOCK_EXCL
) ||
972 (ismrlocked(&ip
->i_lock
, MR_UPDATE
)));
973 mrunlock(&ip
->i_lock
);
976 * Let the AIL know that this item has been unlocked in case
977 * it is in the AIL and anyone is waiting on it. Don't do
978 * this if the caller has asked us not to.
980 if (!(lock_flags
& XFS_IUNLOCK_NONOTIFY
) &&
981 ip
->i_itemp
!= NULL
) {
982 xfs_trans_unlocked_item(ip
->i_mount
,
983 (xfs_log_item_t
*)(ip
->i_itemp
));
986 xfs_ilock_trace(ip
, 3, lock_flags
, (inst_t
*)__return_address
);
990 * give up write locks. the i/o lock cannot be held nested
991 * if it is being demoted.
994 xfs_ilock_demote(xfs_inode_t
*ip
,
997 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
998 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
1000 if (lock_flags
& XFS_ILOCK_EXCL
) {
1001 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
1002 mrdemote(&ip
->i_lock
);
1004 if (lock_flags
& XFS_IOLOCK_EXCL
) {
1005 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
));
1006 mrdemote(&ip
->i_iolock
);
1011 * The following three routines simply manage the i_flock
1012 * semaphore embedded in the inode. This semaphore synchronizes
1013 * processes attempting to flush the in-core inode back to disk.
1016 xfs_iflock(xfs_inode_t
*ip
)
1018 psema(&(ip
->i_flock
), PINOD
|PLTWAIT
);
1022 xfs_iflock_nowait(xfs_inode_t
*ip
)
1024 return (cpsema(&(ip
->i_flock
)));
1028 xfs_ifunlock(xfs_inode_t
*ip
)
1030 ASSERT(issemalocked(&(ip
->i_flock
)));
1031 vsema(&(ip
->i_flock
));