2 * Copyright (c) 2000-2006 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
21 #include "xfs_types.h"
25 #include "xfs_trans.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_dir_leaf.h"
43 #include "xfs_itable.h"
44 #include "xfs_btree.h"
45 #include "xfs_ialloc.h"
46 #include "xfs_alloc.h"
50 #include "xfs_error.h"
51 #include "xfs_quota.h"
52 #include "xfs_utils.h"
53 #include "xfs_rtalloc.h"
54 #include "xfs_refcache.h"
55 #include "xfs_trans_space.h"
56 #include "xfs_log_priv.h"
65 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
66 xfs_inode_t *ip = XFS_BHVTOI(bdp);
68 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
69 return XFS_ERROR(EIO);
72 * If it's a directory with any blocks, read-ahead block 0
73 * as we're almost certain to have the next operation be a read there.
75 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
76 mode = xfs_ilock_map_shared(ip);
77 if (ip->i_d.di_nextents > 0)
78 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
79 xfs_iunlock(ip, mode);
88 lastclose_t lastclose,
91 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
92 xfs_inode_t *ip = XFS_BHVTOI(bdp);
94 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
95 return XFS_ERROR(EIO);
97 if (lastclose != L_TRUE || !VN_ISREG(vp))
101 * If we previously truncated this file and removed old data in
102 * the process, we want to initiate "early" writeout on the last
103 * close. This is an attempt to combat the notorious NULL files
104 * problem which is particularly noticable from a truncate down,
105 * buffered (re-)write (delalloc), followed by a crash. What we
106 * are effectively doing here is significantly reducing the time
107 * window where we'd otherwise be exposed to that problem.
109 if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
110 return bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE);
128 vp = BHV_TO_VNODE(bdp);
129 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
131 ip = XFS_BHVTOI(bdp);
134 if (XFS_FORCED_SHUTDOWN(mp))
135 return XFS_ERROR(EIO);
137 if (!(flags & ATTR_LAZY))
138 xfs_ilock(ip, XFS_ILOCK_SHARED);
140 vap->va_size = ip->i_d.di_size;
141 if (vap->va_mask == XFS_AT_SIZE)
145 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
146 vap->va_nodeid = ip->i_ino;
148 vap->va_nodeid += mp->m_inoadd;
150 vap->va_nlink = ip->i_d.di_nlink;
153 * Quick exit for non-stat callers
156 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
157 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
161 * Copy from in-core inode.
163 vap->va_mode = ip->i_d.di_mode;
164 vap->va_uid = ip->i_d.di_uid;
165 vap->va_gid = ip->i_d.di_gid;
166 vap->va_projid = ip->i_d.di_projid;
169 * Check vnode type block/char vs. everything else.
171 switch (ip->i_d.di_mode & S_IFMT) {
174 vap->va_rdev = ip->i_df.if_u2.if_rdev;
175 vap->va_blocksize = BLKDEV_IOSIZE;
180 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
181 vap->va_blocksize = xfs_preferred_iosize(mp);
185 * If the file blocks are being allocated from a
186 * realtime partition, then return the inode's
187 * realtime extent size or the realtime volume's
190 vap->va_blocksize = ip->i_d.di_extsize ?
191 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
192 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
197 vn_atime_to_timespec(vp, &vap->va_atime);
198 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
199 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
200 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
201 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
204 * Exit for stat callers. See if any of the rest of the fields
205 * to be filled in are needed.
208 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
209 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
213 * Convert di_flags to xflags.
215 vap->va_xflags = xfs_ip2xflags(ip);
218 * Exit for inode revalidate. See if any of the rest of
219 * the fields to be filled in are needed.
222 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
223 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
226 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
228 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
229 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
233 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
234 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
235 ip->i_d.di_anextents;
237 vap->va_anextents = 0;
238 vap->va_gen = ip->i_d.di_gen;
241 if (!(flags & ATTR_LAZY))
242 xfs_iunlock(ip, XFS_ILOCK_SHARED);
268 xfs_prid_t projid=0, iprojid=0;
269 int mandlock_before, mandlock_after;
270 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
274 vp = BHV_TO_VNODE(bdp);
275 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
277 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
278 return XFS_ERROR(EROFS);
281 * Cannot set certain attributes.
284 if (mask & XFS_AT_NOSET) {
285 return XFS_ERROR(EINVAL);
288 ip = XFS_BHVTOI(bdp);
291 if (XFS_FORCED_SHUTDOWN(mp))
292 return XFS_ERROR(EIO);
295 * Timestamps do not need to be logged and hence do not
296 * need to be done within a transaction.
298 if (mask & XFS_AT_UPDTIMES) {
299 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
300 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
301 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
302 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
303 xfs_ichgtime(ip, timeflags);
307 olddquot1 = olddquot2 = NULL;
311 * If disk quotas is on, we make sure that the dquots do exist on disk,
312 * before we start any other transactions. Trying to do this later
313 * is messy. We don't care to take a readlock to look at the ids
314 * in inode here, because we can't hold it across the trans_reserve.
315 * If the IDs do change before we take the ilock, we're covered
316 * because the i_*dquot fields will get updated anyway.
318 if (XFS_IS_QUOTA_ON(mp) &&
319 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
322 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
324 qflags |= XFS_QMOPT_UQUOTA;
326 uid = ip->i_d.di_uid;
328 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
330 qflags |= XFS_QMOPT_GQUOTA;
332 gid = ip->i_d.di_gid;
334 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
335 projid = vap->va_projid;
336 qflags |= XFS_QMOPT_PQUOTA;
338 projid = ip->i_d.di_projid;
341 * We take a reference when we initialize udqp and gdqp,
342 * so it is important that we never blindly double trip on
343 * the same variable. See xfs_create() for an example.
345 ASSERT(udqp == NULL);
346 ASSERT(gdqp == NULL);
347 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
354 * For the other attributes, we acquire the inode lock and
355 * first do an error checking pass.
358 lock_flags = XFS_ILOCK_EXCL;
359 if (flags & ATTR_NOLOCK)
361 if (!(mask & XFS_AT_SIZE)) {
362 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
363 (mp->m_flags & XFS_MOUNT_WSYNC)) {
364 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
366 if ((code = xfs_trans_reserve(tp, 0,
367 XFS_ICHANGE_LOG_RES(mp), 0,
374 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
375 !(flags & ATTR_DMI)) {
376 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
377 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
378 vap->va_size, 0, dmflags, NULL);
385 lock_flags |= XFS_IOLOCK_EXCL;
388 xfs_ilock(ip, lock_flags);
390 /* boolean: are we the file owner? */
391 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
394 * Change various properties of a file.
395 * Only the owner or users with CAP_FOWNER
396 * capability may do these things.
399 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
400 XFS_AT_GID|XFS_AT_PROJID)) {
402 * CAP_FOWNER overrides the following restrictions:
404 * The user ID of the calling process must be equal
405 * to the file owner ID, except in cases where the
406 * CAP_FSETID capability is applicable.
408 if (!file_owner && !capable(CAP_FOWNER)) {
409 code = XFS_ERROR(EPERM);
414 * CAP_FSETID overrides the following restrictions:
416 * The effective user ID of the calling process shall match
417 * the file owner when setting the set-user-ID and
418 * set-group-ID bits on that file.
420 * The effective group ID or one of the supplementary group
421 * IDs of the calling process shall match the group owner of
422 * the file when setting the set-group-ID bit on that file
424 if (mask & XFS_AT_MODE) {
427 if ((vap->va_mode & S_ISUID) && !file_owner)
429 if ((vap->va_mode & S_ISGID) &&
430 !in_group_p((gid_t)ip->i_d.di_gid))
433 /* Linux allows this, Irix doesn't. */
434 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
437 if (m && !capable(CAP_FSETID))
443 * Change file ownership. Must be the owner or privileged.
444 * If the system was configured with the "restricted_chown"
445 * option, the owner is not permitted to give away the file,
446 * and can change the group id only to a group of which he
447 * or she is a member.
449 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
451 * These IDs could have changed since we last looked at them.
452 * But, we're assured that if the ownership did change
453 * while we didn't have the inode locked, inode's dquot(s)
454 * would have changed also.
456 iuid = ip->i_d.di_uid;
457 iprojid = ip->i_d.di_projid;
458 igid = ip->i_d.di_gid;
459 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
460 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
461 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
465 * CAP_CHOWN overrides the following restrictions:
467 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
468 * shall override the restriction that a process cannot
469 * change the user ID of a file it owns and the restriction
470 * that the group ID supplied to the chown() function
471 * shall be equal to either the group ID or one of the
472 * supplementary group IDs of the calling process.
474 if (restricted_chown &&
475 (iuid != uid || (igid != gid &&
476 !in_group_p((gid_t)gid))) &&
477 !capable(CAP_CHOWN)) {
478 code = XFS_ERROR(EPERM);
482 * Do a quota reservation only if uid/projid/gid is actually
485 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
486 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
487 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
489 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
490 capable(CAP_FOWNER) ?
491 XFS_QMOPT_FORCE_RES : 0);
492 if (code) /* out of quota */
498 * Truncate file. Must have write permission and not be a directory.
500 if (mask & XFS_AT_SIZE) {
501 /* Short circuit the truncate case for zero length files */
502 if ((vap->va_size == 0) &&
503 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
504 xfs_iunlock(ip, XFS_ILOCK_EXCL);
505 lock_flags &= ~XFS_ILOCK_EXCL;
506 if (mask & XFS_AT_CTIME)
507 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
513 code = XFS_ERROR(EISDIR);
515 } else if (!VN_ISREG(vp)) {
516 code = XFS_ERROR(EINVAL);
520 * Make sure that the dquots are attached to the inode.
522 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
527 * Change file access or modified times.
529 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
531 if ((flags & ATTR_UTIME) &&
532 !capable(CAP_FOWNER)) {
533 code = XFS_ERROR(EPERM);
540 * Change extent size or realtime flag.
542 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
544 * Can't change extent size if any extents are allocated.
546 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
547 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
549 code = XFS_ERROR(EINVAL); /* EFBIG? */
554 * Can't change realtime flag if any extents are allocated.
556 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
557 (mask & XFS_AT_XFLAGS) &&
558 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
559 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
560 code = XFS_ERROR(EINVAL); /* EFBIG? */
564 * Extent size must be a multiple of the appropriate block
565 * size, if set at all.
567 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
570 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
571 ((mask & XFS_AT_XFLAGS) &&
572 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
573 size = mp->m_sb.sb_rextsize <<
574 mp->m_sb.sb_blocklog;
576 size = mp->m_sb.sb_blocksize;
578 if (vap->va_extsize % size) {
579 code = XFS_ERROR(EINVAL);
584 * If realtime flag is set then must have realtime data.
586 if ((mask & XFS_AT_XFLAGS) &&
587 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
588 if ((mp->m_sb.sb_rblocks == 0) ||
589 (mp->m_sb.sb_rextsize == 0) ||
590 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
591 code = XFS_ERROR(EINVAL);
597 * Can't modify an immutable/append-only file unless
598 * we have appropriate permission.
600 if ((mask & XFS_AT_XFLAGS) &&
602 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
604 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
605 !capable(CAP_LINUX_IMMUTABLE)) {
606 code = XFS_ERROR(EPERM);
612 * Now we can make the changes. Before we join the inode
613 * to the transaction, if XFS_AT_SIZE is set then take care of
614 * the part of the truncation that must be done without the
615 * inode lock. This needs to be done before joining the inode
616 * to the transaction, because the inode cannot be unlocked
617 * once it is a part of the transaction.
619 if (mask & XFS_AT_SIZE) {
621 if ((vap->va_size > ip->i_d.di_size) &&
622 (flags & ATTR_NOSIZETOK) == 0) {
623 code = xfs_igrow_start(ip, vap->va_size, credp);
625 xfs_iunlock(ip, XFS_ILOCK_EXCL);
626 vn_iowait(vp); /* wait for the completion of any pending DIOs */
628 code = xfs_itruncate_data(ip, vap->va_size);
631 lock_flags &= ~XFS_ILOCK_EXCL;
632 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
635 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
636 if ((code = xfs_trans_reserve(tp, 0,
637 XFS_ITRUNCATE_LOG_RES(mp), 0,
638 XFS_TRANS_PERM_LOG_RES,
639 XFS_ITRUNCATE_LOG_COUNT))) {
640 xfs_trans_cancel(tp, 0);
642 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
645 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
646 xfs_ilock(ip, XFS_ILOCK_EXCL);
650 xfs_trans_ijoin(tp, ip, lock_flags);
651 xfs_trans_ihold(tp, ip);
654 /* determine whether mandatory locking mode changes */
655 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
658 * Truncate file. Must have write permission and not be a directory.
660 if (mask & XFS_AT_SIZE) {
661 if (vap->va_size > ip->i_d.di_size) {
662 xfs_igrow_finish(tp, ip, vap->va_size,
663 !(flags & ATTR_DMI));
664 } else if ((vap->va_size <= ip->i_d.di_size) ||
665 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
667 * signal a sync transaction unless
668 * we're truncating an already unlinked
669 * file on a wsync filesystem
671 code = xfs_itruncate_finish(&tp, ip,
672 (xfs_fsize_t)vap->va_size,
674 ((ip->i_d.di_nlink != 0 ||
675 !(mp->m_flags & XFS_MOUNT_WSYNC))
680 * Truncated "down", so we're removing references
681 * to old data here - if we now delay flushing for
682 * a long time, we expose ourselves unduly to the
683 * notorious NULL files problem. So, we mark this
684 * vnode and flush it when the file is closed, and
685 * do not wait the usual (long) time for writeout.
690 * Have to do this even if the file's size doesn't change.
692 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
696 * Change file access modes.
698 if (mask & XFS_AT_MODE) {
699 ip->i_d.di_mode &= S_IFMT;
700 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
702 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
703 timeflags |= XFS_ICHGTIME_CHG;
707 * Change file ownership. Must be the owner or privileged.
708 * If the system was configured with the "restricted_chown"
709 * option, the owner is not permitted to give away the file,
710 * and can change the group id only to a group of which he
711 * or she is a member.
713 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
715 * CAP_FSETID overrides the following restrictions:
717 * The set-user-ID and set-group-ID bits of a file will be
718 * cleared upon successful return from chown()
720 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
721 !capable(CAP_FSETID)) {
722 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
726 * Change the ownerships and register quota modifications
727 * in the transaction.
730 if (XFS_IS_UQUOTA_ON(mp)) {
731 ASSERT(mask & XFS_AT_UID);
733 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
734 &ip->i_udquot, udqp);
736 ip->i_d.di_uid = uid;
739 if (XFS_IS_GQUOTA_ON(mp)) {
740 ASSERT(!XFS_IS_PQUOTA_ON(mp));
741 ASSERT(mask & XFS_AT_GID);
743 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
744 &ip->i_gdquot, gdqp);
746 ip->i_d.di_gid = gid;
748 if (iprojid != projid) {
749 if (XFS_IS_PQUOTA_ON(mp)) {
750 ASSERT(!XFS_IS_GQUOTA_ON(mp));
751 ASSERT(mask & XFS_AT_PROJID);
753 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
754 &ip->i_gdquot, gdqp);
756 ip->i_d.di_projid = projid;
758 * We may have to rev the inode as well as
759 * the superblock version number since projids didn't
760 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
762 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
763 xfs_bump_ino_vers2(tp, ip);
766 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
767 timeflags |= XFS_ICHGTIME_CHG;
772 * Change file access or modified times.
774 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
775 if (mask & XFS_AT_ATIME) {
776 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
777 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
778 ip->i_update_core = 1;
779 timeflags &= ~XFS_ICHGTIME_ACC;
781 if (mask & XFS_AT_MTIME) {
782 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
783 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
784 timeflags &= ~XFS_ICHGTIME_MOD;
785 timeflags |= XFS_ICHGTIME_CHG;
787 if (tp && (flags & ATTR_UTIME))
788 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
792 * Change XFS-added attributes.
794 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
795 if (mask & XFS_AT_EXTSIZE) {
797 * Converting bytes to fs blocks.
799 ip->i_d.di_extsize = vap->va_extsize >>
800 mp->m_sb.sb_blocklog;
802 if (mask & XFS_AT_XFLAGS) {
805 /* can't set PREALLOC this way, just preserve it */
806 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
807 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
808 di_flags |= XFS_DIFLAG_IMMUTABLE;
809 if (vap->va_xflags & XFS_XFLAG_APPEND)
810 di_flags |= XFS_DIFLAG_APPEND;
811 if (vap->va_xflags & XFS_XFLAG_SYNC)
812 di_flags |= XFS_DIFLAG_SYNC;
813 if (vap->va_xflags & XFS_XFLAG_NOATIME)
814 di_flags |= XFS_DIFLAG_NOATIME;
815 if (vap->va_xflags & XFS_XFLAG_NODUMP)
816 di_flags |= XFS_DIFLAG_NODUMP;
817 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
818 di_flags |= XFS_DIFLAG_PROJINHERIT;
819 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
820 di_flags |= XFS_DIFLAG_NODEFRAG;
821 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
822 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
823 di_flags |= XFS_DIFLAG_RTINHERIT;
824 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
825 di_flags |= XFS_DIFLAG_NOSYMLINKS;
826 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
827 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
828 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
829 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
830 di_flags |= XFS_DIFLAG_REALTIME;
831 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
833 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
835 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
836 di_flags |= XFS_DIFLAG_EXTSIZE;
838 ip->i_d.di_flags = di_flags;
840 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
841 timeflags |= XFS_ICHGTIME_CHG;
845 * Change file inode change time only if XFS_AT_CTIME set
846 * AND we have been called by a DMI function.
849 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
850 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
851 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
852 ip->i_update_core = 1;
853 timeflags &= ~XFS_ICHGTIME_CHG;
857 * Send out timestamp changes that need to be set to the
858 * current time. Not done when called by a DMI function.
860 if (timeflags && !(flags & ATTR_DMI))
861 xfs_ichgtime(ip, timeflags);
863 XFS_STATS_INC(xs_ig_attrchg);
866 * If this is a synchronous mount, make sure that the
867 * transaction goes to disk before returning to the user.
868 * This is slightly sub-optimal in that truncates require
869 * two sync transactions instead of one for wsync filesystems.
870 * One for the truncate and one for the timestamps since we
871 * don't want to change the timestamps unless we're sure the
872 * truncate worked. Truncates are less than 1% of the laddis
873 * mix so this probably isn't worth the trouble to optimize.
877 if (mp->m_flags & XFS_MOUNT_WSYNC)
878 xfs_trans_set_sync(tp);
880 code = xfs_trans_commit(tp, commit_flags, NULL);
884 * If the (regular) file's mandatory locking mode changed, then
885 * notify the vnode. We do this under the inode lock to prevent
886 * racing calls to vop_vnode_change.
888 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
889 if (mandlock_before != mandlock_after) {
890 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING,
894 xfs_iunlock(ip, lock_flags);
897 * Release any dquot(s) the inode had kept before chown.
899 XFS_QM_DQRELE(mp, olddquot1);
900 XFS_QM_DQRELE(mp, olddquot2);
901 XFS_QM_DQRELE(mp, udqp);
902 XFS_QM_DQRELE(mp, gdqp);
908 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
909 !(flags & ATTR_DMI)) {
910 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
911 NULL, DM_RIGHT_NULL, NULL, NULL,
912 0, 0, AT_DELAY_FLAG(flags));
917 commit_flags |= XFS_TRANS_ABORT;
920 XFS_QM_DQRELE(mp, udqp);
921 XFS_QM_DQRELE(mp, gdqp);
923 xfs_trans_cancel(tp, commit_flags);
925 if (lock_flags != 0) {
926 xfs_iunlock(ip, lock_flags);
934 * Null conversion from vnode mode bits to inode mode bits, as in efs.
945 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
946 (inst_t *)__return_address);
948 ip = XFS_BHVTOI(bdp);
949 xfs_ilock(ip, XFS_ILOCK_SHARED);
950 error = xfs_iaccess(ip, mode, credp);
951 xfs_iunlock(ip, XFS_ILOCK_SHARED);
957 * The maximum pathlen is 1024 bytes. Since the minimum file system
958 * blocksize is 512 bytes, we can get a max of 2 extents back from
961 #define SYMLINK_MAPS 2
982 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
988 vp = BHV_TO_VNODE(bdp);
989 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
991 ip = XFS_BHVTOI(bdp);
994 if (XFS_FORCED_SHUTDOWN(mp))
995 return XFS_ERROR(EIO);
997 xfs_ilock(ip, XFS_ILOCK_SHARED);
999 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1001 offset = uiop->uio_offset;
1002 count = uiop->uio_resid;
1005 error = XFS_ERROR(EINVAL);
1014 * See if the symlink is stored inline.
1016 pathlen = (int)ip->i_d.di_size;
1018 if (ip->i_df.if_flags & XFS_IFINLINE) {
1019 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1023 * Symlink not inline. Call bmap to get it in.
1025 nmaps = SYMLINK_MAPS;
1027 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1028 0, NULL, 0, mval, &nmaps, NULL, NULL);
1034 for (n = 0; n < nmaps; n++) {
1035 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1036 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1037 bp = xfs_buf_read(mp->m_ddev_targp, d,
1038 BTOBB(byte_cnt), 0);
1039 error = XFS_BUF_GETERROR(bp);
1041 xfs_ioerror_alert("xfs_readlink",
1042 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1046 if (pathlen < byte_cnt)
1048 pathlen -= byte_cnt;
1050 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1057 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1065 * This is called to sync the inode and its data out to disk.
1066 * We need to hold the I/O lock while flushing the data, and
1067 * the inode lock while flushing the inode. The inode lock CANNOT
1068 * be held while flushing the data, so acquire after we're done
1082 int log_flushed = 0, changed = 1;
1084 vn_trace_entry(BHV_TO_VNODE(bdp),
1085 __FUNCTION__, (inst_t *)__return_address);
1087 ip = XFS_BHVTOI(bdp);
1089 ASSERT(start >= 0 && stop >= -1);
1091 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1092 return XFS_ERROR(EIO);
1095 * We always need to make sure that the required inode state
1096 * is safe on disk. The vnode might be clean but because
1097 * of committed transactions that haven't hit the disk yet.
1098 * Likewise, there could be unflushed non-transactional
1099 * changes to the inode core that have to go to disk.
1101 * The following code depends on one assumption: that
1102 * any transaction that changes an inode logs the core
1103 * because it has to change some field in the inode core
1104 * (typically nextents or nblocks). That assumption
1105 * implies that any transactions against an inode will
1106 * catch any non-transactional updates. If inode-altering
1107 * transactions exist that violate this assumption, the
1108 * code breaks. Right now, it figures that if the involved
1109 * update_* field is clear and the inode is unpinned, the
1110 * inode is clean. Either it's been flushed or it's been
1111 * committed and the commit has hit the disk unpinning the inode.
1112 * (Note that xfs_inode_item_format() called at commit clears
1113 * the update_* fields.)
1115 xfs_ilock(ip, XFS_ILOCK_SHARED);
1117 /* If we are flushing data then we care about update_size
1118 * being set, otherwise we care about update_core
1120 if ((flag & FSYNC_DATA) ?
1121 (ip->i_update_size == 0) :
1122 (ip->i_update_core == 0)) {
1124 * Timestamps/size haven't changed since last inode
1125 * flush or inode transaction commit. That means
1126 * either nothing got written or a transaction
1127 * committed which caught the updates. If the
1128 * latter happened and the transaction hasn't
1129 * hit the disk yet, the inode will be still
1130 * be pinned. If it is, force the log.
1133 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1135 if (xfs_ipincount(ip)) {
1136 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1138 ((flag & FSYNC_WAIT)
1139 ? XFS_LOG_SYNC : 0),
1143 * If the inode is not pinned and nothing
1144 * has changed we don't need to flush the
1152 * Kick off a transaction to log the inode
1153 * core to get the updates. Make it
1154 * sync if FSYNC_WAIT is passed in (which
1155 * is done by everybody but specfs). The
1156 * sync transaction will also force the log.
1158 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1159 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1160 if ((error = xfs_trans_reserve(tp, 0,
1161 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1163 xfs_trans_cancel(tp, 0);
1166 xfs_ilock(ip, XFS_ILOCK_EXCL);
1169 * Note - it's possible that we might have pushed
1170 * ourselves out of the way during trans_reserve
1171 * which would flush the inode. But there's no
1172 * guarantee that the inode buffer has actually
1173 * gone out yet (it's delwri). Plus the buffer
1174 * could be pinned anyway if it's part of an
1175 * inode in another recent transaction. So we
1176 * play it safe and fire off the transaction anyway.
1178 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1179 xfs_trans_ihold(tp, ip);
1180 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1181 if (flag & FSYNC_WAIT)
1182 xfs_trans_set_sync(tp);
1183 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1185 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1188 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1190 * If the log write didn't issue an ordered tag we need
1191 * to flush the disk cache for the data device now.
1194 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1197 * If this inode is on the RT dev we need to flush that
1200 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1201 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1208 * This is called by xfs_inactive to free any blocks beyond eof,
1209 * when the link count isn't zero.
1212 xfs_inactive_free_eofblocks(
1218 xfs_fileoff_t end_fsb;
1219 xfs_fileoff_t last_fsb;
1220 xfs_filblks_t map_len;
1222 xfs_bmbt_irec_t imap;
1225 * Figure out if there are any blocks beyond the end
1226 * of the file. If not, then there is nothing to do.
1228 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1229 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1230 map_len = last_fsb - end_fsb;
1235 xfs_ilock(ip, XFS_ILOCK_SHARED);
1236 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1237 NULL, 0, &imap, &nimaps, NULL, NULL);
1238 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1240 if (!error && (nimaps != 0) &&
1241 (imap.br_startblock != HOLESTARTBLOCK ||
1242 ip->i_delayed_blks)) {
1244 * Attach the dquots to the inode up front.
1246 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1250 * There are blocks after the end of file.
1251 * Free them up now by truncating the file to
1254 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1257 * Do the xfs_itruncate_start() call before
1258 * reserving any log space because
1259 * itruncate_start will call into the buffer
1260 * cache and we can't
1261 * do that within a transaction.
1263 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1264 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1267 error = xfs_trans_reserve(tp, 0,
1268 XFS_ITRUNCATE_LOG_RES(mp),
1269 0, XFS_TRANS_PERM_LOG_RES,
1270 XFS_ITRUNCATE_LOG_COUNT);
1272 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1273 xfs_trans_cancel(tp, 0);
1274 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1278 xfs_ilock(ip, XFS_ILOCK_EXCL);
1279 xfs_trans_ijoin(tp, ip,
1282 xfs_trans_ihold(tp, ip);
1284 error = xfs_itruncate_finish(&tp, ip,
1289 * If we get an error at this point we
1290 * simply don't bother truncating the file.
1293 xfs_trans_cancel(tp,
1294 (XFS_TRANS_RELEASE_LOG_RES |
1297 error = xfs_trans_commit(tp,
1298 XFS_TRANS_RELEASE_LOG_RES,
1301 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1307 * Free a symlink that has blocks associated with it.
1310 xfs_inactive_symlink_rmt(
1318 xfs_fsblock_t first_block;
1319 xfs_bmap_free_t free_list;
1322 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1330 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1332 * We're freeing a symlink that has some
1333 * blocks allocated to it. Free the
1334 * blocks here. We know that we've got
1335 * either 1 or 2 extents and that we can
1336 * free them all in one bunmapi call.
1338 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1339 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1340 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1341 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1342 xfs_trans_cancel(tp, 0);
1347 * Lock the inode, fix the size, and join it to the transaction.
1348 * Hold it so in the normal path, we still have it locked for
1349 * the second transaction. In the error paths we need it
1350 * held so the cancel won't rele it, see below.
1352 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1353 size = (int)ip->i_d.di_size;
1354 ip->i_d.di_size = 0;
1355 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1356 xfs_trans_ihold(tp, ip);
1357 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1359 * Find the block(s) so we can inval and unmap them.
1362 XFS_BMAP_INIT(&free_list, &first_block);
1363 nmaps = ARRAY_SIZE(mval);
1364 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1365 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1369 * Invalidate the block(s).
1371 for (i = 0; i < nmaps; i++) {
1372 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1373 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1374 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1375 xfs_trans_binval(tp, bp);
1378 * Unmap the dead block(s) to the free_list.
1380 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1381 &first_block, &free_list, NULL, &done)))
1385 * Commit the first transaction. This logs the EFI and the inode.
1387 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1390 * The transaction must have been committed, since there were
1391 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1392 * The new tp has the extent freeing and EFDs.
1396 * The first xact was committed, so add the inode to the new one.
1397 * Mark it dirty so it will be logged and moved forward in the log as
1398 * part of every commit.
1400 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1401 xfs_trans_ihold(tp, ip);
1402 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1404 * Get a new, empty transaction to return to our caller.
1406 ntp = xfs_trans_dup(tp);
1408 * Commit the transaction containing extent freeing and EFDs.
1409 * If we get an error on the commit here or on the reserve below,
1410 * we need to unlock the inode since the new transaction doesn't
1411 * have the inode attached.
1413 error = xfs_trans_commit(tp, 0, NULL);
1416 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1420 * Remove the memory for extent descriptions (just bookkeeping).
1422 if (ip->i_df.if_bytes)
1423 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1424 ASSERT(ip->i_df.if_bytes == 0);
1426 * Put an itruncate log reservation in the new transaction
1429 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1430 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1431 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1435 * Return with the inode locked but not joined to the transaction.
1441 xfs_bmap_cancel(&free_list);
1444 * Have to come here with the inode locked and either
1445 * (held and in the transaction) or (not in the transaction).
1446 * If the inode isn't held then cancel would iput it, but
1447 * that's wrong since this is inactive and the vnode ref
1448 * count is 0 already.
1449 * Cancel won't do anything to the inode if held, but it still
1450 * needs to be locked until the cancel is done, if it was
1451 * joined to the transaction.
1453 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1454 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1461 xfs_inactive_symlink_local(
1467 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1469 * We're freeing a symlink which fit into
1470 * the inode. Just free the memory used
1471 * to hold the old symlink.
1473 error = xfs_trans_reserve(*tpp, 0,
1474 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1475 0, XFS_TRANS_PERM_LOG_RES,
1476 XFS_ITRUNCATE_LOG_COUNT);
1479 xfs_trans_cancel(*tpp, 0);
1483 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1486 * Zero length symlinks _can_ exist.
1488 if (ip->i_df.if_bytes > 0) {
1489 xfs_idata_realloc(ip,
1490 -(ip->i_df.if_bytes),
1492 ASSERT(ip->i_df.if_bytes == 0);
1506 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1509 ASSERT(ip->i_d.di_forkoff != 0);
1510 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1511 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1513 error = xfs_attr_inactive(ip);
1516 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1517 return error; /* goto out */
1520 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1521 error = xfs_trans_reserve(tp, 0,
1522 XFS_IFREE_LOG_RES(mp),
1523 0, XFS_TRANS_PERM_LOG_RES,
1524 XFS_INACTIVE_LOG_COUNT);
1526 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1527 xfs_trans_cancel(tp, 0);
1529 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1533 xfs_ilock(ip, XFS_ILOCK_EXCL);
1534 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1535 xfs_trans_ihold(tp, ip);
1536 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1538 ASSERT(ip->i_d.di_anextents == 0);
1553 vp = BHV_TO_VNODE(bdp);
1554 ip = XFS_BHVTOI(bdp);
1557 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
1560 /* If this is a read-only mount, don't do this (would generate I/O) */
1561 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1564 #ifdef HAVE_REFCACHE
1565 /* If we are in the NFS reference cache then don't do this now */
1570 if (ip->i_d.di_nlink != 0) {
1571 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1572 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1573 ip->i_delayed_blks > 0)) &&
1574 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1575 (!(ip->i_d.di_flags &
1576 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1577 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1579 /* Update linux inode block count after free above */
1580 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1581 ip->i_d.di_nblocks + ip->i_delayed_blks);
1591 * This is called when the vnode reference count for the vnode
1592 * goes to zero. If the file has been unlinked, then it must
1593 * now be truncated. Also, we clear all of the read-ahead state
1594 * kept for the inode here since the file is now closed.
1603 xfs_bmap_free_t free_list;
1604 xfs_fsblock_t first_block;
1611 vp = BHV_TO_VNODE(bdp);
1612 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1614 ip = XFS_BHVTOI(bdp);
1617 * If the inode is already free, then there can be nothing
1620 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1621 ASSERT(ip->i_df.if_real_bytes == 0);
1622 ASSERT(ip->i_df.if_broot_bytes == 0);
1623 return VN_INACTIVE_CACHE;
1627 * Only do a truncate if it's a regular file with
1628 * some actual space in it. It's OK to look at the
1629 * inode's fields without the lock because we're the
1630 * only one with a reference to the inode.
1632 truncate = ((ip->i_d.di_nlink == 0) &&
1633 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1634 (ip->i_delayed_blks > 0)) &&
1635 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1639 if (ip->i_d.di_nlink == 0 &&
1640 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1641 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1646 /* If this is a read-only mount, don't do this (would generate I/O) */
1647 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1650 if (ip->i_d.di_nlink != 0) {
1651 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1652 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1653 ip->i_delayed_blks > 0)) &&
1654 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1655 (!(ip->i_d.di_flags &
1656 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1657 (ip->i_delayed_blks != 0)))) {
1658 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1659 return VN_INACTIVE_CACHE;
1660 /* Update linux inode block count after free above */
1661 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1662 ip->i_d.di_nblocks + ip->i_delayed_blks);
1667 ASSERT(ip->i_d.di_nlink == 0);
1669 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1670 return VN_INACTIVE_CACHE;
1672 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1675 * Do the xfs_itruncate_start() call before
1676 * reserving any log space because itruncate_start
1677 * will call into the buffer cache and we can't
1678 * do that within a transaction.
1680 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1682 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1684 error = xfs_trans_reserve(tp, 0,
1685 XFS_ITRUNCATE_LOG_RES(mp),
1686 0, XFS_TRANS_PERM_LOG_RES,
1687 XFS_ITRUNCATE_LOG_COUNT);
1689 /* Don't call itruncate_cleanup */
1690 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1691 xfs_trans_cancel(tp, 0);
1692 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1693 return VN_INACTIVE_CACHE;
1696 xfs_ilock(ip, XFS_ILOCK_EXCL);
1697 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1698 xfs_trans_ihold(tp, ip);
1701 * normally, we have to run xfs_itruncate_finish sync.
1702 * But if filesystem is wsync and we're in the inactive
1703 * path, then we know that nlink == 0, and that the
1704 * xaction that made nlink == 0 is permanently committed
1705 * since xfs_remove runs as a synchronous transaction.
1707 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1708 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1711 xfs_trans_cancel(tp,
1712 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1713 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1714 return VN_INACTIVE_CACHE;
1716 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1719 * If we get an error while cleaning up a
1720 * symlink we bail out.
1722 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1723 xfs_inactive_symlink_rmt(ip, &tp) :
1724 xfs_inactive_symlink_local(ip, &tp);
1728 return VN_INACTIVE_CACHE;
1731 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1732 xfs_trans_ihold(tp, ip);
1734 error = xfs_trans_reserve(tp, 0,
1735 XFS_IFREE_LOG_RES(mp),
1736 0, XFS_TRANS_PERM_LOG_RES,
1737 XFS_INACTIVE_LOG_COUNT);
1739 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1740 xfs_trans_cancel(tp, 0);
1741 return VN_INACTIVE_CACHE;
1744 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1745 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1746 xfs_trans_ihold(tp, ip);
1750 * If there are attributes associated with the file
1751 * then blow them away now. The code calls a routine
1752 * that recursively deconstructs the attribute fork.
1753 * We need to just commit the current transaction
1754 * because we can't use it for xfs_attr_inactive().
1756 if (ip->i_d.di_anextents > 0) {
1757 error = xfs_inactive_attrs(ip, &tp);
1759 * If we got an error, the transaction is already
1760 * cancelled, and the inode is unlocked. Just get out.
1763 return VN_INACTIVE_CACHE;
1764 } else if (ip->i_afp) {
1765 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1771 XFS_BMAP_INIT(&free_list, &first_block);
1772 error = xfs_ifree(tp, ip, &free_list);
1775 * If we fail to free the inode, shut down. The cancel
1776 * might do that, we need to make sure. Otherwise the
1777 * inode might be lost for a long time or forever.
1779 if (!XFS_FORCED_SHUTDOWN(mp)) {
1781 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1782 error, mp->m_fsname);
1783 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1785 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1788 * Credit the quota account(s). The inode is gone.
1790 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1793 * Just ignore errors at this point. There is
1794 * nothing we can do except to try to keep going.
1796 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1798 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1801 * Release the dquots held by inode, if any.
1803 XFS_QM_DQDETACH(mp, ip);
1805 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1808 return VN_INACTIVE_CACHE;
1817 bhv_desc_t *dir_bdp,
1824 xfs_inode_t *dp, *ip;
1828 bhv_vnode_t *dir_vp;
1830 dir_vp = BHV_TO_VNODE(dir_bdp);
1831 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1833 dp = XFS_BHVTOI(dir_bdp);
1835 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1836 return XFS_ERROR(EIO);
1838 lock_mode = xfs_ilock_map_shared(dp);
1839 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1841 *vpp = XFS_ITOV(ip);
1844 xfs_iunlock_map_shared(dp, lock_mode);
1850 * xfs_create (create a new file).
1854 bhv_desc_t *dir_bdp,
1860 char *name = VNAME(dentry);
1861 bhv_vnode_t *dir_vp;
1862 xfs_inode_t *dp, *ip;
1863 bhv_vnode_t *vp = NULL;
1868 xfs_bmap_free_t free_list;
1869 xfs_fsblock_t first_block;
1870 boolean_t dp_joined_to_trans;
1871 int dm_event_sent = 0;
1875 struct xfs_dquot *udqp, *gdqp;
1881 dir_vp = BHV_TO_VNODE(dir_bdp);
1882 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1884 dp = XFS_BHVTOI(dir_bdp);
1887 dm_di_mode = vap->va_mode;
1888 namelen = VNAMELEN(dentry);
1890 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1891 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1892 dir_vp, DM_RIGHT_NULL, NULL,
1893 DM_RIGHT_NULL, name, NULL,
1901 if (XFS_FORCED_SHUTDOWN(mp))
1902 return XFS_ERROR(EIO);
1904 /* Return through std_return after this point. */
1907 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1908 prid = dp->i_d.di_projid;
1909 else if (vap->va_mask & XFS_AT_PROJID)
1910 prid = (xfs_prid_t)vap->va_projid;
1912 prid = (xfs_prid_t)dfltprid;
1915 * Make sure that we have allocated dquot(s) on disk.
1917 error = XFS_QM_DQVOPALLOC(mp, dp,
1918 current_fsuid(credp), current_fsgid(credp), prid,
1919 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1924 dp_joined_to_trans = B_FALSE;
1926 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1927 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1928 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1930 * Initially assume that the file does not exist and
1931 * reserve the resources for that case. If that is not
1932 * the case we'll drop the one we have and get a more
1933 * appropriate transaction later.
1935 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1936 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1937 if (error == ENOSPC) {
1939 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1940 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1948 xfs_ilock(dp, XFS_ILOCK_EXCL);
1950 XFS_BMAP_INIT(&free_list, &first_block);
1955 * Reserve disk quota and the inode.
1957 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1962 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1964 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1965 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1966 rdev, credp, prid, resblks > 0,
1969 if (error == ENOSPC)
1976 * At this point, we've gotten a newly allocated inode.
1977 * It is locked (and joined to the transaction).
1980 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1983 * Now we join the directory inode to the transaction.
1984 * We do not do it earlier because xfs_dir_ialloc
1985 * might commit the previous transaction (and release
1990 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1991 dp_joined_to_trans = B_TRUE;
1993 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1994 &first_block, &free_list,
1995 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1997 ASSERT(error != ENOSPC);
2000 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2001 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2004 * If this is a synchronous mount, make sure that the
2005 * create transaction goes to disk before returning to
2008 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2009 xfs_trans_set_sync(tp);
2015 * Attach the dquot(s) to the inodes and modify them incore.
2016 * These ids of the inode couldn't have changed since the new
2017 * inode has been locked ever since it was created.
2019 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2022 * xfs_trans_commit normally decrements the vnode ref count
2023 * when it unlocks the inode. Since we want to return the
2024 * vnode to the caller, we bump the vnode ref count now.
2029 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2031 xfs_bmap_cancel(&free_list);
2035 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2042 XFS_QM_DQRELE(mp, udqp);
2043 XFS_QM_DQRELE(mp, gdqp);
2046 * Propagate the fact that the vnode changed after the
2047 * xfs_inode locks have been released.
2049 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2053 /* Fallthrough to std_return with error = 0 */
2056 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2057 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2058 DM_EVENT_POSTCREATE)) {
2059 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2060 dir_vp, DM_RIGHT_NULL,
2062 DM_RIGHT_NULL, name, NULL,
2063 dm_di_mode, error, 0);
2068 cancel_flags |= XFS_TRANS_ABORT;
2073 xfs_trans_cancel(tp, cancel_flags);
2075 if (!dp_joined_to_trans && (dp != NULL))
2076 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2077 XFS_QM_DQRELE(mp, udqp);
2078 XFS_QM_DQRELE(mp, gdqp);
2084 * Wait until after the current transaction is aborted to
2085 * release the inode. This prevents recursive transactions
2086 * and deadlocks from xfs_inactive.
2088 cancel_flags |= XFS_TRANS_ABORT;
2089 xfs_trans_cancel(tp, cancel_flags);
2092 XFS_QM_DQRELE(mp, udqp);
2093 XFS_QM_DQRELE(mp, gdqp);
2100 * Some counters to see if (and how often) we are hitting some deadlock
2101 * prevention code paths.
2105 int xfs_rm_lock_delays;
2106 int xfs_rm_attempts;
2110 * The following routine will lock the inodes associated with the
2111 * directory and the named entry in the directory. The locks are
2112 * acquired in increasing inode number.
2114 * If the entry is "..", then only the directory is locked. The
2115 * vnode ref count will still include that from the .. entry in
2118 * There is a deadlock we need to worry about. If the locked directory is
2119 * in the AIL, it might be blocking up the log. The next inode we lock
2120 * could be already locked by another thread waiting for log space (e.g
2121 * a permanent log reservation with a long running transaction (see
2122 * xfs_itruncate_finish)). To solve this, we must check if the directory
2123 * is in the ail and use lock_nowait. If we can't lock, we need to
2124 * drop the inode lock on the directory and try again. xfs_iunlock will
2125 * potentially push the tail if we were holding up the log.
2128 xfs_lock_dir_and_entry(
2131 xfs_inode_t *ip) /* inode of entry 'name' */
2135 xfs_inode_t *ips[2];
2144 xfs_ilock(dp, XFS_ILOCK_EXCL);
2151 * We want to lock in increasing inum. Since we've already
2152 * acquired the lock on the directory, we may need to release
2153 * if if the inum of the entry turns out to be less.
2155 if (e_inum > dp->i_ino) {
2157 * We are already in the right order, so just
2158 * lock on the inode of the entry.
2159 * We need to use nowait if dp is in the AIL.
2162 lp = (xfs_log_item_t *)dp->i_itemp;
2163 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2164 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2171 * Unlock dp and try again.
2172 * xfs_iunlock will try to push the tail
2173 * if the inode is in the AIL.
2176 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2178 if ((attempts % 5) == 0) {
2179 delay(1); /* Don't just spin the CPU */
2181 xfs_rm_lock_delays++;
2187 xfs_ilock(ip, XFS_ILOCK_EXCL);
2189 } else if (e_inum < dp->i_ino) {
2190 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2194 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2196 /* else e_inum == dp->i_ino */
2197 /* This can happen if we're asked to lock /x/..
2198 * the entry is "..", which is also the parent directory.
2206 int xfs_small_retries;
2207 int xfs_middle_retries;
2208 int xfs_lots_retries;
2209 int xfs_lock_delays;
2213 * The following routine will lock n inodes in exclusive mode.
2214 * We assume the caller calls us with the inodes in i_ino order.
2216 * We need to detect deadlock where an inode that we lock
2217 * is in the AIL and we start waiting for another inode that is locked
2218 * by a thread in a long running transaction (such as truncate). This can
2219 * result in deadlock since the long running trans might need to wait
2220 * for the inode we just locked in order to push the tail and free space
2230 int attempts = 0, i, j, try_lock;
2233 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2244 for (; i < inodes; i++) {
2247 if (i && (ips[i] == ips[i-1])) /* Already locked */
2251 * If try_lock is not set yet, make sure all locked inodes
2252 * are not in the AIL.
2253 * If any are, set try_lock to be used later.
2257 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2258 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2259 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2266 * If any of the previous locks we have locked is in the AIL,
2267 * we must TRY to get the second and subsequent locks. If
2268 * we can't get any, we must release all we have
2273 /* try_lock must be 0 if i is 0. */
2275 * try_lock means we have an inode locked
2276 * that is in the AIL.
2279 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2283 * Unlock all previous guys and try again.
2284 * xfs_iunlock will try to push the tail
2285 * if the inode is in the AIL.
2288 for(j = i - 1; j >= 0; j--) {
2291 * Check to see if we've already
2292 * unlocked this one.
2293 * Not the first one going back,
2294 * and the inode ptr is the same.
2296 if ((j != (i - 1)) && ips[j] ==
2300 xfs_iunlock(ips[j], lock_mode);
2303 if ((attempts % 5) == 0) {
2304 delay(1); /* Don't just spin the CPU */
2314 xfs_ilock(ips[i], lock_mode);
2320 if (attempts < 5) xfs_small_retries++;
2321 else if (attempts < 100) xfs_middle_retries++;
2322 else xfs_lots_retries++;
2330 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2331 int remove_which_error_return = 0;
2333 #define REMOVE_DEBUG_TRACE(x)
2334 #endif /* ! DEBUG */
2343 bhv_desc_t *dir_bdp,
2347 bhv_vnode_t *dir_vp;
2348 char *name = VNAME(dentry);
2349 xfs_inode_t *dp, *ip;
2350 xfs_trans_t *tp = NULL;
2353 xfs_bmap_free_t free_list;
2354 xfs_fsblock_t first_block;
2362 dir_vp = BHV_TO_VNODE(dir_bdp);
2363 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2365 dp = XFS_BHVTOI(dir_bdp);
2368 if (XFS_FORCED_SHUTDOWN(mp))
2369 return XFS_ERROR(EIO);
2371 namelen = VNAMELEN(dentry);
2373 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2374 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2375 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2376 name, NULL, 0, 0, 0);
2381 /* From this point on, return through std_return */
2385 * We need to get a reference to ip before we get our log
2386 * reservation. The reason for this is that we cannot call
2387 * xfs_iget for an inode for which we do not have a reference
2388 * once we've acquired a log reservation. This is because the
2389 * inode we are trying to get might be in xfs_inactive going
2390 * for a log reservation. Since we'll have to wait for the
2391 * inactive code to complete before returning from xfs_iget,
2392 * we need to make sure that we don't have log space reserved
2393 * when we call xfs_iget. Instead we get an unlocked reference
2394 * to the inode before getting our log reservation.
2396 error = xfs_get_dir_entry(dentry, &ip);
2398 REMOVE_DEBUG_TRACE(__LINE__);
2402 dm_di_mode = ip->i_d.di_mode;
2404 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2408 error = XFS_QM_DQATTACH(mp, dp, 0);
2409 if (!error && dp != ip)
2410 error = XFS_QM_DQATTACH(mp, ip, 0);
2412 REMOVE_DEBUG_TRACE(__LINE__);
2417 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2418 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2420 * We try to get the real space reservation first,
2421 * allowing for directory btree deletion(s) implying
2422 * possible bmap insert(s). If we can't get the space
2423 * reservation then we use 0 instead, and avoid the bmap
2424 * btree insert(s) in the directory code by, if the bmap
2425 * insert tries to happen, instead trimming the LAST
2426 * block from the directory.
2428 resblks = XFS_REMOVE_SPACE_RES(mp);
2429 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2430 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2431 if (error == ENOSPC) {
2433 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2434 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2437 ASSERT(error != ENOSPC);
2438 REMOVE_DEBUG_TRACE(__LINE__);
2439 xfs_trans_cancel(tp, 0);
2444 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2446 REMOVE_DEBUG_TRACE(__LINE__);
2447 xfs_trans_cancel(tp, cancel_flags);
2453 * At this point, we've gotten both the directory and the entry
2456 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2459 * Increment vnode ref count only in this case since
2460 * there's an extra vnode reference in the case where
2464 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2468 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2470 XFS_BMAP_INIT(&free_list, &first_block);
2471 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2472 &first_block, &free_list, 0);
2474 ASSERT(error != ENOENT);
2475 REMOVE_DEBUG_TRACE(__LINE__);
2478 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2481 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2483 error = xfs_droplink(tp, ip);
2485 REMOVE_DEBUG_TRACE(__LINE__);
2489 /* Determine if this is the last link while
2490 * we are in the transaction.
2492 link_zero = (ip)->i_d.di_nlink==0;
2495 * Take an extra ref on the inode so that it doesn't
2496 * go to xfs_inactive() from within the commit.
2501 * If this is a synchronous mount, make sure that the
2502 * remove transaction goes to disk before returning to
2505 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2506 xfs_trans_set_sync(tp);
2509 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2511 REMOVE_DEBUG_TRACE(__LINE__);
2515 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2522 * Before we drop our extra reference to the inode, purge it
2523 * from the refcache if it is there. By waiting until afterwards
2524 * to do the IRELE, we ensure that we won't go inactive in the
2525 * xfs_refcache_purge_ip routine (although that would be OK).
2527 xfs_refcache_purge_ip(ip);
2529 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2532 * Let interposed file systems know about removed links.
2534 bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero);
2538 /* Fall through to std_return with error = 0 */
2540 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2541 DM_EVENT_POSTREMOVE)) {
2542 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2543 dir_vp, DM_RIGHT_NULL,
2544 NULL, DM_RIGHT_NULL,
2545 name, NULL, dm_di_mode, error, 0);
2550 xfs_bmap_cancel(&free_list);
2551 cancel_flags |= XFS_TRANS_ABORT;
2552 xfs_trans_cancel(tp, cancel_flags);
2557 * In this case make sure to not release the inode until after
2558 * the current transaction is aborted. Releasing it beforehand
2559 * can cause us to go to xfs_inactive and start a recursive
2560 * transaction which can easily deadlock with the current one.
2562 xfs_bmap_cancel(&free_list);
2563 cancel_flags |= XFS_TRANS_ABORT;
2564 xfs_trans_cancel(tp, cancel_flags);
2567 * Before we drop our extra reference to the inode, purge it
2568 * from the refcache if it is there. By waiting until afterwards
2569 * to do the IRELE, we ensure that we won't go inactive in the
2570 * xfs_refcache_purge_ip routine (although that would be OK).
2572 xfs_refcache_purge_ip(ip);
2586 bhv_desc_t *target_dir_bdp,
2587 bhv_vnode_t *src_vp,
2591 xfs_inode_t *tdp, *sip;
2594 xfs_inode_t *ips[2];
2596 xfs_bmap_free_t free_list;
2597 xfs_fsblock_t first_block;
2600 bhv_vnode_t *target_dir_vp;
2602 char *target_name = VNAME(dentry);
2605 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2606 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2607 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2609 target_namelen = VNAMELEN(dentry);
2610 if (VN_ISDIR(src_vp))
2611 return XFS_ERROR(EPERM);
2613 sip = xfs_vtoi(src_vp);
2614 tdp = XFS_BHVTOI(target_dir_bdp);
2616 if (XFS_FORCED_SHUTDOWN(mp))
2617 return XFS_ERROR(EIO);
2619 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2620 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2621 target_dir_vp, DM_RIGHT_NULL,
2622 src_vp, DM_RIGHT_NULL,
2623 target_name, NULL, 0, 0, 0);
2628 /* Return through std_return after this point. */
2630 error = XFS_QM_DQATTACH(mp, sip, 0);
2631 if (!error && sip != tdp)
2632 error = XFS_QM_DQATTACH(mp, tdp, 0);
2636 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2637 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2638 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2639 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2640 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2641 if (error == ENOSPC) {
2643 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2644 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2651 if (sip->i_ino < tdp->i_ino) {
2659 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2662 * Increment vnode ref counts since xfs_trans_commit &
2663 * xfs_trans_cancel will both unlock the inodes and
2664 * decrement the associated ref counts.
2667 VN_HOLD(target_dir_vp);
2668 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2669 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2672 * If the source has too many links, we can't make any more to it.
2674 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2675 error = XFS_ERROR(EMLINK);
2680 * If we are using project inheritance, we only allow hard link
2681 * creation in our tree when the project IDs are the same; else
2682 * the tree quota mechanism could be circumvented.
2684 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2685 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2686 error = XFS_ERROR(EXDEV);
2691 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2695 XFS_BMAP_INIT(&free_list, &first_block);
2697 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2698 sip->i_ino, &first_block, &free_list,
2702 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2704 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2706 error = xfs_bumplink(tp, sip);
2712 * If this is a synchronous mount, make sure that the
2713 * link transaction goes to disk before returning to
2716 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2717 xfs_trans_set_sync(tp);
2720 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2722 xfs_bmap_cancel(&free_list);
2726 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2731 /* Fall through to std_return with error = 0. */
2733 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2734 DM_EVENT_POSTLINK)) {
2735 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2736 target_dir_vp, DM_RIGHT_NULL,
2737 src_vp, DM_RIGHT_NULL,
2738 target_name, NULL, 0, error, 0);
2743 cancel_flags |= XFS_TRANS_ABORT;
2747 xfs_trans_cancel(tp, cancel_flags);
2756 bhv_desc_t *dir_bdp,
2762 char *dir_name = VNAME(dentry);
2764 xfs_inode_t *cdp; /* inode of created dir */
2765 bhv_vnode_t *cvp; /* vnode of created dir */
2771 xfs_bmap_free_t free_list;
2772 xfs_fsblock_t first_block;
2773 bhv_vnode_t *dir_vp;
2774 boolean_t dp_joined_to_trans;
2775 boolean_t created = B_FALSE;
2776 int dm_event_sent = 0;
2778 struct xfs_dquot *udqp, *gdqp;
2783 dir_vp = BHV_TO_VNODE(dir_bdp);
2784 dp = XFS_BHVTOI(dir_bdp);
2787 if (XFS_FORCED_SHUTDOWN(mp))
2788 return XFS_ERROR(EIO);
2790 dir_namelen = VNAMELEN(dentry);
2793 dp_joined_to_trans = B_FALSE;
2794 dm_di_mode = vap->va_mode;
2796 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2797 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2798 dir_vp, DM_RIGHT_NULL, NULL,
2799 DM_RIGHT_NULL, dir_name, NULL,
2806 /* Return through std_return after this point. */
2808 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2812 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2813 prid = dp->i_d.di_projid;
2814 else if (vap->va_mask & XFS_AT_PROJID)
2815 prid = (xfs_prid_t)vap->va_projid;
2817 prid = (xfs_prid_t)dfltprid;
2820 * Make sure that we have allocated dquot(s) on disk.
2822 error = XFS_QM_DQVOPALLOC(mp, dp,
2823 current_fsuid(credp), current_fsgid(credp), prid,
2824 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2828 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2829 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2830 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2831 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2832 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2833 if (error == ENOSPC) {
2835 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2836 XFS_TRANS_PERM_LOG_RES,
2837 XFS_MKDIR_LOG_COUNT);
2845 xfs_ilock(dp, XFS_ILOCK_EXCL);
2848 * Check for directory link count overflow.
2850 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2851 error = XFS_ERROR(EMLINK);
2856 * Reserve disk quota and the inode.
2858 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2863 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2866 * create the directory inode.
2868 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2869 0, credp, prid, resblks > 0,
2872 if (error == ENOSPC)
2879 * Now we add the directory inode to the transaction.
2880 * We waited until now since xfs_dir_ialloc might start
2881 * a new transaction. Had we joined the transaction
2882 * earlier, the locks might have gotten released.
2885 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2886 dp_joined_to_trans = B_TRUE;
2888 XFS_BMAP_INIT(&free_list, &first_block);
2890 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2891 cdp->i_ino, &first_block, &free_list,
2892 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2894 ASSERT(error != ENOSPC);
2897 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2900 * Bump the in memory version number of the parent directory
2901 * so that other processes accessing it will recognize that
2902 * the directory has changed.
2906 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2912 error = xfs_bumplink(tp, dp);
2917 cvp = XFS_ITOV(cdp);
2925 * Attach the dquots to the new inode and modify the icount incore.
2927 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2930 * If this is a synchronous mount, make sure that the
2931 * mkdir transaction goes to disk before returning to
2934 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2935 xfs_trans_set_sync(tp);
2938 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2944 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2945 XFS_QM_DQRELE(mp, udqp);
2946 XFS_QM_DQRELE(mp, gdqp);
2951 /* Fall through to std_return with error = 0 or errno from
2952 * xfs_trans_commit. */
2955 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2956 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2957 DM_EVENT_POSTCREATE)) {
2958 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2959 dir_vp, DM_RIGHT_NULL,
2960 created ? XFS_ITOV(cdp):NULL,
2963 dm_di_mode, error, 0);
2969 xfs_bmap_cancel(&free_list);
2971 cancel_flags |= XFS_TRANS_ABORT;
2973 xfs_trans_cancel(tp, cancel_flags);
2974 XFS_QM_DQRELE(mp, udqp);
2975 XFS_QM_DQRELE(mp, gdqp);
2977 if (!dp_joined_to_trans && (dp != NULL)) {
2978 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2991 bhv_desc_t *dir_bdp,
2995 char *name = VNAME(dentry);
2997 xfs_inode_t *cdp; /* child directory */
3001 xfs_bmap_free_t free_list;
3002 xfs_fsblock_t first_block;
3005 bhv_vnode_t *dir_vp;
3011 dir_vp = BHV_TO_VNODE(dir_bdp);
3012 dp = XFS_BHVTOI(dir_bdp);
3015 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3017 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3018 return XFS_ERROR(EIO);
3019 namelen = VNAMELEN(dentry);
3021 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3022 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3023 dir_vp, DM_RIGHT_NULL,
3024 NULL, DM_RIGHT_NULL,
3025 name, NULL, 0, 0, 0);
3027 return XFS_ERROR(error);
3030 /* Return through std_return after this point. */
3035 * We need to get a reference to cdp before we get our log
3036 * reservation. The reason for this is that we cannot call
3037 * xfs_iget for an inode for which we do not have a reference
3038 * once we've acquired a log reservation. This is because the
3039 * inode we are trying to get might be in xfs_inactive going
3040 * for a log reservation. Since we'll have to wait for the
3041 * inactive code to complete before returning from xfs_iget,
3042 * we need to make sure that we don't have log space reserved
3043 * when we call xfs_iget. Instead we get an unlocked reference
3044 * to the inode before getting our log reservation.
3046 error = xfs_get_dir_entry(dentry, &cdp);
3048 REMOVE_DEBUG_TRACE(__LINE__);
3052 dm_di_mode = cdp->i_d.di_mode;
3055 * Get the dquots for the inodes.
3057 error = XFS_QM_DQATTACH(mp, dp, 0);
3058 if (!error && dp != cdp)
3059 error = XFS_QM_DQATTACH(mp, cdp, 0);
3062 REMOVE_DEBUG_TRACE(__LINE__);
3066 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3067 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3069 * We try to get the real space reservation first,
3070 * allowing for directory btree deletion(s) implying
3071 * possible bmap insert(s). If we can't get the space
3072 * reservation then we use 0 instead, and avoid the bmap
3073 * btree insert(s) in the directory code by, if the bmap
3074 * insert tries to happen, instead trimming the LAST
3075 * block from the directory.
3077 resblks = XFS_REMOVE_SPACE_RES(mp);
3078 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3079 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3080 if (error == ENOSPC) {
3082 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3083 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3086 ASSERT(error != ENOSPC);
3091 XFS_BMAP_INIT(&free_list, &first_block);
3094 * Now lock the child directory inode and the parent directory
3095 * inode in the proper order. This will take care of validating
3096 * that the directory entry for the child directory inode has
3097 * not changed while we were obtaining a log reservation.
3099 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3101 xfs_trans_cancel(tp, cancel_flags);
3106 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3109 * Only increment the parent directory vnode count if
3110 * we didn't bump it in looking up cdp. The only time
3111 * we don't bump it is when we're looking up ".".
3117 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3119 ASSERT(cdp->i_d.di_nlink >= 2);
3120 if (cdp->i_d.di_nlink != 2) {
3121 error = XFS_ERROR(ENOTEMPTY);
3124 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3125 error = XFS_ERROR(ENOTEMPTY);
3129 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3130 &first_block, &free_list, resblks);
3135 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3138 * Bump the in memory generation count on the parent
3139 * directory so that other can know that it has changed.
3144 * Drop the link from cdp's "..".
3146 error = xfs_droplink(tp, dp);
3152 * Drop the link from dp to cdp.
3154 error = xfs_droplink(tp, cdp);
3160 * Drop the "." link from cdp to self.
3162 error = xfs_droplink(tp, cdp);
3167 /* Determine these before committing transaction */
3168 last_cdp_link = (cdp)->i_d.di_nlink==0;
3171 * Take an extra ref on the child vnode so that it
3172 * does not go to xfs_inactive() from within the commit.
3177 * If this is a synchronous mount, make sure that the
3178 * rmdir transaction goes to disk before returning to
3181 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3182 xfs_trans_set_sync(tp);
3185 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3187 xfs_bmap_cancel(&free_list);
3188 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3194 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3202 * Let interposed file systems know about removed links.
3204 bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3208 /* Fall through to std_return with error = 0 or the errno
3209 * from xfs_trans_commit. */
3211 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3212 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3213 dir_vp, DM_RIGHT_NULL,
3214 NULL, DM_RIGHT_NULL,
3215 name, NULL, dm_di_mode,
3221 xfs_bmap_cancel(&free_list);
3222 cancel_flags |= XFS_TRANS_ABORT;
3226 xfs_trans_cancel(tp, cancel_flags);
3234 * Read dp's entries starting at uiop->uio_offset and translate them into
3235 * bufsize bytes worth of struct dirents starting at bufbase.
3239 bhv_desc_t *dir_bdp,
3245 xfs_trans_t *tp = NULL;
3249 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3250 (inst_t *)__return_address);
3251 dp = XFS_BHVTOI(dir_bdp);
3253 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3254 return XFS_ERROR(EIO);
3257 lock_mode = xfs_ilock_map_shared(dp);
3258 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3259 xfs_iunlock_map_shared(dp, lock_mode);
3270 bhv_desc_t *dir_bdp,
3283 xfs_bmap_free_t free_list;
3284 xfs_fsblock_t first_block;
3285 boolean_t dp_joined_to_trans;
3286 bhv_vnode_t *dir_vp;
3289 xfs_fileoff_t first_fsb;
3290 xfs_filblks_t fs_blocks;
3292 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3299 struct xfs_dquot *udqp, *gdqp;
3301 char *link_name = VNAME(dentry);
3305 dir_vp = BHV_TO_VNODE(dir_bdp);
3306 dp = XFS_BHVTOI(dir_bdp);
3307 dp_joined_to_trans = B_FALSE;
3312 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3316 if (XFS_FORCED_SHUTDOWN(mp))
3317 return XFS_ERROR(EIO);
3319 link_namelen = VNAMELEN(dentry);
3322 * Check component lengths of the target path name.
3324 pathlen = strlen(target_path);
3325 if (pathlen >= MAXPATHLEN) /* total string too long */
3326 return XFS_ERROR(ENAMETOOLONG);
3327 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3331 for(total = 0, path = target_path; total < pathlen;) {
3335 while(*path == '/') {
3341 * Count up to the next slash or end of path.
3342 * Error out if the component is bigger than MAXNAMELEN.
3344 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3345 if (++len >= MAXNAMELEN) {
3346 error = ENAMETOOLONG;
3353 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3354 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3355 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3356 link_name, target_path, 0, 0, 0);
3361 /* Return through std_return after this point. */
3364 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3365 prid = dp->i_d.di_projid;
3366 else if (vap->va_mask & XFS_AT_PROJID)
3367 prid = (xfs_prid_t)vap->va_projid;
3369 prid = (xfs_prid_t)dfltprid;
3372 * Make sure that we have allocated dquot(s) on disk.
3374 error = XFS_QM_DQVOPALLOC(mp, dp,
3375 current_fsuid(credp), current_fsgid(credp), prid,
3376 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3380 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3381 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3383 * The symlink will fit into the inode data fork?
3384 * There can't be any attributes so we get the whole variable part.
3386 if (pathlen <= XFS_LITINO(mp))
3389 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3390 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3391 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3392 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3393 if (error == ENOSPC && fs_blocks == 0) {
3395 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3396 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3404 xfs_ilock(dp, XFS_ILOCK_EXCL);
3407 * Check whether the directory allows new symlinks or not.
3409 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3410 error = XFS_ERROR(EPERM);
3415 * Reserve disk quota : blocks and inode.
3417 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3422 * Check for ability to enter directory entry, if no space reserved.
3425 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3428 * Initialize the bmap freelist prior to calling either
3429 * bmapi or the directory create code.
3431 XFS_BMAP_INIT(&free_list, &first_block);
3434 * Allocate an inode for the symlink.
3436 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3437 1, 0, credp, prid, resblks > 0, &ip, NULL);
3439 if (error == ENOSPC)
3446 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3447 dp_joined_to_trans = B_TRUE;
3450 * Also attach the dquot(s) to it, if applicable.
3452 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3455 resblks -= XFS_IALLOC_SPACE_RES(mp);
3457 * If the symlink will fit into the inode, write it inline.
3459 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3460 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3461 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3462 ip->i_d.di_size = pathlen;
3465 * The inode was initially created in extent format.
3467 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3468 ip->i_df.if_flags |= XFS_IFINLINE;
3470 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3471 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3475 nmaps = SYMLINK_MAPS;
3477 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3478 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3479 &first_block, resblks, mval, &nmaps,
3486 resblks -= fs_blocks;
3487 ip->i_d.di_size = pathlen;
3488 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3490 cur_chunk = target_path;
3491 for (n = 0; n < nmaps; n++) {
3492 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3493 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3494 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3495 BTOBB(byte_cnt), 0);
3496 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3497 if (pathlen < byte_cnt) {
3500 pathlen -= byte_cnt;
3502 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3503 cur_chunk += byte_cnt;
3505 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3510 * Create the directory entry for the symlink.
3512 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3513 ip->i_ino, &first_block, &free_list, resblks);
3517 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3518 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3521 * Bump the in memory version number of the parent directory
3522 * so that other processes accessing it will recognize that
3523 * the directory has changed.
3528 * If this is a synchronous mount, make sure that the
3529 * symlink transaction goes to disk before returning to
3532 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3533 xfs_trans_set_sync(tp);
3537 * xfs_trans_commit normally decrements the vnode ref count
3538 * when it unlocks the inode. Since we want to return the
3539 * vnode to the caller, we bump the vnode ref count now.
3543 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3547 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3548 XFS_QM_DQRELE(mp, udqp);
3549 XFS_QM_DQRELE(mp, gdqp);
3551 /* Fall through to std_return with error = 0 or errno from
3552 * xfs_trans_commit */
3554 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3555 DM_EVENT_POSTSYMLINK)) {
3556 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3557 dir_vp, DM_RIGHT_NULL,
3558 error ? NULL : XFS_ITOV(ip),
3559 DM_RIGHT_NULL, link_name, target_path,
3575 xfs_bmap_cancel(&free_list);
3576 cancel_flags |= XFS_TRANS_ABORT;
3578 xfs_trans_cancel(tp, cancel_flags);
3579 XFS_QM_DQRELE(mp, udqp);
3580 XFS_QM_DQRELE(mp, gdqp);
3582 if (!dp_joined_to_trans && (dp != NULL)) {
3583 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3593 * A fid routine that takes a pointer to a previously allocated
3594 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3604 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3605 (inst_t *)__return_address);
3606 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3608 xfid = (xfs_fid2_t *)fidp;
3609 ip = XFS_BHVTOI(bdp);
3610 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3613 * use memcpy because the inode is a long long and there's no
3614 * assurance that xfid->fid_ino is properly aligned.
3616 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3617 xfid->fid_gen = ip->i_d.di_gen;
3634 vp = BHV_TO_VNODE(bdp);
3637 ip = XFS_BHVTOI(bdp);
3638 if (locktype == VRWLOCK_WRITE) {
3639 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3640 } else if (locktype == VRWLOCK_TRY_READ) {
3641 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3642 } else if (locktype == VRWLOCK_TRY_WRITE) {
3643 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3645 ASSERT((locktype == VRWLOCK_READ) ||
3646 (locktype == VRWLOCK_WRITE_DIRECT));
3647 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3665 vp = BHV_TO_VNODE(bdp);
3668 ip = XFS_BHVTOI(bdp);
3669 if (locktype == VRWLOCK_WRITE) {
3671 * In the write case, we may have added a new entry to
3672 * the reference cache. This might store a pointer to
3673 * an inode to be released in this inode. If it is there,
3674 * clear the pointer and release the inode after unlocking
3677 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3679 ASSERT((locktype == VRWLOCK_READ) ||
3680 (locktype == VRWLOCK_WRITE_DIRECT));
3681 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3693 xfs_inode_log_item_t *iip;
3696 ip = XFS_BHVTOI(bdp);
3700 if (XFS_FORCED_SHUTDOWN(mp))
3701 return XFS_ERROR(EIO);
3704 * Bypass inodes which have already been cleaned by
3705 * the inode flush clustering code inside xfs_iflush
3707 if ((ip->i_update_core == 0) &&
3708 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3711 if (flags & FLUSH_LOG) {
3712 if (iip && iip->ili_last_lsn) {
3713 xlog_t *log = mp->m_log;
3715 int s, log_flags = XFS_LOG_FORCE;
3717 s = GRANT_LOCK(log);
3718 sync_lsn = log->l_last_sync_lsn;
3719 GRANT_UNLOCK(log, s);
3721 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3724 if (flags & FLUSH_SYNC)
3725 log_flags |= XFS_LOG_SYNC;
3726 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3731 * We make this non-blocking if the inode is contended,
3732 * return EAGAIN to indicate to the caller that they
3733 * did not succeed. This prevents the flush path from
3734 * blocking on inodes inside another operation right
3735 * now, they get caught later by xfs_sync.
3737 if (flags & FLUSH_INODE) {
3740 if (xfs_ipincount(ip))
3743 if (flags & FLUSH_SYNC) {
3744 xfs_ilock(ip, XFS_ILOCK_SHARED);
3746 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3747 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3748 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3755 if (flags & FLUSH_SYNC)
3756 flush_flags = XFS_IFLUSH_SYNC;
3758 flush_flags = XFS_IFLUSH_ASYNC;
3760 error = xfs_iflush(ip, flush_flags);
3761 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3779 if (!capable(CAP_SYS_ADMIN))
3780 return XFS_ERROR(EPERM);
3782 ip = XFS_BHVTOI(bdp);
3785 if (XFS_FORCED_SHUTDOWN(mp))
3786 return XFS_ERROR(EIO);
3788 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3789 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3791 xfs_trans_cancel(tp, 0);
3794 xfs_ilock(ip, XFS_ILOCK_EXCL);
3795 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3797 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3798 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3800 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3802 error = xfs_trans_commit(tp, 0, NULL);
3814 vp = BHV_TO_VNODE(bdp);
3815 ip = XFS_BHVTOI(bdp);
3817 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3819 ASSERT(!VN_MAPPED(vp));
3821 /* bad inode, get out here ASAP */
3829 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3832 * Make sure the atime in the XFS inode is correct before freeing the
3835 xfs_synchronize_atime(ip);
3837 /* If we have nothing to flush with this inode then complete the
3838 * teardown now, otherwise break the link between the xfs inode
3839 * and the linux inode and clean up the xfs inode later. This
3840 * avoids flushing the inode to disk during the delete operation
3843 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3844 xfs_ilock(ip, XFS_ILOCK_EXCL);
3846 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3848 xfs_mount_t *mp = ip->i_mount;
3850 /* Protect sync from us */
3851 XFS_MOUNT_ILOCK(mp);
3852 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3853 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3854 ip->i_flags |= XFS_IRECLAIMABLE;
3855 XFS_MOUNT_IUNLOCK(mp);
3866 xfs_ihash_t *ih = ip->i_hash;
3867 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
3870 if (vp && VN_BAD(vp))
3873 /* The hash lock here protects a thread in xfs_iget_core from
3874 * racing with us on linking the inode back with a vnode.
3875 * Once we have the XFS_IRECLAIM flag set it will not touch
3878 write_lock(&ih->ih_lock);
3879 if ((ip->i_flags & XFS_IRECLAIM) ||
3880 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3881 write_unlock(&ih->ih_lock);
3884 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3888 ip->i_flags |= XFS_IRECLAIM;
3889 write_unlock(&ih->ih_lock);
3892 * If the inode is still dirty, then flush it out. If the inode
3893 * is not in the AIL, then it will be OK to flush it delwri as
3894 * long as xfs_iflush() does not keep any references to the inode.
3895 * We leave that decision up to xfs_iflush() since it has the
3896 * knowledge of whether it's OK to simply do a delwri flush of
3897 * the inode or whether we need to wait until the inode is
3898 * pulled from the AIL.
3899 * We get the flush lock regardless, though, just to make sure
3900 * we don't free it while it is being flushed.
3902 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3904 xfs_ilock(ip, XFS_ILOCK_EXCL);
3908 if (ip->i_update_core ||
3909 ((ip->i_itemp != NULL) &&
3910 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3911 error = xfs_iflush(ip, sync_mode);
3913 * If we hit an error, typically because of filesystem
3914 * shutdown, we don't need to let vn_reclaim to know
3915 * because we're gonna reclaim the inode anyway.
3918 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3921 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3924 ASSERT(ip->i_update_core == 0);
3925 ASSERT(ip->i_itemp == NULL ||
3926 ip->i_itemp->ili_format.ilf_fields == 0);
3927 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3928 } else if (locked) {
3930 * We are not interested in doing an iflush if we're
3931 * in the process of shutting down the filesystem forcibly.
3932 * So, just reclaim the inode.
3935 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3944 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3947 xfs_inode_t *ip, *n;
3952 XFS_MOUNT_ILOCK(mp);
3953 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3955 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3957 if (xfs_ipincount(ip) ||
3958 !xfs_iflock_nowait(ip)) {
3959 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3963 XFS_MOUNT_IUNLOCK(mp);
3964 if (xfs_finish_reclaim(ip, noblock,
3965 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3974 XFS_MOUNT_IUNLOCK(mp);
3979 * xfs_alloc_file_space()
3980 * This routine allocates disk space for the given file.
3982 * If alloc_type == 0, this request is for an ALLOCSP type
3983 * request which will change the file size. In this case, no
3984 * DMAPI event will be generated by the call. A TRUNCATE event
3985 * will be generated later by xfs_setattr.
3987 * If alloc_type != 0, this request is for a RESVSP type
3988 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3989 * lower block boundary byte address is less than the file's
3998 xfs_alloc_file_space(
4005 xfs_mount_t *mp = ip->i_mount;
4007 xfs_filblks_t allocated_fsb;
4008 xfs_filblks_t allocatesize_fsb;
4009 xfs_extlen_t extsz, temp;
4010 xfs_fileoff_t startoffset_fsb;
4011 xfs_fsblock_t firstfsb;
4017 xfs_bmbt_irec_t imaps[1], *imapp;
4018 xfs_bmap_free_t free_list;
4019 uint qblocks, resblks, resrtextents;
4023 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4025 if (XFS_FORCED_SHUTDOWN(mp))
4026 return XFS_ERROR(EIO);
4028 rt = XFS_IS_REALTIME_INODE(ip);
4030 if (!(extsz = ip->i_d.di_extsize))
4031 extsz = mp->m_sb.sb_rextsize;
4033 extsz = ip->i_d.di_extsize;
4036 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4040 return XFS_ERROR(EINVAL);
4046 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4047 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4048 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4050 /* Generate a DMAPI event if needed. */
4051 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4052 (attr_flags&ATTR_DMI) == 0 &&
4053 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4054 xfs_off_t end_dmi_offset;
4056 end_dmi_offset = offset+len;
4057 if (end_dmi_offset > ip->i_d.di_size)
4058 end_dmi_offset = ip->i_d.di_size;
4059 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4060 offset, end_dmi_offset - offset,
4067 * Allocate file space until done or until there is an error
4070 while (allocatesize_fsb && !error) {
4074 * Determine space reservations for data/realtime.
4076 if (unlikely(extsz)) {
4077 s = startoffset_fsb;
4080 e = startoffset_fsb + allocatesize_fsb;
4081 if ((temp = do_mod(startoffset_fsb, extsz)))
4083 if ((temp = do_mod(e, extsz)))
4087 e = allocatesize_fsb;
4091 resrtextents = qblocks = (uint)(e - s);
4092 resrtextents /= mp->m_sb.sb_rextsize;
4093 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4094 quota_flag = XFS_QMOPT_RES_RTBLKS;
4097 resblks = qblocks = \
4098 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4099 quota_flag = XFS_QMOPT_RES_REGBLKS;
4103 * Allocate and setup the transaction.
4105 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4106 error = xfs_trans_reserve(tp, resblks,
4107 XFS_WRITE_LOG_RES(mp), resrtextents,
4108 XFS_TRANS_PERM_LOG_RES,
4109 XFS_WRITE_LOG_COUNT);
4111 * Check for running out of space
4115 * Free the transaction structure.
4117 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4118 xfs_trans_cancel(tp, 0);
4121 xfs_ilock(ip, XFS_ILOCK_EXCL);
4122 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4123 qblocks, 0, quota_flag);
4127 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4128 xfs_trans_ihold(tp, ip);
4131 * Issue the xfs_bmapi() call to allocate the blocks
4133 XFS_BMAP_INIT(&free_list, &firstfsb);
4134 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4135 allocatesize_fsb, bmapi_flag,
4136 &firstfsb, 0, imapp, &nimaps,
4143 * Complete the transaction
4145 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4150 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4151 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4156 allocated_fsb = imapp->br_blockcount;
4159 error = XFS_ERROR(ENOSPC);
4163 startoffset_fsb += allocated_fsb;
4164 allocatesize_fsb -= allocated_fsb;
4167 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4168 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4170 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4171 XFS_ITOV(ip), DM_RIGHT_NULL,
4172 XFS_ITOV(ip), DM_RIGHT_NULL,
4173 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4175 goto retry; /* Maybe DMAPI app. has made space */
4176 /* else fall through with error from XFS_SEND_DATA */
4181 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4182 xfs_bmap_cancel(&free_list);
4183 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4185 error1: /* Just cancel transaction */
4186 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4187 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4188 goto dmapi_enospc_check;
4192 * Zero file bytes between startoff and endoff inclusive.
4193 * The iolock is held exclusive and no blocks are buffered.
4196 xfs_zero_remaining_bytes(
4201 xfs_bmbt_irec_t imap;
4202 xfs_fileoff_t offset_fsb;
4203 xfs_off_t lastoffset;
4206 xfs_mount_t *mp = ip->i_mount;
4210 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4211 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4212 mp->m_rtdev_targp : mp->m_ddev_targp);
4214 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4215 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4217 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4218 NULL, 0, &imap, &nimap, NULL, NULL);
4219 if (error || nimap < 1)
4221 ASSERT(imap.br_blockcount >= 1);
4222 ASSERT(imap.br_startoff == offset_fsb);
4223 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4224 if (lastoffset > endoff)
4225 lastoffset = endoff;
4226 if (imap.br_startblock == HOLESTARTBLOCK)
4228 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4229 if (imap.br_state == XFS_EXT_UNWRITTEN)
4232 XFS_BUF_UNWRITE(bp);
4234 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4236 if ((error = xfs_iowait(bp))) {
4237 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4238 mp, bp, XFS_BUF_ADDR(bp));
4241 memset(XFS_BUF_PTR(bp) +
4242 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4243 0, lastoffset - offset + 1);
4248 if ((error = xfs_iowait(bp))) {
4249 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4250 mp, bp, XFS_BUF_ADDR(bp));
4259 * xfs_free_file_space()
4260 * This routine frees disk space for the given file.
4262 * This routine is only called by xfs_change_file_space
4263 * for an UNRESVSP type call.
4271 xfs_free_file_space(
4280 xfs_off_t end_dmi_offset;
4281 xfs_fileoff_t endoffset_fsb;
4283 xfs_fsblock_t firstfsb;
4284 xfs_bmap_free_t free_list;
4286 xfs_bmbt_irec_t imap;
4294 xfs_fileoff_t startoffset_fsb;
4296 int need_iolock = 1;
4301 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4303 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4307 if (len <= 0) /* if nothing being freed */
4309 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4310 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4311 end_dmi_offset = offset + len;
4312 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4314 if (offset < ip->i_d.di_size &&
4315 (attr_flags & ATTR_DMI) == 0 &&
4316 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4317 if (end_dmi_offset > ip->i_d.di_size)
4318 end_dmi_offset = ip->i_d.di_size;
4319 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4320 offset, end_dmi_offset - offset,
4321 AT_DELAY_FLAG(attr_flags), NULL);
4326 if (attr_flags & ATTR_NOLOCK)
4329 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4330 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4333 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4335 ilen = len + (offset & (rounding - 1));
4336 ioffset = offset & ~(rounding - 1);
4337 if (ilen & (rounding - 1))
4338 ilen = (ilen + rounding) & ~(rounding - 1);
4340 if (VN_CACHED(vp) != 0) {
4341 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4342 ctooff(offtoct(ioffset)), -1);
4343 bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
4344 -1, FI_REMAPF_LOCKED);
4348 * Need to zero the stuff we're not freeing, on disk.
4349 * If its a realtime file & can't use unwritten extents then we
4350 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4351 * will take care of it for us.
4353 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4355 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4356 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4358 goto out_unlock_iolock;
4359 ASSERT(nimap == 0 || nimap == 1);
4360 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4363 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4364 block = imap.br_startblock;
4365 mod = do_div(block, mp->m_sb.sb_rextsize);
4367 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4370 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4371 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4373 goto out_unlock_iolock;
4374 ASSERT(nimap == 0 || nimap == 1);
4375 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4376 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4378 if (mod && (mod != mp->m_sb.sb_rextsize))
4379 endoffset_fsb -= mod;
4382 if ((done = (endoffset_fsb <= startoffset_fsb)))
4384 * One contiguous piece to clear
4386 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4389 * Some full blocks, possibly two pieces to clear
4391 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4392 error = xfs_zero_remaining_bytes(ip, offset,
4393 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4395 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4396 error = xfs_zero_remaining_bytes(ip,
4397 XFS_FSB_TO_B(mp, endoffset_fsb),
4402 * free file space until done or until there is an error
4404 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4405 while (!error && !done) {
4408 * allocate and setup the transaction
4410 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4411 error = xfs_trans_reserve(tp,
4413 XFS_WRITE_LOG_RES(mp),
4415 XFS_TRANS_PERM_LOG_RES,
4416 XFS_WRITE_LOG_COUNT);
4419 * check for running out of space
4423 * Free the transaction structure.
4425 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4426 xfs_trans_cancel(tp, 0);
4429 xfs_ilock(ip, XFS_ILOCK_EXCL);
4430 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4431 ip->i_udquot, ip->i_gdquot, resblks, 0,
4432 XFS_QMOPT_RES_REGBLKS);
4436 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4437 xfs_trans_ihold(tp, ip);
4440 * issue the bunmapi() call to free the blocks
4442 XFS_BMAP_INIT(&free_list, &firstfsb);
4443 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4444 endoffset_fsb - startoffset_fsb,
4445 0, 2, &firstfsb, &free_list, NULL, &done);
4451 * complete the transaction
4453 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4458 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4459 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4464 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4468 xfs_bmap_cancel(&free_list);
4470 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4471 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4477 * xfs_change_file_space()
4478 * This routine allocates or frees disk space for the given file.
4479 * The user specified parameters are checked for alignment and size
4488 xfs_change_file_space(
4502 xfs_off_t startoffset;
4508 vp = BHV_TO_VNODE(bdp);
4509 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4511 ip = XFS_BHVTOI(bdp);
4515 * must be a regular file and have write permission
4518 return XFS_ERROR(EINVAL);
4520 xfs_ilock(ip, XFS_ILOCK_SHARED);
4522 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4523 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4527 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4529 switch (bf->l_whence) {
4530 case 0: /*SEEK_SET*/
4532 case 1: /*SEEK_CUR*/
4533 bf->l_start += offset;
4535 case 2: /*SEEK_END*/
4536 bf->l_start += ip->i_d.di_size;
4539 return XFS_ERROR(EINVAL);
4542 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4544 if ( (bf->l_start < 0)
4545 || (bf->l_start > XFS_MAXIOFFSET(mp))
4546 || (bf->l_start + llen < 0)
4547 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4548 return XFS_ERROR(EINVAL);
4552 startoffset = bf->l_start;
4553 fsize = ip->i_d.di_size;
4556 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4558 * These calls do NOT zero the data space allocated to the file,
4559 * nor do they change the file size.
4561 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4563 * These calls cause the new file data to be zeroed and the file
4564 * size to be changed.
4566 setprealloc = clrprealloc = 0;
4569 case XFS_IOC_RESVSP:
4570 case XFS_IOC_RESVSP64:
4571 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4578 case XFS_IOC_UNRESVSP:
4579 case XFS_IOC_UNRESVSP64:
4580 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4585 case XFS_IOC_ALLOCSP:
4586 case XFS_IOC_ALLOCSP64:
4587 case XFS_IOC_FREESP:
4588 case XFS_IOC_FREESP64:
4589 if (startoffset > fsize) {
4590 error = xfs_alloc_file_space(ip, fsize,
4591 startoffset - fsize, 0, attr_flags);
4596 va.va_mask = XFS_AT_SIZE;
4597 va.va_size = startoffset;
4599 error = xfs_setattr(bdp, &va, attr_flags, credp);
4609 return XFS_ERROR(EINVAL);
4613 * update the inode timestamp, mode, and prealloc flag bits
4615 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4617 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4620 xfs_trans_cancel(tp, 0);
4624 xfs_ilock(ip, XFS_ILOCK_EXCL);
4626 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4627 xfs_trans_ihold(tp, ip);
4629 if ((attr_flags & ATTR_DMI) == 0) {
4630 ip->i_d.di_mode &= ~S_ISUID;
4633 * Note that we don't have to worry about mandatory
4634 * file locking being disabled here because we only
4635 * clear the S_ISGID bit if the Group execute bit is
4636 * on, but if it was on then mandatory locking wouldn't
4637 * have been enabled.
4639 if (ip->i_d.di_mode & S_IXGRP)
4640 ip->i_d.di_mode &= ~S_ISGID;
4642 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4645 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4646 else if (clrprealloc)
4647 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4649 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4650 xfs_trans_set_sync(tp);
4652 error = xfs_trans_commit(tp, 0, NULL);
4654 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4659 bhv_vnodeops_t xfs_vnodeops = {
4660 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4661 .vop_open = xfs_open,
4662 .vop_close = xfs_close,
4663 .vop_read = xfs_read,
4664 #ifdef HAVE_SENDFILE
4665 .vop_sendfile = xfs_sendfile,
4668 .vop_splice_read = xfs_splice_read,
4669 .vop_splice_write = xfs_splice_write,
4671 .vop_write = xfs_write,
4672 .vop_ioctl = xfs_ioctl,
4673 .vop_getattr = xfs_getattr,
4674 .vop_setattr = xfs_setattr,
4675 .vop_access = xfs_access,
4676 .vop_lookup = xfs_lookup,
4677 .vop_create = xfs_create,
4678 .vop_remove = xfs_remove,
4679 .vop_link = xfs_link,
4680 .vop_rename = xfs_rename,
4681 .vop_mkdir = xfs_mkdir,
4682 .vop_rmdir = xfs_rmdir,
4683 .vop_readdir = xfs_readdir,
4684 .vop_symlink = xfs_symlink,
4685 .vop_readlink = xfs_readlink,
4686 .vop_fsync = xfs_fsync,
4687 .vop_inactive = xfs_inactive,
4688 .vop_fid2 = xfs_fid2,
4689 .vop_rwlock = xfs_rwlock,
4690 .vop_rwunlock = xfs_rwunlock,
4691 .vop_bmap = xfs_bmap,
4692 .vop_reclaim = xfs_reclaim,
4693 .vop_attr_get = xfs_attr_get,
4694 .vop_attr_set = xfs_attr_set,
4695 .vop_attr_remove = xfs_attr_remove,
4696 .vop_attr_list = xfs_attr_list,
4697 .vop_link_removed = (vop_link_removed_t)fs_noval,
4698 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4699 .vop_tosspages = fs_tosspages,
4700 .vop_flushinval_pages = fs_flushinval_pages,
4701 .vop_flush_pages = fs_flush_pages,
4702 .vop_release = xfs_release,
4703 .vop_iflush = xfs_inode_flush,