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"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
47 #include "xfs_error.h"
48 #include "xfs_quota.h"
49 #include "xfs_utils.h"
50 #include "xfs_rtalloc.h"
51 #include "xfs_refcache.h"
52 #include "xfs_trans_space.h"
53 #include "xfs_log_priv.h"
54 #include "xfs_filestream.h"
62 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
63 xfs_inode_t *ip = XFS_BHVTOI(bdp);
65 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
66 return XFS_ERROR(EIO);
69 * If it's a directory with any blocks, read-ahead block 0
70 * as we're almost certain to have the next operation be a read there.
72 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
73 mode = xfs_ilock_map_shared(ip);
74 if (ip->i_d.di_nextents > 0)
75 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
76 xfs_iunlock(ip, mode);
95 vp = BHV_TO_VNODE(bdp);
96 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
101 if (XFS_FORCED_SHUTDOWN(mp))
102 return XFS_ERROR(EIO);
104 if (!(flags & ATTR_LAZY))
105 xfs_ilock(ip, XFS_ILOCK_SHARED);
107 vap->va_size = XFS_ISIZE(ip);
108 if (vap->va_mask == XFS_AT_SIZE)
112 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
113 vap->va_nodeid = ip->i_ino;
115 vap->va_nodeid += mp->m_inoadd;
117 vap->va_nlink = ip->i_d.di_nlink;
120 * Quick exit for non-stat callers
123 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
124 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
128 * Copy from in-core inode.
130 vap->va_mode = ip->i_d.di_mode;
131 vap->va_uid = ip->i_d.di_uid;
132 vap->va_gid = ip->i_d.di_gid;
133 vap->va_projid = ip->i_d.di_projid;
136 * Check vnode type block/char vs. everything else.
138 switch (ip->i_d.di_mode & S_IFMT) {
141 vap->va_rdev = ip->i_df.if_u2.if_rdev;
142 vap->va_blocksize = BLKDEV_IOSIZE;
147 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
148 vap->va_blocksize = xfs_preferred_iosize(mp);
152 * If the file blocks are being allocated from a
153 * realtime partition, then return the inode's
154 * realtime extent size or the realtime volume's
158 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
163 vn_atime_to_timespec(vp, &vap->va_atime);
164 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
165 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
166 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
167 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
170 * Exit for stat callers. See if any of the rest of the fields
171 * to be filled in are needed.
174 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
175 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
179 * Convert di_flags to xflags.
181 vap->va_xflags = xfs_ip2xflags(ip);
184 * Exit for inode revalidate. See if any of the rest of
185 * the fields to be filled in are needed.
188 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
189 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
192 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
194 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
195 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
199 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
200 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
201 ip->i_d.di_anextents;
203 vap->va_anextents = 0;
204 vap->va_gen = ip->i_d.di_gen;
207 if (!(flags & ATTR_LAZY))
208 xfs_iunlock(ip, XFS_ILOCK_SHARED);
234 xfs_prid_t projid=0, iprojid=0;
235 int mandlock_before, mandlock_after;
236 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
240 vp = BHV_TO_VNODE(bdp);
241 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
243 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
244 return XFS_ERROR(EROFS);
247 * Cannot set certain attributes.
250 if (mask & XFS_AT_NOSET) {
251 return XFS_ERROR(EINVAL);
254 ip = XFS_BHVTOI(bdp);
257 if (XFS_FORCED_SHUTDOWN(mp))
258 return XFS_ERROR(EIO);
261 * Timestamps do not need to be logged and hence do not
262 * need to be done within a transaction.
264 if (mask & XFS_AT_UPDTIMES) {
265 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
266 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
267 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
268 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
269 xfs_ichgtime(ip, timeflags);
273 olddquot1 = olddquot2 = NULL;
277 * If disk quotas is on, we make sure that the dquots do exist on disk,
278 * before we start any other transactions. Trying to do this later
279 * is messy. We don't care to take a readlock to look at the ids
280 * in inode here, because we can't hold it across the trans_reserve.
281 * If the IDs do change before we take the ilock, we're covered
282 * because the i_*dquot fields will get updated anyway.
284 if (XFS_IS_QUOTA_ON(mp) &&
285 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
288 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
290 qflags |= XFS_QMOPT_UQUOTA;
292 uid = ip->i_d.di_uid;
294 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
296 qflags |= XFS_QMOPT_GQUOTA;
298 gid = ip->i_d.di_gid;
300 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
301 projid = vap->va_projid;
302 qflags |= XFS_QMOPT_PQUOTA;
304 projid = ip->i_d.di_projid;
307 * We take a reference when we initialize udqp and gdqp,
308 * so it is important that we never blindly double trip on
309 * the same variable. See xfs_create() for an example.
311 ASSERT(udqp == NULL);
312 ASSERT(gdqp == NULL);
313 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
320 * For the other attributes, we acquire the inode lock and
321 * first do an error checking pass.
324 lock_flags = XFS_ILOCK_EXCL;
325 if (flags & ATTR_NOLOCK)
327 if (!(mask & XFS_AT_SIZE)) {
328 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
329 (mp->m_flags & XFS_MOUNT_WSYNC)) {
330 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
332 if ((code = xfs_trans_reserve(tp, 0,
333 XFS_ICHANGE_LOG_RES(mp), 0,
340 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
341 !(flags & ATTR_DMI)) {
342 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
343 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
344 vap->va_size, 0, dmflags, NULL);
351 lock_flags |= XFS_IOLOCK_EXCL;
354 xfs_ilock(ip, lock_flags);
356 /* boolean: are we the file owner? */
357 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
360 * Change various properties of a file.
361 * Only the owner or users with CAP_FOWNER
362 * capability may do these things.
365 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
366 XFS_AT_GID|XFS_AT_PROJID)) {
368 * CAP_FOWNER overrides the following restrictions:
370 * The user ID of the calling process must be equal
371 * to the file owner ID, except in cases where the
372 * CAP_FSETID capability is applicable.
374 if (!file_owner && !capable(CAP_FOWNER)) {
375 code = XFS_ERROR(EPERM);
380 * CAP_FSETID overrides the following restrictions:
382 * The effective user ID of the calling process shall match
383 * the file owner when setting the set-user-ID and
384 * set-group-ID bits on that file.
386 * The effective group ID or one of the supplementary group
387 * IDs of the calling process shall match the group owner of
388 * the file when setting the set-group-ID bit on that file
390 if (mask & XFS_AT_MODE) {
393 if ((vap->va_mode & S_ISUID) && !file_owner)
395 if ((vap->va_mode & S_ISGID) &&
396 !in_group_p((gid_t)ip->i_d.di_gid))
399 /* Linux allows this, Irix doesn't. */
400 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
403 if (m && !capable(CAP_FSETID))
409 * Change file ownership. Must be the owner or privileged.
410 * If the system was configured with the "restricted_chown"
411 * option, the owner is not permitted to give away the file,
412 * and can change the group id only to a group of which he
413 * or she is a member.
415 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
417 * These IDs could have changed since we last looked at them.
418 * But, we're assured that if the ownership did change
419 * while we didn't have the inode locked, inode's dquot(s)
420 * would have changed also.
422 iuid = ip->i_d.di_uid;
423 iprojid = ip->i_d.di_projid;
424 igid = ip->i_d.di_gid;
425 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
426 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
427 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
431 * CAP_CHOWN overrides the following restrictions:
433 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
434 * shall override the restriction that a process cannot
435 * change the user ID of a file it owns and the restriction
436 * that the group ID supplied to the chown() function
437 * shall be equal to either the group ID or one of the
438 * supplementary group IDs of the calling process.
440 if (restricted_chown &&
441 (iuid != uid || (igid != gid &&
442 !in_group_p((gid_t)gid))) &&
443 !capable(CAP_CHOWN)) {
444 code = XFS_ERROR(EPERM);
448 * Do a quota reservation only if uid/projid/gid is actually
451 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
452 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
453 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
455 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
456 capable(CAP_FOWNER) ?
457 XFS_QMOPT_FORCE_RES : 0);
458 if (code) /* out of quota */
464 * Truncate file. Must have write permission and not be a directory.
466 if (mask & XFS_AT_SIZE) {
467 /* Short circuit the truncate case for zero length files */
468 if ((vap->va_size == 0) &&
469 (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) {
470 xfs_iunlock(ip, XFS_ILOCK_EXCL);
471 lock_flags &= ~XFS_ILOCK_EXCL;
472 if (mask & XFS_AT_CTIME)
473 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
479 code = XFS_ERROR(EISDIR);
481 } else if (!VN_ISREG(vp)) {
482 code = XFS_ERROR(EINVAL);
486 * Make sure that the dquots are attached to the inode.
488 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
493 * Change file access or modified times.
495 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
497 if ((flags & ATTR_UTIME) &&
498 !capable(CAP_FOWNER)) {
499 code = XFS_ERROR(EPERM);
506 * Change extent size or realtime flag.
508 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
510 * Can't change extent size if any extents are allocated.
512 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
513 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
515 code = XFS_ERROR(EINVAL); /* EFBIG? */
520 * Can't change realtime flag if any extents are allocated.
522 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
523 (mask & XFS_AT_XFLAGS) &&
524 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
525 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
526 code = XFS_ERROR(EINVAL); /* EFBIG? */
530 * Extent size must be a multiple of the appropriate block
531 * size, if set at all.
533 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
536 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
537 ((mask & XFS_AT_XFLAGS) &&
538 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
539 size = mp->m_sb.sb_rextsize <<
540 mp->m_sb.sb_blocklog;
542 size = mp->m_sb.sb_blocksize;
544 if (vap->va_extsize % size) {
545 code = XFS_ERROR(EINVAL);
550 * If realtime flag is set then must have realtime data.
552 if ((mask & XFS_AT_XFLAGS) &&
553 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
554 if ((mp->m_sb.sb_rblocks == 0) ||
555 (mp->m_sb.sb_rextsize == 0) ||
556 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
557 code = XFS_ERROR(EINVAL);
563 * Can't modify an immutable/append-only file unless
564 * we have appropriate permission.
566 if ((mask & XFS_AT_XFLAGS) &&
568 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
570 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
571 !capable(CAP_LINUX_IMMUTABLE)) {
572 code = XFS_ERROR(EPERM);
578 * Now we can make the changes. Before we join the inode
579 * to the transaction, if XFS_AT_SIZE is set then take care of
580 * the part of the truncation that must be done without the
581 * inode lock. This needs to be done before joining the inode
582 * to the transaction, because the inode cannot be unlocked
583 * once it is a part of the transaction.
585 if (mask & XFS_AT_SIZE) {
587 if ((vap->va_size > ip->i_size) &&
588 (flags & ATTR_NOSIZETOK) == 0) {
589 code = xfs_igrow_start(ip, vap->va_size, credp);
591 xfs_iunlock(ip, XFS_ILOCK_EXCL);
594 * We are going to log the inode size change in this
595 * transaction so any previous writes that are beyond the on
596 * disk EOF and the new EOF that have not been written out need
597 * to be written here. If we do not write the data out, we
598 * expose ourselves to the null files problem.
600 * Only flush from the on disk size to the smaller of the in
601 * memory file size or the new size as that's the range we
602 * really care about here and prevents waiting for other data
603 * not within the range we care about here.
606 (ip->i_size != ip->i_d.di_size) &&
607 (vap->va_size > ip->i_d.di_size)) {
608 code = bhv_vop_flush_pages(XFS_ITOV(ip),
609 ip->i_d.di_size, vap->va_size,
610 XFS_B_ASYNC, FI_NONE);
613 /* wait for all I/O to complete */
617 code = xfs_itruncate_data(ip, vap->va_size);
620 lock_flags &= ~XFS_ILOCK_EXCL;
621 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
624 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
625 if ((code = xfs_trans_reserve(tp, 0,
626 XFS_ITRUNCATE_LOG_RES(mp), 0,
627 XFS_TRANS_PERM_LOG_RES,
628 XFS_ITRUNCATE_LOG_COUNT))) {
629 xfs_trans_cancel(tp, 0);
631 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
634 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
635 xfs_ilock(ip, XFS_ILOCK_EXCL);
639 xfs_trans_ijoin(tp, ip, lock_flags);
640 xfs_trans_ihold(tp, ip);
643 /* determine whether mandatory locking mode changes */
644 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
647 * Truncate file. Must have write permission and not be a directory.
649 if (mask & XFS_AT_SIZE) {
650 if (vap->va_size > ip->i_size) {
651 xfs_igrow_finish(tp, ip, vap->va_size,
652 !(flags & ATTR_DMI));
653 } else if ((vap->va_size <= ip->i_size) ||
654 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
656 * signal a sync transaction unless
657 * we're truncating an already unlinked
658 * file on a wsync filesystem
660 code = xfs_itruncate_finish(&tp, ip,
661 (xfs_fsize_t)vap->va_size,
663 ((ip->i_d.di_nlink != 0 ||
664 !(mp->m_flags & XFS_MOUNT_WSYNC))
669 * Truncated "down", so we're removing references
670 * to old data here - if we now delay flushing for
671 * a long time, we expose ourselves unduly to the
672 * notorious NULL files problem. So, we mark this
673 * vnode and flush it when the file is closed, and
674 * do not wait the usual (long) time for writeout.
679 * Have to do this even if the file's size doesn't change.
681 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
685 * Change file access modes.
687 if (mask & XFS_AT_MODE) {
688 ip->i_d.di_mode &= S_IFMT;
689 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
691 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
692 timeflags |= XFS_ICHGTIME_CHG;
696 * Change file ownership. Must be the owner or privileged.
697 * If the system was configured with the "restricted_chown"
698 * option, the owner is not permitted to give away the file,
699 * and can change the group id only to a group of which he
700 * or she is a member.
702 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
704 * CAP_FSETID overrides the following restrictions:
706 * The set-user-ID and set-group-ID bits of a file will be
707 * cleared upon successful return from chown()
709 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
710 !capable(CAP_FSETID)) {
711 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
715 * Change the ownerships and register quota modifications
716 * in the transaction.
719 if (XFS_IS_UQUOTA_ON(mp)) {
720 ASSERT(mask & XFS_AT_UID);
722 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
723 &ip->i_udquot, udqp);
725 ip->i_d.di_uid = uid;
728 if (XFS_IS_GQUOTA_ON(mp)) {
729 ASSERT(!XFS_IS_PQUOTA_ON(mp));
730 ASSERT(mask & XFS_AT_GID);
732 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
733 &ip->i_gdquot, gdqp);
735 ip->i_d.di_gid = gid;
737 if (iprojid != projid) {
738 if (XFS_IS_PQUOTA_ON(mp)) {
739 ASSERT(!XFS_IS_GQUOTA_ON(mp));
740 ASSERT(mask & XFS_AT_PROJID);
742 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
743 &ip->i_gdquot, gdqp);
745 ip->i_d.di_projid = projid;
747 * We may have to rev the inode as well as
748 * the superblock version number since projids didn't
749 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
751 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
752 xfs_bump_ino_vers2(tp, ip);
755 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
756 timeflags |= XFS_ICHGTIME_CHG;
761 * Change file access or modified times.
763 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
764 if (mask & XFS_AT_ATIME) {
765 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
766 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
767 ip->i_update_core = 1;
768 timeflags &= ~XFS_ICHGTIME_ACC;
770 if (mask & XFS_AT_MTIME) {
771 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
772 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
773 timeflags &= ~XFS_ICHGTIME_MOD;
774 timeflags |= XFS_ICHGTIME_CHG;
776 if (tp && (flags & ATTR_UTIME))
777 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
781 * Change XFS-added attributes.
783 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
784 if (mask & XFS_AT_EXTSIZE) {
786 * Converting bytes to fs blocks.
788 ip->i_d.di_extsize = vap->va_extsize >>
789 mp->m_sb.sb_blocklog;
791 if (mask & XFS_AT_XFLAGS) {
794 /* can't set PREALLOC this way, just preserve it */
795 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
796 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
797 di_flags |= XFS_DIFLAG_IMMUTABLE;
798 if (vap->va_xflags & XFS_XFLAG_APPEND)
799 di_flags |= XFS_DIFLAG_APPEND;
800 if (vap->va_xflags & XFS_XFLAG_SYNC)
801 di_flags |= XFS_DIFLAG_SYNC;
802 if (vap->va_xflags & XFS_XFLAG_NOATIME)
803 di_flags |= XFS_DIFLAG_NOATIME;
804 if (vap->va_xflags & XFS_XFLAG_NODUMP)
805 di_flags |= XFS_DIFLAG_NODUMP;
806 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
807 di_flags |= XFS_DIFLAG_PROJINHERIT;
808 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
809 di_flags |= XFS_DIFLAG_NODEFRAG;
810 if (vap->va_xflags & XFS_XFLAG_FILESTREAM)
811 di_flags |= XFS_DIFLAG_FILESTREAM;
812 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
813 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
814 di_flags |= XFS_DIFLAG_RTINHERIT;
815 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
816 di_flags |= XFS_DIFLAG_NOSYMLINKS;
817 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
818 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
819 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
820 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
821 di_flags |= XFS_DIFLAG_REALTIME;
822 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
824 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
826 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
827 di_flags |= XFS_DIFLAG_EXTSIZE;
829 ip->i_d.di_flags = di_flags;
831 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
832 timeflags |= XFS_ICHGTIME_CHG;
836 * Change file inode change time only if XFS_AT_CTIME set
837 * AND we have been called by a DMI function.
840 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
841 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
842 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
843 ip->i_update_core = 1;
844 timeflags &= ~XFS_ICHGTIME_CHG;
848 * Send out timestamp changes that need to be set to the
849 * current time. Not done when called by a DMI function.
851 if (timeflags && !(flags & ATTR_DMI))
852 xfs_ichgtime(ip, timeflags);
854 XFS_STATS_INC(xs_ig_attrchg);
857 * If this is a synchronous mount, make sure that the
858 * transaction goes to disk before returning to the user.
859 * This is slightly sub-optimal in that truncates require
860 * two sync transactions instead of one for wsync filesystems.
861 * One for the truncate and one for the timestamps since we
862 * don't want to change the timestamps unless we're sure the
863 * truncate worked. Truncates are less than 1% of the laddis
864 * mix so this probably isn't worth the trouble to optimize.
868 if (mp->m_flags & XFS_MOUNT_WSYNC)
869 xfs_trans_set_sync(tp);
871 code = xfs_trans_commit(tp, commit_flags);
875 * If the (regular) file's mandatory locking mode changed, then
876 * notify the vnode. We do this under the inode lock to prevent
877 * racing calls to vop_vnode_change.
879 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
880 if (mandlock_before != mandlock_after) {
881 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING,
885 xfs_iunlock(ip, lock_flags);
888 * Release any dquot(s) the inode had kept before chown.
890 XFS_QM_DQRELE(mp, olddquot1);
891 XFS_QM_DQRELE(mp, olddquot2);
892 XFS_QM_DQRELE(mp, udqp);
893 XFS_QM_DQRELE(mp, gdqp);
899 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
900 !(flags & ATTR_DMI)) {
901 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
902 NULL, DM_RIGHT_NULL, NULL, NULL,
903 0, 0, AT_DELAY_FLAG(flags));
908 commit_flags |= XFS_TRANS_ABORT;
911 XFS_QM_DQRELE(mp, udqp);
912 XFS_QM_DQRELE(mp, gdqp);
914 xfs_trans_cancel(tp, commit_flags);
916 if (lock_flags != 0) {
917 xfs_iunlock(ip, lock_flags);
925 * Null conversion from vnode mode bits to inode mode bits, as in efs.
936 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
937 (inst_t *)__return_address);
939 ip = XFS_BHVTOI(bdp);
940 xfs_ilock(ip, XFS_ILOCK_SHARED);
941 error = xfs_iaccess(ip, mode, credp);
942 xfs_iunlock(ip, XFS_ILOCK_SHARED);
948 * The maximum pathlen is 1024 bytes. Since the minimum file system
949 * blocksize is 512 bytes, we can get a max of 2 extents back from
952 #define SYMLINK_MAPS 2
959 xfs_mount_t *mp = ip->i_mount;
960 int pathlen = ip->i_d.di_size;
961 int nmaps = SYMLINK_MAPS;
962 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
969 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
970 mval, &nmaps, NULL, NULL);
974 for (n = 0; n < nmaps; n++) {
975 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
976 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
978 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
979 error = XFS_BUF_GETERROR(bp);
981 xfs_ioerror_alert("xfs_readlink",
982 ip->i_mount, bp, XFS_BUF_ADDR(bp));
986 if (pathlen < byte_cnt)
990 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
994 link[ip->i_d.di_size] = '\0';
1010 xfs_inode_t *ip = XFS_BHVTOI(bdp);
1011 xfs_mount_t *mp = ip->i_mount;
1015 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
1017 if (XFS_FORCED_SHUTDOWN(mp))
1018 return XFS_ERROR(EIO);
1020 xfs_ilock(ip, XFS_ILOCK_SHARED);
1022 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1023 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
1025 pathlen = ip->i_d.di_size;
1029 if (ip->i_df.if_flags & XFS_IFINLINE) {
1030 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
1031 link[pathlen] = '\0';
1033 error = xfs_readlink_bmap(ip, link);
1037 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1044 * This is called to sync the inode and its data out to disk.
1045 * We need to hold the I/O lock while flushing the data, and
1046 * the inode lock while flushing the inode. The inode lock CANNOT
1047 * be held while flushing the data, so acquire after we're done
1061 int log_flushed = 0, changed = 1;
1063 vn_trace_entry(BHV_TO_VNODE(bdp),
1064 __FUNCTION__, (inst_t *)__return_address);
1066 ip = XFS_BHVTOI(bdp);
1068 ASSERT(start >= 0 && stop >= -1);
1070 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1071 return XFS_ERROR(EIO);
1073 if (flag & FSYNC_DATA)
1074 filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping);
1077 * We always need to make sure that the required inode state
1078 * is safe on disk. The vnode might be clean but because
1079 * of committed transactions that haven't hit the disk yet.
1080 * Likewise, there could be unflushed non-transactional
1081 * changes to the inode core that have to go to disk.
1083 * The following code depends on one assumption: that
1084 * any transaction that changes an inode logs the core
1085 * because it has to change some field in the inode core
1086 * (typically nextents or nblocks). That assumption
1087 * implies that any transactions against an inode will
1088 * catch any non-transactional updates. If inode-altering
1089 * transactions exist that violate this assumption, the
1090 * code breaks. Right now, it figures that if the involved
1091 * update_* field is clear and the inode is unpinned, the
1092 * inode is clean. Either it's been flushed or it's been
1093 * committed and the commit has hit the disk unpinning the inode.
1094 * (Note that xfs_inode_item_format() called at commit clears
1095 * the update_* fields.)
1097 xfs_ilock(ip, XFS_ILOCK_SHARED);
1099 /* If we are flushing data then we care about update_size
1100 * being set, otherwise we care about update_core
1102 if ((flag & FSYNC_DATA) ?
1103 (ip->i_update_size == 0) :
1104 (ip->i_update_core == 0)) {
1106 * Timestamps/size haven't changed since last inode
1107 * flush or inode transaction commit. That means
1108 * either nothing got written or a transaction
1109 * committed which caught the updates. If the
1110 * latter happened and the transaction hasn't
1111 * hit the disk yet, the inode will be still
1112 * be pinned. If it is, force the log.
1115 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1117 if (xfs_ipincount(ip)) {
1118 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1120 ((flag & FSYNC_WAIT)
1121 ? XFS_LOG_SYNC : 0),
1125 * If the inode is not pinned and nothing
1126 * has changed we don't need to flush the
1134 * Kick off a transaction to log the inode
1135 * core to get the updates. Make it
1136 * sync if FSYNC_WAIT is passed in (which
1137 * is done by everybody but specfs). The
1138 * sync transaction will also force the log.
1140 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1141 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1142 if ((error = xfs_trans_reserve(tp, 0,
1143 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1145 xfs_trans_cancel(tp, 0);
1148 xfs_ilock(ip, XFS_ILOCK_EXCL);
1151 * Note - it's possible that we might have pushed
1152 * ourselves out of the way during trans_reserve
1153 * which would flush the inode. But there's no
1154 * guarantee that the inode buffer has actually
1155 * gone out yet (it's delwri). Plus the buffer
1156 * could be pinned anyway if it's part of an
1157 * inode in another recent transaction. So we
1158 * play it safe and fire off the transaction anyway.
1160 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1161 xfs_trans_ihold(tp, ip);
1162 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1163 if (flag & FSYNC_WAIT)
1164 xfs_trans_set_sync(tp);
1165 error = _xfs_trans_commit(tp, 0, &log_flushed);
1167 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1170 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1172 * If the log write didn't issue an ordered tag we need
1173 * to flush the disk cache for the data device now.
1176 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1179 * If this inode is on the RT dev we need to flush that
1182 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1183 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1190 * This is called by xfs_inactive to free any blocks beyond eof
1191 * when the link count isn't zero and by xfs_dm_punch_hole() when
1192 * punching a hole to EOF.
1202 xfs_fileoff_t end_fsb;
1203 xfs_fileoff_t last_fsb;
1204 xfs_filblks_t map_len;
1206 xfs_bmbt_irec_t imap;
1207 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
1210 * Figure out if there are any blocks beyond the end
1211 * of the file. If not, then there is nothing to do.
1213 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
1214 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1215 map_len = last_fsb - end_fsb;
1220 xfs_ilock(ip, XFS_ILOCK_SHARED);
1221 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1222 NULL, 0, &imap, &nimaps, NULL, NULL);
1223 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1225 if (!error && (nimaps != 0) &&
1226 (imap.br_startblock != HOLESTARTBLOCK ||
1227 ip->i_delayed_blks)) {
1229 * Attach the dquots to the inode up front.
1231 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1235 * There are blocks after the end of file.
1236 * Free them up now by truncating the file to
1239 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1242 * Do the xfs_itruncate_start() call before
1243 * reserving any log space because
1244 * itruncate_start will call into the buffer
1245 * cache and we can't
1246 * do that within a transaction.
1249 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1250 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1253 xfs_trans_cancel(tp, 0);
1255 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1259 error = xfs_trans_reserve(tp, 0,
1260 XFS_ITRUNCATE_LOG_RES(mp),
1261 0, XFS_TRANS_PERM_LOG_RES,
1262 XFS_ITRUNCATE_LOG_COUNT);
1264 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1265 xfs_trans_cancel(tp, 0);
1266 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1270 xfs_ilock(ip, XFS_ILOCK_EXCL);
1271 xfs_trans_ijoin(tp, ip,
1274 xfs_trans_ihold(tp, ip);
1276 error = xfs_itruncate_finish(&tp, ip,
1281 * If we get an error at this point we
1282 * simply don't bother truncating the file.
1285 xfs_trans_cancel(tp,
1286 (XFS_TRANS_RELEASE_LOG_RES |
1289 error = xfs_trans_commit(tp,
1290 XFS_TRANS_RELEASE_LOG_RES);
1292 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
1299 * Free a symlink that has blocks associated with it.
1302 xfs_inactive_symlink_rmt(
1310 xfs_fsblock_t first_block;
1311 xfs_bmap_free_t free_list;
1314 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1322 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1324 * We're freeing a symlink that has some
1325 * blocks allocated to it. Free the
1326 * blocks here. We know that we've got
1327 * either 1 or 2 extents and that we can
1328 * free them all in one bunmapi call.
1330 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1331 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1332 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1333 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1334 xfs_trans_cancel(tp, 0);
1339 * Lock the inode, fix the size, and join it to the transaction.
1340 * Hold it so in the normal path, we still have it locked for
1341 * the second transaction. In the error paths we need it
1342 * held so the cancel won't rele it, see below.
1344 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1345 size = (int)ip->i_d.di_size;
1346 ip->i_d.di_size = 0;
1347 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1348 xfs_trans_ihold(tp, ip);
1349 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1351 * Find the block(s) so we can inval and unmap them.
1354 XFS_BMAP_INIT(&free_list, &first_block);
1355 nmaps = ARRAY_SIZE(mval);
1356 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1357 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1361 * Invalidate the block(s).
1363 for (i = 0; i < nmaps; i++) {
1364 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1365 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1366 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1367 xfs_trans_binval(tp, bp);
1370 * Unmap the dead block(s) to the free_list.
1372 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1373 &first_block, &free_list, NULL, &done)))
1377 * Commit the first transaction. This logs the EFI and the inode.
1379 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
1382 * The transaction must have been committed, since there were
1383 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1384 * The new tp has the extent freeing and EFDs.
1388 * The first xact was committed, so add the inode to the new one.
1389 * Mark it dirty so it will be logged and moved forward in the log as
1390 * part of every commit.
1392 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1393 xfs_trans_ihold(tp, ip);
1394 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1396 * Get a new, empty transaction to return to our caller.
1398 ntp = xfs_trans_dup(tp);
1400 * Commit the transaction containing extent freeing and EFDs.
1401 * If we get an error on the commit here or on the reserve below,
1402 * we need to unlock the inode since the new transaction doesn't
1403 * have the inode attached.
1405 error = xfs_trans_commit(tp, 0);
1408 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1412 * Remove the memory for extent descriptions (just bookkeeping).
1414 if (ip->i_df.if_bytes)
1415 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1416 ASSERT(ip->i_df.if_bytes == 0);
1418 * Put an itruncate log reservation in the new transaction
1421 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1422 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1423 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1427 * Return with the inode locked but not joined to the transaction.
1433 xfs_bmap_cancel(&free_list);
1436 * Have to come here with the inode locked and either
1437 * (held and in the transaction) or (not in the transaction).
1438 * If the inode isn't held then cancel would iput it, but
1439 * that's wrong since this is inactive and the vnode ref
1440 * count is 0 already.
1441 * Cancel won't do anything to the inode if held, but it still
1442 * needs to be locked until the cancel is done, if it was
1443 * joined to the transaction.
1445 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1446 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1453 xfs_inactive_symlink_local(
1459 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1461 * We're freeing a symlink which fit into
1462 * the inode. Just free the memory used
1463 * to hold the old symlink.
1465 error = xfs_trans_reserve(*tpp, 0,
1466 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1467 0, XFS_TRANS_PERM_LOG_RES,
1468 XFS_ITRUNCATE_LOG_COUNT);
1471 xfs_trans_cancel(*tpp, 0);
1475 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1478 * Zero length symlinks _can_ exist.
1480 if (ip->i_df.if_bytes > 0) {
1481 xfs_idata_realloc(ip,
1482 -(ip->i_df.if_bytes),
1484 ASSERT(ip->i_df.if_bytes == 0);
1498 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1501 ASSERT(ip->i_d.di_forkoff != 0);
1502 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1503 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1505 error = xfs_attr_inactive(ip);
1508 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1509 return error; /* goto out */
1512 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1513 error = xfs_trans_reserve(tp, 0,
1514 XFS_IFREE_LOG_RES(mp),
1515 0, XFS_TRANS_PERM_LOG_RES,
1516 XFS_INACTIVE_LOG_COUNT);
1518 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1519 xfs_trans_cancel(tp, 0);
1521 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1525 xfs_ilock(ip, XFS_ILOCK_EXCL);
1526 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1527 xfs_trans_ihold(tp, ip);
1528 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1530 ASSERT(ip->i_d.di_anextents == 0);
1545 vp = BHV_TO_VNODE(bdp);
1546 ip = XFS_BHVTOI(bdp);
1549 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
1552 /* If this is a read-only mount, don't do this (would generate I/O) */
1553 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1556 if (!XFS_FORCED_SHUTDOWN(mp)) {
1558 * If we are using filestreams, and we have an unlinked
1559 * file that we are processing the last close on, then nothing
1560 * will be able to reopen and write to this file. Purge this
1561 * inode from the filestreams cache so that it doesn't delay
1562 * teardown of the inode.
1564 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1565 xfs_filestream_deassociate(ip);
1568 * If we previously truncated this file and removed old data
1569 * in the process, we want to initiate "early" writeout on
1570 * the last close. This is an attempt to combat the notorious
1571 * NULL files problem which is particularly noticable from a
1572 * truncate down, buffered (re-)write (delalloc), followed by
1573 * a crash. What we are effectively doing here is
1574 * significantly reducing the time window where we'd otherwise
1575 * be exposed to that problem.
1577 if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
1578 bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE);
1581 #ifdef HAVE_REFCACHE
1582 /* If we are in the NFS reference cache then don't do this now */
1587 if (ip->i_d.di_nlink != 0) {
1588 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1589 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1590 ip->i_delayed_blks > 0)) &&
1591 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1592 (!(ip->i_d.di_flags &
1593 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1594 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1597 /* Update linux inode block count after free above */
1598 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1599 ip->i_d.di_nblocks + ip->i_delayed_blks);
1609 * This is called when the vnode reference count for the vnode
1610 * goes to zero. If the file has been unlinked, then it must
1611 * now be truncated. Also, we clear all of the read-ahead state
1612 * kept for the inode here since the file is now closed.
1621 xfs_bmap_free_t free_list;
1622 xfs_fsblock_t first_block;
1629 vp = BHV_TO_VNODE(bdp);
1630 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1632 ip = XFS_BHVTOI(bdp);
1635 * If the inode is already free, then there can be nothing
1638 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1639 ASSERT(ip->i_df.if_real_bytes == 0);
1640 ASSERT(ip->i_df.if_broot_bytes == 0);
1641 return VN_INACTIVE_CACHE;
1645 * Only do a truncate if it's a regular file with
1646 * some actual space in it. It's OK to look at the
1647 * inode's fields without the lock because we're the
1648 * only one with a reference to the inode.
1650 truncate = ((ip->i_d.di_nlink == 0) &&
1651 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1652 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1653 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1657 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) {
1658 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1663 /* If this is a read-only mount, don't do this (would generate I/O) */
1664 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1667 if (ip->i_d.di_nlink != 0) {
1668 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1669 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1670 ip->i_delayed_blks > 0)) &&
1671 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1672 (!(ip->i_d.di_flags &
1673 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1674 (ip->i_delayed_blks != 0)))) {
1675 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1677 return VN_INACTIVE_CACHE;
1678 /* Update linux inode block count after free above */
1679 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1680 ip->i_d.di_nblocks + ip->i_delayed_blks);
1685 ASSERT(ip->i_d.di_nlink == 0);
1687 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1688 return VN_INACTIVE_CACHE;
1690 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1693 * Do the xfs_itruncate_start() call before
1694 * reserving any log space because itruncate_start
1695 * will call into the buffer cache and we can't
1696 * do that within a transaction.
1698 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1700 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1702 xfs_trans_cancel(tp, 0);
1703 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1704 return VN_INACTIVE_CACHE;
1707 error = xfs_trans_reserve(tp, 0,
1708 XFS_ITRUNCATE_LOG_RES(mp),
1709 0, XFS_TRANS_PERM_LOG_RES,
1710 XFS_ITRUNCATE_LOG_COUNT);
1712 /* Don't call itruncate_cleanup */
1713 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1714 xfs_trans_cancel(tp, 0);
1715 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1716 return VN_INACTIVE_CACHE;
1719 xfs_ilock(ip, XFS_ILOCK_EXCL);
1720 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1721 xfs_trans_ihold(tp, ip);
1724 * normally, we have to run xfs_itruncate_finish sync.
1725 * But if filesystem is wsync and we're in the inactive
1726 * path, then we know that nlink == 0, and that the
1727 * xaction that made nlink == 0 is permanently committed
1728 * since xfs_remove runs as a synchronous transaction.
1730 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1731 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1734 xfs_trans_cancel(tp,
1735 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1736 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1737 return VN_INACTIVE_CACHE;
1739 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1742 * If we get an error while cleaning up a
1743 * symlink we bail out.
1745 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1746 xfs_inactive_symlink_rmt(ip, &tp) :
1747 xfs_inactive_symlink_local(ip, &tp);
1751 return VN_INACTIVE_CACHE;
1754 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1755 xfs_trans_ihold(tp, ip);
1757 error = xfs_trans_reserve(tp, 0,
1758 XFS_IFREE_LOG_RES(mp),
1759 0, XFS_TRANS_PERM_LOG_RES,
1760 XFS_INACTIVE_LOG_COUNT);
1762 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1763 xfs_trans_cancel(tp, 0);
1764 return VN_INACTIVE_CACHE;
1767 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1768 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1769 xfs_trans_ihold(tp, ip);
1773 * If there are attributes associated with the file
1774 * then blow them away now. The code calls a routine
1775 * that recursively deconstructs the attribute fork.
1776 * We need to just commit the current transaction
1777 * because we can't use it for xfs_attr_inactive().
1779 if (ip->i_d.di_anextents > 0) {
1780 error = xfs_inactive_attrs(ip, &tp);
1782 * If we got an error, the transaction is already
1783 * cancelled, and the inode is unlocked. Just get out.
1786 return VN_INACTIVE_CACHE;
1787 } else if (ip->i_afp) {
1788 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1794 XFS_BMAP_INIT(&free_list, &first_block);
1795 error = xfs_ifree(tp, ip, &free_list);
1798 * If we fail to free the inode, shut down. The cancel
1799 * might do that, we need to make sure. Otherwise the
1800 * inode might be lost for a long time or forever.
1802 if (!XFS_FORCED_SHUTDOWN(mp)) {
1804 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1805 error, mp->m_fsname);
1806 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1808 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1811 * Credit the quota account(s). The inode is gone.
1813 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1816 * Just ignore errors at this point. There is
1817 * nothing we can do except to try to keep going.
1819 (void) xfs_bmap_finish(&tp, &free_list, &committed);
1820 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1823 * Release the dquots held by inode, if any.
1825 XFS_QM_DQDETACH(mp, ip);
1827 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1830 return VN_INACTIVE_CACHE;
1839 bhv_desc_t *dir_bdp,
1840 bhv_vname_t *dentry,
1846 xfs_inode_t *dp, *ip;
1850 bhv_vnode_t *dir_vp;
1852 dir_vp = BHV_TO_VNODE(dir_bdp);
1853 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1855 dp = XFS_BHVTOI(dir_bdp);
1857 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1858 return XFS_ERROR(EIO);
1860 lock_mode = xfs_ilock_map_shared(dp);
1861 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1863 *vpp = XFS_ITOV(ip);
1866 xfs_iunlock_map_shared(dp, lock_mode);
1872 * xfs_create (create a new file).
1876 bhv_desc_t *dir_bdp,
1877 bhv_vname_t *dentry,
1882 char *name = VNAME(dentry);
1883 bhv_vnode_t *dir_vp;
1884 xfs_inode_t *dp, *ip;
1885 bhv_vnode_t *vp = NULL;
1890 xfs_bmap_free_t free_list;
1891 xfs_fsblock_t first_block;
1892 boolean_t dp_joined_to_trans;
1893 int dm_event_sent = 0;
1897 struct xfs_dquot *udqp, *gdqp;
1903 dir_vp = BHV_TO_VNODE(dir_bdp);
1904 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1906 dp = XFS_BHVTOI(dir_bdp);
1909 dm_di_mode = vap->va_mode;
1910 namelen = VNAMELEN(dentry);
1912 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
1913 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1914 dir_vp, DM_RIGHT_NULL, NULL,
1915 DM_RIGHT_NULL, name, NULL,
1923 if (XFS_FORCED_SHUTDOWN(mp))
1924 return XFS_ERROR(EIO);
1926 /* Return through std_return after this point. */
1929 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1930 prid = dp->i_d.di_projid;
1931 else if (vap->va_mask & XFS_AT_PROJID)
1932 prid = (xfs_prid_t)vap->va_projid;
1934 prid = (xfs_prid_t)dfltprid;
1937 * Make sure that we have allocated dquot(s) on disk.
1939 error = XFS_QM_DQVOPALLOC(mp, dp,
1940 current_fsuid(credp), current_fsgid(credp), prid,
1941 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1946 dp_joined_to_trans = B_FALSE;
1948 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1949 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1950 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1952 * Initially assume that the file does not exist and
1953 * reserve the resources for that case. If that is not
1954 * the case we'll drop the one we have and get a more
1955 * appropriate transaction later.
1957 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1958 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1959 if (error == ENOSPC) {
1961 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1962 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1970 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1972 XFS_BMAP_INIT(&free_list, &first_block);
1977 * Reserve disk quota and the inode.
1979 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1983 if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen)))
1985 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1986 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1987 rdev, credp, prid, resblks > 0,
1990 if (error == ENOSPC)
1997 * At this point, we've gotten a newly allocated inode.
1998 * It is locked (and joined to the transaction).
2001 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
2004 * Now we join the directory inode to the transaction.
2005 * We do not do it earlier because xfs_dir_ialloc
2006 * might commit the previous transaction (and release
2011 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2012 dp_joined_to_trans = B_TRUE;
2014 error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino,
2015 &first_block, &free_list, resblks ?
2016 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2018 ASSERT(error != ENOSPC);
2021 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2022 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2025 * If this is a synchronous mount, make sure that the
2026 * create transaction goes to disk before returning to
2029 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2030 xfs_trans_set_sync(tp);
2036 * Attach the dquot(s) to the inodes and modify them incore.
2037 * These ids of the inode couldn't have changed since the new
2038 * inode has been locked ever since it was created.
2040 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2043 * xfs_trans_commit normally decrements the vnode ref count
2044 * when it unlocks the inode. Since we want to return the
2045 * vnode to the caller, we bump the vnode ref count now.
2050 error = xfs_bmap_finish(&tp, &free_list, &committed);
2052 xfs_bmap_cancel(&free_list);
2056 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2063 XFS_QM_DQRELE(mp, udqp);
2064 XFS_QM_DQRELE(mp, gdqp);
2067 * Propagate the fact that the vnode changed after the
2068 * xfs_inode locks have been released.
2070 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2074 /* Fallthrough to std_return with error = 0 */
2077 if ((*vpp || (error != 0 && dm_event_sent != 0)) &&
2078 DM_EVENT_ENABLED(XFS_BHVTOI(dir_bdp), DM_EVENT_POSTCREATE)) {
2079 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2080 dir_vp, DM_RIGHT_NULL,
2082 DM_RIGHT_NULL, name, NULL,
2083 dm_di_mode, error, 0);
2088 cancel_flags |= XFS_TRANS_ABORT;
2093 xfs_trans_cancel(tp, cancel_flags);
2095 if (!dp_joined_to_trans && (dp != NULL))
2096 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2097 XFS_QM_DQRELE(mp, udqp);
2098 XFS_QM_DQRELE(mp, gdqp);
2104 * Wait until after the current transaction is aborted to
2105 * release the inode. This prevents recursive transactions
2106 * and deadlocks from xfs_inactive.
2108 cancel_flags |= XFS_TRANS_ABORT;
2109 xfs_trans_cancel(tp, cancel_flags);
2112 XFS_QM_DQRELE(mp, udqp);
2113 XFS_QM_DQRELE(mp, gdqp);
2120 * Some counters to see if (and how often) we are hitting some deadlock
2121 * prevention code paths.
2125 int xfs_rm_lock_delays;
2126 int xfs_rm_attempts;
2130 * The following routine will lock the inodes associated with the
2131 * directory and the named entry in the directory. The locks are
2132 * acquired in increasing inode number.
2134 * If the entry is "..", then only the directory is locked. The
2135 * vnode ref count will still include that from the .. entry in
2138 * There is a deadlock we need to worry about. If the locked directory is
2139 * in the AIL, it might be blocking up the log. The next inode we lock
2140 * could be already locked by another thread waiting for log space (e.g
2141 * a permanent log reservation with a long running transaction (see
2142 * xfs_itruncate_finish)). To solve this, we must check if the directory
2143 * is in the ail and use lock_nowait. If we can't lock, we need to
2144 * drop the inode lock on the directory and try again. xfs_iunlock will
2145 * potentially push the tail if we were holding up the log.
2148 xfs_lock_dir_and_entry(
2150 xfs_inode_t *ip) /* inode of entry 'name' */
2154 xfs_inode_t *ips[2];
2163 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2170 * We want to lock in increasing inum. Since we've already
2171 * acquired the lock on the directory, we may need to release
2172 * if if the inum of the entry turns out to be less.
2174 if (e_inum > dp->i_ino) {
2176 * We are already in the right order, so just
2177 * lock on the inode of the entry.
2178 * We need to use nowait if dp is in the AIL.
2181 lp = (xfs_log_item_t *)dp->i_itemp;
2182 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2183 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2190 * Unlock dp and try again.
2191 * xfs_iunlock will try to push the tail
2192 * if the inode is in the AIL.
2195 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2197 if ((attempts % 5) == 0) {
2198 delay(1); /* Don't just spin the CPU */
2200 xfs_rm_lock_delays++;
2206 xfs_ilock(ip, XFS_ILOCK_EXCL);
2208 } else if (e_inum < dp->i_ino) {
2209 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2213 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2215 /* else e_inum == dp->i_ino */
2216 /* This can happen if we're asked to lock /x/..
2217 * the entry is "..", which is also the parent directory.
2225 int xfs_small_retries;
2226 int xfs_middle_retries;
2227 int xfs_lots_retries;
2228 int xfs_lock_delays;
2232 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2236 xfs_lock_inumorder(int lock_mode, int subclass)
2238 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
2239 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
2240 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
2241 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
2247 * The following routine will lock n inodes in exclusive mode.
2248 * We assume the caller calls us with the inodes in i_ino order.
2250 * We need to detect deadlock where an inode that we lock
2251 * is in the AIL and we start waiting for another inode that is locked
2252 * by a thread in a long running transaction (such as truncate). This can
2253 * result in deadlock since the long running trans might need to wait
2254 * for the inode we just locked in order to push the tail and free space
2264 int attempts = 0, i, j, try_lock;
2267 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2278 for (; i < inodes; i++) {
2281 if (i && (ips[i] == ips[i-1])) /* Already locked */
2285 * If try_lock is not set yet, make sure all locked inodes
2286 * are not in the AIL.
2287 * If any are, set try_lock to be used later.
2291 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2292 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2293 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2300 * If any of the previous locks we have locked is in the AIL,
2301 * we must TRY to get the second and subsequent locks. If
2302 * we can't get any, we must release all we have
2307 /* try_lock must be 0 if i is 0. */
2309 * try_lock means we have an inode locked
2310 * that is in the AIL.
2313 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
2317 * Unlock all previous guys and try again.
2318 * xfs_iunlock will try to push the tail
2319 * if the inode is in the AIL.
2322 for(j = i - 1; j >= 0; j--) {
2325 * Check to see if we've already
2326 * unlocked this one.
2327 * Not the first one going back,
2328 * and the inode ptr is the same.
2330 if ((j != (i - 1)) && ips[j] ==
2334 xfs_iunlock(ips[j], lock_mode);
2337 if ((attempts % 5) == 0) {
2338 delay(1); /* Don't just spin the CPU */
2348 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
2354 if (attempts < 5) xfs_small_retries++;
2355 else if (attempts < 100) xfs_middle_retries++;
2356 else xfs_lots_retries++;
2364 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2365 int remove_which_error_return = 0;
2367 #define REMOVE_DEBUG_TRACE(x)
2368 #endif /* ! DEBUG */
2377 bhv_desc_t *dir_bdp,
2378 bhv_vname_t *dentry,
2381 bhv_vnode_t *dir_vp;
2382 char *name = VNAME(dentry);
2383 xfs_inode_t *dp, *ip;
2384 xfs_trans_t *tp = NULL;
2387 xfs_bmap_free_t free_list;
2388 xfs_fsblock_t first_block;
2396 dir_vp = BHV_TO_VNODE(dir_bdp);
2397 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2399 dp = XFS_BHVTOI(dir_bdp);
2402 if (XFS_FORCED_SHUTDOWN(mp))
2403 return XFS_ERROR(EIO);
2405 namelen = VNAMELEN(dentry);
2407 if (!xfs_get_dir_entry(dentry, &ip)) {
2408 dm_di_mode = ip->i_d.di_mode;
2412 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
2413 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2414 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2415 name, NULL, dm_di_mode, 0, 0);
2420 /* From this point on, return through std_return */
2424 * We need to get a reference to ip before we get our log
2425 * reservation. The reason for this is that we cannot call
2426 * xfs_iget for an inode for which we do not have a reference
2427 * once we've acquired a log reservation. This is because the
2428 * inode we are trying to get might be in xfs_inactive going
2429 * for a log reservation. Since we'll have to wait for the
2430 * inactive code to complete before returning from xfs_iget,
2431 * we need to make sure that we don't have log space reserved
2432 * when we call xfs_iget. Instead we get an unlocked reference
2433 * to the inode before getting our log reservation.
2435 error = xfs_get_dir_entry(dentry, &ip);
2437 REMOVE_DEBUG_TRACE(__LINE__);
2441 dm_di_mode = ip->i_d.di_mode;
2443 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2447 error = XFS_QM_DQATTACH(mp, dp, 0);
2448 if (!error && dp != ip)
2449 error = XFS_QM_DQATTACH(mp, ip, 0);
2451 REMOVE_DEBUG_TRACE(__LINE__);
2456 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2457 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2459 * We try to get the real space reservation first,
2460 * allowing for directory btree deletion(s) implying
2461 * possible bmap insert(s). If we can't get the space
2462 * reservation then we use 0 instead, and avoid the bmap
2463 * btree insert(s) in the directory code by, if the bmap
2464 * insert tries to happen, instead trimming the LAST
2465 * block from the directory.
2467 resblks = XFS_REMOVE_SPACE_RES(mp);
2468 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2469 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2470 if (error == ENOSPC) {
2472 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2473 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2476 ASSERT(error != ENOSPC);
2477 REMOVE_DEBUG_TRACE(__LINE__);
2478 xfs_trans_cancel(tp, 0);
2483 error = xfs_lock_dir_and_entry(dp, ip);
2485 REMOVE_DEBUG_TRACE(__LINE__);
2486 xfs_trans_cancel(tp, cancel_flags);
2492 * At this point, we've gotten both the directory and the entry
2495 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2498 * Increment vnode ref count only in this case since
2499 * there's an extra vnode reference in the case where
2503 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2507 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2509 XFS_BMAP_INIT(&free_list, &first_block);
2510 error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino,
2511 &first_block, &free_list, 0);
2513 ASSERT(error != ENOENT);
2514 REMOVE_DEBUG_TRACE(__LINE__);
2517 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2520 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2522 error = xfs_droplink(tp, ip);
2524 REMOVE_DEBUG_TRACE(__LINE__);
2528 /* Determine if this is the last link while
2529 * we are in the transaction.
2531 link_zero = (ip)->i_d.di_nlink==0;
2534 * Take an extra ref on the inode so that it doesn't
2535 * go to xfs_inactive() from within the commit.
2540 * If this is a synchronous mount, make sure that the
2541 * remove transaction goes to disk before returning to
2544 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2545 xfs_trans_set_sync(tp);
2548 error = xfs_bmap_finish(&tp, &free_list, &committed);
2550 REMOVE_DEBUG_TRACE(__LINE__);
2554 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2561 * Before we drop our extra reference to the inode, purge it
2562 * from the refcache if it is there. By waiting until afterwards
2563 * to do the IRELE, we ensure that we won't go inactive in the
2564 * xfs_refcache_purge_ip routine (although that would be OK).
2566 xfs_refcache_purge_ip(ip);
2569 * If we are using filestreams, kill the stream association.
2570 * If the file is still open it may get a new one but that
2571 * will get killed on last close in xfs_close() so we don't
2572 * have to worry about that.
2574 if (link_zero && xfs_inode_is_filestream(ip))
2575 xfs_filestream_deassociate(ip);
2577 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2580 * Let interposed file systems know about removed links.
2582 bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero);
2586 /* Fall through to std_return with error = 0 */
2588 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
2589 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2590 dir_vp, DM_RIGHT_NULL,
2591 NULL, DM_RIGHT_NULL,
2592 name, NULL, dm_di_mode, error, 0);
2597 xfs_bmap_cancel(&free_list);
2598 cancel_flags |= XFS_TRANS_ABORT;
2599 xfs_trans_cancel(tp, cancel_flags);
2604 * In this case make sure to not release the inode until after
2605 * the current transaction is aborted. Releasing it beforehand
2606 * can cause us to go to xfs_inactive and start a recursive
2607 * transaction which can easily deadlock with the current one.
2609 xfs_bmap_cancel(&free_list);
2610 cancel_flags |= XFS_TRANS_ABORT;
2611 xfs_trans_cancel(tp, cancel_flags);
2614 * Before we drop our extra reference to the inode, purge it
2615 * from the refcache if it is there. By waiting until afterwards
2616 * to do the IRELE, we ensure that we won't go inactive in the
2617 * xfs_refcache_purge_ip routine (although that would be OK).
2619 xfs_refcache_purge_ip(ip);
2633 bhv_desc_t *target_dir_bdp,
2634 bhv_vnode_t *src_vp,
2635 bhv_vname_t *dentry,
2638 xfs_inode_t *tdp, *sip;
2641 xfs_inode_t *ips[2];
2643 xfs_bmap_free_t free_list;
2644 xfs_fsblock_t first_block;
2647 bhv_vnode_t *target_dir_vp;
2649 char *target_name = VNAME(dentry);
2652 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2653 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2654 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2656 target_namelen = VNAMELEN(dentry);
2657 ASSERT(!VN_ISDIR(src_vp));
2659 sip = xfs_vtoi(src_vp);
2660 tdp = XFS_BHVTOI(target_dir_bdp);
2662 if (XFS_FORCED_SHUTDOWN(mp))
2663 return XFS_ERROR(EIO);
2665 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
2666 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2667 target_dir_vp, DM_RIGHT_NULL,
2668 src_vp, DM_RIGHT_NULL,
2669 target_name, NULL, 0, 0, 0);
2674 /* Return through std_return after this point. */
2676 error = XFS_QM_DQATTACH(mp, sip, 0);
2677 if (!error && sip != tdp)
2678 error = XFS_QM_DQATTACH(mp, tdp, 0);
2682 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2683 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2684 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2685 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2686 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2687 if (error == ENOSPC) {
2689 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2690 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2697 if (sip->i_ino < tdp->i_ino) {
2705 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2708 * Increment vnode ref counts since xfs_trans_commit &
2709 * xfs_trans_cancel will both unlock the inodes and
2710 * decrement the associated ref counts.
2713 VN_HOLD(target_dir_vp);
2714 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2715 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2718 * If the source has too many links, we can't make any more to it.
2720 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2721 error = XFS_ERROR(EMLINK);
2726 * If we are using project inheritance, we only allow hard link
2727 * creation in our tree when the project IDs are the same; else
2728 * the tree quota mechanism could be circumvented.
2730 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2731 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2732 error = XFS_ERROR(EXDEV);
2737 (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen)))
2740 XFS_BMAP_INIT(&free_list, &first_block);
2742 error = xfs_dir_createname(tp, tdp, target_name, target_namelen,
2743 sip->i_ino, &first_block, &free_list,
2747 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2749 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2751 error = xfs_bumplink(tp, sip);
2756 * If this is a synchronous mount, make sure that the
2757 * link transaction goes to disk before returning to
2760 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2761 xfs_trans_set_sync(tp);
2764 error = xfs_bmap_finish (&tp, &free_list, &committed);
2766 xfs_bmap_cancel(&free_list);
2770 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2774 /* Fall through to std_return with error = 0. */
2776 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
2777 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2778 target_dir_vp, DM_RIGHT_NULL,
2779 src_vp, DM_RIGHT_NULL,
2780 target_name, NULL, 0, error, 0);
2785 cancel_flags |= XFS_TRANS_ABORT;
2789 xfs_trans_cancel(tp, cancel_flags);
2800 bhv_desc_t *dir_bdp,
2801 bhv_vname_t *dentry,
2806 char *dir_name = VNAME(dentry);
2808 xfs_inode_t *cdp; /* inode of created dir */
2809 bhv_vnode_t *cvp; /* vnode of created dir */
2815 xfs_bmap_free_t free_list;
2816 xfs_fsblock_t first_block;
2817 bhv_vnode_t *dir_vp;
2818 boolean_t dp_joined_to_trans;
2819 boolean_t created = B_FALSE;
2820 int dm_event_sent = 0;
2822 struct xfs_dquot *udqp, *gdqp;
2827 dir_vp = BHV_TO_VNODE(dir_bdp);
2828 dp = XFS_BHVTOI(dir_bdp);
2831 if (XFS_FORCED_SHUTDOWN(mp))
2832 return XFS_ERROR(EIO);
2834 dir_namelen = VNAMELEN(dentry);
2837 dp_joined_to_trans = B_FALSE;
2838 dm_di_mode = vap->va_mode;
2840 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
2841 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2842 dir_vp, DM_RIGHT_NULL, NULL,
2843 DM_RIGHT_NULL, dir_name, NULL,
2850 /* Return through std_return after this point. */
2852 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2856 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2857 prid = dp->i_d.di_projid;
2858 else if (vap->va_mask & XFS_AT_PROJID)
2859 prid = (xfs_prid_t)vap->va_projid;
2861 prid = (xfs_prid_t)dfltprid;
2864 * Make sure that we have allocated dquot(s) on disk.
2866 error = XFS_QM_DQVOPALLOC(mp, dp,
2867 current_fsuid(credp), current_fsgid(credp), prid,
2868 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2872 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2873 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2874 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2875 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2876 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2877 if (error == ENOSPC) {
2879 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2880 XFS_TRANS_PERM_LOG_RES,
2881 XFS_MKDIR_LOG_COUNT);
2889 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2892 * Check for directory link count overflow.
2894 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2895 error = XFS_ERROR(EMLINK);
2900 * Reserve disk quota and the inode.
2902 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2907 (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen)))
2910 * create the directory inode.
2912 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2913 0, credp, prid, resblks > 0,
2916 if (error == ENOSPC)
2923 * Now we add the directory inode to the transaction.
2924 * We waited until now since xfs_dir_ialloc might start
2925 * a new transaction. Had we joined the transaction
2926 * earlier, the locks might have gotten released.
2929 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2930 dp_joined_to_trans = B_TRUE;
2932 XFS_BMAP_INIT(&free_list, &first_block);
2934 error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino,
2935 &first_block, &free_list, resblks ?
2936 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2938 ASSERT(error != ENOSPC);
2941 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2944 * Bump the in memory version number of the parent directory
2945 * so that other processes accessing it will recognize that
2946 * the directory has changed.
2950 error = xfs_dir_init(tp, cdp, dp);
2955 error = xfs_bumplink(tp, dp);
2959 cvp = XFS_ITOV(cdp);
2967 * Attach the dquots to the new inode and modify the icount incore.
2969 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2972 * If this is a synchronous mount, make sure that the
2973 * mkdir transaction goes to disk before returning to
2976 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2977 xfs_trans_set_sync(tp);
2980 error = xfs_bmap_finish(&tp, &free_list, &committed);
2986 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2987 XFS_QM_DQRELE(mp, udqp);
2988 XFS_QM_DQRELE(mp, gdqp);
2993 /* Fall through to std_return with error = 0 or errno from
2994 * xfs_trans_commit. */
2997 if ((created || (error != 0 && dm_event_sent != 0)) &&
2998 DM_EVENT_ENABLED(XFS_BHVTOI(dir_bdp), DM_EVENT_POSTCREATE)) {
2999 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
3000 dir_vp, DM_RIGHT_NULL,
3001 created ? XFS_ITOV(cdp):NULL,
3004 dm_di_mode, error, 0);
3010 xfs_bmap_cancel(&free_list);
3012 cancel_flags |= XFS_TRANS_ABORT;
3014 xfs_trans_cancel(tp, cancel_flags);
3015 XFS_QM_DQRELE(mp, udqp);
3016 XFS_QM_DQRELE(mp, gdqp);
3018 if (!dp_joined_to_trans && (dp != NULL)) {
3019 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3032 bhv_desc_t *dir_bdp,
3033 bhv_vname_t *dentry,
3036 char *name = VNAME(dentry);
3038 xfs_inode_t *cdp; /* child directory */
3042 xfs_bmap_free_t free_list;
3043 xfs_fsblock_t first_block;
3046 bhv_vnode_t *dir_vp;
3047 int dm_di_mode = S_IFDIR;
3052 dir_vp = BHV_TO_VNODE(dir_bdp);
3053 dp = XFS_BHVTOI(dir_bdp);
3056 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3058 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3059 return XFS_ERROR(EIO);
3060 namelen = VNAMELEN(dentry);
3062 if (!xfs_get_dir_entry(dentry, &cdp)) {
3063 dm_di_mode = cdp->i_d.di_mode;
3067 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
3068 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3069 dir_vp, DM_RIGHT_NULL,
3070 NULL, DM_RIGHT_NULL,
3071 name, NULL, dm_di_mode, 0, 0);
3073 return XFS_ERROR(error);
3076 /* Return through std_return after this point. */
3081 * We need to get a reference to cdp before we get our log
3082 * reservation. The reason for this is that we cannot call
3083 * xfs_iget for an inode for which we do not have a reference
3084 * once we've acquired a log reservation. This is because the
3085 * inode we are trying to get might be in xfs_inactive going
3086 * for a log reservation. Since we'll have to wait for the
3087 * inactive code to complete before returning from xfs_iget,
3088 * we need to make sure that we don't have log space reserved
3089 * when we call xfs_iget. Instead we get an unlocked reference
3090 * to the inode before getting our log reservation.
3092 error = xfs_get_dir_entry(dentry, &cdp);
3094 REMOVE_DEBUG_TRACE(__LINE__);
3098 dm_di_mode = cdp->i_d.di_mode;
3101 * Get the dquots for the inodes.
3103 error = XFS_QM_DQATTACH(mp, dp, 0);
3104 if (!error && dp != cdp)
3105 error = XFS_QM_DQATTACH(mp, cdp, 0);
3108 REMOVE_DEBUG_TRACE(__LINE__);
3112 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3113 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3115 * We try to get the real space reservation first,
3116 * allowing for directory btree deletion(s) implying
3117 * possible bmap insert(s). If we can't get the space
3118 * reservation then we use 0 instead, and avoid the bmap
3119 * btree insert(s) in the directory code by, if the bmap
3120 * insert tries to happen, instead trimming the LAST
3121 * block from the directory.
3123 resblks = XFS_REMOVE_SPACE_RES(mp);
3124 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3125 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3126 if (error == ENOSPC) {
3128 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3129 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3132 ASSERT(error != ENOSPC);
3137 XFS_BMAP_INIT(&free_list, &first_block);
3140 * Now lock the child directory inode and the parent directory
3141 * inode in the proper order. This will take care of validating
3142 * that the directory entry for the child directory inode has
3143 * not changed while we were obtaining a log reservation.
3145 error = xfs_lock_dir_and_entry(dp, cdp);
3147 xfs_trans_cancel(tp, cancel_flags);
3152 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3155 * Only increment the parent directory vnode count if
3156 * we didn't bump it in looking up cdp. The only time
3157 * we don't bump it is when we're looking up ".".
3163 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3165 ASSERT(cdp->i_d.di_nlink >= 2);
3166 if (cdp->i_d.di_nlink != 2) {
3167 error = XFS_ERROR(ENOTEMPTY);
3170 if (!xfs_dir_isempty(cdp)) {
3171 error = XFS_ERROR(ENOTEMPTY);
3175 error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino,
3176 &first_block, &free_list, resblks);
3180 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3183 * Bump the in memory generation count on the parent
3184 * directory so that other can know that it has changed.
3189 * Drop the link from cdp's "..".
3191 error = xfs_droplink(tp, dp);
3197 * Drop the link from dp to cdp.
3199 error = xfs_droplink(tp, cdp);
3205 * Drop the "." link from cdp to self.
3207 error = xfs_droplink(tp, cdp);
3212 /* Determine these before committing transaction */
3213 last_cdp_link = (cdp)->i_d.di_nlink==0;
3216 * Take an extra ref on the child vnode so that it
3217 * does not go to xfs_inactive() from within the commit.
3222 * If this is a synchronous mount, make sure that the
3223 * rmdir transaction goes to disk before returning to
3226 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3227 xfs_trans_set_sync(tp);
3230 error = xfs_bmap_finish (&tp, &free_list, &committed);
3232 xfs_bmap_cancel(&free_list);
3233 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3239 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3247 * Let interposed file systems know about removed links.
3249 bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3253 /* Fall through to std_return with error = 0 or the errno
3254 * from xfs_trans_commit. */
3256 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
3257 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3258 dir_vp, DM_RIGHT_NULL,
3259 NULL, DM_RIGHT_NULL,
3260 name, NULL, dm_di_mode,
3266 xfs_bmap_cancel(&free_list);
3267 cancel_flags |= XFS_TRANS_ABORT;
3271 xfs_trans_cancel(tp, cancel_flags);
3277 bhv_desc_t *dir_bdp,
3278 bhv_vname_t *dentry,
3290 xfs_bmap_free_t free_list;
3291 xfs_fsblock_t first_block;
3292 boolean_t dp_joined_to_trans;
3293 bhv_vnode_t *dir_vp;
3296 xfs_fileoff_t first_fsb;
3297 xfs_filblks_t fs_blocks;
3299 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3306 struct xfs_dquot *udqp, *gdqp;
3308 char *link_name = VNAME(dentry);
3312 dir_vp = BHV_TO_VNODE(dir_bdp);
3313 dp = XFS_BHVTOI(dir_bdp);
3314 dp_joined_to_trans = B_FALSE;
3319 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3323 if (XFS_FORCED_SHUTDOWN(mp))
3324 return XFS_ERROR(EIO);
3326 link_namelen = VNAMELEN(dentry);
3329 * Check component lengths of the target path name.
3331 pathlen = strlen(target_path);
3332 if (pathlen >= MAXPATHLEN) /* total string too long */
3333 return XFS_ERROR(ENAMETOOLONG);
3334 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3338 for (total = 0, path = target_path; total < pathlen;) {
3342 while(*path == '/') {
3348 * Count up to the next slash or end of path.
3349 * Error out if the component is bigger than MAXNAMELEN.
3351 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3352 if (++len >= MAXNAMELEN) {
3353 error = ENAMETOOLONG;
3360 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
3361 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3362 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3363 link_name, target_path, 0, 0, 0);
3368 /* Return through std_return after this point. */
3371 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3372 prid = dp->i_d.di_projid;
3373 else if (vap->va_mask & XFS_AT_PROJID)
3374 prid = (xfs_prid_t)vap->va_projid;
3376 prid = (xfs_prid_t)dfltprid;
3379 * Make sure that we have allocated dquot(s) on disk.
3381 error = XFS_QM_DQVOPALLOC(mp, dp,
3382 current_fsuid(credp), current_fsgid(credp), prid,
3383 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3387 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3388 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3390 * The symlink will fit into the inode data fork?
3391 * There can't be any attributes so we get the whole variable part.
3393 if (pathlen <= XFS_LITINO(mp))
3396 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3397 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3398 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3399 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3400 if (error == ENOSPC && fs_blocks == 0) {
3402 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3403 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3411 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
3414 * Check whether the directory allows new symlinks or not.
3416 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3417 error = XFS_ERROR(EPERM);
3422 * Reserve disk quota : blocks and inode.
3424 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3429 * Check for ability to enter directory entry, if no space reserved.
3432 (error = xfs_dir_canenter(tp, dp, link_name, link_namelen)))
3435 * Initialize the bmap freelist prior to calling either
3436 * bmapi or the directory create code.
3438 XFS_BMAP_INIT(&free_list, &first_block);
3441 * Allocate an inode for the symlink.
3443 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3444 1, 0, credp, prid, resblks > 0, &ip, NULL);
3446 if (error == ENOSPC)
3453 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3454 dp_joined_to_trans = B_TRUE;
3457 * Also attach the dquot(s) to it, if applicable.
3459 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3462 resblks -= XFS_IALLOC_SPACE_RES(mp);
3464 * If the symlink will fit into the inode, write it inline.
3466 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3467 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3468 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3469 ip->i_d.di_size = pathlen;
3472 * The inode was initially created in extent format.
3474 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3475 ip->i_df.if_flags |= XFS_IFINLINE;
3477 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3478 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3482 nmaps = SYMLINK_MAPS;
3484 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3485 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3486 &first_block, resblks, mval, &nmaps,
3493 resblks -= fs_blocks;
3494 ip->i_d.di_size = pathlen;
3495 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3497 cur_chunk = target_path;
3498 for (n = 0; n < nmaps; n++) {
3499 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3500 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3501 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3502 BTOBB(byte_cnt), 0);
3503 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3504 if (pathlen < byte_cnt) {
3507 pathlen -= byte_cnt;
3509 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3510 cur_chunk += byte_cnt;
3512 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3517 * Create the directory entry for the symlink.
3519 error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino,
3520 &first_block, &free_list, resblks);
3523 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3524 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3527 * Bump the in memory version number of the parent directory
3528 * so that other processes accessing it will recognize that
3529 * the directory has changed.
3534 * If this is a synchronous mount, make sure that the
3535 * symlink transaction goes to disk before returning to
3538 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3539 xfs_trans_set_sync(tp);
3543 * xfs_trans_commit normally decrements the vnode ref count
3544 * when it unlocks the inode. Since we want to return the
3545 * vnode to the caller, we bump the vnode ref count now.
3549 error = xfs_bmap_finish(&tp, &free_list, &committed);
3553 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3554 XFS_QM_DQRELE(mp, udqp);
3555 XFS_QM_DQRELE(mp, gdqp);
3557 /* Fall through to std_return with error = 0 or errno from
3558 * xfs_trans_commit */
3560 if (DM_EVENT_ENABLED(XFS_BHVTOI(dir_bdp), DM_EVENT_POSTSYMLINK)) {
3561 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3562 dir_vp, DM_RIGHT_NULL,
3563 error ? NULL : XFS_ITOV(ip),
3564 DM_RIGHT_NULL, link_name, target_path,
3580 xfs_bmap_cancel(&free_list);
3581 cancel_flags |= XFS_TRANS_ABORT;
3583 xfs_trans_cancel(tp, cancel_flags);
3584 XFS_QM_DQRELE(mp, udqp);
3585 XFS_QM_DQRELE(mp, gdqp);
3587 if (!dp_joined_to_trans && (dp != NULL)) {
3588 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3598 * A fid routine that takes a pointer to a previously allocated
3599 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3609 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3610 (inst_t *)__return_address);
3611 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3613 xfid = (xfs_fid2_t *)fidp;
3614 ip = XFS_BHVTOI(bdp);
3615 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3618 * use memcpy because the inode is a long long and there's no
3619 * assurance that xfid->fid_ino is properly aligned.
3621 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3622 xfid->fid_gen = ip->i_d.di_gen;
3634 bhv_vrwlock_t locktype)
3639 vp = BHV_TO_VNODE(bdp);
3642 ip = XFS_BHVTOI(bdp);
3643 if (locktype == VRWLOCK_WRITE) {
3644 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3645 } else if (locktype == VRWLOCK_TRY_READ) {
3646 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3647 } else if (locktype == VRWLOCK_TRY_WRITE) {
3648 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3650 ASSERT((locktype == VRWLOCK_READ) ||
3651 (locktype == VRWLOCK_WRITE_DIRECT));
3652 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3665 bhv_vrwlock_t locktype)
3670 vp = BHV_TO_VNODE(bdp);
3673 ip = XFS_BHVTOI(bdp);
3674 if (locktype == VRWLOCK_WRITE) {
3676 * In the write case, we may have added a new entry to
3677 * the reference cache. This might store a pointer to
3678 * an inode to be released in this inode. If it is there,
3679 * clear the pointer and release the inode after unlocking
3682 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3684 ASSERT((locktype == VRWLOCK_READ) ||
3685 (locktype == VRWLOCK_WRITE_DIRECT));
3686 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3698 xfs_inode_log_item_t *iip;
3701 ip = XFS_BHVTOI(bdp);
3705 if (XFS_FORCED_SHUTDOWN(mp))
3706 return XFS_ERROR(EIO);
3709 * Bypass inodes which have already been cleaned by
3710 * the inode flush clustering code inside xfs_iflush
3712 if ((ip->i_update_core == 0) &&
3713 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3716 if (flags & FLUSH_LOG) {
3717 if (iip && iip->ili_last_lsn) {
3718 xlog_t *log = mp->m_log;
3720 int s, log_flags = XFS_LOG_FORCE;
3722 s = GRANT_LOCK(log);
3723 sync_lsn = log->l_last_sync_lsn;
3724 GRANT_UNLOCK(log, s);
3726 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) > 0)) {
3727 if (flags & FLUSH_SYNC)
3728 log_flags |= XFS_LOG_SYNC;
3729 error = xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3734 if (ip->i_update_core == 0)
3740 * We make this non-blocking if the inode is contended,
3741 * return EAGAIN to indicate to the caller that they
3742 * did not succeed. This prevents the flush path from
3743 * blocking on inodes inside another operation right
3744 * now, they get caught later by xfs_sync.
3746 if (flags & FLUSH_INODE) {
3749 if (flags & FLUSH_SYNC) {
3750 xfs_ilock(ip, XFS_ILOCK_SHARED);
3752 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3753 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3754 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3761 if (flags & FLUSH_SYNC)
3762 flush_flags = XFS_IFLUSH_SYNC;
3764 flush_flags = XFS_IFLUSH_ASYNC;
3766 error = xfs_iflush(ip, flush_flags);
3767 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3785 if (!capable(CAP_SYS_ADMIN))
3786 return XFS_ERROR(EPERM);
3788 ip = XFS_BHVTOI(bdp);
3791 if (XFS_FORCED_SHUTDOWN(mp))
3792 return XFS_ERROR(EIO);
3794 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3795 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3797 xfs_trans_cancel(tp, 0);
3800 xfs_ilock(ip, XFS_ILOCK_EXCL);
3801 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3803 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3804 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3806 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3808 error = xfs_trans_commit(tp, 0);
3820 vp = BHV_TO_VNODE(bdp);
3821 ip = XFS_BHVTOI(bdp);
3823 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3825 ASSERT(!VN_MAPPED(vp));
3827 /* bad inode, get out here ASAP */
3835 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3838 * Make sure the atime in the XFS inode is correct before freeing the
3841 xfs_synchronize_atime(ip);
3844 * If we have nothing to flush with this inode then complete the
3845 * teardown now, otherwise break the link between the xfs inode and the
3846 * linux inode and clean up the xfs inode later. This avoids flushing
3847 * the inode to disk during the delete operation itself.
3849 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3850 * first to ensure that xfs_iunpin() will never see an xfs inode
3851 * that has a linux inode being reclaimed. Synchronisation is provided
3852 * by the i_flags_lock.
3854 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3855 xfs_ilock(ip, XFS_ILOCK_EXCL);
3857 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3859 xfs_mount_t *mp = ip->i_mount;
3861 /* Protect sync and unpin from us */
3862 XFS_MOUNT_ILOCK(mp);
3863 spin_lock(&ip->i_flags_lock);
3864 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
3865 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3866 spin_unlock(&ip->i_flags_lock);
3867 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3868 XFS_MOUNT_IUNLOCK(mp);
3879 xfs_ihash_t *ih = ip->i_hash;
3880 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
3883 if (vp && VN_BAD(vp))
3886 /* The hash lock here protects a thread in xfs_iget_core from
3887 * racing with us on linking the inode back with a vnode.
3888 * Once we have the XFS_IRECLAIM flag set it will not touch
3891 write_lock(&ih->ih_lock);
3892 spin_lock(&ip->i_flags_lock);
3893 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3894 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
3895 spin_unlock(&ip->i_flags_lock);
3896 write_unlock(&ih->ih_lock);
3899 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3903 __xfs_iflags_set(ip, XFS_IRECLAIM);
3904 spin_unlock(&ip->i_flags_lock);
3905 write_unlock(&ih->ih_lock);
3908 * If the inode is still dirty, then flush it out. If the inode
3909 * is not in the AIL, then it will be OK to flush it delwri as
3910 * long as xfs_iflush() does not keep any references to the inode.
3911 * We leave that decision up to xfs_iflush() since it has the
3912 * knowledge of whether it's OK to simply do a delwri flush of
3913 * the inode or whether we need to wait until the inode is
3914 * pulled from the AIL.
3915 * We get the flush lock regardless, though, just to make sure
3916 * we don't free it while it is being flushed.
3918 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3920 xfs_ilock(ip, XFS_ILOCK_EXCL);
3924 if (ip->i_update_core ||
3925 ((ip->i_itemp != NULL) &&
3926 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3927 error = xfs_iflush(ip, sync_mode);
3929 * If we hit an error, typically because of filesystem
3930 * shutdown, we don't need to let vn_reclaim to know
3931 * because we're gonna reclaim the inode anyway.
3934 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3937 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3940 ASSERT(ip->i_update_core == 0);
3941 ASSERT(ip->i_itemp == NULL ||
3942 ip->i_itemp->ili_format.ilf_fields == 0);
3943 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3944 } else if (locked) {
3946 * We are not interested in doing an iflush if we're
3947 * in the process of shutting down the filesystem forcibly.
3948 * So, just reclaim the inode.
3951 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3960 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3963 xfs_inode_t *ip, *n;
3968 XFS_MOUNT_ILOCK(mp);
3969 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3971 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3973 if (xfs_ipincount(ip) ||
3974 !xfs_iflock_nowait(ip)) {
3975 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3979 XFS_MOUNT_IUNLOCK(mp);
3980 if (xfs_finish_reclaim(ip, noblock,
3981 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3990 XFS_MOUNT_IUNLOCK(mp);
3995 * xfs_alloc_file_space()
3996 * This routine allocates disk space for the given file.
3998 * If alloc_type == 0, this request is for an ALLOCSP type
3999 * request which will change the file size. In this case, no
4000 * DMAPI event will be generated by the call. A TRUNCATE event
4001 * will be generated later by xfs_setattr.
4003 * If alloc_type != 0, this request is for a RESVSP type
4004 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4005 * lower block boundary byte address is less than the file's
4014 xfs_alloc_file_space(
4021 xfs_mount_t *mp = ip->i_mount;
4023 xfs_filblks_t allocated_fsb;
4024 xfs_filblks_t allocatesize_fsb;
4025 xfs_extlen_t extsz, temp;
4026 xfs_fileoff_t startoffset_fsb;
4027 xfs_fsblock_t firstfsb;
4033 xfs_bmbt_irec_t imaps[1], *imapp;
4034 xfs_bmap_free_t free_list;
4035 uint qblocks, resblks, resrtextents;
4039 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4041 if (XFS_FORCED_SHUTDOWN(mp))
4042 return XFS_ERROR(EIO);
4044 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4048 return XFS_ERROR(EINVAL);
4050 rt = XFS_IS_REALTIME_INODE(ip);
4051 extsz = xfs_get_extsz_hint(ip);
4056 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4057 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4058 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4060 /* Generate a DMAPI event if needed. */
4061 if (alloc_type != 0 && offset < ip->i_size &&
4062 (attr_flags&ATTR_DMI) == 0 &&
4063 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
4064 xfs_off_t end_dmi_offset;
4066 end_dmi_offset = offset+len;
4067 if (end_dmi_offset > ip->i_size)
4068 end_dmi_offset = ip->i_size;
4069 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4070 offset, end_dmi_offset - offset,
4077 * Allocate file space until done or until there is an error
4080 while (allocatesize_fsb && !error) {
4084 * Determine space reservations for data/realtime.
4086 if (unlikely(extsz)) {
4087 s = startoffset_fsb;
4090 e = startoffset_fsb + allocatesize_fsb;
4091 if ((temp = do_mod(startoffset_fsb, extsz)))
4093 if ((temp = do_mod(e, extsz)))
4097 e = allocatesize_fsb;
4101 resrtextents = qblocks = (uint)(e - s);
4102 resrtextents /= mp->m_sb.sb_rextsize;
4103 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4104 quota_flag = XFS_QMOPT_RES_RTBLKS;
4107 resblks = qblocks = \
4108 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4109 quota_flag = XFS_QMOPT_RES_REGBLKS;
4113 * Allocate and setup the transaction.
4115 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4116 error = xfs_trans_reserve(tp, resblks,
4117 XFS_WRITE_LOG_RES(mp), resrtextents,
4118 XFS_TRANS_PERM_LOG_RES,
4119 XFS_WRITE_LOG_COUNT);
4121 * Check for running out of space
4125 * Free the transaction structure.
4127 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4128 xfs_trans_cancel(tp, 0);
4131 xfs_ilock(ip, XFS_ILOCK_EXCL);
4132 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4133 qblocks, 0, quota_flag);
4137 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4138 xfs_trans_ihold(tp, ip);
4141 * Issue the xfs_bmapi() call to allocate the blocks
4143 XFS_BMAP_INIT(&free_list, &firstfsb);
4144 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4145 allocatesize_fsb, bmapi_flag,
4146 &firstfsb, 0, imapp, &nimaps,
4153 * Complete the transaction
4155 error = xfs_bmap_finish(&tp, &free_list, &committed);
4160 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4161 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4166 allocated_fsb = imapp->br_blockcount;
4169 error = XFS_ERROR(ENOSPC);
4173 startoffset_fsb += allocated_fsb;
4174 allocatesize_fsb -= allocated_fsb;
4177 if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 &&
4178 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
4179 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4180 XFS_ITOV(ip), DM_RIGHT_NULL,
4181 XFS_ITOV(ip), DM_RIGHT_NULL,
4182 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4184 goto retry; /* Maybe DMAPI app. has made space */
4185 /* else fall through with error from XFS_SEND_DATA */
4190 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4191 xfs_bmap_cancel(&free_list);
4192 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4194 error1: /* Just cancel transaction */
4195 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4196 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4197 goto dmapi_enospc_check;
4201 * Zero file bytes between startoff and endoff inclusive.
4202 * The iolock is held exclusive and no blocks are buffered.
4205 xfs_zero_remaining_bytes(
4210 xfs_bmbt_irec_t imap;
4211 xfs_fileoff_t offset_fsb;
4212 xfs_off_t lastoffset;
4215 xfs_mount_t *mp = ip->i_mount;
4219 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4220 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4221 mp->m_rtdev_targp : mp->m_ddev_targp);
4223 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4224 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4226 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4227 NULL, 0, &imap, &nimap, NULL, NULL);
4228 if (error || nimap < 1)
4230 ASSERT(imap.br_blockcount >= 1);
4231 ASSERT(imap.br_startoff == offset_fsb);
4232 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4233 if (lastoffset > endoff)
4234 lastoffset = endoff;
4235 if (imap.br_startblock == HOLESTARTBLOCK)
4237 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4238 if (imap.br_state == XFS_EXT_UNWRITTEN)
4241 XFS_BUF_UNWRITE(bp);
4243 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4245 if ((error = xfs_iowait(bp))) {
4246 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4247 mp, bp, XFS_BUF_ADDR(bp));
4250 memset(XFS_BUF_PTR(bp) +
4251 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4252 0, lastoffset - offset + 1);
4257 if ((error = xfs_iowait(bp))) {
4258 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4259 mp, bp, XFS_BUF_ADDR(bp));
4268 * xfs_free_file_space()
4269 * This routine frees disk space for the given file.
4271 * This routine is only called by xfs_change_file_space
4272 * for an UNRESVSP type call.
4280 xfs_free_file_space(
4289 xfs_off_t end_dmi_offset;
4290 xfs_fileoff_t endoffset_fsb;
4292 xfs_fsblock_t firstfsb;
4293 xfs_bmap_free_t free_list;
4294 xfs_bmbt_irec_t imap;
4302 xfs_fileoff_t startoffset_fsb;
4304 int need_iolock = 1;
4309 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4311 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4315 if (len <= 0) /* if nothing being freed */
4317 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4318 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4319 end_dmi_offset = offset + len;
4320 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4322 if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 &&
4323 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
4324 if (end_dmi_offset > ip->i_size)
4325 end_dmi_offset = ip->i_size;
4326 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4327 offset, end_dmi_offset - offset,
4328 AT_DELAY_FLAG(attr_flags), NULL);
4333 if (attr_flags & ATTR_NOLOCK)
4336 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4337 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4340 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP);
4341 ioffset = offset & ~(rounding - 1);
4343 if (VN_CACHED(vp) != 0) {
4344 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4345 ctooff(offtoct(ioffset)), -1);
4346 error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
4347 -1, FI_REMAPF_LOCKED);
4349 goto out_unlock_iolock;
4353 * Need to zero the stuff we're not freeing, on disk.
4354 * If its a realtime file & can't use unwritten extents then we
4355 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4356 * will take care of it for us.
4358 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4360 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4361 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4363 goto out_unlock_iolock;
4364 ASSERT(nimap == 0 || nimap == 1);
4365 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4368 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4369 block = imap.br_startblock;
4370 mod = do_div(block, mp->m_sb.sb_rextsize);
4372 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4375 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4376 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4378 goto out_unlock_iolock;
4379 ASSERT(nimap == 0 || nimap == 1);
4380 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4381 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4383 if (mod && (mod != mp->m_sb.sb_rextsize))
4384 endoffset_fsb -= mod;
4387 if ((done = (endoffset_fsb <= startoffset_fsb)))
4389 * One contiguous piece to clear
4391 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4394 * Some full blocks, possibly two pieces to clear
4396 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4397 error = xfs_zero_remaining_bytes(ip, offset,
4398 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4400 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4401 error = xfs_zero_remaining_bytes(ip,
4402 XFS_FSB_TO_B(mp, endoffset_fsb),
4407 * free file space until done or until there is an error
4409 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4410 while (!error && !done) {
4413 * allocate and setup the transaction. Allow this
4414 * transaction to dip into the reserve blocks to ensure
4415 * the freeing of the space succeeds at ENOSPC.
4417 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4418 tp->t_flags |= XFS_TRANS_RESERVE;
4419 error = xfs_trans_reserve(tp,
4421 XFS_WRITE_LOG_RES(mp),
4423 XFS_TRANS_PERM_LOG_RES,
4424 XFS_WRITE_LOG_COUNT);
4427 * check for running out of space
4431 * Free the transaction structure.
4433 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4434 xfs_trans_cancel(tp, 0);
4437 xfs_ilock(ip, XFS_ILOCK_EXCL);
4438 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4439 ip->i_udquot, ip->i_gdquot, resblks, 0,
4440 XFS_QMOPT_RES_REGBLKS);
4444 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4445 xfs_trans_ihold(tp, ip);
4448 * issue the bunmapi() call to free the blocks
4450 XFS_BMAP_INIT(&free_list, &firstfsb);
4451 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4452 endoffset_fsb - startoffset_fsb,
4453 0, 2, &firstfsb, &free_list, NULL, &done);
4459 * complete the transaction
4461 error = xfs_bmap_finish(&tp, &free_list, &committed);
4466 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4467 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4472 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4476 xfs_bmap_cancel(&free_list);
4478 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4479 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4485 * xfs_change_file_space()
4486 * This routine allocates or frees disk space for the given file.
4487 * The user specified parameters are checked for alignment and size
4496 xfs_change_file_space(
4510 xfs_off_t startoffset;
4516 vp = BHV_TO_VNODE(bdp);
4517 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4519 ip = XFS_BHVTOI(bdp);
4523 * must be a regular file and have write permission
4526 return XFS_ERROR(EINVAL);
4528 xfs_ilock(ip, XFS_ILOCK_SHARED);
4530 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4531 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4535 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4537 switch (bf->l_whence) {
4538 case 0: /*SEEK_SET*/
4540 case 1: /*SEEK_CUR*/
4541 bf->l_start += offset;
4543 case 2: /*SEEK_END*/
4544 bf->l_start += ip->i_size;
4547 return XFS_ERROR(EINVAL);
4550 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4552 if ( (bf->l_start < 0)
4553 || (bf->l_start > XFS_MAXIOFFSET(mp))
4554 || (bf->l_start + llen < 0)
4555 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4556 return XFS_ERROR(EINVAL);
4560 startoffset = bf->l_start;
4564 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4566 * These calls do NOT zero the data space allocated to the file,
4567 * nor do they change the file size.
4569 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4571 * These calls cause the new file data to be zeroed and the file
4572 * size to be changed.
4574 setprealloc = clrprealloc = 0;
4577 case XFS_IOC_RESVSP:
4578 case XFS_IOC_RESVSP64:
4579 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4586 case XFS_IOC_UNRESVSP:
4587 case XFS_IOC_UNRESVSP64:
4588 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4593 case XFS_IOC_ALLOCSP:
4594 case XFS_IOC_ALLOCSP64:
4595 case XFS_IOC_FREESP:
4596 case XFS_IOC_FREESP64:
4597 if (startoffset > fsize) {
4598 error = xfs_alloc_file_space(ip, fsize,
4599 startoffset - fsize, 0, attr_flags);
4604 va.va_mask = XFS_AT_SIZE;
4605 va.va_size = startoffset;
4607 error = xfs_setattr(bdp, &va, attr_flags, credp);
4617 return XFS_ERROR(EINVAL);
4621 * update the inode timestamp, mode, and prealloc flag bits
4623 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4625 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4628 xfs_trans_cancel(tp, 0);
4632 xfs_ilock(ip, XFS_ILOCK_EXCL);
4634 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4635 xfs_trans_ihold(tp, ip);
4637 if ((attr_flags & ATTR_DMI) == 0) {
4638 ip->i_d.di_mode &= ~S_ISUID;
4641 * Note that we don't have to worry about mandatory
4642 * file locking being disabled here because we only
4643 * clear the S_ISGID bit if the Group execute bit is
4644 * on, but if it was on then mandatory locking wouldn't
4645 * have been enabled.
4647 if (ip->i_d.di_mode & S_IXGRP)
4648 ip->i_d.di_mode &= ~S_ISGID;
4650 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4653 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4654 else if (clrprealloc)
4655 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4657 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4658 xfs_trans_set_sync(tp);
4660 error = xfs_trans_commit(tp, 0);
4662 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4667 bhv_vnodeops_t xfs_vnodeops = {
4668 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4669 .vop_open = xfs_open,
4670 .vop_read = xfs_read,
4672 .vop_splice_read = xfs_splice_read,
4673 .vop_splice_write = xfs_splice_write,
4675 .vop_write = xfs_write,
4676 .vop_ioctl = xfs_ioctl,
4677 .vop_getattr = xfs_getattr,
4678 .vop_setattr = xfs_setattr,
4679 .vop_access = xfs_access,
4680 .vop_lookup = xfs_lookup,
4681 .vop_create = xfs_create,
4682 .vop_remove = xfs_remove,
4683 .vop_link = xfs_link,
4684 .vop_rename = xfs_rename,
4685 .vop_mkdir = xfs_mkdir,
4686 .vop_rmdir = xfs_rmdir,
4687 .vop_readdir = xfs_readdir,
4688 .vop_symlink = xfs_symlink,
4689 .vop_readlink = xfs_readlink,
4690 .vop_fsync = xfs_fsync,
4691 .vop_inactive = xfs_inactive,
4692 .vop_fid2 = xfs_fid2,
4693 .vop_rwlock = xfs_rwlock,
4694 .vop_rwunlock = xfs_rwunlock,
4695 .vop_bmap = xfs_bmap,
4696 .vop_reclaim = xfs_reclaim,
4697 .vop_attr_get = xfs_attr_get,
4698 .vop_attr_set = xfs_attr_set,
4699 .vop_attr_remove = xfs_attr_remove,
4700 .vop_attr_list = xfs_attr_list,
4701 .vop_link_removed = (vop_link_removed_t)fs_noval,
4702 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4703 .vop_tosspages = fs_tosspages,
4704 .vop_flushinval_pages = fs_flushinval_pages,
4705 .vop_flush_pages = fs_flush_pages,
4706 .vop_release = xfs_release,
4707 .vop_iflush = xfs_inode_flush,