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"
61 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
62 xfs_inode_t *ip = XFS_BHVTOI(bdp);
64 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
65 return XFS_ERROR(EIO);
68 * If it's a directory with any blocks, read-ahead block 0
69 * as we're almost certain to have the next operation be a read there.
71 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
72 mode = xfs_ilock_map_shared(ip);
73 if (ip->i_d.di_nextents > 0)
74 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
75 xfs_iunlock(ip, mode);
84 lastclose_t lastclose,
87 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
88 xfs_inode_t *ip = XFS_BHVTOI(bdp);
90 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
91 return XFS_ERROR(EIO);
93 if (lastclose != L_TRUE || !VN_ISREG(vp))
97 * If we previously truncated this file and removed old data in
98 * the process, we want to initiate "early" writeout on the last
99 * close. This is an attempt to combat the notorious NULL files
100 * problem which is particularly noticable from a truncate down,
101 * buffered (re-)write (delalloc), followed by a crash. What we
102 * are effectively doing here is significantly reducing the time
103 * window where we'd otherwise be exposed to that problem.
105 if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
106 return bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE);
124 vp = BHV_TO_VNODE(bdp);
125 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
127 ip = XFS_BHVTOI(bdp);
130 if (XFS_FORCED_SHUTDOWN(mp))
131 return XFS_ERROR(EIO);
133 if (!(flags & ATTR_LAZY))
134 xfs_ilock(ip, XFS_ILOCK_SHARED);
136 vap->va_size = XFS_ISIZE(ip);
137 if (vap->va_mask == XFS_AT_SIZE)
141 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
142 vap->va_nodeid = ip->i_ino;
144 vap->va_nodeid += mp->m_inoadd;
146 vap->va_nlink = ip->i_d.di_nlink;
149 * Quick exit for non-stat callers
152 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
153 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
157 * Copy from in-core inode.
159 vap->va_mode = ip->i_d.di_mode;
160 vap->va_uid = ip->i_d.di_uid;
161 vap->va_gid = ip->i_d.di_gid;
162 vap->va_projid = ip->i_d.di_projid;
165 * Check vnode type block/char vs. everything else.
167 switch (ip->i_d.di_mode & S_IFMT) {
170 vap->va_rdev = ip->i_df.if_u2.if_rdev;
171 vap->va_blocksize = BLKDEV_IOSIZE;
176 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
177 vap->va_blocksize = xfs_preferred_iosize(mp);
181 * If the file blocks are being allocated from a
182 * realtime partition, then return the inode's
183 * realtime extent size or the realtime volume's
186 vap->va_blocksize = ip->i_d.di_extsize ?
187 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
188 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
193 vn_atime_to_timespec(vp, &vap->va_atime);
194 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
195 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
196 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
197 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
200 * Exit for stat callers. See if any of the rest of the fields
201 * to be filled in are needed.
204 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
205 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
209 * Convert di_flags to xflags.
211 vap->va_xflags = xfs_ip2xflags(ip);
214 * Exit for inode revalidate. See if any of the rest of
215 * the fields to be filled in are needed.
218 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
219 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
222 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
224 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
225 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
229 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
230 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
231 ip->i_d.di_anextents;
233 vap->va_anextents = 0;
234 vap->va_gen = ip->i_d.di_gen;
237 if (!(flags & ATTR_LAZY))
238 xfs_iunlock(ip, XFS_ILOCK_SHARED);
264 xfs_prid_t projid=0, iprojid=0;
265 int mandlock_before, mandlock_after;
266 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
270 vp = BHV_TO_VNODE(bdp);
271 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
273 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
274 return XFS_ERROR(EROFS);
277 * Cannot set certain attributes.
280 if (mask & XFS_AT_NOSET) {
281 return XFS_ERROR(EINVAL);
284 ip = XFS_BHVTOI(bdp);
287 if (XFS_FORCED_SHUTDOWN(mp))
288 return XFS_ERROR(EIO);
291 * Timestamps do not need to be logged and hence do not
292 * need to be done within a transaction.
294 if (mask & XFS_AT_UPDTIMES) {
295 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
296 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
297 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
298 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
299 xfs_ichgtime(ip, timeflags);
303 olddquot1 = olddquot2 = NULL;
307 * If disk quotas is on, we make sure that the dquots do exist on disk,
308 * before we start any other transactions. Trying to do this later
309 * is messy. We don't care to take a readlock to look at the ids
310 * in inode here, because we can't hold it across the trans_reserve.
311 * If the IDs do change before we take the ilock, we're covered
312 * because the i_*dquot fields will get updated anyway.
314 if (XFS_IS_QUOTA_ON(mp) &&
315 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
318 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
320 qflags |= XFS_QMOPT_UQUOTA;
322 uid = ip->i_d.di_uid;
324 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
326 qflags |= XFS_QMOPT_GQUOTA;
328 gid = ip->i_d.di_gid;
330 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
331 projid = vap->va_projid;
332 qflags |= XFS_QMOPT_PQUOTA;
334 projid = ip->i_d.di_projid;
337 * We take a reference when we initialize udqp and gdqp,
338 * so it is important that we never blindly double trip on
339 * the same variable. See xfs_create() for an example.
341 ASSERT(udqp == NULL);
342 ASSERT(gdqp == NULL);
343 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
350 * For the other attributes, we acquire the inode lock and
351 * first do an error checking pass.
354 lock_flags = XFS_ILOCK_EXCL;
355 if (flags & ATTR_NOLOCK)
357 if (!(mask & XFS_AT_SIZE)) {
358 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
359 (mp->m_flags & XFS_MOUNT_WSYNC)) {
360 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
362 if ((code = xfs_trans_reserve(tp, 0,
363 XFS_ICHANGE_LOG_RES(mp), 0,
370 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
371 !(flags & ATTR_DMI)) {
372 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
373 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
374 vap->va_size, 0, dmflags, NULL);
381 lock_flags |= XFS_IOLOCK_EXCL;
384 xfs_ilock(ip, lock_flags);
386 /* boolean: are we the file owner? */
387 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
390 * Change various properties of a file.
391 * Only the owner or users with CAP_FOWNER
392 * capability may do these things.
395 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
396 XFS_AT_GID|XFS_AT_PROJID)) {
398 * CAP_FOWNER overrides the following restrictions:
400 * The user ID of the calling process must be equal
401 * to the file owner ID, except in cases where the
402 * CAP_FSETID capability is applicable.
404 if (!file_owner && !capable(CAP_FOWNER)) {
405 code = XFS_ERROR(EPERM);
410 * CAP_FSETID overrides the following restrictions:
412 * The effective user ID of the calling process shall match
413 * the file owner when setting the set-user-ID and
414 * set-group-ID bits on that file.
416 * The effective group ID or one of the supplementary group
417 * IDs of the calling process shall match the group owner of
418 * the file when setting the set-group-ID bit on that file
420 if (mask & XFS_AT_MODE) {
423 if ((vap->va_mode & S_ISUID) && !file_owner)
425 if ((vap->va_mode & S_ISGID) &&
426 !in_group_p((gid_t)ip->i_d.di_gid))
429 /* Linux allows this, Irix doesn't. */
430 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
433 if (m && !capable(CAP_FSETID))
439 * Change file ownership. Must be the owner or privileged.
440 * If the system was configured with the "restricted_chown"
441 * option, the owner is not permitted to give away the file,
442 * and can change the group id only to a group of which he
443 * or she is a member.
445 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
447 * These IDs could have changed since we last looked at them.
448 * But, we're assured that if the ownership did change
449 * while we didn't have the inode locked, inode's dquot(s)
450 * would have changed also.
452 iuid = ip->i_d.di_uid;
453 iprojid = ip->i_d.di_projid;
454 igid = ip->i_d.di_gid;
455 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
456 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
457 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
461 * CAP_CHOWN overrides the following restrictions:
463 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
464 * shall override the restriction that a process cannot
465 * change the user ID of a file it owns and the restriction
466 * that the group ID supplied to the chown() function
467 * shall be equal to either the group ID or one of the
468 * supplementary group IDs of the calling process.
470 if (restricted_chown &&
471 (iuid != uid || (igid != gid &&
472 !in_group_p((gid_t)gid))) &&
473 !capable(CAP_CHOWN)) {
474 code = XFS_ERROR(EPERM);
478 * Do a quota reservation only if uid/projid/gid is actually
481 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
482 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
483 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
485 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
486 capable(CAP_FOWNER) ?
487 XFS_QMOPT_FORCE_RES : 0);
488 if (code) /* out of quota */
494 * Truncate file. Must have write permission and not be a directory.
496 if (mask & XFS_AT_SIZE) {
497 /* Short circuit the truncate case for zero length files */
498 if ((vap->va_size == 0) &&
499 (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) {
500 xfs_iunlock(ip, XFS_ILOCK_EXCL);
501 lock_flags &= ~XFS_ILOCK_EXCL;
502 if (mask & XFS_AT_CTIME)
503 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
509 code = XFS_ERROR(EISDIR);
511 } else if (!VN_ISREG(vp)) {
512 code = XFS_ERROR(EINVAL);
516 * Make sure that the dquots are attached to the inode.
518 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
523 * Change file access or modified times.
525 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
527 if ((flags & ATTR_UTIME) &&
528 !capable(CAP_FOWNER)) {
529 code = XFS_ERROR(EPERM);
536 * Change extent size or realtime flag.
538 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
540 * Can't change extent size if any extents are allocated.
542 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
543 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
545 code = XFS_ERROR(EINVAL); /* EFBIG? */
550 * Can't change realtime flag if any extents are allocated.
552 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
553 (mask & XFS_AT_XFLAGS) &&
554 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
555 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
556 code = XFS_ERROR(EINVAL); /* EFBIG? */
560 * Extent size must be a multiple of the appropriate block
561 * size, if set at all.
563 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
566 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
567 ((mask & XFS_AT_XFLAGS) &&
568 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
569 size = mp->m_sb.sb_rextsize <<
570 mp->m_sb.sb_blocklog;
572 size = mp->m_sb.sb_blocksize;
574 if (vap->va_extsize % size) {
575 code = XFS_ERROR(EINVAL);
580 * If realtime flag is set then must have realtime data.
582 if ((mask & XFS_AT_XFLAGS) &&
583 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
584 if ((mp->m_sb.sb_rblocks == 0) ||
585 (mp->m_sb.sb_rextsize == 0) ||
586 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
587 code = XFS_ERROR(EINVAL);
593 * Can't modify an immutable/append-only file unless
594 * we have appropriate permission.
596 if ((mask & XFS_AT_XFLAGS) &&
598 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
600 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
601 !capable(CAP_LINUX_IMMUTABLE)) {
602 code = XFS_ERROR(EPERM);
608 * Now we can make the changes. Before we join the inode
609 * to the transaction, if XFS_AT_SIZE is set then take care of
610 * the part of the truncation that must be done without the
611 * inode lock. This needs to be done before joining the inode
612 * to the transaction, because the inode cannot be unlocked
613 * once it is a part of the transaction.
615 if (mask & XFS_AT_SIZE) {
617 if ((vap->va_size > ip->i_size) &&
618 (flags & ATTR_NOSIZETOK) == 0) {
619 code = xfs_igrow_start(ip, vap->va_size, credp);
621 xfs_iunlock(ip, XFS_ILOCK_EXCL);
622 vn_iowait(vp); /* wait for the completion of any pending DIOs */
624 code = xfs_itruncate_data(ip, vap->va_size);
627 lock_flags &= ~XFS_ILOCK_EXCL;
628 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
631 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
632 if ((code = xfs_trans_reserve(tp, 0,
633 XFS_ITRUNCATE_LOG_RES(mp), 0,
634 XFS_TRANS_PERM_LOG_RES,
635 XFS_ITRUNCATE_LOG_COUNT))) {
636 xfs_trans_cancel(tp, 0);
638 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
641 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
642 xfs_ilock(ip, XFS_ILOCK_EXCL);
646 xfs_trans_ijoin(tp, ip, lock_flags);
647 xfs_trans_ihold(tp, ip);
650 /* determine whether mandatory locking mode changes */
651 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
654 * Truncate file. Must have write permission and not be a directory.
656 if (mask & XFS_AT_SIZE) {
657 if (vap->va_size > ip->i_size) {
658 xfs_igrow_finish(tp, ip, vap->va_size,
659 !(flags & ATTR_DMI));
660 } else if ((vap->va_size <= ip->i_size) ||
661 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
663 * signal a sync transaction unless
664 * we're truncating an already unlinked
665 * file on a wsync filesystem
667 code = xfs_itruncate_finish(&tp, ip,
668 (xfs_fsize_t)vap->va_size,
670 ((ip->i_d.di_nlink != 0 ||
671 !(mp->m_flags & XFS_MOUNT_WSYNC))
676 * Truncated "down", so we're removing references
677 * to old data here - if we now delay flushing for
678 * a long time, we expose ourselves unduly to the
679 * notorious NULL files problem. So, we mark this
680 * vnode and flush it when the file is closed, and
681 * do not wait the usual (long) time for writeout.
686 * Have to do this even if the file's size doesn't change.
688 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
692 * Change file access modes.
694 if (mask & XFS_AT_MODE) {
695 ip->i_d.di_mode &= S_IFMT;
696 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
698 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
699 timeflags |= XFS_ICHGTIME_CHG;
703 * Change file ownership. Must be the owner or privileged.
704 * If the system was configured with the "restricted_chown"
705 * option, the owner is not permitted to give away the file,
706 * and can change the group id only to a group of which he
707 * or she is a member.
709 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
711 * CAP_FSETID overrides the following restrictions:
713 * The set-user-ID and set-group-ID bits of a file will be
714 * cleared upon successful return from chown()
716 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
717 !capable(CAP_FSETID)) {
718 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
722 * Change the ownerships and register quota modifications
723 * in the transaction.
726 if (XFS_IS_UQUOTA_ON(mp)) {
727 ASSERT(mask & XFS_AT_UID);
729 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
730 &ip->i_udquot, udqp);
732 ip->i_d.di_uid = uid;
735 if (XFS_IS_GQUOTA_ON(mp)) {
736 ASSERT(!XFS_IS_PQUOTA_ON(mp));
737 ASSERT(mask & XFS_AT_GID);
739 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
740 &ip->i_gdquot, gdqp);
742 ip->i_d.di_gid = gid;
744 if (iprojid != projid) {
745 if (XFS_IS_PQUOTA_ON(mp)) {
746 ASSERT(!XFS_IS_GQUOTA_ON(mp));
747 ASSERT(mask & XFS_AT_PROJID);
749 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
750 &ip->i_gdquot, gdqp);
752 ip->i_d.di_projid = projid;
754 * We may have to rev the inode as well as
755 * the superblock version number since projids didn't
756 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
758 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
759 xfs_bump_ino_vers2(tp, ip);
762 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
763 timeflags |= XFS_ICHGTIME_CHG;
768 * Change file access or modified times.
770 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
771 if (mask & XFS_AT_ATIME) {
772 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
773 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
774 ip->i_update_core = 1;
775 timeflags &= ~XFS_ICHGTIME_ACC;
777 if (mask & XFS_AT_MTIME) {
778 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
779 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
780 timeflags &= ~XFS_ICHGTIME_MOD;
781 timeflags |= XFS_ICHGTIME_CHG;
783 if (tp && (flags & ATTR_UTIME))
784 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
788 * Change XFS-added attributes.
790 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
791 if (mask & XFS_AT_EXTSIZE) {
793 * Converting bytes to fs blocks.
795 ip->i_d.di_extsize = vap->va_extsize >>
796 mp->m_sb.sb_blocklog;
798 if (mask & XFS_AT_XFLAGS) {
801 /* can't set PREALLOC this way, just preserve it */
802 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
803 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
804 di_flags |= XFS_DIFLAG_IMMUTABLE;
805 if (vap->va_xflags & XFS_XFLAG_APPEND)
806 di_flags |= XFS_DIFLAG_APPEND;
807 if (vap->va_xflags & XFS_XFLAG_SYNC)
808 di_flags |= XFS_DIFLAG_SYNC;
809 if (vap->va_xflags & XFS_XFLAG_NOATIME)
810 di_flags |= XFS_DIFLAG_NOATIME;
811 if (vap->va_xflags & XFS_XFLAG_NODUMP)
812 di_flags |= XFS_DIFLAG_NODUMP;
813 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
814 di_flags |= XFS_DIFLAG_PROJINHERIT;
815 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
816 di_flags |= XFS_DIFLAG_NODEFRAG;
817 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
818 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
819 di_flags |= XFS_DIFLAG_RTINHERIT;
820 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
821 di_flags |= XFS_DIFLAG_NOSYMLINKS;
822 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
823 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
824 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
825 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
826 di_flags |= XFS_DIFLAG_REALTIME;
827 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
829 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
831 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
832 di_flags |= XFS_DIFLAG_EXTSIZE;
834 ip->i_d.di_flags = di_flags;
836 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
837 timeflags |= XFS_ICHGTIME_CHG;
841 * Change file inode change time only if XFS_AT_CTIME set
842 * AND we have been called by a DMI function.
845 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
846 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
847 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
848 ip->i_update_core = 1;
849 timeflags &= ~XFS_ICHGTIME_CHG;
853 * Send out timestamp changes that need to be set to the
854 * current time. Not done when called by a DMI function.
856 if (timeflags && !(flags & ATTR_DMI))
857 xfs_ichgtime(ip, timeflags);
859 XFS_STATS_INC(xs_ig_attrchg);
862 * If this is a synchronous mount, make sure that the
863 * transaction goes to disk before returning to the user.
864 * This is slightly sub-optimal in that truncates require
865 * two sync transactions instead of one for wsync filesystems.
866 * One for the truncate and one for the timestamps since we
867 * don't want to change the timestamps unless we're sure the
868 * truncate worked. Truncates are less than 1% of the laddis
869 * mix so this probably isn't worth the trouble to optimize.
873 if (mp->m_flags & XFS_MOUNT_WSYNC)
874 xfs_trans_set_sync(tp);
876 code = xfs_trans_commit(tp, commit_flags);
880 * If the (regular) file's mandatory locking mode changed, then
881 * notify the vnode. We do this under the inode lock to prevent
882 * racing calls to vop_vnode_change.
884 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
885 if (mandlock_before != mandlock_after) {
886 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING,
890 xfs_iunlock(ip, lock_flags);
893 * Release any dquot(s) the inode had kept before chown.
895 XFS_QM_DQRELE(mp, olddquot1);
896 XFS_QM_DQRELE(mp, olddquot2);
897 XFS_QM_DQRELE(mp, udqp);
898 XFS_QM_DQRELE(mp, gdqp);
904 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
905 !(flags & ATTR_DMI)) {
906 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
907 NULL, DM_RIGHT_NULL, NULL, NULL,
908 0, 0, AT_DELAY_FLAG(flags));
913 commit_flags |= XFS_TRANS_ABORT;
916 XFS_QM_DQRELE(mp, udqp);
917 XFS_QM_DQRELE(mp, gdqp);
919 xfs_trans_cancel(tp, commit_flags);
921 if (lock_flags != 0) {
922 xfs_iunlock(ip, lock_flags);
930 * Null conversion from vnode mode bits to inode mode bits, as in efs.
941 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
942 (inst_t *)__return_address);
944 ip = XFS_BHVTOI(bdp);
945 xfs_ilock(ip, XFS_ILOCK_SHARED);
946 error = xfs_iaccess(ip, mode, credp);
947 xfs_iunlock(ip, XFS_ILOCK_SHARED);
953 * The maximum pathlen is 1024 bytes. Since the minimum file system
954 * blocksize is 512 bytes, we can get a max of 2 extents back from
957 #define SYMLINK_MAPS 2
978 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
984 vp = BHV_TO_VNODE(bdp);
985 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
987 ip = XFS_BHVTOI(bdp);
990 if (XFS_FORCED_SHUTDOWN(mp))
991 return XFS_ERROR(EIO);
993 xfs_ilock(ip, XFS_ILOCK_SHARED);
995 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
997 offset = uiop->uio_offset;
998 count = uiop->uio_resid;
1001 error = XFS_ERROR(EINVAL);
1010 * See if the symlink is stored inline.
1012 pathlen = (int)ip->i_d.di_size;
1014 if (ip->i_df.if_flags & XFS_IFINLINE) {
1015 error = xfs_uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1019 * Symlink not inline. Call bmap to get it in.
1021 nmaps = SYMLINK_MAPS;
1023 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1024 0, NULL, 0, mval, &nmaps, NULL, NULL);
1030 for (n = 0; n < nmaps; n++) {
1031 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1032 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1033 bp = xfs_buf_read(mp->m_ddev_targp, d,
1034 BTOBB(byte_cnt), 0);
1035 error = XFS_BUF_GETERROR(bp);
1037 xfs_ioerror_alert("xfs_readlink",
1038 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1042 if (pathlen < byte_cnt)
1044 pathlen -= byte_cnt;
1046 error = xfs_uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1053 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1061 * This is called to sync the inode and its data out to disk.
1062 * We need to hold the I/O lock while flushing the data, and
1063 * the inode lock while flushing the inode. The inode lock CANNOT
1064 * be held while flushing the data, so acquire after we're done
1078 int log_flushed = 0, changed = 1;
1080 vn_trace_entry(BHV_TO_VNODE(bdp),
1081 __FUNCTION__, (inst_t *)__return_address);
1083 ip = XFS_BHVTOI(bdp);
1085 ASSERT(start >= 0 && stop >= -1);
1087 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1088 return XFS_ERROR(EIO);
1091 * We always need to make sure that the required inode state
1092 * is safe on disk. The vnode might be clean but because
1093 * of committed transactions that haven't hit the disk yet.
1094 * Likewise, there could be unflushed non-transactional
1095 * changes to the inode core that have to go to disk.
1097 * The following code depends on one assumption: that
1098 * any transaction that changes an inode logs the core
1099 * because it has to change some field in the inode core
1100 * (typically nextents or nblocks). That assumption
1101 * implies that any transactions against an inode will
1102 * catch any non-transactional updates. If inode-altering
1103 * transactions exist that violate this assumption, the
1104 * code breaks. Right now, it figures that if the involved
1105 * update_* field is clear and the inode is unpinned, the
1106 * inode is clean. Either it's been flushed or it's been
1107 * committed and the commit has hit the disk unpinning the inode.
1108 * (Note that xfs_inode_item_format() called at commit clears
1109 * the update_* fields.)
1111 xfs_ilock(ip, XFS_ILOCK_SHARED);
1113 /* If we are flushing data then we care about update_size
1114 * being set, otherwise we care about update_core
1116 if ((flag & FSYNC_DATA) ?
1117 (ip->i_update_size == 0) :
1118 (ip->i_update_core == 0)) {
1120 * Timestamps/size haven't changed since last inode
1121 * flush or inode transaction commit. That means
1122 * either nothing got written or a transaction
1123 * committed which caught the updates. If the
1124 * latter happened and the transaction hasn't
1125 * hit the disk yet, the inode will be still
1126 * be pinned. If it is, force the log.
1129 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1131 if (xfs_ipincount(ip)) {
1132 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1134 ((flag & FSYNC_WAIT)
1135 ? XFS_LOG_SYNC : 0),
1139 * If the inode is not pinned and nothing
1140 * has changed we don't need to flush the
1148 * Kick off a transaction to log the inode
1149 * core to get the updates. Make it
1150 * sync if FSYNC_WAIT is passed in (which
1151 * is done by everybody but specfs). The
1152 * sync transaction will also force the log.
1154 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1155 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1156 if ((error = xfs_trans_reserve(tp, 0,
1157 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1159 xfs_trans_cancel(tp, 0);
1162 xfs_ilock(ip, XFS_ILOCK_EXCL);
1165 * Note - it's possible that we might have pushed
1166 * ourselves out of the way during trans_reserve
1167 * which would flush the inode. But there's no
1168 * guarantee that the inode buffer has actually
1169 * gone out yet (it's delwri). Plus the buffer
1170 * could be pinned anyway if it's part of an
1171 * inode in another recent transaction. So we
1172 * play it safe and fire off the transaction anyway.
1174 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1175 xfs_trans_ihold(tp, ip);
1176 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1177 if (flag & FSYNC_WAIT)
1178 xfs_trans_set_sync(tp);
1179 error = _xfs_trans_commit(tp, 0, &log_flushed);
1181 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1184 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1186 * If the log write didn't issue an ordered tag we need
1187 * to flush the disk cache for the data device now.
1190 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1193 * If this inode is on the RT dev we need to flush that
1196 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1197 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1204 * This is called by xfs_inactive to free any blocks beyond eof,
1205 * when the link count isn't zero.
1208 xfs_inactive_free_eofblocks(
1214 xfs_fileoff_t end_fsb;
1215 xfs_fileoff_t last_fsb;
1216 xfs_filblks_t map_len;
1218 xfs_bmbt_irec_t imap;
1221 * Figure out if there are any blocks beyond the end
1222 * of the file. If not, then there is nothing to do.
1224 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
1225 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1226 map_len = last_fsb - end_fsb;
1231 xfs_ilock(ip, XFS_ILOCK_SHARED);
1232 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1233 NULL, 0, &imap, &nimaps, NULL, NULL);
1234 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1236 if (!error && (nimaps != 0) &&
1237 (imap.br_startblock != HOLESTARTBLOCK ||
1238 ip->i_delayed_blks)) {
1240 * Attach the dquots to the inode up front.
1242 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1246 * There are blocks after the end of file.
1247 * Free them up now by truncating the file to
1250 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1253 * Do the xfs_itruncate_start() call before
1254 * reserving any log space because
1255 * itruncate_start will call into the buffer
1256 * cache and we can't
1257 * do that within a transaction.
1259 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1260 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1263 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1267 error = xfs_trans_reserve(tp, 0,
1268 XFS_ITRUNCATE_LOG_RES(mp),
1269 0, XFS_TRANS_PERM_LOG_RES,
1270 XFS_ITRUNCATE_LOG_COUNT);
1272 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1273 xfs_trans_cancel(tp, 0);
1274 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1278 xfs_ilock(ip, XFS_ILOCK_EXCL);
1279 xfs_trans_ijoin(tp, ip,
1282 xfs_trans_ihold(tp, ip);
1284 error = xfs_itruncate_finish(&tp, ip,
1289 * If we get an error at this point we
1290 * simply don't bother truncating the file.
1293 xfs_trans_cancel(tp,
1294 (XFS_TRANS_RELEASE_LOG_RES |
1297 error = xfs_trans_commit(tp,
1298 XFS_TRANS_RELEASE_LOG_RES);
1300 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1306 * Free a symlink that has blocks associated with it.
1309 xfs_inactive_symlink_rmt(
1317 xfs_fsblock_t first_block;
1318 xfs_bmap_free_t free_list;
1321 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1329 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1331 * We're freeing a symlink that has some
1332 * blocks allocated to it. Free the
1333 * blocks here. We know that we've got
1334 * either 1 or 2 extents and that we can
1335 * free them all in one bunmapi call.
1337 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1338 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1339 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1340 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1341 xfs_trans_cancel(tp, 0);
1346 * Lock the inode, fix the size, and join it to the transaction.
1347 * Hold it so in the normal path, we still have it locked for
1348 * the second transaction. In the error paths we need it
1349 * held so the cancel won't rele it, see below.
1351 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1352 size = (int)ip->i_d.di_size;
1353 ip->i_d.di_size = 0;
1354 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1355 xfs_trans_ihold(tp, ip);
1356 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1358 * Find the block(s) so we can inval and unmap them.
1361 XFS_BMAP_INIT(&free_list, &first_block);
1362 nmaps = ARRAY_SIZE(mval);
1363 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1364 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1368 * Invalidate the block(s).
1370 for (i = 0; i < nmaps; i++) {
1371 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1372 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1373 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1374 xfs_trans_binval(tp, bp);
1377 * Unmap the dead block(s) to the free_list.
1379 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1380 &first_block, &free_list, NULL, &done)))
1384 * Commit the first transaction. This logs the EFI and the inode.
1386 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
1389 * The transaction must have been committed, since there were
1390 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1391 * The new tp has the extent freeing and EFDs.
1395 * The first xact was committed, so add the inode to the new one.
1396 * Mark it dirty so it will be logged and moved forward in the log as
1397 * part of every commit.
1399 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1400 xfs_trans_ihold(tp, ip);
1401 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1403 * Get a new, empty transaction to return to our caller.
1405 ntp = xfs_trans_dup(tp);
1407 * Commit the transaction containing extent freeing and EFDs.
1408 * If we get an error on the commit here or on the reserve below,
1409 * we need to unlock the inode since the new transaction doesn't
1410 * have the inode attached.
1412 error = xfs_trans_commit(tp, 0);
1415 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1419 * Remove the memory for extent descriptions (just bookkeeping).
1421 if (ip->i_df.if_bytes)
1422 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1423 ASSERT(ip->i_df.if_bytes == 0);
1425 * Put an itruncate log reservation in the new transaction
1428 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1429 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1430 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1434 * Return with the inode locked but not joined to the transaction.
1440 xfs_bmap_cancel(&free_list);
1443 * Have to come here with the inode locked and either
1444 * (held and in the transaction) or (not in the transaction).
1445 * If the inode isn't held then cancel would iput it, but
1446 * that's wrong since this is inactive and the vnode ref
1447 * count is 0 already.
1448 * Cancel won't do anything to the inode if held, but it still
1449 * needs to be locked until the cancel is done, if it was
1450 * joined to the transaction.
1452 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1453 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1460 xfs_inactive_symlink_local(
1466 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1468 * We're freeing a symlink which fit into
1469 * the inode. Just free the memory used
1470 * to hold the old symlink.
1472 error = xfs_trans_reserve(*tpp, 0,
1473 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1474 0, XFS_TRANS_PERM_LOG_RES,
1475 XFS_ITRUNCATE_LOG_COUNT);
1478 xfs_trans_cancel(*tpp, 0);
1482 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1485 * Zero length symlinks _can_ exist.
1487 if (ip->i_df.if_bytes > 0) {
1488 xfs_idata_realloc(ip,
1489 -(ip->i_df.if_bytes),
1491 ASSERT(ip->i_df.if_bytes == 0);
1505 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1508 ASSERT(ip->i_d.di_forkoff != 0);
1509 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1510 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1512 error = xfs_attr_inactive(ip);
1515 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1516 return error; /* goto out */
1519 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1520 error = xfs_trans_reserve(tp, 0,
1521 XFS_IFREE_LOG_RES(mp),
1522 0, XFS_TRANS_PERM_LOG_RES,
1523 XFS_INACTIVE_LOG_COUNT);
1525 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1526 xfs_trans_cancel(tp, 0);
1528 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1532 xfs_ilock(ip, XFS_ILOCK_EXCL);
1533 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1534 xfs_trans_ihold(tp, ip);
1535 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1537 ASSERT(ip->i_d.di_anextents == 0);
1552 vp = BHV_TO_VNODE(bdp);
1553 ip = XFS_BHVTOI(bdp);
1556 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
1559 /* If this is a read-only mount, don't do this (would generate I/O) */
1560 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1563 #ifdef HAVE_REFCACHE
1564 /* If we are in the NFS reference cache then don't do this now */
1569 if (ip->i_d.di_nlink != 0) {
1570 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1571 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1572 ip->i_delayed_blks > 0)) &&
1573 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1574 (!(ip->i_d.di_flags &
1575 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1576 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1578 /* Update linux inode block count after free above */
1579 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1580 ip->i_d.di_nblocks + ip->i_delayed_blks);
1590 * This is called when the vnode reference count for the vnode
1591 * goes to zero. If the file has been unlinked, then it must
1592 * now be truncated. Also, we clear all of the read-ahead state
1593 * kept for the inode here since the file is now closed.
1602 xfs_bmap_free_t free_list;
1603 xfs_fsblock_t first_block;
1610 vp = BHV_TO_VNODE(bdp);
1611 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1613 ip = XFS_BHVTOI(bdp);
1616 * If the inode is already free, then there can be nothing
1619 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1620 ASSERT(ip->i_df.if_real_bytes == 0);
1621 ASSERT(ip->i_df.if_broot_bytes == 0);
1622 return VN_INACTIVE_CACHE;
1626 * Only do a truncate if it's a regular file with
1627 * some actual space in it. It's OK to look at the
1628 * inode's fields without the lock because we're the
1629 * only one with a reference to the inode.
1631 truncate = ((ip->i_d.di_nlink == 0) &&
1632 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1633 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1634 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1638 if (ip->i_d.di_nlink == 0 &&
1639 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1640 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1645 /* If this is a read-only mount, don't do this (would generate I/O) */
1646 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1649 if (ip->i_d.di_nlink != 0) {
1650 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1651 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1652 ip->i_delayed_blks > 0)) &&
1653 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1654 (!(ip->i_d.di_flags &
1655 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1656 (ip->i_delayed_blks != 0)))) {
1657 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1658 return VN_INACTIVE_CACHE;
1659 /* Update linux inode block count after free above */
1660 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1661 ip->i_d.di_nblocks + ip->i_delayed_blks);
1666 ASSERT(ip->i_d.di_nlink == 0);
1668 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1669 return VN_INACTIVE_CACHE;
1671 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1674 * Do the xfs_itruncate_start() call before
1675 * reserving any log space because itruncate_start
1676 * will call into the buffer cache and we can't
1677 * do that within a transaction.
1679 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1681 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1683 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1684 return VN_INACTIVE_CACHE;
1687 error = xfs_trans_reserve(tp, 0,
1688 XFS_ITRUNCATE_LOG_RES(mp),
1689 0, XFS_TRANS_PERM_LOG_RES,
1690 XFS_ITRUNCATE_LOG_COUNT);
1692 /* Don't call itruncate_cleanup */
1693 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1694 xfs_trans_cancel(tp, 0);
1695 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1696 return VN_INACTIVE_CACHE;
1699 xfs_ilock(ip, XFS_ILOCK_EXCL);
1700 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1701 xfs_trans_ihold(tp, ip);
1704 * normally, we have to run xfs_itruncate_finish sync.
1705 * But if filesystem is wsync and we're in the inactive
1706 * path, then we know that nlink == 0, and that the
1707 * xaction that made nlink == 0 is permanently committed
1708 * since xfs_remove runs as a synchronous transaction.
1710 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1711 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1714 xfs_trans_cancel(tp,
1715 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1716 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1717 return VN_INACTIVE_CACHE;
1719 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1722 * If we get an error while cleaning up a
1723 * symlink we bail out.
1725 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1726 xfs_inactive_symlink_rmt(ip, &tp) :
1727 xfs_inactive_symlink_local(ip, &tp);
1731 return VN_INACTIVE_CACHE;
1734 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1735 xfs_trans_ihold(tp, ip);
1737 error = xfs_trans_reserve(tp, 0,
1738 XFS_IFREE_LOG_RES(mp),
1739 0, XFS_TRANS_PERM_LOG_RES,
1740 XFS_INACTIVE_LOG_COUNT);
1742 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1743 xfs_trans_cancel(tp, 0);
1744 return VN_INACTIVE_CACHE;
1747 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1748 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1749 xfs_trans_ihold(tp, ip);
1753 * If there are attributes associated with the file
1754 * then blow them away now. The code calls a routine
1755 * that recursively deconstructs the attribute fork.
1756 * We need to just commit the current transaction
1757 * because we can't use it for xfs_attr_inactive().
1759 if (ip->i_d.di_anextents > 0) {
1760 error = xfs_inactive_attrs(ip, &tp);
1762 * If we got an error, the transaction is already
1763 * cancelled, and the inode is unlocked. Just get out.
1766 return VN_INACTIVE_CACHE;
1767 } else if (ip->i_afp) {
1768 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1774 XFS_BMAP_INIT(&free_list, &first_block);
1775 error = xfs_ifree(tp, ip, &free_list);
1778 * If we fail to free the inode, shut down. The cancel
1779 * might do that, we need to make sure. Otherwise the
1780 * inode might be lost for a long time or forever.
1782 if (!XFS_FORCED_SHUTDOWN(mp)) {
1784 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1785 error, mp->m_fsname);
1786 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1788 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1791 * Credit the quota account(s). The inode is gone.
1793 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1796 * Just ignore errors at this point. There is
1797 * nothing we can do except to try to keep going.
1799 (void) xfs_bmap_finish(&tp, &free_list, &committed);
1800 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1803 * Release the dquots held by inode, if any.
1805 XFS_QM_DQDETACH(mp, ip);
1807 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1810 return VN_INACTIVE_CACHE;
1819 bhv_desc_t *dir_bdp,
1820 bhv_vname_t *dentry,
1826 xfs_inode_t *dp, *ip;
1830 bhv_vnode_t *dir_vp;
1832 dir_vp = BHV_TO_VNODE(dir_bdp);
1833 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1835 dp = XFS_BHVTOI(dir_bdp);
1837 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1838 return XFS_ERROR(EIO);
1840 lock_mode = xfs_ilock_map_shared(dp);
1841 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1843 *vpp = XFS_ITOV(ip);
1846 xfs_iunlock_map_shared(dp, lock_mode);
1852 * xfs_create (create a new file).
1856 bhv_desc_t *dir_bdp,
1857 bhv_vname_t *dentry,
1862 char *name = VNAME(dentry);
1863 bhv_vnode_t *dir_vp;
1864 xfs_inode_t *dp, *ip;
1865 bhv_vnode_t *vp = NULL;
1870 xfs_bmap_free_t free_list;
1871 xfs_fsblock_t first_block;
1872 boolean_t dp_joined_to_trans;
1873 int dm_event_sent = 0;
1877 struct xfs_dquot *udqp, *gdqp;
1883 dir_vp = BHV_TO_VNODE(dir_bdp);
1884 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1886 dp = XFS_BHVTOI(dir_bdp);
1889 dm_di_mode = vap->va_mode;
1890 namelen = VNAMELEN(dentry);
1892 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1893 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1894 dir_vp, DM_RIGHT_NULL, NULL,
1895 DM_RIGHT_NULL, name, NULL,
1903 if (XFS_FORCED_SHUTDOWN(mp))
1904 return XFS_ERROR(EIO);
1906 /* Return through std_return after this point. */
1909 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1910 prid = dp->i_d.di_projid;
1911 else if (vap->va_mask & XFS_AT_PROJID)
1912 prid = (xfs_prid_t)vap->va_projid;
1914 prid = (xfs_prid_t)dfltprid;
1917 * Make sure that we have allocated dquot(s) on disk.
1919 error = XFS_QM_DQVOPALLOC(mp, dp,
1920 current_fsuid(credp), current_fsgid(credp), prid,
1921 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1926 dp_joined_to_trans = B_FALSE;
1928 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1929 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1930 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1932 * Initially assume that the file does not exist and
1933 * reserve the resources for that case. If that is not
1934 * the case we'll drop the one we have and get a more
1935 * appropriate transaction later.
1937 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1938 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1939 if (error == ENOSPC) {
1941 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1942 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1950 xfs_ilock(dp, XFS_ILOCK_EXCL);
1952 XFS_BMAP_INIT(&free_list, &first_block);
1957 * Reserve disk quota and the inode.
1959 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1963 if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen)))
1965 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1966 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1967 rdev, credp, prid, resblks > 0,
1970 if (error == ENOSPC)
1977 * At this point, we've gotten a newly allocated inode.
1978 * It is locked (and joined to the transaction).
1981 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1984 * Now we join the directory inode to the transaction.
1985 * We do not do it earlier because xfs_dir_ialloc
1986 * might commit the previous transaction (and release
1991 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1992 dp_joined_to_trans = B_TRUE;
1994 error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino,
1995 &first_block, &free_list, resblks ?
1996 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1998 ASSERT(error != ENOSPC);
2001 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2002 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2005 * If this is a synchronous mount, make sure that the
2006 * create transaction goes to disk before returning to
2009 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2010 xfs_trans_set_sync(tp);
2016 * Attach the dquot(s) to the inodes and modify them incore.
2017 * These ids of the inode couldn't have changed since the new
2018 * inode has been locked ever since it was created.
2020 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2023 * xfs_trans_commit normally decrements the vnode ref count
2024 * when it unlocks the inode. Since we want to return the
2025 * vnode to the caller, we bump the vnode ref count now.
2030 error = xfs_bmap_finish(&tp, &free_list, &committed);
2032 xfs_bmap_cancel(&free_list);
2036 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2043 XFS_QM_DQRELE(mp, udqp);
2044 XFS_QM_DQRELE(mp, gdqp);
2047 * Propagate the fact that the vnode changed after the
2048 * xfs_inode locks have been released.
2050 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2054 /* Fallthrough to std_return with error = 0 */
2057 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2058 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2059 DM_EVENT_POSTCREATE)) {
2060 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2061 dir_vp, DM_RIGHT_NULL,
2063 DM_RIGHT_NULL, name, NULL,
2064 dm_di_mode, error, 0);
2069 cancel_flags |= XFS_TRANS_ABORT;
2074 xfs_trans_cancel(tp, cancel_flags);
2076 if (!dp_joined_to_trans && (dp != NULL))
2077 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2078 XFS_QM_DQRELE(mp, udqp);
2079 XFS_QM_DQRELE(mp, gdqp);
2085 * Wait until after the current transaction is aborted to
2086 * release the inode. This prevents recursive transactions
2087 * and deadlocks from xfs_inactive.
2089 cancel_flags |= XFS_TRANS_ABORT;
2090 xfs_trans_cancel(tp, cancel_flags);
2093 XFS_QM_DQRELE(mp, udqp);
2094 XFS_QM_DQRELE(mp, gdqp);
2101 * Some counters to see if (and how often) we are hitting some deadlock
2102 * prevention code paths.
2106 int xfs_rm_lock_delays;
2107 int xfs_rm_attempts;
2111 * The following routine will lock the inodes associated with the
2112 * directory and the named entry in the directory. The locks are
2113 * acquired in increasing inode number.
2115 * If the entry is "..", then only the directory is locked. The
2116 * vnode ref count will still include that from the .. entry in
2119 * There is a deadlock we need to worry about. If the locked directory is
2120 * in the AIL, it might be blocking up the log. The next inode we lock
2121 * could be already locked by another thread waiting for log space (e.g
2122 * a permanent log reservation with a long running transaction (see
2123 * xfs_itruncate_finish)). To solve this, we must check if the directory
2124 * is in the ail and use lock_nowait. If we can't lock, we need to
2125 * drop the inode lock on the directory and try again. xfs_iunlock will
2126 * potentially push the tail if we were holding up the log.
2129 xfs_lock_dir_and_entry(
2131 xfs_inode_t *ip) /* inode of entry 'name' */
2135 xfs_inode_t *ips[2];
2144 xfs_ilock(dp, XFS_ILOCK_EXCL);
2151 * We want to lock in increasing inum. Since we've already
2152 * acquired the lock on the directory, we may need to release
2153 * if if the inum of the entry turns out to be less.
2155 if (e_inum > dp->i_ino) {
2157 * We are already in the right order, so just
2158 * lock on the inode of the entry.
2159 * We need to use nowait if dp is in the AIL.
2162 lp = (xfs_log_item_t *)dp->i_itemp;
2163 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2164 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2171 * Unlock dp and try again.
2172 * xfs_iunlock will try to push the tail
2173 * if the inode is in the AIL.
2176 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2178 if ((attempts % 5) == 0) {
2179 delay(1); /* Don't just spin the CPU */
2181 xfs_rm_lock_delays++;
2187 xfs_ilock(ip, XFS_ILOCK_EXCL);
2189 } else if (e_inum < dp->i_ino) {
2190 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2194 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2196 /* else e_inum == dp->i_ino */
2197 /* This can happen if we're asked to lock /x/..
2198 * the entry is "..", which is also the parent directory.
2206 int xfs_small_retries;
2207 int xfs_middle_retries;
2208 int xfs_lots_retries;
2209 int xfs_lock_delays;
2213 * The following routine will lock n inodes in exclusive mode.
2214 * We assume the caller calls us with the inodes in i_ino order.
2216 * We need to detect deadlock where an inode that we lock
2217 * is in the AIL and we start waiting for another inode that is locked
2218 * by a thread in a long running transaction (such as truncate). This can
2219 * result in deadlock since the long running trans might need to wait
2220 * for the inode we just locked in order to push the tail and free space
2230 int attempts = 0, i, j, try_lock;
2233 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2244 for (; i < inodes; i++) {
2247 if (i && (ips[i] == ips[i-1])) /* Already locked */
2251 * If try_lock is not set yet, make sure all locked inodes
2252 * are not in the AIL.
2253 * If any are, set try_lock to be used later.
2257 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2258 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2259 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2266 * If any of the previous locks we have locked is in the AIL,
2267 * we must TRY to get the second and subsequent locks. If
2268 * we can't get any, we must release all we have
2273 /* try_lock must be 0 if i is 0. */
2275 * try_lock means we have an inode locked
2276 * that is in the AIL.
2279 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2283 * Unlock all previous guys and try again.
2284 * xfs_iunlock will try to push the tail
2285 * if the inode is in the AIL.
2288 for(j = i - 1; j >= 0; j--) {
2291 * Check to see if we've already
2292 * unlocked this one.
2293 * Not the first one going back,
2294 * and the inode ptr is the same.
2296 if ((j != (i - 1)) && ips[j] ==
2300 xfs_iunlock(ips[j], lock_mode);
2303 if ((attempts % 5) == 0) {
2304 delay(1); /* Don't just spin the CPU */
2314 xfs_ilock(ips[i], lock_mode);
2320 if (attempts < 5) xfs_small_retries++;
2321 else if (attempts < 100) xfs_middle_retries++;
2322 else xfs_lots_retries++;
2330 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2331 int remove_which_error_return = 0;
2333 #define REMOVE_DEBUG_TRACE(x)
2334 #endif /* ! DEBUG */
2343 bhv_desc_t *dir_bdp,
2344 bhv_vname_t *dentry,
2347 bhv_vnode_t *dir_vp;
2348 char *name = VNAME(dentry);
2349 xfs_inode_t *dp, *ip;
2350 xfs_trans_t *tp = NULL;
2353 xfs_bmap_free_t free_list;
2354 xfs_fsblock_t first_block;
2362 dir_vp = BHV_TO_VNODE(dir_bdp);
2363 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2365 dp = XFS_BHVTOI(dir_bdp);
2368 if (XFS_FORCED_SHUTDOWN(mp))
2369 return XFS_ERROR(EIO);
2371 namelen = VNAMELEN(dentry);
2373 if (!xfs_get_dir_entry(dentry, &ip)) {
2374 dm_di_mode = ip->i_d.di_mode;
2378 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2379 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2380 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2381 name, NULL, dm_di_mode, 0, 0);
2386 /* From this point on, return through std_return */
2390 * We need to get a reference to ip before we get our log
2391 * reservation. The reason for this is that we cannot call
2392 * xfs_iget for an inode for which we do not have a reference
2393 * once we've acquired a log reservation. This is because the
2394 * inode we are trying to get might be in xfs_inactive going
2395 * for a log reservation. Since we'll have to wait for the
2396 * inactive code to complete before returning from xfs_iget,
2397 * we need to make sure that we don't have log space reserved
2398 * when we call xfs_iget. Instead we get an unlocked reference
2399 * to the inode before getting our log reservation.
2401 error = xfs_get_dir_entry(dentry, &ip);
2403 REMOVE_DEBUG_TRACE(__LINE__);
2407 dm_di_mode = ip->i_d.di_mode;
2409 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2413 error = XFS_QM_DQATTACH(mp, dp, 0);
2414 if (!error && dp != ip)
2415 error = XFS_QM_DQATTACH(mp, ip, 0);
2417 REMOVE_DEBUG_TRACE(__LINE__);
2422 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2423 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2425 * We try to get the real space reservation first,
2426 * allowing for directory btree deletion(s) implying
2427 * possible bmap insert(s). If we can't get the space
2428 * reservation then we use 0 instead, and avoid the bmap
2429 * btree insert(s) in the directory code by, if the bmap
2430 * insert tries to happen, instead trimming the LAST
2431 * block from the directory.
2433 resblks = XFS_REMOVE_SPACE_RES(mp);
2434 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2435 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2436 if (error == ENOSPC) {
2438 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2439 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2442 ASSERT(error != ENOSPC);
2443 REMOVE_DEBUG_TRACE(__LINE__);
2444 xfs_trans_cancel(tp, 0);
2449 error = xfs_lock_dir_and_entry(dp, ip);
2451 REMOVE_DEBUG_TRACE(__LINE__);
2452 xfs_trans_cancel(tp, cancel_flags);
2458 * At this point, we've gotten both the directory and the entry
2461 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2464 * Increment vnode ref count only in this case since
2465 * there's an extra vnode reference in the case where
2469 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2473 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2475 XFS_BMAP_INIT(&free_list, &first_block);
2476 error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino,
2477 &first_block, &free_list, 0);
2479 ASSERT(error != ENOENT);
2480 REMOVE_DEBUG_TRACE(__LINE__);
2483 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2486 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2488 error = xfs_droplink(tp, ip);
2490 REMOVE_DEBUG_TRACE(__LINE__);
2494 /* Determine if this is the last link while
2495 * we are in the transaction.
2497 link_zero = (ip)->i_d.di_nlink==0;
2500 * Take an extra ref on the inode so that it doesn't
2501 * go to xfs_inactive() from within the commit.
2506 * If this is a synchronous mount, make sure that the
2507 * remove transaction goes to disk before returning to
2510 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2511 xfs_trans_set_sync(tp);
2514 error = xfs_bmap_finish(&tp, &free_list, &committed);
2516 REMOVE_DEBUG_TRACE(__LINE__);
2520 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2527 * Before we drop our extra reference to the inode, purge it
2528 * from the refcache if it is there. By waiting until afterwards
2529 * to do the IRELE, we ensure that we won't go inactive in the
2530 * xfs_refcache_purge_ip routine (although that would be OK).
2532 xfs_refcache_purge_ip(ip);
2534 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2537 * Let interposed file systems know about removed links.
2539 bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero);
2543 /* Fall through to std_return with error = 0 */
2545 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2546 DM_EVENT_POSTREMOVE)) {
2547 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2548 dir_vp, DM_RIGHT_NULL,
2549 NULL, DM_RIGHT_NULL,
2550 name, NULL, dm_di_mode, error, 0);
2555 xfs_bmap_cancel(&free_list);
2556 cancel_flags |= XFS_TRANS_ABORT;
2557 xfs_trans_cancel(tp, cancel_flags);
2562 * In this case make sure to not release the inode until after
2563 * the current transaction is aborted. Releasing it beforehand
2564 * can cause us to go to xfs_inactive and start a recursive
2565 * transaction which can easily deadlock with the current one.
2567 xfs_bmap_cancel(&free_list);
2568 cancel_flags |= XFS_TRANS_ABORT;
2569 xfs_trans_cancel(tp, cancel_flags);
2572 * Before we drop our extra reference to the inode, purge it
2573 * from the refcache if it is there. By waiting until afterwards
2574 * to do the IRELE, we ensure that we won't go inactive in the
2575 * xfs_refcache_purge_ip routine (although that would be OK).
2577 xfs_refcache_purge_ip(ip);
2591 bhv_desc_t *target_dir_bdp,
2592 bhv_vnode_t *src_vp,
2593 bhv_vname_t *dentry,
2596 xfs_inode_t *tdp, *sip;
2599 xfs_inode_t *ips[2];
2601 xfs_bmap_free_t free_list;
2602 xfs_fsblock_t first_block;
2605 bhv_vnode_t *target_dir_vp;
2607 char *target_name = VNAME(dentry);
2610 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2611 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2612 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2614 target_namelen = VNAMELEN(dentry);
2615 ASSERT(!VN_ISDIR(src_vp));
2617 sip = xfs_vtoi(src_vp);
2618 tdp = XFS_BHVTOI(target_dir_bdp);
2620 if (XFS_FORCED_SHUTDOWN(mp))
2621 return XFS_ERROR(EIO);
2623 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2624 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2625 target_dir_vp, DM_RIGHT_NULL,
2626 src_vp, DM_RIGHT_NULL,
2627 target_name, NULL, 0, 0, 0);
2632 /* Return through std_return after this point. */
2634 error = XFS_QM_DQATTACH(mp, sip, 0);
2635 if (!error && sip != tdp)
2636 error = XFS_QM_DQATTACH(mp, tdp, 0);
2640 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2641 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2642 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2643 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2644 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2645 if (error == ENOSPC) {
2647 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2648 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2655 if (sip->i_ino < tdp->i_ino) {
2663 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2666 * Increment vnode ref counts since xfs_trans_commit &
2667 * xfs_trans_cancel will both unlock the inodes and
2668 * decrement the associated ref counts.
2671 VN_HOLD(target_dir_vp);
2672 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2673 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2676 * If the source has too many links, we can't make any more to it.
2678 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2679 error = XFS_ERROR(EMLINK);
2684 * If we are using project inheritance, we only allow hard link
2685 * creation in our tree when the project IDs are the same; else
2686 * the tree quota mechanism could be circumvented.
2688 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2689 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2690 error = XFS_ERROR(EXDEV);
2695 (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen)))
2698 XFS_BMAP_INIT(&free_list, &first_block);
2700 error = xfs_dir_createname(tp, tdp, target_name, target_namelen,
2701 sip->i_ino, &first_block, &free_list,
2705 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2707 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2709 error = xfs_bumplink(tp, sip);
2714 * If this is a synchronous mount, make sure that the
2715 * link transaction goes to disk before returning to
2718 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2719 xfs_trans_set_sync(tp);
2722 error = xfs_bmap_finish (&tp, &free_list, &committed);
2724 xfs_bmap_cancel(&free_list);
2728 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2732 /* Fall through to std_return with error = 0. */
2734 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2735 DM_EVENT_POSTLINK)) {
2736 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2737 target_dir_vp, DM_RIGHT_NULL,
2738 src_vp, DM_RIGHT_NULL,
2739 target_name, NULL, 0, error, 0);
2744 cancel_flags |= XFS_TRANS_ABORT;
2748 xfs_trans_cancel(tp, cancel_flags);
2759 bhv_desc_t *dir_bdp,
2760 bhv_vname_t *dentry,
2765 char *dir_name = VNAME(dentry);
2767 xfs_inode_t *cdp; /* inode of created dir */
2768 bhv_vnode_t *cvp; /* vnode of created dir */
2774 xfs_bmap_free_t free_list;
2775 xfs_fsblock_t first_block;
2776 bhv_vnode_t *dir_vp;
2777 boolean_t dp_joined_to_trans;
2778 boolean_t created = B_FALSE;
2779 int dm_event_sent = 0;
2781 struct xfs_dquot *udqp, *gdqp;
2786 dir_vp = BHV_TO_VNODE(dir_bdp);
2787 dp = XFS_BHVTOI(dir_bdp);
2790 if (XFS_FORCED_SHUTDOWN(mp))
2791 return XFS_ERROR(EIO);
2793 dir_namelen = VNAMELEN(dentry);
2796 dp_joined_to_trans = B_FALSE;
2797 dm_di_mode = vap->va_mode;
2799 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2800 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2801 dir_vp, DM_RIGHT_NULL, NULL,
2802 DM_RIGHT_NULL, dir_name, NULL,
2809 /* Return through std_return after this point. */
2811 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2815 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2816 prid = dp->i_d.di_projid;
2817 else if (vap->va_mask & XFS_AT_PROJID)
2818 prid = (xfs_prid_t)vap->va_projid;
2820 prid = (xfs_prid_t)dfltprid;
2823 * Make sure that we have allocated dquot(s) on disk.
2825 error = XFS_QM_DQVOPALLOC(mp, dp,
2826 current_fsuid(credp), current_fsgid(credp), prid,
2827 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2831 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2832 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2833 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2834 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2835 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2836 if (error == ENOSPC) {
2838 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2839 XFS_TRANS_PERM_LOG_RES,
2840 XFS_MKDIR_LOG_COUNT);
2848 xfs_ilock(dp, XFS_ILOCK_EXCL);
2851 * Check for directory link count overflow.
2853 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2854 error = XFS_ERROR(EMLINK);
2859 * Reserve disk quota and the inode.
2861 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2866 (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen)))
2869 * create the directory inode.
2871 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2872 0, credp, prid, resblks > 0,
2875 if (error == ENOSPC)
2882 * Now we add the directory inode to the transaction.
2883 * We waited until now since xfs_dir_ialloc might start
2884 * a new transaction. Had we joined the transaction
2885 * earlier, the locks might have gotten released.
2888 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2889 dp_joined_to_trans = B_TRUE;
2891 XFS_BMAP_INIT(&free_list, &first_block);
2893 error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino,
2894 &first_block, &free_list, resblks ?
2895 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2897 ASSERT(error != ENOSPC);
2900 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2903 * Bump the in memory version number of the parent directory
2904 * so that other processes accessing it will recognize that
2905 * the directory has changed.
2909 error = xfs_dir_init(tp, cdp, dp);
2914 error = xfs_bumplink(tp, dp);
2918 cvp = XFS_ITOV(cdp);
2926 * Attach the dquots to the new inode and modify the icount incore.
2928 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2931 * If this is a synchronous mount, make sure that the
2932 * mkdir transaction goes to disk before returning to
2935 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2936 xfs_trans_set_sync(tp);
2939 error = xfs_bmap_finish(&tp, &free_list, &committed);
2945 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2946 XFS_QM_DQRELE(mp, udqp);
2947 XFS_QM_DQRELE(mp, gdqp);
2952 /* Fall through to std_return with error = 0 or errno from
2953 * xfs_trans_commit. */
2956 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2957 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2958 DM_EVENT_POSTCREATE)) {
2959 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2960 dir_vp, DM_RIGHT_NULL,
2961 created ? XFS_ITOV(cdp):NULL,
2964 dm_di_mode, error, 0);
2970 xfs_bmap_cancel(&free_list);
2972 cancel_flags |= XFS_TRANS_ABORT;
2974 xfs_trans_cancel(tp, cancel_flags);
2975 XFS_QM_DQRELE(mp, udqp);
2976 XFS_QM_DQRELE(mp, gdqp);
2978 if (!dp_joined_to_trans && (dp != NULL)) {
2979 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2992 bhv_desc_t *dir_bdp,
2993 bhv_vname_t *dentry,
2996 char *name = VNAME(dentry);
2998 xfs_inode_t *cdp; /* child directory */
3002 xfs_bmap_free_t free_list;
3003 xfs_fsblock_t first_block;
3006 bhv_vnode_t *dir_vp;
3007 int dm_di_mode = S_IFDIR;
3012 dir_vp = BHV_TO_VNODE(dir_bdp);
3013 dp = XFS_BHVTOI(dir_bdp);
3016 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3018 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3019 return XFS_ERROR(EIO);
3020 namelen = VNAMELEN(dentry);
3022 if (!xfs_get_dir_entry(dentry, &cdp)) {
3023 dm_di_mode = cdp->i_d.di_mode;
3027 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3028 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3029 dir_vp, DM_RIGHT_NULL,
3030 NULL, DM_RIGHT_NULL,
3031 name, NULL, dm_di_mode, 0, 0);
3033 return XFS_ERROR(error);
3036 /* Return through std_return after this point. */
3041 * We need to get a reference to cdp before we get our log
3042 * reservation. The reason for this is that we cannot call
3043 * xfs_iget for an inode for which we do not have a reference
3044 * once we've acquired a log reservation. This is because the
3045 * inode we are trying to get might be in xfs_inactive going
3046 * for a log reservation. Since we'll have to wait for the
3047 * inactive code to complete before returning from xfs_iget,
3048 * we need to make sure that we don't have log space reserved
3049 * when we call xfs_iget. Instead we get an unlocked reference
3050 * to the inode before getting our log reservation.
3052 error = xfs_get_dir_entry(dentry, &cdp);
3054 REMOVE_DEBUG_TRACE(__LINE__);
3058 dm_di_mode = cdp->i_d.di_mode;
3061 * Get the dquots for the inodes.
3063 error = XFS_QM_DQATTACH(mp, dp, 0);
3064 if (!error && dp != cdp)
3065 error = XFS_QM_DQATTACH(mp, cdp, 0);
3068 REMOVE_DEBUG_TRACE(__LINE__);
3072 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3073 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3075 * We try to get the real space reservation first,
3076 * allowing for directory btree deletion(s) implying
3077 * possible bmap insert(s). If we can't get the space
3078 * reservation then we use 0 instead, and avoid the bmap
3079 * btree insert(s) in the directory code by, if the bmap
3080 * insert tries to happen, instead trimming the LAST
3081 * block from the directory.
3083 resblks = XFS_REMOVE_SPACE_RES(mp);
3084 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3085 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3086 if (error == ENOSPC) {
3088 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3089 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3092 ASSERT(error != ENOSPC);
3097 XFS_BMAP_INIT(&free_list, &first_block);
3100 * Now lock the child directory inode and the parent directory
3101 * inode in the proper order. This will take care of validating
3102 * that the directory entry for the child directory inode has
3103 * not changed while we were obtaining a log reservation.
3105 error = xfs_lock_dir_and_entry(dp, cdp);
3107 xfs_trans_cancel(tp, cancel_flags);
3112 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3115 * Only increment the parent directory vnode count if
3116 * we didn't bump it in looking up cdp. The only time
3117 * we don't bump it is when we're looking up ".".
3123 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3125 ASSERT(cdp->i_d.di_nlink >= 2);
3126 if (cdp->i_d.di_nlink != 2) {
3127 error = XFS_ERROR(ENOTEMPTY);
3130 if (!xfs_dir_isempty(cdp)) {
3131 error = XFS_ERROR(ENOTEMPTY);
3135 error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino,
3136 &first_block, &free_list, resblks);
3140 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3143 * Bump the in memory generation count on the parent
3144 * directory so that other can know that it has changed.
3149 * Drop the link from cdp's "..".
3151 error = xfs_droplink(tp, dp);
3157 * Drop the link from dp to cdp.
3159 error = xfs_droplink(tp, cdp);
3165 * Drop the "." link from cdp to self.
3167 error = xfs_droplink(tp, cdp);
3172 /* Determine these before committing transaction */
3173 last_cdp_link = (cdp)->i_d.di_nlink==0;
3176 * Take an extra ref on the child vnode so that it
3177 * does not go to xfs_inactive() from within the commit.
3182 * If this is a synchronous mount, make sure that the
3183 * rmdir transaction goes to disk before returning to
3186 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3187 xfs_trans_set_sync(tp);
3190 error = xfs_bmap_finish (&tp, &free_list, &committed);
3192 xfs_bmap_cancel(&free_list);
3193 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3199 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3207 * Let interposed file systems know about removed links.
3209 bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3213 /* Fall through to std_return with error = 0 or the errno
3214 * from xfs_trans_commit. */
3216 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3217 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3218 dir_vp, DM_RIGHT_NULL,
3219 NULL, DM_RIGHT_NULL,
3220 name, NULL, dm_di_mode,
3226 xfs_bmap_cancel(&free_list);
3227 cancel_flags |= XFS_TRANS_ABORT;
3231 xfs_trans_cancel(tp, cancel_flags);
3237 * Read dp's entries starting at uiop->uio_offset and translate them into
3238 * bufsize bytes worth of struct dirents starting at bufbase.
3242 bhv_desc_t *dir_bdp,
3248 xfs_trans_t *tp = NULL;
3252 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3253 (inst_t *)__return_address);
3254 dp = XFS_BHVTOI(dir_bdp);
3256 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
3257 return XFS_ERROR(EIO);
3259 lock_mode = xfs_ilock_map_shared(dp);
3260 error = xfs_dir_getdents(tp, dp, uiop, eofp);
3261 xfs_iunlock_map_shared(dp, lock_mode);
3268 bhv_desc_t *dir_bdp,
3269 bhv_vname_t *dentry,
3281 xfs_bmap_free_t free_list;
3282 xfs_fsblock_t first_block;
3283 boolean_t dp_joined_to_trans;
3284 bhv_vnode_t *dir_vp;
3287 xfs_fileoff_t first_fsb;
3288 xfs_filblks_t fs_blocks;
3290 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3297 struct xfs_dquot *udqp, *gdqp;
3299 char *link_name = VNAME(dentry);
3303 dir_vp = BHV_TO_VNODE(dir_bdp);
3304 dp = XFS_BHVTOI(dir_bdp);
3305 dp_joined_to_trans = B_FALSE;
3310 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3314 if (XFS_FORCED_SHUTDOWN(mp))
3315 return XFS_ERROR(EIO);
3317 link_namelen = VNAMELEN(dentry);
3320 * Check component lengths of the target path name.
3322 pathlen = strlen(target_path);
3323 if (pathlen >= MAXPATHLEN) /* total string too long */
3324 return XFS_ERROR(ENAMETOOLONG);
3325 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3329 for (total = 0, path = target_path; total < pathlen;) {
3333 while(*path == '/') {
3339 * Count up to the next slash or end of path.
3340 * Error out if the component is bigger than MAXNAMELEN.
3342 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3343 if (++len >= MAXNAMELEN) {
3344 error = ENAMETOOLONG;
3351 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3352 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3353 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3354 link_name, target_path, 0, 0, 0);
3359 /* Return through std_return after this point. */
3362 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3363 prid = dp->i_d.di_projid;
3364 else if (vap->va_mask & XFS_AT_PROJID)
3365 prid = (xfs_prid_t)vap->va_projid;
3367 prid = (xfs_prid_t)dfltprid;
3370 * Make sure that we have allocated dquot(s) on disk.
3372 error = XFS_QM_DQVOPALLOC(mp, dp,
3373 current_fsuid(credp), current_fsgid(credp), prid,
3374 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3378 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3379 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3381 * The symlink will fit into the inode data fork?
3382 * There can't be any attributes so we get the whole variable part.
3384 if (pathlen <= XFS_LITINO(mp))
3387 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3388 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3389 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3390 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3391 if (error == ENOSPC && fs_blocks == 0) {
3393 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3394 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3402 xfs_ilock(dp, XFS_ILOCK_EXCL);
3405 * Check whether the directory allows new symlinks or not.
3407 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3408 error = XFS_ERROR(EPERM);
3413 * Reserve disk quota : blocks and inode.
3415 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3420 * Check for ability to enter directory entry, if no space reserved.
3423 (error = xfs_dir_canenter(tp, dp, link_name, link_namelen)))
3426 * Initialize the bmap freelist prior to calling either
3427 * bmapi or the directory create code.
3429 XFS_BMAP_INIT(&free_list, &first_block);
3432 * Allocate an inode for the symlink.
3434 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3435 1, 0, credp, prid, resblks > 0, &ip, NULL);
3437 if (error == ENOSPC)
3444 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3445 dp_joined_to_trans = B_TRUE;
3448 * Also attach the dquot(s) to it, if applicable.
3450 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3453 resblks -= XFS_IALLOC_SPACE_RES(mp);
3455 * If the symlink will fit into the inode, write it inline.
3457 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3458 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3459 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3460 ip->i_d.di_size = pathlen;
3463 * The inode was initially created in extent format.
3465 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3466 ip->i_df.if_flags |= XFS_IFINLINE;
3468 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3469 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3473 nmaps = SYMLINK_MAPS;
3475 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3476 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3477 &first_block, resblks, mval, &nmaps,
3484 resblks -= fs_blocks;
3485 ip->i_d.di_size = pathlen;
3486 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3488 cur_chunk = target_path;
3489 for (n = 0; n < nmaps; n++) {
3490 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3491 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3492 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3493 BTOBB(byte_cnt), 0);
3494 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3495 if (pathlen < byte_cnt) {
3498 pathlen -= byte_cnt;
3500 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3501 cur_chunk += byte_cnt;
3503 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3508 * Create the directory entry for the symlink.
3510 error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino,
3511 &first_block, &free_list, resblks);
3514 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3515 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3518 * Bump the in memory version number of the parent directory
3519 * so that other processes accessing it will recognize that
3520 * the directory has changed.
3525 * If this is a synchronous mount, make sure that the
3526 * symlink transaction goes to disk before returning to
3529 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3530 xfs_trans_set_sync(tp);
3534 * xfs_trans_commit normally decrements the vnode ref count
3535 * when it unlocks the inode. Since we want to return the
3536 * vnode to the caller, we bump the vnode ref count now.
3540 error = xfs_bmap_finish(&tp, &free_list, &committed);
3544 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3545 XFS_QM_DQRELE(mp, udqp);
3546 XFS_QM_DQRELE(mp, gdqp);
3548 /* Fall through to std_return with error = 0 or errno from
3549 * xfs_trans_commit */
3551 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3552 DM_EVENT_POSTSYMLINK)) {
3553 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3554 dir_vp, DM_RIGHT_NULL,
3555 error ? NULL : XFS_ITOV(ip),
3556 DM_RIGHT_NULL, link_name, target_path,
3572 xfs_bmap_cancel(&free_list);
3573 cancel_flags |= XFS_TRANS_ABORT;
3575 xfs_trans_cancel(tp, cancel_flags);
3576 XFS_QM_DQRELE(mp, udqp);
3577 XFS_QM_DQRELE(mp, gdqp);
3579 if (!dp_joined_to_trans && (dp != NULL)) {
3580 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3590 * A fid routine that takes a pointer to a previously allocated
3591 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3601 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3602 (inst_t *)__return_address);
3603 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3605 xfid = (xfs_fid2_t *)fidp;
3606 ip = XFS_BHVTOI(bdp);
3607 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3610 * use memcpy because the inode is a long long and there's no
3611 * assurance that xfid->fid_ino is properly aligned.
3613 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3614 xfid->fid_gen = ip->i_d.di_gen;
3626 bhv_vrwlock_t locktype)
3631 vp = BHV_TO_VNODE(bdp);
3634 ip = XFS_BHVTOI(bdp);
3635 if (locktype == VRWLOCK_WRITE) {
3636 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3637 } else if (locktype == VRWLOCK_TRY_READ) {
3638 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3639 } else if (locktype == VRWLOCK_TRY_WRITE) {
3640 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3642 ASSERT((locktype == VRWLOCK_READ) ||
3643 (locktype == VRWLOCK_WRITE_DIRECT));
3644 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3657 bhv_vrwlock_t locktype)
3662 vp = BHV_TO_VNODE(bdp);
3665 ip = XFS_BHVTOI(bdp);
3666 if (locktype == VRWLOCK_WRITE) {
3668 * In the write case, we may have added a new entry to
3669 * the reference cache. This might store a pointer to
3670 * an inode to be released in this inode. If it is there,
3671 * clear the pointer and release the inode after unlocking
3674 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3676 ASSERT((locktype == VRWLOCK_READ) ||
3677 (locktype == VRWLOCK_WRITE_DIRECT));
3678 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3690 xfs_inode_log_item_t *iip;
3693 ip = XFS_BHVTOI(bdp);
3697 if (XFS_FORCED_SHUTDOWN(mp))
3698 return XFS_ERROR(EIO);
3701 * Bypass inodes which have already been cleaned by
3702 * the inode flush clustering code inside xfs_iflush
3704 if ((ip->i_update_core == 0) &&
3705 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3708 if (flags & FLUSH_LOG) {
3709 if (iip && iip->ili_last_lsn) {
3710 xlog_t *log = mp->m_log;
3712 int s, log_flags = XFS_LOG_FORCE;
3714 s = GRANT_LOCK(log);
3715 sync_lsn = log->l_last_sync_lsn;
3716 GRANT_UNLOCK(log, s);
3718 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3721 if (flags & FLUSH_SYNC)
3722 log_flags |= XFS_LOG_SYNC;
3723 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3728 * We make this non-blocking if the inode is contended,
3729 * return EAGAIN to indicate to the caller that they
3730 * did not succeed. This prevents the flush path from
3731 * blocking on inodes inside another operation right
3732 * now, they get caught later by xfs_sync.
3734 if (flags & FLUSH_INODE) {
3737 if (xfs_ipincount(ip))
3740 if (flags & FLUSH_SYNC) {
3741 xfs_ilock(ip, XFS_ILOCK_SHARED);
3743 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3744 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3745 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3752 if (flags & FLUSH_SYNC)
3753 flush_flags = XFS_IFLUSH_SYNC;
3755 flush_flags = XFS_IFLUSH_ASYNC;
3757 error = xfs_iflush(ip, flush_flags);
3758 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3776 if (!capable(CAP_SYS_ADMIN))
3777 return XFS_ERROR(EPERM);
3779 ip = XFS_BHVTOI(bdp);
3782 if (XFS_FORCED_SHUTDOWN(mp))
3783 return XFS_ERROR(EIO);
3785 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3786 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3788 xfs_trans_cancel(tp, 0);
3791 xfs_ilock(ip, XFS_ILOCK_EXCL);
3792 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3794 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3795 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3797 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3799 error = xfs_trans_commit(tp, 0);
3811 vp = BHV_TO_VNODE(bdp);
3812 ip = XFS_BHVTOI(bdp);
3814 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3816 ASSERT(!VN_MAPPED(vp));
3818 /* bad inode, get out here ASAP */
3826 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3829 * Make sure the atime in the XFS inode is correct before freeing the
3832 xfs_synchronize_atime(ip);
3835 * If we have nothing to flush with this inode then complete the
3836 * teardown now, otherwise break the link between the xfs inode and the
3837 * linux inode and clean up the xfs inode later. This avoids flushing
3838 * the inode to disk during the delete operation itself.
3840 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3841 * first to ensure that xfs_iunpin() will never see an xfs inode
3842 * that has a linux inode being reclaimed. Synchronisation is provided
3843 * by the i_flags_lock.
3845 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3846 xfs_ilock(ip, XFS_ILOCK_EXCL);
3848 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3850 xfs_mount_t *mp = ip->i_mount;
3852 /* Protect sync and unpin from us */
3853 XFS_MOUNT_ILOCK(mp);
3854 spin_lock(&ip->i_flags_lock);
3855 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
3856 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3857 spin_unlock(&ip->i_flags_lock);
3858 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3859 XFS_MOUNT_IUNLOCK(mp);
3870 xfs_ihash_t *ih = ip->i_hash;
3871 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
3874 if (vp && VN_BAD(vp))
3877 /* The hash lock here protects a thread in xfs_iget_core from
3878 * racing with us on linking the inode back with a vnode.
3879 * Once we have the XFS_IRECLAIM flag set it will not touch
3882 write_lock(&ih->ih_lock);
3883 spin_lock(&ip->i_flags_lock);
3884 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3885 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
3886 spin_unlock(&ip->i_flags_lock);
3887 write_unlock(&ih->ih_lock);
3890 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3894 __xfs_iflags_set(ip, XFS_IRECLAIM);
3895 spin_unlock(&ip->i_flags_lock);
3896 write_unlock(&ih->ih_lock);
3899 * If the inode is still dirty, then flush it out. If the inode
3900 * is not in the AIL, then it will be OK to flush it delwri as
3901 * long as xfs_iflush() does not keep any references to the inode.
3902 * We leave that decision up to xfs_iflush() since it has the
3903 * knowledge of whether it's OK to simply do a delwri flush of
3904 * the inode or whether we need to wait until the inode is
3905 * pulled from the AIL.
3906 * We get the flush lock regardless, though, just to make sure
3907 * we don't free it while it is being flushed.
3909 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3911 xfs_ilock(ip, XFS_ILOCK_EXCL);
3915 if (ip->i_update_core ||
3916 ((ip->i_itemp != NULL) &&
3917 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3918 error = xfs_iflush(ip, sync_mode);
3920 * If we hit an error, typically because of filesystem
3921 * shutdown, we don't need to let vn_reclaim to know
3922 * because we're gonna reclaim the inode anyway.
3925 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3928 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3931 ASSERT(ip->i_update_core == 0);
3932 ASSERT(ip->i_itemp == NULL ||
3933 ip->i_itemp->ili_format.ilf_fields == 0);
3934 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3935 } else if (locked) {
3937 * We are not interested in doing an iflush if we're
3938 * in the process of shutting down the filesystem forcibly.
3939 * So, just reclaim the inode.
3942 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3951 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3954 xfs_inode_t *ip, *n;
3959 XFS_MOUNT_ILOCK(mp);
3960 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3962 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3964 if (xfs_ipincount(ip) ||
3965 !xfs_iflock_nowait(ip)) {
3966 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3970 XFS_MOUNT_IUNLOCK(mp);
3971 if (xfs_finish_reclaim(ip, noblock,
3972 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3981 XFS_MOUNT_IUNLOCK(mp);
3986 * xfs_alloc_file_space()
3987 * This routine allocates disk space for the given file.
3989 * If alloc_type == 0, this request is for an ALLOCSP type
3990 * request which will change the file size. In this case, no
3991 * DMAPI event will be generated by the call. A TRUNCATE event
3992 * will be generated later by xfs_setattr.
3994 * If alloc_type != 0, this request is for a RESVSP type
3995 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3996 * lower block boundary byte address is less than the file's
4005 xfs_alloc_file_space(
4012 xfs_mount_t *mp = ip->i_mount;
4014 xfs_filblks_t allocated_fsb;
4015 xfs_filblks_t allocatesize_fsb;
4016 xfs_extlen_t extsz, temp;
4017 xfs_fileoff_t startoffset_fsb;
4018 xfs_fsblock_t firstfsb;
4024 xfs_bmbt_irec_t imaps[1], *imapp;
4025 xfs_bmap_free_t free_list;
4026 uint qblocks, resblks, resrtextents;
4030 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4032 if (XFS_FORCED_SHUTDOWN(mp))
4033 return XFS_ERROR(EIO);
4035 rt = XFS_IS_REALTIME_INODE(ip);
4037 if (!(extsz = ip->i_d.di_extsize))
4038 extsz = mp->m_sb.sb_rextsize;
4040 extsz = ip->i_d.di_extsize;
4043 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4047 return XFS_ERROR(EINVAL);
4053 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4054 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4055 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4057 /* Generate a DMAPI event if needed. */
4058 if (alloc_type != 0 && offset < ip->i_size &&
4059 (attr_flags&ATTR_DMI) == 0 &&
4060 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4061 xfs_off_t end_dmi_offset;
4063 end_dmi_offset = offset+len;
4064 if (end_dmi_offset > ip->i_size)
4065 end_dmi_offset = ip->i_size;
4066 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4067 offset, end_dmi_offset - offset,
4074 * Allocate file space until done or until there is an error
4077 while (allocatesize_fsb && !error) {
4081 * Determine space reservations for data/realtime.
4083 if (unlikely(extsz)) {
4084 s = startoffset_fsb;
4087 e = startoffset_fsb + allocatesize_fsb;
4088 if ((temp = do_mod(startoffset_fsb, extsz)))
4090 if ((temp = do_mod(e, extsz)))
4094 e = allocatesize_fsb;
4098 resrtextents = qblocks = (uint)(e - s);
4099 resrtextents /= mp->m_sb.sb_rextsize;
4100 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4101 quota_flag = XFS_QMOPT_RES_RTBLKS;
4104 resblks = qblocks = \
4105 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4106 quota_flag = XFS_QMOPT_RES_REGBLKS;
4110 * Allocate and setup the transaction.
4112 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4113 error = xfs_trans_reserve(tp, resblks,
4114 XFS_WRITE_LOG_RES(mp), resrtextents,
4115 XFS_TRANS_PERM_LOG_RES,
4116 XFS_WRITE_LOG_COUNT);
4118 * Check for running out of space
4122 * Free the transaction structure.
4124 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4125 xfs_trans_cancel(tp, 0);
4128 xfs_ilock(ip, XFS_ILOCK_EXCL);
4129 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4130 qblocks, 0, quota_flag);
4134 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4135 xfs_trans_ihold(tp, ip);
4138 * Issue the xfs_bmapi() call to allocate the blocks
4140 XFS_BMAP_INIT(&free_list, &firstfsb);
4141 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4142 allocatesize_fsb, bmapi_flag,
4143 &firstfsb, 0, imapp, &nimaps,
4150 * Complete the transaction
4152 error = xfs_bmap_finish(&tp, &free_list, &committed);
4157 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4158 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4163 allocated_fsb = imapp->br_blockcount;
4166 error = XFS_ERROR(ENOSPC);
4170 startoffset_fsb += allocated_fsb;
4171 allocatesize_fsb -= allocated_fsb;
4174 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4175 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4177 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4178 XFS_ITOV(ip), DM_RIGHT_NULL,
4179 XFS_ITOV(ip), DM_RIGHT_NULL,
4180 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4182 goto retry; /* Maybe DMAPI app. has made space */
4183 /* else fall through with error from XFS_SEND_DATA */
4188 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4189 xfs_bmap_cancel(&free_list);
4190 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4192 error1: /* Just cancel transaction */
4193 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4194 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4195 goto dmapi_enospc_check;
4199 * Zero file bytes between startoff and endoff inclusive.
4200 * The iolock is held exclusive and no blocks are buffered.
4203 xfs_zero_remaining_bytes(
4208 xfs_bmbt_irec_t imap;
4209 xfs_fileoff_t offset_fsb;
4210 xfs_off_t lastoffset;
4213 xfs_mount_t *mp = ip->i_mount;
4217 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4218 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4219 mp->m_rtdev_targp : mp->m_ddev_targp);
4221 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4222 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4224 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4225 NULL, 0, &imap, &nimap, NULL, NULL);
4226 if (error || nimap < 1)
4228 ASSERT(imap.br_blockcount >= 1);
4229 ASSERT(imap.br_startoff == offset_fsb);
4230 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4231 if (lastoffset > endoff)
4232 lastoffset = endoff;
4233 if (imap.br_startblock == HOLESTARTBLOCK)
4235 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4236 if (imap.br_state == XFS_EXT_UNWRITTEN)
4239 XFS_BUF_UNWRITE(bp);
4241 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4243 if ((error = xfs_iowait(bp))) {
4244 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4245 mp, bp, XFS_BUF_ADDR(bp));
4248 memset(XFS_BUF_PTR(bp) +
4249 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4250 0, lastoffset - offset + 1);
4255 if ((error = xfs_iowait(bp))) {
4256 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4257 mp, bp, XFS_BUF_ADDR(bp));
4266 * xfs_free_file_space()
4267 * This routine frees disk space for the given file.
4269 * This routine is only called by xfs_change_file_space
4270 * for an UNRESVSP type call.
4278 xfs_free_file_space(
4287 xfs_off_t end_dmi_offset;
4288 xfs_fileoff_t endoffset_fsb;
4290 xfs_fsblock_t firstfsb;
4291 xfs_bmap_free_t free_list;
4293 xfs_bmbt_irec_t imap;
4301 xfs_fileoff_t startoffset_fsb;
4303 int need_iolock = 1;
4308 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4310 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4314 if (len <= 0) /* if nothing being freed */
4316 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4317 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4318 end_dmi_offset = offset + len;
4319 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4321 if (offset < ip->i_size &&
4322 (attr_flags & ATTR_DMI) == 0 &&
4323 DM_EVENT_ENABLED(XFS_MTOVFS(mp), 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 ilen = len + (offset & (rounding - 1));
4342 ioffset = offset & ~(rounding - 1);
4343 if (ilen & (rounding - 1))
4344 ilen = (ilen + rounding) & ~(rounding - 1);
4346 if (VN_CACHED(vp) != 0) {
4347 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4348 ctooff(offtoct(ioffset)), -1);
4349 error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
4350 -1, FI_REMAPF_LOCKED);
4352 goto out_unlock_iolock;
4356 * Need to zero the stuff we're not freeing, on disk.
4357 * If its a realtime file & can't use unwritten extents then we
4358 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4359 * will take care of it for us.
4361 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4363 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4364 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4366 goto out_unlock_iolock;
4367 ASSERT(nimap == 0 || nimap == 1);
4368 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4371 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4372 block = imap.br_startblock;
4373 mod = do_div(block, mp->m_sb.sb_rextsize);
4375 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4378 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4379 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4381 goto out_unlock_iolock;
4382 ASSERT(nimap == 0 || nimap == 1);
4383 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4384 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4386 if (mod && (mod != mp->m_sb.sb_rextsize))
4387 endoffset_fsb -= mod;
4390 if ((done = (endoffset_fsb <= startoffset_fsb)))
4392 * One contiguous piece to clear
4394 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4397 * Some full blocks, possibly two pieces to clear
4399 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4400 error = xfs_zero_remaining_bytes(ip, offset,
4401 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4403 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4404 error = xfs_zero_remaining_bytes(ip,
4405 XFS_FSB_TO_B(mp, endoffset_fsb),
4410 * free file space until done or until there is an error
4412 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4413 while (!error && !done) {
4416 * allocate and setup the transaction
4418 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
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_close = xfs_close,
4671 .vop_read = xfs_read,
4672 #ifdef HAVE_SENDFILE
4673 .vop_sendfile = xfs_sendfile,
4676 .vop_splice_read = xfs_splice_read,
4677 .vop_splice_write = xfs_splice_write,
4679 .vop_write = xfs_write,
4680 .vop_ioctl = xfs_ioctl,
4681 .vop_getattr = xfs_getattr,
4682 .vop_setattr = xfs_setattr,
4683 .vop_access = xfs_access,
4684 .vop_lookup = xfs_lookup,
4685 .vop_create = xfs_create,
4686 .vop_remove = xfs_remove,
4687 .vop_link = xfs_link,
4688 .vop_rename = xfs_rename,
4689 .vop_mkdir = xfs_mkdir,
4690 .vop_rmdir = xfs_rmdir,
4691 .vop_readdir = xfs_readdir,
4692 .vop_symlink = xfs_symlink,
4693 .vop_readlink = xfs_readlink,
4694 .vop_fsync = xfs_fsync,
4695 .vop_inactive = xfs_inactive,
4696 .vop_fid2 = xfs_fid2,
4697 .vop_rwlock = xfs_rwlock,
4698 .vop_rwunlock = xfs_rwunlock,
4699 .vop_bmap = xfs_bmap,
4700 .vop_reclaim = xfs_reclaim,
4701 .vop_attr_get = xfs_attr_get,
4702 .vop_attr_set = xfs_attr_set,
4703 .vop_attr_remove = xfs_attr_remove,
4704 .vop_attr_list = xfs_attr_list,
4705 .vop_link_removed = (vop_link_removed_t)fs_noval,
4706 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4707 .vop_tosspages = fs_tosspages,
4708 .vop_flushinval_pages = fs_flushinval_pages,
4709 .vop_flush_pages = fs_flush_pages,
4710 .vop_release = xfs_release,
4711 .vop_iflush = xfs_inode_flush,