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 = ip->i_d.di_size;
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_d.di_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_d.di_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_d.di_size) {
658 xfs_igrow_finish(tp, ip, vap->va_size,
659 !(flags & ATTR_DMI));
660 } else if ((vap->va_size <= ip->i_d.di_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_d.di_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 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1263 error = xfs_trans_reserve(tp, 0,
1264 XFS_ITRUNCATE_LOG_RES(mp),
1265 0, XFS_TRANS_PERM_LOG_RES,
1266 XFS_ITRUNCATE_LOG_COUNT);
1268 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1269 xfs_trans_cancel(tp, 0);
1270 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1274 xfs_ilock(ip, XFS_ILOCK_EXCL);
1275 xfs_trans_ijoin(tp, ip,
1278 xfs_trans_ihold(tp, ip);
1280 error = xfs_itruncate_finish(&tp, ip,
1285 * If we get an error at this point we
1286 * simply don't bother truncating the file.
1289 xfs_trans_cancel(tp,
1290 (XFS_TRANS_RELEASE_LOG_RES |
1293 error = xfs_trans_commit(tp,
1294 XFS_TRANS_RELEASE_LOG_RES);
1296 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1302 * Free a symlink that has blocks associated with it.
1305 xfs_inactive_symlink_rmt(
1313 xfs_fsblock_t first_block;
1314 xfs_bmap_free_t free_list;
1317 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1325 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1327 * We're freeing a symlink that has some
1328 * blocks allocated to it. Free the
1329 * blocks here. We know that we've got
1330 * either 1 or 2 extents and that we can
1331 * free them all in one bunmapi call.
1333 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1334 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1335 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1336 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1337 xfs_trans_cancel(tp, 0);
1342 * Lock the inode, fix the size, and join it to the transaction.
1343 * Hold it so in the normal path, we still have it locked for
1344 * the second transaction. In the error paths we need it
1345 * held so the cancel won't rele it, see below.
1347 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1348 size = (int)ip->i_d.di_size;
1349 ip->i_d.di_size = 0;
1350 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1351 xfs_trans_ihold(tp, ip);
1352 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1354 * Find the block(s) so we can inval and unmap them.
1357 XFS_BMAP_INIT(&free_list, &first_block);
1358 nmaps = ARRAY_SIZE(mval);
1359 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1360 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1364 * Invalidate the block(s).
1366 for (i = 0; i < nmaps; i++) {
1367 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1368 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1369 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1370 xfs_trans_binval(tp, bp);
1373 * Unmap the dead block(s) to the free_list.
1375 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1376 &first_block, &free_list, NULL, &done)))
1380 * Commit the first transaction. This logs the EFI and the inode.
1382 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
1385 * The transaction must have been committed, since there were
1386 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1387 * The new tp has the extent freeing and EFDs.
1391 * The first xact was committed, so add the inode to the new one.
1392 * Mark it dirty so it will be logged and moved forward in the log as
1393 * part of every commit.
1395 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1396 xfs_trans_ihold(tp, ip);
1397 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1399 * Get a new, empty transaction to return to our caller.
1401 ntp = xfs_trans_dup(tp);
1403 * Commit the transaction containing extent freeing and EFDs.
1404 * If we get an error on the commit here or on the reserve below,
1405 * we need to unlock the inode since the new transaction doesn't
1406 * have the inode attached.
1408 error = xfs_trans_commit(tp, 0);
1411 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1415 * Remove the memory for extent descriptions (just bookkeeping).
1417 if (ip->i_df.if_bytes)
1418 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1419 ASSERT(ip->i_df.if_bytes == 0);
1421 * Put an itruncate log reservation in the new transaction
1424 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1425 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1426 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1430 * Return with the inode locked but not joined to the transaction.
1436 xfs_bmap_cancel(&free_list);
1439 * Have to come here with the inode locked and either
1440 * (held and in the transaction) or (not in the transaction).
1441 * If the inode isn't held then cancel would iput it, but
1442 * that's wrong since this is inactive and the vnode ref
1443 * count is 0 already.
1444 * Cancel won't do anything to the inode if held, but it still
1445 * needs to be locked until the cancel is done, if it was
1446 * joined to the transaction.
1448 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1449 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1456 xfs_inactive_symlink_local(
1462 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1464 * We're freeing a symlink which fit into
1465 * the inode. Just free the memory used
1466 * to hold the old symlink.
1468 error = xfs_trans_reserve(*tpp, 0,
1469 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1470 0, XFS_TRANS_PERM_LOG_RES,
1471 XFS_ITRUNCATE_LOG_COUNT);
1474 xfs_trans_cancel(*tpp, 0);
1478 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1481 * Zero length symlinks _can_ exist.
1483 if (ip->i_df.if_bytes > 0) {
1484 xfs_idata_realloc(ip,
1485 -(ip->i_df.if_bytes),
1487 ASSERT(ip->i_df.if_bytes == 0);
1501 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1504 ASSERT(ip->i_d.di_forkoff != 0);
1505 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1506 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1508 error = xfs_attr_inactive(ip);
1511 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1512 return error; /* goto out */
1515 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1516 error = xfs_trans_reserve(tp, 0,
1517 XFS_IFREE_LOG_RES(mp),
1518 0, XFS_TRANS_PERM_LOG_RES,
1519 XFS_INACTIVE_LOG_COUNT);
1521 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1522 xfs_trans_cancel(tp, 0);
1524 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1528 xfs_ilock(ip, XFS_ILOCK_EXCL);
1529 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1530 xfs_trans_ihold(tp, ip);
1531 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1533 ASSERT(ip->i_d.di_anextents == 0);
1548 vp = BHV_TO_VNODE(bdp);
1549 ip = XFS_BHVTOI(bdp);
1552 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
1555 /* If this is a read-only mount, don't do this (would generate I/O) */
1556 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1559 #ifdef HAVE_REFCACHE
1560 /* If we are in the NFS reference cache then don't do this now */
1565 if (ip->i_d.di_nlink != 0) {
1566 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1567 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1568 ip->i_delayed_blks > 0)) &&
1569 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1570 (!(ip->i_d.di_flags &
1571 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1572 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1574 /* Update linux inode block count after free above */
1575 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1576 ip->i_d.di_nblocks + ip->i_delayed_blks);
1586 * This is called when the vnode reference count for the vnode
1587 * goes to zero. If the file has been unlinked, then it must
1588 * now be truncated. Also, we clear all of the read-ahead state
1589 * kept for the inode here since the file is now closed.
1598 xfs_bmap_free_t free_list;
1599 xfs_fsblock_t first_block;
1606 vp = BHV_TO_VNODE(bdp);
1607 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1609 ip = XFS_BHVTOI(bdp);
1612 * If the inode is already free, then there can be nothing
1615 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1616 ASSERT(ip->i_df.if_real_bytes == 0);
1617 ASSERT(ip->i_df.if_broot_bytes == 0);
1618 return VN_INACTIVE_CACHE;
1622 * Only do a truncate if it's a regular file with
1623 * some actual space in it. It's OK to look at the
1624 * inode's fields without the lock because we're the
1625 * only one with a reference to the inode.
1627 truncate = ((ip->i_d.di_nlink == 0) &&
1628 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1629 (ip->i_delayed_blks > 0)) &&
1630 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1634 if (ip->i_d.di_nlink == 0 &&
1635 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1636 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1641 /* If this is a read-only mount, don't do this (would generate I/O) */
1642 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1645 if (ip->i_d.di_nlink != 0) {
1646 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1647 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1648 ip->i_delayed_blks > 0)) &&
1649 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1650 (!(ip->i_d.di_flags &
1651 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1652 (ip->i_delayed_blks != 0)))) {
1653 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1654 return VN_INACTIVE_CACHE;
1655 /* Update linux inode block count after free above */
1656 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1657 ip->i_d.di_nblocks + ip->i_delayed_blks);
1662 ASSERT(ip->i_d.di_nlink == 0);
1664 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1665 return VN_INACTIVE_CACHE;
1667 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1670 * Do the xfs_itruncate_start() call before
1671 * reserving any log space because itruncate_start
1672 * will call into the buffer cache and we can't
1673 * do that within a transaction.
1675 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1677 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1679 error = xfs_trans_reserve(tp, 0,
1680 XFS_ITRUNCATE_LOG_RES(mp),
1681 0, XFS_TRANS_PERM_LOG_RES,
1682 XFS_ITRUNCATE_LOG_COUNT);
1684 /* Don't call itruncate_cleanup */
1685 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1686 xfs_trans_cancel(tp, 0);
1687 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1688 return VN_INACTIVE_CACHE;
1691 xfs_ilock(ip, XFS_ILOCK_EXCL);
1692 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1693 xfs_trans_ihold(tp, ip);
1696 * normally, we have to run xfs_itruncate_finish sync.
1697 * But if filesystem is wsync and we're in the inactive
1698 * path, then we know that nlink == 0, and that the
1699 * xaction that made nlink == 0 is permanently committed
1700 * since xfs_remove runs as a synchronous transaction.
1702 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1703 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1706 xfs_trans_cancel(tp,
1707 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1708 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1709 return VN_INACTIVE_CACHE;
1711 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1714 * If we get an error while cleaning up a
1715 * symlink we bail out.
1717 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1718 xfs_inactive_symlink_rmt(ip, &tp) :
1719 xfs_inactive_symlink_local(ip, &tp);
1723 return VN_INACTIVE_CACHE;
1726 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1727 xfs_trans_ihold(tp, ip);
1729 error = xfs_trans_reserve(tp, 0,
1730 XFS_IFREE_LOG_RES(mp),
1731 0, XFS_TRANS_PERM_LOG_RES,
1732 XFS_INACTIVE_LOG_COUNT);
1734 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1735 xfs_trans_cancel(tp, 0);
1736 return VN_INACTIVE_CACHE;
1739 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1740 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1741 xfs_trans_ihold(tp, ip);
1745 * If there are attributes associated with the file
1746 * then blow them away now. The code calls a routine
1747 * that recursively deconstructs the attribute fork.
1748 * We need to just commit the current transaction
1749 * because we can't use it for xfs_attr_inactive().
1751 if (ip->i_d.di_anextents > 0) {
1752 error = xfs_inactive_attrs(ip, &tp);
1754 * If we got an error, the transaction is already
1755 * cancelled, and the inode is unlocked. Just get out.
1758 return VN_INACTIVE_CACHE;
1759 } else if (ip->i_afp) {
1760 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1766 XFS_BMAP_INIT(&free_list, &first_block);
1767 error = xfs_ifree(tp, ip, &free_list);
1770 * If we fail to free the inode, shut down. The cancel
1771 * might do that, we need to make sure. Otherwise the
1772 * inode might be lost for a long time or forever.
1774 if (!XFS_FORCED_SHUTDOWN(mp)) {
1776 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1777 error, mp->m_fsname);
1778 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1780 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1783 * Credit the quota account(s). The inode is gone.
1785 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1788 * Just ignore errors at this point. There is
1789 * nothing we can do except to try to keep going.
1791 (void) xfs_bmap_finish(&tp, &free_list, &committed);
1792 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1795 * Release the dquots held by inode, if any.
1797 XFS_QM_DQDETACH(mp, ip);
1799 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1802 return VN_INACTIVE_CACHE;
1811 bhv_desc_t *dir_bdp,
1812 bhv_vname_t *dentry,
1818 xfs_inode_t *dp, *ip;
1822 bhv_vnode_t *dir_vp;
1824 dir_vp = BHV_TO_VNODE(dir_bdp);
1825 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1827 dp = XFS_BHVTOI(dir_bdp);
1829 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1830 return XFS_ERROR(EIO);
1832 lock_mode = xfs_ilock_map_shared(dp);
1833 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1835 *vpp = XFS_ITOV(ip);
1838 xfs_iunlock_map_shared(dp, lock_mode);
1844 * xfs_create (create a new file).
1848 bhv_desc_t *dir_bdp,
1849 bhv_vname_t *dentry,
1854 char *name = VNAME(dentry);
1855 bhv_vnode_t *dir_vp;
1856 xfs_inode_t *dp, *ip;
1857 bhv_vnode_t *vp = NULL;
1862 xfs_bmap_free_t free_list;
1863 xfs_fsblock_t first_block;
1864 boolean_t dp_joined_to_trans;
1865 int dm_event_sent = 0;
1869 struct xfs_dquot *udqp, *gdqp;
1875 dir_vp = BHV_TO_VNODE(dir_bdp);
1876 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1878 dp = XFS_BHVTOI(dir_bdp);
1881 dm_di_mode = vap->va_mode;
1882 namelen = VNAMELEN(dentry);
1884 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1885 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1886 dir_vp, DM_RIGHT_NULL, NULL,
1887 DM_RIGHT_NULL, name, NULL,
1895 if (XFS_FORCED_SHUTDOWN(mp))
1896 return XFS_ERROR(EIO);
1898 /* Return through std_return after this point. */
1901 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1902 prid = dp->i_d.di_projid;
1903 else if (vap->va_mask & XFS_AT_PROJID)
1904 prid = (xfs_prid_t)vap->va_projid;
1906 prid = (xfs_prid_t)dfltprid;
1909 * Make sure that we have allocated dquot(s) on disk.
1911 error = XFS_QM_DQVOPALLOC(mp, dp,
1912 current_fsuid(credp), current_fsgid(credp), prid,
1913 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1918 dp_joined_to_trans = B_FALSE;
1920 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1921 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1922 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1924 * Initially assume that the file does not exist and
1925 * reserve the resources for that case. If that is not
1926 * the case we'll drop the one we have and get a more
1927 * appropriate transaction later.
1929 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1930 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1931 if (error == ENOSPC) {
1933 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1934 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1942 xfs_ilock(dp, XFS_ILOCK_EXCL);
1944 XFS_BMAP_INIT(&free_list, &first_block);
1949 * Reserve disk quota and the inode.
1951 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1955 if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen)))
1957 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1958 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1959 rdev, credp, prid, resblks > 0,
1962 if (error == ENOSPC)
1969 * At this point, we've gotten a newly allocated inode.
1970 * It is locked (and joined to the transaction).
1973 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1976 * Now we join the directory inode to the transaction.
1977 * We do not do it earlier because xfs_dir_ialloc
1978 * might commit the previous transaction (and release
1983 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1984 dp_joined_to_trans = B_TRUE;
1986 error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino,
1987 &first_block, &free_list, resblks ?
1988 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1990 ASSERT(error != ENOSPC);
1993 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1994 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1997 * If this is a synchronous mount, make sure that the
1998 * create transaction goes to disk before returning to
2001 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2002 xfs_trans_set_sync(tp);
2008 * Attach the dquot(s) to the inodes and modify them incore.
2009 * These ids of the inode couldn't have changed since the new
2010 * inode has been locked ever since it was created.
2012 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2015 * xfs_trans_commit normally decrements the vnode ref count
2016 * when it unlocks the inode. Since we want to return the
2017 * vnode to the caller, we bump the vnode ref count now.
2022 error = xfs_bmap_finish(&tp, &free_list, &committed);
2024 xfs_bmap_cancel(&free_list);
2028 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2035 XFS_QM_DQRELE(mp, udqp);
2036 XFS_QM_DQRELE(mp, gdqp);
2039 * Propagate the fact that the vnode changed after the
2040 * xfs_inode locks have been released.
2042 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2046 /* Fallthrough to std_return with error = 0 */
2049 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2050 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2051 DM_EVENT_POSTCREATE)) {
2052 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2053 dir_vp, DM_RIGHT_NULL,
2055 DM_RIGHT_NULL, name, NULL,
2056 dm_di_mode, error, 0);
2061 cancel_flags |= XFS_TRANS_ABORT;
2066 xfs_trans_cancel(tp, cancel_flags);
2068 if (!dp_joined_to_trans && (dp != NULL))
2069 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2070 XFS_QM_DQRELE(mp, udqp);
2071 XFS_QM_DQRELE(mp, gdqp);
2077 * Wait until after the current transaction is aborted to
2078 * release the inode. This prevents recursive transactions
2079 * and deadlocks from xfs_inactive.
2081 cancel_flags |= XFS_TRANS_ABORT;
2082 xfs_trans_cancel(tp, cancel_flags);
2085 XFS_QM_DQRELE(mp, udqp);
2086 XFS_QM_DQRELE(mp, gdqp);
2093 * Some counters to see if (and how often) we are hitting some deadlock
2094 * prevention code paths.
2098 int xfs_rm_lock_delays;
2099 int xfs_rm_attempts;
2103 * The following routine will lock the inodes associated with the
2104 * directory and the named entry in the directory. The locks are
2105 * acquired in increasing inode number.
2107 * If the entry is "..", then only the directory is locked. The
2108 * vnode ref count will still include that from the .. entry in
2111 * There is a deadlock we need to worry about. If the locked directory is
2112 * in the AIL, it might be blocking up the log. The next inode we lock
2113 * could be already locked by another thread waiting for log space (e.g
2114 * a permanent log reservation with a long running transaction (see
2115 * xfs_itruncate_finish)). To solve this, we must check if the directory
2116 * is in the ail and use lock_nowait. If we can't lock, we need to
2117 * drop the inode lock on the directory and try again. xfs_iunlock will
2118 * potentially push the tail if we were holding up the log.
2121 xfs_lock_dir_and_entry(
2123 xfs_inode_t *ip) /* inode of entry 'name' */
2127 xfs_inode_t *ips[2];
2136 xfs_ilock(dp, XFS_ILOCK_EXCL);
2143 * We want to lock in increasing inum. Since we've already
2144 * acquired the lock on the directory, we may need to release
2145 * if if the inum of the entry turns out to be less.
2147 if (e_inum > dp->i_ino) {
2149 * We are already in the right order, so just
2150 * lock on the inode of the entry.
2151 * We need to use nowait if dp is in the AIL.
2154 lp = (xfs_log_item_t *)dp->i_itemp;
2155 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2156 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2163 * Unlock dp and try again.
2164 * xfs_iunlock will try to push the tail
2165 * if the inode is in the AIL.
2168 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2170 if ((attempts % 5) == 0) {
2171 delay(1); /* Don't just spin the CPU */
2173 xfs_rm_lock_delays++;
2179 xfs_ilock(ip, XFS_ILOCK_EXCL);
2181 } else if (e_inum < dp->i_ino) {
2182 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2186 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2188 /* else e_inum == dp->i_ino */
2189 /* This can happen if we're asked to lock /x/..
2190 * the entry is "..", which is also the parent directory.
2198 int xfs_small_retries;
2199 int xfs_middle_retries;
2200 int xfs_lots_retries;
2201 int xfs_lock_delays;
2205 * The following routine will lock n inodes in exclusive mode.
2206 * We assume the caller calls us with the inodes in i_ino order.
2208 * We need to detect deadlock where an inode that we lock
2209 * is in the AIL and we start waiting for another inode that is locked
2210 * by a thread in a long running transaction (such as truncate). This can
2211 * result in deadlock since the long running trans might need to wait
2212 * for the inode we just locked in order to push the tail and free space
2222 int attempts = 0, i, j, try_lock;
2225 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2236 for (; i < inodes; i++) {
2239 if (i && (ips[i] == ips[i-1])) /* Already locked */
2243 * If try_lock is not set yet, make sure all locked inodes
2244 * are not in the AIL.
2245 * If any are, set try_lock to be used later.
2249 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2250 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2251 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2258 * If any of the previous locks we have locked is in the AIL,
2259 * we must TRY to get the second and subsequent locks. If
2260 * we can't get any, we must release all we have
2265 /* try_lock must be 0 if i is 0. */
2267 * try_lock means we have an inode locked
2268 * that is in the AIL.
2271 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2275 * Unlock all previous guys and try again.
2276 * xfs_iunlock will try to push the tail
2277 * if the inode is in the AIL.
2280 for(j = i - 1; j >= 0; j--) {
2283 * Check to see if we've already
2284 * unlocked this one.
2285 * Not the first one going back,
2286 * and the inode ptr is the same.
2288 if ((j != (i - 1)) && ips[j] ==
2292 xfs_iunlock(ips[j], lock_mode);
2295 if ((attempts % 5) == 0) {
2296 delay(1); /* Don't just spin the CPU */
2306 xfs_ilock(ips[i], lock_mode);
2312 if (attempts < 5) xfs_small_retries++;
2313 else if (attempts < 100) xfs_middle_retries++;
2314 else xfs_lots_retries++;
2322 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2323 int remove_which_error_return = 0;
2325 #define REMOVE_DEBUG_TRACE(x)
2326 #endif /* ! DEBUG */
2335 bhv_desc_t *dir_bdp,
2336 bhv_vname_t *dentry,
2339 bhv_vnode_t *dir_vp;
2340 char *name = VNAME(dentry);
2341 xfs_inode_t *dp, *ip;
2342 xfs_trans_t *tp = NULL;
2345 xfs_bmap_free_t free_list;
2346 xfs_fsblock_t first_block;
2354 dir_vp = BHV_TO_VNODE(dir_bdp);
2355 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2357 dp = XFS_BHVTOI(dir_bdp);
2360 if (XFS_FORCED_SHUTDOWN(mp))
2361 return XFS_ERROR(EIO);
2363 namelen = VNAMELEN(dentry);
2365 if (!xfs_get_dir_entry(dentry, &ip)) {
2366 dm_di_mode = ip->i_d.di_mode;
2370 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2371 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2372 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2373 name, NULL, dm_di_mode, 0, 0);
2378 /* From this point on, return through std_return */
2382 * We need to get a reference to ip before we get our log
2383 * reservation. The reason for this is that we cannot call
2384 * xfs_iget for an inode for which we do not have a reference
2385 * once we've acquired a log reservation. This is because the
2386 * inode we are trying to get might be in xfs_inactive going
2387 * for a log reservation. Since we'll have to wait for the
2388 * inactive code to complete before returning from xfs_iget,
2389 * we need to make sure that we don't have log space reserved
2390 * when we call xfs_iget. Instead we get an unlocked reference
2391 * to the inode before getting our log reservation.
2393 error = xfs_get_dir_entry(dentry, &ip);
2395 REMOVE_DEBUG_TRACE(__LINE__);
2399 dm_di_mode = ip->i_d.di_mode;
2401 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2405 error = XFS_QM_DQATTACH(mp, dp, 0);
2406 if (!error && dp != ip)
2407 error = XFS_QM_DQATTACH(mp, ip, 0);
2409 REMOVE_DEBUG_TRACE(__LINE__);
2414 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2415 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2417 * We try to get the real space reservation first,
2418 * allowing for directory btree deletion(s) implying
2419 * possible bmap insert(s). If we can't get the space
2420 * reservation then we use 0 instead, and avoid the bmap
2421 * btree insert(s) in the directory code by, if the bmap
2422 * insert tries to happen, instead trimming the LAST
2423 * block from the directory.
2425 resblks = XFS_REMOVE_SPACE_RES(mp);
2426 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2427 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2428 if (error == ENOSPC) {
2430 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2431 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2434 ASSERT(error != ENOSPC);
2435 REMOVE_DEBUG_TRACE(__LINE__);
2436 xfs_trans_cancel(tp, 0);
2441 error = xfs_lock_dir_and_entry(dp, ip);
2443 REMOVE_DEBUG_TRACE(__LINE__);
2444 xfs_trans_cancel(tp, cancel_flags);
2450 * At this point, we've gotten both the directory and the entry
2453 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2456 * Increment vnode ref count only in this case since
2457 * there's an extra vnode reference in the case where
2461 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2465 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2467 XFS_BMAP_INIT(&free_list, &first_block);
2468 error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino,
2469 &first_block, &free_list, 0);
2471 ASSERT(error != ENOENT);
2472 REMOVE_DEBUG_TRACE(__LINE__);
2475 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2478 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2480 error = xfs_droplink(tp, ip);
2482 REMOVE_DEBUG_TRACE(__LINE__);
2486 /* Determine if this is the last link while
2487 * we are in the transaction.
2489 link_zero = (ip)->i_d.di_nlink==0;
2492 * Take an extra ref on the inode so that it doesn't
2493 * go to xfs_inactive() from within the commit.
2498 * If this is a synchronous mount, make sure that the
2499 * remove transaction goes to disk before returning to
2502 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2503 xfs_trans_set_sync(tp);
2506 error = xfs_bmap_finish(&tp, &free_list, &committed);
2508 REMOVE_DEBUG_TRACE(__LINE__);
2512 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2519 * Before we drop our extra reference to the inode, purge it
2520 * from the refcache if it is there. By waiting until afterwards
2521 * to do the IRELE, we ensure that we won't go inactive in the
2522 * xfs_refcache_purge_ip routine (although that would be OK).
2524 xfs_refcache_purge_ip(ip);
2526 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2529 * Let interposed file systems know about removed links.
2531 bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero);
2535 /* Fall through to std_return with error = 0 */
2537 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2538 DM_EVENT_POSTREMOVE)) {
2539 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2540 dir_vp, DM_RIGHT_NULL,
2541 NULL, DM_RIGHT_NULL,
2542 name, NULL, dm_di_mode, error, 0);
2547 xfs_bmap_cancel(&free_list);
2548 cancel_flags |= XFS_TRANS_ABORT;
2549 xfs_trans_cancel(tp, cancel_flags);
2554 * In this case make sure to not release the inode until after
2555 * the current transaction is aborted. Releasing it beforehand
2556 * can cause us to go to xfs_inactive and start a recursive
2557 * transaction which can easily deadlock with the current one.
2559 xfs_bmap_cancel(&free_list);
2560 cancel_flags |= XFS_TRANS_ABORT;
2561 xfs_trans_cancel(tp, cancel_flags);
2564 * Before we drop our extra reference to the inode, purge it
2565 * from the refcache if it is there. By waiting until afterwards
2566 * to do the IRELE, we ensure that we won't go inactive in the
2567 * xfs_refcache_purge_ip routine (although that would be OK).
2569 xfs_refcache_purge_ip(ip);
2583 bhv_desc_t *target_dir_bdp,
2584 bhv_vnode_t *src_vp,
2585 bhv_vname_t *dentry,
2588 xfs_inode_t *tdp, *sip;
2591 xfs_inode_t *ips[2];
2593 xfs_bmap_free_t free_list;
2594 xfs_fsblock_t first_block;
2597 bhv_vnode_t *target_dir_vp;
2599 char *target_name = VNAME(dentry);
2602 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2603 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2604 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2606 target_namelen = VNAMELEN(dentry);
2607 ASSERT(!VN_ISDIR(src_vp));
2609 sip = xfs_vtoi(src_vp);
2610 tdp = XFS_BHVTOI(target_dir_bdp);
2612 if (XFS_FORCED_SHUTDOWN(mp))
2613 return XFS_ERROR(EIO);
2615 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2616 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2617 target_dir_vp, DM_RIGHT_NULL,
2618 src_vp, DM_RIGHT_NULL,
2619 target_name, NULL, 0, 0, 0);
2624 /* Return through std_return after this point. */
2626 error = XFS_QM_DQATTACH(mp, sip, 0);
2627 if (!error && sip != tdp)
2628 error = XFS_QM_DQATTACH(mp, tdp, 0);
2632 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2633 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2634 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2635 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2636 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2637 if (error == ENOSPC) {
2639 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2640 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2647 if (sip->i_ino < tdp->i_ino) {
2655 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2658 * Increment vnode ref counts since xfs_trans_commit &
2659 * xfs_trans_cancel will both unlock the inodes and
2660 * decrement the associated ref counts.
2663 VN_HOLD(target_dir_vp);
2664 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2665 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2668 * If the source has too many links, we can't make any more to it.
2670 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2671 error = XFS_ERROR(EMLINK);
2676 * If we are using project inheritance, we only allow hard link
2677 * creation in our tree when the project IDs are the same; else
2678 * the tree quota mechanism could be circumvented.
2680 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2681 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2682 error = XFS_ERROR(EXDEV);
2687 (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen)))
2690 XFS_BMAP_INIT(&free_list, &first_block);
2692 error = xfs_dir_createname(tp, tdp, target_name, target_namelen,
2693 sip->i_ino, &first_block, &free_list,
2697 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2699 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2701 error = xfs_bumplink(tp, sip);
2706 * If this is a synchronous mount, make sure that the
2707 * link transaction goes to disk before returning to
2710 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2711 xfs_trans_set_sync(tp);
2714 error = xfs_bmap_finish (&tp, &free_list, &committed);
2716 xfs_bmap_cancel(&free_list);
2720 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2724 /* Fall through to std_return with error = 0. */
2726 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2727 DM_EVENT_POSTLINK)) {
2728 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2729 target_dir_vp, DM_RIGHT_NULL,
2730 src_vp, DM_RIGHT_NULL,
2731 target_name, NULL, 0, error, 0);
2736 cancel_flags |= XFS_TRANS_ABORT;
2740 xfs_trans_cancel(tp, cancel_flags);
2751 bhv_desc_t *dir_bdp,
2752 bhv_vname_t *dentry,
2757 char *dir_name = VNAME(dentry);
2759 xfs_inode_t *cdp; /* inode of created dir */
2760 bhv_vnode_t *cvp; /* vnode of created dir */
2766 xfs_bmap_free_t free_list;
2767 xfs_fsblock_t first_block;
2768 bhv_vnode_t *dir_vp;
2769 boolean_t dp_joined_to_trans;
2770 boolean_t created = B_FALSE;
2771 int dm_event_sent = 0;
2773 struct xfs_dquot *udqp, *gdqp;
2778 dir_vp = BHV_TO_VNODE(dir_bdp);
2779 dp = XFS_BHVTOI(dir_bdp);
2782 if (XFS_FORCED_SHUTDOWN(mp))
2783 return XFS_ERROR(EIO);
2785 dir_namelen = VNAMELEN(dentry);
2788 dp_joined_to_trans = B_FALSE;
2789 dm_di_mode = vap->va_mode;
2791 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2792 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2793 dir_vp, DM_RIGHT_NULL, NULL,
2794 DM_RIGHT_NULL, dir_name, NULL,
2801 /* Return through std_return after this point. */
2803 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2807 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2808 prid = dp->i_d.di_projid;
2809 else if (vap->va_mask & XFS_AT_PROJID)
2810 prid = (xfs_prid_t)vap->va_projid;
2812 prid = (xfs_prid_t)dfltprid;
2815 * Make sure that we have allocated dquot(s) on disk.
2817 error = XFS_QM_DQVOPALLOC(mp, dp,
2818 current_fsuid(credp), current_fsgid(credp), prid,
2819 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2823 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2824 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2825 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2826 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2827 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2828 if (error == ENOSPC) {
2830 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2831 XFS_TRANS_PERM_LOG_RES,
2832 XFS_MKDIR_LOG_COUNT);
2840 xfs_ilock(dp, XFS_ILOCK_EXCL);
2843 * Check for directory link count overflow.
2845 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2846 error = XFS_ERROR(EMLINK);
2851 * Reserve disk quota and the inode.
2853 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2858 (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen)))
2861 * create the directory inode.
2863 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2864 0, credp, prid, resblks > 0,
2867 if (error == ENOSPC)
2874 * Now we add the directory inode to the transaction.
2875 * We waited until now since xfs_dir_ialloc might start
2876 * a new transaction. Had we joined the transaction
2877 * earlier, the locks might have gotten released.
2880 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2881 dp_joined_to_trans = B_TRUE;
2883 XFS_BMAP_INIT(&free_list, &first_block);
2885 error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino,
2886 &first_block, &free_list, resblks ?
2887 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2889 ASSERT(error != ENOSPC);
2892 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2895 * Bump the in memory version number of the parent directory
2896 * so that other processes accessing it will recognize that
2897 * the directory has changed.
2901 error = xfs_dir_init(tp, cdp, dp);
2906 error = xfs_bumplink(tp, dp);
2910 cvp = XFS_ITOV(cdp);
2918 * Attach the dquots to the new inode and modify the icount incore.
2920 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2923 * If this is a synchronous mount, make sure that the
2924 * mkdir transaction goes to disk before returning to
2927 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2928 xfs_trans_set_sync(tp);
2931 error = xfs_bmap_finish(&tp, &free_list, &committed);
2937 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2938 XFS_QM_DQRELE(mp, udqp);
2939 XFS_QM_DQRELE(mp, gdqp);
2944 /* Fall through to std_return with error = 0 or errno from
2945 * xfs_trans_commit. */
2948 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2949 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2950 DM_EVENT_POSTCREATE)) {
2951 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2952 dir_vp, DM_RIGHT_NULL,
2953 created ? XFS_ITOV(cdp):NULL,
2956 dm_di_mode, error, 0);
2962 xfs_bmap_cancel(&free_list);
2964 cancel_flags |= XFS_TRANS_ABORT;
2966 xfs_trans_cancel(tp, cancel_flags);
2967 XFS_QM_DQRELE(mp, udqp);
2968 XFS_QM_DQRELE(mp, gdqp);
2970 if (!dp_joined_to_trans && (dp != NULL)) {
2971 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2984 bhv_desc_t *dir_bdp,
2985 bhv_vname_t *dentry,
2988 char *name = VNAME(dentry);
2990 xfs_inode_t *cdp; /* child directory */
2994 xfs_bmap_free_t free_list;
2995 xfs_fsblock_t first_block;
2998 bhv_vnode_t *dir_vp;
2999 int dm_di_mode = S_IFDIR;
3004 dir_vp = BHV_TO_VNODE(dir_bdp);
3005 dp = XFS_BHVTOI(dir_bdp);
3008 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3010 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3011 return XFS_ERROR(EIO);
3012 namelen = VNAMELEN(dentry);
3014 if (!xfs_get_dir_entry(dentry, &cdp)) {
3015 dm_di_mode = cdp->i_d.di_mode;
3019 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3020 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3021 dir_vp, DM_RIGHT_NULL,
3022 NULL, DM_RIGHT_NULL,
3023 name, NULL, dm_di_mode, 0, 0);
3025 return XFS_ERROR(error);
3028 /* Return through std_return after this point. */
3033 * We need to get a reference to cdp before we get our log
3034 * reservation. The reason for this is that we cannot call
3035 * xfs_iget for an inode for which we do not have a reference
3036 * once we've acquired a log reservation. This is because the
3037 * inode we are trying to get might be in xfs_inactive going
3038 * for a log reservation. Since we'll have to wait for the
3039 * inactive code to complete before returning from xfs_iget,
3040 * we need to make sure that we don't have log space reserved
3041 * when we call xfs_iget. Instead we get an unlocked reference
3042 * to the inode before getting our log reservation.
3044 error = xfs_get_dir_entry(dentry, &cdp);
3046 REMOVE_DEBUG_TRACE(__LINE__);
3050 dm_di_mode = cdp->i_d.di_mode;
3053 * Get the dquots for the inodes.
3055 error = XFS_QM_DQATTACH(mp, dp, 0);
3056 if (!error && dp != cdp)
3057 error = XFS_QM_DQATTACH(mp, cdp, 0);
3060 REMOVE_DEBUG_TRACE(__LINE__);
3064 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3065 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3067 * We try to get the real space reservation first,
3068 * allowing for directory btree deletion(s) implying
3069 * possible bmap insert(s). If we can't get the space
3070 * reservation then we use 0 instead, and avoid the bmap
3071 * btree insert(s) in the directory code by, if the bmap
3072 * insert tries to happen, instead trimming the LAST
3073 * block from the directory.
3075 resblks = XFS_REMOVE_SPACE_RES(mp);
3076 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3077 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3078 if (error == ENOSPC) {
3080 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3081 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3084 ASSERT(error != ENOSPC);
3089 XFS_BMAP_INIT(&free_list, &first_block);
3092 * Now lock the child directory inode and the parent directory
3093 * inode in the proper order. This will take care of validating
3094 * that the directory entry for the child directory inode has
3095 * not changed while we were obtaining a log reservation.
3097 error = xfs_lock_dir_and_entry(dp, cdp);
3099 xfs_trans_cancel(tp, cancel_flags);
3104 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3107 * Only increment the parent directory vnode count if
3108 * we didn't bump it in looking up cdp. The only time
3109 * we don't bump it is when we're looking up ".".
3115 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3117 ASSERT(cdp->i_d.di_nlink >= 2);
3118 if (cdp->i_d.di_nlink != 2) {
3119 error = XFS_ERROR(ENOTEMPTY);
3122 if (!xfs_dir_isempty(cdp)) {
3123 error = XFS_ERROR(ENOTEMPTY);
3127 error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino,
3128 &first_block, &free_list, resblks);
3132 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3135 * Bump the in memory generation count on the parent
3136 * directory so that other can know that it has changed.
3141 * Drop the link from cdp's "..".
3143 error = xfs_droplink(tp, dp);
3149 * Drop the link from dp to cdp.
3151 error = xfs_droplink(tp, cdp);
3157 * Drop the "." link from cdp to self.
3159 error = xfs_droplink(tp, cdp);
3164 /* Determine these before committing transaction */
3165 last_cdp_link = (cdp)->i_d.di_nlink==0;
3168 * Take an extra ref on the child vnode so that it
3169 * does not go to xfs_inactive() from within the commit.
3174 * If this is a synchronous mount, make sure that the
3175 * rmdir transaction goes to disk before returning to
3178 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3179 xfs_trans_set_sync(tp);
3182 error = xfs_bmap_finish (&tp, &free_list, &committed);
3184 xfs_bmap_cancel(&free_list);
3185 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3191 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3199 * Let interposed file systems know about removed links.
3201 bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3205 /* Fall through to std_return with error = 0 or the errno
3206 * from xfs_trans_commit. */
3208 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3209 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3210 dir_vp, DM_RIGHT_NULL,
3211 NULL, DM_RIGHT_NULL,
3212 name, NULL, dm_di_mode,
3218 xfs_bmap_cancel(&free_list);
3219 cancel_flags |= XFS_TRANS_ABORT;
3223 xfs_trans_cancel(tp, cancel_flags);
3229 * Read dp's entries starting at uiop->uio_offset and translate them into
3230 * bufsize bytes worth of struct dirents starting at bufbase.
3234 bhv_desc_t *dir_bdp,
3240 xfs_trans_t *tp = NULL;
3244 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3245 (inst_t *)__return_address);
3246 dp = XFS_BHVTOI(dir_bdp);
3248 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
3249 return XFS_ERROR(EIO);
3251 lock_mode = xfs_ilock_map_shared(dp);
3252 error = xfs_dir_getdents(tp, dp, uiop, eofp);
3253 xfs_iunlock_map_shared(dp, lock_mode);
3260 bhv_desc_t *dir_bdp,
3261 bhv_vname_t *dentry,
3273 xfs_bmap_free_t free_list;
3274 xfs_fsblock_t first_block;
3275 boolean_t dp_joined_to_trans;
3276 bhv_vnode_t *dir_vp;
3279 xfs_fileoff_t first_fsb;
3280 xfs_filblks_t fs_blocks;
3282 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3289 struct xfs_dquot *udqp, *gdqp;
3291 char *link_name = VNAME(dentry);
3295 dir_vp = BHV_TO_VNODE(dir_bdp);
3296 dp = XFS_BHVTOI(dir_bdp);
3297 dp_joined_to_trans = B_FALSE;
3302 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3306 if (XFS_FORCED_SHUTDOWN(mp))
3307 return XFS_ERROR(EIO);
3309 link_namelen = VNAMELEN(dentry);
3312 * Check component lengths of the target path name.
3314 pathlen = strlen(target_path);
3315 if (pathlen >= MAXPATHLEN) /* total string too long */
3316 return XFS_ERROR(ENAMETOOLONG);
3317 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3321 for (total = 0, path = target_path; total < pathlen;) {
3325 while(*path == '/') {
3331 * Count up to the next slash or end of path.
3332 * Error out if the component is bigger than MAXNAMELEN.
3334 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3335 if (++len >= MAXNAMELEN) {
3336 error = ENAMETOOLONG;
3343 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3344 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3345 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3346 link_name, target_path, 0, 0, 0);
3351 /* Return through std_return after this point. */
3354 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3355 prid = dp->i_d.di_projid;
3356 else if (vap->va_mask & XFS_AT_PROJID)
3357 prid = (xfs_prid_t)vap->va_projid;
3359 prid = (xfs_prid_t)dfltprid;
3362 * Make sure that we have allocated dquot(s) on disk.
3364 error = XFS_QM_DQVOPALLOC(mp, dp,
3365 current_fsuid(credp), current_fsgid(credp), prid,
3366 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3370 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3371 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3373 * The symlink will fit into the inode data fork?
3374 * There can't be any attributes so we get the whole variable part.
3376 if (pathlen <= XFS_LITINO(mp))
3379 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3380 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3381 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3382 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3383 if (error == ENOSPC && fs_blocks == 0) {
3385 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3386 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3394 xfs_ilock(dp, XFS_ILOCK_EXCL);
3397 * Check whether the directory allows new symlinks or not.
3399 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3400 error = XFS_ERROR(EPERM);
3405 * Reserve disk quota : blocks and inode.
3407 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3412 * Check for ability to enter directory entry, if no space reserved.
3415 (error = xfs_dir_canenter(tp, dp, link_name, link_namelen)))
3418 * Initialize the bmap freelist prior to calling either
3419 * bmapi or the directory create code.
3421 XFS_BMAP_INIT(&free_list, &first_block);
3424 * Allocate an inode for the symlink.
3426 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3427 1, 0, credp, prid, resblks > 0, &ip, NULL);
3429 if (error == ENOSPC)
3436 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3437 dp_joined_to_trans = B_TRUE;
3440 * Also attach the dquot(s) to it, if applicable.
3442 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3445 resblks -= XFS_IALLOC_SPACE_RES(mp);
3447 * If the symlink will fit into the inode, write it inline.
3449 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3450 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3451 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3452 ip->i_d.di_size = pathlen;
3455 * The inode was initially created in extent format.
3457 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3458 ip->i_df.if_flags |= XFS_IFINLINE;
3460 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3461 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3465 nmaps = SYMLINK_MAPS;
3467 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3468 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3469 &first_block, resblks, mval, &nmaps,
3476 resblks -= fs_blocks;
3477 ip->i_d.di_size = pathlen;
3478 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3480 cur_chunk = target_path;
3481 for (n = 0; n < nmaps; n++) {
3482 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3483 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3484 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3485 BTOBB(byte_cnt), 0);
3486 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3487 if (pathlen < byte_cnt) {
3490 pathlen -= byte_cnt;
3492 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3493 cur_chunk += byte_cnt;
3495 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3500 * Create the directory entry for the symlink.
3502 error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino,
3503 &first_block, &free_list, resblks);
3506 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3507 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3510 * Bump the in memory version number of the parent directory
3511 * so that other processes accessing it will recognize that
3512 * the directory has changed.
3517 * If this is a synchronous mount, make sure that the
3518 * symlink transaction goes to disk before returning to
3521 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3522 xfs_trans_set_sync(tp);
3526 * xfs_trans_commit normally decrements the vnode ref count
3527 * when it unlocks the inode. Since we want to return the
3528 * vnode to the caller, we bump the vnode ref count now.
3532 error = xfs_bmap_finish(&tp, &free_list, &committed);
3536 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3537 XFS_QM_DQRELE(mp, udqp);
3538 XFS_QM_DQRELE(mp, gdqp);
3540 /* Fall through to std_return with error = 0 or errno from
3541 * xfs_trans_commit */
3543 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3544 DM_EVENT_POSTSYMLINK)) {
3545 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3546 dir_vp, DM_RIGHT_NULL,
3547 error ? NULL : XFS_ITOV(ip),
3548 DM_RIGHT_NULL, link_name, target_path,
3564 xfs_bmap_cancel(&free_list);
3565 cancel_flags |= XFS_TRANS_ABORT;
3567 xfs_trans_cancel(tp, cancel_flags);
3568 XFS_QM_DQRELE(mp, udqp);
3569 XFS_QM_DQRELE(mp, gdqp);
3571 if (!dp_joined_to_trans && (dp != NULL)) {
3572 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3582 * A fid routine that takes a pointer to a previously allocated
3583 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3593 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3594 (inst_t *)__return_address);
3595 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3597 xfid = (xfs_fid2_t *)fidp;
3598 ip = XFS_BHVTOI(bdp);
3599 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3602 * use memcpy because the inode is a long long and there's no
3603 * assurance that xfid->fid_ino is properly aligned.
3605 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3606 xfid->fid_gen = ip->i_d.di_gen;
3618 bhv_vrwlock_t locktype)
3623 vp = BHV_TO_VNODE(bdp);
3626 ip = XFS_BHVTOI(bdp);
3627 if (locktype == VRWLOCK_WRITE) {
3628 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3629 } else if (locktype == VRWLOCK_TRY_READ) {
3630 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3631 } else if (locktype == VRWLOCK_TRY_WRITE) {
3632 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3634 ASSERT((locktype == VRWLOCK_READ) ||
3635 (locktype == VRWLOCK_WRITE_DIRECT));
3636 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3649 bhv_vrwlock_t locktype)
3654 vp = BHV_TO_VNODE(bdp);
3657 ip = XFS_BHVTOI(bdp);
3658 if (locktype == VRWLOCK_WRITE) {
3660 * In the write case, we may have added a new entry to
3661 * the reference cache. This might store a pointer to
3662 * an inode to be released in this inode. If it is there,
3663 * clear the pointer and release the inode after unlocking
3666 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3668 ASSERT((locktype == VRWLOCK_READ) ||
3669 (locktype == VRWLOCK_WRITE_DIRECT));
3670 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3682 xfs_inode_log_item_t *iip;
3685 ip = XFS_BHVTOI(bdp);
3689 if (XFS_FORCED_SHUTDOWN(mp))
3690 return XFS_ERROR(EIO);
3693 * Bypass inodes which have already been cleaned by
3694 * the inode flush clustering code inside xfs_iflush
3696 if ((ip->i_update_core == 0) &&
3697 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3700 if (flags & FLUSH_LOG) {
3701 if (iip && iip->ili_last_lsn) {
3702 xlog_t *log = mp->m_log;
3704 int s, log_flags = XFS_LOG_FORCE;
3706 s = GRANT_LOCK(log);
3707 sync_lsn = log->l_last_sync_lsn;
3708 GRANT_UNLOCK(log, s);
3710 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3713 if (flags & FLUSH_SYNC)
3714 log_flags |= XFS_LOG_SYNC;
3715 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3720 * We make this non-blocking if the inode is contended,
3721 * return EAGAIN to indicate to the caller that they
3722 * did not succeed. This prevents the flush path from
3723 * blocking on inodes inside another operation right
3724 * now, they get caught later by xfs_sync.
3726 if (flags & FLUSH_INODE) {
3729 if (xfs_ipincount(ip))
3732 if (flags & FLUSH_SYNC) {
3733 xfs_ilock(ip, XFS_ILOCK_SHARED);
3735 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3736 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3737 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3744 if (flags & FLUSH_SYNC)
3745 flush_flags = XFS_IFLUSH_SYNC;
3747 flush_flags = XFS_IFLUSH_ASYNC;
3749 error = xfs_iflush(ip, flush_flags);
3750 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3768 if (!capable(CAP_SYS_ADMIN))
3769 return XFS_ERROR(EPERM);
3771 ip = XFS_BHVTOI(bdp);
3774 if (XFS_FORCED_SHUTDOWN(mp))
3775 return XFS_ERROR(EIO);
3777 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3778 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3780 xfs_trans_cancel(tp, 0);
3783 xfs_ilock(ip, XFS_ILOCK_EXCL);
3784 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3786 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3787 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3789 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3791 error = xfs_trans_commit(tp, 0);
3803 vp = BHV_TO_VNODE(bdp);
3804 ip = XFS_BHVTOI(bdp);
3806 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3808 ASSERT(!VN_MAPPED(vp));
3810 /* bad inode, get out here ASAP */
3818 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3821 * Make sure the atime in the XFS inode is correct before freeing the
3824 xfs_synchronize_atime(ip);
3827 * If we have nothing to flush with this inode then complete the
3828 * teardown now, otherwise break the link between the xfs inode and the
3829 * linux inode and clean up the xfs inode later. This avoids flushing
3830 * the inode to disk during the delete operation itself.
3832 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3833 * first to ensure that xfs_iunpin() will never see an xfs inode
3834 * that has a linux inode being reclaimed. Synchronisation is provided
3835 * by the i_flags_lock.
3837 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3838 xfs_ilock(ip, XFS_ILOCK_EXCL);
3840 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3842 xfs_mount_t *mp = ip->i_mount;
3844 /* Protect sync and unpin from us */
3845 XFS_MOUNT_ILOCK(mp);
3846 spin_lock(&ip->i_flags_lock);
3847 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
3848 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3849 spin_unlock(&ip->i_flags_lock);
3850 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3851 XFS_MOUNT_IUNLOCK(mp);
3862 xfs_ihash_t *ih = ip->i_hash;
3863 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
3866 if (vp && VN_BAD(vp))
3869 /* The hash lock here protects a thread in xfs_iget_core from
3870 * racing with us on linking the inode back with a vnode.
3871 * Once we have the XFS_IRECLAIM flag set it will not touch
3874 write_lock(&ih->ih_lock);
3875 spin_lock(&ip->i_flags_lock);
3876 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3877 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
3878 spin_unlock(&ip->i_flags_lock);
3879 write_unlock(&ih->ih_lock);
3882 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3886 __xfs_iflags_set(ip, XFS_IRECLAIM);
3887 spin_unlock(&ip->i_flags_lock);
3888 write_unlock(&ih->ih_lock);
3891 * If the inode is still dirty, then flush it out. If the inode
3892 * is not in the AIL, then it will be OK to flush it delwri as
3893 * long as xfs_iflush() does not keep any references to the inode.
3894 * We leave that decision up to xfs_iflush() since it has the
3895 * knowledge of whether it's OK to simply do a delwri flush of
3896 * the inode or whether we need to wait until the inode is
3897 * pulled from the AIL.
3898 * We get the flush lock regardless, though, just to make sure
3899 * we don't free it while it is being flushed.
3901 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3903 xfs_ilock(ip, XFS_ILOCK_EXCL);
3907 if (ip->i_update_core ||
3908 ((ip->i_itemp != NULL) &&
3909 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3910 error = xfs_iflush(ip, sync_mode);
3912 * If we hit an error, typically because of filesystem
3913 * shutdown, we don't need to let vn_reclaim to know
3914 * because we're gonna reclaim the inode anyway.
3917 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3920 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3923 ASSERT(ip->i_update_core == 0);
3924 ASSERT(ip->i_itemp == NULL ||
3925 ip->i_itemp->ili_format.ilf_fields == 0);
3926 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3927 } else if (locked) {
3929 * We are not interested in doing an iflush if we're
3930 * in the process of shutting down the filesystem forcibly.
3931 * So, just reclaim the inode.
3934 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3943 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3946 xfs_inode_t *ip, *n;
3951 XFS_MOUNT_ILOCK(mp);
3952 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3954 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3956 if (xfs_ipincount(ip) ||
3957 !xfs_iflock_nowait(ip)) {
3958 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3962 XFS_MOUNT_IUNLOCK(mp);
3963 if (xfs_finish_reclaim(ip, noblock,
3964 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3973 XFS_MOUNT_IUNLOCK(mp);
3978 * xfs_alloc_file_space()
3979 * This routine allocates disk space for the given file.
3981 * If alloc_type == 0, this request is for an ALLOCSP type
3982 * request which will change the file size. In this case, no
3983 * DMAPI event will be generated by the call. A TRUNCATE event
3984 * will be generated later by xfs_setattr.
3986 * If alloc_type != 0, this request is for a RESVSP type
3987 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3988 * lower block boundary byte address is less than the file's
3997 xfs_alloc_file_space(
4004 xfs_mount_t *mp = ip->i_mount;
4006 xfs_filblks_t allocated_fsb;
4007 xfs_filblks_t allocatesize_fsb;
4008 xfs_extlen_t extsz, temp;
4009 xfs_fileoff_t startoffset_fsb;
4010 xfs_fsblock_t firstfsb;
4016 xfs_bmbt_irec_t imaps[1], *imapp;
4017 xfs_bmap_free_t free_list;
4018 uint qblocks, resblks, resrtextents;
4022 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4024 if (XFS_FORCED_SHUTDOWN(mp))
4025 return XFS_ERROR(EIO);
4027 rt = XFS_IS_REALTIME_INODE(ip);
4029 if (!(extsz = ip->i_d.di_extsize))
4030 extsz = mp->m_sb.sb_rextsize;
4032 extsz = ip->i_d.di_extsize;
4035 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4039 return XFS_ERROR(EINVAL);
4045 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4046 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4047 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4049 /* Generate a DMAPI event if needed. */
4050 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4051 (attr_flags&ATTR_DMI) == 0 &&
4052 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4053 xfs_off_t end_dmi_offset;
4055 end_dmi_offset = offset+len;
4056 if (end_dmi_offset > ip->i_d.di_size)
4057 end_dmi_offset = ip->i_d.di_size;
4058 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4059 offset, end_dmi_offset - offset,
4066 * Allocate file space until done or until there is an error
4069 while (allocatesize_fsb && !error) {
4073 * Determine space reservations for data/realtime.
4075 if (unlikely(extsz)) {
4076 s = startoffset_fsb;
4079 e = startoffset_fsb + allocatesize_fsb;
4080 if ((temp = do_mod(startoffset_fsb, extsz)))
4082 if ((temp = do_mod(e, extsz)))
4086 e = allocatesize_fsb;
4090 resrtextents = qblocks = (uint)(e - s);
4091 resrtextents /= mp->m_sb.sb_rextsize;
4092 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4093 quota_flag = XFS_QMOPT_RES_RTBLKS;
4096 resblks = qblocks = \
4097 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4098 quota_flag = XFS_QMOPT_RES_REGBLKS;
4102 * Allocate and setup the transaction.
4104 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4105 error = xfs_trans_reserve(tp, resblks,
4106 XFS_WRITE_LOG_RES(mp), resrtextents,
4107 XFS_TRANS_PERM_LOG_RES,
4108 XFS_WRITE_LOG_COUNT);
4110 * Check for running out of space
4114 * Free the transaction structure.
4116 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4117 xfs_trans_cancel(tp, 0);
4120 xfs_ilock(ip, XFS_ILOCK_EXCL);
4121 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4122 qblocks, 0, quota_flag);
4126 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4127 xfs_trans_ihold(tp, ip);
4130 * Issue the xfs_bmapi() call to allocate the blocks
4132 XFS_BMAP_INIT(&free_list, &firstfsb);
4133 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4134 allocatesize_fsb, bmapi_flag,
4135 &firstfsb, 0, imapp, &nimaps,
4142 * Complete the transaction
4144 error = xfs_bmap_finish(&tp, &free_list, &committed);
4149 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4150 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4155 allocated_fsb = imapp->br_blockcount;
4158 error = XFS_ERROR(ENOSPC);
4162 startoffset_fsb += allocated_fsb;
4163 allocatesize_fsb -= allocated_fsb;
4166 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4167 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4169 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4170 XFS_ITOV(ip), DM_RIGHT_NULL,
4171 XFS_ITOV(ip), DM_RIGHT_NULL,
4172 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4174 goto retry; /* Maybe DMAPI app. has made space */
4175 /* else fall through with error from XFS_SEND_DATA */
4180 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4181 xfs_bmap_cancel(&free_list);
4182 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4184 error1: /* Just cancel transaction */
4185 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4186 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4187 goto dmapi_enospc_check;
4191 * Zero file bytes between startoff and endoff inclusive.
4192 * The iolock is held exclusive and no blocks are buffered.
4195 xfs_zero_remaining_bytes(
4200 xfs_bmbt_irec_t imap;
4201 xfs_fileoff_t offset_fsb;
4202 xfs_off_t lastoffset;
4205 xfs_mount_t *mp = ip->i_mount;
4209 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4210 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4211 mp->m_rtdev_targp : mp->m_ddev_targp);
4213 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4214 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4216 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4217 NULL, 0, &imap, &nimap, NULL, NULL);
4218 if (error || nimap < 1)
4220 ASSERT(imap.br_blockcount >= 1);
4221 ASSERT(imap.br_startoff == offset_fsb);
4222 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4223 if (lastoffset > endoff)
4224 lastoffset = endoff;
4225 if (imap.br_startblock == HOLESTARTBLOCK)
4227 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4228 if (imap.br_state == XFS_EXT_UNWRITTEN)
4231 XFS_BUF_UNWRITE(bp);
4233 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4235 if ((error = xfs_iowait(bp))) {
4236 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4237 mp, bp, XFS_BUF_ADDR(bp));
4240 memset(XFS_BUF_PTR(bp) +
4241 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4242 0, lastoffset - offset + 1);
4247 if ((error = xfs_iowait(bp))) {
4248 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4249 mp, bp, XFS_BUF_ADDR(bp));
4258 * xfs_free_file_space()
4259 * This routine frees disk space for the given file.
4261 * This routine is only called by xfs_change_file_space
4262 * for an UNRESVSP type call.
4270 xfs_free_file_space(
4279 xfs_off_t end_dmi_offset;
4280 xfs_fileoff_t endoffset_fsb;
4282 xfs_fsblock_t firstfsb;
4283 xfs_bmap_free_t free_list;
4285 xfs_bmbt_irec_t imap;
4293 xfs_fileoff_t startoffset_fsb;
4295 int need_iolock = 1;
4300 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4302 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4306 if (len <= 0) /* if nothing being freed */
4308 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4309 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4310 end_dmi_offset = offset + len;
4311 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4313 if (offset < ip->i_d.di_size &&
4314 (attr_flags & ATTR_DMI) == 0 &&
4315 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4316 if (end_dmi_offset > ip->i_d.di_size)
4317 end_dmi_offset = ip->i_d.di_size;
4318 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4319 offset, end_dmi_offset - offset,
4320 AT_DELAY_FLAG(attr_flags), NULL);
4325 if (attr_flags & ATTR_NOLOCK)
4328 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4329 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4332 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP);
4333 ilen = len + (offset & (rounding - 1));
4334 ioffset = offset & ~(rounding - 1);
4335 if (ilen & (rounding - 1))
4336 ilen = (ilen + rounding) & ~(rounding - 1);
4338 if (VN_CACHED(vp) != 0) {
4339 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4340 ctooff(offtoct(ioffset)), -1);
4341 bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
4342 -1, FI_REMAPF_LOCKED);
4346 * Need to zero the stuff we're not freeing, on disk.
4347 * If its a realtime file & can't use unwritten extents then we
4348 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4349 * will take care of it for us.
4351 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4353 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4354 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4356 goto out_unlock_iolock;
4357 ASSERT(nimap == 0 || nimap == 1);
4358 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4361 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4362 block = imap.br_startblock;
4363 mod = do_div(block, mp->m_sb.sb_rextsize);
4365 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4368 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4369 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4371 goto out_unlock_iolock;
4372 ASSERT(nimap == 0 || nimap == 1);
4373 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4374 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4376 if (mod && (mod != mp->m_sb.sb_rextsize))
4377 endoffset_fsb -= mod;
4380 if ((done = (endoffset_fsb <= startoffset_fsb)))
4382 * One contiguous piece to clear
4384 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4387 * Some full blocks, possibly two pieces to clear
4389 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4390 error = xfs_zero_remaining_bytes(ip, offset,
4391 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4393 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4394 error = xfs_zero_remaining_bytes(ip,
4395 XFS_FSB_TO_B(mp, endoffset_fsb),
4400 * free file space until done or until there is an error
4402 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4403 while (!error && !done) {
4406 * allocate and setup the transaction
4408 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4409 error = xfs_trans_reserve(tp,
4411 XFS_WRITE_LOG_RES(mp),
4413 XFS_TRANS_PERM_LOG_RES,
4414 XFS_WRITE_LOG_COUNT);
4417 * check for running out of space
4421 * Free the transaction structure.
4423 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4424 xfs_trans_cancel(tp, 0);
4427 xfs_ilock(ip, XFS_ILOCK_EXCL);
4428 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4429 ip->i_udquot, ip->i_gdquot, resblks, 0,
4430 XFS_QMOPT_RES_REGBLKS);
4434 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4435 xfs_trans_ihold(tp, ip);
4438 * issue the bunmapi() call to free the blocks
4440 XFS_BMAP_INIT(&free_list, &firstfsb);
4441 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4442 endoffset_fsb - startoffset_fsb,
4443 0, 2, &firstfsb, &free_list, NULL, &done);
4449 * complete the transaction
4451 error = xfs_bmap_finish(&tp, &free_list, &committed);
4456 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4457 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4462 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4466 xfs_bmap_cancel(&free_list);
4468 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4469 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4475 * xfs_change_file_space()
4476 * This routine allocates or frees disk space for the given file.
4477 * The user specified parameters are checked for alignment and size
4486 xfs_change_file_space(
4500 xfs_off_t startoffset;
4506 vp = BHV_TO_VNODE(bdp);
4507 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4509 ip = XFS_BHVTOI(bdp);
4513 * must be a regular file and have write permission
4516 return XFS_ERROR(EINVAL);
4518 xfs_ilock(ip, XFS_ILOCK_SHARED);
4520 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4521 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4525 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4527 switch (bf->l_whence) {
4528 case 0: /*SEEK_SET*/
4530 case 1: /*SEEK_CUR*/
4531 bf->l_start += offset;
4533 case 2: /*SEEK_END*/
4534 bf->l_start += ip->i_d.di_size;
4537 return XFS_ERROR(EINVAL);
4540 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4542 if ( (bf->l_start < 0)
4543 || (bf->l_start > XFS_MAXIOFFSET(mp))
4544 || (bf->l_start + llen < 0)
4545 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4546 return XFS_ERROR(EINVAL);
4550 startoffset = bf->l_start;
4551 fsize = ip->i_d.di_size;
4554 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4556 * These calls do NOT zero the data space allocated to the file,
4557 * nor do they change the file size.
4559 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4561 * These calls cause the new file data to be zeroed and the file
4562 * size to be changed.
4564 setprealloc = clrprealloc = 0;
4567 case XFS_IOC_RESVSP:
4568 case XFS_IOC_RESVSP64:
4569 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4576 case XFS_IOC_UNRESVSP:
4577 case XFS_IOC_UNRESVSP64:
4578 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4583 case XFS_IOC_ALLOCSP:
4584 case XFS_IOC_ALLOCSP64:
4585 case XFS_IOC_FREESP:
4586 case XFS_IOC_FREESP64:
4587 if (startoffset > fsize) {
4588 error = xfs_alloc_file_space(ip, fsize,
4589 startoffset - fsize, 0, attr_flags);
4594 va.va_mask = XFS_AT_SIZE;
4595 va.va_size = startoffset;
4597 error = xfs_setattr(bdp, &va, attr_flags, credp);
4607 return XFS_ERROR(EINVAL);
4611 * update the inode timestamp, mode, and prealloc flag bits
4613 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4615 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4618 xfs_trans_cancel(tp, 0);
4622 xfs_ilock(ip, XFS_ILOCK_EXCL);
4624 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4625 xfs_trans_ihold(tp, ip);
4627 if ((attr_flags & ATTR_DMI) == 0) {
4628 ip->i_d.di_mode &= ~S_ISUID;
4631 * Note that we don't have to worry about mandatory
4632 * file locking being disabled here because we only
4633 * clear the S_ISGID bit if the Group execute bit is
4634 * on, but if it was on then mandatory locking wouldn't
4635 * have been enabled.
4637 if (ip->i_d.di_mode & S_IXGRP)
4638 ip->i_d.di_mode &= ~S_ISGID;
4640 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4643 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4644 else if (clrprealloc)
4645 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4647 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4648 xfs_trans_set_sync(tp);
4650 error = xfs_trans_commit(tp, 0);
4652 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4657 bhv_vnodeops_t xfs_vnodeops = {
4658 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4659 .vop_open = xfs_open,
4660 .vop_close = xfs_close,
4661 .vop_read = xfs_read,
4662 #ifdef HAVE_SENDFILE
4663 .vop_sendfile = xfs_sendfile,
4666 .vop_splice_read = xfs_splice_read,
4667 .vop_splice_write = xfs_splice_write,
4669 .vop_write = xfs_write,
4670 .vop_ioctl = xfs_ioctl,
4671 .vop_getattr = xfs_getattr,
4672 .vop_setattr = xfs_setattr,
4673 .vop_access = xfs_access,
4674 .vop_lookup = xfs_lookup,
4675 .vop_create = xfs_create,
4676 .vop_remove = xfs_remove,
4677 .vop_link = xfs_link,
4678 .vop_rename = xfs_rename,
4679 .vop_mkdir = xfs_mkdir,
4680 .vop_rmdir = xfs_rmdir,
4681 .vop_readdir = xfs_readdir,
4682 .vop_symlink = xfs_symlink,
4683 .vop_readlink = xfs_readlink,
4684 .vop_fsync = xfs_fsync,
4685 .vop_inactive = xfs_inactive,
4686 .vop_fid2 = xfs_fid2,
4687 .vop_rwlock = xfs_rwlock,
4688 .vop_rwunlock = xfs_rwunlock,
4689 .vop_bmap = xfs_bmap,
4690 .vop_reclaim = xfs_reclaim,
4691 .vop_attr_get = xfs_attr_get,
4692 .vop_attr_set = xfs_attr_set,
4693 .vop_attr_remove = xfs_attr_remove,
4694 .vop_attr_list = xfs_attr_list,
4695 .vop_link_removed = (vop_link_removed_t)fs_noval,
4696 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4697 .vop_tosspages = fs_tosspages,
4698 .vop_flushinval_pages = fs_flushinval_pages,
4699 .vop_flush_pages = fs_flush_pages,
4700 .vop_release = xfs_release,
4701 .vop_iflush = xfs_inode_flush,