2 * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_itable.h"
49 #include "xfs_btree.h"
50 #include "xfs_ialloc.h"
51 #include "xfs_alloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
59 #include "xfs_da_btree.h"
62 #include "xfs_refcache.h"
63 #include "xfs_error.h"
65 #include "xfs_rtalloc.h"
66 #include "xfs_quota.h"
67 #include "xfs_utils.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_dir_leaf.h"
71 #include "xfs_log_priv.h"
75 * The maximum pathlen is 1024 bytes. Since the minimum file system
76 * blocksize is 512 bytes, we can get a max of 2 extents back from
79 #define SYMLINK_MAPS 2
82 * For xfs, we check that the file isn't too big to be opened by this kernel.
83 * No other open action is required for regular files. Devices are handled
84 * through the specfs file system, pipes through fifofs. Device and
85 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
86 * when a new vnode is first looked up or created.
97 vp = BHV_TO_VNODE(bdp);
100 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
101 return XFS_ERROR(EIO);
104 * If it's a directory with any blocks, read-ahead block 0
105 * as we're almost certain to have the next operation be a read there.
107 if (vp->v_type == VDIR && ip->i_d.di_nextents > 0) {
108 mode = xfs_ilock_map_shared(ip);
109 if (ip->i_d.di_nextents > 0)
110 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
111 xfs_iunlock(ip, mode);
131 vp = BHV_TO_VNODE(bdp);
132 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
134 ip = XFS_BHVTOI(bdp);
137 if (XFS_FORCED_SHUTDOWN(mp))
138 return XFS_ERROR(EIO);
140 if (!(flags & ATTR_LAZY))
141 xfs_ilock(ip, XFS_ILOCK_SHARED);
143 vap->va_size = ip->i_d.di_size;
144 if (vap->va_mask == XFS_AT_SIZE)
148 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
149 vap->va_nodeid = ip->i_ino;
151 vap->va_nodeid += mp->m_inoadd;
153 vap->va_nlink = ip->i_d.di_nlink;
156 * Quick exit for non-stat callers
159 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
160 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
164 * Copy from in-core inode.
166 vap->va_type = vp->v_type;
167 vap->va_mode = ip->i_d.di_mode & MODEMASK;
168 vap->va_uid = ip->i_d.di_uid;
169 vap->va_gid = ip->i_d.di_gid;
170 vap->va_xid = ip->i_d.di_xid;
171 vap->va_projid = ip->i_d.di_projid;
174 * Check vnode type block/char vs. everything else.
175 * Do it with bitmask because that's faster than looking
176 * for multiple values individually.
178 if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
181 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
184 /* Large block sizes confuse various
185 * user space programs, so letting the
186 * stripe size through is not a good
189 vap->va_blocksize = mp->m_swidth ?
191 * If the underlying volume is a stripe, then
192 * return the stripe width in bytes as the
193 * recommended I/O size.
195 (mp->m_swidth << mp->m_sb.sb_blocklog) :
197 * Return the largest of the preferred buffer
198 * sizes since doing small I/Os into larger
199 * buffers causes buffers to be decommissioned.
200 * The value returned is in bytes.
202 (1 << (int)MAX(mp->m_readio_log,
208 * Return the largest of the preferred buffer
209 * sizes since doing small I/Os into larger
210 * buffers causes buffers to be decommissioned.
211 * The value returned is in bytes.
213 1 << (int)MAX(mp->m_readio_log,
219 * If the file blocks are being allocated from a
220 * realtime partition, then return the inode's
221 * realtime extent size or the realtime volume's
224 vap->va_blocksize = ip->i_d.di_extsize ?
225 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
226 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
229 vap->va_rdev = ip->i_df.if_u2.if_rdev;
230 vap->va_blocksize = BLKDEV_IOSIZE;
233 vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
234 vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
235 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
236 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
237 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
238 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
241 * Exit for stat callers. See if any of the rest of the fields
242 * to be filled in are needed.
245 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
246 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
250 * Convert di_flags to xflags.
252 vap->va_xflags = xfs_ip2xflags(ip);
255 * Exit for inode revalidate. See if any of the rest of
256 * the fields to be filled in are needed.
259 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
260 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
263 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
265 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
266 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
270 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
271 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
272 ip->i_d.di_anextents;
274 vap->va_anextents = 0;
275 vap->va_gen = ip->i_d.di_gen;
278 if (!(flags & ATTR_LAZY))
279 xfs_iunlock(ip, XFS_ILOCK_SHARED);
306 xfs_prid_t projid=0, iprojid=0;
307 int mandlock_before, mandlock_after;
308 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
312 vp = BHV_TO_VNODE(bdp);
313 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
315 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
316 return XFS_ERROR(EROFS);
319 * Cannot set certain attributes.
322 if (mask & XFS_AT_NOSET) {
323 return XFS_ERROR(EINVAL);
326 ip = XFS_BHVTOI(bdp);
329 if (XFS_FORCED_SHUTDOWN(mp))
330 return XFS_ERROR(EIO);
333 * Timestamps do not need to be logged and hence do not
334 * need to be done within a transaction.
336 if (mask & XFS_AT_UPDTIMES) {
337 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
338 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
339 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
340 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
341 xfs_ichgtime(ip, timeflags);
345 olddquot1 = olddquot2 = NULL;
349 * If disk quotas is on, we make sure that the dquots do exist on disk,
350 * before we start any other transactions. Trying to do this later
351 * is messy. We don't care to take a readlock to look at the ids
352 * in inode here, because we can't hold it across the trans_reserve.
353 * If the IDs do change before we take the ilock, we're covered
354 * because the i_*dquot fields will get updated anyway.
356 if (XFS_IS_QUOTA_ON(mp) && (mask & (XFS_AT_UID|XFS_AT_GID))) {
359 /* FIXME: handle xid? */
360 if (mask & XFS_AT_UID) {
362 qflags |= XFS_QMOPT_UQUOTA;
364 uid = ip->i_d.di_uid;
366 if (mask & XFS_AT_GID) {
368 qflags |= XFS_QMOPT_GQUOTA;
370 gid = ip->i_d.di_gid;
373 * We take a reference when we initialize udqp and gdqp,
374 * so it is important that we never blindly double trip on
375 * the same variable. See xfs_create() for an example.
377 ASSERT(udqp == NULL);
378 ASSERT(gdqp == NULL);
379 code = XFS_QM_DQVOPALLOC(mp, ip, uid,gid, qflags, &udqp, &gdqp);
385 * For the other attributes, we acquire the inode lock and
386 * first do an error checking pass.
389 lock_flags = XFS_ILOCK_EXCL;
390 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
391 if (flags & ATTR_NOLOCK)
393 if (!(mask & XFS_AT_SIZE)) {
394 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
395 (mp->m_flags & XFS_MOUNT_WSYNC)) {
396 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
398 if ((code = xfs_trans_reserve(tp, 0,
399 XFS_ICHANGE_LOG_RES(mp), 0,
406 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
407 !(flags & ATTR_DMI)) {
408 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
409 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
410 vap->va_size, 0, dmflags, NULL);
417 lock_flags |= XFS_IOLOCK_EXCL;
420 xfs_ilock(ip, lock_flags);
422 /* boolean: are we the file owner? */
423 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
426 * Change various properties of a file.
427 * Only the owner or users with CAP_FOWNER
428 * capability may do these things.
431 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
432 XFS_AT_GID|XFS_AT_PROJID)) {
433 /* FIXME: handle xid? */
436 * CAP_FOWNER overrides the following restrictions:
438 * The user ID of the calling process must be equal
439 * to the file owner ID, except in cases where the
440 * CAP_FSETID capability is applicable.
442 if (!file_owner && !capable(CAP_FOWNER)) {
443 code = XFS_ERROR(EPERM);
448 * CAP_FSETID overrides the following restrictions:
450 * The effective user ID of the calling process shall match
451 * the file owner when setting the set-user-ID and
452 * set-group-ID bits on that file.
454 * The effective group ID or one of the supplementary group
455 * IDs of the calling process shall match the group owner of
456 * the file when setting the set-group-ID bit on that file
458 if (mask & XFS_AT_MODE) {
461 if ((vap->va_mode & S_ISUID) && !file_owner)
463 if ((vap->va_mode & S_ISGID) &&
464 !in_group_p((gid_t)ip->i_d.di_gid))
467 /* Linux allows this, Irix doesn't. */
468 if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
471 if (m && !capable(CAP_FSETID))
477 * Change file ownership. Must be the owner or privileged.
478 * If the system was configured with the "restricted_chown"
479 * option, the owner is not permitted to give away the file,
480 * and can change the group id only to a group of which he
481 * or she is a member.
483 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_XID|XFS_AT_PROJID)) {
485 * These IDs could have changed since we last looked at them.
486 * But, we're assured that if the ownership did change
487 * while we didn't have the inode locked, inode's dquot(s)
488 * would have changed also.
490 iuid = ip->i_d.di_uid;
491 igid = ip->i_d.di_gid;
492 ixid = ip->i_d.di_xid;
493 iprojid = ip->i_d.di_projid;
494 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
495 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
496 xid = (mask & XFS_AT_XID) ? vap->va_xid : ixid;
497 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
501 * CAP_CHOWN overrides the following restrictions:
503 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
504 * shall override the restriction that a process cannot
505 * change the user ID of a file it owns and the restriction
506 * that the group ID supplied to the chown() function
507 * shall be equal to either the group ID or one of the
508 * supplementary group IDs of the calling process.
510 * XXX: How does restricted_chown affect projid?
512 if (restricted_chown &&
513 (iuid != uid || (igid != gid &&
514 !in_group_p((gid_t)gid))) &&
515 !capable(CAP_CHOWN)) {
516 code = XFS_ERROR(EPERM);
520 * Do a quota reservation only if uid or gid is actually
523 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
524 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
525 /* FIXME: handle xid? */
527 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
528 capable(CAP_FOWNER) ?
529 XFS_QMOPT_FORCE_RES : 0);
530 if (code) /* out of quota */
536 * Truncate file. Must have write permission and not be a directory.
538 if (mask & XFS_AT_SIZE) {
539 /* Short circuit the truncate case for zero length files */
540 if ((vap->va_size == 0) &&
541 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
542 xfs_iunlock(ip, XFS_ILOCK_EXCL);
543 lock_flags &= ~XFS_ILOCK_EXCL;
544 if (mask & XFS_AT_CTIME)
545 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
550 if (vp->v_type == VDIR) {
551 code = XFS_ERROR(EISDIR);
553 } else if (vp->v_type != VREG) {
554 code = XFS_ERROR(EINVAL);
558 * Make sure that the dquots are attached to the inode.
560 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
565 * Change file access or modified times.
567 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
569 if ((flags & ATTR_UTIME) &&
570 !capable(CAP_FOWNER)) {
571 code = XFS_ERROR(EPERM);
578 * Change extent size or realtime flag.
580 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
582 * Can't change extent size if any extents are allocated.
584 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
585 (mask & XFS_AT_EXTSIZE) &&
586 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
588 code = XFS_ERROR(EINVAL); /* EFBIG? */
593 * Can't set extent size unless the file is marked, or
594 * about to be marked as a realtime file.
596 * This check will be removed when fixed size extents
597 * with buffered data writes is implemented.
600 if ((mask & XFS_AT_EXTSIZE) &&
601 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
603 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
604 ((mask & XFS_AT_XFLAGS) &&
605 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
606 code = XFS_ERROR(EINVAL);
611 * Can't change realtime flag if any extents are allocated.
613 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
614 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
615 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
616 code = XFS_ERROR(EINVAL); /* EFBIG? */
620 * Extent size must be a multiple of the appropriate block
621 * size, if set at all.
623 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
626 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
627 ((mask & XFS_AT_XFLAGS) &&
628 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
629 size = mp->m_sb.sb_rextsize <<
630 mp->m_sb.sb_blocklog;
632 size = mp->m_sb.sb_blocksize;
634 if (vap->va_extsize % size) {
635 code = XFS_ERROR(EINVAL);
640 * If realtime flag is set then must have realtime data.
642 if ((mask & XFS_AT_XFLAGS) &&
643 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
644 if ((mp->m_sb.sb_rblocks == 0) ||
645 (mp->m_sb.sb_rextsize == 0) ||
646 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
647 code = XFS_ERROR(EINVAL);
653 * Can't modify an immutable/append-only file unless
654 * we have appropriate permission.
656 if ((mask & XFS_AT_XFLAGS) &&
658 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
660 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
661 !capable(CAP_LINUX_IMMUTABLE)) {
662 code = XFS_ERROR(EPERM);
668 * Now we can make the changes. Before we join the inode
669 * to the transaction, if XFS_AT_SIZE is set then take care of
670 * the part of the truncation that must be done without the
671 * inode lock. This needs to be done before joining the inode
672 * to the transaction, because the inode cannot be unlocked
673 * once it is a part of the transaction.
675 if (mask & XFS_AT_SIZE) {
677 if (vap->va_size > ip->i_d.di_size)
678 code = xfs_igrow_start(ip, vap->va_size, credp);
679 xfs_iunlock(ip, XFS_ILOCK_EXCL);
681 code = xfs_itruncate_data(ip, vap->va_size);
684 lock_flags &= ~XFS_ILOCK_EXCL;
685 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
688 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
689 if ((code = xfs_trans_reserve(tp, 0,
690 XFS_ITRUNCATE_LOG_RES(mp), 0,
691 XFS_TRANS_PERM_LOG_RES,
692 XFS_ITRUNCATE_LOG_COUNT))) {
693 xfs_trans_cancel(tp, 0);
695 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
698 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
699 xfs_ilock(ip, XFS_ILOCK_EXCL);
703 xfs_trans_ijoin(tp, ip, lock_flags);
704 xfs_trans_ihold(tp, ip);
707 /* determine whether mandatory locking mode changes */
708 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
711 * Truncate file. Must have write permission and not be a directory.
713 if (mask & XFS_AT_SIZE) {
714 if (vap->va_size > ip->i_d.di_size) {
715 xfs_igrow_finish(tp, ip, vap->va_size,
716 !(flags & ATTR_DMI));
717 } else if ((vap->va_size <= ip->i_d.di_size) ||
718 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
720 * signal a sync transaction unless
721 * we're truncating an already unlinked
722 * file on a wsync filesystem
724 code = xfs_itruncate_finish(&tp, ip,
725 (xfs_fsize_t)vap->va_size,
727 ((ip->i_d.di_nlink != 0 ||
728 !(mp->m_flags & XFS_MOUNT_WSYNC))
735 * Have to do this even if the file's size doesn't change.
737 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
741 * Change file access modes.
743 if (mask & XFS_AT_MODE) {
744 ip->i_d.di_mode &= S_IFMT;
745 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
747 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
748 timeflags |= XFS_ICHGTIME_CHG;
752 * Change file ownership. Must be the owner or privileged.
753 * If the system was configured with the "restricted_chown"
754 * option, the owner is not permitted to give away the file,
755 * and can change the group id only to a group of which he
756 * or she is a member.
758 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_XID|XFS_AT_PROJID)) {
760 * CAP_FSETID overrides the following restrictions:
762 * The set-user-ID and set-group-ID bits of a file will be
763 * cleared upon successful return from chown()
765 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
766 !capable(CAP_FSETID)) {
767 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
771 * Change the ownerships and register quota modifications
772 * in the transaction.
775 if (XFS_IS_GQUOTA_ON(mp)) {
776 /* FIXME: handle xid quota? */
778 ip->i_d.di_xid = xid;
781 if (XFS_IS_UQUOTA_ON(mp)) {
782 ASSERT(mask & XFS_AT_UID);
784 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
785 &ip->i_udquot, udqp);
787 ip->i_d.di_uid = uid;
790 if (XFS_IS_GQUOTA_ON(mp)) {
791 ASSERT(mask & XFS_AT_GID);
793 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
794 &ip->i_gdquot, gdqp);
796 ip->i_d.di_gid = gid;
798 if (iprojid != projid) {
799 ip->i_d.di_projid = projid;
801 * We may have to rev the inode as well as
802 * the superblock version number since projids didn't
803 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
805 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
806 xfs_bump_ino_vers2(tp, ip);
809 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
810 timeflags |= XFS_ICHGTIME_CHG;
815 * Change file access or modified times.
817 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
818 if (mask & XFS_AT_ATIME) {
819 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
820 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
821 ip->i_update_core = 1;
822 timeflags &= ~XFS_ICHGTIME_ACC;
824 if (mask & XFS_AT_MTIME) {
825 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
826 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
827 timeflags &= ~XFS_ICHGTIME_MOD;
828 timeflags |= XFS_ICHGTIME_CHG;
830 if (tp && (flags & ATTR_UTIME))
831 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
835 * Change XFS-added attributes.
837 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
838 if (mask & XFS_AT_EXTSIZE) {
840 * Converting bytes to fs blocks.
842 ip->i_d.di_extsize = vap->va_extsize >>
843 mp->m_sb.sb_blocklog;
845 if (mask & XFS_AT_XFLAGS) {
848 /* can't set PREALLOC this way, just preserve it */
849 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
850 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
851 di_flags |= XFS_DIFLAG_IMMUTABLE;
852 if (vap->va_xflags & XFS_XFLAG_IUNLINK)
853 di_flags |= XFS_DIFLAG_IUNLINK;
854 if (vap->va_xflags & XFS_XFLAG_BARRIER)
855 di_flags |= XFS_DIFLAG_BARRIER;
856 if (vap->va_xflags & XFS_XFLAG_APPEND)
857 di_flags |= XFS_DIFLAG_APPEND;
858 if (vap->va_xflags & XFS_XFLAG_SYNC)
859 di_flags |= XFS_DIFLAG_SYNC;
860 if (vap->va_xflags & XFS_XFLAG_NOATIME)
861 di_flags |= XFS_DIFLAG_NOATIME;
862 if (vap->va_xflags & XFS_XFLAG_NODUMP)
863 di_flags |= XFS_DIFLAG_NODUMP;
864 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
865 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
866 di_flags |= XFS_DIFLAG_RTINHERIT;
867 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
868 di_flags |= XFS_DIFLAG_NOSYMLINKS;
870 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
871 di_flags |= XFS_DIFLAG_REALTIME;
872 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
874 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
877 ip->i_d.di_flags = di_flags;
879 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
880 timeflags |= XFS_ICHGTIME_CHG;
884 * Change file inode change time only if XFS_AT_CTIME set
885 * AND we have been called by a DMI function.
888 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
889 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
890 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
891 ip->i_update_core = 1;
892 timeflags &= ~XFS_ICHGTIME_CHG;
896 * Send out timestamp changes that need to be set to the
897 * current time. Not done when called by a DMI function.
899 if (timeflags && !(flags & ATTR_DMI))
900 xfs_ichgtime(ip, timeflags);
902 XFS_STATS_INC(xs_ig_attrchg);
905 * If this is a synchronous mount, make sure that the
906 * transaction goes to disk before returning to the user.
907 * This is slightly sub-optimal in that truncates require
908 * two sync transactions instead of one for wsync filesytems.
909 * One for the truncate and one for the timestamps since we
910 * don't want to change the timestamps unless we're sure the
911 * truncate worked. Truncates are less than 1% of the laddis
912 * mix so this probably isn't worth the trouble to optimize.
916 if (mp->m_flags & XFS_MOUNT_WSYNC)
917 xfs_trans_set_sync(tp);
919 code = xfs_trans_commit(tp, commit_flags, NULL);
923 * If the (regular) file's mandatory locking mode changed, then
924 * notify the vnode. We do this under the inode lock to prevent
925 * racing calls to vop_vnode_change.
927 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
928 if (mandlock_before != mandlock_after) {
929 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
933 xfs_iunlock(ip, lock_flags);
936 * Release any dquot(s) the inode had kept before chown.
938 XFS_QM_DQRELE(mp, olddquot1);
939 XFS_QM_DQRELE(mp, olddquot2);
940 XFS_QM_DQRELE(mp, udqp);
941 XFS_QM_DQRELE(mp, gdqp);
947 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
948 !(flags & ATTR_DMI)) {
949 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
950 NULL, DM_RIGHT_NULL, NULL, NULL,
951 0, 0, AT_DELAY_FLAG(flags));
956 commit_flags |= XFS_TRANS_ABORT;
959 XFS_QM_DQRELE(mp, udqp);
960 XFS_QM_DQRELE(mp, gdqp);
962 xfs_trans_cancel(tp, commit_flags);
964 if (lock_flags != 0) {
965 xfs_iunlock(ip, lock_flags);
973 * Null conversion from vnode mode bits to inode mode bits, as in efs.
984 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
985 (inst_t *)__return_address);
987 ip = XFS_BHVTOI(bdp);
988 xfs_ilock(ip, XFS_ILOCK_SHARED);
989 error = xfs_iaccess(ip, mode, credp);
990 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1014 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1020 vp = BHV_TO_VNODE(bdp);
1021 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1023 ip = XFS_BHVTOI(bdp);
1026 if (XFS_FORCED_SHUTDOWN(mp))
1027 return XFS_ERROR(EIO);
1029 xfs_ilock(ip, XFS_ILOCK_SHARED);
1031 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1033 offset = uiop->uio_offset;
1034 count = uiop->uio_resid;
1037 error = XFS_ERROR(EINVAL);
1045 if (!(ioflags & IO_INVIS)) {
1046 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1050 * See if the symlink is stored inline.
1052 pathlen = (int)ip->i_d.di_size;
1054 if (ip->i_df.if_flags & XFS_IFINLINE) {
1055 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1059 * Symlink not inline. Call bmap to get it in.
1061 nmaps = SYMLINK_MAPS;
1063 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1064 0, NULL, 0, mval, &nmaps, NULL);
1070 for (n = 0; n < nmaps; n++) {
1071 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1072 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1073 bp = xfs_buf_read(mp->m_ddev_targp, d,
1074 BTOBB(byte_cnt), 0);
1075 error = XFS_BUF_GETERROR(bp);
1077 xfs_ioerror_alert("xfs_readlink",
1078 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1082 if (pathlen < byte_cnt)
1084 pathlen -= byte_cnt;
1086 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1095 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1104 * This is called to sync the inode and its data out to disk.
1105 * We need to hold the I/O lock while flushing the data, and
1106 * the inode lock while flushing the inode. The inode lock CANNOT
1107 * be held while flushing the data, so acquire after we're done
1122 vn_trace_entry(BHV_TO_VNODE(bdp),
1123 __FUNCTION__, (inst_t *)__return_address);
1125 ip = XFS_BHVTOI(bdp);
1127 ASSERT(start >= 0 && stop >= -1);
1129 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1130 return XFS_ERROR(EIO);
1133 * We always need to make sure that the required inode state
1134 * is safe on disk. The vnode might be clean but because
1135 * of committed transactions that haven't hit the disk yet.
1136 * Likewise, there could be unflushed non-transactional
1137 * changes to the inode core that have to go to disk.
1139 * The following code depends on one assumption: that
1140 * any transaction that changes an inode logs the core
1141 * because it has to change some field in the inode core
1142 * (typically nextents or nblocks). That assumption
1143 * implies that any transactions against an inode will
1144 * catch any non-transactional updates. If inode-altering
1145 * transactions exist that violate this assumption, the
1146 * code breaks. Right now, it figures that if the involved
1147 * update_* field is clear and the inode is unpinned, the
1148 * inode is clean. Either it's been flushed or it's been
1149 * committed and the commit has hit the disk unpinning the inode.
1150 * (Note that xfs_inode_item_format() called at commit clears
1151 * the update_* fields.)
1153 xfs_ilock(ip, XFS_ILOCK_SHARED);
1155 /* If we are flushing data then we care about update_size
1156 * being set, otherwise we care about update_core
1158 if ((flag & FSYNC_DATA) ?
1159 (ip->i_update_size == 0) :
1160 (ip->i_update_core == 0)) {
1162 * Timestamps/size haven't changed since last inode
1163 * flush or inode transaction commit. That means
1164 * either nothing got written or a transaction
1165 * committed which caught the updates. If the
1166 * latter happened and the transaction hasn't
1167 * hit the disk yet, the inode will be still
1168 * be pinned. If it is, force the log.
1171 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1173 if (xfs_ipincount(ip)) {
1174 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1176 ((flag & FSYNC_WAIT)
1177 ? XFS_LOG_SYNC : 0));
1182 * Kick off a transaction to log the inode
1183 * core to get the updates. Make it
1184 * sync if FSYNC_WAIT is passed in (which
1185 * is done by everybody but specfs). The
1186 * sync transaction will also force the log.
1188 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1189 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1190 if ((error = xfs_trans_reserve(tp, 0,
1191 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1193 xfs_trans_cancel(tp, 0);
1196 xfs_ilock(ip, XFS_ILOCK_EXCL);
1199 * Note - it's possible that we might have pushed
1200 * ourselves out of the way during trans_reserve
1201 * which would flush the inode. But there's no
1202 * guarantee that the inode buffer has actually
1203 * gone out yet (it's delwri). Plus the buffer
1204 * could be pinned anyway if it's part of an
1205 * inode in another recent transaction. So we
1206 * play it safe and fire off the transaction anyway.
1208 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1209 xfs_trans_ihold(tp, ip);
1210 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1211 if (flag & FSYNC_WAIT)
1212 xfs_trans_set_sync(tp);
1213 error = xfs_trans_commit(tp, 0, NULL);
1215 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1221 * This is called by xfs_inactive to free any blocks beyond eof,
1222 * when the link count isn't zero.
1225 xfs_inactive_free_eofblocks(
1231 xfs_fileoff_t end_fsb;
1232 xfs_fileoff_t last_fsb;
1233 xfs_filblks_t map_len;
1235 xfs_bmbt_irec_t imap;
1238 * Figure out if there are any blocks beyond the end
1239 * of the file. If not, then there is nothing to do.
1241 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1242 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1243 map_len = last_fsb - end_fsb;
1248 xfs_ilock(ip, XFS_ILOCK_SHARED);
1249 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1250 NULL, 0, &imap, &nimaps, NULL);
1251 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1253 if (!error && (nimaps != 0) &&
1254 (imap.br_startblock != HOLESTARTBLOCK)) {
1256 * Attach the dquots to the inode up front.
1258 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1262 * There are blocks after the end of file.
1263 * Free them up now by truncating the file to
1266 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1269 * Do the xfs_itruncate_start() call before
1270 * reserving any log space because
1271 * itruncate_start will call into the buffer
1272 * cache and we can't
1273 * do that within a transaction.
1275 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1276 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1279 error = xfs_trans_reserve(tp, 0,
1280 XFS_ITRUNCATE_LOG_RES(mp),
1281 0, XFS_TRANS_PERM_LOG_RES,
1282 XFS_ITRUNCATE_LOG_COUNT);
1284 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1285 xfs_trans_cancel(tp, 0);
1286 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1290 xfs_ilock(ip, XFS_ILOCK_EXCL);
1291 xfs_trans_ijoin(tp, ip,
1294 xfs_trans_ihold(tp, ip);
1296 error = xfs_itruncate_finish(&tp, ip,
1301 * If we get an error at this point we
1302 * simply don't bother truncating the file.
1305 xfs_trans_cancel(tp,
1306 (XFS_TRANS_RELEASE_LOG_RES |
1309 error = xfs_trans_commit(tp,
1310 XFS_TRANS_RELEASE_LOG_RES,
1313 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1319 * Free a symlink that has blocks associated with it.
1322 xfs_inactive_symlink_rmt(
1330 xfs_fsblock_t first_block;
1331 xfs_bmap_free_t free_list;
1334 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1342 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1344 * We're freeing a symlink that has some
1345 * blocks allocated to it. Free the
1346 * blocks here. We know that we've got
1347 * either 1 or 2 extents and that we can
1348 * free them all in one bunmapi call.
1350 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1351 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1352 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1353 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1354 xfs_trans_cancel(tp, 0);
1359 * Lock the inode, fix the size, and join it to the transaction.
1360 * Hold it so in the normal path, we still have it locked for
1361 * the second transaction. In the error paths we need it
1362 * held so the cancel won't rele it, see below.
1364 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1365 size = (int)ip->i_d.di_size;
1366 ip->i_d.di_size = 0;
1367 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1368 xfs_trans_ihold(tp, ip);
1369 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1371 * Find the block(s) so we can inval and unmap them.
1374 XFS_BMAP_INIT(&free_list, &first_block);
1375 nmaps = sizeof(mval) / sizeof(mval[0]);
1376 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1377 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1381 * Invalidate the block(s).
1383 for (i = 0; i < nmaps; i++) {
1384 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1385 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1386 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1387 xfs_trans_binval(tp, bp);
1390 * Unmap the dead block(s) to the free_list.
1392 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1393 &first_block, &free_list, &done)))
1397 * Commit the first transaction. This logs the EFI and the inode.
1399 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1402 * The transaction must have been committed, since there were
1403 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1404 * The new tp has the extent freeing and EFDs.
1408 * The first xact was committed, so add the inode to the new one.
1409 * Mark it dirty so it will be logged and moved forward in the log as
1410 * part of every commit.
1412 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1413 xfs_trans_ihold(tp, ip);
1414 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1416 * Get a new, empty transaction to return to our caller.
1418 ntp = xfs_trans_dup(tp);
1420 * Commit the transaction containing extent freeing and EFD's.
1421 * If we get an error on the commit here or on the reserve below,
1422 * we need to unlock the inode since the new transaction doesn't
1423 * have the inode attached.
1425 error = xfs_trans_commit(tp, 0, NULL);
1428 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1432 * Remove the memory for extent descriptions (just bookkeeping).
1434 if (ip->i_df.if_bytes)
1435 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1436 ASSERT(ip->i_df.if_bytes == 0);
1438 * Put an itruncate log reservation in the new transaction
1441 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1442 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1443 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1447 * Return with the inode locked but not joined to the transaction.
1453 xfs_bmap_cancel(&free_list);
1456 * Have to come here with the inode locked and either
1457 * (held and in the transaction) or (not in the transaction).
1458 * If the inode isn't held then cancel would iput it, but
1459 * that's wrong since this is inactive and the vnode ref
1460 * count is 0 already.
1461 * Cancel won't do anything to the inode if held, but it still
1462 * needs to be locked until the cancel is done, if it was
1463 * joined to the transaction.
1465 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1466 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1473 xfs_inactive_symlink_local(
1479 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1481 * We're freeing a symlink which fit into
1482 * the inode. Just free the memory used
1483 * to hold the old symlink.
1485 error = xfs_trans_reserve(*tpp, 0,
1486 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1487 0, XFS_TRANS_PERM_LOG_RES,
1488 XFS_ITRUNCATE_LOG_COUNT);
1491 xfs_trans_cancel(*tpp, 0);
1495 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1498 * Zero length symlinks _can_ exist.
1500 if (ip->i_df.if_bytes > 0) {
1501 xfs_idata_realloc(ip,
1502 -(ip->i_df.if_bytes),
1504 ASSERT(ip->i_df.if_bytes == 0);
1521 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1524 ASSERT(ip->i_d.di_forkoff != 0);
1525 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1526 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1528 error = xfs_attr_inactive(ip);
1531 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1532 return (error); /* goto out*/
1535 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1536 error = xfs_trans_reserve(tp, 0,
1537 XFS_IFREE_LOG_RES(mp),
1538 0, XFS_TRANS_PERM_LOG_RES,
1539 XFS_INACTIVE_LOG_COUNT);
1541 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1542 xfs_trans_cancel(tp, 0);
1544 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1548 xfs_ilock(ip, XFS_ILOCK_EXCL);
1549 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1550 xfs_trans_ihold(tp, ip);
1551 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1553 ASSERT(ip->i_d.di_anextents == 0);
1568 vp = BHV_TO_VNODE(bdp);
1569 ip = XFS_BHVTOI(bdp);
1571 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1575 /* If this is a read-only mount, don't do this (would generate I/O) */
1576 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1579 #ifdef HAVE_REFCACHE
1580 /* If we are in the NFS reference cache then don't do this now */
1587 if (ip->i_d.di_nlink != 0) {
1588 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1589 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1590 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1591 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1592 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1594 /* Update linux inode block count after free above */
1595 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1596 ip->i_d.di_nblocks + ip->i_delayed_blks);
1606 * This is called when the vnode reference count for the vnode
1607 * goes to zero. If the file has been unlinked, then it must
1608 * now be truncated. Also, we clear all of the read-ahead state
1609 * kept for the inode here since the file is now closed.
1618 xfs_bmap_free_t free_list;
1619 xfs_fsblock_t first_block;
1626 vp = BHV_TO_VNODE(bdp);
1627 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1629 ip = XFS_BHVTOI(bdp);
1632 * If the inode is already free, then there can be nothing
1635 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1636 ASSERT(ip->i_df.if_real_bytes == 0);
1637 ASSERT(ip->i_df.if_broot_bytes == 0);
1638 return VN_INACTIVE_CACHE;
1642 * Only do a truncate if it's a regular file with
1643 * some actual space in it. It's OK to look at the
1644 * inode's fields without the lock because we're the
1645 * only one with a reference to the inode.
1647 truncate = ((ip->i_d.di_nlink == 0) &&
1648 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1649 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1653 if (ip->i_d.di_nlink == 0 &&
1654 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1655 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1660 /* If this is a read-only mount, don't do this (would generate I/O) */
1661 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1664 if (ip->i_d.di_nlink != 0) {
1665 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1666 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1667 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1668 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1669 (ip->i_delayed_blks != 0))) {
1670 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1671 return (VN_INACTIVE_CACHE);
1672 /* Update linux inode block count after free above */
1673 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1674 ip->i_d.di_nblocks + ip->i_delayed_blks);
1679 ASSERT(ip->i_d.di_nlink == 0);
1681 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1682 return (VN_INACTIVE_CACHE);
1684 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1687 * Do the xfs_itruncate_start() call before
1688 * reserving any log space because itruncate_start
1689 * will call into the buffer cache and we can't
1690 * do that within a transaction.
1692 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1694 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1696 error = xfs_trans_reserve(tp, 0,
1697 XFS_ITRUNCATE_LOG_RES(mp),
1698 0, XFS_TRANS_PERM_LOG_RES,
1699 XFS_ITRUNCATE_LOG_COUNT);
1701 /* Don't call itruncate_cleanup */
1702 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1703 xfs_trans_cancel(tp, 0);
1704 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1705 return (VN_INACTIVE_CACHE);
1708 xfs_ilock(ip, XFS_ILOCK_EXCL);
1709 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1710 xfs_trans_ihold(tp, ip);
1713 * normally, we have to run xfs_itruncate_finish sync.
1714 * But if filesystem is wsync and we're in the inactive
1715 * path, then we know that nlink == 0, and that the
1716 * xaction that made nlink == 0 is permanently committed
1717 * since xfs_remove runs as a synchronous transaction.
1719 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1720 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1723 xfs_trans_cancel(tp,
1724 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1725 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1726 return (VN_INACTIVE_CACHE);
1728 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1731 * If we get an error while cleaning up a
1732 * symlink we bail out.
1734 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1735 xfs_inactive_symlink_rmt(ip, &tp) :
1736 xfs_inactive_symlink_local(ip, &tp);
1740 return (VN_INACTIVE_CACHE);
1743 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1744 xfs_trans_ihold(tp, ip);
1746 error = xfs_trans_reserve(tp, 0,
1747 XFS_IFREE_LOG_RES(mp),
1748 0, XFS_TRANS_PERM_LOG_RES,
1749 XFS_INACTIVE_LOG_COUNT);
1751 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1752 xfs_trans_cancel(tp, 0);
1753 return (VN_INACTIVE_CACHE);
1756 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1757 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1758 xfs_trans_ihold(tp, ip);
1762 * If there are attributes associated with the file
1763 * then blow them away now. The code calls a routine
1764 * that recursively deconstructs the attribute fork.
1765 * We need to just commit the current transaction
1766 * because we can't use it for xfs_attr_inactive().
1768 if (ip->i_d.di_anextents > 0) {
1769 error = xfs_inactive_attrs(ip, &tp);
1771 * If we got an error, the transaction is already
1772 * cancelled, and the inode is unlocked. Just get out.
1775 return (VN_INACTIVE_CACHE);
1776 } else if (ip->i_afp) {
1777 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1783 XFS_BMAP_INIT(&free_list, &first_block);
1784 error = xfs_ifree(tp, ip, &free_list);
1787 * If we fail to free the inode, shut down. The cancel
1788 * might do that, we need to make sure. Otherwise the
1789 * inode might be lost for a long time or forever.
1791 if (!XFS_FORCED_SHUTDOWN(mp)) {
1793 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1794 error, mp->m_fsname);
1795 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1797 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1800 * Credit the quota account(s). The inode is gone.
1802 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1805 * Just ignore errors at this point. There is
1806 * nothing we can do except to try to keep going.
1808 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1810 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1813 * Release the dquots held by inode, if any.
1815 XFS_QM_DQDETACH(mp, ip);
1817 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1820 return VN_INACTIVE_CACHE;
1829 bhv_desc_t *dir_bdp,
1836 xfs_inode_t *dp, *ip;
1842 dir_vp = BHV_TO_VNODE(dir_bdp);
1843 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1845 dp = XFS_BHVTOI(dir_bdp);
1847 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1848 return XFS_ERROR(EIO);
1850 lock_mode = xfs_ilock_map_shared(dp);
1851 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1853 *vpp = XFS_ITOV(ip);
1856 xfs_iunlock_map_shared(dp, lock_mode);
1862 * xfs_create (create a new file).
1866 bhv_desc_t *dir_bdp,
1872 char *name = VNAME(dentry);
1874 xfs_inode_t *dp, *ip;
1880 xfs_bmap_free_t free_list;
1881 xfs_fsblock_t first_block;
1882 boolean_t dp_joined_to_trans;
1883 int dm_event_sent = 0;
1887 struct xfs_dquot *udqp, *gdqp;
1893 dir_vp = BHV_TO_VNODE(dir_bdp);
1894 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1896 dp = XFS_BHVTOI(dir_bdp);
1899 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1900 namelen = VNAMELEN(dentry);
1902 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1903 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1904 dir_vp, DM_RIGHT_NULL, NULL,
1905 DM_RIGHT_NULL, name, NULL,
1913 if (XFS_FORCED_SHUTDOWN(mp))
1914 return XFS_ERROR(EIO);
1916 /* Return through std_return after this point. */
1919 if (vap->va_mask & XFS_AT_PROJID)
1920 prid = (xfs_prid_t)vap->va_projid;
1922 prid = (xfs_prid_t)dfltprid;
1925 * Make sure that we have allocated dquot(s) on disk.
1927 error = XFS_QM_DQVOPALLOC(mp, dp,
1928 current_fsuid(credp), current_fsgid(credp),
1929 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1934 dp_joined_to_trans = B_FALSE;
1936 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1937 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1938 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1940 * Initially assume that the file does not exist and
1941 * reserve the resources for that case. If that is not
1942 * the case we'll drop the one we have and get a more
1943 * appropriate transaction later.
1945 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1946 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1947 if (error == ENOSPC) {
1949 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1950 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1958 xfs_ilock(dp, XFS_ILOCK_EXCL);
1960 XFS_BMAP_INIT(&free_list, &first_block);
1965 * Reserve disk quota and the inode.
1967 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1972 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1974 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1975 error = xfs_dir_ialloc(&tp, dp,
1976 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1977 rdev, credp, prid, resblks > 0,
1980 if (error == ENOSPC)
1987 * At this point, we've gotten a newly allocated inode.
1988 * It is locked (and joined to the transaction).
1991 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1994 * Now we join the directory inode to the transaction.
1995 * We do not do it earlier because xfs_dir_ialloc
1996 * might commit the previous transaction (and release
2001 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2002 dp_joined_to_trans = B_TRUE;
2004 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
2005 &first_block, &free_list,
2006 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2008 ASSERT(error != ENOSPC);
2011 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2012 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2015 * If this is a synchronous mount, make sure that the
2016 * create transaction goes to disk before returning to
2019 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2020 xfs_trans_set_sync(tp);
2026 * Attach the dquot(s) to the inodes and modify them incore.
2027 * These ids of the inode couldn't have changed since the new
2028 * inode has been locked ever since it was created.
2030 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2033 * xfs_trans_commit normally decrements the vnode ref count
2034 * when it unlocks the inode. Since we want to return the
2035 * vnode to the caller, we bump the vnode ref count now.
2040 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2042 xfs_bmap_cancel(&free_list);
2046 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2053 XFS_QM_DQRELE(mp, udqp);
2054 XFS_QM_DQRELE(mp, gdqp);
2057 * Propogate the fact that the vnode changed after the
2058 * xfs_inode locks have been released.
2060 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2064 /* Fallthrough to std_return with error = 0 */
2067 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2068 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2069 DM_EVENT_POSTCREATE)) {
2070 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2071 dir_vp, DM_RIGHT_NULL,
2073 DM_RIGHT_NULL, name, NULL,
2074 dm_di_mode, error, 0);
2079 cancel_flags |= XFS_TRANS_ABORT;
2084 xfs_trans_cancel(tp, cancel_flags);
2086 if (!dp_joined_to_trans && (dp != NULL))
2087 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2088 XFS_QM_DQRELE(mp, udqp);
2089 XFS_QM_DQRELE(mp, gdqp);
2095 * Wait until after the current transaction is aborted to
2096 * release the inode. This prevents recursive transactions
2097 * and deadlocks from xfs_inactive.
2099 cancel_flags |= XFS_TRANS_ABORT;
2100 xfs_trans_cancel(tp, cancel_flags);
2103 XFS_QM_DQRELE(mp, udqp);
2104 XFS_QM_DQRELE(mp, gdqp);
2111 * Some counters to see if (and how often) we are hitting some deadlock
2112 * prevention code paths.
2116 int xfs_rm_lock_delays;
2117 int xfs_rm_attempts;
2121 * The following routine will lock the inodes associated with the
2122 * directory and the named entry in the directory. The locks are
2123 * acquired in increasing inode number.
2125 * If the entry is "..", then only the directory is locked. The
2126 * vnode ref count will still include that from the .. entry in
2129 * There is a deadlock we need to worry about. If the locked directory is
2130 * in the AIL, it might be blocking up the log. The next inode we lock
2131 * could be already locked by another thread waiting for log space (e.g
2132 * a permanent log reservation with a long running transaction (see
2133 * xfs_itruncate_finish)). To solve this, we must check if the directory
2134 * is in the ail and use lock_nowait. If we can't lock, we need to
2135 * drop the inode lock on the directory and try again. xfs_iunlock will
2136 * potentially push the tail if we were holding up the log.
2139 xfs_lock_dir_and_entry(
2142 xfs_inode_t *ip) /* inode of entry 'name' */
2146 xfs_inode_t *ips[2];
2155 xfs_ilock(dp, XFS_ILOCK_EXCL);
2162 * We want to lock in increasing inum. Since we've already
2163 * acquired the lock on the directory, we may need to release
2164 * if if the inum of the entry turns out to be less.
2166 if (e_inum > dp->i_ino) {
2168 * We are already in the right order, so just
2169 * lock on the inode of the entry.
2170 * We need to use nowait if dp is in the AIL.
2173 lp = (xfs_log_item_t *)dp->i_itemp;
2174 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2175 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2182 * Unlock dp and try again.
2183 * xfs_iunlock will try to push the tail
2184 * if the inode is in the AIL.
2187 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2189 if ((attempts % 5) == 0) {
2190 delay(1); /* Don't just spin the CPU */
2192 xfs_rm_lock_delays++;
2198 xfs_ilock(ip, XFS_ILOCK_EXCL);
2200 } else if (e_inum < dp->i_ino) {
2201 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2205 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2207 /* else e_inum == dp->i_ino */
2208 /* This can happen if we're asked to lock /x/..
2209 * the entry is "..", which is also the parent directory.
2217 int xfs_small_retries;
2218 int xfs_middle_retries;
2219 int xfs_lots_retries;
2220 int xfs_lock_delays;
2224 * The following routine will lock n inodes in exclusive mode.
2225 * We assume the caller calls us with the inodes in i_ino order.
2227 * We need to detect deadlock where an inode that we lock
2228 * is in the AIL and we start waiting for another inode that is locked
2229 * by a thread in a long running transaction (such as truncate). This can
2230 * result in deadlock since the long running trans might need to wait
2231 * for the inode we just locked in order to push the tail and free space
2241 int attempts = 0, i, j, try_lock;
2244 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2255 for (; i < inodes; i++) {
2258 if (i && (ips[i] == ips[i-1])) /* Already locked */
2262 * If try_lock is not set yet, make sure all locked inodes
2263 * are not in the AIL.
2264 * If any are, set try_lock to be used later.
2268 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2269 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2270 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2277 * If any of the previous locks we have locked is in the AIL,
2278 * we must TRY to get the second and subsequent locks. If
2279 * we can't get any, we must release all we have
2284 /* try_lock must be 0 if i is 0. */
2286 * try_lock means we have an inode locked
2287 * that is in the AIL.
2290 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2294 * Unlock all previous guys and try again.
2295 * xfs_iunlock will try to push the tail
2296 * if the inode is in the AIL.
2299 for(j = i - 1; j >= 0; j--) {
2302 * Check to see if we've already
2303 * unlocked this one.
2304 * Not the first one going back,
2305 * and the inode ptr is the same.
2307 if ((j != (i - 1)) && ips[j] ==
2311 xfs_iunlock(ips[j], lock_mode);
2314 if ((attempts % 5) == 0) {
2315 delay(1); /* Don't just spin the CPU */
2325 xfs_ilock(ips[i], lock_mode);
2331 if (attempts < 5) xfs_small_retries++;
2332 else if (attempts < 100) xfs_middle_retries++;
2333 else xfs_lots_retries++;
2341 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2342 int remove_which_error_return = 0;
2344 #define REMOVE_DEBUG_TRACE(x)
2345 #endif /* ! DEBUG */
2354 bhv_desc_t *dir_bdp,
2359 char *name = VNAME(dentry);
2360 xfs_inode_t *dp, *ip;
2361 xfs_trans_t *tp = NULL;
2364 xfs_bmap_free_t free_list;
2365 xfs_fsblock_t first_block;
2373 dir_vp = BHV_TO_VNODE(dir_bdp);
2374 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2376 dp = XFS_BHVTOI(dir_bdp);
2379 if (XFS_FORCED_SHUTDOWN(mp))
2380 return XFS_ERROR(EIO);
2382 namelen = VNAMELEN(dentry);
2384 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2385 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2386 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2387 name, NULL, 0, 0, 0);
2392 /* From this point on, return through std_return */
2396 * We need to get a reference to ip before we get our log
2397 * reservation. The reason for this is that we cannot call
2398 * xfs_iget for an inode for which we do not have a reference
2399 * once we've acquired a log reservation. This is because the
2400 * inode we are trying to get might be in xfs_inactive going
2401 * for a log reservation. Since we'll have to wait for the
2402 * inactive code to complete before returning from xfs_iget,
2403 * we need to make sure that we don't have log space reserved
2404 * when we call xfs_iget. Instead we get an unlocked referece
2405 * to the inode before getting our log reservation.
2407 error = xfs_get_dir_entry(dentry, &ip);
2409 REMOVE_DEBUG_TRACE(__LINE__);
2413 dm_di_mode = ip->i_d.di_mode;
2415 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2419 error = XFS_QM_DQATTACH(mp, dp, 0);
2420 if (!error && dp != ip)
2421 error = XFS_QM_DQATTACH(mp, ip, 0);
2423 REMOVE_DEBUG_TRACE(__LINE__);
2428 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2429 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2431 * We try to get the real space reservation first,
2432 * allowing for directory btree deletion(s) implying
2433 * possible bmap insert(s). If we can't get the space
2434 * reservation then we use 0 instead, and avoid the bmap
2435 * btree insert(s) in the directory code by, if the bmap
2436 * insert tries to happen, instead trimming the LAST
2437 * block from the directory.
2439 resblks = XFS_REMOVE_SPACE_RES(mp);
2440 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2441 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2442 if (error == ENOSPC) {
2444 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2445 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2448 ASSERT(error != ENOSPC);
2449 REMOVE_DEBUG_TRACE(__LINE__);
2450 xfs_trans_cancel(tp, 0);
2455 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2457 REMOVE_DEBUG_TRACE(__LINE__);
2458 xfs_trans_cancel(tp, cancel_flags);
2464 * At this point, we've gotten both the directory and the entry
2467 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2470 * Increment vnode ref count only in this case since
2471 * there's an extra vnode reference in the case where
2475 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2479 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2481 XFS_BMAP_INIT(&free_list, &first_block);
2482 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2483 &first_block, &free_list, 0);
2485 ASSERT(error != ENOENT);
2486 REMOVE_DEBUG_TRACE(__LINE__);
2489 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2492 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2494 error = xfs_droplink(tp, ip);
2496 REMOVE_DEBUG_TRACE(__LINE__);
2500 /* Determine if this is the last link while
2501 * we are in the transaction.
2503 link_zero = (ip)->i_d.di_nlink==0;
2506 * Take an extra ref on the inode so that it doesn't
2507 * go to xfs_inactive() from within the commit.
2512 * If this is a synchronous mount, make sure that the
2513 * remove transaction goes to disk before returning to
2516 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2517 xfs_trans_set_sync(tp);
2520 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2522 REMOVE_DEBUG_TRACE(__LINE__);
2526 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2533 * Before we drop our extra reference to the inode, purge it
2534 * from the refcache if it is there. By waiting until afterwards
2535 * to do the IRELE, we ensure that we won't go inactive in the
2536 * xfs_refcache_purge_ip routine (although that would be OK).
2538 xfs_refcache_purge_ip(ip);
2540 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2543 * Let interposed file systems know about removed links.
2545 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2549 /* Fall through to std_return with error = 0 */
2551 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2552 DM_EVENT_POSTREMOVE)) {
2553 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2554 dir_vp, DM_RIGHT_NULL,
2555 NULL, DM_RIGHT_NULL,
2556 name, NULL, dm_di_mode, error, 0);
2561 xfs_bmap_cancel(&free_list);
2562 cancel_flags |= XFS_TRANS_ABORT;
2563 xfs_trans_cancel(tp, cancel_flags);
2568 * In this case make sure to not release the inode until after
2569 * the current transaction is aborted. Releasing it beforehand
2570 * can cause us to go to xfs_inactive and start a recursive
2571 * transaction which can easily deadlock with the current one.
2573 xfs_bmap_cancel(&free_list);
2574 cancel_flags |= XFS_TRANS_ABORT;
2575 xfs_trans_cancel(tp, cancel_flags);
2578 * Before we drop our extra reference to the inode, purge it
2579 * from the refcache if it is there. By waiting until afterwards
2580 * to do the IRELE, we ensure that we won't go inactive in the
2581 * xfs_refcache_purge_ip routine (although that would be OK).
2583 xfs_refcache_purge_ip(ip);
2597 bhv_desc_t *target_dir_bdp,
2602 xfs_inode_t *tdp, *sip;
2605 xfs_inode_t *ips[2];
2607 xfs_bmap_free_t free_list;
2608 xfs_fsblock_t first_block;
2611 vnode_t *target_dir_vp;
2612 bhv_desc_t *src_bdp;
2614 char *target_name = VNAME(dentry);
2617 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2618 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2619 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2621 target_namelen = VNAMELEN(dentry);
2622 if (src_vp->v_type == VDIR)
2623 return XFS_ERROR(EPERM);
2626 * For now, manually find the XFS behavior descriptor for
2627 * the source vnode. If it doesn't exist then something
2628 * is wrong and we should just return an error.
2629 * Eventually we need to figure out how link is going to
2630 * work in the face of stacked vnodes.
2632 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2633 if (src_bdp == NULL) {
2634 return XFS_ERROR(EXDEV);
2636 sip = XFS_BHVTOI(src_bdp);
2637 tdp = XFS_BHVTOI(target_dir_bdp);
2639 if (XFS_FORCED_SHUTDOWN(mp))
2640 return XFS_ERROR(EIO);
2642 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2643 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2644 target_dir_vp, DM_RIGHT_NULL,
2645 src_vp, DM_RIGHT_NULL,
2646 target_name, NULL, 0, 0, 0);
2651 /* Return through std_return after this point. */
2653 error = XFS_QM_DQATTACH(mp, sip, 0);
2654 if (!error && sip != tdp)
2655 error = XFS_QM_DQATTACH(mp, tdp, 0);
2659 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2660 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2661 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2662 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2663 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2664 if (error == ENOSPC) {
2666 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2667 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2674 if (sip->i_ino < tdp->i_ino) {
2682 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2685 * Increment vnode ref counts since xfs_trans_commit &
2686 * xfs_trans_cancel will both unlock the inodes and
2687 * decrement the associated ref counts.
2690 VN_HOLD(target_dir_vp);
2691 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2692 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2695 * If the source has too many links, we can't make any more to it.
2697 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2698 error = XFS_ERROR(EMLINK);
2703 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2707 XFS_BMAP_INIT(&free_list, &first_block);
2709 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2710 sip->i_ino, &first_block, &free_list,
2714 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2716 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2718 error = xfs_bumplink(tp, sip);
2724 * If this is a synchronous mount, make sure that the
2725 * link transaction goes to disk before returning to
2728 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2729 xfs_trans_set_sync(tp);
2732 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2734 xfs_bmap_cancel(&free_list);
2738 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2743 /* Fall through to std_return with error = 0. */
2745 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2746 DM_EVENT_POSTLINK)) {
2747 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2748 target_dir_vp, DM_RIGHT_NULL,
2749 src_vp, DM_RIGHT_NULL,
2750 target_name, NULL, 0, error, 0);
2755 cancel_flags |= XFS_TRANS_ABORT;
2758 xfs_trans_cancel(tp, cancel_flags);
2768 bhv_desc_t *dir_bdp,
2774 char *dir_name = VNAME(dentry);
2776 xfs_inode_t *cdp; /* inode of created dir */
2777 vnode_t *cvp; /* vnode of created dir */
2783 xfs_bmap_free_t free_list;
2784 xfs_fsblock_t first_block;
2786 boolean_t dp_joined_to_trans;
2787 boolean_t created = B_FALSE;
2788 int dm_event_sent = 0;
2790 struct xfs_dquot *udqp, *gdqp;
2795 dir_vp = BHV_TO_VNODE(dir_bdp);
2796 dp = XFS_BHVTOI(dir_bdp);
2799 if (XFS_FORCED_SHUTDOWN(mp))
2800 return XFS_ERROR(EIO);
2802 dir_namelen = VNAMELEN(dentry);
2805 dp_joined_to_trans = B_FALSE;
2806 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2808 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2809 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2810 dir_vp, DM_RIGHT_NULL, NULL,
2811 DM_RIGHT_NULL, dir_name, NULL,
2818 /* Return through std_return after this point. */
2820 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2824 if (vap->va_mask & XFS_AT_PROJID)
2825 prid = (xfs_prid_t)vap->va_projid;
2827 prid = (xfs_prid_t)dfltprid;
2830 * Make sure that we have allocated dquot(s) on disk.
2832 error = XFS_QM_DQVOPALLOC(mp, dp,
2833 current_fsuid(credp), current_fsgid(credp),
2834 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2838 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2839 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2840 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2841 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2842 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2843 if (error == ENOSPC) {
2845 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2846 XFS_TRANS_PERM_LOG_RES,
2847 XFS_MKDIR_LOG_COUNT);
2855 xfs_ilock(dp, XFS_ILOCK_EXCL);
2858 * Check for directory link count overflow.
2860 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2861 error = XFS_ERROR(EMLINK);
2866 * Reserve disk quota and the inode.
2868 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2873 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2876 * create the directory inode.
2878 error = xfs_dir_ialloc(&tp, dp,
2879 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2880 0, credp, prid, resblks > 0,
2883 if (error == ENOSPC)
2890 * Now we add the directory inode to the transaction.
2891 * We waited until now since xfs_dir_ialloc might start
2892 * a new transaction. Had we joined the transaction
2893 * earlier, the locks might have gotten released.
2896 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2897 dp_joined_to_trans = B_TRUE;
2899 XFS_BMAP_INIT(&free_list, &first_block);
2901 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2902 cdp->i_ino, &first_block, &free_list,
2903 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2905 ASSERT(error != ENOSPC);
2908 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2911 * Bump the in memory version number of the parent directory
2912 * so that other processes accessing it will recognize that
2913 * the directory has changed.
2917 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2923 error = xfs_bumplink(tp, dp);
2928 cvp = XFS_ITOV(cdp);
2936 * Attach the dquots to the new inode and modify the icount incore.
2938 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2941 * If this is a synchronous mount, make sure that the
2942 * mkdir transaction goes to disk before returning to
2945 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2946 xfs_trans_set_sync(tp);
2949 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2955 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2956 XFS_QM_DQRELE(mp, udqp);
2957 XFS_QM_DQRELE(mp, gdqp);
2962 /* Fall through to std_return with error = 0 or errno from
2963 * xfs_trans_commit. */
2966 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2967 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2968 DM_EVENT_POSTCREATE)) {
2969 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2970 dir_vp, DM_RIGHT_NULL,
2971 created ? XFS_ITOV(cdp):NULL,
2974 dm_di_mode, error, 0);
2980 xfs_bmap_cancel(&free_list);
2982 cancel_flags |= XFS_TRANS_ABORT;
2984 xfs_trans_cancel(tp, cancel_flags);
2985 XFS_QM_DQRELE(mp, udqp);
2986 XFS_QM_DQRELE(mp, gdqp);
2988 if (!dp_joined_to_trans && (dp != NULL)) {
2989 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3002 bhv_desc_t *dir_bdp,
3006 char *name = VNAME(dentry);
3008 xfs_inode_t *cdp; /* child directory */
3012 xfs_bmap_free_t free_list;
3013 xfs_fsblock_t first_block;
3022 dir_vp = BHV_TO_VNODE(dir_bdp);
3023 dp = XFS_BHVTOI(dir_bdp);
3026 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3028 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3029 return XFS_ERROR(EIO);
3030 namelen = VNAMELEN(dentry);
3032 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3033 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3034 dir_vp, DM_RIGHT_NULL,
3035 NULL, DM_RIGHT_NULL,
3036 name, NULL, 0, 0, 0);
3038 return XFS_ERROR(error);
3041 /* Return through std_return after this point. */
3046 * We need to get a reference to cdp before we get our log
3047 * reservation. The reason for this is that we cannot call
3048 * xfs_iget for an inode for which we do not have a reference
3049 * once we've acquired a log reservation. This is because the
3050 * inode we are trying to get might be in xfs_inactive going
3051 * for a log reservation. Since we'll have to wait for the
3052 * inactive code to complete before returning from xfs_iget,
3053 * we need to make sure that we don't have log space reserved
3054 * when we call xfs_iget. Instead we get an unlocked referece
3055 * to the inode before getting our log reservation.
3057 error = xfs_get_dir_entry(dentry, &cdp);
3059 REMOVE_DEBUG_TRACE(__LINE__);
3063 dm_di_mode = cdp->i_d.di_mode;
3066 * Get the dquots for the inodes.
3068 error = XFS_QM_DQATTACH(mp, dp, 0);
3069 if (!error && dp != cdp)
3070 error = XFS_QM_DQATTACH(mp, cdp, 0);
3073 REMOVE_DEBUG_TRACE(__LINE__);
3077 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3078 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3080 * We try to get the real space reservation first,
3081 * allowing for directory btree deletion(s) implying
3082 * possible bmap insert(s). If we can't get the space
3083 * reservation then we use 0 instead, and avoid the bmap
3084 * btree insert(s) in the directory code by, if the bmap
3085 * insert tries to happen, instead trimming the LAST
3086 * block from the directory.
3088 resblks = XFS_REMOVE_SPACE_RES(mp);
3089 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3090 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3091 if (error == ENOSPC) {
3093 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3094 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3097 ASSERT(error != ENOSPC);
3102 XFS_BMAP_INIT(&free_list, &first_block);
3105 * Now lock the child directory inode and the parent directory
3106 * inode in the proper order. This will take care of validating
3107 * that the directory entry for the child directory inode has
3108 * not changed while we were obtaining a log reservation.
3110 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3112 xfs_trans_cancel(tp, cancel_flags);
3117 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3120 * Only increment the parent directory vnode count if
3121 * we didn't bump it in looking up cdp. The only time
3122 * we don't bump it is when we're looking up ".".
3128 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3130 ASSERT(cdp->i_d.di_nlink >= 2);
3131 if (cdp->i_d.di_nlink != 2) {
3132 error = XFS_ERROR(ENOTEMPTY);
3135 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3136 error = XFS_ERROR(ENOTEMPTY);
3140 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3141 &first_block, &free_list, resblks);
3146 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3149 * Bump the in memory generation count on the parent
3150 * directory so that other can know that it has changed.
3155 * Drop the link from cdp's "..".
3157 error = xfs_droplink(tp, dp);
3163 * Drop the link from dp to cdp.
3165 error = xfs_droplink(tp, cdp);
3171 * Drop the "." link from cdp to self.
3173 error = xfs_droplink(tp, cdp);
3178 /* Determine these before committing transaction */
3179 last_cdp_link = (cdp)->i_d.di_nlink==0;
3182 * Take an extra ref on the child vnode so that it
3183 * does not go to xfs_inactive() from within the commit.
3188 * If this is a synchronous mount, make sure that the
3189 * rmdir transaction goes to disk before returning to
3192 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3193 xfs_trans_set_sync(tp);
3196 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3198 xfs_bmap_cancel(&free_list);
3199 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3205 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3213 * Let interposed file systems know about removed links.
3215 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3219 /* Fall through to std_return with error = 0 or the errno
3220 * from xfs_trans_commit. */
3222 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3223 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3224 dir_vp, DM_RIGHT_NULL,
3225 NULL, DM_RIGHT_NULL,
3226 name, NULL, dm_di_mode,
3232 xfs_bmap_cancel(&free_list);
3233 cancel_flags |= XFS_TRANS_ABORT;
3235 xfs_trans_cancel(tp, cancel_flags);
3243 * Read dp's entries starting at uiop->uio_offset and translate them into
3244 * bufsize bytes worth of struct dirents starting at bufbase.
3248 bhv_desc_t *dir_bdp,
3254 xfs_trans_t *tp = NULL;
3257 xfs_off_t start_offset;
3259 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3260 (inst_t *)__return_address);
3261 dp = XFS_BHVTOI(dir_bdp);
3263 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3264 return XFS_ERROR(EIO);
3267 lock_mode = xfs_ilock_map_shared(dp);
3268 start_offset = uiop->uio_offset;
3269 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3270 if (start_offset != uiop->uio_offset) {
3271 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3273 xfs_iunlock_map_shared(dp, lock_mode);
3284 bhv_desc_t *dir_bdp,
3297 xfs_bmap_free_t free_list;
3298 xfs_fsblock_t first_block;
3299 boolean_t dp_joined_to_trans;
3303 xfs_fileoff_t first_fsb;
3304 xfs_filblks_t fs_blocks;
3306 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3313 struct xfs_dquot *udqp, *gdqp;
3315 char *link_name = VNAME(dentry);
3319 dir_vp = BHV_TO_VNODE(dir_bdp);
3320 dp = XFS_BHVTOI(dir_bdp);
3321 dp_joined_to_trans = B_FALSE;
3326 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3330 if (XFS_FORCED_SHUTDOWN(mp))
3331 return XFS_ERROR(EIO);
3333 link_namelen = VNAMELEN(dentry);
3336 * Check component lengths of the target path name.
3338 pathlen = strlen(target_path);
3339 if (pathlen >= MAXPATHLEN) /* total string too long */
3340 return XFS_ERROR(ENAMETOOLONG);
3341 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3345 for(total = 0, path = target_path; total < pathlen;) {
3349 while(*path == '/') {
3355 * Count up to the next slash or end of path.
3356 * Error out if the component is bigger than MAXNAMELEN.
3358 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3359 if (++len >= MAXNAMELEN) {
3360 error = ENAMETOOLONG;
3367 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3368 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3369 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3370 link_name, target_path, 0, 0, 0);
3375 /* Return through std_return after this point. */
3378 if (vap->va_mask & XFS_AT_PROJID)
3379 prid = (xfs_prid_t)vap->va_projid;
3381 prid = (xfs_prid_t)dfltprid;
3384 * Make sure that we have allocated dquot(s) on disk.
3386 error = XFS_QM_DQVOPALLOC(mp, dp,
3387 current_fsuid(credp), current_fsgid(credp),
3388 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3392 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3393 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3395 * The symlink will fit into the inode data fork?
3396 * There can't be any attributes so we get the whole variable part.
3398 if (pathlen <= XFS_LITINO(mp))
3401 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3402 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3403 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3404 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3405 if (error == ENOSPC && fs_blocks == 0) {
3407 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3408 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3416 xfs_ilock(dp, XFS_ILOCK_EXCL);
3419 * Check whether the directory allows new symlinks or not.
3421 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3422 error = XFS_ERROR(EPERM);
3427 * Reserve disk quota : blocks and inode.
3429 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3434 * Check for ability to enter directory entry, if no space reserved.
3437 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3440 * Initialize the bmap freelist prior to calling either
3441 * bmapi or the directory create code.
3443 XFS_BMAP_INIT(&free_list, &first_block);
3446 * Allocate an inode for the symlink.
3448 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3449 1, 0, credp, prid, resblks > 0, &ip, NULL);
3451 if (error == ENOSPC)
3458 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3459 dp_joined_to_trans = B_TRUE;
3462 * Also attach the dquot(s) to it, if applicable.
3464 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3467 resblks -= XFS_IALLOC_SPACE_RES(mp);
3469 * If the symlink will fit into the inode, write it inline.
3471 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3472 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3473 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3474 ip->i_d.di_size = pathlen;
3477 * The inode was initially created in extent format.
3479 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3480 ip->i_df.if_flags |= XFS_IFINLINE;
3482 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3483 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3487 nmaps = SYMLINK_MAPS;
3489 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3490 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3491 &first_block, resblks, mval, &nmaps,
3498 resblks -= fs_blocks;
3499 ip->i_d.di_size = pathlen;
3500 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3502 cur_chunk = target_path;
3503 for (n = 0; n < nmaps; n++) {
3504 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3505 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3506 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3507 BTOBB(byte_cnt), 0);
3508 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3509 if (pathlen < byte_cnt) {
3512 pathlen -= byte_cnt;
3514 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3515 cur_chunk += byte_cnt;
3517 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3522 * Create the directory entry for the symlink.
3524 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3525 ip->i_ino, &first_block, &free_list, resblks);
3529 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3530 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3533 * Bump the in memory version number of the parent directory
3534 * so that other processes accessing it will recognize that
3535 * the directory has changed.
3540 * If this is a synchronous mount, make sure that the
3541 * symlink transaction goes to disk before returning to
3544 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3545 xfs_trans_set_sync(tp);
3549 * xfs_trans_commit normally decrements the vnode ref count
3550 * when it unlocks the inode. Since we want to return the
3551 * vnode to the caller, we bump the vnode ref count now.
3555 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3559 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3560 XFS_QM_DQRELE(mp, udqp);
3561 XFS_QM_DQRELE(mp, gdqp);
3563 /* Fall through to std_return with error = 0 or errno from
3564 * xfs_trans_commit */
3566 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3567 DM_EVENT_POSTSYMLINK)) {
3568 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3569 dir_vp, DM_RIGHT_NULL,
3570 error ? NULL : XFS_ITOV(ip),
3571 DM_RIGHT_NULL, link_name, target_path,
3587 xfs_bmap_cancel(&free_list);
3588 cancel_flags |= XFS_TRANS_ABORT;
3590 xfs_trans_cancel(tp, cancel_flags);
3591 XFS_QM_DQRELE(mp, udqp);
3592 XFS_QM_DQRELE(mp, gdqp);
3594 if (!dp_joined_to_trans && (dp != NULL)) {
3595 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3605 * A fid routine that takes a pointer to a previously allocated
3606 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3616 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3617 (inst_t *)__return_address);
3618 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3620 xfid = (xfs_fid2_t *)fidp;
3621 ip = XFS_BHVTOI(bdp);
3622 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3625 * use memcpy because the inode is a long long and there's no
3626 * assurance that xfid->fid_ino is properly aligned.
3628 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3629 xfid->fid_gen = ip->i_d.di_gen;
3646 vp = BHV_TO_VNODE(bdp);
3647 if (vp->v_type == VDIR)
3649 ip = XFS_BHVTOI(bdp);
3650 if (locktype == VRWLOCK_WRITE) {
3651 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3652 } else if (locktype == VRWLOCK_TRY_READ) {
3653 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3654 } else if (locktype == VRWLOCK_TRY_WRITE) {
3655 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3657 ASSERT((locktype == VRWLOCK_READ) ||
3658 (locktype == VRWLOCK_WRITE_DIRECT));
3659 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3677 vp = BHV_TO_VNODE(bdp);
3678 if (vp->v_type == VDIR)
3680 ip = XFS_BHVTOI(bdp);
3681 if (locktype == VRWLOCK_WRITE) {
3683 * In the write case, we may have added a new entry to
3684 * the reference cache. This might store a pointer to
3685 * an inode to be released in this inode. If it is there,
3686 * clear the pointer and release the inode after unlocking
3689 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3691 ASSERT((locktype == VRWLOCK_READ) ||
3692 (locktype == VRWLOCK_WRITE_DIRECT));
3693 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3705 xfs_inode_log_item_t *iip;
3708 ip = XFS_BHVTOI(bdp);
3712 if (XFS_FORCED_SHUTDOWN(mp))
3713 return XFS_ERROR(EIO);
3716 * Bypass inodes which have already been cleaned by
3717 * the inode flush clustering code inside xfs_iflush
3719 if ((ip->i_update_core == 0) &&
3720 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3723 if (flags & FLUSH_LOG) {
3724 if (iip && iip->ili_last_lsn) {
3725 xlog_t *log = mp->m_log;
3727 int s, log_flags = XFS_LOG_FORCE;
3729 s = GRANT_LOCK(log);
3730 sync_lsn = log->l_last_sync_lsn;
3731 GRANT_UNLOCK(log, s);
3733 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3736 if (flags & FLUSH_SYNC)
3737 log_flags |= XFS_LOG_SYNC;
3738 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3743 * We make this non-blocking if the inode is contended,
3744 * return EAGAIN to indicate to the caller that they
3745 * did not succeed. This prevents the flush path from
3746 * blocking on inodes inside another operation right
3747 * now, they get caught later by xfs_sync.
3749 if (flags & FLUSH_INODE) {
3752 if (xfs_ipincount(ip))
3755 if (flags & FLUSH_SYNC) {
3756 xfs_ilock(ip, XFS_ILOCK_SHARED);
3758 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3759 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3760 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3767 if (flags & FLUSH_SYNC)
3768 flush_flags = XFS_IFLUSH_SYNC;
3770 flush_flags = XFS_IFLUSH_ASYNC;
3772 error = xfs_iflush(ip, flush_flags);
3773 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3792 if (!capable(CAP_SYS_ADMIN))
3793 return XFS_ERROR(EPERM);
3795 ip = XFS_BHVTOI(bdp);
3798 if (XFS_FORCED_SHUTDOWN(mp))
3799 return XFS_ERROR(EIO);
3801 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3802 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3804 xfs_trans_cancel(tp, 0);
3807 xfs_ilock(ip, XFS_ILOCK_EXCL);
3808 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3810 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3811 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3813 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3815 error = xfs_trans_commit(tp, 0, NULL);
3831 vp = BHV_TO_VNODE(bdp);
3832 ip = XFS_BHVTOI(bdp);
3834 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3836 ASSERT(!VN_MAPPED(vp));
3838 /* bad inode, get out here ASAP */
3844 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3845 if (ip->i_d.di_size > 0) {
3847 * Flush and invalidate any data left around that is
3848 * a part of this file.
3850 * Get the inode's i/o lock so that buffers are pushed
3851 * out while holding the proper lock. We can't hold
3852 * the inode lock here since flushing out buffers may
3853 * cause us to try to get the lock in xfs_strategy().
3855 * We don't have to call remapf() here, because there
3856 * cannot be any mapped file references to this vnode
3857 * since it is being reclaimed.
3859 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3862 * If we hit an IO error, we need to make sure that the
3863 * buffer and page caches of file data for
3864 * the file are tossed away. We don't want to use
3865 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3866 * pages to stay attached to the vnode, but be
3867 * marked P_BAD. pdflush/vnode_pagebad
3870 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3871 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3873 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3876 ASSERT(VN_CACHED(vp) == 0);
3877 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3878 ip->i_delayed_blks == 0);
3879 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3880 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3882 * di_size field may not be quite accurate if we're
3885 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3886 ASSERT(VN_CACHED(vp) == 0);
3890 /* If we have nothing to flush with this inode then complete the
3891 * teardown now, otherwise break the link between the xfs inode
3892 * and the linux inode and clean up the xfs inode later. This
3893 * avoids flushing the inode to disk during the delete operation
3896 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3897 xfs_ilock(ip, XFS_ILOCK_EXCL);
3899 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3901 xfs_mount_t *mp = ip->i_mount;
3903 /* Protect sync from us */
3904 XFS_MOUNT_ILOCK(mp);
3905 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3906 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3907 ip->i_flags |= XFS_IRECLAIMABLE;
3908 XFS_MOUNT_IUNLOCK(mp);
3919 xfs_ihash_t *ih = ip->i_hash;
3920 vnode_t *vp = XFS_ITOV_NULL(ip);
3923 if (vp && VN_BAD(vp))
3926 /* The hash lock here protects a thread in xfs_iget_core from
3927 * racing with us on linking the inode back with a vnode.
3928 * Once we have the XFS_IRECLAIM flag set it will not touch
3931 write_lock(&ih->ih_lock);
3932 if ((ip->i_flags & XFS_IRECLAIM) ||
3933 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3934 write_unlock(&ih->ih_lock);
3937 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3941 ip->i_flags |= XFS_IRECLAIM;
3942 write_unlock(&ih->ih_lock);
3945 * If the inode is still dirty, then flush it out. If the inode
3946 * is not in the AIL, then it will be OK to flush it delwri as
3947 * long as xfs_iflush() does not keep any references to the inode.
3948 * We leave that decision up to xfs_iflush() since it has the
3949 * knowledge of whether it's OK to simply do a delwri flush of
3950 * the inode or whether we need to wait until the inode is
3951 * pulled from the AIL.
3952 * We get the flush lock regardless, though, just to make sure
3953 * we don't free it while it is being flushed.
3955 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3957 xfs_ilock(ip, XFS_ILOCK_EXCL);
3961 if (ip->i_update_core ||
3962 ((ip->i_itemp != NULL) &&
3963 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3964 error = xfs_iflush(ip, sync_mode);
3966 * If we hit an error, typically because of filesystem
3967 * shutdown, we don't need to let vn_reclaim to know
3968 * because we're gonna reclaim the inode anyway.
3971 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3974 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3977 ASSERT(ip->i_update_core == 0);
3978 ASSERT(ip->i_itemp == NULL ||
3979 ip->i_itemp->ili_format.ilf_fields == 0);
3980 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3981 } else if (locked) {
3983 * We are not interested in doing an iflush if we're
3984 * in the process of shutting down the filesystem forcibly.
3985 * So, just reclaim the inode.
3988 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3997 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
4000 xfs_inode_t *ip, *n;
4005 XFS_MOUNT_ILOCK(mp);
4006 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
4008 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
4010 if (xfs_ipincount(ip) ||
4011 !xfs_iflock_nowait(ip)) {
4012 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4016 XFS_MOUNT_IUNLOCK(mp);
4017 xfs_finish_reclaim(ip, noblock,
4018 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
4026 XFS_MOUNT_IUNLOCK(mp);
4031 * xfs_alloc_file_space()
4032 * This routine allocates disk space for the given file.
4034 * If alloc_type == 0, this request is for an ALLOCSP type
4035 * request which will change the file size. In this case, no
4036 * DMAPI event will be generated by the call. A TRUNCATE event
4037 * will be generated later by xfs_setattr.
4039 * If alloc_type != 0, this request is for a RESVSP type
4040 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4041 * lower block boundary byte address is less than the file's
4050 xfs_alloc_file_space(
4057 xfs_filblks_t allocated_fsb;
4058 xfs_filblks_t allocatesize_fsb;
4061 xfs_filblks_t datablocks;
4063 xfs_fsblock_t firstfsb;
4064 xfs_bmap_free_t free_list;
4065 xfs_bmbt_irec_t *imapp;
4066 xfs_bmbt_irec_t imaps[1];
4073 xfs_fileoff_t startoffset_fsb;
4075 int xfs_bmapi_flags;
4077 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4080 if (XFS_FORCED_SHUTDOWN(mp))
4081 return XFS_ERROR(EIO);
4084 * determine if this is a realtime file
4086 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4087 if (ip->i_d.di_extsize)
4088 rtextsize = ip->i_d.di_extsize;
4090 rtextsize = mp->m_sb.sb_rextsize;
4094 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4098 return XFS_ERROR(EINVAL);
4104 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4105 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4106 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4108 /* Generate a DMAPI event if needed. */
4109 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4110 (attr_flags&ATTR_DMI) == 0 &&
4111 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4112 xfs_off_t end_dmi_offset;
4114 end_dmi_offset = offset+len;
4115 if (end_dmi_offset > ip->i_d.di_size)
4116 end_dmi_offset = ip->i_d.di_size;
4117 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4118 offset, end_dmi_offset - offset,
4125 * allocate file space until done or until there is an error
4128 while (allocatesize_fsb && !error) {
4130 * determine if reserving space on
4131 * the data or realtime partition.
4136 s = startoffset_fsb;
4137 do_div(s, rtextsize);
4139 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4141 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4144 datablocks = allocatesize_fsb;
4149 * allocate and setup the transaction
4151 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4152 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4153 error = xfs_trans_reserve(tp,
4155 XFS_WRITE_LOG_RES(mp),
4157 XFS_TRANS_PERM_LOG_RES,
4158 XFS_WRITE_LOG_COUNT);
4161 * check for running out of space
4165 * Free the transaction structure.
4167 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4168 xfs_trans_cancel(tp, 0);
4171 xfs_ilock(ip, XFS_ILOCK_EXCL);
4172 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4173 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4174 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4178 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4179 xfs_trans_ihold(tp, ip);
4182 * issue the bmapi() call to allocate the blocks
4184 XFS_BMAP_INIT(&free_list, &firstfsb);
4185 error = xfs_bmapi(tp, ip, startoffset_fsb,
4186 allocatesize_fsb, xfs_bmapi_flags,
4187 &firstfsb, 0, imapp, &reccount,
4194 * complete the transaction
4196 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4201 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4202 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4207 allocated_fsb = imapp->br_blockcount;
4209 if (reccount == 0) {
4210 error = XFS_ERROR(ENOSPC);
4214 startoffset_fsb += allocated_fsb;
4215 allocatesize_fsb -= allocated_fsb;
4218 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4219 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4221 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4222 XFS_ITOV(ip), DM_RIGHT_NULL,
4223 XFS_ITOV(ip), DM_RIGHT_NULL,
4224 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4226 goto retry; /* Maybe DMAPI app. has made space */
4227 /* else fall through with error from XFS_SEND_DATA */
4233 xfs_bmap_cancel(&free_list);
4235 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4236 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4237 goto dmapi_enospc_check;
4241 * Zero file bytes between startoff and endoff inclusive.
4242 * The iolock is held exclusive and no blocks are buffered.
4245 xfs_zero_remaining_bytes(
4250 xfs_bmbt_irec_t imap;
4251 xfs_fileoff_t offset_fsb;
4252 xfs_off_t lastoffset;
4255 xfs_mount_t *mp = ip->i_mount;
4259 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4260 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4261 mp->m_rtdev_targp : mp->m_ddev_targp);
4263 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4264 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4266 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4268 if (error || nimap < 1)
4270 ASSERT(imap.br_blockcount >= 1);
4271 ASSERT(imap.br_startoff == offset_fsb);
4272 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4273 if (lastoffset > endoff)
4274 lastoffset = endoff;
4275 if (imap.br_startblock == HOLESTARTBLOCK)
4277 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4278 if (imap.br_state == XFS_EXT_UNWRITTEN)
4281 XFS_BUF_UNWRITE(bp);
4283 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4285 if ((error = xfs_iowait(bp))) {
4286 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4287 mp, bp, XFS_BUF_ADDR(bp));
4290 memset(XFS_BUF_PTR(bp) +
4291 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4292 0, lastoffset - offset + 1);
4297 if ((error = xfs_iowait(bp))) {
4298 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4299 mp, bp, XFS_BUF_ADDR(bp));
4308 * xfs_free_file_space()
4309 * This routine frees disk space for the given file.
4311 * This routine is only called by xfs_change_file_space
4312 * for an UNRESVSP type call.
4320 xfs_free_file_space(
4328 xfs_off_t end_dmi_offset;
4329 xfs_fileoff_t endoffset_fsb;
4331 xfs_fsblock_t firstfsb;
4332 xfs_bmap_free_t free_list;
4334 xfs_bmbt_irec_t imap;
4342 xfs_fileoff_t startoffset_fsb;
4344 int need_iolock = 1;
4346 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4349 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4353 if (len <= 0) /* if nothing being freed */
4355 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4356 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4357 end_dmi_offset = offset + len;
4358 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4360 if (offset < ip->i_d.di_size &&
4361 (attr_flags & ATTR_DMI) == 0 &&
4362 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4363 if (end_dmi_offset > ip->i_d.di_size)
4364 end_dmi_offset = ip->i_d.di_size;
4365 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4366 offset, end_dmi_offset - offset,
4367 AT_DELAY_FLAG(attr_flags), NULL);
4372 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4373 if (attr_flags & ATTR_NOLOCK)
4376 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4378 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4380 ilen = len + (offset & (rounding - 1));
4381 ioffset = offset & ~(rounding - 1);
4382 if (ilen & (rounding - 1))
4383 ilen = (ilen + rounding) & ~(rounding - 1);
4384 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4386 * Need to zero the stuff we're not freeing, on disk.
4387 * If its a realtime file & can't use unwritten extents then we
4388 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4389 * will take care of it for us.
4391 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4393 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4394 &imap, &nimap, NULL);
4396 goto out_unlock_iolock;
4397 ASSERT(nimap == 0 || nimap == 1);
4398 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4401 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4402 block = imap.br_startblock;
4403 mod = do_div(block, mp->m_sb.sb_rextsize);
4405 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4408 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4409 &imap, &nimap, NULL);
4411 goto out_unlock_iolock;
4412 ASSERT(nimap == 0 || nimap == 1);
4413 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4414 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4416 if (mod && (mod != mp->m_sb.sb_rextsize))
4417 endoffset_fsb -= mod;
4420 if ((done = (endoffset_fsb <= startoffset_fsb)))
4422 * One contiguous piece to clear
4424 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4427 * Some full blocks, possibly two pieces to clear
4429 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4430 error = xfs_zero_remaining_bytes(ip, offset,
4431 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4433 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4434 error = xfs_zero_remaining_bytes(ip,
4435 XFS_FSB_TO_B(mp, endoffset_fsb),
4440 * free file space until done or until there is an error
4442 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4443 while (!error && !done) {
4446 * allocate and setup the transaction
4448 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4449 error = xfs_trans_reserve(tp,
4451 XFS_WRITE_LOG_RES(mp),
4453 XFS_TRANS_PERM_LOG_RES,
4454 XFS_WRITE_LOG_COUNT);
4457 * check for running out of space
4461 * Free the transaction structure.
4463 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4464 xfs_trans_cancel(tp, 0);
4467 xfs_ilock(ip, XFS_ILOCK_EXCL);
4468 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4469 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4470 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4474 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4475 xfs_trans_ihold(tp, ip);
4478 * issue the bunmapi() call to free the blocks
4480 XFS_BMAP_INIT(&free_list, &firstfsb);
4481 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4482 endoffset_fsb - startoffset_fsb,
4483 0, 2, &firstfsb, &free_list, &done);
4489 * complete the transaction
4491 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4496 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4497 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4502 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4506 xfs_bmap_cancel(&free_list);
4508 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4509 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4515 * xfs_change_file_space()
4516 * This routine allocates or frees disk space for the given file.
4517 * The user specified parameters are checked for alignment and size
4526 xfs_change_file_space(
4540 xfs_off_t startoffset;
4546 vp = BHV_TO_VNODE(bdp);
4547 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4549 ip = XFS_BHVTOI(bdp);
4553 * must be a regular file and have write permission
4555 if (vp->v_type != VREG)
4556 return XFS_ERROR(EINVAL);
4558 xfs_ilock(ip, XFS_ILOCK_SHARED);
4560 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4561 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4565 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4567 switch (bf->l_whence) {
4568 case 0: /*SEEK_SET*/
4570 case 1: /*SEEK_CUR*/
4571 bf->l_start += offset;
4573 case 2: /*SEEK_END*/
4574 bf->l_start += ip->i_d.di_size;
4577 return XFS_ERROR(EINVAL);
4580 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4582 if ( (bf->l_start < 0)
4583 || (bf->l_start > XFS_MAXIOFFSET(mp))
4584 || (bf->l_start + llen < 0)
4585 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4586 return XFS_ERROR(EINVAL);
4590 startoffset = bf->l_start;
4591 fsize = ip->i_d.di_size;
4594 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4596 * These calls do NOT zero the data space allocated to the file,
4597 * nor do they change the file size.
4599 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4601 * These calls cause the new file data to be zeroed and the file
4602 * size to be changed.
4604 setprealloc = clrprealloc = 0;
4607 case XFS_IOC_RESVSP:
4608 case XFS_IOC_RESVSP64:
4609 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4616 case XFS_IOC_UNRESVSP:
4617 case XFS_IOC_UNRESVSP64:
4618 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4623 case XFS_IOC_ALLOCSP:
4624 case XFS_IOC_ALLOCSP64:
4625 case XFS_IOC_FREESP:
4626 case XFS_IOC_FREESP64:
4627 if (startoffset > fsize) {
4628 error = xfs_alloc_file_space(ip, fsize,
4629 startoffset - fsize, 0, attr_flags);
4634 va.va_mask = XFS_AT_SIZE;
4635 va.va_size = startoffset;
4637 error = xfs_setattr(bdp, &va, attr_flags, credp);
4647 return XFS_ERROR(EINVAL);
4651 * update the inode timestamp, mode, and prealloc flag bits
4653 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4655 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4658 xfs_trans_cancel(tp, 0);
4662 xfs_ilock(ip, XFS_ILOCK_EXCL);
4664 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4665 xfs_trans_ihold(tp, ip);
4667 if ((attr_flags & ATTR_DMI) == 0) {
4668 ip->i_d.di_mode &= ~S_ISUID;
4671 * Note that we don't have to worry about mandatory
4672 * file locking being disabled here because we only
4673 * clear the S_ISGID bit if the Group execute bit is
4674 * on, but if it was on then mandatory locking wouldn't
4675 * have been enabled.
4677 if (ip->i_d.di_mode & S_IXGRP)
4678 ip->i_d.di_mode &= ~S_ISGID;
4680 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4683 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4684 else if (clrprealloc)
4685 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4687 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4688 xfs_trans_set_sync(tp);
4690 error = xfs_trans_commit(tp, 0, NULL);
4692 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4697 vnodeops_t xfs_vnodeops = {
4698 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4699 .vop_open = xfs_open,
4700 .vop_read = xfs_read,
4701 #ifdef HAVE_SENDFILE
4702 .vop_sendfile = xfs_sendfile,
4704 .vop_write = xfs_write,
4705 .vop_ioctl = xfs_ioctl,
4706 .vop_getattr = xfs_getattr,
4707 .vop_setattr = xfs_setattr,
4708 .vop_access = xfs_access,
4709 .vop_lookup = xfs_lookup,
4710 .vop_create = xfs_create,
4711 .vop_remove = xfs_remove,
4712 .vop_link = xfs_link,
4713 .vop_rename = xfs_rename,
4714 .vop_mkdir = xfs_mkdir,
4715 .vop_rmdir = xfs_rmdir,
4716 .vop_readdir = xfs_readdir,
4717 .vop_symlink = xfs_symlink,
4718 .vop_readlink = xfs_readlink,
4719 .vop_fsync = xfs_fsync,
4720 .vop_inactive = xfs_inactive,
4721 .vop_fid2 = xfs_fid2,
4722 .vop_rwlock = xfs_rwlock,
4723 .vop_rwunlock = xfs_rwunlock,
4724 .vop_bmap = xfs_bmap,
4725 .vop_reclaim = xfs_reclaim,
4726 .vop_attr_get = xfs_attr_get,
4727 .vop_attr_set = xfs_attr_set,
4728 .vop_attr_remove = xfs_attr_remove,
4729 .vop_attr_list = xfs_attr_list,
4730 .vop_link_removed = (vop_link_removed_t)fs_noval,
4731 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4732 .vop_tosspages = fs_tosspages,
4733 .vop_flushinval_pages = fs_flushinval_pages,
4734 .vop_flush_pages = fs_flush_pages,
4735 .vop_release = xfs_release,
4736 .vop_iflush = xfs_inode_flush,