2 * Copyright (c) 2000-2003 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_projid = ip->i_d.di_projid;
173 * Check vnode type block/char vs. everything else.
174 * Do it with bitmask because that's faster than looking
175 * for multiple values individually.
177 if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
180 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
183 /* Large block sizes confuse various
184 * user space programs, so letting the
185 * stripe size through is not a good
188 vap->va_blocksize = mp->m_swidth ?
190 * If the underlying volume is a stripe, then
191 * return the stripe width in bytes as the
192 * recommended I/O size.
194 (mp->m_swidth << mp->m_sb.sb_blocklog) :
196 * Return the largest of the preferred buffer
197 * sizes since doing small I/Os into larger
198 * buffers causes buffers to be decommissioned.
199 * The value returned is in bytes.
201 (1 << (int)MAX(mp->m_readio_log,
207 * Return the largest of the preferred buffer
208 * sizes since doing small I/Os into larger
209 * buffers causes buffers to be decommissioned.
210 * The value returned is in bytes.
212 1 << (int)MAX(mp->m_readio_log,
218 * If the file blocks are being allocated from a
219 * realtime partition, then return the inode's
220 * realtime extent size or the realtime volume's
223 vap->va_blocksize = ip->i_d.di_extsize ?
224 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
225 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
228 vap->va_rdev = ip->i_df.if_u2.if_rdev;
229 vap->va_blocksize = BLKDEV_IOSIZE;
232 vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
233 vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
234 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
235 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
236 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
237 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
240 * Exit for stat callers. See if any of the rest of the fields
241 * to be filled in are needed.
244 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
245 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
249 * Convert di_flags to xflags.
251 vap->va_xflags = xfs_dic2xflags(&ip->i_d, ARCH_NOCONVERT);
254 * Exit for inode revalidate. See if any of the rest of
255 * the fields to be filled in are needed.
258 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
259 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
262 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
264 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
265 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
269 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
270 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
271 ip->i_d.di_anextents;
273 vap->va_anextents = 0;
274 vap->va_gen = ip->i_d.di_gen;
277 if (!(flags & ATTR_LAZY))
278 xfs_iunlock(ip, XFS_ILOCK_SHARED);
304 xfs_prid_t projid=0, iprojid=0;
305 int mandlock_before, mandlock_after;
306 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
309 vp = BHV_TO_VNODE(bdp);
310 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
312 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
313 return XFS_ERROR(EROFS);
316 * Cannot set certain attributes.
319 if (mask & XFS_AT_NOSET) {
320 return XFS_ERROR(EINVAL);
323 ip = XFS_BHVTOI(bdp);
326 if (XFS_FORCED_SHUTDOWN(mp))
327 return XFS_ERROR(EIO);
330 * Timestamps do not need to be logged and hence do not
331 * need to be done within a transaction.
333 if (mask & XFS_AT_UPDTIMES) {
334 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
335 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
336 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
337 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
338 xfs_ichgtime(ip, timeflags);
342 olddquot1 = olddquot2 = NULL;
346 * If disk quotas is on, we make sure that the dquots do exist on disk,
347 * before we start any other transactions. Trying to do this later
348 * is messy. We don't care to take a readlock to look at the ids
349 * in inode here, because we can't hold it across the trans_reserve.
350 * If the IDs do change before we take the ilock, we're covered
351 * because the i_*dquot fields will get updated anyway.
353 if (XFS_IS_QUOTA_ON(mp) && (mask & (XFS_AT_UID|XFS_AT_GID))) {
356 if (mask & XFS_AT_UID) {
358 qflags |= XFS_QMOPT_UQUOTA;
360 uid = ip->i_d.di_uid;
362 if (mask & XFS_AT_GID) {
364 qflags |= XFS_QMOPT_GQUOTA;
366 gid = ip->i_d.di_gid;
369 * We take a reference when we initialize udqp and gdqp,
370 * so it is important that we never blindly double trip on
371 * the same variable. See xfs_create() for an example.
373 ASSERT(udqp == NULL);
374 ASSERT(gdqp == NULL);
375 code = XFS_QM_DQVOPALLOC(mp, ip, uid,gid, qflags, &udqp, &gdqp);
381 * For the other attributes, we acquire the inode lock and
382 * first do an error checking pass.
385 lock_flags = XFS_ILOCK_EXCL;
386 if (!(mask & XFS_AT_SIZE)) {
387 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
388 (mp->m_flags & XFS_MOUNT_WSYNC)) {
389 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
391 if ((code = xfs_trans_reserve(tp, 0,
392 XFS_ICHANGE_LOG_RES(mp), 0,
399 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
400 !(flags & ATTR_DMI)) {
401 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
402 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
403 vap->va_size, 0, dmflags, NULL);
409 lock_flags |= XFS_IOLOCK_EXCL;
412 xfs_ilock(ip, lock_flags);
414 /* boolean: are we the file owner? */
415 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
418 * Change various properties of a file.
419 * Only the owner or users with CAP_FOWNER
420 * capability may do these things.
423 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
424 XFS_AT_GID|XFS_AT_PROJID)) {
426 * CAP_FOWNER overrides the following restrictions:
428 * The user ID of the calling process must be equal
429 * to the file owner ID, except in cases where the
430 * CAP_FSETID capability is applicable.
432 if (!file_owner && !capable(CAP_FOWNER)) {
433 code = XFS_ERROR(EPERM);
438 * CAP_FSETID overrides the following restrictions:
440 * The effective user ID of the calling process shall match
441 * the file owner when setting the set-user-ID and
442 * set-group-ID bits on that file.
444 * The effective group ID or one of the supplementary group
445 * IDs of the calling process shall match the group owner of
446 * the file when setting the set-group-ID bit on that file
448 if (mask & XFS_AT_MODE) {
451 if ((vap->va_mode & S_ISUID) && !file_owner)
453 if ((vap->va_mode & S_ISGID) &&
454 !in_group_p((gid_t)ip->i_d.di_gid))
457 /* Linux allows this, Irix doesn't. */
458 if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
461 if (m && !capable(CAP_FSETID))
467 * Change file ownership. Must be the owner or privileged.
468 * If the system was configured with the "restricted_chown"
469 * option, the owner is not permitted to give away the file,
470 * and can change the group id only to a group of which he
471 * or she is a member.
473 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
475 * These IDs could have changed since we last looked at them.
476 * But, we're assured that if the ownership did change
477 * while we didn't have the inode locked, inode's dquot(s)
478 * would have changed also.
480 iuid = ip->i_d.di_uid;
481 iprojid = ip->i_d.di_projid;
482 igid = ip->i_d.di_gid;
483 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
484 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
485 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
489 * CAP_CHOWN overrides the following restrictions:
491 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
492 * shall override the restriction that a process cannot
493 * change the user ID of a file it owns and the restriction
494 * that the group ID supplied to the chown() function
495 * shall be equal to either the group ID or one of the
496 * supplementary group IDs of the calling process.
498 * XXX: How does restricted_chown affect projid?
500 if (restricted_chown &&
501 (iuid != uid || (igid != gid &&
502 !in_group_p((gid_t)gid))) &&
503 !capable(CAP_CHOWN)) {
504 code = XFS_ERROR(EPERM);
508 * Do a quota reservation only if uid or gid is actually
511 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
512 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
514 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
515 capable(CAP_FOWNER) ?
516 XFS_QMOPT_FORCE_RES : 0);
517 if (code) /* out of quota */
523 * Truncate file. Must have write permission and not be a directory.
525 if (mask & XFS_AT_SIZE) {
526 /* Short circuit the truncate case for zero length files */
527 if ((vap->va_size == 0) &&
528 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
529 xfs_iunlock(ip, XFS_ILOCK_EXCL);
530 lock_flags &= ~XFS_ILOCK_EXCL;
531 if (mask & XFS_AT_CTIME)
532 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
537 if (vp->v_type == VDIR) {
538 code = XFS_ERROR(EISDIR);
540 } else if (vp->v_type != VREG) {
541 code = XFS_ERROR(EINVAL);
545 * Make sure that the dquots are attached to the inode.
547 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
552 * Change file access or modified times.
554 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
556 if ((flags & ATTR_UTIME) &&
557 !capable(CAP_FOWNER)) {
558 code = XFS_ERROR(EPERM);
565 * Change extent size or realtime flag.
567 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
569 * Can't change extent size if any extents are allocated.
571 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
572 (mask & XFS_AT_EXTSIZE) &&
573 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
575 code = XFS_ERROR(EINVAL); /* EFBIG? */
580 * Can't set extent size unless the file is marked, or
581 * about to be marked as a realtime file.
583 * This check will be removed when fixed size extents
584 * with buffered data writes is implemented.
587 if ((mask & XFS_AT_EXTSIZE) &&
588 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
590 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
591 ((mask & XFS_AT_XFLAGS) &&
592 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
593 code = XFS_ERROR(EINVAL);
598 * Can't change realtime flag if any extents are allocated.
600 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
601 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
602 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
603 code = XFS_ERROR(EINVAL); /* EFBIG? */
607 * Extent size must be a multiple of the appropriate block
608 * size, if set at all.
610 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
613 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
614 ((mask & XFS_AT_XFLAGS) &&
615 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
616 size = mp->m_sb.sb_rextsize <<
617 mp->m_sb.sb_blocklog;
619 size = mp->m_sb.sb_blocksize;
621 if (vap->va_extsize % size) {
622 code = XFS_ERROR(EINVAL);
627 * If realtime flag is set then must have realtime data.
629 if ((mask & XFS_AT_XFLAGS) &&
630 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
631 if ((mp->m_sb.sb_rblocks == 0) ||
632 (mp->m_sb.sb_rextsize == 0) ||
633 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
634 code = XFS_ERROR(EINVAL);
640 * Can't modify an immutable/append-only file unless
641 * we have appropriate permission.
643 if ((mask & XFS_AT_XFLAGS) &&
645 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
647 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
648 !capable(CAP_LINUX_IMMUTABLE)) {
649 code = XFS_ERROR(EPERM);
655 * Now we can make the changes. Before we join the inode
656 * to the transaction, if XFS_AT_SIZE is set then take care of
657 * the part of the truncation that must be done without the
658 * inode lock. This needs to be done before joining the inode
659 * to the transaction, because the inode cannot be unlocked
660 * once it is a part of the transaction.
662 if (mask & XFS_AT_SIZE) {
664 if (vap->va_size > ip->i_d.di_size)
665 code = xfs_igrow_start(ip, vap->va_size, credp);
666 xfs_iunlock(ip, XFS_ILOCK_EXCL);
668 code = xfs_itruncate_data(ip, vap->va_size);
671 lock_flags &= ~XFS_ILOCK_EXCL;
672 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
675 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
676 if ((code = xfs_trans_reserve(tp, 0,
677 XFS_ITRUNCATE_LOG_RES(mp), 0,
678 XFS_TRANS_PERM_LOG_RES,
679 XFS_ITRUNCATE_LOG_COUNT))) {
680 xfs_trans_cancel(tp, 0);
681 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
684 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
685 xfs_ilock(ip, XFS_ILOCK_EXCL);
689 xfs_trans_ijoin(tp, ip, lock_flags);
690 xfs_trans_ihold(tp, ip);
693 /* determine whether mandatory locking mode changes */
694 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
697 * Truncate file. Must have write permission and not be a directory.
699 if (mask & XFS_AT_SIZE) {
700 if (vap->va_size > ip->i_d.di_size) {
701 xfs_igrow_finish(tp, ip, vap->va_size,
702 !(flags & ATTR_DMI));
703 } else if ((vap->va_size <= ip->i_d.di_size) ||
704 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
706 * signal a sync transaction unless
707 * we're truncating an already unlinked
708 * file on a wsync filesystem
710 code = xfs_itruncate_finish(&tp, ip,
711 (xfs_fsize_t)vap->va_size,
713 ((ip->i_d.di_nlink != 0 ||
714 !(mp->m_flags & XFS_MOUNT_WSYNC))
721 * Have to do this even if the file's size doesn't change.
723 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
727 * Change file access modes.
729 if (mask & XFS_AT_MODE) {
730 ip->i_d.di_mode &= S_IFMT;
731 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
733 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
734 timeflags |= XFS_ICHGTIME_CHG;
738 * Change file ownership. Must be the owner or privileged.
739 * If the system was configured with the "restricted_chown"
740 * option, the owner is not permitted to give away the file,
741 * and can change the group id only to a group of which he
742 * or she is a member.
744 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
746 * CAP_FSETID overrides the following restrictions:
748 * The set-user-ID and set-group-ID bits of a file will be
749 * cleared upon successful return from chown()
751 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
752 !capable(CAP_FSETID)) {
753 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
757 * Change the ownerships and register quota modifications
758 * in the transaction.
761 if (XFS_IS_UQUOTA_ON(mp)) {
762 ASSERT(mask & XFS_AT_UID);
764 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
765 &ip->i_udquot, udqp);
767 ip->i_d.di_uid = uid;
770 if (XFS_IS_GQUOTA_ON(mp)) {
771 ASSERT(mask & XFS_AT_GID);
773 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
774 &ip->i_gdquot, gdqp);
776 ip->i_d.di_gid = gid;
778 if (iprojid != projid) {
779 ip->i_d.di_projid = projid;
781 * We may have to rev the inode as well as
782 * the superblock version number since projids didn't
783 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
785 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
786 xfs_bump_ino_vers2(tp, ip);
789 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
790 timeflags |= XFS_ICHGTIME_CHG;
795 * Change file access or modified times.
797 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
798 if (mask & XFS_AT_ATIME) {
799 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
800 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
801 ip->i_update_core = 1;
802 timeflags &= ~XFS_ICHGTIME_ACC;
804 if (mask & XFS_AT_MTIME) {
805 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
806 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
807 timeflags &= ~XFS_ICHGTIME_MOD;
808 timeflags |= XFS_ICHGTIME_CHG;
810 if (tp && (flags & ATTR_UTIME))
811 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
815 * Change XFS-added attributes.
817 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
818 if (mask & XFS_AT_EXTSIZE) {
820 * Converting bytes to fs blocks.
822 ip->i_d.di_extsize = vap->va_extsize >>
823 mp->m_sb.sb_blocklog;
825 if (mask & XFS_AT_XFLAGS) {
826 /* can't set PREALLOC this way, just preserve it */
828 (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
829 if (vap->va_xflags & XFS_XFLAG_REALTIME &&
830 (ip->i_d.di_mode & S_IFMT) == S_IFREG) {
831 ip->i_d.di_flags |= XFS_DIFLAG_REALTIME;
832 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
834 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
836 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
837 ip->i_d.di_flags |= XFS_DIFLAG_IMMUTABLE;
838 if (vap->va_xflags & XFS_XFLAG_APPEND)
839 ip->i_d.di_flags |= XFS_DIFLAG_APPEND;
840 if (vap->va_xflags & XFS_XFLAG_SYNC)
841 ip->i_d.di_flags |= XFS_DIFLAG_SYNC;
842 if (vap->va_xflags & XFS_XFLAG_NOATIME)
843 ip->i_d.di_flags |= XFS_DIFLAG_NOATIME;
844 if (vap->va_xflags & XFS_XFLAG_NODUMP)
845 ip->i_d.di_flags |= XFS_DIFLAG_NODUMP;
846 if ((vap->va_xflags & XFS_XFLAG_RTINHERIT) &&
847 (ip->i_d.di_mode & S_IFMT) == S_IFDIR)
848 ip->i_d.di_flags |= XFS_DIFLAG_RTINHERIT;
850 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
851 timeflags |= XFS_ICHGTIME_CHG;
855 * Change file inode change time only if XFS_AT_CTIME set
856 * AND we have been called by a DMI function.
859 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
860 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
861 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
862 ip->i_update_core = 1;
863 timeflags &= ~XFS_ICHGTIME_CHG;
867 * Send out timestamp changes that need to be set to the
868 * current time. Not done when called by a DMI function.
870 if (timeflags && !(flags & ATTR_DMI))
871 xfs_ichgtime(ip, timeflags);
873 XFS_STATS_INC(xs_ig_attrchg);
876 * If this is a synchronous mount, make sure that the
877 * transaction goes to disk before returning to the user.
878 * This is slightly sub-optimal in that truncates require
879 * two sync transactions instead of one for wsync filesytems.
880 * One for the truncate and one for the timestamps since we
881 * don't want to change the timestamps unless we're sure the
882 * truncate worked. Truncates are less than 1% of the laddis
883 * mix so this probably isn't worth the trouble to optimize.
887 if (mp->m_flags & XFS_MOUNT_WSYNC)
888 xfs_trans_set_sync(tp);
890 code = xfs_trans_commit(tp, commit_flags, NULL);
894 * If the (regular) file's mandatory locking mode changed, then
895 * notify the vnode. We do this under the inode lock to prevent
896 * racing calls to vop_vnode_change.
898 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
899 if (mandlock_before != mandlock_after) {
900 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
904 xfs_iunlock(ip, lock_flags);
907 * Release any dquot(s) the inode had kept before chown.
909 XFS_QM_DQRELE(mp, olddquot1);
910 XFS_QM_DQRELE(mp, olddquot2);
911 XFS_QM_DQRELE(mp, udqp);
912 XFS_QM_DQRELE(mp, gdqp);
918 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
919 !(flags & ATTR_DMI)) {
920 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
921 NULL, DM_RIGHT_NULL, NULL, NULL,
922 0, 0, AT_DELAY_FLAG(flags));
927 commit_flags |= XFS_TRANS_ABORT;
930 XFS_QM_DQRELE(mp, udqp);
931 XFS_QM_DQRELE(mp, gdqp);
933 xfs_trans_cancel(tp, commit_flags);
935 if (lock_flags != 0) {
936 xfs_iunlock(ip, lock_flags);
944 * Null conversion from vnode mode bits to inode mode bits, as in efs.
955 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
956 (inst_t *)__return_address);
958 ip = XFS_BHVTOI(bdp);
959 xfs_ilock(ip, XFS_ILOCK_SHARED);
960 error = xfs_iaccess(ip, mode, credp);
961 xfs_iunlock(ip, XFS_ILOCK_SHARED);
985 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
991 vp = BHV_TO_VNODE(bdp);
992 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
994 ip = XFS_BHVTOI(bdp);
997 if (XFS_FORCED_SHUTDOWN(mp))
998 return XFS_ERROR(EIO);
1000 xfs_ilock(ip, XFS_ILOCK_SHARED);
1002 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1004 offset = uiop->uio_offset;
1005 count = uiop->uio_resid;
1008 error = XFS_ERROR(EINVAL);
1016 if (!(ioflags & IO_INVIS)) {
1017 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1021 * See if the symlink is stored inline.
1023 pathlen = (int)ip->i_d.di_size;
1025 if (ip->i_df.if_flags & XFS_IFINLINE) {
1026 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1030 * Symlink not inline. Call bmap to get it in.
1032 nmaps = SYMLINK_MAPS;
1034 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1035 0, NULL, 0, mval, &nmaps, NULL);
1041 for (n = 0; n < nmaps; n++) {
1042 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1043 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1044 bp = xfs_buf_read(mp->m_ddev_targp, d,
1045 BTOBB(byte_cnt), 0);
1046 error = XFS_BUF_GETERROR(bp);
1048 xfs_ioerror_alert("xfs_readlink",
1049 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1053 if (pathlen < byte_cnt)
1055 pathlen -= byte_cnt;
1057 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1066 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1075 * This is called to sync the inode and its data out to disk.
1076 * We need to hold the I/O lock while flushing the data, and
1077 * the inode lock while flushing the inode. The inode lock CANNOT
1078 * be held while flushing the data, so acquire after we're done
1093 vn_trace_entry(BHV_TO_VNODE(bdp),
1094 __FUNCTION__, (inst_t *)__return_address);
1096 ip = XFS_BHVTOI(bdp);
1098 ASSERT(start >= 0 && stop >= -1);
1100 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1101 return XFS_ERROR(EIO);
1104 * We always need to make sure that the required inode state
1105 * is safe on disk. The vnode might be clean but because
1106 * of committed transactions that haven't hit the disk yet.
1107 * Likewise, there could be unflushed non-transactional
1108 * changes to the inode core that have to go to disk.
1110 * The following code depends on one assumption: that
1111 * any transaction that changes an inode logs the core
1112 * because it has to change some field in the inode core
1113 * (typically nextents or nblocks). That assumption
1114 * implies that any transactions against an inode will
1115 * catch any non-transactional updates. If inode-altering
1116 * transactions exist that violate this assumption, the
1117 * code breaks. Right now, it figures that if the involved
1118 * update_* field is clear and the inode is unpinned, the
1119 * inode is clean. Either it's been flushed or it's been
1120 * committed and the commit has hit the disk unpinning the inode.
1121 * (Note that xfs_inode_item_format() called at commit clears
1122 * the update_* fields.)
1124 xfs_ilock(ip, XFS_ILOCK_SHARED);
1126 /* If we are flushing data then we care about update_size
1127 * being set, otherwise we care about update_core
1129 if ((flag & FSYNC_DATA) ?
1130 (ip->i_update_size == 0) :
1131 (ip->i_update_core == 0)) {
1133 * Timestamps/size haven't changed since last inode
1134 * flush or inode transaction commit. That means
1135 * either nothing got written or a transaction
1136 * committed which caught the updates. If the
1137 * latter happened and the transaction hasn't
1138 * hit the disk yet, the inode will be still
1139 * be pinned. If it is, force the log.
1142 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1144 if (xfs_ipincount(ip)) {
1145 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1147 ((flag & FSYNC_WAIT)
1148 ? XFS_LOG_SYNC : 0));
1153 * Kick off a transaction to log the inode
1154 * core to get the updates. Make it
1155 * sync if FSYNC_WAIT is passed in (which
1156 * is done by everybody but specfs). The
1157 * sync transaction will also force the log.
1159 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1160 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1161 if ((error = xfs_trans_reserve(tp, 0,
1162 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1164 xfs_trans_cancel(tp, 0);
1167 xfs_ilock(ip, XFS_ILOCK_EXCL);
1170 * Note - it's possible that we might have pushed
1171 * ourselves out of the way during trans_reserve
1172 * which would flush the inode. But there's no
1173 * guarantee that the inode buffer has actually
1174 * gone out yet (it's delwri). Plus the buffer
1175 * could be pinned anyway if it's part of an
1176 * inode in another recent transaction. So we
1177 * play it safe and fire off the transaction anyway.
1179 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1180 xfs_trans_ihold(tp, ip);
1181 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1182 if (flag & FSYNC_WAIT)
1183 xfs_trans_set_sync(tp);
1184 error = xfs_trans_commit(tp, 0, NULL);
1186 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1192 * This is called by xfs_inactive to free any blocks beyond eof,
1193 * when the link count isn't zero.
1196 xfs_inactive_free_eofblocks(
1202 xfs_fileoff_t end_fsb;
1203 xfs_fileoff_t last_fsb;
1204 xfs_filblks_t map_len;
1206 xfs_bmbt_irec_t imap;
1209 * Figure out if there are any blocks beyond the end
1210 * of the file. If not, then there is nothing to do.
1212 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1213 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1214 map_len = last_fsb - end_fsb;
1219 xfs_ilock(ip, XFS_ILOCK_SHARED);
1220 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1221 NULL, 0, &imap, &nimaps, NULL);
1222 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1224 if (!error && (nimaps != 0) &&
1225 (imap.br_startblock != HOLESTARTBLOCK)) {
1227 * Attach the dquots to the inode up front.
1229 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1233 * There are blocks after the end of file.
1234 * Free them up now by truncating the file to
1237 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1240 * Do the xfs_itruncate_start() call before
1241 * reserving any log space because
1242 * itruncate_start will call into the buffer
1243 * cache and we can't
1244 * do that within a transaction.
1246 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1247 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1250 error = xfs_trans_reserve(tp, 0,
1251 XFS_ITRUNCATE_LOG_RES(mp),
1252 0, XFS_TRANS_PERM_LOG_RES,
1253 XFS_ITRUNCATE_LOG_COUNT);
1255 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1256 xfs_trans_cancel(tp, 0);
1257 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1261 xfs_ilock(ip, XFS_ILOCK_EXCL);
1262 xfs_trans_ijoin(tp, ip,
1265 xfs_trans_ihold(tp, ip);
1267 error = xfs_itruncate_finish(&tp, ip,
1272 * If we get an error at this point we
1273 * simply don't bother truncating the file.
1276 xfs_trans_cancel(tp,
1277 (XFS_TRANS_RELEASE_LOG_RES |
1280 error = xfs_trans_commit(tp,
1281 XFS_TRANS_RELEASE_LOG_RES,
1284 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1290 * Free a symlink that has blocks associated with it.
1293 xfs_inactive_symlink_rmt(
1301 xfs_fsblock_t first_block;
1302 xfs_bmap_free_t free_list;
1305 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1313 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1315 * We're freeing a symlink that has some
1316 * blocks allocated to it. Free the
1317 * blocks here. We know that we've got
1318 * either 1 or 2 extents and that we can
1319 * free them all in one bunmapi call.
1321 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1322 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1323 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1324 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1325 xfs_trans_cancel(tp, 0);
1330 * Lock the inode, fix the size, and join it to the transaction.
1331 * Hold it so in the normal path, we still have it locked for
1332 * the second transaction. In the error paths we need it
1333 * held so the cancel won't rele it, see below.
1335 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1336 size = (int)ip->i_d.di_size;
1337 ip->i_d.di_size = 0;
1338 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1339 xfs_trans_ihold(tp, ip);
1340 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1342 * Find the block(s) so we can inval and unmap them.
1345 XFS_BMAP_INIT(&free_list, &first_block);
1346 nmaps = sizeof(mval) / sizeof(mval[0]);
1347 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1348 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1352 * Invalidate the block(s).
1354 for (i = 0; i < nmaps; i++) {
1355 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1356 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1357 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1358 xfs_trans_binval(tp, bp);
1361 * Unmap the dead block(s) to the free_list.
1363 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1364 &first_block, &free_list, &done)))
1368 * Commit the first transaction. This logs the EFI and the inode.
1370 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1373 * The transaction must have been committed, since there were
1374 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1375 * The new tp has the extent freeing and EFDs.
1379 * The first xact was committed, so add the inode to the new one.
1380 * Mark it dirty so it will be logged and moved forward in the log as
1381 * part of every commit.
1383 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1384 xfs_trans_ihold(tp, ip);
1385 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1387 * Get a new, empty transaction to return to our caller.
1389 ntp = xfs_trans_dup(tp);
1391 * Commit the transaction containing extent freeing and EFD's.
1392 * If we get an error on the commit here or on the reserve below,
1393 * we need to unlock the inode since the new transaction doesn't
1394 * have the inode attached.
1396 error = xfs_trans_commit(tp, 0, NULL);
1399 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1403 * Remove the memory for extent descriptions (just bookkeeping).
1405 if (ip->i_df.if_bytes)
1406 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1407 ASSERT(ip->i_df.if_bytes == 0);
1409 * Put an itruncate log reservation in the new transaction
1412 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1413 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1414 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1418 * Return with the inode locked but not joined to the transaction.
1424 xfs_bmap_cancel(&free_list);
1427 * Have to come here with the inode locked and either
1428 * (held and in the transaction) or (not in the transaction).
1429 * If the inode isn't held then cancel would iput it, but
1430 * that's wrong since this is inactive and the vnode ref
1431 * count is 0 already.
1432 * Cancel won't do anything to the inode if held, but it still
1433 * needs to be locked until the cancel is done, if it was
1434 * joined to the transaction.
1436 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1437 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1444 xfs_inactive_symlink_local(
1450 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1452 * We're freeing a symlink which fit into
1453 * the inode. Just free the memory used
1454 * to hold the old symlink.
1456 error = xfs_trans_reserve(*tpp, 0,
1457 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1458 0, XFS_TRANS_PERM_LOG_RES,
1459 XFS_ITRUNCATE_LOG_COUNT);
1462 xfs_trans_cancel(*tpp, 0);
1466 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1469 * Zero length symlinks _can_ exist.
1471 if (ip->i_df.if_bytes > 0) {
1472 xfs_idata_realloc(ip,
1473 -(ip->i_df.if_bytes),
1475 ASSERT(ip->i_df.if_bytes == 0);
1492 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1495 ASSERT(ip->i_d.di_forkoff != 0);
1496 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1497 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1499 error = xfs_attr_inactive(ip);
1502 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1503 return (error); /* goto out*/
1506 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1507 error = xfs_trans_reserve(tp, 0,
1508 XFS_IFREE_LOG_RES(mp),
1509 0, XFS_TRANS_PERM_LOG_RES,
1510 XFS_INACTIVE_LOG_COUNT);
1512 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1513 xfs_trans_cancel(tp, 0);
1515 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1519 xfs_ilock(ip, XFS_ILOCK_EXCL);
1520 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1521 xfs_trans_ihold(tp, ip);
1522 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1524 ASSERT(ip->i_d.di_anextents == 0);
1539 vp = BHV_TO_VNODE(bdp);
1540 ip = XFS_BHVTOI(bdp);
1542 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1546 /* If this is a read-only mount, don't do this (would generate I/O) */
1547 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1550 #ifdef HAVE_REFCACHE
1551 /* If we are in the NFS reference cache then don't do this now */
1558 if (ip->i_d.di_nlink != 0) {
1559 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1560 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1561 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1562 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1563 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1565 /* Update linux inode block count after free above */
1566 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1567 ip->i_d.di_nblocks + ip->i_delayed_blks);
1577 * This is called when the vnode reference count for the vnode
1578 * goes to zero. If the file has been unlinked, then it must
1579 * now be truncated. Also, we clear all of the read-ahead state
1580 * kept for the inode here since the file is now closed.
1589 xfs_bmap_free_t free_list;
1590 xfs_fsblock_t first_block;
1597 vp = BHV_TO_VNODE(bdp);
1598 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1600 ip = XFS_BHVTOI(bdp);
1603 * If the inode is already free, then there can be nothing
1606 if (ip->i_d.di_mode == 0) {
1607 ASSERT(ip->i_df.if_real_bytes == 0);
1608 ASSERT(ip->i_df.if_broot_bytes == 0);
1609 return VN_INACTIVE_CACHE;
1613 * Only do a truncate if it's a regular file with
1614 * some actual space in it. It's OK to look at the
1615 * inode's fields without the lock because we're the
1616 * only one with a reference to the inode.
1618 truncate = ((ip->i_d.di_nlink == 0) &&
1619 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1620 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1624 if (ip->i_d.di_nlink == 0 &&
1625 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1626 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1631 /* If this is a read-only mount, don't do this (would generate I/O) */
1632 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1635 if (ip->i_d.di_nlink != 0) {
1636 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1637 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1638 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1639 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1640 (ip->i_delayed_blks != 0))) {
1641 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1642 return (VN_INACTIVE_CACHE);
1643 /* Update linux inode block count after free above */
1644 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1645 ip->i_d.di_nblocks + ip->i_delayed_blks);
1650 ASSERT(ip->i_d.di_nlink == 0);
1652 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1653 return (VN_INACTIVE_CACHE);
1655 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1658 * Do the xfs_itruncate_start() call before
1659 * reserving any log space because itruncate_start
1660 * will call into the buffer cache and we can't
1661 * do that within a transaction.
1663 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1665 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1667 error = xfs_trans_reserve(tp, 0,
1668 XFS_ITRUNCATE_LOG_RES(mp),
1669 0, XFS_TRANS_PERM_LOG_RES,
1670 XFS_ITRUNCATE_LOG_COUNT);
1672 /* Don't call itruncate_cleanup */
1673 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1674 xfs_trans_cancel(tp, 0);
1675 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1676 return (VN_INACTIVE_CACHE);
1679 xfs_ilock(ip, XFS_ILOCK_EXCL);
1680 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1681 xfs_trans_ihold(tp, ip);
1684 * normally, we have to run xfs_itruncate_finish sync.
1685 * But if filesystem is wsync and we're in the inactive
1686 * path, then we know that nlink == 0, and that the
1687 * xaction that made nlink == 0 is permanently committed
1688 * since xfs_remove runs as a synchronous transaction.
1690 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1691 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1694 xfs_trans_cancel(tp,
1695 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1696 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1697 return (VN_INACTIVE_CACHE);
1699 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1702 * If we get an error while cleaning up a
1703 * symlink we bail out.
1705 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1706 xfs_inactive_symlink_rmt(ip, &tp) :
1707 xfs_inactive_symlink_local(ip, &tp);
1711 return (VN_INACTIVE_CACHE);
1714 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1715 xfs_trans_ihold(tp, ip);
1717 error = xfs_trans_reserve(tp, 0,
1718 XFS_IFREE_LOG_RES(mp),
1719 0, XFS_TRANS_PERM_LOG_RES,
1720 XFS_INACTIVE_LOG_COUNT);
1722 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1723 xfs_trans_cancel(tp, 0);
1724 return (VN_INACTIVE_CACHE);
1727 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1728 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1729 xfs_trans_ihold(tp, ip);
1733 * If there are attributes associated with the file
1734 * then blow them away now. The code calls a routine
1735 * that recursively deconstructs the attribute fork.
1736 * We need to just commit the current transaction
1737 * because we can't use it for xfs_attr_inactive().
1739 if (ip->i_d.di_anextents > 0) {
1740 error = xfs_inactive_attrs(ip, &tp);
1742 * If we got an error, the transaction is already
1743 * cancelled, and the inode is unlocked. Just get out.
1746 return (VN_INACTIVE_CACHE);
1747 } else if (ip->i_afp) {
1748 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1754 XFS_BMAP_INIT(&free_list, &first_block);
1755 error = xfs_ifree(tp, ip, &free_list);
1758 * If we fail to free the inode, shut down. The cancel
1759 * might do that, we need to make sure. Otherwise the
1760 * inode might be lost for a long time or forever.
1762 if (!XFS_FORCED_SHUTDOWN(mp)) {
1764 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1765 error, mp->m_fsname);
1766 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1768 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1771 * Credit the quota account(s). The inode is gone.
1773 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1776 * Just ignore errors at this point. There is
1777 * nothing we can do except to try to keep going.
1779 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1781 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1784 * Release the dquots held by inode, if any.
1786 XFS_QM_DQDETACH(mp, ip);
1788 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1791 return VN_INACTIVE_CACHE;
1800 bhv_desc_t *dir_bdp,
1807 xfs_inode_t *dp, *ip;
1813 dir_vp = BHV_TO_VNODE(dir_bdp);
1814 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1816 dp = XFS_BHVTOI(dir_bdp);
1818 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1819 return XFS_ERROR(EIO);
1821 lock_mode = xfs_ilock_map_shared(dp);
1822 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1824 *vpp = XFS_ITOV(ip);
1827 xfs_iunlock_map_shared(dp, lock_mode);
1833 * xfs_create (create a new file).
1837 bhv_desc_t *dir_bdp,
1843 char *name = VNAME(dentry);
1845 xfs_inode_t *dp, *ip;
1851 xfs_bmap_free_t free_list;
1852 xfs_fsblock_t first_block;
1853 boolean_t dp_joined_to_trans;
1854 int dm_event_sent = 0;
1858 struct xfs_dquot *udqp, *gdqp;
1864 dir_vp = BHV_TO_VNODE(dir_bdp);
1865 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1867 dp = XFS_BHVTOI(dir_bdp);
1870 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1871 namelen = VNAMELEN(dentry);
1873 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1874 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1875 dir_vp, DM_RIGHT_NULL, NULL,
1876 DM_RIGHT_NULL, name, NULL,
1884 if (XFS_FORCED_SHUTDOWN(mp))
1885 return XFS_ERROR(EIO);
1887 /* Return through std_return after this point. */
1890 if (vap->va_mask & XFS_AT_PROJID)
1891 prid = (xfs_prid_t)vap->va_projid;
1893 prid = (xfs_prid_t)dfltprid;
1896 * Make sure that we have allocated dquot(s) on disk.
1898 error = XFS_QM_DQVOPALLOC(mp, dp,
1899 current_fsuid(credp), current_fsgid(credp),
1900 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1905 dp_joined_to_trans = B_FALSE;
1907 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1908 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1909 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1911 * Initially assume that the file does not exist and
1912 * reserve the resources for that case. If that is not
1913 * the case we'll drop the one we have and get a more
1914 * appropriate transaction later.
1916 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1917 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1918 if (error == ENOSPC) {
1920 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1921 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1929 xfs_ilock(dp, XFS_ILOCK_EXCL);
1931 XFS_BMAP_INIT(&free_list, &first_block);
1936 * Reserve disk quota and the inode.
1938 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1943 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1945 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1946 error = xfs_dir_ialloc(&tp, dp,
1947 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1948 rdev, credp, prid, resblks > 0,
1951 if (error == ENOSPC)
1958 * At this point, we've gotten a newly allocated inode.
1959 * It is locked (and joined to the transaction).
1962 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1965 * Now we join the directory inode to the transaction.
1966 * We do not do it earlier because xfs_dir_ialloc
1967 * might commit the previous transaction (and release
1972 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1973 dp_joined_to_trans = B_TRUE;
1975 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1976 &first_block, &free_list,
1977 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1979 ASSERT(error != ENOSPC);
1982 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1983 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1986 * If this is a synchronous mount, make sure that the
1987 * create transaction goes to disk before returning to
1990 if (mp->m_flags & XFS_MOUNT_WSYNC) {
1991 xfs_trans_set_sync(tp);
1997 * Attach the dquot(s) to the inodes and modify them incore.
1998 * These ids of the inode couldn't have changed since the new
1999 * inode has been locked ever since it was created.
2001 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2004 * xfs_trans_commit normally decrements the vnode ref count
2005 * when it unlocks the inode. Since we want to return the
2006 * vnode to the caller, we bump the vnode ref count now.
2011 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2013 xfs_bmap_cancel(&free_list);
2017 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2024 XFS_QM_DQRELE(mp, udqp);
2025 XFS_QM_DQRELE(mp, gdqp);
2028 * Propogate the fact that the vnode changed after the
2029 * xfs_inode locks have been released.
2031 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2035 /* Fallthrough to std_return with error = 0 */
2038 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2039 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2040 DM_EVENT_POSTCREATE)) {
2041 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2042 dir_vp, DM_RIGHT_NULL,
2044 DM_RIGHT_NULL, name, NULL,
2045 dm_di_mode, error, 0);
2050 cancel_flags |= XFS_TRANS_ABORT;
2055 xfs_trans_cancel(tp, cancel_flags);
2057 if (!dp_joined_to_trans && (dp != NULL))
2058 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2059 XFS_QM_DQRELE(mp, udqp);
2060 XFS_QM_DQRELE(mp, gdqp);
2066 * Wait until after the current transaction is aborted to
2067 * release the inode. This prevents recursive transactions
2068 * and deadlocks from xfs_inactive.
2070 cancel_flags |= XFS_TRANS_ABORT;
2071 xfs_trans_cancel(tp, cancel_flags);
2074 XFS_QM_DQRELE(mp, udqp);
2075 XFS_QM_DQRELE(mp, gdqp);
2082 * Some counters to see if (and how often) we are hitting some deadlock
2083 * prevention code paths.
2087 int xfs_rm_lock_delays;
2088 int xfs_rm_attempts;
2092 * The following routine will lock the inodes associated with the
2093 * directory and the named entry in the directory. The locks are
2094 * acquired in increasing inode number.
2096 * If the entry is "..", then only the directory is locked. The
2097 * vnode ref count will still include that from the .. entry in
2100 * There is a deadlock we need to worry about. If the locked directory is
2101 * in the AIL, it might be blocking up the log. The next inode we lock
2102 * could be already locked by another thread waiting for log space (e.g
2103 * a permanent log reservation with a long running transaction (see
2104 * xfs_itruncate_finish)). To solve this, we must check if the directory
2105 * is in the ail and use lock_nowait. If we can't lock, we need to
2106 * drop the inode lock on the directory and try again. xfs_iunlock will
2107 * potentially push the tail if we were holding up the log.
2110 xfs_lock_dir_and_entry(
2113 xfs_inode_t *ip) /* inode of entry 'name' */
2117 xfs_inode_t *ips[2];
2126 xfs_ilock(dp, XFS_ILOCK_EXCL);
2133 * We want to lock in increasing inum. Since we've already
2134 * acquired the lock on the directory, we may need to release
2135 * if if the inum of the entry turns out to be less.
2137 if (e_inum > dp->i_ino) {
2139 * We are already in the right order, so just
2140 * lock on the inode of the entry.
2141 * We need to use nowait if dp is in the AIL.
2144 lp = (xfs_log_item_t *)dp->i_itemp;
2145 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2146 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2153 * Unlock dp and try again.
2154 * xfs_iunlock will try to push the tail
2155 * if the inode is in the AIL.
2158 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2160 if ((attempts % 5) == 0) {
2161 delay(1); /* Don't just spin the CPU */
2163 xfs_rm_lock_delays++;
2169 xfs_ilock(ip, XFS_ILOCK_EXCL);
2171 } else if (e_inum < dp->i_ino) {
2172 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2176 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2178 /* else e_inum == dp->i_ino */
2179 /* This can happen if we're asked to lock /x/..
2180 * the entry is "..", which is also the parent directory.
2188 int xfs_small_retries;
2189 int xfs_middle_retries;
2190 int xfs_lots_retries;
2191 int xfs_lock_delays;
2195 * The following routine will lock n inodes in exclusive mode.
2196 * We assume the caller calls us with the inodes in i_ino order.
2198 * We need to detect deadlock where an inode that we lock
2199 * is in the AIL and we start waiting for another inode that is locked
2200 * by a thread in a long running transaction (such as truncate). This can
2201 * result in deadlock since the long running trans might need to wait
2202 * for the inode we just locked in order to push the tail and free space
2212 int attempts = 0, i, j, try_lock;
2215 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2226 for (; i < inodes; i++) {
2229 if (i && (ips[i] == ips[i-1])) /* Already locked */
2233 * If try_lock is not set yet, make sure all locked inodes
2234 * are not in the AIL.
2235 * If any are, set try_lock to be used later.
2239 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2240 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2241 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2248 * If any of the previous locks we have locked is in the AIL,
2249 * we must TRY to get the second and subsequent locks. If
2250 * we can't get any, we must release all we have
2255 /* try_lock must be 0 if i is 0. */
2257 * try_lock means we have an inode locked
2258 * that is in the AIL.
2261 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2265 * Unlock all previous guys and try again.
2266 * xfs_iunlock will try to push the tail
2267 * if the inode is in the AIL.
2270 for(j = i - 1; j >= 0; j--) {
2273 * Check to see if we've already
2274 * unlocked this one.
2275 * Not the first one going back,
2276 * and the inode ptr is the same.
2278 if ((j != (i - 1)) && ips[j] ==
2282 xfs_iunlock(ips[j], lock_mode);
2285 if ((attempts % 5) == 0) {
2286 delay(1); /* Don't just spin the CPU */
2296 xfs_ilock(ips[i], lock_mode);
2302 if (attempts < 5) xfs_small_retries++;
2303 else if (attempts < 100) xfs_middle_retries++;
2304 else xfs_lots_retries++;
2312 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2313 int remove_which_error_return = 0;
2315 #define REMOVE_DEBUG_TRACE(x)
2316 #endif /* ! DEBUG */
2325 bhv_desc_t *dir_bdp,
2330 char *name = VNAME(dentry);
2331 xfs_inode_t *dp, *ip;
2332 xfs_trans_t *tp = NULL;
2335 xfs_bmap_free_t free_list;
2336 xfs_fsblock_t first_block;
2344 dir_vp = BHV_TO_VNODE(dir_bdp);
2345 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2347 dp = XFS_BHVTOI(dir_bdp);
2350 if (XFS_FORCED_SHUTDOWN(mp))
2351 return XFS_ERROR(EIO);
2353 namelen = VNAMELEN(dentry);
2355 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2356 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2357 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2358 name, NULL, 0, 0, 0);
2363 /* From this point on, return through std_return */
2367 * We need to get a reference to ip before we get our log
2368 * reservation. The reason for this is that we cannot call
2369 * xfs_iget for an inode for which we do not have a reference
2370 * once we've acquired a log reservation. This is because the
2371 * inode we are trying to get might be in xfs_inactive going
2372 * for a log reservation. Since we'll have to wait for the
2373 * inactive code to complete before returning from xfs_iget,
2374 * we need to make sure that we don't have log space reserved
2375 * when we call xfs_iget. Instead we get an unlocked referece
2376 * to the inode before getting our log reservation.
2378 error = xfs_get_dir_entry(dentry, &ip);
2380 REMOVE_DEBUG_TRACE(__LINE__);
2384 dm_di_mode = ip->i_d.di_mode;
2386 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2390 error = XFS_QM_DQATTACH(mp, dp, 0);
2391 if (!error && dp != ip)
2392 error = XFS_QM_DQATTACH(mp, ip, 0);
2394 REMOVE_DEBUG_TRACE(__LINE__);
2399 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2400 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2402 * We try to get the real space reservation first,
2403 * allowing for directory btree deletion(s) implying
2404 * possible bmap insert(s). If we can't get the space
2405 * reservation then we use 0 instead, and avoid the bmap
2406 * btree insert(s) in the directory code by, if the bmap
2407 * insert tries to happen, instead trimming the LAST
2408 * block from the directory.
2410 resblks = XFS_REMOVE_SPACE_RES(mp);
2411 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2412 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2413 if (error == ENOSPC) {
2415 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2416 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2419 ASSERT(error != ENOSPC);
2420 REMOVE_DEBUG_TRACE(__LINE__);
2421 xfs_trans_cancel(tp, 0);
2426 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2428 REMOVE_DEBUG_TRACE(__LINE__);
2429 xfs_trans_cancel(tp, cancel_flags);
2435 * At this point, we've gotten both the directory and the entry
2438 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2441 * Increment vnode ref count only in this case since
2442 * there's an extra vnode reference in the case where
2446 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2450 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2452 XFS_BMAP_INIT(&free_list, &first_block);
2453 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2454 &first_block, &free_list, 0);
2456 ASSERT(error != ENOENT);
2457 REMOVE_DEBUG_TRACE(__LINE__);
2460 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2463 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2465 error = xfs_droplink(tp, ip);
2467 REMOVE_DEBUG_TRACE(__LINE__);
2471 /* Determine if this is the last link while
2472 * we are in the transaction.
2474 link_zero = (ip)->i_d.di_nlink==0;
2477 * Take an extra ref on the inode so that it doesn't
2478 * go to xfs_inactive() from within the commit.
2483 * If this is a synchronous mount, make sure that the
2484 * remove transaction goes to disk before returning to
2487 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2488 xfs_trans_set_sync(tp);
2491 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2493 REMOVE_DEBUG_TRACE(__LINE__);
2497 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2504 * Before we drop our extra reference to the inode, purge it
2505 * from the refcache if it is there. By waiting until afterwards
2506 * to do the IRELE, we ensure that we won't go inactive in the
2507 * xfs_refcache_purge_ip routine (although that would be OK).
2509 xfs_refcache_purge_ip(ip);
2511 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2514 * Let interposed file systems know about removed links.
2516 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2520 /* Fall through to std_return with error = 0 */
2522 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2523 DM_EVENT_POSTREMOVE)) {
2524 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2525 dir_vp, DM_RIGHT_NULL,
2526 NULL, DM_RIGHT_NULL,
2527 name, NULL, dm_di_mode, error, 0);
2532 xfs_bmap_cancel(&free_list);
2533 cancel_flags |= XFS_TRANS_ABORT;
2534 xfs_trans_cancel(tp, cancel_flags);
2539 * In this case make sure to not release the inode until after
2540 * the current transaction is aborted. Releasing it beforehand
2541 * can cause us to go to xfs_inactive and start a recursive
2542 * transaction which can easily deadlock with the current one.
2544 xfs_bmap_cancel(&free_list);
2545 cancel_flags |= XFS_TRANS_ABORT;
2546 xfs_trans_cancel(tp, cancel_flags);
2549 * Before we drop our extra reference to the inode, purge it
2550 * from the refcache if it is there. By waiting until afterwards
2551 * to do the IRELE, we ensure that we won't go inactive in the
2552 * xfs_refcache_purge_ip routine (although that would be OK).
2554 xfs_refcache_purge_ip(ip);
2568 bhv_desc_t *target_dir_bdp,
2573 xfs_inode_t *tdp, *sip;
2576 xfs_inode_t *ips[2];
2578 xfs_bmap_free_t free_list;
2579 xfs_fsblock_t first_block;
2582 vnode_t *target_dir_vp;
2583 bhv_desc_t *src_bdp;
2585 char *target_name = VNAME(dentry);
2588 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2589 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2590 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2592 target_namelen = VNAMELEN(dentry);
2593 if (src_vp->v_type == VDIR)
2594 return XFS_ERROR(EPERM);
2597 * For now, manually find the XFS behavior descriptor for
2598 * the source vnode. If it doesn't exist then something
2599 * is wrong and we should just return an error.
2600 * Eventually we need to figure out how link is going to
2601 * work in the face of stacked vnodes.
2603 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2604 if (src_bdp == NULL) {
2605 return XFS_ERROR(EXDEV);
2607 sip = XFS_BHVTOI(src_bdp);
2608 tdp = XFS_BHVTOI(target_dir_bdp);
2610 if (XFS_FORCED_SHUTDOWN(mp))
2611 return XFS_ERROR(EIO);
2613 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2614 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2615 target_dir_vp, DM_RIGHT_NULL,
2616 src_vp, DM_RIGHT_NULL,
2617 target_name, NULL, 0, 0, 0);
2622 /* Return through std_return after this point. */
2624 error = XFS_QM_DQATTACH(mp, sip, 0);
2625 if (!error && sip != tdp)
2626 error = XFS_QM_DQATTACH(mp, tdp, 0);
2630 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2631 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2632 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2633 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2634 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2635 if (error == ENOSPC) {
2637 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2638 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2645 if (sip->i_ino < tdp->i_ino) {
2653 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2656 * Increment vnode ref counts since xfs_trans_commit &
2657 * xfs_trans_cancel will both unlock the inodes and
2658 * decrement the associated ref counts.
2661 VN_HOLD(target_dir_vp);
2662 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2663 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2666 * If the source has too many links, we can't make any more to it.
2668 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2669 error = XFS_ERROR(EMLINK);
2674 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2678 XFS_BMAP_INIT(&free_list, &first_block);
2680 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2681 sip->i_ino, &first_block, &free_list,
2685 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2687 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2689 error = xfs_bumplink(tp, sip);
2695 * If this is a synchronous mount, make sure that the
2696 * link transaction goes to disk before returning to
2699 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2700 xfs_trans_set_sync(tp);
2703 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2705 xfs_bmap_cancel(&free_list);
2709 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2714 /* Fall through to std_return with error = 0. */
2716 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2717 DM_EVENT_POSTLINK)) {
2718 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2719 target_dir_vp, DM_RIGHT_NULL,
2720 src_vp, DM_RIGHT_NULL,
2721 target_name, NULL, 0, error, 0);
2726 cancel_flags |= XFS_TRANS_ABORT;
2729 xfs_trans_cancel(tp, cancel_flags);
2739 bhv_desc_t *dir_bdp,
2745 char *dir_name = VNAME(dentry);
2747 xfs_inode_t *cdp; /* inode of created dir */
2748 vnode_t *cvp; /* vnode of created dir */
2754 xfs_bmap_free_t free_list;
2755 xfs_fsblock_t first_block;
2757 boolean_t dp_joined_to_trans;
2758 boolean_t created = B_FALSE;
2759 int dm_event_sent = 0;
2761 struct xfs_dquot *udqp, *gdqp;
2766 dir_vp = BHV_TO_VNODE(dir_bdp);
2767 dp = XFS_BHVTOI(dir_bdp);
2770 if (XFS_FORCED_SHUTDOWN(mp))
2771 return XFS_ERROR(EIO);
2773 dir_namelen = VNAMELEN(dentry);
2776 dp_joined_to_trans = B_FALSE;
2777 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2779 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2780 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2781 dir_vp, DM_RIGHT_NULL, NULL,
2782 DM_RIGHT_NULL, dir_name, NULL,
2789 /* Return through std_return after this point. */
2791 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2795 if (vap->va_mask & XFS_AT_PROJID)
2796 prid = (xfs_prid_t)vap->va_projid;
2798 prid = (xfs_prid_t)dfltprid;
2801 * Make sure that we have allocated dquot(s) on disk.
2803 error = XFS_QM_DQVOPALLOC(mp, dp,
2804 current_fsuid(credp), current_fsgid(credp),
2805 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2809 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2810 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2811 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2812 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2813 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2814 if (error == ENOSPC) {
2816 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2817 XFS_TRANS_PERM_LOG_RES,
2818 XFS_MKDIR_LOG_COUNT);
2826 xfs_ilock(dp, XFS_ILOCK_EXCL);
2829 * Check for directory link count overflow.
2831 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2832 error = XFS_ERROR(EMLINK);
2837 * Reserve disk quota and the inode.
2839 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2844 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2847 * create the directory inode.
2849 error = xfs_dir_ialloc(&tp, dp,
2850 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2851 0, credp, prid, resblks > 0,
2854 if (error == ENOSPC)
2861 * Now we add the directory inode to the transaction.
2862 * We waited until now since xfs_dir_ialloc might start
2863 * a new transaction. Had we joined the transaction
2864 * earlier, the locks might have gotten released.
2867 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2868 dp_joined_to_trans = B_TRUE;
2870 XFS_BMAP_INIT(&free_list, &first_block);
2872 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2873 cdp->i_ino, &first_block, &free_list,
2874 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2876 ASSERT(error != ENOSPC);
2879 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2882 * Bump the in memory version number of the parent directory
2883 * so that other processes accessing it will recognize that
2884 * the directory has changed.
2888 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2894 error = xfs_bumplink(tp, dp);
2899 cvp = XFS_ITOV(cdp);
2907 * Attach the dquots to the new inode and modify the icount incore.
2909 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2912 * If this is a synchronous mount, make sure that the
2913 * mkdir transaction goes to disk before returning to
2916 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2917 xfs_trans_set_sync(tp);
2920 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2926 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2927 XFS_QM_DQRELE(mp, udqp);
2928 XFS_QM_DQRELE(mp, gdqp);
2933 /* Fall through to std_return with error = 0 or errno from
2934 * xfs_trans_commit. */
2937 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2938 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2939 DM_EVENT_POSTCREATE)) {
2940 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2941 dir_vp, DM_RIGHT_NULL,
2942 created ? XFS_ITOV(cdp):NULL,
2945 dm_di_mode, error, 0);
2951 xfs_bmap_cancel(&free_list);
2953 cancel_flags |= XFS_TRANS_ABORT;
2955 xfs_trans_cancel(tp, cancel_flags);
2956 XFS_QM_DQRELE(mp, udqp);
2957 XFS_QM_DQRELE(mp, gdqp);
2959 if (!dp_joined_to_trans && (dp != NULL)) {
2960 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2973 bhv_desc_t *dir_bdp,
2977 char *name = VNAME(dentry);
2979 xfs_inode_t *cdp; /* child directory */
2983 xfs_bmap_free_t free_list;
2984 xfs_fsblock_t first_block;
2993 dir_vp = BHV_TO_VNODE(dir_bdp);
2994 dp = XFS_BHVTOI(dir_bdp);
2997 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2999 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3000 return XFS_ERROR(EIO);
3001 namelen = VNAMELEN(dentry);
3003 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3004 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3005 dir_vp, DM_RIGHT_NULL,
3006 NULL, DM_RIGHT_NULL,
3007 name, NULL, 0, 0, 0);
3009 return XFS_ERROR(error);
3012 /* Return through std_return after this point. */
3017 * We need to get a reference to cdp before we get our log
3018 * reservation. The reason for this is that we cannot call
3019 * xfs_iget for an inode for which we do not have a reference
3020 * once we've acquired a log reservation. This is because the
3021 * inode we are trying to get might be in xfs_inactive going
3022 * for a log reservation. Since we'll have to wait for the
3023 * inactive code to complete before returning from xfs_iget,
3024 * we need to make sure that we don't have log space reserved
3025 * when we call xfs_iget. Instead we get an unlocked referece
3026 * to the inode before getting our log reservation.
3028 error = xfs_get_dir_entry(dentry, &cdp);
3030 REMOVE_DEBUG_TRACE(__LINE__);
3034 dm_di_mode = cdp->i_d.di_mode;
3037 * Get the dquots for the inodes.
3039 error = XFS_QM_DQATTACH(mp, dp, 0);
3040 if (!error && dp != cdp)
3041 error = XFS_QM_DQATTACH(mp, cdp, 0);
3044 REMOVE_DEBUG_TRACE(__LINE__);
3048 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3049 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3051 * We try to get the real space reservation first,
3052 * allowing for directory btree deletion(s) implying
3053 * possible bmap insert(s). If we can't get the space
3054 * reservation then we use 0 instead, and avoid the bmap
3055 * btree insert(s) in the directory code by, if the bmap
3056 * insert tries to happen, instead trimming the LAST
3057 * block from the directory.
3059 resblks = XFS_REMOVE_SPACE_RES(mp);
3060 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3061 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3062 if (error == ENOSPC) {
3064 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3065 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3068 ASSERT(error != ENOSPC);
3073 XFS_BMAP_INIT(&free_list, &first_block);
3076 * Now lock the child directory inode and the parent directory
3077 * inode in the proper order. This will take care of validating
3078 * that the directory entry for the child directory inode has
3079 * not changed while we were obtaining a log reservation.
3081 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3083 xfs_trans_cancel(tp, cancel_flags);
3088 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3091 * Only increment the parent directory vnode count if
3092 * we didn't bump it in looking up cdp. The only time
3093 * we don't bump it is when we're looking up ".".
3099 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3101 ASSERT(cdp->i_d.di_nlink >= 2);
3102 if (cdp->i_d.di_nlink != 2) {
3103 error = XFS_ERROR(ENOTEMPTY);
3106 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3107 error = XFS_ERROR(ENOTEMPTY);
3111 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3112 &first_block, &free_list, resblks);
3117 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3120 * Bump the in memory generation count on the parent
3121 * directory so that other can know that it has changed.
3126 * Drop the link from cdp's "..".
3128 error = xfs_droplink(tp, dp);
3134 * Drop the link from dp to cdp.
3136 error = xfs_droplink(tp, cdp);
3142 * Drop the "." link from cdp to self.
3144 error = xfs_droplink(tp, cdp);
3149 /* Determine these before committing transaction */
3150 last_cdp_link = (cdp)->i_d.di_nlink==0;
3153 * Take an extra ref on the child vnode so that it
3154 * does not go to xfs_inactive() from within the commit.
3159 * If this is a synchronous mount, make sure that the
3160 * rmdir transaction goes to disk before returning to
3163 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3164 xfs_trans_set_sync(tp);
3167 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3169 xfs_bmap_cancel(&free_list);
3170 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3176 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3184 * Let interposed file systems know about removed links.
3186 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3190 /* Fall through to std_return with error = 0 or the errno
3191 * from xfs_trans_commit. */
3193 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3194 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3195 dir_vp, DM_RIGHT_NULL,
3196 NULL, DM_RIGHT_NULL,
3197 name, NULL, dm_di_mode,
3203 xfs_bmap_cancel(&free_list);
3204 cancel_flags |= XFS_TRANS_ABORT;
3206 xfs_trans_cancel(tp, cancel_flags);
3214 * Read dp's entries starting at uiop->uio_offset and translate them into
3215 * bufsize bytes worth of struct dirents starting at bufbase.
3219 bhv_desc_t *dir_bdp,
3225 xfs_trans_t *tp = NULL;
3228 xfs_off_t start_offset;
3230 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3231 (inst_t *)__return_address);
3232 dp = XFS_BHVTOI(dir_bdp);
3234 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3235 return XFS_ERROR(EIO);
3238 lock_mode = xfs_ilock_map_shared(dp);
3239 start_offset = uiop->uio_offset;
3240 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3241 if (start_offset != uiop->uio_offset) {
3242 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3244 xfs_iunlock_map_shared(dp, lock_mode);
3255 bhv_desc_t *dir_bdp,
3268 xfs_bmap_free_t free_list;
3269 xfs_fsblock_t first_block;
3270 boolean_t dp_joined_to_trans;
3274 xfs_fileoff_t first_fsb;
3275 xfs_filblks_t fs_blocks;
3277 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3284 struct xfs_dquot *udqp, *gdqp;
3286 char *link_name = VNAME(dentry);
3290 dir_vp = BHV_TO_VNODE(dir_bdp);
3291 dp = XFS_BHVTOI(dir_bdp);
3292 dp_joined_to_trans = B_FALSE;
3297 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3301 if (XFS_FORCED_SHUTDOWN(mp))
3302 return XFS_ERROR(EIO);
3304 link_namelen = VNAMELEN(dentry);
3307 * Check component lengths of the target path name.
3309 pathlen = strlen(target_path);
3310 if (pathlen >= MAXPATHLEN) /* total string too long */
3311 return XFS_ERROR(ENAMETOOLONG);
3312 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3316 for(total = 0, path = target_path; total < pathlen;) {
3320 while(*path == '/') {
3326 * Count up to the next slash or end of path.
3327 * Error out if the component is bigger than MAXNAMELEN.
3329 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3330 if (++len >= MAXNAMELEN) {
3331 error = ENAMETOOLONG;
3338 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3339 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3340 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3341 link_name, target_path, 0, 0, 0);
3346 /* Return through std_return after this point. */
3349 if (vap->va_mask & XFS_AT_PROJID)
3350 prid = (xfs_prid_t)vap->va_projid;
3352 prid = (xfs_prid_t)dfltprid;
3355 * Make sure that we have allocated dquot(s) on disk.
3357 error = XFS_QM_DQVOPALLOC(mp, dp,
3358 current_fsuid(credp), current_fsgid(credp),
3359 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3363 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3364 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3366 * The symlink will fit into the inode data fork?
3367 * There can't be any attributes so we get the whole variable part.
3369 if (pathlen <= XFS_LITINO(mp))
3372 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3373 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3374 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3375 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3376 if (error == ENOSPC && fs_blocks == 0) {
3378 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3379 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3387 xfs_ilock(dp, XFS_ILOCK_EXCL);
3390 * Reserve disk quota : blocks and inode.
3392 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3397 * Check for ability to enter directory entry, if no space reserved.
3400 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3403 * Initialize the bmap freelist prior to calling either
3404 * bmapi or the directory create code.
3406 XFS_BMAP_INIT(&free_list, &first_block);
3409 * Allocate an inode for the symlink.
3411 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3412 1, 0, credp, prid, resblks > 0, &ip, NULL);
3414 if (error == ENOSPC)
3421 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3422 dp_joined_to_trans = B_TRUE;
3425 * Also attach the dquot(s) to it, if applicable.
3427 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3430 resblks -= XFS_IALLOC_SPACE_RES(mp);
3432 * If the symlink will fit into the inode, write it inline.
3434 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3435 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3436 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3437 ip->i_d.di_size = pathlen;
3440 * The inode was initially created in extent format.
3442 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3443 ip->i_df.if_flags |= XFS_IFINLINE;
3445 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3446 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3450 nmaps = SYMLINK_MAPS;
3452 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3453 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3454 &first_block, resblks, mval, &nmaps,
3461 resblks -= fs_blocks;
3462 ip->i_d.di_size = pathlen;
3463 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3465 cur_chunk = target_path;
3466 for (n = 0; n < nmaps; n++) {
3467 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3468 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3469 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3470 BTOBB(byte_cnt), 0);
3471 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3472 if (pathlen < byte_cnt) {
3475 pathlen -= byte_cnt;
3477 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3478 cur_chunk += byte_cnt;
3480 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3485 * Create the directory entry for the symlink.
3487 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3488 ip->i_ino, &first_block, &free_list, resblks);
3492 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3493 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3496 * Bump the in memory version number of the parent directory
3497 * so that other processes accessing it will recognize that
3498 * the directory has changed.
3503 * If this is a synchronous mount, make sure that the
3504 * symlink transaction goes to disk before returning to
3507 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3508 xfs_trans_set_sync(tp);
3512 * xfs_trans_commit normally decrements the vnode ref count
3513 * when it unlocks the inode. Since we want to return the
3514 * vnode to the caller, we bump the vnode ref count now.
3518 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3522 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3523 XFS_QM_DQRELE(mp, udqp);
3524 XFS_QM_DQRELE(mp, gdqp);
3526 /* Fall through to std_return with error = 0 or errno from
3527 * xfs_trans_commit */
3529 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3530 DM_EVENT_POSTSYMLINK)) {
3531 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3532 dir_vp, DM_RIGHT_NULL,
3533 error ? NULL : XFS_ITOV(ip),
3534 DM_RIGHT_NULL, link_name, target_path,
3550 xfs_bmap_cancel(&free_list);
3551 cancel_flags |= XFS_TRANS_ABORT;
3553 xfs_trans_cancel(tp, cancel_flags);
3554 XFS_QM_DQRELE(mp, udqp);
3555 XFS_QM_DQRELE(mp, gdqp);
3557 if (!dp_joined_to_trans && (dp != NULL)) {
3558 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3568 * A fid routine that takes a pointer to a previously allocated
3569 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3579 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3580 (inst_t *)__return_address);
3581 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3583 xfid = (xfs_fid2_t *)fidp;
3584 ip = XFS_BHVTOI(bdp);
3585 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3588 * use memcpy because the inode is a long long and there's no
3589 * assurance that xfid->fid_ino is properly aligned.
3591 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3592 xfid->fid_gen = ip->i_d.di_gen;
3609 vp = BHV_TO_VNODE(bdp);
3610 if (vp->v_type == VDIR)
3612 ip = XFS_BHVTOI(bdp);
3613 if (locktype == VRWLOCK_WRITE) {
3614 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3615 } else if (locktype == VRWLOCK_TRY_READ) {
3616 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3617 } else if (locktype == VRWLOCK_TRY_WRITE) {
3618 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3620 ASSERT((locktype == VRWLOCK_READ) ||
3621 (locktype == VRWLOCK_WRITE_DIRECT));
3622 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3640 vp = BHV_TO_VNODE(bdp);
3641 if (vp->v_type == VDIR)
3643 ip = XFS_BHVTOI(bdp);
3644 if (locktype == VRWLOCK_WRITE) {
3646 * In the write case, we may have added a new entry to
3647 * the reference cache. This might store a pointer to
3648 * an inode to be released in this inode. If it is there,
3649 * clear the pointer and release the inode after unlocking
3652 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3654 ASSERT((locktype == VRWLOCK_READ) ||
3655 (locktype == VRWLOCK_WRITE_DIRECT));
3656 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3670 ip = XFS_BHVTOI(bdp);
3673 if (XFS_FORCED_SHUTDOWN(mp))
3674 return XFS_ERROR(EIO);
3676 /* Bypass inodes which have already been cleaned by
3677 * the inode flush clustering code inside xfs_iflush
3679 if ((ip->i_update_core == 0) &&
3680 ((ip->i_itemp == NULL) ||
3681 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)))
3684 if (flags & FLUSH_LOG) {
3685 xfs_inode_log_item_t *iip = ip->i_itemp;
3687 if (iip && iip->ili_last_lsn) {
3688 xlog_t *log = mp->m_log;
3690 int s, log_flags = XFS_LOG_FORCE;
3692 s = GRANT_LOCK(log);
3693 sync_lsn = log->l_last_sync_lsn;
3694 GRANT_UNLOCK(log, s);
3696 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3699 if (flags & FLUSH_SYNC)
3700 log_flags |= XFS_LOG_SYNC;
3701 return xfs_log_force(mp, iip->ili_last_lsn,
3706 /* We make this non-blocking if the inode is contended,
3707 * return EAGAIN to indicate to the caller that they
3708 * did not succeed. This prevents the flush path from
3709 * blocking on inodes inside another operation right
3710 * now, they get caught later by xfs_sync.
3712 if (flags & FLUSH_INODE) {
3715 if (xfs_ipincount(ip))
3718 if (flags & FLUSH_SYNC) {
3719 xfs_ilock(ip, XFS_ILOCK_SHARED);
3721 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3722 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3723 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3730 if (flags & FLUSH_SYNC)
3731 flush_flags = XFS_IFLUSH_SYNC;
3733 flush_flags = XFS_IFLUSH_ASYNC;
3735 error = xfs_iflush(ip, flush_flags);
3736 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3755 if (!capable(CAP_SYS_ADMIN))
3756 return XFS_ERROR(EPERM);
3758 ip = XFS_BHVTOI(bdp);
3761 if (XFS_FORCED_SHUTDOWN(mp))
3762 return XFS_ERROR(EIO);
3764 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3765 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3767 xfs_trans_cancel(tp, 0);
3770 xfs_ilock(ip, XFS_ILOCK_EXCL);
3771 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3773 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3774 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3776 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3778 error = xfs_trans_commit(tp, 0, NULL);
3794 vp = BHV_TO_VNODE(bdp);
3796 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3798 ASSERT(!VN_MAPPED(vp));
3799 ip = XFS_BHVTOI(bdp);
3801 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3802 if (ip->i_d.di_size > 0) {
3804 * Flush and invalidate any data left around that is
3805 * a part of this file.
3807 * Get the inode's i/o lock so that buffers are pushed
3808 * out while holding the proper lock. We can't hold
3809 * the inode lock here since flushing out buffers may
3810 * cause us to try to get the lock in xfs_strategy().
3812 * We don't have to call remapf() here, because there
3813 * cannot be any mapped file references to this vnode
3814 * since it is being reclaimed.
3816 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3819 * If we hit an IO error, we need to make sure that the
3820 * buffer and page caches of file data for
3821 * the file are tossed away. We don't want to use
3822 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3823 * pages to stay attached to the vnode, but be
3824 * marked P_BAD. pdflush/vnode_pagebad
3827 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3828 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3830 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3833 ASSERT(VN_CACHED(vp) == 0);
3834 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3835 ip->i_delayed_blks == 0);
3836 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3837 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3839 * di_size field may not be quite accurate if we're
3842 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3843 ASSERT(VN_CACHED(vp) == 0);
3847 /* If we have nothing to flush with this inode then complete the
3848 * teardown now, otherwise break the link between the xfs inode
3849 * and the linux inode and clean up the xfs inode later. This
3850 * avoids flushing the inode to disk during the delete operation
3853 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3854 xfs_ilock(ip, XFS_ILOCK_EXCL);
3856 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3858 xfs_mount_t *mp = ip->i_mount;
3860 /* Protect sync from us */
3861 XFS_MOUNT_ILOCK(mp);
3862 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3863 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3864 ip->i_flags |= XFS_IRECLAIMABLE;
3865 XFS_MOUNT_IUNLOCK(mp);
3876 xfs_ihash_t *ih = ip->i_hash;
3879 /* The hash lock here protects a thread in xfs_iget_core from
3880 * racing with us on linking the inode back with a vnode.
3881 * Once we have the XFS_IRECLAIM flag set it will not touch
3884 write_lock(&ih->ih_lock);
3885 if ((ip->i_flags & XFS_IRECLAIM) ||
3886 (!(ip->i_flags & XFS_IRECLAIMABLE) &&
3887 (XFS_ITOV_NULL(ip) == NULL))) {
3888 write_unlock(&ih->ih_lock);
3891 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3895 ip->i_flags |= XFS_IRECLAIM;
3896 write_unlock(&ih->ih_lock);
3899 * If the inode is still dirty, then flush it out. If the inode
3900 * is not in the AIL, then it will be OK to flush it delwri as
3901 * long as xfs_iflush() does not keep any references to the inode.
3902 * We leave that decision up to xfs_iflush() since it has the
3903 * knowledge of whether it's OK to simply do a delwri flush of
3904 * the inode or whether we need to wait until the inode is
3905 * pulled from the AIL.
3906 * We get the flush lock regardless, though, just to make sure
3907 * we don't free it while it is being flushed.
3909 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3911 xfs_ilock(ip, XFS_ILOCK_EXCL);
3915 if (ip->i_update_core ||
3916 ((ip->i_itemp != NULL) &&
3917 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3918 error = xfs_iflush(ip, sync_mode);
3920 * If we hit an error, typically because of filesystem
3921 * shutdown, we don't need to let vn_reclaim to know
3922 * because we're gonna reclaim the inode anyway.
3925 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3929 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3932 ASSERT(ip->i_update_core == 0);
3933 ASSERT(ip->i_itemp == NULL ||
3934 ip->i_itemp->ili_format.ilf_fields == 0);
3935 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3936 } else if (locked) {
3938 * We are not interested in doing an iflush if we're
3939 * in the process of shutting down the filesystem forcibly.
3940 * So, just reclaim the inode.
3943 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3951 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3954 struct list_head *curr, *next;
3960 XFS_MOUNT_ILOCK(mp);
3961 list_for_each_safe(curr, next, &mp->m_del_inodes) {
3962 ip = list_entry(curr, xfs_inode_t, i_reclaim);
3964 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3966 if (xfs_ipincount(ip) ||
3967 !xfs_iflock_nowait(ip)) {
3968 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3972 XFS_MOUNT_IUNLOCK(mp);
3973 xfs_finish_reclaim(ip, noblock,
3974 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
3982 XFS_MOUNT_IUNLOCK(mp);
3987 * xfs_alloc_file_space()
3988 * This routine allocates disk space for the given file.
3990 * If alloc_type == 0, this request is for an ALLOCSP type
3991 * request which will change the file size. In this case, no
3992 * DMAPI event will be generated by the call. A TRUNCATE event
3993 * will be generated later by xfs_setattr.
3995 * If alloc_type != 0, this request is for a RESVSP type
3996 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3997 * lower block boundary byte address is less than the file's
4006 xfs_alloc_file_space(
4013 xfs_filblks_t allocated_fsb;
4014 xfs_filblks_t allocatesize_fsb;
4017 xfs_filblks_t datablocks;
4019 xfs_fsblock_t firstfsb;
4020 xfs_bmap_free_t free_list;
4021 xfs_bmbt_irec_t *imapp;
4022 xfs_bmbt_irec_t imaps[1];
4029 xfs_fileoff_t startoffset_fsb;
4031 int xfs_bmapi_flags;
4033 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4036 if (XFS_FORCED_SHUTDOWN(mp))
4037 return XFS_ERROR(EIO);
4040 * determine if this is a realtime file
4042 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4043 if (ip->i_d.di_extsize)
4044 rtextsize = ip->i_d.di_extsize;
4046 rtextsize = mp->m_sb.sb_rextsize;
4050 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4054 return XFS_ERROR(EINVAL);
4060 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4061 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4062 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4064 /* Generate a DMAPI event if needed. */
4065 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4066 (attr_flags&ATTR_DMI) == 0 &&
4067 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4068 xfs_off_t end_dmi_offset;
4070 end_dmi_offset = offset+len;
4071 if (end_dmi_offset > ip->i_d.di_size)
4072 end_dmi_offset = ip->i_d.di_size;
4073 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4074 offset, end_dmi_offset - offset,
4081 * allocate file space until done or until there is an error
4084 while (allocatesize_fsb && !error) {
4086 * determine if reserving space on
4087 * the data or realtime partition.
4092 s = startoffset_fsb;
4093 do_div(s, rtextsize);
4095 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4097 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4100 datablocks = allocatesize_fsb;
4105 * allocate and setup the transaction
4107 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4108 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4109 error = xfs_trans_reserve(tp,
4111 XFS_WRITE_LOG_RES(mp),
4113 XFS_TRANS_PERM_LOG_RES,
4114 XFS_WRITE_LOG_COUNT);
4117 * check for running out of space
4121 * Free the transaction structure.
4123 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4124 xfs_trans_cancel(tp, 0);
4127 xfs_ilock(ip, XFS_ILOCK_EXCL);
4128 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4129 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4130 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4134 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4135 xfs_trans_ihold(tp, ip);
4138 * issue the bmapi() call to allocate the blocks
4140 XFS_BMAP_INIT(&free_list, &firstfsb);
4141 error = xfs_bmapi(tp, ip, startoffset_fsb,
4142 allocatesize_fsb, xfs_bmapi_flags,
4143 &firstfsb, 0, imapp, &reccount,
4150 * complete the transaction
4152 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4157 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4158 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4163 allocated_fsb = imapp->br_blockcount;
4165 if (reccount == 0) {
4166 error = XFS_ERROR(ENOSPC);
4170 startoffset_fsb += allocated_fsb;
4171 allocatesize_fsb -= allocated_fsb;
4174 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4175 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4177 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4178 XFS_ITOV(ip), DM_RIGHT_NULL,
4179 XFS_ITOV(ip), DM_RIGHT_NULL,
4180 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4182 goto retry; /* Maybe DMAPI app. has made space */
4183 /* else fall through with error from XFS_SEND_DATA */
4189 xfs_bmap_cancel(&free_list);
4191 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4192 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4193 goto dmapi_enospc_check;
4197 * Zero file bytes between startoff and endoff inclusive.
4198 * The iolock is held exclusive and no blocks are buffered.
4201 xfs_zero_remaining_bytes(
4206 xfs_bmbt_irec_t imap;
4207 xfs_fileoff_t offset_fsb;
4208 xfs_off_t lastoffset;
4211 xfs_mount_t *mp = ip->i_mount;
4215 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4216 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4217 mp->m_rtdev_targp : mp->m_ddev_targp);
4219 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4220 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4222 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4224 if (error || nimap < 1)
4226 ASSERT(imap.br_blockcount >= 1);
4227 ASSERT(imap.br_startoff == offset_fsb);
4228 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4229 if (lastoffset > endoff)
4230 lastoffset = endoff;
4231 if (imap.br_startblock == HOLESTARTBLOCK)
4233 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4234 if (imap.br_state == XFS_EXT_UNWRITTEN)
4237 XFS_BUF_UNWRITE(bp);
4239 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4241 if ((error = xfs_iowait(bp))) {
4242 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4243 mp, bp, XFS_BUF_ADDR(bp));
4246 memset(XFS_BUF_PTR(bp) +
4247 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4248 0, lastoffset - offset + 1);
4253 if ((error = xfs_iowait(bp))) {
4254 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4255 mp, bp, XFS_BUF_ADDR(bp));
4264 * xfs_free_file_space()
4265 * This routine frees disk space for the given file.
4267 * This routine is only called by xfs_change_file_space
4268 * for an UNRESVSP type call.
4276 xfs_free_file_space(
4284 xfs_off_t end_dmi_offset;
4285 xfs_fileoff_t endoffset_fsb;
4287 xfs_fsblock_t firstfsb;
4288 xfs_bmap_free_t free_list;
4290 xfs_bmbt_irec_t imap;
4298 xfs_fileoff_t startoffset_fsb;
4301 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4304 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4308 if (len <= 0) /* if nothing being freed */
4310 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4311 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4312 end_dmi_offset = offset + len;
4313 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4315 if (offset < ip->i_d.di_size &&
4316 (attr_flags & ATTR_DMI) == 0 &&
4317 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4318 if (end_dmi_offset > ip->i_d.di_size)
4319 end_dmi_offset = ip->i_d.di_size;
4320 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4321 offset, end_dmi_offset - offset,
4322 AT_DELAY_FLAG(attr_flags), NULL);
4327 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4328 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4330 ilen = len + (offset & (rounding - 1));
4331 ioffset = offset & ~(rounding - 1);
4332 if (ilen & (rounding - 1))
4333 ilen = (ilen + rounding) & ~(rounding - 1);
4334 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4336 * Need to zero the stuff we're not freeing, on disk.
4337 * If its a realtime file & can't use unwritten extents then we
4338 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4339 * will take care of it for us.
4341 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4343 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4344 &imap, &nimap, NULL);
4346 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4349 ASSERT(nimap == 0 || nimap == 1);
4350 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4353 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4354 block = imap.br_startblock;
4355 mod = do_div(block, mp->m_sb.sb_rextsize);
4357 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4360 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4361 &imap, &nimap, NULL);
4363 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4366 ASSERT(nimap == 0 || nimap == 1);
4367 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4368 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4370 if (mod && (mod != mp->m_sb.sb_rextsize))
4371 endoffset_fsb -= mod;
4374 if ((done = (endoffset_fsb <= startoffset_fsb)))
4376 * One contiguous piece to clear
4378 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4381 * Some full blocks, possibly two pieces to clear
4383 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4384 error = xfs_zero_remaining_bytes(ip, offset,
4385 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4387 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4388 error = xfs_zero_remaining_bytes(ip,
4389 XFS_FSB_TO_B(mp, endoffset_fsb),
4394 * free file space until done or until there is an error
4396 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4397 while (!error && !done) {
4400 * allocate and setup the transaction
4402 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4403 error = xfs_trans_reserve(tp,
4405 XFS_WRITE_LOG_RES(mp),
4407 XFS_TRANS_PERM_LOG_RES,
4408 XFS_WRITE_LOG_COUNT);
4411 * check for running out of space
4415 * Free the transaction structure.
4417 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4418 xfs_trans_cancel(tp, 0);
4421 xfs_ilock(ip, XFS_ILOCK_EXCL);
4422 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4423 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4424 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4428 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4429 xfs_trans_ihold(tp, ip);
4432 * issue the bunmapi() call to free the blocks
4434 XFS_BMAP_INIT(&free_list, &firstfsb);
4435 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4436 endoffset_fsb - startoffset_fsb,
4437 0, 2, &firstfsb, &free_list, &done);
4443 * complete the transaction
4445 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4450 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4451 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4454 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4458 xfs_bmap_cancel(&free_list);
4460 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4461 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
4466 * xfs_change_file_space()
4467 * This routine allocates or frees disk space for the given file.
4468 * The user specified parameters are checked for alignment and size
4477 xfs_change_file_space(
4491 xfs_off_t startoffset;
4497 vp = BHV_TO_VNODE(bdp);
4498 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4500 ip = XFS_BHVTOI(bdp);
4504 * must be a regular file and have write permission
4506 if (vp->v_type != VREG)
4507 return XFS_ERROR(EINVAL);
4509 xfs_ilock(ip, XFS_ILOCK_SHARED);
4511 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4512 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4516 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4518 switch (bf->l_whence) {
4519 case 0: /*SEEK_SET*/
4521 case 1: /*SEEK_CUR*/
4522 bf->l_start += offset;
4524 case 2: /*SEEK_END*/
4525 bf->l_start += ip->i_d.di_size;
4528 return XFS_ERROR(EINVAL);
4531 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4533 if ( (bf->l_start < 0)
4534 || (bf->l_start > XFS_MAXIOFFSET(mp))
4535 || (bf->l_start + llen < 0)
4536 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4537 return XFS_ERROR(EINVAL);
4541 startoffset = bf->l_start;
4542 fsize = ip->i_d.di_size;
4545 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4547 * These calls do NOT zero the data space allocated to the file,
4548 * nor do they change the file size.
4550 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4552 * These calls cause the new file data to be zeroed and the file
4553 * size to be changed.
4555 setprealloc = clrprealloc = 0;
4558 case XFS_IOC_RESVSP:
4559 case XFS_IOC_RESVSP64:
4560 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4567 case XFS_IOC_UNRESVSP:
4568 case XFS_IOC_UNRESVSP64:
4569 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4574 case XFS_IOC_ALLOCSP:
4575 case XFS_IOC_ALLOCSP64:
4576 case XFS_IOC_FREESP:
4577 case XFS_IOC_FREESP64:
4578 if (startoffset > fsize) {
4579 error = xfs_alloc_file_space(ip, fsize,
4580 startoffset - fsize, 0, attr_flags);
4585 va.va_mask = XFS_AT_SIZE;
4586 va.va_size = startoffset;
4588 error = xfs_setattr(bdp, &va, attr_flags, credp);
4598 return XFS_ERROR(EINVAL);
4602 * update the inode timestamp, mode, and prealloc flag bits
4604 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4606 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4609 xfs_trans_cancel(tp, 0);
4613 xfs_ilock(ip, XFS_ILOCK_EXCL);
4615 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4616 xfs_trans_ihold(tp, ip);
4618 ip->i_d.di_mode &= ~S_ISUID;
4621 * Note that we don't have to worry about mandatory
4622 * file locking being disabled here because we only
4623 * clear the S_ISGID bit if the Group execute bit is
4624 * on, but if it was on then mandatory locking wouldn't
4625 * have been enabled.
4627 if (ip->i_d.di_mode & S_IXGRP)
4628 ip->i_d.di_mode &= ~S_ISGID;
4630 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4633 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4634 else if (clrprealloc)
4635 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4637 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4638 xfs_trans_set_sync(tp);
4640 error = xfs_trans_commit(tp, 0, NULL);
4642 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4647 vnodeops_t xfs_vnodeops = {
4648 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4649 .vop_open = xfs_open,
4650 .vop_read = xfs_read,
4651 #ifdef HAVE_SENDFILE
4652 .vop_sendfile = xfs_sendfile,
4654 .vop_write = xfs_write,
4655 .vop_ioctl = xfs_ioctl,
4656 .vop_getattr = xfs_getattr,
4657 .vop_setattr = xfs_setattr,
4658 .vop_access = xfs_access,
4659 .vop_lookup = xfs_lookup,
4660 .vop_create = xfs_create,
4661 .vop_remove = xfs_remove,
4662 .vop_link = xfs_link,
4663 .vop_rename = xfs_rename,
4664 .vop_mkdir = xfs_mkdir,
4665 .vop_rmdir = xfs_rmdir,
4666 .vop_readdir = xfs_readdir,
4667 .vop_symlink = xfs_symlink,
4668 .vop_readlink = xfs_readlink,
4669 .vop_fsync = xfs_fsync,
4670 .vop_inactive = xfs_inactive,
4671 .vop_fid2 = xfs_fid2,
4672 .vop_rwlock = xfs_rwlock,
4673 .vop_rwunlock = xfs_rwunlock,
4674 .vop_bmap = xfs_bmap,
4675 .vop_reclaim = xfs_reclaim,
4676 .vop_attr_get = xfs_attr_get,
4677 .vop_attr_set = xfs_attr_set,
4678 .vop_attr_remove = xfs_attr_remove,
4679 .vop_attr_list = xfs_attr_list,
4680 .vop_link_removed = (vop_link_removed_t)fs_noval,
4681 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4682 .vop_tosspages = fs_tosspages,
4683 .vop_flushinval_pages = fs_flushinval_pages,
4684 .vop_flush_pages = fs_flush_pages,
4685 .vop_release = xfs_release,
4686 .vop_iflush = xfs_inode_flush,