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 ip->i_d.di_flags = 0;
827 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
828 ip->i_d.di_flags |= XFS_DIFLAG_REALTIME;
829 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
831 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
832 ip->i_d.di_flags |= XFS_DIFLAG_IMMUTABLE;
833 if (vap->va_xflags & XFS_XFLAG_APPEND)
834 ip->i_d.di_flags |= XFS_DIFLAG_APPEND;
835 if (vap->va_xflags & XFS_XFLAG_SYNC)
836 ip->i_d.di_flags |= XFS_DIFLAG_SYNC;
837 if (vap->va_xflags & XFS_XFLAG_NOATIME)
838 ip->i_d.di_flags |= XFS_DIFLAG_NOATIME;
839 if (vap->va_xflags & XFS_XFLAG_NODUMP)
840 ip->i_d.di_flags |= XFS_DIFLAG_NODUMP;
841 /* can't set PREALLOC this way, just ignore it */
843 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
844 timeflags |= XFS_ICHGTIME_CHG;
848 * Change file inode change time only if XFS_AT_CTIME set
849 * AND we have been called by a DMI function.
852 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
853 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
854 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
855 ip->i_update_core = 1;
856 timeflags &= ~XFS_ICHGTIME_CHG;
860 * Send out timestamp changes that need to be set to the
861 * current time. Not done when called by a DMI function.
863 if (timeflags && !(flags & ATTR_DMI))
864 xfs_ichgtime(ip, timeflags);
866 XFS_STATS_INC(xs_ig_attrchg);
869 * If this is a synchronous mount, make sure that the
870 * transaction goes to disk before returning to the user.
871 * This is slightly sub-optimal in that truncates require
872 * two sync transactions instead of one for wsync filesytems.
873 * One for the truncate and one for the timestamps since we
874 * don't want to change the timestamps unless we're sure the
875 * truncate worked. Truncates are less than 1% of the laddis
876 * mix so this probably isn't worth the trouble to optimize.
880 if (mp->m_flags & XFS_MOUNT_WSYNC)
881 xfs_trans_set_sync(tp);
883 code = xfs_trans_commit(tp, commit_flags, NULL);
887 * If the (regular) file's mandatory locking mode changed, then
888 * notify the vnode. We do this under the inode lock to prevent
889 * racing calls to vop_vnode_change.
891 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
892 if (mandlock_before != mandlock_after) {
893 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
897 xfs_iunlock(ip, lock_flags);
900 * Release any dquot(s) the inode had kept before chown.
902 XFS_QM_DQRELE(mp, olddquot1);
903 XFS_QM_DQRELE(mp, olddquot2);
904 XFS_QM_DQRELE(mp, udqp);
905 XFS_QM_DQRELE(mp, gdqp);
911 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
912 !(flags & ATTR_DMI)) {
913 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
914 NULL, DM_RIGHT_NULL, NULL, NULL,
915 0, 0, AT_DELAY_FLAG(flags));
920 commit_flags |= XFS_TRANS_ABORT;
923 XFS_QM_DQRELE(mp, udqp);
924 XFS_QM_DQRELE(mp, gdqp);
926 xfs_trans_cancel(tp, commit_flags);
928 if (lock_flags != 0) {
929 xfs_iunlock(ip, lock_flags);
937 * Null conversion from vnode mode bits to inode mode bits, as in efs.
948 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
949 (inst_t *)__return_address);
951 ip = XFS_BHVTOI(bdp);
952 xfs_ilock(ip, XFS_ILOCK_SHARED);
953 error = xfs_iaccess(ip, mode, credp);
954 xfs_iunlock(ip, XFS_ILOCK_SHARED);
978 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
984 vp = BHV_TO_VNODE(bdp);
985 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
987 ip = XFS_BHVTOI(bdp);
990 if (XFS_FORCED_SHUTDOWN(mp))
991 return XFS_ERROR(EIO);
993 xfs_ilock(ip, XFS_ILOCK_SHARED);
995 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
997 offset = uiop->uio_offset;
998 count = uiop->uio_resid;
1001 error = XFS_ERROR(EINVAL);
1009 if (!(ioflags & IO_INVIS)) {
1010 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1014 * See if the symlink is stored inline.
1016 pathlen = (int)ip->i_d.di_size;
1018 if (ip->i_df.if_flags & XFS_IFINLINE) {
1019 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1023 * Symlink not inline. Call bmap to get it in.
1025 nmaps = SYMLINK_MAPS;
1027 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1028 0, NULL, 0, mval, &nmaps, NULL);
1034 for (n = 0; n < nmaps; n++) {
1035 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1036 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1037 bp = xfs_buf_read(mp->m_ddev_targp, d,
1038 BTOBB(byte_cnt), 0);
1039 error = XFS_BUF_GETERROR(bp);
1041 xfs_ioerror_alert("xfs_readlink",
1042 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1046 if (pathlen < byte_cnt)
1048 pathlen -= byte_cnt;
1050 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1059 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1068 * This is called to sync the inode and its data out to disk.
1069 * We need to hold the I/O lock while flushing the data, and
1070 * the inode lock while flushing the inode. The inode lock CANNOT
1071 * be held while flushing the data, so acquire after we're done
1086 vn_trace_entry(BHV_TO_VNODE(bdp),
1087 __FUNCTION__, (inst_t *)__return_address);
1089 ip = XFS_BHVTOI(bdp);
1091 ASSERT(start >= 0 && stop >= -1);
1093 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1094 return XFS_ERROR(EIO);
1097 * We always need to make sure that the required inode state
1098 * is safe on disk. The vnode might be clean but because
1099 * of committed transactions that haven't hit the disk yet.
1100 * Likewise, there could be unflushed non-transactional
1101 * changes to the inode core that have to go to disk.
1103 * The following code depends on one assumption: that
1104 * any transaction that changes an inode logs the core
1105 * because it has to change some field in the inode core
1106 * (typically nextents or nblocks). That assumption
1107 * implies that any transactions against an inode will
1108 * catch any non-transactional updates. If inode-altering
1109 * transactions exist that violate this assumption, the
1110 * code breaks. Right now, it figures that if the involved
1111 * update_* field is clear and the inode is unpinned, the
1112 * inode is clean. Either it's been flushed or it's been
1113 * committed and the commit has hit the disk unpinning the inode.
1114 * (Note that xfs_inode_item_format() called at commit clears
1115 * the update_* fields.)
1117 xfs_ilock(ip, XFS_ILOCK_SHARED);
1119 /* If we are flushing data then we care about update_size
1120 * being set, otherwise we care about update_core
1122 if ((flag & FSYNC_DATA) ?
1123 (ip->i_update_size == 0) :
1124 (ip->i_update_core == 0)) {
1126 * Timestamps/size haven't changed since last inode
1127 * flush or inode transaction commit. That means
1128 * either nothing got written or a transaction
1129 * committed which caught the updates. If the
1130 * latter happened and the transaction hasn't
1131 * hit the disk yet, the inode will be still
1132 * be pinned. If it is, force the log.
1135 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1137 if (xfs_ipincount(ip)) {
1138 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1140 ((flag & FSYNC_WAIT)
1141 ? XFS_LOG_SYNC : 0));
1146 * Kick off a transaction to log the inode
1147 * core to get the updates. Make it
1148 * sync if FSYNC_WAIT is passed in (which
1149 * is done by everybody but specfs). The
1150 * sync transaction will also force the log.
1152 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1153 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1154 if ((error = xfs_trans_reserve(tp, 0,
1155 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1157 xfs_trans_cancel(tp, 0);
1160 xfs_ilock(ip, XFS_ILOCK_EXCL);
1163 * Note - it's possible that we might have pushed
1164 * ourselves out of the way during trans_reserve
1165 * which would flush the inode. But there's no
1166 * guarantee that the inode buffer has actually
1167 * gone out yet (it's delwri). Plus the buffer
1168 * could be pinned anyway if it's part of an
1169 * inode in another recent transaction. So we
1170 * play it safe and fire off the transaction anyway.
1172 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1173 xfs_trans_ihold(tp, ip);
1174 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1175 if (flag & FSYNC_WAIT)
1176 xfs_trans_set_sync(tp);
1177 error = xfs_trans_commit(tp, 0, NULL);
1179 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1185 * This is called by xfs_inactive to free any blocks beyond eof,
1186 * when the link count isn't zero.
1189 xfs_inactive_free_eofblocks(
1195 xfs_fileoff_t end_fsb;
1196 xfs_fileoff_t last_fsb;
1197 xfs_filblks_t map_len;
1199 xfs_bmbt_irec_t imap;
1202 * Figure out if there are any blocks beyond the end
1203 * of the file. If not, then there is nothing to do.
1205 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1206 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1207 map_len = last_fsb - end_fsb;
1212 xfs_ilock(ip, XFS_ILOCK_SHARED);
1213 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1214 NULL, 0, &imap, &nimaps, NULL);
1215 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1217 if (!error && (nimaps != 0) &&
1218 (imap.br_startblock != HOLESTARTBLOCK)) {
1220 * Attach the dquots to the inode up front.
1222 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1226 * There are blocks after the end of file.
1227 * Free them up now by truncating the file to
1230 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1233 * Do the xfs_itruncate_start() call before
1234 * reserving any log space because
1235 * itruncate_start will call into the buffer
1236 * cache and we can't
1237 * do that within a transaction.
1239 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1240 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1243 error = xfs_trans_reserve(tp, 0,
1244 XFS_ITRUNCATE_LOG_RES(mp),
1245 0, XFS_TRANS_PERM_LOG_RES,
1246 XFS_ITRUNCATE_LOG_COUNT);
1248 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1249 xfs_trans_cancel(tp, 0);
1250 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1254 xfs_ilock(ip, XFS_ILOCK_EXCL);
1255 xfs_trans_ijoin(tp, ip,
1258 xfs_trans_ihold(tp, ip);
1260 error = xfs_itruncate_finish(&tp, ip,
1265 * If we get an error at this point we
1266 * simply don't bother truncating the file.
1269 xfs_trans_cancel(tp,
1270 (XFS_TRANS_RELEASE_LOG_RES |
1273 error = xfs_trans_commit(tp,
1274 XFS_TRANS_RELEASE_LOG_RES,
1277 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1283 * Free a symlink that has blocks associated with it.
1286 xfs_inactive_symlink_rmt(
1294 xfs_fsblock_t first_block;
1295 xfs_bmap_free_t free_list;
1298 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1306 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1308 * We're freeing a symlink that has some
1309 * blocks allocated to it. Free the
1310 * blocks here. We know that we've got
1311 * either 1 or 2 extents and that we can
1312 * free them all in one bunmapi call.
1314 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1315 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1316 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1317 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1318 xfs_trans_cancel(tp, 0);
1323 * Lock the inode, fix the size, and join it to the transaction.
1324 * Hold it so in the normal path, we still have it locked for
1325 * the second transaction. In the error paths we need it
1326 * held so the cancel won't rele it, see below.
1328 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1329 size = (int)ip->i_d.di_size;
1330 ip->i_d.di_size = 0;
1331 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1332 xfs_trans_ihold(tp, ip);
1333 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1335 * Find the block(s) so we can inval and unmap them.
1338 XFS_BMAP_INIT(&free_list, &first_block);
1339 nmaps = sizeof(mval) / sizeof(mval[0]);
1340 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1341 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1345 * Invalidate the block(s).
1347 for (i = 0; i < nmaps; i++) {
1348 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1349 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1350 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1351 xfs_trans_binval(tp, bp);
1354 * Unmap the dead block(s) to the free_list.
1356 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1357 &first_block, &free_list, &done)))
1361 * Commit the first transaction. This logs the EFI and the inode.
1363 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1366 * The transaction must have been committed, since there were
1367 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1368 * The new tp has the extent freeing and EFDs.
1372 * The first xact was committed, so add the inode to the new one.
1373 * Mark it dirty so it will be logged and moved forward in the log as
1374 * part of every commit.
1376 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1377 xfs_trans_ihold(tp, ip);
1378 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1380 * Get a new, empty transaction to return to our caller.
1382 ntp = xfs_trans_dup(tp);
1384 * Commit the transaction containing extent freeing and EFD's.
1385 * If we get an error on the commit here or on the reserve below,
1386 * we need to unlock the inode since the new transaction doesn't
1387 * have the inode attached.
1389 error = xfs_trans_commit(tp, 0, NULL);
1392 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1396 * Remove the memory for extent descriptions (just bookkeeping).
1398 if (ip->i_df.if_bytes)
1399 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1400 ASSERT(ip->i_df.if_bytes == 0);
1402 * Put an itruncate log reservation in the new transaction
1405 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1406 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1407 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1411 * Return with the inode locked but not joined to the transaction.
1417 xfs_bmap_cancel(&free_list);
1420 * Have to come here with the inode locked and either
1421 * (held and in the transaction) or (not in the transaction).
1422 * If the inode isn't held then cancel would iput it, but
1423 * that's wrong since this is inactive and the vnode ref
1424 * count is 0 already.
1425 * Cancel won't do anything to the inode if held, but it still
1426 * needs to be locked until the cancel is done, if it was
1427 * joined to the transaction.
1429 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1430 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1437 xfs_inactive_symlink_local(
1443 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1445 * We're freeing a symlink which fit into
1446 * the inode. Just free the memory used
1447 * to hold the old symlink.
1449 error = xfs_trans_reserve(*tpp, 0,
1450 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1451 0, XFS_TRANS_PERM_LOG_RES,
1452 XFS_ITRUNCATE_LOG_COUNT);
1455 xfs_trans_cancel(*tpp, 0);
1459 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1462 * Zero length symlinks _can_ exist.
1464 if (ip->i_df.if_bytes > 0) {
1465 xfs_idata_realloc(ip,
1466 -(ip->i_df.if_bytes),
1468 ASSERT(ip->i_df.if_bytes == 0);
1485 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1488 ASSERT(ip->i_d.di_forkoff != 0);
1489 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1490 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1492 error = xfs_attr_inactive(ip);
1495 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1496 return (error); /* goto out*/
1499 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1500 error = xfs_trans_reserve(tp, 0,
1501 XFS_IFREE_LOG_RES(mp),
1502 0, XFS_TRANS_PERM_LOG_RES,
1503 XFS_INACTIVE_LOG_COUNT);
1505 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1506 xfs_trans_cancel(tp, 0);
1508 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1512 xfs_ilock(ip, XFS_ILOCK_EXCL);
1513 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1514 xfs_trans_ihold(tp, ip);
1515 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1517 ASSERT(ip->i_d.di_anextents == 0);
1532 vp = BHV_TO_VNODE(bdp);
1533 ip = XFS_BHVTOI(bdp);
1535 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1539 /* If this is a read-only mount, don't do this (would generate I/O) */
1540 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1543 #ifdef HAVE_REFCACHE
1544 /* If we are in the NFS reference cache then don't do this now */
1551 if (ip->i_d.di_nlink != 0) {
1552 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1553 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1554 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1555 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1556 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1558 /* Update linux inode block count after free above */
1559 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1560 ip->i_d.di_nblocks + ip->i_delayed_blks);
1570 * This is called when the vnode reference count for the vnode
1571 * goes to zero. If the file has been unlinked, then it must
1572 * now be truncated. Also, we clear all of the read-ahead state
1573 * kept for the inode here since the file is now closed.
1582 xfs_bmap_free_t free_list;
1583 xfs_fsblock_t first_block;
1590 vp = BHV_TO_VNODE(bdp);
1591 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1593 ip = XFS_BHVTOI(bdp);
1596 * If the inode is already free, then there can be nothing
1599 if (ip->i_d.di_mode == 0) {
1600 ASSERT(ip->i_df.if_real_bytes == 0);
1601 ASSERT(ip->i_df.if_broot_bytes == 0);
1602 return VN_INACTIVE_CACHE;
1606 * Only do a truncate if it's a regular file with
1607 * some actual space in it. It's OK to look at the
1608 * inode's fields without the lock because we're the
1609 * only one with a reference to the inode.
1611 truncate = ((ip->i_d.di_nlink == 0) &&
1612 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1613 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1617 if (ip->i_d.di_nlink == 0 &&
1618 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1619 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1624 /* If this is a read-only mount, don't do this (would generate I/O) */
1625 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1628 if (ip->i_d.di_nlink != 0) {
1629 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1630 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1631 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1632 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1633 (ip->i_delayed_blks != 0))) {
1634 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1635 return (VN_INACTIVE_CACHE);
1636 /* Update linux inode block count after free above */
1637 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1638 ip->i_d.di_nblocks + ip->i_delayed_blks);
1643 ASSERT(ip->i_d.di_nlink == 0);
1645 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1646 return (VN_INACTIVE_CACHE);
1648 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1651 * Do the xfs_itruncate_start() call before
1652 * reserving any log space because itruncate_start
1653 * will call into the buffer cache and we can't
1654 * do that within a transaction.
1656 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1658 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1660 error = xfs_trans_reserve(tp, 0,
1661 XFS_ITRUNCATE_LOG_RES(mp),
1662 0, XFS_TRANS_PERM_LOG_RES,
1663 XFS_ITRUNCATE_LOG_COUNT);
1665 /* Don't call itruncate_cleanup */
1666 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1667 xfs_trans_cancel(tp, 0);
1668 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1669 return (VN_INACTIVE_CACHE);
1672 xfs_ilock(ip, XFS_ILOCK_EXCL);
1673 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1674 xfs_trans_ihold(tp, ip);
1677 * normally, we have to run xfs_itruncate_finish sync.
1678 * But if filesystem is wsync and we're in the inactive
1679 * path, then we know that nlink == 0, and that the
1680 * xaction that made nlink == 0 is permanently committed
1681 * since xfs_remove runs as a synchronous transaction.
1683 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1684 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1687 xfs_trans_cancel(tp,
1688 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1689 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1690 return (VN_INACTIVE_CACHE);
1692 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1695 * If we get an error while cleaning up a
1696 * symlink we bail out.
1698 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1699 xfs_inactive_symlink_rmt(ip, &tp) :
1700 xfs_inactive_symlink_local(ip, &tp);
1704 return (VN_INACTIVE_CACHE);
1707 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1708 xfs_trans_ihold(tp, ip);
1710 error = xfs_trans_reserve(tp, 0,
1711 XFS_IFREE_LOG_RES(mp),
1712 0, XFS_TRANS_PERM_LOG_RES,
1713 XFS_INACTIVE_LOG_COUNT);
1715 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1716 xfs_trans_cancel(tp, 0);
1717 return (VN_INACTIVE_CACHE);
1720 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1721 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1722 xfs_trans_ihold(tp, ip);
1726 * If there are attributes associated with the file
1727 * then blow them away now. The code calls a routine
1728 * that recursively deconstructs the attribute fork.
1729 * We need to just commit the current transaction
1730 * because we can't use it for xfs_attr_inactive().
1732 if (ip->i_d.di_anextents > 0) {
1733 error = xfs_inactive_attrs(ip, &tp);
1735 * If we got an error, the transaction is already
1736 * cancelled, and the inode is unlocked. Just get out.
1739 return (VN_INACTIVE_CACHE);
1740 } else if (ip->i_afp) {
1741 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1747 XFS_BMAP_INIT(&free_list, &first_block);
1748 error = xfs_ifree(tp, ip, &free_list);
1751 * If we fail to free the inode, shut down. The cancel
1752 * might do that, we need to make sure. Otherwise the
1753 * inode might be lost for a long time or forever.
1755 if (!XFS_FORCED_SHUTDOWN(mp)) {
1757 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1758 error, mp->m_fsname);
1759 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1761 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1764 * Credit the quota account(s). The inode is gone.
1766 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1769 * Just ignore errors at this point. There is
1770 * nothing we can do except to try to keep going.
1772 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1774 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1777 * Release the dquots held by inode, if any.
1779 XFS_QM_DQDETACH(mp, ip);
1781 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1784 return VN_INACTIVE_CACHE;
1793 bhv_desc_t *dir_bdp,
1800 xfs_inode_t *dp, *ip;
1806 dir_vp = BHV_TO_VNODE(dir_bdp);
1807 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1809 dp = XFS_BHVTOI(dir_bdp);
1811 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1812 return XFS_ERROR(EIO);
1814 lock_mode = xfs_ilock_map_shared(dp);
1815 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1817 *vpp = XFS_ITOV(ip);
1820 xfs_iunlock_map_shared(dp, lock_mode);
1825 #define XFS_CREATE_NEW_MAXTRIES 10000
1828 * xfs_create (create a new file).
1832 bhv_desc_t *dir_bdp,
1838 char *name = VNAME(dentry);
1840 xfs_inode_t *dp, *ip;
1846 xfs_bmap_free_t free_list;
1847 xfs_fsblock_t first_block;
1848 boolean_t dp_joined_to_trans;
1849 int dm_event_sent = 0;
1853 struct xfs_dquot *udqp, *gdqp;
1859 dir_vp = BHV_TO_VNODE(dir_bdp);
1860 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1862 dp = XFS_BHVTOI(dir_bdp);
1865 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1866 namelen = VNAMELEN(dentry);
1868 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1869 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1870 dir_vp, DM_RIGHT_NULL, NULL,
1871 DM_RIGHT_NULL, name, NULL,
1879 if (XFS_FORCED_SHUTDOWN(mp))
1880 return XFS_ERROR(EIO);
1882 /* Return through std_return after this point. */
1885 if (vap->va_mask & XFS_AT_PROJID)
1886 prid = (xfs_prid_t)vap->va_projid;
1888 prid = (xfs_prid_t)dfltprid;
1891 * Make sure that we have allocated dquot(s) on disk.
1893 error = XFS_QM_DQVOPALLOC(mp, dp,
1894 current_fsuid(credp), current_fsgid(credp),
1895 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1900 dp_joined_to_trans = B_FALSE;
1902 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1903 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1904 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1906 * Initially assume that the file does not exist and
1907 * reserve the resources for that case. If that is not
1908 * the case we'll drop the one we have and get a more
1909 * appropriate transaction later.
1911 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1912 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1913 if (error == ENOSPC) {
1915 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1916 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1924 xfs_ilock(dp, XFS_ILOCK_EXCL);
1926 XFS_BMAP_INIT(&free_list, &first_block);
1931 * Reserve disk quota and the inode.
1933 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1938 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1940 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1941 error = xfs_dir_ialloc(&tp, dp,
1942 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1943 rdev, credp, prid, resblks > 0,
1946 if (error == ENOSPC)
1953 * At this point, we've gotten a newly allocated inode.
1954 * It is locked (and joined to the transaction).
1957 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1960 * Now we join the directory inode to the transaction.
1961 * We do not do it earlier because xfs_dir_ialloc
1962 * might commit the previous transaction (and release
1967 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1968 dp_joined_to_trans = B_TRUE;
1970 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1971 &first_block, &free_list,
1972 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1974 ASSERT(error != ENOSPC);
1977 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1978 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1981 * If this is a synchronous mount, make sure that the
1982 * create transaction goes to disk before returning to
1985 if (mp->m_flags & XFS_MOUNT_WSYNC) {
1986 xfs_trans_set_sync(tp);
1992 * Attach the dquot(s) to the inodes and modify them incore.
1993 * These ids of the inode couldn't have changed since the new
1994 * inode has been locked ever since it was created.
1996 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
1999 * xfs_trans_commit normally decrements the vnode ref count
2000 * when it unlocks the inode. Since we want to return the
2001 * vnode to the caller, we bump the vnode ref count now.
2006 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2008 xfs_bmap_cancel(&free_list);
2012 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2019 XFS_QM_DQRELE(mp, udqp);
2020 XFS_QM_DQRELE(mp, gdqp);
2023 * Propogate the fact that the vnode changed after the
2024 * xfs_inode locks have been released.
2026 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2030 /* Fallthrough to std_return with error = 0 */
2033 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2034 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2035 DM_EVENT_POSTCREATE)) {
2036 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2037 dir_vp, DM_RIGHT_NULL,
2039 DM_RIGHT_NULL, name, NULL,
2040 dm_di_mode, error, 0);
2045 cancel_flags |= XFS_TRANS_ABORT;
2050 xfs_trans_cancel(tp, cancel_flags);
2052 if (!dp_joined_to_trans && (dp != NULL))
2053 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2054 XFS_QM_DQRELE(mp, udqp);
2055 XFS_QM_DQRELE(mp, gdqp);
2061 * Wait until after the current transaction is aborted to
2062 * release the inode. This prevents recursive transactions
2063 * and deadlocks from xfs_inactive.
2065 cancel_flags |= XFS_TRANS_ABORT;
2066 xfs_trans_cancel(tp, cancel_flags);
2069 XFS_QM_DQRELE(mp, udqp);
2070 XFS_QM_DQRELE(mp, gdqp);
2077 * Some counters to see if (and how often) we are hitting some deadlock
2078 * prevention code paths.
2082 int xfs_rm_lock_delays;
2083 int xfs_rm_attempts;
2087 * The following routine will lock the inodes associated with the
2088 * directory and the named entry in the directory. The locks are
2089 * acquired in increasing inode number.
2091 * If the entry is "..", then only the directory is locked. The
2092 * vnode ref count will still include that from the .. entry in
2095 * There is a deadlock we need to worry about. If the locked directory is
2096 * in the AIL, it might be blocking up the log. The next inode we lock
2097 * could be already locked by another thread waiting for log space (e.g
2098 * a permanent log reservation with a long running transaction (see
2099 * xfs_itruncate_finish)). To solve this, we must check if the directory
2100 * is in the ail and use lock_nowait. If we can't lock, we need to
2101 * drop the inode lock on the directory and try again. xfs_iunlock will
2102 * potentially push the tail if we were holding up the log.
2105 xfs_lock_dir_and_entry(
2108 xfs_inode_t *ip) /* inode of entry 'name' */
2112 xfs_inode_t *ips[2];
2121 xfs_ilock(dp, XFS_ILOCK_EXCL);
2128 * We want to lock in increasing inum. Since we've already
2129 * acquired the lock on the directory, we may need to release
2130 * if if the inum of the entry turns out to be less.
2132 if (e_inum > dp->i_ino) {
2134 * We are already in the right order, so just
2135 * lock on the inode of the entry.
2136 * We need to use nowait if dp is in the AIL.
2139 lp = (xfs_log_item_t *)dp->i_itemp;
2140 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2141 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2148 * Unlock dp and try again.
2149 * xfs_iunlock will try to push the tail
2150 * if the inode is in the AIL.
2153 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2155 if ((attempts % 5) == 0) {
2156 delay(1); /* Don't just spin the CPU */
2158 xfs_rm_lock_delays++;
2164 xfs_ilock(ip, XFS_ILOCK_EXCL);
2166 } else if (e_inum < dp->i_ino) {
2167 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2171 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2173 /* else e_inum == dp->i_ino */
2174 /* This can happen if we're asked to lock /x/..
2175 * the entry is "..", which is also the parent directory.
2183 int xfs_small_retries;
2184 int xfs_middle_retries;
2185 int xfs_lots_retries;
2186 int xfs_lock_delays;
2190 * The following routine will lock n inodes in exclusive mode.
2191 * We assume the caller calls us with the inodes in i_ino order.
2193 * We need to detect deadlock where an inode that we lock
2194 * is in the AIL and we start waiting for another inode that is locked
2195 * by a thread in a long running transaction (such as truncate). This can
2196 * result in deadlock since the long running trans might need to wait
2197 * for the inode we just locked in order to push the tail and free space
2207 int attempts = 0, i, j, try_lock;
2210 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2221 for (; i < inodes; i++) {
2224 if (i && (ips[i] == ips[i-1])) /* Already locked */
2228 * If try_lock is not set yet, make sure all locked inodes
2229 * are not in the AIL.
2230 * If any are, set try_lock to be used later.
2234 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2235 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2236 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2243 * If any of the previous locks we have locked is in the AIL,
2244 * we must TRY to get the second and subsequent locks. If
2245 * we can't get any, we must release all we have
2250 /* try_lock must be 0 if i is 0. */
2252 * try_lock means we have an inode locked
2253 * that is in the AIL.
2256 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2260 * Unlock all previous guys and try again.
2261 * xfs_iunlock will try to push the tail
2262 * if the inode is in the AIL.
2265 for(j = i - 1; j >= 0; j--) {
2268 * Check to see if we've already
2269 * unlocked this one.
2270 * Not the first one going back,
2271 * and the inode ptr is the same.
2273 if ((j != (i - 1)) && ips[j] ==
2277 xfs_iunlock(ips[j], lock_mode);
2280 if ((attempts % 5) == 0) {
2281 delay(1); /* Don't just spin the CPU */
2291 xfs_ilock(ips[i], lock_mode);
2297 if (attempts < 5) xfs_small_retries++;
2298 else if (attempts < 100) xfs_middle_retries++;
2299 else xfs_lots_retries++;
2307 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2308 int remove_which_error_return = 0;
2310 #define REMOVE_DEBUG_TRACE(x)
2311 #endif /* ! DEBUG */
2320 bhv_desc_t *dir_bdp,
2325 char *name = VNAME(dentry);
2326 xfs_inode_t *dp, *ip;
2327 xfs_trans_t *tp = NULL;
2330 xfs_bmap_free_t free_list;
2331 xfs_fsblock_t first_block;
2339 dir_vp = BHV_TO_VNODE(dir_bdp);
2340 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2342 dp = XFS_BHVTOI(dir_bdp);
2345 if (XFS_FORCED_SHUTDOWN(mp))
2346 return XFS_ERROR(EIO);
2348 namelen = VNAMELEN(dentry);
2350 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2351 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2352 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2353 name, NULL, 0, 0, 0);
2358 /* From this point on, return through std_return */
2362 * We need to get a reference to ip before we get our log
2363 * reservation. The reason for this is that we cannot call
2364 * xfs_iget for an inode for which we do not have a reference
2365 * once we've acquired a log reservation. This is because the
2366 * inode we are trying to get might be in xfs_inactive going
2367 * for a log reservation. Since we'll have to wait for the
2368 * inactive code to complete before returning from xfs_iget,
2369 * we need to make sure that we don't have log space reserved
2370 * when we call xfs_iget. Instead we get an unlocked referece
2371 * to the inode before getting our log reservation.
2373 error = xfs_get_dir_entry(dentry, &ip);
2375 REMOVE_DEBUG_TRACE(__LINE__);
2379 dm_di_mode = ip->i_d.di_mode;
2381 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2385 error = XFS_QM_DQATTACH(mp, dp, 0);
2386 if (!error && dp != ip)
2387 error = XFS_QM_DQATTACH(mp, ip, 0);
2389 REMOVE_DEBUG_TRACE(__LINE__);
2394 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2395 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2397 * We try to get the real space reservation first,
2398 * allowing for directory btree deletion(s) implying
2399 * possible bmap insert(s). If we can't get the space
2400 * reservation then we use 0 instead, and avoid the bmap
2401 * btree insert(s) in the directory code by, if the bmap
2402 * insert tries to happen, instead trimming the LAST
2403 * block from the directory.
2405 resblks = XFS_REMOVE_SPACE_RES(mp);
2406 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2407 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2408 if (error == ENOSPC) {
2410 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2411 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2414 ASSERT(error != ENOSPC);
2415 REMOVE_DEBUG_TRACE(__LINE__);
2416 xfs_trans_cancel(tp, 0);
2421 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2423 REMOVE_DEBUG_TRACE(__LINE__);
2424 xfs_trans_cancel(tp, cancel_flags);
2430 * At this point, we've gotten both the directory and the entry
2433 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2436 * Increment vnode ref count only in this case since
2437 * there's an extra vnode reference in the case where
2441 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2445 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2447 XFS_BMAP_INIT(&free_list, &first_block);
2448 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2449 &first_block, &free_list, 0);
2451 ASSERT(error != ENOENT);
2452 REMOVE_DEBUG_TRACE(__LINE__);
2455 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2458 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2460 error = xfs_droplink(tp, ip);
2462 REMOVE_DEBUG_TRACE(__LINE__);
2466 /* Determine if this is the last link while
2467 * we are in the transaction.
2469 link_zero = (ip)->i_d.di_nlink==0;
2472 * Take an extra ref on the inode so that it doesn't
2473 * go to xfs_inactive() from within the commit.
2478 * If this is a synchronous mount, make sure that the
2479 * remove transaction goes to disk before returning to
2482 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2483 xfs_trans_set_sync(tp);
2486 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2488 REMOVE_DEBUG_TRACE(__LINE__);
2492 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2499 * Before we drop our extra reference to the inode, purge it
2500 * from the refcache if it is there. By waiting until afterwards
2501 * to do the IRELE, we ensure that we won't go inactive in the
2502 * xfs_refcache_purge_ip routine (although that would be OK).
2504 xfs_refcache_purge_ip(ip);
2506 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2509 * Let interposed file systems know about removed links.
2511 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2515 /* Fall through to std_return with error = 0 */
2517 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2518 DM_EVENT_POSTREMOVE)) {
2519 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2520 dir_vp, DM_RIGHT_NULL,
2521 NULL, DM_RIGHT_NULL,
2522 name, NULL, dm_di_mode, error, 0);
2527 xfs_bmap_cancel(&free_list);
2528 cancel_flags |= XFS_TRANS_ABORT;
2529 xfs_trans_cancel(tp, cancel_flags);
2534 * In this case make sure to not release the inode until after
2535 * the current transaction is aborted. Releasing it beforehand
2536 * can cause us to go to xfs_inactive and start a recursive
2537 * transaction which can easily deadlock with the current one.
2539 xfs_bmap_cancel(&free_list);
2540 cancel_flags |= XFS_TRANS_ABORT;
2541 xfs_trans_cancel(tp, cancel_flags);
2544 * Before we drop our extra reference to the inode, purge it
2545 * from the refcache if it is there. By waiting until afterwards
2546 * to do the IRELE, we ensure that we won't go inactive in the
2547 * xfs_refcache_purge_ip routine (although that would be OK).
2549 xfs_refcache_purge_ip(ip);
2563 bhv_desc_t *target_dir_bdp,
2568 xfs_inode_t *tdp, *sip;
2571 xfs_inode_t *ips[2];
2573 xfs_bmap_free_t free_list;
2574 xfs_fsblock_t first_block;
2577 vnode_t *target_dir_vp;
2578 bhv_desc_t *src_bdp;
2580 char *target_name = VNAME(dentry);
2583 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2584 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2585 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2587 target_namelen = VNAMELEN(dentry);
2588 if (src_vp->v_type == VDIR)
2589 return XFS_ERROR(EPERM);
2592 * For now, manually find the XFS behavior descriptor for
2593 * the source vnode. If it doesn't exist then something
2594 * is wrong and we should just return an error.
2595 * Eventually we need to figure out how link is going to
2596 * work in the face of stacked vnodes.
2598 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2599 if (src_bdp == NULL) {
2600 return XFS_ERROR(EXDEV);
2602 sip = XFS_BHVTOI(src_bdp);
2603 tdp = XFS_BHVTOI(target_dir_bdp);
2605 if (XFS_FORCED_SHUTDOWN(mp))
2606 return XFS_ERROR(EIO);
2608 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2609 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2610 target_dir_vp, DM_RIGHT_NULL,
2611 src_vp, DM_RIGHT_NULL,
2612 target_name, NULL, 0, 0, 0);
2617 /* Return through std_return after this point. */
2619 error = XFS_QM_DQATTACH(mp, sip, 0);
2620 if (!error && sip != tdp)
2621 error = XFS_QM_DQATTACH(mp, tdp, 0);
2625 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2626 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2627 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2628 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2629 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2630 if (error == ENOSPC) {
2632 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2633 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2640 if (sip->i_ino < tdp->i_ino) {
2648 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2651 * Increment vnode ref counts since xfs_trans_commit &
2652 * xfs_trans_cancel will both unlock the inodes and
2653 * decrement the associated ref counts.
2656 VN_HOLD(target_dir_vp);
2657 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2658 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2661 * If the source has too many links, we can't make any more to it.
2663 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2664 error = XFS_ERROR(EMLINK);
2669 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2673 XFS_BMAP_INIT(&free_list, &first_block);
2675 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2676 sip->i_ino, &first_block, &free_list,
2680 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2682 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2684 error = xfs_bumplink(tp, sip);
2690 * If this is a synchronous mount, make sure that the
2691 * link transaction goes to disk before returning to
2694 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2695 xfs_trans_set_sync(tp);
2698 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2700 xfs_bmap_cancel(&free_list);
2704 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2709 /* Fall through to std_return with error = 0. */
2711 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2712 DM_EVENT_POSTLINK)) {
2713 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2714 target_dir_vp, DM_RIGHT_NULL,
2715 src_vp, DM_RIGHT_NULL,
2716 target_name, NULL, 0, error, 0);
2721 cancel_flags |= XFS_TRANS_ABORT;
2724 xfs_trans_cancel(tp, cancel_flags);
2734 bhv_desc_t *dir_bdp,
2740 char *dir_name = VNAME(dentry);
2742 xfs_inode_t *cdp; /* inode of created dir */
2743 vnode_t *cvp; /* vnode of created dir */
2749 xfs_bmap_free_t free_list;
2750 xfs_fsblock_t first_block;
2752 boolean_t dp_joined_to_trans;
2753 boolean_t created = B_FALSE;
2754 int dm_event_sent = 0;
2756 struct xfs_dquot *udqp, *gdqp;
2761 dir_vp = BHV_TO_VNODE(dir_bdp);
2762 dp = XFS_BHVTOI(dir_bdp);
2765 if (XFS_FORCED_SHUTDOWN(mp))
2766 return XFS_ERROR(EIO);
2768 dir_namelen = VNAMELEN(dentry);
2771 dp_joined_to_trans = B_FALSE;
2772 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2774 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2775 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2776 dir_vp, DM_RIGHT_NULL, NULL,
2777 DM_RIGHT_NULL, dir_name, NULL,
2784 /* Return through std_return after this point. */
2786 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2790 if (vap->va_mask & XFS_AT_PROJID)
2791 prid = (xfs_prid_t)vap->va_projid;
2793 prid = (xfs_prid_t)dfltprid;
2796 * Make sure that we have allocated dquot(s) on disk.
2798 error = XFS_QM_DQVOPALLOC(mp, dp,
2799 current_fsuid(credp), current_fsgid(credp),
2800 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2804 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2805 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2806 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2807 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2808 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2809 if (error == ENOSPC) {
2811 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2812 XFS_TRANS_PERM_LOG_RES,
2813 XFS_MKDIR_LOG_COUNT);
2821 xfs_ilock(dp, XFS_ILOCK_EXCL);
2824 * Check for directory link count overflow.
2826 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2827 error = XFS_ERROR(EMLINK);
2832 * Reserve disk quota and the inode.
2834 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2839 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2842 * create the directory inode.
2844 error = xfs_dir_ialloc(&tp, dp,
2845 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2846 0, credp, prid, resblks > 0,
2849 if (error == ENOSPC)
2856 * Now we add the directory inode to the transaction.
2857 * We waited until now since xfs_dir_ialloc might start
2858 * a new transaction. Had we joined the transaction
2859 * earlier, the locks might have gotten released.
2862 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2863 dp_joined_to_trans = B_TRUE;
2865 XFS_BMAP_INIT(&free_list, &first_block);
2867 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2868 cdp->i_ino, &first_block, &free_list,
2869 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2871 ASSERT(error != ENOSPC);
2874 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2877 * Bump the in memory version number of the parent directory
2878 * so that other processes accessing it will recognize that
2879 * the directory has changed.
2883 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2889 error = xfs_bumplink(tp, dp);
2894 cvp = XFS_ITOV(cdp);
2902 * Attach the dquots to the new inode and modify the icount incore.
2904 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2907 * If this is a synchronous mount, make sure that the
2908 * mkdir transaction goes to disk before returning to
2911 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2912 xfs_trans_set_sync(tp);
2915 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2921 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2922 XFS_QM_DQRELE(mp, udqp);
2923 XFS_QM_DQRELE(mp, gdqp);
2928 /* Fall through to std_return with error = 0 or errno from
2929 * xfs_trans_commit. */
2932 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2933 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2934 DM_EVENT_POSTCREATE)) {
2935 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2936 dir_vp, DM_RIGHT_NULL,
2937 created ? XFS_ITOV(cdp):NULL,
2940 dm_di_mode, error, 0);
2946 xfs_bmap_cancel(&free_list);
2948 cancel_flags |= XFS_TRANS_ABORT;
2950 xfs_trans_cancel(tp, cancel_flags);
2951 XFS_QM_DQRELE(mp, udqp);
2952 XFS_QM_DQRELE(mp, gdqp);
2954 if (!dp_joined_to_trans && (dp != NULL)) {
2955 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2968 bhv_desc_t *dir_bdp,
2972 char *name = VNAME(dentry);
2974 xfs_inode_t *cdp; /* child directory */
2978 xfs_bmap_free_t free_list;
2979 xfs_fsblock_t first_block;
2988 dir_vp = BHV_TO_VNODE(dir_bdp);
2989 dp = XFS_BHVTOI(dir_bdp);
2992 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2994 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
2995 return XFS_ERROR(EIO);
2996 namelen = VNAMELEN(dentry);
2998 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2999 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3000 dir_vp, DM_RIGHT_NULL,
3001 NULL, DM_RIGHT_NULL,
3002 name, NULL, 0, 0, 0);
3004 return XFS_ERROR(error);
3007 /* Return through std_return after this point. */
3012 * We need to get a reference to cdp before we get our log
3013 * reservation. The reason for this is that we cannot call
3014 * xfs_iget for an inode for which we do not have a reference
3015 * once we've acquired a log reservation. This is because the
3016 * inode we are trying to get might be in xfs_inactive going
3017 * for a log reservation. Since we'll have to wait for the
3018 * inactive code to complete before returning from xfs_iget,
3019 * we need to make sure that we don't have log space reserved
3020 * when we call xfs_iget. Instead we get an unlocked referece
3021 * to the inode before getting our log reservation.
3023 error = xfs_get_dir_entry(dentry, &cdp);
3025 REMOVE_DEBUG_TRACE(__LINE__);
3029 dm_di_mode = cdp->i_d.di_mode;
3032 * Get the dquots for the inodes.
3034 error = XFS_QM_DQATTACH(mp, dp, 0);
3035 if (!error && dp != cdp)
3036 error = XFS_QM_DQATTACH(mp, cdp, 0);
3039 REMOVE_DEBUG_TRACE(__LINE__);
3043 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3044 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3046 * We try to get the real space reservation first,
3047 * allowing for directory btree deletion(s) implying
3048 * possible bmap insert(s). If we can't get the space
3049 * reservation then we use 0 instead, and avoid the bmap
3050 * btree insert(s) in the directory code by, if the bmap
3051 * insert tries to happen, instead trimming the LAST
3052 * block from the directory.
3054 resblks = XFS_REMOVE_SPACE_RES(mp);
3055 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3056 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3057 if (error == ENOSPC) {
3059 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3060 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3063 ASSERT(error != ENOSPC);
3068 XFS_BMAP_INIT(&free_list, &first_block);
3071 * Now lock the child directory inode and the parent directory
3072 * inode in the proper order. This will take care of validating
3073 * that the directory entry for the child directory inode has
3074 * not changed while we were obtaining a log reservation.
3076 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3078 xfs_trans_cancel(tp, cancel_flags);
3083 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3086 * Only increment the parent directory vnode count if
3087 * we didn't bump it in looking up cdp. The only time
3088 * we don't bump it is when we're looking up ".".
3094 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3096 ASSERT(cdp->i_d.di_nlink >= 2);
3097 if (cdp->i_d.di_nlink != 2) {
3098 error = XFS_ERROR(ENOTEMPTY);
3101 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3102 error = XFS_ERROR(ENOTEMPTY);
3106 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3107 &first_block, &free_list, resblks);
3112 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3115 * Bump the in memory generation count on the parent
3116 * directory so that other can know that it has changed.
3121 * Drop the link from cdp's "..".
3123 error = xfs_droplink(tp, dp);
3129 * Drop the link from dp to cdp.
3131 error = xfs_droplink(tp, cdp);
3137 * Drop the "." link from cdp to self.
3139 error = xfs_droplink(tp, cdp);
3144 /* Determine these before committing transaction */
3145 last_cdp_link = (cdp)->i_d.di_nlink==0;
3148 * Take an extra ref on the child vnode so that it
3149 * does not go to xfs_inactive() from within the commit.
3154 * If this is a synchronous mount, make sure that the
3155 * rmdir transaction goes to disk before returning to
3158 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3159 xfs_trans_set_sync(tp);
3162 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3164 xfs_bmap_cancel(&free_list);
3165 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3171 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3179 * Let interposed file systems know about removed links.
3181 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3185 /* Fall through to std_return with error = 0 or the errno
3186 * from xfs_trans_commit. */
3188 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3189 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3190 dir_vp, DM_RIGHT_NULL,
3191 NULL, DM_RIGHT_NULL,
3192 name, NULL, dm_di_mode,
3198 xfs_bmap_cancel(&free_list);
3199 cancel_flags |= XFS_TRANS_ABORT;
3201 xfs_trans_cancel(tp, cancel_flags);
3209 * Read dp's entries starting at uiop->uio_offset and translate them into
3210 * bufsize bytes worth of struct dirents starting at bufbase.
3214 bhv_desc_t *dir_bdp,
3220 xfs_trans_t *tp = NULL;
3223 xfs_off_t start_offset;
3225 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3226 (inst_t *)__return_address);
3227 dp = XFS_BHVTOI(dir_bdp);
3229 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3230 return XFS_ERROR(EIO);
3233 lock_mode = xfs_ilock_map_shared(dp);
3234 start_offset = uiop->uio_offset;
3235 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3236 if (start_offset != uiop->uio_offset) {
3237 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3239 xfs_iunlock_map_shared(dp, lock_mode);
3250 bhv_desc_t *dir_bdp,
3263 xfs_bmap_free_t free_list;
3264 xfs_fsblock_t first_block;
3265 boolean_t dp_joined_to_trans;
3269 xfs_fileoff_t first_fsb;
3270 xfs_filblks_t fs_blocks;
3272 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3279 struct xfs_dquot *udqp, *gdqp;
3281 char *link_name = VNAME(dentry);
3285 dir_vp = BHV_TO_VNODE(dir_bdp);
3286 dp = XFS_BHVTOI(dir_bdp);
3287 dp_joined_to_trans = B_FALSE;
3292 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3296 if (XFS_FORCED_SHUTDOWN(mp))
3297 return XFS_ERROR(EIO);
3299 link_namelen = VNAMELEN(dentry);
3302 * Check component lengths of the target path name.
3304 pathlen = strlen(target_path);
3305 if (pathlen >= MAXPATHLEN) /* total string too long */
3306 return XFS_ERROR(ENAMETOOLONG);
3307 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3311 for(total = 0, path = target_path; total < pathlen;) {
3315 while(*path == '/') {
3321 * Count up to the next slash or end of path.
3322 * Error out if the component is bigger than MAXNAMELEN.
3324 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3325 if (++len >= MAXNAMELEN) {
3326 error = ENAMETOOLONG;
3333 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3334 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3335 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3336 link_name, target_path, 0, 0, 0);
3341 /* Return through std_return after this point. */
3344 if (vap->va_mask & XFS_AT_PROJID)
3345 prid = (xfs_prid_t)vap->va_projid;
3347 prid = (xfs_prid_t)dfltprid;
3350 * Make sure that we have allocated dquot(s) on disk.
3352 error = XFS_QM_DQVOPALLOC(mp, dp,
3353 current_fsuid(credp), current_fsgid(credp),
3354 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3358 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3359 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3361 * The symlink will fit into the inode data fork?
3362 * There can't be any attributes so we get the whole variable part.
3364 if (pathlen <= XFS_LITINO(mp))
3367 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3368 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3369 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3370 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3371 if (error == ENOSPC && fs_blocks == 0) {
3373 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3374 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3382 xfs_ilock(dp, XFS_ILOCK_EXCL);
3385 * Reserve disk quota : blocks and inode.
3387 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3392 * Check for ability to enter directory entry, if no space reserved.
3395 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3398 * Initialize the bmap freelist prior to calling either
3399 * bmapi or the directory create code.
3401 XFS_BMAP_INIT(&free_list, &first_block);
3404 * Allocate an inode for the symlink.
3406 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3407 1, 0, credp, prid, resblks > 0, &ip, NULL);
3409 if (error == ENOSPC)
3416 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3417 dp_joined_to_trans = B_TRUE;
3420 * Also attach the dquot(s) to it, if applicable.
3422 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3425 resblks -= XFS_IALLOC_SPACE_RES(mp);
3427 * If the symlink will fit into the inode, write it inline.
3429 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3430 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3431 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3432 ip->i_d.di_size = pathlen;
3435 * The inode was initially created in extent format.
3437 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3438 ip->i_df.if_flags |= XFS_IFINLINE;
3440 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3441 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3445 nmaps = SYMLINK_MAPS;
3447 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3448 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3449 &first_block, resblks, mval, &nmaps,
3456 resblks -= fs_blocks;
3457 ip->i_d.di_size = pathlen;
3458 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3460 cur_chunk = target_path;
3461 for (n = 0; n < nmaps; n++) {
3462 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3463 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3464 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3465 BTOBB(byte_cnt), 0);
3466 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3467 if (pathlen < byte_cnt) {
3470 pathlen -= byte_cnt;
3472 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3473 cur_chunk += byte_cnt;
3475 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3480 * Create the directory entry for the symlink.
3482 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3483 ip->i_ino, &first_block, &free_list, resblks);
3487 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3488 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3491 * Bump the in memory version number of the parent directory
3492 * so that other processes accessing it will recognize that
3493 * the directory has changed.
3498 * If this is a synchronous mount, make sure that the
3499 * symlink transaction goes to disk before returning to
3502 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3503 xfs_trans_set_sync(tp);
3507 * xfs_trans_commit normally decrements the vnode ref count
3508 * when it unlocks the inode. Since we want to return the
3509 * vnode to the caller, we bump the vnode ref count now.
3513 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3517 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3518 XFS_QM_DQRELE(mp, udqp);
3519 XFS_QM_DQRELE(mp, gdqp);
3521 /* Fall through to std_return with error = 0 or errno from
3522 * xfs_trans_commit */
3524 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3525 DM_EVENT_POSTSYMLINK)) {
3526 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3527 dir_vp, DM_RIGHT_NULL,
3528 error ? NULL : XFS_ITOV(ip),
3529 DM_RIGHT_NULL, link_name, target_path,
3545 xfs_bmap_cancel(&free_list);
3546 cancel_flags |= XFS_TRANS_ABORT;
3548 xfs_trans_cancel(tp, cancel_flags);
3549 XFS_QM_DQRELE(mp, udqp);
3550 XFS_QM_DQRELE(mp, gdqp);
3552 if (!dp_joined_to_trans && (dp != NULL)) {
3553 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3563 * A fid routine that takes a pointer to a previously allocated
3564 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3574 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3575 (inst_t *)__return_address);
3576 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3578 xfid = (xfs_fid2_t *)fidp;
3579 ip = XFS_BHVTOI(bdp);
3580 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3583 * use memcpy because the inode is a long long and there's no
3584 * assurance that xfid->fid_ino is properly aligned.
3586 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3587 xfid->fid_gen = ip->i_d.di_gen;
3604 vp = BHV_TO_VNODE(bdp);
3605 if (vp->v_type == VDIR)
3607 ip = XFS_BHVTOI(bdp);
3608 if (locktype == VRWLOCK_WRITE) {
3609 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3610 } else if (locktype == VRWLOCK_TRY_READ) {
3611 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3612 } else if (locktype == VRWLOCK_TRY_WRITE) {
3613 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3615 ASSERT((locktype == VRWLOCK_READ) ||
3616 (locktype == VRWLOCK_WRITE_DIRECT));
3617 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3635 vp = BHV_TO_VNODE(bdp);
3636 if (vp->v_type == VDIR)
3638 ip = XFS_BHVTOI(bdp);
3639 if (locktype == VRWLOCK_WRITE) {
3641 * In the write case, we may have added a new entry to
3642 * the reference cache. This might store a pointer to
3643 * an inode to be released in this inode. If it is there,
3644 * clear the pointer and release the inode after unlocking
3647 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3649 ASSERT((locktype == VRWLOCK_READ) ||
3650 (locktype == VRWLOCK_WRITE_DIRECT));
3651 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3665 ip = XFS_BHVTOI(bdp);
3668 if (XFS_FORCED_SHUTDOWN(mp))
3669 return XFS_ERROR(EIO);
3671 /* Bypass inodes which have already been cleaned by
3672 * the inode flush clustering code inside xfs_iflush
3674 if ((ip->i_update_core == 0) &&
3675 ((ip->i_itemp == NULL) ||
3676 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)))
3679 if (flags & FLUSH_LOG) {
3680 xfs_inode_log_item_t *iip = ip->i_itemp;
3682 if (iip && iip->ili_last_lsn) {
3683 xlog_t *log = mp->m_log;
3685 int s, log_flags = XFS_LOG_FORCE;
3687 s = GRANT_LOCK(log);
3688 sync_lsn = log->l_last_sync_lsn;
3689 GRANT_UNLOCK(log, s);
3691 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3694 if (flags & FLUSH_SYNC)
3695 log_flags |= XFS_LOG_SYNC;
3696 return xfs_log_force(mp, iip->ili_last_lsn,
3701 /* We make this non-blocking if the inode is contended,
3702 * return EAGAIN to indicate to the caller that they
3703 * did not succeed. This prevents the flush path from
3704 * blocking on inodes inside another operation right
3705 * now, they get caught later by xfs_sync.
3707 if (flags & FLUSH_INODE) {
3710 if (xfs_ipincount(ip))
3713 if (flags & FLUSH_SYNC) {
3714 xfs_ilock(ip, XFS_ILOCK_SHARED);
3716 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3717 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3718 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3725 if (flags & FLUSH_SYNC)
3726 flush_flags = XFS_IFLUSH_SYNC;
3728 flush_flags = XFS_IFLUSH_ASYNC;
3730 error = xfs_iflush(ip, flush_flags);
3731 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3750 if (!capable(CAP_SYS_ADMIN))
3751 return XFS_ERROR(EPERM);
3753 ip = XFS_BHVTOI(bdp);
3756 if (XFS_FORCED_SHUTDOWN(mp))
3757 return XFS_ERROR(EIO);
3759 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3760 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3762 xfs_trans_cancel(tp, 0);
3765 xfs_ilock(ip, XFS_ILOCK_EXCL);
3766 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3768 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3769 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3771 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3773 error = xfs_trans_commit(tp, 0, NULL);
3789 vp = BHV_TO_VNODE(bdp);
3791 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3793 ASSERT(!VN_MAPPED(vp));
3794 ip = XFS_BHVTOI(bdp);
3796 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3797 if (ip->i_d.di_size > 0) {
3799 * Flush and invalidate any data left around that is
3800 * a part of this file.
3802 * Get the inode's i/o lock so that buffers are pushed
3803 * out while holding the proper lock. We can't hold
3804 * the inode lock here since flushing out buffers may
3805 * cause us to try to get the lock in xfs_strategy().
3807 * We don't have to call remapf() here, because there
3808 * cannot be any mapped file references to this vnode
3809 * since it is being reclaimed.
3811 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3814 * If we hit an IO error, we need to make sure that the
3815 * buffer and page caches of file data for
3816 * the file are tossed away. We don't want to use
3817 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3818 * pages to stay attached to the vnode, but be
3819 * marked P_BAD. pdflush/vnode_pagebad
3822 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3823 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3825 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3828 ASSERT(VN_CACHED(vp) == 0);
3829 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3830 ip->i_delayed_blks == 0);
3831 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3832 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3834 * di_size field may not be quite accurate if we're
3837 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3838 ASSERT(VN_CACHED(vp) == 0);
3842 /* If we have nothing to flush with this inode then complete the
3843 * teardown now, otherwise break the link between the xfs inode
3844 * and the linux inode and clean up the xfs inode later. This
3845 * avoids flushing the inode to disk during the delete operation
3848 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3849 xfs_ilock(ip, XFS_ILOCK_EXCL);
3851 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3853 xfs_mount_t *mp = ip->i_mount;
3855 /* Protect sync from us */
3856 XFS_MOUNT_ILOCK(mp);
3857 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3858 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3859 ip->i_flags |= XFS_IRECLAIMABLE;
3860 XFS_MOUNT_IUNLOCK(mp);
3871 xfs_ihash_t *ih = ip->i_hash;
3874 /* The hash lock here protects a thread in xfs_iget_core from
3875 * racing with us on linking the inode back with a vnode.
3876 * Once we have the XFS_IRECLAIM flag set it will not touch
3879 write_lock(&ih->ih_lock);
3880 if ((ip->i_flags & XFS_IRECLAIM) ||
3881 (!(ip->i_flags & XFS_IRECLAIMABLE) &&
3882 (XFS_ITOV_NULL(ip) == NULL))) {
3883 write_unlock(&ih->ih_lock);
3886 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3890 ip->i_flags |= XFS_IRECLAIM;
3891 write_unlock(&ih->ih_lock);
3894 * If the inode is still dirty, then flush it out. If the inode
3895 * is not in the AIL, then it will be OK to flush it delwri as
3896 * long as xfs_iflush() does not keep any references to the inode.
3897 * We leave that decision up to xfs_iflush() since it has the
3898 * knowledge of whether it's OK to simply do a delwri flush of
3899 * the inode or whether we need to wait until the inode is
3900 * pulled from the AIL.
3901 * We get the flush lock regardless, though, just to make sure
3902 * we don't free it while it is being flushed.
3904 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3906 xfs_ilock(ip, XFS_ILOCK_EXCL);
3910 if (ip->i_update_core ||
3911 ((ip->i_itemp != NULL) &&
3912 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3913 error = xfs_iflush(ip, sync_mode);
3915 * If we hit an error, typically because of filesystem
3916 * shutdown, we don't need to let vn_reclaim to know
3917 * because we're gonna reclaim the inode anyway.
3920 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3924 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3927 ASSERT(ip->i_update_core == 0);
3928 ASSERT(ip->i_itemp == NULL ||
3929 ip->i_itemp->ili_format.ilf_fields == 0);
3930 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3931 } else if (locked) {
3933 * We are not interested in doing an iflush if we're
3934 * in the process of shutting down the filesystem forcibly.
3935 * So, just reclaim the inode.
3938 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3946 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3949 struct list_head *curr, *next;
3955 XFS_MOUNT_ILOCK(mp);
3956 list_for_each_safe(curr, next, &mp->m_del_inodes) {
3957 ip = list_entry(curr, xfs_inode_t, i_reclaim);
3959 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3961 if (xfs_ipincount(ip) ||
3962 !xfs_iflock_nowait(ip)) {
3963 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3967 XFS_MOUNT_IUNLOCK(mp);
3968 xfs_finish_reclaim(ip, noblock,
3969 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
3977 XFS_MOUNT_IUNLOCK(mp);
3982 * xfs_alloc_file_space()
3983 * This routine allocates disk space for the given file.
3985 * If alloc_type == 0, this request is for an ALLOCSP type
3986 * request which will change the file size. In this case, no
3987 * DMAPI event will be generated by the call. A TRUNCATE event
3988 * will be generated later by xfs_setattr.
3990 * If alloc_type != 0, this request is for a RESVSP type
3991 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3992 * lower block boundary byte address is less than the file's
4001 xfs_alloc_file_space(
4008 xfs_filblks_t allocated_fsb;
4009 xfs_filblks_t allocatesize_fsb;
4012 xfs_filblks_t datablocks;
4014 xfs_fsblock_t firstfsb;
4015 xfs_bmap_free_t free_list;
4016 xfs_bmbt_irec_t *imapp;
4017 xfs_bmbt_irec_t imaps[1];
4024 xfs_fileoff_t startoffset_fsb;
4026 int xfs_bmapi_flags;
4028 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4031 if (XFS_FORCED_SHUTDOWN(mp))
4032 return XFS_ERROR(EIO);
4035 * determine if this is a realtime file
4037 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4038 if (ip->i_d.di_extsize)
4039 rtextsize = ip->i_d.di_extsize;
4041 rtextsize = mp->m_sb.sb_rextsize;
4045 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4049 return XFS_ERROR(EINVAL);
4055 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4056 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4057 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4059 /* Generate a DMAPI event if needed. */
4060 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4061 (attr_flags&ATTR_DMI) == 0 &&
4062 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4063 xfs_off_t end_dmi_offset;
4065 end_dmi_offset = offset+len;
4066 if (end_dmi_offset > ip->i_d.di_size)
4067 end_dmi_offset = ip->i_d.di_size;
4068 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4069 offset, end_dmi_offset - offset,
4076 * allocate file space until done or until there is an error
4079 while (allocatesize_fsb && !error) {
4081 * determine if reserving space on
4082 * the data or realtime partition.
4087 s = startoffset_fsb;
4088 do_div(s, rtextsize);
4090 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4092 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4095 datablocks = allocatesize_fsb;
4100 * allocate and setup the transaction
4102 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4103 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4104 error = xfs_trans_reserve(tp,
4106 XFS_WRITE_LOG_RES(mp),
4108 XFS_TRANS_PERM_LOG_RES,
4109 XFS_WRITE_LOG_COUNT);
4112 * check for running out of space
4116 * Free the transaction structure.
4118 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4119 xfs_trans_cancel(tp, 0);
4122 xfs_ilock(ip, XFS_ILOCK_EXCL);
4123 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4124 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4125 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4129 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4130 xfs_trans_ihold(tp, ip);
4133 * issue the bmapi() call to allocate the blocks
4135 XFS_BMAP_INIT(&free_list, &firstfsb);
4136 error = xfs_bmapi(tp, ip, startoffset_fsb,
4137 allocatesize_fsb, xfs_bmapi_flags,
4138 &firstfsb, 0, imapp, &reccount,
4145 * complete the transaction
4147 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4152 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4153 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4158 allocated_fsb = imapp->br_blockcount;
4160 if (reccount == 0) {
4161 error = XFS_ERROR(ENOSPC);
4165 startoffset_fsb += allocated_fsb;
4166 allocatesize_fsb -= allocated_fsb;
4169 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4170 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4172 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4173 XFS_ITOV(ip), DM_RIGHT_NULL,
4174 XFS_ITOV(ip), DM_RIGHT_NULL,
4175 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4177 goto retry; /* Maybe DMAPI app. has made space */
4178 /* else fall through with error from XFS_SEND_DATA */
4184 xfs_bmap_cancel(&free_list);
4186 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4187 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4188 goto dmapi_enospc_check;
4192 * Zero file bytes between startoff and endoff inclusive.
4193 * The iolock is held exclusive and no blocks are buffered.
4196 xfs_zero_remaining_bytes(
4201 xfs_bmbt_irec_t imap;
4202 xfs_fileoff_t offset_fsb;
4203 xfs_off_t lastoffset;
4206 xfs_mount_t *mp = ip->i_mount;
4210 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4211 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4212 mp->m_rtdev_targp : mp->m_ddev_targp);
4214 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4215 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4217 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4219 if (error || nimap < 1)
4221 ASSERT(imap.br_blockcount >= 1);
4222 ASSERT(imap.br_startoff == offset_fsb);
4223 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4224 if (lastoffset > endoff)
4225 lastoffset = endoff;
4226 if (imap.br_startblock == HOLESTARTBLOCK)
4228 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4229 if (imap.br_state == XFS_EXT_UNWRITTEN)
4232 XFS_BUF_UNWRITE(bp);
4234 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4236 if ((error = xfs_iowait(bp))) {
4237 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4238 mp, bp, XFS_BUF_ADDR(bp));
4241 memset(XFS_BUF_PTR(bp) +
4242 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4243 0, lastoffset - offset + 1);
4248 if ((error = xfs_iowait(bp))) {
4249 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4250 mp, bp, XFS_BUF_ADDR(bp));
4259 * xfs_free_file_space()
4260 * This routine frees disk space for the given file.
4262 * This routine is only called by xfs_change_file_space
4263 * for an UNRESVSP type call.
4271 xfs_free_file_space(
4279 xfs_off_t end_dmi_offset;
4280 xfs_fileoff_t endoffset_fsb;
4282 xfs_fsblock_t firstfsb;
4283 xfs_bmap_free_t free_list;
4285 xfs_bmbt_irec_t imap;
4293 xfs_fileoff_t startoffset_fsb;
4296 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4299 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4303 if (len <= 0) /* if nothing being freed */
4305 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4306 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4307 end_dmi_offset = offset + len;
4308 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4310 if (offset < ip->i_d.di_size &&
4311 (attr_flags & ATTR_DMI) == 0 &&
4312 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4313 if (end_dmi_offset > ip->i_d.di_size)
4314 end_dmi_offset = ip->i_d.di_size;
4315 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4316 offset, end_dmi_offset - offset,
4317 AT_DELAY_FLAG(attr_flags), NULL);
4322 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4323 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4325 ilen = len + (offset & (rounding - 1));
4326 ioffset = offset & ~(rounding - 1);
4327 if (ilen & (rounding - 1))
4328 ilen = (ilen + rounding) & ~(rounding - 1);
4329 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4331 * Need to zero the stuff we're not freeing, on disk.
4332 * If its a realtime file & can't use unwritten extents then we
4333 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4334 * will take care of it for us.
4336 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4338 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4339 &imap, &nimap, NULL);
4342 ASSERT(nimap == 0 || nimap == 1);
4343 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4346 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4347 block = imap.br_startblock;
4348 mod = do_div(block, mp->m_sb.sb_rextsize);
4350 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4353 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4354 &imap, &nimap, NULL);
4357 ASSERT(nimap == 0 || nimap == 1);
4358 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4359 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4361 if (mod && (mod != mp->m_sb.sb_rextsize))
4362 endoffset_fsb -= mod;
4365 if ((done = (endoffset_fsb <= startoffset_fsb)))
4367 * One contiguous piece to clear
4369 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4372 * Some full blocks, possibly two pieces to clear
4374 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4375 error = xfs_zero_remaining_bytes(ip, offset,
4376 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4378 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4379 error = xfs_zero_remaining_bytes(ip,
4380 XFS_FSB_TO_B(mp, endoffset_fsb),
4385 * free file space until done or until there is an error
4387 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4388 while (!error && !done) {
4391 * allocate and setup the transaction
4393 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4394 error = xfs_trans_reserve(tp,
4396 XFS_WRITE_LOG_RES(mp),
4398 XFS_TRANS_PERM_LOG_RES,
4399 XFS_WRITE_LOG_COUNT);
4402 * check for running out of space
4406 * Free the transaction structure.
4408 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4409 xfs_trans_cancel(tp, 0);
4412 xfs_ilock(ip, XFS_ILOCK_EXCL);
4413 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4414 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4415 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4419 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4420 xfs_trans_ihold(tp, ip);
4423 * issue the bunmapi() call to free the blocks
4425 XFS_BMAP_INIT(&free_list, &firstfsb);
4426 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4427 endoffset_fsb - startoffset_fsb,
4428 0, 2, &firstfsb, &free_list, &done);
4434 * complete the transaction
4436 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4441 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4442 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4445 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4449 xfs_bmap_cancel(&free_list);
4451 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4452 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
4457 * xfs_change_file_space()
4458 * This routine allocates or frees disk space for the given file.
4459 * The user specified parameters are checked for alignment and size
4468 xfs_change_file_space(
4482 xfs_off_t startoffset;
4488 vp = BHV_TO_VNODE(bdp);
4489 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4491 ip = XFS_BHVTOI(bdp);
4495 * must be a regular file and have write permission
4497 if (vp->v_type != VREG)
4498 return XFS_ERROR(EINVAL);
4500 xfs_ilock(ip, XFS_ILOCK_SHARED);
4502 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4503 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4507 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4509 switch (bf->l_whence) {
4510 case 0: /*SEEK_SET*/
4512 case 1: /*SEEK_CUR*/
4513 bf->l_start += offset;
4515 case 2: /*SEEK_END*/
4516 bf->l_start += ip->i_d.di_size;
4519 return XFS_ERROR(EINVAL);
4522 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4524 if ( (bf->l_start < 0)
4525 || (bf->l_start > XFS_MAXIOFFSET(mp))
4526 || (bf->l_start + llen < 0)
4527 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4528 return XFS_ERROR(EINVAL);
4532 startoffset = bf->l_start;
4533 fsize = ip->i_d.di_size;
4536 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4538 * These calls do NOT zero the data space allocated to the file,
4539 * nor do they change the file size.
4541 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4543 * These calls cause the new file data to be zeroed and the file
4544 * size to be changed.
4546 setprealloc = clrprealloc = 0;
4549 case XFS_IOC_RESVSP:
4550 case XFS_IOC_RESVSP64:
4551 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4558 case XFS_IOC_UNRESVSP:
4559 case XFS_IOC_UNRESVSP64:
4560 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4565 case XFS_IOC_ALLOCSP:
4566 case XFS_IOC_ALLOCSP64:
4567 case XFS_IOC_FREESP:
4568 case XFS_IOC_FREESP64:
4569 if (startoffset > fsize) {
4570 error = xfs_alloc_file_space(ip, fsize,
4571 startoffset - fsize, 0, attr_flags);
4576 va.va_mask = XFS_AT_SIZE;
4577 va.va_size = startoffset;
4579 error = xfs_setattr(bdp, &va, attr_flags, credp);
4589 return XFS_ERROR(EINVAL);
4593 * update the inode timestamp, mode, and prealloc flag bits
4595 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4597 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4600 xfs_trans_cancel(tp, 0);
4604 xfs_ilock(ip, XFS_ILOCK_EXCL);
4606 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4607 xfs_trans_ihold(tp, ip);
4609 ip->i_d.di_mode &= ~S_ISUID;
4612 * Note that we don't have to worry about mandatory
4613 * file locking being disabled here because we only
4614 * clear the S_ISGID bit if the Group execute bit is
4615 * on, but if it was on then mandatory locking wouldn't
4616 * have been enabled.
4618 if (ip->i_d.di_mode & S_IXGRP)
4619 ip->i_d.di_mode &= ~S_ISGID;
4621 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4624 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4625 else if (clrprealloc)
4626 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4628 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4629 xfs_trans_set_sync(tp);
4631 error = xfs_trans_commit(tp, 0, NULL);
4633 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4638 vnodeops_t xfs_vnodeops = {
4639 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4640 .vop_open = xfs_open,
4641 .vop_read = xfs_read,
4642 #ifdef HAVE_SENDFILE
4643 .vop_sendfile = xfs_sendfile,
4645 .vop_write = xfs_write,
4646 .vop_ioctl = xfs_ioctl,
4647 .vop_getattr = xfs_getattr,
4648 .vop_setattr = xfs_setattr,
4649 .vop_access = xfs_access,
4650 .vop_lookup = xfs_lookup,
4651 .vop_create = xfs_create,
4652 .vop_remove = xfs_remove,
4653 .vop_link = xfs_link,
4654 .vop_rename = xfs_rename,
4655 .vop_mkdir = xfs_mkdir,
4656 .vop_rmdir = xfs_rmdir,
4657 .vop_readdir = xfs_readdir,
4658 .vop_symlink = xfs_symlink,
4659 .vop_readlink = xfs_readlink,
4660 .vop_fsync = xfs_fsync,
4661 .vop_inactive = xfs_inactive,
4662 .vop_fid2 = xfs_fid2,
4663 .vop_rwlock = xfs_rwlock,
4664 .vop_rwunlock = xfs_rwunlock,
4665 .vop_bmap = xfs_bmap,
4666 .vop_reclaim = xfs_reclaim,
4667 .vop_attr_get = xfs_attr_get,
4668 .vop_attr_set = xfs_attr_set,
4669 .vop_attr_remove = xfs_attr_remove,
4670 .vop_attr_list = xfs_attr_list,
4671 .vop_link_removed = (vop_link_removed_t)fs_noval,
4672 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4673 .vop_tosspages = fs_tosspages,
4674 .vop_flushinval_pages = fs_flushinval_pages,
4675 .vop_flush_pages = fs_flush_pages,
4676 .vop_release = xfs_release,
4677 .vop_iflush = xfs_inode_flush,