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 if (_MAC_XFS_IACCESS(ip, MACWRITE, credp)) {
415 code = XFS_ERROR(EACCES);
419 /* boolean: are we the file owner? */
420 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
423 * Change various properties of a file.
424 * Only the owner or users with CAP_FOWNER
425 * capability may do these things.
428 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
429 XFS_AT_GID|XFS_AT_PROJID)) {
431 * CAP_FOWNER overrides the following restrictions:
433 * The user ID of the calling process must be equal
434 * to the file owner ID, except in cases where the
435 * CAP_FSETID capability is applicable.
437 if (!file_owner && !capable(CAP_FOWNER)) {
438 code = XFS_ERROR(EPERM);
443 * CAP_FSETID overrides the following restrictions:
445 * The effective user ID of the calling process shall match
446 * the file owner when setting the set-user-ID and
447 * set-group-ID bits on that file.
449 * The effective group ID or one of the supplementary group
450 * IDs of the calling process shall match the group owner of
451 * the file when setting the set-group-ID bit on that file
453 if (mask & XFS_AT_MODE) {
456 if ((vap->va_mode & S_ISUID) && !file_owner)
458 if ((vap->va_mode & S_ISGID) &&
459 !in_group_p((gid_t)ip->i_d.di_gid))
462 /* Linux allows this, Irix doesn't. */
463 if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
466 if (m && !capable(CAP_FSETID))
472 * Change file ownership. Must be the owner or privileged.
473 * If the system was configured with the "restricted_chown"
474 * option, the owner is not permitted to give away the file,
475 * and can change the group id only to a group of which he
476 * or she is a member.
478 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
480 * These IDs could have changed since we last looked at them.
481 * But, we're assured that if the ownership did change
482 * while we didn't have the inode locked, inode's dquot(s)
483 * would have changed also.
485 iuid = ip->i_d.di_uid;
486 iprojid = ip->i_d.di_projid;
487 igid = ip->i_d.di_gid;
488 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
489 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
490 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
494 * CAP_CHOWN overrides the following restrictions:
496 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
497 * shall override the restriction that a process cannot
498 * change the user ID of a file it owns and the restriction
499 * that the group ID supplied to the chown() function
500 * shall be equal to either the group ID or one of the
501 * supplementary group IDs of the calling process.
503 * XXX: How does restricted_chown affect projid?
505 if (restricted_chown &&
506 (iuid != uid || (igid != gid &&
507 !in_group_p((gid_t)gid))) &&
508 !capable(CAP_CHOWN)) {
509 code = XFS_ERROR(EPERM);
513 * Do a quota reservation only if uid or gid is actually
516 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
517 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
519 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
520 capable(CAP_FOWNER) ?
521 XFS_QMOPT_FORCE_RES : 0);
522 if (code) /* out of quota */
528 * Truncate file. Must have write permission and not be a directory.
530 if (mask & XFS_AT_SIZE) {
531 /* Short circuit the truncate case for zero length files */
532 if ((vap->va_size == 0) &&
533 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
534 xfs_iunlock(ip, XFS_ILOCK_EXCL);
535 lock_flags &= ~XFS_ILOCK_EXCL;
536 if (mask & XFS_AT_CTIME)
537 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
542 if (vp->v_type == VDIR) {
543 code = XFS_ERROR(EISDIR);
545 } else if (vp->v_type != VREG) {
546 code = XFS_ERROR(EINVAL);
550 * Make sure that the dquots are attached to the inode.
552 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
557 * Change file access or modified times.
559 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
561 if ((flags & ATTR_UTIME) &&
562 !capable(CAP_FOWNER)) {
563 code = XFS_ERROR(EPERM);
570 * Change extent size or realtime flag.
572 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
574 * Can't change extent size if any extents are allocated.
576 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
577 (mask & XFS_AT_EXTSIZE) &&
578 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
580 code = XFS_ERROR(EINVAL); /* EFBIG? */
585 * Can't set extent size unless the file is marked, or
586 * about to be marked as a realtime file.
588 * This check will be removed when fixed size extents
589 * with buffered data writes is implemented.
592 if ((mask & XFS_AT_EXTSIZE) &&
593 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
595 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
596 ((mask & XFS_AT_XFLAGS) &&
597 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
598 code = XFS_ERROR(EINVAL);
603 * Can't change realtime flag if any extents are allocated.
605 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
606 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
607 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
608 code = XFS_ERROR(EINVAL); /* EFBIG? */
612 * Extent size must be a multiple of the appropriate block
613 * size, if set at all.
615 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
618 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
619 ((mask & XFS_AT_XFLAGS) &&
620 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
621 size = mp->m_sb.sb_rextsize <<
622 mp->m_sb.sb_blocklog;
624 size = mp->m_sb.sb_blocksize;
626 if (vap->va_extsize % size) {
627 code = XFS_ERROR(EINVAL);
632 * If realtime flag is set then must have realtime data.
634 if ((mask & XFS_AT_XFLAGS) &&
635 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
636 if ((mp->m_sb.sb_rblocks == 0) ||
637 (mp->m_sb.sb_rextsize == 0) ||
638 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
639 code = XFS_ERROR(EINVAL);
645 * Can't modify an immutable/append-only file unless
646 * we have appropriate permission.
648 if ((mask & XFS_AT_XFLAGS) &&
650 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
652 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
653 !capable(CAP_LINUX_IMMUTABLE)) {
654 code = XFS_ERROR(EPERM);
660 * Now we can make the changes. Before we join the inode
661 * to the transaction, if XFS_AT_SIZE is set then take care of
662 * the part of the truncation that must be done without the
663 * inode lock. This needs to be done before joining the inode
664 * to the transaction, because the inode cannot be unlocked
665 * once it is a part of the transaction.
667 if (mask & XFS_AT_SIZE) {
669 if (vap->va_size > ip->i_d.di_size)
670 code = xfs_igrow_start(ip, vap->va_size, credp);
671 xfs_iunlock(ip, XFS_ILOCK_EXCL);
673 code = xfs_itruncate_data(ip, vap->va_size);
676 lock_flags &= ~XFS_ILOCK_EXCL;
677 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
680 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
681 if ((code = xfs_trans_reserve(tp, 0,
682 XFS_ITRUNCATE_LOG_RES(mp), 0,
683 XFS_TRANS_PERM_LOG_RES,
684 XFS_ITRUNCATE_LOG_COUNT))) {
685 xfs_trans_cancel(tp, 0);
686 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
689 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
690 xfs_ilock(ip, XFS_ILOCK_EXCL);
694 xfs_trans_ijoin(tp, ip, lock_flags);
695 xfs_trans_ihold(tp, ip);
698 /* determine whether mandatory locking mode changes */
699 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
702 * Truncate file. Must have write permission and not be a directory.
704 if (mask & XFS_AT_SIZE) {
705 if (vap->va_size > ip->i_d.di_size) {
706 xfs_igrow_finish(tp, ip, vap->va_size,
707 !(flags & ATTR_DMI));
708 } else if ((vap->va_size <= ip->i_d.di_size) ||
709 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
711 * signal a sync transaction unless
712 * we're truncating an already unlinked
713 * file on a wsync filesystem
715 code = xfs_itruncate_finish(&tp, ip,
716 (xfs_fsize_t)vap->va_size,
718 ((ip->i_d.di_nlink != 0 ||
719 !(mp->m_flags & XFS_MOUNT_WSYNC))
726 * Have to do this even if the file's size doesn't change.
728 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
732 * Change file access modes.
734 if (mask & XFS_AT_MODE) {
735 ip->i_d.di_mode &= S_IFMT;
736 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
738 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
739 timeflags |= XFS_ICHGTIME_CHG;
743 * Change file ownership. Must be the owner or privileged.
744 * If the system was configured with the "restricted_chown"
745 * option, the owner is not permitted to give away the file,
746 * and can change the group id only to a group of which he
747 * or she is a member.
749 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
751 * CAP_FSETID overrides the following restrictions:
753 * The set-user-ID and set-group-ID bits of a file will be
754 * cleared upon successful return from chown()
756 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
757 !capable(CAP_FSETID)) {
758 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
762 * Change the ownerships and register quota modifications
763 * in the transaction.
766 if (XFS_IS_UQUOTA_ON(mp)) {
767 ASSERT(mask & XFS_AT_UID);
769 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
770 &ip->i_udquot, udqp);
772 ip->i_d.di_uid = uid;
775 if (XFS_IS_GQUOTA_ON(mp)) {
776 ASSERT(mask & XFS_AT_GID);
778 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
779 &ip->i_gdquot, gdqp);
781 ip->i_d.di_gid = gid;
783 if (iprojid != projid) {
784 ip->i_d.di_projid = projid;
786 * We may have to rev the inode as well as
787 * the superblock version number since projids didn't
788 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
790 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
791 xfs_bump_ino_vers2(tp, ip);
794 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
795 timeflags |= XFS_ICHGTIME_CHG;
800 * Change file access or modified times.
802 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
803 if (mask & XFS_AT_ATIME) {
804 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
805 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
806 ip->i_update_core = 1;
807 timeflags &= ~XFS_ICHGTIME_ACC;
809 if (mask & XFS_AT_MTIME) {
810 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
811 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
812 timeflags &= ~XFS_ICHGTIME_MOD;
813 timeflags |= XFS_ICHGTIME_CHG;
815 if (tp && (flags & ATTR_UTIME))
816 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
820 * Change XFS-added attributes.
822 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
823 if (mask & XFS_AT_EXTSIZE) {
825 * Converting bytes to fs blocks.
827 ip->i_d.di_extsize = vap->va_extsize >>
828 mp->m_sb.sb_blocklog;
830 if (mask & XFS_AT_XFLAGS) {
831 ip->i_d.di_flags = 0;
832 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
833 ip->i_d.di_flags |= XFS_DIFLAG_REALTIME;
834 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
836 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
837 ip->i_d.di_flags |= XFS_DIFLAG_IMMUTABLE;
838 if (vap->va_xflags & XFS_XFLAG_APPEND)
839 ip->i_d.di_flags |= XFS_DIFLAG_APPEND;
840 if (vap->va_xflags & XFS_XFLAG_SYNC)
841 ip->i_d.di_flags |= XFS_DIFLAG_SYNC;
842 if (vap->va_xflags & XFS_XFLAG_NOATIME)
843 ip->i_d.di_flags |= XFS_DIFLAG_NOATIME;
844 if (vap->va_xflags & XFS_XFLAG_NODUMP)
845 ip->i_d.di_flags |= XFS_DIFLAG_NODUMP;
846 /* can't set PREALLOC this way, just ignore it */
848 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
849 timeflags |= XFS_ICHGTIME_CHG;
853 * Change file inode change time only if XFS_AT_CTIME set
854 * AND we have been called by a DMI function.
857 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
858 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
859 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
860 ip->i_update_core = 1;
861 timeflags &= ~XFS_ICHGTIME_CHG;
865 * Send out timestamp changes that need to be set to the
866 * current time. Not done when called by a DMI function.
868 if (timeflags && !(flags & ATTR_DMI))
869 xfs_ichgtime(ip, timeflags);
871 XFS_STATS_INC(xs_ig_attrchg);
874 * If this is a synchronous mount, make sure that the
875 * transaction goes to disk before returning to the user.
876 * This is slightly sub-optimal in that truncates require
877 * two sync transactions instead of one for wsync filesytems.
878 * One for the truncate and one for the timestamps since we
879 * don't want to change the timestamps unless we're sure the
880 * truncate worked. Truncates are less than 1% of the laddis
881 * mix so this probably isn't worth the trouble to optimize.
885 if (mp->m_flags & XFS_MOUNT_WSYNC)
886 xfs_trans_set_sync(tp);
888 code = xfs_trans_commit(tp, commit_flags, NULL);
892 * If the (regular) file's mandatory locking mode changed, then
893 * notify the vnode. We do this under the inode lock to prevent
894 * racing calls to vop_vnode_change.
896 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
897 if (mandlock_before != mandlock_after) {
898 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
902 xfs_iunlock(ip, lock_flags);
905 * Release any dquot(s) the inode had kept before chown.
907 XFS_QM_DQRELE(mp, olddquot1);
908 XFS_QM_DQRELE(mp, olddquot2);
909 XFS_QM_DQRELE(mp, udqp);
910 XFS_QM_DQRELE(mp, gdqp);
916 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
917 !(flags & ATTR_DMI)) {
918 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
919 NULL, DM_RIGHT_NULL, NULL, NULL,
920 0, 0, AT_DELAY_FLAG(flags));
925 commit_flags |= XFS_TRANS_ABORT;
928 XFS_QM_DQRELE(mp, udqp);
929 XFS_QM_DQRELE(mp, gdqp);
931 xfs_trans_cancel(tp, commit_flags);
933 if (lock_flags != 0) {
934 xfs_iunlock(ip, lock_flags);
942 * Null conversion from vnode mode bits to inode mode bits, as in efs.
953 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
954 (inst_t *)__return_address);
956 ip = XFS_BHVTOI(bdp);
957 xfs_ilock(ip, XFS_ILOCK_SHARED);
958 error = xfs_iaccess(ip, mode, credp);
959 xfs_iunlock(ip, XFS_ILOCK_SHARED);
983 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
989 vp = BHV_TO_VNODE(bdp);
990 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
992 ip = XFS_BHVTOI(bdp);
995 if (XFS_FORCED_SHUTDOWN(mp))
996 return XFS_ERROR(EIO);
998 xfs_ilock(ip, XFS_ILOCK_SHARED);
1000 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1002 offset = uiop->uio_offset;
1003 count = uiop->uio_resid;
1006 error = XFS_ERROR(EINVAL);
1014 if (!(ioflags & IO_INVIS)) {
1015 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1019 * See if the symlink is stored inline.
1021 pathlen = (int)ip->i_d.di_size;
1023 if (ip->i_df.if_flags & XFS_IFINLINE) {
1024 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1028 * Symlink not inline. Call bmap to get it in.
1030 nmaps = SYMLINK_MAPS;
1032 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1033 0, NULL, 0, mval, &nmaps, NULL);
1039 for (n = 0; n < nmaps; n++) {
1040 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1041 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1042 bp = xfs_buf_read(mp->m_ddev_targp, d,
1043 BTOBB(byte_cnt), 0);
1044 error = XFS_BUF_GETERROR(bp);
1046 xfs_ioerror_alert("xfs_readlink",
1047 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1051 if (pathlen < byte_cnt)
1053 pathlen -= byte_cnt;
1055 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1064 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1073 * This is called to sync the inode and its data out to disk.
1074 * We need to hold the I/O lock while flushing the data, and
1075 * the inode lock while flushing the inode. The inode lock CANNOT
1076 * be held while flushing the data, so acquire after we're done
1091 vn_trace_entry(BHV_TO_VNODE(bdp),
1092 __FUNCTION__, (inst_t *)__return_address);
1094 ip = XFS_BHVTOI(bdp);
1096 ASSERT(start >= 0 && stop >= -1);
1098 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1099 return XFS_ERROR(EIO);
1102 * We always need to make sure that the required inode state
1103 * is safe on disk. The vnode might be clean but because
1104 * of committed transactions that haven't hit the disk yet.
1105 * Likewise, there could be unflushed non-transactional
1106 * changes to the inode core that have to go to disk.
1108 * The following code depends on one assumption: that
1109 * any transaction that changes an inode logs the core
1110 * because it has to change some field in the inode core
1111 * (typically nextents or nblocks). That assumption
1112 * implies that any transactions against an inode will
1113 * catch any non-transactional updates. If inode-altering
1114 * transactions exist that violate this assumption, the
1115 * code breaks. Right now, it figures that if the involved
1116 * update_* field is clear and the inode is unpinned, the
1117 * inode is clean. Either it's been flushed or it's been
1118 * committed and the commit has hit the disk unpinning the inode.
1119 * (Note that xfs_inode_item_format() called at commit clears
1120 * the update_* fields.)
1122 xfs_ilock(ip, XFS_ILOCK_SHARED);
1124 /* If we are flushing data then we care about update_size
1125 * being set, otherwise we care about update_core
1127 if ((flag & FSYNC_DATA) ?
1128 (ip->i_update_size == 0) :
1129 (ip->i_update_core == 0)) {
1131 * Timestamps/size haven't changed since last inode
1132 * flush or inode transaction commit. That means
1133 * either nothing got written or a transaction
1134 * committed which caught the updates. If the
1135 * latter happened and the transaction hasn't
1136 * hit the disk yet, the inode will be still
1137 * be pinned. If it is, force the log.
1140 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1142 if (xfs_ipincount(ip)) {
1143 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1145 ((flag & FSYNC_WAIT)
1146 ? XFS_LOG_SYNC : 0));
1151 * Kick off a transaction to log the inode
1152 * core to get the updates. Make it
1153 * sync if FSYNC_WAIT is passed in (which
1154 * is done by everybody but specfs). The
1155 * sync transaction will also force the log.
1157 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1158 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1159 if ((error = xfs_trans_reserve(tp, 0,
1160 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1162 xfs_trans_cancel(tp, 0);
1165 xfs_ilock(ip, XFS_ILOCK_EXCL);
1168 * Note - it's possible that we might have pushed
1169 * ourselves out of the way during trans_reserve
1170 * which would flush the inode. But there's no
1171 * guarantee that the inode buffer has actually
1172 * gone out yet (it's delwri). Plus the buffer
1173 * could be pinned anyway if it's part of an
1174 * inode in another recent transaction. So we
1175 * play it safe and fire off the transaction anyway.
1177 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1178 xfs_trans_ihold(tp, ip);
1179 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1180 if (flag & FSYNC_WAIT)
1181 xfs_trans_set_sync(tp);
1182 error = xfs_trans_commit(tp, 0, NULL);
1184 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1190 * This is called by xfs_inactive to free any blocks beyond eof,
1191 * when the link count isn't zero.
1194 xfs_inactive_free_eofblocks(
1200 xfs_fileoff_t end_fsb;
1201 xfs_fileoff_t last_fsb;
1202 xfs_filblks_t map_len;
1204 xfs_bmbt_irec_t imap;
1207 * Figure out if there are any blocks beyond the end
1208 * of the file. If not, then there is nothing to do.
1210 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1211 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1212 map_len = last_fsb - end_fsb;
1217 xfs_ilock(ip, XFS_ILOCK_SHARED);
1218 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1219 NULL, 0, &imap, &nimaps, NULL);
1220 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1222 if (!error && (nimaps != 0) &&
1223 (imap.br_startblock != HOLESTARTBLOCK)) {
1225 * Attach the dquots to the inode up front.
1227 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1231 * There are blocks after the end of file.
1232 * Free them up now by truncating the file to
1235 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1238 * Do the xfs_itruncate_start() call before
1239 * reserving any log space because
1240 * itruncate_start will call into the buffer
1241 * cache and we can't
1242 * do that within a transaction.
1244 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1245 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1248 error = xfs_trans_reserve(tp, 0,
1249 XFS_ITRUNCATE_LOG_RES(mp),
1250 0, XFS_TRANS_PERM_LOG_RES,
1251 XFS_ITRUNCATE_LOG_COUNT);
1253 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1254 xfs_trans_cancel(tp, 0);
1255 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1259 xfs_ilock(ip, XFS_ILOCK_EXCL);
1260 xfs_trans_ijoin(tp, ip,
1263 xfs_trans_ihold(tp, ip);
1265 error = xfs_itruncate_finish(&tp, ip,
1270 * If we get an error at this point we
1271 * simply don't bother truncating the file.
1274 xfs_trans_cancel(tp,
1275 (XFS_TRANS_RELEASE_LOG_RES |
1278 error = xfs_trans_commit(tp,
1279 XFS_TRANS_RELEASE_LOG_RES,
1282 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1288 * Free a symlink that has blocks associated with it.
1291 xfs_inactive_symlink_rmt(
1299 xfs_fsblock_t first_block;
1300 xfs_bmap_free_t free_list;
1303 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1311 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1313 * We're freeing a symlink that has some
1314 * blocks allocated to it. Free the
1315 * blocks here. We know that we've got
1316 * either 1 or 2 extents and that we can
1317 * free them all in one bunmapi call.
1319 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1320 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1321 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1322 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1323 xfs_trans_cancel(tp, 0);
1328 * Lock the inode, fix the size, and join it to the transaction.
1329 * Hold it so in the normal path, we still have it locked for
1330 * the second transaction. In the error paths we need it
1331 * held so the cancel won't rele it, see below.
1333 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1334 size = (int)ip->i_d.di_size;
1335 ip->i_d.di_size = 0;
1336 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1337 xfs_trans_ihold(tp, ip);
1338 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1340 * Find the block(s) so we can inval and unmap them.
1343 XFS_BMAP_INIT(&free_list, &first_block);
1344 nmaps = sizeof(mval) / sizeof(mval[0]);
1345 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1346 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1350 * Invalidate the block(s).
1352 for (i = 0; i < nmaps; i++) {
1353 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1354 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1355 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1356 xfs_trans_binval(tp, bp);
1359 * Unmap the dead block(s) to the free_list.
1361 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1362 &first_block, &free_list, &done)))
1366 * Commit the first transaction. This logs the EFI and the inode.
1368 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1371 * The transaction must have been committed, since there were
1372 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1373 * The new tp has the extent freeing and EFDs.
1377 * The first xact was committed, so add the inode to the new one.
1378 * Mark it dirty so it will be logged and moved forward in the log as
1379 * part of every commit.
1381 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1382 xfs_trans_ihold(tp, ip);
1383 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1385 * Get a new, empty transaction to return to our caller.
1387 ntp = xfs_trans_dup(tp);
1389 * Commit the transaction containing extent freeing and EFD's.
1390 * If we get an error on the commit here or on the reserve below,
1391 * we need to unlock the inode since the new transaction doesn't
1392 * have the inode attached.
1394 error = xfs_trans_commit(tp, 0, NULL);
1397 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1401 * Remove the memory for extent descriptions (just bookkeeping).
1403 if (ip->i_df.if_bytes)
1404 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1405 ASSERT(ip->i_df.if_bytes == 0);
1407 * Put an itruncate log reservation in the new transaction
1410 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1411 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1412 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1416 * Return with the inode locked but not joined to the transaction.
1422 xfs_bmap_cancel(&free_list);
1425 * Have to come here with the inode locked and either
1426 * (held and in the transaction) or (not in the transaction).
1427 * If the inode isn't held then cancel would iput it, but
1428 * that's wrong since this is inactive and the vnode ref
1429 * count is 0 already.
1430 * Cancel won't do anything to the inode if held, but it still
1431 * needs to be locked until the cancel is done, if it was
1432 * joined to the transaction.
1434 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1435 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1442 xfs_inactive_symlink_local(
1448 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1450 * We're freeing a symlink which fit into
1451 * the inode. Just free the memory used
1452 * to hold the old symlink.
1454 error = xfs_trans_reserve(*tpp, 0,
1455 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1456 0, XFS_TRANS_PERM_LOG_RES,
1457 XFS_ITRUNCATE_LOG_COUNT);
1460 xfs_trans_cancel(*tpp, 0);
1464 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1467 * Zero length symlinks _can_ exist.
1469 if (ip->i_df.if_bytes > 0) {
1470 xfs_idata_realloc(ip,
1471 -(ip->i_df.if_bytes),
1473 ASSERT(ip->i_df.if_bytes == 0);
1490 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1493 ASSERT(ip->i_d.di_forkoff != 0);
1494 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1495 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1497 error = xfs_attr_inactive(ip);
1500 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1501 return (error); /* goto out*/
1504 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1505 error = xfs_trans_reserve(tp, 0,
1506 XFS_IFREE_LOG_RES(mp),
1507 0, XFS_TRANS_PERM_LOG_RES,
1508 XFS_INACTIVE_LOG_COUNT);
1510 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1511 xfs_trans_cancel(tp, 0);
1513 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1517 xfs_ilock(ip, XFS_ILOCK_EXCL);
1518 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1519 xfs_trans_ihold(tp, ip);
1520 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1522 ASSERT(ip->i_d.di_anextents == 0);
1537 vp = BHV_TO_VNODE(bdp);
1538 ip = XFS_BHVTOI(bdp);
1540 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1544 /* If this is a read-only mount, don't do this (would generate I/O) */
1545 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1548 #ifdef HAVE_REFCACHE
1549 /* If we are in the NFS reference cache then don't do this now */
1556 if (ip->i_d.di_nlink != 0) {
1557 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1558 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1559 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1560 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1561 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1563 /* Update linux inode block count after free above */
1564 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1565 ip->i_d.di_nblocks + ip->i_delayed_blks);
1575 * This is called when the vnode reference count for the vnode
1576 * goes to zero. If the file has been unlinked, then it must
1577 * now be truncated. Also, we clear all of the read-ahead state
1578 * kept for the inode here since the file is now closed.
1587 xfs_bmap_free_t free_list;
1588 xfs_fsblock_t first_block;
1595 vp = BHV_TO_VNODE(bdp);
1596 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1598 ip = XFS_BHVTOI(bdp);
1601 * If the inode is already free, then there can be nothing
1604 if (ip->i_d.di_mode == 0) {
1605 ASSERT(ip->i_df.if_real_bytes == 0);
1606 ASSERT(ip->i_df.if_broot_bytes == 0);
1607 return VN_INACTIVE_CACHE;
1611 * Only do a truncate if it's a regular file with
1612 * some actual space in it. It's OK to look at the
1613 * inode's fields without the lock because we're the
1614 * only one with a reference to the inode.
1616 truncate = ((ip->i_d.di_nlink == 0) &&
1617 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1618 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1622 if (ip->i_d.di_nlink == 0 &&
1623 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1624 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1629 /* If this is a read-only mount, don't do this (would generate I/O) */
1630 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1633 if (ip->i_d.di_nlink != 0) {
1634 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1635 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1636 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1637 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1638 (ip->i_delayed_blks != 0))) {
1639 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1640 return (VN_INACTIVE_CACHE);
1641 /* Update linux inode block count after free above */
1642 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1643 ip->i_d.di_nblocks + ip->i_delayed_blks);
1648 ASSERT(ip->i_d.di_nlink == 0);
1650 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1651 return (VN_INACTIVE_CACHE);
1653 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1656 * Do the xfs_itruncate_start() call before
1657 * reserving any log space because itruncate_start
1658 * will call into the buffer cache and we can't
1659 * do that within a transaction.
1661 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1663 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1665 error = xfs_trans_reserve(tp, 0,
1666 XFS_ITRUNCATE_LOG_RES(mp),
1667 0, XFS_TRANS_PERM_LOG_RES,
1668 XFS_ITRUNCATE_LOG_COUNT);
1670 /* Don't call itruncate_cleanup */
1671 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1672 xfs_trans_cancel(tp, 0);
1673 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1674 return (VN_INACTIVE_CACHE);
1677 xfs_ilock(ip, XFS_ILOCK_EXCL);
1678 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1679 xfs_trans_ihold(tp, ip);
1682 * normally, we have to run xfs_itruncate_finish sync.
1683 * But if filesystem is wsync and we're in the inactive
1684 * path, then we know that nlink == 0, and that the
1685 * xaction that made nlink == 0 is permanently committed
1686 * since xfs_remove runs as a synchronous transaction.
1688 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1689 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1692 xfs_trans_cancel(tp,
1693 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1694 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1695 return (VN_INACTIVE_CACHE);
1697 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1700 * If we get an error while cleaning up a
1701 * symlink we bail out.
1703 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1704 xfs_inactive_symlink_rmt(ip, &tp) :
1705 xfs_inactive_symlink_local(ip, &tp);
1709 return (VN_INACTIVE_CACHE);
1712 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1713 xfs_trans_ihold(tp, ip);
1715 error = xfs_trans_reserve(tp, 0,
1716 XFS_IFREE_LOG_RES(mp),
1717 0, XFS_TRANS_PERM_LOG_RES,
1718 XFS_INACTIVE_LOG_COUNT);
1720 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1721 xfs_trans_cancel(tp, 0);
1722 return (VN_INACTIVE_CACHE);
1725 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1726 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1727 xfs_trans_ihold(tp, ip);
1731 * If there are attributes associated with the file
1732 * then blow them away now. The code calls a routine
1733 * that recursively deconstructs the attribute fork.
1734 * We need to just commit the current transaction
1735 * because we can't use it for xfs_attr_inactive().
1737 if (ip->i_d.di_anextents > 0) {
1738 error = xfs_inactive_attrs(ip, &tp);
1740 * If we got an error, the transaction is already
1741 * cancelled, and the inode is unlocked. Just get out.
1744 return (VN_INACTIVE_CACHE);
1745 } else if (ip->i_afp) {
1746 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1752 XFS_BMAP_INIT(&free_list, &first_block);
1753 error = xfs_ifree(tp, ip, &free_list);
1756 * If we fail to free the inode, shut down. The cancel
1757 * might do that, we need to make sure. Otherwise the
1758 * inode might be lost for a long time or forever.
1760 if (!XFS_FORCED_SHUTDOWN(mp)) {
1762 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1763 error, mp->m_fsname);
1764 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1766 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1769 * Credit the quota account(s). The inode is gone.
1771 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1774 * Just ignore errors at this point. There is
1775 * nothing we can do except to try to keep going.
1777 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1779 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1782 * Release the dquots held by inode, if any.
1784 XFS_QM_DQDETACH(mp, ip);
1786 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1789 return VN_INACTIVE_CACHE;
1798 bhv_desc_t *dir_bdp,
1805 xfs_inode_t *dp, *ip;
1811 dir_vp = BHV_TO_VNODE(dir_bdp);
1812 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1814 dp = XFS_BHVTOI(dir_bdp);
1816 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1817 return XFS_ERROR(EIO);
1819 lock_mode = xfs_ilock_map_shared(dp);
1820 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1822 *vpp = XFS_ITOV(ip);
1825 xfs_iunlock_map_shared(dp, lock_mode);
1830 #define XFS_CREATE_NEW_MAXTRIES 10000
1833 * xfs_create (create a new file).
1837 bhv_desc_t *dir_bdp,
1843 char *name = VNAME(dentry);
1845 xfs_inode_t *dp, *ip;
1851 xfs_bmap_free_t free_list;
1852 xfs_fsblock_t first_block;
1853 boolean_t dp_joined_to_trans;
1854 int dm_event_sent = 0;
1858 struct xfs_dquot *udqp, *gdqp;
1864 dir_vp = BHV_TO_VNODE(dir_bdp);
1865 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1867 dp = XFS_BHVTOI(dir_bdp);
1870 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1871 namelen = VNAMELEN(dentry);
1873 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1874 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1875 dir_vp, DM_RIGHT_NULL, NULL,
1876 DM_RIGHT_NULL, name, NULL,
1884 if (XFS_FORCED_SHUTDOWN(mp))
1885 return XFS_ERROR(EIO);
1887 /* Return through std_return after this point. */
1890 if (vap->va_mask & XFS_AT_PROJID)
1891 prid = (xfs_prid_t)vap->va_projid;
1893 prid = (xfs_prid_t)dfltprid;
1896 * Make sure that we have allocated dquot(s) on disk.
1898 error = XFS_QM_DQVOPALLOC(mp, dp,
1899 current_fsuid(credp), current_fsgid(credp),
1900 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1905 dp_joined_to_trans = B_FALSE;
1907 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1908 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1909 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1911 * Initially assume that the file does not exist and
1912 * reserve the resources for that case. If that is not
1913 * the case we'll drop the one we have and get a more
1914 * appropriate transaction later.
1916 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1917 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1918 if (error == ENOSPC) {
1920 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1921 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1929 xfs_ilock(dp, XFS_ILOCK_EXCL);
1931 XFS_BMAP_INIT(&free_list, &first_block);
1936 * Reserve disk quota and the inode.
1938 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1943 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1945 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1946 error = xfs_dir_ialloc(&tp, dp,
1947 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1948 rdev, credp, prid, resblks > 0,
1951 if (error == ENOSPC)
1958 * At this point, we've gotten a newly allocated inode.
1959 * It is locked (and joined to the transaction).
1962 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1965 * Now we join the directory inode to the transaction.
1966 * We do not do it earlier because xfs_dir_ialloc
1967 * might commit the previous transaction (and release
1972 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1973 dp_joined_to_trans = B_TRUE;
1975 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1976 &first_block, &free_list,
1977 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1979 ASSERT(error != ENOSPC);
1982 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1983 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1986 * If this is a synchronous mount, make sure that the
1987 * create transaction goes to disk before returning to
1990 if (mp->m_flags & XFS_MOUNT_WSYNC) {
1991 xfs_trans_set_sync(tp);
1997 * Attach the dquot(s) to the inodes and modify them incore.
1998 * These ids of the inode couldn't have changed since the new
1999 * inode has been locked ever since it was created.
2001 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2004 * xfs_trans_commit normally decrements the vnode ref count
2005 * when it unlocks the inode. Since we want to return the
2006 * vnode to the caller, we bump the vnode ref count now.
2011 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2013 xfs_bmap_cancel(&free_list);
2017 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2024 XFS_QM_DQRELE(mp, udqp);
2025 XFS_QM_DQRELE(mp, gdqp);
2028 * Propogate the fact that the vnode changed after the
2029 * xfs_inode locks have been released.
2031 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2035 /* Fallthrough to std_return with error = 0 */
2038 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2039 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2040 DM_EVENT_POSTCREATE)) {
2041 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2042 dir_vp, DM_RIGHT_NULL,
2044 DM_RIGHT_NULL, name, NULL,
2045 dm_di_mode, error, 0);
2050 cancel_flags |= XFS_TRANS_ABORT;
2055 xfs_trans_cancel(tp, cancel_flags);
2057 if (!dp_joined_to_trans && (dp != NULL))
2058 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2059 XFS_QM_DQRELE(mp, udqp);
2060 XFS_QM_DQRELE(mp, gdqp);
2066 * Wait until after the current transaction is aborted to
2067 * release the inode. This prevents recursive transactions
2068 * and deadlocks from xfs_inactive.
2070 cancel_flags |= XFS_TRANS_ABORT;
2071 xfs_trans_cancel(tp, cancel_flags);
2074 XFS_QM_DQRELE(mp, udqp);
2075 XFS_QM_DQRELE(mp, gdqp);
2082 * Some counters to see if (and how often) we are hitting some deadlock
2083 * prevention code paths.
2087 int xfs_rm_lock_delays;
2088 int xfs_rm_attempts;
2092 * The following routine will lock the inodes associated with the
2093 * directory and the named entry in the directory. The locks are
2094 * acquired in increasing inode number.
2096 * If the entry is "..", then only the directory is locked. The
2097 * vnode ref count will still include that from the .. entry in
2100 * There is a deadlock we need to worry about. If the locked directory is
2101 * in the AIL, it might be blocking up the log. The next inode we lock
2102 * could be already locked by another thread waiting for log space (e.g
2103 * a permanent log reservation with a long running transaction (see
2104 * xfs_itruncate_finish)). To solve this, we must check if the directory
2105 * is in the ail and use lock_nowait. If we can't lock, we need to
2106 * drop the inode lock on the directory and try again. xfs_iunlock will
2107 * potentially push the tail if we were holding up the log.
2110 xfs_lock_dir_and_entry(
2113 xfs_inode_t *ip) /* inode of entry 'name' */
2117 xfs_inode_t *ips[2];
2126 xfs_ilock(dp, XFS_ILOCK_EXCL);
2133 * We want to lock in increasing inum. Since we've already
2134 * acquired the lock on the directory, we may need to release
2135 * if if the inum of the entry turns out to be less.
2137 if (e_inum > dp->i_ino) {
2139 * We are already in the right order, so just
2140 * lock on the inode of the entry.
2141 * We need to use nowait if dp is in the AIL.
2144 lp = (xfs_log_item_t *)dp->i_itemp;
2145 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2146 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2153 * Unlock dp and try again.
2154 * xfs_iunlock will try to push the tail
2155 * if the inode is in the AIL.
2158 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2160 if ((attempts % 5) == 0) {
2161 delay(1); /* Don't just spin the CPU */
2163 xfs_rm_lock_delays++;
2169 xfs_ilock(ip, XFS_ILOCK_EXCL);
2171 } else if (e_inum < dp->i_ino) {
2172 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2176 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2178 /* else e_inum == dp->i_ino */
2179 /* This can happen if we're asked to lock /x/..
2180 * the entry is "..", which is also the parent directory.
2188 int xfs_small_retries;
2189 int xfs_middle_retries;
2190 int xfs_lots_retries;
2191 int xfs_lock_delays;
2195 * The following routine will lock n inodes in exclusive mode.
2196 * We assume the caller calls us with the inodes in i_ino order.
2198 * We need to detect deadlock where an inode that we lock
2199 * is in the AIL and we start waiting for another inode that is locked
2200 * by a thread in a long running transaction (such as truncate). This can
2201 * result in deadlock since the long running trans might need to wait
2202 * for the inode we just locked in order to push the tail and free space
2212 int attempts = 0, i, j, try_lock;
2215 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2226 for (; i < inodes; i++) {
2229 if (i && (ips[i] == ips[i-1])) /* Already locked */
2233 * If try_lock is not set yet, make sure all locked inodes
2234 * are not in the AIL.
2235 * If any are, set try_lock to be used later.
2239 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2240 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2241 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2248 * If any of the previous locks we have locked is in the AIL,
2249 * we must TRY to get the second and subsequent locks. If
2250 * we can't get any, we must release all we have
2255 /* try_lock must be 0 if i is 0. */
2257 * try_lock means we have an inode locked
2258 * that is in the AIL.
2261 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2265 * Unlock all previous guys and try again.
2266 * xfs_iunlock will try to push the tail
2267 * if the inode is in the AIL.
2270 for(j = i - 1; j >= 0; j--) {
2273 * Check to see if we've already
2274 * unlocked this one.
2275 * Not the first one going back,
2276 * and the inode ptr is the same.
2278 if ((j != (i - 1)) && ips[j] ==
2282 xfs_iunlock(ips[j], lock_mode);
2285 if ((attempts % 5) == 0) {
2286 delay(1); /* Don't just spin the CPU */
2296 xfs_ilock(ips[i], lock_mode);
2302 if (attempts < 5) xfs_small_retries++;
2303 else if (attempts < 100) xfs_middle_retries++;
2304 else xfs_lots_retries++;
2312 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2313 int remove_which_error_return = 0;
2315 #define REMOVE_DEBUG_TRACE(x)
2316 #endif /* ! DEBUG */
2325 bhv_desc_t *dir_bdp,
2330 char *name = VNAME(dentry);
2331 xfs_inode_t *dp, *ip;
2332 xfs_trans_t *tp = NULL;
2335 xfs_bmap_free_t free_list;
2336 xfs_fsblock_t first_block;
2344 dir_vp = BHV_TO_VNODE(dir_bdp);
2345 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2347 dp = XFS_BHVTOI(dir_bdp);
2350 if (XFS_FORCED_SHUTDOWN(mp))
2351 return XFS_ERROR(EIO);
2353 namelen = VNAMELEN(dentry);
2355 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2356 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2357 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2358 name, NULL, 0, 0, 0);
2363 /* From this point on, return through std_return */
2367 * We need to get a reference to ip before we get our log
2368 * reservation. The reason for this is that we cannot call
2369 * xfs_iget for an inode for which we do not have a reference
2370 * once we've acquired a log reservation. This is because the
2371 * inode we are trying to get might be in xfs_inactive going
2372 * for a log reservation. Since we'll have to wait for the
2373 * inactive code to complete before returning from xfs_iget,
2374 * we need to make sure that we don't have log space reserved
2375 * when we call xfs_iget. Instead we get an unlocked referece
2376 * to the inode before getting our log reservation.
2378 error = xfs_get_dir_entry(dentry, &ip);
2380 REMOVE_DEBUG_TRACE(__LINE__);
2384 dm_di_mode = ip->i_d.di_mode;
2386 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2390 error = XFS_QM_DQATTACH(mp, dp, 0);
2391 if (!error && dp != ip)
2392 error = XFS_QM_DQATTACH(mp, ip, 0);
2394 REMOVE_DEBUG_TRACE(__LINE__);
2399 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2400 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2402 * We try to get the real space reservation first,
2403 * allowing for directory btree deletion(s) implying
2404 * possible bmap insert(s). If we can't get the space
2405 * reservation then we use 0 instead, and avoid the bmap
2406 * btree insert(s) in the directory code by, if the bmap
2407 * insert tries to happen, instead trimming the LAST
2408 * block from the directory.
2410 resblks = XFS_REMOVE_SPACE_RES(mp);
2411 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2412 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2413 if (error == ENOSPC) {
2415 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2416 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2419 ASSERT(error != ENOSPC);
2420 REMOVE_DEBUG_TRACE(__LINE__);
2421 xfs_trans_cancel(tp, 0);
2426 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2428 REMOVE_DEBUG_TRACE(__LINE__);
2429 xfs_trans_cancel(tp, cancel_flags);
2435 * At this point, we've gotten both the directory and the entry
2438 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2441 * Increment vnode ref count only in this case since
2442 * there's an extra vnode reference in the case where
2446 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2449 if ((error = _MAC_XFS_IACCESS(ip, MACWRITE, credp))) {
2450 REMOVE_DEBUG_TRACE(__LINE__);
2455 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2457 XFS_BMAP_INIT(&free_list, &first_block);
2458 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2459 &first_block, &free_list, 0);
2461 ASSERT(error != ENOENT);
2462 REMOVE_DEBUG_TRACE(__LINE__);
2465 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2468 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2470 error = xfs_droplink(tp, ip);
2472 REMOVE_DEBUG_TRACE(__LINE__);
2476 /* Determine if this is the last link while
2477 * we are in the transaction.
2479 link_zero = (ip)->i_d.di_nlink==0;
2482 * Take an extra ref on the inode so that it doesn't
2483 * go to xfs_inactive() from within the commit.
2488 * If this is a synchronous mount, make sure that the
2489 * remove transaction goes to disk before returning to
2492 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2493 xfs_trans_set_sync(tp);
2496 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2498 REMOVE_DEBUG_TRACE(__LINE__);
2502 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2509 * Before we drop our extra reference to the inode, purge it
2510 * from the refcache if it is there. By waiting until afterwards
2511 * to do the IRELE, we ensure that we won't go inactive in the
2512 * xfs_refcache_purge_ip routine (although that would be OK).
2514 xfs_refcache_purge_ip(ip);
2516 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2519 * Let interposed file systems know about removed links.
2521 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2525 /* Fall through to std_return with error = 0 */
2527 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2528 DM_EVENT_POSTREMOVE)) {
2529 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2530 dir_vp, DM_RIGHT_NULL,
2531 NULL, DM_RIGHT_NULL,
2532 name, NULL, dm_di_mode, error, 0);
2537 xfs_bmap_cancel(&free_list);
2538 cancel_flags |= XFS_TRANS_ABORT;
2541 xfs_trans_cancel(tp, cancel_flags);
2546 * In this case make sure to not release the inode until after
2547 * the current transaction is aborted. Releasing it beforehand
2548 * can cause us to go to xfs_inactive and start a recursive
2549 * transaction which can easily deadlock with the current one.
2551 xfs_bmap_cancel(&free_list);
2552 cancel_flags |= XFS_TRANS_ABORT;
2553 xfs_trans_cancel(tp, cancel_flags);
2556 * Before we drop our extra reference to the inode, purge it
2557 * from the refcache if it is there. By waiting until afterwards
2558 * to do the IRELE, we ensure that we won't go inactive in the
2559 * xfs_refcache_purge_ip routine (although that would be OK).
2561 xfs_refcache_purge_ip(ip);
2575 bhv_desc_t *target_dir_bdp,
2580 xfs_inode_t *tdp, *sip;
2583 xfs_inode_t *ips[2];
2585 xfs_bmap_free_t free_list;
2586 xfs_fsblock_t first_block;
2589 vnode_t *target_dir_vp;
2590 bhv_desc_t *src_bdp;
2592 char *target_name = VNAME(dentry);
2595 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2596 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2597 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2599 target_namelen = VNAMELEN(dentry);
2600 if (src_vp->v_type == VDIR)
2601 return XFS_ERROR(EPERM);
2604 * For now, manually find the XFS behavior descriptor for
2605 * the source vnode. If it doesn't exist then something
2606 * is wrong and we should just return an error.
2607 * Eventually we need to figure out how link is going to
2608 * work in the face of stacked vnodes.
2610 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2611 if (src_bdp == NULL) {
2612 return XFS_ERROR(EXDEV);
2614 sip = XFS_BHVTOI(src_bdp);
2615 tdp = XFS_BHVTOI(target_dir_bdp);
2617 if (XFS_FORCED_SHUTDOWN(mp))
2618 return XFS_ERROR(EIO);
2620 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2621 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2622 target_dir_vp, DM_RIGHT_NULL,
2623 src_vp, DM_RIGHT_NULL,
2624 target_name, NULL, 0, 0, 0);
2629 /* Return through std_return after this point. */
2631 error = XFS_QM_DQATTACH(mp, sip, 0);
2632 if (!error && sip != tdp)
2633 error = XFS_QM_DQATTACH(mp, tdp, 0);
2637 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2638 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2639 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2640 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2641 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2642 if (error == ENOSPC) {
2644 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2645 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2652 if (sip->i_ino < tdp->i_ino) {
2660 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2663 * Increment vnode ref counts since xfs_trans_commit &
2664 * xfs_trans_cancel will both unlock the inodes and
2665 * decrement the associated ref counts.
2668 VN_HOLD(target_dir_vp);
2669 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2670 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2673 * If the source has too many links, we can't make any more to it.
2675 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2676 error = XFS_ERROR(EMLINK);
2681 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2685 XFS_BMAP_INIT(&free_list, &first_block);
2687 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2688 sip->i_ino, &first_block, &free_list,
2692 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2694 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2696 error = xfs_bumplink(tp, sip);
2702 * If this is a synchronous mount, make sure that the
2703 * link transaction goes to disk before returning to
2706 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2707 xfs_trans_set_sync(tp);
2710 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2712 xfs_bmap_cancel(&free_list);
2716 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2721 /* Fall through to std_return with error = 0. */
2723 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2724 DM_EVENT_POSTLINK)) {
2725 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2726 target_dir_vp, DM_RIGHT_NULL,
2727 src_vp, DM_RIGHT_NULL,
2728 target_name, NULL, 0, error, 0);
2733 cancel_flags |= XFS_TRANS_ABORT;
2736 xfs_trans_cancel(tp, cancel_flags);
2746 bhv_desc_t *dir_bdp,
2752 char *dir_name = VNAME(dentry);
2754 xfs_inode_t *cdp; /* inode of created dir */
2755 vnode_t *cvp; /* vnode of created dir */
2761 xfs_bmap_free_t free_list;
2762 xfs_fsblock_t first_block;
2764 boolean_t dp_joined_to_trans;
2765 boolean_t created = B_FALSE;
2766 int dm_event_sent = 0;
2768 struct xfs_dquot *udqp, *gdqp;
2773 dir_vp = BHV_TO_VNODE(dir_bdp);
2774 dp = XFS_BHVTOI(dir_bdp);
2777 if (XFS_FORCED_SHUTDOWN(mp))
2778 return XFS_ERROR(EIO);
2780 dir_namelen = VNAMELEN(dentry);
2783 dp_joined_to_trans = B_FALSE;
2784 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2786 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2787 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2788 dir_vp, DM_RIGHT_NULL, NULL,
2789 DM_RIGHT_NULL, dir_name, NULL,
2796 /* Return through std_return after this point. */
2798 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2802 if (vap->va_mask & XFS_AT_PROJID)
2803 prid = (xfs_prid_t)vap->va_projid;
2805 prid = (xfs_prid_t)dfltprid;
2808 * Make sure that we have allocated dquot(s) on disk.
2810 error = XFS_QM_DQVOPALLOC(mp, dp,
2811 current_fsuid(credp), current_fsgid(credp),
2812 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2816 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2817 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2818 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2819 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2820 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2821 if (error == ENOSPC) {
2823 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2824 XFS_TRANS_PERM_LOG_RES,
2825 XFS_MKDIR_LOG_COUNT);
2833 xfs_ilock(dp, XFS_ILOCK_EXCL);
2836 * Check for directory link count overflow.
2838 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2839 error = XFS_ERROR(EMLINK);
2844 * Reserve disk quota and the inode.
2846 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2851 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2854 * create the directory inode.
2856 error = xfs_dir_ialloc(&tp, dp,
2857 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2858 0, credp, prid, resblks > 0,
2861 if (error == ENOSPC)
2868 * Now we add the directory inode to the transaction.
2869 * We waited until now since xfs_dir_ialloc might start
2870 * a new transaction. Had we joined the transaction
2871 * earlier, the locks might have gotten released.
2874 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2875 dp_joined_to_trans = B_TRUE;
2877 XFS_BMAP_INIT(&free_list, &first_block);
2879 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2880 cdp->i_ino, &first_block, &free_list,
2881 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2883 ASSERT(error != ENOSPC);
2886 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2889 * Bump the in memory version number of the parent directory
2890 * so that other processes accessing it will recognize that
2891 * the directory has changed.
2895 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2901 error = xfs_bumplink(tp, dp);
2906 cvp = XFS_ITOV(cdp);
2914 * Attach the dquots to the new inode and modify the icount incore.
2916 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2919 * If this is a synchronous mount, make sure that the
2920 * mkdir transaction goes to disk before returning to
2923 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2924 xfs_trans_set_sync(tp);
2927 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2933 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2934 XFS_QM_DQRELE(mp, udqp);
2935 XFS_QM_DQRELE(mp, gdqp);
2940 /* Fall through to std_return with error = 0 or errno from
2941 * xfs_trans_commit. */
2944 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2945 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2946 DM_EVENT_POSTCREATE)) {
2947 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2948 dir_vp, DM_RIGHT_NULL,
2949 created ? XFS_ITOV(cdp):NULL,
2952 dm_di_mode, error, 0);
2958 xfs_bmap_cancel(&free_list);
2960 cancel_flags |= XFS_TRANS_ABORT;
2962 xfs_trans_cancel(tp, cancel_flags);
2963 XFS_QM_DQRELE(mp, udqp);
2964 XFS_QM_DQRELE(mp, gdqp);
2966 if (!dp_joined_to_trans && (dp != NULL)) {
2967 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2980 bhv_desc_t *dir_bdp,
2984 char *name = VNAME(dentry);
2986 xfs_inode_t *cdp; /* child directory */
2990 xfs_bmap_free_t free_list;
2991 xfs_fsblock_t first_block;
3000 dir_vp = BHV_TO_VNODE(dir_bdp);
3001 dp = XFS_BHVTOI(dir_bdp);
3004 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3006 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3007 return XFS_ERROR(EIO);
3008 namelen = VNAMELEN(dentry);
3010 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3011 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3012 dir_vp, DM_RIGHT_NULL,
3013 NULL, DM_RIGHT_NULL,
3014 name, NULL, 0, 0, 0);
3016 return XFS_ERROR(error);
3019 /* Return through std_return after this point. */
3024 * We need to get a reference to cdp before we get our log
3025 * reservation. The reason for this is that we cannot call
3026 * xfs_iget for an inode for which we do not have a reference
3027 * once we've acquired a log reservation. This is because the
3028 * inode we are trying to get might be in xfs_inactive going
3029 * for a log reservation. Since we'll have to wait for the
3030 * inactive code to complete before returning from xfs_iget,
3031 * we need to make sure that we don't have log space reserved
3032 * when we call xfs_iget. Instead we get an unlocked referece
3033 * to the inode before getting our log reservation.
3035 error = xfs_get_dir_entry(dentry, &cdp);
3037 REMOVE_DEBUG_TRACE(__LINE__);
3041 dm_di_mode = cdp->i_d.di_mode;
3044 * Get the dquots for the inodes.
3046 error = XFS_QM_DQATTACH(mp, dp, 0);
3047 if (!error && dp != cdp)
3048 error = XFS_QM_DQATTACH(mp, cdp, 0);
3051 REMOVE_DEBUG_TRACE(__LINE__);
3055 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3056 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3058 * We try to get the real space reservation first,
3059 * allowing for directory btree deletion(s) implying
3060 * possible bmap insert(s). If we can't get the space
3061 * reservation then we use 0 instead, and avoid the bmap
3062 * btree insert(s) in the directory code by, if the bmap
3063 * insert tries to happen, instead trimming the LAST
3064 * block from the directory.
3066 resblks = XFS_REMOVE_SPACE_RES(mp);
3067 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3068 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3069 if (error == ENOSPC) {
3071 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3072 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3075 ASSERT(error != ENOSPC);
3080 XFS_BMAP_INIT(&free_list, &first_block);
3083 * Now lock the child directory inode and the parent directory
3084 * inode in the proper order. This will take care of validating
3085 * that the directory entry for the child directory inode has
3086 * not changed while we were obtaining a log reservation.
3088 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3090 xfs_trans_cancel(tp, cancel_flags);
3095 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3098 * Only increment the parent directory vnode count if
3099 * we didn't bump it in looking up cdp. The only time
3100 * we don't bump it is when we're looking up ".".
3106 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3108 if ((error = _MAC_XFS_IACCESS(cdp, MACWRITE, credp))) {
3112 ASSERT(cdp->i_d.di_nlink >= 2);
3113 if (cdp->i_d.di_nlink != 2) {
3114 error = XFS_ERROR(ENOTEMPTY);
3117 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3118 error = XFS_ERROR(ENOTEMPTY);
3122 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3123 &first_block, &free_list, resblks);
3128 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3131 * Bump the in memory generation count on the parent
3132 * directory so that other can know that it has changed.
3137 * Drop the link from cdp's "..".
3139 error = xfs_droplink(tp, dp);
3145 * Drop the link from dp to cdp.
3147 error = xfs_droplink(tp, cdp);
3153 * Drop the "." link from cdp to self.
3155 error = xfs_droplink(tp, cdp);
3160 /* Determine these before committing transaction */
3161 last_cdp_link = (cdp)->i_d.di_nlink==0;
3164 * Take an extra ref on the child vnode so that it
3165 * does not go to xfs_inactive() from within the commit.
3170 * If this is a synchronous mount, make sure that the
3171 * rmdir transaction goes to disk before returning to
3174 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3175 xfs_trans_set_sync(tp);
3178 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3180 xfs_bmap_cancel(&free_list);
3181 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3187 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3195 * Let interposed file systems know about removed links.
3197 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3201 /* Fall through to std_return with error = 0 or the errno
3202 * from xfs_trans_commit. */
3204 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3205 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3206 dir_vp, DM_RIGHT_NULL,
3207 NULL, DM_RIGHT_NULL,
3208 name, NULL, dm_di_mode,
3214 xfs_bmap_cancel(&free_list);
3215 cancel_flags |= XFS_TRANS_ABORT;
3217 xfs_trans_cancel(tp, cancel_flags);
3225 * Read dp's entries starting at uiop->uio_offset and translate them into
3226 * bufsize bytes worth of struct dirents starting at bufbase.
3230 bhv_desc_t *dir_bdp,
3236 xfs_trans_t *tp = NULL;
3239 xfs_off_t start_offset;
3241 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3242 (inst_t *)__return_address);
3243 dp = XFS_BHVTOI(dir_bdp);
3245 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3246 return XFS_ERROR(EIO);
3249 lock_mode = xfs_ilock_map_shared(dp);
3250 start_offset = uiop->uio_offset;
3251 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3252 if (start_offset != uiop->uio_offset) {
3253 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3255 xfs_iunlock_map_shared(dp, lock_mode);
3266 bhv_desc_t *dir_bdp,
3279 xfs_bmap_free_t free_list;
3280 xfs_fsblock_t first_block;
3281 boolean_t dp_joined_to_trans;
3285 xfs_fileoff_t first_fsb;
3286 xfs_filblks_t fs_blocks;
3288 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3295 struct xfs_dquot *udqp, *gdqp;
3297 char *link_name = VNAME(dentry);
3301 dir_vp = BHV_TO_VNODE(dir_bdp);
3302 dp = XFS_BHVTOI(dir_bdp);
3303 dp_joined_to_trans = B_FALSE;
3308 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3312 if (XFS_FORCED_SHUTDOWN(mp))
3313 return XFS_ERROR(EIO);
3315 link_namelen = VNAMELEN(dentry);
3318 * Check component lengths of the target path name.
3320 pathlen = strlen(target_path);
3321 if (pathlen >= MAXPATHLEN) /* total string too long */
3322 return XFS_ERROR(ENAMETOOLONG);
3323 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3327 for(total = 0, path = target_path; total < pathlen;) {
3331 while(*path == '/') {
3337 * Count up to the next slash or end of path.
3338 * Error out if the component is bigger than MAXNAMELEN.
3340 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3341 if (++len >= MAXNAMELEN) {
3342 error = ENAMETOOLONG;
3349 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3350 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3351 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3352 link_name, target_path, 0, 0, 0);
3357 /* Return through std_return after this point. */
3360 if (vap->va_mask & XFS_AT_PROJID)
3361 prid = (xfs_prid_t)vap->va_projid;
3363 prid = (xfs_prid_t)dfltprid;
3366 * Make sure that we have allocated dquot(s) on disk.
3368 error = XFS_QM_DQVOPALLOC(mp, dp,
3369 current_fsuid(credp), current_fsgid(credp),
3370 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3374 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3375 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3377 * The symlink will fit into the inode data fork?
3378 * There can't be any attributes so we get the whole variable part.
3380 if (pathlen <= XFS_LITINO(mp))
3383 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3384 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3385 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3386 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3387 if (error == ENOSPC && fs_blocks == 0) {
3389 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3390 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3398 xfs_ilock(dp, XFS_ILOCK_EXCL);
3401 * Reserve disk quota : blocks and inode.
3403 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3408 * Check for ability to enter directory entry, if no space reserved.
3411 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3414 * Initialize the bmap freelist prior to calling either
3415 * bmapi or the directory create code.
3417 XFS_BMAP_INIT(&free_list, &first_block);
3420 * Allocate an inode for the symlink.
3422 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3423 1, 0, credp, prid, resblks > 0, &ip, NULL);
3425 if (error == ENOSPC)
3432 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3433 dp_joined_to_trans = B_TRUE;
3436 * Also attach the dquot(s) to it, if applicable.
3438 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3441 resblks -= XFS_IALLOC_SPACE_RES(mp);
3443 * If the symlink will fit into the inode, write it inline.
3445 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3446 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3447 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3448 ip->i_d.di_size = pathlen;
3451 * The inode was initially created in extent format.
3453 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3454 ip->i_df.if_flags |= XFS_IFINLINE;
3456 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3457 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3461 nmaps = SYMLINK_MAPS;
3463 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3464 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3465 &first_block, resblks, mval, &nmaps,
3472 resblks -= fs_blocks;
3473 ip->i_d.di_size = pathlen;
3474 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3476 cur_chunk = target_path;
3477 for (n = 0; n < nmaps; n++) {
3478 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3479 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3480 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3481 BTOBB(byte_cnt), 0);
3482 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3483 if (pathlen < byte_cnt) {
3486 pathlen -= byte_cnt;
3488 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3489 cur_chunk += byte_cnt;
3491 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3496 * Create the directory entry for the symlink.
3498 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3499 ip->i_ino, &first_block, &free_list, resblks);
3503 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3504 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3507 * Bump the in memory version number of the parent directory
3508 * so that other processes accessing it will recognize that
3509 * the directory has changed.
3514 * If this is a synchronous mount, make sure that the
3515 * symlink transaction goes to disk before returning to
3518 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3519 xfs_trans_set_sync(tp);
3523 * xfs_trans_commit normally decrements the vnode ref count
3524 * when it unlocks the inode. Since we want to return the
3525 * vnode to the caller, we bump the vnode ref count now.
3529 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3533 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3534 XFS_QM_DQRELE(mp, udqp);
3535 XFS_QM_DQRELE(mp, gdqp);
3537 /* Fall through to std_return with error = 0 or errno from
3538 * xfs_trans_commit */
3540 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3541 DM_EVENT_POSTSYMLINK)) {
3542 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3543 dir_vp, DM_RIGHT_NULL,
3544 error ? NULL : XFS_ITOV(ip),
3545 DM_RIGHT_NULL, link_name, target_path,
3561 xfs_bmap_cancel(&free_list);
3562 cancel_flags |= XFS_TRANS_ABORT;
3564 xfs_trans_cancel(tp, cancel_flags);
3565 XFS_QM_DQRELE(mp, udqp);
3566 XFS_QM_DQRELE(mp, gdqp);
3568 if (!dp_joined_to_trans && (dp != NULL)) {
3569 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3579 * A fid routine that takes a pointer to a previously allocated
3580 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3590 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3591 (inst_t *)__return_address);
3592 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3594 xfid = (xfs_fid2_t *)fidp;
3595 ip = XFS_BHVTOI(bdp);
3596 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3599 * use memcpy because the inode is a long long and there's no
3600 * assurance that xfid->fid_ino is properly aligned.
3602 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3603 xfid->fid_gen = ip->i_d.di_gen;
3620 vp = BHV_TO_VNODE(bdp);
3621 if (vp->v_type == VDIR)
3623 ip = XFS_BHVTOI(bdp);
3624 if (locktype == VRWLOCK_WRITE) {
3625 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3626 } else if (locktype == VRWLOCK_TRY_READ) {
3627 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3628 } else if (locktype == VRWLOCK_TRY_WRITE) {
3629 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3631 ASSERT((locktype == VRWLOCK_READ) ||
3632 (locktype == VRWLOCK_WRITE_DIRECT));
3633 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3651 vp = BHV_TO_VNODE(bdp);
3652 if (vp->v_type == VDIR)
3654 ip = XFS_BHVTOI(bdp);
3655 if (locktype == VRWLOCK_WRITE) {
3657 * In the write case, we may have added a new entry to
3658 * the reference cache. This might store a pointer to
3659 * an inode to be released in this inode. If it is there,
3660 * clear the pointer and release the inode after unlocking
3663 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3665 ASSERT((locktype == VRWLOCK_READ) ||
3666 (locktype == VRWLOCK_WRITE_DIRECT));
3667 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3681 ip = XFS_BHVTOI(bdp);
3684 if (XFS_FORCED_SHUTDOWN(mp))
3685 return XFS_ERROR(EIO);
3687 /* Bypass inodes which have already been cleaned by
3688 * the inode flush clustering code inside xfs_iflush
3690 if ((ip->i_update_core == 0) &&
3691 ((ip->i_itemp == NULL) ||
3692 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)))
3695 if (flags & FLUSH_LOG) {
3696 xfs_inode_log_item_t *iip = ip->i_itemp;
3698 if (iip && iip->ili_last_lsn) {
3699 xlog_t *log = mp->m_log;
3701 int s, log_flags = XFS_LOG_FORCE;
3703 s = GRANT_LOCK(log);
3704 sync_lsn = log->l_last_sync_lsn;
3705 GRANT_UNLOCK(log, s);
3707 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3710 if (flags & FLUSH_SYNC)
3711 log_flags |= XFS_LOG_SYNC;
3712 return xfs_log_force(mp, iip->ili_last_lsn,
3717 /* We make this non-blocking if the inode is contended,
3718 * return EAGAIN to indicate to the caller that they
3719 * did not succeed. This prevents the flush path from
3720 * blocking on inodes inside another operation right
3721 * now, they get caught later by xfs_sync.
3723 if (flags & FLUSH_INODE) {
3726 if (xfs_ipincount(ip))
3729 if (flags & FLUSH_SYNC) {
3730 xfs_ilock(ip, XFS_ILOCK_SHARED);
3732 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3733 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3734 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3741 if (flags & FLUSH_SYNC)
3742 flush_flags = XFS_IFLUSH_SYNC;
3744 flush_flags = XFS_IFLUSH_ASYNC;
3746 error = xfs_iflush(ip, flush_flags);
3747 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3766 if (!capable(CAP_SYS_ADMIN))
3767 return XFS_ERROR(EPERM);
3769 ip = XFS_BHVTOI(bdp);
3772 if (XFS_FORCED_SHUTDOWN(mp))
3773 return XFS_ERROR(EIO);
3775 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3776 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3778 xfs_trans_cancel(tp, 0);
3781 xfs_ilock(ip, XFS_ILOCK_EXCL);
3782 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3784 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3785 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3787 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3789 error = xfs_trans_commit(tp, 0, NULL);
3805 vp = BHV_TO_VNODE(bdp);
3807 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3809 ASSERT(!VN_MAPPED(vp));
3810 ip = XFS_BHVTOI(bdp);
3812 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3813 if (ip->i_d.di_size > 0) {
3815 * Flush and invalidate any data left around that is
3816 * a part of this file.
3818 * Get the inode's i/o lock so that buffers are pushed
3819 * out while holding the proper lock. We can't hold
3820 * the inode lock here since flushing out buffers may
3821 * cause us to try to get the lock in xfs_strategy().
3823 * We don't have to call remapf() here, because there
3824 * cannot be any mapped file references to this vnode
3825 * since it is being reclaimed.
3827 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3830 * If we hit an IO error, we need to make sure that the
3831 * buffer and page caches of file data for
3832 * the file are tossed away. We don't want to use
3833 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3834 * pages to stay attached to the vnode, but be
3835 * marked P_BAD. pdflush/vnode_pagebad
3838 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3839 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3841 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3844 ASSERT(VN_CACHED(vp) == 0);
3845 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3846 ip->i_delayed_blks == 0);
3847 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3848 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3850 * di_size field may not be quite accurate if we're
3853 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3854 ASSERT(VN_CACHED(vp) == 0);
3858 /* If we have nothing to flush with this inode then complete the
3859 * teardown now, otherwise break the link between the xfs inode
3860 * and the linux inode and clean up the xfs inode later. This
3861 * avoids flushing the inode to disk during the delete operation
3864 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3865 xfs_ilock(ip, XFS_ILOCK_EXCL);
3867 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3869 xfs_mount_t *mp = ip->i_mount;
3871 /* Protect sync from us */
3872 XFS_MOUNT_ILOCK(mp);
3873 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3874 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3875 ip->i_flags |= XFS_IRECLAIMABLE;
3876 XFS_MOUNT_IUNLOCK(mp);
3887 xfs_ihash_t *ih = ip->i_hash;
3890 /* The hash lock here protects a thread in xfs_iget_core from
3891 * racing with us on linking the inode back with a vnode.
3892 * Once we have the XFS_IRECLAIM flag set it will not touch
3895 write_lock(&ih->ih_lock);
3896 if ((ip->i_flags & XFS_IRECLAIM) ||
3897 (!(ip->i_flags & XFS_IRECLAIMABLE) &&
3898 (XFS_ITOV_NULL(ip) == NULL))) {
3899 write_unlock(&ih->ih_lock);
3902 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3906 ip->i_flags |= XFS_IRECLAIM;
3907 write_unlock(&ih->ih_lock);
3910 * If the inode is still dirty, then flush it out. If the inode
3911 * is not in the AIL, then it will be OK to flush it delwri as
3912 * long as xfs_iflush() does not keep any references to the inode.
3913 * We leave that decision up to xfs_iflush() since it has the
3914 * knowledge of whether it's OK to simply do a delwri flush of
3915 * the inode or whether we need to wait until the inode is
3916 * pulled from the AIL.
3917 * We get the flush lock regardless, though, just to make sure
3918 * we don't free it while it is being flushed.
3920 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3922 xfs_ilock(ip, XFS_ILOCK_EXCL);
3926 if (ip->i_update_core ||
3927 ((ip->i_itemp != NULL) &&
3928 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3929 error = xfs_iflush(ip, sync_mode);
3931 * If we hit an error, typically because of filesystem
3932 * shutdown, we don't need to let vn_reclaim to know
3933 * because we're gonna reclaim the inode anyway.
3936 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3940 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3943 ASSERT(ip->i_update_core == 0);
3944 ASSERT(ip->i_itemp == NULL ||
3945 ip->i_itemp->ili_format.ilf_fields == 0);
3946 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3947 } else if (locked) {
3949 * We are not interested in doing an iflush if we're
3950 * in the process of shutting down the filesystem forcibly.
3951 * So, just reclaim the inode.
3954 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3962 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3965 struct list_head *curr, *next;
3971 XFS_MOUNT_ILOCK(mp);
3972 list_for_each_safe(curr, next, &mp->m_del_inodes) {
3973 ip = list_entry(curr, xfs_inode_t, i_reclaim);
3975 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3977 if (xfs_ipincount(ip) ||
3978 !xfs_iflock_nowait(ip)) {
3979 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3983 XFS_MOUNT_IUNLOCK(mp);
3984 xfs_finish_reclaim(ip, noblock,
3985 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
3993 XFS_MOUNT_IUNLOCK(mp);
3998 * xfs_alloc_file_space()
3999 * This routine allocates disk space for the given file.
4001 * If alloc_type == 0, this request is for an ALLOCSP type
4002 * request which will change the file size. In this case, no
4003 * DMAPI event will be generated by the call. A TRUNCATE event
4004 * will be generated later by xfs_setattr.
4006 * If alloc_type != 0, this request is for a RESVSP type
4007 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4008 * lower block boundary byte address is less than the file's
4017 xfs_alloc_file_space(
4024 xfs_filblks_t allocated_fsb;
4025 xfs_filblks_t allocatesize_fsb;
4028 xfs_filblks_t datablocks;
4030 xfs_fsblock_t firstfsb;
4031 xfs_bmap_free_t free_list;
4032 xfs_bmbt_irec_t *imapp;
4033 xfs_bmbt_irec_t imaps[1];
4040 xfs_fileoff_t startoffset_fsb;
4042 int xfs_bmapi_flags;
4044 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4047 if (XFS_FORCED_SHUTDOWN(mp))
4048 return XFS_ERROR(EIO);
4051 * determine if this is a realtime file
4053 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4054 if (ip->i_d.di_extsize)
4055 rtextsize = ip->i_d.di_extsize;
4057 rtextsize = mp->m_sb.sb_rextsize;
4061 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4065 return XFS_ERROR(EINVAL);
4071 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4072 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4073 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4075 /* Generate a DMAPI event if needed. */
4076 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4077 (attr_flags&ATTR_DMI) == 0 &&
4078 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4079 xfs_off_t end_dmi_offset;
4081 end_dmi_offset = offset+len;
4082 if (end_dmi_offset > ip->i_d.di_size)
4083 end_dmi_offset = ip->i_d.di_size;
4084 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4085 offset, end_dmi_offset - offset,
4092 * allocate file space until done or until there is an error
4095 while (allocatesize_fsb && !error) {
4097 * determine if reserving space on
4098 * the data or realtime partition.
4103 s = startoffset_fsb;
4104 do_div(s, rtextsize);
4106 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4108 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4111 datablocks = allocatesize_fsb;
4116 * allocate and setup the transaction
4118 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4119 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4120 error = xfs_trans_reserve(tp,
4122 XFS_WRITE_LOG_RES(mp),
4124 XFS_TRANS_PERM_LOG_RES,
4125 XFS_WRITE_LOG_COUNT);
4128 * check for running out of space
4132 * Free the transaction structure.
4134 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4135 xfs_trans_cancel(tp, 0);
4138 xfs_ilock(ip, XFS_ILOCK_EXCL);
4139 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4140 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4141 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4145 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4146 xfs_trans_ihold(tp, ip);
4149 * issue the bmapi() call to allocate the blocks
4151 XFS_BMAP_INIT(&free_list, &firstfsb);
4152 error = xfs_bmapi(tp, ip, startoffset_fsb,
4153 allocatesize_fsb, xfs_bmapi_flags,
4154 &firstfsb, 0, imapp, &reccount,
4161 * complete the transaction
4163 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4168 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4169 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4174 allocated_fsb = imapp->br_blockcount;
4176 if (reccount == 0) {
4177 error = XFS_ERROR(ENOSPC);
4181 startoffset_fsb += allocated_fsb;
4182 allocatesize_fsb -= allocated_fsb;
4185 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4186 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4188 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4189 XFS_ITOV(ip), DM_RIGHT_NULL,
4190 XFS_ITOV(ip), DM_RIGHT_NULL,
4191 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4193 goto retry; /* Maybe DMAPI app. has made space */
4194 /* else fall through with error from XFS_SEND_DATA */
4200 xfs_bmap_cancel(&free_list);
4202 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4203 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4204 goto dmapi_enospc_check;
4208 * Zero file bytes between startoff and endoff inclusive.
4209 * The iolock is held exclusive and no blocks are buffered.
4212 xfs_zero_remaining_bytes(
4217 xfs_bmbt_irec_t imap;
4218 xfs_fileoff_t offset_fsb;
4219 xfs_off_t lastoffset;
4222 xfs_mount_t *mp = ip->i_mount;
4226 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4227 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4228 mp->m_rtdev_targp : mp->m_ddev_targp);
4230 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4231 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4233 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4235 if (error || nimap < 1)
4237 ASSERT(imap.br_blockcount >= 1);
4238 ASSERT(imap.br_startoff == offset_fsb);
4239 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4240 if (lastoffset > endoff)
4241 lastoffset = endoff;
4242 if (imap.br_startblock == HOLESTARTBLOCK)
4244 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4245 if (imap.br_state == XFS_EXT_UNWRITTEN)
4248 XFS_BUF_UNWRITE(bp);
4250 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4252 if ((error = xfs_iowait(bp))) {
4253 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4254 mp, bp, XFS_BUF_ADDR(bp));
4257 memset(XFS_BUF_PTR(bp) +
4258 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4259 0, lastoffset - offset + 1);
4264 if ((error = xfs_iowait(bp))) {
4265 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4266 mp, bp, XFS_BUF_ADDR(bp));
4275 * xfs_free_file_space()
4276 * This routine frees disk space for the given file.
4278 * This routine is only called by xfs_change_file_space
4279 * for an UNRESVSP type call.
4287 xfs_free_file_space(
4295 xfs_off_t end_dmi_offset;
4296 xfs_fileoff_t endoffset_fsb;
4298 xfs_fsblock_t firstfsb;
4299 xfs_bmap_free_t free_list;
4301 xfs_bmbt_irec_t imap;
4309 xfs_fileoff_t startoffset_fsb;
4312 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4315 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4319 if (len <= 0) /* if nothing being freed */
4321 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4322 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4323 end_dmi_offset = offset + len;
4324 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4326 if (offset < ip->i_d.di_size &&
4327 (attr_flags & ATTR_DMI) == 0 &&
4328 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4329 if (end_dmi_offset > ip->i_d.di_size)
4330 end_dmi_offset = ip->i_d.di_size;
4331 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4332 offset, end_dmi_offset - offset,
4333 AT_DELAY_FLAG(attr_flags), NULL);
4338 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4339 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4341 ilen = len + (offset & (rounding - 1));
4342 ioffset = offset & ~(rounding - 1);
4343 if (ilen & (rounding - 1))
4344 ilen = (ilen + rounding) & ~(rounding - 1);
4345 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4347 * Need to zero the stuff we're not freeing, on disk.
4348 * If its a realtime file & can't use unwritten extents then we
4349 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4350 * will take care of it for us.
4352 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4354 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4355 &imap, &nimap, NULL);
4358 ASSERT(nimap == 0 || nimap == 1);
4359 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4362 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4363 block = imap.br_startblock;
4364 mod = do_div(block, mp->m_sb.sb_rextsize);
4366 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4369 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4370 &imap, &nimap, NULL);
4373 ASSERT(nimap == 0 || nimap == 1);
4374 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4375 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4377 if (mod && (mod != mp->m_sb.sb_rextsize))
4378 endoffset_fsb -= mod;
4381 if ((done = (endoffset_fsb <= startoffset_fsb)))
4383 * One contiguous piece to clear
4385 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4388 * Some full blocks, possibly two pieces to clear
4390 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4391 error = xfs_zero_remaining_bytes(ip, offset,
4392 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4394 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4395 error = xfs_zero_remaining_bytes(ip,
4396 XFS_FSB_TO_B(mp, endoffset_fsb),
4401 * free file space until done or until there is an error
4403 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4404 while (!error && !done) {
4407 * allocate and setup the transaction
4409 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4410 error = xfs_trans_reserve(tp,
4412 XFS_WRITE_LOG_RES(mp),
4414 XFS_TRANS_PERM_LOG_RES,
4415 XFS_WRITE_LOG_COUNT);
4418 * check for running out of space
4422 * Free the transaction structure.
4424 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4425 xfs_trans_cancel(tp, 0);
4428 xfs_ilock(ip, XFS_ILOCK_EXCL);
4429 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4430 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4431 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4435 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4436 xfs_trans_ihold(tp, ip);
4439 * issue the bunmapi() call to free the blocks
4441 XFS_BMAP_INIT(&free_list, &firstfsb);
4442 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4443 endoffset_fsb - startoffset_fsb,
4444 0, 2, &firstfsb, &free_list, &done);
4450 * complete the transaction
4452 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4457 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4458 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4461 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4465 xfs_bmap_cancel(&free_list);
4467 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4468 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
4473 * xfs_change_file_space()
4474 * This routine allocates or frees disk space for the given file.
4475 * The user specified parameters are checked for alignment and size
4484 xfs_change_file_space(
4498 xfs_off_t startoffset;
4504 vp = BHV_TO_VNODE(bdp);
4505 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4507 ip = XFS_BHVTOI(bdp);
4511 * must be a regular file and have write permission
4513 if (vp->v_type != VREG)
4514 return XFS_ERROR(EINVAL);
4516 xfs_ilock(ip, XFS_ILOCK_SHARED);
4518 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4519 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4523 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4525 switch (bf->l_whence) {
4526 case 0: /*SEEK_SET*/
4528 case 1: /*SEEK_CUR*/
4529 bf->l_start += offset;
4531 case 2: /*SEEK_END*/
4532 bf->l_start += ip->i_d.di_size;
4535 return XFS_ERROR(EINVAL);
4538 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4540 if ( (bf->l_start < 0)
4541 || (bf->l_start > XFS_MAXIOFFSET(mp))
4542 || (bf->l_start + llen < 0)
4543 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4544 return XFS_ERROR(EINVAL);
4548 startoffset = bf->l_start;
4549 fsize = ip->i_d.di_size;
4552 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4554 * These calls do NOT zero the data space allocated to the file,
4555 * nor do they change the file size.
4557 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4559 * These calls cause the new file data to be zeroed and the file
4560 * size to be changed.
4562 setprealloc = clrprealloc = 0;
4565 case XFS_IOC_RESVSP:
4566 case XFS_IOC_RESVSP64:
4567 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4574 case XFS_IOC_UNRESVSP:
4575 case XFS_IOC_UNRESVSP64:
4576 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4581 case XFS_IOC_ALLOCSP:
4582 case XFS_IOC_ALLOCSP64:
4583 case XFS_IOC_FREESP:
4584 case XFS_IOC_FREESP64:
4585 if (startoffset > fsize) {
4586 error = xfs_alloc_file_space(ip, fsize,
4587 startoffset - fsize, 0, attr_flags);
4592 va.va_mask = XFS_AT_SIZE;
4593 va.va_size = startoffset;
4595 error = xfs_setattr(bdp, &va, attr_flags, credp);
4605 return XFS_ERROR(EINVAL);
4609 * update the inode timestamp, mode, and prealloc flag bits
4611 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4613 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4616 xfs_trans_cancel(tp, 0);
4620 xfs_ilock(ip, XFS_ILOCK_EXCL);
4622 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4623 xfs_trans_ihold(tp, ip);
4625 ip->i_d.di_mode &= ~S_ISUID;
4628 * Note that we don't have to worry about mandatory
4629 * file locking being disabled here because we only
4630 * clear the S_ISGID bit if the Group execute bit is
4631 * on, but if it was on then mandatory locking wouldn't
4632 * have been enabled.
4634 if (ip->i_d.di_mode & S_IXGRP)
4635 ip->i_d.di_mode &= ~S_ISGID;
4637 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4640 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4641 else if (clrprealloc)
4642 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4644 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4645 xfs_trans_set_sync(tp);
4647 error = xfs_trans_commit(tp, 0, NULL);
4649 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4654 vnodeops_t xfs_vnodeops = {
4655 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4656 .vop_open = xfs_open,
4657 .vop_read = xfs_read,
4658 #ifdef HAVE_SENDFILE
4659 .vop_sendfile = xfs_sendfile,
4661 .vop_write = xfs_write,
4662 .vop_ioctl = xfs_ioctl,
4663 .vop_getattr = xfs_getattr,
4664 .vop_setattr = xfs_setattr,
4665 .vop_access = xfs_access,
4666 .vop_lookup = xfs_lookup,
4667 .vop_create = xfs_create,
4668 .vop_remove = xfs_remove,
4669 .vop_link = xfs_link,
4670 .vop_rename = xfs_rename,
4671 .vop_mkdir = xfs_mkdir,
4672 .vop_rmdir = xfs_rmdir,
4673 .vop_readdir = xfs_readdir,
4674 .vop_symlink = xfs_symlink,
4675 .vop_readlink = xfs_readlink,
4676 .vop_fsync = xfs_fsync,
4677 .vop_inactive = xfs_inactive,
4678 .vop_fid2 = xfs_fid2,
4679 .vop_rwlock = xfs_rwlock,
4680 .vop_rwunlock = xfs_rwunlock,
4681 .vop_bmap = xfs_bmap,
4682 .vop_reclaim = xfs_reclaim,
4683 .vop_attr_get = xfs_attr_get,
4684 .vop_attr_set = xfs_attr_set,
4685 .vop_attr_remove = xfs_attr_remove,
4686 .vop_attr_list = xfs_attr_list,
4687 .vop_link_removed = (vop_link_removed_t)fs_noval,
4688 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4689 .vop_tosspages = fs_tosspages,
4690 .vop_flushinval_pages = fs_flushinval_pages,
4691 .vop_flush_pages = fs_flush_pages,
4692 .vop_release = xfs_release,
4693 .vop_iflush = xfs_inode_flush,