2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
25 #include "xfs_trans.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_dir_leaf.h"
43 #include "xfs_itable.h"
44 #include "xfs_btree.h"
45 #include "xfs_ialloc.h"
46 #include "xfs_alloc.h"
50 #include "xfs_error.h"
51 #include "xfs_quota.h"
52 #include "xfs_utils.h"
53 #include "xfs_rtalloc.h"
54 #include "xfs_refcache.h"
55 #include "xfs_trans_space.h"
56 #include "xfs_log_priv.h"
62 * The maximum pathlen is 1024 bytes. Since the minimum file system
63 * blocksize is 512 bytes, we can get a max of 2 extents back from
66 #define SYMLINK_MAPS 2
69 * For xfs, we check that the file isn't too big to be opened by this kernel.
70 * No other open action is required for regular files. Devices are handled
71 * through the specfs file system, pipes through fifofs. Device and
72 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
73 * when a new vnode is first looked up or created.
84 vp = BHV_TO_VNODE(bdp);
87 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
88 return XFS_ERROR(EIO);
91 * If it's a directory with any blocks, read-ahead block 0
92 * as we're almost certain to have the next operation be a read there.
94 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
95 mode = xfs_ilock_map_shared(ip);
96 if (ip->i_d.di_nextents > 0)
97 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
98 xfs_iunlock(ip, mode);
118 vp = BHV_TO_VNODE(bdp);
119 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
121 ip = XFS_BHVTOI(bdp);
124 if (XFS_FORCED_SHUTDOWN(mp))
125 return XFS_ERROR(EIO);
127 if (!(flags & ATTR_LAZY))
128 xfs_ilock(ip, XFS_ILOCK_SHARED);
130 vap->va_size = ip->i_d.di_size;
131 if (vap->va_mask == XFS_AT_SIZE)
135 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
136 vap->va_nodeid = ip->i_ino;
138 vap->va_nodeid += mp->m_inoadd;
140 vap->va_nlink = ip->i_d.di_nlink;
143 * Quick exit for non-stat callers
146 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
147 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
151 * Copy from in-core inode.
153 vap->va_mode = ip->i_d.di_mode;
154 vap->va_uid = ip->i_d.di_uid;
155 vap->va_gid = ip->i_d.di_gid;
156 vap->va_projid = ip->i_d.di_projid;
159 * Check vnode type block/char vs. everything else.
161 switch (ip->i_d.di_mode & S_IFMT) {
164 vap->va_rdev = ip->i_df.if_u2.if_rdev;
165 vap->va_blocksize = BLKDEV_IOSIZE;
170 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
171 vap->va_blocksize = xfs_preferred_iosize(mp);
175 * If the file blocks are being allocated from a
176 * realtime partition, then return the inode's
177 * realtime extent size or the realtime volume's
180 vap->va_blocksize = ip->i_d.di_extsize ?
181 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
182 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
187 vn_atime_to_timespec(vp, &vap->va_atime);
188 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
189 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
190 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
191 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
194 * Exit for stat callers. See if any of the rest of the fields
195 * to be filled in are needed.
198 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
199 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
203 * Convert di_flags to xflags.
205 vap->va_xflags = xfs_ip2xflags(ip);
208 * Exit for inode revalidate. See if any of the rest of
209 * the fields to be filled in are needed.
212 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
213 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
216 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
218 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
219 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
223 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
224 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
225 ip->i_d.di_anextents;
227 vap->va_anextents = 0;
228 vap->va_gen = ip->i_d.di_gen;
231 if (!(flags & ATTR_LAZY))
232 xfs_iunlock(ip, XFS_ILOCK_SHARED);
258 xfs_prid_t projid=0, iprojid=0;
259 int mandlock_before, mandlock_after;
260 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
264 vp = BHV_TO_VNODE(bdp);
265 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
267 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
268 return XFS_ERROR(EROFS);
271 * Cannot set certain attributes.
274 if (mask & XFS_AT_NOSET) {
275 return XFS_ERROR(EINVAL);
278 ip = XFS_BHVTOI(bdp);
281 if (XFS_FORCED_SHUTDOWN(mp))
282 return XFS_ERROR(EIO);
285 * Timestamps do not need to be logged and hence do not
286 * need to be done within a transaction.
288 if (mask & XFS_AT_UPDTIMES) {
289 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
290 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
291 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
292 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
293 xfs_ichgtime(ip, timeflags);
297 olddquot1 = olddquot2 = NULL;
301 * If disk quotas is on, we make sure that the dquots do exist on disk,
302 * before we start any other transactions. Trying to do this later
303 * is messy. We don't care to take a readlock to look at the ids
304 * in inode here, because we can't hold it across the trans_reserve.
305 * If the IDs do change before we take the ilock, we're covered
306 * because the i_*dquot fields will get updated anyway.
308 if (XFS_IS_QUOTA_ON(mp) &&
309 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
312 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
314 qflags |= XFS_QMOPT_UQUOTA;
316 uid = ip->i_d.di_uid;
318 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
320 qflags |= XFS_QMOPT_GQUOTA;
322 gid = ip->i_d.di_gid;
324 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
325 projid = vap->va_projid;
326 qflags |= XFS_QMOPT_PQUOTA;
328 projid = ip->i_d.di_projid;
331 * We take a reference when we initialize udqp and gdqp,
332 * so it is important that we never blindly double trip on
333 * the same variable. See xfs_create() for an example.
335 ASSERT(udqp == NULL);
336 ASSERT(gdqp == NULL);
337 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
344 * For the other attributes, we acquire the inode lock and
345 * first do an error checking pass.
348 lock_flags = XFS_ILOCK_EXCL;
349 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
350 if (flags & ATTR_NOLOCK)
352 if (!(mask & XFS_AT_SIZE)) {
353 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
354 (mp->m_flags & XFS_MOUNT_WSYNC)) {
355 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
357 if ((code = xfs_trans_reserve(tp, 0,
358 XFS_ICHANGE_LOG_RES(mp), 0,
365 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
366 !(flags & ATTR_DMI)) {
367 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
368 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
369 vap->va_size, 0, dmflags, NULL);
376 lock_flags |= XFS_IOLOCK_EXCL;
379 xfs_ilock(ip, lock_flags);
381 /* boolean: are we the file owner? */
383 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
385 file_owner = (credp->cr_uid == ip->i_d.di_uid);
389 * Change various properties of a file.
390 * Only the owner or users with CAP_FOWNER
391 * capability may do these things.
394 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
395 XFS_AT_GID|XFS_AT_PROJID)) {
397 * CAP_FOWNER overrides the following restrictions:
399 * The user ID of the calling process must be equal
400 * to the file owner ID, except in cases where the
401 * CAP_FSETID capability is applicable.
403 if (!file_owner && !capable(CAP_FOWNER)) {
404 code = XFS_ERROR(EPERM);
409 * CAP_FSETID overrides the following restrictions:
411 * The effective user ID of the calling process shall match
412 * the file owner when setting the set-user-ID and
413 * set-group-ID bits on that file.
415 * The effective group ID or one of the supplementary group
416 * IDs of the calling process shall match the group owner of
417 * the file when setting the set-group-ID bit on that file
419 if (mask & XFS_AT_MODE) {
422 if ((vap->va_mode & S_ISUID) && !file_owner)
424 if ((vap->va_mode & S_ISGID) &&
425 !groupmember((gid_t)ip->i_d.di_gid, credp))
428 /* Linux allows this, Irix doesn't. */
429 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
432 if (m && !capable(CAP_FSETID))
438 * Change file ownership. Must be the owner or privileged.
439 * If the system was configured with the "restricted_chown"
440 * option, the owner is not permitted to give away the file,
441 * and can change the group id only to a group of which he
442 * or she is a member.
444 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
446 * These IDs could have changed since we last looked at them.
447 * But, we're assured that if the ownership did change
448 * while we didn't have the inode locked, inode's dquot(s)
449 * would have changed also.
451 iuid = ip->i_d.di_uid;
452 iprojid = ip->i_d.di_projid;
453 igid = ip->i_d.di_gid;
454 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
455 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
457 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
461 * CAP_CHOWN overrides the following restrictions:
463 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
464 * shall override the restriction that a process cannot
465 * change the user ID of a file it owns and the restriction
466 * that the group ID supplied to the chown() function
467 * shall be equal to either the group ID or one of the
468 * supplementary group IDs of the calling process.
470 if (restricted_chown &&
471 (iuid != uid || (igid != gid &&
472 !groupmember((gid_t)gid, credp))) &&
473 !capable(CAP_CHOWN)) {
474 code = XFS_ERROR(EPERM);
478 * Do a quota reservation only if uid/projid/gid is actually
481 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
482 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
483 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
485 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
486 capable(CAP_FOWNER) ?
487 XFS_QMOPT_FORCE_RES : 0);
488 if (code) /* out of quota */
494 * Truncate file. Must have write permission and not be a directory.
496 if (mask & XFS_AT_SIZE) {
497 /* Short circuit the truncate case for zero length files */
498 if ((vap->va_size == 0) &&
499 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
500 xfs_iunlock(ip, XFS_ILOCK_EXCL);
501 lock_flags &= ~XFS_ILOCK_EXCL;
502 if (mask & XFS_AT_CTIME)
503 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
509 code = XFS_ERROR(EISDIR);
511 } else if (!VN_ISREG(vp)) {
512 code = XFS_ERROR(EINVAL);
516 * Make sure that the dquots are attached to the inode.
518 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
523 * Change file access or modified times.
525 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
527 if ((flags & ATTR_UTIME) &&
528 !capable(CAP_FOWNER)) {
529 code = XFS_ERROR(EPERM);
536 * Change extent size or realtime flag.
538 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
540 * Can't change extent size if any extents are allocated.
542 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
543 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
545 code = XFS_ERROR(EINVAL); /* EFBIG? */
549 * Can't change realtime flag if any extents are allocated.
551 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
552 (mask & XFS_AT_XFLAGS) &&
553 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
554 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
555 code = XFS_ERROR(EINVAL); /* EFBIG? */
560 * Extent size must be a multiple of the appropriate block
561 * size, if set at all.
563 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
566 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
567 ((mask & XFS_AT_XFLAGS) &&
568 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
569 size = mp->m_sb.sb_rextsize <<
570 mp->m_sb.sb_blocklog;
572 size = mp->m_sb.sb_blocksize;
574 if (vap->va_extsize % size) {
575 code = XFS_ERROR(EINVAL);
580 * If realtime flag is set then must have realtime data.
582 if ((mask & XFS_AT_XFLAGS) &&
583 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
584 if ((mp->m_sb.sb_rblocks == 0) ||
585 (mp->m_sb.sb_rextsize == 0) ||
586 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
587 code = XFS_ERROR(EINVAL);
593 * Can't modify an immutable/append-only file unless
594 * we have appropriate permission.
596 if ((mask & XFS_AT_XFLAGS) &&
598 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
600 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
601 !capable(CAP_LINUX_IMMUTABLE)) {
602 code = XFS_ERROR(EPERM);
608 * Now we can make the changes. Before we join the inode
609 * to the transaction, if XFS_AT_SIZE is set then take care of
610 * the part of the truncation that must be done without the
611 * inode lock. This needs to be done before joining the inode
612 * to the transaction, because the inode cannot be unlocked
613 * once it is a part of the transaction.
615 if (mask & XFS_AT_SIZE) {
617 if ((vap->va_size > ip->i_d.di_size) &&
618 (flags & ATTR_NOSIZETOK) == 0) {
619 code = xfs_igrow_start(ip, vap->va_size, credp);
621 xfs_iunlock(ip, XFS_ILOCK_EXCL);
622 vn_iowait(vp); /* wait for the completion of any pending DIOs */
624 code = xfs_itruncate_data(ip, vap->va_size);
627 lock_flags &= ~XFS_ILOCK_EXCL;
628 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
631 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
632 if ((code = xfs_trans_reserve(tp, 0,
633 XFS_ITRUNCATE_LOG_RES(mp), 0,
634 XFS_TRANS_PERM_LOG_RES,
635 XFS_ITRUNCATE_LOG_COUNT))) {
636 xfs_trans_cancel(tp, 0);
638 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
641 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
642 xfs_ilock(ip, XFS_ILOCK_EXCL);
646 xfs_trans_ijoin(tp, ip, lock_flags);
647 xfs_trans_ihold(tp, ip);
650 /* determine whether mandatory locking mode changes */
651 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
654 * Truncate file. Must have write permission and not be a directory.
656 if (mask & XFS_AT_SIZE) {
657 if (vap->va_size > ip->i_d.di_size) {
658 xfs_igrow_finish(tp, ip, vap->va_size,
659 !(flags & ATTR_DMI));
660 } else if ((vap->va_size <= ip->i_d.di_size) ||
661 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
663 * signal a sync transaction unless
664 * we're truncating an already unlinked
665 * file on a wsync filesystem
667 code = xfs_itruncate_finish(&tp, ip,
668 (xfs_fsize_t)vap->va_size,
670 ((ip->i_d.di_nlink != 0 ||
671 !(mp->m_flags & XFS_MOUNT_WSYNC))
678 * Have to do this even if the file's size doesn't change.
680 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
684 * Change file access modes.
686 if (mask & XFS_AT_MODE) {
687 ip->i_d.di_mode &= S_IFMT;
688 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
690 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
691 timeflags |= XFS_ICHGTIME_CHG;
695 * Change file ownership. Must be the owner or privileged.
696 * If the system was configured with the "restricted_chown"
697 * option, the owner is not permitted to give away the file,
698 * and can change the group id only to a group of which he
699 * or she is a member.
701 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
703 * CAP_FSETID overrides the following restrictions:
705 * The set-user-ID and set-group-ID bits of a file will be
706 * cleared upon successful return from chown()
708 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
709 !capable(CAP_FSETID)) {
710 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
714 * Change the ownerships and register quota modifications
715 * in the transaction.
718 if (XFS_IS_UQUOTA_ON(mp)) {
719 ASSERT(mask & XFS_AT_UID);
721 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
722 &ip->i_udquot, udqp);
724 ip->i_d.di_uid = uid;
727 if (XFS_IS_GQUOTA_ON(mp)) {
728 ASSERT(!XFS_IS_PQUOTA_ON(mp));
729 ASSERT(mask & XFS_AT_GID);
731 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
732 &ip->i_gdquot, gdqp);
734 ip->i_d.di_gid = gid;
736 if (iprojid != projid) {
737 if (XFS_IS_PQUOTA_ON(mp)) {
738 ASSERT(!XFS_IS_GQUOTA_ON(mp));
739 ASSERT(mask & XFS_AT_PROJID);
741 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
742 &ip->i_gdquot, gdqp);
744 ip->i_d.di_projid = projid;
746 * We may have to rev the inode as well as
747 * the superblock version number since projids didn't
748 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
750 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
751 xfs_bump_ino_vers2(tp, ip);
754 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
755 timeflags |= XFS_ICHGTIME_CHG;
760 * Change file access or modified times.
762 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
763 if (mask & XFS_AT_ATIME) {
764 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
765 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
766 ip->i_update_core = 1;
767 //timeflags &= ~XFS_ICHGTIME_ACC;
769 if (mask & XFS_AT_MTIME) {
770 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
771 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
772 timeflags &= ~XFS_ICHGTIME_MOD;
773 timeflags |= XFS_ICHGTIME_CHG;
775 if (tp && (flags & ATTR_UTIME))
776 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
780 * Change XFS-added attributes.
782 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
783 if (mask & XFS_AT_EXTSIZE) {
785 * Converting bytes to fs blocks.
787 ip->i_d.di_extsize = vap->va_extsize >>
788 mp->m_sb.sb_blocklog;
790 if (mask & XFS_AT_XFLAGS) {
793 /* can't set PREALLOC this way, just preserve it */
794 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
795 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
796 di_flags |= XFS_DIFLAG_IMMUTABLE;
797 if (vap->va_xflags & XFS_XFLAG_APPEND)
798 di_flags |= XFS_DIFLAG_APPEND;
799 if (vap->va_xflags & XFS_XFLAG_SYNC)
800 di_flags |= XFS_DIFLAG_SYNC;
801 if (vap->va_xflags & XFS_XFLAG_NOATIME)
802 di_flags |= XFS_DIFLAG_NOATIME;
803 if (vap->va_xflags & XFS_XFLAG_NODUMP)
804 di_flags |= XFS_DIFLAG_NODUMP;
805 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
806 di_flags |= XFS_DIFLAG_PROJINHERIT;
807 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
808 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
809 di_flags |= XFS_DIFLAG_RTINHERIT;
810 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
811 di_flags |= XFS_DIFLAG_NOSYMLINKS;
812 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
813 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
814 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
815 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
816 di_flags |= XFS_DIFLAG_REALTIME;
817 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
819 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
821 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
822 di_flags |= XFS_DIFLAG_EXTSIZE;
824 ip->i_d.di_flags = di_flags;
826 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
827 timeflags |= XFS_ICHGTIME_CHG;
831 * Change file inode change time only if XFS_AT_CTIME set
832 * AND we have been called by a DMI function.
835 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
836 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
837 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
838 ip->i_update_core = 1;
839 timeflags &= ~XFS_ICHGTIME_CHG;
843 * Send out timestamp changes that need to be set to the
844 * current time. Not done when called by a DMI function.
846 if (timeflags && !(flags & ATTR_DMI))
847 xfs_ichgtime(ip, timeflags);
849 XFS_STATS_INC(xs_ig_attrchg);
852 * If this is a synchronous mount, make sure that the
853 * transaction goes to disk before returning to the user.
854 * This is slightly sub-optimal in that truncates require
855 * two sync transactions instead of one for wsync filesystems.
856 * One for the truncate and one for the timestamps since we
857 * don't want to change the timestamps unless we're sure the
858 * truncate worked. Truncates are less than 1% of the laddis
859 * mix so this probably isn't worth the trouble to optimize.
863 if (mp->m_flags & XFS_MOUNT_WSYNC)
864 xfs_trans_set_sync(tp);
866 code = xfs_trans_commit(tp, commit_flags, NULL);
870 * If the (regular) file's mandatory locking mode changed, then
871 * notify the vnode. We do this under the inode lock to prevent
872 * racing calls to vop_vnode_change.
874 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
875 if (mandlock_before != mandlock_after) {
876 XVOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
880 xfs_iunlock(ip, lock_flags);
883 * Release any dquot(s) the inode had kept before chown.
885 XFS_QM_DQRELE(mp, olddquot1);
886 XFS_QM_DQRELE(mp, olddquot2);
887 XFS_QM_DQRELE(mp, udqp);
888 XFS_QM_DQRELE(mp, gdqp);
894 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
895 !(flags & ATTR_DMI)) {
896 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
897 NULL, DM_RIGHT_NULL, NULL, NULL,
898 0, 0, AT_DELAY_FLAG(flags));
903 commit_flags |= XFS_TRANS_ABORT;
906 XFS_QM_DQRELE(mp, udqp);
907 XFS_QM_DQRELE(mp, gdqp);
909 xfs_trans_cancel(tp, commit_flags);
911 if (lock_flags != 0) {
912 xfs_iunlock(ip, lock_flags);
920 * Null conversion from vnode mode bits to inode mode bits, as in efs.
931 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
932 (inst_t *)__return_address);
934 ip = XFS_BHVTOI(bdp);
935 xfs_ilock(ip, XFS_ILOCK_SHARED);
936 error = xfs_iaccess(ip, mode, credp);
937 xfs_iunlock(ip, XFS_ILOCK_SHARED);
961 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
967 vp = BHV_TO_VNODE(bdp);
968 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
970 ip = XFS_BHVTOI(bdp);
973 if (XFS_FORCED_SHUTDOWN(mp))
974 return XFS_ERROR(EIO);
976 xfs_ilock(ip, XFS_ILOCK_SHARED);
978 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
980 offset = uiop->uio_offset;
981 count = uiop->uio_resid;
984 error = XFS_ERROR(EINVAL);
993 * See if the symlink is stored inline.
995 pathlen = (int)ip->i_d.di_size;
997 if (ip->i_df.if_flags & XFS_IFINLINE) {
998 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1002 * Symlink not inline. Call bmap to get it in.
1004 nmaps = SYMLINK_MAPS;
1006 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1007 0, NULL, 0, mval, &nmaps, NULL, NULL);
1013 for (n = 0; n < nmaps; n++) {
1014 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1015 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1016 bp = xfs_buf_read(mp->m_ddev_targp, d,
1017 BTOBB(byte_cnt), 0);
1018 error = XFS_BUF_GETERROR(bp);
1020 xfs_ioerror_alert("xfs_readlink",
1021 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1025 if (pathlen < byte_cnt)
1027 pathlen -= byte_cnt;
1029 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1036 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1044 * This is called to sync the inode and its data out to disk.
1045 * We need to hold the I/O lock while flushing the data, and
1046 * the inode lock while flushing the inode. The inode lock CANNOT
1047 * be held while flushing the data, so acquire after we're done
1061 int log_flushed = 0, changed = 1;
1063 vn_trace_entry(BHV_TO_VNODE(bdp),
1064 __FUNCTION__, (inst_t *)__return_address);
1066 ip = XFS_BHVTOI(bdp);
1068 ASSERT(start >= 0 && stop >= -1);
1070 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1071 return XFS_ERROR(EIO);
1074 * We always need to make sure that the required inode state
1075 * is safe on disk. The vnode might be clean but because
1076 * of committed transactions that haven't hit the disk yet.
1077 * Likewise, there could be unflushed non-transactional
1078 * changes to the inode core that have to go to disk.
1080 * The following code depends on one assumption: that
1081 * any transaction that changes an inode logs the core
1082 * because it has to change some field in the inode core
1083 * (typically nextents or nblocks). That assumption
1084 * implies that any transactions against an inode will
1085 * catch any non-transactional updates. If inode-altering
1086 * transactions exist that violate this assumption, the
1087 * code breaks. Right now, it figures that if the involved
1088 * update_* field is clear and the inode is unpinned, the
1089 * inode is clean. Either it's been flushed or it's been
1090 * committed and the commit has hit the disk unpinning the inode.
1091 * (Note that xfs_inode_item_format() called at commit clears
1092 * the update_* fields.)
1094 xfs_ilock(ip, XFS_ILOCK_SHARED);
1096 /* If we are flushing data then we care about update_size
1097 * being set, otherwise we care about update_core
1099 if ((flag & FSYNC_DATA) ?
1100 (ip->i_update_size == 0) :
1101 (ip->i_update_core == 0)) {
1103 * Timestamps/size haven't changed since last inode
1104 * flush or inode transaction commit. That means
1105 * either nothing got written or a transaction
1106 * committed which caught the updates. If the
1107 * latter happened and the transaction hasn't
1108 * hit the disk yet, the inode will be still
1109 * be pinned. If it is, force the log.
1112 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1114 if (xfs_ipincount(ip)) {
1115 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1117 ((flag & FSYNC_WAIT)
1118 ? XFS_LOG_SYNC : 0),
1122 * If the inode is not pinned and nothing
1123 * has changed we don't need to flush the
1131 * Kick off a transaction to log the inode
1132 * core to get the updates. Make it
1133 * sync if FSYNC_WAIT is passed in (which
1134 * is done by everybody but specfs). The
1135 * sync transaction will also force the log.
1137 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1138 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1139 if ((error = xfs_trans_reserve(tp, 0,
1140 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1142 xfs_trans_cancel(tp, 0);
1145 xfs_ilock(ip, XFS_ILOCK_EXCL);
1148 * Note - it's possible that we might have pushed
1149 * ourselves out of the way during trans_reserve
1150 * which would flush the inode. But there's no
1151 * guarantee that the inode buffer has actually
1152 * gone out yet (it's delwri). Plus the buffer
1153 * could be pinned anyway if it's part of an
1154 * inode in another recent transaction. So we
1155 * play it safe and fire off the transaction anyway.
1157 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1158 xfs_trans_ihold(tp, ip);
1159 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1160 if (flag & FSYNC_WAIT)
1161 xfs_trans_set_sync(tp);
1162 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1164 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1167 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1169 * If the log write didn't issue an ordered tag we need
1170 * to flush the disk cache for the data device now.
1173 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1176 * If this inode is on the RT dev we need to flush that
1179 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1180 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1187 * This is called by xfs_inactive to free any blocks beyond eof,
1188 * when the link count isn't zero.
1191 xfs_inactive_free_eofblocks(
1197 xfs_fileoff_t end_fsb;
1198 xfs_fileoff_t last_fsb;
1199 xfs_filblks_t map_len;
1201 xfs_bmbt_irec_t imap;
1204 * Figure out if there are any blocks beyond the end
1205 * of the file. If not, then there is nothing to do.
1207 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1208 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1209 map_len = last_fsb - end_fsb;
1214 xfs_ilock(ip, XFS_ILOCK_SHARED);
1215 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1216 NULL, 0, &imap, &nimaps, NULL, NULL);
1217 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1219 if (!error && (nimaps != 0) &&
1220 (imap.br_startblock != HOLESTARTBLOCK ||
1221 ip->i_delayed_blks)) {
1223 * Attach the dquots to the inode up front.
1225 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1229 * There are blocks after the end of file.
1230 * Free them up now by truncating the file to
1233 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1236 * Do the xfs_itruncate_start() call before
1237 * reserving any log space because
1238 * itruncate_start will call into the buffer
1239 * cache and we can't
1240 * do that within a transaction.
1242 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1243 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1246 error = xfs_trans_reserve(tp, 0,
1247 XFS_ITRUNCATE_LOG_RES(mp),
1248 0, XFS_TRANS_PERM_LOG_RES,
1249 XFS_ITRUNCATE_LOG_COUNT);
1251 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1252 xfs_trans_cancel(tp, 0);
1253 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1257 xfs_ilock(ip, XFS_ILOCK_EXCL);
1258 xfs_trans_ijoin(tp, ip,
1261 xfs_trans_ihold(tp, ip);
1263 error = xfs_itruncate_finish(&tp, ip,
1268 * If we get an error at this point we
1269 * simply don't bother truncating the file.
1272 xfs_trans_cancel(tp,
1273 (XFS_TRANS_RELEASE_LOG_RES |
1276 error = xfs_trans_commit(tp,
1277 XFS_TRANS_RELEASE_LOG_RES,
1280 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1286 * Free a symlink that has blocks associated with it.
1289 xfs_inactive_symlink_rmt(
1297 xfs_fsblock_t first_block;
1298 xfs_bmap_free_t free_list;
1301 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1309 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1311 * We're freeing a symlink that has some
1312 * blocks allocated to it. Free the
1313 * blocks here. We know that we've got
1314 * either 1 or 2 extents and that we can
1315 * free them all in one bunmapi call.
1317 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1318 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1319 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1320 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1321 xfs_trans_cancel(tp, 0);
1326 * Lock the inode, fix the size, and join it to the transaction.
1327 * Hold it so in the normal path, we still have it locked for
1328 * the second transaction. In the error paths we need it
1329 * held so the cancel won't rele it, see below.
1331 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1332 size = (int)ip->i_d.di_size;
1333 ip->i_d.di_size = 0;
1334 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1335 xfs_trans_ihold(tp, ip);
1336 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1338 * Find the block(s) so we can inval and unmap them.
1341 XFS_BMAP_INIT(&free_list, &first_block);
1342 nmaps = sizeof(mval) / sizeof(mval[0]);
1343 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1344 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1348 * Invalidate the block(s).
1350 for (i = 0; i < nmaps; i++) {
1351 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1352 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1353 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1354 xfs_trans_binval(tp, bp);
1357 * Unmap the dead block(s) to the free_list.
1359 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1360 &first_block, &free_list, NULL, &done)))
1364 * Commit the first transaction. This logs the EFI and the inode.
1366 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1369 * The transaction must have been committed, since there were
1370 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1371 * The new tp has the extent freeing and EFDs.
1375 * The first xact was committed, so add the inode to the new one.
1376 * Mark it dirty so it will be logged and moved forward in the log as
1377 * part of every commit.
1379 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1380 xfs_trans_ihold(tp, ip);
1381 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1383 * Get a new, empty transaction to return to our caller.
1385 ntp = xfs_trans_dup(tp);
1387 * Commit the transaction containing extent freeing and EFDs.
1388 * If we get an error on the commit here or on the reserve below,
1389 * we need to unlock the inode since the new transaction doesn't
1390 * have the inode attached.
1392 error = xfs_trans_commit(tp, 0, NULL);
1395 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1399 * Remove the memory for extent descriptions (just bookkeeping).
1401 if (ip->i_df.if_bytes)
1402 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1403 ASSERT(ip->i_df.if_bytes == 0);
1405 * Put an itruncate log reservation in the new transaction
1408 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1409 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1410 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1414 * Return with the inode locked but not joined to the transaction.
1420 xfs_bmap_cancel(&free_list);
1423 * Have to come here with the inode locked and either
1424 * (held and in the transaction) or (not in the transaction).
1425 * If the inode isn't held then cancel would iput it, but
1426 * that's wrong since this is inactive and the vnode ref
1427 * count is 0 already.
1428 * Cancel won't do anything to the inode if held, but it still
1429 * needs to be locked until the cancel is done, if it was
1430 * joined to the transaction.
1432 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1433 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1440 xfs_inactive_symlink_local(
1446 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1448 * We're freeing a symlink which fit into
1449 * the inode. Just free the memory used
1450 * to hold the old symlink.
1452 error = xfs_trans_reserve(*tpp, 0,
1453 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1454 0, XFS_TRANS_PERM_LOG_RES,
1455 XFS_ITRUNCATE_LOG_COUNT);
1458 xfs_trans_cancel(*tpp, 0);
1462 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1465 * Zero length symlinks _can_ exist.
1467 if (ip->i_df.if_bytes > 0) {
1468 xfs_idata_realloc(ip,
1469 -(ip->i_df.if_bytes),
1471 ASSERT(ip->i_df.if_bytes == 0);
1488 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1491 ASSERT(ip->i_d.di_forkoff != 0);
1492 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1493 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1495 error = xfs_attr_inactive(ip);
1498 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1499 return error; /* goto out */
1502 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1503 error = xfs_trans_reserve(tp, 0,
1504 XFS_IFREE_LOG_RES(mp),
1505 0, XFS_TRANS_PERM_LOG_RES,
1506 XFS_INACTIVE_LOG_COUNT);
1508 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1509 xfs_trans_cancel(tp, 0);
1511 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1515 xfs_ilock(ip, XFS_ILOCK_EXCL);
1516 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1517 xfs_trans_ihold(tp, ip);
1518 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1520 ASSERT(ip->i_d.di_anextents == 0);
1535 vp = BHV_TO_VNODE(bdp);
1536 ip = XFS_BHVTOI(bdp);
1538 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1542 /* If this is a read-only mount, don't do this (would generate I/O) */
1543 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1546 #ifdef HAVE_REFCACHE
1547 /* If we are in the NFS reference cache then don't do this now */
1554 if (ip->i_d.di_nlink != 0) {
1555 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1556 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1557 ip->i_delayed_blks > 0)) &&
1558 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1559 (!(ip->i_d.di_flags &
1560 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1561 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1564 #ifdef RMC /* Update linux inode block count after free above */
1565 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1566 ip->i_d.di_nblocks + ip->i_delayed_blks);
1577 * This is called when the vnode reference count for the vnode
1578 * goes to zero. If the file has been unlinked, then it must
1579 * now be truncated. Also, we clear all of the read-ahead state
1580 * kept for the inode here since the file is now closed.
1590 xfs_bmap_free_t free_list;
1591 xfs_fsblock_t first_block;
1598 vp = BHV_TO_VNODE(bdp);
1599 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1601 ip = XFS_BHVTOI(bdp);
1604 * If the inode is already free, then there can be nothing
1607 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1608 ASSERT(ip->i_df.if_real_bytes == 0);
1609 ASSERT(ip->i_df.if_broot_bytes == 0);
1610 return VN_INACTIVE_CACHE;
1614 * Only do a truncate if it's a regular file with
1615 * some actual space in it. It's OK to look at the
1616 * inode's fields without the lock because we're the
1617 * only one with a reference to the inode.
1619 truncate = ((ip->i_d.di_nlink == 0) &&
1620 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1621 (ip->i_delayed_blks > 0)) &&
1622 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1626 if (ip->i_d.di_nlink == 0 &&
1627 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1628 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1633 /* If this is a read-only mount, don't do this (would generate I/O) */
1634 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1637 if (ip->i_d.di_nlink != 0) {
1638 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1639 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1640 ip->i_delayed_blks > 0)) &&
1641 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1642 (!(ip->i_d.di_flags &
1643 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1644 (ip->i_delayed_blks != 0)))) {
1645 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1646 return VN_INACTIVE_CACHE;
1648 /* Update linux inode block count after free above */
1649 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1650 ip->i_d.di_nblocks + ip->i_delayed_blks);
1656 ASSERT(ip->i_d.di_nlink == 0);
1658 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1659 return VN_INACTIVE_CACHE;
1661 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1664 * Do the xfs_itruncate_start() call before
1665 * reserving any log space because itruncate_start
1666 * will call into the buffer cache and we can't
1667 * do that within a transaction.
1669 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1671 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1673 error = xfs_trans_reserve(tp, 0,
1674 XFS_ITRUNCATE_LOG_RES(mp),
1675 0, XFS_TRANS_PERM_LOG_RES,
1676 XFS_ITRUNCATE_LOG_COUNT);
1678 /* Don't call itruncate_cleanup */
1679 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1680 xfs_trans_cancel(tp, 0);
1681 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1682 return VN_INACTIVE_CACHE;
1685 xfs_ilock(ip, XFS_ILOCK_EXCL);
1686 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1687 xfs_trans_ihold(tp, ip);
1690 * normally, we have to run xfs_itruncate_finish sync.
1691 * But if filesystem is wsync and we're in the inactive
1692 * path, then we know that nlink == 0, and that the
1693 * xaction that made nlink == 0 is permanently committed
1694 * since xfs_remove runs as a synchronous transaction.
1696 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1697 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1700 xfs_trans_cancel(tp,
1701 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1702 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1703 return VN_INACTIVE_CACHE;
1705 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1708 * If we get an error while cleaning up a
1709 * symlink we bail out.
1711 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1712 xfs_inactive_symlink_rmt(ip, &tp) :
1713 xfs_inactive_symlink_local(ip, &tp);
1717 return VN_INACTIVE_CACHE;
1720 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1721 xfs_trans_ihold(tp, ip);
1723 error = xfs_trans_reserve(tp, 0,
1724 XFS_IFREE_LOG_RES(mp),
1725 0, XFS_TRANS_PERM_LOG_RES,
1726 XFS_INACTIVE_LOG_COUNT);
1728 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1729 xfs_trans_cancel(tp, 0);
1730 return VN_INACTIVE_CACHE;
1733 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1734 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1735 xfs_trans_ihold(tp, ip);
1739 * If there are attributes associated with the file
1740 * then blow them away now. The code calls a routine
1741 * that recursively deconstructs the attribute fork.
1742 * We need to just commit the current transaction
1743 * because we can't use it for xfs_attr_inactive().
1745 if (ip->i_d.di_anextents > 0) {
1746 error = xfs_inactive_attrs(ip, &tp);
1748 * If we got an error, the transaction is already
1749 * cancelled, and the inode is unlocked. Just get out.
1752 return VN_INACTIVE_CACHE;
1753 } else if (ip->i_afp) {
1754 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1760 XFS_BMAP_INIT(&free_list, &first_block);
1761 error = xfs_ifree(tp, ip, &free_list);
1764 * If we fail to free the inode, shut down. The cancel
1765 * might do that, we need to make sure. Otherwise the
1766 * inode might be lost for a long time or forever.
1768 if (!XFS_FORCED_SHUTDOWN(mp)) {
1770 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1771 error, mp->m_fsname);
1772 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1774 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1777 * Credit the quota account(s). The inode is gone.
1779 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1782 * Just ignore errors at this point. There is
1783 * nothing we can do except to try to keep going.
1785 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1787 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1790 * Release the dquots held by inode, if any.
1792 XFS_QM_DQDETACH(mp, ip);
1794 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1797 return VN_INACTIVE_CACHE;
1806 bhv_desc_t *dir_bdp,
1813 xfs_inode_t *dp, *ip;
1817 xfs_vnode_t *dir_vp;
1819 dir_vp = BHV_TO_VNODE(dir_bdp);
1820 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1822 dp = XFS_BHVTOI(dir_bdp);
1824 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1825 return XFS_ERROR(EIO);
1827 lock_mode = xfs_ilock_map_shared(dp);
1828 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1830 *vpp = XFS_ITOV(ip);
1833 xfs_iunlock_map_shared(dp, lock_mode);
1839 * xfs_create (create a new file).
1843 bhv_desc_t *dir_bdp,
1849 char *name = VNAME(dentry);
1850 xfs_vnode_t *dir_vp;
1851 xfs_inode_t *dp, *ip;
1852 xfs_vnode_t *vp=NULL;
1857 xfs_bmap_free_t free_list;
1858 xfs_fsblock_t first_block;
1859 boolean_t dp_joined_to_trans;
1860 int dm_event_sent = 0;
1864 struct xfs_dquot *udqp, *gdqp;
1870 dir_vp = BHV_TO_VNODE(dir_bdp);
1871 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1873 dp = XFS_BHVTOI(dir_bdp);
1876 dm_di_mode = vap->va_mode;
1877 namelen = VNAMELEN(dentry);
1879 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1880 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1881 dir_vp, DM_RIGHT_NULL, NULL,
1882 DM_RIGHT_NULL, name, NULL,
1890 if (XFS_FORCED_SHUTDOWN(mp))
1891 return XFS_ERROR(EIO);
1893 /* Return through std_return after this point. */
1897 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1898 prid = dp->i_d.di_projid;
1899 else if (vap->va_mask & XFS_AT_PROJID)
1900 prid = (xfs_prid_t)vap->va_projid;
1902 prid = (xfs_prid_t)dfltprid;
1905 * Make sure that we have allocated dquot(s) on disk.
1907 error = XFS_QM_DQVOPALLOC(mp, dp,
1908 current_fsuid(credp), current_fsgid(credp), prid,
1909 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1914 dp_joined_to_trans = B_FALSE;
1916 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1917 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1918 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1920 * Initially assume that the file does not exist and
1921 * reserve the resources for that case. If that is not
1922 * the case we'll drop the one we have and get a more
1923 * appropriate transaction later.
1925 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1926 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1927 if (error == ENOSPC) {
1929 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1930 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1938 xfs_ilock(dp, XFS_ILOCK_EXCL);
1940 XFS_BMAP_INIT(&free_list, &first_block);
1945 * Reserve disk quota and the inode.
1947 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1952 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1954 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1955 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1956 rdev, credp, prid, resblks > 0,
1959 if (error == ENOSPC)
1966 * At this point, we've gotten a newly allocated inode.
1967 * It is locked (and joined to the transaction).
1970 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1973 * Now we join the directory inode to the transaction.
1974 * We do not do it earlier because xfs_dir_ialloc
1975 * might commit the previous transaction (and release
1980 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1981 dp_joined_to_trans = B_TRUE;
1983 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1984 &first_block, &free_list,
1985 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1987 ASSERT(error != ENOSPC);
1990 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1991 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1994 * If this is a synchronous mount, make sure that the
1995 * create transaction goes to disk before returning to
1998 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1999 xfs_trans_set_sync(tp);
2005 * Attach the dquot(s) to the inodes and modify them incore.
2006 * These ids of the inode couldn't have changed since the new
2007 * inode has been locked ever since it was created.
2009 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2012 * xfs_trans_commit normally decrements the vnode ref count
2013 * when it unlocks the inode. Since we want to return the
2014 * vnode to the caller, we bump the vnode ref count now.
2019 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2021 xfs_bmap_cancel(&free_list);
2025 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2032 XFS_QM_DQRELE(mp, udqp);
2033 XFS_QM_DQRELE(mp, gdqp);
2036 * Propagate the fact that the vnode changed after the
2037 * xfs_inode locks have been released.
2039 XVOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2043 /* Fallthrough to std_return with error = 0 */
2046 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2047 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2048 DM_EVENT_POSTCREATE)) {
2049 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2050 dir_vp, DM_RIGHT_NULL,
2052 DM_RIGHT_NULL, name, NULL,
2053 dm_di_mode, error, 0);
2058 cancel_flags |= XFS_TRANS_ABORT;
2063 xfs_trans_cancel(tp, cancel_flags);
2065 if (!dp_joined_to_trans && (dp != NULL))
2066 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2067 XFS_QM_DQRELE(mp, udqp);
2068 XFS_QM_DQRELE(mp, gdqp);
2074 * Wait until after the current transaction is aborted to
2075 * release the inode. This prevents recursive transactions
2076 * and deadlocks from xfs_inactive.
2078 cancel_flags |= XFS_TRANS_ABORT;
2079 xfs_trans_cancel(tp, cancel_flags);
2082 XFS_QM_DQRELE(mp, udqp);
2083 XFS_QM_DQRELE(mp, gdqp);
2090 * Some counters to see if (and how often) we are hitting some deadlock
2091 * prevention code paths.
2095 int xfs_rm_lock_delays;
2096 int xfs_rm_attempts;
2100 * The following routine will lock the inodes associated with the
2101 * directory and the named entry in the directory. The locks are
2102 * acquired in increasing inode number.
2104 * If the entry is "..", then only the directory is locked. The
2105 * vnode ref count will still include that from the .. entry in
2108 * There is a deadlock we need to worry about. If the locked directory is
2109 * in the AIL, it might be blocking up the log. The next inode we lock
2110 * could be already locked by another thread waiting for log space (e.g
2111 * a permanent log reservation with a long running transaction (see
2112 * xfs_itruncate_finish)). To solve this, we must check if the directory
2113 * is in the ail and use lock_nowait. If we can't lock, we need to
2114 * drop the inode lock on the directory and try again. xfs_iunlock will
2115 * potentially push the tail if we were holding up the log.
2118 xfs_lock_dir_and_entry(
2121 xfs_inode_t *ip) /* inode of entry 'name' */
2125 xfs_inode_t *ips[2];
2134 xfs_ilock(dp, XFS_ILOCK_EXCL);
2141 * We want to lock in increasing inum. Since we've already
2142 * acquired the lock on the directory, we may need to release
2143 * if if the inum of the entry turns out to be less.
2145 if (e_inum > dp->i_ino) {
2147 * We are already in the right order, so just
2148 * lock on the inode of the entry.
2149 * We need to use nowait if dp is in the AIL.
2152 lp = (xfs_log_item_t *)dp->i_itemp;
2153 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2154 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2161 * Unlock dp and try again.
2162 * xfs_iunlock will try to push the tail
2163 * if the inode is in the AIL.
2166 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2168 if ((attempts % 5) == 0) {
2169 delay(1); /* Don't just spin the CPU */
2171 xfs_rm_lock_delays++;
2177 xfs_ilock(ip, XFS_ILOCK_EXCL);
2179 } else if (e_inum < dp->i_ino) {
2180 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2184 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2186 /* else e_inum == dp->i_ino */
2187 /* This can happen if we're asked to lock /x/..
2188 * the entry is "..", which is also the parent directory.
2196 int xfs_small_retries;
2197 int xfs_middle_retries;
2198 int xfs_lots_retries;
2199 int xfs_lock_delays;
2203 * The following routine will lock n inodes in exclusive mode.
2204 * We assume the caller calls us with the inodes in i_ino order.
2206 * We need to detect deadlock where an inode that we lock
2207 * is in the AIL and we start waiting for another inode that is locked
2208 * by a thread in a long running transaction (such as truncate). This can
2209 * result in deadlock since the long running trans might need to wait
2210 * for the inode we just locked in order to push the tail and free space
2220 int attempts = 0, i, j, try_lock;
2223 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2234 for (; i < inodes; i++) {
2237 if (i && (ips[i] == ips[i-1])) /* Already locked */
2241 * If try_lock is not set yet, make sure all locked inodes
2242 * are not in the AIL.
2243 * If any are, set try_lock to be used later.
2247 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2248 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2249 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2256 * If any of the previous locks we have locked is in the AIL,
2257 * we must TRY to get the second and subsequent locks. If
2258 * we can't get any, we must release all we have
2263 /* try_lock must be 0 if i is 0. */
2265 * try_lock means we have an inode locked
2266 * that is in the AIL.
2269 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2273 * Unlock all previous guys and try again.
2274 * xfs_iunlock will try to push the tail
2275 * if the inode is in the AIL.
2278 for(j = i - 1; j >= 0; j--) {
2281 * Check to see if we've already
2282 * unlocked this one.
2283 * Not the first one going back,
2284 * and the inode ptr is the same.
2286 if ((j != (i - 1)) && ips[j] ==
2290 xfs_iunlock(ips[j], lock_mode);
2293 if ((attempts % 5) == 0) {
2294 delay(1); /* Don't just spin the CPU */
2304 xfs_ilock(ips[i], lock_mode);
2310 if (attempts < 5) xfs_small_retries++;
2311 else if (attempts < 100) xfs_middle_retries++;
2312 else xfs_lots_retries++;
2320 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2321 int remove_which_error_return = 0;
2323 #define REMOVE_DEBUG_TRACE(x)
2324 #endif /* ! DEBUG */
2326 extern int xfs_remove(bhv_desc_t *, bhv_desc_t *, vname_t *, cred_t *);
2333 bhv_desc_t *dir_bdp,
2338 xfs_vnode_t *dir_vp;
2340 char *name = VNAME(dentry);
2341 xfs_inode_t *dp, *ip;
2342 xfs_trans_t *tp = NULL;
2345 xfs_bmap_free_t free_list;
2346 xfs_fsblock_t first_block;
2354 dir_vp = BHV_TO_VNODE(dir_bdp);
2355 xvp = BHV_TO_VNODE(vp_bdp);
2357 printf("xfs_remove: dvp %p vp %p\n",dir_vp,xvp);
2358 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2360 dp = XFS_BHVTOI(dir_bdp);
2363 if (XFS_FORCED_SHUTDOWN(mp))
2364 return XFS_ERROR(EIO);
2366 namelen = VNAMELEN(dentry);
2368 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2369 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2370 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2371 name, NULL, 0, 0, 0);
2376 /* From this point on, return through std_return */
2380 * We need to get a reference to ip before we get our log
2381 * reservation. The reason for this is that we cannot call
2382 * xfs_iget for an inode for which we do not have a reference
2383 * once we've acquired a log reservation. This is because the
2384 * inode we are trying to get might be in xfs_inactive going
2385 * for a log reservation. Since we'll have to wait for the
2386 * inactive code to complete before returning from xfs_iget,
2387 * we need to make sure that we don't have log space reserved
2388 * when we call xfs_iget. Instead we get an unlocked reference
2389 * to the inode before getting our log reservation.
2392 error = xfs_get_dir_entry(dentry, &ip);
2394 /* FreeBSD has already done the lookup */
2399 REMOVE_DEBUG_TRACE(__LINE__);
2403 dm_di_mode = ip->i_d.di_mode;
2405 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2409 error = XFS_QM_DQATTACH(mp, dp, 0);
2410 if (!error && dp != ip)
2411 error = XFS_QM_DQATTACH(mp, ip, 0);
2413 REMOVE_DEBUG_TRACE(__LINE__);
2418 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2419 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2421 * We try to get the real space reservation first,
2422 * allowing for directory btree deletion(s) implying
2423 * possible bmap insert(s). If we can't get the space
2424 * reservation then we use 0 instead, and avoid the bmap
2425 * btree insert(s) in the directory code by, if the bmap
2426 * insert tries to happen, instead trimming the LAST
2427 * block from the directory.
2429 resblks = XFS_REMOVE_SPACE_RES(mp);
2430 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2431 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2432 if (error == ENOSPC) {
2434 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2435 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2438 ASSERT(error != ENOSPC);
2439 REMOVE_DEBUG_TRACE(__LINE__);
2440 xfs_trans_cancel(tp, 0);
2445 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2447 REMOVE_DEBUG_TRACE(__LINE__);
2448 xfs_trans_cancel(tp, cancel_flags);
2454 * At this point, we've gotten both the directory and the entry
2457 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2460 * Increment vnode ref count only in this case since
2461 * there's an extra vnode reference in the case where
2465 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2469 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2471 XFS_BMAP_INIT(&free_list, &first_block);
2472 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2473 &first_block, &free_list, 0);
2475 ASSERT(error != ENOENT);
2476 REMOVE_DEBUG_TRACE(__LINE__);
2479 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2482 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2484 error = xfs_droplink(tp, ip);
2486 REMOVE_DEBUG_TRACE(__LINE__);
2490 /* Determine if this is the last link while
2491 * we are in the transaction.
2493 link_zero = (ip)->i_d.di_nlink==0;
2496 * Take an extra ref on the inode so that it doesn't
2497 * go to xfs_inactive() from within the commit.
2502 * If this is a synchronous mount, make sure that the
2503 * remove transaction goes to disk before returning to
2506 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2507 xfs_trans_set_sync(tp);
2510 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2512 REMOVE_DEBUG_TRACE(__LINE__);
2516 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2523 * Before we drop our extra reference to the inode, purge it
2524 * from the refcache if it is there. By waiting until afterwards
2525 * to do the IRELE, we ensure that we won't go inactive in the
2526 * xfs_refcache_purge_ip routine (although that would be OK).
2528 xfs_refcache_purge_ip(ip);
2530 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2533 * Let interposed file systems know about removed links.
2535 XVOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2539 /* Fall through to std_return with error = 0 */
2541 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2542 DM_EVENT_POSTREMOVE)) {
2543 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2544 dir_vp, DM_RIGHT_NULL,
2545 NULL, DM_RIGHT_NULL,
2546 name, NULL, dm_di_mode, error, 0);
2551 xfs_bmap_cancel(&free_list);
2552 cancel_flags |= XFS_TRANS_ABORT;
2553 xfs_trans_cancel(tp, cancel_flags);
2558 * In this case make sure to not release the inode until after
2559 * the current transaction is aborted. Releasing it beforehand
2560 * can cause us to go to xfs_inactive and start a recursive
2561 * transaction which can easily deadlock with the current one.
2563 xfs_bmap_cancel(&free_list);
2564 cancel_flags |= XFS_TRANS_ABORT;
2565 xfs_trans_cancel(tp, cancel_flags);
2568 * Before we drop our extra reference to the inode, purge it
2569 * from the refcache if it is there. By waiting until afterwards
2570 * to do the IRELE, we ensure that we won't go inactive in the
2571 * xfs_refcache_purge_ip routine (although that would be OK).
2573 xfs_refcache_purge_ip(ip);
2587 bhv_desc_t *target_dir_bdp,
2588 xfs_vnode_t *src_vp,
2592 xfs_inode_t *tdp, *sip;
2595 xfs_inode_t *ips[2];
2597 xfs_bmap_free_t free_list;
2598 xfs_fsblock_t first_block;
2601 xfs_vnode_t *target_dir_vp;
2603 char *target_name = VNAME(dentry);
2606 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2607 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2608 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2610 target_namelen = VNAMELEN(dentry);
2611 if (VN_ISDIR(src_vp))
2612 return XFS_ERROR(EPERM);
2614 sip = xfs_vtoi(src_vp);
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 * If we are using project inheritance, we only allow hard link
2682 * creation in our tree when the project IDs are the same; else
2683 * the tree quota mechanism could be circumvented.
2685 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2686 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2687 error = XFS_ERROR(EPERM);
2692 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2696 XFS_BMAP_INIT(&free_list, &first_block);
2698 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2699 sip->i_ino, &first_block, &free_list,
2703 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2705 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2707 error = xfs_bumplink(tp, sip);
2713 * If this is a synchronous mount, make sure that the
2714 * link transaction goes to disk before returning to
2717 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2718 xfs_trans_set_sync(tp);
2721 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2723 xfs_bmap_cancel(&free_list);
2727 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2732 /* Fall through to std_return with error = 0. */
2734 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2735 DM_EVENT_POSTLINK)) {
2736 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2737 target_dir_vp, DM_RIGHT_NULL,
2738 src_vp, DM_RIGHT_NULL,
2739 target_name, NULL, 0, error, 0);
2744 cancel_flags |= XFS_TRANS_ABORT;
2748 xfs_trans_cancel(tp, cancel_flags);
2757 bhv_desc_t *dir_bdp,
2763 char *dir_name = VNAME(dentry);
2765 xfs_inode_t *cdp; /* inode of created dir */
2766 xfs_vnode_t *cvp; /* vnode of created dir */
2772 xfs_bmap_free_t free_list;
2773 xfs_fsblock_t first_block;
2774 xfs_vnode_t *dir_vp;
2775 boolean_t dp_joined_to_trans;
2776 boolean_t created = B_FALSE;
2777 int dm_event_sent = 0;
2779 struct xfs_dquot *udqp, *gdqp;
2784 dir_vp = BHV_TO_VNODE(dir_bdp);
2785 dp = XFS_BHVTOI(dir_bdp);
2788 if (XFS_FORCED_SHUTDOWN(mp))
2789 return XFS_ERROR(EIO);
2791 dir_namelen = VNAMELEN(dentry);
2794 dp_joined_to_trans = B_FALSE;
2795 dm_di_mode = vap->va_mode;
2797 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2798 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2799 dir_vp, DM_RIGHT_NULL, NULL,
2800 DM_RIGHT_NULL, dir_name, NULL,
2807 /* Return through std_return after this point. */
2809 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2814 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2815 prid = dp->i_d.di_projid;
2816 else if (vap->va_mask & XFS_AT_PROJID)
2817 prid = (xfs_prid_t)vap->va_projid;
2819 prid = (xfs_prid_t)dfltprid;
2822 * Make sure that we have allocated dquot(s) on disk.
2824 error = XFS_QM_DQVOPALLOC(mp, dp,
2825 current_fsuid(credp), current_fsgid(credp), prid,
2826 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2830 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2831 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2832 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2833 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2834 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2835 if (error == ENOSPC) {
2837 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2838 XFS_TRANS_PERM_LOG_RES,
2839 XFS_MKDIR_LOG_COUNT);
2847 xfs_ilock(dp, XFS_ILOCK_EXCL);
2850 * Check for directory link count overflow.
2852 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2853 error = XFS_ERROR(EMLINK);
2858 * Reserve disk quota and the inode.
2860 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2865 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2868 * create the directory inode.
2870 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2871 0, credp, prid, resblks > 0,
2874 if (error == ENOSPC)
2881 * Now we add the directory inode to the transaction.
2882 * We waited until now since xfs_dir_ialloc might start
2883 * a new transaction. Had we joined the transaction
2884 * earlier, the locks might have gotten released.
2887 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2888 dp_joined_to_trans = B_TRUE;
2890 XFS_BMAP_INIT(&free_list, &first_block);
2892 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2893 cdp->i_ino, &first_block, &free_list,
2894 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2896 ASSERT(error != ENOSPC);
2899 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2902 * Bump the in memory version number of the parent directory
2903 * so that other processes accessing it will recognize that
2904 * the directory has changed.
2908 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2914 error = xfs_bumplink(tp, dp);
2919 cvp = XFS_ITOV(cdp);
2927 * Attach the dquots to the new inode and modify the icount incore.
2929 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2932 * If this is a synchronous mount, make sure that the
2933 * mkdir transaction goes to disk before returning to
2936 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2937 xfs_trans_set_sync(tp);
2940 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2946 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2947 XFS_QM_DQRELE(mp, udqp);
2948 XFS_QM_DQRELE(mp, gdqp);
2953 /* Fall through to std_return with error = 0 or errno from
2954 * xfs_trans_commit. */
2957 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2958 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2959 DM_EVENT_POSTCREATE)) {
2960 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2961 dir_vp, DM_RIGHT_NULL,
2962 created ? XFS_ITOV(cdp):NULL,
2965 dm_di_mode, error, 0);
2971 xfs_bmap_cancel(&free_list);
2973 cancel_flags |= XFS_TRANS_ABORT;
2975 xfs_trans_cancel(tp, cancel_flags);
2976 XFS_QM_DQRELE(mp, udqp);
2977 XFS_QM_DQRELE(mp, gdqp);
2979 if (!dp_joined_to_trans && (dp != NULL)) {
2980 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2993 bhv_desc_t *dir_bdp,
2997 char *name = VNAME(dentry);
2999 xfs_inode_t *cdp; /* child directory */
3003 xfs_bmap_free_t free_list;
3004 xfs_fsblock_t first_block;
3007 xfs_vnode_t *dir_vp;
3013 dir_vp = BHV_TO_VNODE(dir_bdp);
3014 dp = XFS_BHVTOI(dir_bdp);
3017 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3019 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3020 return XFS_ERROR(EIO);
3021 namelen = VNAMELEN(dentry);
3023 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3024 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3025 dir_vp, DM_RIGHT_NULL,
3026 NULL, DM_RIGHT_NULL,
3027 name, NULL, 0, 0, 0);
3029 return XFS_ERROR(error);
3032 /* Return through std_return after this point. */
3037 * We need to get a reference to cdp before we get our log
3038 * reservation. The reason for this is that we cannot call
3039 * xfs_iget for an inode for which we do not have a reference
3040 * once we've acquired a log reservation. This is because the
3041 * inode we are trying to get might be in xfs_inactive going
3042 * for a log reservation. Since we'll have to wait for the
3043 * inactive code to complete before returning from xfs_iget,
3044 * we need to make sure that we don't have log space reserved
3045 * when we call xfs_iget. Instead we get an unlocked reference
3046 * to the inode before getting our log reservation.
3048 error = xfs_get_dir_entry(dentry, &cdp);
3050 REMOVE_DEBUG_TRACE(__LINE__);
3054 dm_di_mode = cdp->i_d.di_mode;
3057 * Get the dquots for the inodes.
3059 error = XFS_QM_DQATTACH(mp, dp, 0);
3060 if (!error && dp != cdp)
3061 error = XFS_QM_DQATTACH(mp, cdp, 0);
3064 REMOVE_DEBUG_TRACE(__LINE__);
3068 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3069 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3071 * We try to get the real space reservation first,
3072 * allowing for directory btree deletion(s) implying
3073 * possible bmap insert(s). If we can't get the space
3074 * reservation then we use 0 instead, and avoid the bmap
3075 * btree insert(s) in the directory code by, if the bmap
3076 * insert tries to happen, instead trimming the LAST
3077 * block from the directory.
3079 resblks = XFS_REMOVE_SPACE_RES(mp);
3080 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3081 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3082 if (error == ENOSPC) {
3084 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3085 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3088 ASSERT(error != ENOSPC);
3093 XFS_BMAP_INIT(&free_list, &first_block);
3096 * Now lock the child directory inode and the parent directory
3097 * inode in the proper order. This will take care of validating
3098 * that the directory entry for the child directory inode has
3099 * not changed while we were obtaining a log reservation.
3101 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3103 xfs_trans_cancel(tp, cancel_flags);
3108 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3111 * Only increment the parent directory vnode count if
3112 * we didn't bump it in looking up cdp. The only time
3113 * we don't bump it is when we're looking up ".".
3119 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3121 ASSERT(cdp->i_d.di_nlink >= 2);
3122 if (cdp->i_d.di_nlink != 2) {
3123 error = XFS_ERROR(ENOTEMPTY);
3126 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3127 error = XFS_ERROR(ENOTEMPTY);
3131 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3132 &first_block, &free_list, resblks);
3137 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3140 * Bump the in memory generation count on the parent
3141 * directory so that other can know that it has changed.
3146 * Drop the link from cdp's "..".
3148 error = xfs_droplink(tp, dp);
3154 * Drop the link from dp to cdp.
3156 error = xfs_droplink(tp, cdp);
3162 * Drop the "." link from cdp to self.
3164 error = xfs_droplink(tp, cdp);
3169 /* Determine these before committing transaction */
3170 last_cdp_link = (cdp)->i_d.di_nlink==0;
3173 * Take an extra ref on the child vnode so that it
3174 * does not go to xfs_inactive() from within the commit.
3179 * If this is a synchronous mount, make sure that the
3180 * rmdir transaction goes to disk before returning to
3183 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3184 xfs_trans_set_sync(tp);
3187 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3189 xfs_bmap_cancel(&free_list);
3190 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3196 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3204 * Let interposed file systems know about removed links.
3206 XVOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3210 /* Fall through to std_return with error = 0 or the errno
3211 * from xfs_trans_commit. */
3213 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3214 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3215 dir_vp, DM_RIGHT_NULL,
3216 NULL, DM_RIGHT_NULL,
3217 name, NULL, dm_di_mode,
3223 xfs_bmap_cancel(&free_list);
3224 cancel_flags |= XFS_TRANS_ABORT;
3228 xfs_trans_cancel(tp, cancel_flags);
3236 * Read dp's entries starting at uiop->uio_offset and translate them into
3237 * bufsize bytes worth of struct dirents starting at bufbase.
3241 bhv_desc_t *dir_bdp,
3247 xfs_trans_t *tp = NULL;
3251 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3252 (inst_t *)__return_address);
3253 dp = XFS_BHVTOI(dir_bdp);
3255 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3256 return XFS_ERROR(EIO);
3259 lock_mode = xfs_ilock_map_shared(dp);
3260 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3261 xfs_iunlock_map_shared(dp, lock_mode);
3272 bhv_desc_t *dir_bdp,
3285 xfs_bmap_free_t free_list;
3286 xfs_fsblock_t first_block;
3287 boolean_t dp_joined_to_trans;
3288 xfs_vnode_t *dir_vp;
3291 xfs_fileoff_t first_fsb;
3292 xfs_filblks_t fs_blocks;
3294 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3301 struct xfs_dquot *udqp, *gdqp;
3303 char *link_name = VNAME(dentry);
3305 struct thread *current = curthread;
3308 dir_vp = BHV_TO_VNODE(dir_bdp);
3309 dp = XFS_BHVTOI(dir_bdp);
3310 dp_joined_to_trans = B_FALSE;
3315 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3319 if (XFS_FORCED_SHUTDOWN(mp))
3320 return XFS_ERROR(EIO);
3322 link_namelen = VNAMELEN(dentry);
3325 * Check component lengths of the target path name.
3327 pathlen = strlen(target_path);
3328 if (pathlen >= MAXPATHLEN) /* total string too long */
3329 return XFS_ERROR(ENAMETOOLONG);
3330 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3334 for(total = 0, path = target_path; total < pathlen;) {
3338 while(*path == '/') {
3344 * Count up to the next slash or end of path.
3345 * Error out if the component is bigger than MAXNAMELEN.
3347 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3348 if (++len >= MAXNAMELEN) {
3349 error = ENAMETOOLONG;
3356 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3357 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3358 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3359 link_name, target_path, 0, 0, 0);
3364 /* Return through std_return after this point. */
3369 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3370 prid = dp->i_d.di_projid;
3371 else if (vap->va_mask & XFS_AT_PROJID)
3372 prid = (xfs_prid_t)vap->va_projid;
3375 prid = (xfs_prid_t)dfltprid;
3378 * Make sure that we have allocated dquot(s) on disk.
3380 error = XFS_QM_DQVOPALLOC(mp, dp,
3381 current->td_ucred->cr_uid,
3382 current->td_ucred->cr_groups[0],
3384 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3388 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3389 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3391 * The symlink will fit into the inode data fork?
3392 * There can't be any attributes so we get the whole variable part.
3394 if (pathlen <= XFS_LITINO(mp))
3397 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3398 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3399 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3400 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3401 if (error == ENOSPC && fs_blocks == 0) {
3403 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3404 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3412 xfs_ilock(dp, XFS_ILOCK_EXCL);
3415 * Check whether the directory allows new symlinks or not.
3417 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3418 error = XFS_ERROR(EPERM);
3423 * Reserve disk quota : blocks and inode.
3425 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3430 * Check for ability to enter directory entry, if no space reserved.
3433 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3436 * Initialize the bmap freelist prior to calling either
3437 * bmapi or the directory create code.
3439 XFS_BMAP_INIT(&free_list, &first_block);
3442 * Allocate an inode for the symlink.
3444 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3445 1, 0, credp, prid, resblks > 0, &ip, NULL);
3447 if (error == ENOSPC)
3454 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3455 dp_joined_to_trans = B_TRUE;
3458 * Also attach the dquot(s) to it, if applicable.
3460 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3463 resblks -= XFS_IALLOC_SPACE_RES(mp);
3465 * If the symlink will fit into the inode, write it inline.
3467 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3468 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3469 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3470 ip->i_d.di_size = pathlen;
3473 * The inode was initially created in extent format.
3475 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3476 ip->i_df.if_flags |= XFS_IFINLINE;
3478 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3479 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3483 nmaps = SYMLINK_MAPS;
3485 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3486 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3487 &first_block, resblks, mval, &nmaps,
3494 resblks -= fs_blocks;
3495 ip->i_d.di_size = pathlen;
3496 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3498 cur_chunk = target_path;
3499 for (n = 0; n < nmaps; n++) {
3500 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3501 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3502 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3503 BTOBB(byte_cnt), 0);
3504 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3505 if (pathlen < byte_cnt) {
3508 pathlen -= byte_cnt;
3510 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3511 cur_chunk += byte_cnt;
3513 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3518 * Create the directory entry for the symlink.
3520 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3521 ip->i_ino, &first_block, &free_list, resblks);
3525 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3526 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3529 * Bump the in memory version number of the parent directory
3530 * so that other processes accessing it will recognize that
3531 * the directory has changed.
3536 * If this is a synchronous mount, make sure that the
3537 * symlink transaction goes to disk before returning to
3540 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3541 xfs_trans_set_sync(tp);
3545 * xfs_trans_commit normally decrements the vnode ref count
3546 * when it unlocks the inode. Since we want to return the
3547 * vnode to the caller, we bump the vnode ref count now.
3551 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3555 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3556 XFS_QM_DQRELE(mp, udqp);
3557 XFS_QM_DQRELE(mp, gdqp);
3559 /* Fall through to std_return with error = 0 or errno from
3560 * xfs_trans_commit */
3562 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3563 DM_EVENT_POSTSYMLINK)) {
3564 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3565 dir_vp, DM_RIGHT_NULL,
3566 error ? NULL : XFS_ITOV(ip),
3567 DM_RIGHT_NULL, link_name, target_path,
3583 xfs_bmap_cancel(&free_list);
3584 cancel_flags |= XFS_TRANS_ABORT;
3586 xfs_trans_cancel(tp, cancel_flags);
3587 XFS_QM_DQRELE(mp, udqp);
3588 XFS_QM_DQRELE(mp, gdqp);
3590 if (!dp_joined_to_trans && (dp != NULL)) {
3591 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3601 * A fid routine that takes a pointer to a previously allocated
3602 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3612 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3613 (inst_t *)__return_address);
3614 ASSERT(sizeof(xfs_fid_t) >= sizeof(xfs_fid2_t));
3616 xfid = (xfs_fid2_t *)fidp;
3617 ip = XFS_BHVTOI(bdp);
3618 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3621 * use memcpy because the inode is a long long and there's no
3622 * assurance that xfid->fid_ino is properly aligned.
3624 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3625 xfid->fid_gen = ip->i_d.di_gen;
3642 vp = BHV_TO_VNODE(bdp);
3645 ip = XFS_BHVTOI(bdp);
3646 if (locktype == VRWLOCK_WRITE) {
3647 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3648 } else if (locktype == VRWLOCK_TRY_READ) {
3649 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3650 } else if (locktype == VRWLOCK_TRY_WRITE) {
3651 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3653 ASSERT((locktype == VRWLOCK_READ) ||
3654 (locktype == VRWLOCK_WRITE_DIRECT));
3655 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3673 vp = BHV_TO_VNODE(bdp);
3676 ip = XFS_BHVTOI(bdp);
3677 if (locktype == VRWLOCK_WRITE) {
3679 * In the write case, we may have added a new entry to
3680 * the reference cache. This might store a pointer to
3681 * an inode to be released in this inode. If it is there,
3682 * clear the pointer and release the inode after unlocking
3685 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3687 ASSERT((locktype == VRWLOCK_READ) ||
3688 (locktype == VRWLOCK_WRITE_DIRECT));
3689 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3701 xfs_inode_log_item_t *iip;
3704 ip = XFS_BHVTOI(bdp);
3708 if (XFS_FORCED_SHUTDOWN(mp))
3709 return XFS_ERROR(EIO);
3712 * Bypass inodes which have already been cleaned by
3713 * the inode flush clustering code inside xfs_iflush
3715 if ((ip->i_update_core == 0) &&
3716 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3719 if (flags & FLUSH_LOG) {
3720 if (iip && iip->ili_last_lsn) {
3721 xlog_t *log = mp->m_log;
3723 int s, log_flags = XFS_LOG_FORCE;
3725 s = GRANT_LOCK(log);
3726 sync_lsn = log->l_last_sync_lsn;
3727 GRANT_UNLOCK(log, s);
3729 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3732 if (flags & FLUSH_SYNC)
3733 log_flags |= XFS_LOG_SYNC;
3734 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3739 * We make this non-blocking if the inode is contended,
3740 * return EAGAIN to indicate to the caller that they
3741 * did not succeed. This prevents the flush path from
3742 * blocking on inodes inside another operation right
3743 * now, they get caught later by xfs_sync.
3745 if (flags & FLUSH_INODE) {
3748 if (xfs_ipincount(ip))
3751 if (flags & FLUSH_SYNC) {
3752 xfs_ilock(ip, XFS_ILOCK_SHARED);
3754 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3755 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3756 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3763 if (flags & FLUSH_SYNC)
3764 flush_flags = XFS_IFLUSH_SYNC;
3766 flush_flags = XFS_IFLUSH_ASYNC;
3768 error = xfs_iflush(ip, flush_flags);
3769 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3788 if (!capable(CAP_SYS_ADMIN))
3789 return XFS_ERROR(EPERM);
3791 ip = XFS_BHVTOI(bdp);
3794 if (XFS_FORCED_SHUTDOWN(mp))
3795 return XFS_ERROR(EIO);
3797 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3798 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3800 xfs_trans_cancel(tp, 0);
3803 xfs_ilock(ip, XFS_ILOCK_EXCL);
3804 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3806 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3807 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3809 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3811 error = xfs_trans_commit(tp, 0, NULL);
3827 vp = BHV_TO_VNODE(bdp);
3828 ip = XFS_BHVTOI(bdp);
3830 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3832 ASSERT(!VN_MAPPED(vp));
3834 /* bad inode, get out here ASAP */
3842 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3845 * Make sure the atime in the XFS inode is correct before freeing the
3848 xfs_synchronize_atime(ip);
3850 vnode_destroy_vobject(vp->v_vnode);
3852 /* If we have nothing to flush with this inode then complete the
3853 * teardown now, otherwise break the link between the xfs inode
3854 * and the linux inode and clean up the xfs inode later. This
3855 * avoids flushing the inode to disk during the delete operation
3858 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3859 xfs_ilock(ip, XFS_ILOCK_EXCL);
3861 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3863 xfs_mount_t *mp = ip->i_mount;
3865 /* Protect sync from us */
3866 XFS_MOUNT_ILOCK(mp);
3867 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3868 TAILQ_INSERT_TAIL(&mp->m_del_inodes, ip, i_reclaim);
3869 ip->i_flags |= XFS_IRECLAIMABLE;
3870 XFS_MOUNT_IUNLOCK(mp);
3881 xfs_ihash_t *ih = ip->i_hash;
3882 xfs_vnode_t *vp = XFS_ITOV_NULL(ip);
3885 if (vp && VN_BAD(vp))
3888 /* The hash lock here protects a thread in xfs_iget_core from
3889 * racing with us on linking the inode back with a vnode.
3890 * Once we have the XFS_IRECLAIM flag set it will not touch
3893 write_lock(&ih->ih_lock);
3894 if ((ip->i_flags & XFS_IRECLAIM) ||
3895 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3896 write_unlock(&ih->ih_lock);
3899 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3903 ip->i_flags |= XFS_IRECLAIM;
3904 write_unlock(&ih->ih_lock);
3907 * If the inode is still dirty, then flush it out. If the inode
3908 * is not in the AIL, then it will be OK to flush it delwri as
3909 * long as xfs_iflush() does not keep any references to the inode.
3910 * We leave that decision up to xfs_iflush() since it has the
3911 * knowledge of whether it's OK to simply do a delwri flush of
3912 * the inode or whether we need to wait until the inode is
3913 * pulled from the AIL.
3914 * We get the flush lock regardless, though, just to make sure
3915 * we don't free it while it is being flushed.
3917 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3919 xfs_ilock(ip, XFS_ILOCK_EXCL);
3923 if (ip->i_update_core ||
3924 ((ip->i_itemp != NULL) &&
3925 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3926 error = xfs_iflush(ip, sync_mode);
3928 * If we hit an error, typically because of filesystem
3929 * shutdown, we don't need to let vn_reclaim to know
3930 * because we're gonna reclaim the inode anyway.
3933 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3936 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3939 ASSERT(ip->i_update_core == 0);
3940 ASSERT(ip->i_itemp == NULL ||
3941 ip->i_itemp->ili_format.ilf_fields == 0);
3942 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3943 } else if (locked) {
3945 * We are not interested in doing an iflush if we're
3946 * in the process of shutting down the filesystem forcibly.
3947 * So, just reclaim the inode.
3950 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3959 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3963 xfs_inode_t *ip, *n;
3968 XFS_MOUNT_ILOCK(mp);
3969 TAILQ_FOREACH_SAFE(curr, &mp->m_del_inodes, i_reclaim, next) {
3972 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3974 if (xfs_ipincount(ip) ||
3975 !xfs_iflock_nowait(ip)) {
3976 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3980 XFS_MOUNT_IUNLOCK(mp);
3981 if (xfs_finish_reclaim(ip, noblock,
3982 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3991 XFS_MOUNT_IUNLOCK(mp);
3997 * xfs_alloc_file_space()
3998 * This routine allocates disk space for the given file.
4000 * If alloc_type == 0, this request is for an ALLOCSP type
4001 * request which will change the file size. In this case, no
4002 * DMAPI event will be generated by the call. A TRUNCATE event
4003 * will be generated later by xfs_setattr.
4005 * If alloc_type != 0, this request is for a RESVSP type
4006 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4007 * lower block boundary byte address is less than the file's
4016 xfs_alloc_file_space(
4023 xfs_mount_t *mp = ip->i_mount;
4025 xfs_filblks_t allocated_fsb;
4026 xfs_filblks_t allocatesize_fsb;
4027 xfs_extlen_t extsz, temp;
4028 xfs_fileoff_t startoffset_fsb;
4029 xfs_fsblock_t firstfsb;
4035 xfs_bmbt_irec_t imaps[1], *imapp;
4036 xfs_bmap_free_t free_list;
4037 uint qblocks, resblks, resrtextents;
4041 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4043 if (XFS_FORCED_SHUTDOWN(mp))
4044 return XFS_ERROR(EIO);
4046 rt = XFS_IS_REALTIME_INODE(ip);
4048 if (!(extsz = ip->i_d.di_extsize))
4049 extsz = mp->m_sb.sb_rextsize;
4051 extsz = ip->i_d.di_extsize;
4054 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4058 return XFS_ERROR(EINVAL);
4064 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4065 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4066 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4068 /* Generate a DMAPI event if needed. */
4069 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4070 (attr_flags&ATTR_DMI) == 0 &&
4071 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4072 xfs_off_t end_dmi_offset;
4074 end_dmi_offset = offset+len;
4075 if (end_dmi_offset > ip->i_d.di_size)
4076 end_dmi_offset = ip->i_d.di_size;
4077 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4078 offset, end_dmi_offset - offset,
4085 * Allocate file space until done or until there is an error
4088 while (allocatesize_fsb && !error) {
4092 * Determine space reservations for data/realtime.
4094 if (unlikely(extsz)) {
4095 s = startoffset_fsb;
4098 e = startoffset_fsb + allocatesize_fsb;
4099 if ((temp = do_mod(startoffset_fsb, extsz)))
4101 if ((temp = do_mod(e, extsz)))
4105 e = allocatesize_fsb;
4109 resrtextents = qblocks = (uint)(e - s);
4110 resrtextents /= mp->m_sb.sb_rextsize;
4111 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4112 quota_flag = XFS_QMOPT_RES_RTBLKS;
4115 resblks = qblocks = \
4116 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4117 quota_flag = XFS_QMOPT_RES_REGBLKS;
4121 * Allocate and setup the transaction.
4123 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4124 error = xfs_trans_reserve(tp, resblks,
4125 XFS_WRITE_LOG_RES(mp), resrtextents,
4126 XFS_TRANS_PERM_LOG_RES,
4127 XFS_WRITE_LOG_COUNT);
4129 * Check for running out of space
4133 * Free the transaction structure.
4135 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4136 xfs_trans_cancel(tp, 0);
4139 xfs_ilock(ip, XFS_ILOCK_EXCL);
4140 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4141 qblocks, 0, quota_flag);
4145 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4146 xfs_trans_ihold(tp, ip);
4149 * Issue the xfs_bmapi() call to allocate the blocks
4151 XFS_BMAP_INIT(&free_list, &firstfsb);
4152 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4153 allocatesize_fsb, bmapi_flag,
4154 &firstfsb, 0, imapp, &nimaps,
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;
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 */
4199 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4200 xfs_bmap_cancel(&free_list);
4201 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4203 error1: /* Just cancel transaction */
4204 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4205 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4206 goto dmapi_enospc_check;
4210 * Zero file bytes between startoff and endoff inclusive.
4211 * The iolock is held exclusive and no blocks are buffered.
4214 xfs_zero_remaining_bytes(
4219 xfs_bmbt_irec_t imap;
4220 xfs_fileoff_t offset_fsb;
4221 xfs_off_t lastoffset;
4224 xfs_mount_t *mp = ip->i_mount;
4228 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4229 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4230 mp->m_rtdev_targp : mp->m_ddev_targp);
4232 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4233 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4235 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4236 NULL, 0, &imap, &nimap, NULL, NULL);
4237 if (error || nimap < 1)
4239 ASSERT(imap.br_blockcount >= 1);
4240 ASSERT(imap.br_startoff == offset_fsb);
4241 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4242 if (lastoffset > endoff)
4243 lastoffset = endoff;
4244 if (imap.br_startblock == HOLESTARTBLOCK)
4246 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4247 if (imap.br_state == XFS_EXT_UNWRITTEN)
4250 XFS_BUF_UNWRITE(bp);
4252 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4254 if ((error = xfs_iowait(bp))) {
4255 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4256 mp, bp, XFS_BUF_ADDR(bp));
4259 memset(XFS_BUF_PTR(bp) +
4260 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4261 0, lastoffset - offset + 1);
4266 if ((error = xfs_iowait(bp))) {
4267 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4268 mp, bp, XFS_BUF_ADDR(bp));
4277 * xfs_free_file_space()
4278 * This routine frees disk space for the given file.
4280 * This routine is only called by xfs_change_file_space
4281 * for an UNRESVSP type call.
4289 xfs_free_file_space(
4298 xfs_off_t end_dmi_offset;
4299 xfs_fileoff_t endoffset_fsb;
4301 xfs_fsblock_t firstfsb;
4302 xfs_bmap_free_t free_list;
4304 xfs_bmbt_irec_t imap;
4312 xfs_fileoff_t startoffset_fsb;
4314 int need_iolock = 1;
4319 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4321 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4325 if (len <= 0) /* if nothing being freed */
4327 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4328 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4329 end_dmi_offset = offset + len;
4330 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4332 if (offset < ip->i_d.di_size &&
4333 (attr_flags & ATTR_DMI) == 0 &&
4334 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4335 if (end_dmi_offset > ip->i_d.di_size)
4336 end_dmi_offset = ip->i_d.di_size;
4337 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4338 offset, end_dmi_offset - offset,
4339 AT_DELAY_FLAG(attr_flags), NULL);
4344 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4345 if (attr_flags & ATTR_NOLOCK)
4348 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4349 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4352 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4354 ilen = len + (offset & (rounding - 1));
4355 ioffset = offset & ~(rounding - 1);
4356 if (ilen & (rounding - 1))
4357 ilen = (ilen + rounding) & ~(rounding - 1);
4359 if (VN_CACHED(vp) != 0) {
4360 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4361 ctooff(offtoct(ioffset)), -1);
4362 XVOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4363 -1, FI_REMAPF_LOCKED);
4367 * Need to zero the stuff we're not freeing, on disk.
4368 * If its a realtime file & can't use unwritten extents then we
4369 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4370 * will take care of it for us.
4372 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4374 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4375 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4377 goto out_unlock_iolock;
4378 ASSERT(nimap == 0 || nimap == 1);
4379 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4382 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4383 block = imap.br_startblock;
4384 mod = do_div(block, mp->m_sb.sb_rextsize);
4386 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4389 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4390 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4392 goto out_unlock_iolock;
4393 ASSERT(nimap == 0 || nimap == 1);
4394 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4395 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4397 if (mod && (mod != mp->m_sb.sb_rextsize))
4398 endoffset_fsb -= mod;
4401 if ((done = (endoffset_fsb <= startoffset_fsb)))
4403 * One contiguous piece to clear
4405 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4408 * Some full blocks, possibly two pieces to clear
4410 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4411 error = xfs_zero_remaining_bytes(ip, offset,
4412 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4414 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4415 error = xfs_zero_remaining_bytes(ip,
4416 XFS_FSB_TO_B(mp, endoffset_fsb),
4421 * free file space until done or until there is an error
4423 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4424 while (!error && !done) {
4427 * allocate and setup the transaction
4429 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4430 error = xfs_trans_reserve(tp,
4432 XFS_WRITE_LOG_RES(mp),
4434 XFS_TRANS_PERM_LOG_RES,
4435 XFS_WRITE_LOG_COUNT);
4438 * check for running out of space
4442 * Free the transaction structure.
4444 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4445 xfs_trans_cancel(tp, 0);
4448 xfs_ilock(ip, XFS_ILOCK_EXCL);
4449 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4450 ip->i_udquot, ip->i_gdquot, resblks, 0,
4451 XFS_QMOPT_RES_REGBLKS);
4455 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4456 xfs_trans_ihold(tp, ip);
4459 * issue the bunmapi() call to free the blocks
4461 XFS_BMAP_INIT(&free_list, &firstfsb);
4462 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4463 endoffset_fsb - startoffset_fsb,
4464 0, 2, &firstfsb, &free_list, NULL, &done);
4470 * complete the transaction
4472 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4477 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4478 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4483 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4487 xfs_bmap_cancel(&free_list);
4489 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4490 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4496 * xfs_change_file_space()
4497 * This routine allocates or frees disk space for the given file.
4498 * The user specified parameters are checked for alignment and size
4507 xfs_change_file_space(
4521 xfs_off_t startoffset;
4527 vp = BHV_TO_VNODE(bdp);
4528 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4530 ip = XFS_BHVTOI(bdp);
4534 * must be a regular file and have write permission
4537 return XFS_ERROR(EINVAL);
4539 xfs_ilock(ip, XFS_ILOCK_SHARED);
4541 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4542 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4546 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4548 switch (bf->l_whence) {
4549 case 0: /*SEEK_SET*/
4551 case 1: /*SEEK_CUR*/
4552 bf->l_start += offset;
4554 case 2: /*SEEK_END*/
4555 bf->l_start += ip->i_d.di_size;
4558 return XFS_ERROR(EINVAL);
4561 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4563 if ( (bf->l_start < 0)
4564 || (bf->l_start > XFS_MAXIOFFSET(mp))
4565 || (bf->l_start + llen < 0)
4566 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4567 return XFS_ERROR(EINVAL);
4571 startoffset = bf->l_start;
4572 fsize = ip->i_d.di_size;
4575 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4577 * These calls do NOT zero the data space allocated to the file,
4578 * nor do they change the file size.
4580 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4582 * These calls cause the new file data to be zeroed and the file
4583 * size to be changed.
4585 setprealloc = clrprealloc = 0;
4588 case XFS_IOC_RESVSP:
4589 case XFS_IOC_RESVSP64:
4590 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4597 case XFS_IOC_UNRESVSP:
4598 case XFS_IOC_UNRESVSP64:
4599 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4604 case XFS_IOC_ALLOCSP:
4605 case XFS_IOC_ALLOCSP64:
4606 case XFS_IOC_FREESP:
4607 case XFS_IOC_FREESP64:
4608 if (startoffset > fsize) {
4609 error = xfs_alloc_file_space(ip, fsize,
4610 startoffset - fsize, 0, attr_flags);
4615 va.va_mask = XFS_AT_SIZE;
4616 va.va_size = startoffset;
4618 error = xfs_setattr(bdp, &va, attr_flags, credp);
4628 return XFS_ERROR(EINVAL);
4632 * update the inode timestamp, mode, and prealloc flag bits
4634 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4636 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4639 xfs_trans_cancel(tp, 0);
4643 xfs_ilock(ip, XFS_ILOCK_EXCL);
4645 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4646 xfs_trans_ihold(tp, ip);
4648 if ((attr_flags & ATTR_DMI) == 0) {
4649 ip->i_d.di_mode &= ~S_ISUID;
4652 * Note that we don't have to worry about mandatory
4653 * file locking being disabled here because we only
4654 * clear the S_ISGID bit if the Group execute bit is
4655 * on, but if it was on then mandatory locking wouldn't
4656 * have been enabled.
4658 if (ip->i_d.di_mode & S_IXGRP)
4659 ip->i_d.di_mode &= ~S_ISGID;
4661 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4664 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4665 else if (clrprealloc)
4666 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4668 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4669 xfs_trans_set_sync(tp);
4671 error = xfs_trans_commit(tp, 0, NULL);
4673 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4679 xfs_vnodeops_t xfs_vnodeops = {
4680 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4681 .vop_open = xfs_open,
4682 .vop_read = xfs_read,
4683 #ifdef HAVE_SENDFILE
4684 .vop_sendfile = xfs_sendfile,
4686 .vop_write = xfs_write,
4687 .vop_ioctl = xfs_ioctl,
4688 .vop_getattr = xfs_getattr,
4689 .vop_setattr = xfs_setattr,
4690 .vop_access = xfs_access,
4691 .vop_lookup = xfs_lookup,
4692 .vop_create = xfs_create,
4693 .vop_remove = xfs_remove,
4694 .vop_link = xfs_link,
4695 .vop_rename = xfs_rename,
4696 .vop_mkdir = xfs_mkdir,
4697 .vop_rmdir = xfs_rmdir,
4698 .vop_readdir = xfs_readdir,
4699 .vop_symlink = xfs_symlink,
4700 .vop_readlink = xfs_readlink,
4701 .vop_fsync = xfs_fsync,
4702 .vop_inactive = xfs_inactive,
4703 .vop_fid2 = xfs_fid2,
4704 .vop_rwlock = xfs_rwlock,
4705 .vop_rwunlock = xfs_rwunlock,
4706 .vop_bmap = xfs_bmap,
4707 .vop_reclaim = xfs_reclaim,
4708 .vop_attr_get = xfs_attr_get,
4709 .vop_attr_set = xfs_attr_set,
4710 .vop_attr_remove = xfs_attr_remove,
4711 .vop_attr_list = xfs_attr_list,
4712 .vop_link_removed = (xfs_vop_link_removed_t)fs_noval,
4713 .vop_vnode_change = (xfs_vop_vnode_change_t)fs_noval,
4714 .vop_tosspages = fs_tosspages,
4715 .vop_flushinval_pages = fs_flushinval_pages,
4716 .vop_flush_pages = fs_flush_pages,
4717 .vop_release = xfs_release,
4718 .vop_iflush = xfs_inode_flush,