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
19 #include <linux/capability.h>
23 #include "xfs_types.h"
27 #include "xfs_trans.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_da_btree.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_alloc_btree.h"
37 #include "xfs_ialloc_btree.h"
38 #include "xfs_dir_sf.h"
39 #include "xfs_dir2_sf.h"
40 #include "xfs_attr_sf.h"
41 #include "xfs_dinode.h"
42 #include "xfs_inode.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_dir_leaf.h"
45 #include "xfs_itable.h"
46 #include "xfs_btree.h"
47 #include "xfs_ialloc.h"
48 #include "xfs_alloc.h"
52 #include "xfs_error.h"
53 #include "xfs_quota.h"
54 #include "xfs_utils.h"
55 #include "xfs_rtalloc.h"
56 #include "xfs_refcache.h"
57 #include "xfs_trans_space.h"
58 #include "xfs_log_priv.h"
63 * The maximum pathlen is 1024 bytes. Since the minimum file system
64 * blocksize is 512 bytes, we can get a max of 2 extents back from
67 #define SYMLINK_MAPS 2
70 * For xfs, we check that the file isn't too big to be opened by this kernel.
71 * No other open action is required for regular files. Devices are handled
72 * through the specfs file system, pipes through fifofs. Device and
73 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
74 * when a new vnode is first looked up or created.
85 vp = BHV_TO_VNODE(bdp);
88 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
89 return XFS_ERROR(EIO);
92 * If it's a directory with any blocks, read-ahead block 0
93 * as we're almost certain to have the next operation be a read there.
95 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
96 mode = xfs_ilock_map_shared(ip);
97 if (ip->i_d.di_nextents > 0)
98 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
99 xfs_iunlock(ip, mode);
119 vp = BHV_TO_VNODE(bdp);
120 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
122 ip = XFS_BHVTOI(bdp);
125 if (XFS_FORCED_SHUTDOWN(mp))
126 return XFS_ERROR(EIO);
128 if (!(flags & ATTR_LAZY))
129 xfs_ilock(ip, XFS_ILOCK_SHARED);
131 vap->va_size = ip->i_d.di_size;
132 if (vap->va_mask == XFS_AT_SIZE)
136 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
137 vap->va_nodeid = ip->i_ino;
139 vap->va_nodeid += mp->m_inoadd;
141 vap->va_nlink = ip->i_d.di_nlink;
144 * Quick exit for non-stat callers
147 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
148 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
152 * Copy from in-core inode.
154 vap->va_mode = ip->i_d.di_mode;
155 vap->va_uid = ip->i_d.di_uid;
156 vap->va_gid = ip->i_d.di_gid;
157 vap->va_projid = ip->i_d.di_projid;
160 * Check vnode type block/char vs. everything else.
162 switch (ip->i_d.di_mode & S_IFMT) {
165 vap->va_rdev = ip->i_df.if_u2.if_rdev;
166 vap->va_blocksize = BLKDEV_IOSIZE;
171 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
172 vap->va_blocksize = xfs_preferred_iosize(mp);
176 * If the file blocks are being allocated from a
177 * realtime partition, then return the inode's
178 * realtime extent size or the realtime volume's
181 vap->va_blocksize = ip->i_d.di_extsize ?
182 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
183 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
188 vn_atime_to_timespec(vp, &vap->va_atime);
189 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
190 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
191 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
192 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
195 * Exit for stat callers. See if any of the rest of the fields
196 * to be filled in are needed.
199 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
200 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
204 * Convert di_flags to xflags.
206 vap->va_xflags = xfs_ip2xflags(ip);
209 * Exit for inode revalidate. See if any of the rest of
210 * the fields to be filled in are needed.
213 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
214 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
217 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
219 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
220 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
224 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
225 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
226 ip->i_d.di_anextents;
228 vap->va_anextents = 0;
229 vap->va_gen = ip->i_d.di_gen;
232 if (!(flags & ATTR_LAZY))
233 xfs_iunlock(ip, XFS_ILOCK_SHARED);
259 xfs_prid_t projid=0, iprojid=0;
260 int mandlock_before, mandlock_after;
261 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
265 vp = BHV_TO_VNODE(bdp);
266 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
268 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
269 return XFS_ERROR(EROFS);
272 * Cannot set certain attributes.
275 if (mask & XFS_AT_NOSET) {
276 return XFS_ERROR(EINVAL);
279 ip = XFS_BHVTOI(bdp);
282 if (XFS_FORCED_SHUTDOWN(mp))
283 return XFS_ERROR(EIO);
286 * Timestamps do not need to be logged and hence do not
287 * need to be done within a transaction.
289 if (mask & XFS_AT_UPDTIMES) {
290 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
291 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
292 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
293 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
294 xfs_ichgtime(ip, timeflags);
298 olddquot1 = olddquot2 = NULL;
302 * If disk quotas is on, we make sure that the dquots do exist on disk,
303 * before we start any other transactions. Trying to do this later
304 * is messy. We don't care to take a readlock to look at the ids
305 * in inode here, because we can't hold it across the trans_reserve.
306 * If the IDs do change before we take the ilock, we're covered
307 * because the i_*dquot fields will get updated anyway.
309 if (XFS_IS_QUOTA_ON(mp) &&
310 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
313 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
315 qflags |= XFS_QMOPT_UQUOTA;
317 uid = ip->i_d.di_uid;
319 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
321 qflags |= XFS_QMOPT_GQUOTA;
323 gid = ip->i_d.di_gid;
325 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
326 projid = vap->va_projid;
327 qflags |= XFS_QMOPT_PQUOTA;
329 projid = ip->i_d.di_projid;
332 * We take a reference when we initialize udqp and gdqp,
333 * so it is important that we never blindly double trip on
334 * the same variable. See xfs_create() for an example.
336 ASSERT(udqp == NULL);
337 ASSERT(gdqp == NULL);
338 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
345 * For the other attributes, we acquire the inode lock and
346 * first do an error checking pass.
349 lock_flags = XFS_ILOCK_EXCL;
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? */
382 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
385 * Change various properties of a file.
386 * Only the owner or users with CAP_FOWNER
387 * capability may do these things.
390 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
391 XFS_AT_GID|XFS_AT_PROJID)) {
393 * CAP_FOWNER overrides the following restrictions:
395 * The user ID of the calling process must be equal
396 * to the file owner ID, except in cases where the
397 * CAP_FSETID capability is applicable.
399 if (!file_owner && !capable(CAP_FOWNER)) {
400 code = XFS_ERROR(EPERM);
405 * CAP_FSETID overrides the following restrictions:
407 * The effective user ID of the calling process shall match
408 * the file owner when setting the set-user-ID and
409 * set-group-ID bits on that file.
411 * The effective group ID or one of the supplementary group
412 * IDs of the calling process shall match the group owner of
413 * the file when setting the set-group-ID bit on that file
415 if (mask & XFS_AT_MODE) {
418 if ((vap->va_mode & S_ISUID) && !file_owner)
420 if ((vap->va_mode & S_ISGID) &&
421 !in_group_p((gid_t)ip->i_d.di_gid))
424 /* Linux allows this, Irix doesn't. */
425 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
428 if (m && !capable(CAP_FSETID))
434 * Change file ownership. Must be the owner or privileged.
435 * If the system was configured with the "restricted_chown"
436 * option, the owner is not permitted to give away the file,
437 * and can change the group id only to a group of which he
438 * or she is a member.
440 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
442 * These IDs could have changed since we last looked at them.
443 * But, we're assured that if the ownership did change
444 * while we didn't have the inode locked, inode's dquot(s)
445 * would have changed also.
447 iuid = ip->i_d.di_uid;
448 iprojid = ip->i_d.di_projid;
449 igid = ip->i_d.di_gid;
450 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
451 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
452 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
456 * CAP_CHOWN overrides the following restrictions:
458 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
459 * shall override the restriction that a process cannot
460 * change the user ID of a file it owns and the restriction
461 * that the group ID supplied to the chown() function
462 * shall be equal to either the group ID or one of the
463 * supplementary group IDs of the calling process.
465 if (restricted_chown &&
466 (iuid != uid || (igid != gid &&
467 !in_group_p((gid_t)gid))) &&
468 !capable(CAP_CHOWN)) {
469 code = XFS_ERROR(EPERM);
473 * Do a quota reservation only if uid/projid/gid is actually
476 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
477 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
478 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
480 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
481 capable(CAP_FOWNER) ?
482 XFS_QMOPT_FORCE_RES : 0);
483 if (code) /* out of quota */
489 * Truncate file. Must have write permission and not be a directory.
491 if (mask & XFS_AT_SIZE) {
492 /* Short circuit the truncate case for zero length files */
493 if ((vap->va_size == 0) &&
494 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
495 xfs_iunlock(ip, XFS_ILOCK_EXCL);
496 lock_flags &= ~XFS_ILOCK_EXCL;
497 if (mask & XFS_AT_CTIME)
498 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
504 code = XFS_ERROR(EISDIR);
506 } else if (!VN_ISREG(vp)) {
507 code = XFS_ERROR(EINVAL);
511 * Make sure that the dquots are attached to the inode.
513 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
518 * Change file access or modified times.
520 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
522 if ((flags & ATTR_UTIME) &&
523 !capable(CAP_FOWNER)) {
524 code = XFS_ERROR(EPERM);
531 * Change extent size or realtime flag.
533 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
535 * Can't change extent size if any extents are allocated.
537 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
538 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
540 code = XFS_ERROR(EINVAL); /* EFBIG? */
545 * Can't change realtime flag if any extents are allocated.
547 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
548 (mask & XFS_AT_XFLAGS) &&
549 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
550 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
551 code = XFS_ERROR(EINVAL); /* EFBIG? */
555 * Extent size must be a multiple of the appropriate block
556 * size, if set at all.
558 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
561 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
562 ((mask & XFS_AT_XFLAGS) &&
563 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
564 size = mp->m_sb.sb_rextsize <<
565 mp->m_sb.sb_blocklog;
567 size = mp->m_sb.sb_blocksize;
569 if (vap->va_extsize % size) {
570 code = XFS_ERROR(EINVAL);
575 * If realtime flag is set then must have realtime data.
577 if ((mask & XFS_AT_XFLAGS) &&
578 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
579 if ((mp->m_sb.sb_rblocks == 0) ||
580 (mp->m_sb.sb_rextsize == 0) ||
581 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
582 code = XFS_ERROR(EINVAL);
588 * Can't modify an immutable/append-only file unless
589 * we have appropriate permission.
591 if ((mask & XFS_AT_XFLAGS) &&
593 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
595 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
596 !capable(CAP_LINUX_IMMUTABLE)) {
597 code = XFS_ERROR(EPERM);
603 * Now we can make the changes. Before we join the inode
604 * to the transaction, if XFS_AT_SIZE is set then take care of
605 * the part of the truncation that must be done without the
606 * inode lock. This needs to be done before joining the inode
607 * to the transaction, because the inode cannot be unlocked
608 * once it is a part of the transaction.
610 if (mask & XFS_AT_SIZE) {
612 if ((vap->va_size > ip->i_d.di_size) &&
613 (flags & ATTR_NOSIZETOK) == 0) {
614 code = xfs_igrow_start(ip, vap->va_size, credp);
616 xfs_iunlock(ip, XFS_ILOCK_EXCL);
617 vn_iowait(vp); /* wait for the completion of any pending DIOs */
619 code = xfs_itruncate_data(ip, vap->va_size);
622 lock_flags &= ~XFS_ILOCK_EXCL;
623 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
626 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
627 if ((code = xfs_trans_reserve(tp, 0,
628 XFS_ITRUNCATE_LOG_RES(mp), 0,
629 XFS_TRANS_PERM_LOG_RES,
630 XFS_ITRUNCATE_LOG_COUNT))) {
631 xfs_trans_cancel(tp, 0);
633 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
636 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
637 xfs_ilock(ip, XFS_ILOCK_EXCL);
641 xfs_trans_ijoin(tp, ip, lock_flags);
642 xfs_trans_ihold(tp, ip);
645 /* determine whether mandatory locking mode changes */
646 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
649 * Truncate file. Must have write permission and not be a directory.
651 if (mask & XFS_AT_SIZE) {
652 if (vap->va_size > ip->i_d.di_size) {
653 xfs_igrow_finish(tp, ip, vap->va_size,
654 !(flags & ATTR_DMI));
655 } else if ((vap->va_size <= ip->i_d.di_size) ||
656 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
658 * signal a sync transaction unless
659 * we're truncating an already unlinked
660 * file on a wsync filesystem
662 code = xfs_itruncate_finish(&tp, ip,
663 (xfs_fsize_t)vap->va_size,
665 ((ip->i_d.di_nlink != 0 ||
666 !(mp->m_flags & XFS_MOUNT_WSYNC))
673 * Have to do this even if the file's size doesn't change.
675 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
679 * Change file access modes.
681 if (mask & XFS_AT_MODE) {
682 ip->i_d.di_mode &= S_IFMT;
683 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
685 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
686 timeflags |= XFS_ICHGTIME_CHG;
690 * Change file ownership. Must be the owner or privileged.
691 * If the system was configured with the "restricted_chown"
692 * option, the owner is not permitted to give away the file,
693 * and can change the group id only to a group of which he
694 * or she is a member.
696 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
698 * CAP_FSETID overrides the following restrictions:
700 * The set-user-ID and set-group-ID bits of a file will be
701 * cleared upon successful return from chown()
703 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
704 !capable(CAP_FSETID)) {
705 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
709 * Change the ownerships and register quota modifications
710 * in the transaction.
713 if (XFS_IS_UQUOTA_ON(mp)) {
714 ASSERT(mask & XFS_AT_UID);
716 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
717 &ip->i_udquot, udqp);
719 ip->i_d.di_uid = uid;
722 if (XFS_IS_GQUOTA_ON(mp)) {
723 ASSERT(!XFS_IS_PQUOTA_ON(mp));
724 ASSERT(mask & XFS_AT_GID);
726 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
727 &ip->i_gdquot, gdqp);
729 ip->i_d.di_gid = gid;
731 if (iprojid != projid) {
732 if (XFS_IS_PQUOTA_ON(mp)) {
733 ASSERT(!XFS_IS_GQUOTA_ON(mp));
734 ASSERT(mask & XFS_AT_PROJID);
736 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
737 &ip->i_gdquot, gdqp);
739 ip->i_d.di_projid = projid;
741 * We may have to rev the inode as well as
742 * the superblock version number since projids didn't
743 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
745 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
746 xfs_bump_ino_vers2(tp, ip);
749 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
750 timeflags |= XFS_ICHGTIME_CHG;
755 * Change file access or modified times.
757 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
758 if (mask & XFS_AT_ATIME) {
759 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
760 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
761 ip->i_update_core = 1;
762 timeflags &= ~XFS_ICHGTIME_ACC;
764 if (mask & XFS_AT_MTIME) {
765 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
766 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
767 timeflags &= ~XFS_ICHGTIME_MOD;
768 timeflags |= XFS_ICHGTIME_CHG;
770 if (tp && (flags & ATTR_UTIME))
771 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
775 * Change XFS-added attributes.
777 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
778 if (mask & XFS_AT_EXTSIZE) {
780 * Converting bytes to fs blocks.
782 ip->i_d.di_extsize = vap->va_extsize >>
783 mp->m_sb.sb_blocklog;
785 if (mask & XFS_AT_XFLAGS) {
788 /* can't set PREALLOC this way, just preserve it */
789 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
790 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
791 di_flags |= XFS_DIFLAG_IMMUTABLE;
792 if (vap->va_xflags & XFS_XFLAG_APPEND)
793 di_flags |= XFS_DIFLAG_APPEND;
794 if (vap->va_xflags & XFS_XFLAG_SYNC)
795 di_flags |= XFS_DIFLAG_SYNC;
796 if (vap->va_xflags & XFS_XFLAG_NOATIME)
797 di_flags |= XFS_DIFLAG_NOATIME;
798 if (vap->va_xflags & XFS_XFLAG_NODUMP)
799 di_flags |= XFS_DIFLAG_NODUMP;
800 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
801 di_flags |= XFS_DIFLAG_PROJINHERIT;
802 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
803 di_flags |= XFS_DIFLAG_NODEFRAG;
804 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
805 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
806 di_flags |= XFS_DIFLAG_RTINHERIT;
807 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
808 di_flags |= XFS_DIFLAG_NOSYMLINKS;
809 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
810 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
811 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
812 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
813 di_flags |= XFS_DIFLAG_REALTIME;
814 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
816 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
818 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
819 di_flags |= XFS_DIFLAG_EXTSIZE;
821 ip->i_d.di_flags = di_flags;
823 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
824 timeflags |= XFS_ICHGTIME_CHG;
828 * Change file inode change time only if XFS_AT_CTIME set
829 * AND we have been called by a DMI function.
832 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
833 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
834 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
835 ip->i_update_core = 1;
836 timeflags &= ~XFS_ICHGTIME_CHG;
840 * Send out timestamp changes that need to be set to the
841 * current time. Not done when called by a DMI function.
843 if (timeflags && !(flags & ATTR_DMI))
844 xfs_ichgtime(ip, timeflags);
846 XFS_STATS_INC(xs_ig_attrchg);
849 * If this is a synchronous mount, make sure that the
850 * transaction goes to disk before returning to the user.
851 * This is slightly sub-optimal in that truncates require
852 * two sync transactions instead of one for wsync filesystems.
853 * One for the truncate and one for the timestamps since we
854 * don't want to change the timestamps unless we're sure the
855 * truncate worked. Truncates are less than 1% of the laddis
856 * mix so this probably isn't worth the trouble to optimize.
860 if (mp->m_flags & XFS_MOUNT_WSYNC)
861 xfs_trans_set_sync(tp);
863 code = xfs_trans_commit(tp, commit_flags, NULL);
867 * If the (regular) file's mandatory locking mode changed, then
868 * notify the vnode. We do this under the inode lock to prevent
869 * racing calls to vop_vnode_change.
871 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
872 if (mandlock_before != mandlock_after) {
873 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
877 xfs_iunlock(ip, lock_flags);
880 * Release any dquot(s) the inode had kept before chown.
882 XFS_QM_DQRELE(mp, olddquot1);
883 XFS_QM_DQRELE(mp, olddquot2);
884 XFS_QM_DQRELE(mp, udqp);
885 XFS_QM_DQRELE(mp, gdqp);
891 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
892 !(flags & ATTR_DMI)) {
893 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
894 NULL, DM_RIGHT_NULL, NULL, NULL,
895 0, 0, AT_DELAY_FLAG(flags));
900 commit_flags |= XFS_TRANS_ABORT;
903 XFS_QM_DQRELE(mp, udqp);
904 XFS_QM_DQRELE(mp, gdqp);
906 xfs_trans_cancel(tp, commit_flags);
908 if (lock_flags != 0) {
909 xfs_iunlock(ip, lock_flags);
917 * Null conversion from vnode mode bits to inode mode bits, as in efs.
928 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
929 (inst_t *)__return_address);
931 ip = XFS_BHVTOI(bdp);
932 xfs_ilock(ip, XFS_ILOCK_SHARED);
933 error = xfs_iaccess(ip, mode, credp);
934 xfs_iunlock(ip, XFS_ILOCK_SHARED);
958 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
964 vp = BHV_TO_VNODE(bdp);
965 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
967 ip = XFS_BHVTOI(bdp);
970 if (XFS_FORCED_SHUTDOWN(mp))
971 return XFS_ERROR(EIO);
973 xfs_ilock(ip, XFS_ILOCK_SHARED);
975 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
977 offset = uiop->uio_offset;
978 count = uiop->uio_resid;
981 error = XFS_ERROR(EINVAL);
990 * See if the symlink is stored inline.
992 pathlen = (int)ip->i_d.di_size;
994 if (ip->i_df.if_flags & XFS_IFINLINE) {
995 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
999 * Symlink not inline. Call bmap to get it in.
1001 nmaps = SYMLINK_MAPS;
1003 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1004 0, NULL, 0, mval, &nmaps, NULL, NULL);
1010 for (n = 0; n < nmaps; n++) {
1011 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1012 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1013 bp = xfs_buf_read(mp->m_ddev_targp, d,
1014 BTOBB(byte_cnt), 0);
1015 error = XFS_BUF_GETERROR(bp);
1017 xfs_ioerror_alert("xfs_readlink",
1018 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1022 if (pathlen < byte_cnt)
1024 pathlen -= byte_cnt;
1026 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1033 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1041 * This is called to sync the inode and its data out to disk.
1042 * We need to hold the I/O lock while flushing the data, and
1043 * the inode lock while flushing the inode. The inode lock CANNOT
1044 * be held while flushing the data, so acquire after we're done
1058 int log_flushed = 0, changed = 1;
1060 vn_trace_entry(BHV_TO_VNODE(bdp),
1061 __FUNCTION__, (inst_t *)__return_address);
1063 ip = XFS_BHVTOI(bdp);
1065 ASSERT(start >= 0 && stop >= -1);
1067 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1068 return XFS_ERROR(EIO);
1071 * We always need to make sure that the required inode state
1072 * is safe on disk. The vnode might be clean but because
1073 * of committed transactions that haven't hit the disk yet.
1074 * Likewise, there could be unflushed non-transactional
1075 * changes to the inode core that have to go to disk.
1077 * The following code depends on one assumption: that
1078 * any transaction that changes an inode logs the core
1079 * because it has to change some field in the inode core
1080 * (typically nextents or nblocks). That assumption
1081 * implies that any transactions against an inode will
1082 * catch any non-transactional updates. If inode-altering
1083 * transactions exist that violate this assumption, the
1084 * code breaks. Right now, it figures that if the involved
1085 * update_* field is clear and the inode is unpinned, the
1086 * inode is clean. Either it's been flushed or it's been
1087 * committed and the commit has hit the disk unpinning the inode.
1088 * (Note that xfs_inode_item_format() called at commit clears
1089 * the update_* fields.)
1091 xfs_ilock(ip, XFS_ILOCK_SHARED);
1093 /* If we are flushing data then we care about update_size
1094 * being set, otherwise we care about update_core
1096 if ((flag & FSYNC_DATA) ?
1097 (ip->i_update_size == 0) :
1098 (ip->i_update_core == 0)) {
1100 * Timestamps/size haven't changed since last inode
1101 * flush or inode transaction commit. That means
1102 * either nothing got written or a transaction
1103 * committed which caught the updates. If the
1104 * latter happened and the transaction hasn't
1105 * hit the disk yet, the inode will be still
1106 * be pinned. If it is, force the log.
1109 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1111 if (xfs_ipincount(ip)) {
1112 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1114 ((flag & FSYNC_WAIT)
1115 ? XFS_LOG_SYNC : 0),
1119 * If the inode is not pinned and nothing
1120 * has changed we don't need to flush the
1128 * Kick off a transaction to log the inode
1129 * core to get the updates. Make it
1130 * sync if FSYNC_WAIT is passed in (which
1131 * is done by everybody but specfs). The
1132 * sync transaction will also force the log.
1134 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1135 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1136 if ((error = xfs_trans_reserve(tp, 0,
1137 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1139 xfs_trans_cancel(tp, 0);
1142 xfs_ilock(ip, XFS_ILOCK_EXCL);
1145 * Note - it's possible that we might have pushed
1146 * ourselves out of the way during trans_reserve
1147 * which would flush the inode. But there's no
1148 * guarantee that the inode buffer has actually
1149 * gone out yet (it's delwri). Plus the buffer
1150 * could be pinned anyway if it's part of an
1151 * inode in another recent transaction. So we
1152 * play it safe and fire off the transaction anyway.
1154 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1155 xfs_trans_ihold(tp, ip);
1156 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1157 if (flag & FSYNC_WAIT)
1158 xfs_trans_set_sync(tp);
1159 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1161 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1164 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1166 * If the log write didn't issue an ordered tag we need
1167 * to flush the disk cache for the data device now.
1170 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1173 * If this inode is on the RT dev we need to flush that
1176 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1177 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1184 * This is called by xfs_inactive to free any blocks beyond eof,
1185 * when the link count isn't zero.
1188 xfs_inactive_free_eofblocks(
1194 xfs_fileoff_t end_fsb;
1195 xfs_fileoff_t last_fsb;
1196 xfs_filblks_t map_len;
1198 xfs_bmbt_irec_t imap;
1201 * Figure out if there are any blocks beyond the end
1202 * of the file. If not, then there is nothing to do.
1204 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1205 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1206 map_len = last_fsb - end_fsb;
1211 xfs_ilock(ip, XFS_ILOCK_SHARED);
1212 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1213 NULL, 0, &imap, &nimaps, NULL, NULL);
1214 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1216 if (!error && (nimaps != 0) &&
1217 (imap.br_startblock != HOLESTARTBLOCK ||
1218 ip->i_delayed_blks)) {
1220 * Attach the dquots to the inode up front.
1222 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1226 * There are blocks after the end of file.
1227 * Free them up now by truncating the file to
1230 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1233 * Do the xfs_itruncate_start() call before
1234 * reserving any log space because
1235 * itruncate_start will call into the buffer
1236 * cache and we can't
1237 * do that within a transaction.
1239 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1240 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1243 error = xfs_trans_reserve(tp, 0,
1244 XFS_ITRUNCATE_LOG_RES(mp),
1245 0, XFS_TRANS_PERM_LOG_RES,
1246 XFS_ITRUNCATE_LOG_COUNT);
1248 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1249 xfs_trans_cancel(tp, 0);
1250 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1254 xfs_ilock(ip, XFS_ILOCK_EXCL);
1255 xfs_trans_ijoin(tp, ip,
1258 xfs_trans_ihold(tp, ip);
1260 error = xfs_itruncate_finish(&tp, ip,
1265 * If we get an error at this point we
1266 * simply don't bother truncating the file.
1269 xfs_trans_cancel(tp,
1270 (XFS_TRANS_RELEASE_LOG_RES |
1273 error = xfs_trans_commit(tp,
1274 XFS_TRANS_RELEASE_LOG_RES,
1277 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1283 * Free a symlink that has blocks associated with it.
1286 xfs_inactive_symlink_rmt(
1294 xfs_fsblock_t first_block;
1295 xfs_bmap_free_t free_list;
1298 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1306 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1308 * We're freeing a symlink that has some
1309 * blocks allocated to it. Free the
1310 * blocks here. We know that we've got
1311 * either 1 or 2 extents and that we can
1312 * free them all in one bunmapi call.
1314 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1315 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1316 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1317 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1318 xfs_trans_cancel(tp, 0);
1323 * Lock the inode, fix the size, and join it to the transaction.
1324 * Hold it so in the normal path, we still have it locked for
1325 * the second transaction. In the error paths we need it
1326 * held so the cancel won't rele it, see below.
1328 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1329 size = (int)ip->i_d.di_size;
1330 ip->i_d.di_size = 0;
1331 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1332 xfs_trans_ihold(tp, ip);
1333 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1335 * Find the block(s) so we can inval and unmap them.
1338 XFS_BMAP_INIT(&free_list, &first_block);
1339 nmaps = ARRAY_SIZE(mval);
1340 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1341 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1345 * Invalidate the block(s).
1347 for (i = 0; i < nmaps; i++) {
1348 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1349 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1350 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1351 xfs_trans_binval(tp, bp);
1354 * Unmap the dead block(s) to the free_list.
1356 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1357 &first_block, &free_list, NULL, &done)))
1361 * Commit the first transaction. This logs the EFI and the inode.
1363 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1366 * The transaction must have been committed, since there were
1367 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1368 * The new tp has the extent freeing and EFDs.
1372 * The first xact was committed, so add the inode to the new one.
1373 * Mark it dirty so it will be logged and moved forward in the log as
1374 * part of every commit.
1376 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1377 xfs_trans_ihold(tp, ip);
1378 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1380 * Get a new, empty transaction to return to our caller.
1382 ntp = xfs_trans_dup(tp);
1384 * Commit the transaction containing extent freeing and EFDs.
1385 * If we get an error on the commit here or on the reserve below,
1386 * we need to unlock the inode since the new transaction doesn't
1387 * have the inode attached.
1389 error = xfs_trans_commit(tp, 0, NULL);
1392 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1396 * Remove the memory for extent descriptions (just bookkeeping).
1398 if (ip->i_df.if_bytes)
1399 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1400 ASSERT(ip->i_df.if_bytes == 0);
1402 * Put an itruncate log reservation in the new transaction
1405 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1406 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1407 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1411 * Return with the inode locked but not joined to the transaction.
1417 xfs_bmap_cancel(&free_list);
1420 * Have to come here with the inode locked and either
1421 * (held and in the transaction) or (not in the transaction).
1422 * If the inode isn't held then cancel would iput it, but
1423 * that's wrong since this is inactive and the vnode ref
1424 * count is 0 already.
1425 * Cancel won't do anything to the inode if held, but it still
1426 * needs to be locked until the cancel is done, if it was
1427 * joined to the transaction.
1429 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1430 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1437 xfs_inactive_symlink_local(
1443 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1445 * We're freeing a symlink which fit into
1446 * the inode. Just free the memory used
1447 * to hold the old symlink.
1449 error = xfs_trans_reserve(*tpp, 0,
1450 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1451 0, XFS_TRANS_PERM_LOG_RES,
1452 XFS_ITRUNCATE_LOG_COUNT);
1455 xfs_trans_cancel(*tpp, 0);
1459 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1462 * Zero length symlinks _can_ exist.
1464 if (ip->i_df.if_bytes > 0) {
1465 xfs_idata_realloc(ip,
1466 -(ip->i_df.if_bytes),
1468 ASSERT(ip->i_df.if_bytes == 0);
1485 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1488 ASSERT(ip->i_d.di_forkoff != 0);
1489 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1490 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1492 error = xfs_attr_inactive(ip);
1495 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1496 return error; /* goto out */
1499 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1500 error = xfs_trans_reserve(tp, 0,
1501 XFS_IFREE_LOG_RES(mp),
1502 0, XFS_TRANS_PERM_LOG_RES,
1503 XFS_INACTIVE_LOG_COUNT);
1505 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1506 xfs_trans_cancel(tp, 0);
1508 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1512 xfs_ilock(ip, XFS_ILOCK_EXCL);
1513 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1514 xfs_trans_ihold(tp, ip);
1515 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1517 ASSERT(ip->i_d.di_anextents == 0);
1532 vp = BHV_TO_VNODE(bdp);
1533 ip = XFS_BHVTOI(bdp);
1535 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1539 /* If this is a read-only mount, don't do this (would generate I/O) */
1540 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1543 #ifdef HAVE_REFCACHE
1544 /* If we are in the NFS reference cache then don't do this now */
1551 if (ip->i_d.di_nlink != 0) {
1552 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1553 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1554 ip->i_delayed_blks > 0)) &&
1555 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1556 (!(ip->i_d.di_flags &
1557 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1558 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1560 /* Update linux inode block count after free above */
1561 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1562 ip->i_d.di_nblocks + ip->i_delayed_blks);
1572 * This is called when the vnode reference count for the vnode
1573 * goes to zero. If the file has been unlinked, then it must
1574 * now be truncated. Also, we clear all of the read-ahead state
1575 * kept for the inode here since the file is now closed.
1584 xfs_bmap_free_t free_list;
1585 xfs_fsblock_t first_block;
1592 vp = BHV_TO_VNODE(bdp);
1593 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1595 ip = XFS_BHVTOI(bdp);
1598 * If the inode is already free, then there can be nothing
1601 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1602 ASSERT(ip->i_df.if_real_bytes == 0);
1603 ASSERT(ip->i_df.if_broot_bytes == 0);
1604 return VN_INACTIVE_CACHE;
1608 * Only do a truncate if it's a regular file with
1609 * some actual space in it. It's OK to look at the
1610 * inode's fields without the lock because we're the
1611 * only one with a reference to the inode.
1613 truncate = ((ip->i_d.di_nlink == 0) &&
1614 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1615 (ip->i_delayed_blks > 0)) &&
1616 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1620 if (ip->i_d.di_nlink == 0 &&
1621 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1622 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1627 /* If this is a read-only mount, don't do this (would generate I/O) */
1628 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1631 if (ip->i_d.di_nlink != 0) {
1632 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1633 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1634 ip->i_delayed_blks > 0)) &&
1635 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1636 (!(ip->i_d.di_flags &
1637 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1638 (ip->i_delayed_blks != 0)))) {
1639 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1640 return VN_INACTIVE_CACHE;
1641 /* Update linux inode block count after free above */
1642 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1643 ip->i_d.di_nblocks + ip->i_delayed_blks);
1648 ASSERT(ip->i_d.di_nlink == 0);
1650 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1651 return VN_INACTIVE_CACHE;
1653 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1656 * Do the xfs_itruncate_start() call before
1657 * reserving any log space because itruncate_start
1658 * will call into the buffer cache and we can't
1659 * do that within a transaction.
1661 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1663 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1665 error = xfs_trans_reserve(tp, 0,
1666 XFS_ITRUNCATE_LOG_RES(mp),
1667 0, XFS_TRANS_PERM_LOG_RES,
1668 XFS_ITRUNCATE_LOG_COUNT);
1670 /* Don't call itruncate_cleanup */
1671 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1672 xfs_trans_cancel(tp, 0);
1673 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1674 return VN_INACTIVE_CACHE;
1677 xfs_ilock(ip, XFS_ILOCK_EXCL);
1678 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1679 xfs_trans_ihold(tp, ip);
1682 * normally, we have to run xfs_itruncate_finish sync.
1683 * But if filesystem is wsync and we're in the inactive
1684 * path, then we know that nlink == 0, and that the
1685 * xaction that made nlink == 0 is permanently committed
1686 * since xfs_remove runs as a synchronous transaction.
1688 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1689 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1692 xfs_trans_cancel(tp,
1693 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1694 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1695 return VN_INACTIVE_CACHE;
1697 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1700 * If we get an error while cleaning up a
1701 * symlink we bail out.
1703 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1704 xfs_inactive_symlink_rmt(ip, &tp) :
1705 xfs_inactive_symlink_local(ip, &tp);
1709 return VN_INACTIVE_CACHE;
1712 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1713 xfs_trans_ihold(tp, ip);
1715 error = xfs_trans_reserve(tp, 0,
1716 XFS_IFREE_LOG_RES(mp),
1717 0, XFS_TRANS_PERM_LOG_RES,
1718 XFS_INACTIVE_LOG_COUNT);
1720 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1721 xfs_trans_cancel(tp, 0);
1722 return VN_INACTIVE_CACHE;
1725 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1726 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1727 xfs_trans_ihold(tp, ip);
1731 * If there are attributes associated with the file
1732 * then blow them away now. The code calls a routine
1733 * that recursively deconstructs the attribute fork.
1734 * We need to just commit the current transaction
1735 * because we can't use it for xfs_attr_inactive().
1737 if (ip->i_d.di_anextents > 0) {
1738 error = xfs_inactive_attrs(ip, &tp);
1740 * If we got an error, the transaction is already
1741 * cancelled, and the inode is unlocked. Just get out.
1744 return VN_INACTIVE_CACHE;
1745 } else if (ip->i_afp) {
1746 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1752 XFS_BMAP_INIT(&free_list, &first_block);
1753 error = xfs_ifree(tp, ip, &free_list);
1756 * If we fail to free the inode, shut down. The cancel
1757 * might do that, we need to make sure. Otherwise the
1758 * inode might be lost for a long time or forever.
1760 if (!XFS_FORCED_SHUTDOWN(mp)) {
1762 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1763 error, mp->m_fsname);
1764 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1766 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1769 * Credit the quota account(s). The inode is gone.
1771 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1774 * Just ignore errors at this point. There is
1775 * nothing we can do except to try to keep going.
1777 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1779 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1782 * Release the dquots held by inode, if any.
1784 XFS_QM_DQDETACH(mp, ip);
1786 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1789 return VN_INACTIVE_CACHE;
1798 bhv_desc_t *dir_bdp,
1805 xfs_inode_t *dp, *ip;
1811 dir_vp = BHV_TO_VNODE(dir_bdp);
1812 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1814 dp = XFS_BHVTOI(dir_bdp);
1816 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1817 return XFS_ERROR(EIO);
1819 lock_mode = xfs_ilock_map_shared(dp);
1820 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1822 *vpp = XFS_ITOV(ip);
1825 xfs_iunlock_map_shared(dp, lock_mode);
1831 * xfs_create (create a new file).
1835 bhv_desc_t *dir_bdp,
1841 char *name = VNAME(dentry);
1843 xfs_inode_t *dp, *ip;
1849 xfs_bmap_free_t free_list;
1850 xfs_fsblock_t first_block;
1851 boolean_t dp_joined_to_trans;
1852 int dm_event_sent = 0;
1856 struct xfs_dquot *udqp, *gdqp;
1862 dir_vp = BHV_TO_VNODE(dir_bdp);
1863 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1865 dp = XFS_BHVTOI(dir_bdp);
1868 dm_di_mode = vap->va_mode;
1869 namelen = VNAMELEN(dentry);
1871 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1872 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1873 dir_vp, DM_RIGHT_NULL, NULL,
1874 DM_RIGHT_NULL, name, NULL,
1882 if (XFS_FORCED_SHUTDOWN(mp))
1883 return XFS_ERROR(EIO);
1885 /* Return through std_return after this point. */
1888 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1889 prid = dp->i_d.di_projid;
1890 else if (vap->va_mask & XFS_AT_PROJID)
1891 prid = (xfs_prid_t)vap->va_projid;
1893 prid = (xfs_prid_t)dfltprid;
1896 * Make sure that we have allocated dquot(s) on disk.
1898 error = XFS_QM_DQVOPALLOC(mp, dp,
1899 current_fsuid(credp), current_fsgid(credp), prid,
1900 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1905 dp_joined_to_trans = B_FALSE;
1907 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1908 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1909 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1911 * Initially assume that the file does not exist and
1912 * reserve the resources for that case. If that is not
1913 * the case we'll drop the one we have and get a more
1914 * appropriate transaction later.
1916 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1917 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1918 if (error == ENOSPC) {
1920 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1921 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1929 xfs_ilock(dp, XFS_ILOCK_EXCL);
1931 XFS_BMAP_INIT(&free_list, &first_block);
1936 * Reserve disk quota and the inode.
1938 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1943 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1945 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1946 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1947 rdev, credp, prid, resblks > 0,
1950 if (error == ENOSPC)
1957 * At this point, we've gotten a newly allocated inode.
1958 * It is locked (and joined to the transaction).
1961 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1964 * Now we join the directory inode to the transaction.
1965 * We do not do it earlier because xfs_dir_ialloc
1966 * might commit the previous transaction (and release
1971 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1972 dp_joined_to_trans = B_TRUE;
1974 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1975 &first_block, &free_list,
1976 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1978 ASSERT(error != ENOSPC);
1981 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1982 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1985 * If this is a synchronous mount, make sure that the
1986 * create transaction goes to disk before returning to
1989 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1990 xfs_trans_set_sync(tp);
1996 * Attach the dquot(s) to the inodes and modify them incore.
1997 * These ids of the inode couldn't have changed since the new
1998 * inode has been locked ever since it was created.
2000 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2003 * xfs_trans_commit normally decrements the vnode ref count
2004 * when it unlocks the inode. Since we want to return the
2005 * vnode to the caller, we bump the vnode ref count now.
2010 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2012 xfs_bmap_cancel(&free_list);
2016 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2023 XFS_QM_DQRELE(mp, udqp);
2024 XFS_QM_DQRELE(mp, gdqp);
2027 * Propagate the fact that the vnode changed after the
2028 * xfs_inode locks have been released.
2030 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2034 /* Fallthrough to std_return with error = 0 */
2037 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2038 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2039 DM_EVENT_POSTCREATE)) {
2040 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2041 dir_vp, DM_RIGHT_NULL,
2043 DM_RIGHT_NULL, name, NULL,
2044 dm_di_mode, error, 0);
2049 cancel_flags |= XFS_TRANS_ABORT;
2054 xfs_trans_cancel(tp, cancel_flags);
2056 if (!dp_joined_to_trans && (dp != NULL))
2057 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2058 XFS_QM_DQRELE(mp, udqp);
2059 XFS_QM_DQRELE(mp, gdqp);
2065 * Wait until after the current transaction is aborted to
2066 * release the inode. This prevents recursive transactions
2067 * and deadlocks from xfs_inactive.
2069 cancel_flags |= XFS_TRANS_ABORT;
2070 xfs_trans_cancel(tp, cancel_flags);
2073 XFS_QM_DQRELE(mp, udqp);
2074 XFS_QM_DQRELE(mp, gdqp);
2081 * Some counters to see if (and how often) we are hitting some deadlock
2082 * prevention code paths.
2086 int xfs_rm_lock_delays;
2087 int xfs_rm_attempts;
2091 * The following routine will lock the inodes associated with the
2092 * directory and the named entry in the directory. The locks are
2093 * acquired in increasing inode number.
2095 * If the entry is "..", then only the directory is locked. The
2096 * vnode ref count will still include that from the .. entry in
2099 * There is a deadlock we need to worry about. If the locked directory is
2100 * in the AIL, it might be blocking up the log. The next inode we lock
2101 * could be already locked by another thread waiting for log space (e.g
2102 * a permanent log reservation with a long running transaction (see
2103 * xfs_itruncate_finish)). To solve this, we must check if the directory
2104 * is in the ail and use lock_nowait. If we can't lock, we need to
2105 * drop the inode lock on the directory and try again. xfs_iunlock will
2106 * potentially push the tail if we were holding up the log.
2109 xfs_lock_dir_and_entry(
2112 xfs_inode_t *ip) /* inode of entry 'name' */
2116 xfs_inode_t *ips[2];
2125 xfs_ilock(dp, XFS_ILOCK_EXCL);
2132 * We want to lock in increasing inum. Since we've already
2133 * acquired the lock on the directory, we may need to release
2134 * if if the inum of the entry turns out to be less.
2136 if (e_inum > dp->i_ino) {
2138 * We are already in the right order, so just
2139 * lock on the inode of the entry.
2140 * We need to use nowait if dp is in the AIL.
2143 lp = (xfs_log_item_t *)dp->i_itemp;
2144 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2145 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2152 * Unlock dp and try again.
2153 * xfs_iunlock will try to push the tail
2154 * if the inode is in the AIL.
2157 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2159 if ((attempts % 5) == 0) {
2160 delay(1); /* Don't just spin the CPU */
2162 xfs_rm_lock_delays++;
2168 xfs_ilock(ip, XFS_ILOCK_EXCL);
2170 } else if (e_inum < dp->i_ino) {
2171 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2175 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2177 /* else e_inum == dp->i_ino */
2178 /* This can happen if we're asked to lock /x/..
2179 * the entry is "..", which is also the parent directory.
2187 int xfs_small_retries;
2188 int xfs_middle_retries;
2189 int xfs_lots_retries;
2190 int xfs_lock_delays;
2194 * The following routine will lock n inodes in exclusive mode.
2195 * We assume the caller calls us with the inodes in i_ino order.
2197 * We need to detect deadlock where an inode that we lock
2198 * is in the AIL and we start waiting for another inode that is locked
2199 * by a thread in a long running transaction (such as truncate). This can
2200 * result in deadlock since the long running trans might need to wait
2201 * for the inode we just locked in order to push the tail and free space
2211 int attempts = 0, i, j, try_lock;
2214 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2225 for (; i < inodes; i++) {
2228 if (i && (ips[i] == ips[i-1])) /* Already locked */
2232 * If try_lock is not set yet, make sure all locked inodes
2233 * are not in the AIL.
2234 * If any are, set try_lock to be used later.
2238 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2239 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2240 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2247 * If any of the previous locks we have locked is in the AIL,
2248 * we must TRY to get the second and subsequent locks. If
2249 * we can't get any, we must release all we have
2254 /* try_lock must be 0 if i is 0. */
2256 * try_lock means we have an inode locked
2257 * that is in the AIL.
2260 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2264 * Unlock all previous guys and try again.
2265 * xfs_iunlock will try to push the tail
2266 * if the inode is in the AIL.
2269 for(j = i - 1; j >= 0; j--) {
2272 * Check to see if we've already
2273 * unlocked this one.
2274 * Not the first one going back,
2275 * and the inode ptr is the same.
2277 if ((j != (i - 1)) && ips[j] ==
2281 xfs_iunlock(ips[j], lock_mode);
2284 if ((attempts % 5) == 0) {
2285 delay(1); /* Don't just spin the CPU */
2295 xfs_ilock(ips[i], lock_mode);
2301 if (attempts < 5) xfs_small_retries++;
2302 else if (attempts < 100) xfs_middle_retries++;
2303 else xfs_lots_retries++;
2311 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2312 int remove_which_error_return = 0;
2314 #define REMOVE_DEBUG_TRACE(x)
2315 #endif /* ! DEBUG */
2324 bhv_desc_t *dir_bdp,
2329 char *name = VNAME(dentry);
2330 xfs_inode_t *dp, *ip;
2331 xfs_trans_t *tp = NULL;
2334 xfs_bmap_free_t free_list;
2335 xfs_fsblock_t first_block;
2343 dir_vp = BHV_TO_VNODE(dir_bdp);
2344 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2346 dp = XFS_BHVTOI(dir_bdp);
2349 if (XFS_FORCED_SHUTDOWN(mp))
2350 return XFS_ERROR(EIO);
2352 namelen = VNAMELEN(dentry);
2354 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2355 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2356 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2357 name, NULL, 0, 0, 0);
2362 /* From this point on, return through std_return */
2366 * We need to get a reference to ip before we get our log
2367 * reservation. The reason for this is that we cannot call
2368 * xfs_iget for an inode for which we do not have a reference
2369 * once we've acquired a log reservation. This is because the
2370 * inode we are trying to get might be in xfs_inactive going
2371 * for a log reservation. Since we'll have to wait for the
2372 * inactive code to complete before returning from xfs_iget,
2373 * we need to make sure that we don't have log space reserved
2374 * when we call xfs_iget. Instead we get an unlocked reference
2375 * to the inode before getting our log reservation.
2377 error = xfs_get_dir_entry(dentry, &ip);
2379 REMOVE_DEBUG_TRACE(__LINE__);
2383 dm_di_mode = ip->i_d.di_mode;
2385 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2389 error = XFS_QM_DQATTACH(mp, dp, 0);
2390 if (!error && dp != ip)
2391 error = XFS_QM_DQATTACH(mp, ip, 0);
2393 REMOVE_DEBUG_TRACE(__LINE__);
2398 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2399 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2401 * We try to get the real space reservation first,
2402 * allowing for directory btree deletion(s) implying
2403 * possible bmap insert(s). If we can't get the space
2404 * reservation then we use 0 instead, and avoid the bmap
2405 * btree insert(s) in the directory code by, if the bmap
2406 * insert tries to happen, instead trimming the LAST
2407 * block from the directory.
2409 resblks = XFS_REMOVE_SPACE_RES(mp);
2410 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2411 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2412 if (error == ENOSPC) {
2414 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2415 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2418 ASSERT(error != ENOSPC);
2419 REMOVE_DEBUG_TRACE(__LINE__);
2420 xfs_trans_cancel(tp, 0);
2425 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2427 REMOVE_DEBUG_TRACE(__LINE__);
2428 xfs_trans_cancel(tp, cancel_flags);
2434 * At this point, we've gotten both the directory and the entry
2437 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2440 * Increment vnode ref count only in this case since
2441 * there's an extra vnode reference in the case where
2445 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2449 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2451 XFS_BMAP_INIT(&free_list, &first_block);
2452 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2453 &first_block, &free_list, 0);
2455 ASSERT(error != ENOENT);
2456 REMOVE_DEBUG_TRACE(__LINE__);
2459 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2462 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2464 error = xfs_droplink(tp, ip);
2466 REMOVE_DEBUG_TRACE(__LINE__);
2470 /* Determine if this is the last link while
2471 * we are in the transaction.
2473 link_zero = (ip)->i_d.di_nlink==0;
2476 * Take an extra ref on the inode so that it doesn't
2477 * go to xfs_inactive() from within the commit.
2482 * If this is a synchronous mount, make sure that the
2483 * remove transaction goes to disk before returning to
2486 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2487 xfs_trans_set_sync(tp);
2490 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2492 REMOVE_DEBUG_TRACE(__LINE__);
2496 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2503 * Before we drop our extra reference to the inode, purge it
2504 * from the refcache if it is there. By waiting until afterwards
2505 * to do the IRELE, we ensure that we won't go inactive in the
2506 * xfs_refcache_purge_ip routine (although that would be OK).
2508 xfs_refcache_purge_ip(ip);
2510 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2513 * Let interposed file systems know about removed links.
2515 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2519 /* Fall through to std_return with error = 0 */
2521 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2522 DM_EVENT_POSTREMOVE)) {
2523 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2524 dir_vp, DM_RIGHT_NULL,
2525 NULL, DM_RIGHT_NULL,
2526 name, NULL, dm_di_mode, error, 0);
2531 xfs_bmap_cancel(&free_list);
2532 cancel_flags |= XFS_TRANS_ABORT;
2533 xfs_trans_cancel(tp, cancel_flags);
2538 * In this case make sure to not release the inode until after
2539 * the current transaction is aborted. Releasing it beforehand
2540 * can cause us to go to xfs_inactive and start a recursive
2541 * transaction which can easily deadlock with the current one.
2543 xfs_bmap_cancel(&free_list);
2544 cancel_flags |= XFS_TRANS_ABORT;
2545 xfs_trans_cancel(tp, cancel_flags);
2548 * Before we drop our extra reference to the inode, purge it
2549 * from the refcache if it is there. By waiting until afterwards
2550 * to do the IRELE, we ensure that we won't go inactive in the
2551 * xfs_refcache_purge_ip routine (although that would be OK).
2553 xfs_refcache_purge_ip(ip);
2567 bhv_desc_t *target_dir_bdp,
2572 xfs_inode_t *tdp, *sip;
2575 xfs_inode_t *ips[2];
2577 xfs_bmap_free_t free_list;
2578 xfs_fsblock_t first_block;
2581 vnode_t *target_dir_vp;
2583 char *target_name = VNAME(dentry);
2586 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2587 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2588 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2590 target_namelen = VNAMELEN(dentry);
2591 if (VN_ISDIR(src_vp))
2592 return XFS_ERROR(EPERM);
2594 sip = xfs_vtoi(src_vp);
2595 tdp = XFS_BHVTOI(target_dir_bdp);
2597 if (XFS_FORCED_SHUTDOWN(mp))
2598 return XFS_ERROR(EIO);
2600 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2601 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2602 target_dir_vp, DM_RIGHT_NULL,
2603 src_vp, DM_RIGHT_NULL,
2604 target_name, NULL, 0, 0, 0);
2609 /* Return through std_return after this point. */
2611 error = XFS_QM_DQATTACH(mp, sip, 0);
2612 if (!error && sip != tdp)
2613 error = XFS_QM_DQATTACH(mp, tdp, 0);
2617 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2618 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2619 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2620 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2621 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2622 if (error == ENOSPC) {
2624 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2625 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2632 if (sip->i_ino < tdp->i_ino) {
2640 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2643 * Increment vnode ref counts since xfs_trans_commit &
2644 * xfs_trans_cancel will both unlock the inodes and
2645 * decrement the associated ref counts.
2648 VN_HOLD(target_dir_vp);
2649 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2650 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2653 * If the source has too many links, we can't make any more to it.
2655 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2656 error = XFS_ERROR(EMLINK);
2661 * If we are using project inheritance, we only allow hard link
2662 * creation in our tree when the project IDs are the same; else
2663 * the tree quota mechanism could be circumvented.
2665 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2666 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2667 error = XFS_ERROR(EXDEV);
2672 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2676 XFS_BMAP_INIT(&free_list, &first_block);
2678 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2679 sip->i_ino, &first_block, &free_list,
2683 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2685 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2687 error = xfs_bumplink(tp, sip);
2693 * If this is a synchronous mount, make sure that the
2694 * link transaction goes to disk before returning to
2697 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2698 xfs_trans_set_sync(tp);
2701 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2703 xfs_bmap_cancel(&free_list);
2707 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2712 /* Fall through to std_return with error = 0. */
2714 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2715 DM_EVENT_POSTLINK)) {
2716 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2717 target_dir_vp, DM_RIGHT_NULL,
2718 src_vp, DM_RIGHT_NULL,
2719 target_name, NULL, 0, error, 0);
2724 cancel_flags |= XFS_TRANS_ABORT;
2728 xfs_trans_cancel(tp, cancel_flags);
2737 bhv_desc_t *dir_bdp,
2743 char *dir_name = VNAME(dentry);
2745 xfs_inode_t *cdp; /* inode of created dir */
2746 vnode_t *cvp; /* vnode of created dir */
2752 xfs_bmap_free_t free_list;
2753 xfs_fsblock_t first_block;
2755 boolean_t dp_joined_to_trans;
2756 boolean_t created = B_FALSE;
2757 int dm_event_sent = 0;
2759 struct xfs_dquot *udqp, *gdqp;
2764 dir_vp = BHV_TO_VNODE(dir_bdp);
2765 dp = XFS_BHVTOI(dir_bdp);
2768 if (XFS_FORCED_SHUTDOWN(mp))
2769 return XFS_ERROR(EIO);
2771 dir_namelen = VNAMELEN(dentry);
2774 dp_joined_to_trans = B_FALSE;
2775 dm_di_mode = vap->va_mode;
2777 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2778 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2779 dir_vp, DM_RIGHT_NULL, NULL,
2780 DM_RIGHT_NULL, dir_name, NULL,
2787 /* Return through std_return after this point. */
2789 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2793 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2794 prid = dp->i_d.di_projid;
2795 else if (vap->va_mask & XFS_AT_PROJID)
2796 prid = (xfs_prid_t)vap->va_projid;
2798 prid = (xfs_prid_t)dfltprid;
2801 * Make sure that we have allocated dquot(s) on disk.
2803 error = XFS_QM_DQVOPALLOC(mp, dp,
2804 current_fsuid(credp), current_fsgid(credp), prid,
2805 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2809 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2810 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2811 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2812 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2813 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2814 if (error == ENOSPC) {
2816 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2817 XFS_TRANS_PERM_LOG_RES,
2818 XFS_MKDIR_LOG_COUNT);
2826 xfs_ilock(dp, XFS_ILOCK_EXCL);
2829 * Check for directory link count overflow.
2831 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2832 error = XFS_ERROR(EMLINK);
2837 * Reserve disk quota and the inode.
2839 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2844 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2847 * create the directory inode.
2849 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2850 0, credp, prid, resblks > 0,
2853 if (error == ENOSPC)
2860 * Now we add the directory inode to the transaction.
2861 * We waited until now since xfs_dir_ialloc might start
2862 * a new transaction. Had we joined the transaction
2863 * earlier, the locks might have gotten released.
2866 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2867 dp_joined_to_trans = B_TRUE;
2869 XFS_BMAP_INIT(&free_list, &first_block);
2871 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2872 cdp->i_ino, &first_block, &free_list,
2873 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2875 ASSERT(error != ENOSPC);
2878 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2881 * Bump the in memory version number of the parent directory
2882 * so that other processes accessing it will recognize that
2883 * the directory has changed.
2887 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2893 error = xfs_bumplink(tp, dp);
2898 cvp = XFS_ITOV(cdp);
2906 * Attach the dquots to the new inode and modify the icount incore.
2908 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2911 * If this is a synchronous mount, make sure that the
2912 * mkdir transaction goes to disk before returning to
2915 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2916 xfs_trans_set_sync(tp);
2919 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2925 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2926 XFS_QM_DQRELE(mp, udqp);
2927 XFS_QM_DQRELE(mp, gdqp);
2932 /* Fall through to std_return with error = 0 or errno from
2933 * xfs_trans_commit. */
2936 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2937 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2938 DM_EVENT_POSTCREATE)) {
2939 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2940 dir_vp, DM_RIGHT_NULL,
2941 created ? XFS_ITOV(cdp):NULL,
2944 dm_di_mode, error, 0);
2950 xfs_bmap_cancel(&free_list);
2952 cancel_flags |= XFS_TRANS_ABORT;
2954 xfs_trans_cancel(tp, cancel_flags);
2955 XFS_QM_DQRELE(mp, udqp);
2956 XFS_QM_DQRELE(mp, gdqp);
2958 if (!dp_joined_to_trans && (dp != NULL)) {
2959 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2972 bhv_desc_t *dir_bdp,
2976 char *name = VNAME(dentry);
2978 xfs_inode_t *cdp; /* child directory */
2982 xfs_bmap_free_t free_list;
2983 xfs_fsblock_t first_block;
2992 dir_vp = BHV_TO_VNODE(dir_bdp);
2993 dp = XFS_BHVTOI(dir_bdp);
2996 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2998 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
2999 return XFS_ERROR(EIO);
3000 namelen = VNAMELEN(dentry);
3002 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3003 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3004 dir_vp, DM_RIGHT_NULL,
3005 NULL, DM_RIGHT_NULL,
3006 name, NULL, 0, 0, 0);
3008 return XFS_ERROR(error);
3011 /* Return through std_return after this point. */
3016 * We need to get a reference to cdp before we get our log
3017 * reservation. The reason for this is that we cannot call
3018 * xfs_iget for an inode for which we do not have a reference
3019 * once we've acquired a log reservation. This is because the
3020 * inode we are trying to get might be in xfs_inactive going
3021 * for a log reservation. Since we'll have to wait for the
3022 * inactive code to complete before returning from xfs_iget,
3023 * we need to make sure that we don't have log space reserved
3024 * when we call xfs_iget. Instead we get an unlocked reference
3025 * to the inode before getting our log reservation.
3027 error = xfs_get_dir_entry(dentry, &cdp);
3029 REMOVE_DEBUG_TRACE(__LINE__);
3033 dm_di_mode = cdp->i_d.di_mode;
3036 * Get the dquots for the inodes.
3038 error = XFS_QM_DQATTACH(mp, dp, 0);
3039 if (!error && dp != cdp)
3040 error = XFS_QM_DQATTACH(mp, cdp, 0);
3043 REMOVE_DEBUG_TRACE(__LINE__);
3047 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3048 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3050 * We try to get the real space reservation first,
3051 * allowing for directory btree deletion(s) implying
3052 * possible bmap insert(s). If we can't get the space
3053 * reservation then we use 0 instead, and avoid the bmap
3054 * btree insert(s) in the directory code by, if the bmap
3055 * insert tries to happen, instead trimming the LAST
3056 * block from the directory.
3058 resblks = XFS_REMOVE_SPACE_RES(mp);
3059 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3060 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3061 if (error == ENOSPC) {
3063 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3064 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3067 ASSERT(error != ENOSPC);
3072 XFS_BMAP_INIT(&free_list, &first_block);
3075 * Now lock the child directory inode and the parent directory
3076 * inode in the proper order. This will take care of validating
3077 * that the directory entry for the child directory inode has
3078 * not changed while we were obtaining a log reservation.
3080 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3082 xfs_trans_cancel(tp, cancel_flags);
3087 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3090 * Only increment the parent directory vnode count if
3091 * we didn't bump it in looking up cdp. The only time
3092 * we don't bump it is when we're looking up ".".
3098 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3100 ASSERT(cdp->i_d.di_nlink >= 2);
3101 if (cdp->i_d.di_nlink != 2) {
3102 error = XFS_ERROR(ENOTEMPTY);
3105 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3106 error = XFS_ERROR(ENOTEMPTY);
3110 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3111 &first_block, &free_list, resblks);
3116 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3119 * Bump the in memory generation count on the parent
3120 * directory so that other can know that it has changed.
3125 * Drop the link from cdp's "..".
3127 error = xfs_droplink(tp, dp);
3133 * Drop the link from dp to cdp.
3135 error = xfs_droplink(tp, cdp);
3141 * Drop the "." link from cdp to self.
3143 error = xfs_droplink(tp, cdp);
3148 /* Determine these before committing transaction */
3149 last_cdp_link = (cdp)->i_d.di_nlink==0;
3152 * Take an extra ref on the child vnode so that it
3153 * does not go to xfs_inactive() from within the commit.
3158 * If this is a synchronous mount, make sure that the
3159 * rmdir transaction goes to disk before returning to
3162 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3163 xfs_trans_set_sync(tp);
3166 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3168 xfs_bmap_cancel(&free_list);
3169 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3175 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3183 * Let interposed file systems know about removed links.
3185 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3189 /* Fall through to std_return with error = 0 or the errno
3190 * from xfs_trans_commit. */
3192 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3193 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3194 dir_vp, DM_RIGHT_NULL,
3195 NULL, DM_RIGHT_NULL,
3196 name, NULL, dm_di_mode,
3202 xfs_bmap_cancel(&free_list);
3203 cancel_flags |= XFS_TRANS_ABORT;
3207 xfs_trans_cancel(tp, cancel_flags);
3215 * Read dp's entries starting at uiop->uio_offset and translate them into
3216 * bufsize bytes worth of struct dirents starting at bufbase.
3220 bhv_desc_t *dir_bdp,
3226 xfs_trans_t *tp = NULL;
3230 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3231 (inst_t *)__return_address);
3232 dp = XFS_BHVTOI(dir_bdp);
3234 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3235 return XFS_ERROR(EIO);
3238 lock_mode = xfs_ilock_map_shared(dp);
3239 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3240 xfs_iunlock_map_shared(dp, lock_mode);
3251 bhv_desc_t *dir_bdp,
3264 xfs_bmap_free_t free_list;
3265 xfs_fsblock_t first_block;
3266 boolean_t dp_joined_to_trans;
3270 xfs_fileoff_t first_fsb;
3271 xfs_filblks_t fs_blocks;
3273 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3280 struct xfs_dquot *udqp, *gdqp;
3282 char *link_name = VNAME(dentry);
3286 dir_vp = BHV_TO_VNODE(dir_bdp);
3287 dp = XFS_BHVTOI(dir_bdp);
3288 dp_joined_to_trans = B_FALSE;
3293 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3297 if (XFS_FORCED_SHUTDOWN(mp))
3298 return XFS_ERROR(EIO);
3300 link_namelen = VNAMELEN(dentry);
3303 * Check component lengths of the target path name.
3305 pathlen = strlen(target_path);
3306 if (pathlen >= MAXPATHLEN) /* total string too long */
3307 return XFS_ERROR(ENAMETOOLONG);
3308 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3312 for(total = 0, path = target_path; total < pathlen;) {
3316 while(*path == '/') {
3322 * Count up to the next slash or end of path.
3323 * Error out if the component is bigger than MAXNAMELEN.
3325 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3326 if (++len >= MAXNAMELEN) {
3327 error = ENAMETOOLONG;
3334 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3335 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3336 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3337 link_name, target_path, 0, 0, 0);
3342 /* Return through std_return after this point. */
3345 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3346 prid = dp->i_d.di_projid;
3347 else if (vap->va_mask & XFS_AT_PROJID)
3348 prid = (xfs_prid_t)vap->va_projid;
3350 prid = (xfs_prid_t)dfltprid;
3353 * Make sure that we have allocated dquot(s) on disk.
3355 error = XFS_QM_DQVOPALLOC(mp, dp,
3356 current_fsuid(credp), current_fsgid(credp), prid,
3357 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3361 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3362 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3364 * The symlink will fit into the inode data fork?
3365 * There can't be any attributes so we get the whole variable part.
3367 if (pathlen <= XFS_LITINO(mp))
3370 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3371 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3372 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3373 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3374 if (error == ENOSPC && fs_blocks == 0) {
3376 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3377 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3385 xfs_ilock(dp, XFS_ILOCK_EXCL);
3388 * Check whether the directory allows new symlinks or not.
3390 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3391 error = XFS_ERROR(EPERM);
3396 * Reserve disk quota : blocks and inode.
3398 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3403 * Check for ability to enter directory entry, if no space reserved.
3406 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3409 * Initialize the bmap freelist prior to calling either
3410 * bmapi or the directory create code.
3412 XFS_BMAP_INIT(&free_list, &first_block);
3415 * Allocate an inode for the symlink.
3417 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3418 1, 0, credp, prid, resblks > 0, &ip, NULL);
3420 if (error == ENOSPC)
3427 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3428 dp_joined_to_trans = B_TRUE;
3431 * Also attach the dquot(s) to it, if applicable.
3433 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3436 resblks -= XFS_IALLOC_SPACE_RES(mp);
3438 * If the symlink will fit into the inode, write it inline.
3440 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3441 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3442 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3443 ip->i_d.di_size = pathlen;
3446 * The inode was initially created in extent format.
3448 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3449 ip->i_df.if_flags |= XFS_IFINLINE;
3451 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3452 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3456 nmaps = SYMLINK_MAPS;
3458 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3459 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3460 &first_block, resblks, mval, &nmaps,
3467 resblks -= fs_blocks;
3468 ip->i_d.di_size = pathlen;
3469 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3471 cur_chunk = target_path;
3472 for (n = 0; n < nmaps; n++) {
3473 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3474 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3475 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3476 BTOBB(byte_cnt), 0);
3477 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3478 if (pathlen < byte_cnt) {
3481 pathlen -= byte_cnt;
3483 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3484 cur_chunk += byte_cnt;
3486 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3491 * Create the directory entry for the symlink.
3493 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3494 ip->i_ino, &first_block, &free_list, resblks);
3498 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3499 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3502 * Bump the in memory version number of the parent directory
3503 * so that other processes accessing it will recognize that
3504 * the directory has changed.
3509 * If this is a synchronous mount, make sure that the
3510 * symlink transaction goes to disk before returning to
3513 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3514 xfs_trans_set_sync(tp);
3518 * xfs_trans_commit normally decrements the vnode ref count
3519 * when it unlocks the inode. Since we want to return the
3520 * vnode to the caller, we bump the vnode ref count now.
3524 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3528 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3529 XFS_QM_DQRELE(mp, udqp);
3530 XFS_QM_DQRELE(mp, gdqp);
3532 /* Fall through to std_return with error = 0 or errno from
3533 * xfs_trans_commit */
3535 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3536 DM_EVENT_POSTSYMLINK)) {
3537 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3538 dir_vp, DM_RIGHT_NULL,
3539 error ? NULL : XFS_ITOV(ip),
3540 DM_RIGHT_NULL, link_name, target_path,
3556 xfs_bmap_cancel(&free_list);
3557 cancel_flags |= XFS_TRANS_ABORT;
3559 xfs_trans_cancel(tp, cancel_flags);
3560 XFS_QM_DQRELE(mp, udqp);
3561 XFS_QM_DQRELE(mp, gdqp);
3563 if (!dp_joined_to_trans && (dp != NULL)) {
3564 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3574 * A fid routine that takes a pointer to a previously allocated
3575 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3585 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3586 (inst_t *)__return_address);
3587 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3589 xfid = (xfs_fid2_t *)fidp;
3590 ip = XFS_BHVTOI(bdp);
3591 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3594 * use memcpy because the inode is a long long and there's no
3595 * assurance that xfid->fid_ino is properly aligned.
3597 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3598 xfid->fid_gen = ip->i_d.di_gen;
3615 vp = BHV_TO_VNODE(bdp);
3618 ip = XFS_BHVTOI(bdp);
3619 if (locktype == VRWLOCK_WRITE) {
3620 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3621 } else if (locktype == VRWLOCK_TRY_READ) {
3622 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3623 } else if (locktype == VRWLOCK_TRY_WRITE) {
3624 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3626 ASSERT((locktype == VRWLOCK_READ) ||
3627 (locktype == VRWLOCK_WRITE_DIRECT));
3628 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3646 vp = BHV_TO_VNODE(bdp);
3649 ip = XFS_BHVTOI(bdp);
3650 if (locktype == VRWLOCK_WRITE) {
3652 * In the write case, we may have added a new entry to
3653 * the reference cache. This might store a pointer to
3654 * an inode to be released in this inode. If it is there,
3655 * clear the pointer and release the inode after unlocking
3658 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3660 ASSERT((locktype == VRWLOCK_READ) ||
3661 (locktype == VRWLOCK_WRITE_DIRECT));
3662 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3674 xfs_inode_log_item_t *iip;
3677 ip = XFS_BHVTOI(bdp);
3681 if (XFS_FORCED_SHUTDOWN(mp))
3682 return XFS_ERROR(EIO);
3685 * Bypass inodes which have already been cleaned by
3686 * the inode flush clustering code inside xfs_iflush
3688 if ((ip->i_update_core == 0) &&
3689 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3692 if (flags & FLUSH_LOG) {
3693 if (iip && iip->ili_last_lsn) {
3694 xlog_t *log = mp->m_log;
3696 int s, log_flags = XFS_LOG_FORCE;
3698 s = GRANT_LOCK(log);
3699 sync_lsn = log->l_last_sync_lsn;
3700 GRANT_UNLOCK(log, s);
3702 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3705 if (flags & FLUSH_SYNC)
3706 log_flags |= XFS_LOG_SYNC;
3707 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3712 * We make this non-blocking if the inode is contended,
3713 * return EAGAIN to indicate to the caller that they
3714 * did not succeed. This prevents the flush path from
3715 * blocking on inodes inside another operation right
3716 * now, they get caught later by xfs_sync.
3718 if (flags & FLUSH_INODE) {
3721 if (xfs_ipincount(ip))
3724 if (flags & FLUSH_SYNC) {
3725 xfs_ilock(ip, XFS_ILOCK_SHARED);
3727 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3728 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3729 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3736 if (flags & FLUSH_SYNC)
3737 flush_flags = XFS_IFLUSH_SYNC;
3739 flush_flags = XFS_IFLUSH_ASYNC;
3741 error = xfs_iflush(ip, flush_flags);
3742 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3761 if (!capable(CAP_SYS_ADMIN))
3762 return XFS_ERROR(EPERM);
3764 ip = XFS_BHVTOI(bdp);
3767 if (XFS_FORCED_SHUTDOWN(mp))
3768 return XFS_ERROR(EIO);
3770 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3771 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3773 xfs_trans_cancel(tp, 0);
3776 xfs_ilock(ip, XFS_ILOCK_EXCL);
3777 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3779 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3780 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3782 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3784 error = xfs_trans_commit(tp, 0, NULL);
3800 vp = BHV_TO_VNODE(bdp);
3801 ip = XFS_BHVTOI(bdp);
3803 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3805 ASSERT(!VN_MAPPED(vp));
3807 /* bad inode, get out here ASAP */
3815 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3818 * Make sure the atime in the XFS inode is correct before freeing the
3821 xfs_synchronize_atime(ip);
3823 /* If we have nothing to flush with this inode then complete the
3824 * teardown now, otherwise break the link between the xfs inode
3825 * and the linux inode and clean up the xfs inode later. This
3826 * avoids flushing the inode to disk during the delete operation
3829 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3830 xfs_ilock(ip, XFS_ILOCK_EXCL);
3832 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3834 xfs_mount_t *mp = ip->i_mount;
3836 /* Protect sync from us */
3837 XFS_MOUNT_ILOCK(mp);
3838 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3839 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3840 ip->i_flags |= XFS_IRECLAIMABLE;
3841 XFS_MOUNT_IUNLOCK(mp);
3852 xfs_ihash_t *ih = ip->i_hash;
3853 vnode_t *vp = XFS_ITOV_NULL(ip);
3856 if (vp && VN_BAD(vp))
3859 /* The hash lock here protects a thread in xfs_iget_core from
3860 * racing with us on linking the inode back with a vnode.
3861 * Once we have the XFS_IRECLAIM flag set it will not touch
3864 write_lock(&ih->ih_lock);
3865 if ((ip->i_flags & XFS_IRECLAIM) ||
3866 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3867 write_unlock(&ih->ih_lock);
3870 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3874 ip->i_flags |= XFS_IRECLAIM;
3875 write_unlock(&ih->ih_lock);
3878 * If the inode is still dirty, then flush it out. If the inode
3879 * is not in the AIL, then it will be OK to flush it delwri as
3880 * long as xfs_iflush() does not keep any references to the inode.
3881 * We leave that decision up to xfs_iflush() since it has the
3882 * knowledge of whether it's OK to simply do a delwri flush of
3883 * the inode or whether we need to wait until the inode is
3884 * pulled from the AIL.
3885 * We get the flush lock regardless, though, just to make sure
3886 * we don't free it while it is being flushed.
3888 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3890 xfs_ilock(ip, XFS_ILOCK_EXCL);
3894 if (ip->i_update_core ||
3895 ((ip->i_itemp != NULL) &&
3896 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3897 error = xfs_iflush(ip, sync_mode);
3899 * If we hit an error, typically because of filesystem
3900 * shutdown, we don't need to let vn_reclaim to know
3901 * because we're gonna reclaim the inode anyway.
3904 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3907 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3910 ASSERT(ip->i_update_core == 0);
3911 ASSERT(ip->i_itemp == NULL ||
3912 ip->i_itemp->ili_format.ilf_fields == 0);
3913 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3914 } else if (locked) {
3916 * We are not interested in doing an iflush if we're
3917 * in the process of shutting down the filesystem forcibly.
3918 * So, just reclaim the inode.
3921 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3930 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3933 xfs_inode_t *ip, *n;
3938 XFS_MOUNT_ILOCK(mp);
3939 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3941 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3943 if (xfs_ipincount(ip) ||
3944 !xfs_iflock_nowait(ip)) {
3945 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3949 XFS_MOUNT_IUNLOCK(mp);
3950 if (xfs_finish_reclaim(ip, noblock,
3951 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3960 XFS_MOUNT_IUNLOCK(mp);
3965 * xfs_alloc_file_space()
3966 * This routine allocates disk space for the given file.
3968 * If alloc_type == 0, this request is for an ALLOCSP type
3969 * request which will change the file size. In this case, no
3970 * DMAPI event will be generated by the call. A TRUNCATE event
3971 * will be generated later by xfs_setattr.
3973 * If alloc_type != 0, this request is for a RESVSP type
3974 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3975 * lower block boundary byte address is less than the file's
3984 xfs_alloc_file_space(
3991 xfs_mount_t *mp = ip->i_mount;
3993 xfs_filblks_t allocated_fsb;
3994 xfs_filblks_t allocatesize_fsb;
3995 xfs_extlen_t extsz, temp;
3996 xfs_fileoff_t startoffset_fsb;
3997 xfs_fsblock_t firstfsb;
4003 xfs_bmbt_irec_t imaps[1], *imapp;
4004 xfs_bmap_free_t free_list;
4005 uint qblocks, resblks, resrtextents;
4009 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4011 if (XFS_FORCED_SHUTDOWN(mp))
4012 return XFS_ERROR(EIO);
4014 rt = XFS_IS_REALTIME_INODE(ip);
4016 if (!(extsz = ip->i_d.di_extsize))
4017 extsz = mp->m_sb.sb_rextsize;
4019 extsz = ip->i_d.di_extsize;
4022 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4026 return XFS_ERROR(EINVAL);
4032 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4033 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4034 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4036 /* Generate a DMAPI event if needed. */
4037 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4038 (attr_flags&ATTR_DMI) == 0 &&
4039 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4040 xfs_off_t end_dmi_offset;
4042 end_dmi_offset = offset+len;
4043 if (end_dmi_offset > ip->i_d.di_size)
4044 end_dmi_offset = ip->i_d.di_size;
4045 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4046 offset, end_dmi_offset - offset,
4053 * Allocate file space until done or until there is an error
4056 while (allocatesize_fsb && !error) {
4060 * Determine space reservations for data/realtime.
4062 if (unlikely(extsz)) {
4063 s = startoffset_fsb;
4066 e = startoffset_fsb + allocatesize_fsb;
4067 if ((temp = do_mod(startoffset_fsb, extsz)))
4069 if ((temp = do_mod(e, extsz)))
4073 e = allocatesize_fsb;
4077 resrtextents = qblocks = (uint)(e - s);
4078 resrtextents /= mp->m_sb.sb_rextsize;
4079 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4080 quota_flag = XFS_QMOPT_RES_RTBLKS;
4083 resblks = qblocks = \
4084 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4085 quota_flag = XFS_QMOPT_RES_REGBLKS;
4089 * Allocate and setup the transaction.
4091 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4092 error = xfs_trans_reserve(tp, resblks,
4093 XFS_WRITE_LOG_RES(mp), resrtextents,
4094 XFS_TRANS_PERM_LOG_RES,
4095 XFS_WRITE_LOG_COUNT);
4097 * Check for running out of space
4101 * Free the transaction structure.
4103 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4104 xfs_trans_cancel(tp, 0);
4107 xfs_ilock(ip, XFS_ILOCK_EXCL);
4108 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4109 qblocks, 0, quota_flag);
4113 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4114 xfs_trans_ihold(tp, ip);
4117 * Issue the xfs_bmapi() call to allocate the blocks
4119 XFS_BMAP_INIT(&free_list, &firstfsb);
4120 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4121 allocatesize_fsb, bmapi_flag,
4122 &firstfsb, 0, imapp, &nimaps,
4129 * Complete the transaction
4131 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4136 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4137 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4142 allocated_fsb = imapp->br_blockcount;
4145 error = XFS_ERROR(ENOSPC);
4149 startoffset_fsb += allocated_fsb;
4150 allocatesize_fsb -= allocated_fsb;
4153 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4154 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4156 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4157 XFS_ITOV(ip), DM_RIGHT_NULL,
4158 XFS_ITOV(ip), DM_RIGHT_NULL,
4159 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4161 goto retry; /* Maybe DMAPI app. has made space */
4162 /* else fall through with error from XFS_SEND_DATA */
4167 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4168 xfs_bmap_cancel(&free_list);
4169 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4171 error1: /* Just cancel transaction */
4172 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4173 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4174 goto dmapi_enospc_check;
4178 * Zero file bytes between startoff and endoff inclusive.
4179 * The iolock is held exclusive and no blocks are buffered.
4182 xfs_zero_remaining_bytes(
4187 xfs_bmbt_irec_t imap;
4188 xfs_fileoff_t offset_fsb;
4189 xfs_off_t lastoffset;
4192 xfs_mount_t *mp = ip->i_mount;
4196 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4197 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4198 mp->m_rtdev_targp : mp->m_ddev_targp);
4200 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4201 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4203 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4204 NULL, 0, &imap, &nimap, NULL, NULL);
4205 if (error || nimap < 1)
4207 ASSERT(imap.br_blockcount >= 1);
4208 ASSERT(imap.br_startoff == offset_fsb);
4209 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4210 if (lastoffset > endoff)
4211 lastoffset = endoff;
4212 if (imap.br_startblock == HOLESTARTBLOCK)
4214 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4215 if (imap.br_state == XFS_EXT_UNWRITTEN)
4218 XFS_BUF_UNWRITE(bp);
4220 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4222 if ((error = xfs_iowait(bp))) {
4223 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4224 mp, bp, XFS_BUF_ADDR(bp));
4227 memset(XFS_BUF_PTR(bp) +
4228 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4229 0, lastoffset - offset + 1);
4234 if ((error = xfs_iowait(bp))) {
4235 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4236 mp, bp, XFS_BUF_ADDR(bp));
4245 * xfs_free_file_space()
4246 * This routine frees disk space for the given file.
4248 * This routine is only called by xfs_change_file_space
4249 * for an UNRESVSP type call.
4257 xfs_free_file_space(
4266 xfs_off_t end_dmi_offset;
4267 xfs_fileoff_t endoffset_fsb;
4269 xfs_fsblock_t firstfsb;
4270 xfs_bmap_free_t free_list;
4272 xfs_bmbt_irec_t imap;
4280 xfs_fileoff_t startoffset_fsb;
4282 int need_iolock = 1;
4287 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4289 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4293 if (len <= 0) /* if nothing being freed */
4295 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4296 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4297 end_dmi_offset = offset + len;
4298 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4300 if (offset < ip->i_d.di_size &&
4301 (attr_flags & ATTR_DMI) == 0 &&
4302 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4303 if (end_dmi_offset > ip->i_d.di_size)
4304 end_dmi_offset = ip->i_d.di_size;
4305 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4306 offset, end_dmi_offset - offset,
4307 AT_DELAY_FLAG(attr_flags), NULL);
4312 if (attr_flags & ATTR_NOLOCK)
4315 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4316 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4319 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4321 ilen = len + (offset & (rounding - 1));
4322 ioffset = offset & ~(rounding - 1);
4323 if (ilen & (rounding - 1))
4324 ilen = (ilen + rounding) & ~(rounding - 1);
4326 if (VN_CACHED(vp) != 0) {
4327 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4328 ctooff(offtoct(ioffset)), -1);
4329 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4330 -1, FI_REMAPF_LOCKED);
4334 * Need to zero the stuff we're not freeing, on disk.
4335 * If its a realtime file & can't use unwritten extents then we
4336 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4337 * will take care of it for us.
4339 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4341 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4342 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4344 goto out_unlock_iolock;
4345 ASSERT(nimap == 0 || nimap == 1);
4346 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4349 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4350 block = imap.br_startblock;
4351 mod = do_div(block, mp->m_sb.sb_rextsize);
4353 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4356 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4357 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4359 goto out_unlock_iolock;
4360 ASSERT(nimap == 0 || nimap == 1);
4361 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4362 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4364 if (mod && (mod != mp->m_sb.sb_rextsize))
4365 endoffset_fsb -= mod;
4368 if ((done = (endoffset_fsb <= startoffset_fsb)))
4370 * One contiguous piece to clear
4372 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4375 * Some full blocks, possibly two pieces to clear
4377 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4378 error = xfs_zero_remaining_bytes(ip, offset,
4379 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4381 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4382 error = xfs_zero_remaining_bytes(ip,
4383 XFS_FSB_TO_B(mp, endoffset_fsb),
4388 * free file space until done or until there is an error
4390 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4391 while (!error && !done) {
4394 * allocate and setup the transaction
4396 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4397 error = xfs_trans_reserve(tp,
4399 XFS_WRITE_LOG_RES(mp),
4401 XFS_TRANS_PERM_LOG_RES,
4402 XFS_WRITE_LOG_COUNT);
4405 * check for running out of space
4409 * Free the transaction structure.
4411 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4412 xfs_trans_cancel(tp, 0);
4415 xfs_ilock(ip, XFS_ILOCK_EXCL);
4416 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4417 ip->i_udquot, ip->i_gdquot, resblks, 0,
4418 XFS_QMOPT_RES_REGBLKS);
4422 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4423 xfs_trans_ihold(tp, ip);
4426 * issue the bunmapi() call to free the blocks
4428 XFS_BMAP_INIT(&free_list, &firstfsb);
4429 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4430 endoffset_fsb - startoffset_fsb,
4431 0, 2, &firstfsb, &free_list, NULL, &done);
4437 * complete the transaction
4439 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4444 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4445 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4450 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4454 xfs_bmap_cancel(&free_list);
4456 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4457 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4463 * xfs_change_file_space()
4464 * This routine allocates or frees disk space for the given file.
4465 * The user specified parameters are checked for alignment and size
4474 xfs_change_file_space(
4488 xfs_off_t startoffset;
4494 vp = BHV_TO_VNODE(bdp);
4495 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4497 ip = XFS_BHVTOI(bdp);
4501 * must be a regular file and have write permission
4504 return XFS_ERROR(EINVAL);
4506 xfs_ilock(ip, XFS_ILOCK_SHARED);
4508 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4509 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4513 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4515 switch (bf->l_whence) {
4516 case 0: /*SEEK_SET*/
4518 case 1: /*SEEK_CUR*/
4519 bf->l_start += offset;
4521 case 2: /*SEEK_END*/
4522 bf->l_start += ip->i_d.di_size;
4525 return XFS_ERROR(EINVAL);
4528 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4530 if ( (bf->l_start < 0)
4531 || (bf->l_start > XFS_MAXIOFFSET(mp))
4532 || (bf->l_start + llen < 0)
4533 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4534 return XFS_ERROR(EINVAL);
4538 startoffset = bf->l_start;
4539 fsize = ip->i_d.di_size;
4542 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4544 * These calls do NOT zero the data space allocated to the file,
4545 * nor do they change the file size.
4547 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4549 * These calls cause the new file data to be zeroed and the file
4550 * size to be changed.
4552 setprealloc = clrprealloc = 0;
4555 case XFS_IOC_RESVSP:
4556 case XFS_IOC_RESVSP64:
4557 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4564 case XFS_IOC_UNRESVSP:
4565 case XFS_IOC_UNRESVSP64:
4566 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4571 case XFS_IOC_ALLOCSP:
4572 case XFS_IOC_ALLOCSP64:
4573 case XFS_IOC_FREESP:
4574 case XFS_IOC_FREESP64:
4575 if (startoffset > fsize) {
4576 error = xfs_alloc_file_space(ip, fsize,
4577 startoffset - fsize, 0, attr_flags);
4582 va.va_mask = XFS_AT_SIZE;
4583 va.va_size = startoffset;
4585 error = xfs_setattr(bdp, &va, attr_flags, credp);
4595 return XFS_ERROR(EINVAL);
4599 * update the inode timestamp, mode, and prealloc flag bits
4601 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4603 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4606 xfs_trans_cancel(tp, 0);
4610 xfs_ilock(ip, XFS_ILOCK_EXCL);
4612 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4613 xfs_trans_ihold(tp, ip);
4615 if ((attr_flags & ATTR_DMI) == 0) {
4616 ip->i_d.di_mode &= ~S_ISUID;
4619 * Note that we don't have to worry about mandatory
4620 * file locking being disabled here because we only
4621 * clear the S_ISGID bit if the Group execute bit is
4622 * on, but if it was on then mandatory locking wouldn't
4623 * have been enabled.
4625 if (ip->i_d.di_mode & S_IXGRP)
4626 ip->i_d.di_mode &= ~S_ISGID;
4628 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4631 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4632 else if (clrprealloc)
4633 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4635 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4636 xfs_trans_set_sync(tp);
4638 error = xfs_trans_commit(tp, 0, NULL);
4640 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4645 vnodeops_t xfs_vnodeops = {
4646 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4647 .vop_open = xfs_open,
4648 .vop_read = xfs_read,
4649 #ifdef HAVE_SENDFILE
4650 .vop_sendfile = xfs_sendfile,
4653 .vop_splice_read = xfs_splice_read,
4654 .vop_splice_write = xfs_splice_write,
4656 .vop_write = xfs_write,
4657 .vop_ioctl = xfs_ioctl,
4658 .vop_getattr = xfs_getattr,
4659 .vop_setattr = xfs_setattr,
4660 .vop_access = xfs_access,
4661 .vop_lookup = xfs_lookup,
4662 .vop_create = xfs_create,
4663 .vop_remove = xfs_remove,
4664 .vop_link = xfs_link,
4665 .vop_rename = xfs_rename,
4666 .vop_mkdir = xfs_mkdir,
4667 .vop_rmdir = xfs_rmdir,
4668 .vop_readdir = xfs_readdir,
4669 .vop_symlink = xfs_symlink,
4670 .vop_readlink = xfs_readlink,
4671 .vop_fsync = xfs_fsync,
4672 .vop_inactive = xfs_inactive,
4673 .vop_fid2 = xfs_fid2,
4674 .vop_rwlock = xfs_rwlock,
4675 .vop_rwunlock = xfs_rwunlock,
4676 .vop_bmap = xfs_bmap,
4677 .vop_reclaim = xfs_reclaim,
4678 .vop_attr_get = xfs_attr_get,
4679 .vop_attr_set = xfs_attr_set,
4680 .vop_attr_remove = xfs_attr_remove,
4681 .vop_attr_list = xfs_attr_list,
4682 .vop_link_removed = (vop_link_removed_t)fs_noval,
4683 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4684 .vop_tosspages = fs_tosspages,
4685 .vop_flushinval_pages = fs_flushinval_pages,
4686 .vop_flush_pages = fs_flush_pages,
4687 .vop_release = xfs_release,
4688 .vop_iflush = xfs_inode_flush,