2 * Copyright (c) 2000-2005 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_xid = ip->i_d.di_xid;
158 vap->va_projid = ip->i_d.di_projid;
161 * Check vnode type block/char vs. everything else.
163 switch (ip->i_d.di_mode & S_IFMT) {
166 vap->va_rdev = ip->i_df.if_u2.if_rdev;
167 vap->va_blocksize = BLKDEV_IOSIZE;
172 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
173 vap->va_blocksize = xfs_preferred_iosize(mp);
177 * If the file blocks are being allocated from a
178 * realtime partition, then return the inode's
179 * realtime extent size or the realtime volume's
182 vap->va_blocksize = ip->i_d.di_extsize ?
183 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
184 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
189 vn_atime_to_timespec(vp, &vap->va_atime);
190 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
191 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
192 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
193 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
196 * Exit for stat callers. See if any of the rest of the fields
197 * to be filled in are needed.
200 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
201 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
205 * Convert di_flags to xflags.
207 vap->va_xflags = xfs_ip2xflags(ip);
210 * Exit for inode revalidate. See if any of the rest of
211 * the fields to be filled in are needed.
214 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
215 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
218 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
220 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
221 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
225 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
226 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
227 ip->i_d.di_anextents;
229 vap->va_anextents = 0;
230 vap->va_gen = ip->i_d.di_gen;
233 if (!(flags & ATTR_LAZY))
234 xfs_iunlock(ip, XFS_ILOCK_SHARED);
261 xfs_prid_t projid=0, iprojid=0;
262 int mandlock_before, mandlock_after;
263 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
267 vp = BHV_TO_VNODE(bdp);
268 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
270 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
271 return XFS_ERROR(EROFS);
274 * Cannot set certain attributes.
277 if (mask & XFS_AT_NOSET) {
278 return XFS_ERROR(EINVAL);
281 ip = XFS_BHVTOI(bdp);
284 if (XFS_FORCED_SHUTDOWN(mp))
285 return XFS_ERROR(EIO);
288 * Timestamps do not need to be logged and hence do not
289 * need to be done within a transaction.
291 if (mask & XFS_AT_UPDTIMES) {
292 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
293 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
294 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
295 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
296 xfs_ichgtime(ip, timeflags);
300 olddquot1 = olddquot2 = NULL;
304 * If disk quotas is on, we make sure that the dquots do exist on disk,
305 * before we start any other transactions. Trying to do this later
306 * is messy. We don't care to take a readlock to look at the ids
307 * in inode here, because we can't hold it across the trans_reserve.
308 * If the IDs do change before we take the ilock, we're covered
309 * because the i_*dquot fields will get updated anyway.
311 if (XFS_IS_QUOTA_ON(mp) &&
312 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
315 /* FIXME: handle xid? */
316 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
318 qflags |= XFS_QMOPT_UQUOTA;
320 uid = ip->i_d.di_uid;
322 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
324 qflags |= XFS_QMOPT_GQUOTA;
326 gid = ip->i_d.di_gid;
328 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
329 projid = vap->va_projid;
330 qflags |= XFS_QMOPT_PQUOTA;
332 projid = ip->i_d.di_projid;
335 * We take a reference when we initialize udqp and gdqp,
336 * so it is important that we never blindly double trip on
337 * the same variable. See xfs_create() for an example.
339 ASSERT(udqp == NULL);
340 ASSERT(gdqp == NULL);
341 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
348 * For the other attributes, we acquire the inode lock and
349 * first do an error checking pass.
352 lock_flags = XFS_ILOCK_EXCL;
353 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
354 if (flags & ATTR_NOLOCK)
356 if (!(mask & XFS_AT_SIZE)) {
357 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
358 (mp->m_flags & XFS_MOUNT_WSYNC)) {
359 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
361 if ((code = xfs_trans_reserve(tp, 0,
362 XFS_ICHANGE_LOG_RES(mp), 0,
369 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
370 !(flags & ATTR_DMI)) {
371 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
372 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
373 vap->va_size, 0, dmflags, NULL);
380 lock_flags |= XFS_IOLOCK_EXCL;
383 xfs_ilock(ip, lock_flags);
385 /* boolean: are we the file owner? */
386 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
389 * Change various properties of a file.
390 * Only the owner or users with CAP_FOWNER
391 * capability may do these things.
394 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
395 XFS_AT_GID|XFS_AT_PROJID)) {
396 /* FIXME: handle xid? */
399 * CAP_FOWNER overrides the following restrictions:
401 * The user ID of the calling process must be equal
402 * to the file owner ID, except in cases where the
403 * CAP_FSETID capability is applicable.
405 if (!file_owner && !capable(CAP_FOWNER)) {
406 code = XFS_ERROR(EPERM);
411 * CAP_FSETID overrides the following restrictions:
413 * The effective user ID of the calling process shall match
414 * the file owner when setting the set-user-ID and
415 * set-group-ID bits on that file.
417 * The effective group ID or one of the supplementary group
418 * IDs of the calling process shall match the group owner of
419 * the file when setting the set-group-ID bit on that file
421 if (mask & XFS_AT_MODE) {
424 if ((vap->va_mode & S_ISUID) && !file_owner)
426 if ((vap->va_mode & S_ISGID) &&
427 !in_group_p((gid_t)ip->i_d.di_gid))
430 /* Linux allows this, Irix doesn't. */
431 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
434 if (m && !capable(CAP_FSETID))
440 * Change file ownership. Must be the owner or privileged.
441 * If the system was configured with the "restricted_chown"
442 * option, the owner is not permitted to give away the file,
443 * and can change the group id only to a group of which he
444 * or she is a member.
446 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_XID|XFS_AT_PROJID)) {
448 * These IDs could have changed since we last looked at them.
449 * But, we're assured that if the ownership did change
450 * while we didn't have the inode locked, inode's dquot(s)
451 * would have changed also.
453 iuid = ip->i_d.di_uid;
454 igid = ip->i_d.di_gid;
455 ixid = ip->i_d.di_xid;
456 iprojid = ip->i_d.di_projid;
457 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
458 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
459 xid = (mask & XFS_AT_XID) ? vap->va_xid : ixid;
460 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
464 * CAP_CHOWN overrides the following restrictions:
466 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
467 * shall override the restriction that a process cannot
468 * change the user ID of a file it owns and the restriction
469 * that the group ID supplied to the chown() function
470 * shall be equal to either the group ID or one of the
471 * supplementary group IDs of the calling process.
473 if (restricted_chown &&
474 (iuid != uid || (igid != gid &&
475 !in_group_p((gid_t)gid))) &&
476 !capable(CAP_CHOWN)) {
477 code = XFS_ERROR(EPERM);
481 * Do a quota reservation only if uid/projid/gid is actually
484 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
485 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
486 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
487 /* FIXME: handle xid? */
489 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
490 capable(CAP_FOWNER) ?
491 XFS_QMOPT_FORCE_RES : 0);
492 if (code) /* out of quota */
498 * Truncate file. Must have write permission and not be a directory.
500 if (mask & XFS_AT_SIZE) {
501 /* Short circuit the truncate case for zero length files */
502 if ((vap->va_size == 0) &&
503 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
504 xfs_iunlock(ip, XFS_ILOCK_EXCL);
505 lock_flags &= ~XFS_ILOCK_EXCL;
506 if (mask & XFS_AT_CTIME)
507 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
513 code = XFS_ERROR(EISDIR);
515 } else if (!VN_ISREG(vp)) {
516 code = XFS_ERROR(EINVAL);
520 * Make sure that the dquots are attached to the inode.
522 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
527 * Change file access or modified times.
529 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
531 if ((flags & ATTR_UTIME) &&
532 !capable(CAP_FOWNER)) {
533 code = XFS_ERROR(EPERM);
540 * Change extent size or realtime flag.
542 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
544 * Can't change extent size if any extents are allocated.
546 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
547 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
549 code = XFS_ERROR(EINVAL); /* EFBIG? */
554 * Can't change realtime flag if any extents are allocated.
556 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
557 (mask & XFS_AT_XFLAGS) &&
558 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
559 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
560 code = XFS_ERROR(EINVAL); /* EFBIG? */
564 * Extent size must be a multiple of the appropriate block
565 * size, if set at all.
567 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
570 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
571 ((mask & XFS_AT_XFLAGS) &&
572 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
573 size = mp->m_sb.sb_rextsize <<
574 mp->m_sb.sb_blocklog;
576 size = mp->m_sb.sb_blocksize;
578 if (vap->va_extsize % size) {
579 code = XFS_ERROR(EINVAL);
584 * If realtime flag is set then must have realtime data.
586 if ((mask & XFS_AT_XFLAGS) &&
587 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
588 if ((mp->m_sb.sb_rblocks == 0) ||
589 (mp->m_sb.sb_rextsize == 0) ||
590 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
591 code = XFS_ERROR(EINVAL);
597 * Can't modify an immutable/append-only file unless
598 * we have appropriate permission.
600 if ((mask & XFS_AT_XFLAGS) &&
602 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
604 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
605 !capable(CAP_LINUX_IMMUTABLE)) {
606 code = XFS_ERROR(EPERM);
612 * Now we can make the changes. Before we join the inode
613 * to the transaction, if XFS_AT_SIZE is set then take care of
614 * the part of the truncation that must be done without the
615 * inode lock. This needs to be done before joining the inode
616 * to the transaction, because the inode cannot be unlocked
617 * once it is a part of the transaction.
619 if (mask & XFS_AT_SIZE) {
621 if ((vap->va_size > ip->i_d.di_size) &&
622 (flags & ATTR_NOSIZETOK) == 0) {
623 code = xfs_igrow_start(ip, vap->va_size, credp);
625 xfs_iunlock(ip, XFS_ILOCK_EXCL);
627 code = xfs_itruncate_data(ip, vap->va_size);
630 lock_flags &= ~XFS_ILOCK_EXCL;
631 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
634 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
635 if ((code = xfs_trans_reserve(tp, 0,
636 XFS_ITRUNCATE_LOG_RES(mp), 0,
637 XFS_TRANS_PERM_LOG_RES,
638 XFS_ITRUNCATE_LOG_COUNT))) {
639 xfs_trans_cancel(tp, 0);
641 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
644 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
645 xfs_ilock(ip, XFS_ILOCK_EXCL);
649 xfs_trans_ijoin(tp, ip, lock_flags);
650 xfs_trans_ihold(tp, ip);
653 /* determine whether mandatory locking mode changes */
654 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
657 * Truncate file. Must have write permission and not be a directory.
659 if (mask & XFS_AT_SIZE) {
660 if (vap->va_size > ip->i_d.di_size) {
661 xfs_igrow_finish(tp, ip, vap->va_size,
662 !(flags & ATTR_DMI));
663 } else if ((vap->va_size <= ip->i_d.di_size) ||
664 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
666 * signal a sync transaction unless
667 * we're truncating an already unlinked
668 * file on a wsync filesystem
670 code = xfs_itruncate_finish(&tp, ip,
671 (xfs_fsize_t)vap->va_size,
673 ((ip->i_d.di_nlink != 0 ||
674 !(mp->m_flags & XFS_MOUNT_WSYNC))
681 * Have to do this even if the file's size doesn't change.
683 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
687 * Change file access modes.
689 if (mask & XFS_AT_MODE) {
690 ip->i_d.di_mode &= S_IFMT;
691 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
693 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
694 timeflags |= XFS_ICHGTIME_CHG;
698 * Change file ownership. Must be the owner or privileged.
699 * If the system was configured with the "restricted_chown"
700 * option, the owner is not permitted to give away the file,
701 * and can change the group id only to a group of which he
702 * or she is a member.
704 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_XID|XFS_AT_PROJID)) {
706 * CAP_FSETID overrides the following restrictions:
708 * The set-user-ID and set-group-ID bits of a file will be
709 * cleared upon successful return from chown()
711 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
712 !capable(CAP_FSETID)) {
713 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
717 * Change the ownerships and register quota modifications
718 * in the transaction.
721 if (XFS_IS_GQUOTA_ON(mp)) {
722 /* FIXME: handle xid quota? */
724 ip->i_d.di_xid = xid;
727 if (XFS_IS_UQUOTA_ON(mp)) {
728 ASSERT(mask & XFS_AT_UID);
730 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
731 &ip->i_udquot, udqp);
733 ip->i_d.di_uid = uid;
736 if (XFS_IS_GQUOTA_ON(mp)) {
737 ASSERT(!XFS_IS_PQUOTA_ON(mp));
738 ASSERT(mask & XFS_AT_GID);
740 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
741 &ip->i_gdquot, gdqp);
743 ip->i_d.di_gid = gid;
745 if (iprojid != projid) {
746 if (XFS_IS_PQUOTA_ON(mp)) {
747 ASSERT(!XFS_IS_GQUOTA_ON(mp));
748 ASSERT(mask & XFS_AT_PROJID);
750 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
751 &ip->i_gdquot, gdqp);
753 ip->i_d.di_projid = projid;
755 * We may have to rev the inode as well as
756 * the superblock version number since projids didn't
757 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
759 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
760 xfs_bump_ino_vers2(tp, ip);
763 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
764 timeflags |= XFS_ICHGTIME_CHG;
769 * Change file access or modified times.
771 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
772 if (mask & XFS_AT_ATIME) {
773 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
774 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
775 ip->i_update_core = 1;
776 timeflags &= ~XFS_ICHGTIME_ACC;
778 if (mask & XFS_AT_MTIME) {
779 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
780 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
781 timeflags &= ~XFS_ICHGTIME_MOD;
782 timeflags |= XFS_ICHGTIME_CHG;
784 if (tp && (flags & ATTR_UTIME))
785 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
789 * Change XFS-added attributes.
791 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
792 if (mask & XFS_AT_EXTSIZE) {
794 * Converting bytes to fs blocks.
796 ip->i_d.di_extsize = vap->va_extsize >>
797 mp->m_sb.sb_blocklog;
799 if (mask & XFS_AT_XFLAGS) {
802 /* can't set PREALLOC this way, just preserve it */
803 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
804 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
805 di_flags |= XFS_DIFLAG_IMMUTABLE;
806 if (vap->va_xflags & XFS_XFLAG_IUNLINK)
807 di_flags |= XFS_DIFLAG_IUNLINK;
808 if (vap->va_xflags & XFS_XFLAG_BARRIER)
809 di_flags |= XFS_DIFLAG_BARRIER;
810 if (vap->va_xflags & XFS_XFLAG_APPEND)
811 di_flags |= XFS_DIFLAG_APPEND;
812 if (vap->va_xflags & XFS_XFLAG_SYNC)
813 di_flags |= XFS_DIFLAG_SYNC;
814 if (vap->va_xflags & XFS_XFLAG_NOATIME)
815 di_flags |= XFS_DIFLAG_NOATIME;
816 if (vap->va_xflags & XFS_XFLAG_NODUMP)
817 di_flags |= XFS_DIFLAG_NODUMP;
818 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
819 di_flags |= XFS_DIFLAG_PROJINHERIT;
820 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
821 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
822 di_flags |= XFS_DIFLAG_RTINHERIT;
823 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
824 di_flags |= XFS_DIFLAG_NOSYMLINKS;
825 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
826 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
827 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
828 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
829 di_flags |= XFS_DIFLAG_REALTIME;
830 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
832 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
834 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
835 di_flags |= XFS_DIFLAG_EXTSIZE;
837 ip->i_d.di_flags = di_flags;
839 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
840 timeflags |= XFS_ICHGTIME_CHG;
844 * Change file inode change time only if XFS_AT_CTIME set
845 * AND we have been called by a DMI function.
848 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
849 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
850 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
851 ip->i_update_core = 1;
852 timeflags &= ~XFS_ICHGTIME_CHG;
856 * Send out timestamp changes that need to be set to the
857 * current time. Not done when called by a DMI function.
859 if (timeflags && !(flags & ATTR_DMI))
860 xfs_ichgtime(ip, timeflags);
862 XFS_STATS_INC(xs_ig_attrchg);
865 * If this is a synchronous mount, make sure that the
866 * transaction goes to disk before returning to the user.
867 * This is slightly sub-optimal in that truncates require
868 * two sync transactions instead of one for wsync filesytems.
869 * One for the truncate and one for the timestamps since we
870 * don't want to change the timestamps unless we're sure the
871 * truncate worked. Truncates are less than 1% of the laddis
872 * mix so this probably isn't worth the trouble to optimize.
876 if (mp->m_flags & XFS_MOUNT_WSYNC)
877 xfs_trans_set_sync(tp);
879 code = xfs_trans_commit(tp, commit_flags, NULL);
883 * If the (regular) file's mandatory locking mode changed, then
884 * notify the vnode. We do this under the inode lock to prevent
885 * racing calls to vop_vnode_change.
887 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
888 if (mandlock_before != mandlock_after) {
889 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
893 xfs_iunlock(ip, lock_flags);
896 * Release any dquot(s) the inode had kept before chown.
898 XFS_QM_DQRELE(mp, olddquot1);
899 XFS_QM_DQRELE(mp, olddquot2);
900 XFS_QM_DQRELE(mp, udqp);
901 XFS_QM_DQRELE(mp, gdqp);
907 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
908 !(flags & ATTR_DMI)) {
909 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
910 NULL, DM_RIGHT_NULL, NULL, NULL,
911 0, 0, AT_DELAY_FLAG(flags));
916 commit_flags |= XFS_TRANS_ABORT;
919 XFS_QM_DQRELE(mp, udqp);
920 XFS_QM_DQRELE(mp, gdqp);
922 xfs_trans_cancel(tp, commit_flags);
924 if (lock_flags != 0) {
925 xfs_iunlock(ip, lock_flags);
933 * Null conversion from vnode mode bits to inode mode bits, as in efs.
944 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
945 (inst_t *)__return_address);
947 ip = XFS_BHVTOI(bdp);
948 xfs_ilock(ip, XFS_ILOCK_SHARED);
949 error = xfs_iaccess(ip, mode, credp);
950 xfs_iunlock(ip, XFS_ILOCK_SHARED);
974 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
980 vp = BHV_TO_VNODE(bdp);
981 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
983 ip = XFS_BHVTOI(bdp);
986 if (XFS_FORCED_SHUTDOWN(mp))
987 return XFS_ERROR(EIO);
989 xfs_ilock(ip, XFS_ILOCK_SHARED);
991 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
993 offset = uiop->uio_offset;
994 count = uiop->uio_resid;
997 error = XFS_ERROR(EINVAL);
1006 * See if the symlink is stored inline.
1008 pathlen = (int)ip->i_d.di_size;
1010 if (ip->i_df.if_flags & XFS_IFINLINE) {
1011 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1015 * Symlink not inline. Call bmap to get it in.
1017 nmaps = SYMLINK_MAPS;
1019 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1020 0, NULL, 0, mval, &nmaps, NULL);
1026 for (n = 0; n < nmaps; n++) {
1027 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1028 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1029 bp = xfs_buf_read(mp->m_ddev_targp, d,
1030 BTOBB(byte_cnt), 0);
1031 error = XFS_BUF_GETERROR(bp);
1033 xfs_ioerror_alert("xfs_readlink",
1034 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1038 if (pathlen < byte_cnt)
1040 pathlen -= byte_cnt;
1042 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1049 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1057 * This is called to sync the inode and its data out to disk.
1058 * We need to hold the I/O lock while flushing the data, and
1059 * the inode lock while flushing the inode. The inode lock CANNOT
1060 * be held while flushing the data, so acquire after we're done
1074 int log_flushed = 0, changed = 1;
1076 vn_trace_entry(BHV_TO_VNODE(bdp),
1077 __FUNCTION__, (inst_t *)__return_address);
1079 ip = XFS_BHVTOI(bdp);
1081 ASSERT(start >= 0 && stop >= -1);
1083 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1084 return XFS_ERROR(EIO);
1087 * We always need to make sure that the required inode state
1088 * is safe on disk. The vnode might be clean but because
1089 * of committed transactions that haven't hit the disk yet.
1090 * Likewise, there could be unflushed non-transactional
1091 * changes to the inode core that have to go to disk.
1093 * The following code depends on one assumption: that
1094 * any transaction that changes an inode logs the core
1095 * because it has to change some field in the inode core
1096 * (typically nextents or nblocks). That assumption
1097 * implies that any transactions against an inode will
1098 * catch any non-transactional updates. If inode-altering
1099 * transactions exist that violate this assumption, the
1100 * code breaks. Right now, it figures that if the involved
1101 * update_* field is clear and the inode is unpinned, the
1102 * inode is clean. Either it's been flushed or it's been
1103 * committed and the commit has hit the disk unpinning the inode.
1104 * (Note that xfs_inode_item_format() called at commit clears
1105 * the update_* fields.)
1107 xfs_ilock(ip, XFS_ILOCK_SHARED);
1109 /* If we are flushing data then we care about update_size
1110 * being set, otherwise we care about update_core
1112 if ((flag & FSYNC_DATA) ?
1113 (ip->i_update_size == 0) :
1114 (ip->i_update_core == 0)) {
1116 * Timestamps/size haven't changed since last inode
1117 * flush or inode transaction commit. That means
1118 * either nothing got written or a transaction
1119 * committed which caught the updates. If the
1120 * latter happened and the transaction hasn't
1121 * hit the disk yet, the inode will be still
1122 * be pinned. If it is, force the log.
1125 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1127 if (xfs_ipincount(ip)) {
1128 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1130 ((flag & FSYNC_WAIT)
1131 ? XFS_LOG_SYNC : 0),
1135 * If the inode is not pinned and nothing
1136 * has changed we don't need to flush the
1144 * Kick off a transaction to log the inode
1145 * core to get the updates. Make it
1146 * sync if FSYNC_WAIT is passed in (which
1147 * is done by everybody but specfs). The
1148 * sync transaction will also force the log.
1150 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1151 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1152 if ((error = xfs_trans_reserve(tp, 0,
1153 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1155 xfs_trans_cancel(tp, 0);
1158 xfs_ilock(ip, XFS_ILOCK_EXCL);
1161 * Note - it's possible that we might have pushed
1162 * ourselves out of the way during trans_reserve
1163 * which would flush the inode. But there's no
1164 * guarantee that the inode buffer has actually
1165 * gone out yet (it's delwri). Plus the buffer
1166 * could be pinned anyway if it's part of an
1167 * inode in another recent transaction. So we
1168 * play it safe and fire off the transaction anyway.
1170 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1171 xfs_trans_ihold(tp, ip);
1172 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1173 if (flag & FSYNC_WAIT)
1174 xfs_trans_set_sync(tp);
1175 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1177 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1180 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1182 * If the log write didn't issue an ordered tag we need
1183 * to flush the disk cache for the data device now.
1186 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1189 * If this inode is on the RT dev we need to flush that
1192 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1193 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1200 * This is called by xfs_inactive to free any blocks beyond eof,
1201 * when the link count isn't zero.
1204 xfs_inactive_free_eofblocks(
1210 xfs_fileoff_t end_fsb;
1211 xfs_fileoff_t last_fsb;
1212 xfs_filblks_t map_len;
1214 xfs_bmbt_irec_t imap;
1217 * Figure out if there are any blocks beyond the end
1218 * of the file. If not, then there is nothing to do.
1220 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1221 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1222 map_len = last_fsb - end_fsb;
1227 xfs_ilock(ip, XFS_ILOCK_SHARED);
1228 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1229 NULL, 0, &imap, &nimaps, NULL);
1230 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1232 if (!error && (nimaps != 0) &&
1233 (imap.br_startblock != HOLESTARTBLOCK ||
1234 ip->i_delayed_blks)) {
1236 * Attach the dquots to the inode up front.
1238 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1242 * There are blocks after the end of file.
1243 * Free them up now by truncating the file to
1246 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1249 * Do the xfs_itruncate_start() call before
1250 * reserving any log space because
1251 * itruncate_start will call into the buffer
1252 * cache and we can't
1253 * do that within a transaction.
1255 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1256 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1259 error = xfs_trans_reserve(tp, 0,
1260 XFS_ITRUNCATE_LOG_RES(mp),
1261 0, XFS_TRANS_PERM_LOG_RES,
1262 XFS_ITRUNCATE_LOG_COUNT);
1264 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1265 xfs_trans_cancel(tp, 0);
1266 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1270 xfs_ilock(ip, XFS_ILOCK_EXCL);
1271 xfs_trans_ijoin(tp, ip,
1274 xfs_trans_ihold(tp, ip);
1276 error = xfs_itruncate_finish(&tp, ip,
1281 * If we get an error at this point we
1282 * simply don't bother truncating the file.
1285 xfs_trans_cancel(tp,
1286 (XFS_TRANS_RELEASE_LOG_RES |
1289 error = xfs_trans_commit(tp,
1290 XFS_TRANS_RELEASE_LOG_RES,
1293 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1299 * Free a symlink that has blocks associated with it.
1302 xfs_inactive_symlink_rmt(
1310 xfs_fsblock_t first_block;
1311 xfs_bmap_free_t free_list;
1314 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1322 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1324 * We're freeing a symlink that has some
1325 * blocks allocated to it. Free the
1326 * blocks here. We know that we've got
1327 * either 1 or 2 extents and that we can
1328 * free them all in one bunmapi call.
1330 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1331 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1332 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1333 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1334 xfs_trans_cancel(tp, 0);
1339 * Lock the inode, fix the size, and join it to the transaction.
1340 * Hold it so in the normal path, we still have it locked for
1341 * the second transaction. In the error paths we need it
1342 * held so the cancel won't rele it, see below.
1344 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1345 size = (int)ip->i_d.di_size;
1346 ip->i_d.di_size = 0;
1347 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1348 xfs_trans_ihold(tp, ip);
1349 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1351 * Find the block(s) so we can inval and unmap them.
1354 XFS_BMAP_INIT(&free_list, &first_block);
1355 nmaps = sizeof(mval) / sizeof(mval[0]);
1356 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1357 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1361 * Invalidate the block(s).
1363 for (i = 0; i < nmaps; i++) {
1364 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1365 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1366 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1367 xfs_trans_binval(tp, bp);
1370 * Unmap the dead block(s) to the free_list.
1372 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1373 &first_block, &free_list, &done)))
1377 * Commit the first transaction. This logs the EFI and the inode.
1379 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1382 * The transaction must have been committed, since there were
1383 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1384 * The new tp has the extent freeing and EFDs.
1388 * The first xact was committed, so add the inode to the new one.
1389 * Mark it dirty so it will be logged and moved forward in the log as
1390 * part of every commit.
1392 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1393 xfs_trans_ihold(tp, ip);
1394 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1396 * Get a new, empty transaction to return to our caller.
1398 ntp = xfs_trans_dup(tp);
1400 * Commit the transaction containing extent freeing and EFD's.
1401 * If we get an error on the commit here or on the reserve below,
1402 * we need to unlock the inode since the new transaction doesn't
1403 * have the inode attached.
1405 error = xfs_trans_commit(tp, 0, NULL);
1408 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1412 * Remove the memory for extent descriptions (just bookkeeping).
1414 if (ip->i_df.if_bytes)
1415 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1416 ASSERT(ip->i_df.if_bytes == 0);
1418 * Put an itruncate log reservation in the new transaction
1421 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1422 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1423 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1427 * Return with the inode locked but not joined to the transaction.
1433 xfs_bmap_cancel(&free_list);
1436 * Have to come here with the inode locked and either
1437 * (held and in the transaction) or (not in the transaction).
1438 * If the inode isn't held then cancel would iput it, but
1439 * that's wrong since this is inactive and the vnode ref
1440 * count is 0 already.
1441 * Cancel won't do anything to the inode if held, but it still
1442 * needs to be locked until the cancel is done, if it was
1443 * joined to the transaction.
1445 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1446 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1453 xfs_inactive_symlink_local(
1459 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1461 * We're freeing a symlink which fit into
1462 * the inode. Just free the memory used
1463 * to hold the old symlink.
1465 error = xfs_trans_reserve(*tpp, 0,
1466 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1467 0, XFS_TRANS_PERM_LOG_RES,
1468 XFS_ITRUNCATE_LOG_COUNT);
1471 xfs_trans_cancel(*tpp, 0);
1475 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1478 * Zero length symlinks _can_ exist.
1480 if (ip->i_df.if_bytes > 0) {
1481 xfs_idata_realloc(ip,
1482 -(ip->i_df.if_bytes),
1484 ASSERT(ip->i_df.if_bytes == 0);
1501 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1504 ASSERT(ip->i_d.di_forkoff != 0);
1505 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1506 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1508 error = xfs_attr_inactive(ip);
1511 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1512 return error; /* goto out */
1515 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1516 error = xfs_trans_reserve(tp, 0,
1517 XFS_IFREE_LOG_RES(mp),
1518 0, XFS_TRANS_PERM_LOG_RES,
1519 XFS_INACTIVE_LOG_COUNT);
1521 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1522 xfs_trans_cancel(tp, 0);
1524 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1528 xfs_ilock(ip, XFS_ILOCK_EXCL);
1529 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1530 xfs_trans_ihold(tp, ip);
1531 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1533 ASSERT(ip->i_d.di_anextents == 0);
1548 vp = BHV_TO_VNODE(bdp);
1549 ip = XFS_BHVTOI(bdp);
1551 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1555 /* If this is a read-only mount, don't do this (would generate I/O) */
1556 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1559 #ifdef HAVE_REFCACHE
1560 /* If we are in the NFS reference cache then don't do this now */
1567 if (ip->i_d.di_nlink != 0) {
1568 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1569 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1570 ip->i_delayed_blks > 0)) &&
1571 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1572 (!(ip->i_d.di_flags &
1573 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1574 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1576 /* Update linux inode block count after free above */
1577 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1578 ip->i_d.di_nblocks + ip->i_delayed_blks);
1588 * This is called when the vnode reference count for the vnode
1589 * goes to zero. If the file has been unlinked, then it must
1590 * now be truncated. Also, we clear all of the read-ahead state
1591 * kept for the inode here since the file is now closed.
1600 xfs_bmap_free_t free_list;
1601 xfs_fsblock_t first_block;
1608 vp = BHV_TO_VNODE(bdp);
1609 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1611 ip = XFS_BHVTOI(bdp);
1614 * If the inode is already free, then there can be nothing
1617 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1618 ASSERT(ip->i_df.if_real_bytes == 0);
1619 ASSERT(ip->i_df.if_broot_bytes == 0);
1620 return VN_INACTIVE_CACHE;
1624 * Only do a truncate if it's a regular file with
1625 * some actual space in it. It's OK to look at the
1626 * inode's fields without the lock because we're the
1627 * only one with a reference to the inode.
1629 truncate = ((ip->i_d.di_nlink == 0) &&
1630 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1631 (ip->i_delayed_blks > 0)) &&
1632 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1636 if (ip->i_d.di_nlink == 0 &&
1637 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1638 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1643 /* If this is a read-only mount, don't do this (would generate I/O) */
1644 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1647 if (ip->i_d.di_nlink != 0) {
1648 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1649 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1650 ip->i_delayed_blks > 0)) &&
1651 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1652 (!(ip->i_d.di_flags &
1653 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1654 (ip->i_delayed_blks != 0)))) {
1655 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1656 return VN_INACTIVE_CACHE;
1657 /* Update linux inode block count after free above */
1658 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1659 ip->i_d.di_nblocks + ip->i_delayed_blks);
1664 ASSERT(ip->i_d.di_nlink == 0);
1666 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1667 return VN_INACTIVE_CACHE;
1669 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1672 * Do the xfs_itruncate_start() call before
1673 * reserving any log space because itruncate_start
1674 * will call into the buffer cache and we can't
1675 * do that within a transaction.
1677 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1679 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1681 error = xfs_trans_reserve(tp, 0,
1682 XFS_ITRUNCATE_LOG_RES(mp),
1683 0, XFS_TRANS_PERM_LOG_RES,
1684 XFS_ITRUNCATE_LOG_COUNT);
1686 /* Don't call itruncate_cleanup */
1687 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1688 xfs_trans_cancel(tp, 0);
1689 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1690 return VN_INACTIVE_CACHE;
1693 xfs_ilock(ip, XFS_ILOCK_EXCL);
1694 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1695 xfs_trans_ihold(tp, ip);
1698 * normally, we have to run xfs_itruncate_finish sync.
1699 * But if filesystem is wsync and we're in the inactive
1700 * path, then we know that nlink == 0, and that the
1701 * xaction that made nlink == 0 is permanently committed
1702 * since xfs_remove runs as a synchronous transaction.
1704 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1705 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1708 xfs_trans_cancel(tp,
1709 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1710 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1711 return VN_INACTIVE_CACHE;
1713 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1716 * If we get an error while cleaning up a
1717 * symlink we bail out.
1719 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1720 xfs_inactive_symlink_rmt(ip, &tp) :
1721 xfs_inactive_symlink_local(ip, &tp);
1725 return VN_INACTIVE_CACHE;
1728 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1729 xfs_trans_ihold(tp, ip);
1731 error = xfs_trans_reserve(tp, 0,
1732 XFS_IFREE_LOG_RES(mp),
1733 0, XFS_TRANS_PERM_LOG_RES,
1734 XFS_INACTIVE_LOG_COUNT);
1736 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1737 xfs_trans_cancel(tp, 0);
1738 return VN_INACTIVE_CACHE;
1741 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1742 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1743 xfs_trans_ihold(tp, ip);
1747 * If there are attributes associated with the file
1748 * then blow them away now. The code calls a routine
1749 * that recursively deconstructs the attribute fork.
1750 * We need to just commit the current transaction
1751 * because we can't use it for xfs_attr_inactive().
1753 if (ip->i_d.di_anextents > 0) {
1754 error = xfs_inactive_attrs(ip, &tp);
1756 * If we got an error, the transaction is already
1757 * cancelled, and the inode is unlocked. Just get out.
1760 return VN_INACTIVE_CACHE;
1761 } else if (ip->i_afp) {
1762 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1768 XFS_BMAP_INIT(&free_list, &first_block);
1769 error = xfs_ifree(tp, ip, &free_list);
1772 * If we fail to free the inode, shut down. The cancel
1773 * might do that, we need to make sure. Otherwise the
1774 * inode might be lost for a long time or forever.
1776 if (!XFS_FORCED_SHUTDOWN(mp)) {
1778 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1779 error, mp->m_fsname);
1780 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1782 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1785 * Credit the quota account(s). The inode is gone.
1787 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1790 * Just ignore errors at this point. There is
1791 * nothing we can do except to try to keep going.
1793 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1795 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1798 * Release the dquots held by inode, if any.
1800 XFS_QM_DQDETACH(mp, ip);
1802 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1805 return VN_INACTIVE_CACHE;
1814 bhv_desc_t *dir_bdp,
1821 xfs_inode_t *dp, *ip;
1827 dir_vp = BHV_TO_VNODE(dir_bdp);
1828 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1830 dp = XFS_BHVTOI(dir_bdp);
1832 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1833 return XFS_ERROR(EIO);
1835 lock_mode = xfs_ilock_map_shared(dp);
1836 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1838 *vpp = XFS_ITOV(ip);
1841 xfs_iunlock_map_shared(dp, lock_mode);
1847 * xfs_create (create a new file).
1851 bhv_desc_t *dir_bdp,
1857 char *name = VNAME(dentry);
1859 xfs_inode_t *dp, *ip;
1865 xfs_bmap_free_t free_list;
1866 xfs_fsblock_t first_block;
1867 boolean_t dp_joined_to_trans;
1868 int dm_event_sent = 0;
1872 struct xfs_dquot *udqp, *gdqp;
1878 dir_vp = BHV_TO_VNODE(dir_bdp);
1879 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1881 dp = XFS_BHVTOI(dir_bdp);
1884 dm_di_mode = vap->va_mode;
1885 namelen = VNAMELEN(dentry);
1887 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1888 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1889 dir_vp, DM_RIGHT_NULL, NULL,
1890 DM_RIGHT_NULL, name, NULL,
1898 if (XFS_FORCED_SHUTDOWN(mp))
1899 return XFS_ERROR(EIO);
1901 /* Return through std_return after this point. */
1904 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1905 prid = dp->i_d.di_projid;
1906 else if (vap->va_mask & XFS_AT_PROJID)
1907 prid = (xfs_prid_t)vap->va_projid;
1909 prid = (xfs_prid_t)dfltprid;
1912 * Make sure that we have allocated dquot(s) on disk.
1914 error = XFS_QM_DQVOPALLOC(mp, dp,
1915 current_fsuid(credp), current_fsgid(credp), prid,
1916 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1921 dp_joined_to_trans = B_FALSE;
1923 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1924 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1925 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1927 * Initially assume that the file does not exist and
1928 * reserve the resources for that case. If that is not
1929 * the case we'll drop the one we have and get a more
1930 * appropriate transaction later.
1932 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1933 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1934 if (error == ENOSPC) {
1936 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1937 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1945 xfs_ilock(dp, XFS_ILOCK_EXCL);
1947 XFS_BMAP_INIT(&free_list, &first_block);
1952 * Reserve disk quota and the inode.
1954 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1959 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1961 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1962 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1963 rdev, credp, prid, resblks > 0,
1966 if (error == ENOSPC)
1973 * At this point, we've gotten a newly allocated inode.
1974 * It is locked (and joined to the transaction).
1977 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1980 * Now we join the directory inode to the transaction.
1981 * We do not do it earlier because xfs_dir_ialloc
1982 * might commit the previous transaction (and release
1987 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1988 dp_joined_to_trans = B_TRUE;
1990 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1991 &first_block, &free_list,
1992 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1994 ASSERT(error != ENOSPC);
1997 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1998 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2001 * If this is a synchronous mount, make sure that the
2002 * create transaction goes to disk before returning to
2005 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2006 xfs_trans_set_sync(tp);
2012 * Attach the dquot(s) to the inodes and modify them incore.
2013 * These ids of the inode couldn't have changed since the new
2014 * inode has been locked ever since it was created.
2016 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2019 * xfs_trans_commit normally decrements the vnode ref count
2020 * when it unlocks the inode. Since we want to return the
2021 * vnode to the caller, we bump the vnode ref count now.
2026 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2028 xfs_bmap_cancel(&free_list);
2032 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2039 XFS_QM_DQRELE(mp, udqp);
2040 XFS_QM_DQRELE(mp, gdqp);
2043 * Propogate the fact that the vnode changed after the
2044 * xfs_inode locks have been released.
2046 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2050 /* Fallthrough to std_return with error = 0 */
2053 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2054 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2055 DM_EVENT_POSTCREATE)) {
2056 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2057 dir_vp, DM_RIGHT_NULL,
2059 DM_RIGHT_NULL, name, NULL,
2060 dm_di_mode, error, 0);
2065 cancel_flags |= XFS_TRANS_ABORT;
2070 xfs_trans_cancel(tp, cancel_flags);
2072 if (!dp_joined_to_trans && (dp != NULL))
2073 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2074 XFS_QM_DQRELE(mp, udqp);
2075 XFS_QM_DQRELE(mp, gdqp);
2081 * Wait until after the current transaction is aborted to
2082 * release the inode. This prevents recursive transactions
2083 * and deadlocks from xfs_inactive.
2085 cancel_flags |= XFS_TRANS_ABORT;
2086 xfs_trans_cancel(tp, cancel_flags);
2089 XFS_QM_DQRELE(mp, udqp);
2090 XFS_QM_DQRELE(mp, gdqp);
2097 * Some counters to see if (and how often) we are hitting some deadlock
2098 * prevention code paths.
2102 int xfs_rm_lock_delays;
2103 int xfs_rm_attempts;
2107 * The following routine will lock the inodes associated with the
2108 * directory and the named entry in the directory. The locks are
2109 * acquired in increasing inode number.
2111 * If the entry is "..", then only the directory is locked. The
2112 * vnode ref count will still include that from the .. entry in
2115 * There is a deadlock we need to worry about. If the locked directory is
2116 * in the AIL, it might be blocking up the log. The next inode we lock
2117 * could be already locked by another thread waiting for log space (e.g
2118 * a permanent log reservation with a long running transaction (see
2119 * xfs_itruncate_finish)). To solve this, we must check if the directory
2120 * is in the ail and use lock_nowait. If we can't lock, we need to
2121 * drop the inode lock on the directory and try again. xfs_iunlock will
2122 * potentially push the tail if we were holding up the log.
2125 xfs_lock_dir_and_entry(
2128 xfs_inode_t *ip) /* inode of entry 'name' */
2132 xfs_inode_t *ips[2];
2141 xfs_ilock(dp, XFS_ILOCK_EXCL);
2148 * We want to lock in increasing inum. Since we've already
2149 * acquired the lock on the directory, we may need to release
2150 * if if the inum of the entry turns out to be less.
2152 if (e_inum > dp->i_ino) {
2154 * We are already in the right order, so just
2155 * lock on the inode of the entry.
2156 * We need to use nowait if dp is in the AIL.
2159 lp = (xfs_log_item_t *)dp->i_itemp;
2160 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2161 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2168 * Unlock dp and try again.
2169 * xfs_iunlock will try to push the tail
2170 * if the inode is in the AIL.
2173 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2175 if ((attempts % 5) == 0) {
2176 delay(1); /* Don't just spin the CPU */
2178 xfs_rm_lock_delays++;
2184 xfs_ilock(ip, XFS_ILOCK_EXCL);
2186 } else if (e_inum < dp->i_ino) {
2187 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2191 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2193 /* else e_inum == dp->i_ino */
2194 /* This can happen if we're asked to lock /x/..
2195 * the entry is "..", which is also the parent directory.
2203 int xfs_small_retries;
2204 int xfs_middle_retries;
2205 int xfs_lots_retries;
2206 int xfs_lock_delays;
2210 * The following routine will lock n inodes in exclusive mode.
2211 * We assume the caller calls us with the inodes in i_ino order.
2213 * We need to detect deadlock where an inode that we lock
2214 * is in the AIL and we start waiting for another inode that is locked
2215 * by a thread in a long running transaction (such as truncate). This can
2216 * result in deadlock since the long running trans might need to wait
2217 * for the inode we just locked in order to push the tail and free space
2227 int attempts = 0, i, j, try_lock;
2230 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2241 for (; i < inodes; i++) {
2244 if (i && (ips[i] == ips[i-1])) /* Already locked */
2248 * If try_lock is not set yet, make sure all locked inodes
2249 * are not in the AIL.
2250 * If any are, set try_lock to be used later.
2254 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2255 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2256 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2263 * If any of the previous locks we have locked is in the AIL,
2264 * we must TRY to get the second and subsequent locks. If
2265 * we can't get any, we must release all we have
2270 /* try_lock must be 0 if i is 0. */
2272 * try_lock means we have an inode locked
2273 * that is in the AIL.
2276 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2280 * Unlock all previous guys and try again.
2281 * xfs_iunlock will try to push the tail
2282 * if the inode is in the AIL.
2285 for(j = i - 1; j >= 0; j--) {
2288 * Check to see if we've already
2289 * unlocked this one.
2290 * Not the first one going back,
2291 * and the inode ptr is the same.
2293 if ((j != (i - 1)) && ips[j] ==
2297 xfs_iunlock(ips[j], lock_mode);
2300 if ((attempts % 5) == 0) {
2301 delay(1); /* Don't just spin the CPU */
2311 xfs_ilock(ips[i], lock_mode);
2317 if (attempts < 5) xfs_small_retries++;
2318 else if (attempts < 100) xfs_middle_retries++;
2319 else xfs_lots_retries++;
2327 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2328 int remove_which_error_return = 0;
2330 #define REMOVE_DEBUG_TRACE(x)
2331 #endif /* ! DEBUG */
2340 bhv_desc_t *dir_bdp,
2345 char *name = VNAME(dentry);
2346 xfs_inode_t *dp, *ip;
2347 xfs_trans_t *tp = NULL;
2350 xfs_bmap_free_t free_list;
2351 xfs_fsblock_t first_block;
2359 dir_vp = BHV_TO_VNODE(dir_bdp);
2360 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2362 dp = XFS_BHVTOI(dir_bdp);
2365 if (XFS_FORCED_SHUTDOWN(mp))
2366 return XFS_ERROR(EIO);
2368 namelen = VNAMELEN(dentry);
2370 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2371 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2372 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2373 name, NULL, 0, 0, 0);
2378 /* From this point on, return through std_return */
2382 * We need to get a reference to ip before we get our log
2383 * reservation. The reason for this is that we cannot call
2384 * xfs_iget for an inode for which we do not have a reference
2385 * once we've acquired a log reservation. This is because the
2386 * inode we are trying to get might be in xfs_inactive going
2387 * for a log reservation. Since we'll have to wait for the
2388 * inactive code to complete before returning from xfs_iget,
2389 * we need to make sure that we don't have log space reserved
2390 * when we call xfs_iget. Instead we get an unlocked referece
2391 * to the inode before getting our log reservation.
2393 error = xfs_get_dir_entry(dentry, &ip);
2395 REMOVE_DEBUG_TRACE(__LINE__);
2399 dm_di_mode = ip->i_d.di_mode;
2401 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2405 error = XFS_QM_DQATTACH(mp, dp, 0);
2406 if (!error && dp != ip)
2407 error = XFS_QM_DQATTACH(mp, ip, 0);
2409 REMOVE_DEBUG_TRACE(__LINE__);
2414 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2415 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2417 * We try to get the real space reservation first,
2418 * allowing for directory btree deletion(s) implying
2419 * possible bmap insert(s). If we can't get the space
2420 * reservation then we use 0 instead, and avoid the bmap
2421 * btree insert(s) in the directory code by, if the bmap
2422 * insert tries to happen, instead trimming the LAST
2423 * block from the directory.
2425 resblks = XFS_REMOVE_SPACE_RES(mp);
2426 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2427 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2428 if (error == ENOSPC) {
2430 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2431 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2434 ASSERT(error != ENOSPC);
2435 REMOVE_DEBUG_TRACE(__LINE__);
2436 xfs_trans_cancel(tp, 0);
2441 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2443 REMOVE_DEBUG_TRACE(__LINE__);
2444 xfs_trans_cancel(tp, cancel_flags);
2450 * At this point, we've gotten both the directory and the entry
2453 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2456 * Increment vnode ref count only in this case since
2457 * there's an extra vnode reference in the case where
2461 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2465 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2467 XFS_BMAP_INIT(&free_list, &first_block);
2468 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2469 &first_block, &free_list, 0);
2471 ASSERT(error != ENOENT);
2472 REMOVE_DEBUG_TRACE(__LINE__);
2475 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2478 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2480 error = xfs_droplink(tp, ip);
2482 REMOVE_DEBUG_TRACE(__LINE__);
2486 /* Determine if this is the last link while
2487 * we are in the transaction.
2489 link_zero = (ip)->i_d.di_nlink==0;
2492 * Take an extra ref on the inode so that it doesn't
2493 * go to xfs_inactive() from within the commit.
2498 * If this is a synchronous mount, make sure that the
2499 * remove transaction goes to disk before returning to
2502 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2503 xfs_trans_set_sync(tp);
2506 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2508 REMOVE_DEBUG_TRACE(__LINE__);
2512 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2519 * Before we drop our extra reference to the inode, purge it
2520 * from the refcache if it is there. By waiting until afterwards
2521 * to do the IRELE, we ensure that we won't go inactive in the
2522 * xfs_refcache_purge_ip routine (although that would be OK).
2524 xfs_refcache_purge_ip(ip);
2526 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2529 * Let interposed file systems know about removed links.
2531 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2535 /* Fall through to std_return with error = 0 */
2537 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2538 DM_EVENT_POSTREMOVE)) {
2539 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2540 dir_vp, DM_RIGHT_NULL,
2541 NULL, DM_RIGHT_NULL,
2542 name, NULL, dm_di_mode, error, 0);
2547 xfs_bmap_cancel(&free_list);
2548 cancel_flags |= XFS_TRANS_ABORT;
2549 xfs_trans_cancel(tp, cancel_flags);
2554 * In this case make sure to not release the inode until after
2555 * the current transaction is aborted. Releasing it beforehand
2556 * can cause us to go to xfs_inactive and start a recursive
2557 * transaction which can easily deadlock with the current one.
2559 xfs_bmap_cancel(&free_list);
2560 cancel_flags |= XFS_TRANS_ABORT;
2561 xfs_trans_cancel(tp, cancel_flags);
2564 * Before we drop our extra reference to the inode, purge it
2565 * from the refcache if it is there. By waiting until afterwards
2566 * to do the IRELE, we ensure that we won't go inactive in the
2567 * xfs_refcache_purge_ip routine (although that would be OK).
2569 xfs_refcache_purge_ip(ip);
2583 bhv_desc_t *target_dir_bdp,
2588 xfs_inode_t *tdp, *sip;
2591 xfs_inode_t *ips[2];
2593 xfs_bmap_free_t free_list;
2594 xfs_fsblock_t first_block;
2597 vnode_t *target_dir_vp;
2599 char *target_name = VNAME(dentry);
2602 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2603 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2604 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2606 target_namelen = VNAMELEN(dentry);
2607 if (VN_ISDIR(src_vp))
2608 return XFS_ERROR(EPERM);
2610 sip = xfs_vtoi(src_vp);
2611 tdp = XFS_BHVTOI(target_dir_bdp);
2613 if (XFS_FORCED_SHUTDOWN(mp))
2614 return XFS_ERROR(EIO);
2616 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2617 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2618 target_dir_vp, DM_RIGHT_NULL,
2619 src_vp, DM_RIGHT_NULL,
2620 target_name, NULL, 0, 0, 0);
2625 /* Return through std_return after this point. */
2627 error = XFS_QM_DQATTACH(mp, sip, 0);
2628 if (!error && sip != tdp)
2629 error = XFS_QM_DQATTACH(mp, tdp, 0);
2633 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2634 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2635 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2636 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2637 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2638 if (error == ENOSPC) {
2640 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2641 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2648 if (sip->i_ino < tdp->i_ino) {
2656 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2659 * Increment vnode ref counts since xfs_trans_commit &
2660 * xfs_trans_cancel will both unlock the inodes and
2661 * decrement the associated ref counts.
2664 VN_HOLD(target_dir_vp);
2665 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2666 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2669 * If the source has too many links, we can't make any more to it.
2671 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2672 error = XFS_ERROR(EMLINK);
2677 * If we are using project inheritance, we only allow hard link
2678 * creation in our tree when the project IDs are the same; else
2679 * the tree quota mechanism could be circumvented.
2681 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2682 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2683 error = XFS_ERROR(EPERM);
2688 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2692 XFS_BMAP_INIT(&free_list, &first_block);
2694 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2695 sip->i_ino, &first_block, &free_list,
2699 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2701 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2703 error = xfs_bumplink(tp, sip);
2709 * If this is a synchronous mount, make sure that the
2710 * link transaction goes to disk before returning to
2713 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2714 xfs_trans_set_sync(tp);
2717 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2719 xfs_bmap_cancel(&free_list);
2723 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2728 /* Fall through to std_return with error = 0. */
2730 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2731 DM_EVENT_POSTLINK)) {
2732 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2733 target_dir_vp, DM_RIGHT_NULL,
2734 src_vp, DM_RIGHT_NULL,
2735 target_name, NULL, 0, error, 0);
2740 cancel_flags |= XFS_TRANS_ABORT;
2744 xfs_trans_cancel(tp, cancel_flags);
2753 bhv_desc_t *dir_bdp,
2759 char *dir_name = VNAME(dentry);
2761 xfs_inode_t *cdp; /* inode of created dir */
2762 vnode_t *cvp; /* vnode of created dir */
2768 xfs_bmap_free_t free_list;
2769 xfs_fsblock_t first_block;
2771 boolean_t dp_joined_to_trans;
2772 boolean_t created = B_FALSE;
2773 int dm_event_sent = 0;
2775 struct xfs_dquot *udqp, *gdqp;
2780 dir_vp = BHV_TO_VNODE(dir_bdp);
2781 dp = XFS_BHVTOI(dir_bdp);
2784 if (XFS_FORCED_SHUTDOWN(mp))
2785 return XFS_ERROR(EIO);
2787 dir_namelen = VNAMELEN(dentry);
2790 dp_joined_to_trans = B_FALSE;
2791 dm_di_mode = vap->va_mode;
2793 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2794 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2795 dir_vp, DM_RIGHT_NULL, NULL,
2796 DM_RIGHT_NULL, dir_name, NULL,
2803 /* Return through std_return after this point. */
2805 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2809 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2810 prid = dp->i_d.di_projid;
2811 else if (vap->va_mask & XFS_AT_PROJID)
2812 prid = (xfs_prid_t)vap->va_projid;
2814 prid = (xfs_prid_t)dfltprid;
2817 * Make sure that we have allocated dquot(s) on disk.
2819 error = XFS_QM_DQVOPALLOC(mp, dp,
2820 current_fsuid(credp), current_fsgid(credp), prid,
2821 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2825 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2826 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2827 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2828 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2829 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2830 if (error == ENOSPC) {
2832 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2833 XFS_TRANS_PERM_LOG_RES,
2834 XFS_MKDIR_LOG_COUNT);
2842 xfs_ilock(dp, XFS_ILOCK_EXCL);
2845 * Check for directory link count overflow.
2847 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2848 error = XFS_ERROR(EMLINK);
2853 * Reserve disk quota and the inode.
2855 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2860 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2863 * create the directory inode.
2865 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2866 0, credp, prid, resblks > 0,
2869 if (error == ENOSPC)
2876 * Now we add the directory inode to the transaction.
2877 * We waited until now since xfs_dir_ialloc might start
2878 * a new transaction. Had we joined the transaction
2879 * earlier, the locks might have gotten released.
2882 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2883 dp_joined_to_trans = B_TRUE;
2885 XFS_BMAP_INIT(&free_list, &first_block);
2887 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2888 cdp->i_ino, &first_block, &free_list,
2889 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2891 ASSERT(error != ENOSPC);
2894 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2897 * Bump the in memory version number of the parent directory
2898 * so that other processes accessing it will recognize that
2899 * the directory has changed.
2903 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2909 error = xfs_bumplink(tp, dp);
2914 cvp = XFS_ITOV(cdp);
2922 * Attach the dquots to the new inode and modify the icount incore.
2924 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2927 * If this is a synchronous mount, make sure that the
2928 * mkdir transaction goes to disk before returning to
2931 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2932 xfs_trans_set_sync(tp);
2935 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2941 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2942 XFS_QM_DQRELE(mp, udqp);
2943 XFS_QM_DQRELE(mp, gdqp);
2948 /* Fall through to std_return with error = 0 or errno from
2949 * xfs_trans_commit. */
2952 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2953 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2954 DM_EVENT_POSTCREATE)) {
2955 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2956 dir_vp, DM_RIGHT_NULL,
2957 created ? XFS_ITOV(cdp):NULL,
2960 dm_di_mode, error, 0);
2966 xfs_bmap_cancel(&free_list);
2968 cancel_flags |= XFS_TRANS_ABORT;
2970 xfs_trans_cancel(tp, cancel_flags);
2971 XFS_QM_DQRELE(mp, udqp);
2972 XFS_QM_DQRELE(mp, gdqp);
2974 if (!dp_joined_to_trans && (dp != NULL)) {
2975 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2988 bhv_desc_t *dir_bdp,
2992 char *name = VNAME(dentry);
2994 xfs_inode_t *cdp; /* child directory */
2998 xfs_bmap_free_t free_list;
2999 xfs_fsblock_t first_block;
3008 dir_vp = BHV_TO_VNODE(dir_bdp);
3009 dp = XFS_BHVTOI(dir_bdp);
3012 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3014 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3015 return XFS_ERROR(EIO);
3016 namelen = VNAMELEN(dentry);
3018 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3019 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3020 dir_vp, DM_RIGHT_NULL,
3021 NULL, DM_RIGHT_NULL,
3022 name, NULL, 0, 0, 0);
3024 return XFS_ERROR(error);
3027 /* Return through std_return after this point. */
3032 * We need to get a reference to cdp before we get our log
3033 * reservation. The reason for this is that we cannot call
3034 * xfs_iget for an inode for which we do not have a reference
3035 * once we've acquired a log reservation. This is because the
3036 * inode we are trying to get might be in xfs_inactive going
3037 * for a log reservation. Since we'll have to wait for the
3038 * inactive code to complete before returning from xfs_iget,
3039 * we need to make sure that we don't have log space reserved
3040 * when we call xfs_iget. Instead we get an unlocked referece
3041 * to the inode before getting our log reservation.
3043 error = xfs_get_dir_entry(dentry, &cdp);
3045 REMOVE_DEBUG_TRACE(__LINE__);
3049 dm_di_mode = cdp->i_d.di_mode;
3052 * Get the dquots for the inodes.
3054 error = XFS_QM_DQATTACH(mp, dp, 0);
3055 if (!error && dp != cdp)
3056 error = XFS_QM_DQATTACH(mp, cdp, 0);
3059 REMOVE_DEBUG_TRACE(__LINE__);
3063 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3064 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3066 * We try to get the real space reservation first,
3067 * allowing for directory btree deletion(s) implying
3068 * possible bmap insert(s). If we can't get the space
3069 * reservation then we use 0 instead, and avoid the bmap
3070 * btree insert(s) in the directory code by, if the bmap
3071 * insert tries to happen, instead trimming the LAST
3072 * block from the directory.
3074 resblks = XFS_REMOVE_SPACE_RES(mp);
3075 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3076 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3077 if (error == ENOSPC) {
3079 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3080 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3083 ASSERT(error != ENOSPC);
3088 XFS_BMAP_INIT(&free_list, &first_block);
3091 * Now lock the child directory inode and the parent directory
3092 * inode in the proper order. This will take care of validating
3093 * that the directory entry for the child directory inode has
3094 * not changed while we were obtaining a log reservation.
3096 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3098 xfs_trans_cancel(tp, cancel_flags);
3103 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3106 * Only increment the parent directory vnode count if
3107 * we didn't bump it in looking up cdp. The only time
3108 * we don't bump it is when we're looking up ".".
3114 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3116 ASSERT(cdp->i_d.di_nlink >= 2);
3117 if (cdp->i_d.di_nlink != 2) {
3118 error = XFS_ERROR(ENOTEMPTY);
3121 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3122 error = XFS_ERROR(ENOTEMPTY);
3126 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3127 &first_block, &free_list, resblks);
3132 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3135 * Bump the in memory generation count on the parent
3136 * directory so that other can know that it has changed.
3141 * Drop the link from cdp's "..".
3143 error = xfs_droplink(tp, dp);
3149 * Drop the link from dp to cdp.
3151 error = xfs_droplink(tp, cdp);
3157 * Drop the "." link from cdp to self.
3159 error = xfs_droplink(tp, cdp);
3164 /* Determine these before committing transaction */
3165 last_cdp_link = (cdp)->i_d.di_nlink==0;
3168 * Take an extra ref on the child vnode so that it
3169 * does not go to xfs_inactive() from within the commit.
3174 * If this is a synchronous mount, make sure that the
3175 * rmdir transaction goes to disk before returning to
3178 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3179 xfs_trans_set_sync(tp);
3182 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3184 xfs_bmap_cancel(&free_list);
3185 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3191 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3199 * Let interposed file systems know about removed links.
3201 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3205 /* Fall through to std_return with error = 0 or the errno
3206 * from xfs_trans_commit. */
3208 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3209 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3210 dir_vp, DM_RIGHT_NULL,
3211 NULL, DM_RIGHT_NULL,
3212 name, NULL, dm_di_mode,
3218 xfs_bmap_cancel(&free_list);
3219 cancel_flags |= XFS_TRANS_ABORT;
3223 xfs_trans_cancel(tp, cancel_flags);
3231 * Read dp's entries starting at uiop->uio_offset and translate them into
3232 * bufsize bytes worth of struct dirents starting at bufbase.
3236 bhv_desc_t *dir_bdp,
3242 xfs_trans_t *tp = NULL;
3246 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3247 (inst_t *)__return_address);
3248 dp = XFS_BHVTOI(dir_bdp);
3250 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3251 return XFS_ERROR(EIO);
3254 lock_mode = xfs_ilock_map_shared(dp);
3255 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3256 xfs_iunlock_map_shared(dp, lock_mode);
3267 bhv_desc_t *dir_bdp,
3280 xfs_bmap_free_t free_list;
3281 xfs_fsblock_t first_block;
3282 boolean_t dp_joined_to_trans;
3286 xfs_fileoff_t first_fsb;
3287 xfs_filblks_t fs_blocks;
3289 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3296 struct xfs_dquot *udqp, *gdqp;
3298 char *link_name = VNAME(dentry);
3302 dir_vp = BHV_TO_VNODE(dir_bdp);
3303 dp = XFS_BHVTOI(dir_bdp);
3304 dp_joined_to_trans = B_FALSE;
3309 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3313 if (XFS_FORCED_SHUTDOWN(mp))
3314 return XFS_ERROR(EIO);
3316 link_namelen = VNAMELEN(dentry);
3319 * Check component lengths of the target path name.
3321 pathlen = strlen(target_path);
3322 if (pathlen >= MAXPATHLEN) /* total string too long */
3323 return XFS_ERROR(ENAMETOOLONG);
3324 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3328 for(total = 0, path = target_path; total < pathlen;) {
3332 while(*path == '/') {
3338 * Count up to the next slash or end of path.
3339 * Error out if the component is bigger than MAXNAMELEN.
3341 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3342 if (++len >= MAXNAMELEN) {
3343 error = ENAMETOOLONG;
3350 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3351 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3352 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3353 link_name, target_path, 0, 0, 0);
3358 /* Return through std_return after this point. */
3361 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3362 prid = dp->i_d.di_projid;
3363 else if (vap->va_mask & XFS_AT_PROJID)
3364 prid = (xfs_prid_t)vap->va_projid;
3366 prid = (xfs_prid_t)dfltprid;
3369 * Make sure that we have allocated dquot(s) on disk.
3371 error = XFS_QM_DQVOPALLOC(mp, dp,
3372 current_fsuid(credp), current_fsgid(credp), prid,
3373 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3377 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3378 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3380 * The symlink will fit into the inode data fork?
3381 * There can't be any attributes so we get the whole variable part.
3383 if (pathlen <= XFS_LITINO(mp))
3386 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3387 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3388 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3389 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3390 if (error == ENOSPC && fs_blocks == 0) {
3392 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3393 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3401 xfs_ilock(dp, XFS_ILOCK_EXCL);
3404 * Check whether the directory allows new symlinks or not.
3406 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3407 error = XFS_ERROR(EPERM);
3412 * Reserve disk quota : blocks and inode.
3414 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3419 * Check for ability to enter directory entry, if no space reserved.
3422 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3425 * Initialize the bmap freelist prior to calling either
3426 * bmapi or the directory create code.
3428 XFS_BMAP_INIT(&free_list, &first_block);
3431 * Allocate an inode for the symlink.
3433 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3434 1, 0, credp, prid, resblks > 0, &ip, NULL);
3436 if (error == ENOSPC)
3443 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3444 dp_joined_to_trans = B_TRUE;
3447 * Also attach the dquot(s) to it, if applicable.
3449 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3452 resblks -= XFS_IALLOC_SPACE_RES(mp);
3454 * If the symlink will fit into the inode, write it inline.
3456 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3457 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3458 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3459 ip->i_d.di_size = pathlen;
3462 * The inode was initially created in extent format.
3464 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3465 ip->i_df.if_flags |= XFS_IFINLINE;
3467 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3468 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3472 nmaps = SYMLINK_MAPS;
3474 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3475 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3476 &first_block, resblks, mval, &nmaps,
3483 resblks -= fs_blocks;
3484 ip->i_d.di_size = pathlen;
3485 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3487 cur_chunk = target_path;
3488 for (n = 0; n < nmaps; n++) {
3489 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3490 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3491 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3492 BTOBB(byte_cnt), 0);
3493 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3494 if (pathlen < byte_cnt) {
3497 pathlen -= byte_cnt;
3499 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3500 cur_chunk += byte_cnt;
3502 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3507 * Create the directory entry for the symlink.
3509 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3510 ip->i_ino, &first_block, &free_list, resblks);
3514 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3515 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3518 * Bump the in memory version number of the parent directory
3519 * so that other processes accessing it will recognize that
3520 * the directory has changed.
3525 * If this is a synchronous mount, make sure that the
3526 * symlink transaction goes to disk before returning to
3529 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3530 xfs_trans_set_sync(tp);
3534 * xfs_trans_commit normally decrements the vnode ref count
3535 * when it unlocks the inode. Since we want to return the
3536 * vnode to the caller, we bump the vnode ref count now.
3540 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3544 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3545 XFS_QM_DQRELE(mp, udqp);
3546 XFS_QM_DQRELE(mp, gdqp);
3548 /* Fall through to std_return with error = 0 or errno from
3549 * xfs_trans_commit */
3551 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3552 DM_EVENT_POSTSYMLINK)) {
3553 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3554 dir_vp, DM_RIGHT_NULL,
3555 error ? NULL : XFS_ITOV(ip),
3556 DM_RIGHT_NULL, link_name, target_path,
3572 xfs_bmap_cancel(&free_list);
3573 cancel_flags |= XFS_TRANS_ABORT;
3575 xfs_trans_cancel(tp, cancel_flags);
3576 XFS_QM_DQRELE(mp, udqp);
3577 XFS_QM_DQRELE(mp, gdqp);
3579 if (!dp_joined_to_trans && (dp != NULL)) {
3580 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3590 * A fid routine that takes a pointer to a previously allocated
3591 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3601 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3602 (inst_t *)__return_address);
3603 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3605 xfid = (xfs_fid2_t *)fidp;
3606 ip = XFS_BHVTOI(bdp);
3607 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3610 * use memcpy because the inode is a long long and there's no
3611 * assurance that xfid->fid_ino is properly aligned.
3613 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3614 xfid->fid_gen = ip->i_d.di_gen;
3631 vp = BHV_TO_VNODE(bdp);
3634 ip = XFS_BHVTOI(bdp);
3635 if (locktype == VRWLOCK_WRITE) {
3636 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3637 } else if (locktype == VRWLOCK_TRY_READ) {
3638 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3639 } else if (locktype == VRWLOCK_TRY_WRITE) {
3640 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3642 ASSERT((locktype == VRWLOCK_READ) ||
3643 (locktype == VRWLOCK_WRITE_DIRECT));
3644 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3662 vp = BHV_TO_VNODE(bdp);
3665 ip = XFS_BHVTOI(bdp);
3666 if (locktype == VRWLOCK_WRITE) {
3668 * In the write case, we may have added a new entry to
3669 * the reference cache. This might store a pointer to
3670 * an inode to be released in this inode. If it is there,
3671 * clear the pointer and release the inode after unlocking
3674 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3676 ASSERT((locktype == VRWLOCK_READ) ||
3677 (locktype == VRWLOCK_WRITE_DIRECT));
3678 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3690 xfs_inode_log_item_t *iip;
3693 ip = XFS_BHVTOI(bdp);
3697 if (XFS_FORCED_SHUTDOWN(mp))
3698 return XFS_ERROR(EIO);
3701 * Bypass inodes which have already been cleaned by
3702 * the inode flush clustering code inside xfs_iflush
3704 if ((ip->i_update_core == 0) &&
3705 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3708 if (flags & FLUSH_LOG) {
3709 if (iip && iip->ili_last_lsn) {
3710 xlog_t *log = mp->m_log;
3712 int s, log_flags = XFS_LOG_FORCE;
3714 s = GRANT_LOCK(log);
3715 sync_lsn = log->l_last_sync_lsn;
3716 GRANT_UNLOCK(log, s);
3718 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3721 if (flags & FLUSH_SYNC)
3722 log_flags |= XFS_LOG_SYNC;
3723 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3728 * We make this non-blocking if the inode is contended,
3729 * return EAGAIN to indicate to the caller that they
3730 * did not succeed. This prevents the flush path from
3731 * blocking on inodes inside another operation right
3732 * now, they get caught later by xfs_sync.
3734 if (flags & FLUSH_INODE) {
3737 if (xfs_ipincount(ip))
3740 if (flags & FLUSH_SYNC) {
3741 xfs_ilock(ip, XFS_ILOCK_SHARED);
3743 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3744 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3745 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3752 if (flags & FLUSH_SYNC)
3753 flush_flags = XFS_IFLUSH_SYNC;
3755 flush_flags = XFS_IFLUSH_ASYNC;
3757 error = xfs_iflush(ip, flush_flags);
3758 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3777 if (!capable(CAP_SYS_ADMIN))
3778 return XFS_ERROR(EPERM);
3780 ip = XFS_BHVTOI(bdp);
3783 if (XFS_FORCED_SHUTDOWN(mp))
3784 return XFS_ERROR(EIO);
3786 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3787 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3789 xfs_trans_cancel(tp, 0);
3792 xfs_ilock(ip, XFS_ILOCK_EXCL);
3793 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3795 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3796 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3798 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3800 error = xfs_trans_commit(tp, 0, NULL);
3816 vp = BHV_TO_VNODE(bdp);
3817 ip = XFS_BHVTOI(bdp);
3819 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3821 ASSERT(!VN_MAPPED(vp));
3823 /* bad inode, get out here ASAP */
3831 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3834 * Make sure the atime in the XFS inode is correct before freeing the
3837 xfs_synchronize_atime(ip);
3839 /* If we have nothing to flush with this inode then complete the
3840 * teardown now, otherwise break the link between the xfs inode
3841 * and the linux inode and clean up the xfs inode later. This
3842 * avoids flushing the inode to disk during the delete operation
3845 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3846 xfs_ilock(ip, XFS_ILOCK_EXCL);
3848 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3850 xfs_mount_t *mp = ip->i_mount;
3852 /* Protect sync from us */
3853 XFS_MOUNT_ILOCK(mp);
3854 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3855 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3856 ip->i_flags |= XFS_IRECLAIMABLE;
3857 XFS_MOUNT_IUNLOCK(mp);
3868 xfs_ihash_t *ih = ip->i_hash;
3869 vnode_t *vp = XFS_ITOV_NULL(ip);
3872 if (vp && VN_BAD(vp))
3875 /* The hash lock here protects a thread in xfs_iget_core from
3876 * racing with us on linking the inode back with a vnode.
3877 * Once we have the XFS_IRECLAIM flag set it will not touch
3880 write_lock(&ih->ih_lock);
3881 if ((ip->i_flags & XFS_IRECLAIM) ||
3882 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3883 write_unlock(&ih->ih_lock);
3886 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3890 ip->i_flags |= XFS_IRECLAIM;
3891 write_unlock(&ih->ih_lock);
3894 * If the inode is still dirty, then flush it out. If the inode
3895 * is not in the AIL, then it will be OK to flush it delwri as
3896 * long as xfs_iflush() does not keep any references to the inode.
3897 * We leave that decision up to xfs_iflush() since it has the
3898 * knowledge of whether it's OK to simply do a delwri flush of
3899 * the inode or whether we need to wait until the inode is
3900 * pulled from the AIL.
3901 * We get the flush lock regardless, though, just to make sure
3902 * we don't free it while it is being flushed.
3904 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3906 xfs_ilock(ip, XFS_ILOCK_EXCL);
3910 if (ip->i_update_core ||
3911 ((ip->i_itemp != NULL) &&
3912 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3913 error = xfs_iflush(ip, sync_mode);
3915 * If we hit an error, typically because of filesystem
3916 * shutdown, we don't need to let vn_reclaim to know
3917 * because we're gonna reclaim the inode anyway.
3920 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3923 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3926 ASSERT(ip->i_update_core == 0);
3927 ASSERT(ip->i_itemp == NULL ||
3928 ip->i_itemp->ili_format.ilf_fields == 0);
3929 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3930 } else if (locked) {
3932 * We are not interested in doing an iflush if we're
3933 * in the process of shutting down the filesystem forcibly.
3934 * So, just reclaim the inode.
3937 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3946 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3949 xfs_inode_t *ip, *n;
3954 XFS_MOUNT_ILOCK(mp);
3955 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3957 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3959 if (xfs_ipincount(ip) ||
3960 !xfs_iflock_nowait(ip)) {
3961 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3965 XFS_MOUNT_IUNLOCK(mp);
3966 if (xfs_finish_reclaim(ip, noblock,
3967 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3976 XFS_MOUNT_IUNLOCK(mp);
3981 * xfs_alloc_file_space()
3982 * This routine allocates disk space for the given file.
3984 * If alloc_type == 0, this request is for an ALLOCSP type
3985 * request which will change the file size. In this case, no
3986 * DMAPI event will be generated by the call. A TRUNCATE event
3987 * will be generated later by xfs_setattr.
3989 * If alloc_type != 0, this request is for a RESVSP type
3990 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3991 * lower block boundary byte address is less than the file's
4000 xfs_alloc_file_space(
4007 xfs_mount_t *mp = ip->i_mount;
4009 xfs_filblks_t allocated_fsb;
4010 xfs_filblks_t allocatesize_fsb;
4011 xfs_extlen_t extsz, temp;
4012 xfs_fileoff_t startoffset_fsb;
4013 xfs_fsblock_t firstfsb;
4019 xfs_bmbt_irec_t imaps[1], *imapp;
4020 xfs_bmap_free_t free_list;
4021 uint qblocks, resblks, resrtextents;
4025 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4027 if (XFS_FORCED_SHUTDOWN(mp))
4028 return XFS_ERROR(EIO);
4030 rt = XFS_IS_REALTIME_INODE(ip);
4032 if (!(extsz = ip->i_d.di_extsize))
4033 extsz = mp->m_sb.sb_rextsize;
4035 extsz = ip->i_d.di_extsize;
4038 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4042 return XFS_ERROR(EINVAL);
4048 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4049 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4050 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4052 /* Generate a DMAPI event if needed. */
4053 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4054 (attr_flags&ATTR_DMI) == 0 &&
4055 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4056 xfs_off_t end_dmi_offset;
4058 end_dmi_offset = offset+len;
4059 if (end_dmi_offset > ip->i_d.di_size)
4060 end_dmi_offset = ip->i_d.di_size;
4061 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4062 offset, end_dmi_offset - offset,
4069 * Allocate file space until done or until there is an error
4072 while (allocatesize_fsb && !error) {
4076 * Determine space reservations for data/realtime.
4078 if (unlikely(extsz)) {
4079 s = startoffset_fsb;
4082 e = startoffset_fsb + allocatesize_fsb;
4083 if ((temp = do_mod(startoffset_fsb, extsz)))
4085 if ((temp = do_mod(e, extsz)))
4089 e = allocatesize_fsb;
4093 resrtextents = qblocks = (uint)(e - s);
4094 resrtextents /= mp->m_sb.sb_rextsize;
4095 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4096 quota_flag = XFS_QMOPT_RES_RTBLKS;
4099 resblks = qblocks = \
4100 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4101 quota_flag = XFS_QMOPT_RES_REGBLKS;
4105 * Allocate and setup the transaction.
4107 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4108 error = xfs_trans_reserve(tp, resblks,
4109 XFS_WRITE_LOG_RES(mp), resrtextents,
4110 XFS_TRANS_PERM_LOG_RES,
4111 XFS_WRITE_LOG_COUNT);
4113 * Check for running out of space
4117 * Free the transaction structure.
4119 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4120 xfs_trans_cancel(tp, 0);
4123 xfs_ilock(ip, XFS_ILOCK_EXCL);
4124 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4125 qblocks, 0, quota_flag);
4129 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4130 xfs_trans_ihold(tp, ip);
4133 * Issue the xfs_bmapi() call to allocate the blocks
4135 XFS_BMAP_INIT(&free_list, &firstfsb);
4136 error = xfs_bmapi(tp, ip, startoffset_fsb,
4137 allocatesize_fsb, bmapi_flag,
4138 &firstfsb, 0, imapp, &nimaps,
4145 * Complete the transaction
4147 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4152 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4153 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4158 allocated_fsb = imapp->br_blockcount;
4161 error = XFS_ERROR(ENOSPC);
4165 startoffset_fsb += allocated_fsb;
4166 allocatesize_fsb -= allocated_fsb;
4169 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4170 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4172 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4173 XFS_ITOV(ip), DM_RIGHT_NULL,
4174 XFS_ITOV(ip), DM_RIGHT_NULL,
4175 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4177 goto retry; /* Maybe DMAPI app. has made space */
4178 /* else fall through with error from XFS_SEND_DATA */
4183 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4184 xfs_bmap_cancel(&free_list);
4185 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4187 error1: /* Just cancel transaction */
4188 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4189 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4190 goto dmapi_enospc_check;
4194 * Zero file bytes between startoff and endoff inclusive.
4195 * The iolock is held exclusive and no blocks are buffered.
4198 xfs_zero_remaining_bytes(
4203 xfs_bmbt_irec_t imap;
4204 xfs_fileoff_t offset_fsb;
4205 xfs_off_t lastoffset;
4208 xfs_mount_t *mp = ip->i_mount;
4212 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4213 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4214 mp->m_rtdev_targp : mp->m_ddev_targp);
4216 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4217 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4219 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4221 if (error || nimap < 1)
4223 ASSERT(imap.br_blockcount >= 1);
4224 ASSERT(imap.br_startoff == offset_fsb);
4225 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4226 if (lastoffset > endoff)
4227 lastoffset = endoff;
4228 if (imap.br_startblock == HOLESTARTBLOCK)
4230 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4231 if (imap.br_state == XFS_EXT_UNWRITTEN)
4234 XFS_BUF_UNWRITE(bp);
4236 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4238 if ((error = xfs_iowait(bp))) {
4239 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4240 mp, bp, XFS_BUF_ADDR(bp));
4243 memset(XFS_BUF_PTR(bp) +
4244 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4245 0, lastoffset - offset + 1);
4250 if ((error = xfs_iowait(bp))) {
4251 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4252 mp, bp, XFS_BUF_ADDR(bp));
4261 * xfs_free_file_space()
4262 * This routine frees disk space for the given file.
4264 * This routine is only called by xfs_change_file_space
4265 * for an UNRESVSP type call.
4273 xfs_free_file_space(
4282 xfs_off_t end_dmi_offset;
4283 xfs_fileoff_t endoffset_fsb;
4285 xfs_fsblock_t firstfsb;
4286 xfs_bmap_free_t free_list;
4288 xfs_bmbt_irec_t imap;
4296 xfs_fileoff_t startoffset_fsb;
4298 int need_iolock = 1;
4303 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4305 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4309 if (len <= 0) /* if nothing being freed */
4311 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4312 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4313 end_dmi_offset = offset + len;
4314 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4316 if (offset < ip->i_d.di_size &&
4317 (attr_flags & ATTR_DMI) == 0 &&
4318 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4319 if (end_dmi_offset > ip->i_d.di_size)
4320 end_dmi_offset = ip->i_d.di_size;
4321 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4322 offset, end_dmi_offset - offset,
4323 AT_DELAY_FLAG(attr_flags), NULL);
4328 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4329 if (attr_flags & ATTR_NOLOCK)
4332 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4334 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4336 ilen = len + (offset & (rounding - 1));
4337 ioffset = offset & ~(rounding - 1);
4338 if (ilen & (rounding - 1))
4339 ilen = (ilen + rounding) & ~(rounding - 1);
4341 if (VN_CACHED(vp) != 0) {
4342 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4343 ctooff(offtoct(ioffset)), -1);
4344 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4345 -1, FI_REMAPF_LOCKED);
4349 * Need to zero the stuff we're not freeing, on disk.
4350 * If its a realtime file & can't use unwritten extents then we
4351 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4352 * will take care of it for us.
4354 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4356 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4357 &imap, &nimap, NULL);
4359 goto out_unlock_iolock;
4360 ASSERT(nimap == 0 || nimap == 1);
4361 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4364 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4365 block = imap.br_startblock;
4366 mod = do_div(block, mp->m_sb.sb_rextsize);
4368 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4371 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4372 &imap, &nimap, NULL);
4374 goto out_unlock_iolock;
4375 ASSERT(nimap == 0 || nimap == 1);
4376 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4377 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4379 if (mod && (mod != mp->m_sb.sb_rextsize))
4380 endoffset_fsb -= mod;
4383 if ((done = (endoffset_fsb <= startoffset_fsb)))
4385 * One contiguous piece to clear
4387 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4390 * Some full blocks, possibly two pieces to clear
4392 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4393 error = xfs_zero_remaining_bytes(ip, offset,
4394 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4396 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4397 error = xfs_zero_remaining_bytes(ip,
4398 XFS_FSB_TO_B(mp, endoffset_fsb),
4403 * free file space until done or until there is an error
4405 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4406 while (!error && !done) {
4409 * allocate and setup the transaction
4411 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4412 error = xfs_trans_reserve(tp,
4414 XFS_WRITE_LOG_RES(mp),
4416 XFS_TRANS_PERM_LOG_RES,
4417 XFS_WRITE_LOG_COUNT);
4420 * check for running out of space
4424 * Free the transaction structure.
4426 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4427 xfs_trans_cancel(tp, 0);
4430 xfs_ilock(ip, XFS_ILOCK_EXCL);
4431 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4432 ip->i_udquot, ip->i_gdquot, resblks, 0,
4433 XFS_QMOPT_RES_REGBLKS);
4437 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4438 xfs_trans_ihold(tp, ip);
4441 * issue the bunmapi() call to free the blocks
4443 XFS_BMAP_INIT(&free_list, &firstfsb);
4444 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4445 endoffset_fsb - startoffset_fsb,
4446 0, 2, &firstfsb, &free_list, &done);
4452 * complete the transaction
4454 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4459 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4460 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4465 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4469 xfs_bmap_cancel(&free_list);
4471 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4472 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4478 * xfs_change_file_space()
4479 * This routine allocates or frees disk space for the given file.
4480 * The user specified parameters are checked for alignment and size
4489 xfs_change_file_space(
4503 xfs_off_t startoffset;
4509 vp = BHV_TO_VNODE(bdp);
4510 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4512 ip = XFS_BHVTOI(bdp);
4516 * must be a regular file and have write permission
4519 return XFS_ERROR(EINVAL);
4521 xfs_ilock(ip, XFS_ILOCK_SHARED);
4523 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4524 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4528 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4530 switch (bf->l_whence) {
4531 case 0: /*SEEK_SET*/
4533 case 1: /*SEEK_CUR*/
4534 bf->l_start += offset;
4536 case 2: /*SEEK_END*/
4537 bf->l_start += ip->i_d.di_size;
4540 return XFS_ERROR(EINVAL);
4543 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4545 if ( (bf->l_start < 0)
4546 || (bf->l_start > XFS_MAXIOFFSET(mp))
4547 || (bf->l_start + llen < 0)
4548 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4549 return XFS_ERROR(EINVAL);
4553 startoffset = bf->l_start;
4554 fsize = ip->i_d.di_size;
4557 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4559 * These calls do NOT zero the data space allocated to the file,
4560 * nor do they change the file size.
4562 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4564 * These calls cause the new file data to be zeroed and the file
4565 * size to be changed.
4567 setprealloc = clrprealloc = 0;
4570 case XFS_IOC_RESVSP:
4571 case XFS_IOC_RESVSP64:
4572 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4579 case XFS_IOC_UNRESVSP:
4580 case XFS_IOC_UNRESVSP64:
4581 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4586 case XFS_IOC_ALLOCSP:
4587 case XFS_IOC_ALLOCSP64:
4588 case XFS_IOC_FREESP:
4589 case XFS_IOC_FREESP64:
4590 if (startoffset > fsize) {
4591 error = xfs_alloc_file_space(ip, fsize,
4592 startoffset - fsize, 0, attr_flags);
4597 va.va_mask = XFS_AT_SIZE;
4598 va.va_size = startoffset;
4600 error = xfs_setattr(bdp, &va, attr_flags, credp);
4610 return XFS_ERROR(EINVAL);
4614 * update the inode timestamp, mode, and prealloc flag bits
4616 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4618 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4621 xfs_trans_cancel(tp, 0);
4625 xfs_ilock(ip, XFS_ILOCK_EXCL);
4627 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4628 xfs_trans_ihold(tp, ip);
4630 if ((attr_flags & ATTR_DMI) == 0) {
4631 ip->i_d.di_mode &= ~S_ISUID;
4634 * Note that we don't have to worry about mandatory
4635 * file locking being disabled here because we only
4636 * clear the S_ISGID bit if the Group execute bit is
4637 * on, but if it was on then mandatory locking wouldn't
4638 * have been enabled.
4640 if (ip->i_d.di_mode & S_IXGRP)
4641 ip->i_d.di_mode &= ~S_ISGID;
4643 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4646 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4647 else if (clrprealloc)
4648 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4650 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4651 xfs_trans_set_sync(tp);
4653 error = xfs_trans_commit(tp, 0, NULL);
4655 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4660 vnodeops_t xfs_vnodeops = {
4661 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4662 .vop_open = xfs_open,
4663 .vop_read = xfs_read,
4664 #ifdef HAVE_SENDFILE
4665 .vop_sendfile = xfs_sendfile,
4667 .vop_write = xfs_write,
4668 .vop_ioctl = xfs_ioctl,
4669 .vop_getattr = xfs_getattr,
4670 .vop_setattr = xfs_setattr,
4671 .vop_access = xfs_access,
4672 .vop_lookup = xfs_lookup,
4673 .vop_create = xfs_create,
4674 .vop_remove = xfs_remove,
4675 .vop_link = xfs_link,
4676 .vop_rename = xfs_rename,
4677 .vop_mkdir = xfs_mkdir,
4678 .vop_rmdir = xfs_rmdir,
4679 .vop_readdir = xfs_readdir,
4680 .vop_symlink = xfs_symlink,
4681 .vop_readlink = xfs_readlink,
4682 .vop_fsync = xfs_fsync,
4683 .vop_inactive = xfs_inactive,
4684 .vop_fid2 = xfs_fid2,
4685 .vop_rwlock = xfs_rwlock,
4686 .vop_rwunlock = xfs_rwunlock,
4687 .vop_bmap = xfs_bmap,
4688 .vop_reclaim = xfs_reclaim,
4689 .vop_attr_get = xfs_attr_get,
4690 .vop_attr_set = xfs_attr_set,
4691 .vop_attr_remove = xfs_attr_remove,
4692 .vop_attr_list = xfs_attr_list,
4693 .vop_link_removed = (vop_link_removed_t)fs_noval,
4694 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4695 .vop_tosspages = fs_tosspages,
4696 .vop_flushinval_pages = fs_flushinval_pages,
4697 .vop_flush_pages = fs_flush_pages,
4698 .vop_release = xfs_release,
4699 .vop_iflush = xfs_inode_flush,