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);
626 vn_iowait(vp); /* wait for the completion of any pending DIOs */
628 code = xfs_itruncate_data(ip, vap->va_size);
631 lock_flags &= ~XFS_ILOCK_EXCL;
632 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
635 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
636 if ((code = xfs_trans_reserve(tp, 0,
637 XFS_ITRUNCATE_LOG_RES(mp), 0,
638 XFS_TRANS_PERM_LOG_RES,
639 XFS_ITRUNCATE_LOG_COUNT))) {
640 xfs_trans_cancel(tp, 0);
642 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
645 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
646 xfs_ilock(ip, XFS_ILOCK_EXCL);
650 xfs_trans_ijoin(tp, ip, lock_flags);
651 xfs_trans_ihold(tp, ip);
654 /* determine whether mandatory locking mode changes */
655 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
658 * Truncate file. Must have write permission and not be a directory.
660 if (mask & XFS_AT_SIZE) {
661 if (vap->va_size > ip->i_d.di_size) {
662 xfs_igrow_finish(tp, ip, vap->va_size,
663 !(flags & ATTR_DMI));
664 } else if ((vap->va_size <= ip->i_d.di_size) ||
665 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
667 * signal a sync transaction unless
668 * we're truncating an already unlinked
669 * file on a wsync filesystem
671 code = xfs_itruncate_finish(&tp, ip,
672 (xfs_fsize_t)vap->va_size,
674 ((ip->i_d.di_nlink != 0 ||
675 !(mp->m_flags & XFS_MOUNT_WSYNC))
682 * Have to do this even if the file's size doesn't change.
684 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
688 * Change file access modes.
690 if (mask & XFS_AT_MODE) {
691 ip->i_d.di_mode &= S_IFMT;
692 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
694 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
695 timeflags |= XFS_ICHGTIME_CHG;
699 * Change file ownership. Must be the owner or privileged.
700 * If the system was configured with the "restricted_chown"
701 * option, the owner is not permitted to give away the file,
702 * and can change the group id only to a group of which he
703 * or she is a member.
705 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_XID|XFS_AT_PROJID)) {
707 * CAP_FSETID overrides the following restrictions:
709 * The set-user-ID and set-group-ID bits of a file will be
710 * cleared upon successful return from chown()
712 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
713 !capable(CAP_FSETID)) {
714 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
718 * Change the ownerships and register quota modifications
719 * in the transaction.
722 if (XFS_IS_GQUOTA_ON(mp)) {
723 /* FIXME: handle xid quota? */
725 ip->i_d.di_xid = xid;
728 if (XFS_IS_UQUOTA_ON(mp)) {
729 ASSERT(mask & XFS_AT_UID);
731 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
732 &ip->i_udquot, udqp);
734 ip->i_d.di_uid = uid;
737 if (XFS_IS_GQUOTA_ON(mp)) {
738 ASSERT(!XFS_IS_PQUOTA_ON(mp));
739 ASSERT(mask & XFS_AT_GID);
741 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
742 &ip->i_gdquot, gdqp);
744 ip->i_d.di_gid = gid;
746 if (iprojid != projid) {
747 if (XFS_IS_PQUOTA_ON(mp)) {
748 ASSERT(!XFS_IS_GQUOTA_ON(mp));
749 ASSERT(mask & XFS_AT_PROJID);
751 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
752 &ip->i_gdquot, gdqp);
754 ip->i_d.di_projid = projid;
756 * We may have to rev the inode as well as
757 * the superblock version number since projids didn't
758 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
760 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
761 xfs_bump_ino_vers2(tp, ip);
764 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
765 timeflags |= XFS_ICHGTIME_CHG;
770 * Change file access or modified times.
772 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
773 if (mask & XFS_AT_ATIME) {
774 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
775 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
776 ip->i_update_core = 1;
777 timeflags &= ~XFS_ICHGTIME_ACC;
779 if (mask & XFS_AT_MTIME) {
780 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
781 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
782 timeflags &= ~XFS_ICHGTIME_MOD;
783 timeflags |= XFS_ICHGTIME_CHG;
785 if (tp && (flags & ATTR_UTIME))
786 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
790 * Change XFS-added attributes.
792 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
793 if (mask & XFS_AT_EXTSIZE) {
795 * Converting bytes to fs blocks.
797 ip->i_d.di_extsize = vap->va_extsize >>
798 mp->m_sb.sb_blocklog;
800 if (mask & XFS_AT_XFLAGS) {
803 /* can't set PREALLOC this way, just preserve it */
804 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
805 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
806 di_flags |= XFS_DIFLAG_IMMUTABLE;
807 if (vap->va_xflags & XFS_XFLAG_IUNLINK)
808 di_flags |= XFS_DIFLAG_IUNLINK;
809 if (vap->va_xflags & XFS_XFLAG_BARRIER)
810 di_flags |= XFS_DIFLAG_BARRIER;
811 if (vap->va_xflags & XFS_XFLAG_APPEND)
812 di_flags |= XFS_DIFLAG_APPEND;
813 if (vap->va_xflags & XFS_XFLAG_SYNC)
814 di_flags |= XFS_DIFLAG_SYNC;
815 if (vap->va_xflags & XFS_XFLAG_NOATIME)
816 di_flags |= XFS_DIFLAG_NOATIME;
817 if (vap->va_xflags & XFS_XFLAG_NODUMP)
818 di_flags |= XFS_DIFLAG_NODUMP;
819 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
820 di_flags |= XFS_DIFLAG_PROJINHERIT;
821 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
822 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
823 di_flags |= XFS_DIFLAG_RTINHERIT;
824 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
825 di_flags |= XFS_DIFLAG_NOSYMLINKS;
826 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
827 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
828 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
829 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
830 di_flags |= XFS_DIFLAG_REALTIME;
831 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
833 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
835 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
836 di_flags |= XFS_DIFLAG_EXTSIZE;
838 ip->i_d.di_flags = di_flags;
840 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
841 timeflags |= XFS_ICHGTIME_CHG;
845 * Change file inode change time only if XFS_AT_CTIME set
846 * AND we have been called by a DMI function.
849 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
850 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
851 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
852 ip->i_update_core = 1;
853 timeflags &= ~XFS_ICHGTIME_CHG;
857 * Send out timestamp changes that need to be set to the
858 * current time. Not done when called by a DMI function.
860 if (timeflags && !(flags & ATTR_DMI))
861 xfs_ichgtime(ip, timeflags);
863 XFS_STATS_INC(xs_ig_attrchg);
866 * If this is a synchronous mount, make sure that the
867 * transaction goes to disk before returning to the user.
868 * This is slightly sub-optimal in that truncates require
869 * two sync transactions instead of one for wsync filesystems.
870 * One for the truncate and one for the timestamps since we
871 * don't want to change the timestamps unless we're sure the
872 * truncate worked. Truncates are less than 1% of the laddis
873 * mix so this probably isn't worth the trouble to optimize.
877 if (mp->m_flags & XFS_MOUNT_WSYNC)
878 xfs_trans_set_sync(tp);
880 code = xfs_trans_commit(tp, commit_flags, NULL);
884 * If the (regular) file's mandatory locking mode changed, then
885 * notify the vnode. We do this under the inode lock to prevent
886 * racing calls to vop_vnode_change.
888 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
889 if (mandlock_before != mandlock_after) {
890 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
894 xfs_iunlock(ip, lock_flags);
897 * Release any dquot(s) the inode had kept before chown.
899 XFS_QM_DQRELE(mp, olddquot1);
900 XFS_QM_DQRELE(mp, olddquot2);
901 XFS_QM_DQRELE(mp, udqp);
902 XFS_QM_DQRELE(mp, gdqp);
908 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
909 !(flags & ATTR_DMI)) {
910 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
911 NULL, DM_RIGHT_NULL, NULL, NULL,
912 0, 0, AT_DELAY_FLAG(flags));
917 commit_flags |= XFS_TRANS_ABORT;
920 XFS_QM_DQRELE(mp, udqp);
921 XFS_QM_DQRELE(mp, gdqp);
923 xfs_trans_cancel(tp, commit_flags);
925 if (lock_flags != 0) {
926 xfs_iunlock(ip, lock_flags);
934 * Null conversion from vnode mode bits to inode mode bits, as in efs.
945 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
946 (inst_t *)__return_address);
948 ip = XFS_BHVTOI(bdp);
949 xfs_ilock(ip, XFS_ILOCK_SHARED);
950 error = xfs_iaccess(ip, mode, credp);
951 xfs_iunlock(ip, XFS_ILOCK_SHARED);
975 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
981 vp = BHV_TO_VNODE(bdp);
982 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
984 ip = XFS_BHVTOI(bdp);
987 if (XFS_FORCED_SHUTDOWN(mp))
988 return XFS_ERROR(EIO);
990 xfs_ilock(ip, XFS_ILOCK_SHARED);
992 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
994 offset = uiop->uio_offset;
995 count = uiop->uio_resid;
998 error = XFS_ERROR(EINVAL);
1007 * See if the symlink is stored inline.
1009 pathlen = (int)ip->i_d.di_size;
1011 if (ip->i_df.if_flags & XFS_IFINLINE) {
1012 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1016 * Symlink not inline. Call bmap to get it in.
1018 nmaps = SYMLINK_MAPS;
1020 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1021 0, NULL, 0, mval, &nmaps, NULL);
1027 for (n = 0; n < nmaps; n++) {
1028 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1029 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1030 bp = xfs_buf_read(mp->m_ddev_targp, d,
1031 BTOBB(byte_cnt), 0);
1032 error = XFS_BUF_GETERROR(bp);
1034 xfs_ioerror_alert("xfs_readlink",
1035 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1039 if (pathlen < byte_cnt)
1041 pathlen -= byte_cnt;
1043 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1050 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1058 * This is called to sync the inode and its data out to disk.
1059 * We need to hold the I/O lock while flushing the data, and
1060 * the inode lock while flushing the inode. The inode lock CANNOT
1061 * be held while flushing the data, so acquire after we're done
1075 int log_flushed = 0, changed = 1;
1077 vn_trace_entry(BHV_TO_VNODE(bdp),
1078 __FUNCTION__, (inst_t *)__return_address);
1080 ip = XFS_BHVTOI(bdp);
1082 ASSERT(start >= 0 && stop >= -1);
1084 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1085 return XFS_ERROR(EIO);
1088 * We always need to make sure that the required inode state
1089 * is safe on disk. The vnode might be clean but because
1090 * of committed transactions that haven't hit the disk yet.
1091 * Likewise, there could be unflushed non-transactional
1092 * changes to the inode core that have to go to disk.
1094 * The following code depends on one assumption: that
1095 * any transaction that changes an inode logs the core
1096 * because it has to change some field in the inode core
1097 * (typically nextents or nblocks). That assumption
1098 * implies that any transactions against an inode will
1099 * catch any non-transactional updates. If inode-altering
1100 * transactions exist that violate this assumption, the
1101 * code breaks. Right now, it figures that if the involved
1102 * update_* field is clear and the inode is unpinned, the
1103 * inode is clean. Either it's been flushed or it's been
1104 * committed and the commit has hit the disk unpinning the inode.
1105 * (Note that xfs_inode_item_format() called at commit clears
1106 * the update_* fields.)
1108 xfs_ilock(ip, XFS_ILOCK_SHARED);
1110 /* If we are flushing data then we care about update_size
1111 * being set, otherwise we care about update_core
1113 if ((flag & FSYNC_DATA) ?
1114 (ip->i_update_size == 0) :
1115 (ip->i_update_core == 0)) {
1117 * Timestamps/size haven't changed since last inode
1118 * flush or inode transaction commit. That means
1119 * either nothing got written or a transaction
1120 * committed which caught the updates. If the
1121 * latter happened and the transaction hasn't
1122 * hit the disk yet, the inode will be still
1123 * be pinned. If it is, force the log.
1126 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1128 if (xfs_ipincount(ip)) {
1129 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1131 ((flag & FSYNC_WAIT)
1132 ? XFS_LOG_SYNC : 0),
1136 * If the inode is not pinned and nothing
1137 * has changed we don't need to flush the
1145 * Kick off a transaction to log the inode
1146 * core to get the updates. Make it
1147 * sync if FSYNC_WAIT is passed in (which
1148 * is done by everybody but specfs). The
1149 * sync transaction will also force the log.
1151 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1152 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1153 if ((error = xfs_trans_reserve(tp, 0,
1154 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1156 xfs_trans_cancel(tp, 0);
1159 xfs_ilock(ip, XFS_ILOCK_EXCL);
1162 * Note - it's possible that we might have pushed
1163 * ourselves out of the way during trans_reserve
1164 * which would flush the inode. But there's no
1165 * guarantee that the inode buffer has actually
1166 * gone out yet (it's delwri). Plus the buffer
1167 * could be pinned anyway if it's part of an
1168 * inode in another recent transaction. So we
1169 * play it safe and fire off the transaction anyway.
1171 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1172 xfs_trans_ihold(tp, ip);
1173 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1174 if (flag & FSYNC_WAIT)
1175 xfs_trans_set_sync(tp);
1176 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1178 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1181 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1183 * If the log write didn't issue an ordered tag we need
1184 * to flush the disk cache for the data device now.
1187 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1190 * If this inode is on the RT dev we need to flush that
1193 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1194 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1201 * This is called by xfs_inactive to free any blocks beyond eof,
1202 * when the link count isn't zero.
1205 xfs_inactive_free_eofblocks(
1211 xfs_fileoff_t end_fsb;
1212 xfs_fileoff_t last_fsb;
1213 xfs_filblks_t map_len;
1215 xfs_bmbt_irec_t imap;
1218 * Figure out if there are any blocks beyond the end
1219 * of the file. If not, then there is nothing to do.
1221 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1222 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1223 map_len = last_fsb - end_fsb;
1228 xfs_ilock(ip, XFS_ILOCK_SHARED);
1229 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1230 NULL, 0, &imap, &nimaps, NULL);
1231 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1233 if (!error && (nimaps != 0) &&
1234 (imap.br_startblock != HOLESTARTBLOCK ||
1235 ip->i_delayed_blks)) {
1237 * Attach the dquots to the inode up front.
1239 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1243 * There are blocks after the end of file.
1244 * Free them up now by truncating the file to
1247 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1250 * Do the xfs_itruncate_start() call before
1251 * reserving any log space because
1252 * itruncate_start will call into the buffer
1253 * cache and we can't
1254 * do that within a transaction.
1256 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1257 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1260 error = xfs_trans_reserve(tp, 0,
1261 XFS_ITRUNCATE_LOG_RES(mp),
1262 0, XFS_TRANS_PERM_LOG_RES,
1263 XFS_ITRUNCATE_LOG_COUNT);
1265 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1266 xfs_trans_cancel(tp, 0);
1267 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1271 xfs_ilock(ip, XFS_ILOCK_EXCL);
1272 xfs_trans_ijoin(tp, ip,
1275 xfs_trans_ihold(tp, ip);
1277 error = xfs_itruncate_finish(&tp, ip,
1282 * If we get an error at this point we
1283 * simply don't bother truncating the file.
1286 xfs_trans_cancel(tp,
1287 (XFS_TRANS_RELEASE_LOG_RES |
1290 error = xfs_trans_commit(tp,
1291 XFS_TRANS_RELEASE_LOG_RES,
1294 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1300 * Free a symlink that has blocks associated with it.
1303 xfs_inactive_symlink_rmt(
1311 xfs_fsblock_t first_block;
1312 xfs_bmap_free_t free_list;
1315 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1323 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1325 * We're freeing a symlink that has some
1326 * blocks allocated to it. Free the
1327 * blocks here. We know that we've got
1328 * either 1 or 2 extents and that we can
1329 * free them all in one bunmapi call.
1331 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1332 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1333 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1334 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1335 xfs_trans_cancel(tp, 0);
1340 * Lock the inode, fix the size, and join it to the transaction.
1341 * Hold it so in the normal path, we still have it locked for
1342 * the second transaction. In the error paths we need it
1343 * held so the cancel won't rele it, see below.
1345 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1346 size = (int)ip->i_d.di_size;
1347 ip->i_d.di_size = 0;
1348 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1349 xfs_trans_ihold(tp, ip);
1350 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1352 * Find the block(s) so we can inval and unmap them.
1355 XFS_BMAP_INIT(&free_list, &first_block);
1356 nmaps = ARRAY_SIZE(mval);
1357 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1358 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1362 * Invalidate the block(s).
1364 for (i = 0; i < nmaps; i++) {
1365 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1366 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1367 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1368 xfs_trans_binval(tp, bp);
1371 * Unmap the dead block(s) to the free_list.
1373 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1374 &first_block, &free_list, &done)))
1378 * Commit the first transaction. This logs the EFI and the inode.
1380 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1383 * The transaction must have been committed, since there were
1384 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1385 * The new tp has the extent freeing and EFDs.
1389 * The first xact was committed, so add the inode to the new one.
1390 * Mark it dirty so it will be logged and moved forward in the log as
1391 * part of every commit.
1393 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1394 xfs_trans_ihold(tp, ip);
1395 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1397 * Get a new, empty transaction to return to our caller.
1399 ntp = xfs_trans_dup(tp);
1401 * Commit the transaction containing extent freeing and EFDs.
1402 * If we get an error on the commit here or on the reserve below,
1403 * we need to unlock the inode since the new transaction doesn't
1404 * have the inode attached.
1406 error = xfs_trans_commit(tp, 0, NULL);
1409 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1413 * Remove the memory for extent descriptions (just bookkeeping).
1415 if (ip->i_df.if_bytes)
1416 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1417 ASSERT(ip->i_df.if_bytes == 0);
1419 * Put an itruncate log reservation in the new transaction
1422 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1423 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1424 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1428 * Return with the inode locked but not joined to the transaction.
1434 xfs_bmap_cancel(&free_list);
1437 * Have to come here with the inode locked and either
1438 * (held and in the transaction) or (not in the transaction).
1439 * If the inode isn't held then cancel would iput it, but
1440 * that's wrong since this is inactive and the vnode ref
1441 * count is 0 already.
1442 * Cancel won't do anything to the inode if held, but it still
1443 * needs to be locked until the cancel is done, if it was
1444 * joined to the transaction.
1446 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1447 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1454 xfs_inactive_symlink_local(
1460 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1462 * We're freeing a symlink which fit into
1463 * the inode. Just free the memory used
1464 * to hold the old symlink.
1466 error = xfs_trans_reserve(*tpp, 0,
1467 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1468 0, XFS_TRANS_PERM_LOG_RES,
1469 XFS_ITRUNCATE_LOG_COUNT);
1472 xfs_trans_cancel(*tpp, 0);
1476 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1479 * Zero length symlinks _can_ exist.
1481 if (ip->i_df.if_bytes > 0) {
1482 xfs_idata_realloc(ip,
1483 -(ip->i_df.if_bytes),
1485 ASSERT(ip->i_df.if_bytes == 0);
1502 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1505 ASSERT(ip->i_d.di_forkoff != 0);
1506 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1507 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1509 error = xfs_attr_inactive(ip);
1512 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1513 return error; /* goto out */
1516 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1517 error = xfs_trans_reserve(tp, 0,
1518 XFS_IFREE_LOG_RES(mp),
1519 0, XFS_TRANS_PERM_LOG_RES,
1520 XFS_INACTIVE_LOG_COUNT);
1522 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1523 xfs_trans_cancel(tp, 0);
1525 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1529 xfs_ilock(ip, XFS_ILOCK_EXCL);
1530 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1531 xfs_trans_ihold(tp, ip);
1532 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1534 ASSERT(ip->i_d.di_anextents == 0);
1549 vp = BHV_TO_VNODE(bdp);
1550 ip = XFS_BHVTOI(bdp);
1552 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1556 /* If this is a read-only mount, don't do this (would generate I/O) */
1557 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1560 #ifdef HAVE_REFCACHE
1561 /* If we are in the NFS reference cache then don't do this now */
1568 if (ip->i_d.di_nlink != 0) {
1569 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1570 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1571 ip->i_delayed_blks > 0)) &&
1572 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1573 (!(ip->i_d.di_flags &
1574 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1575 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1577 /* Update linux inode block count after free above */
1578 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1579 ip->i_d.di_nblocks + ip->i_delayed_blks);
1589 * This is called when the vnode reference count for the vnode
1590 * goes to zero. If the file has been unlinked, then it must
1591 * now be truncated. Also, we clear all of the read-ahead state
1592 * kept for the inode here since the file is now closed.
1601 xfs_bmap_free_t free_list;
1602 xfs_fsblock_t first_block;
1609 vp = BHV_TO_VNODE(bdp);
1610 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1612 ip = XFS_BHVTOI(bdp);
1615 * If the inode is already free, then there can be nothing
1618 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1619 ASSERT(ip->i_df.if_real_bytes == 0);
1620 ASSERT(ip->i_df.if_broot_bytes == 0);
1621 return VN_INACTIVE_CACHE;
1625 * Only do a truncate if it's a regular file with
1626 * some actual space in it. It's OK to look at the
1627 * inode's fields without the lock because we're the
1628 * only one with a reference to the inode.
1630 truncate = ((ip->i_d.di_nlink == 0) &&
1631 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1632 (ip->i_delayed_blks > 0)) &&
1633 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1637 if (ip->i_d.di_nlink == 0 &&
1638 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1639 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1644 /* If this is a read-only mount, don't do this (would generate I/O) */
1645 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1648 if (ip->i_d.di_nlink != 0) {
1649 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1650 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1651 ip->i_delayed_blks > 0)) &&
1652 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1653 (!(ip->i_d.di_flags &
1654 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1655 (ip->i_delayed_blks != 0)))) {
1656 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1657 return VN_INACTIVE_CACHE;
1658 /* Update linux inode block count after free above */
1659 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1660 ip->i_d.di_nblocks + ip->i_delayed_blks);
1665 ASSERT(ip->i_d.di_nlink == 0);
1667 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1668 return VN_INACTIVE_CACHE;
1670 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1673 * Do the xfs_itruncate_start() call before
1674 * reserving any log space because itruncate_start
1675 * will call into the buffer cache and we can't
1676 * do that within a transaction.
1678 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1680 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1682 error = xfs_trans_reserve(tp, 0,
1683 XFS_ITRUNCATE_LOG_RES(mp),
1684 0, XFS_TRANS_PERM_LOG_RES,
1685 XFS_ITRUNCATE_LOG_COUNT);
1687 /* Don't call itruncate_cleanup */
1688 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1689 xfs_trans_cancel(tp, 0);
1690 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1691 return VN_INACTIVE_CACHE;
1694 xfs_ilock(ip, XFS_ILOCK_EXCL);
1695 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1696 xfs_trans_ihold(tp, ip);
1699 * normally, we have to run xfs_itruncate_finish sync.
1700 * But if filesystem is wsync and we're in the inactive
1701 * path, then we know that nlink == 0, and that the
1702 * xaction that made nlink == 0 is permanently committed
1703 * since xfs_remove runs as a synchronous transaction.
1705 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1706 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1709 xfs_trans_cancel(tp,
1710 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1711 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1712 return VN_INACTIVE_CACHE;
1714 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1717 * If we get an error while cleaning up a
1718 * symlink we bail out.
1720 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1721 xfs_inactive_symlink_rmt(ip, &tp) :
1722 xfs_inactive_symlink_local(ip, &tp);
1726 return VN_INACTIVE_CACHE;
1729 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1730 xfs_trans_ihold(tp, ip);
1732 error = xfs_trans_reserve(tp, 0,
1733 XFS_IFREE_LOG_RES(mp),
1734 0, XFS_TRANS_PERM_LOG_RES,
1735 XFS_INACTIVE_LOG_COUNT);
1737 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1738 xfs_trans_cancel(tp, 0);
1739 return VN_INACTIVE_CACHE;
1742 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1743 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1744 xfs_trans_ihold(tp, ip);
1748 * If there are attributes associated with the file
1749 * then blow them away now. The code calls a routine
1750 * that recursively deconstructs the attribute fork.
1751 * We need to just commit the current transaction
1752 * because we can't use it for xfs_attr_inactive().
1754 if (ip->i_d.di_anextents > 0) {
1755 error = xfs_inactive_attrs(ip, &tp);
1757 * If we got an error, the transaction is already
1758 * cancelled, and the inode is unlocked. Just get out.
1761 return VN_INACTIVE_CACHE;
1762 } else if (ip->i_afp) {
1763 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1769 XFS_BMAP_INIT(&free_list, &first_block);
1770 error = xfs_ifree(tp, ip, &free_list);
1773 * If we fail to free the inode, shut down. The cancel
1774 * might do that, we need to make sure. Otherwise the
1775 * inode might be lost for a long time or forever.
1777 if (!XFS_FORCED_SHUTDOWN(mp)) {
1779 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1780 error, mp->m_fsname);
1781 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1783 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1786 * Credit the quota account(s). The inode is gone.
1788 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1791 * Just ignore errors at this point. There is
1792 * nothing we can do except to try to keep going.
1794 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1796 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1799 * Release the dquots held by inode, if any.
1801 XFS_QM_DQDETACH(mp, ip);
1803 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1806 return VN_INACTIVE_CACHE;
1815 bhv_desc_t *dir_bdp,
1822 xfs_inode_t *dp, *ip;
1828 dir_vp = BHV_TO_VNODE(dir_bdp);
1829 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1831 dp = XFS_BHVTOI(dir_bdp);
1833 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1834 return XFS_ERROR(EIO);
1836 lock_mode = xfs_ilock_map_shared(dp);
1837 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1839 *vpp = XFS_ITOV(ip);
1842 xfs_iunlock_map_shared(dp, lock_mode);
1848 * xfs_create (create a new file).
1852 bhv_desc_t *dir_bdp,
1858 char *name = VNAME(dentry);
1860 xfs_inode_t *dp, *ip;
1866 xfs_bmap_free_t free_list;
1867 xfs_fsblock_t first_block;
1868 boolean_t dp_joined_to_trans;
1869 int dm_event_sent = 0;
1873 struct xfs_dquot *udqp, *gdqp;
1879 dir_vp = BHV_TO_VNODE(dir_bdp);
1880 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1882 dp = XFS_BHVTOI(dir_bdp);
1885 dm_di_mode = vap->va_mode;
1886 namelen = VNAMELEN(dentry);
1888 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1889 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1890 dir_vp, DM_RIGHT_NULL, NULL,
1891 DM_RIGHT_NULL, name, NULL,
1899 if (XFS_FORCED_SHUTDOWN(mp))
1900 return XFS_ERROR(EIO);
1902 /* Return through std_return after this point. */
1905 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1906 prid = dp->i_d.di_projid;
1907 else if (vap->va_mask & XFS_AT_PROJID)
1908 prid = (xfs_prid_t)vap->va_projid;
1910 prid = (xfs_prid_t)dfltprid;
1913 * Make sure that we have allocated dquot(s) on disk.
1915 error = XFS_QM_DQVOPALLOC(mp, dp,
1916 current_fsuid(credp), current_fsgid(credp), prid,
1917 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1922 dp_joined_to_trans = B_FALSE;
1924 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1925 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1926 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1928 * Initially assume that the file does not exist and
1929 * reserve the resources for that case. If that is not
1930 * the case we'll drop the one we have and get a more
1931 * appropriate transaction later.
1933 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1934 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1935 if (error == ENOSPC) {
1937 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1938 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1946 xfs_ilock(dp, XFS_ILOCK_EXCL);
1948 XFS_BMAP_INIT(&free_list, &first_block);
1953 * Reserve disk quota and the inode.
1955 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1960 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1962 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1963 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1964 rdev, credp, prid, resblks > 0,
1967 if (error == ENOSPC)
1974 * At this point, we've gotten a newly allocated inode.
1975 * It is locked (and joined to the transaction).
1978 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1981 * Now we join the directory inode to the transaction.
1982 * We do not do it earlier because xfs_dir_ialloc
1983 * might commit the previous transaction (and release
1988 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1989 dp_joined_to_trans = B_TRUE;
1991 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1992 &first_block, &free_list,
1993 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1995 ASSERT(error != ENOSPC);
1998 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1999 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2002 * If this is a synchronous mount, make sure that the
2003 * create transaction goes to disk before returning to
2006 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2007 xfs_trans_set_sync(tp);
2013 * Attach the dquot(s) to the inodes and modify them incore.
2014 * These ids of the inode couldn't have changed since the new
2015 * inode has been locked ever since it was created.
2017 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2020 * xfs_trans_commit normally decrements the vnode ref count
2021 * when it unlocks the inode. Since we want to return the
2022 * vnode to the caller, we bump the vnode ref count now.
2027 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2029 xfs_bmap_cancel(&free_list);
2033 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2040 XFS_QM_DQRELE(mp, udqp);
2041 XFS_QM_DQRELE(mp, gdqp);
2044 * Propagate the fact that the vnode changed after the
2045 * xfs_inode locks have been released.
2047 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2051 /* Fallthrough to std_return with error = 0 */
2054 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2055 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2056 DM_EVENT_POSTCREATE)) {
2057 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2058 dir_vp, DM_RIGHT_NULL,
2060 DM_RIGHT_NULL, name, NULL,
2061 dm_di_mode, error, 0);
2066 cancel_flags |= XFS_TRANS_ABORT;
2071 xfs_trans_cancel(tp, cancel_flags);
2073 if (!dp_joined_to_trans && (dp != NULL))
2074 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2075 XFS_QM_DQRELE(mp, udqp);
2076 XFS_QM_DQRELE(mp, gdqp);
2082 * Wait until after the current transaction is aborted to
2083 * release the inode. This prevents recursive transactions
2084 * and deadlocks from xfs_inactive.
2086 cancel_flags |= XFS_TRANS_ABORT;
2087 xfs_trans_cancel(tp, cancel_flags);
2090 XFS_QM_DQRELE(mp, udqp);
2091 XFS_QM_DQRELE(mp, gdqp);
2098 * Some counters to see if (and how often) we are hitting some deadlock
2099 * prevention code paths.
2103 int xfs_rm_lock_delays;
2104 int xfs_rm_attempts;
2108 * The following routine will lock the inodes associated with the
2109 * directory and the named entry in the directory. The locks are
2110 * acquired in increasing inode number.
2112 * If the entry is "..", then only the directory is locked. The
2113 * vnode ref count will still include that from the .. entry in
2116 * There is a deadlock we need to worry about. If the locked directory is
2117 * in the AIL, it might be blocking up the log. The next inode we lock
2118 * could be already locked by another thread waiting for log space (e.g
2119 * a permanent log reservation with a long running transaction (see
2120 * xfs_itruncate_finish)). To solve this, we must check if the directory
2121 * is in the ail and use lock_nowait. If we can't lock, we need to
2122 * drop the inode lock on the directory and try again. xfs_iunlock will
2123 * potentially push the tail if we were holding up the log.
2126 xfs_lock_dir_and_entry(
2129 xfs_inode_t *ip) /* inode of entry 'name' */
2133 xfs_inode_t *ips[2];
2142 xfs_ilock(dp, XFS_ILOCK_EXCL);
2149 * We want to lock in increasing inum. Since we've already
2150 * acquired the lock on the directory, we may need to release
2151 * if if the inum of the entry turns out to be less.
2153 if (e_inum > dp->i_ino) {
2155 * We are already in the right order, so just
2156 * lock on the inode of the entry.
2157 * We need to use nowait if dp is in the AIL.
2160 lp = (xfs_log_item_t *)dp->i_itemp;
2161 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2162 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2169 * Unlock dp and try again.
2170 * xfs_iunlock will try to push the tail
2171 * if the inode is in the AIL.
2174 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2176 if ((attempts % 5) == 0) {
2177 delay(1); /* Don't just spin the CPU */
2179 xfs_rm_lock_delays++;
2185 xfs_ilock(ip, XFS_ILOCK_EXCL);
2187 } else if (e_inum < dp->i_ino) {
2188 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2192 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2194 /* else e_inum == dp->i_ino */
2195 /* This can happen if we're asked to lock /x/..
2196 * the entry is "..", which is also the parent directory.
2204 int xfs_small_retries;
2205 int xfs_middle_retries;
2206 int xfs_lots_retries;
2207 int xfs_lock_delays;
2211 * The following routine will lock n inodes in exclusive mode.
2212 * We assume the caller calls us with the inodes in i_ino order.
2214 * We need to detect deadlock where an inode that we lock
2215 * is in the AIL and we start waiting for another inode that is locked
2216 * by a thread in a long running transaction (such as truncate). This can
2217 * result in deadlock since the long running trans might need to wait
2218 * for the inode we just locked in order to push the tail and free space
2228 int attempts = 0, i, j, try_lock;
2231 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2242 for (; i < inodes; i++) {
2245 if (i && (ips[i] == ips[i-1])) /* Already locked */
2249 * If try_lock is not set yet, make sure all locked inodes
2250 * are not in the AIL.
2251 * If any are, set try_lock to be used later.
2255 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2256 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2257 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2264 * If any of the previous locks we have locked is in the AIL,
2265 * we must TRY to get the second and subsequent locks. If
2266 * we can't get any, we must release all we have
2271 /* try_lock must be 0 if i is 0. */
2273 * try_lock means we have an inode locked
2274 * that is in the AIL.
2277 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2281 * Unlock all previous guys and try again.
2282 * xfs_iunlock will try to push the tail
2283 * if the inode is in the AIL.
2286 for(j = i - 1; j >= 0; j--) {
2289 * Check to see if we've already
2290 * unlocked this one.
2291 * Not the first one going back,
2292 * and the inode ptr is the same.
2294 if ((j != (i - 1)) && ips[j] ==
2298 xfs_iunlock(ips[j], lock_mode);
2301 if ((attempts % 5) == 0) {
2302 delay(1); /* Don't just spin the CPU */
2312 xfs_ilock(ips[i], lock_mode);
2318 if (attempts < 5) xfs_small_retries++;
2319 else if (attempts < 100) xfs_middle_retries++;
2320 else xfs_lots_retries++;
2328 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2329 int remove_which_error_return = 0;
2331 #define REMOVE_DEBUG_TRACE(x)
2332 #endif /* ! DEBUG */
2341 bhv_desc_t *dir_bdp,
2346 char *name = VNAME(dentry);
2347 xfs_inode_t *dp, *ip;
2348 xfs_trans_t *tp = NULL;
2351 xfs_bmap_free_t free_list;
2352 xfs_fsblock_t first_block;
2360 dir_vp = BHV_TO_VNODE(dir_bdp);
2361 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2363 dp = XFS_BHVTOI(dir_bdp);
2366 if (XFS_FORCED_SHUTDOWN(mp))
2367 return XFS_ERROR(EIO);
2369 namelen = VNAMELEN(dentry);
2371 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2372 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2373 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2374 name, NULL, 0, 0, 0);
2379 /* From this point on, return through std_return */
2383 * We need to get a reference to ip before we get our log
2384 * reservation. The reason for this is that we cannot call
2385 * xfs_iget for an inode for which we do not have a reference
2386 * once we've acquired a log reservation. This is because the
2387 * inode we are trying to get might be in xfs_inactive going
2388 * for a log reservation. Since we'll have to wait for the
2389 * inactive code to complete before returning from xfs_iget,
2390 * we need to make sure that we don't have log space reserved
2391 * when we call xfs_iget. Instead we get an unlocked reference
2392 * to the inode before getting our log reservation.
2394 error = xfs_get_dir_entry(dentry, &ip);
2396 REMOVE_DEBUG_TRACE(__LINE__);
2400 dm_di_mode = ip->i_d.di_mode;
2402 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2406 error = XFS_QM_DQATTACH(mp, dp, 0);
2407 if (!error && dp != ip)
2408 error = XFS_QM_DQATTACH(mp, ip, 0);
2410 REMOVE_DEBUG_TRACE(__LINE__);
2415 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2416 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2418 * We try to get the real space reservation first,
2419 * allowing for directory btree deletion(s) implying
2420 * possible bmap insert(s). If we can't get the space
2421 * reservation then we use 0 instead, and avoid the bmap
2422 * btree insert(s) in the directory code by, if the bmap
2423 * insert tries to happen, instead trimming the LAST
2424 * block from the directory.
2426 resblks = XFS_REMOVE_SPACE_RES(mp);
2427 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2428 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2429 if (error == ENOSPC) {
2431 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2432 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2435 ASSERT(error != ENOSPC);
2436 REMOVE_DEBUG_TRACE(__LINE__);
2437 xfs_trans_cancel(tp, 0);
2442 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2444 REMOVE_DEBUG_TRACE(__LINE__);
2445 xfs_trans_cancel(tp, cancel_flags);
2451 * At this point, we've gotten both the directory and the entry
2454 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2457 * Increment vnode ref count only in this case since
2458 * there's an extra vnode reference in the case where
2462 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2466 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2468 XFS_BMAP_INIT(&free_list, &first_block);
2469 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2470 &first_block, &free_list, 0);
2472 ASSERT(error != ENOENT);
2473 REMOVE_DEBUG_TRACE(__LINE__);
2476 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2479 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2481 error = xfs_droplink(tp, ip);
2483 REMOVE_DEBUG_TRACE(__LINE__);
2487 /* Determine if this is the last link while
2488 * we are in the transaction.
2490 link_zero = (ip)->i_d.di_nlink==0;
2493 * Take an extra ref on the inode so that it doesn't
2494 * go to xfs_inactive() from within the commit.
2499 * If this is a synchronous mount, make sure that the
2500 * remove transaction goes to disk before returning to
2503 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2504 xfs_trans_set_sync(tp);
2507 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2509 REMOVE_DEBUG_TRACE(__LINE__);
2513 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2520 * Before we drop our extra reference to the inode, purge it
2521 * from the refcache if it is there. By waiting until afterwards
2522 * to do the IRELE, we ensure that we won't go inactive in the
2523 * xfs_refcache_purge_ip routine (although that would be OK).
2525 xfs_refcache_purge_ip(ip);
2527 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2530 * Let interposed file systems know about removed links.
2532 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2536 /* Fall through to std_return with error = 0 */
2538 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2539 DM_EVENT_POSTREMOVE)) {
2540 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2541 dir_vp, DM_RIGHT_NULL,
2542 NULL, DM_RIGHT_NULL,
2543 name, NULL, dm_di_mode, error, 0);
2548 xfs_bmap_cancel(&free_list);
2549 cancel_flags |= XFS_TRANS_ABORT;
2550 xfs_trans_cancel(tp, cancel_flags);
2555 * In this case make sure to not release the inode until after
2556 * the current transaction is aborted. Releasing it beforehand
2557 * can cause us to go to xfs_inactive and start a recursive
2558 * transaction which can easily deadlock with the current one.
2560 xfs_bmap_cancel(&free_list);
2561 cancel_flags |= XFS_TRANS_ABORT;
2562 xfs_trans_cancel(tp, cancel_flags);
2565 * Before we drop our extra reference to the inode, purge it
2566 * from the refcache if it is there. By waiting until afterwards
2567 * to do the IRELE, we ensure that we won't go inactive in the
2568 * xfs_refcache_purge_ip routine (although that would be OK).
2570 xfs_refcache_purge_ip(ip);
2584 bhv_desc_t *target_dir_bdp,
2589 xfs_inode_t *tdp, *sip;
2592 xfs_inode_t *ips[2];
2594 xfs_bmap_free_t free_list;
2595 xfs_fsblock_t first_block;
2598 vnode_t *target_dir_vp;
2600 char *target_name = VNAME(dentry);
2603 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2604 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2605 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2607 target_namelen = VNAMELEN(dentry);
2608 if (VN_ISDIR(src_vp))
2609 return XFS_ERROR(EPERM);
2611 sip = xfs_vtoi(src_vp);
2612 tdp = XFS_BHVTOI(target_dir_bdp);
2614 if (XFS_FORCED_SHUTDOWN(mp))
2615 return XFS_ERROR(EIO);
2617 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2618 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2619 target_dir_vp, DM_RIGHT_NULL,
2620 src_vp, DM_RIGHT_NULL,
2621 target_name, NULL, 0, 0, 0);
2626 /* Return through std_return after this point. */
2628 error = XFS_QM_DQATTACH(mp, sip, 0);
2629 if (!error && sip != tdp)
2630 error = XFS_QM_DQATTACH(mp, tdp, 0);
2634 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2635 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2636 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2637 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2638 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2639 if (error == ENOSPC) {
2641 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2642 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2649 if (sip->i_ino < tdp->i_ino) {
2657 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2660 * Increment vnode ref counts since xfs_trans_commit &
2661 * xfs_trans_cancel will both unlock the inodes and
2662 * decrement the associated ref counts.
2665 VN_HOLD(target_dir_vp);
2666 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2667 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2670 * If the source has too many links, we can't make any more to it.
2672 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2673 error = XFS_ERROR(EMLINK);
2678 * If we are using project inheritance, we only allow hard link
2679 * creation in our tree when the project IDs are the same; else
2680 * the tree quota mechanism could be circumvented.
2682 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2683 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2684 error = XFS_ERROR(EXDEV);
2689 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2693 XFS_BMAP_INIT(&free_list, &first_block);
2695 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2696 sip->i_ino, &first_block, &free_list,
2700 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2702 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2704 error = xfs_bumplink(tp, sip);
2710 * If this is a synchronous mount, make sure that the
2711 * link transaction goes to disk before returning to
2714 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2715 xfs_trans_set_sync(tp);
2718 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2720 xfs_bmap_cancel(&free_list);
2724 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2729 /* Fall through to std_return with error = 0. */
2731 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2732 DM_EVENT_POSTLINK)) {
2733 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2734 target_dir_vp, DM_RIGHT_NULL,
2735 src_vp, DM_RIGHT_NULL,
2736 target_name, NULL, 0, error, 0);
2741 cancel_flags |= XFS_TRANS_ABORT;
2745 xfs_trans_cancel(tp, cancel_flags);
2754 bhv_desc_t *dir_bdp,
2760 char *dir_name = VNAME(dentry);
2762 xfs_inode_t *cdp; /* inode of created dir */
2763 vnode_t *cvp; /* vnode of created dir */
2769 xfs_bmap_free_t free_list;
2770 xfs_fsblock_t first_block;
2772 boolean_t dp_joined_to_trans;
2773 boolean_t created = B_FALSE;
2774 int dm_event_sent = 0;
2776 struct xfs_dquot *udqp, *gdqp;
2781 dir_vp = BHV_TO_VNODE(dir_bdp);
2782 dp = XFS_BHVTOI(dir_bdp);
2785 if (XFS_FORCED_SHUTDOWN(mp))
2786 return XFS_ERROR(EIO);
2788 dir_namelen = VNAMELEN(dentry);
2791 dp_joined_to_trans = B_FALSE;
2792 dm_di_mode = vap->va_mode;
2794 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2795 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2796 dir_vp, DM_RIGHT_NULL, NULL,
2797 DM_RIGHT_NULL, dir_name, NULL,
2804 /* Return through std_return after this point. */
2806 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2810 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2811 prid = dp->i_d.di_projid;
2812 else if (vap->va_mask & XFS_AT_PROJID)
2813 prid = (xfs_prid_t)vap->va_projid;
2815 prid = (xfs_prid_t)dfltprid;
2818 * Make sure that we have allocated dquot(s) on disk.
2820 error = XFS_QM_DQVOPALLOC(mp, dp,
2821 current_fsuid(credp), current_fsgid(credp), prid,
2822 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2826 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2827 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2828 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2829 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2830 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2831 if (error == ENOSPC) {
2833 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2834 XFS_TRANS_PERM_LOG_RES,
2835 XFS_MKDIR_LOG_COUNT);
2843 xfs_ilock(dp, XFS_ILOCK_EXCL);
2846 * Check for directory link count overflow.
2848 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2849 error = XFS_ERROR(EMLINK);
2854 * Reserve disk quota and the inode.
2856 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2861 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2864 * create the directory inode.
2866 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2867 0, credp, prid, resblks > 0,
2870 if (error == ENOSPC)
2877 * Now we add the directory inode to the transaction.
2878 * We waited until now since xfs_dir_ialloc might start
2879 * a new transaction. Had we joined the transaction
2880 * earlier, the locks might have gotten released.
2883 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2884 dp_joined_to_trans = B_TRUE;
2886 XFS_BMAP_INIT(&free_list, &first_block);
2888 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2889 cdp->i_ino, &first_block, &free_list,
2890 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2892 ASSERT(error != ENOSPC);
2895 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2898 * Bump the in memory version number of the parent directory
2899 * so that other processes accessing it will recognize that
2900 * the directory has changed.
2904 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2910 error = xfs_bumplink(tp, dp);
2915 cvp = XFS_ITOV(cdp);
2923 * Attach the dquots to the new inode and modify the icount incore.
2925 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2928 * If this is a synchronous mount, make sure that the
2929 * mkdir transaction goes to disk before returning to
2932 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2933 xfs_trans_set_sync(tp);
2936 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2942 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2943 XFS_QM_DQRELE(mp, udqp);
2944 XFS_QM_DQRELE(mp, gdqp);
2949 /* Fall through to std_return with error = 0 or errno from
2950 * xfs_trans_commit. */
2953 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2954 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2955 DM_EVENT_POSTCREATE)) {
2956 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2957 dir_vp, DM_RIGHT_NULL,
2958 created ? XFS_ITOV(cdp):NULL,
2961 dm_di_mode, error, 0);
2967 xfs_bmap_cancel(&free_list);
2969 cancel_flags |= XFS_TRANS_ABORT;
2971 xfs_trans_cancel(tp, cancel_flags);
2972 XFS_QM_DQRELE(mp, udqp);
2973 XFS_QM_DQRELE(mp, gdqp);
2975 if (!dp_joined_to_trans && (dp != NULL)) {
2976 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2989 bhv_desc_t *dir_bdp,
2993 char *name = VNAME(dentry);
2995 xfs_inode_t *cdp; /* child directory */
2999 xfs_bmap_free_t free_list;
3000 xfs_fsblock_t first_block;
3009 dir_vp = BHV_TO_VNODE(dir_bdp);
3010 dp = XFS_BHVTOI(dir_bdp);
3013 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3015 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3016 return XFS_ERROR(EIO);
3017 namelen = VNAMELEN(dentry);
3019 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3020 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3021 dir_vp, DM_RIGHT_NULL,
3022 NULL, DM_RIGHT_NULL,
3023 name, NULL, 0, 0, 0);
3025 return XFS_ERROR(error);
3028 /* Return through std_return after this point. */
3033 * We need to get a reference to cdp before we get our log
3034 * reservation. The reason for this is that we cannot call
3035 * xfs_iget for an inode for which we do not have a reference
3036 * once we've acquired a log reservation. This is because the
3037 * inode we are trying to get might be in xfs_inactive going
3038 * for a log reservation. Since we'll have to wait for the
3039 * inactive code to complete before returning from xfs_iget,
3040 * we need to make sure that we don't have log space reserved
3041 * when we call xfs_iget. Instead we get an unlocked reference
3042 * to the inode before getting our log reservation.
3044 error = xfs_get_dir_entry(dentry, &cdp);
3046 REMOVE_DEBUG_TRACE(__LINE__);
3050 dm_di_mode = cdp->i_d.di_mode;
3053 * Get the dquots for the inodes.
3055 error = XFS_QM_DQATTACH(mp, dp, 0);
3056 if (!error && dp != cdp)
3057 error = XFS_QM_DQATTACH(mp, cdp, 0);
3060 REMOVE_DEBUG_TRACE(__LINE__);
3064 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3065 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3067 * We try to get the real space reservation first,
3068 * allowing for directory btree deletion(s) implying
3069 * possible bmap insert(s). If we can't get the space
3070 * reservation then we use 0 instead, and avoid the bmap
3071 * btree insert(s) in the directory code by, if the bmap
3072 * insert tries to happen, instead trimming the LAST
3073 * block from the directory.
3075 resblks = XFS_REMOVE_SPACE_RES(mp);
3076 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3077 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3078 if (error == ENOSPC) {
3080 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3081 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3084 ASSERT(error != ENOSPC);
3089 XFS_BMAP_INIT(&free_list, &first_block);
3092 * Now lock the child directory inode and the parent directory
3093 * inode in the proper order. This will take care of validating
3094 * that the directory entry for the child directory inode has
3095 * not changed while we were obtaining a log reservation.
3097 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3099 xfs_trans_cancel(tp, cancel_flags);
3104 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3107 * Only increment the parent directory vnode count if
3108 * we didn't bump it in looking up cdp. The only time
3109 * we don't bump it is when we're looking up ".".
3115 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3117 ASSERT(cdp->i_d.di_nlink >= 2);
3118 if (cdp->i_d.di_nlink != 2) {
3119 error = XFS_ERROR(ENOTEMPTY);
3122 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3123 error = XFS_ERROR(ENOTEMPTY);
3127 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3128 &first_block, &free_list, resblks);
3133 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3136 * Bump the in memory generation count on the parent
3137 * directory so that other can know that it has changed.
3142 * Drop the link from cdp's "..".
3144 error = xfs_droplink(tp, dp);
3150 * Drop the link from dp to cdp.
3152 error = xfs_droplink(tp, cdp);
3158 * Drop the "." link from cdp to self.
3160 error = xfs_droplink(tp, cdp);
3165 /* Determine these before committing transaction */
3166 last_cdp_link = (cdp)->i_d.di_nlink==0;
3169 * Take an extra ref on the child vnode so that it
3170 * does not go to xfs_inactive() from within the commit.
3175 * If this is a synchronous mount, make sure that the
3176 * rmdir transaction goes to disk before returning to
3179 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3180 xfs_trans_set_sync(tp);
3183 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3185 xfs_bmap_cancel(&free_list);
3186 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3192 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3200 * Let interposed file systems know about removed links.
3202 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3206 /* Fall through to std_return with error = 0 or the errno
3207 * from xfs_trans_commit. */
3209 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3210 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3211 dir_vp, DM_RIGHT_NULL,
3212 NULL, DM_RIGHT_NULL,
3213 name, NULL, dm_di_mode,
3219 xfs_bmap_cancel(&free_list);
3220 cancel_flags |= XFS_TRANS_ABORT;
3224 xfs_trans_cancel(tp, cancel_flags);
3232 * Read dp's entries starting at uiop->uio_offset and translate them into
3233 * bufsize bytes worth of struct dirents starting at bufbase.
3237 bhv_desc_t *dir_bdp,
3243 xfs_trans_t *tp = NULL;
3247 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3248 (inst_t *)__return_address);
3249 dp = XFS_BHVTOI(dir_bdp);
3251 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3252 return XFS_ERROR(EIO);
3255 lock_mode = xfs_ilock_map_shared(dp);
3256 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3257 xfs_iunlock_map_shared(dp, lock_mode);
3268 bhv_desc_t *dir_bdp,
3281 xfs_bmap_free_t free_list;
3282 xfs_fsblock_t first_block;
3283 boolean_t dp_joined_to_trans;
3287 xfs_fileoff_t first_fsb;
3288 xfs_filblks_t fs_blocks;
3290 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3297 struct xfs_dquot *udqp, *gdqp;
3299 char *link_name = VNAME(dentry);
3303 dir_vp = BHV_TO_VNODE(dir_bdp);
3304 dp = XFS_BHVTOI(dir_bdp);
3305 dp_joined_to_trans = B_FALSE;
3310 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3314 if (XFS_FORCED_SHUTDOWN(mp))
3315 return XFS_ERROR(EIO);
3317 link_namelen = VNAMELEN(dentry);
3320 * Check component lengths of the target path name.
3322 pathlen = strlen(target_path);
3323 if (pathlen >= MAXPATHLEN) /* total string too long */
3324 return XFS_ERROR(ENAMETOOLONG);
3325 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3329 for(total = 0, path = target_path; total < pathlen;) {
3333 while(*path == '/') {
3339 * Count up to the next slash or end of path.
3340 * Error out if the component is bigger than MAXNAMELEN.
3342 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3343 if (++len >= MAXNAMELEN) {
3344 error = ENAMETOOLONG;
3351 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3352 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3353 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3354 link_name, target_path, 0, 0, 0);
3359 /* Return through std_return after this point. */
3362 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3363 prid = dp->i_d.di_projid;
3364 else if (vap->va_mask & XFS_AT_PROJID)
3365 prid = (xfs_prid_t)vap->va_projid;
3367 prid = (xfs_prid_t)dfltprid;
3370 * Make sure that we have allocated dquot(s) on disk.
3372 error = XFS_QM_DQVOPALLOC(mp, dp,
3373 current_fsuid(credp), current_fsgid(credp), prid,
3374 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3378 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3379 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3381 * The symlink will fit into the inode data fork?
3382 * There can't be any attributes so we get the whole variable part.
3384 if (pathlen <= XFS_LITINO(mp))
3387 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3388 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3389 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3390 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3391 if (error == ENOSPC && fs_blocks == 0) {
3393 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3394 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3402 xfs_ilock(dp, XFS_ILOCK_EXCL);
3405 * Check whether the directory allows new symlinks or not.
3407 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3408 error = XFS_ERROR(EPERM);
3413 * Reserve disk quota : blocks and inode.
3415 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3420 * Check for ability to enter directory entry, if no space reserved.
3423 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3426 * Initialize the bmap freelist prior to calling either
3427 * bmapi or the directory create code.
3429 XFS_BMAP_INIT(&free_list, &first_block);
3432 * Allocate an inode for the symlink.
3434 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3435 1, 0, credp, prid, resblks > 0, &ip, NULL);
3437 if (error == ENOSPC)
3444 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3445 dp_joined_to_trans = B_TRUE;
3448 * Also attach the dquot(s) to it, if applicable.
3450 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3453 resblks -= XFS_IALLOC_SPACE_RES(mp);
3455 * If the symlink will fit into the inode, write it inline.
3457 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3458 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3459 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3460 ip->i_d.di_size = pathlen;
3463 * The inode was initially created in extent format.
3465 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3466 ip->i_df.if_flags |= XFS_IFINLINE;
3468 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3469 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3473 nmaps = SYMLINK_MAPS;
3475 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3476 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3477 &first_block, resblks, mval, &nmaps,
3484 resblks -= fs_blocks;
3485 ip->i_d.di_size = pathlen;
3486 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3488 cur_chunk = target_path;
3489 for (n = 0; n < nmaps; n++) {
3490 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3491 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3492 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3493 BTOBB(byte_cnt), 0);
3494 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3495 if (pathlen < byte_cnt) {
3498 pathlen -= byte_cnt;
3500 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3501 cur_chunk += byte_cnt;
3503 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3508 * Create the directory entry for the symlink.
3510 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3511 ip->i_ino, &first_block, &free_list, resblks);
3515 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3516 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3519 * Bump the in memory version number of the parent directory
3520 * so that other processes accessing it will recognize that
3521 * the directory has changed.
3526 * If this is a synchronous mount, make sure that the
3527 * symlink transaction goes to disk before returning to
3530 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3531 xfs_trans_set_sync(tp);
3535 * xfs_trans_commit normally decrements the vnode ref count
3536 * when it unlocks the inode. Since we want to return the
3537 * vnode to the caller, we bump the vnode ref count now.
3541 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3545 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3546 XFS_QM_DQRELE(mp, udqp);
3547 XFS_QM_DQRELE(mp, gdqp);
3549 /* Fall through to std_return with error = 0 or errno from
3550 * xfs_trans_commit */
3552 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3553 DM_EVENT_POSTSYMLINK)) {
3554 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3555 dir_vp, DM_RIGHT_NULL,
3556 error ? NULL : XFS_ITOV(ip),
3557 DM_RIGHT_NULL, link_name, target_path,
3573 xfs_bmap_cancel(&free_list);
3574 cancel_flags |= XFS_TRANS_ABORT;
3576 xfs_trans_cancel(tp, cancel_flags);
3577 XFS_QM_DQRELE(mp, udqp);
3578 XFS_QM_DQRELE(mp, gdqp);
3580 if (!dp_joined_to_trans && (dp != NULL)) {
3581 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3591 * A fid routine that takes a pointer to a previously allocated
3592 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3602 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3603 (inst_t *)__return_address);
3604 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3606 xfid = (xfs_fid2_t *)fidp;
3607 ip = XFS_BHVTOI(bdp);
3608 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3611 * use memcpy because the inode is a long long and there's no
3612 * assurance that xfid->fid_ino is properly aligned.
3614 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3615 xfid->fid_gen = ip->i_d.di_gen;
3632 vp = BHV_TO_VNODE(bdp);
3635 ip = XFS_BHVTOI(bdp);
3636 if (locktype == VRWLOCK_WRITE) {
3637 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3638 } else if (locktype == VRWLOCK_TRY_READ) {
3639 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3640 } else if (locktype == VRWLOCK_TRY_WRITE) {
3641 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3643 ASSERT((locktype == VRWLOCK_READ) ||
3644 (locktype == VRWLOCK_WRITE_DIRECT));
3645 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3663 vp = BHV_TO_VNODE(bdp);
3666 ip = XFS_BHVTOI(bdp);
3667 if (locktype == VRWLOCK_WRITE) {
3669 * In the write case, we may have added a new entry to
3670 * the reference cache. This might store a pointer to
3671 * an inode to be released in this inode. If it is there,
3672 * clear the pointer and release the inode after unlocking
3675 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3677 ASSERT((locktype == VRWLOCK_READ) ||
3678 (locktype == VRWLOCK_WRITE_DIRECT));
3679 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3691 xfs_inode_log_item_t *iip;
3694 ip = XFS_BHVTOI(bdp);
3698 if (XFS_FORCED_SHUTDOWN(mp))
3699 return XFS_ERROR(EIO);
3702 * Bypass inodes which have already been cleaned by
3703 * the inode flush clustering code inside xfs_iflush
3705 if ((ip->i_update_core == 0) &&
3706 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3709 if (flags & FLUSH_LOG) {
3710 if (iip && iip->ili_last_lsn) {
3711 xlog_t *log = mp->m_log;
3713 int s, log_flags = XFS_LOG_FORCE;
3715 s = GRANT_LOCK(log);
3716 sync_lsn = log->l_last_sync_lsn;
3717 GRANT_UNLOCK(log, s);
3719 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3722 if (flags & FLUSH_SYNC)
3723 log_flags |= XFS_LOG_SYNC;
3724 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3729 * We make this non-blocking if the inode is contended,
3730 * return EAGAIN to indicate to the caller that they
3731 * did not succeed. This prevents the flush path from
3732 * blocking on inodes inside another operation right
3733 * now, they get caught later by xfs_sync.
3735 if (flags & FLUSH_INODE) {
3738 if (xfs_ipincount(ip))
3741 if (flags & FLUSH_SYNC) {
3742 xfs_ilock(ip, XFS_ILOCK_SHARED);
3744 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3745 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3746 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3753 if (flags & FLUSH_SYNC)
3754 flush_flags = XFS_IFLUSH_SYNC;
3756 flush_flags = XFS_IFLUSH_ASYNC;
3758 error = xfs_iflush(ip, flush_flags);
3759 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3778 if (!capable(CAP_SYS_ADMIN))
3779 return XFS_ERROR(EPERM);
3781 ip = XFS_BHVTOI(bdp);
3784 if (XFS_FORCED_SHUTDOWN(mp))
3785 return XFS_ERROR(EIO);
3787 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3788 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3790 xfs_trans_cancel(tp, 0);
3793 xfs_ilock(ip, XFS_ILOCK_EXCL);
3794 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3796 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3797 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3799 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3801 error = xfs_trans_commit(tp, 0, NULL);
3817 vp = BHV_TO_VNODE(bdp);
3818 ip = XFS_BHVTOI(bdp);
3820 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3822 ASSERT(!VN_MAPPED(vp));
3824 /* bad inode, get out here ASAP */
3832 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3835 * Make sure the atime in the XFS inode is correct before freeing the
3838 xfs_synchronize_atime(ip);
3840 /* If we have nothing to flush with this inode then complete the
3841 * teardown now, otherwise break the link between the xfs inode
3842 * and the linux inode and clean up the xfs inode later. This
3843 * avoids flushing the inode to disk during the delete operation
3846 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3847 xfs_ilock(ip, XFS_ILOCK_EXCL);
3849 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3851 xfs_mount_t *mp = ip->i_mount;
3853 /* Protect sync from us */
3854 XFS_MOUNT_ILOCK(mp);
3855 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3856 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3857 ip->i_flags |= XFS_IRECLAIMABLE;
3858 XFS_MOUNT_IUNLOCK(mp);
3869 xfs_ihash_t *ih = ip->i_hash;
3870 vnode_t *vp = XFS_ITOV_NULL(ip);
3873 if (vp && VN_BAD(vp))
3876 /* The hash lock here protects a thread in xfs_iget_core from
3877 * racing with us on linking the inode back with a vnode.
3878 * Once we have the XFS_IRECLAIM flag set it will not touch
3881 write_lock(&ih->ih_lock);
3882 if ((ip->i_flags & XFS_IRECLAIM) ||
3883 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3884 write_unlock(&ih->ih_lock);
3887 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3891 ip->i_flags |= XFS_IRECLAIM;
3892 write_unlock(&ih->ih_lock);
3895 * If the inode is still dirty, then flush it out. If the inode
3896 * is not in the AIL, then it will be OK to flush it delwri as
3897 * long as xfs_iflush() does not keep any references to the inode.
3898 * We leave that decision up to xfs_iflush() since it has the
3899 * knowledge of whether it's OK to simply do a delwri flush of
3900 * the inode or whether we need to wait until the inode is
3901 * pulled from the AIL.
3902 * We get the flush lock regardless, though, just to make sure
3903 * we don't free it while it is being flushed.
3905 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3907 xfs_ilock(ip, XFS_ILOCK_EXCL);
3911 if (ip->i_update_core ||
3912 ((ip->i_itemp != NULL) &&
3913 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3914 error = xfs_iflush(ip, sync_mode);
3916 * If we hit an error, typically because of filesystem
3917 * shutdown, we don't need to let vn_reclaim to know
3918 * because we're gonna reclaim the inode anyway.
3921 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3924 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3927 ASSERT(ip->i_update_core == 0);
3928 ASSERT(ip->i_itemp == NULL ||
3929 ip->i_itemp->ili_format.ilf_fields == 0);
3930 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3931 } else if (locked) {
3933 * We are not interested in doing an iflush if we're
3934 * in the process of shutting down the filesystem forcibly.
3935 * So, just reclaim the inode.
3938 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3947 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3950 xfs_inode_t *ip, *n;
3955 XFS_MOUNT_ILOCK(mp);
3956 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3958 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3960 if (xfs_ipincount(ip) ||
3961 !xfs_iflock_nowait(ip)) {
3962 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3966 XFS_MOUNT_IUNLOCK(mp);
3967 if (xfs_finish_reclaim(ip, noblock,
3968 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3977 XFS_MOUNT_IUNLOCK(mp);
3982 * xfs_alloc_file_space()
3983 * This routine allocates disk space for the given file.
3985 * If alloc_type == 0, this request is for an ALLOCSP type
3986 * request which will change the file size. In this case, no
3987 * DMAPI event will be generated by the call. A TRUNCATE event
3988 * will be generated later by xfs_setattr.
3990 * If alloc_type != 0, this request is for a RESVSP type
3991 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3992 * lower block boundary byte address is less than the file's
4001 xfs_alloc_file_space(
4008 xfs_mount_t *mp = ip->i_mount;
4010 xfs_filblks_t allocated_fsb;
4011 xfs_filblks_t allocatesize_fsb;
4012 xfs_extlen_t extsz, temp;
4013 xfs_fileoff_t startoffset_fsb;
4014 xfs_fsblock_t firstfsb;
4020 xfs_bmbt_irec_t imaps[1], *imapp;
4021 xfs_bmap_free_t free_list;
4022 uint qblocks, resblks, resrtextents;
4026 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4028 if (XFS_FORCED_SHUTDOWN(mp))
4029 return XFS_ERROR(EIO);
4031 rt = XFS_IS_REALTIME_INODE(ip);
4033 if (!(extsz = ip->i_d.di_extsize))
4034 extsz = mp->m_sb.sb_rextsize;
4036 extsz = ip->i_d.di_extsize;
4039 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4043 return XFS_ERROR(EINVAL);
4049 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4050 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4051 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4053 /* Generate a DMAPI event if needed. */
4054 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4055 (attr_flags&ATTR_DMI) == 0 &&
4056 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4057 xfs_off_t end_dmi_offset;
4059 end_dmi_offset = offset+len;
4060 if (end_dmi_offset > ip->i_d.di_size)
4061 end_dmi_offset = ip->i_d.di_size;
4062 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4063 offset, end_dmi_offset - offset,
4070 * Allocate file space until done or until there is an error
4073 while (allocatesize_fsb && !error) {
4077 * Determine space reservations for data/realtime.
4079 if (unlikely(extsz)) {
4080 s = startoffset_fsb;
4083 e = startoffset_fsb + allocatesize_fsb;
4084 if ((temp = do_mod(startoffset_fsb, extsz)))
4086 if ((temp = do_mod(e, extsz)))
4090 e = allocatesize_fsb;
4094 resrtextents = qblocks = (uint)(e - s);
4095 resrtextents /= mp->m_sb.sb_rextsize;
4096 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4097 quota_flag = XFS_QMOPT_RES_RTBLKS;
4100 resblks = qblocks = \
4101 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4102 quota_flag = XFS_QMOPT_RES_REGBLKS;
4106 * Allocate and setup the transaction.
4108 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4109 error = xfs_trans_reserve(tp, resblks,
4110 XFS_WRITE_LOG_RES(mp), resrtextents,
4111 XFS_TRANS_PERM_LOG_RES,
4112 XFS_WRITE_LOG_COUNT);
4114 * Check for running out of space
4118 * Free the transaction structure.
4120 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4121 xfs_trans_cancel(tp, 0);
4124 xfs_ilock(ip, XFS_ILOCK_EXCL);
4125 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4126 qblocks, 0, quota_flag);
4130 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4131 xfs_trans_ihold(tp, ip);
4134 * Issue the xfs_bmapi() call to allocate the blocks
4136 XFS_BMAP_INIT(&free_list, &firstfsb);
4137 error = xfs_bmapi(tp, ip, startoffset_fsb,
4138 allocatesize_fsb, bmapi_flag,
4139 &firstfsb, 0, imapp, &nimaps,
4146 * Complete the transaction
4148 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4153 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4154 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4159 allocated_fsb = imapp->br_blockcount;
4162 error = XFS_ERROR(ENOSPC);
4166 startoffset_fsb += allocated_fsb;
4167 allocatesize_fsb -= allocated_fsb;
4170 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4171 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4173 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4174 XFS_ITOV(ip), DM_RIGHT_NULL,
4175 XFS_ITOV(ip), DM_RIGHT_NULL,
4176 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4178 goto retry; /* Maybe DMAPI app. has made space */
4179 /* else fall through with error from XFS_SEND_DATA */
4184 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4185 xfs_bmap_cancel(&free_list);
4186 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4188 error1: /* Just cancel transaction */
4189 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4190 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4191 goto dmapi_enospc_check;
4195 * Zero file bytes between startoff and endoff inclusive.
4196 * The iolock is held exclusive and no blocks are buffered.
4199 xfs_zero_remaining_bytes(
4204 xfs_bmbt_irec_t imap;
4205 xfs_fileoff_t offset_fsb;
4206 xfs_off_t lastoffset;
4209 xfs_mount_t *mp = ip->i_mount;
4213 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4214 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4215 mp->m_rtdev_targp : mp->m_ddev_targp);
4217 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4218 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4220 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4222 if (error || nimap < 1)
4224 ASSERT(imap.br_blockcount >= 1);
4225 ASSERT(imap.br_startoff == offset_fsb);
4226 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4227 if (lastoffset > endoff)
4228 lastoffset = endoff;
4229 if (imap.br_startblock == HOLESTARTBLOCK)
4231 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4232 if (imap.br_state == XFS_EXT_UNWRITTEN)
4235 XFS_BUF_UNWRITE(bp);
4237 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4239 if ((error = xfs_iowait(bp))) {
4240 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4241 mp, bp, XFS_BUF_ADDR(bp));
4244 memset(XFS_BUF_PTR(bp) +
4245 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4246 0, lastoffset - offset + 1);
4251 if ((error = xfs_iowait(bp))) {
4252 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4253 mp, bp, XFS_BUF_ADDR(bp));
4262 * xfs_free_file_space()
4263 * This routine frees disk space for the given file.
4265 * This routine is only called by xfs_change_file_space
4266 * for an UNRESVSP type call.
4274 xfs_free_file_space(
4283 xfs_off_t end_dmi_offset;
4284 xfs_fileoff_t endoffset_fsb;
4286 xfs_fsblock_t firstfsb;
4287 xfs_bmap_free_t free_list;
4289 xfs_bmbt_irec_t imap;
4297 xfs_fileoff_t startoffset_fsb;
4299 int need_iolock = 1;
4304 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4306 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4310 if (len <= 0) /* if nothing being freed */
4312 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4313 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4314 end_dmi_offset = offset + len;
4315 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4317 if (offset < ip->i_d.di_size &&
4318 (attr_flags & ATTR_DMI) == 0 &&
4319 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4320 if (end_dmi_offset > ip->i_d.di_size)
4321 end_dmi_offset = ip->i_d.di_size;
4322 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4323 offset, end_dmi_offset - offset,
4324 AT_DELAY_FLAG(attr_flags), NULL);
4329 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4330 if (attr_flags & ATTR_NOLOCK)
4333 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4334 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4337 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4339 ilen = len + (offset & (rounding - 1));
4340 ioffset = offset & ~(rounding - 1);
4341 if (ilen & (rounding - 1))
4342 ilen = (ilen + rounding) & ~(rounding - 1);
4344 if (VN_CACHED(vp) != 0) {
4345 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4346 ctooff(offtoct(ioffset)), -1);
4347 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4348 -1, FI_REMAPF_LOCKED);
4352 * Need to zero the stuff we're not freeing, on disk.
4353 * If its a realtime file & can't use unwritten extents then we
4354 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4355 * will take care of it for us.
4357 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4359 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4360 &imap, &nimap, NULL);
4362 goto out_unlock_iolock;
4363 ASSERT(nimap == 0 || nimap == 1);
4364 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4367 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4368 block = imap.br_startblock;
4369 mod = do_div(block, mp->m_sb.sb_rextsize);
4371 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4374 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4375 &imap, &nimap, NULL);
4377 goto out_unlock_iolock;
4378 ASSERT(nimap == 0 || nimap == 1);
4379 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4380 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4382 if (mod && (mod != mp->m_sb.sb_rextsize))
4383 endoffset_fsb -= mod;
4386 if ((done = (endoffset_fsb <= startoffset_fsb)))
4388 * One contiguous piece to clear
4390 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4393 * Some full blocks, possibly two pieces to clear
4395 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4396 error = xfs_zero_remaining_bytes(ip, offset,
4397 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4399 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4400 error = xfs_zero_remaining_bytes(ip,
4401 XFS_FSB_TO_B(mp, endoffset_fsb),
4406 * free file space until done or until there is an error
4408 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4409 while (!error && !done) {
4412 * allocate and setup the transaction
4414 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4415 error = xfs_trans_reserve(tp,
4417 XFS_WRITE_LOG_RES(mp),
4419 XFS_TRANS_PERM_LOG_RES,
4420 XFS_WRITE_LOG_COUNT);
4423 * check for running out of space
4427 * Free the transaction structure.
4429 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4430 xfs_trans_cancel(tp, 0);
4433 xfs_ilock(ip, XFS_ILOCK_EXCL);
4434 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4435 ip->i_udquot, ip->i_gdquot, resblks, 0,
4436 XFS_QMOPT_RES_REGBLKS);
4440 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4441 xfs_trans_ihold(tp, ip);
4444 * issue the bunmapi() call to free the blocks
4446 XFS_BMAP_INIT(&free_list, &firstfsb);
4447 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4448 endoffset_fsb - startoffset_fsb,
4449 0, 2, &firstfsb, &free_list, &done);
4455 * complete the transaction
4457 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4462 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4463 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4468 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4472 xfs_bmap_cancel(&free_list);
4474 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4475 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4481 * xfs_change_file_space()
4482 * This routine allocates or frees disk space for the given file.
4483 * The user specified parameters are checked for alignment and size
4492 xfs_change_file_space(
4506 xfs_off_t startoffset;
4512 vp = BHV_TO_VNODE(bdp);
4513 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4515 ip = XFS_BHVTOI(bdp);
4519 * must be a regular file and have write permission
4522 return XFS_ERROR(EINVAL);
4524 xfs_ilock(ip, XFS_ILOCK_SHARED);
4526 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4527 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4531 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4533 switch (bf->l_whence) {
4534 case 0: /*SEEK_SET*/
4536 case 1: /*SEEK_CUR*/
4537 bf->l_start += offset;
4539 case 2: /*SEEK_END*/
4540 bf->l_start += ip->i_d.di_size;
4543 return XFS_ERROR(EINVAL);
4546 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4548 if ( (bf->l_start < 0)
4549 || (bf->l_start > XFS_MAXIOFFSET(mp))
4550 || (bf->l_start + llen < 0)
4551 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4552 return XFS_ERROR(EINVAL);
4556 startoffset = bf->l_start;
4557 fsize = ip->i_d.di_size;
4560 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4562 * These calls do NOT zero the data space allocated to the file,
4563 * nor do they change the file size.
4565 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4567 * These calls cause the new file data to be zeroed and the file
4568 * size to be changed.
4570 setprealloc = clrprealloc = 0;
4573 case XFS_IOC_RESVSP:
4574 case XFS_IOC_RESVSP64:
4575 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4582 case XFS_IOC_UNRESVSP:
4583 case XFS_IOC_UNRESVSP64:
4584 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4589 case XFS_IOC_ALLOCSP:
4590 case XFS_IOC_ALLOCSP64:
4591 case XFS_IOC_FREESP:
4592 case XFS_IOC_FREESP64:
4593 if (startoffset > fsize) {
4594 error = xfs_alloc_file_space(ip, fsize,
4595 startoffset - fsize, 0, attr_flags);
4600 va.va_mask = XFS_AT_SIZE;
4601 va.va_size = startoffset;
4603 error = xfs_setattr(bdp, &va, attr_flags, credp);
4613 return XFS_ERROR(EINVAL);
4617 * update the inode timestamp, mode, and prealloc flag bits
4619 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4621 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4624 xfs_trans_cancel(tp, 0);
4628 xfs_ilock(ip, XFS_ILOCK_EXCL);
4630 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4631 xfs_trans_ihold(tp, ip);
4633 if ((attr_flags & ATTR_DMI) == 0) {
4634 ip->i_d.di_mode &= ~S_ISUID;
4637 * Note that we don't have to worry about mandatory
4638 * file locking being disabled here because we only
4639 * clear the S_ISGID bit if the Group execute bit is
4640 * on, but if it was on then mandatory locking wouldn't
4641 * have been enabled.
4643 if (ip->i_d.di_mode & S_IXGRP)
4644 ip->i_d.di_mode &= ~S_ISGID;
4646 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4649 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4650 else if (clrprealloc)
4651 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4653 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4654 xfs_trans_set_sync(tp);
4656 error = xfs_trans_commit(tp, 0, NULL);
4658 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4663 vnodeops_t xfs_vnodeops = {
4664 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4665 .vop_open = xfs_open,
4666 .vop_read = xfs_read,
4667 #ifdef HAVE_SENDFILE
4668 .vop_sendfile = xfs_sendfile,
4671 .vop_splice_read = xfs_splice_read,
4672 .vop_splice_write = xfs_splice_write,
4674 .vop_write = xfs_write,
4675 .vop_ioctl = xfs_ioctl,
4676 .vop_getattr = xfs_getattr,
4677 .vop_setattr = xfs_setattr,
4678 .vop_access = xfs_access,
4679 .vop_lookup = xfs_lookup,
4680 .vop_create = xfs_create,
4681 .vop_remove = xfs_remove,
4682 .vop_link = xfs_link,
4683 .vop_rename = xfs_rename,
4684 .vop_mkdir = xfs_mkdir,
4685 .vop_rmdir = xfs_rmdir,
4686 .vop_readdir = xfs_readdir,
4687 .vop_symlink = xfs_symlink,
4688 .vop_readlink = xfs_readlink,
4689 .vop_fsync = xfs_fsync,
4690 .vop_inactive = xfs_inactive,
4691 .vop_fid2 = xfs_fid2,
4692 .vop_rwlock = xfs_rwlock,
4693 .vop_rwunlock = xfs_rwunlock,
4694 .vop_bmap = xfs_bmap,
4695 .vop_reclaim = xfs_reclaim,
4696 .vop_attr_get = xfs_attr_get,
4697 .vop_attr_set = xfs_attr_set,
4698 .vop_attr_remove = xfs_attr_remove,
4699 .vop_attr_list = xfs_attr_list,
4700 .vop_link_removed = (vop_link_removed_t)fs_noval,
4701 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4702 .vop_tosspages = fs_tosspages,
4703 .vop_flushinval_pages = fs_flushinval_pages,
4704 .vop_flush_pages = fs_flush_pages,
4705 .vop_release = xfs_release,
4706 .vop_iflush = xfs_inode_flush,