2 * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_itable.h"
49 #include "xfs_btree.h"
50 #include "xfs_ialloc.h"
51 #include "xfs_alloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
59 #include "xfs_da_btree.h"
62 #include "xfs_refcache.h"
63 #include "xfs_error.h"
65 #include "xfs_rtalloc.h"
66 #include "xfs_quota.h"
67 #include "xfs_utils.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_dir_leaf.h"
71 #include "xfs_log_priv.h"
75 * The maximum pathlen is 1024 bytes. Since the minimum file system
76 * blocksize is 512 bytes, we can get a max of 2 extents back from
79 #define SYMLINK_MAPS 2
82 * For xfs, we check that the file isn't too big to be opened by this kernel.
83 * No other open action is required for regular files. Devices are handled
84 * through the specfs file system, pipes through fifofs. Device and
85 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
86 * when a new vnode is first looked up or created.
97 vp = BHV_TO_VNODE(bdp);
100 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
101 return XFS_ERROR(EIO);
104 * If it's a directory with any blocks, read-ahead block 0
105 * as we're almost certain to have the next operation be a read there.
107 if (vp->v_type == VDIR && ip->i_d.di_nextents > 0) {
108 mode = xfs_ilock_map_shared(ip);
109 if (ip->i_d.di_nextents > 0)
110 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
111 xfs_iunlock(ip, mode);
131 vp = BHV_TO_VNODE(bdp);
132 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
134 ip = XFS_BHVTOI(bdp);
137 if (XFS_FORCED_SHUTDOWN(mp))
138 return XFS_ERROR(EIO);
140 if (!(flags & ATTR_LAZY))
141 xfs_ilock(ip, XFS_ILOCK_SHARED);
143 vap->va_size = ip->i_d.di_size;
144 if (vap->va_mask == XFS_AT_SIZE)
148 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
149 vap->va_nodeid = ip->i_ino;
151 vap->va_nodeid += mp->m_inoadd;
153 vap->va_nlink = ip->i_d.di_nlink;
156 * Quick exit for non-stat callers
159 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
160 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
164 * Copy from in-core inode.
166 vap->va_type = vp->v_type;
167 vap->va_mode = ip->i_d.di_mode & MODEMASK;
168 vap->va_uid = ip->i_d.di_uid;
169 vap->va_gid = ip->i_d.di_gid;
170 vap->va_projid = ip->i_d.di_projid;
173 * Check vnode type block/char vs. everything else.
174 * Do it with bitmask because that's faster than looking
175 * for multiple values individually.
177 if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
180 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
183 /* Large block sizes confuse various
184 * user space programs, so letting the
185 * stripe size through is not a good
188 vap->va_blocksize = mp->m_swidth ?
190 * If the underlying volume is a stripe, then
191 * return the stripe width in bytes as the
192 * recommended I/O size.
194 (mp->m_swidth << mp->m_sb.sb_blocklog) :
196 * Return the largest of the preferred buffer
197 * sizes since doing small I/Os into larger
198 * buffers causes buffers to be decommissioned.
199 * The value returned is in bytes.
201 (1 << (int)MAX(mp->m_readio_log,
207 * Return the largest of the preferred buffer
208 * sizes since doing small I/Os into larger
209 * buffers causes buffers to be decommissioned.
210 * The value returned is in bytes.
212 1 << (int)MAX(mp->m_readio_log,
218 * If the file blocks are being allocated from a
219 * realtime partition, then return the inode's
220 * realtime extent size or the realtime volume's
223 vap->va_blocksize = ip->i_d.di_extsize ?
224 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
225 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
228 vap->va_rdev = ip->i_df.if_u2.if_rdev;
229 vap->va_blocksize = BLKDEV_IOSIZE;
232 vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
233 vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
234 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
235 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
236 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
237 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
240 * Exit for stat callers. See if any of the rest of the fields
241 * to be filled in are needed.
244 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
245 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
249 * Convert di_flags to xflags.
251 vap->va_xflags = xfs_ip2xflags(ip);
254 * Exit for inode revalidate. See if any of the rest of
255 * the fields to be filled in are needed.
258 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
259 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
262 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
264 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
265 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
269 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
270 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
271 ip->i_d.di_anextents;
273 vap->va_anextents = 0;
274 vap->va_gen = ip->i_d.di_gen;
277 if (!(flags & ATTR_LAZY))
278 xfs_iunlock(ip, XFS_ILOCK_SHARED);
304 xfs_prid_t projid=0, iprojid=0;
305 int mandlock_before, mandlock_after;
306 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
310 vp = BHV_TO_VNODE(bdp);
311 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
313 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
314 return XFS_ERROR(EROFS);
317 * Cannot set certain attributes.
320 if (mask & XFS_AT_NOSET) {
321 return XFS_ERROR(EINVAL);
324 ip = XFS_BHVTOI(bdp);
327 if (XFS_FORCED_SHUTDOWN(mp))
328 return XFS_ERROR(EIO);
331 * Timestamps do not need to be logged and hence do not
332 * need to be done within a transaction.
334 if (mask & XFS_AT_UPDTIMES) {
335 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
336 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
337 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
338 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
339 xfs_ichgtime(ip, timeflags);
343 olddquot1 = olddquot2 = NULL;
347 * If disk quotas is on, we make sure that the dquots do exist on disk,
348 * before we start any other transactions. Trying to do this later
349 * is messy. We don't care to take a readlock to look at the ids
350 * in inode here, because we can't hold it across the trans_reserve.
351 * If the IDs do change before we take the ilock, we're covered
352 * because the i_*dquot fields will get updated anyway.
354 if (XFS_IS_QUOTA_ON(mp) &&
355 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
358 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
360 qflags |= XFS_QMOPT_UQUOTA;
362 uid = ip->i_d.di_uid;
364 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
366 qflags |= XFS_QMOPT_GQUOTA;
368 gid = ip->i_d.di_gid;
370 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
371 projid = vap->va_projid;
372 qflags |= XFS_QMOPT_PQUOTA;
374 projid = ip->i_d.di_projid;
377 * We take a reference when we initialize udqp and gdqp,
378 * so it is important that we never blindly double trip on
379 * the same variable. See xfs_create() for an example.
381 ASSERT(udqp == NULL);
382 ASSERT(gdqp == NULL);
383 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
390 * For the other attributes, we acquire the inode lock and
391 * first do an error checking pass.
394 lock_flags = XFS_ILOCK_EXCL;
395 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
396 if (flags & ATTR_NOLOCK)
398 if (!(mask & XFS_AT_SIZE)) {
399 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
400 (mp->m_flags & XFS_MOUNT_WSYNC)) {
401 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
403 if ((code = xfs_trans_reserve(tp, 0,
404 XFS_ICHANGE_LOG_RES(mp), 0,
411 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
412 !(flags & ATTR_DMI)) {
413 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
414 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
415 vap->va_size, 0, dmflags, NULL);
422 lock_flags |= XFS_IOLOCK_EXCL;
425 xfs_ilock(ip, lock_flags);
427 /* boolean: are we the file owner? */
428 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
431 * Change various properties of a file.
432 * Only the owner or users with CAP_FOWNER
433 * capability may do these things.
436 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
437 XFS_AT_GID|XFS_AT_PROJID)) {
439 * CAP_FOWNER overrides the following restrictions:
441 * The user ID of the calling process must be equal
442 * to the file owner ID, except in cases where the
443 * CAP_FSETID capability is applicable.
445 if (!file_owner && !capable(CAP_FOWNER)) {
446 code = XFS_ERROR(EPERM);
451 * CAP_FSETID overrides the following restrictions:
453 * The effective user ID of the calling process shall match
454 * the file owner when setting the set-user-ID and
455 * set-group-ID bits on that file.
457 * The effective group ID or one of the supplementary group
458 * IDs of the calling process shall match the group owner of
459 * the file when setting the set-group-ID bit on that file
461 if (mask & XFS_AT_MODE) {
464 if ((vap->va_mode & S_ISUID) && !file_owner)
466 if ((vap->va_mode & S_ISGID) &&
467 !in_group_p((gid_t)ip->i_d.di_gid))
470 /* Linux allows this, Irix doesn't. */
471 if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
474 if (m && !capable(CAP_FSETID))
480 * Change file ownership. Must be the owner or privileged.
481 * If the system was configured with the "restricted_chown"
482 * option, the owner is not permitted to give away the file,
483 * and can change the group id only to a group of which he
484 * or she is a member.
486 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
488 * These IDs could have changed since we last looked at them.
489 * But, we're assured that if the ownership did change
490 * while we didn't have the inode locked, inode's dquot(s)
491 * would have changed also.
493 iuid = ip->i_d.di_uid;
494 iprojid = ip->i_d.di_projid;
495 igid = ip->i_d.di_gid;
496 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
497 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
498 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
502 * CAP_CHOWN overrides the following restrictions:
504 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
505 * shall override the restriction that a process cannot
506 * change the user ID of a file it owns and the restriction
507 * that the group ID supplied to the chown() function
508 * shall be equal to either the group ID or one of the
509 * supplementary group IDs of the calling process.
511 * XXX: How does restricted_chown affect projid?
513 if (restricted_chown &&
514 (iuid != uid || (igid != gid &&
515 !in_group_p((gid_t)gid))) &&
516 !capable(CAP_CHOWN)) {
517 code = XFS_ERROR(EPERM);
521 * Do a quota reservation only if uid/projid/gid is actually
524 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
525 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
526 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
528 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
529 capable(CAP_FOWNER) ?
530 XFS_QMOPT_FORCE_RES : 0);
531 if (code) /* out of quota */
537 * Truncate file. Must have write permission and not be a directory.
539 if (mask & XFS_AT_SIZE) {
540 /* Short circuit the truncate case for zero length files */
541 if ((vap->va_size == 0) &&
542 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
543 xfs_iunlock(ip, XFS_ILOCK_EXCL);
544 lock_flags &= ~XFS_ILOCK_EXCL;
545 if (mask & XFS_AT_CTIME)
546 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
551 if (vp->v_type == VDIR) {
552 code = XFS_ERROR(EISDIR);
554 } else if (vp->v_type != VREG) {
555 code = XFS_ERROR(EINVAL);
559 * Make sure that the dquots are attached to the inode.
561 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
566 * Change file access or modified times.
568 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
570 if ((flags & ATTR_UTIME) &&
571 !capable(CAP_FOWNER)) {
572 code = XFS_ERROR(EPERM);
579 * Change extent size or realtime flag.
581 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
583 * Can't change extent size if any extents are allocated.
585 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
586 (mask & XFS_AT_EXTSIZE) &&
587 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
589 code = XFS_ERROR(EINVAL); /* EFBIG? */
594 * Can't set extent size unless the file is marked, or
595 * about to be marked as a realtime file.
597 * This check will be removed when fixed size extents
598 * with buffered data writes is implemented.
601 if ((mask & XFS_AT_EXTSIZE) &&
602 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
604 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
605 ((mask & XFS_AT_XFLAGS) &&
606 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
607 code = XFS_ERROR(EINVAL);
612 * Can't change realtime flag if any extents are allocated.
614 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
615 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
616 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
617 code = XFS_ERROR(EINVAL); /* EFBIG? */
621 * Extent size must be a multiple of the appropriate block
622 * size, if set at all.
624 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
627 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
628 ((mask & XFS_AT_XFLAGS) &&
629 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
630 size = mp->m_sb.sb_rextsize <<
631 mp->m_sb.sb_blocklog;
633 size = mp->m_sb.sb_blocksize;
635 if (vap->va_extsize % size) {
636 code = XFS_ERROR(EINVAL);
641 * If realtime flag is set then must have realtime data.
643 if ((mask & XFS_AT_XFLAGS) &&
644 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
645 if ((mp->m_sb.sb_rblocks == 0) ||
646 (mp->m_sb.sb_rextsize == 0) ||
647 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
648 code = XFS_ERROR(EINVAL);
654 * Can't modify an immutable/append-only file unless
655 * we have appropriate permission.
657 if ((mask & XFS_AT_XFLAGS) &&
659 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
661 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
662 !capable(CAP_LINUX_IMMUTABLE)) {
663 code = XFS_ERROR(EPERM);
669 * Now we can make the changes. Before we join the inode
670 * to the transaction, if XFS_AT_SIZE is set then take care of
671 * the part of the truncation that must be done without the
672 * inode lock. This needs to be done before joining the inode
673 * to the transaction, because the inode cannot be unlocked
674 * once it is a part of the transaction.
676 if (mask & XFS_AT_SIZE) {
678 if (vap->va_size > ip->i_d.di_size)
679 code = xfs_igrow_start(ip, vap->va_size, credp);
680 xfs_iunlock(ip, XFS_ILOCK_EXCL);
682 code = xfs_itruncate_data(ip, vap->va_size);
685 lock_flags &= ~XFS_ILOCK_EXCL;
686 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
689 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
690 if ((code = xfs_trans_reserve(tp, 0,
691 XFS_ITRUNCATE_LOG_RES(mp), 0,
692 XFS_TRANS_PERM_LOG_RES,
693 XFS_ITRUNCATE_LOG_COUNT))) {
694 xfs_trans_cancel(tp, 0);
696 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
699 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
700 xfs_ilock(ip, XFS_ILOCK_EXCL);
704 xfs_trans_ijoin(tp, ip, lock_flags);
705 xfs_trans_ihold(tp, ip);
708 /* determine whether mandatory locking mode changes */
709 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
712 * Truncate file. Must have write permission and not be a directory.
714 if (mask & XFS_AT_SIZE) {
715 if (vap->va_size > ip->i_d.di_size) {
716 xfs_igrow_finish(tp, ip, vap->va_size,
717 !(flags & ATTR_DMI));
718 } else if ((vap->va_size <= ip->i_d.di_size) ||
719 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
721 * signal a sync transaction unless
722 * we're truncating an already unlinked
723 * file on a wsync filesystem
725 code = xfs_itruncate_finish(&tp, ip,
726 (xfs_fsize_t)vap->va_size,
728 ((ip->i_d.di_nlink != 0 ||
729 !(mp->m_flags & XFS_MOUNT_WSYNC))
736 * Have to do this even if the file's size doesn't change.
738 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
742 * Change file access modes.
744 if (mask & XFS_AT_MODE) {
745 ip->i_d.di_mode &= S_IFMT;
746 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
748 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
749 timeflags |= XFS_ICHGTIME_CHG;
753 * Change file ownership. Must be the owner or privileged.
754 * If the system was configured with the "restricted_chown"
755 * option, the owner is not permitted to give away the file,
756 * and can change the group id only to a group of which he
757 * or she is a member.
759 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
761 * CAP_FSETID overrides the following restrictions:
763 * The set-user-ID and set-group-ID bits of a file will be
764 * cleared upon successful return from chown()
766 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
767 !capable(CAP_FSETID)) {
768 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
772 * Change the ownerships and register quota modifications
773 * in the transaction.
776 if (XFS_IS_UQUOTA_ON(mp)) {
777 ASSERT(mask & XFS_AT_UID);
779 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
780 &ip->i_udquot, udqp);
782 ip->i_d.di_uid = uid;
785 if (XFS_IS_GQUOTA_ON(mp)) {
786 ASSERT(!XFS_IS_PQUOTA_ON(mp));
787 ASSERT(mask & XFS_AT_GID);
789 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
790 &ip->i_gdquot, gdqp);
792 ip->i_d.di_gid = gid;
794 if (iprojid != projid) {
795 if (XFS_IS_PQUOTA_ON(mp)) {
796 ASSERT(!XFS_IS_GQUOTA_ON(mp));
797 ASSERT(mask & XFS_AT_PROJID);
799 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
800 &ip->i_gdquot, gdqp);
802 ip->i_d.di_projid = projid;
804 * We may have to rev the inode as well as
805 * the superblock version number since projids didn't
806 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
808 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
809 xfs_bump_ino_vers2(tp, ip);
812 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
813 timeflags |= XFS_ICHGTIME_CHG;
818 * Change file access or modified times.
820 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
821 if (mask & XFS_AT_ATIME) {
822 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
823 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
824 ip->i_update_core = 1;
825 timeflags &= ~XFS_ICHGTIME_ACC;
827 if (mask & XFS_AT_MTIME) {
828 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
829 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
830 timeflags &= ~XFS_ICHGTIME_MOD;
831 timeflags |= XFS_ICHGTIME_CHG;
833 if (tp && (flags & ATTR_UTIME))
834 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
838 * Change XFS-added attributes.
840 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
841 if (mask & XFS_AT_EXTSIZE) {
843 * Converting bytes to fs blocks.
845 ip->i_d.di_extsize = vap->va_extsize >>
846 mp->m_sb.sb_blocklog;
848 if (mask & XFS_AT_XFLAGS) {
851 /* can't set PREALLOC this way, just preserve it */
852 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
853 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
854 di_flags |= XFS_DIFLAG_IMMUTABLE;
855 if (vap->va_xflags & XFS_XFLAG_APPEND)
856 di_flags |= XFS_DIFLAG_APPEND;
857 if (vap->va_xflags & XFS_XFLAG_SYNC)
858 di_flags |= XFS_DIFLAG_SYNC;
859 if (vap->va_xflags & XFS_XFLAG_NOATIME)
860 di_flags |= XFS_DIFLAG_NOATIME;
861 if (vap->va_xflags & XFS_XFLAG_NODUMP)
862 di_flags |= XFS_DIFLAG_NODUMP;
863 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
864 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
865 di_flags |= XFS_DIFLAG_RTINHERIT;
866 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
867 di_flags |= XFS_DIFLAG_NOSYMLINKS;
869 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
870 di_flags |= XFS_DIFLAG_REALTIME;
871 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
873 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
876 ip->i_d.di_flags = di_flags;
878 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
879 timeflags |= XFS_ICHGTIME_CHG;
883 * Change file inode change time only if XFS_AT_CTIME set
884 * AND we have been called by a DMI function.
887 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
888 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
889 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
890 ip->i_update_core = 1;
891 timeflags &= ~XFS_ICHGTIME_CHG;
895 * Send out timestamp changes that need to be set to the
896 * current time. Not done when called by a DMI function.
898 if (timeflags && !(flags & ATTR_DMI))
899 xfs_ichgtime(ip, timeflags);
901 XFS_STATS_INC(xs_ig_attrchg);
904 * If this is a synchronous mount, make sure that the
905 * transaction goes to disk before returning to the user.
906 * This is slightly sub-optimal in that truncates require
907 * two sync transactions instead of one for wsync filesytems.
908 * One for the truncate and one for the timestamps since we
909 * don't want to change the timestamps unless we're sure the
910 * truncate worked. Truncates are less than 1% of the laddis
911 * mix so this probably isn't worth the trouble to optimize.
915 if (mp->m_flags & XFS_MOUNT_WSYNC)
916 xfs_trans_set_sync(tp);
918 code = xfs_trans_commit(tp, commit_flags, NULL);
922 * If the (regular) file's mandatory locking mode changed, then
923 * notify the vnode. We do this under the inode lock to prevent
924 * racing calls to vop_vnode_change.
926 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
927 if (mandlock_before != mandlock_after) {
928 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
932 xfs_iunlock(ip, lock_flags);
935 * Release any dquot(s) the inode had kept before chown.
937 XFS_QM_DQRELE(mp, olddquot1);
938 XFS_QM_DQRELE(mp, olddquot2);
939 XFS_QM_DQRELE(mp, udqp);
940 XFS_QM_DQRELE(mp, gdqp);
946 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
947 !(flags & ATTR_DMI)) {
948 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
949 NULL, DM_RIGHT_NULL, NULL, NULL,
950 0, 0, AT_DELAY_FLAG(flags));
955 commit_flags |= XFS_TRANS_ABORT;
958 XFS_QM_DQRELE(mp, udqp);
959 XFS_QM_DQRELE(mp, gdqp);
961 xfs_trans_cancel(tp, commit_flags);
963 if (lock_flags != 0) {
964 xfs_iunlock(ip, lock_flags);
972 * Null conversion from vnode mode bits to inode mode bits, as in efs.
983 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
984 (inst_t *)__return_address);
986 ip = XFS_BHVTOI(bdp);
987 xfs_ilock(ip, XFS_ILOCK_SHARED);
988 error = xfs_iaccess(ip, mode, credp);
989 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1013 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1019 vp = BHV_TO_VNODE(bdp);
1020 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1022 ip = XFS_BHVTOI(bdp);
1025 if (XFS_FORCED_SHUTDOWN(mp))
1026 return XFS_ERROR(EIO);
1028 xfs_ilock(ip, XFS_ILOCK_SHARED);
1030 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1032 offset = uiop->uio_offset;
1033 count = uiop->uio_resid;
1036 error = XFS_ERROR(EINVAL);
1044 if (!(ioflags & IO_INVIS)) {
1045 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1049 * See if the symlink is stored inline.
1051 pathlen = (int)ip->i_d.di_size;
1053 if (ip->i_df.if_flags & XFS_IFINLINE) {
1054 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1058 * Symlink not inline. Call bmap to get it in.
1060 nmaps = SYMLINK_MAPS;
1062 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1063 0, NULL, 0, mval, &nmaps, NULL);
1069 for (n = 0; n < nmaps; n++) {
1070 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1071 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1072 bp = xfs_buf_read(mp->m_ddev_targp, d,
1073 BTOBB(byte_cnt), 0);
1074 error = XFS_BUF_GETERROR(bp);
1076 xfs_ioerror_alert("xfs_readlink",
1077 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1081 if (pathlen < byte_cnt)
1083 pathlen -= byte_cnt;
1085 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1094 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1103 * This is called to sync the inode and its data out to disk.
1104 * We need to hold the I/O lock while flushing the data, and
1105 * the inode lock while flushing the inode. The inode lock CANNOT
1106 * be held while flushing the data, so acquire after we're done
1121 vn_trace_entry(BHV_TO_VNODE(bdp),
1122 __FUNCTION__, (inst_t *)__return_address);
1124 ip = XFS_BHVTOI(bdp);
1126 ASSERT(start >= 0 && stop >= -1);
1128 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1129 return XFS_ERROR(EIO);
1132 * We always need to make sure that the required inode state
1133 * is safe on disk. The vnode might be clean but because
1134 * of committed transactions that haven't hit the disk yet.
1135 * Likewise, there could be unflushed non-transactional
1136 * changes to the inode core that have to go to disk.
1138 * The following code depends on one assumption: that
1139 * any transaction that changes an inode logs the core
1140 * because it has to change some field in the inode core
1141 * (typically nextents or nblocks). That assumption
1142 * implies that any transactions against an inode will
1143 * catch any non-transactional updates. If inode-altering
1144 * transactions exist that violate this assumption, the
1145 * code breaks. Right now, it figures that if the involved
1146 * update_* field is clear and the inode is unpinned, the
1147 * inode is clean. Either it's been flushed or it's been
1148 * committed and the commit has hit the disk unpinning the inode.
1149 * (Note that xfs_inode_item_format() called at commit clears
1150 * the update_* fields.)
1152 xfs_ilock(ip, XFS_ILOCK_SHARED);
1154 /* If we are flushing data then we care about update_size
1155 * being set, otherwise we care about update_core
1157 if ((flag & FSYNC_DATA) ?
1158 (ip->i_update_size == 0) :
1159 (ip->i_update_core == 0)) {
1161 * Timestamps/size haven't changed since last inode
1162 * flush or inode transaction commit. That means
1163 * either nothing got written or a transaction
1164 * committed which caught the updates. If the
1165 * latter happened and the transaction hasn't
1166 * hit the disk yet, the inode will be still
1167 * be pinned. If it is, force the log.
1170 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1172 if (xfs_ipincount(ip)) {
1173 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1175 ((flag & FSYNC_WAIT)
1176 ? XFS_LOG_SYNC : 0));
1181 * Kick off a transaction to log the inode
1182 * core to get the updates. Make it
1183 * sync if FSYNC_WAIT is passed in (which
1184 * is done by everybody but specfs). The
1185 * sync transaction will also force the log.
1187 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1188 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1189 if ((error = xfs_trans_reserve(tp, 0,
1190 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1192 xfs_trans_cancel(tp, 0);
1195 xfs_ilock(ip, XFS_ILOCK_EXCL);
1198 * Note - it's possible that we might have pushed
1199 * ourselves out of the way during trans_reserve
1200 * which would flush the inode. But there's no
1201 * guarantee that the inode buffer has actually
1202 * gone out yet (it's delwri). Plus the buffer
1203 * could be pinned anyway if it's part of an
1204 * inode in another recent transaction. So we
1205 * play it safe and fire off the transaction anyway.
1207 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1208 xfs_trans_ihold(tp, ip);
1209 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1210 if (flag & FSYNC_WAIT)
1211 xfs_trans_set_sync(tp);
1212 error = xfs_trans_commit(tp, 0, NULL);
1214 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1220 * This is called by xfs_inactive to free any blocks beyond eof,
1221 * when the link count isn't zero.
1224 xfs_inactive_free_eofblocks(
1230 xfs_fileoff_t end_fsb;
1231 xfs_fileoff_t last_fsb;
1232 xfs_filblks_t map_len;
1234 xfs_bmbt_irec_t imap;
1237 * Figure out if there are any blocks beyond the end
1238 * of the file. If not, then there is nothing to do.
1240 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1241 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1242 map_len = last_fsb - end_fsb;
1247 xfs_ilock(ip, XFS_ILOCK_SHARED);
1248 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1249 NULL, 0, &imap, &nimaps, NULL);
1250 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1252 if (!error && (nimaps != 0) &&
1253 (imap.br_startblock != HOLESTARTBLOCK)) {
1255 * Attach the dquots to the inode up front.
1257 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1261 * There are blocks after the end of file.
1262 * Free them up now by truncating the file to
1265 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1268 * Do the xfs_itruncate_start() call before
1269 * reserving any log space because
1270 * itruncate_start will call into the buffer
1271 * cache and we can't
1272 * do that within a transaction.
1274 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1275 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1278 error = xfs_trans_reserve(tp, 0,
1279 XFS_ITRUNCATE_LOG_RES(mp),
1280 0, XFS_TRANS_PERM_LOG_RES,
1281 XFS_ITRUNCATE_LOG_COUNT);
1283 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1284 xfs_trans_cancel(tp, 0);
1285 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1289 xfs_ilock(ip, XFS_ILOCK_EXCL);
1290 xfs_trans_ijoin(tp, ip,
1293 xfs_trans_ihold(tp, ip);
1295 error = xfs_itruncate_finish(&tp, ip,
1300 * If we get an error at this point we
1301 * simply don't bother truncating the file.
1304 xfs_trans_cancel(tp,
1305 (XFS_TRANS_RELEASE_LOG_RES |
1308 error = xfs_trans_commit(tp,
1309 XFS_TRANS_RELEASE_LOG_RES,
1312 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1318 * Free a symlink that has blocks associated with it.
1321 xfs_inactive_symlink_rmt(
1329 xfs_fsblock_t first_block;
1330 xfs_bmap_free_t free_list;
1333 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1341 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1343 * We're freeing a symlink that has some
1344 * blocks allocated to it. Free the
1345 * blocks here. We know that we've got
1346 * either 1 or 2 extents and that we can
1347 * free them all in one bunmapi call.
1349 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1350 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1351 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1352 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1353 xfs_trans_cancel(tp, 0);
1358 * Lock the inode, fix the size, and join it to the transaction.
1359 * Hold it so in the normal path, we still have it locked for
1360 * the second transaction. In the error paths we need it
1361 * held so the cancel won't rele it, see below.
1363 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1364 size = (int)ip->i_d.di_size;
1365 ip->i_d.di_size = 0;
1366 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1367 xfs_trans_ihold(tp, ip);
1368 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1370 * Find the block(s) so we can inval and unmap them.
1373 XFS_BMAP_INIT(&free_list, &first_block);
1374 nmaps = sizeof(mval) / sizeof(mval[0]);
1375 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1376 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1380 * Invalidate the block(s).
1382 for (i = 0; i < nmaps; i++) {
1383 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1384 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1385 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1386 xfs_trans_binval(tp, bp);
1389 * Unmap the dead block(s) to the free_list.
1391 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1392 &first_block, &free_list, &done)))
1396 * Commit the first transaction. This logs the EFI and the inode.
1398 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1401 * The transaction must have been committed, since there were
1402 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1403 * The new tp has the extent freeing and EFDs.
1407 * The first xact was committed, so add the inode to the new one.
1408 * Mark it dirty so it will be logged and moved forward in the log as
1409 * part of every commit.
1411 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1412 xfs_trans_ihold(tp, ip);
1413 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1415 * Get a new, empty transaction to return to our caller.
1417 ntp = xfs_trans_dup(tp);
1419 * Commit the transaction containing extent freeing and EFD's.
1420 * If we get an error on the commit here or on the reserve below,
1421 * we need to unlock the inode since the new transaction doesn't
1422 * have the inode attached.
1424 error = xfs_trans_commit(tp, 0, NULL);
1427 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1431 * Remove the memory for extent descriptions (just bookkeeping).
1433 if (ip->i_df.if_bytes)
1434 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1435 ASSERT(ip->i_df.if_bytes == 0);
1437 * Put an itruncate log reservation in the new transaction
1440 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1441 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1442 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1446 * Return with the inode locked but not joined to the transaction.
1452 xfs_bmap_cancel(&free_list);
1455 * Have to come here with the inode locked and either
1456 * (held and in the transaction) or (not in the transaction).
1457 * If the inode isn't held then cancel would iput it, but
1458 * that's wrong since this is inactive and the vnode ref
1459 * count is 0 already.
1460 * Cancel won't do anything to the inode if held, but it still
1461 * needs to be locked until the cancel is done, if it was
1462 * joined to the transaction.
1464 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1465 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1472 xfs_inactive_symlink_local(
1478 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1480 * We're freeing a symlink which fit into
1481 * the inode. Just free the memory used
1482 * to hold the old symlink.
1484 error = xfs_trans_reserve(*tpp, 0,
1485 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1486 0, XFS_TRANS_PERM_LOG_RES,
1487 XFS_ITRUNCATE_LOG_COUNT);
1490 xfs_trans_cancel(*tpp, 0);
1494 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1497 * Zero length symlinks _can_ exist.
1499 if (ip->i_df.if_bytes > 0) {
1500 xfs_idata_realloc(ip,
1501 -(ip->i_df.if_bytes),
1503 ASSERT(ip->i_df.if_bytes == 0);
1520 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1523 ASSERT(ip->i_d.di_forkoff != 0);
1524 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1525 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1527 error = xfs_attr_inactive(ip);
1530 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1531 return (error); /* goto out*/
1534 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1535 error = xfs_trans_reserve(tp, 0,
1536 XFS_IFREE_LOG_RES(mp),
1537 0, XFS_TRANS_PERM_LOG_RES,
1538 XFS_INACTIVE_LOG_COUNT);
1540 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1541 xfs_trans_cancel(tp, 0);
1543 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1547 xfs_ilock(ip, XFS_ILOCK_EXCL);
1548 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1549 xfs_trans_ihold(tp, ip);
1550 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1552 ASSERT(ip->i_d.di_anextents == 0);
1567 vp = BHV_TO_VNODE(bdp);
1568 ip = XFS_BHVTOI(bdp);
1570 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1574 /* If this is a read-only mount, don't do this (would generate I/O) */
1575 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1578 #ifdef HAVE_REFCACHE
1579 /* If we are in the NFS reference cache then don't do this now */
1586 if (ip->i_d.di_nlink != 0) {
1587 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1588 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1589 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1590 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1591 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1593 /* Update linux inode block count after free above */
1594 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1595 ip->i_d.di_nblocks + ip->i_delayed_blks);
1605 * This is called when the vnode reference count for the vnode
1606 * goes to zero. If the file has been unlinked, then it must
1607 * now be truncated. Also, we clear all of the read-ahead state
1608 * kept for the inode here since the file is now closed.
1617 xfs_bmap_free_t free_list;
1618 xfs_fsblock_t first_block;
1625 vp = BHV_TO_VNODE(bdp);
1626 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1628 ip = XFS_BHVTOI(bdp);
1631 * If the inode is already free, then there can be nothing
1634 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1635 ASSERT(ip->i_df.if_real_bytes == 0);
1636 ASSERT(ip->i_df.if_broot_bytes == 0);
1637 return VN_INACTIVE_CACHE;
1641 * Only do a truncate if it's a regular file with
1642 * some actual space in it. It's OK to look at the
1643 * inode's fields without the lock because we're the
1644 * only one with a reference to the inode.
1646 truncate = ((ip->i_d.di_nlink == 0) &&
1647 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1648 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1652 if (ip->i_d.di_nlink == 0 &&
1653 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1654 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1659 /* If this is a read-only mount, don't do this (would generate I/O) */
1660 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1663 if (ip->i_d.di_nlink != 0) {
1664 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1665 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1666 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1667 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1668 (ip->i_delayed_blks != 0))) {
1669 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1670 return (VN_INACTIVE_CACHE);
1671 /* Update linux inode block count after free above */
1672 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1673 ip->i_d.di_nblocks + ip->i_delayed_blks);
1678 ASSERT(ip->i_d.di_nlink == 0);
1680 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1681 return (VN_INACTIVE_CACHE);
1683 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1686 * Do the xfs_itruncate_start() call before
1687 * reserving any log space because itruncate_start
1688 * will call into the buffer cache and we can't
1689 * do that within a transaction.
1691 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1693 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1695 error = xfs_trans_reserve(tp, 0,
1696 XFS_ITRUNCATE_LOG_RES(mp),
1697 0, XFS_TRANS_PERM_LOG_RES,
1698 XFS_ITRUNCATE_LOG_COUNT);
1700 /* Don't call itruncate_cleanup */
1701 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1702 xfs_trans_cancel(tp, 0);
1703 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1704 return (VN_INACTIVE_CACHE);
1707 xfs_ilock(ip, XFS_ILOCK_EXCL);
1708 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1709 xfs_trans_ihold(tp, ip);
1712 * normally, we have to run xfs_itruncate_finish sync.
1713 * But if filesystem is wsync and we're in the inactive
1714 * path, then we know that nlink == 0, and that the
1715 * xaction that made nlink == 0 is permanently committed
1716 * since xfs_remove runs as a synchronous transaction.
1718 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1719 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1722 xfs_trans_cancel(tp,
1723 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1724 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1725 return (VN_INACTIVE_CACHE);
1727 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1730 * If we get an error while cleaning up a
1731 * symlink we bail out.
1733 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1734 xfs_inactive_symlink_rmt(ip, &tp) :
1735 xfs_inactive_symlink_local(ip, &tp);
1739 return (VN_INACTIVE_CACHE);
1742 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1743 xfs_trans_ihold(tp, ip);
1745 error = xfs_trans_reserve(tp, 0,
1746 XFS_IFREE_LOG_RES(mp),
1747 0, XFS_TRANS_PERM_LOG_RES,
1748 XFS_INACTIVE_LOG_COUNT);
1750 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1751 xfs_trans_cancel(tp, 0);
1752 return (VN_INACTIVE_CACHE);
1755 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1756 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1757 xfs_trans_ihold(tp, ip);
1761 * If there are attributes associated with the file
1762 * then blow them away now. The code calls a routine
1763 * that recursively deconstructs the attribute fork.
1764 * We need to just commit the current transaction
1765 * because we can't use it for xfs_attr_inactive().
1767 if (ip->i_d.di_anextents > 0) {
1768 error = xfs_inactive_attrs(ip, &tp);
1770 * If we got an error, the transaction is already
1771 * cancelled, and the inode is unlocked. Just get out.
1774 return (VN_INACTIVE_CACHE);
1775 } else if (ip->i_afp) {
1776 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1782 XFS_BMAP_INIT(&free_list, &first_block);
1783 error = xfs_ifree(tp, ip, &free_list);
1786 * If we fail to free the inode, shut down. The cancel
1787 * might do that, we need to make sure. Otherwise the
1788 * inode might be lost for a long time or forever.
1790 if (!XFS_FORCED_SHUTDOWN(mp)) {
1792 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1793 error, mp->m_fsname);
1794 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1796 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1799 * Credit the quota account(s). The inode is gone.
1801 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1804 * Just ignore errors at this point. There is
1805 * nothing we can do except to try to keep going.
1807 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1809 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1812 * Release the dquots held by inode, if any.
1814 XFS_QM_DQDETACH(mp, ip);
1816 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1819 return VN_INACTIVE_CACHE;
1828 bhv_desc_t *dir_bdp,
1835 xfs_inode_t *dp, *ip;
1841 dir_vp = BHV_TO_VNODE(dir_bdp);
1842 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1844 dp = XFS_BHVTOI(dir_bdp);
1846 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1847 return XFS_ERROR(EIO);
1849 lock_mode = xfs_ilock_map_shared(dp);
1850 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1852 *vpp = XFS_ITOV(ip);
1855 xfs_iunlock_map_shared(dp, lock_mode);
1861 * xfs_create (create a new file).
1865 bhv_desc_t *dir_bdp,
1871 char *name = VNAME(dentry);
1873 xfs_inode_t *dp, *ip;
1879 xfs_bmap_free_t free_list;
1880 xfs_fsblock_t first_block;
1881 boolean_t dp_joined_to_trans;
1882 int dm_event_sent = 0;
1886 struct xfs_dquot *udqp, *gdqp;
1892 dir_vp = BHV_TO_VNODE(dir_bdp);
1893 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1895 dp = XFS_BHVTOI(dir_bdp);
1898 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1899 namelen = VNAMELEN(dentry);
1901 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1902 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1903 dir_vp, DM_RIGHT_NULL, NULL,
1904 DM_RIGHT_NULL, name, NULL,
1912 if (XFS_FORCED_SHUTDOWN(mp))
1913 return XFS_ERROR(EIO);
1915 /* Return through std_return after this point. */
1918 if (vap->va_mask & XFS_AT_PROJID)
1919 prid = (xfs_prid_t)vap->va_projid;
1921 prid = (xfs_prid_t)dfltprid;
1924 * Make sure that we have allocated dquot(s) on disk.
1926 error = XFS_QM_DQVOPALLOC(mp, dp,
1927 current_fsuid(credp), current_fsgid(credp), prid,
1928 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1933 dp_joined_to_trans = B_FALSE;
1935 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1936 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1937 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1939 * Initially assume that the file does not exist and
1940 * reserve the resources for that case. If that is not
1941 * the case we'll drop the one we have and get a more
1942 * appropriate transaction later.
1944 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1945 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1946 if (error == ENOSPC) {
1948 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1949 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1957 xfs_ilock(dp, XFS_ILOCK_EXCL);
1959 XFS_BMAP_INIT(&free_list, &first_block);
1964 * Reserve disk quota and the inode.
1966 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1971 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1973 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1974 error = xfs_dir_ialloc(&tp, dp,
1975 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1976 rdev, credp, prid, resblks > 0,
1979 if (error == ENOSPC)
1986 * At this point, we've gotten a newly allocated inode.
1987 * It is locked (and joined to the transaction).
1990 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1993 * Now we join the directory inode to the transaction.
1994 * We do not do it earlier because xfs_dir_ialloc
1995 * might commit the previous transaction (and release
2000 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2001 dp_joined_to_trans = B_TRUE;
2003 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
2004 &first_block, &free_list,
2005 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2007 ASSERT(error != ENOSPC);
2010 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2011 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2014 * If this is a synchronous mount, make sure that the
2015 * create transaction goes to disk before returning to
2018 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2019 xfs_trans_set_sync(tp);
2025 * Attach the dquot(s) to the inodes and modify them incore.
2026 * These ids of the inode couldn't have changed since the new
2027 * inode has been locked ever since it was created.
2029 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2032 * xfs_trans_commit normally decrements the vnode ref count
2033 * when it unlocks the inode. Since we want to return the
2034 * vnode to the caller, we bump the vnode ref count now.
2039 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2041 xfs_bmap_cancel(&free_list);
2045 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2052 XFS_QM_DQRELE(mp, udqp);
2053 XFS_QM_DQRELE(mp, gdqp);
2056 * Propogate the fact that the vnode changed after the
2057 * xfs_inode locks have been released.
2059 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2063 /* Fallthrough to std_return with error = 0 */
2066 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2067 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2068 DM_EVENT_POSTCREATE)) {
2069 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2070 dir_vp, DM_RIGHT_NULL,
2072 DM_RIGHT_NULL, name, NULL,
2073 dm_di_mode, error, 0);
2078 cancel_flags |= XFS_TRANS_ABORT;
2083 xfs_trans_cancel(tp, cancel_flags);
2085 if (!dp_joined_to_trans && (dp != NULL))
2086 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2087 XFS_QM_DQRELE(mp, udqp);
2088 XFS_QM_DQRELE(mp, gdqp);
2094 * Wait until after the current transaction is aborted to
2095 * release the inode. This prevents recursive transactions
2096 * and deadlocks from xfs_inactive.
2098 cancel_flags |= XFS_TRANS_ABORT;
2099 xfs_trans_cancel(tp, cancel_flags);
2102 XFS_QM_DQRELE(mp, udqp);
2103 XFS_QM_DQRELE(mp, gdqp);
2110 * Some counters to see if (and how often) we are hitting some deadlock
2111 * prevention code paths.
2115 int xfs_rm_lock_delays;
2116 int xfs_rm_attempts;
2120 * The following routine will lock the inodes associated with the
2121 * directory and the named entry in the directory. The locks are
2122 * acquired in increasing inode number.
2124 * If the entry is "..", then only the directory is locked. The
2125 * vnode ref count will still include that from the .. entry in
2128 * There is a deadlock we need to worry about. If the locked directory is
2129 * in the AIL, it might be blocking up the log. The next inode we lock
2130 * could be already locked by another thread waiting for log space (e.g
2131 * a permanent log reservation with a long running transaction (see
2132 * xfs_itruncate_finish)). To solve this, we must check if the directory
2133 * is in the ail and use lock_nowait. If we can't lock, we need to
2134 * drop the inode lock on the directory and try again. xfs_iunlock will
2135 * potentially push the tail if we were holding up the log.
2138 xfs_lock_dir_and_entry(
2141 xfs_inode_t *ip) /* inode of entry 'name' */
2145 xfs_inode_t *ips[2];
2154 xfs_ilock(dp, XFS_ILOCK_EXCL);
2161 * We want to lock in increasing inum. Since we've already
2162 * acquired the lock on the directory, we may need to release
2163 * if if the inum of the entry turns out to be less.
2165 if (e_inum > dp->i_ino) {
2167 * We are already in the right order, so just
2168 * lock on the inode of the entry.
2169 * We need to use nowait if dp is in the AIL.
2172 lp = (xfs_log_item_t *)dp->i_itemp;
2173 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2174 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2181 * Unlock dp and try again.
2182 * xfs_iunlock will try to push the tail
2183 * if the inode is in the AIL.
2186 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2188 if ((attempts % 5) == 0) {
2189 delay(1); /* Don't just spin the CPU */
2191 xfs_rm_lock_delays++;
2197 xfs_ilock(ip, XFS_ILOCK_EXCL);
2199 } else if (e_inum < dp->i_ino) {
2200 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2204 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2206 /* else e_inum == dp->i_ino */
2207 /* This can happen if we're asked to lock /x/..
2208 * the entry is "..", which is also the parent directory.
2216 int xfs_small_retries;
2217 int xfs_middle_retries;
2218 int xfs_lots_retries;
2219 int xfs_lock_delays;
2223 * The following routine will lock n inodes in exclusive mode.
2224 * We assume the caller calls us with the inodes in i_ino order.
2226 * We need to detect deadlock where an inode that we lock
2227 * is in the AIL and we start waiting for another inode that is locked
2228 * by a thread in a long running transaction (such as truncate). This can
2229 * result in deadlock since the long running trans might need to wait
2230 * for the inode we just locked in order to push the tail and free space
2240 int attempts = 0, i, j, try_lock;
2243 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2254 for (; i < inodes; i++) {
2257 if (i && (ips[i] == ips[i-1])) /* Already locked */
2261 * If try_lock is not set yet, make sure all locked inodes
2262 * are not in the AIL.
2263 * If any are, set try_lock to be used later.
2267 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2268 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2269 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2276 * If any of the previous locks we have locked is in the AIL,
2277 * we must TRY to get the second and subsequent locks. If
2278 * we can't get any, we must release all we have
2283 /* try_lock must be 0 if i is 0. */
2285 * try_lock means we have an inode locked
2286 * that is in the AIL.
2289 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2293 * Unlock all previous guys and try again.
2294 * xfs_iunlock will try to push the tail
2295 * if the inode is in the AIL.
2298 for(j = i - 1; j >= 0; j--) {
2301 * Check to see if we've already
2302 * unlocked this one.
2303 * Not the first one going back,
2304 * and the inode ptr is the same.
2306 if ((j != (i - 1)) && ips[j] ==
2310 xfs_iunlock(ips[j], lock_mode);
2313 if ((attempts % 5) == 0) {
2314 delay(1); /* Don't just spin the CPU */
2324 xfs_ilock(ips[i], lock_mode);
2330 if (attempts < 5) xfs_small_retries++;
2331 else if (attempts < 100) xfs_middle_retries++;
2332 else xfs_lots_retries++;
2340 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2341 int remove_which_error_return = 0;
2343 #define REMOVE_DEBUG_TRACE(x)
2344 #endif /* ! DEBUG */
2353 bhv_desc_t *dir_bdp,
2358 char *name = VNAME(dentry);
2359 xfs_inode_t *dp, *ip;
2360 xfs_trans_t *tp = NULL;
2363 xfs_bmap_free_t free_list;
2364 xfs_fsblock_t first_block;
2372 dir_vp = BHV_TO_VNODE(dir_bdp);
2373 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2375 dp = XFS_BHVTOI(dir_bdp);
2378 if (XFS_FORCED_SHUTDOWN(mp))
2379 return XFS_ERROR(EIO);
2381 namelen = VNAMELEN(dentry);
2383 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2384 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2385 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2386 name, NULL, 0, 0, 0);
2391 /* From this point on, return through std_return */
2395 * We need to get a reference to ip before we get our log
2396 * reservation. The reason for this is that we cannot call
2397 * xfs_iget for an inode for which we do not have a reference
2398 * once we've acquired a log reservation. This is because the
2399 * inode we are trying to get might be in xfs_inactive going
2400 * for a log reservation. Since we'll have to wait for the
2401 * inactive code to complete before returning from xfs_iget,
2402 * we need to make sure that we don't have log space reserved
2403 * when we call xfs_iget. Instead we get an unlocked referece
2404 * to the inode before getting our log reservation.
2406 error = xfs_get_dir_entry(dentry, &ip);
2408 REMOVE_DEBUG_TRACE(__LINE__);
2412 dm_di_mode = ip->i_d.di_mode;
2414 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2418 error = XFS_QM_DQATTACH(mp, dp, 0);
2419 if (!error && dp != ip)
2420 error = XFS_QM_DQATTACH(mp, ip, 0);
2422 REMOVE_DEBUG_TRACE(__LINE__);
2427 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2428 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2430 * We try to get the real space reservation first,
2431 * allowing for directory btree deletion(s) implying
2432 * possible bmap insert(s). If we can't get the space
2433 * reservation then we use 0 instead, and avoid the bmap
2434 * btree insert(s) in the directory code by, if the bmap
2435 * insert tries to happen, instead trimming the LAST
2436 * block from the directory.
2438 resblks = XFS_REMOVE_SPACE_RES(mp);
2439 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2440 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2441 if (error == ENOSPC) {
2443 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2444 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2447 ASSERT(error != ENOSPC);
2448 REMOVE_DEBUG_TRACE(__LINE__);
2449 xfs_trans_cancel(tp, 0);
2454 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2456 REMOVE_DEBUG_TRACE(__LINE__);
2457 xfs_trans_cancel(tp, cancel_flags);
2463 * At this point, we've gotten both the directory and the entry
2466 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2469 * Increment vnode ref count only in this case since
2470 * there's an extra vnode reference in the case where
2474 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2478 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2480 XFS_BMAP_INIT(&free_list, &first_block);
2481 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2482 &first_block, &free_list, 0);
2484 ASSERT(error != ENOENT);
2485 REMOVE_DEBUG_TRACE(__LINE__);
2488 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2491 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2493 error = xfs_droplink(tp, ip);
2495 REMOVE_DEBUG_TRACE(__LINE__);
2499 /* Determine if this is the last link while
2500 * we are in the transaction.
2502 link_zero = (ip)->i_d.di_nlink==0;
2505 * Take an extra ref on the inode so that it doesn't
2506 * go to xfs_inactive() from within the commit.
2511 * If this is a synchronous mount, make sure that the
2512 * remove transaction goes to disk before returning to
2515 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2516 xfs_trans_set_sync(tp);
2519 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2521 REMOVE_DEBUG_TRACE(__LINE__);
2525 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2532 * Before we drop our extra reference to the inode, purge it
2533 * from the refcache if it is there. By waiting until afterwards
2534 * to do the IRELE, we ensure that we won't go inactive in the
2535 * xfs_refcache_purge_ip routine (although that would be OK).
2537 xfs_refcache_purge_ip(ip);
2539 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2542 * Let interposed file systems know about removed links.
2544 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2548 /* Fall through to std_return with error = 0 */
2550 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2551 DM_EVENT_POSTREMOVE)) {
2552 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2553 dir_vp, DM_RIGHT_NULL,
2554 NULL, DM_RIGHT_NULL,
2555 name, NULL, dm_di_mode, error, 0);
2560 xfs_bmap_cancel(&free_list);
2561 cancel_flags |= XFS_TRANS_ABORT;
2562 xfs_trans_cancel(tp, cancel_flags);
2567 * In this case make sure to not release the inode until after
2568 * the current transaction is aborted. Releasing it beforehand
2569 * can cause us to go to xfs_inactive and start a recursive
2570 * transaction which can easily deadlock with the current one.
2572 xfs_bmap_cancel(&free_list);
2573 cancel_flags |= XFS_TRANS_ABORT;
2574 xfs_trans_cancel(tp, cancel_flags);
2577 * Before we drop our extra reference to the inode, purge it
2578 * from the refcache if it is there. By waiting until afterwards
2579 * to do the IRELE, we ensure that we won't go inactive in the
2580 * xfs_refcache_purge_ip routine (although that would be OK).
2582 xfs_refcache_purge_ip(ip);
2596 bhv_desc_t *target_dir_bdp,
2601 xfs_inode_t *tdp, *sip;
2604 xfs_inode_t *ips[2];
2606 xfs_bmap_free_t free_list;
2607 xfs_fsblock_t first_block;
2610 vnode_t *target_dir_vp;
2611 bhv_desc_t *src_bdp;
2613 char *target_name = VNAME(dentry);
2616 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2617 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2618 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2620 target_namelen = VNAMELEN(dentry);
2621 if (src_vp->v_type == VDIR)
2622 return XFS_ERROR(EPERM);
2625 * For now, manually find the XFS behavior descriptor for
2626 * the source vnode. If it doesn't exist then something
2627 * is wrong and we should just return an error.
2628 * Eventually we need to figure out how link is going to
2629 * work in the face of stacked vnodes.
2631 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2632 if (src_bdp == NULL) {
2633 return XFS_ERROR(EXDEV);
2635 sip = XFS_BHVTOI(src_bdp);
2636 tdp = XFS_BHVTOI(target_dir_bdp);
2638 if (XFS_FORCED_SHUTDOWN(mp))
2639 return XFS_ERROR(EIO);
2641 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2642 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2643 target_dir_vp, DM_RIGHT_NULL,
2644 src_vp, DM_RIGHT_NULL,
2645 target_name, NULL, 0, 0, 0);
2650 /* Return through std_return after this point. */
2652 error = XFS_QM_DQATTACH(mp, sip, 0);
2653 if (!error && sip != tdp)
2654 error = XFS_QM_DQATTACH(mp, tdp, 0);
2658 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2659 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2660 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2661 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2662 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2663 if (error == ENOSPC) {
2665 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2666 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2673 if (sip->i_ino < tdp->i_ino) {
2681 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2684 * Increment vnode ref counts since xfs_trans_commit &
2685 * xfs_trans_cancel will both unlock the inodes and
2686 * decrement the associated ref counts.
2689 VN_HOLD(target_dir_vp);
2690 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2691 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2694 * If the source has too many links, we can't make any more to it.
2696 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2697 error = XFS_ERROR(EMLINK);
2702 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2706 XFS_BMAP_INIT(&free_list, &first_block);
2708 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2709 sip->i_ino, &first_block, &free_list,
2713 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2715 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2717 error = xfs_bumplink(tp, sip);
2723 * If this is a synchronous mount, make sure that the
2724 * link transaction goes to disk before returning to
2727 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2728 xfs_trans_set_sync(tp);
2731 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2733 xfs_bmap_cancel(&free_list);
2737 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2742 /* Fall through to std_return with error = 0. */
2744 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2745 DM_EVENT_POSTLINK)) {
2746 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2747 target_dir_vp, DM_RIGHT_NULL,
2748 src_vp, DM_RIGHT_NULL,
2749 target_name, NULL, 0, error, 0);
2754 cancel_flags |= XFS_TRANS_ABORT;
2757 xfs_trans_cancel(tp, cancel_flags);
2767 bhv_desc_t *dir_bdp,
2773 char *dir_name = VNAME(dentry);
2775 xfs_inode_t *cdp; /* inode of created dir */
2776 vnode_t *cvp; /* vnode of created dir */
2782 xfs_bmap_free_t free_list;
2783 xfs_fsblock_t first_block;
2785 boolean_t dp_joined_to_trans;
2786 boolean_t created = B_FALSE;
2787 int dm_event_sent = 0;
2789 struct xfs_dquot *udqp, *gdqp;
2794 dir_vp = BHV_TO_VNODE(dir_bdp);
2795 dp = XFS_BHVTOI(dir_bdp);
2798 if (XFS_FORCED_SHUTDOWN(mp))
2799 return XFS_ERROR(EIO);
2801 dir_namelen = VNAMELEN(dentry);
2804 dp_joined_to_trans = B_FALSE;
2805 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2807 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2808 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2809 dir_vp, DM_RIGHT_NULL, NULL,
2810 DM_RIGHT_NULL, dir_name, NULL,
2817 /* Return through std_return after this point. */
2819 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2823 if (vap->va_mask & XFS_AT_PROJID)
2824 prid = (xfs_prid_t)vap->va_projid;
2826 prid = (xfs_prid_t)dfltprid;
2829 * Make sure that we have allocated dquot(s) on disk.
2831 error = XFS_QM_DQVOPALLOC(mp, dp,
2832 current_fsuid(credp), current_fsgid(credp), prid,
2833 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2837 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2838 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2839 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2840 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2841 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2842 if (error == ENOSPC) {
2844 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2845 XFS_TRANS_PERM_LOG_RES,
2846 XFS_MKDIR_LOG_COUNT);
2854 xfs_ilock(dp, XFS_ILOCK_EXCL);
2857 * Check for directory link count overflow.
2859 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2860 error = XFS_ERROR(EMLINK);
2865 * Reserve disk quota and the inode.
2867 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2872 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2875 * create the directory inode.
2877 error = xfs_dir_ialloc(&tp, dp,
2878 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2879 0, credp, prid, resblks > 0,
2882 if (error == ENOSPC)
2889 * Now we add the directory inode to the transaction.
2890 * We waited until now since xfs_dir_ialloc might start
2891 * a new transaction. Had we joined the transaction
2892 * earlier, the locks might have gotten released.
2895 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2896 dp_joined_to_trans = B_TRUE;
2898 XFS_BMAP_INIT(&free_list, &first_block);
2900 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2901 cdp->i_ino, &first_block, &free_list,
2902 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2904 ASSERT(error != ENOSPC);
2907 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2910 * Bump the in memory version number of the parent directory
2911 * so that other processes accessing it will recognize that
2912 * the directory has changed.
2916 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2922 error = xfs_bumplink(tp, dp);
2927 cvp = XFS_ITOV(cdp);
2935 * Attach the dquots to the new inode and modify the icount incore.
2937 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2940 * If this is a synchronous mount, make sure that the
2941 * mkdir transaction goes to disk before returning to
2944 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2945 xfs_trans_set_sync(tp);
2948 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2954 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2955 XFS_QM_DQRELE(mp, udqp);
2956 XFS_QM_DQRELE(mp, gdqp);
2961 /* Fall through to std_return with error = 0 or errno from
2962 * xfs_trans_commit. */
2965 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2966 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2967 DM_EVENT_POSTCREATE)) {
2968 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2969 dir_vp, DM_RIGHT_NULL,
2970 created ? XFS_ITOV(cdp):NULL,
2973 dm_di_mode, error, 0);
2979 xfs_bmap_cancel(&free_list);
2981 cancel_flags |= XFS_TRANS_ABORT;
2983 xfs_trans_cancel(tp, cancel_flags);
2984 XFS_QM_DQRELE(mp, udqp);
2985 XFS_QM_DQRELE(mp, gdqp);
2987 if (!dp_joined_to_trans && (dp != NULL)) {
2988 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3001 bhv_desc_t *dir_bdp,
3005 char *name = VNAME(dentry);
3007 xfs_inode_t *cdp; /* child directory */
3011 xfs_bmap_free_t free_list;
3012 xfs_fsblock_t first_block;
3021 dir_vp = BHV_TO_VNODE(dir_bdp);
3022 dp = XFS_BHVTOI(dir_bdp);
3025 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3027 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3028 return XFS_ERROR(EIO);
3029 namelen = VNAMELEN(dentry);
3031 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3032 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3033 dir_vp, DM_RIGHT_NULL,
3034 NULL, DM_RIGHT_NULL,
3035 name, NULL, 0, 0, 0);
3037 return XFS_ERROR(error);
3040 /* Return through std_return after this point. */
3045 * We need to get a reference to cdp before we get our log
3046 * reservation. The reason for this is that we cannot call
3047 * xfs_iget for an inode for which we do not have a reference
3048 * once we've acquired a log reservation. This is because the
3049 * inode we are trying to get might be in xfs_inactive going
3050 * for a log reservation. Since we'll have to wait for the
3051 * inactive code to complete before returning from xfs_iget,
3052 * we need to make sure that we don't have log space reserved
3053 * when we call xfs_iget. Instead we get an unlocked referece
3054 * to the inode before getting our log reservation.
3056 error = xfs_get_dir_entry(dentry, &cdp);
3058 REMOVE_DEBUG_TRACE(__LINE__);
3062 dm_di_mode = cdp->i_d.di_mode;
3065 * Get the dquots for the inodes.
3067 error = XFS_QM_DQATTACH(mp, dp, 0);
3068 if (!error && dp != cdp)
3069 error = XFS_QM_DQATTACH(mp, cdp, 0);
3072 REMOVE_DEBUG_TRACE(__LINE__);
3076 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3077 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3079 * We try to get the real space reservation first,
3080 * allowing for directory btree deletion(s) implying
3081 * possible bmap insert(s). If we can't get the space
3082 * reservation then we use 0 instead, and avoid the bmap
3083 * btree insert(s) in the directory code by, if the bmap
3084 * insert tries to happen, instead trimming the LAST
3085 * block from the directory.
3087 resblks = XFS_REMOVE_SPACE_RES(mp);
3088 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3089 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3090 if (error == ENOSPC) {
3092 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3093 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3096 ASSERT(error != ENOSPC);
3101 XFS_BMAP_INIT(&free_list, &first_block);
3104 * Now lock the child directory inode and the parent directory
3105 * inode in the proper order. This will take care of validating
3106 * that the directory entry for the child directory inode has
3107 * not changed while we were obtaining a log reservation.
3109 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3111 xfs_trans_cancel(tp, cancel_flags);
3116 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3119 * Only increment the parent directory vnode count if
3120 * we didn't bump it in looking up cdp. The only time
3121 * we don't bump it is when we're looking up ".".
3127 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3129 ASSERT(cdp->i_d.di_nlink >= 2);
3130 if (cdp->i_d.di_nlink != 2) {
3131 error = XFS_ERROR(ENOTEMPTY);
3134 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3135 error = XFS_ERROR(ENOTEMPTY);
3139 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3140 &first_block, &free_list, resblks);
3145 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3148 * Bump the in memory generation count on the parent
3149 * directory so that other can know that it has changed.
3154 * Drop the link from cdp's "..".
3156 error = xfs_droplink(tp, dp);
3162 * Drop the link from dp to cdp.
3164 error = xfs_droplink(tp, cdp);
3170 * Drop the "." link from cdp to self.
3172 error = xfs_droplink(tp, cdp);
3177 /* Determine these before committing transaction */
3178 last_cdp_link = (cdp)->i_d.di_nlink==0;
3181 * Take an extra ref on the child vnode so that it
3182 * does not go to xfs_inactive() from within the commit.
3187 * If this is a synchronous mount, make sure that the
3188 * rmdir transaction goes to disk before returning to
3191 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3192 xfs_trans_set_sync(tp);
3195 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3197 xfs_bmap_cancel(&free_list);
3198 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3204 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3212 * Let interposed file systems know about removed links.
3214 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3218 /* Fall through to std_return with error = 0 or the errno
3219 * from xfs_trans_commit. */
3221 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3222 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3223 dir_vp, DM_RIGHT_NULL,
3224 NULL, DM_RIGHT_NULL,
3225 name, NULL, dm_di_mode,
3231 xfs_bmap_cancel(&free_list);
3232 cancel_flags |= XFS_TRANS_ABORT;
3234 xfs_trans_cancel(tp, cancel_flags);
3242 * Read dp's entries starting at uiop->uio_offset and translate them into
3243 * bufsize bytes worth of struct dirents starting at bufbase.
3247 bhv_desc_t *dir_bdp,
3253 xfs_trans_t *tp = NULL;
3256 xfs_off_t start_offset;
3258 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3259 (inst_t *)__return_address);
3260 dp = XFS_BHVTOI(dir_bdp);
3262 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3263 return XFS_ERROR(EIO);
3266 lock_mode = xfs_ilock_map_shared(dp);
3267 start_offset = uiop->uio_offset;
3268 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3269 if (start_offset != uiop->uio_offset) {
3270 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3272 xfs_iunlock_map_shared(dp, lock_mode);
3283 bhv_desc_t *dir_bdp,
3296 xfs_bmap_free_t free_list;
3297 xfs_fsblock_t first_block;
3298 boolean_t dp_joined_to_trans;
3302 xfs_fileoff_t first_fsb;
3303 xfs_filblks_t fs_blocks;
3305 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3312 struct xfs_dquot *udqp, *gdqp;
3314 char *link_name = VNAME(dentry);
3318 dir_vp = BHV_TO_VNODE(dir_bdp);
3319 dp = XFS_BHVTOI(dir_bdp);
3320 dp_joined_to_trans = B_FALSE;
3325 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3329 if (XFS_FORCED_SHUTDOWN(mp))
3330 return XFS_ERROR(EIO);
3332 link_namelen = VNAMELEN(dentry);
3335 * Check component lengths of the target path name.
3337 pathlen = strlen(target_path);
3338 if (pathlen >= MAXPATHLEN) /* total string too long */
3339 return XFS_ERROR(ENAMETOOLONG);
3340 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3344 for(total = 0, path = target_path; total < pathlen;) {
3348 while(*path == '/') {
3354 * Count up to the next slash or end of path.
3355 * Error out if the component is bigger than MAXNAMELEN.
3357 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3358 if (++len >= MAXNAMELEN) {
3359 error = ENAMETOOLONG;
3366 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3367 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3368 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3369 link_name, target_path, 0, 0, 0);
3374 /* Return through std_return after this point. */
3377 if (vap->va_mask & XFS_AT_PROJID)
3378 prid = (xfs_prid_t)vap->va_projid;
3380 prid = (xfs_prid_t)dfltprid;
3383 * Make sure that we have allocated dquot(s) on disk.
3385 error = XFS_QM_DQVOPALLOC(mp, dp,
3386 current_fsuid(credp), current_fsgid(credp), prid,
3387 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3391 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3392 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3394 * The symlink will fit into the inode data fork?
3395 * There can't be any attributes so we get the whole variable part.
3397 if (pathlen <= XFS_LITINO(mp))
3400 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3401 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3402 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3403 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3404 if (error == ENOSPC && fs_blocks == 0) {
3406 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3407 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3415 xfs_ilock(dp, XFS_ILOCK_EXCL);
3418 * Check whether the directory allows new symlinks or not.
3420 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3421 error = XFS_ERROR(EPERM);
3426 * Reserve disk quota : blocks and inode.
3428 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3433 * Check for ability to enter directory entry, if no space reserved.
3436 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3439 * Initialize the bmap freelist prior to calling either
3440 * bmapi or the directory create code.
3442 XFS_BMAP_INIT(&free_list, &first_block);
3445 * Allocate an inode for the symlink.
3447 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3448 1, 0, credp, prid, resblks > 0, &ip, NULL);
3450 if (error == ENOSPC)
3457 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3458 dp_joined_to_trans = B_TRUE;
3461 * Also attach the dquot(s) to it, if applicable.
3463 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3466 resblks -= XFS_IALLOC_SPACE_RES(mp);
3468 * If the symlink will fit into the inode, write it inline.
3470 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3471 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3472 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3473 ip->i_d.di_size = pathlen;
3476 * The inode was initially created in extent format.
3478 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3479 ip->i_df.if_flags |= XFS_IFINLINE;
3481 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3482 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3486 nmaps = SYMLINK_MAPS;
3488 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3489 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3490 &first_block, resblks, mval, &nmaps,
3497 resblks -= fs_blocks;
3498 ip->i_d.di_size = pathlen;
3499 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3501 cur_chunk = target_path;
3502 for (n = 0; n < nmaps; n++) {
3503 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3504 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3505 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3506 BTOBB(byte_cnt), 0);
3507 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3508 if (pathlen < byte_cnt) {
3511 pathlen -= byte_cnt;
3513 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3514 cur_chunk += byte_cnt;
3516 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3521 * Create the directory entry for the symlink.
3523 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3524 ip->i_ino, &first_block, &free_list, resblks);
3528 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3529 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3532 * Bump the in memory version number of the parent directory
3533 * so that other processes accessing it will recognize that
3534 * the directory has changed.
3539 * If this is a synchronous mount, make sure that the
3540 * symlink transaction goes to disk before returning to
3543 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3544 xfs_trans_set_sync(tp);
3548 * xfs_trans_commit normally decrements the vnode ref count
3549 * when it unlocks the inode. Since we want to return the
3550 * vnode to the caller, we bump the vnode ref count now.
3554 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3558 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3559 XFS_QM_DQRELE(mp, udqp);
3560 XFS_QM_DQRELE(mp, gdqp);
3562 /* Fall through to std_return with error = 0 or errno from
3563 * xfs_trans_commit */
3565 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3566 DM_EVENT_POSTSYMLINK)) {
3567 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3568 dir_vp, DM_RIGHT_NULL,
3569 error ? NULL : XFS_ITOV(ip),
3570 DM_RIGHT_NULL, link_name, target_path,
3586 xfs_bmap_cancel(&free_list);
3587 cancel_flags |= XFS_TRANS_ABORT;
3589 xfs_trans_cancel(tp, cancel_flags);
3590 XFS_QM_DQRELE(mp, udqp);
3591 XFS_QM_DQRELE(mp, gdqp);
3593 if (!dp_joined_to_trans && (dp != NULL)) {
3594 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3604 * A fid routine that takes a pointer to a previously allocated
3605 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3615 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3616 (inst_t *)__return_address);
3617 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3619 xfid = (xfs_fid2_t *)fidp;
3620 ip = XFS_BHVTOI(bdp);
3621 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3624 * use memcpy because the inode is a long long and there's no
3625 * assurance that xfid->fid_ino is properly aligned.
3627 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3628 xfid->fid_gen = ip->i_d.di_gen;
3645 vp = BHV_TO_VNODE(bdp);
3646 if (vp->v_type == VDIR)
3648 ip = XFS_BHVTOI(bdp);
3649 if (locktype == VRWLOCK_WRITE) {
3650 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3651 } else if (locktype == VRWLOCK_TRY_READ) {
3652 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3653 } else if (locktype == VRWLOCK_TRY_WRITE) {
3654 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3656 ASSERT((locktype == VRWLOCK_READ) ||
3657 (locktype == VRWLOCK_WRITE_DIRECT));
3658 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3676 vp = BHV_TO_VNODE(bdp);
3677 if (vp->v_type == VDIR)
3679 ip = XFS_BHVTOI(bdp);
3680 if (locktype == VRWLOCK_WRITE) {
3682 * In the write case, we may have added a new entry to
3683 * the reference cache. This might store a pointer to
3684 * an inode to be released in this inode. If it is there,
3685 * clear the pointer and release the inode after unlocking
3688 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3690 ASSERT((locktype == VRWLOCK_READ) ||
3691 (locktype == VRWLOCK_WRITE_DIRECT));
3692 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3704 xfs_inode_log_item_t *iip;
3707 ip = XFS_BHVTOI(bdp);
3711 if (XFS_FORCED_SHUTDOWN(mp))
3712 return XFS_ERROR(EIO);
3715 * Bypass inodes which have already been cleaned by
3716 * the inode flush clustering code inside xfs_iflush
3718 if ((ip->i_update_core == 0) &&
3719 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3722 if (flags & FLUSH_LOG) {
3723 if (iip && iip->ili_last_lsn) {
3724 xlog_t *log = mp->m_log;
3726 int s, log_flags = XFS_LOG_FORCE;
3728 s = GRANT_LOCK(log);
3729 sync_lsn = log->l_last_sync_lsn;
3730 GRANT_UNLOCK(log, s);
3732 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3735 if (flags & FLUSH_SYNC)
3736 log_flags |= XFS_LOG_SYNC;
3737 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3742 * We make this non-blocking if the inode is contended,
3743 * return EAGAIN to indicate to the caller that they
3744 * did not succeed. This prevents the flush path from
3745 * blocking on inodes inside another operation right
3746 * now, they get caught later by xfs_sync.
3748 if (flags & FLUSH_INODE) {
3751 if (xfs_ipincount(ip))
3754 if (flags & FLUSH_SYNC) {
3755 xfs_ilock(ip, XFS_ILOCK_SHARED);
3757 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3758 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3759 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3766 if (flags & FLUSH_SYNC)
3767 flush_flags = XFS_IFLUSH_SYNC;
3769 flush_flags = XFS_IFLUSH_ASYNC;
3771 error = xfs_iflush(ip, flush_flags);
3772 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3791 if (!capable(CAP_SYS_ADMIN))
3792 return XFS_ERROR(EPERM);
3794 ip = XFS_BHVTOI(bdp);
3797 if (XFS_FORCED_SHUTDOWN(mp))
3798 return XFS_ERROR(EIO);
3800 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3801 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3803 xfs_trans_cancel(tp, 0);
3806 xfs_ilock(ip, XFS_ILOCK_EXCL);
3807 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3809 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3810 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3812 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3814 error = xfs_trans_commit(tp, 0, NULL);
3830 vp = BHV_TO_VNODE(bdp);
3831 ip = XFS_BHVTOI(bdp);
3833 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3835 ASSERT(!VN_MAPPED(vp));
3837 /* bad inode, get out here ASAP */
3843 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3844 if (ip->i_d.di_size > 0) {
3846 * Flush and invalidate any data left around that is
3847 * a part of this file.
3849 * Get the inode's i/o lock so that buffers are pushed
3850 * out while holding the proper lock. We can't hold
3851 * the inode lock here since flushing out buffers may
3852 * cause us to try to get the lock in xfs_strategy().
3854 * We don't have to call remapf() here, because there
3855 * cannot be any mapped file references to this vnode
3856 * since it is being reclaimed.
3858 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3861 * If we hit an IO error, we need to make sure that the
3862 * buffer and page caches of file data for
3863 * the file are tossed away. We don't want to use
3864 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3865 * pages to stay attached to the vnode, but be
3866 * marked P_BAD. pdflush/vnode_pagebad
3869 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3870 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3872 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3875 ASSERT(VN_CACHED(vp) == 0);
3876 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3877 ip->i_delayed_blks == 0);
3878 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3879 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3881 * di_size field may not be quite accurate if we're
3884 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3885 ASSERT(VN_CACHED(vp) == 0);
3889 /* If we have nothing to flush with this inode then complete the
3890 * teardown now, otherwise break the link between the xfs inode
3891 * and the linux inode and clean up the xfs inode later. This
3892 * avoids flushing the inode to disk during the delete operation
3895 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3896 xfs_ilock(ip, XFS_ILOCK_EXCL);
3898 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3900 xfs_mount_t *mp = ip->i_mount;
3902 /* Protect sync from us */
3903 XFS_MOUNT_ILOCK(mp);
3904 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3905 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3906 ip->i_flags |= XFS_IRECLAIMABLE;
3907 XFS_MOUNT_IUNLOCK(mp);
3918 xfs_ihash_t *ih = ip->i_hash;
3919 vnode_t *vp = XFS_ITOV_NULL(ip);
3922 if (vp && VN_BAD(vp))
3925 /* The hash lock here protects a thread in xfs_iget_core from
3926 * racing with us on linking the inode back with a vnode.
3927 * Once we have the XFS_IRECLAIM flag set it will not touch
3930 write_lock(&ih->ih_lock);
3931 if ((ip->i_flags & XFS_IRECLAIM) ||
3932 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3933 write_unlock(&ih->ih_lock);
3936 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3940 ip->i_flags |= XFS_IRECLAIM;
3941 write_unlock(&ih->ih_lock);
3944 * If the inode is still dirty, then flush it out. If the inode
3945 * is not in the AIL, then it will be OK to flush it delwri as
3946 * long as xfs_iflush() does not keep any references to the inode.
3947 * We leave that decision up to xfs_iflush() since it has the
3948 * knowledge of whether it's OK to simply do a delwri flush of
3949 * the inode or whether we need to wait until the inode is
3950 * pulled from the AIL.
3951 * We get the flush lock regardless, though, just to make sure
3952 * we don't free it while it is being flushed.
3954 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3956 xfs_ilock(ip, XFS_ILOCK_EXCL);
3960 if (ip->i_update_core ||
3961 ((ip->i_itemp != NULL) &&
3962 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3963 error = xfs_iflush(ip, sync_mode);
3965 * If we hit an error, typically because of filesystem
3966 * shutdown, we don't need to let vn_reclaim to know
3967 * because we're gonna reclaim the inode anyway.
3970 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3973 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3976 ASSERT(ip->i_update_core == 0);
3977 ASSERT(ip->i_itemp == NULL ||
3978 ip->i_itemp->ili_format.ilf_fields == 0);
3979 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3980 } else if (locked) {
3982 * We are not interested in doing an iflush if we're
3983 * in the process of shutting down the filesystem forcibly.
3984 * So, just reclaim the inode.
3987 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3996 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3999 xfs_inode_t *ip, *n;
4004 XFS_MOUNT_ILOCK(mp);
4005 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
4007 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
4009 if (xfs_ipincount(ip) ||
4010 !xfs_iflock_nowait(ip)) {
4011 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4015 XFS_MOUNT_IUNLOCK(mp);
4016 xfs_finish_reclaim(ip, noblock,
4017 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
4025 XFS_MOUNT_IUNLOCK(mp);
4030 * xfs_alloc_file_space()
4031 * This routine allocates disk space for the given file.
4033 * If alloc_type == 0, this request is for an ALLOCSP type
4034 * request which will change the file size. In this case, no
4035 * DMAPI event will be generated by the call. A TRUNCATE event
4036 * will be generated later by xfs_setattr.
4038 * If alloc_type != 0, this request is for a RESVSP type
4039 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4040 * lower block boundary byte address is less than the file's
4049 xfs_alloc_file_space(
4056 xfs_filblks_t allocated_fsb;
4057 xfs_filblks_t allocatesize_fsb;
4060 xfs_filblks_t datablocks;
4062 xfs_fsblock_t firstfsb;
4063 xfs_bmap_free_t free_list;
4064 xfs_bmbt_irec_t *imapp;
4065 xfs_bmbt_irec_t imaps[1];
4072 xfs_fileoff_t startoffset_fsb;
4074 int xfs_bmapi_flags;
4076 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4079 if (XFS_FORCED_SHUTDOWN(mp))
4080 return XFS_ERROR(EIO);
4083 * determine if this is a realtime file
4085 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4086 if (ip->i_d.di_extsize)
4087 rtextsize = ip->i_d.di_extsize;
4089 rtextsize = mp->m_sb.sb_rextsize;
4093 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4097 return XFS_ERROR(EINVAL);
4103 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4104 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4105 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4107 /* Generate a DMAPI event if needed. */
4108 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4109 (attr_flags&ATTR_DMI) == 0 &&
4110 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4111 xfs_off_t end_dmi_offset;
4113 end_dmi_offset = offset+len;
4114 if (end_dmi_offset > ip->i_d.di_size)
4115 end_dmi_offset = ip->i_d.di_size;
4116 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4117 offset, end_dmi_offset - offset,
4124 * allocate file space until done or until there is an error
4127 while (allocatesize_fsb && !error) {
4129 * determine if reserving space on
4130 * the data or realtime partition.
4135 s = startoffset_fsb;
4136 do_div(s, rtextsize);
4138 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4140 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4143 datablocks = allocatesize_fsb;
4148 * allocate and setup the transaction
4150 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4151 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4152 error = xfs_trans_reserve(tp,
4154 XFS_WRITE_LOG_RES(mp),
4156 XFS_TRANS_PERM_LOG_RES,
4157 XFS_WRITE_LOG_COUNT);
4160 * check for running out of space
4164 * Free the transaction structure.
4166 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4167 xfs_trans_cancel(tp, 0);
4170 xfs_ilock(ip, XFS_ILOCK_EXCL);
4171 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4172 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4173 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4177 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4178 xfs_trans_ihold(tp, ip);
4181 * issue the bmapi() call to allocate the blocks
4183 XFS_BMAP_INIT(&free_list, &firstfsb);
4184 error = xfs_bmapi(tp, ip, startoffset_fsb,
4185 allocatesize_fsb, xfs_bmapi_flags,
4186 &firstfsb, 0, imapp, &reccount,
4193 * complete the transaction
4195 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4200 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4201 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4206 allocated_fsb = imapp->br_blockcount;
4208 if (reccount == 0) {
4209 error = XFS_ERROR(ENOSPC);
4213 startoffset_fsb += allocated_fsb;
4214 allocatesize_fsb -= allocated_fsb;
4217 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4218 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4220 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4221 XFS_ITOV(ip), DM_RIGHT_NULL,
4222 XFS_ITOV(ip), DM_RIGHT_NULL,
4223 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4225 goto retry; /* Maybe DMAPI app. has made space */
4226 /* else fall through with error from XFS_SEND_DATA */
4232 xfs_bmap_cancel(&free_list);
4234 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4235 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4236 goto dmapi_enospc_check;
4240 * Zero file bytes between startoff and endoff inclusive.
4241 * The iolock is held exclusive and no blocks are buffered.
4244 xfs_zero_remaining_bytes(
4249 xfs_bmbt_irec_t imap;
4250 xfs_fileoff_t offset_fsb;
4251 xfs_off_t lastoffset;
4254 xfs_mount_t *mp = ip->i_mount;
4258 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4259 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4260 mp->m_rtdev_targp : mp->m_ddev_targp);
4262 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4263 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4265 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4267 if (error || nimap < 1)
4269 ASSERT(imap.br_blockcount >= 1);
4270 ASSERT(imap.br_startoff == offset_fsb);
4271 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4272 if (lastoffset > endoff)
4273 lastoffset = endoff;
4274 if (imap.br_startblock == HOLESTARTBLOCK)
4276 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4277 if (imap.br_state == XFS_EXT_UNWRITTEN)
4280 XFS_BUF_UNWRITE(bp);
4282 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4284 if ((error = xfs_iowait(bp))) {
4285 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4286 mp, bp, XFS_BUF_ADDR(bp));
4289 memset(XFS_BUF_PTR(bp) +
4290 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4291 0, lastoffset - offset + 1);
4296 if ((error = xfs_iowait(bp))) {
4297 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4298 mp, bp, XFS_BUF_ADDR(bp));
4307 * xfs_free_file_space()
4308 * This routine frees disk space for the given file.
4310 * This routine is only called by xfs_change_file_space
4311 * for an UNRESVSP type call.
4319 xfs_free_file_space(
4327 xfs_off_t end_dmi_offset;
4328 xfs_fileoff_t endoffset_fsb;
4330 xfs_fsblock_t firstfsb;
4331 xfs_bmap_free_t free_list;
4333 xfs_bmbt_irec_t imap;
4341 xfs_fileoff_t startoffset_fsb;
4343 int need_iolock = 1;
4345 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4348 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4352 if (len <= 0) /* if nothing being freed */
4354 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4355 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4356 end_dmi_offset = offset + len;
4357 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4359 if (offset < ip->i_d.di_size &&
4360 (attr_flags & ATTR_DMI) == 0 &&
4361 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4362 if (end_dmi_offset > ip->i_d.di_size)
4363 end_dmi_offset = ip->i_d.di_size;
4364 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4365 offset, end_dmi_offset - offset,
4366 AT_DELAY_FLAG(attr_flags), NULL);
4371 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4372 if (attr_flags & ATTR_NOLOCK)
4375 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4377 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4379 ilen = len + (offset & (rounding - 1));
4380 ioffset = offset & ~(rounding - 1);
4381 if (ilen & (rounding - 1))
4382 ilen = (ilen + rounding) & ~(rounding - 1);
4383 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4385 * Need to zero the stuff we're not freeing, on disk.
4386 * If its a realtime file & can't use unwritten extents then we
4387 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4388 * will take care of it for us.
4390 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4392 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4393 &imap, &nimap, NULL);
4395 goto out_unlock_iolock;
4396 ASSERT(nimap == 0 || nimap == 1);
4397 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4400 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4401 block = imap.br_startblock;
4402 mod = do_div(block, mp->m_sb.sb_rextsize);
4404 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4407 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4408 &imap, &nimap, NULL);
4410 goto out_unlock_iolock;
4411 ASSERT(nimap == 0 || nimap == 1);
4412 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4413 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4415 if (mod && (mod != mp->m_sb.sb_rextsize))
4416 endoffset_fsb -= mod;
4419 if ((done = (endoffset_fsb <= startoffset_fsb)))
4421 * One contiguous piece to clear
4423 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4426 * Some full blocks, possibly two pieces to clear
4428 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4429 error = xfs_zero_remaining_bytes(ip, offset,
4430 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4432 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4433 error = xfs_zero_remaining_bytes(ip,
4434 XFS_FSB_TO_B(mp, endoffset_fsb),
4439 * free file space until done or until there is an error
4441 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4442 while (!error && !done) {
4445 * allocate and setup the transaction
4447 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4448 error = xfs_trans_reserve(tp,
4450 XFS_WRITE_LOG_RES(mp),
4452 XFS_TRANS_PERM_LOG_RES,
4453 XFS_WRITE_LOG_COUNT);
4456 * check for running out of space
4460 * Free the transaction structure.
4462 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4463 xfs_trans_cancel(tp, 0);
4466 xfs_ilock(ip, XFS_ILOCK_EXCL);
4467 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4468 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4469 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4473 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4474 xfs_trans_ihold(tp, ip);
4477 * issue the bunmapi() call to free the blocks
4479 XFS_BMAP_INIT(&free_list, &firstfsb);
4480 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4481 endoffset_fsb - startoffset_fsb,
4482 0, 2, &firstfsb, &free_list, &done);
4488 * complete the transaction
4490 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4495 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4496 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4501 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4505 xfs_bmap_cancel(&free_list);
4507 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4508 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4514 * xfs_change_file_space()
4515 * This routine allocates or frees disk space for the given file.
4516 * The user specified parameters are checked for alignment and size
4525 xfs_change_file_space(
4539 xfs_off_t startoffset;
4545 vp = BHV_TO_VNODE(bdp);
4546 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4548 ip = XFS_BHVTOI(bdp);
4552 * must be a regular file and have write permission
4554 if (vp->v_type != VREG)
4555 return XFS_ERROR(EINVAL);
4557 xfs_ilock(ip, XFS_ILOCK_SHARED);
4559 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4560 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4564 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4566 switch (bf->l_whence) {
4567 case 0: /*SEEK_SET*/
4569 case 1: /*SEEK_CUR*/
4570 bf->l_start += offset;
4572 case 2: /*SEEK_END*/
4573 bf->l_start += ip->i_d.di_size;
4576 return XFS_ERROR(EINVAL);
4579 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4581 if ( (bf->l_start < 0)
4582 || (bf->l_start > XFS_MAXIOFFSET(mp))
4583 || (bf->l_start + llen < 0)
4584 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4585 return XFS_ERROR(EINVAL);
4589 startoffset = bf->l_start;
4590 fsize = ip->i_d.di_size;
4593 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4595 * These calls do NOT zero the data space allocated to the file,
4596 * nor do they change the file size.
4598 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4600 * These calls cause the new file data to be zeroed and the file
4601 * size to be changed.
4603 setprealloc = clrprealloc = 0;
4606 case XFS_IOC_RESVSP:
4607 case XFS_IOC_RESVSP64:
4608 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4615 case XFS_IOC_UNRESVSP:
4616 case XFS_IOC_UNRESVSP64:
4617 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4622 case XFS_IOC_ALLOCSP:
4623 case XFS_IOC_ALLOCSP64:
4624 case XFS_IOC_FREESP:
4625 case XFS_IOC_FREESP64:
4626 if (startoffset > fsize) {
4627 error = xfs_alloc_file_space(ip, fsize,
4628 startoffset - fsize, 0, attr_flags);
4633 va.va_mask = XFS_AT_SIZE;
4634 va.va_size = startoffset;
4636 error = xfs_setattr(bdp, &va, attr_flags, credp);
4646 return XFS_ERROR(EINVAL);
4650 * update the inode timestamp, mode, and prealloc flag bits
4652 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4654 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4657 xfs_trans_cancel(tp, 0);
4661 xfs_ilock(ip, XFS_ILOCK_EXCL);
4663 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4664 xfs_trans_ihold(tp, ip);
4666 if ((attr_flags & ATTR_DMI) == 0) {
4667 ip->i_d.di_mode &= ~S_ISUID;
4670 * Note that we don't have to worry about mandatory
4671 * file locking being disabled here because we only
4672 * clear the S_ISGID bit if the Group execute bit is
4673 * on, but if it was on then mandatory locking wouldn't
4674 * have been enabled.
4676 if (ip->i_d.di_mode & S_IXGRP)
4677 ip->i_d.di_mode &= ~S_ISGID;
4679 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4682 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4683 else if (clrprealloc)
4684 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4686 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4687 xfs_trans_set_sync(tp);
4689 error = xfs_trans_commit(tp, 0, NULL);
4691 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4696 vnodeops_t xfs_vnodeops = {
4697 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4698 .vop_open = xfs_open,
4699 .vop_read = xfs_read,
4700 #ifdef HAVE_SENDFILE
4701 .vop_sendfile = xfs_sendfile,
4703 .vop_write = xfs_write,
4704 .vop_ioctl = xfs_ioctl,
4705 .vop_getattr = xfs_getattr,
4706 .vop_setattr = xfs_setattr,
4707 .vop_access = xfs_access,
4708 .vop_lookup = xfs_lookup,
4709 .vop_create = xfs_create,
4710 .vop_remove = xfs_remove,
4711 .vop_link = xfs_link,
4712 .vop_rename = xfs_rename,
4713 .vop_mkdir = xfs_mkdir,
4714 .vop_rmdir = xfs_rmdir,
4715 .vop_readdir = xfs_readdir,
4716 .vop_symlink = xfs_symlink,
4717 .vop_readlink = xfs_readlink,
4718 .vop_fsync = xfs_fsync,
4719 .vop_inactive = xfs_inactive,
4720 .vop_fid2 = xfs_fid2,
4721 .vop_rwlock = xfs_rwlock,
4722 .vop_rwunlock = xfs_rwunlock,
4723 .vop_bmap = xfs_bmap,
4724 .vop_reclaim = xfs_reclaim,
4725 .vop_attr_get = xfs_attr_get,
4726 .vop_attr_set = xfs_attr_set,
4727 .vop_attr_remove = xfs_attr_remove,
4728 .vop_attr_list = xfs_attr_list,
4729 .vop_link_removed = (vop_link_removed_t)fs_noval,
4730 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4731 .vop_tosspages = fs_tosspages,
4732 .vop_flushinval_pages = fs_flushinval_pages,
4733 .vop_flush_pages = fs_flush_pages,
4734 .vop_release = xfs_release,
4735 .vop_iflush = xfs_inode_flush,