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 (VN_ISDIR(vp) && 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_mode = ip->i_d.di_mode;
167 vap->va_uid = ip->i_d.di_uid;
168 vap->va_gid = ip->i_d.di_gid;
169 vap->va_projid = ip->i_d.di_projid;
172 * Check vnode type block/char vs. everything else.
174 switch (ip->i_d.di_mode & S_IFMT) {
177 vap->va_rdev = ip->i_df.if_u2.if_rdev;
178 vap->va_blocksize = BLKDEV_IOSIZE;
183 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
184 vap->va_blocksize = xfs_preferred_iosize(mp);
188 * If the file blocks are being allocated from a
189 * realtime partition, then return the inode's
190 * realtime extent size or the realtime volume's
193 vap->va_blocksize = ip->i_d.di_extsize ?
194 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
195 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
200 vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
201 vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
202 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
203 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
204 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
205 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
208 * Exit for stat callers. See if any of the rest of the fields
209 * to be filled in are needed.
212 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
213 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
217 * Convert di_flags to xflags.
219 vap->va_xflags = xfs_ip2xflags(ip);
222 * Exit for inode revalidate. See if any of the rest of
223 * the fields to be filled in are needed.
226 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
227 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
230 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
232 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
233 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
237 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
238 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
239 ip->i_d.di_anextents;
241 vap->va_anextents = 0;
242 vap->va_gen = ip->i_d.di_gen;
245 if (!(flags & ATTR_LAZY))
246 xfs_iunlock(ip, XFS_ILOCK_SHARED);
272 xfs_prid_t projid=0, iprojid=0;
273 int mandlock_before, mandlock_after;
274 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
278 vp = BHV_TO_VNODE(bdp);
279 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
281 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
282 return XFS_ERROR(EROFS);
285 * Cannot set certain attributes.
288 if (mask & XFS_AT_NOSET) {
289 return XFS_ERROR(EINVAL);
292 ip = XFS_BHVTOI(bdp);
295 if (XFS_FORCED_SHUTDOWN(mp))
296 return XFS_ERROR(EIO);
299 * Timestamps do not need to be logged and hence do not
300 * need to be done within a transaction.
302 if (mask & XFS_AT_UPDTIMES) {
303 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
304 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
305 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
306 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
307 xfs_ichgtime(ip, timeflags);
311 olddquot1 = olddquot2 = NULL;
315 * If disk quotas is on, we make sure that the dquots do exist on disk,
316 * before we start any other transactions. Trying to do this later
317 * is messy. We don't care to take a readlock to look at the ids
318 * in inode here, because we can't hold it across the trans_reserve.
319 * If the IDs do change before we take the ilock, we're covered
320 * because the i_*dquot fields will get updated anyway.
322 if (XFS_IS_QUOTA_ON(mp) &&
323 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
326 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
328 qflags |= XFS_QMOPT_UQUOTA;
330 uid = ip->i_d.di_uid;
332 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
334 qflags |= XFS_QMOPT_GQUOTA;
336 gid = ip->i_d.di_gid;
338 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
339 projid = vap->va_projid;
340 qflags |= XFS_QMOPT_PQUOTA;
342 projid = ip->i_d.di_projid;
345 * We take a reference when we initialize udqp and gdqp,
346 * so it is important that we never blindly double trip on
347 * the same variable. See xfs_create() for an example.
349 ASSERT(udqp == NULL);
350 ASSERT(gdqp == NULL);
351 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
358 * For the other attributes, we acquire the inode lock and
359 * first do an error checking pass.
362 lock_flags = XFS_ILOCK_EXCL;
363 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
364 if (flags & ATTR_NOLOCK)
366 if (!(mask & XFS_AT_SIZE)) {
367 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
368 (mp->m_flags & XFS_MOUNT_WSYNC)) {
369 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
371 if ((code = xfs_trans_reserve(tp, 0,
372 XFS_ICHANGE_LOG_RES(mp), 0,
379 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
380 !(flags & ATTR_DMI)) {
381 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
382 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
383 vap->va_size, 0, dmflags, NULL);
390 lock_flags |= XFS_IOLOCK_EXCL;
393 xfs_ilock(ip, lock_flags);
395 /* boolean: are we the file owner? */
396 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
399 * Change various properties of a file.
400 * Only the owner or users with CAP_FOWNER
401 * capability may do these things.
404 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
405 XFS_AT_GID|XFS_AT_PROJID)) {
407 * CAP_FOWNER overrides the following restrictions:
409 * The user ID of the calling process must be equal
410 * to the file owner ID, except in cases where the
411 * CAP_FSETID capability is applicable.
413 if (!file_owner && !capable(CAP_FOWNER)) {
414 code = XFS_ERROR(EPERM);
419 * CAP_FSETID overrides the following restrictions:
421 * The effective user ID of the calling process shall match
422 * the file owner when setting the set-user-ID and
423 * set-group-ID bits on that file.
425 * The effective group ID or one of the supplementary group
426 * IDs of the calling process shall match the group owner of
427 * the file when setting the set-group-ID bit on that file
429 if (mask & XFS_AT_MODE) {
432 if ((vap->va_mode & S_ISUID) && !file_owner)
434 if ((vap->va_mode & S_ISGID) &&
435 !in_group_p((gid_t)ip->i_d.di_gid))
438 /* Linux allows this, Irix doesn't. */
439 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
442 if (m && !capable(CAP_FSETID))
448 * Change file ownership. Must be the owner or privileged.
449 * If the system was configured with the "restricted_chown"
450 * option, the owner is not permitted to give away the file,
451 * and can change the group id only to a group of which he
452 * or she is a member.
454 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
456 * These IDs could have changed since we last looked at them.
457 * But, we're assured that if the ownership did change
458 * while we didn't have the inode locked, inode's dquot(s)
459 * would have changed also.
461 iuid = ip->i_d.di_uid;
462 iprojid = ip->i_d.di_projid;
463 igid = ip->i_d.di_gid;
464 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
465 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
466 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
470 * CAP_CHOWN overrides the following restrictions:
472 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
473 * shall override the restriction that a process cannot
474 * change the user ID of a file it owns and the restriction
475 * that the group ID supplied to the chown() function
476 * shall be equal to either the group ID or one of the
477 * supplementary group IDs of the calling process.
479 if (restricted_chown &&
480 (iuid != uid || (igid != gid &&
481 !in_group_p((gid_t)gid))) &&
482 !capable(CAP_CHOWN)) {
483 code = XFS_ERROR(EPERM);
487 * Do a quota reservation only if uid/projid/gid is actually
490 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
491 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
492 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
494 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
495 capable(CAP_FOWNER) ?
496 XFS_QMOPT_FORCE_RES : 0);
497 if (code) /* out of quota */
503 * Truncate file. Must have write permission and not be a directory.
505 if (mask & XFS_AT_SIZE) {
506 /* Short circuit the truncate case for zero length files */
507 if ((vap->va_size == 0) &&
508 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
509 xfs_iunlock(ip, XFS_ILOCK_EXCL);
510 lock_flags &= ~XFS_ILOCK_EXCL;
511 if (mask & XFS_AT_CTIME)
512 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
518 code = XFS_ERROR(EISDIR);
520 } else if (!VN_ISREG(vp)) {
521 code = XFS_ERROR(EINVAL);
525 * Make sure that the dquots are attached to the inode.
527 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
532 * Change file access or modified times.
534 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
536 if ((flags & ATTR_UTIME) &&
537 !capable(CAP_FOWNER)) {
538 code = XFS_ERROR(EPERM);
545 * Change extent size or realtime flag.
547 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
549 * Can't change extent size if any extents are allocated.
551 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
552 (mask & XFS_AT_EXTSIZE) &&
553 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
555 code = XFS_ERROR(EINVAL); /* EFBIG? */
560 * Can't set extent size unless the file is marked, or
561 * about to be marked as a realtime file.
563 * This check will be removed when fixed size extents
564 * with buffered data writes is implemented.
567 if ((mask & XFS_AT_EXTSIZE) &&
568 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
570 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
571 ((mask & XFS_AT_XFLAGS) &&
572 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
573 code = XFS_ERROR(EINVAL);
578 * Can't change realtime flag if any extents are allocated.
580 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
581 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
582 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
583 code = XFS_ERROR(EINVAL); /* EFBIG? */
587 * Extent size must be a multiple of the appropriate block
588 * size, if set at all.
590 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
593 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
594 ((mask & XFS_AT_XFLAGS) &&
595 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
596 size = mp->m_sb.sb_rextsize <<
597 mp->m_sb.sb_blocklog;
599 size = mp->m_sb.sb_blocksize;
601 if (vap->va_extsize % size) {
602 code = XFS_ERROR(EINVAL);
607 * If realtime flag is set then must have realtime data.
609 if ((mask & XFS_AT_XFLAGS) &&
610 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
611 if ((mp->m_sb.sb_rblocks == 0) ||
612 (mp->m_sb.sb_rextsize == 0) ||
613 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
614 code = XFS_ERROR(EINVAL);
620 * Can't modify an immutable/append-only file unless
621 * we have appropriate permission.
623 if ((mask & XFS_AT_XFLAGS) &&
625 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
627 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
628 !capable(CAP_LINUX_IMMUTABLE)) {
629 code = XFS_ERROR(EPERM);
635 * Now we can make the changes. Before we join the inode
636 * to the transaction, if XFS_AT_SIZE is set then take care of
637 * the part of the truncation that must be done without the
638 * inode lock. This needs to be done before joining the inode
639 * to the transaction, because the inode cannot be unlocked
640 * once it is a part of the transaction.
642 if (mask & XFS_AT_SIZE) {
644 if (vap->va_size > ip->i_d.di_size)
645 code = xfs_igrow_start(ip, vap->va_size, credp);
646 xfs_iunlock(ip, XFS_ILOCK_EXCL);
648 code = xfs_itruncate_data(ip, vap->va_size);
651 lock_flags &= ~XFS_ILOCK_EXCL;
652 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
655 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
656 if ((code = xfs_trans_reserve(tp, 0,
657 XFS_ITRUNCATE_LOG_RES(mp), 0,
658 XFS_TRANS_PERM_LOG_RES,
659 XFS_ITRUNCATE_LOG_COUNT))) {
660 xfs_trans_cancel(tp, 0);
662 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
665 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
666 xfs_ilock(ip, XFS_ILOCK_EXCL);
670 xfs_trans_ijoin(tp, ip, lock_flags);
671 xfs_trans_ihold(tp, ip);
674 /* determine whether mandatory locking mode changes */
675 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
678 * Truncate file. Must have write permission and not be a directory.
680 if (mask & XFS_AT_SIZE) {
681 if (vap->va_size > ip->i_d.di_size) {
682 xfs_igrow_finish(tp, ip, vap->va_size,
683 !(flags & ATTR_DMI));
684 } else if ((vap->va_size <= ip->i_d.di_size) ||
685 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
687 * signal a sync transaction unless
688 * we're truncating an already unlinked
689 * file on a wsync filesystem
691 code = xfs_itruncate_finish(&tp, ip,
692 (xfs_fsize_t)vap->va_size,
694 ((ip->i_d.di_nlink != 0 ||
695 !(mp->m_flags & XFS_MOUNT_WSYNC))
702 * Have to do this even if the file's size doesn't change.
704 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
708 * Change file access modes.
710 if (mask & XFS_AT_MODE) {
711 ip->i_d.di_mode &= S_IFMT;
712 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
714 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
715 timeflags |= XFS_ICHGTIME_CHG;
719 * Change file ownership. Must be the owner or privileged.
720 * If the system was configured with the "restricted_chown"
721 * option, the owner is not permitted to give away the file,
722 * and can change the group id only to a group of which he
723 * or she is a member.
725 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
727 * CAP_FSETID overrides the following restrictions:
729 * The set-user-ID and set-group-ID bits of a file will be
730 * cleared upon successful return from chown()
732 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
733 !capable(CAP_FSETID)) {
734 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
738 * Change the ownerships and register quota modifications
739 * in the transaction.
742 if (XFS_IS_UQUOTA_ON(mp)) {
743 ASSERT(mask & XFS_AT_UID);
745 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
746 &ip->i_udquot, udqp);
748 ip->i_d.di_uid = uid;
751 if (XFS_IS_GQUOTA_ON(mp)) {
752 ASSERT(!XFS_IS_PQUOTA_ON(mp));
753 ASSERT(mask & XFS_AT_GID);
755 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
756 &ip->i_gdquot, gdqp);
758 ip->i_d.di_gid = gid;
760 if (iprojid != projid) {
761 if (XFS_IS_PQUOTA_ON(mp)) {
762 ASSERT(!XFS_IS_GQUOTA_ON(mp));
763 ASSERT(mask & XFS_AT_PROJID);
765 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
766 &ip->i_gdquot, gdqp);
768 ip->i_d.di_projid = projid;
770 * We may have to rev the inode as well as
771 * the superblock version number since projids didn't
772 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
774 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
775 xfs_bump_ino_vers2(tp, ip);
778 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
779 timeflags |= XFS_ICHGTIME_CHG;
784 * Change file access or modified times.
786 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
787 if (mask & XFS_AT_ATIME) {
788 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
789 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
790 ip->i_update_core = 1;
791 timeflags &= ~XFS_ICHGTIME_ACC;
793 if (mask & XFS_AT_MTIME) {
794 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
795 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
796 timeflags &= ~XFS_ICHGTIME_MOD;
797 timeflags |= XFS_ICHGTIME_CHG;
799 if (tp && (flags & ATTR_UTIME))
800 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
804 * Change XFS-added attributes.
806 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
807 if (mask & XFS_AT_EXTSIZE) {
809 * Converting bytes to fs blocks.
811 ip->i_d.di_extsize = vap->va_extsize >>
812 mp->m_sb.sb_blocklog;
814 if (mask & XFS_AT_XFLAGS) {
817 /* can't set PREALLOC this way, just preserve it */
818 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
819 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
820 di_flags |= XFS_DIFLAG_IMMUTABLE;
821 if (vap->va_xflags & XFS_XFLAG_APPEND)
822 di_flags |= XFS_DIFLAG_APPEND;
823 if (vap->va_xflags & XFS_XFLAG_SYNC)
824 di_flags |= XFS_DIFLAG_SYNC;
825 if (vap->va_xflags & XFS_XFLAG_NOATIME)
826 di_flags |= XFS_DIFLAG_NOATIME;
827 if (vap->va_xflags & XFS_XFLAG_NODUMP)
828 di_flags |= XFS_DIFLAG_NODUMP;
829 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
830 di_flags |= XFS_DIFLAG_PROJINHERIT;
831 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
832 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
833 di_flags |= XFS_DIFLAG_RTINHERIT;
834 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
835 di_flags |= XFS_DIFLAG_NOSYMLINKS;
837 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
838 di_flags |= XFS_DIFLAG_REALTIME;
839 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
841 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
844 ip->i_d.di_flags = di_flags;
846 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
847 timeflags |= XFS_ICHGTIME_CHG;
851 * Change file inode change time only if XFS_AT_CTIME set
852 * AND we have been called by a DMI function.
855 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
856 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
857 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
858 ip->i_update_core = 1;
859 timeflags &= ~XFS_ICHGTIME_CHG;
863 * Send out timestamp changes that need to be set to the
864 * current time. Not done when called by a DMI function.
866 if (timeflags && !(flags & ATTR_DMI))
867 xfs_ichgtime(ip, timeflags);
869 XFS_STATS_INC(xs_ig_attrchg);
872 * If this is a synchronous mount, make sure that the
873 * transaction goes to disk before returning to the user.
874 * This is slightly sub-optimal in that truncates require
875 * two sync transactions instead of one for wsync filesytems.
876 * One for the truncate and one for the timestamps since we
877 * don't want to change the timestamps unless we're sure the
878 * truncate worked. Truncates are less than 1% of the laddis
879 * mix so this probably isn't worth the trouble to optimize.
883 if (mp->m_flags & XFS_MOUNT_WSYNC)
884 xfs_trans_set_sync(tp);
886 code = xfs_trans_commit(tp, commit_flags, NULL);
890 * If the (regular) file's mandatory locking mode changed, then
891 * notify the vnode. We do this under the inode lock to prevent
892 * racing calls to vop_vnode_change.
894 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
895 if (mandlock_before != mandlock_after) {
896 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
900 xfs_iunlock(ip, lock_flags);
903 * Release any dquot(s) the inode had kept before chown.
905 XFS_QM_DQRELE(mp, olddquot1);
906 XFS_QM_DQRELE(mp, olddquot2);
907 XFS_QM_DQRELE(mp, udqp);
908 XFS_QM_DQRELE(mp, gdqp);
914 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
915 !(flags & ATTR_DMI)) {
916 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
917 NULL, DM_RIGHT_NULL, NULL, NULL,
918 0, 0, AT_DELAY_FLAG(flags));
923 commit_flags |= XFS_TRANS_ABORT;
926 XFS_QM_DQRELE(mp, udqp);
927 XFS_QM_DQRELE(mp, gdqp);
929 xfs_trans_cancel(tp, commit_flags);
931 if (lock_flags != 0) {
932 xfs_iunlock(ip, lock_flags);
940 * Null conversion from vnode mode bits to inode mode bits, as in efs.
951 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
952 (inst_t *)__return_address);
954 ip = XFS_BHVTOI(bdp);
955 xfs_ilock(ip, XFS_ILOCK_SHARED);
956 error = xfs_iaccess(ip, mode, credp);
957 xfs_iunlock(ip, XFS_ILOCK_SHARED);
981 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
987 vp = BHV_TO_VNODE(bdp);
988 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
990 ip = XFS_BHVTOI(bdp);
993 if (XFS_FORCED_SHUTDOWN(mp))
994 return XFS_ERROR(EIO);
996 xfs_ilock(ip, XFS_ILOCK_SHARED);
998 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1000 offset = uiop->uio_offset;
1001 count = uiop->uio_resid;
1004 error = XFS_ERROR(EINVAL);
1012 if (!(ioflags & IO_INVIS)) {
1013 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1017 * See if the symlink is stored inline.
1019 pathlen = (int)ip->i_d.di_size;
1021 if (ip->i_df.if_flags & XFS_IFINLINE) {
1022 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1026 * Symlink not inline. Call bmap to get it in.
1028 nmaps = SYMLINK_MAPS;
1030 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1031 0, NULL, 0, mval, &nmaps, NULL);
1037 for (n = 0; n < nmaps; n++) {
1038 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1039 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1040 bp = xfs_buf_read(mp->m_ddev_targp, d,
1041 BTOBB(byte_cnt), 0);
1042 error = XFS_BUF_GETERROR(bp);
1044 xfs_ioerror_alert("xfs_readlink",
1045 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1049 if (pathlen < byte_cnt)
1051 pathlen -= byte_cnt;
1053 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1062 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1071 * This is called to sync the inode and its data out to disk.
1072 * We need to hold the I/O lock while flushing the data, and
1073 * the inode lock while flushing the inode. The inode lock CANNOT
1074 * be held while flushing the data, so acquire after we're done
1088 int log_flushed = 0, changed = 1;
1090 vn_trace_entry(BHV_TO_VNODE(bdp),
1091 __FUNCTION__, (inst_t *)__return_address);
1093 ip = XFS_BHVTOI(bdp);
1095 ASSERT(start >= 0 && stop >= -1);
1097 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1098 return XFS_ERROR(EIO);
1101 * We always need to make sure that the required inode state
1102 * is safe on disk. The vnode might be clean but because
1103 * of committed transactions that haven't hit the disk yet.
1104 * Likewise, there could be unflushed non-transactional
1105 * changes to the inode core that have to go to disk.
1107 * The following code depends on one assumption: that
1108 * any transaction that changes an inode logs the core
1109 * because it has to change some field in the inode core
1110 * (typically nextents or nblocks). That assumption
1111 * implies that any transactions against an inode will
1112 * catch any non-transactional updates. If inode-altering
1113 * transactions exist that violate this assumption, the
1114 * code breaks. Right now, it figures that if the involved
1115 * update_* field is clear and the inode is unpinned, the
1116 * inode is clean. Either it's been flushed or it's been
1117 * committed and the commit has hit the disk unpinning the inode.
1118 * (Note that xfs_inode_item_format() called at commit clears
1119 * the update_* fields.)
1121 xfs_ilock(ip, XFS_ILOCK_SHARED);
1123 /* If we are flushing data then we care about update_size
1124 * being set, otherwise we care about update_core
1126 if ((flag & FSYNC_DATA) ?
1127 (ip->i_update_size == 0) :
1128 (ip->i_update_core == 0)) {
1130 * Timestamps/size haven't changed since last inode
1131 * flush or inode transaction commit. That means
1132 * either nothing got written or a transaction
1133 * committed which caught the updates. If the
1134 * latter happened and the transaction hasn't
1135 * hit the disk yet, the inode will be still
1136 * be pinned. If it is, force the log.
1139 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1141 if (xfs_ipincount(ip)) {
1142 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1144 ((flag & FSYNC_WAIT)
1145 ? XFS_LOG_SYNC : 0),
1149 * If the inode is not pinned and nothing
1150 * has changed we don't need to flush the
1158 * Kick off a transaction to log the inode
1159 * core to get the updates. Make it
1160 * sync if FSYNC_WAIT is passed in (which
1161 * is done by everybody but specfs). The
1162 * sync transaction will also force the log.
1164 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1165 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1166 if ((error = xfs_trans_reserve(tp, 0,
1167 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1169 xfs_trans_cancel(tp, 0);
1172 xfs_ilock(ip, XFS_ILOCK_EXCL);
1175 * Note - it's possible that we might have pushed
1176 * ourselves out of the way during trans_reserve
1177 * which would flush the inode. But there's no
1178 * guarantee that the inode buffer has actually
1179 * gone out yet (it's delwri). Plus the buffer
1180 * could be pinned anyway if it's part of an
1181 * inode in another recent transaction. So we
1182 * play it safe and fire off the transaction anyway.
1184 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1185 xfs_trans_ihold(tp, ip);
1186 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1187 if (flag & FSYNC_WAIT)
1188 xfs_trans_set_sync(tp);
1189 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1191 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1194 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1196 * If the log write didn't issue an ordered tag we need
1197 * to flush the disk cache for the data device now.
1200 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1203 * If this inode is on the RT dev we need to flush that
1206 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1207 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1214 * This is called by xfs_inactive to free any blocks beyond eof,
1215 * when the link count isn't zero.
1218 xfs_inactive_free_eofblocks(
1224 xfs_fileoff_t end_fsb;
1225 xfs_fileoff_t last_fsb;
1226 xfs_filblks_t map_len;
1228 xfs_bmbt_irec_t imap;
1231 * Figure out if there are any blocks beyond the end
1232 * of the file. If not, then there is nothing to do.
1234 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1235 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1236 map_len = last_fsb - end_fsb;
1241 xfs_ilock(ip, XFS_ILOCK_SHARED);
1242 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1243 NULL, 0, &imap, &nimaps, NULL);
1244 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1246 if (!error && (nimaps != 0) &&
1247 (imap.br_startblock != HOLESTARTBLOCK)) {
1249 * Attach the dquots to the inode up front.
1251 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1255 * There are blocks after the end of file.
1256 * Free them up now by truncating the file to
1259 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1262 * Do the xfs_itruncate_start() call before
1263 * reserving any log space because
1264 * itruncate_start will call into the buffer
1265 * cache and we can't
1266 * do that within a transaction.
1268 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1269 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1272 error = xfs_trans_reserve(tp, 0,
1273 XFS_ITRUNCATE_LOG_RES(mp),
1274 0, XFS_TRANS_PERM_LOG_RES,
1275 XFS_ITRUNCATE_LOG_COUNT);
1277 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1278 xfs_trans_cancel(tp, 0);
1279 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1283 xfs_ilock(ip, XFS_ILOCK_EXCL);
1284 xfs_trans_ijoin(tp, ip,
1287 xfs_trans_ihold(tp, ip);
1289 error = xfs_itruncate_finish(&tp, ip,
1294 * If we get an error at this point we
1295 * simply don't bother truncating the file.
1298 xfs_trans_cancel(tp,
1299 (XFS_TRANS_RELEASE_LOG_RES |
1302 error = xfs_trans_commit(tp,
1303 XFS_TRANS_RELEASE_LOG_RES,
1306 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1312 * Free a symlink that has blocks associated with it.
1315 xfs_inactive_symlink_rmt(
1323 xfs_fsblock_t first_block;
1324 xfs_bmap_free_t free_list;
1327 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1335 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1337 * We're freeing a symlink that has some
1338 * blocks allocated to it. Free the
1339 * blocks here. We know that we've got
1340 * either 1 or 2 extents and that we can
1341 * free them all in one bunmapi call.
1343 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1344 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1345 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1346 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1347 xfs_trans_cancel(tp, 0);
1352 * Lock the inode, fix the size, and join it to the transaction.
1353 * Hold it so in the normal path, we still have it locked for
1354 * the second transaction. In the error paths we need it
1355 * held so the cancel won't rele it, see below.
1357 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1358 size = (int)ip->i_d.di_size;
1359 ip->i_d.di_size = 0;
1360 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1361 xfs_trans_ihold(tp, ip);
1362 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1364 * Find the block(s) so we can inval and unmap them.
1367 XFS_BMAP_INIT(&free_list, &first_block);
1368 nmaps = sizeof(mval) / sizeof(mval[0]);
1369 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1370 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1374 * Invalidate the block(s).
1376 for (i = 0; i < nmaps; i++) {
1377 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1378 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1379 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1380 xfs_trans_binval(tp, bp);
1383 * Unmap the dead block(s) to the free_list.
1385 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1386 &first_block, &free_list, &done)))
1390 * Commit the first transaction. This logs the EFI and the inode.
1392 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1395 * The transaction must have been committed, since there were
1396 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1397 * The new tp has the extent freeing and EFDs.
1401 * The first xact was committed, so add the inode to the new one.
1402 * Mark it dirty so it will be logged and moved forward in the log as
1403 * part of every commit.
1405 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1406 xfs_trans_ihold(tp, ip);
1407 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1409 * Get a new, empty transaction to return to our caller.
1411 ntp = xfs_trans_dup(tp);
1413 * Commit the transaction containing extent freeing and EFD's.
1414 * If we get an error on the commit here or on the reserve below,
1415 * we need to unlock the inode since the new transaction doesn't
1416 * have the inode attached.
1418 error = xfs_trans_commit(tp, 0, NULL);
1421 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1425 * Remove the memory for extent descriptions (just bookkeeping).
1427 if (ip->i_df.if_bytes)
1428 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1429 ASSERT(ip->i_df.if_bytes == 0);
1431 * Put an itruncate log reservation in the new transaction
1434 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1435 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1436 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1440 * Return with the inode locked but not joined to the transaction.
1446 xfs_bmap_cancel(&free_list);
1449 * Have to come here with the inode locked and either
1450 * (held and in the transaction) or (not in the transaction).
1451 * If the inode isn't held then cancel would iput it, but
1452 * that's wrong since this is inactive and the vnode ref
1453 * count is 0 already.
1454 * Cancel won't do anything to the inode if held, but it still
1455 * needs to be locked until the cancel is done, if it was
1456 * joined to the transaction.
1458 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1459 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1466 xfs_inactive_symlink_local(
1472 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1474 * We're freeing a symlink which fit into
1475 * the inode. Just free the memory used
1476 * to hold the old symlink.
1478 error = xfs_trans_reserve(*tpp, 0,
1479 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1480 0, XFS_TRANS_PERM_LOG_RES,
1481 XFS_ITRUNCATE_LOG_COUNT);
1484 xfs_trans_cancel(*tpp, 0);
1488 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1491 * Zero length symlinks _can_ exist.
1493 if (ip->i_df.if_bytes > 0) {
1494 xfs_idata_realloc(ip,
1495 -(ip->i_df.if_bytes),
1497 ASSERT(ip->i_df.if_bytes == 0);
1514 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1517 ASSERT(ip->i_d.di_forkoff != 0);
1518 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1519 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1521 error = xfs_attr_inactive(ip);
1524 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1525 return (error); /* goto out*/
1528 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1529 error = xfs_trans_reserve(tp, 0,
1530 XFS_IFREE_LOG_RES(mp),
1531 0, XFS_TRANS_PERM_LOG_RES,
1532 XFS_INACTIVE_LOG_COUNT);
1534 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1535 xfs_trans_cancel(tp, 0);
1537 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1541 xfs_ilock(ip, XFS_ILOCK_EXCL);
1542 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1543 xfs_trans_ihold(tp, ip);
1544 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1546 ASSERT(ip->i_d.di_anextents == 0);
1561 vp = BHV_TO_VNODE(bdp);
1562 ip = XFS_BHVTOI(bdp);
1564 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1568 /* If this is a read-only mount, don't do this (would generate I/O) */
1569 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1572 #ifdef HAVE_REFCACHE
1573 /* If we are in the NFS reference cache then don't do this now */
1580 if (ip->i_d.di_nlink != 0) {
1581 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1582 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1583 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1584 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1585 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1587 /* Update linux inode block count after free above */
1588 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1589 ip->i_d.di_nblocks + ip->i_delayed_blks);
1599 * This is called when the vnode reference count for the vnode
1600 * goes to zero. If the file has been unlinked, then it must
1601 * now be truncated. Also, we clear all of the read-ahead state
1602 * kept for the inode here since the file is now closed.
1611 xfs_bmap_free_t free_list;
1612 xfs_fsblock_t first_block;
1619 vp = BHV_TO_VNODE(bdp);
1620 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1622 ip = XFS_BHVTOI(bdp);
1625 * If the inode is already free, then there can be nothing
1628 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1629 ASSERT(ip->i_df.if_real_bytes == 0);
1630 ASSERT(ip->i_df.if_broot_bytes == 0);
1631 return VN_INACTIVE_CACHE;
1635 * Only do a truncate if it's a regular file with
1636 * some actual space in it. It's OK to look at the
1637 * inode's fields without the lock because we're the
1638 * only one with a reference to the inode.
1640 truncate = ((ip->i_d.di_nlink == 0) &&
1641 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1642 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1646 if (ip->i_d.di_nlink == 0 &&
1647 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1648 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1653 /* If this is a read-only mount, don't do this (would generate I/O) */
1654 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1657 if (ip->i_d.di_nlink != 0) {
1658 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1659 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1660 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1661 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1662 (ip->i_delayed_blks != 0))) {
1663 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1664 return (VN_INACTIVE_CACHE);
1665 /* Update linux inode block count after free above */
1666 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1667 ip->i_d.di_nblocks + ip->i_delayed_blks);
1672 ASSERT(ip->i_d.di_nlink == 0);
1674 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1675 return (VN_INACTIVE_CACHE);
1677 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1680 * Do the xfs_itruncate_start() call before
1681 * reserving any log space because itruncate_start
1682 * will call into the buffer cache and we can't
1683 * do that within a transaction.
1685 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1687 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1689 error = xfs_trans_reserve(tp, 0,
1690 XFS_ITRUNCATE_LOG_RES(mp),
1691 0, XFS_TRANS_PERM_LOG_RES,
1692 XFS_ITRUNCATE_LOG_COUNT);
1694 /* Don't call itruncate_cleanup */
1695 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1696 xfs_trans_cancel(tp, 0);
1697 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1698 return (VN_INACTIVE_CACHE);
1701 xfs_ilock(ip, XFS_ILOCK_EXCL);
1702 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1703 xfs_trans_ihold(tp, ip);
1706 * normally, we have to run xfs_itruncate_finish sync.
1707 * But if filesystem is wsync and we're in the inactive
1708 * path, then we know that nlink == 0, and that the
1709 * xaction that made nlink == 0 is permanently committed
1710 * since xfs_remove runs as a synchronous transaction.
1712 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1713 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1716 xfs_trans_cancel(tp,
1717 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1718 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1719 return (VN_INACTIVE_CACHE);
1721 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1724 * If we get an error while cleaning up a
1725 * symlink we bail out.
1727 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1728 xfs_inactive_symlink_rmt(ip, &tp) :
1729 xfs_inactive_symlink_local(ip, &tp);
1733 return (VN_INACTIVE_CACHE);
1736 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1737 xfs_trans_ihold(tp, ip);
1739 error = xfs_trans_reserve(tp, 0,
1740 XFS_IFREE_LOG_RES(mp),
1741 0, XFS_TRANS_PERM_LOG_RES,
1742 XFS_INACTIVE_LOG_COUNT);
1744 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1745 xfs_trans_cancel(tp, 0);
1746 return (VN_INACTIVE_CACHE);
1749 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1750 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1751 xfs_trans_ihold(tp, ip);
1755 * If there are attributes associated with the file
1756 * then blow them away now. The code calls a routine
1757 * that recursively deconstructs the attribute fork.
1758 * We need to just commit the current transaction
1759 * because we can't use it for xfs_attr_inactive().
1761 if (ip->i_d.di_anextents > 0) {
1762 error = xfs_inactive_attrs(ip, &tp);
1764 * If we got an error, the transaction is already
1765 * cancelled, and the inode is unlocked. Just get out.
1768 return (VN_INACTIVE_CACHE);
1769 } else if (ip->i_afp) {
1770 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1776 XFS_BMAP_INIT(&free_list, &first_block);
1777 error = xfs_ifree(tp, ip, &free_list);
1780 * If we fail to free the inode, shut down. The cancel
1781 * might do that, we need to make sure. Otherwise the
1782 * inode might be lost for a long time or forever.
1784 if (!XFS_FORCED_SHUTDOWN(mp)) {
1786 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1787 error, mp->m_fsname);
1788 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1790 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1793 * Credit the quota account(s). The inode is gone.
1795 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1798 * Just ignore errors at this point. There is
1799 * nothing we can do except to try to keep going.
1801 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1803 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1806 * Release the dquots held by inode, if any.
1808 XFS_QM_DQDETACH(mp, ip);
1810 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1813 return VN_INACTIVE_CACHE;
1822 bhv_desc_t *dir_bdp,
1829 xfs_inode_t *dp, *ip;
1835 dir_vp = BHV_TO_VNODE(dir_bdp);
1836 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1838 dp = XFS_BHVTOI(dir_bdp);
1840 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1841 return XFS_ERROR(EIO);
1843 lock_mode = xfs_ilock_map_shared(dp);
1844 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1846 *vpp = XFS_ITOV(ip);
1849 xfs_iunlock_map_shared(dp, lock_mode);
1855 * xfs_create (create a new file).
1859 bhv_desc_t *dir_bdp,
1865 char *name = VNAME(dentry);
1867 xfs_inode_t *dp, *ip;
1873 xfs_bmap_free_t free_list;
1874 xfs_fsblock_t first_block;
1875 boolean_t dp_joined_to_trans;
1876 int dm_event_sent = 0;
1880 struct xfs_dquot *udqp, *gdqp;
1886 dir_vp = BHV_TO_VNODE(dir_bdp);
1887 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1889 dp = XFS_BHVTOI(dir_bdp);
1892 dm_di_mode = vap->va_mode;
1893 namelen = VNAMELEN(dentry);
1895 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1896 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1897 dir_vp, DM_RIGHT_NULL, NULL,
1898 DM_RIGHT_NULL, name, NULL,
1906 if (XFS_FORCED_SHUTDOWN(mp))
1907 return XFS_ERROR(EIO);
1909 /* Return through std_return after this point. */
1912 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1913 prid = dp->i_d.di_projid;
1914 else if (vap->va_mask & XFS_AT_PROJID)
1915 prid = (xfs_prid_t)vap->va_projid;
1917 prid = (xfs_prid_t)dfltprid;
1920 * Make sure that we have allocated dquot(s) on disk.
1922 error = XFS_QM_DQVOPALLOC(mp, dp,
1923 current_fsuid(credp), current_fsgid(credp), prid,
1924 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1929 dp_joined_to_trans = B_FALSE;
1931 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1932 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1933 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1935 * Initially assume that the file does not exist and
1936 * reserve the resources for that case. If that is not
1937 * the case we'll drop the one we have and get a more
1938 * appropriate transaction later.
1940 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1941 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1942 if (error == ENOSPC) {
1944 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1945 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1953 xfs_ilock(dp, XFS_ILOCK_EXCL);
1955 XFS_BMAP_INIT(&free_list, &first_block);
1960 * Reserve disk quota and the inode.
1962 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1967 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1969 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1970 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1971 rdev, credp, prid, resblks > 0,
1974 if (error == ENOSPC)
1981 * At this point, we've gotten a newly allocated inode.
1982 * It is locked (and joined to the transaction).
1985 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1988 * Now we join the directory inode to the transaction.
1989 * We do not do it earlier because xfs_dir_ialloc
1990 * might commit the previous transaction (and release
1995 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1996 dp_joined_to_trans = B_TRUE;
1998 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1999 &first_block, &free_list,
2000 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2002 ASSERT(error != ENOSPC);
2005 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2006 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2009 * If this is a synchronous mount, make sure that the
2010 * create transaction goes to disk before returning to
2013 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2014 xfs_trans_set_sync(tp);
2020 * Attach the dquot(s) to the inodes and modify them incore.
2021 * These ids of the inode couldn't have changed since the new
2022 * inode has been locked ever since it was created.
2024 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2027 * xfs_trans_commit normally decrements the vnode ref count
2028 * when it unlocks the inode. Since we want to return the
2029 * vnode to the caller, we bump the vnode ref count now.
2034 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2036 xfs_bmap_cancel(&free_list);
2040 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2047 XFS_QM_DQRELE(mp, udqp);
2048 XFS_QM_DQRELE(mp, gdqp);
2051 * Propogate the fact that the vnode changed after the
2052 * xfs_inode locks have been released.
2054 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2058 /* Fallthrough to std_return with error = 0 */
2061 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2062 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2063 DM_EVENT_POSTCREATE)) {
2064 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2065 dir_vp, DM_RIGHT_NULL,
2067 DM_RIGHT_NULL, name, NULL,
2068 dm_di_mode, error, 0);
2073 cancel_flags |= XFS_TRANS_ABORT;
2078 xfs_trans_cancel(tp, cancel_flags);
2080 if (!dp_joined_to_trans && (dp != NULL))
2081 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2082 XFS_QM_DQRELE(mp, udqp);
2083 XFS_QM_DQRELE(mp, gdqp);
2089 * Wait until after the current transaction is aborted to
2090 * release the inode. This prevents recursive transactions
2091 * and deadlocks from xfs_inactive.
2093 cancel_flags |= XFS_TRANS_ABORT;
2094 xfs_trans_cancel(tp, cancel_flags);
2097 XFS_QM_DQRELE(mp, udqp);
2098 XFS_QM_DQRELE(mp, gdqp);
2105 * Some counters to see if (and how often) we are hitting some deadlock
2106 * prevention code paths.
2110 int xfs_rm_lock_delays;
2111 int xfs_rm_attempts;
2115 * The following routine will lock the inodes associated with the
2116 * directory and the named entry in the directory. The locks are
2117 * acquired in increasing inode number.
2119 * If the entry is "..", then only the directory is locked. The
2120 * vnode ref count will still include that from the .. entry in
2123 * There is a deadlock we need to worry about. If the locked directory is
2124 * in the AIL, it might be blocking up the log. The next inode we lock
2125 * could be already locked by another thread waiting for log space (e.g
2126 * a permanent log reservation with a long running transaction (see
2127 * xfs_itruncate_finish)). To solve this, we must check if the directory
2128 * is in the ail and use lock_nowait. If we can't lock, we need to
2129 * drop the inode lock on the directory and try again. xfs_iunlock will
2130 * potentially push the tail if we were holding up the log.
2133 xfs_lock_dir_and_entry(
2136 xfs_inode_t *ip) /* inode of entry 'name' */
2140 xfs_inode_t *ips[2];
2149 xfs_ilock(dp, XFS_ILOCK_EXCL);
2156 * We want to lock in increasing inum. Since we've already
2157 * acquired the lock on the directory, we may need to release
2158 * if if the inum of the entry turns out to be less.
2160 if (e_inum > dp->i_ino) {
2162 * We are already in the right order, so just
2163 * lock on the inode of the entry.
2164 * We need to use nowait if dp is in the AIL.
2167 lp = (xfs_log_item_t *)dp->i_itemp;
2168 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2169 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2176 * Unlock dp and try again.
2177 * xfs_iunlock will try to push the tail
2178 * if the inode is in the AIL.
2181 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2183 if ((attempts % 5) == 0) {
2184 delay(1); /* Don't just spin the CPU */
2186 xfs_rm_lock_delays++;
2192 xfs_ilock(ip, XFS_ILOCK_EXCL);
2194 } else if (e_inum < dp->i_ino) {
2195 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2199 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2201 /* else e_inum == dp->i_ino */
2202 /* This can happen if we're asked to lock /x/..
2203 * the entry is "..", which is also the parent directory.
2211 int xfs_small_retries;
2212 int xfs_middle_retries;
2213 int xfs_lots_retries;
2214 int xfs_lock_delays;
2218 * The following routine will lock n inodes in exclusive mode.
2219 * We assume the caller calls us with the inodes in i_ino order.
2221 * We need to detect deadlock where an inode that we lock
2222 * is in the AIL and we start waiting for another inode that is locked
2223 * by a thread in a long running transaction (such as truncate). This can
2224 * result in deadlock since the long running trans might need to wait
2225 * for the inode we just locked in order to push the tail and free space
2235 int attempts = 0, i, j, try_lock;
2238 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2249 for (; i < inodes; i++) {
2252 if (i && (ips[i] == ips[i-1])) /* Already locked */
2256 * If try_lock is not set yet, make sure all locked inodes
2257 * are not in the AIL.
2258 * If any are, set try_lock to be used later.
2262 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2263 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2264 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2271 * If any of the previous locks we have locked is in the AIL,
2272 * we must TRY to get the second and subsequent locks. If
2273 * we can't get any, we must release all we have
2278 /* try_lock must be 0 if i is 0. */
2280 * try_lock means we have an inode locked
2281 * that is in the AIL.
2284 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2288 * Unlock all previous guys and try again.
2289 * xfs_iunlock will try to push the tail
2290 * if the inode is in the AIL.
2293 for(j = i - 1; j >= 0; j--) {
2296 * Check to see if we've already
2297 * unlocked this one.
2298 * Not the first one going back,
2299 * and the inode ptr is the same.
2301 if ((j != (i - 1)) && ips[j] ==
2305 xfs_iunlock(ips[j], lock_mode);
2308 if ((attempts % 5) == 0) {
2309 delay(1); /* Don't just spin the CPU */
2319 xfs_ilock(ips[i], lock_mode);
2325 if (attempts < 5) xfs_small_retries++;
2326 else if (attempts < 100) xfs_middle_retries++;
2327 else xfs_lots_retries++;
2335 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2336 int remove_which_error_return = 0;
2338 #define REMOVE_DEBUG_TRACE(x)
2339 #endif /* ! DEBUG */
2348 bhv_desc_t *dir_bdp,
2353 char *name = VNAME(dentry);
2354 xfs_inode_t *dp, *ip;
2355 xfs_trans_t *tp = NULL;
2358 xfs_bmap_free_t free_list;
2359 xfs_fsblock_t first_block;
2367 dir_vp = BHV_TO_VNODE(dir_bdp);
2368 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2370 dp = XFS_BHVTOI(dir_bdp);
2373 if (XFS_FORCED_SHUTDOWN(mp))
2374 return XFS_ERROR(EIO);
2376 namelen = VNAMELEN(dentry);
2378 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2379 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2380 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2381 name, NULL, 0, 0, 0);
2386 /* From this point on, return through std_return */
2390 * We need to get a reference to ip before we get our log
2391 * reservation. The reason for this is that we cannot call
2392 * xfs_iget for an inode for which we do not have a reference
2393 * once we've acquired a log reservation. This is because the
2394 * inode we are trying to get might be in xfs_inactive going
2395 * for a log reservation. Since we'll have to wait for the
2396 * inactive code to complete before returning from xfs_iget,
2397 * we need to make sure that we don't have log space reserved
2398 * when we call xfs_iget. Instead we get an unlocked referece
2399 * to the inode before getting our log reservation.
2401 error = xfs_get_dir_entry(dentry, &ip);
2403 REMOVE_DEBUG_TRACE(__LINE__);
2407 dm_di_mode = ip->i_d.di_mode;
2409 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2413 error = XFS_QM_DQATTACH(mp, dp, 0);
2414 if (!error && dp != ip)
2415 error = XFS_QM_DQATTACH(mp, ip, 0);
2417 REMOVE_DEBUG_TRACE(__LINE__);
2422 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2423 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2425 * We try to get the real space reservation first,
2426 * allowing for directory btree deletion(s) implying
2427 * possible bmap insert(s). If we can't get the space
2428 * reservation then we use 0 instead, and avoid the bmap
2429 * btree insert(s) in the directory code by, if the bmap
2430 * insert tries to happen, instead trimming the LAST
2431 * block from the directory.
2433 resblks = XFS_REMOVE_SPACE_RES(mp);
2434 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2435 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2436 if (error == ENOSPC) {
2438 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2439 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2442 ASSERT(error != ENOSPC);
2443 REMOVE_DEBUG_TRACE(__LINE__);
2444 xfs_trans_cancel(tp, 0);
2449 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2451 REMOVE_DEBUG_TRACE(__LINE__);
2452 xfs_trans_cancel(tp, cancel_flags);
2458 * At this point, we've gotten both the directory and the entry
2461 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2464 * Increment vnode ref count only in this case since
2465 * there's an extra vnode reference in the case where
2469 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2473 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2475 XFS_BMAP_INIT(&free_list, &first_block);
2476 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2477 &first_block, &free_list, 0);
2479 ASSERT(error != ENOENT);
2480 REMOVE_DEBUG_TRACE(__LINE__);
2483 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2486 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2488 error = xfs_droplink(tp, ip);
2490 REMOVE_DEBUG_TRACE(__LINE__);
2494 /* Determine if this is the last link while
2495 * we are in the transaction.
2497 link_zero = (ip)->i_d.di_nlink==0;
2500 * Take an extra ref on the inode so that it doesn't
2501 * go to xfs_inactive() from within the commit.
2506 * If this is a synchronous mount, make sure that the
2507 * remove transaction goes to disk before returning to
2510 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2511 xfs_trans_set_sync(tp);
2514 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2516 REMOVE_DEBUG_TRACE(__LINE__);
2520 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2527 * Before we drop our extra reference to the inode, purge it
2528 * from the refcache if it is there. By waiting until afterwards
2529 * to do the IRELE, we ensure that we won't go inactive in the
2530 * xfs_refcache_purge_ip routine (although that would be OK).
2532 xfs_refcache_purge_ip(ip);
2534 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2537 * Let interposed file systems know about removed links.
2539 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2543 /* Fall through to std_return with error = 0 */
2545 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2546 DM_EVENT_POSTREMOVE)) {
2547 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2548 dir_vp, DM_RIGHT_NULL,
2549 NULL, DM_RIGHT_NULL,
2550 name, NULL, dm_di_mode, error, 0);
2555 xfs_bmap_cancel(&free_list);
2556 cancel_flags |= XFS_TRANS_ABORT;
2557 xfs_trans_cancel(tp, cancel_flags);
2562 * In this case make sure to not release the inode until after
2563 * the current transaction is aborted. Releasing it beforehand
2564 * can cause us to go to xfs_inactive and start a recursive
2565 * transaction which can easily deadlock with the current one.
2567 xfs_bmap_cancel(&free_list);
2568 cancel_flags |= XFS_TRANS_ABORT;
2569 xfs_trans_cancel(tp, cancel_flags);
2572 * Before we drop our extra reference to the inode, purge it
2573 * from the refcache if it is there. By waiting until afterwards
2574 * to do the IRELE, we ensure that we won't go inactive in the
2575 * xfs_refcache_purge_ip routine (although that would be OK).
2577 xfs_refcache_purge_ip(ip);
2591 bhv_desc_t *target_dir_bdp,
2596 xfs_inode_t *tdp, *sip;
2599 xfs_inode_t *ips[2];
2601 xfs_bmap_free_t free_list;
2602 xfs_fsblock_t first_block;
2605 vnode_t *target_dir_vp;
2606 bhv_desc_t *src_bdp;
2608 char *target_name = VNAME(dentry);
2611 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2612 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2613 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2615 target_namelen = VNAMELEN(dentry);
2616 if (VN_ISDIR(src_vp))
2617 return XFS_ERROR(EPERM);
2619 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2620 sip = XFS_BHVTOI(src_bdp);
2621 tdp = XFS_BHVTOI(target_dir_bdp);
2623 if (XFS_FORCED_SHUTDOWN(mp))
2624 return XFS_ERROR(EIO);
2626 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2627 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2628 target_dir_vp, DM_RIGHT_NULL,
2629 src_vp, DM_RIGHT_NULL,
2630 target_name, NULL, 0, 0, 0);
2635 /* Return through std_return after this point. */
2637 error = XFS_QM_DQATTACH(mp, sip, 0);
2638 if (!error && sip != tdp)
2639 error = XFS_QM_DQATTACH(mp, tdp, 0);
2643 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2644 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2645 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2646 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2647 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2648 if (error == ENOSPC) {
2650 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2651 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2658 if (sip->i_ino < tdp->i_ino) {
2666 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2669 * Increment vnode ref counts since xfs_trans_commit &
2670 * xfs_trans_cancel will both unlock the inodes and
2671 * decrement the associated ref counts.
2674 VN_HOLD(target_dir_vp);
2675 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2676 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2679 * If the source has too many links, we can't make any more to it.
2681 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2682 error = XFS_ERROR(EMLINK);
2687 * If we are using project inheritance, we only allow hard link
2688 * creation in our tree when the project IDs are the same; else
2689 * the tree quota mechanism could be circumvented.
2691 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2692 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2693 error = XFS_ERROR(EPERM);
2698 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2702 XFS_BMAP_INIT(&free_list, &first_block);
2704 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2705 sip->i_ino, &first_block, &free_list,
2709 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2711 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2713 error = xfs_bumplink(tp, sip);
2719 * If this is a synchronous mount, make sure that the
2720 * link transaction goes to disk before returning to
2723 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2724 xfs_trans_set_sync(tp);
2727 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2729 xfs_bmap_cancel(&free_list);
2733 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2738 /* Fall through to std_return with error = 0. */
2740 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2741 DM_EVENT_POSTLINK)) {
2742 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2743 target_dir_vp, DM_RIGHT_NULL,
2744 src_vp, DM_RIGHT_NULL,
2745 target_name, NULL, 0, error, 0);
2750 cancel_flags |= XFS_TRANS_ABORT;
2753 xfs_trans_cancel(tp, cancel_flags);
2763 bhv_desc_t *dir_bdp,
2769 char *dir_name = VNAME(dentry);
2771 xfs_inode_t *cdp; /* inode of created dir */
2772 vnode_t *cvp; /* vnode of created dir */
2778 xfs_bmap_free_t free_list;
2779 xfs_fsblock_t first_block;
2781 boolean_t dp_joined_to_trans;
2782 boolean_t created = B_FALSE;
2783 int dm_event_sent = 0;
2785 struct xfs_dquot *udqp, *gdqp;
2790 dir_vp = BHV_TO_VNODE(dir_bdp);
2791 dp = XFS_BHVTOI(dir_bdp);
2794 if (XFS_FORCED_SHUTDOWN(mp))
2795 return XFS_ERROR(EIO);
2797 dir_namelen = VNAMELEN(dentry);
2800 dp_joined_to_trans = B_FALSE;
2801 dm_di_mode = vap->va_mode;
2803 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2804 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2805 dir_vp, DM_RIGHT_NULL, NULL,
2806 DM_RIGHT_NULL, dir_name, NULL,
2813 /* Return through std_return after this point. */
2815 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2819 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2820 prid = dp->i_d.di_projid;
2821 else if (vap->va_mask & XFS_AT_PROJID)
2822 prid = (xfs_prid_t)vap->va_projid;
2824 prid = (xfs_prid_t)dfltprid;
2827 * Make sure that we have allocated dquot(s) on disk.
2829 error = XFS_QM_DQVOPALLOC(mp, dp,
2830 current_fsuid(credp), current_fsgid(credp), prid,
2831 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2835 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2836 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2837 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2838 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2839 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2840 if (error == ENOSPC) {
2842 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2843 XFS_TRANS_PERM_LOG_RES,
2844 XFS_MKDIR_LOG_COUNT);
2852 xfs_ilock(dp, XFS_ILOCK_EXCL);
2855 * Check for directory link count overflow.
2857 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2858 error = XFS_ERROR(EMLINK);
2863 * Reserve disk quota and the inode.
2865 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2870 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2873 * create the directory inode.
2875 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2876 0, credp, prid, resblks > 0,
2879 if (error == ENOSPC)
2886 * Now we add the directory inode to the transaction.
2887 * We waited until now since xfs_dir_ialloc might start
2888 * a new transaction. Had we joined the transaction
2889 * earlier, the locks might have gotten released.
2892 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2893 dp_joined_to_trans = B_TRUE;
2895 XFS_BMAP_INIT(&free_list, &first_block);
2897 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2898 cdp->i_ino, &first_block, &free_list,
2899 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2901 ASSERT(error != ENOSPC);
2904 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2907 * Bump the in memory version number of the parent directory
2908 * so that other processes accessing it will recognize that
2909 * the directory has changed.
2913 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2919 error = xfs_bumplink(tp, dp);
2924 cvp = XFS_ITOV(cdp);
2932 * Attach the dquots to the new inode and modify the icount incore.
2934 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2937 * If this is a synchronous mount, make sure that the
2938 * mkdir transaction goes to disk before returning to
2941 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2942 xfs_trans_set_sync(tp);
2945 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2951 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2952 XFS_QM_DQRELE(mp, udqp);
2953 XFS_QM_DQRELE(mp, gdqp);
2958 /* Fall through to std_return with error = 0 or errno from
2959 * xfs_trans_commit. */
2962 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2963 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2964 DM_EVENT_POSTCREATE)) {
2965 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2966 dir_vp, DM_RIGHT_NULL,
2967 created ? XFS_ITOV(cdp):NULL,
2970 dm_di_mode, error, 0);
2976 xfs_bmap_cancel(&free_list);
2978 cancel_flags |= XFS_TRANS_ABORT;
2980 xfs_trans_cancel(tp, cancel_flags);
2981 XFS_QM_DQRELE(mp, udqp);
2982 XFS_QM_DQRELE(mp, gdqp);
2984 if (!dp_joined_to_trans && (dp != NULL)) {
2985 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2998 bhv_desc_t *dir_bdp,
3002 char *name = VNAME(dentry);
3004 xfs_inode_t *cdp; /* child directory */
3008 xfs_bmap_free_t free_list;
3009 xfs_fsblock_t first_block;
3018 dir_vp = BHV_TO_VNODE(dir_bdp);
3019 dp = XFS_BHVTOI(dir_bdp);
3022 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3024 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3025 return XFS_ERROR(EIO);
3026 namelen = VNAMELEN(dentry);
3028 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3029 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3030 dir_vp, DM_RIGHT_NULL,
3031 NULL, DM_RIGHT_NULL,
3032 name, NULL, 0, 0, 0);
3034 return XFS_ERROR(error);
3037 /* Return through std_return after this point. */
3042 * We need to get a reference to cdp before we get our log
3043 * reservation. The reason for this is that we cannot call
3044 * xfs_iget for an inode for which we do not have a reference
3045 * once we've acquired a log reservation. This is because the
3046 * inode we are trying to get might be in xfs_inactive going
3047 * for a log reservation. Since we'll have to wait for the
3048 * inactive code to complete before returning from xfs_iget,
3049 * we need to make sure that we don't have log space reserved
3050 * when we call xfs_iget. Instead we get an unlocked referece
3051 * to the inode before getting our log reservation.
3053 error = xfs_get_dir_entry(dentry, &cdp);
3055 REMOVE_DEBUG_TRACE(__LINE__);
3059 dm_di_mode = cdp->i_d.di_mode;
3062 * Get the dquots for the inodes.
3064 error = XFS_QM_DQATTACH(mp, dp, 0);
3065 if (!error && dp != cdp)
3066 error = XFS_QM_DQATTACH(mp, cdp, 0);
3069 REMOVE_DEBUG_TRACE(__LINE__);
3073 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3074 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3076 * We try to get the real space reservation first,
3077 * allowing for directory btree deletion(s) implying
3078 * possible bmap insert(s). If we can't get the space
3079 * reservation then we use 0 instead, and avoid the bmap
3080 * btree insert(s) in the directory code by, if the bmap
3081 * insert tries to happen, instead trimming the LAST
3082 * block from the directory.
3084 resblks = XFS_REMOVE_SPACE_RES(mp);
3085 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3086 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3087 if (error == ENOSPC) {
3089 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3090 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3093 ASSERT(error != ENOSPC);
3098 XFS_BMAP_INIT(&free_list, &first_block);
3101 * Now lock the child directory inode and the parent directory
3102 * inode in the proper order. This will take care of validating
3103 * that the directory entry for the child directory inode has
3104 * not changed while we were obtaining a log reservation.
3106 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3108 xfs_trans_cancel(tp, cancel_flags);
3113 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3116 * Only increment the parent directory vnode count if
3117 * we didn't bump it in looking up cdp. The only time
3118 * we don't bump it is when we're looking up ".".
3124 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3126 ASSERT(cdp->i_d.di_nlink >= 2);
3127 if (cdp->i_d.di_nlink != 2) {
3128 error = XFS_ERROR(ENOTEMPTY);
3131 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3132 error = XFS_ERROR(ENOTEMPTY);
3136 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3137 &first_block, &free_list, resblks);
3142 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3145 * Bump the in memory generation count on the parent
3146 * directory so that other can know that it has changed.
3151 * Drop the link from cdp's "..".
3153 error = xfs_droplink(tp, dp);
3159 * Drop the link from dp to cdp.
3161 error = xfs_droplink(tp, cdp);
3167 * Drop the "." link from cdp to self.
3169 error = xfs_droplink(tp, cdp);
3174 /* Determine these before committing transaction */
3175 last_cdp_link = (cdp)->i_d.di_nlink==0;
3178 * Take an extra ref on the child vnode so that it
3179 * does not go to xfs_inactive() from within the commit.
3184 * If this is a synchronous mount, make sure that the
3185 * rmdir transaction goes to disk before returning to
3188 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3189 xfs_trans_set_sync(tp);
3192 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3194 xfs_bmap_cancel(&free_list);
3195 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3201 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3209 * Let interposed file systems know about removed links.
3211 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3215 /* Fall through to std_return with error = 0 or the errno
3216 * from xfs_trans_commit. */
3218 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3219 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3220 dir_vp, DM_RIGHT_NULL,
3221 NULL, DM_RIGHT_NULL,
3222 name, NULL, dm_di_mode,
3228 xfs_bmap_cancel(&free_list);
3229 cancel_flags |= XFS_TRANS_ABORT;
3231 xfs_trans_cancel(tp, cancel_flags);
3239 * Read dp's entries starting at uiop->uio_offset and translate them into
3240 * bufsize bytes worth of struct dirents starting at bufbase.
3244 bhv_desc_t *dir_bdp,
3250 xfs_trans_t *tp = NULL;
3253 xfs_off_t start_offset;
3255 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3256 (inst_t *)__return_address);
3257 dp = XFS_BHVTOI(dir_bdp);
3259 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3260 return XFS_ERROR(EIO);
3263 lock_mode = xfs_ilock_map_shared(dp);
3264 start_offset = uiop->uio_offset;
3265 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3266 if (start_offset != uiop->uio_offset) {
3267 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3269 xfs_iunlock_map_shared(dp, lock_mode);
3280 bhv_desc_t *dir_bdp,
3293 xfs_bmap_free_t free_list;
3294 xfs_fsblock_t first_block;
3295 boolean_t dp_joined_to_trans;
3299 xfs_fileoff_t first_fsb;
3300 xfs_filblks_t fs_blocks;
3302 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3309 struct xfs_dquot *udqp, *gdqp;
3311 char *link_name = VNAME(dentry);
3315 dir_vp = BHV_TO_VNODE(dir_bdp);
3316 dp = XFS_BHVTOI(dir_bdp);
3317 dp_joined_to_trans = B_FALSE;
3322 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3326 if (XFS_FORCED_SHUTDOWN(mp))
3327 return XFS_ERROR(EIO);
3329 link_namelen = VNAMELEN(dentry);
3332 * Check component lengths of the target path name.
3334 pathlen = strlen(target_path);
3335 if (pathlen >= MAXPATHLEN) /* total string too long */
3336 return XFS_ERROR(ENAMETOOLONG);
3337 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3341 for(total = 0, path = target_path; total < pathlen;) {
3345 while(*path == '/') {
3351 * Count up to the next slash or end of path.
3352 * Error out if the component is bigger than MAXNAMELEN.
3354 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3355 if (++len >= MAXNAMELEN) {
3356 error = ENAMETOOLONG;
3363 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3364 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3365 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3366 link_name, target_path, 0, 0, 0);
3371 /* Return through std_return after this point. */
3374 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3375 prid = dp->i_d.di_projid;
3376 else if (vap->va_mask & XFS_AT_PROJID)
3377 prid = (xfs_prid_t)vap->va_projid;
3379 prid = (xfs_prid_t)dfltprid;
3382 * Make sure that we have allocated dquot(s) on disk.
3384 error = XFS_QM_DQVOPALLOC(mp, dp,
3385 current_fsuid(credp), current_fsgid(credp), prid,
3386 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3390 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3391 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3393 * The symlink will fit into the inode data fork?
3394 * There can't be any attributes so we get the whole variable part.
3396 if (pathlen <= XFS_LITINO(mp))
3399 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3400 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3401 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3402 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3403 if (error == ENOSPC && fs_blocks == 0) {
3405 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3406 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3414 xfs_ilock(dp, XFS_ILOCK_EXCL);
3417 * Check whether the directory allows new symlinks or not.
3419 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3420 error = XFS_ERROR(EPERM);
3425 * Reserve disk quota : blocks and inode.
3427 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3432 * Check for ability to enter directory entry, if no space reserved.
3435 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3438 * Initialize the bmap freelist prior to calling either
3439 * bmapi or the directory create code.
3441 XFS_BMAP_INIT(&free_list, &first_block);
3444 * Allocate an inode for the symlink.
3446 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3447 1, 0, credp, prid, resblks > 0, &ip, NULL);
3449 if (error == ENOSPC)
3456 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3457 dp_joined_to_trans = B_TRUE;
3460 * Also attach the dquot(s) to it, if applicable.
3462 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3465 resblks -= XFS_IALLOC_SPACE_RES(mp);
3467 * If the symlink will fit into the inode, write it inline.
3469 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3470 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3471 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3472 ip->i_d.di_size = pathlen;
3475 * The inode was initially created in extent format.
3477 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3478 ip->i_df.if_flags |= XFS_IFINLINE;
3480 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3481 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3485 nmaps = SYMLINK_MAPS;
3487 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3488 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3489 &first_block, resblks, mval, &nmaps,
3496 resblks -= fs_blocks;
3497 ip->i_d.di_size = pathlen;
3498 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3500 cur_chunk = target_path;
3501 for (n = 0; n < nmaps; n++) {
3502 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3503 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3504 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3505 BTOBB(byte_cnt), 0);
3506 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3507 if (pathlen < byte_cnt) {
3510 pathlen -= byte_cnt;
3512 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3513 cur_chunk += byte_cnt;
3515 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3520 * Create the directory entry for the symlink.
3522 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3523 ip->i_ino, &first_block, &free_list, resblks);
3527 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3528 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3531 * Bump the in memory version number of the parent directory
3532 * so that other processes accessing it will recognize that
3533 * the directory has changed.
3538 * If this is a synchronous mount, make sure that the
3539 * symlink transaction goes to disk before returning to
3542 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3543 xfs_trans_set_sync(tp);
3547 * xfs_trans_commit normally decrements the vnode ref count
3548 * when it unlocks the inode. Since we want to return the
3549 * vnode to the caller, we bump the vnode ref count now.
3553 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3557 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3558 XFS_QM_DQRELE(mp, udqp);
3559 XFS_QM_DQRELE(mp, gdqp);
3561 /* Fall through to std_return with error = 0 or errno from
3562 * xfs_trans_commit */
3564 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3565 DM_EVENT_POSTSYMLINK)) {
3566 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3567 dir_vp, DM_RIGHT_NULL,
3568 error ? NULL : XFS_ITOV(ip),
3569 DM_RIGHT_NULL, link_name, target_path,
3585 xfs_bmap_cancel(&free_list);
3586 cancel_flags |= XFS_TRANS_ABORT;
3588 xfs_trans_cancel(tp, cancel_flags);
3589 XFS_QM_DQRELE(mp, udqp);
3590 XFS_QM_DQRELE(mp, gdqp);
3592 if (!dp_joined_to_trans && (dp != NULL)) {
3593 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3603 * A fid routine that takes a pointer to a previously allocated
3604 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3614 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3615 (inst_t *)__return_address);
3616 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3618 xfid = (xfs_fid2_t *)fidp;
3619 ip = XFS_BHVTOI(bdp);
3620 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3623 * use memcpy because the inode is a long long and there's no
3624 * assurance that xfid->fid_ino is properly aligned.
3626 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3627 xfid->fid_gen = ip->i_d.di_gen;
3644 vp = BHV_TO_VNODE(bdp);
3647 ip = XFS_BHVTOI(bdp);
3648 if (locktype == VRWLOCK_WRITE) {
3649 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3650 } else if (locktype == VRWLOCK_TRY_READ) {
3651 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3652 } else if (locktype == VRWLOCK_TRY_WRITE) {
3653 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3655 ASSERT((locktype == VRWLOCK_READ) ||
3656 (locktype == VRWLOCK_WRITE_DIRECT));
3657 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3675 vp = BHV_TO_VNODE(bdp);
3678 ip = XFS_BHVTOI(bdp);
3679 if (locktype == VRWLOCK_WRITE) {
3681 * In the write case, we may have added a new entry to
3682 * the reference cache. This might store a pointer to
3683 * an inode to be released in this inode. If it is there,
3684 * clear the pointer and release the inode after unlocking
3687 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3689 ASSERT((locktype == VRWLOCK_READ) ||
3690 (locktype == VRWLOCK_WRITE_DIRECT));
3691 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3703 xfs_inode_log_item_t *iip;
3706 ip = XFS_BHVTOI(bdp);
3710 if (XFS_FORCED_SHUTDOWN(mp))
3711 return XFS_ERROR(EIO);
3714 * Bypass inodes which have already been cleaned by
3715 * the inode flush clustering code inside xfs_iflush
3717 if ((ip->i_update_core == 0) &&
3718 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3721 if (flags & FLUSH_LOG) {
3722 if (iip && iip->ili_last_lsn) {
3723 xlog_t *log = mp->m_log;
3725 int s, log_flags = XFS_LOG_FORCE;
3727 s = GRANT_LOCK(log);
3728 sync_lsn = log->l_last_sync_lsn;
3729 GRANT_UNLOCK(log, s);
3731 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3734 if (flags & FLUSH_SYNC)
3735 log_flags |= XFS_LOG_SYNC;
3736 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3741 * We make this non-blocking if the inode is contended,
3742 * return EAGAIN to indicate to the caller that they
3743 * did not succeed. This prevents the flush path from
3744 * blocking on inodes inside another operation right
3745 * now, they get caught later by xfs_sync.
3747 if (flags & FLUSH_INODE) {
3750 if (xfs_ipincount(ip))
3753 if (flags & FLUSH_SYNC) {
3754 xfs_ilock(ip, XFS_ILOCK_SHARED);
3756 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3757 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3758 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3765 if (flags & FLUSH_SYNC)
3766 flush_flags = XFS_IFLUSH_SYNC;
3768 flush_flags = XFS_IFLUSH_ASYNC;
3770 error = xfs_iflush(ip, flush_flags);
3771 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3790 if (!capable(CAP_SYS_ADMIN))
3791 return XFS_ERROR(EPERM);
3793 ip = XFS_BHVTOI(bdp);
3796 if (XFS_FORCED_SHUTDOWN(mp))
3797 return XFS_ERROR(EIO);
3799 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3800 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3802 xfs_trans_cancel(tp, 0);
3805 xfs_ilock(ip, XFS_ILOCK_EXCL);
3806 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3808 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3809 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3811 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3813 error = xfs_trans_commit(tp, 0, NULL);
3829 vp = BHV_TO_VNODE(bdp);
3830 ip = XFS_BHVTOI(bdp);
3832 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3834 ASSERT(!VN_MAPPED(vp));
3836 /* bad inode, get out here ASAP */
3844 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3845 ASSERT(VN_CACHED(vp) == 0);
3847 /* If we have nothing to flush with this inode then complete the
3848 * teardown now, otherwise break the link between the xfs inode
3849 * and the linux inode and clean up the xfs inode later. This
3850 * avoids flushing the inode to disk during the delete operation
3853 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3854 xfs_ilock(ip, XFS_ILOCK_EXCL);
3856 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3858 xfs_mount_t *mp = ip->i_mount;
3860 /* Protect sync from us */
3861 XFS_MOUNT_ILOCK(mp);
3862 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3863 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3864 ip->i_flags |= XFS_IRECLAIMABLE;
3865 XFS_MOUNT_IUNLOCK(mp);
3876 xfs_ihash_t *ih = ip->i_hash;
3877 vnode_t *vp = XFS_ITOV_NULL(ip);
3880 if (vp && VN_BAD(vp))
3883 /* The hash lock here protects a thread in xfs_iget_core from
3884 * racing with us on linking the inode back with a vnode.
3885 * Once we have the XFS_IRECLAIM flag set it will not touch
3888 write_lock(&ih->ih_lock);
3889 if ((ip->i_flags & XFS_IRECLAIM) ||
3890 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3891 write_unlock(&ih->ih_lock);
3894 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3898 ip->i_flags |= XFS_IRECLAIM;
3899 write_unlock(&ih->ih_lock);
3902 * If the inode is still dirty, then flush it out. If the inode
3903 * is not in the AIL, then it will be OK to flush it delwri as
3904 * long as xfs_iflush() does not keep any references to the inode.
3905 * We leave that decision up to xfs_iflush() since it has the
3906 * knowledge of whether it's OK to simply do a delwri flush of
3907 * the inode or whether we need to wait until the inode is
3908 * pulled from the AIL.
3909 * We get the flush lock regardless, though, just to make sure
3910 * we don't free it while it is being flushed.
3912 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3914 xfs_ilock(ip, XFS_ILOCK_EXCL);
3918 if (ip->i_update_core ||
3919 ((ip->i_itemp != NULL) &&
3920 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3921 error = xfs_iflush(ip, sync_mode);
3923 * If we hit an error, typically because of filesystem
3924 * shutdown, we don't need to let vn_reclaim to know
3925 * because we're gonna reclaim the inode anyway.
3928 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3931 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3934 ASSERT(ip->i_update_core == 0);
3935 ASSERT(ip->i_itemp == NULL ||
3936 ip->i_itemp->ili_format.ilf_fields == 0);
3937 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3938 } else if (locked) {
3940 * We are not interested in doing an iflush if we're
3941 * in the process of shutting down the filesystem forcibly.
3942 * So, just reclaim the inode.
3945 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3954 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3957 xfs_inode_t *ip, *n;
3962 XFS_MOUNT_ILOCK(mp);
3963 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3965 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3967 if (xfs_ipincount(ip) ||
3968 !xfs_iflock_nowait(ip)) {
3969 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3973 XFS_MOUNT_IUNLOCK(mp);
3974 xfs_finish_reclaim(ip, noblock,
3975 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
3983 XFS_MOUNT_IUNLOCK(mp);
3988 * xfs_alloc_file_space()
3989 * This routine allocates disk space for the given file.
3991 * If alloc_type == 0, this request is for an ALLOCSP type
3992 * request which will change the file size. In this case, no
3993 * DMAPI event will be generated by the call. A TRUNCATE event
3994 * will be generated later by xfs_setattr.
3996 * If alloc_type != 0, this request is for a RESVSP type
3997 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3998 * lower block boundary byte address is less than the file's
4007 xfs_alloc_file_space(
4014 xfs_filblks_t allocated_fsb;
4015 xfs_filblks_t allocatesize_fsb;
4018 xfs_filblks_t datablocks;
4020 xfs_fsblock_t firstfsb;
4021 xfs_bmap_free_t free_list;
4022 xfs_bmbt_irec_t *imapp;
4023 xfs_bmbt_irec_t imaps[1];
4030 xfs_fileoff_t startoffset_fsb;
4032 int xfs_bmapi_flags;
4034 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4037 if (XFS_FORCED_SHUTDOWN(mp))
4038 return XFS_ERROR(EIO);
4041 * determine if this is a realtime file
4043 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4044 if (ip->i_d.di_extsize)
4045 rtextsize = ip->i_d.di_extsize;
4047 rtextsize = mp->m_sb.sb_rextsize;
4051 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4055 return XFS_ERROR(EINVAL);
4061 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4062 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4063 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4065 /* Generate a DMAPI event if needed. */
4066 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4067 (attr_flags&ATTR_DMI) == 0 &&
4068 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4069 xfs_off_t end_dmi_offset;
4071 end_dmi_offset = offset+len;
4072 if (end_dmi_offset > ip->i_d.di_size)
4073 end_dmi_offset = ip->i_d.di_size;
4074 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4075 offset, end_dmi_offset - offset,
4082 * allocate file space until done or until there is an error
4085 while (allocatesize_fsb && !error) {
4087 * determine if reserving space on
4088 * the data or realtime partition.
4093 s = startoffset_fsb;
4094 do_div(s, rtextsize);
4096 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4098 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4101 datablocks = allocatesize_fsb;
4106 * allocate and setup the transaction
4108 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4109 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4110 error = xfs_trans_reserve(tp,
4112 XFS_WRITE_LOG_RES(mp),
4114 XFS_TRANS_PERM_LOG_RES,
4115 XFS_WRITE_LOG_COUNT);
4118 * check for running out of space
4122 * Free the transaction structure.
4124 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4125 xfs_trans_cancel(tp, 0);
4128 xfs_ilock(ip, XFS_ILOCK_EXCL);
4129 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4130 ip->i_udquot, ip->i_gdquot, resblks, 0, 0);
4134 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4135 xfs_trans_ihold(tp, ip);
4138 * issue the bmapi() call to allocate the blocks
4140 XFS_BMAP_INIT(&free_list, &firstfsb);
4141 error = xfs_bmapi(tp, ip, startoffset_fsb,
4142 allocatesize_fsb, xfs_bmapi_flags,
4143 &firstfsb, 0, imapp, &reccount,
4150 * complete the transaction
4152 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4157 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4158 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4163 allocated_fsb = imapp->br_blockcount;
4165 if (reccount == 0) {
4166 error = XFS_ERROR(ENOSPC);
4170 startoffset_fsb += allocated_fsb;
4171 allocatesize_fsb -= allocated_fsb;
4174 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4175 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4177 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4178 XFS_ITOV(ip), DM_RIGHT_NULL,
4179 XFS_ITOV(ip), DM_RIGHT_NULL,
4180 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4182 goto retry; /* Maybe DMAPI app. has made space */
4183 /* else fall through with error from XFS_SEND_DATA */
4189 xfs_bmap_cancel(&free_list);
4191 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4192 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4193 goto dmapi_enospc_check;
4197 * Zero file bytes between startoff and endoff inclusive.
4198 * The iolock is held exclusive and no blocks are buffered.
4201 xfs_zero_remaining_bytes(
4206 xfs_bmbt_irec_t imap;
4207 xfs_fileoff_t offset_fsb;
4208 xfs_off_t lastoffset;
4211 xfs_mount_t *mp = ip->i_mount;
4215 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4216 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4217 mp->m_rtdev_targp : mp->m_ddev_targp);
4219 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4220 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4222 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4224 if (error || nimap < 1)
4226 ASSERT(imap.br_blockcount >= 1);
4227 ASSERT(imap.br_startoff == offset_fsb);
4228 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4229 if (lastoffset > endoff)
4230 lastoffset = endoff;
4231 if (imap.br_startblock == HOLESTARTBLOCK)
4233 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4234 if (imap.br_state == XFS_EXT_UNWRITTEN)
4237 XFS_BUF_UNWRITE(bp);
4239 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4241 if ((error = xfs_iowait(bp))) {
4242 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4243 mp, bp, XFS_BUF_ADDR(bp));
4246 memset(XFS_BUF_PTR(bp) +
4247 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4248 0, lastoffset - offset + 1);
4253 if ((error = xfs_iowait(bp))) {
4254 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4255 mp, bp, XFS_BUF_ADDR(bp));
4264 * xfs_free_file_space()
4265 * This routine frees disk space for the given file.
4267 * This routine is only called by xfs_change_file_space
4268 * for an UNRESVSP type call.
4276 xfs_free_file_space(
4285 xfs_off_t end_dmi_offset;
4286 xfs_fileoff_t endoffset_fsb;
4288 xfs_fsblock_t firstfsb;
4289 xfs_bmap_free_t free_list;
4291 xfs_bmbt_irec_t imap;
4299 xfs_fileoff_t startoffset_fsb;
4301 int need_iolock = 1;
4306 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4308 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4312 if (len <= 0) /* if nothing being freed */
4314 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4315 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4316 end_dmi_offset = offset + len;
4317 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4319 if (offset < ip->i_d.di_size &&
4320 (attr_flags & ATTR_DMI) == 0 &&
4321 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4322 if (end_dmi_offset > ip->i_d.di_size)
4323 end_dmi_offset = ip->i_d.di_size;
4324 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4325 offset, end_dmi_offset - offset,
4326 AT_DELAY_FLAG(attr_flags), NULL);
4331 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4332 if (attr_flags & ATTR_NOLOCK)
4335 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4337 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4339 ilen = len + (offset & (rounding - 1));
4340 ioffset = offset & ~(rounding - 1);
4341 if (ilen & (rounding - 1))
4342 ilen = (ilen + rounding) & ~(rounding - 1);
4344 if (VN_CACHED(vp) != 0) {
4345 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4346 ctooff(offtoct(ioffset)), -1);
4347 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4348 -1, FI_REMAPF_LOCKED);
4352 * Need to zero the stuff we're not freeing, on disk.
4353 * If its a realtime file & can't use unwritten extents then we
4354 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4355 * will take care of it for us.
4357 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4359 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4360 &imap, &nimap, NULL);
4362 goto out_unlock_iolock;
4363 ASSERT(nimap == 0 || nimap == 1);
4364 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4367 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4368 block = imap.br_startblock;
4369 mod = do_div(block, mp->m_sb.sb_rextsize);
4371 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4374 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4375 &imap, &nimap, NULL);
4377 goto out_unlock_iolock;
4378 ASSERT(nimap == 0 || nimap == 1);
4379 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4380 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4382 if (mod && (mod != mp->m_sb.sb_rextsize))
4383 endoffset_fsb -= mod;
4386 if ((done = (endoffset_fsb <= startoffset_fsb)))
4388 * One contiguous piece to clear
4390 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4393 * Some full blocks, possibly two pieces to clear
4395 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4396 error = xfs_zero_remaining_bytes(ip, offset,
4397 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4399 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4400 error = xfs_zero_remaining_bytes(ip,
4401 XFS_FSB_TO_B(mp, endoffset_fsb),
4406 * free file space until done or until there is an error
4408 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4409 while (!error && !done) {
4412 * allocate and setup the transaction
4414 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4415 error = xfs_trans_reserve(tp,
4417 XFS_WRITE_LOG_RES(mp),
4419 XFS_TRANS_PERM_LOG_RES,
4420 XFS_WRITE_LOG_COUNT);
4423 * check for running out of space
4427 * Free the transaction structure.
4429 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4430 xfs_trans_cancel(tp, 0);
4433 xfs_ilock(ip, XFS_ILOCK_EXCL);
4434 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4435 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4436 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4440 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4441 xfs_trans_ihold(tp, ip);
4444 * issue the bunmapi() call to free the blocks
4446 XFS_BMAP_INIT(&free_list, &firstfsb);
4447 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4448 endoffset_fsb - startoffset_fsb,
4449 0, 2, &firstfsb, &free_list, &done);
4455 * complete the transaction
4457 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4462 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4463 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4468 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4472 xfs_bmap_cancel(&free_list);
4474 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4475 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4481 * xfs_change_file_space()
4482 * This routine allocates or frees disk space for the given file.
4483 * The user specified parameters are checked for alignment and size
4492 xfs_change_file_space(
4506 xfs_off_t startoffset;
4512 vp = BHV_TO_VNODE(bdp);
4513 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4515 ip = XFS_BHVTOI(bdp);
4519 * must be a regular file and have write permission
4522 return XFS_ERROR(EINVAL);
4524 xfs_ilock(ip, XFS_ILOCK_SHARED);
4526 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4527 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4531 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4533 switch (bf->l_whence) {
4534 case 0: /*SEEK_SET*/
4536 case 1: /*SEEK_CUR*/
4537 bf->l_start += offset;
4539 case 2: /*SEEK_END*/
4540 bf->l_start += ip->i_d.di_size;
4543 return XFS_ERROR(EINVAL);
4546 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4548 if ( (bf->l_start < 0)
4549 || (bf->l_start > XFS_MAXIOFFSET(mp))
4550 || (bf->l_start + llen < 0)
4551 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4552 return XFS_ERROR(EINVAL);
4556 startoffset = bf->l_start;
4557 fsize = ip->i_d.di_size;
4560 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4562 * These calls do NOT zero the data space allocated to the file,
4563 * nor do they change the file size.
4565 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4567 * These calls cause the new file data to be zeroed and the file
4568 * size to be changed.
4570 setprealloc = clrprealloc = 0;
4573 case XFS_IOC_RESVSP:
4574 case XFS_IOC_RESVSP64:
4575 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4582 case XFS_IOC_UNRESVSP:
4583 case XFS_IOC_UNRESVSP64:
4584 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4589 case XFS_IOC_ALLOCSP:
4590 case XFS_IOC_ALLOCSP64:
4591 case XFS_IOC_FREESP:
4592 case XFS_IOC_FREESP64:
4593 if (startoffset > fsize) {
4594 error = xfs_alloc_file_space(ip, fsize,
4595 startoffset - fsize, 0, attr_flags);
4600 va.va_mask = XFS_AT_SIZE;
4601 va.va_size = startoffset;
4603 error = xfs_setattr(bdp, &va, attr_flags, credp);
4613 return XFS_ERROR(EINVAL);
4617 * update the inode timestamp, mode, and prealloc flag bits
4619 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4621 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4624 xfs_trans_cancel(tp, 0);
4628 xfs_ilock(ip, XFS_ILOCK_EXCL);
4630 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4631 xfs_trans_ihold(tp, ip);
4633 if ((attr_flags & ATTR_DMI) == 0) {
4634 ip->i_d.di_mode &= ~S_ISUID;
4637 * Note that we don't have to worry about mandatory
4638 * file locking being disabled here because we only
4639 * clear the S_ISGID bit if the Group execute bit is
4640 * on, but if it was on then mandatory locking wouldn't
4641 * have been enabled.
4643 if (ip->i_d.di_mode & S_IXGRP)
4644 ip->i_d.di_mode &= ~S_ISGID;
4646 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4649 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4650 else if (clrprealloc)
4651 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4653 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4654 xfs_trans_set_sync(tp);
4656 error = xfs_trans_commit(tp, 0, NULL);
4658 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4663 vnodeops_t xfs_vnodeops = {
4664 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4665 .vop_open = xfs_open,
4666 .vop_read = xfs_read,
4667 #ifdef HAVE_SENDFILE
4668 .vop_sendfile = xfs_sendfile,
4670 .vop_write = xfs_write,
4671 .vop_ioctl = xfs_ioctl,
4672 .vop_getattr = xfs_getattr,
4673 .vop_setattr = xfs_setattr,
4674 .vop_access = xfs_access,
4675 .vop_lookup = xfs_lookup,
4676 .vop_create = xfs_create,
4677 .vop_remove = xfs_remove,
4678 .vop_link = xfs_link,
4679 .vop_rename = xfs_rename,
4680 .vop_mkdir = xfs_mkdir,
4681 .vop_rmdir = xfs_rmdir,
4682 .vop_readdir = xfs_readdir,
4683 .vop_symlink = xfs_symlink,
4684 .vop_readlink = xfs_readlink,
4685 .vop_fsync = xfs_fsync,
4686 .vop_inactive = xfs_inactive,
4687 .vop_fid2 = xfs_fid2,
4688 .vop_rwlock = xfs_rwlock,
4689 .vop_rwunlock = xfs_rwunlock,
4690 .vop_bmap = xfs_bmap,
4691 .vop_reclaim = xfs_reclaim,
4692 .vop_attr_get = xfs_attr_get,
4693 .vop_attr_set = xfs_attr_set,
4694 .vop_attr_remove = xfs_attr_remove,
4695 .vop_attr_list = xfs_attr_list,
4696 .vop_link_removed = (vop_link_removed_t)fs_noval,
4697 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4698 .vop_tosspages = fs_tosspages,
4699 .vop_flushinval_pages = fs_flushinval_pages,
4700 .vop_flush_pages = fs_flush_pages,
4701 .vop_release = xfs_release,
4702 .vop_iflush = xfs_inode_flush,