2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_dir_leaf.h"
42 #include "xfs_itable.h"
43 #include "xfs_btree.h"
44 #include "xfs_ialloc.h"
45 #include "xfs_alloc.h"
49 #include "xfs_error.h"
50 #include "xfs_quota.h"
51 #include "xfs_utils.h"
52 #include "xfs_rtalloc.h"
53 #include "xfs_refcache.h"
54 #include "xfs_trans_space.h"
55 #include "xfs_log_priv.h"
60 * The maximum pathlen is 1024 bytes. Since the minimum file system
61 * blocksize is 512 bytes, we can get a max of 2 extents back from
64 #define SYMLINK_MAPS 2
67 * For xfs, we check that the file isn't too big to be opened by this kernel.
68 * No other open action is required for regular files. Devices are handled
69 * through the specfs file system, pipes through fifofs. Device and
70 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
71 * when a new vnode is first looked up or created.
82 vp = BHV_TO_VNODE(bdp);
85 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
86 return XFS_ERROR(EIO);
89 * If it's a directory with any blocks, read-ahead block 0
90 * as we're almost certain to have the next operation be a read there.
92 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
93 mode = xfs_ilock_map_shared(ip);
94 if (ip->i_d.di_nextents > 0)
95 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
96 xfs_iunlock(ip, mode);
116 vp = BHV_TO_VNODE(bdp);
117 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
119 ip = XFS_BHVTOI(bdp);
122 if (XFS_FORCED_SHUTDOWN(mp))
123 return XFS_ERROR(EIO);
125 if (!(flags & ATTR_LAZY))
126 xfs_ilock(ip, XFS_ILOCK_SHARED);
128 vap->va_size = ip->i_d.di_size;
129 if (vap->va_mask == XFS_AT_SIZE)
133 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
134 vap->va_nodeid = ip->i_ino;
136 vap->va_nodeid += mp->m_inoadd;
138 vap->va_nlink = ip->i_d.di_nlink;
141 * Quick exit for non-stat callers
144 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
145 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
149 * Copy from in-core inode.
151 vap->va_mode = ip->i_d.di_mode;
152 vap->va_uid = ip->i_d.di_uid;
153 vap->va_gid = ip->i_d.di_gid;
154 vap->va_projid = ip->i_d.di_projid;
157 * Check vnode type block/char vs. everything else.
159 switch (ip->i_d.di_mode & S_IFMT) {
162 vap->va_rdev = ip->i_df.if_u2.if_rdev;
163 vap->va_blocksize = BLKDEV_IOSIZE;
168 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
169 vap->va_blocksize = xfs_preferred_iosize(mp);
173 * If the file blocks are being allocated from a
174 * realtime partition, then return the inode's
175 * realtime extent size or the realtime volume's
178 vap->va_blocksize = ip->i_d.di_extsize ?
179 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
180 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
185 vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
186 vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
187 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
188 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
189 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
190 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
193 * Exit for stat callers. See if any of the rest of the fields
194 * to be filled in are needed.
197 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
198 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
202 * Convert di_flags to xflags.
204 vap->va_xflags = xfs_ip2xflags(ip);
207 * Exit for inode revalidate. See if any of the rest of
208 * the fields to be filled in are needed.
211 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
212 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
215 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
217 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
218 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
222 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
223 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
224 ip->i_d.di_anextents;
226 vap->va_anextents = 0;
227 vap->va_gen = ip->i_d.di_gen;
230 if (!(flags & ATTR_LAZY))
231 xfs_iunlock(ip, XFS_ILOCK_SHARED);
257 xfs_prid_t projid=0, iprojid=0;
258 int mandlock_before, mandlock_after;
259 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
263 vp = BHV_TO_VNODE(bdp);
264 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
266 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
267 return XFS_ERROR(EROFS);
270 * Cannot set certain attributes.
273 if (mask & XFS_AT_NOSET) {
274 return XFS_ERROR(EINVAL);
277 ip = XFS_BHVTOI(bdp);
280 if (XFS_FORCED_SHUTDOWN(mp))
281 return XFS_ERROR(EIO);
284 * Timestamps do not need to be logged and hence do not
285 * need to be done within a transaction.
287 if (mask & XFS_AT_UPDTIMES) {
288 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
289 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
290 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
291 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
292 xfs_ichgtime(ip, timeflags);
296 olddquot1 = olddquot2 = NULL;
300 * If disk quotas is on, we make sure that the dquots do exist on disk,
301 * before we start any other transactions. Trying to do this later
302 * is messy. We don't care to take a readlock to look at the ids
303 * in inode here, because we can't hold it across the trans_reserve.
304 * If the IDs do change before we take the ilock, we're covered
305 * because the i_*dquot fields will get updated anyway.
307 if (XFS_IS_QUOTA_ON(mp) &&
308 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
311 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
313 qflags |= XFS_QMOPT_UQUOTA;
315 uid = ip->i_d.di_uid;
317 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
319 qflags |= XFS_QMOPT_GQUOTA;
321 gid = ip->i_d.di_gid;
323 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
324 projid = vap->va_projid;
325 qflags |= XFS_QMOPT_PQUOTA;
327 projid = ip->i_d.di_projid;
330 * We take a reference when we initialize udqp and gdqp,
331 * so it is important that we never blindly double trip on
332 * the same variable. See xfs_create() for an example.
334 ASSERT(udqp == NULL);
335 ASSERT(gdqp == NULL);
336 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
343 * For the other attributes, we acquire the inode lock and
344 * first do an error checking pass.
347 lock_flags = XFS_ILOCK_EXCL;
348 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
349 if (flags & ATTR_NOLOCK)
351 if (!(mask & XFS_AT_SIZE)) {
352 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
353 (mp->m_flags & XFS_MOUNT_WSYNC)) {
354 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
356 if ((code = xfs_trans_reserve(tp, 0,
357 XFS_ICHANGE_LOG_RES(mp), 0,
364 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
365 !(flags & ATTR_DMI)) {
366 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
367 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
368 vap->va_size, 0, dmflags, NULL);
375 lock_flags |= XFS_IOLOCK_EXCL;
378 xfs_ilock(ip, lock_flags);
380 /* boolean: are we the file owner? */
381 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
384 * Change various properties of a file.
385 * Only the owner or users with CAP_FOWNER
386 * capability may do these things.
389 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
390 XFS_AT_GID|XFS_AT_PROJID)) {
392 * CAP_FOWNER overrides the following restrictions:
394 * The user ID of the calling process must be equal
395 * to the file owner ID, except in cases where the
396 * CAP_FSETID capability is applicable.
398 if (!file_owner && !capable(CAP_FOWNER)) {
399 code = XFS_ERROR(EPERM);
404 * CAP_FSETID overrides the following restrictions:
406 * The effective user ID of the calling process shall match
407 * the file owner when setting the set-user-ID and
408 * set-group-ID bits on that file.
410 * The effective group ID or one of the supplementary group
411 * IDs of the calling process shall match the group owner of
412 * the file when setting the set-group-ID bit on that file
414 if (mask & XFS_AT_MODE) {
417 if ((vap->va_mode & S_ISUID) && !file_owner)
419 if ((vap->va_mode & S_ISGID) &&
420 !in_group_p((gid_t)ip->i_d.di_gid))
423 /* Linux allows this, Irix doesn't. */
424 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
427 if (m && !capable(CAP_FSETID))
433 * Change file ownership. Must be the owner or privileged.
434 * If the system was configured with the "restricted_chown"
435 * option, the owner is not permitted to give away the file,
436 * and can change the group id only to a group of which he
437 * or she is a member.
439 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
441 * These IDs could have changed since we last looked at them.
442 * But, we're assured that if the ownership did change
443 * while we didn't have the inode locked, inode's dquot(s)
444 * would have changed also.
446 iuid = ip->i_d.di_uid;
447 iprojid = ip->i_d.di_projid;
448 igid = ip->i_d.di_gid;
449 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
450 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
451 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
455 * CAP_CHOWN overrides the following restrictions:
457 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
458 * shall override the restriction that a process cannot
459 * change the user ID of a file it owns and the restriction
460 * that the group ID supplied to the chown() function
461 * shall be equal to either the group ID or one of the
462 * supplementary group IDs of the calling process.
464 if (restricted_chown &&
465 (iuid != uid || (igid != gid &&
466 !in_group_p((gid_t)gid))) &&
467 !capable(CAP_CHOWN)) {
468 code = XFS_ERROR(EPERM);
472 * Do a quota reservation only if uid/projid/gid is actually
475 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
476 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
477 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
479 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
480 capable(CAP_FOWNER) ?
481 XFS_QMOPT_FORCE_RES : 0);
482 if (code) /* out of quota */
488 * Truncate file. Must have write permission and not be a directory.
490 if (mask & XFS_AT_SIZE) {
491 /* Short circuit the truncate case for zero length files */
492 if ((vap->va_size == 0) &&
493 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
494 xfs_iunlock(ip, XFS_ILOCK_EXCL);
495 lock_flags &= ~XFS_ILOCK_EXCL;
496 if (mask & XFS_AT_CTIME)
497 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
503 code = XFS_ERROR(EISDIR);
505 } else if (!VN_ISREG(vp)) {
506 code = XFS_ERROR(EINVAL);
510 * Make sure that the dquots are attached to the inode.
512 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
517 * Change file access or modified times.
519 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
521 if ((flags & ATTR_UTIME) &&
522 !capable(CAP_FOWNER)) {
523 code = XFS_ERROR(EPERM);
530 * Change extent size or realtime flag.
532 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
534 * Can't change extent size if any extents are allocated.
536 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
537 (mask & XFS_AT_EXTSIZE) &&
538 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
540 code = XFS_ERROR(EINVAL); /* EFBIG? */
545 * Can't set extent size unless the file is marked, or
546 * about to be marked as a realtime file.
548 * This check will be removed when fixed size extents
549 * with buffered data writes is implemented.
552 if ((mask & XFS_AT_EXTSIZE) &&
553 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
555 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
556 ((mask & XFS_AT_XFLAGS) &&
557 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
558 code = XFS_ERROR(EINVAL);
563 * Can't change realtime flag if any extents are allocated.
565 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
566 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
567 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
568 code = XFS_ERROR(EINVAL); /* EFBIG? */
572 * Extent size must be a multiple of the appropriate block
573 * size, if set at all.
575 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
578 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
579 ((mask & XFS_AT_XFLAGS) &&
580 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
581 size = mp->m_sb.sb_rextsize <<
582 mp->m_sb.sb_blocklog;
584 size = mp->m_sb.sb_blocksize;
586 if (vap->va_extsize % size) {
587 code = XFS_ERROR(EINVAL);
592 * If realtime flag is set then must have realtime data.
594 if ((mask & XFS_AT_XFLAGS) &&
595 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
596 if ((mp->m_sb.sb_rblocks == 0) ||
597 (mp->m_sb.sb_rextsize == 0) ||
598 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
599 code = XFS_ERROR(EINVAL);
605 * Can't modify an immutable/append-only file unless
606 * we have appropriate permission.
608 if ((mask & XFS_AT_XFLAGS) &&
610 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
612 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
613 !capable(CAP_LINUX_IMMUTABLE)) {
614 code = XFS_ERROR(EPERM);
620 * Now we can make the changes. Before we join the inode
621 * to the transaction, if XFS_AT_SIZE is set then take care of
622 * the part of the truncation that must be done without the
623 * inode lock. This needs to be done before joining the inode
624 * to the transaction, because the inode cannot be unlocked
625 * once it is a part of the transaction.
627 if (mask & XFS_AT_SIZE) {
629 if (vap->va_size > ip->i_d.di_size)
630 code = xfs_igrow_start(ip, vap->va_size, credp);
631 xfs_iunlock(ip, XFS_ILOCK_EXCL);
633 code = xfs_itruncate_data(ip, vap->va_size);
636 lock_flags &= ~XFS_ILOCK_EXCL;
637 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
640 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
641 if ((code = xfs_trans_reserve(tp, 0,
642 XFS_ITRUNCATE_LOG_RES(mp), 0,
643 XFS_TRANS_PERM_LOG_RES,
644 XFS_ITRUNCATE_LOG_COUNT))) {
645 xfs_trans_cancel(tp, 0);
647 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
650 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
651 xfs_ilock(ip, XFS_ILOCK_EXCL);
655 xfs_trans_ijoin(tp, ip, lock_flags);
656 xfs_trans_ihold(tp, ip);
659 /* determine whether mandatory locking mode changes */
660 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
663 * Truncate file. Must have write permission and not be a directory.
665 if (mask & XFS_AT_SIZE) {
666 if (vap->va_size > ip->i_d.di_size) {
667 xfs_igrow_finish(tp, ip, vap->va_size,
668 !(flags & ATTR_DMI));
669 } else if ((vap->va_size <= ip->i_d.di_size) ||
670 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
672 * signal a sync transaction unless
673 * we're truncating an already unlinked
674 * file on a wsync filesystem
676 code = xfs_itruncate_finish(&tp, ip,
677 (xfs_fsize_t)vap->va_size,
679 ((ip->i_d.di_nlink != 0 ||
680 !(mp->m_flags & XFS_MOUNT_WSYNC))
687 * Have to do this even if the file's size doesn't change.
689 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
693 * Change file access modes.
695 if (mask & XFS_AT_MODE) {
696 ip->i_d.di_mode &= S_IFMT;
697 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
699 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
700 timeflags |= XFS_ICHGTIME_CHG;
704 * Change file ownership. Must be the owner or privileged.
705 * If the system was configured with the "restricted_chown"
706 * option, the owner is not permitted to give away the file,
707 * and can change the group id only to a group of which he
708 * or she is a member.
710 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
712 * CAP_FSETID overrides the following restrictions:
714 * The set-user-ID and set-group-ID bits of a file will be
715 * cleared upon successful return from chown()
717 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
718 !capable(CAP_FSETID)) {
719 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
723 * Change the ownerships and register quota modifications
724 * in the transaction.
727 if (XFS_IS_UQUOTA_ON(mp)) {
728 ASSERT(mask & XFS_AT_UID);
730 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
731 &ip->i_udquot, udqp);
733 ip->i_d.di_uid = uid;
736 if (XFS_IS_GQUOTA_ON(mp)) {
737 ASSERT(!XFS_IS_PQUOTA_ON(mp));
738 ASSERT(mask & XFS_AT_GID);
740 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
741 &ip->i_gdquot, gdqp);
743 ip->i_d.di_gid = gid;
745 if (iprojid != projid) {
746 if (XFS_IS_PQUOTA_ON(mp)) {
747 ASSERT(!XFS_IS_GQUOTA_ON(mp));
748 ASSERT(mask & XFS_AT_PROJID);
750 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
751 &ip->i_gdquot, gdqp);
753 ip->i_d.di_projid = projid;
755 * We may have to rev the inode as well as
756 * the superblock version number since projids didn't
757 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
759 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
760 xfs_bump_ino_vers2(tp, ip);
763 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
764 timeflags |= XFS_ICHGTIME_CHG;
769 * Change file access or modified times.
771 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
772 if (mask & XFS_AT_ATIME) {
773 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
774 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
775 ip->i_update_core = 1;
776 timeflags &= ~XFS_ICHGTIME_ACC;
778 if (mask & XFS_AT_MTIME) {
779 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
780 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
781 timeflags &= ~XFS_ICHGTIME_MOD;
782 timeflags |= XFS_ICHGTIME_CHG;
784 if (tp && (flags & ATTR_UTIME))
785 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
789 * Change XFS-added attributes.
791 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
792 if (mask & XFS_AT_EXTSIZE) {
794 * Converting bytes to fs blocks.
796 ip->i_d.di_extsize = vap->va_extsize >>
797 mp->m_sb.sb_blocklog;
799 if (mask & XFS_AT_XFLAGS) {
802 /* can't set PREALLOC this way, just preserve it */
803 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
804 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
805 di_flags |= XFS_DIFLAG_IMMUTABLE;
806 if (vap->va_xflags & XFS_XFLAG_APPEND)
807 di_flags |= XFS_DIFLAG_APPEND;
808 if (vap->va_xflags & XFS_XFLAG_SYNC)
809 di_flags |= XFS_DIFLAG_SYNC;
810 if (vap->va_xflags & XFS_XFLAG_NOATIME)
811 di_flags |= XFS_DIFLAG_NOATIME;
812 if (vap->va_xflags & XFS_XFLAG_NODUMP)
813 di_flags |= XFS_DIFLAG_NODUMP;
814 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
815 di_flags |= XFS_DIFLAG_PROJINHERIT;
816 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
817 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
818 di_flags |= XFS_DIFLAG_RTINHERIT;
819 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
820 di_flags |= XFS_DIFLAG_NOSYMLINKS;
822 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
823 di_flags |= XFS_DIFLAG_REALTIME;
824 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
826 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
829 ip->i_d.di_flags = di_flags;
831 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
832 timeflags |= XFS_ICHGTIME_CHG;
836 * Change file inode change time only if XFS_AT_CTIME set
837 * AND we have been called by a DMI function.
840 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
841 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
842 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
843 ip->i_update_core = 1;
844 timeflags &= ~XFS_ICHGTIME_CHG;
848 * Send out timestamp changes that need to be set to the
849 * current time. Not done when called by a DMI function.
851 if (timeflags && !(flags & ATTR_DMI))
852 xfs_ichgtime(ip, timeflags);
854 XFS_STATS_INC(xs_ig_attrchg);
857 * If this is a synchronous mount, make sure that the
858 * transaction goes to disk before returning to the user.
859 * This is slightly sub-optimal in that truncates require
860 * two sync transactions instead of one for wsync filesytems.
861 * One for the truncate and one for the timestamps since we
862 * don't want to change the timestamps unless we're sure the
863 * truncate worked. Truncates are less than 1% of the laddis
864 * mix so this probably isn't worth the trouble to optimize.
868 if (mp->m_flags & XFS_MOUNT_WSYNC)
869 xfs_trans_set_sync(tp);
871 code = xfs_trans_commit(tp, commit_flags, NULL);
875 * If the (regular) file's mandatory locking mode changed, then
876 * notify the vnode. We do this under the inode lock to prevent
877 * racing calls to vop_vnode_change.
879 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
880 if (mandlock_before != mandlock_after) {
881 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
885 xfs_iunlock(ip, lock_flags);
888 * Release any dquot(s) the inode had kept before chown.
890 XFS_QM_DQRELE(mp, olddquot1);
891 XFS_QM_DQRELE(mp, olddquot2);
892 XFS_QM_DQRELE(mp, udqp);
893 XFS_QM_DQRELE(mp, gdqp);
899 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
900 !(flags & ATTR_DMI)) {
901 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
902 NULL, DM_RIGHT_NULL, NULL, NULL,
903 0, 0, AT_DELAY_FLAG(flags));
908 commit_flags |= XFS_TRANS_ABORT;
911 XFS_QM_DQRELE(mp, udqp);
912 XFS_QM_DQRELE(mp, gdqp);
914 xfs_trans_cancel(tp, commit_flags);
916 if (lock_flags != 0) {
917 xfs_iunlock(ip, lock_flags);
925 * Null conversion from vnode mode bits to inode mode bits, as in efs.
936 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
937 (inst_t *)__return_address);
939 ip = XFS_BHVTOI(bdp);
940 xfs_ilock(ip, XFS_ILOCK_SHARED);
941 error = xfs_iaccess(ip, mode, credp);
942 xfs_iunlock(ip, XFS_ILOCK_SHARED);
966 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
972 vp = BHV_TO_VNODE(bdp);
973 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
975 ip = XFS_BHVTOI(bdp);
978 if (XFS_FORCED_SHUTDOWN(mp))
979 return XFS_ERROR(EIO);
981 xfs_ilock(ip, XFS_ILOCK_SHARED);
983 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
985 offset = uiop->uio_offset;
986 count = uiop->uio_resid;
989 error = XFS_ERROR(EINVAL);
997 if (!(ioflags & IO_INVIS)) {
998 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1002 * See if the symlink is stored inline.
1004 pathlen = (int)ip->i_d.di_size;
1006 if (ip->i_df.if_flags & XFS_IFINLINE) {
1007 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1011 * Symlink not inline. Call bmap to get it in.
1013 nmaps = SYMLINK_MAPS;
1015 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1016 0, NULL, 0, mval, &nmaps, NULL);
1022 for (n = 0; n < nmaps; n++) {
1023 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1024 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1025 bp = xfs_buf_read(mp->m_ddev_targp, d,
1026 BTOBB(byte_cnt), 0);
1027 error = XFS_BUF_GETERROR(bp);
1029 xfs_ioerror_alert("xfs_readlink",
1030 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1034 if (pathlen < byte_cnt)
1036 pathlen -= byte_cnt;
1038 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1047 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1056 * This is called to sync the inode and its data out to disk.
1057 * We need to hold the I/O lock while flushing the data, and
1058 * the inode lock while flushing the inode. The inode lock CANNOT
1059 * be held while flushing the data, so acquire after we're done
1073 int log_flushed = 0, changed = 1;
1075 vn_trace_entry(BHV_TO_VNODE(bdp),
1076 __FUNCTION__, (inst_t *)__return_address);
1078 ip = XFS_BHVTOI(bdp);
1080 ASSERT(start >= 0 && stop >= -1);
1082 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1083 return XFS_ERROR(EIO);
1086 * We always need to make sure that the required inode state
1087 * is safe on disk. The vnode might be clean but because
1088 * of committed transactions that haven't hit the disk yet.
1089 * Likewise, there could be unflushed non-transactional
1090 * changes to the inode core that have to go to disk.
1092 * The following code depends on one assumption: that
1093 * any transaction that changes an inode logs the core
1094 * because it has to change some field in the inode core
1095 * (typically nextents or nblocks). That assumption
1096 * implies that any transactions against an inode will
1097 * catch any non-transactional updates. If inode-altering
1098 * transactions exist that violate this assumption, the
1099 * code breaks. Right now, it figures that if the involved
1100 * update_* field is clear and the inode is unpinned, the
1101 * inode is clean. Either it's been flushed or it's been
1102 * committed and the commit has hit the disk unpinning the inode.
1103 * (Note that xfs_inode_item_format() called at commit clears
1104 * the update_* fields.)
1106 xfs_ilock(ip, XFS_ILOCK_SHARED);
1108 /* If we are flushing data then we care about update_size
1109 * being set, otherwise we care about update_core
1111 if ((flag & FSYNC_DATA) ?
1112 (ip->i_update_size == 0) :
1113 (ip->i_update_core == 0)) {
1115 * Timestamps/size haven't changed since last inode
1116 * flush or inode transaction commit. That means
1117 * either nothing got written or a transaction
1118 * committed which caught the updates. If the
1119 * latter happened and the transaction hasn't
1120 * hit the disk yet, the inode will be still
1121 * be pinned. If it is, force the log.
1124 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1126 if (xfs_ipincount(ip)) {
1127 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1129 ((flag & FSYNC_WAIT)
1130 ? XFS_LOG_SYNC : 0),
1134 * If the inode is not pinned and nothing
1135 * has changed we don't need to flush the
1143 * Kick off a transaction to log the inode
1144 * core to get the updates. Make it
1145 * sync if FSYNC_WAIT is passed in (which
1146 * is done by everybody but specfs). The
1147 * sync transaction will also force the log.
1149 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1150 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1151 if ((error = xfs_trans_reserve(tp, 0,
1152 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1154 xfs_trans_cancel(tp, 0);
1157 xfs_ilock(ip, XFS_ILOCK_EXCL);
1160 * Note - it's possible that we might have pushed
1161 * ourselves out of the way during trans_reserve
1162 * which would flush the inode. But there's no
1163 * guarantee that the inode buffer has actually
1164 * gone out yet (it's delwri). Plus the buffer
1165 * could be pinned anyway if it's part of an
1166 * inode in another recent transaction. So we
1167 * play it safe and fire off the transaction anyway.
1169 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1170 xfs_trans_ihold(tp, ip);
1171 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1172 if (flag & FSYNC_WAIT)
1173 xfs_trans_set_sync(tp);
1174 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1176 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1179 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1181 * If the log write didn't issue an ordered tag we need
1182 * to flush the disk cache for the data device now.
1185 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1188 * If this inode is on the RT dev we need to flush that
1191 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1192 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1199 * This is called by xfs_inactive to free any blocks beyond eof,
1200 * when the link count isn't zero.
1203 xfs_inactive_free_eofblocks(
1209 xfs_fileoff_t end_fsb;
1210 xfs_fileoff_t last_fsb;
1211 xfs_filblks_t map_len;
1213 xfs_bmbt_irec_t imap;
1216 * Figure out if there are any blocks beyond the end
1217 * of the file. If not, then there is nothing to do.
1219 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1220 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1221 map_len = last_fsb - end_fsb;
1226 xfs_ilock(ip, XFS_ILOCK_SHARED);
1227 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1228 NULL, 0, &imap, &nimaps, NULL);
1229 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1231 if (!error && (nimaps != 0) &&
1232 (imap.br_startblock != HOLESTARTBLOCK)) {
1234 * Attach the dquots to the inode up front.
1236 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1240 * There are blocks after the end of file.
1241 * Free them up now by truncating the file to
1244 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1247 * Do the xfs_itruncate_start() call before
1248 * reserving any log space because
1249 * itruncate_start will call into the buffer
1250 * cache and we can't
1251 * do that within a transaction.
1253 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1254 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1257 error = xfs_trans_reserve(tp, 0,
1258 XFS_ITRUNCATE_LOG_RES(mp),
1259 0, XFS_TRANS_PERM_LOG_RES,
1260 XFS_ITRUNCATE_LOG_COUNT);
1262 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1263 xfs_trans_cancel(tp, 0);
1264 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1268 xfs_ilock(ip, XFS_ILOCK_EXCL);
1269 xfs_trans_ijoin(tp, ip,
1272 xfs_trans_ihold(tp, ip);
1274 error = xfs_itruncate_finish(&tp, ip,
1279 * If we get an error at this point we
1280 * simply don't bother truncating the file.
1283 xfs_trans_cancel(tp,
1284 (XFS_TRANS_RELEASE_LOG_RES |
1287 error = xfs_trans_commit(tp,
1288 XFS_TRANS_RELEASE_LOG_RES,
1291 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1297 * Free a symlink that has blocks associated with it.
1300 xfs_inactive_symlink_rmt(
1308 xfs_fsblock_t first_block;
1309 xfs_bmap_free_t free_list;
1312 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1320 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1322 * We're freeing a symlink that has some
1323 * blocks allocated to it. Free the
1324 * blocks here. We know that we've got
1325 * either 1 or 2 extents and that we can
1326 * free them all in one bunmapi call.
1328 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1329 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1330 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1331 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1332 xfs_trans_cancel(tp, 0);
1337 * Lock the inode, fix the size, and join it to the transaction.
1338 * Hold it so in the normal path, we still have it locked for
1339 * the second transaction. In the error paths we need it
1340 * held so the cancel won't rele it, see below.
1342 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1343 size = (int)ip->i_d.di_size;
1344 ip->i_d.di_size = 0;
1345 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1346 xfs_trans_ihold(tp, ip);
1347 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1349 * Find the block(s) so we can inval and unmap them.
1352 XFS_BMAP_INIT(&free_list, &first_block);
1353 nmaps = sizeof(mval) / sizeof(mval[0]);
1354 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1355 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1359 * Invalidate the block(s).
1361 for (i = 0; i < nmaps; i++) {
1362 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1363 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1364 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1365 xfs_trans_binval(tp, bp);
1368 * Unmap the dead block(s) to the free_list.
1370 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1371 &first_block, &free_list, &done)))
1375 * Commit the first transaction. This logs the EFI and the inode.
1377 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1380 * The transaction must have been committed, since there were
1381 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1382 * The new tp has the extent freeing and EFDs.
1386 * The first xact was committed, so add the inode to the new one.
1387 * Mark it dirty so it will be logged and moved forward in the log as
1388 * part of every commit.
1390 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1391 xfs_trans_ihold(tp, ip);
1392 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1394 * Get a new, empty transaction to return to our caller.
1396 ntp = xfs_trans_dup(tp);
1398 * Commit the transaction containing extent freeing and EFD's.
1399 * If we get an error on the commit here or on the reserve below,
1400 * we need to unlock the inode since the new transaction doesn't
1401 * have the inode attached.
1403 error = xfs_trans_commit(tp, 0, NULL);
1406 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1410 * Remove the memory for extent descriptions (just bookkeeping).
1412 if (ip->i_df.if_bytes)
1413 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1414 ASSERT(ip->i_df.if_bytes == 0);
1416 * Put an itruncate log reservation in the new transaction
1419 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1420 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1421 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1425 * Return with the inode locked but not joined to the transaction.
1431 xfs_bmap_cancel(&free_list);
1434 * Have to come here with the inode locked and either
1435 * (held and in the transaction) or (not in the transaction).
1436 * If the inode isn't held then cancel would iput it, but
1437 * that's wrong since this is inactive and the vnode ref
1438 * count is 0 already.
1439 * Cancel won't do anything to the inode if held, but it still
1440 * needs to be locked until the cancel is done, if it was
1441 * joined to the transaction.
1443 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1444 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1451 xfs_inactive_symlink_local(
1457 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1459 * We're freeing a symlink which fit into
1460 * the inode. Just free the memory used
1461 * to hold the old symlink.
1463 error = xfs_trans_reserve(*tpp, 0,
1464 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1465 0, XFS_TRANS_PERM_LOG_RES,
1466 XFS_ITRUNCATE_LOG_COUNT);
1469 xfs_trans_cancel(*tpp, 0);
1473 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1476 * Zero length symlinks _can_ exist.
1478 if (ip->i_df.if_bytes > 0) {
1479 xfs_idata_realloc(ip,
1480 -(ip->i_df.if_bytes),
1482 ASSERT(ip->i_df.if_bytes == 0);
1499 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1502 ASSERT(ip->i_d.di_forkoff != 0);
1503 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1504 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1506 error = xfs_attr_inactive(ip);
1509 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1510 return (error); /* goto out*/
1513 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1514 error = xfs_trans_reserve(tp, 0,
1515 XFS_IFREE_LOG_RES(mp),
1516 0, XFS_TRANS_PERM_LOG_RES,
1517 XFS_INACTIVE_LOG_COUNT);
1519 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1520 xfs_trans_cancel(tp, 0);
1522 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1526 xfs_ilock(ip, XFS_ILOCK_EXCL);
1527 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1528 xfs_trans_ihold(tp, ip);
1529 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1531 ASSERT(ip->i_d.di_anextents == 0);
1546 vp = BHV_TO_VNODE(bdp);
1547 ip = XFS_BHVTOI(bdp);
1549 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1553 /* If this is a read-only mount, don't do this (would generate I/O) */
1554 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1557 #ifdef HAVE_REFCACHE
1558 /* If we are in the NFS reference cache then don't do this now */
1565 if (ip->i_d.di_nlink != 0) {
1566 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1567 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1568 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1569 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1570 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1572 /* Update linux inode block count after free above */
1573 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1574 ip->i_d.di_nblocks + ip->i_delayed_blks);
1584 * This is called when the vnode reference count for the vnode
1585 * goes to zero. If the file has been unlinked, then it must
1586 * now be truncated. Also, we clear all of the read-ahead state
1587 * kept for the inode here since the file is now closed.
1596 xfs_bmap_free_t free_list;
1597 xfs_fsblock_t first_block;
1604 vp = BHV_TO_VNODE(bdp);
1605 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1607 ip = XFS_BHVTOI(bdp);
1610 * If the inode is already free, then there can be nothing
1613 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1614 ASSERT(ip->i_df.if_real_bytes == 0);
1615 ASSERT(ip->i_df.if_broot_bytes == 0);
1616 return VN_INACTIVE_CACHE;
1620 * Only do a truncate if it's a regular file with
1621 * some actual space in it. It's OK to look at the
1622 * inode's fields without the lock because we're the
1623 * only one with a reference to the inode.
1625 truncate = ((ip->i_d.di_nlink == 0) &&
1626 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1627 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1631 if (ip->i_d.di_nlink == 0 &&
1632 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1633 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1638 /* If this is a read-only mount, don't do this (would generate I/O) */
1639 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1642 if (ip->i_d.di_nlink != 0) {
1643 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1644 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1645 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1646 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1647 (ip->i_delayed_blks != 0))) {
1648 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1649 return (VN_INACTIVE_CACHE);
1650 /* Update linux inode block count after free above */
1651 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1652 ip->i_d.di_nblocks + ip->i_delayed_blks);
1657 ASSERT(ip->i_d.di_nlink == 0);
1659 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1660 return (VN_INACTIVE_CACHE);
1662 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1665 * Do the xfs_itruncate_start() call before
1666 * reserving any log space because itruncate_start
1667 * will call into the buffer cache and we can't
1668 * do that within a transaction.
1670 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1672 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1674 error = xfs_trans_reserve(tp, 0,
1675 XFS_ITRUNCATE_LOG_RES(mp),
1676 0, XFS_TRANS_PERM_LOG_RES,
1677 XFS_ITRUNCATE_LOG_COUNT);
1679 /* Don't call itruncate_cleanup */
1680 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1681 xfs_trans_cancel(tp, 0);
1682 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1683 return (VN_INACTIVE_CACHE);
1686 xfs_ilock(ip, XFS_ILOCK_EXCL);
1687 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1688 xfs_trans_ihold(tp, ip);
1691 * normally, we have to run xfs_itruncate_finish sync.
1692 * But if filesystem is wsync and we're in the inactive
1693 * path, then we know that nlink == 0, and that the
1694 * xaction that made nlink == 0 is permanently committed
1695 * since xfs_remove runs as a synchronous transaction.
1697 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1698 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1701 xfs_trans_cancel(tp,
1702 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1703 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1704 return (VN_INACTIVE_CACHE);
1706 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1709 * If we get an error while cleaning up a
1710 * symlink we bail out.
1712 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1713 xfs_inactive_symlink_rmt(ip, &tp) :
1714 xfs_inactive_symlink_local(ip, &tp);
1718 return (VN_INACTIVE_CACHE);
1721 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1722 xfs_trans_ihold(tp, ip);
1724 error = xfs_trans_reserve(tp, 0,
1725 XFS_IFREE_LOG_RES(mp),
1726 0, XFS_TRANS_PERM_LOG_RES,
1727 XFS_INACTIVE_LOG_COUNT);
1729 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1730 xfs_trans_cancel(tp, 0);
1731 return (VN_INACTIVE_CACHE);
1734 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1735 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1736 xfs_trans_ihold(tp, ip);
1740 * If there are attributes associated with the file
1741 * then blow them away now. The code calls a routine
1742 * that recursively deconstructs the attribute fork.
1743 * We need to just commit the current transaction
1744 * because we can't use it for xfs_attr_inactive().
1746 if (ip->i_d.di_anextents > 0) {
1747 error = xfs_inactive_attrs(ip, &tp);
1749 * If we got an error, the transaction is already
1750 * cancelled, and the inode is unlocked. Just get out.
1753 return (VN_INACTIVE_CACHE);
1754 } else if (ip->i_afp) {
1755 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1761 XFS_BMAP_INIT(&free_list, &first_block);
1762 error = xfs_ifree(tp, ip, &free_list);
1765 * If we fail to free the inode, shut down. The cancel
1766 * might do that, we need to make sure. Otherwise the
1767 * inode might be lost for a long time or forever.
1769 if (!XFS_FORCED_SHUTDOWN(mp)) {
1771 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1772 error, mp->m_fsname);
1773 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1775 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1778 * Credit the quota account(s). The inode is gone.
1780 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1783 * Just ignore errors at this point. There is
1784 * nothing we can do except to try to keep going.
1786 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1788 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1791 * Release the dquots held by inode, if any.
1793 XFS_QM_DQDETACH(mp, ip);
1795 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1798 return VN_INACTIVE_CACHE;
1807 bhv_desc_t *dir_bdp,
1814 xfs_inode_t *dp, *ip;
1820 dir_vp = BHV_TO_VNODE(dir_bdp);
1821 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1823 dp = XFS_BHVTOI(dir_bdp);
1825 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1826 return XFS_ERROR(EIO);
1828 lock_mode = xfs_ilock_map_shared(dp);
1829 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1831 *vpp = XFS_ITOV(ip);
1834 xfs_iunlock_map_shared(dp, lock_mode);
1840 * xfs_create (create a new file).
1844 bhv_desc_t *dir_bdp,
1850 char *name = VNAME(dentry);
1852 xfs_inode_t *dp, *ip;
1858 xfs_bmap_free_t free_list;
1859 xfs_fsblock_t first_block;
1860 boolean_t dp_joined_to_trans;
1861 int dm_event_sent = 0;
1865 struct xfs_dquot *udqp, *gdqp;
1871 dir_vp = BHV_TO_VNODE(dir_bdp);
1872 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1874 dp = XFS_BHVTOI(dir_bdp);
1877 dm_di_mode = vap->va_mode;
1878 namelen = VNAMELEN(dentry);
1880 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1881 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1882 dir_vp, DM_RIGHT_NULL, NULL,
1883 DM_RIGHT_NULL, name, NULL,
1891 if (XFS_FORCED_SHUTDOWN(mp))
1892 return XFS_ERROR(EIO);
1894 /* Return through std_return after this point. */
1897 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1898 prid = dp->i_d.di_projid;
1899 else if (vap->va_mask & XFS_AT_PROJID)
1900 prid = (xfs_prid_t)vap->va_projid;
1902 prid = (xfs_prid_t)dfltprid;
1905 * Make sure that we have allocated dquot(s) on disk.
1907 error = XFS_QM_DQVOPALLOC(mp, dp,
1908 current_fsuid(credp), current_fsgid(credp), prid,
1909 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1914 dp_joined_to_trans = B_FALSE;
1916 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1917 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1918 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1920 * Initially assume that the file does not exist and
1921 * reserve the resources for that case. If that is not
1922 * the case we'll drop the one we have and get a more
1923 * appropriate transaction later.
1925 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1926 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1927 if (error == ENOSPC) {
1929 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1930 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1938 xfs_ilock(dp, XFS_ILOCK_EXCL);
1940 XFS_BMAP_INIT(&free_list, &first_block);
1945 * Reserve disk quota and the inode.
1947 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1952 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1954 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1955 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1956 rdev, credp, prid, resblks > 0,
1959 if (error == ENOSPC)
1966 * At this point, we've gotten a newly allocated inode.
1967 * It is locked (and joined to the transaction).
1970 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1973 * Now we join the directory inode to the transaction.
1974 * We do not do it earlier because xfs_dir_ialloc
1975 * might commit the previous transaction (and release
1980 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1981 dp_joined_to_trans = B_TRUE;
1983 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1984 &first_block, &free_list,
1985 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1987 ASSERT(error != ENOSPC);
1990 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1991 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1994 * If this is a synchronous mount, make sure that the
1995 * create transaction goes to disk before returning to
1998 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1999 xfs_trans_set_sync(tp);
2005 * Attach the dquot(s) to the inodes and modify them incore.
2006 * These ids of the inode couldn't have changed since the new
2007 * inode has been locked ever since it was created.
2009 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2012 * xfs_trans_commit normally decrements the vnode ref count
2013 * when it unlocks the inode. Since we want to return the
2014 * vnode to the caller, we bump the vnode ref count now.
2019 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2021 xfs_bmap_cancel(&free_list);
2025 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2032 XFS_QM_DQRELE(mp, udqp);
2033 XFS_QM_DQRELE(mp, gdqp);
2036 * Propogate the fact that the vnode changed after the
2037 * xfs_inode locks have been released.
2039 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2043 /* Fallthrough to std_return with error = 0 */
2046 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2047 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2048 DM_EVENT_POSTCREATE)) {
2049 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2050 dir_vp, DM_RIGHT_NULL,
2052 DM_RIGHT_NULL, name, NULL,
2053 dm_di_mode, error, 0);
2058 cancel_flags |= XFS_TRANS_ABORT;
2063 xfs_trans_cancel(tp, cancel_flags);
2065 if (!dp_joined_to_trans && (dp != NULL))
2066 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2067 XFS_QM_DQRELE(mp, udqp);
2068 XFS_QM_DQRELE(mp, gdqp);
2074 * Wait until after the current transaction is aborted to
2075 * release the inode. This prevents recursive transactions
2076 * and deadlocks from xfs_inactive.
2078 cancel_flags |= XFS_TRANS_ABORT;
2079 xfs_trans_cancel(tp, cancel_flags);
2082 XFS_QM_DQRELE(mp, udqp);
2083 XFS_QM_DQRELE(mp, gdqp);
2090 * Some counters to see if (and how often) we are hitting some deadlock
2091 * prevention code paths.
2095 int xfs_rm_lock_delays;
2096 int xfs_rm_attempts;
2100 * The following routine will lock the inodes associated with the
2101 * directory and the named entry in the directory. The locks are
2102 * acquired in increasing inode number.
2104 * If the entry is "..", then only the directory is locked. The
2105 * vnode ref count will still include that from the .. entry in
2108 * There is a deadlock we need to worry about. If the locked directory is
2109 * in the AIL, it might be blocking up the log. The next inode we lock
2110 * could be already locked by another thread waiting for log space (e.g
2111 * a permanent log reservation with a long running transaction (see
2112 * xfs_itruncate_finish)). To solve this, we must check if the directory
2113 * is in the ail and use lock_nowait. If we can't lock, we need to
2114 * drop the inode lock on the directory and try again. xfs_iunlock will
2115 * potentially push the tail if we were holding up the log.
2118 xfs_lock_dir_and_entry(
2121 xfs_inode_t *ip) /* inode of entry 'name' */
2125 xfs_inode_t *ips[2];
2134 xfs_ilock(dp, XFS_ILOCK_EXCL);
2141 * We want to lock in increasing inum. Since we've already
2142 * acquired the lock on the directory, we may need to release
2143 * if if the inum of the entry turns out to be less.
2145 if (e_inum > dp->i_ino) {
2147 * We are already in the right order, so just
2148 * lock on the inode of the entry.
2149 * We need to use nowait if dp is in the AIL.
2152 lp = (xfs_log_item_t *)dp->i_itemp;
2153 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2154 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2161 * Unlock dp and try again.
2162 * xfs_iunlock will try to push the tail
2163 * if the inode is in the AIL.
2166 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2168 if ((attempts % 5) == 0) {
2169 delay(1); /* Don't just spin the CPU */
2171 xfs_rm_lock_delays++;
2177 xfs_ilock(ip, XFS_ILOCK_EXCL);
2179 } else if (e_inum < dp->i_ino) {
2180 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2184 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2186 /* else e_inum == dp->i_ino */
2187 /* This can happen if we're asked to lock /x/..
2188 * the entry is "..", which is also the parent directory.
2196 int xfs_small_retries;
2197 int xfs_middle_retries;
2198 int xfs_lots_retries;
2199 int xfs_lock_delays;
2203 * The following routine will lock n inodes in exclusive mode.
2204 * We assume the caller calls us with the inodes in i_ino order.
2206 * We need to detect deadlock where an inode that we lock
2207 * is in the AIL and we start waiting for another inode that is locked
2208 * by a thread in a long running transaction (such as truncate). This can
2209 * result in deadlock since the long running trans might need to wait
2210 * for the inode we just locked in order to push the tail and free space
2220 int attempts = 0, i, j, try_lock;
2223 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2234 for (; i < inodes; i++) {
2237 if (i && (ips[i] == ips[i-1])) /* Already locked */
2241 * If try_lock is not set yet, make sure all locked inodes
2242 * are not in the AIL.
2243 * If any are, set try_lock to be used later.
2247 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2248 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2249 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2256 * If any of the previous locks we have locked is in the AIL,
2257 * we must TRY to get the second and subsequent locks. If
2258 * we can't get any, we must release all we have
2263 /* try_lock must be 0 if i is 0. */
2265 * try_lock means we have an inode locked
2266 * that is in the AIL.
2269 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2273 * Unlock all previous guys and try again.
2274 * xfs_iunlock will try to push the tail
2275 * if the inode is in the AIL.
2278 for(j = i - 1; j >= 0; j--) {
2281 * Check to see if we've already
2282 * unlocked this one.
2283 * Not the first one going back,
2284 * and the inode ptr is the same.
2286 if ((j != (i - 1)) && ips[j] ==
2290 xfs_iunlock(ips[j], lock_mode);
2293 if ((attempts % 5) == 0) {
2294 delay(1); /* Don't just spin the CPU */
2304 xfs_ilock(ips[i], lock_mode);
2310 if (attempts < 5) xfs_small_retries++;
2311 else if (attempts < 100) xfs_middle_retries++;
2312 else xfs_lots_retries++;
2320 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2321 int remove_which_error_return = 0;
2323 #define REMOVE_DEBUG_TRACE(x)
2324 #endif /* ! DEBUG */
2333 bhv_desc_t *dir_bdp,
2338 char *name = VNAME(dentry);
2339 xfs_inode_t *dp, *ip;
2340 xfs_trans_t *tp = NULL;
2343 xfs_bmap_free_t free_list;
2344 xfs_fsblock_t first_block;
2352 dir_vp = BHV_TO_VNODE(dir_bdp);
2353 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2355 dp = XFS_BHVTOI(dir_bdp);
2358 if (XFS_FORCED_SHUTDOWN(mp))
2359 return XFS_ERROR(EIO);
2361 namelen = VNAMELEN(dentry);
2363 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2364 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2365 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2366 name, NULL, 0, 0, 0);
2371 /* From this point on, return through std_return */
2375 * We need to get a reference to ip before we get our log
2376 * reservation. The reason for this is that we cannot call
2377 * xfs_iget for an inode for which we do not have a reference
2378 * once we've acquired a log reservation. This is because the
2379 * inode we are trying to get might be in xfs_inactive going
2380 * for a log reservation. Since we'll have to wait for the
2381 * inactive code to complete before returning from xfs_iget,
2382 * we need to make sure that we don't have log space reserved
2383 * when we call xfs_iget. Instead we get an unlocked referece
2384 * to the inode before getting our log reservation.
2386 error = xfs_get_dir_entry(dentry, &ip);
2388 REMOVE_DEBUG_TRACE(__LINE__);
2392 dm_di_mode = ip->i_d.di_mode;
2394 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2398 error = XFS_QM_DQATTACH(mp, dp, 0);
2399 if (!error && dp != ip)
2400 error = XFS_QM_DQATTACH(mp, ip, 0);
2402 REMOVE_DEBUG_TRACE(__LINE__);
2407 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2408 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2410 * We try to get the real space reservation first,
2411 * allowing for directory btree deletion(s) implying
2412 * possible bmap insert(s). If we can't get the space
2413 * reservation then we use 0 instead, and avoid the bmap
2414 * btree insert(s) in the directory code by, if the bmap
2415 * insert tries to happen, instead trimming the LAST
2416 * block from the directory.
2418 resblks = XFS_REMOVE_SPACE_RES(mp);
2419 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2420 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2421 if (error == ENOSPC) {
2423 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2424 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2427 ASSERT(error != ENOSPC);
2428 REMOVE_DEBUG_TRACE(__LINE__);
2429 xfs_trans_cancel(tp, 0);
2434 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2436 REMOVE_DEBUG_TRACE(__LINE__);
2437 xfs_trans_cancel(tp, cancel_flags);
2443 * At this point, we've gotten both the directory and the entry
2446 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2449 * Increment vnode ref count only in this case since
2450 * there's an extra vnode reference in the case where
2454 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2458 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2460 XFS_BMAP_INIT(&free_list, &first_block);
2461 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2462 &first_block, &free_list, 0);
2464 ASSERT(error != ENOENT);
2465 REMOVE_DEBUG_TRACE(__LINE__);
2468 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2471 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2473 error = xfs_droplink(tp, ip);
2475 REMOVE_DEBUG_TRACE(__LINE__);
2479 /* Determine if this is the last link while
2480 * we are in the transaction.
2482 link_zero = (ip)->i_d.di_nlink==0;
2485 * Take an extra ref on the inode so that it doesn't
2486 * go to xfs_inactive() from within the commit.
2491 * If this is a synchronous mount, make sure that the
2492 * remove transaction goes to disk before returning to
2495 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2496 xfs_trans_set_sync(tp);
2499 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2501 REMOVE_DEBUG_TRACE(__LINE__);
2505 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2512 * Before we drop our extra reference to the inode, purge it
2513 * from the refcache if it is there. By waiting until afterwards
2514 * to do the IRELE, we ensure that we won't go inactive in the
2515 * xfs_refcache_purge_ip routine (although that would be OK).
2517 xfs_refcache_purge_ip(ip);
2519 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2522 * Let interposed file systems know about removed links.
2524 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2528 /* Fall through to std_return with error = 0 */
2530 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2531 DM_EVENT_POSTREMOVE)) {
2532 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2533 dir_vp, DM_RIGHT_NULL,
2534 NULL, DM_RIGHT_NULL,
2535 name, NULL, dm_di_mode, error, 0);
2540 xfs_bmap_cancel(&free_list);
2541 cancel_flags |= XFS_TRANS_ABORT;
2542 xfs_trans_cancel(tp, cancel_flags);
2547 * In this case make sure to not release the inode until after
2548 * the current transaction is aborted. Releasing it beforehand
2549 * can cause us to go to xfs_inactive and start a recursive
2550 * transaction which can easily deadlock with the current one.
2552 xfs_bmap_cancel(&free_list);
2553 cancel_flags |= XFS_TRANS_ABORT;
2554 xfs_trans_cancel(tp, cancel_flags);
2557 * Before we drop our extra reference to the inode, purge it
2558 * from the refcache if it is there. By waiting until afterwards
2559 * to do the IRELE, we ensure that we won't go inactive in the
2560 * xfs_refcache_purge_ip routine (although that would be OK).
2562 xfs_refcache_purge_ip(ip);
2576 bhv_desc_t *target_dir_bdp,
2581 xfs_inode_t *tdp, *sip;
2584 xfs_inode_t *ips[2];
2586 xfs_bmap_free_t free_list;
2587 xfs_fsblock_t first_block;
2590 vnode_t *target_dir_vp;
2591 bhv_desc_t *src_bdp;
2593 char *target_name = VNAME(dentry);
2596 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2597 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2598 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2600 target_namelen = VNAMELEN(dentry);
2601 if (VN_ISDIR(src_vp))
2602 return XFS_ERROR(EPERM);
2604 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2605 sip = XFS_BHVTOI(src_bdp);
2606 tdp = XFS_BHVTOI(target_dir_bdp);
2608 if (XFS_FORCED_SHUTDOWN(mp))
2609 return XFS_ERROR(EIO);
2611 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2612 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2613 target_dir_vp, DM_RIGHT_NULL,
2614 src_vp, DM_RIGHT_NULL,
2615 target_name, NULL, 0, 0, 0);
2620 /* Return through std_return after this point. */
2622 error = XFS_QM_DQATTACH(mp, sip, 0);
2623 if (!error && sip != tdp)
2624 error = XFS_QM_DQATTACH(mp, tdp, 0);
2628 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2629 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2630 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2631 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2632 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2633 if (error == ENOSPC) {
2635 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2636 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2643 if (sip->i_ino < tdp->i_ino) {
2651 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2654 * Increment vnode ref counts since xfs_trans_commit &
2655 * xfs_trans_cancel will both unlock the inodes and
2656 * decrement the associated ref counts.
2659 VN_HOLD(target_dir_vp);
2660 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2661 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2664 * If the source has too many links, we can't make any more to it.
2666 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2667 error = XFS_ERROR(EMLINK);
2672 * If we are using project inheritance, we only allow hard link
2673 * creation in our tree when the project IDs are the same; else
2674 * the tree quota mechanism could be circumvented.
2676 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2677 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2678 error = XFS_ERROR(EPERM);
2683 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2687 XFS_BMAP_INIT(&free_list, &first_block);
2689 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2690 sip->i_ino, &first_block, &free_list,
2694 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2696 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2698 error = xfs_bumplink(tp, sip);
2704 * If this is a synchronous mount, make sure that the
2705 * link transaction goes to disk before returning to
2708 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2709 xfs_trans_set_sync(tp);
2712 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2714 xfs_bmap_cancel(&free_list);
2718 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2723 /* Fall through to std_return with error = 0. */
2725 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2726 DM_EVENT_POSTLINK)) {
2727 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2728 target_dir_vp, DM_RIGHT_NULL,
2729 src_vp, DM_RIGHT_NULL,
2730 target_name, NULL, 0, error, 0);
2735 cancel_flags |= XFS_TRANS_ABORT;
2738 xfs_trans_cancel(tp, cancel_flags);
2748 bhv_desc_t *dir_bdp,
2754 char *dir_name = VNAME(dentry);
2756 xfs_inode_t *cdp; /* inode of created dir */
2757 vnode_t *cvp; /* vnode of created dir */
2763 xfs_bmap_free_t free_list;
2764 xfs_fsblock_t first_block;
2766 boolean_t dp_joined_to_trans;
2767 boolean_t created = B_FALSE;
2768 int dm_event_sent = 0;
2770 struct xfs_dquot *udqp, *gdqp;
2775 dir_vp = BHV_TO_VNODE(dir_bdp);
2776 dp = XFS_BHVTOI(dir_bdp);
2779 if (XFS_FORCED_SHUTDOWN(mp))
2780 return XFS_ERROR(EIO);
2782 dir_namelen = VNAMELEN(dentry);
2785 dp_joined_to_trans = B_FALSE;
2786 dm_di_mode = vap->va_mode;
2788 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2789 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2790 dir_vp, DM_RIGHT_NULL, NULL,
2791 DM_RIGHT_NULL, dir_name, NULL,
2798 /* Return through std_return after this point. */
2800 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2804 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2805 prid = dp->i_d.di_projid;
2806 else if (vap->va_mask & XFS_AT_PROJID)
2807 prid = (xfs_prid_t)vap->va_projid;
2809 prid = (xfs_prid_t)dfltprid;
2812 * Make sure that we have allocated dquot(s) on disk.
2814 error = XFS_QM_DQVOPALLOC(mp, dp,
2815 current_fsuid(credp), current_fsgid(credp), prid,
2816 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2820 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2821 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2822 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2823 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2824 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2825 if (error == ENOSPC) {
2827 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2828 XFS_TRANS_PERM_LOG_RES,
2829 XFS_MKDIR_LOG_COUNT);
2837 xfs_ilock(dp, XFS_ILOCK_EXCL);
2840 * Check for directory link count overflow.
2842 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2843 error = XFS_ERROR(EMLINK);
2848 * Reserve disk quota and the inode.
2850 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2855 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2858 * create the directory inode.
2860 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2861 0, credp, prid, resblks > 0,
2864 if (error == ENOSPC)
2871 * Now we add the directory inode to the transaction.
2872 * We waited until now since xfs_dir_ialloc might start
2873 * a new transaction. Had we joined the transaction
2874 * earlier, the locks might have gotten released.
2877 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2878 dp_joined_to_trans = B_TRUE;
2880 XFS_BMAP_INIT(&free_list, &first_block);
2882 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2883 cdp->i_ino, &first_block, &free_list,
2884 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2886 ASSERT(error != ENOSPC);
2889 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2892 * Bump the in memory version number of the parent directory
2893 * so that other processes accessing it will recognize that
2894 * the directory has changed.
2898 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2904 error = xfs_bumplink(tp, dp);
2909 cvp = XFS_ITOV(cdp);
2917 * Attach the dquots to the new inode and modify the icount incore.
2919 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2922 * If this is a synchronous mount, make sure that the
2923 * mkdir transaction goes to disk before returning to
2926 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2927 xfs_trans_set_sync(tp);
2930 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2936 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2937 XFS_QM_DQRELE(mp, udqp);
2938 XFS_QM_DQRELE(mp, gdqp);
2943 /* Fall through to std_return with error = 0 or errno from
2944 * xfs_trans_commit. */
2947 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2948 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2949 DM_EVENT_POSTCREATE)) {
2950 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2951 dir_vp, DM_RIGHT_NULL,
2952 created ? XFS_ITOV(cdp):NULL,
2955 dm_di_mode, error, 0);
2961 xfs_bmap_cancel(&free_list);
2963 cancel_flags |= XFS_TRANS_ABORT;
2965 xfs_trans_cancel(tp, cancel_flags);
2966 XFS_QM_DQRELE(mp, udqp);
2967 XFS_QM_DQRELE(mp, gdqp);
2969 if (!dp_joined_to_trans && (dp != NULL)) {
2970 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2983 bhv_desc_t *dir_bdp,
2987 char *name = VNAME(dentry);
2989 xfs_inode_t *cdp; /* child directory */
2993 xfs_bmap_free_t free_list;
2994 xfs_fsblock_t first_block;
3003 dir_vp = BHV_TO_VNODE(dir_bdp);
3004 dp = XFS_BHVTOI(dir_bdp);
3007 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3009 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3010 return XFS_ERROR(EIO);
3011 namelen = VNAMELEN(dentry);
3013 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3014 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3015 dir_vp, DM_RIGHT_NULL,
3016 NULL, DM_RIGHT_NULL,
3017 name, NULL, 0, 0, 0);
3019 return XFS_ERROR(error);
3022 /* Return through std_return after this point. */
3027 * We need to get a reference to cdp before we get our log
3028 * reservation. The reason for this is that we cannot call
3029 * xfs_iget for an inode for which we do not have a reference
3030 * once we've acquired a log reservation. This is because the
3031 * inode we are trying to get might be in xfs_inactive going
3032 * for a log reservation. Since we'll have to wait for the
3033 * inactive code to complete before returning from xfs_iget,
3034 * we need to make sure that we don't have log space reserved
3035 * when we call xfs_iget. Instead we get an unlocked referece
3036 * to the inode before getting our log reservation.
3038 error = xfs_get_dir_entry(dentry, &cdp);
3040 REMOVE_DEBUG_TRACE(__LINE__);
3044 dm_di_mode = cdp->i_d.di_mode;
3047 * Get the dquots for the inodes.
3049 error = XFS_QM_DQATTACH(mp, dp, 0);
3050 if (!error && dp != cdp)
3051 error = XFS_QM_DQATTACH(mp, cdp, 0);
3054 REMOVE_DEBUG_TRACE(__LINE__);
3058 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3059 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3061 * We try to get the real space reservation first,
3062 * allowing for directory btree deletion(s) implying
3063 * possible bmap insert(s). If we can't get the space
3064 * reservation then we use 0 instead, and avoid the bmap
3065 * btree insert(s) in the directory code by, if the bmap
3066 * insert tries to happen, instead trimming the LAST
3067 * block from the directory.
3069 resblks = XFS_REMOVE_SPACE_RES(mp);
3070 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3071 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3072 if (error == ENOSPC) {
3074 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3075 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3078 ASSERT(error != ENOSPC);
3083 XFS_BMAP_INIT(&free_list, &first_block);
3086 * Now lock the child directory inode and the parent directory
3087 * inode in the proper order. This will take care of validating
3088 * that the directory entry for the child directory inode has
3089 * not changed while we were obtaining a log reservation.
3091 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3093 xfs_trans_cancel(tp, cancel_flags);
3098 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3101 * Only increment the parent directory vnode count if
3102 * we didn't bump it in looking up cdp. The only time
3103 * we don't bump it is when we're looking up ".".
3109 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3111 ASSERT(cdp->i_d.di_nlink >= 2);
3112 if (cdp->i_d.di_nlink != 2) {
3113 error = XFS_ERROR(ENOTEMPTY);
3116 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3117 error = XFS_ERROR(ENOTEMPTY);
3121 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3122 &first_block, &free_list, resblks);
3127 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3130 * Bump the in memory generation count on the parent
3131 * directory so that other can know that it has changed.
3136 * Drop the link from cdp's "..".
3138 error = xfs_droplink(tp, dp);
3144 * Drop the link from dp to cdp.
3146 error = xfs_droplink(tp, cdp);
3152 * Drop the "." link from cdp to self.
3154 error = xfs_droplink(tp, cdp);
3159 /* Determine these before committing transaction */
3160 last_cdp_link = (cdp)->i_d.di_nlink==0;
3163 * Take an extra ref on the child vnode so that it
3164 * does not go to xfs_inactive() from within the commit.
3169 * If this is a synchronous mount, make sure that the
3170 * rmdir transaction goes to disk before returning to
3173 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3174 xfs_trans_set_sync(tp);
3177 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3179 xfs_bmap_cancel(&free_list);
3180 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3186 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3194 * Let interposed file systems know about removed links.
3196 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3200 /* Fall through to std_return with error = 0 or the errno
3201 * from xfs_trans_commit. */
3203 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3204 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3205 dir_vp, DM_RIGHT_NULL,
3206 NULL, DM_RIGHT_NULL,
3207 name, NULL, dm_di_mode,
3213 xfs_bmap_cancel(&free_list);
3214 cancel_flags |= XFS_TRANS_ABORT;
3216 xfs_trans_cancel(tp, cancel_flags);
3224 * Read dp's entries starting at uiop->uio_offset and translate them into
3225 * bufsize bytes worth of struct dirents starting at bufbase.
3229 bhv_desc_t *dir_bdp,
3235 xfs_trans_t *tp = NULL;
3238 xfs_off_t start_offset;
3240 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3241 (inst_t *)__return_address);
3242 dp = XFS_BHVTOI(dir_bdp);
3244 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3245 return XFS_ERROR(EIO);
3248 lock_mode = xfs_ilock_map_shared(dp);
3249 start_offset = uiop->uio_offset;
3250 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3251 if (start_offset != uiop->uio_offset) {
3252 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3254 xfs_iunlock_map_shared(dp, lock_mode);
3265 bhv_desc_t *dir_bdp,
3278 xfs_bmap_free_t free_list;
3279 xfs_fsblock_t first_block;
3280 boolean_t dp_joined_to_trans;
3284 xfs_fileoff_t first_fsb;
3285 xfs_filblks_t fs_blocks;
3287 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3294 struct xfs_dquot *udqp, *gdqp;
3296 char *link_name = VNAME(dentry);
3300 dir_vp = BHV_TO_VNODE(dir_bdp);
3301 dp = XFS_BHVTOI(dir_bdp);
3302 dp_joined_to_trans = B_FALSE;
3307 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3311 if (XFS_FORCED_SHUTDOWN(mp))
3312 return XFS_ERROR(EIO);
3314 link_namelen = VNAMELEN(dentry);
3317 * Check component lengths of the target path name.
3319 pathlen = strlen(target_path);
3320 if (pathlen >= MAXPATHLEN) /* total string too long */
3321 return XFS_ERROR(ENAMETOOLONG);
3322 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3326 for(total = 0, path = target_path; total < pathlen;) {
3330 while(*path == '/') {
3336 * Count up to the next slash or end of path.
3337 * Error out if the component is bigger than MAXNAMELEN.
3339 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3340 if (++len >= MAXNAMELEN) {
3341 error = ENAMETOOLONG;
3348 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3349 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3350 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3351 link_name, target_path, 0, 0, 0);
3356 /* Return through std_return after this point. */
3359 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3360 prid = dp->i_d.di_projid;
3361 else if (vap->va_mask & XFS_AT_PROJID)
3362 prid = (xfs_prid_t)vap->va_projid;
3364 prid = (xfs_prid_t)dfltprid;
3367 * Make sure that we have allocated dquot(s) on disk.
3369 error = XFS_QM_DQVOPALLOC(mp, dp,
3370 current_fsuid(credp), current_fsgid(credp), prid,
3371 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3375 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3376 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3378 * The symlink will fit into the inode data fork?
3379 * There can't be any attributes so we get the whole variable part.
3381 if (pathlen <= XFS_LITINO(mp))
3384 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3385 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3386 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3387 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3388 if (error == ENOSPC && fs_blocks == 0) {
3390 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3391 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3399 xfs_ilock(dp, XFS_ILOCK_EXCL);
3402 * Check whether the directory allows new symlinks or not.
3404 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3405 error = XFS_ERROR(EPERM);
3410 * Reserve disk quota : blocks and inode.
3412 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3417 * Check for ability to enter directory entry, if no space reserved.
3420 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3423 * Initialize the bmap freelist prior to calling either
3424 * bmapi or the directory create code.
3426 XFS_BMAP_INIT(&free_list, &first_block);
3429 * Allocate an inode for the symlink.
3431 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3432 1, 0, credp, prid, resblks > 0, &ip, NULL);
3434 if (error == ENOSPC)
3441 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3442 dp_joined_to_trans = B_TRUE;
3445 * Also attach the dquot(s) to it, if applicable.
3447 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3450 resblks -= XFS_IALLOC_SPACE_RES(mp);
3452 * If the symlink will fit into the inode, write it inline.
3454 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3455 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3456 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3457 ip->i_d.di_size = pathlen;
3460 * The inode was initially created in extent format.
3462 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3463 ip->i_df.if_flags |= XFS_IFINLINE;
3465 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3466 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3470 nmaps = SYMLINK_MAPS;
3472 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3473 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3474 &first_block, resblks, mval, &nmaps,
3481 resblks -= fs_blocks;
3482 ip->i_d.di_size = pathlen;
3483 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3485 cur_chunk = target_path;
3486 for (n = 0; n < nmaps; n++) {
3487 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3488 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3489 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3490 BTOBB(byte_cnt), 0);
3491 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3492 if (pathlen < byte_cnt) {
3495 pathlen -= byte_cnt;
3497 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3498 cur_chunk += byte_cnt;
3500 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3505 * Create the directory entry for the symlink.
3507 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3508 ip->i_ino, &first_block, &free_list, resblks);
3512 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3513 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3516 * Bump the in memory version number of the parent directory
3517 * so that other processes accessing it will recognize that
3518 * the directory has changed.
3523 * If this is a synchronous mount, make sure that the
3524 * symlink transaction goes to disk before returning to
3527 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3528 xfs_trans_set_sync(tp);
3532 * xfs_trans_commit normally decrements the vnode ref count
3533 * when it unlocks the inode. Since we want to return the
3534 * vnode to the caller, we bump the vnode ref count now.
3538 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3542 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3543 XFS_QM_DQRELE(mp, udqp);
3544 XFS_QM_DQRELE(mp, gdqp);
3546 /* Fall through to std_return with error = 0 or errno from
3547 * xfs_trans_commit */
3549 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3550 DM_EVENT_POSTSYMLINK)) {
3551 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3552 dir_vp, DM_RIGHT_NULL,
3553 error ? NULL : XFS_ITOV(ip),
3554 DM_RIGHT_NULL, link_name, target_path,
3570 xfs_bmap_cancel(&free_list);
3571 cancel_flags |= XFS_TRANS_ABORT;
3573 xfs_trans_cancel(tp, cancel_flags);
3574 XFS_QM_DQRELE(mp, udqp);
3575 XFS_QM_DQRELE(mp, gdqp);
3577 if (!dp_joined_to_trans && (dp != NULL)) {
3578 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3588 * A fid routine that takes a pointer to a previously allocated
3589 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3599 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3600 (inst_t *)__return_address);
3601 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3603 xfid = (xfs_fid2_t *)fidp;
3604 ip = XFS_BHVTOI(bdp);
3605 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3608 * use memcpy because the inode is a long long and there's no
3609 * assurance that xfid->fid_ino is properly aligned.
3611 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3612 xfid->fid_gen = ip->i_d.di_gen;
3629 vp = BHV_TO_VNODE(bdp);
3632 ip = XFS_BHVTOI(bdp);
3633 if (locktype == VRWLOCK_WRITE) {
3634 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3635 } else if (locktype == VRWLOCK_TRY_READ) {
3636 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3637 } else if (locktype == VRWLOCK_TRY_WRITE) {
3638 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3640 ASSERT((locktype == VRWLOCK_READ) ||
3641 (locktype == VRWLOCK_WRITE_DIRECT));
3642 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3660 vp = BHV_TO_VNODE(bdp);
3663 ip = XFS_BHVTOI(bdp);
3664 if (locktype == VRWLOCK_WRITE) {
3666 * In the write case, we may have added a new entry to
3667 * the reference cache. This might store a pointer to
3668 * an inode to be released in this inode. If it is there,
3669 * clear the pointer and release the inode after unlocking
3672 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3674 ASSERT((locktype == VRWLOCK_READ) ||
3675 (locktype == VRWLOCK_WRITE_DIRECT));
3676 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3688 xfs_inode_log_item_t *iip;
3691 ip = XFS_BHVTOI(bdp);
3695 if (XFS_FORCED_SHUTDOWN(mp))
3696 return XFS_ERROR(EIO);
3699 * Bypass inodes which have already been cleaned by
3700 * the inode flush clustering code inside xfs_iflush
3702 if ((ip->i_update_core == 0) &&
3703 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3706 if (flags & FLUSH_LOG) {
3707 if (iip && iip->ili_last_lsn) {
3708 xlog_t *log = mp->m_log;
3710 int s, log_flags = XFS_LOG_FORCE;
3712 s = GRANT_LOCK(log);
3713 sync_lsn = log->l_last_sync_lsn;
3714 GRANT_UNLOCK(log, s);
3716 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3719 if (flags & FLUSH_SYNC)
3720 log_flags |= XFS_LOG_SYNC;
3721 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3726 * We make this non-blocking if the inode is contended,
3727 * return EAGAIN to indicate to the caller that they
3728 * did not succeed. This prevents the flush path from
3729 * blocking on inodes inside another operation right
3730 * now, they get caught later by xfs_sync.
3732 if (flags & FLUSH_INODE) {
3735 if (xfs_ipincount(ip))
3738 if (flags & FLUSH_SYNC) {
3739 xfs_ilock(ip, XFS_ILOCK_SHARED);
3741 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3742 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3743 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3750 if (flags & FLUSH_SYNC)
3751 flush_flags = XFS_IFLUSH_SYNC;
3753 flush_flags = XFS_IFLUSH_ASYNC;
3755 error = xfs_iflush(ip, flush_flags);
3756 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3775 if (!capable(CAP_SYS_ADMIN))
3776 return XFS_ERROR(EPERM);
3778 ip = XFS_BHVTOI(bdp);
3781 if (XFS_FORCED_SHUTDOWN(mp))
3782 return XFS_ERROR(EIO);
3784 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3785 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3787 xfs_trans_cancel(tp, 0);
3790 xfs_ilock(ip, XFS_ILOCK_EXCL);
3791 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3793 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3794 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3796 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3798 error = xfs_trans_commit(tp, 0, NULL);
3814 vp = BHV_TO_VNODE(bdp);
3815 ip = XFS_BHVTOI(bdp);
3817 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3819 ASSERT(!VN_MAPPED(vp));
3821 /* bad inode, get out here ASAP */
3829 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3830 ASSERT(VN_CACHED(vp) == 0);
3832 /* If we have nothing to flush with this inode then complete the
3833 * teardown now, otherwise break the link between the xfs inode
3834 * and the linux inode and clean up the xfs inode later. This
3835 * avoids flushing the inode to disk during the delete operation
3838 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3839 xfs_ilock(ip, XFS_ILOCK_EXCL);
3841 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3843 xfs_mount_t *mp = ip->i_mount;
3845 /* Protect sync from us */
3846 XFS_MOUNT_ILOCK(mp);
3847 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3848 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3849 ip->i_flags |= XFS_IRECLAIMABLE;
3850 XFS_MOUNT_IUNLOCK(mp);
3861 xfs_ihash_t *ih = ip->i_hash;
3862 vnode_t *vp = XFS_ITOV_NULL(ip);
3865 if (vp && VN_BAD(vp))
3868 /* The hash lock here protects a thread in xfs_iget_core from
3869 * racing with us on linking the inode back with a vnode.
3870 * Once we have the XFS_IRECLAIM flag set it will not touch
3873 write_lock(&ih->ih_lock);
3874 if ((ip->i_flags & XFS_IRECLAIM) ||
3875 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3876 write_unlock(&ih->ih_lock);
3879 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3883 ip->i_flags |= XFS_IRECLAIM;
3884 write_unlock(&ih->ih_lock);
3887 * If the inode is still dirty, then flush it out. If the inode
3888 * is not in the AIL, then it will be OK to flush it delwri as
3889 * long as xfs_iflush() does not keep any references to the inode.
3890 * We leave that decision up to xfs_iflush() since it has the
3891 * knowledge of whether it's OK to simply do a delwri flush of
3892 * the inode or whether we need to wait until the inode is
3893 * pulled from the AIL.
3894 * We get the flush lock regardless, though, just to make sure
3895 * we don't free it while it is being flushed.
3897 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3899 xfs_ilock(ip, XFS_ILOCK_EXCL);
3903 if (ip->i_update_core ||
3904 ((ip->i_itemp != NULL) &&
3905 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3906 error = xfs_iflush(ip, sync_mode);
3908 * If we hit an error, typically because of filesystem
3909 * shutdown, we don't need to let vn_reclaim to know
3910 * because we're gonna reclaim the inode anyway.
3913 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3916 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3919 ASSERT(ip->i_update_core == 0);
3920 ASSERT(ip->i_itemp == NULL ||
3921 ip->i_itemp->ili_format.ilf_fields == 0);
3922 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3923 } else if (locked) {
3925 * We are not interested in doing an iflush if we're
3926 * in the process of shutting down the filesystem forcibly.
3927 * So, just reclaim the inode.
3930 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3939 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3942 xfs_inode_t *ip, *n;
3947 XFS_MOUNT_ILOCK(mp);
3948 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3950 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3952 if (xfs_ipincount(ip) ||
3953 !xfs_iflock_nowait(ip)) {
3954 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3958 XFS_MOUNT_IUNLOCK(mp);
3959 xfs_finish_reclaim(ip, noblock,
3960 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
3968 XFS_MOUNT_IUNLOCK(mp);
3973 * xfs_alloc_file_space()
3974 * This routine allocates disk space for the given file.
3976 * If alloc_type == 0, this request is for an ALLOCSP type
3977 * request which will change the file size. In this case, no
3978 * DMAPI event will be generated by the call. A TRUNCATE event
3979 * will be generated later by xfs_setattr.
3981 * If alloc_type != 0, this request is for a RESVSP type
3982 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3983 * lower block boundary byte address is less than the file's
3992 xfs_alloc_file_space(
3999 xfs_filblks_t allocated_fsb;
4000 xfs_filblks_t allocatesize_fsb;
4003 xfs_filblks_t datablocks;
4005 xfs_fsblock_t firstfsb;
4006 xfs_bmap_free_t free_list;
4007 xfs_bmbt_irec_t *imapp;
4008 xfs_bmbt_irec_t imaps[1];
4015 xfs_fileoff_t startoffset_fsb;
4017 int xfs_bmapi_flags;
4019 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4022 if (XFS_FORCED_SHUTDOWN(mp))
4023 return XFS_ERROR(EIO);
4026 * determine if this is a realtime file
4028 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4029 if (ip->i_d.di_extsize)
4030 rtextsize = ip->i_d.di_extsize;
4032 rtextsize = mp->m_sb.sb_rextsize;
4036 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4040 return XFS_ERROR(EINVAL);
4046 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4047 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4048 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4050 /* Generate a DMAPI event if needed. */
4051 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4052 (attr_flags&ATTR_DMI) == 0 &&
4053 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4054 xfs_off_t end_dmi_offset;
4056 end_dmi_offset = offset+len;
4057 if (end_dmi_offset > ip->i_d.di_size)
4058 end_dmi_offset = ip->i_d.di_size;
4059 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4060 offset, end_dmi_offset - offset,
4067 * allocate file space until done or until there is an error
4070 while (allocatesize_fsb && !error) {
4072 * determine if reserving space on
4073 * the data or realtime partition.
4078 s = startoffset_fsb;
4079 do_div(s, rtextsize);
4081 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4083 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4086 datablocks = allocatesize_fsb;
4091 * allocate and setup the transaction
4093 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4094 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4095 error = xfs_trans_reserve(tp,
4097 XFS_WRITE_LOG_RES(mp),
4099 XFS_TRANS_PERM_LOG_RES,
4100 XFS_WRITE_LOG_COUNT);
4103 * check for running out of space
4107 * Free the transaction structure.
4109 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4110 xfs_trans_cancel(tp, 0);
4113 xfs_ilock(ip, XFS_ILOCK_EXCL);
4114 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4115 ip->i_udquot, ip->i_gdquot, resblks, 0, 0);
4119 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4120 xfs_trans_ihold(tp, ip);
4123 * issue the bmapi() call to allocate the blocks
4125 XFS_BMAP_INIT(&free_list, &firstfsb);
4126 error = xfs_bmapi(tp, ip, startoffset_fsb,
4127 allocatesize_fsb, xfs_bmapi_flags,
4128 &firstfsb, 0, imapp, &reccount,
4135 * complete the transaction
4137 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4142 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4143 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4148 allocated_fsb = imapp->br_blockcount;
4150 if (reccount == 0) {
4151 error = XFS_ERROR(ENOSPC);
4155 startoffset_fsb += allocated_fsb;
4156 allocatesize_fsb -= allocated_fsb;
4159 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4160 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4162 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4163 XFS_ITOV(ip), DM_RIGHT_NULL,
4164 XFS_ITOV(ip), DM_RIGHT_NULL,
4165 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4167 goto retry; /* Maybe DMAPI app. has made space */
4168 /* else fall through with error from XFS_SEND_DATA */
4174 xfs_bmap_cancel(&free_list);
4176 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4177 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4178 goto dmapi_enospc_check;
4182 * Zero file bytes between startoff and endoff inclusive.
4183 * The iolock is held exclusive and no blocks are buffered.
4186 xfs_zero_remaining_bytes(
4191 xfs_bmbt_irec_t imap;
4192 xfs_fileoff_t offset_fsb;
4193 xfs_off_t lastoffset;
4196 xfs_mount_t *mp = ip->i_mount;
4200 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4201 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4202 mp->m_rtdev_targp : mp->m_ddev_targp);
4204 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4205 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4207 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4209 if (error || nimap < 1)
4211 ASSERT(imap.br_blockcount >= 1);
4212 ASSERT(imap.br_startoff == offset_fsb);
4213 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4214 if (lastoffset > endoff)
4215 lastoffset = endoff;
4216 if (imap.br_startblock == HOLESTARTBLOCK)
4218 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4219 if (imap.br_state == XFS_EXT_UNWRITTEN)
4222 XFS_BUF_UNWRITE(bp);
4224 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4226 if ((error = xfs_iowait(bp))) {
4227 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4228 mp, bp, XFS_BUF_ADDR(bp));
4231 memset(XFS_BUF_PTR(bp) +
4232 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4233 0, lastoffset - offset + 1);
4238 if ((error = xfs_iowait(bp))) {
4239 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4240 mp, bp, XFS_BUF_ADDR(bp));
4249 * xfs_free_file_space()
4250 * This routine frees disk space for the given file.
4252 * This routine is only called by xfs_change_file_space
4253 * for an UNRESVSP type call.
4261 xfs_free_file_space(
4270 xfs_off_t end_dmi_offset;
4271 xfs_fileoff_t endoffset_fsb;
4273 xfs_fsblock_t firstfsb;
4274 xfs_bmap_free_t free_list;
4276 xfs_bmbt_irec_t imap;
4284 xfs_fileoff_t startoffset_fsb;
4286 int need_iolock = 1;
4291 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4293 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4297 if (len <= 0) /* if nothing being freed */
4299 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4300 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4301 end_dmi_offset = offset + len;
4302 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4304 if (offset < ip->i_d.di_size &&
4305 (attr_flags & ATTR_DMI) == 0 &&
4306 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4307 if (end_dmi_offset > ip->i_d.di_size)
4308 end_dmi_offset = ip->i_d.di_size;
4309 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4310 offset, end_dmi_offset - offset,
4311 AT_DELAY_FLAG(attr_flags), NULL);
4316 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4317 if (attr_flags & ATTR_NOLOCK)
4320 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4322 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4324 ilen = len + (offset & (rounding - 1));
4325 ioffset = offset & ~(rounding - 1);
4326 if (ilen & (rounding - 1))
4327 ilen = (ilen + rounding) & ~(rounding - 1);
4329 if (VN_CACHED(vp) != 0) {
4330 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4331 ctooff(offtoct(ioffset)), -1);
4332 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4333 -1, FI_REMAPF_LOCKED);
4337 * Need to zero the stuff we're not freeing, on disk.
4338 * If its a realtime file & can't use unwritten extents then we
4339 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4340 * will take care of it for us.
4342 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4344 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4345 &imap, &nimap, NULL);
4347 goto out_unlock_iolock;
4348 ASSERT(nimap == 0 || nimap == 1);
4349 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4352 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4353 block = imap.br_startblock;
4354 mod = do_div(block, mp->m_sb.sb_rextsize);
4356 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4359 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 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) {
4365 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4367 if (mod && (mod != mp->m_sb.sb_rextsize))
4368 endoffset_fsb -= mod;
4371 if ((done = (endoffset_fsb <= startoffset_fsb)))
4373 * One contiguous piece to clear
4375 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4378 * Some full blocks, possibly two pieces to clear
4380 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4381 error = xfs_zero_remaining_bytes(ip, offset,
4382 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4384 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4385 error = xfs_zero_remaining_bytes(ip,
4386 XFS_FSB_TO_B(mp, endoffset_fsb),
4391 * free file space until done or until there is an error
4393 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4394 while (!error && !done) {
4397 * allocate and setup the transaction
4399 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4400 error = xfs_trans_reserve(tp,
4402 XFS_WRITE_LOG_RES(mp),
4404 XFS_TRANS_PERM_LOG_RES,
4405 XFS_WRITE_LOG_COUNT);
4408 * check for running out of space
4412 * Free the transaction structure.
4414 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4415 xfs_trans_cancel(tp, 0);
4418 xfs_ilock(ip, XFS_ILOCK_EXCL);
4419 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4420 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4421 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4425 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4426 xfs_trans_ihold(tp, ip);
4429 * issue the bunmapi() call to free the blocks
4431 XFS_BMAP_INIT(&free_list, &firstfsb);
4432 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4433 endoffset_fsb - startoffset_fsb,
4434 0, 2, &firstfsb, &free_list, &done);
4440 * complete the transaction
4442 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4447 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4448 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4453 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4457 xfs_bmap_cancel(&free_list);
4459 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4460 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4466 * xfs_change_file_space()
4467 * This routine allocates or frees disk space for the given file.
4468 * The user specified parameters are checked for alignment and size
4477 xfs_change_file_space(
4491 xfs_off_t startoffset;
4497 vp = BHV_TO_VNODE(bdp);
4498 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4500 ip = XFS_BHVTOI(bdp);
4504 * must be a regular file and have write permission
4507 return XFS_ERROR(EINVAL);
4509 xfs_ilock(ip, XFS_ILOCK_SHARED);
4511 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4512 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4516 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4518 switch (bf->l_whence) {
4519 case 0: /*SEEK_SET*/
4521 case 1: /*SEEK_CUR*/
4522 bf->l_start += offset;
4524 case 2: /*SEEK_END*/
4525 bf->l_start += ip->i_d.di_size;
4528 return XFS_ERROR(EINVAL);
4531 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4533 if ( (bf->l_start < 0)
4534 || (bf->l_start > XFS_MAXIOFFSET(mp))
4535 || (bf->l_start + llen < 0)
4536 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4537 return XFS_ERROR(EINVAL);
4541 startoffset = bf->l_start;
4542 fsize = ip->i_d.di_size;
4545 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4547 * These calls do NOT zero the data space allocated to the file,
4548 * nor do they change the file size.
4550 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4552 * These calls cause the new file data to be zeroed and the file
4553 * size to be changed.
4555 setprealloc = clrprealloc = 0;
4558 case XFS_IOC_RESVSP:
4559 case XFS_IOC_RESVSP64:
4560 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4567 case XFS_IOC_UNRESVSP:
4568 case XFS_IOC_UNRESVSP64:
4569 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4574 case XFS_IOC_ALLOCSP:
4575 case XFS_IOC_ALLOCSP64:
4576 case XFS_IOC_FREESP:
4577 case XFS_IOC_FREESP64:
4578 if (startoffset > fsize) {
4579 error = xfs_alloc_file_space(ip, fsize,
4580 startoffset - fsize, 0, attr_flags);
4585 va.va_mask = XFS_AT_SIZE;
4586 va.va_size = startoffset;
4588 error = xfs_setattr(bdp, &va, attr_flags, credp);
4598 return XFS_ERROR(EINVAL);
4602 * update the inode timestamp, mode, and prealloc flag bits
4604 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4606 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4609 xfs_trans_cancel(tp, 0);
4613 xfs_ilock(ip, XFS_ILOCK_EXCL);
4615 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4616 xfs_trans_ihold(tp, ip);
4618 if ((attr_flags & ATTR_DMI) == 0) {
4619 ip->i_d.di_mode &= ~S_ISUID;
4622 * Note that we don't have to worry about mandatory
4623 * file locking being disabled here because we only
4624 * clear the S_ISGID bit if the Group execute bit is
4625 * on, but if it was on then mandatory locking wouldn't
4626 * have been enabled.
4628 if (ip->i_d.di_mode & S_IXGRP)
4629 ip->i_d.di_mode &= ~S_ISGID;
4631 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4634 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4635 else if (clrprealloc)
4636 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4638 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4639 xfs_trans_set_sync(tp);
4641 error = xfs_trans_commit(tp, 0, NULL);
4643 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4648 vnodeops_t xfs_vnodeops = {
4649 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4650 .vop_open = xfs_open,
4651 .vop_read = xfs_read,
4652 #ifdef HAVE_SENDFILE
4653 .vop_sendfile = xfs_sendfile,
4655 .vop_write = xfs_write,
4656 .vop_ioctl = xfs_ioctl,
4657 .vop_getattr = xfs_getattr,
4658 .vop_setattr = xfs_setattr,
4659 .vop_access = xfs_access,
4660 .vop_lookup = xfs_lookup,
4661 .vop_create = xfs_create,
4662 .vop_remove = xfs_remove,
4663 .vop_link = xfs_link,
4664 .vop_rename = xfs_rename,
4665 .vop_mkdir = xfs_mkdir,
4666 .vop_rmdir = xfs_rmdir,
4667 .vop_readdir = xfs_readdir,
4668 .vop_symlink = xfs_symlink,
4669 .vop_readlink = xfs_readlink,
4670 .vop_fsync = xfs_fsync,
4671 .vop_inactive = xfs_inactive,
4672 .vop_fid2 = xfs_fid2,
4673 .vop_rwlock = xfs_rwlock,
4674 .vop_rwunlock = xfs_rwunlock,
4675 .vop_bmap = xfs_bmap,
4676 .vop_reclaim = xfs_reclaim,
4677 .vop_attr_get = xfs_attr_get,
4678 .vop_attr_set = xfs_attr_set,
4679 .vop_attr_remove = xfs_attr_remove,
4680 .vop_attr_list = xfs_attr_list,
4681 .vop_link_removed = (vop_link_removed_t)fs_noval,
4682 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4683 .vop_tosspages = fs_tosspages,
4684 .vop_flushinval_pages = fs_flushinval_pages,
4685 .vop_flush_pages = fs_flush_pages,
4686 .vop_release = xfs_release,
4687 .vop_iflush = xfs_inode_flush,