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 && (mask & XFS_AT_EXTSIZE) &&
537 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
539 code = XFS_ERROR(EINVAL); /* EFBIG? */
544 * Can't change realtime flag if any extents are allocated.
546 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
547 (mask & XFS_AT_XFLAGS) &&
548 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
549 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
550 code = XFS_ERROR(EINVAL); /* EFBIG? */
554 * Extent size must be a multiple of the appropriate block
555 * size, if set at all.
557 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
560 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
561 ((mask & XFS_AT_XFLAGS) &&
562 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
563 size = mp->m_sb.sb_rextsize <<
564 mp->m_sb.sb_blocklog;
566 size = mp->m_sb.sb_blocksize;
568 if (vap->va_extsize % size) {
569 code = XFS_ERROR(EINVAL);
574 * If realtime flag is set then must have realtime data.
576 if ((mask & XFS_AT_XFLAGS) &&
577 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
578 if ((mp->m_sb.sb_rblocks == 0) ||
579 (mp->m_sb.sb_rextsize == 0) ||
580 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
581 code = XFS_ERROR(EINVAL);
587 * Can't modify an immutable/append-only file unless
588 * we have appropriate permission.
590 if ((mask & XFS_AT_XFLAGS) &&
592 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
594 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
595 !capable(CAP_LINUX_IMMUTABLE)) {
596 code = XFS_ERROR(EPERM);
602 * Now we can make the changes. Before we join the inode
603 * to the transaction, if XFS_AT_SIZE is set then take care of
604 * the part of the truncation that must be done without the
605 * inode lock. This needs to be done before joining the inode
606 * to the transaction, because the inode cannot be unlocked
607 * once it is a part of the transaction.
609 if (mask & XFS_AT_SIZE) {
611 if ((vap->va_size > ip->i_d.di_size) &&
612 (flags & ATTR_NOSIZETOK) == 0) {
613 code = xfs_igrow_start(ip, vap->va_size, credp);
615 xfs_iunlock(ip, XFS_ILOCK_EXCL);
617 code = xfs_itruncate_data(ip, vap->va_size);
620 lock_flags &= ~XFS_ILOCK_EXCL;
621 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
624 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
625 if ((code = xfs_trans_reserve(tp, 0,
626 XFS_ITRUNCATE_LOG_RES(mp), 0,
627 XFS_TRANS_PERM_LOG_RES,
628 XFS_ITRUNCATE_LOG_COUNT))) {
629 xfs_trans_cancel(tp, 0);
631 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
634 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
635 xfs_ilock(ip, XFS_ILOCK_EXCL);
639 xfs_trans_ijoin(tp, ip, lock_flags);
640 xfs_trans_ihold(tp, ip);
643 /* determine whether mandatory locking mode changes */
644 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
647 * Truncate file. Must have write permission and not be a directory.
649 if (mask & XFS_AT_SIZE) {
650 if (vap->va_size > ip->i_d.di_size) {
651 xfs_igrow_finish(tp, ip, vap->va_size,
652 !(flags & ATTR_DMI));
653 } else if ((vap->va_size <= ip->i_d.di_size) ||
654 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
656 * signal a sync transaction unless
657 * we're truncating an already unlinked
658 * file on a wsync filesystem
660 code = xfs_itruncate_finish(&tp, ip,
661 (xfs_fsize_t)vap->va_size,
663 ((ip->i_d.di_nlink != 0 ||
664 !(mp->m_flags & XFS_MOUNT_WSYNC))
671 * Have to do this even if the file's size doesn't change.
673 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
677 * Change file access modes.
679 if (mask & XFS_AT_MODE) {
680 ip->i_d.di_mode &= S_IFMT;
681 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
683 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
684 timeflags |= XFS_ICHGTIME_CHG;
688 * Change file ownership. Must be the owner or privileged.
689 * If the system was configured with the "restricted_chown"
690 * option, the owner is not permitted to give away the file,
691 * and can change the group id only to a group of which he
692 * or she is a member.
694 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
696 * CAP_FSETID overrides the following restrictions:
698 * The set-user-ID and set-group-ID bits of a file will be
699 * cleared upon successful return from chown()
701 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
702 !capable(CAP_FSETID)) {
703 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
707 * Change the ownerships and register quota modifications
708 * in the transaction.
711 if (XFS_IS_UQUOTA_ON(mp)) {
712 ASSERT(mask & XFS_AT_UID);
714 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
715 &ip->i_udquot, udqp);
717 ip->i_d.di_uid = uid;
720 if (XFS_IS_GQUOTA_ON(mp)) {
721 ASSERT(!XFS_IS_PQUOTA_ON(mp));
722 ASSERT(mask & XFS_AT_GID);
724 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
725 &ip->i_gdquot, gdqp);
727 ip->i_d.di_gid = gid;
729 if (iprojid != projid) {
730 if (XFS_IS_PQUOTA_ON(mp)) {
731 ASSERT(!XFS_IS_GQUOTA_ON(mp));
732 ASSERT(mask & XFS_AT_PROJID);
734 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
735 &ip->i_gdquot, gdqp);
737 ip->i_d.di_projid = projid;
739 * We may have to rev the inode as well as
740 * the superblock version number since projids didn't
741 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
743 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
744 xfs_bump_ino_vers2(tp, ip);
747 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
748 timeflags |= XFS_ICHGTIME_CHG;
753 * Change file access or modified times.
755 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
756 if (mask & XFS_AT_ATIME) {
757 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
758 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
759 ip->i_update_core = 1;
760 timeflags &= ~XFS_ICHGTIME_ACC;
762 if (mask & XFS_AT_MTIME) {
763 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
764 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
765 timeflags &= ~XFS_ICHGTIME_MOD;
766 timeflags |= XFS_ICHGTIME_CHG;
768 if (tp && (flags & ATTR_UTIME))
769 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
773 * Change XFS-added attributes.
775 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
776 if (mask & XFS_AT_EXTSIZE) {
778 * Converting bytes to fs blocks.
780 ip->i_d.di_extsize = vap->va_extsize >>
781 mp->m_sb.sb_blocklog;
783 if (mask & XFS_AT_XFLAGS) {
786 /* can't set PREALLOC this way, just preserve it */
787 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
788 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
789 di_flags |= XFS_DIFLAG_IMMUTABLE;
790 if (vap->va_xflags & XFS_XFLAG_APPEND)
791 di_flags |= XFS_DIFLAG_APPEND;
792 if (vap->va_xflags & XFS_XFLAG_SYNC)
793 di_flags |= XFS_DIFLAG_SYNC;
794 if (vap->va_xflags & XFS_XFLAG_NOATIME)
795 di_flags |= XFS_DIFLAG_NOATIME;
796 if (vap->va_xflags & XFS_XFLAG_NODUMP)
797 di_flags |= XFS_DIFLAG_NODUMP;
798 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
799 di_flags |= XFS_DIFLAG_PROJINHERIT;
800 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
801 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
802 di_flags |= XFS_DIFLAG_RTINHERIT;
803 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
804 di_flags |= XFS_DIFLAG_NOSYMLINKS;
805 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
806 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
807 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
808 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
809 di_flags |= XFS_DIFLAG_REALTIME;
810 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
812 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
814 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
815 di_flags |= XFS_DIFLAG_EXTSIZE;
817 ip->i_d.di_flags = di_flags;
819 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
820 timeflags |= XFS_ICHGTIME_CHG;
824 * Change file inode change time only if XFS_AT_CTIME set
825 * AND we have been called by a DMI function.
828 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
829 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
830 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
831 ip->i_update_core = 1;
832 timeflags &= ~XFS_ICHGTIME_CHG;
836 * Send out timestamp changes that need to be set to the
837 * current time. Not done when called by a DMI function.
839 if (timeflags && !(flags & ATTR_DMI))
840 xfs_ichgtime(ip, timeflags);
842 XFS_STATS_INC(xs_ig_attrchg);
845 * If this is a synchronous mount, make sure that the
846 * transaction goes to disk before returning to the user.
847 * This is slightly sub-optimal in that truncates require
848 * two sync transactions instead of one for wsync filesytems.
849 * One for the truncate and one for the timestamps since we
850 * don't want to change the timestamps unless we're sure the
851 * truncate worked. Truncates are less than 1% of the laddis
852 * mix so this probably isn't worth the trouble to optimize.
856 if (mp->m_flags & XFS_MOUNT_WSYNC)
857 xfs_trans_set_sync(tp);
859 code = xfs_trans_commit(tp, commit_flags, NULL);
863 * If the (regular) file's mandatory locking mode changed, then
864 * notify the vnode. We do this under the inode lock to prevent
865 * racing calls to vop_vnode_change.
867 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
868 if (mandlock_before != mandlock_after) {
869 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
873 xfs_iunlock(ip, lock_flags);
876 * Release any dquot(s) the inode had kept before chown.
878 XFS_QM_DQRELE(mp, olddquot1);
879 XFS_QM_DQRELE(mp, olddquot2);
880 XFS_QM_DQRELE(mp, udqp);
881 XFS_QM_DQRELE(mp, gdqp);
887 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
888 !(flags & ATTR_DMI)) {
889 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
890 NULL, DM_RIGHT_NULL, NULL, NULL,
891 0, 0, AT_DELAY_FLAG(flags));
896 commit_flags |= XFS_TRANS_ABORT;
899 XFS_QM_DQRELE(mp, udqp);
900 XFS_QM_DQRELE(mp, gdqp);
902 xfs_trans_cancel(tp, commit_flags);
904 if (lock_flags != 0) {
905 xfs_iunlock(ip, lock_flags);
913 * Null conversion from vnode mode bits to inode mode bits, as in efs.
924 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
925 (inst_t *)__return_address);
927 ip = XFS_BHVTOI(bdp);
928 xfs_ilock(ip, XFS_ILOCK_SHARED);
929 error = xfs_iaccess(ip, mode, credp);
930 xfs_iunlock(ip, XFS_ILOCK_SHARED);
954 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
960 vp = BHV_TO_VNODE(bdp);
961 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
963 ip = XFS_BHVTOI(bdp);
966 if (XFS_FORCED_SHUTDOWN(mp))
967 return XFS_ERROR(EIO);
969 xfs_ilock(ip, XFS_ILOCK_SHARED);
971 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
973 offset = uiop->uio_offset;
974 count = uiop->uio_resid;
977 error = XFS_ERROR(EINVAL);
985 if (!(ioflags & IO_INVIS)) {
986 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
990 * See if the symlink is stored inline.
992 pathlen = (int)ip->i_d.di_size;
994 if (ip->i_df.if_flags & XFS_IFINLINE) {
995 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
999 * Symlink not inline. Call bmap to get it in.
1001 nmaps = SYMLINK_MAPS;
1003 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1004 0, NULL, 0, mval, &nmaps, NULL);
1010 for (n = 0; n < nmaps; n++) {
1011 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1012 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1013 bp = xfs_buf_read(mp->m_ddev_targp, d,
1014 BTOBB(byte_cnt), 0);
1015 error = XFS_BUF_GETERROR(bp);
1017 xfs_ioerror_alert("xfs_readlink",
1018 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1022 if (pathlen < byte_cnt)
1024 pathlen -= byte_cnt;
1026 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1035 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1044 * This is called to sync the inode and its data out to disk.
1045 * We need to hold the I/O lock while flushing the data, and
1046 * the inode lock while flushing the inode. The inode lock CANNOT
1047 * be held while flushing the data, so acquire after we're done
1061 int log_flushed = 0, changed = 1;
1063 vn_trace_entry(BHV_TO_VNODE(bdp),
1064 __FUNCTION__, (inst_t *)__return_address);
1066 ip = XFS_BHVTOI(bdp);
1068 ASSERT(start >= 0 && stop >= -1);
1070 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1071 return XFS_ERROR(EIO);
1074 * We always need to make sure that the required inode state
1075 * is safe on disk. The vnode might be clean but because
1076 * of committed transactions that haven't hit the disk yet.
1077 * Likewise, there could be unflushed non-transactional
1078 * changes to the inode core that have to go to disk.
1080 * The following code depends on one assumption: that
1081 * any transaction that changes an inode logs the core
1082 * because it has to change some field in the inode core
1083 * (typically nextents or nblocks). That assumption
1084 * implies that any transactions against an inode will
1085 * catch any non-transactional updates. If inode-altering
1086 * transactions exist that violate this assumption, the
1087 * code breaks. Right now, it figures that if the involved
1088 * update_* field is clear and the inode is unpinned, the
1089 * inode is clean. Either it's been flushed or it's been
1090 * committed and the commit has hit the disk unpinning the inode.
1091 * (Note that xfs_inode_item_format() called at commit clears
1092 * the update_* fields.)
1094 xfs_ilock(ip, XFS_ILOCK_SHARED);
1096 /* If we are flushing data then we care about update_size
1097 * being set, otherwise we care about update_core
1099 if ((flag & FSYNC_DATA) ?
1100 (ip->i_update_size == 0) :
1101 (ip->i_update_core == 0)) {
1103 * Timestamps/size haven't changed since last inode
1104 * flush or inode transaction commit. That means
1105 * either nothing got written or a transaction
1106 * committed which caught the updates. If the
1107 * latter happened and the transaction hasn't
1108 * hit the disk yet, the inode will be still
1109 * be pinned. If it is, force the log.
1112 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1114 if (xfs_ipincount(ip)) {
1115 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1117 ((flag & FSYNC_WAIT)
1118 ? XFS_LOG_SYNC : 0),
1122 * If the inode is not pinned and nothing
1123 * has changed we don't need to flush the
1131 * Kick off a transaction to log the inode
1132 * core to get the updates. Make it
1133 * sync if FSYNC_WAIT is passed in (which
1134 * is done by everybody but specfs). The
1135 * sync transaction will also force the log.
1137 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1138 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1139 if ((error = xfs_trans_reserve(tp, 0,
1140 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1142 xfs_trans_cancel(tp, 0);
1145 xfs_ilock(ip, XFS_ILOCK_EXCL);
1148 * Note - it's possible that we might have pushed
1149 * ourselves out of the way during trans_reserve
1150 * which would flush the inode. But there's no
1151 * guarantee that the inode buffer has actually
1152 * gone out yet (it's delwri). Plus the buffer
1153 * could be pinned anyway if it's part of an
1154 * inode in another recent transaction. So we
1155 * play it safe and fire off the transaction anyway.
1157 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1158 xfs_trans_ihold(tp, ip);
1159 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1160 if (flag & FSYNC_WAIT)
1161 xfs_trans_set_sync(tp);
1162 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1164 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1167 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1169 * If the log write didn't issue an ordered tag we need
1170 * to flush the disk cache for the data device now.
1173 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1176 * If this inode is on the RT dev we need to flush that
1179 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1180 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1187 * This is called by xfs_inactive to free any blocks beyond eof,
1188 * when the link count isn't zero.
1191 xfs_inactive_free_eofblocks(
1197 xfs_fileoff_t end_fsb;
1198 xfs_fileoff_t last_fsb;
1199 xfs_filblks_t map_len;
1201 xfs_bmbt_irec_t imap;
1204 * Figure out if there are any blocks beyond the end
1205 * of the file. If not, then there is nothing to do.
1207 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1208 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1209 map_len = last_fsb - end_fsb;
1214 xfs_ilock(ip, XFS_ILOCK_SHARED);
1215 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1216 NULL, 0, &imap, &nimaps, NULL);
1217 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1219 if (!error && (nimaps != 0) &&
1220 (imap.br_startblock != HOLESTARTBLOCK)) {
1222 * Attach the dquots to the inode up front.
1224 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1228 * There are blocks after the end of file.
1229 * Free them up now by truncating the file to
1232 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1235 * Do the xfs_itruncate_start() call before
1236 * reserving any log space because
1237 * itruncate_start will call into the buffer
1238 * cache and we can't
1239 * do that within a transaction.
1241 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1242 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1245 error = xfs_trans_reserve(tp, 0,
1246 XFS_ITRUNCATE_LOG_RES(mp),
1247 0, XFS_TRANS_PERM_LOG_RES,
1248 XFS_ITRUNCATE_LOG_COUNT);
1250 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1251 xfs_trans_cancel(tp, 0);
1252 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1256 xfs_ilock(ip, XFS_ILOCK_EXCL);
1257 xfs_trans_ijoin(tp, ip,
1260 xfs_trans_ihold(tp, ip);
1262 error = xfs_itruncate_finish(&tp, ip,
1267 * If we get an error at this point we
1268 * simply don't bother truncating the file.
1271 xfs_trans_cancel(tp,
1272 (XFS_TRANS_RELEASE_LOG_RES |
1275 error = xfs_trans_commit(tp,
1276 XFS_TRANS_RELEASE_LOG_RES,
1279 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1285 * Free a symlink that has blocks associated with it.
1288 xfs_inactive_symlink_rmt(
1296 xfs_fsblock_t first_block;
1297 xfs_bmap_free_t free_list;
1300 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1308 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1310 * We're freeing a symlink that has some
1311 * blocks allocated to it. Free the
1312 * blocks here. We know that we've got
1313 * either 1 or 2 extents and that we can
1314 * free them all in one bunmapi call.
1316 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1317 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1318 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1319 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1320 xfs_trans_cancel(tp, 0);
1325 * Lock the inode, fix the size, and join it to the transaction.
1326 * Hold it so in the normal path, we still have it locked for
1327 * the second transaction. In the error paths we need it
1328 * held so the cancel won't rele it, see below.
1330 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1331 size = (int)ip->i_d.di_size;
1332 ip->i_d.di_size = 0;
1333 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1334 xfs_trans_ihold(tp, ip);
1335 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1337 * Find the block(s) so we can inval and unmap them.
1340 XFS_BMAP_INIT(&free_list, &first_block);
1341 nmaps = sizeof(mval) / sizeof(mval[0]);
1342 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1343 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1347 * Invalidate the block(s).
1349 for (i = 0; i < nmaps; i++) {
1350 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1351 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1352 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1353 xfs_trans_binval(tp, bp);
1356 * Unmap the dead block(s) to the free_list.
1358 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1359 &first_block, &free_list, &done)))
1363 * Commit the first transaction. This logs the EFI and the inode.
1365 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1368 * The transaction must have been committed, since there were
1369 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1370 * The new tp has the extent freeing and EFDs.
1374 * The first xact was committed, so add the inode to the new one.
1375 * Mark it dirty so it will be logged and moved forward in the log as
1376 * part of every commit.
1378 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1379 xfs_trans_ihold(tp, ip);
1380 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1382 * Get a new, empty transaction to return to our caller.
1384 ntp = xfs_trans_dup(tp);
1386 * Commit the transaction containing extent freeing and EFD's.
1387 * If we get an error on the commit here or on the reserve below,
1388 * we need to unlock the inode since the new transaction doesn't
1389 * have the inode attached.
1391 error = xfs_trans_commit(tp, 0, NULL);
1394 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1398 * Remove the memory for extent descriptions (just bookkeeping).
1400 if (ip->i_df.if_bytes)
1401 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1402 ASSERT(ip->i_df.if_bytes == 0);
1404 * Put an itruncate log reservation in the new transaction
1407 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1408 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1409 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1413 * Return with the inode locked but not joined to the transaction.
1419 xfs_bmap_cancel(&free_list);
1422 * Have to come here with the inode locked and either
1423 * (held and in the transaction) or (not in the transaction).
1424 * If the inode isn't held then cancel would iput it, but
1425 * that's wrong since this is inactive and the vnode ref
1426 * count is 0 already.
1427 * Cancel won't do anything to the inode if held, but it still
1428 * needs to be locked until the cancel is done, if it was
1429 * joined to the transaction.
1431 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1432 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1439 xfs_inactive_symlink_local(
1445 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1447 * We're freeing a symlink which fit into
1448 * the inode. Just free the memory used
1449 * to hold the old symlink.
1451 error = xfs_trans_reserve(*tpp, 0,
1452 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1453 0, XFS_TRANS_PERM_LOG_RES,
1454 XFS_ITRUNCATE_LOG_COUNT);
1457 xfs_trans_cancel(*tpp, 0);
1461 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1464 * Zero length symlinks _can_ exist.
1466 if (ip->i_df.if_bytes > 0) {
1467 xfs_idata_realloc(ip,
1468 -(ip->i_df.if_bytes),
1470 ASSERT(ip->i_df.if_bytes == 0);
1487 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1490 ASSERT(ip->i_d.di_forkoff != 0);
1491 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1492 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1494 error = xfs_attr_inactive(ip);
1497 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1498 return (error); /* goto out*/
1501 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1502 error = xfs_trans_reserve(tp, 0,
1503 XFS_IFREE_LOG_RES(mp),
1504 0, XFS_TRANS_PERM_LOG_RES,
1505 XFS_INACTIVE_LOG_COUNT);
1507 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1508 xfs_trans_cancel(tp, 0);
1510 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1514 xfs_ilock(ip, XFS_ILOCK_EXCL);
1515 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1516 xfs_trans_ihold(tp, ip);
1517 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1519 ASSERT(ip->i_d.di_anextents == 0);
1534 vp = BHV_TO_VNODE(bdp);
1535 ip = XFS_BHVTOI(bdp);
1537 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1541 /* If this is a read-only mount, don't do this (would generate I/O) */
1542 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1545 #ifdef HAVE_REFCACHE
1546 /* If we are in the NFS reference cache then don't do this now */
1553 if (ip->i_d.di_nlink != 0) {
1554 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1555 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1556 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1557 (!(ip->i_d.di_flags &
1558 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1559 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1561 /* Update linux inode block count after free above */
1562 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1563 ip->i_d.di_nblocks + ip->i_delayed_blks);
1573 * This is called when the vnode reference count for the vnode
1574 * goes to zero. If the file has been unlinked, then it must
1575 * now be truncated. Also, we clear all of the read-ahead state
1576 * kept for the inode here since the file is now closed.
1585 xfs_bmap_free_t free_list;
1586 xfs_fsblock_t first_block;
1593 vp = BHV_TO_VNODE(bdp);
1594 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1596 ip = XFS_BHVTOI(bdp);
1599 * If the inode is already free, then there can be nothing
1602 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1603 ASSERT(ip->i_df.if_real_bytes == 0);
1604 ASSERT(ip->i_df.if_broot_bytes == 0);
1605 return VN_INACTIVE_CACHE;
1609 * Only do a truncate if it's a regular file with
1610 * some actual space in it. It's OK to look at the
1611 * inode's fields without the lock because we're the
1612 * only one with a reference to the inode.
1614 truncate = ((ip->i_d.di_nlink == 0) &&
1615 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1616 (ip->i_delayed_blks > 0)) &&
1617 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1621 if (ip->i_d.di_nlink == 0 &&
1622 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1623 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1628 /* If this is a read-only mount, don't do this (would generate I/O) */
1629 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1632 if (ip->i_d.di_nlink != 0) {
1633 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1634 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1635 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1636 (!(ip->i_d.di_flags &
1637 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1638 (ip->i_delayed_blks != 0)))) {
1639 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1640 return (VN_INACTIVE_CACHE);
1641 /* Update linux inode block count after free above */
1642 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1643 ip->i_d.di_nblocks + ip->i_delayed_blks);
1648 ASSERT(ip->i_d.di_nlink == 0);
1650 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1651 return (VN_INACTIVE_CACHE);
1653 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1656 * Do the xfs_itruncate_start() call before
1657 * reserving any log space because itruncate_start
1658 * will call into the buffer cache and we can't
1659 * do that within a transaction.
1661 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1663 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1665 error = xfs_trans_reserve(tp, 0,
1666 XFS_ITRUNCATE_LOG_RES(mp),
1667 0, XFS_TRANS_PERM_LOG_RES,
1668 XFS_ITRUNCATE_LOG_COUNT);
1670 /* Don't call itruncate_cleanup */
1671 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1672 xfs_trans_cancel(tp, 0);
1673 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1674 return (VN_INACTIVE_CACHE);
1677 xfs_ilock(ip, XFS_ILOCK_EXCL);
1678 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1679 xfs_trans_ihold(tp, ip);
1682 * normally, we have to run xfs_itruncate_finish sync.
1683 * But if filesystem is wsync and we're in the inactive
1684 * path, then we know that nlink == 0, and that the
1685 * xaction that made nlink == 0 is permanently committed
1686 * since xfs_remove runs as a synchronous transaction.
1688 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1689 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1692 xfs_trans_cancel(tp,
1693 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1694 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1695 return (VN_INACTIVE_CACHE);
1697 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1700 * If we get an error while cleaning up a
1701 * symlink we bail out.
1703 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1704 xfs_inactive_symlink_rmt(ip, &tp) :
1705 xfs_inactive_symlink_local(ip, &tp);
1709 return (VN_INACTIVE_CACHE);
1712 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1713 xfs_trans_ihold(tp, ip);
1715 error = xfs_trans_reserve(tp, 0,
1716 XFS_IFREE_LOG_RES(mp),
1717 0, XFS_TRANS_PERM_LOG_RES,
1718 XFS_INACTIVE_LOG_COUNT);
1720 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1721 xfs_trans_cancel(tp, 0);
1722 return (VN_INACTIVE_CACHE);
1725 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1726 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1727 xfs_trans_ihold(tp, ip);
1731 * If there are attributes associated with the file
1732 * then blow them away now. The code calls a routine
1733 * that recursively deconstructs the attribute fork.
1734 * We need to just commit the current transaction
1735 * because we can't use it for xfs_attr_inactive().
1737 if (ip->i_d.di_anextents > 0) {
1738 error = xfs_inactive_attrs(ip, &tp);
1740 * If we got an error, the transaction is already
1741 * cancelled, and the inode is unlocked. Just get out.
1744 return (VN_INACTIVE_CACHE);
1745 } else if (ip->i_afp) {
1746 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1752 XFS_BMAP_INIT(&free_list, &first_block);
1753 error = xfs_ifree(tp, ip, &free_list);
1756 * If we fail to free the inode, shut down. The cancel
1757 * might do that, we need to make sure. Otherwise the
1758 * inode might be lost for a long time or forever.
1760 if (!XFS_FORCED_SHUTDOWN(mp)) {
1762 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1763 error, mp->m_fsname);
1764 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1766 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1769 * Credit the quota account(s). The inode is gone.
1771 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1774 * Just ignore errors at this point. There is
1775 * nothing we can do except to try to keep going.
1777 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1779 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1782 * Release the dquots held by inode, if any.
1784 XFS_QM_DQDETACH(mp, ip);
1786 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1789 return VN_INACTIVE_CACHE;
1798 bhv_desc_t *dir_bdp,
1805 xfs_inode_t *dp, *ip;
1811 dir_vp = BHV_TO_VNODE(dir_bdp);
1812 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1814 dp = XFS_BHVTOI(dir_bdp);
1816 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1817 return XFS_ERROR(EIO);
1819 lock_mode = xfs_ilock_map_shared(dp);
1820 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1822 *vpp = XFS_ITOV(ip);
1825 xfs_iunlock_map_shared(dp, lock_mode);
1831 * xfs_create (create a new file).
1835 bhv_desc_t *dir_bdp,
1841 char *name = VNAME(dentry);
1843 xfs_inode_t *dp, *ip;
1849 xfs_bmap_free_t free_list;
1850 xfs_fsblock_t first_block;
1851 boolean_t dp_joined_to_trans;
1852 int dm_event_sent = 0;
1856 struct xfs_dquot *udqp, *gdqp;
1862 dir_vp = BHV_TO_VNODE(dir_bdp);
1863 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1865 dp = XFS_BHVTOI(dir_bdp);
1868 dm_di_mode = vap->va_mode;
1869 namelen = VNAMELEN(dentry);
1871 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1872 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1873 dir_vp, DM_RIGHT_NULL, NULL,
1874 DM_RIGHT_NULL, name, NULL,
1882 if (XFS_FORCED_SHUTDOWN(mp))
1883 return XFS_ERROR(EIO);
1885 /* Return through std_return after this point. */
1888 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1889 prid = dp->i_d.di_projid;
1890 else if (vap->va_mask & XFS_AT_PROJID)
1891 prid = (xfs_prid_t)vap->va_projid;
1893 prid = (xfs_prid_t)dfltprid;
1896 * Make sure that we have allocated dquot(s) on disk.
1898 error = XFS_QM_DQVOPALLOC(mp, dp,
1899 current_fsuid(credp), current_fsgid(credp), prid,
1900 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1905 dp_joined_to_trans = B_FALSE;
1907 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1908 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1909 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1911 * Initially assume that the file does not exist and
1912 * reserve the resources for that case. If that is not
1913 * the case we'll drop the one we have and get a more
1914 * appropriate transaction later.
1916 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1917 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1918 if (error == ENOSPC) {
1920 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1921 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1929 xfs_ilock(dp, XFS_ILOCK_EXCL);
1931 XFS_BMAP_INIT(&free_list, &first_block);
1936 * Reserve disk quota and the inode.
1938 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1943 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1945 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1946 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1947 rdev, credp, prid, resblks > 0,
1950 if (error == ENOSPC)
1957 * At this point, we've gotten a newly allocated inode.
1958 * It is locked (and joined to the transaction).
1961 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1964 * Now we join the directory inode to the transaction.
1965 * We do not do it earlier because xfs_dir_ialloc
1966 * might commit the previous transaction (and release
1971 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1972 dp_joined_to_trans = B_TRUE;
1974 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1975 &first_block, &free_list,
1976 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1978 ASSERT(error != ENOSPC);
1981 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1982 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1985 * If this is a synchronous mount, make sure that the
1986 * create transaction goes to disk before returning to
1989 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1990 xfs_trans_set_sync(tp);
1996 * Attach the dquot(s) to the inodes and modify them incore.
1997 * These ids of the inode couldn't have changed since the new
1998 * inode has been locked ever since it was created.
2000 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2003 * xfs_trans_commit normally decrements the vnode ref count
2004 * when it unlocks the inode. Since we want to return the
2005 * vnode to the caller, we bump the vnode ref count now.
2010 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2012 xfs_bmap_cancel(&free_list);
2016 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2023 XFS_QM_DQRELE(mp, udqp);
2024 XFS_QM_DQRELE(mp, gdqp);
2027 * Propogate the fact that the vnode changed after the
2028 * xfs_inode locks have been released.
2030 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2034 /* Fallthrough to std_return with error = 0 */
2037 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2038 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2039 DM_EVENT_POSTCREATE)) {
2040 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2041 dir_vp, DM_RIGHT_NULL,
2043 DM_RIGHT_NULL, name, NULL,
2044 dm_di_mode, error, 0);
2049 cancel_flags |= XFS_TRANS_ABORT;
2054 xfs_trans_cancel(tp, cancel_flags);
2056 if (!dp_joined_to_trans && (dp != NULL))
2057 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2058 XFS_QM_DQRELE(mp, udqp);
2059 XFS_QM_DQRELE(mp, gdqp);
2065 * Wait until after the current transaction is aborted to
2066 * release the inode. This prevents recursive transactions
2067 * and deadlocks from xfs_inactive.
2069 cancel_flags |= XFS_TRANS_ABORT;
2070 xfs_trans_cancel(tp, cancel_flags);
2073 XFS_QM_DQRELE(mp, udqp);
2074 XFS_QM_DQRELE(mp, gdqp);
2081 * Some counters to see if (and how often) we are hitting some deadlock
2082 * prevention code paths.
2086 int xfs_rm_lock_delays;
2087 int xfs_rm_attempts;
2091 * The following routine will lock the inodes associated with the
2092 * directory and the named entry in the directory. The locks are
2093 * acquired in increasing inode number.
2095 * If the entry is "..", then only the directory is locked. The
2096 * vnode ref count will still include that from the .. entry in
2099 * There is a deadlock we need to worry about. If the locked directory is
2100 * in the AIL, it might be blocking up the log. The next inode we lock
2101 * could be already locked by another thread waiting for log space (e.g
2102 * a permanent log reservation with a long running transaction (see
2103 * xfs_itruncate_finish)). To solve this, we must check if the directory
2104 * is in the ail and use lock_nowait. If we can't lock, we need to
2105 * drop the inode lock on the directory and try again. xfs_iunlock will
2106 * potentially push the tail if we were holding up the log.
2109 xfs_lock_dir_and_entry(
2112 xfs_inode_t *ip) /* inode of entry 'name' */
2116 xfs_inode_t *ips[2];
2125 xfs_ilock(dp, XFS_ILOCK_EXCL);
2132 * We want to lock in increasing inum. Since we've already
2133 * acquired the lock on the directory, we may need to release
2134 * if if the inum of the entry turns out to be less.
2136 if (e_inum > dp->i_ino) {
2138 * We are already in the right order, so just
2139 * lock on the inode of the entry.
2140 * We need to use nowait if dp is in the AIL.
2143 lp = (xfs_log_item_t *)dp->i_itemp;
2144 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2145 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2152 * Unlock dp and try again.
2153 * xfs_iunlock will try to push the tail
2154 * if the inode is in the AIL.
2157 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2159 if ((attempts % 5) == 0) {
2160 delay(1); /* Don't just spin the CPU */
2162 xfs_rm_lock_delays++;
2168 xfs_ilock(ip, XFS_ILOCK_EXCL);
2170 } else if (e_inum < dp->i_ino) {
2171 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2175 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2177 /* else e_inum == dp->i_ino */
2178 /* This can happen if we're asked to lock /x/..
2179 * the entry is "..", which is also the parent directory.
2187 int xfs_small_retries;
2188 int xfs_middle_retries;
2189 int xfs_lots_retries;
2190 int xfs_lock_delays;
2194 * The following routine will lock n inodes in exclusive mode.
2195 * We assume the caller calls us with the inodes in i_ino order.
2197 * We need to detect deadlock where an inode that we lock
2198 * is in the AIL and we start waiting for another inode that is locked
2199 * by a thread in a long running transaction (such as truncate). This can
2200 * result in deadlock since the long running trans might need to wait
2201 * for the inode we just locked in order to push the tail and free space
2211 int attempts = 0, i, j, try_lock;
2214 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2225 for (; i < inodes; i++) {
2228 if (i && (ips[i] == ips[i-1])) /* Already locked */
2232 * If try_lock is not set yet, make sure all locked inodes
2233 * are not in the AIL.
2234 * If any are, set try_lock to be used later.
2238 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2239 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2240 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2247 * If any of the previous locks we have locked is in the AIL,
2248 * we must TRY to get the second and subsequent locks. If
2249 * we can't get any, we must release all we have
2254 /* try_lock must be 0 if i is 0. */
2256 * try_lock means we have an inode locked
2257 * that is in the AIL.
2260 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2264 * Unlock all previous guys and try again.
2265 * xfs_iunlock will try to push the tail
2266 * if the inode is in the AIL.
2269 for(j = i - 1; j >= 0; j--) {
2272 * Check to see if we've already
2273 * unlocked this one.
2274 * Not the first one going back,
2275 * and the inode ptr is the same.
2277 if ((j != (i - 1)) && ips[j] ==
2281 xfs_iunlock(ips[j], lock_mode);
2284 if ((attempts % 5) == 0) {
2285 delay(1); /* Don't just spin the CPU */
2295 xfs_ilock(ips[i], lock_mode);
2301 if (attempts < 5) xfs_small_retries++;
2302 else if (attempts < 100) xfs_middle_retries++;
2303 else xfs_lots_retries++;
2311 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2312 int remove_which_error_return = 0;
2314 #define REMOVE_DEBUG_TRACE(x)
2315 #endif /* ! DEBUG */
2324 bhv_desc_t *dir_bdp,
2329 char *name = VNAME(dentry);
2330 xfs_inode_t *dp, *ip;
2331 xfs_trans_t *tp = NULL;
2334 xfs_bmap_free_t free_list;
2335 xfs_fsblock_t first_block;
2343 dir_vp = BHV_TO_VNODE(dir_bdp);
2344 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2346 dp = XFS_BHVTOI(dir_bdp);
2349 if (XFS_FORCED_SHUTDOWN(mp))
2350 return XFS_ERROR(EIO);
2352 namelen = VNAMELEN(dentry);
2354 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2355 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2356 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2357 name, NULL, 0, 0, 0);
2362 /* From this point on, return through std_return */
2366 * We need to get a reference to ip before we get our log
2367 * reservation. The reason for this is that we cannot call
2368 * xfs_iget for an inode for which we do not have a reference
2369 * once we've acquired a log reservation. This is because the
2370 * inode we are trying to get might be in xfs_inactive going
2371 * for a log reservation. Since we'll have to wait for the
2372 * inactive code to complete before returning from xfs_iget,
2373 * we need to make sure that we don't have log space reserved
2374 * when we call xfs_iget. Instead we get an unlocked referece
2375 * to the inode before getting our log reservation.
2377 error = xfs_get_dir_entry(dentry, &ip);
2379 REMOVE_DEBUG_TRACE(__LINE__);
2383 dm_di_mode = ip->i_d.di_mode;
2385 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2389 error = XFS_QM_DQATTACH(mp, dp, 0);
2390 if (!error && dp != ip)
2391 error = XFS_QM_DQATTACH(mp, ip, 0);
2393 REMOVE_DEBUG_TRACE(__LINE__);
2398 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2399 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2401 * We try to get the real space reservation first,
2402 * allowing for directory btree deletion(s) implying
2403 * possible bmap insert(s). If we can't get the space
2404 * reservation then we use 0 instead, and avoid the bmap
2405 * btree insert(s) in the directory code by, if the bmap
2406 * insert tries to happen, instead trimming the LAST
2407 * block from the directory.
2409 resblks = XFS_REMOVE_SPACE_RES(mp);
2410 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2411 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2412 if (error == ENOSPC) {
2414 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2415 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2418 ASSERT(error != ENOSPC);
2419 REMOVE_DEBUG_TRACE(__LINE__);
2420 xfs_trans_cancel(tp, 0);
2425 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2427 REMOVE_DEBUG_TRACE(__LINE__);
2428 xfs_trans_cancel(tp, cancel_flags);
2434 * At this point, we've gotten both the directory and the entry
2437 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2440 * Increment vnode ref count only in this case since
2441 * there's an extra vnode reference in the case where
2445 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2449 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2451 XFS_BMAP_INIT(&free_list, &first_block);
2452 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2453 &first_block, &free_list, 0);
2455 ASSERT(error != ENOENT);
2456 REMOVE_DEBUG_TRACE(__LINE__);
2459 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2462 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2464 error = xfs_droplink(tp, ip);
2466 REMOVE_DEBUG_TRACE(__LINE__);
2470 /* Determine if this is the last link while
2471 * we are in the transaction.
2473 link_zero = (ip)->i_d.di_nlink==0;
2476 * Take an extra ref on the inode so that it doesn't
2477 * go to xfs_inactive() from within the commit.
2482 * If this is a synchronous mount, make sure that the
2483 * remove transaction goes to disk before returning to
2486 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2487 xfs_trans_set_sync(tp);
2490 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2492 REMOVE_DEBUG_TRACE(__LINE__);
2496 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2503 * Before we drop our extra reference to the inode, purge it
2504 * from the refcache if it is there. By waiting until afterwards
2505 * to do the IRELE, we ensure that we won't go inactive in the
2506 * xfs_refcache_purge_ip routine (although that would be OK).
2508 xfs_refcache_purge_ip(ip);
2510 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2513 * Let interposed file systems know about removed links.
2515 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2519 /* Fall through to std_return with error = 0 */
2521 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2522 DM_EVENT_POSTREMOVE)) {
2523 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2524 dir_vp, DM_RIGHT_NULL,
2525 NULL, DM_RIGHT_NULL,
2526 name, NULL, dm_di_mode, error, 0);
2531 xfs_bmap_cancel(&free_list);
2532 cancel_flags |= XFS_TRANS_ABORT;
2533 xfs_trans_cancel(tp, cancel_flags);
2538 * In this case make sure to not release the inode until after
2539 * the current transaction is aborted. Releasing it beforehand
2540 * can cause us to go to xfs_inactive and start a recursive
2541 * transaction which can easily deadlock with the current one.
2543 xfs_bmap_cancel(&free_list);
2544 cancel_flags |= XFS_TRANS_ABORT;
2545 xfs_trans_cancel(tp, cancel_flags);
2548 * Before we drop our extra reference to the inode, purge it
2549 * from the refcache if it is there. By waiting until afterwards
2550 * to do the IRELE, we ensure that we won't go inactive in the
2551 * xfs_refcache_purge_ip routine (although that would be OK).
2553 xfs_refcache_purge_ip(ip);
2567 bhv_desc_t *target_dir_bdp,
2572 xfs_inode_t *tdp, *sip;
2575 xfs_inode_t *ips[2];
2577 xfs_bmap_free_t free_list;
2578 xfs_fsblock_t first_block;
2581 vnode_t *target_dir_vp;
2582 bhv_desc_t *src_bdp;
2584 char *target_name = VNAME(dentry);
2587 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2588 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2589 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2591 target_namelen = VNAMELEN(dentry);
2592 if (VN_ISDIR(src_vp))
2593 return XFS_ERROR(EPERM);
2595 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2596 sip = XFS_BHVTOI(src_bdp);
2597 tdp = XFS_BHVTOI(target_dir_bdp);
2599 if (XFS_FORCED_SHUTDOWN(mp))
2600 return XFS_ERROR(EIO);
2602 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2603 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2604 target_dir_vp, DM_RIGHT_NULL,
2605 src_vp, DM_RIGHT_NULL,
2606 target_name, NULL, 0, 0, 0);
2611 /* Return through std_return after this point. */
2613 error = XFS_QM_DQATTACH(mp, sip, 0);
2614 if (!error && sip != tdp)
2615 error = XFS_QM_DQATTACH(mp, tdp, 0);
2619 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2620 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2621 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2622 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2623 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2624 if (error == ENOSPC) {
2626 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2627 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2634 if (sip->i_ino < tdp->i_ino) {
2642 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2645 * Increment vnode ref counts since xfs_trans_commit &
2646 * xfs_trans_cancel will both unlock the inodes and
2647 * decrement the associated ref counts.
2650 VN_HOLD(target_dir_vp);
2651 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2652 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2655 * If the source has too many links, we can't make any more to it.
2657 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2658 error = XFS_ERROR(EMLINK);
2663 * If we are using project inheritance, we only allow hard link
2664 * creation in our tree when the project IDs are the same; else
2665 * the tree quota mechanism could be circumvented.
2667 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2668 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2669 error = XFS_ERROR(EPERM);
2674 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2678 XFS_BMAP_INIT(&free_list, &first_block);
2680 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2681 sip->i_ino, &first_block, &free_list,
2685 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2687 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2689 error = xfs_bumplink(tp, sip);
2695 * If this is a synchronous mount, make sure that the
2696 * link transaction goes to disk before returning to
2699 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2700 xfs_trans_set_sync(tp);
2703 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2705 xfs_bmap_cancel(&free_list);
2709 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2714 /* Fall through to std_return with error = 0. */
2716 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2717 DM_EVENT_POSTLINK)) {
2718 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2719 target_dir_vp, DM_RIGHT_NULL,
2720 src_vp, DM_RIGHT_NULL,
2721 target_name, NULL, 0, error, 0);
2726 cancel_flags |= XFS_TRANS_ABORT;
2729 xfs_trans_cancel(tp, cancel_flags);
2739 bhv_desc_t *dir_bdp,
2745 char *dir_name = VNAME(dentry);
2747 xfs_inode_t *cdp; /* inode of created dir */
2748 vnode_t *cvp; /* vnode of created dir */
2754 xfs_bmap_free_t free_list;
2755 xfs_fsblock_t first_block;
2757 boolean_t dp_joined_to_trans;
2758 boolean_t created = B_FALSE;
2759 int dm_event_sent = 0;
2761 struct xfs_dquot *udqp, *gdqp;
2766 dir_vp = BHV_TO_VNODE(dir_bdp);
2767 dp = XFS_BHVTOI(dir_bdp);
2770 if (XFS_FORCED_SHUTDOWN(mp))
2771 return XFS_ERROR(EIO);
2773 dir_namelen = VNAMELEN(dentry);
2776 dp_joined_to_trans = B_FALSE;
2777 dm_di_mode = vap->va_mode;
2779 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2780 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2781 dir_vp, DM_RIGHT_NULL, NULL,
2782 DM_RIGHT_NULL, dir_name, NULL,
2789 /* Return through std_return after this point. */
2791 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2795 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2796 prid = dp->i_d.di_projid;
2797 else if (vap->va_mask & XFS_AT_PROJID)
2798 prid = (xfs_prid_t)vap->va_projid;
2800 prid = (xfs_prid_t)dfltprid;
2803 * Make sure that we have allocated dquot(s) on disk.
2805 error = XFS_QM_DQVOPALLOC(mp, dp,
2806 current_fsuid(credp), current_fsgid(credp), prid,
2807 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2811 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2812 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2813 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2814 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2815 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2816 if (error == ENOSPC) {
2818 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2819 XFS_TRANS_PERM_LOG_RES,
2820 XFS_MKDIR_LOG_COUNT);
2828 xfs_ilock(dp, XFS_ILOCK_EXCL);
2831 * Check for directory link count overflow.
2833 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2834 error = XFS_ERROR(EMLINK);
2839 * Reserve disk quota and the inode.
2841 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2846 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2849 * create the directory inode.
2851 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2852 0, credp, prid, resblks > 0,
2855 if (error == ENOSPC)
2862 * Now we add the directory inode to the transaction.
2863 * We waited until now since xfs_dir_ialloc might start
2864 * a new transaction. Had we joined the transaction
2865 * earlier, the locks might have gotten released.
2868 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2869 dp_joined_to_trans = B_TRUE;
2871 XFS_BMAP_INIT(&free_list, &first_block);
2873 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2874 cdp->i_ino, &first_block, &free_list,
2875 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2877 ASSERT(error != ENOSPC);
2880 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2883 * Bump the in memory version number of the parent directory
2884 * so that other processes accessing it will recognize that
2885 * the directory has changed.
2889 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2895 error = xfs_bumplink(tp, dp);
2900 cvp = XFS_ITOV(cdp);
2908 * Attach the dquots to the new inode and modify the icount incore.
2910 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2913 * If this is a synchronous mount, make sure that the
2914 * mkdir transaction goes to disk before returning to
2917 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2918 xfs_trans_set_sync(tp);
2921 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2927 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2928 XFS_QM_DQRELE(mp, udqp);
2929 XFS_QM_DQRELE(mp, gdqp);
2934 /* Fall through to std_return with error = 0 or errno from
2935 * xfs_trans_commit. */
2938 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2939 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2940 DM_EVENT_POSTCREATE)) {
2941 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2942 dir_vp, DM_RIGHT_NULL,
2943 created ? XFS_ITOV(cdp):NULL,
2946 dm_di_mode, error, 0);
2952 xfs_bmap_cancel(&free_list);
2954 cancel_flags |= XFS_TRANS_ABORT;
2956 xfs_trans_cancel(tp, cancel_flags);
2957 XFS_QM_DQRELE(mp, udqp);
2958 XFS_QM_DQRELE(mp, gdqp);
2960 if (!dp_joined_to_trans && (dp != NULL)) {
2961 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2974 bhv_desc_t *dir_bdp,
2978 char *name = VNAME(dentry);
2980 xfs_inode_t *cdp; /* child directory */
2984 xfs_bmap_free_t free_list;
2985 xfs_fsblock_t first_block;
2994 dir_vp = BHV_TO_VNODE(dir_bdp);
2995 dp = XFS_BHVTOI(dir_bdp);
2998 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3000 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3001 return XFS_ERROR(EIO);
3002 namelen = VNAMELEN(dentry);
3004 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3005 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3006 dir_vp, DM_RIGHT_NULL,
3007 NULL, DM_RIGHT_NULL,
3008 name, NULL, 0, 0, 0);
3010 return XFS_ERROR(error);
3013 /* Return through std_return after this point. */
3018 * We need to get a reference to cdp before we get our log
3019 * reservation. The reason for this is that we cannot call
3020 * xfs_iget for an inode for which we do not have a reference
3021 * once we've acquired a log reservation. This is because the
3022 * inode we are trying to get might be in xfs_inactive going
3023 * for a log reservation. Since we'll have to wait for the
3024 * inactive code to complete before returning from xfs_iget,
3025 * we need to make sure that we don't have log space reserved
3026 * when we call xfs_iget. Instead we get an unlocked referece
3027 * to the inode before getting our log reservation.
3029 error = xfs_get_dir_entry(dentry, &cdp);
3031 REMOVE_DEBUG_TRACE(__LINE__);
3035 dm_di_mode = cdp->i_d.di_mode;
3038 * Get the dquots for the inodes.
3040 error = XFS_QM_DQATTACH(mp, dp, 0);
3041 if (!error && dp != cdp)
3042 error = XFS_QM_DQATTACH(mp, cdp, 0);
3045 REMOVE_DEBUG_TRACE(__LINE__);
3049 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3050 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3052 * We try to get the real space reservation first,
3053 * allowing for directory btree deletion(s) implying
3054 * possible bmap insert(s). If we can't get the space
3055 * reservation then we use 0 instead, and avoid the bmap
3056 * btree insert(s) in the directory code by, if the bmap
3057 * insert tries to happen, instead trimming the LAST
3058 * block from the directory.
3060 resblks = XFS_REMOVE_SPACE_RES(mp);
3061 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3062 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3063 if (error == ENOSPC) {
3065 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3066 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3069 ASSERT(error != ENOSPC);
3074 XFS_BMAP_INIT(&free_list, &first_block);
3077 * Now lock the child directory inode and the parent directory
3078 * inode in the proper order. This will take care of validating
3079 * that the directory entry for the child directory inode has
3080 * not changed while we were obtaining a log reservation.
3082 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3084 xfs_trans_cancel(tp, cancel_flags);
3089 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3092 * Only increment the parent directory vnode count if
3093 * we didn't bump it in looking up cdp. The only time
3094 * we don't bump it is when we're looking up ".".
3100 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3102 ASSERT(cdp->i_d.di_nlink >= 2);
3103 if (cdp->i_d.di_nlink != 2) {
3104 error = XFS_ERROR(ENOTEMPTY);
3107 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3108 error = XFS_ERROR(ENOTEMPTY);
3112 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3113 &first_block, &free_list, resblks);
3118 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3121 * Bump the in memory generation count on the parent
3122 * directory so that other can know that it has changed.
3127 * Drop the link from cdp's "..".
3129 error = xfs_droplink(tp, dp);
3135 * Drop the link from dp to cdp.
3137 error = xfs_droplink(tp, cdp);
3143 * Drop the "." link from cdp to self.
3145 error = xfs_droplink(tp, cdp);
3150 /* Determine these before committing transaction */
3151 last_cdp_link = (cdp)->i_d.di_nlink==0;
3154 * Take an extra ref on the child vnode so that it
3155 * does not go to xfs_inactive() from within the commit.
3160 * If this is a synchronous mount, make sure that the
3161 * rmdir transaction goes to disk before returning to
3164 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3165 xfs_trans_set_sync(tp);
3168 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3170 xfs_bmap_cancel(&free_list);
3171 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3177 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3185 * Let interposed file systems know about removed links.
3187 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3191 /* Fall through to std_return with error = 0 or the errno
3192 * from xfs_trans_commit. */
3194 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3195 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3196 dir_vp, DM_RIGHT_NULL,
3197 NULL, DM_RIGHT_NULL,
3198 name, NULL, dm_di_mode,
3204 xfs_bmap_cancel(&free_list);
3205 cancel_flags |= XFS_TRANS_ABORT;
3207 xfs_trans_cancel(tp, cancel_flags);
3215 * Read dp's entries starting at uiop->uio_offset and translate them into
3216 * bufsize bytes worth of struct dirents starting at bufbase.
3220 bhv_desc_t *dir_bdp,
3226 xfs_trans_t *tp = NULL;
3229 xfs_off_t start_offset;
3231 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3232 (inst_t *)__return_address);
3233 dp = XFS_BHVTOI(dir_bdp);
3235 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3236 return XFS_ERROR(EIO);
3239 lock_mode = xfs_ilock_map_shared(dp);
3240 start_offset = uiop->uio_offset;
3241 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3242 if (start_offset != uiop->uio_offset) {
3243 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3245 xfs_iunlock_map_shared(dp, lock_mode);
3256 bhv_desc_t *dir_bdp,
3269 xfs_bmap_free_t free_list;
3270 xfs_fsblock_t first_block;
3271 boolean_t dp_joined_to_trans;
3275 xfs_fileoff_t first_fsb;
3276 xfs_filblks_t fs_blocks;
3278 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3285 struct xfs_dquot *udqp, *gdqp;
3287 char *link_name = VNAME(dentry);
3291 dir_vp = BHV_TO_VNODE(dir_bdp);
3292 dp = XFS_BHVTOI(dir_bdp);
3293 dp_joined_to_trans = B_FALSE;
3298 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3302 if (XFS_FORCED_SHUTDOWN(mp))
3303 return XFS_ERROR(EIO);
3305 link_namelen = VNAMELEN(dentry);
3308 * Check component lengths of the target path name.
3310 pathlen = strlen(target_path);
3311 if (pathlen >= MAXPATHLEN) /* total string too long */
3312 return XFS_ERROR(ENAMETOOLONG);
3313 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3317 for(total = 0, path = target_path; total < pathlen;) {
3321 while(*path == '/') {
3327 * Count up to the next slash or end of path.
3328 * Error out if the component is bigger than MAXNAMELEN.
3330 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3331 if (++len >= MAXNAMELEN) {
3332 error = ENAMETOOLONG;
3339 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3340 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3341 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3342 link_name, target_path, 0, 0, 0);
3347 /* Return through std_return after this point. */
3350 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3351 prid = dp->i_d.di_projid;
3352 else if (vap->va_mask & XFS_AT_PROJID)
3353 prid = (xfs_prid_t)vap->va_projid;
3355 prid = (xfs_prid_t)dfltprid;
3358 * Make sure that we have allocated dquot(s) on disk.
3360 error = XFS_QM_DQVOPALLOC(mp, dp,
3361 current_fsuid(credp), current_fsgid(credp), prid,
3362 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3366 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3367 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3369 * The symlink will fit into the inode data fork?
3370 * There can't be any attributes so we get the whole variable part.
3372 if (pathlen <= XFS_LITINO(mp))
3375 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3376 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3377 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3378 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3379 if (error == ENOSPC && fs_blocks == 0) {
3381 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3382 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3390 xfs_ilock(dp, XFS_ILOCK_EXCL);
3393 * Check whether the directory allows new symlinks or not.
3395 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3396 error = XFS_ERROR(EPERM);
3401 * Reserve disk quota : blocks and inode.
3403 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3408 * Check for ability to enter directory entry, if no space reserved.
3411 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3414 * Initialize the bmap freelist prior to calling either
3415 * bmapi or the directory create code.
3417 XFS_BMAP_INIT(&free_list, &first_block);
3420 * Allocate an inode for the symlink.
3422 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3423 1, 0, credp, prid, resblks > 0, &ip, NULL);
3425 if (error == ENOSPC)
3432 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3433 dp_joined_to_trans = B_TRUE;
3436 * Also attach the dquot(s) to it, if applicable.
3438 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3441 resblks -= XFS_IALLOC_SPACE_RES(mp);
3443 * If the symlink will fit into the inode, write it inline.
3445 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3446 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3447 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3448 ip->i_d.di_size = pathlen;
3451 * The inode was initially created in extent format.
3453 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3454 ip->i_df.if_flags |= XFS_IFINLINE;
3456 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3457 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3461 nmaps = SYMLINK_MAPS;
3463 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3464 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3465 &first_block, resblks, mval, &nmaps,
3472 resblks -= fs_blocks;
3473 ip->i_d.di_size = pathlen;
3474 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3476 cur_chunk = target_path;
3477 for (n = 0; n < nmaps; n++) {
3478 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3479 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3480 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3481 BTOBB(byte_cnt), 0);
3482 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3483 if (pathlen < byte_cnt) {
3486 pathlen -= byte_cnt;
3488 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3489 cur_chunk += byte_cnt;
3491 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3496 * Create the directory entry for the symlink.
3498 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3499 ip->i_ino, &first_block, &free_list, resblks);
3503 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3504 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3507 * Bump the in memory version number of the parent directory
3508 * so that other processes accessing it will recognize that
3509 * the directory has changed.
3514 * If this is a synchronous mount, make sure that the
3515 * symlink transaction goes to disk before returning to
3518 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3519 xfs_trans_set_sync(tp);
3523 * xfs_trans_commit normally decrements the vnode ref count
3524 * when it unlocks the inode. Since we want to return the
3525 * vnode to the caller, we bump the vnode ref count now.
3529 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3533 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3534 XFS_QM_DQRELE(mp, udqp);
3535 XFS_QM_DQRELE(mp, gdqp);
3537 /* Fall through to std_return with error = 0 or errno from
3538 * xfs_trans_commit */
3540 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3541 DM_EVENT_POSTSYMLINK)) {
3542 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3543 dir_vp, DM_RIGHT_NULL,
3544 error ? NULL : XFS_ITOV(ip),
3545 DM_RIGHT_NULL, link_name, target_path,
3561 xfs_bmap_cancel(&free_list);
3562 cancel_flags |= XFS_TRANS_ABORT;
3564 xfs_trans_cancel(tp, cancel_flags);
3565 XFS_QM_DQRELE(mp, udqp);
3566 XFS_QM_DQRELE(mp, gdqp);
3568 if (!dp_joined_to_trans && (dp != NULL)) {
3569 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3579 * A fid routine that takes a pointer to a previously allocated
3580 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3590 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3591 (inst_t *)__return_address);
3592 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3594 xfid = (xfs_fid2_t *)fidp;
3595 ip = XFS_BHVTOI(bdp);
3596 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3599 * use memcpy because the inode is a long long and there's no
3600 * assurance that xfid->fid_ino is properly aligned.
3602 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3603 xfid->fid_gen = ip->i_d.di_gen;
3620 vp = BHV_TO_VNODE(bdp);
3623 ip = XFS_BHVTOI(bdp);
3624 if (locktype == VRWLOCK_WRITE) {
3625 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3626 } else if (locktype == VRWLOCK_TRY_READ) {
3627 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3628 } else if (locktype == VRWLOCK_TRY_WRITE) {
3629 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3631 ASSERT((locktype == VRWLOCK_READ) ||
3632 (locktype == VRWLOCK_WRITE_DIRECT));
3633 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3651 vp = BHV_TO_VNODE(bdp);
3654 ip = XFS_BHVTOI(bdp);
3655 if (locktype == VRWLOCK_WRITE) {
3657 * In the write case, we may have added a new entry to
3658 * the reference cache. This might store a pointer to
3659 * an inode to be released in this inode. If it is there,
3660 * clear the pointer and release the inode after unlocking
3663 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3665 ASSERT((locktype == VRWLOCK_READ) ||
3666 (locktype == VRWLOCK_WRITE_DIRECT));
3667 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3679 xfs_inode_log_item_t *iip;
3682 ip = XFS_BHVTOI(bdp);
3686 if (XFS_FORCED_SHUTDOWN(mp))
3687 return XFS_ERROR(EIO);
3690 * Bypass inodes which have already been cleaned by
3691 * the inode flush clustering code inside xfs_iflush
3693 if ((ip->i_update_core == 0) &&
3694 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3697 if (flags & FLUSH_LOG) {
3698 if (iip && iip->ili_last_lsn) {
3699 xlog_t *log = mp->m_log;
3701 int s, log_flags = XFS_LOG_FORCE;
3703 s = GRANT_LOCK(log);
3704 sync_lsn = log->l_last_sync_lsn;
3705 GRANT_UNLOCK(log, s);
3707 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3710 if (flags & FLUSH_SYNC)
3711 log_flags |= XFS_LOG_SYNC;
3712 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3717 * We make this non-blocking if the inode is contended,
3718 * return EAGAIN to indicate to the caller that they
3719 * did not succeed. This prevents the flush path from
3720 * blocking on inodes inside another operation right
3721 * now, they get caught later by xfs_sync.
3723 if (flags & FLUSH_INODE) {
3726 if (xfs_ipincount(ip))
3729 if (flags & FLUSH_SYNC) {
3730 xfs_ilock(ip, XFS_ILOCK_SHARED);
3732 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3733 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3734 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3741 if (flags & FLUSH_SYNC)
3742 flush_flags = XFS_IFLUSH_SYNC;
3744 flush_flags = XFS_IFLUSH_ASYNC;
3746 error = xfs_iflush(ip, flush_flags);
3747 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3766 if (!capable(CAP_SYS_ADMIN))
3767 return XFS_ERROR(EPERM);
3769 ip = XFS_BHVTOI(bdp);
3772 if (XFS_FORCED_SHUTDOWN(mp))
3773 return XFS_ERROR(EIO);
3775 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3776 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3778 xfs_trans_cancel(tp, 0);
3781 xfs_ilock(ip, XFS_ILOCK_EXCL);
3782 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3784 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3785 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3787 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3789 error = xfs_trans_commit(tp, 0, NULL);
3805 vp = BHV_TO_VNODE(bdp);
3806 ip = XFS_BHVTOI(bdp);
3808 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3810 ASSERT(!VN_MAPPED(vp));
3812 /* bad inode, get out here ASAP */
3820 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3822 /* If we have nothing to flush with this inode then complete the
3823 * teardown now, otherwise break the link between the xfs inode
3824 * and the linux inode and clean up the xfs inode later. This
3825 * avoids flushing the inode to disk during the delete operation
3828 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3829 xfs_ilock(ip, XFS_ILOCK_EXCL);
3831 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3833 xfs_mount_t *mp = ip->i_mount;
3835 /* Protect sync from us */
3836 XFS_MOUNT_ILOCK(mp);
3837 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3838 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3839 ip->i_flags |= XFS_IRECLAIMABLE;
3840 XFS_MOUNT_IUNLOCK(mp);
3851 xfs_ihash_t *ih = ip->i_hash;
3852 vnode_t *vp = XFS_ITOV_NULL(ip);
3855 if (vp && VN_BAD(vp))
3858 /* The hash lock here protects a thread in xfs_iget_core from
3859 * racing with us on linking the inode back with a vnode.
3860 * Once we have the XFS_IRECLAIM flag set it will not touch
3863 write_lock(&ih->ih_lock);
3864 if ((ip->i_flags & XFS_IRECLAIM) ||
3865 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3866 write_unlock(&ih->ih_lock);
3869 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3873 ip->i_flags |= XFS_IRECLAIM;
3874 write_unlock(&ih->ih_lock);
3877 * If the inode is still dirty, then flush it out. If the inode
3878 * is not in the AIL, then it will be OK to flush it delwri as
3879 * long as xfs_iflush() does not keep any references to the inode.
3880 * We leave that decision up to xfs_iflush() since it has the
3881 * knowledge of whether it's OK to simply do a delwri flush of
3882 * the inode or whether we need to wait until the inode is
3883 * pulled from the AIL.
3884 * We get the flush lock regardless, though, just to make sure
3885 * we don't free it while it is being flushed.
3887 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3889 xfs_ilock(ip, XFS_ILOCK_EXCL);
3893 if (ip->i_update_core ||
3894 ((ip->i_itemp != NULL) &&
3895 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3896 error = xfs_iflush(ip, sync_mode);
3898 * If we hit an error, typically because of filesystem
3899 * shutdown, we don't need to let vn_reclaim to know
3900 * because we're gonna reclaim the inode anyway.
3903 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3906 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3909 ASSERT(ip->i_update_core == 0);
3910 ASSERT(ip->i_itemp == NULL ||
3911 ip->i_itemp->ili_format.ilf_fields == 0);
3912 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3913 } else if (locked) {
3915 * We are not interested in doing an iflush if we're
3916 * in the process of shutting down the filesystem forcibly.
3917 * So, just reclaim the inode.
3920 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3929 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3932 xfs_inode_t *ip, *n;
3937 XFS_MOUNT_ILOCK(mp);
3938 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3940 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3942 if (xfs_ipincount(ip) ||
3943 !xfs_iflock_nowait(ip)) {
3944 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3948 XFS_MOUNT_IUNLOCK(mp);
3949 if (xfs_finish_reclaim(ip, noblock,
3950 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3959 XFS_MOUNT_IUNLOCK(mp);
3964 * xfs_alloc_file_space()
3965 * This routine allocates disk space for the given file.
3967 * If alloc_type == 0, this request is for an ALLOCSP type
3968 * request which will change the file size. In this case, no
3969 * DMAPI event will be generated by the call. A TRUNCATE event
3970 * will be generated later by xfs_setattr.
3972 * If alloc_type != 0, this request is for a RESVSP type
3973 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3974 * lower block boundary byte address is less than the file's
3983 xfs_alloc_file_space(
3990 xfs_mount_t *mp = ip->i_mount;
3992 xfs_filblks_t allocated_fsb;
3993 xfs_filblks_t allocatesize_fsb;
3994 xfs_extlen_t extsz, temp;
3995 xfs_fileoff_t startoffset_fsb;
3996 xfs_fsblock_t firstfsb;
4002 xfs_bmbt_irec_t imaps[1], *imapp;
4003 xfs_bmap_free_t free_list;
4004 uint qblocks, resblks, resrtextents;
4008 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4010 if (XFS_FORCED_SHUTDOWN(mp))
4011 return XFS_ERROR(EIO);
4013 rt = XFS_IS_REALTIME_INODE(ip);
4015 if (!(extsz = ip->i_d.di_extsize))
4016 extsz = mp->m_sb.sb_rextsize;
4018 extsz = ip->i_d.di_extsize;
4021 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4025 return XFS_ERROR(EINVAL);
4031 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4032 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4033 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4035 /* Generate a DMAPI event if needed. */
4036 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4037 (attr_flags&ATTR_DMI) == 0 &&
4038 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4039 xfs_off_t end_dmi_offset;
4041 end_dmi_offset = offset+len;
4042 if (end_dmi_offset > ip->i_d.di_size)
4043 end_dmi_offset = ip->i_d.di_size;
4044 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4045 offset, end_dmi_offset - offset,
4052 * Allocate file space until done or until there is an error
4055 while (allocatesize_fsb && !error) {
4059 * Determine space reservations for data/realtime.
4061 if (unlikely(extsz)) {
4062 s = startoffset_fsb;
4065 e = startoffset_fsb + allocatesize_fsb;
4066 if ((temp = do_mod(startoffset_fsb, extsz)))
4068 if ((temp = do_mod(e, extsz)))
4072 e = allocatesize_fsb;
4076 resrtextents = qblocks = (uint)(e - s);
4077 resrtextents /= mp->m_sb.sb_rextsize;
4078 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4079 quota_flag = XFS_QMOPT_RES_RTBLKS;
4082 resblks = qblocks = \
4083 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4084 quota_flag = XFS_QMOPT_RES_REGBLKS;
4088 * Allocate and setup the transaction.
4090 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4091 error = xfs_trans_reserve(tp, resblks,
4092 XFS_WRITE_LOG_RES(mp), resrtextents,
4093 XFS_TRANS_PERM_LOG_RES,
4094 XFS_WRITE_LOG_COUNT);
4096 * Check for running out of space
4100 * Free the transaction structure.
4102 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4103 xfs_trans_cancel(tp, 0);
4106 xfs_ilock(ip, XFS_ILOCK_EXCL);
4107 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4108 qblocks, 0, quota_flag);
4112 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4113 xfs_trans_ihold(tp, ip);
4116 * Issue the xfs_bmapi() call to allocate the blocks
4118 XFS_BMAP_INIT(&free_list, &firstfsb);
4119 error = xfs_bmapi(tp, ip, startoffset_fsb,
4120 allocatesize_fsb, bmapi_flag,
4121 &firstfsb, 0, imapp, &nimaps,
4128 * Complete the transaction
4130 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4135 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4136 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4141 allocated_fsb = imapp->br_blockcount;
4144 error = XFS_ERROR(ENOSPC);
4148 startoffset_fsb += allocated_fsb;
4149 allocatesize_fsb -= allocated_fsb;
4152 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4153 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4155 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4156 XFS_ITOV(ip), DM_RIGHT_NULL,
4157 XFS_ITOV(ip), DM_RIGHT_NULL,
4158 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4160 goto retry; /* Maybe DMAPI app. has made space */
4161 /* else fall through with error from XFS_SEND_DATA */
4166 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4167 xfs_bmap_cancel(&free_list);
4168 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4170 error1: /* Just cancel transaction */
4171 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4172 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4173 goto dmapi_enospc_check;
4177 * Zero file bytes between startoff and endoff inclusive.
4178 * The iolock is held exclusive and no blocks are buffered.
4181 xfs_zero_remaining_bytes(
4186 xfs_bmbt_irec_t imap;
4187 xfs_fileoff_t offset_fsb;
4188 xfs_off_t lastoffset;
4191 xfs_mount_t *mp = ip->i_mount;
4195 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4196 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4197 mp->m_rtdev_targp : mp->m_ddev_targp);
4199 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4200 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4202 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4204 if (error || nimap < 1)
4206 ASSERT(imap.br_blockcount >= 1);
4207 ASSERT(imap.br_startoff == offset_fsb);
4208 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4209 if (lastoffset > endoff)
4210 lastoffset = endoff;
4211 if (imap.br_startblock == HOLESTARTBLOCK)
4213 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4214 if (imap.br_state == XFS_EXT_UNWRITTEN)
4217 XFS_BUF_UNWRITE(bp);
4219 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4221 if ((error = xfs_iowait(bp))) {
4222 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4223 mp, bp, XFS_BUF_ADDR(bp));
4226 memset(XFS_BUF_PTR(bp) +
4227 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4228 0, lastoffset - offset + 1);
4233 if ((error = xfs_iowait(bp))) {
4234 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4235 mp, bp, XFS_BUF_ADDR(bp));
4244 * xfs_free_file_space()
4245 * This routine frees disk space for the given file.
4247 * This routine is only called by xfs_change_file_space
4248 * for an UNRESVSP type call.
4256 xfs_free_file_space(
4265 xfs_off_t end_dmi_offset;
4266 xfs_fileoff_t endoffset_fsb;
4268 xfs_fsblock_t firstfsb;
4269 xfs_bmap_free_t free_list;
4271 xfs_bmbt_irec_t imap;
4279 xfs_fileoff_t startoffset_fsb;
4281 int need_iolock = 1;
4286 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4288 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4292 if (len <= 0) /* if nothing being freed */
4294 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4295 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4296 end_dmi_offset = offset + len;
4297 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4299 if (offset < ip->i_d.di_size &&
4300 (attr_flags & ATTR_DMI) == 0 &&
4301 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4302 if (end_dmi_offset > ip->i_d.di_size)
4303 end_dmi_offset = ip->i_d.di_size;
4304 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4305 offset, end_dmi_offset - offset,
4306 AT_DELAY_FLAG(attr_flags), NULL);
4311 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4312 if (attr_flags & ATTR_NOLOCK)
4315 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4317 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4319 ilen = len + (offset & (rounding - 1));
4320 ioffset = offset & ~(rounding - 1);
4321 if (ilen & (rounding - 1))
4322 ilen = (ilen + rounding) & ~(rounding - 1);
4324 if (VN_CACHED(vp) != 0) {
4325 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4326 ctooff(offtoct(ioffset)), -1);
4327 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4328 -1, FI_REMAPF_LOCKED);
4332 * Need to zero the stuff we're not freeing, on disk.
4333 * If its a realtime file & can't use unwritten extents then we
4334 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4335 * will take care of it for us.
4337 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4339 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4340 &imap, &nimap, NULL);
4342 goto out_unlock_iolock;
4343 ASSERT(nimap == 0 || nimap == 1);
4344 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4347 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4348 block = imap.br_startblock;
4349 mod = do_div(block, mp->m_sb.sb_rextsize);
4351 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4354 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4355 &imap, &nimap, NULL);
4357 goto out_unlock_iolock;
4358 ASSERT(nimap == 0 || nimap == 1);
4359 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4360 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4362 if (mod && (mod != mp->m_sb.sb_rextsize))
4363 endoffset_fsb -= mod;
4366 if ((done = (endoffset_fsb <= startoffset_fsb)))
4368 * One contiguous piece to clear
4370 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4373 * Some full blocks, possibly two pieces to clear
4375 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4376 error = xfs_zero_remaining_bytes(ip, offset,
4377 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4379 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4380 error = xfs_zero_remaining_bytes(ip,
4381 XFS_FSB_TO_B(mp, endoffset_fsb),
4386 * free file space until done or until there is an error
4388 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4389 while (!error && !done) {
4392 * allocate and setup the transaction
4394 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4395 error = xfs_trans_reserve(tp,
4397 XFS_WRITE_LOG_RES(mp),
4399 XFS_TRANS_PERM_LOG_RES,
4400 XFS_WRITE_LOG_COUNT);
4403 * check for running out of space
4407 * Free the transaction structure.
4409 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4410 xfs_trans_cancel(tp, 0);
4413 xfs_ilock(ip, XFS_ILOCK_EXCL);
4414 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4415 ip->i_udquot, ip->i_gdquot, resblks, 0,
4416 XFS_QMOPT_RES_REGBLKS);
4420 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4421 xfs_trans_ihold(tp, ip);
4424 * issue the bunmapi() call to free the blocks
4426 XFS_BMAP_INIT(&free_list, &firstfsb);
4427 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4428 endoffset_fsb - startoffset_fsb,
4429 0, 2, &firstfsb, &free_list, &done);
4435 * complete the transaction
4437 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4442 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4443 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4448 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4452 xfs_bmap_cancel(&free_list);
4454 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4455 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4461 * xfs_change_file_space()
4462 * This routine allocates or frees disk space for the given file.
4463 * The user specified parameters are checked for alignment and size
4472 xfs_change_file_space(
4486 xfs_off_t startoffset;
4492 vp = BHV_TO_VNODE(bdp);
4493 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4495 ip = XFS_BHVTOI(bdp);
4499 * must be a regular file and have write permission
4502 return XFS_ERROR(EINVAL);
4504 xfs_ilock(ip, XFS_ILOCK_SHARED);
4506 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4507 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4511 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4513 switch (bf->l_whence) {
4514 case 0: /*SEEK_SET*/
4516 case 1: /*SEEK_CUR*/
4517 bf->l_start += offset;
4519 case 2: /*SEEK_END*/
4520 bf->l_start += ip->i_d.di_size;
4523 return XFS_ERROR(EINVAL);
4526 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4528 if ( (bf->l_start < 0)
4529 || (bf->l_start > XFS_MAXIOFFSET(mp))
4530 || (bf->l_start + llen < 0)
4531 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4532 return XFS_ERROR(EINVAL);
4536 startoffset = bf->l_start;
4537 fsize = ip->i_d.di_size;
4540 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4542 * These calls do NOT zero the data space allocated to the file,
4543 * nor do they change the file size.
4545 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4547 * These calls cause the new file data to be zeroed and the file
4548 * size to be changed.
4550 setprealloc = clrprealloc = 0;
4553 case XFS_IOC_RESVSP:
4554 case XFS_IOC_RESVSP64:
4555 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4562 case XFS_IOC_UNRESVSP:
4563 case XFS_IOC_UNRESVSP64:
4564 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4569 case XFS_IOC_ALLOCSP:
4570 case XFS_IOC_ALLOCSP64:
4571 case XFS_IOC_FREESP:
4572 case XFS_IOC_FREESP64:
4573 if (startoffset > fsize) {
4574 error = xfs_alloc_file_space(ip, fsize,
4575 startoffset - fsize, 0, attr_flags);
4580 va.va_mask = XFS_AT_SIZE;
4581 va.va_size = startoffset;
4583 error = xfs_setattr(bdp, &va, attr_flags, credp);
4593 return XFS_ERROR(EINVAL);
4597 * update the inode timestamp, mode, and prealloc flag bits
4599 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4601 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4604 xfs_trans_cancel(tp, 0);
4608 xfs_ilock(ip, XFS_ILOCK_EXCL);
4610 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4611 xfs_trans_ihold(tp, ip);
4613 if ((attr_flags & ATTR_DMI) == 0) {
4614 ip->i_d.di_mode &= ~S_ISUID;
4617 * Note that we don't have to worry about mandatory
4618 * file locking being disabled here because we only
4619 * clear the S_ISGID bit if the Group execute bit is
4620 * on, but if it was on then mandatory locking wouldn't
4621 * have been enabled.
4623 if (ip->i_d.di_mode & S_IXGRP)
4624 ip->i_d.di_mode &= ~S_ISGID;
4626 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4629 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4630 else if (clrprealloc)
4631 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4633 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4634 xfs_trans_set_sync(tp);
4636 error = xfs_trans_commit(tp, 0, NULL);
4638 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4643 vnodeops_t xfs_vnodeops = {
4644 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4645 .vop_open = xfs_open,
4646 .vop_read = xfs_read,
4647 #ifdef HAVE_SENDFILE
4648 .vop_sendfile = xfs_sendfile,
4650 .vop_write = xfs_write,
4651 .vop_ioctl = xfs_ioctl,
4652 .vop_getattr = xfs_getattr,
4653 .vop_setattr = xfs_setattr,
4654 .vop_access = xfs_access,
4655 .vop_lookup = xfs_lookup,
4656 .vop_create = xfs_create,
4657 .vop_remove = xfs_remove,
4658 .vop_link = xfs_link,
4659 .vop_rename = xfs_rename,
4660 .vop_mkdir = xfs_mkdir,
4661 .vop_rmdir = xfs_rmdir,
4662 .vop_readdir = xfs_readdir,
4663 .vop_symlink = xfs_symlink,
4664 .vop_readlink = xfs_readlink,
4665 .vop_fsync = xfs_fsync,
4666 .vop_inactive = xfs_inactive,
4667 .vop_fid2 = xfs_fid2,
4668 .vop_rwlock = xfs_rwlock,
4669 .vop_rwunlock = xfs_rwunlock,
4670 .vop_bmap = xfs_bmap,
4671 .vop_reclaim = xfs_reclaim,
4672 .vop_attr_get = xfs_attr_get,
4673 .vop_attr_set = xfs_attr_set,
4674 .vop_attr_remove = xfs_attr_remove,
4675 .vop_attr_list = xfs_attr_list,
4676 .vop_link_removed = (vop_link_removed_t)fs_noval,
4677 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4678 .vop_tosspages = fs_tosspages,
4679 .vop_flushinval_pages = fs_flushinval_pages,
4680 .vop_flush_pages = fs_flush_pages,
4681 .vop_release = xfs_release,
4682 .vop_iflush = xfs_inode_flush,