2 * Copyright (c) 2000-2006 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
21 #include "xfs_types.h"
25 #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_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
47 #include "xfs_error.h"
48 #include "xfs_quota.h"
49 #include "xfs_utils.h"
50 #include "xfs_rtalloc.h"
51 #include "xfs_trans_space.h"
52 #include "xfs_log_priv.h"
53 #include "xfs_filestream.h"
54 #include "xfs_vnodeops.h"
62 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
63 return XFS_ERROR(EIO);
66 * If it's a directory with any blocks, read-ahead block 0
67 * as we're almost certain to have the next operation be a read there.
69 if (S_ISDIR(ip->i_d.di_mode) && ip->i_d.di_nextents > 0) {
70 mode = xfs_ilock_map_shared(ip);
71 if (ip->i_d.di_nextents > 0)
72 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
73 xfs_iunlock(ip, mode);
88 xfs_mount_t *mp = ip->i_mount;
97 xfs_prid_t projid=0, iprojid=0;
98 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
102 xfs_itrace_entry(ip);
104 if (mp->m_flags & XFS_MOUNT_RDONLY)
105 return XFS_ERROR(EROFS);
108 * Cannot set certain attributes.
111 if (mask & XFS_AT_NOSET) {
112 return XFS_ERROR(EINVAL);
115 if (XFS_FORCED_SHUTDOWN(mp))
116 return XFS_ERROR(EIO);
119 * Timestamps do not need to be logged and hence do not
120 * need to be done within a transaction.
122 if (mask & XFS_AT_UPDTIMES) {
123 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
124 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
125 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
126 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
127 xfs_ichgtime(ip, timeflags);
131 olddquot1 = olddquot2 = NULL;
135 * If disk quotas is on, we make sure that the dquots do exist on disk,
136 * before we start any other transactions. Trying to do this later
137 * is messy. We don't care to take a readlock to look at the ids
138 * in inode here, because we can't hold it across the trans_reserve.
139 * If the IDs do change before we take the ilock, we're covered
140 * because the i_*dquot fields will get updated anyway.
142 if (XFS_IS_QUOTA_ON(mp) &&
143 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
146 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
148 qflags |= XFS_QMOPT_UQUOTA;
150 uid = ip->i_d.di_uid;
152 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
154 qflags |= XFS_QMOPT_GQUOTA;
156 gid = ip->i_d.di_gid;
158 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
159 projid = vap->va_projid;
160 qflags |= XFS_QMOPT_PQUOTA;
162 projid = ip->i_d.di_projid;
165 * We take a reference when we initialize udqp and gdqp,
166 * so it is important that we never blindly double trip on
167 * the same variable. See xfs_create() for an example.
169 ASSERT(udqp == NULL);
170 ASSERT(gdqp == NULL);
171 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
178 * For the other attributes, we acquire the inode lock and
179 * first do an error checking pass.
182 lock_flags = XFS_ILOCK_EXCL;
183 if (flags & ATTR_NOLOCK)
185 if (!(mask & XFS_AT_SIZE)) {
186 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
187 (mp->m_flags & XFS_MOUNT_WSYNC)) {
188 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
190 if ((code = xfs_trans_reserve(tp, 0,
191 XFS_ICHANGE_LOG_RES(mp), 0,
198 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
199 !(flags & ATTR_DMI)) {
200 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
201 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip,
202 vap->va_size, 0, dmflags, NULL);
209 lock_flags |= XFS_IOLOCK_EXCL;
212 xfs_ilock(ip, lock_flags);
214 /* boolean: are we the file owner? */
215 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
218 * Change various properties of a file.
219 * Only the owner or users with CAP_FOWNER
220 * capability may do these things.
223 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
224 XFS_AT_GID|XFS_AT_PROJID)) {
226 * CAP_FOWNER overrides the following restrictions:
228 * The user ID of the calling process must be equal
229 * to the file owner ID, except in cases where the
230 * CAP_FSETID capability is applicable.
232 if (!file_owner && !capable(CAP_FOWNER)) {
233 code = XFS_ERROR(EPERM);
238 * CAP_FSETID overrides the following restrictions:
240 * The effective user ID of the calling process shall match
241 * the file owner when setting the set-user-ID and
242 * set-group-ID bits on that file.
244 * The effective group ID or one of the supplementary group
245 * IDs of the calling process shall match the group owner of
246 * the file when setting the set-group-ID bit on that file
248 if (mask & XFS_AT_MODE) {
251 if ((vap->va_mode & S_ISUID) && !file_owner)
253 if ((vap->va_mode & S_ISGID) &&
254 !in_group_p((gid_t)ip->i_d.di_gid))
257 /* Linux allows this, Irix doesn't. */
258 if ((vap->va_mode & S_ISVTX) && !S_ISDIR(ip->i_d.di_mode))
261 if (m && !capable(CAP_FSETID))
267 * Change file ownership. Must be the owner or privileged.
268 * If the system was configured with the "restricted_chown"
269 * option, the owner is not permitted to give away the file,
270 * and can change the group id only to a group of which he
271 * or she is a member.
273 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
275 * These IDs could have changed since we last looked at them.
276 * But, we're assured that if the ownership did change
277 * while we didn't have the inode locked, inode's dquot(s)
278 * would have changed also.
280 iuid = ip->i_d.di_uid;
281 iprojid = ip->i_d.di_projid;
282 igid = ip->i_d.di_gid;
283 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
284 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
285 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
289 * CAP_CHOWN overrides the following restrictions:
291 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
292 * shall override the restriction that a process cannot
293 * change the user ID of a file it owns and the restriction
294 * that the group ID supplied to the chown() function
295 * shall be equal to either the group ID or one of the
296 * supplementary group IDs of the calling process.
298 if (restricted_chown &&
299 (iuid != uid || (igid != gid &&
300 !in_group_p((gid_t)gid))) &&
301 !capable(CAP_CHOWN)) {
302 code = XFS_ERROR(EPERM);
306 * Do a quota reservation only if uid/projid/gid is actually
309 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
310 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
311 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
313 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
314 capable(CAP_FOWNER) ?
315 XFS_QMOPT_FORCE_RES : 0);
316 if (code) /* out of quota */
322 * Truncate file. Must have write permission and not be a directory.
324 if (mask & XFS_AT_SIZE) {
325 /* Short circuit the truncate case for zero length files */
326 if ((vap->va_size == 0) &&
327 (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) {
328 xfs_iunlock(ip, XFS_ILOCK_EXCL);
329 lock_flags &= ~XFS_ILOCK_EXCL;
330 if (mask & XFS_AT_CTIME)
331 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
336 if (S_ISDIR(ip->i_d.di_mode)) {
337 code = XFS_ERROR(EISDIR);
339 } else if (!S_ISREG(ip->i_d.di_mode)) {
340 code = XFS_ERROR(EINVAL);
344 * Make sure that the dquots are attached to the inode.
346 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
351 * Change file access or modified times.
353 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
355 if ((flags & ATTR_UTIME) &&
356 !capable(CAP_FOWNER)) {
357 code = XFS_ERROR(EPERM);
364 * Change extent size or realtime flag.
366 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
368 * Can't change extent size if any extents are allocated.
370 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
371 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
373 code = XFS_ERROR(EINVAL); /* EFBIG? */
378 * Can't change realtime flag if any extents are allocated.
380 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
381 (mask & XFS_AT_XFLAGS) &&
382 (XFS_IS_REALTIME_INODE(ip)) !=
383 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
384 code = XFS_ERROR(EINVAL); /* EFBIG? */
388 * Extent size must be a multiple of the appropriate block
389 * size, if set at all.
391 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
394 if (XFS_IS_REALTIME_INODE(ip) ||
395 ((mask & XFS_AT_XFLAGS) &&
396 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
397 size = mp->m_sb.sb_rextsize <<
398 mp->m_sb.sb_blocklog;
400 size = mp->m_sb.sb_blocksize;
402 if (vap->va_extsize % size) {
403 code = XFS_ERROR(EINVAL);
408 * If realtime flag is set then must have realtime data.
410 if ((mask & XFS_AT_XFLAGS) &&
411 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
412 if ((mp->m_sb.sb_rblocks == 0) ||
413 (mp->m_sb.sb_rextsize == 0) ||
414 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
415 code = XFS_ERROR(EINVAL);
421 * Can't modify an immutable/append-only file unless
422 * we have appropriate permission.
424 if ((mask & XFS_AT_XFLAGS) &&
426 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
428 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
429 !capable(CAP_LINUX_IMMUTABLE)) {
430 code = XFS_ERROR(EPERM);
436 * Now we can make the changes. Before we join the inode
437 * to the transaction, if XFS_AT_SIZE is set then take care of
438 * the part of the truncation that must be done without the
439 * inode lock. This needs to be done before joining the inode
440 * to the transaction, because the inode cannot be unlocked
441 * once it is a part of the transaction.
443 if (mask & XFS_AT_SIZE) {
445 if ((vap->va_size > ip->i_size) &&
446 (flags & ATTR_NOSIZETOK) == 0) {
448 * Do the first part of growing a file: zero any data
449 * in the last block that is beyond the old EOF. We
450 * need to do this before the inode is joined to the
451 * transaction to modify the i_size.
453 code = xfs_zero_eof(ip, vap->va_size, ip->i_size);
455 xfs_iunlock(ip, XFS_ILOCK_EXCL);
458 * We are going to log the inode size change in this
459 * transaction so any previous writes that are beyond the on
460 * disk EOF and the new EOF that have not been written out need
461 * to be written here. If we do not write the data out, we
462 * expose ourselves to the null files problem.
464 * Only flush from the on disk size to the smaller of the in
465 * memory file size or the new size as that's the range we
466 * really care about here and prevents waiting for other data
467 * not within the range we care about here.
470 (ip->i_size != ip->i_d.di_size) &&
471 (vap->va_size > ip->i_d.di_size)) {
472 code = xfs_flush_pages(ip,
473 ip->i_d.di_size, vap->va_size,
474 XFS_B_ASYNC, FI_NONE);
477 /* wait for all I/O to complete */
481 code = xfs_itruncate_data(ip, vap->va_size);
484 lock_flags &= ~XFS_ILOCK_EXCL;
485 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
488 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
489 if ((code = xfs_trans_reserve(tp, 0,
490 XFS_ITRUNCATE_LOG_RES(mp), 0,
491 XFS_TRANS_PERM_LOG_RES,
492 XFS_ITRUNCATE_LOG_COUNT))) {
493 xfs_trans_cancel(tp, 0);
495 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
498 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
499 xfs_ilock(ip, XFS_ILOCK_EXCL);
503 xfs_trans_ijoin(tp, ip, lock_flags);
504 xfs_trans_ihold(tp, ip);
508 * Truncate file. Must have write permission and not be a directory.
510 if (mask & XFS_AT_SIZE) {
512 * Only change the c/mtime if we are changing the size
513 * or we are explicitly asked to change it. This handles
514 * the semantic difference between truncate() and ftruncate()
515 * as implemented in the VFS.
517 if (vap->va_size != ip->i_size || (mask & XFS_AT_CTIME))
518 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
520 if (vap->va_size > ip->i_size) {
521 ip->i_d.di_size = vap->va_size;
522 ip->i_size = vap->va_size;
523 if (!(flags & ATTR_DMI))
524 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
525 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
526 } else if ((vap->va_size <= ip->i_size) ||
527 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
529 * signal a sync transaction unless
530 * we're truncating an already unlinked
531 * file on a wsync filesystem
533 code = xfs_itruncate_finish(&tp, ip,
534 (xfs_fsize_t)vap->va_size,
536 ((ip->i_d.di_nlink != 0 ||
537 !(mp->m_flags & XFS_MOUNT_WSYNC))
542 * Truncated "down", so we're removing references
543 * to old data here - if we now delay flushing for
544 * a long time, we expose ourselves unduly to the
545 * notorious NULL files problem. So, we mark this
546 * vnode and flush it when the file is closed, and
547 * do not wait the usual (long) time for writeout.
549 xfs_iflags_set(ip, XFS_ITRUNCATED);
554 * Change file access modes.
556 if (mask & XFS_AT_MODE) {
557 ip->i_d.di_mode &= S_IFMT;
558 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
560 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
561 timeflags |= XFS_ICHGTIME_CHG;
565 * Change file ownership. Must be the owner or privileged.
566 * If the system was configured with the "restricted_chown"
567 * option, the owner is not permitted to give away the file,
568 * and can change the group id only to a group of which he
569 * or she is a member.
571 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
573 * CAP_FSETID overrides the following restrictions:
575 * The set-user-ID and set-group-ID bits of a file will be
576 * cleared upon successful return from chown()
578 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
579 !capable(CAP_FSETID)) {
580 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
584 * Change the ownerships and register quota modifications
585 * in the transaction.
588 if (XFS_IS_UQUOTA_ON(mp)) {
589 ASSERT(mask & XFS_AT_UID);
591 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
592 &ip->i_udquot, udqp);
594 ip->i_d.di_uid = uid;
597 if (XFS_IS_GQUOTA_ON(mp)) {
598 ASSERT(!XFS_IS_PQUOTA_ON(mp));
599 ASSERT(mask & XFS_AT_GID);
601 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
602 &ip->i_gdquot, gdqp);
604 ip->i_d.di_gid = gid;
606 if (iprojid != projid) {
607 if (XFS_IS_PQUOTA_ON(mp)) {
608 ASSERT(!XFS_IS_GQUOTA_ON(mp));
609 ASSERT(mask & XFS_AT_PROJID);
611 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
612 &ip->i_gdquot, gdqp);
614 ip->i_d.di_projid = projid;
616 * We may have to rev the inode as well as
617 * the superblock version number since projids didn't
618 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
620 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
621 xfs_bump_ino_vers2(tp, ip);
624 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
625 timeflags |= XFS_ICHGTIME_CHG;
630 * Change file access or modified times.
632 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
633 if (mask & XFS_AT_ATIME) {
634 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
635 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
636 ip->i_update_core = 1;
637 timeflags &= ~XFS_ICHGTIME_ACC;
639 if (mask & XFS_AT_MTIME) {
640 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
641 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
642 timeflags &= ~XFS_ICHGTIME_MOD;
643 timeflags |= XFS_ICHGTIME_CHG;
645 if (tp && (flags & ATTR_UTIME))
646 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
650 * Change XFS-added attributes.
652 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
653 if (mask & XFS_AT_EXTSIZE) {
655 * Converting bytes to fs blocks.
657 ip->i_d.di_extsize = vap->va_extsize >>
658 mp->m_sb.sb_blocklog;
660 if (mask & XFS_AT_XFLAGS) {
663 /* can't set PREALLOC this way, just preserve it */
664 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
665 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
666 di_flags |= XFS_DIFLAG_IMMUTABLE;
667 if (vap->va_xflags & XFS_XFLAG_APPEND)
668 di_flags |= XFS_DIFLAG_APPEND;
669 if (vap->va_xflags & XFS_XFLAG_SYNC)
670 di_flags |= XFS_DIFLAG_SYNC;
671 if (vap->va_xflags & XFS_XFLAG_NOATIME)
672 di_flags |= XFS_DIFLAG_NOATIME;
673 if (vap->va_xflags & XFS_XFLAG_NODUMP)
674 di_flags |= XFS_DIFLAG_NODUMP;
675 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
676 di_flags |= XFS_DIFLAG_PROJINHERIT;
677 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
678 di_flags |= XFS_DIFLAG_NODEFRAG;
679 if (vap->va_xflags & XFS_XFLAG_FILESTREAM)
680 di_flags |= XFS_DIFLAG_FILESTREAM;
681 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
682 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
683 di_flags |= XFS_DIFLAG_RTINHERIT;
684 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
685 di_flags |= XFS_DIFLAG_NOSYMLINKS;
686 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
687 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
688 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
689 if (vap->va_xflags & XFS_XFLAG_REALTIME)
690 di_flags |= XFS_DIFLAG_REALTIME;
691 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
692 di_flags |= XFS_DIFLAG_EXTSIZE;
694 ip->i_d.di_flags = di_flags;
696 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
697 timeflags |= XFS_ICHGTIME_CHG;
701 * Change file inode change time only if XFS_AT_CTIME set
702 * AND we have been called by a DMI function.
705 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
706 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
707 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
708 ip->i_update_core = 1;
709 timeflags &= ~XFS_ICHGTIME_CHG;
713 * Send out timestamp changes that need to be set to the
714 * current time. Not done when called by a DMI function.
716 if (timeflags && !(flags & ATTR_DMI))
717 xfs_ichgtime(ip, timeflags);
719 XFS_STATS_INC(xs_ig_attrchg);
722 * If this is a synchronous mount, make sure that the
723 * transaction goes to disk before returning to the user.
724 * This is slightly sub-optimal in that truncates require
725 * two sync transactions instead of one for wsync filesystems.
726 * One for the truncate and one for the timestamps since we
727 * don't want to change the timestamps unless we're sure the
728 * truncate worked. Truncates are less than 1% of the laddis
729 * mix so this probably isn't worth the trouble to optimize.
733 if (mp->m_flags & XFS_MOUNT_WSYNC)
734 xfs_trans_set_sync(tp);
736 code = xfs_trans_commit(tp, commit_flags);
739 xfs_iunlock(ip, lock_flags);
742 * Release any dquot(s) the inode had kept before chown.
744 XFS_QM_DQRELE(mp, olddquot1);
745 XFS_QM_DQRELE(mp, olddquot2);
746 XFS_QM_DQRELE(mp, udqp);
747 XFS_QM_DQRELE(mp, gdqp);
753 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
754 !(flags & ATTR_DMI)) {
755 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
756 NULL, DM_RIGHT_NULL, NULL, NULL,
757 0, 0, AT_DELAY_FLAG(flags));
762 commit_flags |= XFS_TRANS_ABORT;
765 XFS_QM_DQRELE(mp, udqp);
766 XFS_QM_DQRELE(mp, gdqp);
768 xfs_trans_cancel(tp, commit_flags);
770 if (lock_flags != 0) {
771 xfs_iunlock(ip, lock_flags);
777 * The maximum pathlen is 1024 bytes. Since the minimum file system
778 * blocksize is 512 bytes, we can get a max of 2 extents back from
781 #define SYMLINK_MAPS 2
788 xfs_mount_t *mp = ip->i_mount;
789 int pathlen = ip->i_d.di_size;
790 int nmaps = SYMLINK_MAPS;
791 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
798 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
799 mval, &nmaps, NULL, NULL);
803 for (n = 0; n < nmaps; n++) {
804 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
805 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
807 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
808 error = XFS_BUF_GETERROR(bp);
810 xfs_ioerror_alert("xfs_readlink",
811 ip->i_mount, bp, XFS_BUF_ADDR(bp));
815 if (pathlen < byte_cnt)
819 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
823 link[ip->i_d.di_size] = '\0';
835 xfs_mount_t *mp = ip->i_mount;
839 xfs_itrace_entry(ip);
841 if (XFS_FORCED_SHUTDOWN(mp))
842 return XFS_ERROR(EIO);
844 xfs_ilock(ip, XFS_ILOCK_SHARED);
846 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
847 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
849 pathlen = ip->i_d.di_size;
853 if (ip->i_df.if_flags & XFS_IFINLINE) {
854 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
855 link[pathlen] = '\0';
857 error = xfs_readlink_bmap(ip, link);
861 xfs_iunlock(ip, XFS_ILOCK_SHARED);
868 * This is called to sync the inode and its data out to disk. We need to hold
869 * the I/O lock while flushing the data, and the inode lock while flushing the
870 * inode. The inode lock CANNOT be held while flushing the data, so acquire
871 * after we're done with that.
879 int log_flushed = 0, changed = 1;
881 xfs_itrace_entry(ip);
883 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
884 return XFS_ERROR(EIO);
886 /* capture size updates in I/O completion before writing the inode. */
887 error = filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping);
889 return XFS_ERROR(error);
892 * We always need to make sure that the required inode state is safe on
893 * disk. The vnode might be clean but we still might need to force the
894 * log because of committed transactions that haven't hit the disk yet.
895 * Likewise, there could be unflushed non-transactional changes to the
896 * inode core that have to go to disk and this requires us to issue
897 * a synchronous transaction to capture these changes correctly.
899 * This code relies on the assumption that if the update_* fields
900 * of the inode are clear and the inode is unpinned then it is clean
901 * and no action is required.
903 xfs_ilock(ip, XFS_ILOCK_SHARED);
905 if (!(ip->i_update_size || ip->i_update_core)) {
907 * Timestamps/size haven't changed since last inode flush or
908 * inode transaction commit. That means either nothing got
909 * written or a transaction committed which caught the updates.
910 * If the latter happened and the transaction hasn't hit the
911 * disk yet, the inode will be still be pinned. If it is,
915 xfs_iunlock(ip, XFS_ILOCK_SHARED);
917 if (xfs_ipincount(ip)) {
918 error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
919 XFS_LOG_FORCE | XFS_LOG_SYNC,
923 * If the inode is not pinned and nothing has changed
924 * we don't need to flush the cache.
930 * Kick off a transaction to log the inode core to get the
931 * updates. The sync transaction will also force the log.
933 xfs_iunlock(ip, XFS_ILOCK_SHARED);
934 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
935 error = xfs_trans_reserve(tp, 0,
936 XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0);
938 xfs_trans_cancel(tp, 0);
941 xfs_ilock(ip, XFS_ILOCK_EXCL);
944 * Note - it's possible that we might have pushed ourselves out
945 * of the way during trans_reserve which would flush the inode.
946 * But there's no guarantee that the inode buffer has actually
947 * gone out yet (it's delwri). Plus the buffer could be pinned
948 * anyway if it's part of an inode in another recent
949 * transaction. So we play it safe and fire off the
950 * transaction anyway.
952 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
953 xfs_trans_ihold(tp, ip);
954 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
955 xfs_trans_set_sync(tp);
956 error = _xfs_trans_commit(tp, 0, &log_flushed);
958 xfs_iunlock(ip, XFS_ILOCK_EXCL);
961 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
963 * If the log write didn't issue an ordered tag we need
964 * to flush the disk cache for the data device now.
967 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
970 * If this inode is on the RT dev we need to flush that
973 if (XFS_IS_REALTIME_INODE(ip))
974 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
981 * This is called by xfs_inactive to free any blocks beyond eof
982 * when the link count isn't zero and by xfs_dm_punch_hole() when
983 * punching a hole to EOF.
993 xfs_fileoff_t end_fsb;
994 xfs_fileoff_t last_fsb;
995 xfs_filblks_t map_len;
997 xfs_bmbt_irec_t imap;
998 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
1001 * Figure out if there are any blocks beyond the end
1002 * of the file. If not, then there is nothing to do.
1004 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
1005 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1006 map_len = last_fsb - end_fsb;
1011 xfs_ilock(ip, XFS_ILOCK_SHARED);
1012 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1013 NULL, 0, &imap, &nimaps, NULL, NULL);
1014 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1016 if (!error && (nimaps != 0) &&
1017 (imap.br_startblock != HOLESTARTBLOCK ||
1018 ip->i_delayed_blks)) {
1020 * Attach the dquots to the inode up front.
1022 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1026 * There are blocks after the end of file.
1027 * Free them up now by truncating the file to
1030 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1033 * Do the xfs_itruncate_start() call before
1034 * reserving any log space because
1035 * itruncate_start will call into the buffer
1036 * cache and we can't
1037 * do that within a transaction.
1040 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1041 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1044 xfs_trans_cancel(tp, 0);
1046 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1050 error = xfs_trans_reserve(tp, 0,
1051 XFS_ITRUNCATE_LOG_RES(mp),
1052 0, XFS_TRANS_PERM_LOG_RES,
1053 XFS_ITRUNCATE_LOG_COUNT);
1055 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1056 xfs_trans_cancel(tp, 0);
1057 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1061 xfs_ilock(ip, XFS_ILOCK_EXCL);
1062 xfs_trans_ijoin(tp, ip,
1065 xfs_trans_ihold(tp, ip);
1067 error = xfs_itruncate_finish(&tp, ip,
1072 * If we get an error at this point we
1073 * simply don't bother truncating the file.
1076 xfs_trans_cancel(tp,
1077 (XFS_TRANS_RELEASE_LOG_RES |
1080 error = xfs_trans_commit(tp,
1081 XFS_TRANS_RELEASE_LOG_RES);
1083 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
1090 * Free a symlink that has blocks associated with it.
1093 xfs_inactive_symlink_rmt(
1101 xfs_fsblock_t first_block;
1102 xfs_bmap_free_t free_list;
1105 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1113 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1115 * We're freeing a symlink that has some
1116 * blocks allocated to it. Free the
1117 * blocks here. We know that we've got
1118 * either 1 or 2 extents and that we can
1119 * free them all in one bunmapi call.
1121 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1122 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1123 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1124 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1125 xfs_trans_cancel(tp, 0);
1130 * Lock the inode, fix the size, and join it to the transaction.
1131 * Hold it so in the normal path, we still have it locked for
1132 * the second transaction. In the error paths we need it
1133 * held so the cancel won't rele it, see below.
1135 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1136 size = (int)ip->i_d.di_size;
1137 ip->i_d.di_size = 0;
1138 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1139 xfs_trans_ihold(tp, ip);
1140 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1142 * Find the block(s) so we can inval and unmap them.
1145 XFS_BMAP_INIT(&free_list, &first_block);
1146 nmaps = ARRAY_SIZE(mval);
1147 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1148 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1152 * Invalidate the block(s).
1154 for (i = 0; i < nmaps; i++) {
1155 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1156 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1157 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1158 xfs_trans_binval(tp, bp);
1161 * Unmap the dead block(s) to the free_list.
1163 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1164 &first_block, &free_list, NULL, &done)))
1168 * Commit the first transaction. This logs the EFI and the inode.
1170 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
1173 * The transaction must have been committed, since there were
1174 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1175 * The new tp has the extent freeing and EFDs.
1179 * The first xact was committed, so add the inode to the new one.
1180 * Mark it dirty so it will be logged and moved forward in the log as
1181 * part of every commit.
1183 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1184 xfs_trans_ihold(tp, ip);
1185 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1187 * Get a new, empty transaction to return to our caller.
1189 ntp = xfs_trans_dup(tp);
1191 * Commit the transaction containing extent freeing and EFDs.
1192 * If we get an error on the commit here or on the reserve below,
1193 * we need to unlock the inode since the new transaction doesn't
1194 * have the inode attached.
1196 error = xfs_trans_commit(tp, 0);
1199 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1203 * Remove the memory for extent descriptions (just bookkeeping).
1205 if (ip->i_df.if_bytes)
1206 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1207 ASSERT(ip->i_df.if_bytes == 0);
1209 * Put an itruncate log reservation in the new transaction
1212 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1213 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1214 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1218 * Return with the inode locked but not joined to the transaction.
1224 xfs_bmap_cancel(&free_list);
1227 * Have to come here with the inode locked and either
1228 * (held and in the transaction) or (not in the transaction).
1229 * If the inode isn't held then cancel would iput it, but
1230 * that's wrong since this is inactive and the vnode ref
1231 * count is 0 already.
1232 * Cancel won't do anything to the inode if held, but it still
1233 * needs to be locked until the cancel is done, if it was
1234 * joined to the transaction.
1236 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1237 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1244 xfs_inactive_symlink_local(
1250 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1252 * We're freeing a symlink which fit into
1253 * the inode. Just free the memory used
1254 * to hold the old symlink.
1256 error = xfs_trans_reserve(*tpp, 0,
1257 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1258 0, XFS_TRANS_PERM_LOG_RES,
1259 XFS_ITRUNCATE_LOG_COUNT);
1262 xfs_trans_cancel(*tpp, 0);
1266 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1269 * Zero length symlinks _can_ exist.
1271 if (ip->i_df.if_bytes > 0) {
1272 xfs_idata_realloc(ip,
1273 -(ip->i_df.if_bytes),
1275 ASSERT(ip->i_df.if_bytes == 0);
1289 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1292 ASSERT(ip->i_d.di_forkoff != 0);
1293 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1294 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1298 error = xfs_attr_inactive(ip);
1302 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1303 error = xfs_trans_reserve(tp, 0,
1304 XFS_IFREE_LOG_RES(mp),
1305 0, XFS_TRANS_PERM_LOG_RES,
1306 XFS_INACTIVE_LOG_COUNT);
1310 xfs_ilock(ip, XFS_ILOCK_EXCL);
1311 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1312 xfs_trans_ihold(tp, ip);
1313 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1315 ASSERT(ip->i_d.di_anextents == 0);
1321 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1322 xfs_trans_cancel(tp, 0);
1325 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1333 bhv_vnode_t *vp = XFS_ITOV(ip);
1334 xfs_mount_t *mp = ip->i_mount;
1337 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1340 /* If this is a read-only mount, don't do this (would generate I/O) */
1341 if (mp->m_flags & XFS_MOUNT_RDONLY)
1344 if (!XFS_FORCED_SHUTDOWN(mp)) {
1348 * If we are using filestreams, and we have an unlinked
1349 * file that we are processing the last close on, then nothing
1350 * will be able to reopen and write to this file. Purge this
1351 * inode from the filestreams cache so that it doesn't delay
1352 * teardown of the inode.
1354 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1355 xfs_filestream_deassociate(ip);
1358 * If we previously truncated this file and removed old data
1359 * in the process, we want to initiate "early" writeout on
1360 * the last close. This is an attempt to combat the notorious
1361 * NULL files problem which is particularly noticable from a
1362 * truncate down, buffered (re-)write (delalloc), followed by
1363 * a crash. What we are effectively doing here is
1364 * significantly reducing the time window where we'd otherwise
1365 * be exposed to that problem.
1367 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1368 if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
1369 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
1372 if (ip->i_d.di_nlink != 0) {
1373 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1374 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1375 ip->i_delayed_blks > 0)) &&
1376 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1377 (!(ip->i_d.di_flags &
1378 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1379 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1391 * This is called when the vnode reference count for the vnode
1392 * goes to zero. If the file has been unlinked, then it must
1393 * now be truncated. Also, we clear all of the read-ahead state
1394 * kept for the inode here since the file is now closed.
1400 bhv_vnode_t *vp = XFS_ITOV(ip);
1401 xfs_bmap_free_t free_list;
1402 xfs_fsblock_t first_block;
1409 xfs_itrace_entry(ip);
1412 * If the inode is already free, then there can be nothing
1415 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1416 ASSERT(ip->i_df.if_real_bytes == 0);
1417 ASSERT(ip->i_df.if_broot_bytes == 0);
1418 return VN_INACTIVE_CACHE;
1422 * Only do a truncate if it's a regular file with
1423 * some actual space in it. It's OK to look at the
1424 * inode's fields without the lock because we're the
1425 * only one with a reference to the inode.
1427 truncate = ((ip->i_d.di_nlink == 0) &&
1428 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1429 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1430 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1434 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1435 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
1439 /* If this is a read-only mount, don't do this (would generate I/O) */
1440 if (mp->m_flags & XFS_MOUNT_RDONLY)
1443 if (ip->i_d.di_nlink != 0) {
1444 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1445 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1446 ip->i_delayed_blks > 0)) &&
1447 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1448 (!(ip->i_d.di_flags &
1449 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1450 (ip->i_delayed_blks != 0)))) {
1451 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1453 return VN_INACTIVE_CACHE;
1458 ASSERT(ip->i_d.di_nlink == 0);
1460 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1461 return VN_INACTIVE_CACHE;
1463 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1466 * Do the xfs_itruncate_start() call before
1467 * reserving any log space because itruncate_start
1468 * will call into the buffer cache and we can't
1469 * do that within a transaction.
1471 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1473 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1475 xfs_trans_cancel(tp, 0);
1476 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1477 return VN_INACTIVE_CACHE;
1480 error = xfs_trans_reserve(tp, 0,
1481 XFS_ITRUNCATE_LOG_RES(mp),
1482 0, XFS_TRANS_PERM_LOG_RES,
1483 XFS_ITRUNCATE_LOG_COUNT);
1485 /* Don't call itruncate_cleanup */
1486 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1487 xfs_trans_cancel(tp, 0);
1488 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1489 return VN_INACTIVE_CACHE;
1492 xfs_ilock(ip, XFS_ILOCK_EXCL);
1493 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1494 xfs_trans_ihold(tp, ip);
1497 * normally, we have to run xfs_itruncate_finish sync.
1498 * But if filesystem is wsync and we're in the inactive
1499 * path, then we know that nlink == 0, and that the
1500 * xaction that made nlink == 0 is permanently committed
1501 * since xfs_remove runs as a synchronous transaction.
1503 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1504 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1507 xfs_trans_cancel(tp,
1508 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1509 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1510 return VN_INACTIVE_CACHE;
1512 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1515 * If we get an error while cleaning up a
1516 * symlink we bail out.
1518 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1519 xfs_inactive_symlink_rmt(ip, &tp) :
1520 xfs_inactive_symlink_local(ip, &tp);
1524 return VN_INACTIVE_CACHE;
1527 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1528 xfs_trans_ihold(tp, ip);
1530 error = xfs_trans_reserve(tp, 0,
1531 XFS_IFREE_LOG_RES(mp),
1532 0, XFS_TRANS_PERM_LOG_RES,
1533 XFS_INACTIVE_LOG_COUNT);
1535 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1536 xfs_trans_cancel(tp, 0);
1537 return VN_INACTIVE_CACHE;
1540 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1541 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1542 xfs_trans_ihold(tp, ip);
1546 * If there are attributes associated with the file
1547 * then blow them away now. The code calls a routine
1548 * that recursively deconstructs the attribute fork.
1549 * We need to just commit the current transaction
1550 * because we can't use it for xfs_attr_inactive().
1552 if (ip->i_d.di_anextents > 0) {
1553 error = xfs_inactive_attrs(ip, &tp);
1555 * If we got an error, the transaction is already
1556 * cancelled, and the inode is unlocked. Just get out.
1559 return VN_INACTIVE_CACHE;
1560 } else if (ip->i_afp) {
1561 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1567 XFS_BMAP_INIT(&free_list, &first_block);
1568 error = xfs_ifree(tp, ip, &free_list);
1571 * If we fail to free the inode, shut down. The cancel
1572 * might do that, we need to make sure. Otherwise the
1573 * inode might be lost for a long time or forever.
1575 if (!XFS_FORCED_SHUTDOWN(mp)) {
1577 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1578 error, mp->m_fsname);
1579 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1581 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1584 * Credit the quota account(s). The inode is gone.
1586 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1589 * Just ignore errors at this point. There is nothing we can
1590 * do except to try to keep going. Make sure it's not a silent
1593 error = xfs_bmap_finish(&tp, &free_list, &committed);
1595 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1596 "xfs_bmap_finish() returned error %d", error);
1597 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1599 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1600 "xfs_trans_commit() returned error %d", error);
1603 * Release the dquots held by inode, if any.
1605 XFS_QM_DQDETACH(mp, ip);
1607 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1610 return VN_INACTIVE_CACHE;
1614 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1615 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1616 * ci_name->name will point to a the actual name (caller must free) or
1617 * will be set to NULL if an exact match is found.
1622 struct xfs_name *name,
1624 struct xfs_name *ci_name)
1630 xfs_itrace_entry(dp);
1632 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1633 return XFS_ERROR(EIO);
1635 lock_mode = xfs_ilock_map_shared(dp);
1636 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
1637 xfs_iunlock_map_shared(dp, lock_mode);
1642 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
1646 xfs_itrace_ref(*ipp);
1651 kmem_free(ci_name->name);
1660 struct xfs_name *name,
1666 xfs_mount_t *mp = dp->i_mount;
1670 xfs_bmap_free_t free_list;
1671 xfs_fsblock_t first_block;
1672 boolean_t unlock_dp_on_error = B_FALSE;
1673 int dm_event_sent = 0;
1677 struct xfs_dquot *udqp, *gdqp;
1681 xfs_itrace_entry(dp);
1683 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
1684 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1685 dp, DM_RIGHT_NULL, NULL,
1686 DM_RIGHT_NULL, name->name, NULL,
1694 if (XFS_FORCED_SHUTDOWN(mp))
1695 return XFS_ERROR(EIO);
1697 /* Return through std_return after this point. */
1700 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1701 prid = dp->i_d.di_projid;
1703 prid = (xfs_prid_t)dfltprid;
1706 * Make sure that we have allocated dquot(s) on disk.
1708 error = XFS_QM_DQVOPALLOC(mp, dp,
1709 current_fsuid(credp), current_fsgid(credp), prid,
1710 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1716 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1717 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1718 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1720 * Initially assume that the file does not exist and
1721 * reserve the resources for that case. If that is not
1722 * the case we'll drop the one we have and get a more
1723 * appropriate transaction later.
1725 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1726 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1727 if (error == ENOSPC) {
1729 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1730 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1737 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1738 unlock_dp_on_error = B_TRUE;
1740 XFS_BMAP_INIT(&free_list, &first_block);
1745 * Reserve disk quota and the inode.
1747 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1751 error = xfs_dir_canenter(tp, dp, name, resblks);
1754 error = xfs_dir_ialloc(&tp, dp, mode, 1,
1755 rdev, credp, prid, resblks > 0,
1758 if (error == ENOSPC)
1765 * At this point, we've gotten a newly allocated inode.
1766 * It is locked (and joined to the transaction).
1769 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1772 * Now we join the directory inode to the transaction. We do not do it
1773 * earlier because xfs_dir_ialloc might commit the previous transaction
1774 * (and release all the locks). An error from here on will result in
1775 * the transaction cancel unlocking dp so don't do it explicitly in the
1779 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1780 unlock_dp_on_error = B_FALSE;
1782 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1783 &first_block, &free_list, resblks ?
1784 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1786 ASSERT(error != ENOSPC);
1789 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1790 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1793 * If this is a synchronous mount, make sure that the
1794 * create transaction goes to disk before returning to
1797 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1798 xfs_trans_set_sync(tp);
1804 * Attach the dquot(s) to the inodes and modify them incore.
1805 * These ids of the inode couldn't have changed since the new
1806 * inode has been locked ever since it was created.
1808 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
1811 * xfs_trans_commit normally decrements the vnode ref count
1812 * when it unlocks the inode. Since we want to return the
1813 * vnode to the caller, we bump the vnode ref count now.
1817 error = xfs_bmap_finish(&tp, &free_list, &committed);
1819 xfs_bmap_cancel(&free_list);
1823 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1830 XFS_QM_DQRELE(mp, udqp);
1831 XFS_QM_DQRELE(mp, gdqp);
1835 /* Fallthrough to std_return with error = 0 */
1838 if ((*ipp || (error != 0 && dm_event_sent != 0)) &&
1839 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
1840 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
1843 DM_RIGHT_NULL, name->name, NULL,
1849 cancel_flags |= XFS_TRANS_ABORT;
1854 xfs_trans_cancel(tp, cancel_flags);
1856 XFS_QM_DQRELE(mp, udqp);
1857 XFS_QM_DQRELE(mp, gdqp);
1859 if (unlock_dp_on_error)
1860 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1866 * Wait until after the current transaction is aborted to
1867 * release the inode. This prevents recursive transactions
1868 * and deadlocks from xfs_inactive.
1870 cancel_flags |= XFS_TRANS_ABORT;
1871 xfs_trans_cancel(tp, cancel_flags);
1874 XFS_QM_DQRELE(mp, udqp);
1875 XFS_QM_DQRELE(mp, gdqp);
1882 * Some counters to see if (and how often) we are hitting some deadlock
1883 * prevention code paths.
1887 int xfs_rm_lock_delays;
1888 int xfs_rm_attempts;
1892 * The following routine will lock the inodes associated with the
1893 * directory and the named entry in the directory. The locks are
1894 * acquired in increasing inode number.
1896 * If the entry is "..", then only the directory is locked. The
1897 * vnode ref count will still include that from the .. entry in
1900 * There is a deadlock we need to worry about. If the locked directory is
1901 * in the AIL, it might be blocking up the log. The next inode we lock
1902 * could be already locked by another thread waiting for log space (e.g
1903 * a permanent log reservation with a long running transaction (see
1904 * xfs_itruncate_finish)). To solve this, we must check if the directory
1905 * is in the ail and use lock_nowait. If we can't lock, we need to
1906 * drop the inode lock on the directory and try again. xfs_iunlock will
1907 * potentially push the tail if we were holding up the log.
1910 xfs_lock_dir_and_entry(
1912 xfs_inode_t *ip) /* inode of entry 'name' */
1916 xfs_inode_t *ips[2];
1925 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1932 * We want to lock in increasing inum. Since we've already
1933 * acquired the lock on the directory, we may need to release
1934 * if if the inum of the entry turns out to be less.
1936 if (e_inum > dp->i_ino) {
1938 * We are already in the right order, so just
1939 * lock on the inode of the entry.
1940 * We need to use nowait if dp is in the AIL.
1943 lp = (xfs_log_item_t *)dp->i_itemp;
1944 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1945 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1952 * Unlock dp and try again.
1953 * xfs_iunlock will try to push the tail
1954 * if the inode is in the AIL.
1957 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1959 if ((attempts % 5) == 0) {
1960 delay(1); /* Don't just spin the CPU */
1962 xfs_rm_lock_delays++;
1968 xfs_ilock(ip, XFS_ILOCK_EXCL);
1970 } else if (e_inum < dp->i_ino) {
1971 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1975 xfs_lock_inodes(ips, 2, XFS_ILOCK_EXCL);
1977 /* else e_inum == dp->i_ino */
1978 /* This can happen if we're asked to lock /x/..
1979 * the entry is "..", which is also the parent directory.
1987 int xfs_small_retries;
1988 int xfs_middle_retries;
1989 int xfs_lots_retries;
1990 int xfs_lock_delays;
1994 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1998 xfs_lock_inumorder(int lock_mode, int subclass)
2000 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
2001 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
2002 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
2003 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
2009 * The following routine will lock n inodes in exclusive mode.
2010 * We assume the caller calls us with the inodes in i_ino order.
2012 * We need to detect deadlock where an inode that we lock
2013 * is in the AIL and we start waiting for another inode that is locked
2014 * by a thread in a long running transaction (such as truncate). This can
2015 * result in deadlock since the long running trans might need to wait
2016 * for the inode we just locked in order to push the tail and free space
2025 int attempts = 0, i, j, try_lock;
2028 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2034 for (; i < inodes; i++) {
2037 if (i && (ips[i] == ips[i-1])) /* Already locked */
2041 * If try_lock is not set yet, make sure all locked inodes
2042 * are not in the AIL.
2043 * If any are, set try_lock to be used later.
2047 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2048 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2049 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2056 * If any of the previous locks we have locked is in the AIL,
2057 * we must TRY to get the second and subsequent locks. If
2058 * we can't get any, we must release all we have
2063 /* try_lock must be 0 if i is 0. */
2065 * try_lock means we have an inode locked
2066 * that is in the AIL.
2069 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
2073 * Unlock all previous guys and try again.
2074 * xfs_iunlock will try to push the tail
2075 * if the inode is in the AIL.
2078 for(j = i - 1; j >= 0; j--) {
2081 * Check to see if we've already
2082 * unlocked this one.
2083 * Not the first one going back,
2084 * and the inode ptr is the same.
2086 if ((j != (i - 1)) && ips[j] ==
2090 xfs_iunlock(ips[j], lock_mode);
2093 if ((attempts % 5) == 0) {
2094 delay(1); /* Don't just spin the CPU */
2104 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
2110 if (attempts < 5) xfs_small_retries++;
2111 else if (attempts < 100) xfs_middle_retries++;
2112 else xfs_lots_retries++;
2120 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2121 int remove_which_error_return = 0;
2123 #define REMOVE_DEBUG_TRACE(x)
2124 #endif /* ! DEBUG */
2129 struct xfs_name *name,
2132 xfs_mount_t *mp = dp->i_mount;
2133 xfs_trans_t *tp = NULL;
2135 xfs_bmap_free_t free_list;
2136 xfs_fsblock_t first_block;
2142 xfs_itrace_entry(dp);
2144 if (XFS_FORCED_SHUTDOWN(mp))
2145 return XFS_ERROR(EIO);
2147 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
2148 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL,
2149 NULL, DM_RIGHT_NULL, name->name, NULL,
2150 ip->i_d.di_mode, 0, 0);
2155 xfs_itrace_entry(ip);
2158 error = XFS_QM_DQATTACH(mp, dp, 0);
2160 error = XFS_QM_DQATTACH(mp, ip, 0);
2162 REMOVE_DEBUG_TRACE(__LINE__);
2166 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2167 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2169 * We try to get the real space reservation first,
2170 * allowing for directory btree deletion(s) implying
2171 * possible bmap insert(s). If we can't get the space
2172 * reservation then we use 0 instead, and avoid the bmap
2173 * btree insert(s) in the directory code by, if the bmap
2174 * insert tries to happen, instead trimming the LAST
2175 * block from the directory.
2177 resblks = XFS_REMOVE_SPACE_RES(mp);
2178 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2179 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2180 if (error == ENOSPC) {
2182 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2183 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2186 ASSERT(error != ENOSPC);
2187 REMOVE_DEBUG_TRACE(__LINE__);
2188 xfs_trans_cancel(tp, 0);
2192 error = xfs_lock_dir_and_entry(dp, ip);
2194 REMOVE_DEBUG_TRACE(__LINE__);
2195 xfs_trans_cancel(tp, cancel_flags);
2200 * At this point, we've gotten both the directory and the entry
2204 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2207 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2210 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2212 XFS_BMAP_INIT(&free_list, &first_block);
2213 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2214 &first_block, &free_list, resblks);
2216 ASSERT(error != ENOENT);
2217 REMOVE_DEBUG_TRACE(__LINE__);
2220 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2223 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2225 error = xfs_droplink(tp, ip);
2227 REMOVE_DEBUG_TRACE(__LINE__);
2231 /* Determine if this is the last link while
2232 * we are in the transaction.
2234 link_zero = (ip)->i_d.di_nlink==0;
2237 * If this is a synchronous mount, make sure that the
2238 * remove transaction goes to disk before returning to
2241 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2242 xfs_trans_set_sync(tp);
2245 error = xfs_bmap_finish(&tp, &free_list, &committed);
2247 REMOVE_DEBUG_TRACE(__LINE__);
2251 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2256 * If we are using filestreams, kill the stream association.
2257 * If the file is still open it may get a new one but that
2258 * will get killed on last close in xfs_close() so we don't
2259 * have to worry about that.
2261 if (link_zero && xfs_inode_is_filestream(ip))
2262 xfs_filestream_deassociate(ip);
2264 xfs_itrace_exit(ip);
2266 /* Fall through to std_return with error = 0 */
2268 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
2269 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2271 NULL, DM_RIGHT_NULL,
2272 name->name, NULL, ip->i_d.di_mode, error, 0);
2277 xfs_bmap_cancel(&free_list);
2278 cancel_flags |= XFS_TRANS_ABORT;
2279 xfs_trans_cancel(tp, cancel_flags);
2284 * In this case make sure to not release the inode until after
2285 * the current transaction is aborted. Releasing it beforehand
2286 * can cause us to go to xfs_inactive and start a recursive
2287 * transaction which can easily deadlock with the current one.
2289 xfs_bmap_cancel(&free_list);
2290 cancel_flags |= XFS_TRANS_ABORT;
2291 xfs_trans_cancel(tp, cancel_flags);
2300 struct xfs_name *target_name)
2302 xfs_mount_t *mp = tdp->i_mount;
2304 xfs_inode_t *ips[2];
2306 xfs_bmap_free_t free_list;
2307 xfs_fsblock_t first_block;
2312 xfs_itrace_entry(tdp);
2313 xfs_itrace_entry(sip);
2315 ASSERT(!S_ISDIR(sip->i_d.di_mode));
2317 if (XFS_FORCED_SHUTDOWN(mp))
2318 return XFS_ERROR(EIO);
2320 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
2321 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2324 target_name->name, NULL, 0, 0, 0);
2329 /* Return through std_return after this point. */
2331 error = XFS_QM_DQATTACH(mp, sip, 0);
2332 if (!error && sip != tdp)
2333 error = XFS_QM_DQATTACH(mp, tdp, 0);
2337 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2338 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2339 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
2340 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2341 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2342 if (error == ENOSPC) {
2344 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2345 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2352 if (sip->i_ino < tdp->i_ino) {
2360 xfs_lock_inodes(ips, 2, XFS_ILOCK_EXCL);
2363 * Increment vnode ref counts since xfs_trans_commit &
2364 * xfs_trans_cancel will both unlock the inodes and
2365 * decrement the associated ref counts.
2369 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2370 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2373 * If the source has too many links, we can't make any more to it.
2375 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2376 error = XFS_ERROR(EMLINK);
2381 * If we are using project inheritance, we only allow hard link
2382 * creation in our tree when the project IDs are the same; else
2383 * the tree quota mechanism could be circumvented.
2385 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2386 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2387 error = XFS_ERROR(EXDEV);
2391 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
2395 XFS_BMAP_INIT(&free_list, &first_block);
2397 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
2398 &first_block, &free_list, resblks);
2401 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2403 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2405 error = xfs_bumplink(tp, sip);
2410 * If this is a synchronous mount, make sure that the
2411 * link transaction goes to disk before returning to
2414 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2415 xfs_trans_set_sync(tp);
2418 error = xfs_bmap_finish (&tp, &free_list, &committed);
2420 xfs_bmap_cancel(&free_list);
2424 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2428 /* Fall through to std_return with error = 0. */
2430 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
2431 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2434 target_name->name, NULL, 0, error, 0);
2439 cancel_flags |= XFS_TRANS_ABORT;
2443 xfs_trans_cancel(tp, cancel_flags);
2451 struct xfs_name *dir_name,
2456 xfs_mount_t *mp = dp->i_mount;
2457 xfs_inode_t *cdp; /* inode of created dir */
2462 xfs_bmap_free_t free_list;
2463 xfs_fsblock_t first_block;
2464 boolean_t unlock_dp_on_error = B_FALSE;
2465 boolean_t created = B_FALSE;
2466 int dm_event_sent = 0;
2468 struct xfs_dquot *udqp, *gdqp;
2471 if (XFS_FORCED_SHUTDOWN(mp))
2472 return XFS_ERROR(EIO);
2476 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
2477 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2478 dp, DM_RIGHT_NULL, NULL,
2479 DM_RIGHT_NULL, dir_name->name, NULL,
2486 /* Return through std_return after this point. */
2488 xfs_itrace_entry(dp);
2492 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2493 prid = dp->i_d.di_projid;
2495 prid = (xfs_prid_t)dfltprid;
2498 * Make sure that we have allocated dquot(s) on disk.
2500 error = XFS_QM_DQVOPALLOC(mp, dp,
2501 current_fsuid(credp), current_fsgid(credp), prid,
2502 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2506 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2507 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2508 resblks = XFS_MKDIR_SPACE_RES(mp, dir_name->len);
2509 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2510 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2511 if (error == ENOSPC) {
2513 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2514 XFS_TRANS_PERM_LOG_RES,
2515 XFS_MKDIR_LOG_COUNT);
2522 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2523 unlock_dp_on_error = B_TRUE;
2526 * Check for directory link count overflow.
2528 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2529 error = XFS_ERROR(EMLINK);
2534 * Reserve disk quota and the inode.
2536 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2540 error = xfs_dir_canenter(tp, dp, dir_name, resblks);
2544 * create the directory inode.
2546 error = xfs_dir_ialloc(&tp, dp, mode, 2,
2547 0, credp, prid, resblks > 0,
2550 if (error == ENOSPC)
2554 xfs_itrace_ref(cdp);
2557 * Now we add the directory inode to the transaction.
2558 * We waited until now since xfs_dir_ialloc might start
2559 * a new transaction. Had we joined the transaction
2560 * earlier, the locks might have gotten released. An error
2561 * from here on will result in the transaction cancel
2562 * unlocking dp so don't do it explicitly in the error path.
2565 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2566 unlock_dp_on_error = B_FALSE;
2568 XFS_BMAP_INIT(&free_list, &first_block);
2570 error = xfs_dir_createname(tp, dp, dir_name, cdp->i_ino,
2571 &first_block, &free_list, resblks ?
2572 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2574 ASSERT(error != ENOSPC);
2577 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2580 * Bump the in memory version number of the parent directory
2581 * so that other processes accessing it will recognize that
2582 * the directory has changed.
2586 error = xfs_dir_init(tp, cdp, dp);
2591 error = xfs_bumplink(tp, dp);
2601 * Attach the dquots to the new inode and modify the icount incore.
2603 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2606 * If this is a synchronous mount, make sure that the
2607 * mkdir transaction goes to disk before returning to
2610 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2611 xfs_trans_set_sync(tp);
2614 error = xfs_bmap_finish(&tp, &free_list, &committed);
2620 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2621 XFS_QM_DQRELE(mp, udqp);
2622 XFS_QM_DQRELE(mp, gdqp);
2627 /* Fall through to std_return with error = 0 or errno from
2628 * xfs_trans_commit. */
2631 if ((created || (error != 0 && dm_event_sent != 0)) &&
2632 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
2633 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2635 created ? cdp : NULL,
2637 dir_name->name, NULL,
2644 xfs_bmap_cancel(&free_list);
2646 cancel_flags |= XFS_TRANS_ABORT;
2648 xfs_trans_cancel(tp, cancel_flags);
2649 XFS_QM_DQRELE(mp, udqp);
2650 XFS_QM_DQRELE(mp, gdqp);
2652 if (unlock_dp_on_error)
2653 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2661 struct xfs_name *name,
2664 xfs_mount_t *mp = dp->i_mount;
2667 xfs_bmap_free_t free_list;
2668 xfs_fsblock_t first_block;
2674 xfs_itrace_entry(dp);
2676 if (XFS_FORCED_SHUTDOWN(mp))
2677 return XFS_ERROR(EIO);
2679 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
2680 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
2682 NULL, DM_RIGHT_NULL, name->name,
2683 NULL, cdp->i_d.di_mode, 0, 0);
2685 return XFS_ERROR(error);
2689 * Get the dquots for the inodes.
2691 error = XFS_QM_DQATTACH(mp, dp, 0);
2693 error = XFS_QM_DQATTACH(mp, cdp, 0);
2695 REMOVE_DEBUG_TRACE(__LINE__);
2699 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2700 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2702 * We try to get the real space reservation first,
2703 * allowing for directory btree deletion(s) implying
2704 * possible bmap insert(s). If we can't get the space
2705 * reservation then we use 0 instead, and avoid the bmap
2706 * btree insert(s) in the directory code by, if the bmap
2707 * insert tries to happen, instead trimming the LAST
2708 * block from the directory.
2710 resblks = XFS_REMOVE_SPACE_RES(mp);
2711 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2712 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
2713 if (error == ENOSPC) {
2715 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2716 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
2719 ASSERT(error != ENOSPC);
2723 XFS_BMAP_INIT(&free_list, &first_block);
2726 * Now lock the child directory inode and the parent directory
2727 * inode in the proper order. This will take care of validating
2728 * that the directory entry for the child directory inode has
2729 * not changed while we were obtaining a log reservation.
2731 error = xfs_lock_dir_and_entry(dp, cdp);
2733 xfs_trans_cancel(tp, cancel_flags);
2738 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2741 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
2743 ASSERT(cdp->i_d.di_nlink >= 2);
2744 if (cdp->i_d.di_nlink != 2) {
2745 error = XFS_ERROR(ENOTEMPTY);
2748 if (!xfs_dir_isempty(cdp)) {
2749 error = XFS_ERROR(ENOTEMPTY);
2753 error = xfs_dir_removename(tp, dp, name, cdp->i_ino,
2754 &first_block, &free_list, resblks);
2758 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2761 * Bump the in memory generation count on the parent
2762 * directory so that other can know that it has changed.
2767 * Drop the link from cdp's "..".
2769 error = xfs_droplink(tp, dp);
2775 * Drop the link from dp to cdp.
2777 error = xfs_droplink(tp, cdp);
2783 * Drop the "." link from cdp to self.
2785 error = xfs_droplink(tp, cdp);
2790 /* Determine these before committing transaction */
2791 last_cdp_link = (cdp)->i_d.di_nlink==0;
2794 * If this is a synchronous mount, make sure that the
2795 * rmdir transaction goes to disk before returning to
2798 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2799 xfs_trans_set_sync(tp);
2802 error = xfs_bmap_finish (&tp, &free_list, &committed);
2804 xfs_bmap_cancel(&free_list);
2805 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2810 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2816 /* Fall through to std_return with error = 0 or the errno
2817 * from xfs_trans_commit. */
2819 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
2820 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2822 NULL, DM_RIGHT_NULL,
2823 name->name, NULL, cdp->i_d.di_mode,
2829 xfs_bmap_cancel(&free_list);
2830 cancel_flags |= XFS_TRANS_ABORT;
2834 xfs_trans_cancel(tp, cancel_flags);
2841 struct xfs_name *link_name,
2842 const char *target_path,
2847 xfs_mount_t *mp = dp->i_mount;
2852 xfs_bmap_free_t free_list;
2853 xfs_fsblock_t first_block;
2854 boolean_t unlock_dp_on_error = B_FALSE;
2857 xfs_fileoff_t first_fsb;
2858 xfs_filblks_t fs_blocks;
2860 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
2862 const char *cur_chunk;
2867 struct xfs_dquot *udqp, *gdqp;
2875 xfs_itrace_entry(dp);
2877 if (XFS_FORCED_SHUTDOWN(mp))
2878 return XFS_ERROR(EIO);
2881 * Check component lengths of the target path name.
2883 pathlen = strlen(target_path);
2884 if (pathlen >= MAXPATHLEN) /* total string too long */
2885 return XFS_ERROR(ENAMETOOLONG);
2887 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
2888 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
2889 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2890 link_name->name, target_path, 0, 0, 0);
2895 /* Return through std_return after this point. */
2898 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2899 prid = dp->i_d.di_projid;
2901 prid = (xfs_prid_t)dfltprid;
2904 * Make sure that we have allocated dquot(s) on disk.
2906 error = XFS_QM_DQVOPALLOC(mp, dp,
2907 current_fsuid(credp), current_fsgid(credp), prid,
2908 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2912 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
2913 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2915 * The symlink will fit into the inode data fork?
2916 * There can't be any attributes so we get the whole variable part.
2918 if (pathlen <= XFS_LITINO(mp))
2921 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
2922 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
2923 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2924 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2925 if (error == ENOSPC && fs_blocks == 0) {
2927 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2928 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2935 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2936 unlock_dp_on_error = B_TRUE;
2939 * Check whether the directory allows new symlinks or not.
2941 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2942 error = XFS_ERROR(EPERM);
2947 * Reserve disk quota : blocks and inode.
2949 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2954 * Check for ability to enter directory entry, if no space reserved.
2956 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2960 * Initialize the bmap freelist prior to calling either
2961 * bmapi or the directory create code.
2963 XFS_BMAP_INIT(&free_list, &first_block);
2966 * Allocate an inode for the symlink.
2968 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
2969 1, 0, credp, prid, resblks > 0, &ip, NULL);
2971 if (error == ENOSPC)
2978 * An error after we've joined dp to the transaction will result in the
2979 * transaction cancel unlocking dp so don't do it explicitly in the
2983 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2984 unlock_dp_on_error = B_FALSE;
2987 * Also attach the dquot(s) to it, if applicable.
2989 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2992 resblks -= XFS_IALLOC_SPACE_RES(mp);
2994 * If the symlink will fit into the inode, write it inline.
2996 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2997 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2998 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2999 ip->i_d.di_size = pathlen;
3002 * The inode was initially created in extent format.
3004 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3005 ip->i_df.if_flags |= XFS_IFINLINE;
3007 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3008 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3012 nmaps = SYMLINK_MAPS;
3014 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3015 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3016 &first_block, resblks, mval, &nmaps,
3023 resblks -= fs_blocks;
3024 ip->i_d.di_size = pathlen;
3025 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3027 cur_chunk = target_path;
3028 for (n = 0; n < nmaps; n++) {
3029 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3030 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3031 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3032 BTOBB(byte_cnt), 0);
3033 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3034 if (pathlen < byte_cnt) {
3037 pathlen -= byte_cnt;
3039 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3040 cur_chunk += byte_cnt;
3042 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3047 * Create the directory entry for the symlink.
3049 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
3050 &first_block, &free_list, resblks);
3053 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3054 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3057 * Bump the in memory version number of the parent directory
3058 * so that other processes accessing it will recognize that
3059 * the directory has changed.
3064 * If this is a synchronous mount, make sure that the
3065 * symlink transaction goes to disk before returning to
3068 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3069 xfs_trans_set_sync(tp);
3073 * xfs_trans_commit normally decrements the vnode ref count
3074 * when it unlocks the inode. Since we want to return the
3075 * vnode to the caller, we bump the vnode ref count now.
3079 error = xfs_bmap_finish(&tp, &free_list, &committed);
3083 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3084 XFS_QM_DQRELE(mp, udqp);
3085 XFS_QM_DQRELE(mp, gdqp);
3087 /* Fall through to std_return with error = 0 or errno from
3088 * xfs_trans_commit */
3090 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
3091 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3094 DM_RIGHT_NULL, link_name->name,
3095 target_path, 0, error, 0);
3105 xfs_bmap_cancel(&free_list);
3106 cancel_flags |= XFS_TRANS_ABORT;
3108 xfs_trans_cancel(tp, cancel_flags);
3109 XFS_QM_DQRELE(mp, udqp);
3110 XFS_QM_DQRELE(mp, gdqp);
3112 if (unlock_dp_on_error)
3113 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3123 xfs_mount_t *mp = ip->i_mount;
3126 if (XFS_FORCED_SHUTDOWN(mp))
3127 return XFS_ERROR(EIO);
3130 * Bypass inodes which have already been cleaned by
3131 * the inode flush clustering code inside xfs_iflush
3133 if (xfs_inode_clean(ip))
3137 * We make this non-blocking if the inode is contended,
3138 * return EAGAIN to indicate to the caller that they
3139 * did not succeed. This prevents the flush path from
3140 * blocking on inodes inside another operation right
3141 * now, they get caught later by xfs_sync.
3143 if (flags & FLUSH_SYNC) {
3144 xfs_ilock(ip, XFS_ILOCK_SHARED);
3146 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3147 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3148 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3155 error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC
3156 : XFS_IFLUSH_ASYNC_NOBLOCK);
3157 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3169 xfs_mount_t *mp = ip->i_mount;
3173 if (!capable(CAP_SYS_ADMIN))
3174 return XFS_ERROR(EPERM);
3176 if (XFS_FORCED_SHUTDOWN(mp))
3177 return XFS_ERROR(EIO);
3179 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3180 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3182 xfs_trans_cancel(tp, 0);
3185 xfs_ilock(ip, XFS_ILOCK_EXCL);
3186 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3188 ip->i_d.di_dmevmask = evmask;
3189 ip->i_d.di_dmstate = state;
3191 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3193 error = xfs_trans_commit(tp, 0);
3202 bhv_vnode_t *vp = XFS_ITOV(ip);
3204 xfs_itrace_entry(ip);
3206 ASSERT(!VN_MAPPED(vp));
3208 /* bad inode, get out here ASAP */
3216 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3219 * Make sure the atime in the XFS inode is correct before freeing the
3222 xfs_synchronize_atime(ip);
3225 * If we have nothing to flush with this inode then complete the
3226 * teardown now, otherwise break the link between the xfs inode and the
3227 * linux inode and clean up the xfs inode later. This avoids flushing
3228 * the inode to disk during the delete operation itself.
3230 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3231 * first to ensure that xfs_iunpin() will never see an xfs inode
3232 * that has a linux inode being reclaimed. Synchronisation is provided
3233 * by the i_flags_lock.
3235 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3236 xfs_ilock(ip, XFS_ILOCK_EXCL);
3238 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3240 xfs_mount_t *mp = ip->i_mount;
3242 /* Protect sync and unpin from us */
3243 XFS_MOUNT_ILOCK(mp);
3244 spin_lock(&ip->i_flags_lock);
3245 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
3246 vn_to_inode(vp)->i_private = NULL;
3248 spin_unlock(&ip->i_flags_lock);
3249 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3250 XFS_MOUNT_IUNLOCK(mp);
3261 xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino);
3262 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
3265 if (vp && VN_BAD(vp))
3268 /* The hash lock here protects a thread in xfs_iget_core from
3269 * racing with us on linking the inode back with a vnode.
3270 * Once we have the XFS_IRECLAIM flag set it will not touch
3273 write_lock(&pag->pag_ici_lock);
3274 spin_lock(&ip->i_flags_lock);
3275 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3276 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
3277 spin_unlock(&ip->i_flags_lock);
3278 write_unlock(&pag->pag_ici_lock);
3281 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3285 __xfs_iflags_set(ip, XFS_IRECLAIM);
3286 spin_unlock(&ip->i_flags_lock);
3287 write_unlock(&pag->pag_ici_lock);
3288 xfs_put_perag(ip->i_mount, pag);
3291 * If the inode is still dirty, then flush it out. If the inode
3292 * is not in the AIL, then it will be OK to flush it delwri as
3293 * long as xfs_iflush() does not keep any references to the inode.
3294 * We leave that decision up to xfs_iflush() since it has the
3295 * knowledge of whether it's OK to simply do a delwri flush of
3296 * the inode or whether we need to wait until the inode is
3297 * pulled from the AIL.
3298 * We get the flush lock regardless, though, just to make sure
3299 * we don't free it while it is being flushed.
3302 xfs_ilock(ip, XFS_ILOCK_EXCL);
3306 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3307 if (ip->i_update_core ||
3308 ((ip->i_itemp != NULL) &&
3309 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3310 error = xfs_iflush(ip, sync_mode);
3312 * If we hit an error, typically because of filesystem
3313 * shutdown, we don't need to let vn_reclaim to know
3314 * because we're gonna reclaim the inode anyway.
3317 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3320 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3323 ASSERT(ip->i_update_core == 0);
3324 ASSERT(ip->i_itemp == NULL ||
3325 ip->i_itemp->ili_format.ilf_fields == 0);
3329 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3337 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3340 xfs_inode_t *ip, *n;
3345 XFS_MOUNT_ILOCK(mp);
3346 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3348 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3350 if (xfs_ipincount(ip) ||
3351 !xfs_iflock_nowait(ip)) {
3352 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3356 XFS_MOUNT_IUNLOCK(mp);
3357 if (xfs_finish_reclaim(ip, noblock,
3358 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3367 XFS_MOUNT_IUNLOCK(mp);
3372 * xfs_alloc_file_space()
3373 * This routine allocates disk space for the given file.
3375 * If alloc_type == 0, this request is for an ALLOCSP type
3376 * request which will change the file size. In this case, no
3377 * DMAPI event will be generated by the call. A TRUNCATE event
3378 * will be generated later by xfs_setattr.
3380 * If alloc_type != 0, this request is for a RESVSP type
3381 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3382 * lower block boundary byte address is less than the file's
3391 xfs_alloc_file_space(
3398 xfs_mount_t *mp = ip->i_mount;
3400 xfs_filblks_t allocated_fsb;
3401 xfs_filblks_t allocatesize_fsb;
3402 xfs_extlen_t extsz, temp;
3403 xfs_fileoff_t startoffset_fsb;
3404 xfs_fsblock_t firstfsb;
3410 xfs_bmbt_irec_t imaps[1], *imapp;
3411 xfs_bmap_free_t free_list;
3412 uint qblocks, resblks, resrtextents;
3416 xfs_itrace_entry(ip);
3418 if (XFS_FORCED_SHUTDOWN(mp))
3419 return XFS_ERROR(EIO);
3421 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3425 return XFS_ERROR(EINVAL);
3427 rt = XFS_IS_REALTIME_INODE(ip);
3428 extsz = xfs_get_extsz_hint(ip);
3433 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
3434 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
3435 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
3437 /* Generate a DMAPI event if needed. */
3438 if (alloc_type != 0 && offset < ip->i_size &&
3439 (attr_flags&ATTR_DMI) == 0 &&
3440 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
3441 xfs_off_t end_dmi_offset;
3443 end_dmi_offset = offset+len;
3444 if (end_dmi_offset > ip->i_size)
3445 end_dmi_offset = ip->i_size;
3446 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
3447 end_dmi_offset - offset, 0, NULL);
3453 * Allocate file space until done or until there is an error
3456 while (allocatesize_fsb && !error) {
3460 * Determine space reservations for data/realtime.
3462 if (unlikely(extsz)) {
3463 s = startoffset_fsb;
3466 e = startoffset_fsb + allocatesize_fsb;
3467 if ((temp = do_mod(startoffset_fsb, extsz)))
3469 if ((temp = do_mod(e, extsz)))
3473 e = allocatesize_fsb;
3477 resrtextents = qblocks = (uint)(e - s);
3478 resrtextents /= mp->m_sb.sb_rextsize;
3479 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3480 quota_flag = XFS_QMOPT_RES_RTBLKS;
3483 resblks = qblocks = \
3484 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
3485 quota_flag = XFS_QMOPT_RES_REGBLKS;
3489 * Allocate and setup the transaction.
3491 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
3492 error = xfs_trans_reserve(tp, resblks,
3493 XFS_WRITE_LOG_RES(mp), resrtextents,
3494 XFS_TRANS_PERM_LOG_RES,
3495 XFS_WRITE_LOG_COUNT);
3497 * Check for running out of space
3501 * Free the transaction structure.
3503 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3504 xfs_trans_cancel(tp, 0);
3507 xfs_ilock(ip, XFS_ILOCK_EXCL);
3508 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
3509 qblocks, 0, quota_flag);
3513 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3514 xfs_trans_ihold(tp, ip);
3517 * Issue the xfs_bmapi() call to allocate the blocks
3519 XFS_BMAP_INIT(&free_list, &firstfsb);
3520 error = xfs_bmapi(tp, ip, startoffset_fsb,
3521 allocatesize_fsb, bmapi_flag,
3522 &firstfsb, 0, imapp, &nimaps,
3529 * Complete the transaction
3531 error = xfs_bmap_finish(&tp, &free_list, &committed);
3536 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3537 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3542 allocated_fsb = imapp->br_blockcount;
3545 error = XFS_ERROR(ENOSPC);
3549 startoffset_fsb += allocated_fsb;
3550 allocatesize_fsb -= allocated_fsb;
3553 if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 &&
3554 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
3555 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
3558 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
3560 goto retry; /* Maybe DMAPI app. has made space */
3561 /* else fall through with error from XFS_SEND_DATA */
3566 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
3567 xfs_bmap_cancel(&free_list);
3568 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
3570 error1: /* Just cancel transaction */
3571 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3572 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3573 goto dmapi_enospc_check;
3577 * Zero file bytes between startoff and endoff inclusive.
3578 * The iolock is held exclusive and no blocks are buffered.
3581 xfs_zero_remaining_bytes(
3586 xfs_bmbt_irec_t imap;
3587 xfs_fileoff_t offset_fsb;
3588 xfs_off_t lastoffset;
3591 xfs_mount_t *mp = ip->i_mount;
3595 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
3596 XFS_IS_REALTIME_INODE(ip) ?
3597 mp->m_rtdev_targp : mp->m_ddev_targp);
3599 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
3600 offset_fsb = XFS_B_TO_FSBT(mp, offset);
3602 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
3603 NULL, 0, &imap, &nimap, NULL, NULL);
3604 if (error || nimap < 1)
3606 ASSERT(imap.br_blockcount >= 1);
3607 ASSERT(imap.br_startoff == offset_fsb);
3608 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
3609 if (lastoffset > endoff)
3610 lastoffset = endoff;
3611 if (imap.br_startblock == HOLESTARTBLOCK)
3613 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3614 if (imap.br_state == XFS_EXT_UNWRITTEN)
3617 XFS_BUF_UNWRITE(bp);
3619 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
3621 error = xfs_iowait(bp);
3623 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
3624 mp, bp, XFS_BUF_ADDR(bp));
3627 memset(XFS_BUF_PTR(bp) +
3628 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
3629 0, lastoffset - offset + 1);
3634 error = xfs_iowait(bp);
3636 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
3637 mp, bp, XFS_BUF_ADDR(bp));
3646 * xfs_free_file_space()
3647 * This routine frees disk space for the given file.
3649 * This routine is only called by xfs_change_file_space
3650 * for an UNRESVSP type call.
3658 xfs_free_file_space(
3667 xfs_off_t end_dmi_offset;
3668 xfs_fileoff_t endoffset_fsb;
3670 xfs_fsblock_t firstfsb;
3671 xfs_bmap_free_t free_list;
3672 xfs_bmbt_irec_t imap;
3680 xfs_fileoff_t startoffset_fsb;
3682 int need_iolock = 1;
3687 xfs_itrace_entry(ip);
3689 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3693 if (len <= 0) /* if nothing being freed */
3695 rt = XFS_IS_REALTIME_INODE(ip);
3696 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
3697 end_dmi_offset = offset + len;
3698 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
3700 if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 &&
3701 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
3702 if (end_dmi_offset > ip->i_size)
3703 end_dmi_offset = ip->i_size;
3704 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
3705 offset, end_dmi_offset - offset,
3706 AT_DELAY_FLAG(attr_flags), NULL);
3711 if (attr_flags & ATTR_NOLOCK)
3714 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3715 vn_iowait(ip); /* wait for the completion of any pending DIOs */
3718 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
3719 ioffset = offset & ~(rounding - 1);
3721 if (VN_CACHED(vp) != 0) {
3722 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
3723 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
3725 goto out_unlock_iolock;
3729 * Need to zero the stuff we're not freeing, on disk.
3730 * If its a realtime file & can't use unwritten extents then we
3731 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3732 * will take care of it for us.
3734 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
3736 error = xfs_bmapi(NULL, ip, startoffset_fsb,
3737 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
3739 goto out_unlock_iolock;
3740 ASSERT(nimap == 0 || nimap == 1);
3741 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3744 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3745 block = imap.br_startblock;
3746 mod = do_div(block, mp->m_sb.sb_rextsize);
3748 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
3751 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
3752 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
3754 goto out_unlock_iolock;
3755 ASSERT(nimap == 0 || nimap == 1);
3756 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3757 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3759 if (mod && (mod != mp->m_sb.sb_rextsize))
3760 endoffset_fsb -= mod;
3763 if ((done = (endoffset_fsb <= startoffset_fsb)))
3765 * One contiguous piece to clear
3767 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
3770 * Some full blocks, possibly two pieces to clear
3772 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
3773 error = xfs_zero_remaining_bytes(ip, offset,
3774 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
3776 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
3777 error = xfs_zero_remaining_bytes(ip,
3778 XFS_FSB_TO_B(mp, endoffset_fsb),
3783 * free file space until done or until there is an error
3785 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3786 while (!error && !done) {
3789 * allocate and setup the transaction. Allow this
3790 * transaction to dip into the reserve blocks to ensure
3791 * the freeing of the space succeeds at ENOSPC.
3793 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
3794 tp->t_flags |= XFS_TRANS_RESERVE;
3795 error = xfs_trans_reserve(tp,
3797 XFS_WRITE_LOG_RES(mp),
3799 XFS_TRANS_PERM_LOG_RES,
3800 XFS_WRITE_LOG_COUNT);
3803 * check for running out of space
3807 * Free the transaction structure.
3809 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3810 xfs_trans_cancel(tp, 0);
3813 xfs_ilock(ip, XFS_ILOCK_EXCL);
3814 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
3815 ip->i_udquot, ip->i_gdquot, resblks, 0,
3816 XFS_QMOPT_RES_REGBLKS);
3820 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3821 xfs_trans_ihold(tp, ip);
3824 * issue the bunmapi() call to free the blocks
3826 XFS_BMAP_INIT(&free_list, &firstfsb);
3827 error = xfs_bunmapi(tp, ip, startoffset_fsb,
3828 endoffset_fsb - startoffset_fsb,
3829 0, 2, &firstfsb, &free_list, NULL, &done);
3835 * complete the transaction
3837 error = xfs_bmap_finish(&tp, &free_list, &committed);
3842 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3843 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3848 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3852 xfs_bmap_cancel(&free_list);
3854 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3855 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
3861 * xfs_change_file_space()
3862 * This routine allocates or frees disk space for the given file.
3863 * The user specified parameters are checked for alignment and size
3872 xfs_change_file_space(
3880 xfs_mount_t *mp = ip->i_mount;
3885 xfs_off_t startoffset;
3890 xfs_itrace_entry(ip);
3892 if (!S_ISREG(ip->i_d.di_mode))
3893 return XFS_ERROR(EINVAL);
3895 switch (bf->l_whence) {
3896 case 0: /*SEEK_SET*/
3898 case 1: /*SEEK_CUR*/
3899 bf->l_start += offset;
3901 case 2: /*SEEK_END*/
3902 bf->l_start += ip->i_size;
3905 return XFS_ERROR(EINVAL);
3908 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
3910 if ( (bf->l_start < 0)
3911 || (bf->l_start > XFS_MAXIOFFSET(mp))
3912 || (bf->l_start + llen < 0)
3913 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
3914 return XFS_ERROR(EINVAL);
3918 startoffset = bf->l_start;
3922 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3924 * These calls do NOT zero the data space allocated to the file,
3925 * nor do they change the file size.
3927 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3929 * These calls cause the new file data to be zeroed and the file
3930 * size to be changed.
3932 setprealloc = clrprealloc = 0;
3935 case XFS_IOC_RESVSP:
3936 case XFS_IOC_RESVSP64:
3937 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
3944 case XFS_IOC_UNRESVSP:
3945 case XFS_IOC_UNRESVSP64:
3946 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
3951 case XFS_IOC_ALLOCSP:
3952 case XFS_IOC_ALLOCSP64:
3953 case XFS_IOC_FREESP:
3954 case XFS_IOC_FREESP64:
3955 if (startoffset > fsize) {
3956 error = xfs_alloc_file_space(ip, fsize,
3957 startoffset - fsize, 0, attr_flags);
3962 va.va_mask = XFS_AT_SIZE;
3963 va.va_size = startoffset;
3965 error = xfs_setattr(ip, &va, attr_flags, credp);
3975 return XFS_ERROR(EINVAL);
3979 * update the inode timestamp, mode, and prealloc flag bits
3981 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
3983 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
3986 xfs_trans_cancel(tp, 0);
3990 xfs_ilock(ip, XFS_ILOCK_EXCL);
3992 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3993 xfs_trans_ihold(tp, ip);
3995 if ((attr_flags & ATTR_DMI) == 0) {
3996 ip->i_d.di_mode &= ~S_ISUID;
3999 * Note that we don't have to worry about mandatory
4000 * file locking being disabled here because we only
4001 * clear the S_ISGID bit if the Group execute bit is
4002 * on, but if it was on then mandatory locking wouldn't
4003 * have been enabled.
4005 if (ip->i_d.di_mode & S_IXGRP)
4006 ip->i_d.di_mode &= ~S_ISGID;
4008 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4011 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4012 else if (clrprealloc)
4013 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4015 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4016 xfs_trans_set_sync(tp);
4018 error = xfs_trans_commit(tp, 0);
4020 xfs_iunlock(ip, XFS_ILOCK_EXCL);