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 xfs_mount_t *mp = ip->i_mount;
63 struct inode *inode = VFS_I(ip);
64 int mask = iattr->ia_valid;
72 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
78 if (mp->m_flags & XFS_MOUNT_RDONLY)
79 return XFS_ERROR(EROFS);
81 if (XFS_FORCED_SHUTDOWN(mp))
82 return XFS_ERROR(EIO);
84 olddquot1 = olddquot2 = NULL;
88 * If disk quotas is on, we make sure that the dquots do exist on disk,
89 * before we start any other transactions. Trying to do this later
90 * is messy. We don't care to take a readlock to look at the ids
91 * in inode here, because we can't hold it across the trans_reserve.
92 * If the IDs do change before we take the ilock, we're covered
93 * because the i_*dquot fields will get updated anyway.
95 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
98 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
100 qflags |= XFS_QMOPT_UQUOTA;
102 uid = ip->i_d.di_uid;
104 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
106 qflags |= XFS_QMOPT_GQUOTA;
108 gid = ip->i_d.di_gid;
112 * We take a reference when we initialize udqp and gdqp,
113 * so it is important that we never blindly double trip on
114 * the same variable. See xfs_create() for an example.
116 ASSERT(udqp == NULL);
117 ASSERT(gdqp == NULL);
118 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, ip->i_d.di_projid,
119 qflags, &udqp, &gdqp);
125 * For the other attributes, we acquire the inode lock and
126 * first do an error checking pass.
129 lock_flags = XFS_ILOCK_EXCL;
130 if (flags & XFS_ATTR_NOLOCK)
132 if (!(mask & ATTR_SIZE)) {
133 if ((mask != (ATTR_CTIME|ATTR_ATIME|ATTR_MTIME)) ||
134 (mp->m_flags & XFS_MOUNT_WSYNC)) {
135 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
137 if ((code = xfs_trans_reserve(tp, 0,
138 XFS_ICHANGE_LOG_RES(mp), 0,
145 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
146 !(flags & XFS_ATTR_DMI)) {
147 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
148 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip,
149 iattr->ia_size, 0, dmflags, NULL);
156 lock_flags |= XFS_IOLOCK_EXCL;
159 xfs_ilock(ip, lock_flags);
161 /* boolean: are we the file owner? */
162 file_owner = (current_fsuid() == ip->i_d.di_uid);
165 * Change various properties of a file.
166 * Only the owner or users with CAP_FOWNER
167 * capability may do these things.
169 if (mask & (ATTR_MODE|ATTR_UID|ATTR_GID)) {
171 * CAP_FOWNER overrides the following restrictions:
173 * The user ID of the calling process must be equal
174 * to the file owner ID, except in cases where the
175 * CAP_FSETID capability is applicable.
177 if (!file_owner && !capable(CAP_FOWNER)) {
178 code = XFS_ERROR(EPERM);
183 * CAP_FSETID overrides the following restrictions:
185 * The effective user ID of the calling process shall match
186 * the file owner when setting the set-user-ID and
187 * set-group-ID bits on that file.
189 * The effective group ID or one of the supplementary group
190 * IDs of the calling process shall match the group owner of
191 * the file when setting the set-group-ID bit on that file
193 if (mask & ATTR_MODE) {
196 if ((iattr->ia_mode & S_ISUID) && !file_owner)
198 if ((iattr->ia_mode & S_ISGID) &&
199 !in_group_p((gid_t)ip->i_d.di_gid))
202 /* Linux allows this, Irix doesn't. */
203 if ((iattr->ia_mode & S_ISVTX) && !S_ISDIR(ip->i_d.di_mode))
206 if (m && !capable(CAP_FSETID))
207 iattr->ia_mode &= ~m;
212 * Change file ownership. Must be the owner or privileged.
214 if (mask & (ATTR_UID|ATTR_GID)) {
216 * These IDs could have changed since we last looked at them.
217 * But, we're assured that if the ownership did change
218 * while we didn't have the inode locked, inode's dquot(s)
219 * would have changed also.
221 iuid = ip->i_d.di_uid;
222 igid = ip->i_d.di_gid;
223 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
224 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
227 * CAP_CHOWN overrides the following restrictions:
229 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
230 * shall override the restriction that a process cannot
231 * change the user ID of a file it owns and the restriction
232 * that the group ID supplied to the chown() function
233 * shall be equal to either the group ID or one of the
234 * supplementary group IDs of the calling process.
237 (igid != gid && !in_group_p((gid_t)gid))) &&
238 !capable(CAP_CHOWN)) {
239 code = XFS_ERROR(EPERM);
243 * Do a quota reservation only if uid/gid is actually
246 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
247 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
249 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
250 capable(CAP_FOWNER) ?
251 XFS_QMOPT_FORCE_RES : 0);
252 if (code) /* out of quota */
258 * Truncate file. Must have write permission and not be a directory.
260 if (mask & ATTR_SIZE) {
261 /* Short circuit the truncate case for zero length files */
262 if (iattr->ia_size == 0 &&
263 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
264 xfs_iunlock(ip, XFS_ILOCK_EXCL);
265 lock_flags &= ~XFS_ILOCK_EXCL;
266 if (mask & ATTR_CTIME)
267 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
272 if (S_ISDIR(ip->i_d.di_mode)) {
273 code = XFS_ERROR(EISDIR);
275 } else if (!S_ISREG(ip->i_d.di_mode)) {
276 code = XFS_ERROR(EINVAL);
280 * Make sure that the dquots are attached to the inode.
282 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
287 * Change file access or modified times.
289 if (mask & (ATTR_ATIME|ATTR_MTIME)) {
291 if ((mask & (ATTR_MTIME_SET|ATTR_ATIME_SET)) &&
292 !capable(CAP_FOWNER)) {
293 code = XFS_ERROR(EPERM);
300 * Now we can make the changes. Before we join the inode
301 * to the transaction, if ATTR_SIZE is set then take care of
302 * the part of the truncation that must be done without the
303 * inode lock. This needs to be done before joining the inode
304 * to the transaction, because the inode cannot be unlocked
305 * once it is a part of the transaction.
307 if (mask & ATTR_SIZE) {
309 if (iattr->ia_size > ip->i_size) {
311 * Do the first part of growing a file: zero any data
312 * in the last block that is beyond the old EOF. We
313 * need to do this before the inode is joined to the
314 * transaction to modify the i_size.
316 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
318 xfs_iunlock(ip, XFS_ILOCK_EXCL);
321 * We are going to log the inode size change in this
322 * transaction so any previous writes that are beyond the on
323 * disk EOF and the new EOF that have not been written out need
324 * to be written here. If we do not write the data out, we
325 * expose ourselves to the null files problem.
327 * Only flush from the on disk size to the smaller of the in
328 * memory file size or the new size as that's the range we
329 * really care about here and prevents waiting for other data
330 * not within the range we care about here.
333 ip->i_size != ip->i_d.di_size &&
334 iattr->ia_size > ip->i_d.di_size) {
335 code = xfs_flush_pages(ip,
336 ip->i_d.di_size, iattr->ia_size,
337 XFS_B_ASYNC, FI_NONE);
340 /* wait for all I/O to complete */
344 code = xfs_itruncate_data(ip, iattr->ia_size);
347 lock_flags &= ~XFS_ILOCK_EXCL;
348 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
351 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
352 if ((code = xfs_trans_reserve(tp, 0,
353 XFS_ITRUNCATE_LOG_RES(mp), 0,
354 XFS_TRANS_PERM_LOG_RES,
355 XFS_ITRUNCATE_LOG_COUNT))) {
356 xfs_trans_cancel(tp, 0);
358 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
361 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
362 xfs_ilock(ip, XFS_ILOCK_EXCL);
366 xfs_trans_ijoin(tp, ip, lock_flags);
367 xfs_trans_ihold(tp, ip);
371 * Truncate file. Must have write permission and not be a directory.
373 if (mask & ATTR_SIZE) {
375 * Only change the c/mtime if we are changing the size
376 * or we are explicitly asked to change it. This handles
377 * the semantic difference between truncate() and ftruncate()
378 * as implemented in the VFS.
380 if (iattr->ia_size != ip->i_size || (mask & ATTR_CTIME))
381 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
383 if (iattr->ia_size > ip->i_size) {
384 ip->i_d.di_size = iattr->ia_size;
385 ip->i_size = iattr->ia_size;
386 if (!(flags & XFS_ATTR_DMI))
387 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
388 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
389 } else if (iattr->ia_size <= ip->i_size ||
390 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
392 * signal a sync transaction unless
393 * we're truncating an already unlinked
394 * file on a wsync filesystem
396 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
398 ((ip->i_d.di_nlink != 0 ||
399 !(mp->m_flags & XFS_MOUNT_WSYNC))
404 * Truncated "down", so we're removing references
405 * to old data here - if we now delay flushing for
406 * a long time, we expose ourselves unduly to the
407 * notorious NULL files problem. So, we mark this
408 * vnode and flush it when the file is closed, and
409 * do not wait the usual (long) time for writeout.
411 xfs_iflags_set(ip, XFS_ITRUNCATED);
416 * Change file access modes.
418 if (mask & ATTR_MODE) {
419 ip->i_d.di_mode &= S_IFMT;
420 ip->i_d.di_mode |= iattr->ia_mode & ~S_IFMT;
422 inode->i_mode &= S_IFMT;
423 inode->i_mode |= iattr->ia_mode & ~S_IFMT;
425 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
426 timeflags |= XFS_ICHGTIME_CHG;
430 * Change file ownership. Must be the owner or privileged.
432 if (mask & (ATTR_UID|ATTR_GID)) {
434 * CAP_FSETID overrides the following restrictions:
436 * The set-user-ID and set-group-ID bits of a file will be
437 * cleared upon successful return from chown()
439 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
440 !capable(CAP_FSETID)) {
441 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
445 * Change the ownerships and register quota modifications
446 * in the transaction.
449 if (XFS_IS_UQUOTA_ON(mp)) {
450 ASSERT(mask & ATTR_UID);
452 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
453 &ip->i_udquot, udqp);
455 ip->i_d.di_uid = uid;
459 if (XFS_IS_GQUOTA_ON(mp)) {
460 ASSERT(!XFS_IS_PQUOTA_ON(mp));
461 ASSERT(mask & ATTR_GID);
463 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
464 &ip->i_gdquot, gdqp);
466 ip->i_d.di_gid = gid;
470 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
471 timeflags |= XFS_ICHGTIME_CHG;
476 * Change file access or modified times.
478 if (mask & (ATTR_ATIME|ATTR_MTIME)) {
479 if (mask & ATTR_ATIME) {
480 inode->i_atime = iattr->ia_atime;
481 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
482 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
483 ip->i_update_core = 1;
485 if (mask & ATTR_MTIME) {
486 inode->i_mtime = iattr->ia_mtime;
487 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
488 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
489 timeflags &= ~XFS_ICHGTIME_MOD;
490 timeflags |= XFS_ICHGTIME_CHG;
492 if (tp && (mask & (ATTR_MTIME_SET|ATTR_ATIME_SET)))
493 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
497 * Change file inode change time only if ATTR_CTIME set
498 * AND we have been called by a DMI function.
501 if ((flags & XFS_ATTR_DMI) && (mask & ATTR_CTIME)) {
502 inode->i_ctime = iattr->ia_ctime;
503 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
504 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
505 ip->i_update_core = 1;
506 timeflags &= ~XFS_ICHGTIME_CHG;
510 * Send out timestamp changes that need to be set to the
511 * current time. Not done when called by a DMI function.
513 if (timeflags && !(flags & XFS_ATTR_DMI))
514 xfs_ichgtime(ip, timeflags);
516 XFS_STATS_INC(xs_ig_attrchg);
519 * If this is a synchronous mount, make sure that the
520 * transaction goes to disk before returning to the user.
521 * This is slightly sub-optimal in that truncates require
522 * two sync transactions instead of one for wsync filesystems.
523 * One for the truncate and one for the timestamps since we
524 * don't want to change the timestamps unless we're sure the
525 * truncate worked. Truncates are less than 1% of the laddis
526 * mix so this probably isn't worth the trouble to optimize.
530 if (mp->m_flags & XFS_MOUNT_WSYNC)
531 xfs_trans_set_sync(tp);
533 code = xfs_trans_commit(tp, commit_flags);
536 xfs_iunlock(ip, lock_flags);
539 * Release any dquot(s) the inode had kept before chown.
541 XFS_QM_DQRELE(mp, olddquot1);
542 XFS_QM_DQRELE(mp, olddquot2);
543 XFS_QM_DQRELE(mp, udqp);
544 XFS_QM_DQRELE(mp, gdqp);
550 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
551 !(flags & XFS_ATTR_DMI)) {
552 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
553 NULL, DM_RIGHT_NULL, NULL, NULL,
554 0, 0, AT_DELAY_FLAG(flags));
559 commit_flags |= XFS_TRANS_ABORT;
562 XFS_QM_DQRELE(mp, udqp);
563 XFS_QM_DQRELE(mp, gdqp);
565 xfs_trans_cancel(tp, commit_flags);
567 if (lock_flags != 0) {
568 xfs_iunlock(ip, lock_flags);
574 * The maximum pathlen is 1024 bytes. Since the minimum file system
575 * blocksize is 512 bytes, we can get a max of 2 extents back from
578 #define SYMLINK_MAPS 2
585 xfs_mount_t *mp = ip->i_mount;
586 int pathlen = ip->i_d.di_size;
587 int nmaps = SYMLINK_MAPS;
588 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
595 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
596 mval, &nmaps, NULL, NULL);
600 for (n = 0; n < nmaps; n++) {
601 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
602 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
604 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
605 error = XFS_BUF_GETERROR(bp);
607 xfs_ioerror_alert("xfs_readlink",
608 ip->i_mount, bp, XFS_BUF_ADDR(bp));
612 if (pathlen < byte_cnt)
616 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
620 link[ip->i_d.di_size] = '\0';
632 xfs_mount_t *mp = ip->i_mount;
636 xfs_itrace_entry(ip);
638 if (XFS_FORCED_SHUTDOWN(mp))
639 return XFS_ERROR(EIO);
641 xfs_ilock(ip, XFS_ILOCK_SHARED);
643 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
644 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
646 pathlen = ip->i_d.di_size;
650 if (ip->i_df.if_flags & XFS_IFINLINE) {
651 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
652 link[pathlen] = '\0';
654 error = xfs_readlink_bmap(ip, link);
658 xfs_iunlock(ip, XFS_ILOCK_SHARED);
665 * This is called to sync the inode and its data out to disk. We need to hold
666 * the I/O lock while flushing the data, and the inode lock while flushing the
667 * inode. The inode lock CANNOT be held while flushing the data, so acquire
668 * after we're done with that.
676 int log_flushed = 0, changed = 1;
678 xfs_itrace_entry(ip);
680 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
681 return XFS_ERROR(EIO);
683 /* capture size updates in I/O completion before writing the inode. */
684 error = xfs_wait_on_pages(ip, 0, -1);
686 return XFS_ERROR(error);
689 * We always need to make sure that the required inode state is safe on
690 * disk. The vnode might be clean but we still might need to force the
691 * log because of committed transactions that haven't hit the disk yet.
692 * Likewise, there could be unflushed non-transactional changes to the
693 * inode core that have to go to disk and this requires us to issue
694 * a synchronous transaction to capture these changes correctly.
696 * This code relies on the assumption that if the update_* fields
697 * of the inode are clear and the inode is unpinned then it is clean
698 * and no action is required.
700 xfs_ilock(ip, XFS_ILOCK_SHARED);
702 if (!(ip->i_update_size || ip->i_update_core)) {
704 * Timestamps/size haven't changed since last inode flush or
705 * inode transaction commit. That means either nothing got
706 * written or a transaction committed which caught the updates.
707 * If the latter happened and the transaction hasn't hit the
708 * disk yet, the inode will be still be pinned. If it is,
712 xfs_iunlock(ip, XFS_ILOCK_SHARED);
714 if (xfs_ipincount(ip)) {
715 error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
716 XFS_LOG_FORCE | XFS_LOG_SYNC,
720 * If the inode is not pinned and nothing has changed
721 * we don't need to flush the cache.
727 * Kick off a transaction to log the inode core to get the
728 * updates. The sync transaction will also force the log.
730 xfs_iunlock(ip, XFS_ILOCK_SHARED);
731 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
732 error = xfs_trans_reserve(tp, 0,
733 XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0);
735 xfs_trans_cancel(tp, 0);
738 xfs_ilock(ip, XFS_ILOCK_EXCL);
741 * Note - it's possible that we might have pushed ourselves out
742 * of the way during trans_reserve which would flush the inode.
743 * But there's no guarantee that the inode buffer has actually
744 * gone out yet (it's delwri). Plus the buffer could be pinned
745 * anyway if it's part of an inode in another recent
746 * transaction. So we play it safe and fire off the
747 * transaction anyway.
749 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
750 xfs_trans_ihold(tp, ip);
751 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
752 xfs_trans_set_sync(tp);
753 error = _xfs_trans_commit(tp, 0, &log_flushed);
755 xfs_iunlock(ip, XFS_ILOCK_EXCL);
758 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
760 * If the log write didn't issue an ordered tag we need
761 * to flush the disk cache for the data device now.
764 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
767 * If this inode is on the RT dev we need to flush that
770 if (XFS_IS_REALTIME_INODE(ip))
771 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
778 * This is called by xfs_inactive to free any blocks beyond eof
779 * when the link count isn't zero and by xfs_dm_punch_hole() when
780 * punching a hole to EOF.
790 xfs_fileoff_t end_fsb;
791 xfs_fileoff_t last_fsb;
792 xfs_filblks_t map_len;
794 xfs_bmbt_irec_t imap;
795 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
798 * Figure out if there are any blocks beyond the end
799 * of the file. If not, then there is nothing to do.
801 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
802 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
803 map_len = last_fsb - end_fsb;
808 xfs_ilock(ip, XFS_ILOCK_SHARED);
809 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
810 NULL, 0, &imap, &nimaps, NULL, NULL);
811 xfs_iunlock(ip, XFS_ILOCK_SHARED);
813 if (!error && (nimaps != 0) &&
814 (imap.br_startblock != HOLESTARTBLOCK ||
815 ip->i_delayed_blks)) {
817 * Attach the dquots to the inode up front.
819 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
823 * There are blocks after the end of file.
824 * Free them up now by truncating the file to
827 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
830 * Do the xfs_itruncate_start() call before
831 * reserving any log space because
832 * itruncate_start will call into the buffer
834 * do that within a transaction.
837 xfs_ilock(ip, XFS_IOLOCK_EXCL);
838 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
841 xfs_trans_cancel(tp, 0);
843 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
847 error = xfs_trans_reserve(tp, 0,
848 XFS_ITRUNCATE_LOG_RES(mp),
849 0, XFS_TRANS_PERM_LOG_RES,
850 XFS_ITRUNCATE_LOG_COUNT);
852 ASSERT(XFS_FORCED_SHUTDOWN(mp));
853 xfs_trans_cancel(tp, 0);
854 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
858 xfs_ilock(ip, XFS_ILOCK_EXCL);
859 xfs_trans_ijoin(tp, ip,
862 xfs_trans_ihold(tp, ip);
864 error = xfs_itruncate_finish(&tp, ip,
869 * If we get an error at this point we
870 * simply don't bother truncating the file.
874 (XFS_TRANS_RELEASE_LOG_RES |
877 error = xfs_trans_commit(tp,
878 XFS_TRANS_RELEASE_LOG_RES);
880 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
887 * Free a symlink that has blocks associated with it.
890 xfs_inactive_symlink_rmt(
898 xfs_fsblock_t first_block;
899 xfs_bmap_free_t free_list;
902 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
910 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
912 * We're freeing a symlink that has some
913 * blocks allocated to it. Free the
914 * blocks here. We know that we've got
915 * either 1 or 2 extents and that we can
916 * free them all in one bunmapi call.
918 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
919 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
920 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
921 ASSERT(XFS_FORCED_SHUTDOWN(mp));
922 xfs_trans_cancel(tp, 0);
927 * Lock the inode, fix the size, and join it to the transaction.
928 * Hold it so in the normal path, we still have it locked for
929 * the second transaction. In the error paths we need it
930 * held so the cancel won't rele it, see below.
932 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
933 size = (int)ip->i_d.di_size;
935 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
936 xfs_trans_ihold(tp, ip);
937 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
939 * Find the block(s) so we can inval and unmap them.
942 XFS_BMAP_INIT(&free_list, &first_block);
943 nmaps = ARRAY_SIZE(mval);
944 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
945 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
949 * Invalidate the block(s).
951 for (i = 0; i < nmaps; i++) {
952 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
953 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
954 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
955 xfs_trans_binval(tp, bp);
958 * Unmap the dead block(s) to the free_list.
960 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
961 &first_block, &free_list, NULL, &done)))
965 * Commit the first transaction. This logs the EFI and the inode.
967 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
970 * The transaction must have been committed, since there were
971 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
972 * The new tp has the extent freeing and EFDs.
976 * The first xact was committed, so add the inode to the new one.
977 * Mark it dirty so it will be logged and moved forward in the log as
978 * part of every commit.
980 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
981 xfs_trans_ihold(tp, ip);
982 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
984 * Get a new, empty transaction to return to our caller.
986 ntp = xfs_trans_dup(tp);
988 * Commit the transaction containing extent freeing and EFDs.
989 * If we get an error on the commit here or on the reserve below,
990 * we need to unlock the inode since the new transaction doesn't
991 * have the inode attached.
993 error = xfs_trans_commit(tp, 0);
996 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1000 * transaction commit worked ok so we can drop the extra ticket
1001 * reference that we gained in xfs_trans_dup()
1003 xfs_log_ticket_put(tp->t_ticket);
1006 * Remove the memory for extent descriptions (just bookkeeping).
1008 if (ip->i_df.if_bytes)
1009 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1010 ASSERT(ip->i_df.if_bytes == 0);
1012 * Put an itruncate log reservation in the new transaction
1015 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1016 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1017 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1021 * Return with the inode locked but not joined to the transaction.
1027 xfs_bmap_cancel(&free_list);
1030 * Have to come here with the inode locked and either
1031 * (held and in the transaction) or (not in the transaction).
1032 * If the inode isn't held then cancel would iput it, but
1033 * that's wrong since this is inactive and the vnode ref
1034 * count is 0 already.
1035 * Cancel won't do anything to the inode if held, but it still
1036 * needs to be locked until the cancel is done, if it was
1037 * joined to the transaction.
1039 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1040 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1047 xfs_inactive_symlink_local(
1053 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1055 * We're freeing a symlink which fit into
1056 * the inode. Just free the memory used
1057 * to hold the old symlink.
1059 error = xfs_trans_reserve(*tpp, 0,
1060 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1061 0, XFS_TRANS_PERM_LOG_RES,
1062 XFS_ITRUNCATE_LOG_COUNT);
1065 xfs_trans_cancel(*tpp, 0);
1069 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1072 * Zero length symlinks _can_ exist.
1074 if (ip->i_df.if_bytes > 0) {
1075 xfs_idata_realloc(ip,
1076 -(ip->i_df.if_bytes),
1078 ASSERT(ip->i_df.if_bytes == 0);
1092 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1095 ASSERT(ip->i_d.di_forkoff != 0);
1096 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1097 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1101 error = xfs_attr_inactive(ip);
1105 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1106 error = xfs_trans_reserve(tp, 0,
1107 XFS_IFREE_LOG_RES(mp),
1108 0, XFS_TRANS_PERM_LOG_RES,
1109 XFS_INACTIVE_LOG_COUNT);
1113 xfs_ilock(ip, XFS_ILOCK_EXCL);
1114 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1115 xfs_trans_ihold(tp, ip);
1116 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1118 ASSERT(ip->i_d.di_anextents == 0);
1124 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1125 xfs_trans_cancel(tp, 0);
1128 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1136 xfs_mount_t *mp = ip->i_mount;
1139 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1142 /* If this is a read-only mount, don't do this (would generate I/O) */
1143 if (mp->m_flags & XFS_MOUNT_RDONLY)
1146 if (!XFS_FORCED_SHUTDOWN(mp)) {
1150 * If we are using filestreams, and we have an unlinked
1151 * file that we are processing the last close on, then nothing
1152 * will be able to reopen and write to this file. Purge this
1153 * inode from the filestreams cache so that it doesn't delay
1154 * teardown of the inode.
1156 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1157 xfs_filestream_deassociate(ip);
1160 * If we previously truncated this file and removed old data
1161 * in the process, we want to initiate "early" writeout on
1162 * the last close. This is an attempt to combat the notorious
1163 * NULL files problem which is particularly noticable from a
1164 * truncate down, buffered (re-)write (delalloc), followed by
1165 * a crash. What we are effectively doing here is
1166 * significantly reducing the time window where we'd otherwise
1167 * be exposed to that problem.
1169 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1170 if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
1171 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
1174 if (ip->i_d.di_nlink != 0) {
1175 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1176 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
1177 ip->i_delayed_blks > 0)) &&
1178 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1179 (!(ip->i_d.di_flags &
1180 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1181 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1193 * This is called when the vnode reference count for the vnode
1194 * goes to zero. If the file has been unlinked, then it must
1195 * now be truncated. Also, we clear all of the read-ahead state
1196 * kept for the inode here since the file is now closed.
1202 xfs_bmap_free_t free_list;
1203 xfs_fsblock_t first_block;
1210 xfs_itrace_entry(ip);
1213 * If the inode is already free, then there can be nothing
1216 if (ip->i_d.di_mode == 0 || VN_BAD(VFS_I(ip))) {
1217 ASSERT(ip->i_df.if_real_bytes == 0);
1218 ASSERT(ip->i_df.if_broot_bytes == 0);
1219 return VN_INACTIVE_CACHE;
1223 * Only do a truncate if it's a regular file with
1224 * some actual space in it. It's OK to look at the
1225 * inode's fields without the lock because we're the
1226 * only one with a reference to the inode.
1228 truncate = ((ip->i_d.di_nlink == 0) &&
1229 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1230 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1231 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1235 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1236 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
1240 /* If this is a read-only mount, don't do this (would generate I/O) */
1241 if (mp->m_flags & XFS_MOUNT_RDONLY)
1244 if (ip->i_d.di_nlink != 0) {
1245 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1246 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
1247 ip->i_delayed_blks > 0)) &&
1248 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1249 (!(ip->i_d.di_flags &
1250 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1251 (ip->i_delayed_blks != 0)))) {
1252 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1254 return VN_INACTIVE_CACHE;
1259 ASSERT(ip->i_d.di_nlink == 0);
1261 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1262 return VN_INACTIVE_CACHE;
1264 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1267 * Do the xfs_itruncate_start() call before
1268 * reserving any log space because itruncate_start
1269 * will call into the buffer cache and we can't
1270 * do that within a transaction.
1272 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1274 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1276 xfs_trans_cancel(tp, 0);
1277 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1278 return VN_INACTIVE_CACHE;
1281 error = xfs_trans_reserve(tp, 0,
1282 XFS_ITRUNCATE_LOG_RES(mp),
1283 0, XFS_TRANS_PERM_LOG_RES,
1284 XFS_ITRUNCATE_LOG_COUNT);
1286 /* Don't call itruncate_cleanup */
1287 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1288 xfs_trans_cancel(tp, 0);
1289 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1290 return VN_INACTIVE_CACHE;
1293 xfs_ilock(ip, XFS_ILOCK_EXCL);
1294 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1295 xfs_trans_ihold(tp, ip);
1298 * normally, we have to run xfs_itruncate_finish sync.
1299 * But if filesystem is wsync and we're in the inactive
1300 * path, then we know that nlink == 0, and that the
1301 * xaction that made nlink == 0 is permanently committed
1302 * since xfs_remove runs as a synchronous transaction.
1304 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1305 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1308 xfs_trans_cancel(tp,
1309 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1310 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1311 return VN_INACTIVE_CACHE;
1313 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1316 * If we get an error while cleaning up a
1317 * symlink we bail out.
1319 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1320 xfs_inactive_symlink_rmt(ip, &tp) :
1321 xfs_inactive_symlink_local(ip, &tp);
1325 return VN_INACTIVE_CACHE;
1328 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1329 xfs_trans_ihold(tp, ip);
1331 error = xfs_trans_reserve(tp, 0,
1332 XFS_IFREE_LOG_RES(mp),
1333 0, XFS_TRANS_PERM_LOG_RES,
1334 XFS_INACTIVE_LOG_COUNT);
1336 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1337 xfs_trans_cancel(tp, 0);
1338 return VN_INACTIVE_CACHE;
1341 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1342 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1343 xfs_trans_ihold(tp, ip);
1347 * If there are attributes associated with the file
1348 * then blow them away now. The code calls a routine
1349 * that recursively deconstructs the attribute fork.
1350 * We need to just commit the current transaction
1351 * because we can't use it for xfs_attr_inactive().
1353 if (ip->i_d.di_anextents > 0) {
1354 error = xfs_inactive_attrs(ip, &tp);
1356 * If we got an error, the transaction is already
1357 * cancelled, and the inode is unlocked. Just get out.
1360 return VN_INACTIVE_CACHE;
1361 } else if (ip->i_afp) {
1362 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1368 XFS_BMAP_INIT(&free_list, &first_block);
1369 error = xfs_ifree(tp, ip, &free_list);
1372 * If we fail to free the inode, shut down. The cancel
1373 * might do that, we need to make sure. Otherwise the
1374 * inode might be lost for a long time or forever.
1376 if (!XFS_FORCED_SHUTDOWN(mp)) {
1378 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1379 error, mp->m_fsname);
1380 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1382 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1385 * Credit the quota account(s). The inode is gone.
1387 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1390 * Just ignore errors at this point. There is nothing we can
1391 * do except to try to keep going. Make sure it's not a silent
1394 error = xfs_bmap_finish(&tp, &free_list, &committed);
1396 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1397 "xfs_bmap_finish() returned error %d", error);
1398 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1400 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1401 "xfs_trans_commit() returned error %d", error);
1404 * Release the dquots held by inode, if any.
1406 XFS_QM_DQDETACH(mp, ip);
1408 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1411 return VN_INACTIVE_CACHE;
1415 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1416 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1417 * ci_name->name will point to a the actual name (caller must free) or
1418 * will be set to NULL if an exact match is found.
1423 struct xfs_name *name,
1425 struct xfs_name *ci_name)
1431 xfs_itrace_entry(dp);
1433 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1434 return XFS_ERROR(EIO);
1436 lock_mode = xfs_ilock_map_shared(dp);
1437 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
1438 xfs_iunlock_map_shared(dp, lock_mode);
1443 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
1447 xfs_itrace_ref(*ipp);
1452 kmem_free(ci_name->name);
1461 struct xfs_name *name,
1467 xfs_mount_t *mp = dp->i_mount;
1471 xfs_bmap_free_t free_list;
1472 xfs_fsblock_t first_block;
1473 boolean_t unlock_dp_on_error = B_FALSE;
1474 int dm_event_sent = 0;
1478 struct xfs_dquot *udqp, *gdqp;
1482 xfs_itrace_entry(dp);
1484 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
1485 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1486 dp, DM_RIGHT_NULL, NULL,
1487 DM_RIGHT_NULL, name->name, NULL,
1495 if (XFS_FORCED_SHUTDOWN(mp))
1496 return XFS_ERROR(EIO);
1498 /* Return through std_return after this point. */
1501 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1502 prid = dp->i_d.di_projid;
1504 prid = (xfs_prid_t)dfltprid;
1507 * Make sure that we have allocated dquot(s) on disk.
1509 error = XFS_QM_DQVOPALLOC(mp, dp,
1510 current_fsuid(), current_fsgid(), prid,
1511 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1517 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1518 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1519 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1521 * Initially assume that the file does not exist and
1522 * reserve the resources for that case. If that is not
1523 * the case we'll drop the one we have and get a more
1524 * appropriate transaction later.
1526 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1527 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1528 if (error == ENOSPC) {
1530 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1531 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1538 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1539 unlock_dp_on_error = B_TRUE;
1541 XFS_BMAP_INIT(&free_list, &first_block);
1546 * Reserve disk quota and the inode.
1548 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1552 error = xfs_dir_canenter(tp, dp, name, resblks);
1555 error = xfs_dir_ialloc(&tp, dp, mode, 1,
1556 rdev, credp, prid, resblks > 0,
1559 if (error == ENOSPC)
1566 * At this point, we've gotten a newly allocated inode.
1567 * It is locked (and joined to the transaction).
1570 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1573 * Now we join the directory inode to the transaction. We do not do it
1574 * earlier because xfs_dir_ialloc might commit the previous transaction
1575 * (and release all the locks). An error from here on will result in
1576 * the transaction cancel unlocking dp so don't do it explicitly in the
1580 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1581 unlock_dp_on_error = B_FALSE;
1583 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1584 &first_block, &free_list, resblks ?
1585 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1587 ASSERT(error != ENOSPC);
1590 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1591 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1594 * If this is a synchronous mount, make sure that the
1595 * create transaction goes to disk before returning to
1598 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1599 xfs_trans_set_sync(tp);
1603 * Attach the dquot(s) to the inodes and modify them incore.
1604 * These ids of the inode couldn't have changed since the new
1605 * inode has been locked ever since it was created.
1607 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
1610 * xfs_trans_commit normally decrements the vnode ref count
1611 * when it unlocks the inode. Since we want to return the
1612 * vnode to the caller, we bump the vnode ref count now.
1616 error = xfs_bmap_finish(&tp, &free_list, &committed);
1618 xfs_bmap_cancel(&free_list);
1622 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1629 XFS_QM_DQRELE(mp, udqp);
1630 XFS_QM_DQRELE(mp, gdqp);
1634 /* Fallthrough to std_return with error = 0 */
1637 if ((*ipp || (error != 0 && dm_event_sent != 0)) &&
1638 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
1639 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
1642 DM_RIGHT_NULL, name->name, NULL,
1648 cancel_flags |= XFS_TRANS_ABORT;
1653 xfs_trans_cancel(tp, cancel_flags);
1655 XFS_QM_DQRELE(mp, udqp);
1656 XFS_QM_DQRELE(mp, gdqp);
1658 if (unlock_dp_on_error)
1659 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1665 * Wait until after the current transaction is aborted to
1666 * release the inode. This prevents recursive transactions
1667 * and deadlocks from xfs_inactive.
1669 cancel_flags |= XFS_TRANS_ABORT;
1670 xfs_trans_cancel(tp, cancel_flags);
1673 XFS_QM_DQRELE(mp, udqp);
1674 XFS_QM_DQRELE(mp, gdqp);
1681 int xfs_small_retries;
1682 int xfs_middle_retries;
1683 int xfs_lots_retries;
1684 int xfs_lock_delays;
1688 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1692 xfs_lock_inumorder(int lock_mode, int subclass)
1694 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1695 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1696 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1697 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1703 * The following routine will lock n inodes in exclusive mode.
1704 * We assume the caller calls us with the inodes in i_ino order.
1706 * We need to detect deadlock where an inode that we lock
1707 * is in the AIL and we start waiting for another inode that is locked
1708 * by a thread in a long running transaction (such as truncate). This can
1709 * result in deadlock since the long running trans might need to wait
1710 * for the inode we just locked in order to push the tail and free space
1719 int attempts = 0, i, j, try_lock;
1722 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1728 for (; i < inodes; i++) {
1731 if (i && (ips[i] == ips[i-1])) /* Already locked */
1735 * If try_lock is not set yet, make sure all locked inodes
1736 * are not in the AIL.
1737 * If any are, set try_lock to be used later.
1741 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1742 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1743 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1750 * If any of the previous locks we have locked is in the AIL,
1751 * we must TRY to get the second and subsequent locks. If
1752 * we can't get any, we must release all we have
1757 /* try_lock must be 0 if i is 0. */
1759 * try_lock means we have an inode locked
1760 * that is in the AIL.
1763 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1767 * Unlock all previous guys and try again.
1768 * xfs_iunlock will try to push the tail
1769 * if the inode is in the AIL.
1772 for(j = i - 1; j >= 0; j--) {
1775 * Check to see if we've already
1776 * unlocked this one.
1777 * Not the first one going back,
1778 * and the inode ptr is the same.
1780 if ((j != (i - 1)) && ips[j] ==
1784 xfs_iunlock(ips[j], lock_mode);
1787 if ((attempts % 5) == 0) {
1788 delay(1); /* Don't just spin the CPU */
1798 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1804 if (attempts < 5) xfs_small_retries++;
1805 else if (attempts < 100) xfs_middle_retries++;
1806 else xfs_lots_retries++;
1814 * xfs_lock_two_inodes() can only be used to lock one type of lock
1815 * at a time - the iolock or the ilock, but not both at once. If
1816 * we lock both at once, lockdep will report false positives saying
1817 * we have violated locking orders.
1820 xfs_lock_two_inodes(
1829 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1830 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1831 ASSERT(ip0->i_ino != ip1->i_ino);
1833 if (ip0->i_ino > ip1->i_ino) {
1840 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1843 * If the first lock we have locked is in the AIL, we must TRY to get
1844 * the second lock. If we can't get it, we must release the first one
1847 lp = (xfs_log_item_t *)ip0->i_itemp;
1848 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1849 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1850 xfs_iunlock(ip0, lock_mode);
1851 if ((++attempts % 5) == 0)
1852 delay(1); /* Don't just spin the CPU */
1856 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1863 struct xfs_name *name,
1866 xfs_mount_t *mp = dp->i_mount;
1867 xfs_trans_t *tp = NULL;
1868 int is_dir = S_ISDIR(ip->i_d.di_mode);
1870 xfs_bmap_free_t free_list;
1871 xfs_fsblock_t first_block;
1878 xfs_itrace_entry(dp);
1879 xfs_itrace_entry(ip);
1881 if (XFS_FORCED_SHUTDOWN(mp))
1882 return XFS_ERROR(EIO);
1884 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
1885 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL,
1886 NULL, DM_RIGHT_NULL, name->name, NULL,
1887 ip->i_d.di_mode, 0, 0);
1892 error = XFS_QM_DQATTACH(mp, dp, 0);
1896 error = XFS_QM_DQATTACH(mp, ip, 0);
1901 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1902 log_count = XFS_DEFAULT_LOG_COUNT;
1904 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1905 log_count = XFS_REMOVE_LOG_COUNT;
1907 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1910 * We try to get the real space reservation first,
1911 * allowing for directory btree deletion(s) implying
1912 * possible bmap insert(s). If we can't get the space
1913 * reservation then we use 0 instead, and avoid the bmap
1914 * btree insert(s) in the directory code by, if the bmap
1915 * insert tries to happen, instead trimming the LAST
1916 * block from the directory.
1918 resblks = XFS_REMOVE_SPACE_RES(mp);
1919 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1920 XFS_TRANS_PERM_LOG_RES, log_count);
1921 if (error == ENOSPC) {
1923 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1924 XFS_TRANS_PERM_LOG_RES, log_count);
1927 ASSERT(error != ENOSPC);
1929 goto out_trans_cancel;
1932 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1935 * At this point, we've gotten both the directory and the entry
1939 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1942 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1945 * If we're removing a directory perform some additional validation.
1948 ASSERT(ip->i_d.di_nlink >= 2);
1949 if (ip->i_d.di_nlink != 2) {
1950 error = XFS_ERROR(ENOTEMPTY);
1951 goto out_trans_cancel;
1953 if (!xfs_dir_isempty(ip)) {
1954 error = XFS_ERROR(ENOTEMPTY);
1955 goto out_trans_cancel;
1959 XFS_BMAP_INIT(&free_list, &first_block);
1960 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1961 &first_block, &free_list, resblks);
1963 ASSERT(error != ENOENT);
1964 goto out_bmap_cancel;
1966 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1970 * Drop the link from ip's "..".
1972 error = xfs_droplink(tp, dp);
1974 goto out_bmap_cancel;
1977 * Drop the "." link from ip to self.
1979 error = xfs_droplink(tp, ip);
1981 goto out_bmap_cancel;
1984 * When removing a non-directory we need to log the parent
1985 * inode here. For a directory this is done implicitly
1986 * by the xfs_droplink call for the ".." entry.
1988 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1992 * Drop the link from dp to ip.
1994 error = xfs_droplink(tp, ip);
1996 goto out_bmap_cancel;
1999 * Determine if this is the last link while
2000 * we are in the transaction.
2002 link_zero = (ip->i_d.di_nlink == 0);
2005 * If this is a synchronous mount, make sure that the
2006 * remove transaction goes to disk before returning to
2009 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2010 xfs_trans_set_sync(tp);
2012 error = xfs_bmap_finish(&tp, &free_list, &committed);
2014 goto out_bmap_cancel;
2016 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2021 * If we are using filestreams, kill the stream association.
2022 * If the file is still open it may get a new one but that
2023 * will get killed on last close in xfs_close() so we don't
2024 * have to worry about that.
2026 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2027 xfs_filestream_deassociate(ip);
2029 xfs_itrace_exit(ip);
2030 xfs_itrace_exit(dp);
2033 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
2034 XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL,
2035 NULL, DM_RIGHT_NULL, name->name, NULL,
2036 ip->i_d.di_mode, error, 0);
2042 xfs_bmap_cancel(&free_list);
2043 cancel_flags |= XFS_TRANS_ABORT;
2045 xfs_trans_cancel(tp, cancel_flags);
2053 struct xfs_name *target_name)
2055 xfs_mount_t *mp = tdp->i_mount;
2058 xfs_bmap_free_t free_list;
2059 xfs_fsblock_t first_block;
2064 xfs_itrace_entry(tdp);
2065 xfs_itrace_entry(sip);
2067 ASSERT(!S_ISDIR(sip->i_d.di_mode));
2069 if (XFS_FORCED_SHUTDOWN(mp))
2070 return XFS_ERROR(EIO);
2072 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
2073 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2076 target_name->name, NULL, 0, 0, 0);
2081 /* Return through std_return after this point. */
2083 error = XFS_QM_DQATTACH(mp, sip, 0);
2084 if (!error && sip != tdp)
2085 error = XFS_QM_DQATTACH(mp, tdp, 0);
2089 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2090 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2091 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
2092 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2093 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2094 if (error == ENOSPC) {
2096 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2097 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2104 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
2107 * Increment vnode ref counts since xfs_trans_commit &
2108 * xfs_trans_cancel will both unlock the inodes and
2109 * decrement the associated ref counts.
2113 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2114 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2117 * If the source has too many links, we can't make any more to it.
2119 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2120 error = XFS_ERROR(EMLINK);
2125 * If we are using project inheritance, we only allow hard link
2126 * creation in our tree when the project IDs are the same; else
2127 * the tree quota mechanism could be circumvented.
2129 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2130 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2131 error = XFS_ERROR(EXDEV);
2135 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
2139 XFS_BMAP_INIT(&free_list, &first_block);
2141 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
2142 &first_block, &free_list, resblks);
2145 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2146 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2148 error = xfs_bumplink(tp, sip);
2153 * If this is a synchronous mount, make sure that the
2154 * link transaction goes to disk before returning to
2157 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2158 xfs_trans_set_sync(tp);
2161 error = xfs_bmap_finish (&tp, &free_list, &committed);
2163 xfs_bmap_cancel(&free_list);
2167 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2171 /* Fall through to std_return with error = 0. */
2173 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
2174 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2177 target_name->name, NULL, 0, error, 0);
2182 cancel_flags |= XFS_TRANS_ABORT;
2186 xfs_trans_cancel(tp, cancel_flags);
2194 struct xfs_name *dir_name,
2199 xfs_mount_t *mp = dp->i_mount;
2200 xfs_inode_t *cdp; /* inode of created dir */
2205 xfs_bmap_free_t free_list;
2206 xfs_fsblock_t first_block;
2207 boolean_t unlock_dp_on_error = B_FALSE;
2208 boolean_t created = B_FALSE;
2209 int dm_event_sent = 0;
2211 struct xfs_dquot *udqp, *gdqp;
2214 if (XFS_FORCED_SHUTDOWN(mp))
2215 return XFS_ERROR(EIO);
2219 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
2220 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2221 dp, DM_RIGHT_NULL, NULL,
2222 DM_RIGHT_NULL, dir_name->name, NULL,
2229 /* Return through std_return after this point. */
2231 xfs_itrace_entry(dp);
2235 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2236 prid = dp->i_d.di_projid;
2238 prid = (xfs_prid_t)dfltprid;
2241 * Make sure that we have allocated dquot(s) on disk.
2243 error = XFS_QM_DQVOPALLOC(mp, dp,
2244 current_fsuid(), current_fsgid(), prid,
2245 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2249 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2250 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2251 resblks = XFS_MKDIR_SPACE_RES(mp, dir_name->len);
2252 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2253 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2254 if (error == ENOSPC) {
2256 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2257 XFS_TRANS_PERM_LOG_RES,
2258 XFS_MKDIR_LOG_COUNT);
2265 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2266 unlock_dp_on_error = B_TRUE;
2269 * Check for directory link count overflow.
2271 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2272 error = XFS_ERROR(EMLINK);
2277 * Reserve disk quota and the inode.
2279 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2283 error = xfs_dir_canenter(tp, dp, dir_name, resblks);
2287 * create the directory inode.
2289 error = xfs_dir_ialloc(&tp, dp, mode, 2,
2290 0, credp, prid, resblks > 0,
2293 if (error == ENOSPC)
2297 xfs_itrace_ref(cdp);
2300 * Now we add the directory inode to the transaction.
2301 * We waited until now since xfs_dir_ialloc might start
2302 * a new transaction. Had we joined the transaction
2303 * earlier, the locks might have gotten released. An error
2304 * from here on will result in the transaction cancel
2305 * unlocking dp so don't do it explicitly in the error path.
2308 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2309 unlock_dp_on_error = B_FALSE;
2311 XFS_BMAP_INIT(&free_list, &first_block);
2313 error = xfs_dir_createname(tp, dp, dir_name, cdp->i_ino,
2314 &first_block, &free_list, resblks ?
2315 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2317 ASSERT(error != ENOSPC);
2320 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2322 error = xfs_dir_init(tp, cdp, dp);
2326 error = xfs_bumplink(tp, dp);
2336 * Attach the dquots to the new inode and modify the icount incore.
2338 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2341 * If this is a synchronous mount, make sure that the
2342 * mkdir transaction goes to disk before returning to
2345 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2346 xfs_trans_set_sync(tp);
2349 error = xfs_bmap_finish(&tp, &free_list, &committed);
2355 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2356 XFS_QM_DQRELE(mp, udqp);
2357 XFS_QM_DQRELE(mp, gdqp);
2362 /* Fall through to std_return with error = 0 or errno from
2363 * xfs_trans_commit. */
2366 if ((created || (error != 0 && dm_event_sent != 0)) &&
2367 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
2368 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2370 created ? cdp : NULL,
2372 dir_name->name, NULL,
2379 xfs_bmap_cancel(&free_list);
2381 cancel_flags |= XFS_TRANS_ABORT;
2383 xfs_trans_cancel(tp, cancel_flags);
2384 XFS_QM_DQRELE(mp, udqp);
2385 XFS_QM_DQRELE(mp, gdqp);
2387 if (unlock_dp_on_error)
2388 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2396 struct xfs_name *link_name,
2397 const char *target_path,
2402 xfs_mount_t *mp = dp->i_mount;
2407 xfs_bmap_free_t free_list;
2408 xfs_fsblock_t first_block;
2409 boolean_t unlock_dp_on_error = B_FALSE;
2412 xfs_fileoff_t first_fsb;
2413 xfs_filblks_t fs_blocks;
2415 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
2417 const char *cur_chunk;
2422 struct xfs_dquot *udqp, *gdqp;
2430 xfs_itrace_entry(dp);
2432 if (XFS_FORCED_SHUTDOWN(mp))
2433 return XFS_ERROR(EIO);
2436 * Check component lengths of the target path name.
2438 pathlen = strlen(target_path);
2439 if (pathlen >= MAXPATHLEN) /* total string too long */
2440 return XFS_ERROR(ENAMETOOLONG);
2442 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
2443 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
2444 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2445 link_name->name, target_path, 0, 0, 0);
2450 /* Return through std_return after this point. */
2453 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2454 prid = dp->i_d.di_projid;
2456 prid = (xfs_prid_t)dfltprid;
2459 * Make sure that we have allocated dquot(s) on disk.
2461 error = XFS_QM_DQVOPALLOC(mp, dp,
2462 current_fsuid(), current_fsgid(), prid,
2463 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2467 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
2468 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2470 * The symlink will fit into the inode data fork?
2471 * There can't be any attributes so we get the whole variable part.
2473 if (pathlen <= XFS_LITINO(mp))
2476 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
2477 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
2478 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2479 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2480 if (error == ENOSPC && fs_blocks == 0) {
2482 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2483 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2490 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2491 unlock_dp_on_error = B_TRUE;
2494 * Check whether the directory allows new symlinks or not.
2496 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2497 error = XFS_ERROR(EPERM);
2502 * Reserve disk quota : blocks and inode.
2504 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2509 * Check for ability to enter directory entry, if no space reserved.
2511 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2515 * Initialize the bmap freelist prior to calling either
2516 * bmapi or the directory create code.
2518 XFS_BMAP_INIT(&free_list, &first_block);
2521 * Allocate an inode for the symlink.
2523 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
2524 1, 0, credp, prid, resblks > 0, &ip, NULL);
2526 if (error == ENOSPC)
2533 * An error after we've joined dp to the transaction will result in the
2534 * transaction cancel unlocking dp so don't do it explicitly in the
2538 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2539 unlock_dp_on_error = B_FALSE;
2542 * Also attach the dquot(s) to it, if applicable.
2544 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2547 resblks -= XFS_IALLOC_SPACE_RES(mp);
2549 * If the symlink will fit into the inode, write it inline.
2551 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2552 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2553 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2554 ip->i_d.di_size = pathlen;
2557 * The inode was initially created in extent format.
2559 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2560 ip->i_df.if_flags |= XFS_IFINLINE;
2562 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2563 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2567 nmaps = SYMLINK_MAPS;
2569 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2570 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2571 &first_block, resblks, mval, &nmaps,
2578 resblks -= fs_blocks;
2579 ip->i_d.di_size = pathlen;
2580 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2582 cur_chunk = target_path;
2583 for (n = 0; n < nmaps; n++) {
2584 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2585 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2586 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2587 BTOBB(byte_cnt), 0);
2588 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2589 if (pathlen < byte_cnt) {
2592 pathlen -= byte_cnt;
2594 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2595 cur_chunk += byte_cnt;
2597 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2602 * Create the directory entry for the symlink.
2604 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2605 &first_block, &free_list, resblks);
2608 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2609 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2612 * If this is a synchronous mount, make sure that the
2613 * symlink transaction goes to disk before returning to
2616 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2617 xfs_trans_set_sync(tp);
2621 * xfs_trans_commit normally decrements the vnode ref count
2622 * when it unlocks the inode. Since we want to return the
2623 * vnode to the caller, we bump the vnode ref count now.
2627 error = xfs_bmap_finish(&tp, &free_list, &committed);
2631 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2632 XFS_QM_DQRELE(mp, udqp);
2633 XFS_QM_DQRELE(mp, gdqp);
2635 /* Fall through to std_return with error = 0 or errno from
2636 * xfs_trans_commit */
2638 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
2639 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
2642 DM_RIGHT_NULL, link_name->name,
2643 target_path, 0, error, 0);
2653 xfs_bmap_cancel(&free_list);
2654 cancel_flags |= XFS_TRANS_ABORT;
2656 xfs_trans_cancel(tp, cancel_flags);
2657 XFS_QM_DQRELE(mp, udqp);
2658 XFS_QM_DQRELE(mp, gdqp);
2660 if (unlock_dp_on_error)
2661 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2671 xfs_mount_t *mp = ip->i_mount;
2674 if (XFS_FORCED_SHUTDOWN(mp))
2675 return XFS_ERROR(EIO);
2678 * Bypass inodes which have already been cleaned by
2679 * the inode flush clustering code inside xfs_iflush
2681 if (xfs_inode_clean(ip))
2685 * We make this non-blocking if the inode is contended,
2686 * return EAGAIN to indicate to the caller that they
2687 * did not succeed. This prevents the flush path from
2688 * blocking on inodes inside another operation right
2689 * now, they get caught later by xfs_sync.
2691 if (flags & FLUSH_SYNC) {
2692 xfs_ilock(ip, XFS_ILOCK_SHARED);
2694 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
2695 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
2696 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2703 error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC
2704 : XFS_IFLUSH_ASYNC_NOBLOCK);
2705 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2717 xfs_mount_t *mp = ip->i_mount;
2721 if (!capable(CAP_SYS_ADMIN))
2722 return XFS_ERROR(EPERM);
2724 if (XFS_FORCED_SHUTDOWN(mp))
2725 return XFS_ERROR(EIO);
2727 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2728 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2730 xfs_trans_cancel(tp, 0);
2733 xfs_ilock(ip, XFS_ILOCK_EXCL);
2734 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2736 ip->i_d.di_dmevmask = evmask;
2737 ip->i_d.di_dmstate = state;
2739 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2741 error = xfs_trans_commit(tp, 0);
2751 xfs_itrace_entry(ip);
2753 ASSERT(!VN_MAPPED(VFS_I(ip)));
2755 /* bad inode, get out here ASAP */
2756 if (VN_BAD(VFS_I(ip))) {
2763 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
2766 * Make sure the atime in the XFS inode is correct before freeing the
2769 xfs_synchronize_atime(ip);
2772 * If we have nothing to flush with this inode then complete the
2773 * teardown now, otherwise break the link between the xfs inode and the
2774 * linux inode and clean up the xfs inode later. This avoids flushing
2775 * the inode to disk during the delete operation itself.
2777 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2778 * first to ensure that xfs_iunpin() will never see an xfs inode
2779 * that has a linux inode being reclaimed. Synchronisation is provided
2780 * by the i_flags_lock.
2782 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
2783 xfs_ilock(ip, XFS_ILOCK_EXCL);
2785 xfs_iflags_set(ip, XFS_IRECLAIMABLE);
2786 return xfs_reclaim_inode(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
2788 xfs_inode_set_reclaim_tag(ip);
2793 * xfs_alloc_file_space()
2794 * This routine allocates disk space for the given file.
2796 * If alloc_type == 0, this request is for an ALLOCSP type
2797 * request which will change the file size. In this case, no
2798 * DMAPI event will be generated by the call. A TRUNCATE event
2799 * will be generated later by xfs_setattr.
2801 * If alloc_type != 0, this request is for a RESVSP type
2802 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2803 * lower block boundary byte address is less than the file's
2812 xfs_alloc_file_space(
2819 xfs_mount_t *mp = ip->i_mount;
2821 xfs_filblks_t allocated_fsb;
2822 xfs_filblks_t allocatesize_fsb;
2823 xfs_extlen_t extsz, temp;
2824 xfs_fileoff_t startoffset_fsb;
2825 xfs_fsblock_t firstfsb;
2831 xfs_bmbt_irec_t imaps[1], *imapp;
2832 xfs_bmap_free_t free_list;
2833 uint qblocks, resblks, resrtextents;
2837 xfs_itrace_entry(ip);
2839 if (XFS_FORCED_SHUTDOWN(mp))
2840 return XFS_ERROR(EIO);
2842 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
2846 return XFS_ERROR(EINVAL);
2848 rt = XFS_IS_REALTIME_INODE(ip);
2849 extsz = xfs_get_extsz_hint(ip);
2854 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
2855 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
2856 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
2858 /* Generate a DMAPI event if needed. */
2859 if (alloc_type != 0 && offset < ip->i_size &&
2860 (attr_flags & XFS_ATTR_DMI) == 0 &&
2861 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
2862 xfs_off_t end_dmi_offset;
2864 end_dmi_offset = offset+len;
2865 if (end_dmi_offset > ip->i_size)
2866 end_dmi_offset = ip->i_size;
2867 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
2868 end_dmi_offset - offset, 0, NULL);
2874 * Allocate file space until done or until there is an error
2877 while (allocatesize_fsb && !error) {
2881 * Determine space reservations for data/realtime.
2883 if (unlikely(extsz)) {
2884 s = startoffset_fsb;
2887 e = startoffset_fsb + allocatesize_fsb;
2888 if ((temp = do_mod(startoffset_fsb, extsz)))
2890 if ((temp = do_mod(e, extsz)))
2894 e = allocatesize_fsb;
2898 resrtextents = qblocks = (uint)(e - s);
2899 resrtextents /= mp->m_sb.sb_rextsize;
2900 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2901 quota_flag = XFS_QMOPT_RES_RTBLKS;
2904 resblks = qblocks = \
2905 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
2906 quota_flag = XFS_QMOPT_RES_REGBLKS;
2910 * Allocate and setup the transaction.
2912 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2913 error = xfs_trans_reserve(tp, resblks,
2914 XFS_WRITE_LOG_RES(mp), resrtextents,
2915 XFS_TRANS_PERM_LOG_RES,
2916 XFS_WRITE_LOG_COUNT);
2918 * Check for running out of space
2922 * Free the transaction structure.
2924 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2925 xfs_trans_cancel(tp, 0);
2928 xfs_ilock(ip, XFS_ILOCK_EXCL);
2929 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
2930 qblocks, 0, quota_flag);
2934 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2935 xfs_trans_ihold(tp, ip);
2938 * Issue the xfs_bmapi() call to allocate the blocks
2940 XFS_BMAP_INIT(&free_list, &firstfsb);
2941 error = xfs_bmapi(tp, ip, startoffset_fsb,
2942 allocatesize_fsb, bmapi_flag,
2943 &firstfsb, 0, imapp, &nimaps,
2950 * Complete the transaction
2952 error = xfs_bmap_finish(&tp, &free_list, &committed);
2957 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2958 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2963 allocated_fsb = imapp->br_blockcount;
2966 error = XFS_ERROR(ENOSPC);
2970 startoffset_fsb += allocated_fsb;
2971 allocatesize_fsb -= allocated_fsb;
2974 if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 &&
2975 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
2976 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
2979 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
2981 goto retry; /* Maybe DMAPI app. has made space */
2982 /* else fall through with error from XFS_SEND_DATA */
2987 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
2988 xfs_bmap_cancel(&free_list);
2989 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
2991 error1: /* Just cancel transaction */
2992 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2993 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2994 goto dmapi_enospc_check;
2998 * Zero file bytes between startoff and endoff inclusive.
2999 * The iolock is held exclusive and no blocks are buffered.
3001 * This function is used by xfs_free_file_space() to zero
3002 * partial blocks when the range to free is not block aligned.
3003 * When unreserving space with boundaries that are not block
3004 * aligned we round up the start and round down the end
3005 * boundaries and then use this function to zero the parts of
3006 * the blocks that got dropped during the rounding.
3009 xfs_zero_remaining_bytes(
3014 xfs_bmbt_irec_t imap;
3015 xfs_fileoff_t offset_fsb;
3016 xfs_off_t lastoffset;
3019 xfs_mount_t *mp = ip->i_mount;
3024 * Avoid doing I/O beyond eof - it's not necessary
3025 * since nothing can read beyond eof. The space will
3026 * be zeroed when the file is extended anyway.
3028 if (startoff >= ip->i_size)
3031 if (endoff > ip->i_size)
3032 endoff = ip->i_size;
3034 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
3035 XFS_IS_REALTIME_INODE(ip) ?
3036 mp->m_rtdev_targp : mp->m_ddev_targp);
3038 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
3039 offset_fsb = XFS_B_TO_FSBT(mp, offset);
3041 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
3042 NULL, 0, &imap, &nimap, NULL, NULL);
3043 if (error || nimap < 1)
3045 ASSERT(imap.br_blockcount >= 1);
3046 ASSERT(imap.br_startoff == offset_fsb);
3047 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
3048 if (lastoffset > endoff)
3049 lastoffset = endoff;
3050 if (imap.br_startblock == HOLESTARTBLOCK)
3052 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3053 if (imap.br_state == XFS_EXT_UNWRITTEN)
3056 XFS_BUF_UNWRITE(bp);
3058 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
3060 error = xfs_iowait(bp);
3062 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
3063 mp, bp, XFS_BUF_ADDR(bp));
3066 memset(XFS_BUF_PTR(bp) +
3067 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
3068 0, lastoffset - offset + 1);
3073 error = xfs_iowait(bp);
3075 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
3076 mp, bp, XFS_BUF_ADDR(bp));
3085 * xfs_free_file_space()
3086 * This routine frees disk space for the given file.
3088 * This routine is only called by xfs_change_file_space
3089 * for an UNRESVSP type call.
3097 xfs_free_file_space(
3105 xfs_off_t end_dmi_offset;
3106 xfs_fileoff_t endoffset_fsb;
3108 xfs_fsblock_t firstfsb;
3109 xfs_bmap_free_t free_list;
3110 xfs_bmbt_irec_t imap;
3118 xfs_fileoff_t startoffset_fsb;
3120 int need_iolock = 1;
3124 xfs_itrace_entry(ip);
3126 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3130 if (len <= 0) /* if nothing being freed */
3132 rt = XFS_IS_REALTIME_INODE(ip);
3133 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
3134 end_dmi_offset = offset + len;
3135 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
3137 if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 &&
3138 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
3139 if (end_dmi_offset > ip->i_size)
3140 end_dmi_offset = ip->i_size;
3141 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
3142 offset, end_dmi_offset - offset,
3143 AT_DELAY_FLAG(attr_flags), NULL);
3148 if (attr_flags & XFS_ATTR_NOLOCK)
3151 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3152 vn_iowait(ip); /* wait for the completion of any pending DIOs */
3155 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
3156 ioffset = offset & ~(rounding - 1);
3158 if (VN_CACHED(VFS_I(ip)) != 0) {
3159 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
3160 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
3162 goto out_unlock_iolock;
3166 * Need to zero the stuff we're not freeing, on disk.
3167 * If its a realtime file & can't use unwritten extents then we
3168 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3169 * will take care of it for us.
3171 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
3173 error = xfs_bmapi(NULL, ip, startoffset_fsb,
3174 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
3176 goto out_unlock_iolock;
3177 ASSERT(nimap == 0 || nimap == 1);
3178 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3181 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3182 block = imap.br_startblock;
3183 mod = do_div(block, mp->m_sb.sb_rextsize);
3185 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
3188 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
3189 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
3191 goto out_unlock_iolock;
3192 ASSERT(nimap == 0 || nimap == 1);
3193 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3194 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3196 if (mod && (mod != mp->m_sb.sb_rextsize))
3197 endoffset_fsb -= mod;
3200 if ((done = (endoffset_fsb <= startoffset_fsb)))
3202 * One contiguous piece to clear
3204 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
3207 * Some full blocks, possibly two pieces to clear
3209 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
3210 error = xfs_zero_remaining_bytes(ip, offset,
3211 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
3213 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
3214 error = xfs_zero_remaining_bytes(ip,
3215 XFS_FSB_TO_B(mp, endoffset_fsb),
3220 * free file space until done or until there is an error
3222 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3223 while (!error && !done) {
3226 * allocate and setup the transaction. Allow this
3227 * transaction to dip into the reserve blocks to ensure
3228 * the freeing of the space succeeds at ENOSPC.
3230 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
3231 tp->t_flags |= XFS_TRANS_RESERVE;
3232 error = xfs_trans_reserve(tp,
3234 XFS_WRITE_LOG_RES(mp),
3236 XFS_TRANS_PERM_LOG_RES,
3237 XFS_WRITE_LOG_COUNT);
3240 * check for running out of space
3244 * Free the transaction structure.
3246 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3247 xfs_trans_cancel(tp, 0);
3250 xfs_ilock(ip, XFS_ILOCK_EXCL);
3251 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
3252 ip->i_udquot, ip->i_gdquot, resblks, 0,
3253 XFS_QMOPT_RES_REGBLKS);
3257 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3258 xfs_trans_ihold(tp, ip);
3261 * issue the bunmapi() call to free the blocks
3263 XFS_BMAP_INIT(&free_list, &firstfsb);
3264 error = xfs_bunmapi(tp, ip, startoffset_fsb,
3265 endoffset_fsb - startoffset_fsb,
3266 0, 2, &firstfsb, &free_list, NULL, &done);
3272 * complete the transaction
3274 error = xfs_bmap_finish(&tp, &free_list, &committed);
3279 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3280 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3285 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3289 xfs_bmap_cancel(&free_list);
3291 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3292 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
3298 * xfs_change_file_space()
3299 * This routine allocates or frees disk space for the given file.
3300 * The user specified parameters are checked for alignment and size
3309 xfs_change_file_space(
3316 xfs_mount_t *mp = ip->i_mount;
3321 xfs_off_t startoffset;
3326 xfs_itrace_entry(ip);
3328 if (!S_ISREG(ip->i_d.di_mode))
3329 return XFS_ERROR(EINVAL);
3331 switch (bf->l_whence) {
3332 case 0: /*SEEK_SET*/
3334 case 1: /*SEEK_CUR*/
3335 bf->l_start += offset;
3337 case 2: /*SEEK_END*/
3338 bf->l_start += ip->i_size;
3341 return XFS_ERROR(EINVAL);
3344 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
3346 if ( (bf->l_start < 0)
3347 || (bf->l_start > XFS_MAXIOFFSET(mp))
3348 || (bf->l_start + llen < 0)
3349 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
3350 return XFS_ERROR(EINVAL);
3354 startoffset = bf->l_start;
3358 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3360 * These calls do NOT zero the data space allocated to the file,
3361 * nor do they change the file size.
3363 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3365 * These calls cause the new file data to be zeroed and the file
3366 * size to be changed.
3368 setprealloc = clrprealloc = 0;
3371 case XFS_IOC_RESVSP:
3372 case XFS_IOC_RESVSP64:
3373 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
3380 case XFS_IOC_UNRESVSP:
3381 case XFS_IOC_UNRESVSP64:
3382 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
3387 case XFS_IOC_ALLOCSP:
3388 case XFS_IOC_ALLOCSP64:
3389 case XFS_IOC_FREESP:
3390 case XFS_IOC_FREESP64:
3391 if (startoffset > fsize) {
3392 error = xfs_alloc_file_space(ip, fsize,
3393 startoffset - fsize, 0, attr_flags);
3398 iattr.ia_valid = ATTR_SIZE;
3399 iattr.ia_size = startoffset;
3401 error = xfs_setattr(ip, &iattr, attr_flags);
3411 return XFS_ERROR(EINVAL);
3415 * update the inode timestamp, mode, and prealloc flag bits
3417 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
3419 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
3422 xfs_trans_cancel(tp, 0);
3426 xfs_ilock(ip, XFS_ILOCK_EXCL);
3428 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3429 xfs_trans_ihold(tp, ip);
3431 if ((attr_flags & XFS_ATTR_DMI) == 0) {
3432 ip->i_d.di_mode &= ~S_ISUID;
3435 * Note that we don't have to worry about mandatory
3436 * file locking being disabled here because we only
3437 * clear the S_ISGID bit if the Group execute bit is
3438 * on, but if it was on then mandatory locking wouldn't
3439 * have been enabled.
3441 if (ip->i_d.di_mode & S_IXGRP)
3442 ip->i_d.di_mode &= ~S_ISGID;
3444 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3447 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
3448 else if (clrprealloc)
3449 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
3451 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3452 xfs_trans_set_sync(tp);
3454 error = xfs_trans_commit(tp, 0);
3456 xfs_iunlock(ip, XFS_ILOCK_EXCL);