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);
85 xfs_mount_t *mp = ip->i_mount;
86 struct inode *inode = VFS_I(ip);
87 int mask = iattr->ia_valid;
95 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
101 if (mp->m_flags & XFS_MOUNT_RDONLY)
102 return XFS_ERROR(EROFS);
104 if (XFS_FORCED_SHUTDOWN(mp))
105 return XFS_ERROR(EIO);
107 olddquot1 = olddquot2 = NULL;
111 * If disk quotas is on, we make sure that the dquots do exist on disk,
112 * before we start any other transactions. Trying to do this later
113 * is messy. We don't care to take a readlock to look at the ids
114 * in inode here, because we can't hold it across the trans_reserve.
115 * If the IDs do change before we take the ilock, we're covered
116 * because the i_*dquot fields will get updated anyway.
118 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
121 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
123 qflags |= XFS_QMOPT_UQUOTA;
125 uid = ip->i_d.di_uid;
127 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
129 qflags |= XFS_QMOPT_GQUOTA;
131 gid = ip->i_d.di_gid;
135 * We take a reference when we initialize udqp and gdqp,
136 * so it is important that we never blindly double trip on
137 * the same variable. See xfs_create() for an example.
139 ASSERT(udqp == NULL);
140 ASSERT(gdqp == NULL);
141 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, ip->i_d.di_projid,
142 qflags, &udqp, &gdqp);
148 * For the other attributes, we acquire the inode lock and
149 * first do an error checking pass.
152 lock_flags = XFS_ILOCK_EXCL;
153 if (flags & XFS_ATTR_NOLOCK)
155 if (!(mask & ATTR_SIZE)) {
156 if ((mask != (ATTR_CTIME|ATTR_ATIME|ATTR_MTIME)) ||
157 (mp->m_flags & XFS_MOUNT_WSYNC)) {
158 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
160 if ((code = xfs_trans_reserve(tp, 0,
161 XFS_ICHANGE_LOG_RES(mp), 0,
168 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
169 !(flags & XFS_ATTR_DMI)) {
170 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
171 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip,
172 iattr->ia_size, 0, dmflags, NULL);
179 lock_flags |= XFS_IOLOCK_EXCL;
182 xfs_ilock(ip, lock_flags);
184 /* boolean: are we the file owner? */
185 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
188 * Change various properties of a file.
189 * Only the owner or users with CAP_FOWNER
190 * capability may do these things.
192 if (mask & (ATTR_MODE|ATTR_UID|ATTR_GID)) {
194 * CAP_FOWNER overrides the following restrictions:
196 * The user ID of the calling process must be equal
197 * to the file owner ID, except in cases where the
198 * CAP_FSETID capability is applicable.
200 if (!file_owner && !capable(CAP_FOWNER)) {
201 code = XFS_ERROR(EPERM);
206 * CAP_FSETID overrides the following restrictions:
208 * The effective user ID of the calling process shall match
209 * the file owner when setting the set-user-ID and
210 * set-group-ID bits on that file.
212 * The effective group ID or one of the supplementary group
213 * IDs of the calling process shall match the group owner of
214 * the file when setting the set-group-ID bit on that file
216 if (mask & ATTR_MODE) {
219 if ((iattr->ia_mode & S_ISUID) && !file_owner)
221 if ((iattr->ia_mode & S_ISGID) &&
222 !in_group_p((gid_t)ip->i_d.di_gid))
225 /* Linux allows this, Irix doesn't. */
226 if ((iattr->ia_mode & S_ISVTX) && !S_ISDIR(ip->i_d.di_mode))
229 if (m && !capable(CAP_FSETID))
230 iattr->ia_mode &= ~m;
235 * Change file ownership. Must be the owner or privileged.
236 * If the system was configured with the "restricted_chown"
237 * option, the owner is not permitted to give away the file,
238 * and can change the group id only to a group of which he
239 * or she is a member.
241 if (mask & (ATTR_UID|ATTR_GID)) {
243 * These IDs could have changed since we last looked at them.
244 * But, we're assured that if the ownership did change
245 * while we didn't have the inode locked, inode's dquot(s)
246 * would have changed also.
248 iuid = ip->i_d.di_uid;
249 igid = ip->i_d.di_gid;
250 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
251 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
254 * CAP_CHOWN overrides the following restrictions:
256 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
257 * shall override the restriction that a process cannot
258 * change the user ID of a file it owns and the restriction
259 * that the group ID supplied to the chown() function
260 * shall be equal to either the group ID or one of the
261 * supplementary group IDs of the calling process.
263 if (restricted_chown &&
264 (iuid != uid || (igid != gid &&
265 !in_group_p((gid_t)gid))) &&
266 !capable(CAP_CHOWN)) {
267 code = XFS_ERROR(EPERM);
271 * Do a quota reservation only if uid/gid is actually
274 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
275 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
277 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
278 capable(CAP_FOWNER) ?
279 XFS_QMOPT_FORCE_RES : 0);
280 if (code) /* out of quota */
286 * Truncate file. Must have write permission and not be a directory.
288 if (mask & ATTR_SIZE) {
289 /* Short circuit the truncate case for zero length files */
290 if (iattr->ia_size == 0 &&
291 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
292 xfs_iunlock(ip, XFS_ILOCK_EXCL);
293 lock_flags &= ~XFS_ILOCK_EXCL;
294 if (mask & ATTR_CTIME)
295 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
300 if (S_ISDIR(ip->i_d.di_mode)) {
301 code = XFS_ERROR(EISDIR);
303 } else if (!S_ISREG(ip->i_d.di_mode)) {
304 code = XFS_ERROR(EINVAL);
308 * Make sure that the dquots are attached to the inode.
310 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
315 * Change file access or modified times.
317 if (mask & (ATTR_ATIME|ATTR_MTIME)) {
319 if ((mask & (ATTR_MTIME_SET|ATTR_ATIME_SET)) &&
320 !capable(CAP_FOWNER)) {
321 code = XFS_ERROR(EPERM);
328 * Now we can make the changes. Before we join the inode
329 * to the transaction, if ATTR_SIZE is set then take care of
330 * the part of the truncation that must be done without the
331 * inode lock. This needs to be done before joining the inode
332 * to the transaction, because the inode cannot be unlocked
333 * once it is a part of the transaction.
335 if (mask & ATTR_SIZE) {
337 if (iattr->ia_size > ip->i_size) {
339 * Do the first part of growing a file: zero any data
340 * in the last block that is beyond the old EOF. We
341 * need to do this before the inode is joined to the
342 * transaction to modify the i_size.
344 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
346 xfs_iunlock(ip, XFS_ILOCK_EXCL);
349 * We are going to log the inode size change in this
350 * transaction so any previous writes that are beyond the on
351 * disk EOF and the new EOF that have not been written out need
352 * to be written here. If we do not write the data out, we
353 * expose ourselves to the null files problem.
355 * Only flush from the on disk size to the smaller of the in
356 * memory file size or the new size as that's the range we
357 * really care about here and prevents waiting for other data
358 * not within the range we care about here.
361 ip->i_size != ip->i_d.di_size &&
362 iattr->ia_size > ip->i_d.di_size) {
363 code = xfs_flush_pages(ip,
364 ip->i_d.di_size, iattr->ia_size,
365 XFS_B_ASYNC, FI_NONE);
368 /* wait for all I/O to complete */
372 code = xfs_itruncate_data(ip, iattr->ia_size);
375 lock_flags &= ~XFS_ILOCK_EXCL;
376 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
379 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
380 if ((code = xfs_trans_reserve(tp, 0,
381 XFS_ITRUNCATE_LOG_RES(mp), 0,
382 XFS_TRANS_PERM_LOG_RES,
383 XFS_ITRUNCATE_LOG_COUNT))) {
384 xfs_trans_cancel(tp, 0);
386 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
389 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
390 xfs_ilock(ip, XFS_ILOCK_EXCL);
394 xfs_trans_ijoin(tp, ip, lock_flags);
395 xfs_trans_ihold(tp, ip);
399 * Truncate file. Must have write permission and not be a directory.
401 if (mask & ATTR_SIZE) {
403 * Only change the c/mtime if we are changing the size
404 * or we are explicitly asked to change it. This handles
405 * the semantic difference between truncate() and ftruncate()
406 * as implemented in the VFS.
408 if (iattr->ia_size != ip->i_size || (mask & ATTR_CTIME))
409 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
411 if (iattr->ia_size > ip->i_size) {
412 ip->i_d.di_size = iattr->ia_size;
413 ip->i_size = iattr->ia_size;
414 if (!(flags & XFS_ATTR_DMI))
415 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
416 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
417 } else if (iattr->ia_size <= ip->i_size ||
418 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
420 * signal a sync transaction unless
421 * we're truncating an already unlinked
422 * file on a wsync filesystem
424 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
426 ((ip->i_d.di_nlink != 0 ||
427 !(mp->m_flags & XFS_MOUNT_WSYNC))
432 * Truncated "down", so we're removing references
433 * to old data here - if we now delay flushing for
434 * a long time, we expose ourselves unduly to the
435 * notorious NULL files problem. So, we mark this
436 * vnode and flush it when the file is closed, and
437 * do not wait the usual (long) time for writeout.
439 xfs_iflags_set(ip, XFS_ITRUNCATED);
444 * Change file access modes.
446 if (mask & ATTR_MODE) {
447 ip->i_d.di_mode &= S_IFMT;
448 ip->i_d.di_mode |= iattr->ia_mode & ~S_IFMT;
450 inode->i_mode &= S_IFMT;
451 inode->i_mode |= iattr->ia_mode & ~S_IFMT;
453 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
454 timeflags |= XFS_ICHGTIME_CHG;
458 * Change file ownership. Must be the owner or privileged.
459 * If the system was configured with the "restricted_chown"
460 * option, the owner is not permitted to give away the file,
461 * and can change the group id only to a group of which he
462 * or she is a member.
464 if (mask & (ATTR_UID|ATTR_GID)) {
466 * CAP_FSETID overrides the following restrictions:
468 * The set-user-ID and set-group-ID bits of a file will be
469 * cleared upon successful return from chown()
471 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
472 !capable(CAP_FSETID)) {
473 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
477 * Change the ownerships and register quota modifications
478 * in the transaction.
481 if (XFS_IS_UQUOTA_ON(mp)) {
482 ASSERT(mask & ATTR_UID);
484 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
485 &ip->i_udquot, udqp);
487 ip->i_d.di_uid = uid;
491 if (XFS_IS_GQUOTA_ON(mp)) {
492 ASSERT(!XFS_IS_PQUOTA_ON(mp));
493 ASSERT(mask & ATTR_GID);
495 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
496 &ip->i_gdquot, gdqp);
498 ip->i_d.di_gid = gid;
502 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
503 timeflags |= XFS_ICHGTIME_CHG;
508 * Change file access or modified times.
510 if (mask & (ATTR_ATIME|ATTR_MTIME)) {
511 if (mask & ATTR_ATIME) {
512 inode->i_atime = iattr->ia_atime;
513 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
514 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
515 ip->i_update_core = 1;
516 timeflags &= ~XFS_ICHGTIME_ACC;
518 if (mask & ATTR_MTIME) {
519 inode->i_mtime = iattr->ia_mtime;
520 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
521 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
522 timeflags &= ~XFS_ICHGTIME_MOD;
523 timeflags |= XFS_ICHGTIME_CHG;
525 if (tp && (mask & (ATTR_MTIME_SET|ATTR_ATIME_SET)))
526 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
530 * Change file inode change time only if ATTR_CTIME set
531 * AND we have been called by a DMI function.
534 if ((flags & XFS_ATTR_DMI) && (mask & ATTR_CTIME)) {
535 inode->i_ctime = iattr->ia_ctime;
536 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
537 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
538 ip->i_update_core = 1;
539 timeflags &= ~XFS_ICHGTIME_CHG;
543 * Send out timestamp changes that need to be set to the
544 * current time. Not done when called by a DMI function.
546 if (timeflags && !(flags & XFS_ATTR_DMI))
547 xfs_ichgtime(ip, timeflags);
549 XFS_STATS_INC(xs_ig_attrchg);
552 * If this is a synchronous mount, make sure that the
553 * transaction goes to disk before returning to the user.
554 * This is slightly sub-optimal in that truncates require
555 * two sync transactions instead of one for wsync filesystems.
556 * One for the truncate and one for the timestamps since we
557 * don't want to change the timestamps unless we're sure the
558 * truncate worked. Truncates are less than 1% of the laddis
559 * mix so this probably isn't worth the trouble to optimize.
563 if (mp->m_flags & XFS_MOUNT_WSYNC)
564 xfs_trans_set_sync(tp);
566 code = xfs_trans_commit(tp, commit_flags);
569 xfs_iunlock(ip, lock_flags);
572 * Release any dquot(s) the inode had kept before chown.
574 XFS_QM_DQRELE(mp, olddquot1);
575 XFS_QM_DQRELE(mp, olddquot2);
576 XFS_QM_DQRELE(mp, udqp);
577 XFS_QM_DQRELE(mp, gdqp);
583 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
584 !(flags & XFS_ATTR_DMI)) {
585 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
586 NULL, DM_RIGHT_NULL, NULL, NULL,
587 0, 0, AT_DELAY_FLAG(flags));
592 commit_flags |= XFS_TRANS_ABORT;
595 XFS_QM_DQRELE(mp, udqp);
596 XFS_QM_DQRELE(mp, gdqp);
598 xfs_trans_cancel(tp, commit_flags);
600 if (lock_flags != 0) {
601 xfs_iunlock(ip, lock_flags);
607 * The maximum pathlen is 1024 bytes. Since the minimum file system
608 * blocksize is 512 bytes, we can get a max of 2 extents back from
611 #define SYMLINK_MAPS 2
618 xfs_mount_t *mp = ip->i_mount;
619 int pathlen = ip->i_d.di_size;
620 int nmaps = SYMLINK_MAPS;
621 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
628 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
629 mval, &nmaps, NULL, NULL);
633 for (n = 0; n < nmaps; n++) {
634 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
635 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
637 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
638 error = XFS_BUF_GETERROR(bp);
640 xfs_ioerror_alert("xfs_readlink",
641 ip->i_mount, bp, XFS_BUF_ADDR(bp));
645 if (pathlen < byte_cnt)
649 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
653 link[ip->i_d.di_size] = '\0';
665 xfs_mount_t *mp = ip->i_mount;
669 xfs_itrace_entry(ip);
671 if (XFS_FORCED_SHUTDOWN(mp))
672 return XFS_ERROR(EIO);
674 xfs_ilock(ip, XFS_ILOCK_SHARED);
676 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
677 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
679 pathlen = ip->i_d.di_size;
683 if (ip->i_df.if_flags & XFS_IFINLINE) {
684 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
685 link[pathlen] = '\0';
687 error = xfs_readlink_bmap(ip, link);
691 xfs_iunlock(ip, XFS_ILOCK_SHARED);
698 * This is called to sync the inode and its data out to disk. We need to hold
699 * the I/O lock while flushing the data, and the inode lock while flushing the
700 * inode. The inode lock CANNOT be held while flushing the data, so acquire
701 * after we're done with that.
709 int log_flushed = 0, changed = 1;
711 xfs_itrace_entry(ip);
713 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
714 return XFS_ERROR(EIO);
716 /* capture size updates in I/O completion before writing the inode. */
717 error = filemap_fdatawait(VFS_I(ip)->i_mapping);
719 return XFS_ERROR(error);
722 * We always need to make sure that the required inode state is safe on
723 * disk. The vnode might be clean but we still might need to force the
724 * log because of committed transactions that haven't hit the disk yet.
725 * Likewise, there could be unflushed non-transactional changes to the
726 * inode core that have to go to disk and this requires us to issue
727 * a synchronous transaction to capture these changes correctly.
729 * This code relies on the assumption that if the update_* fields
730 * of the inode are clear and the inode is unpinned then it is clean
731 * and no action is required.
733 xfs_ilock(ip, XFS_ILOCK_SHARED);
735 if (!(ip->i_update_size || ip->i_update_core)) {
737 * Timestamps/size haven't changed since last inode flush or
738 * inode transaction commit. That means either nothing got
739 * written or a transaction committed which caught the updates.
740 * If the latter happened and the transaction hasn't hit the
741 * disk yet, the inode will be still be pinned. If it is,
745 xfs_iunlock(ip, XFS_ILOCK_SHARED);
747 if (xfs_ipincount(ip)) {
748 error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
749 XFS_LOG_FORCE | XFS_LOG_SYNC,
753 * If the inode is not pinned and nothing has changed
754 * we don't need to flush the cache.
760 * Kick off a transaction to log the inode core to get the
761 * updates. The sync transaction will also force the log.
763 xfs_iunlock(ip, XFS_ILOCK_SHARED);
764 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
765 error = xfs_trans_reserve(tp, 0,
766 XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0);
768 xfs_trans_cancel(tp, 0);
771 xfs_ilock(ip, XFS_ILOCK_EXCL);
774 * Note - it's possible that we might have pushed ourselves out
775 * of the way during trans_reserve which would flush the inode.
776 * But there's no guarantee that the inode buffer has actually
777 * gone out yet (it's delwri). Plus the buffer could be pinned
778 * anyway if it's part of an inode in another recent
779 * transaction. So we play it safe and fire off the
780 * transaction anyway.
782 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
783 xfs_trans_ihold(tp, ip);
784 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
785 xfs_trans_set_sync(tp);
786 error = _xfs_trans_commit(tp, 0, &log_flushed);
788 xfs_iunlock(ip, XFS_ILOCK_EXCL);
791 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
793 * If the log write didn't issue an ordered tag we need
794 * to flush the disk cache for the data device now.
797 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
800 * If this inode is on the RT dev we need to flush that
803 if (XFS_IS_REALTIME_INODE(ip))
804 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
811 * This is called by xfs_inactive to free any blocks beyond eof
812 * when the link count isn't zero and by xfs_dm_punch_hole() when
813 * punching a hole to EOF.
823 xfs_fileoff_t end_fsb;
824 xfs_fileoff_t last_fsb;
825 xfs_filblks_t map_len;
827 xfs_bmbt_irec_t imap;
828 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
831 * Figure out if there are any blocks beyond the end
832 * of the file. If not, then there is nothing to do.
834 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
835 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
836 map_len = last_fsb - end_fsb;
841 xfs_ilock(ip, XFS_ILOCK_SHARED);
842 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
843 NULL, 0, &imap, &nimaps, NULL, NULL);
844 xfs_iunlock(ip, XFS_ILOCK_SHARED);
846 if (!error && (nimaps != 0) &&
847 (imap.br_startblock != HOLESTARTBLOCK ||
848 ip->i_delayed_blks)) {
850 * Attach the dquots to the inode up front.
852 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
856 * There are blocks after the end of file.
857 * Free them up now by truncating the file to
860 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
863 * Do the xfs_itruncate_start() call before
864 * reserving any log space because
865 * itruncate_start will call into the buffer
867 * do that within a transaction.
870 xfs_ilock(ip, XFS_IOLOCK_EXCL);
871 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
874 xfs_trans_cancel(tp, 0);
876 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
880 error = xfs_trans_reserve(tp, 0,
881 XFS_ITRUNCATE_LOG_RES(mp),
882 0, XFS_TRANS_PERM_LOG_RES,
883 XFS_ITRUNCATE_LOG_COUNT);
885 ASSERT(XFS_FORCED_SHUTDOWN(mp));
886 xfs_trans_cancel(tp, 0);
887 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
891 xfs_ilock(ip, XFS_ILOCK_EXCL);
892 xfs_trans_ijoin(tp, ip,
895 xfs_trans_ihold(tp, ip);
897 error = xfs_itruncate_finish(&tp, ip,
902 * If we get an error at this point we
903 * simply don't bother truncating the file.
907 (XFS_TRANS_RELEASE_LOG_RES |
910 error = xfs_trans_commit(tp,
911 XFS_TRANS_RELEASE_LOG_RES);
913 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
920 * Free a symlink that has blocks associated with it.
923 xfs_inactive_symlink_rmt(
931 xfs_fsblock_t first_block;
932 xfs_bmap_free_t free_list;
935 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
943 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
945 * We're freeing a symlink that has some
946 * blocks allocated to it. Free the
947 * blocks here. We know that we've got
948 * either 1 or 2 extents and that we can
949 * free them all in one bunmapi call.
951 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
952 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
953 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
954 ASSERT(XFS_FORCED_SHUTDOWN(mp));
955 xfs_trans_cancel(tp, 0);
960 * Lock the inode, fix the size, and join it to the transaction.
961 * Hold it so in the normal path, we still have it locked for
962 * the second transaction. In the error paths we need it
963 * held so the cancel won't rele it, see below.
965 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
966 size = (int)ip->i_d.di_size;
968 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
969 xfs_trans_ihold(tp, ip);
970 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
972 * Find the block(s) so we can inval and unmap them.
975 XFS_BMAP_INIT(&free_list, &first_block);
976 nmaps = ARRAY_SIZE(mval);
977 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
978 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
982 * Invalidate the block(s).
984 for (i = 0; i < nmaps; i++) {
985 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
986 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
987 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
988 xfs_trans_binval(tp, bp);
991 * Unmap the dead block(s) to the free_list.
993 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
994 &first_block, &free_list, NULL, &done)))
998 * Commit the first transaction. This logs the EFI and the inode.
1000 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
1003 * The transaction must have been committed, since there were
1004 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1005 * The new tp has the extent freeing and EFDs.
1009 * The first xact was committed, so add the inode to the new one.
1010 * Mark it dirty so it will be logged and moved forward in the log as
1011 * part of every commit.
1013 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1014 xfs_trans_ihold(tp, ip);
1015 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1017 * Get a new, empty transaction to return to our caller.
1019 ntp = xfs_trans_dup(tp);
1021 * Commit the transaction containing extent freeing and EFDs.
1022 * If we get an error on the commit here or on the reserve below,
1023 * we need to unlock the inode since the new transaction doesn't
1024 * have the inode attached.
1026 error = xfs_trans_commit(tp, 0);
1029 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1033 * Remove the memory for extent descriptions (just bookkeeping).
1035 if (ip->i_df.if_bytes)
1036 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1037 ASSERT(ip->i_df.if_bytes == 0);
1039 * Put an itruncate log reservation in the new transaction
1042 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1043 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1044 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1048 * Return with the inode locked but not joined to the transaction.
1054 xfs_bmap_cancel(&free_list);
1057 * Have to come here with the inode locked and either
1058 * (held and in the transaction) or (not in the transaction).
1059 * If the inode isn't held then cancel would iput it, but
1060 * that's wrong since this is inactive and the vnode ref
1061 * count is 0 already.
1062 * Cancel won't do anything to the inode if held, but it still
1063 * needs to be locked until the cancel is done, if it was
1064 * joined to the transaction.
1066 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1067 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1074 xfs_inactive_symlink_local(
1080 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1082 * We're freeing a symlink which fit into
1083 * the inode. Just free the memory used
1084 * to hold the old symlink.
1086 error = xfs_trans_reserve(*tpp, 0,
1087 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1088 0, XFS_TRANS_PERM_LOG_RES,
1089 XFS_ITRUNCATE_LOG_COUNT);
1092 xfs_trans_cancel(*tpp, 0);
1096 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1099 * Zero length symlinks _can_ exist.
1101 if (ip->i_df.if_bytes > 0) {
1102 xfs_idata_realloc(ip,
1103 -(ip->i_df.if_bytes),
1105 ASSERT(ip->i_df.if_bytes == 0);
1119 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1122 ASSERT(ip->i_d.di_forkoff != 0);
1123 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1124 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1128 error = xfs_attr_inactive(ip);
1132 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1133 error = xfs_trans_reserve(tp, 0,
1134 XFS_IFREE_LOG_RES(mp),
1135 0, XFS_TRANS_PERM_LOG_RES,
1136 XFS_INACTIVE_LOG_COUNT);
1140 xfs_ilock(ip, XFS_ILOCK_EXCL);
1141 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1142 xfs_trans_ihold(tp, ip);
1143 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1145 ASSERT(ip->i_d.di_anextents == 0);
1151 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1152 xfs_trans_cancel(tp, 0);
1155 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1163 bhv_vnode_t *vp = VFS_I(ip);
1164 xfs_mount_t *mp = ip->i_mount;
1167 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1170 /* If this is a read-only mount, don't do this (would generate I/O) */
1171 if (mp->m_flags & XFS_MOUNT_RDONLY)
1174 if (!XFS_FORCED_SHUTDOWN(mp)) {
1178 * If we are using filestreams, and we have an unlinked
1179 * file that we are processing the last close on, then nothing
1180 * will be able to reopen and write to this file. Purge this
1181 * inode from the filestreams cache so that it doesn't delay
1182 * teardown of the inode.
1184 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1185 xfs_filestream_deassociate(ip);
1188 * If we previously truncated this file and removed old data
1189 * in the process, we want to initiate "early" writeout on
1190 * the last close. This is an attempt to combat the notorious
1191 * NULL files problem which is particularly noticable from a
1192 * truncate down, buffered (re-)write (delalloc), followed by
1193 * a crash. What we are effectively doing here is
1194 * significantly reducing the time window where we'd otherwise
1195 * be exposed to that problem.
1197 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1198 if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
1199 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
1202 if (ip->i_d.di_nlink != 0) {
1203 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1204 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1205 ip->i_delayed_blks > 0)) &&
1206 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1207 (!(ip->i_d.di_flags &
1208 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1209 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1221 * This is called when the vnode reference count for the vnode
1222 * goes to zero. If the file has been unlinked, then it must
1223 * now be truncated. Also, we clear all of the read-ahead state
1224 * kept for the inode here since the file is now closed.
1230 bhv_vnode_t *vp = VFS_I(ip);
1231 xfs_bmap_free_t free_list;
1232 xfs_fsblock_t first_block;
1239 xfs_itrace_entry(ip);
1242 * If the inode is already free, then there can be nothing
1245 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1246 ASSERT(ip->i_df.if_real_bytes == 0);
1247 ASSERT(ip->i_df.if_broot_bytes == 0);
1248 return VN_INACTIVE_CACHE;
1252 * Only do a truncate if it's a regular file with
1253 * some actual space in it. It's OK to look at the
1254 * inode's fields without the lock because we're the
1255 * only one with a reference to the inode.
1257 truncate = ((ip->i_d.di_nlink == 0) &&
1258 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1259 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1260 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1264 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1265 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
1269 /* If this is a read-only mount, don't do this (would generate I/O) */
1270 if (mp->m_flags & XFS_MOUNT_RDONLY)
1273 if (ip->i_d.di_nlink != 0) {
1274 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1275 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1276 ip->i_delayed_blks > 0)) &&
1277 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1278 (!(ip->i_d.di_flags &
1279 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1280 (ip->i_delayed_blks != 0)))) {
1281 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1283 return VN_INACTIVE_CACHE;
1288 ASSERT(ip->i_d.di_nlink == 0);
1290 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1291 return VN_INACTIVE_CACHE;
1293 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1296 * Do the xfs_itruncate_start() call before
1297 * reserving any log space because itruncate_start
1298 * will call into the buffer cache and we can't
1299 * do that within a transaction.
1301 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1303 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1305 xfs_trans_cancel(tp, 0);
1306 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1307 return VN_INACTIVE_CACHE;
1310 error = xfs_trans_reserve(tp, 0,
1311 XFS_ITRUNCATE_LOG_RES(mp),
1312 0, XFS_TRANS_PERM_LOG_RES,
1313 XFS_ITRUNCATE_LOG_COUNT);
1315 /* Don't call itruncate_cleanup */
1316 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1317 xfs_trans_cancel(tp, 0);
1318 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1319 return VN_INACTIVE_CACHE;
1322 xfs_ilock(ip, XFS_ILOCK_EXCL);
1323 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1324 xfs_trans_ihold(tp, ip);
1327 * normally, we have to run xfs_itruncate_finish sync.
1328 * But if filesystem is wsync and we're in the inactive
1329 * path, then we know that nlink == 0, and that the
1330 * xaction that made nlink == 0 is permanently committed
1331 * since xfs_remove runs as a synchronous transaction.
1333 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1334 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1337 xfs_trans_cancel(tp,
1338 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1339 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1340 return VN_INACTIVE_CACHE;
1342 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1345 * If we get an error while cleaning up a
1346 * symlink we bail out.
1348 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1349 xfs_inactive_symlink_rmt(ip, &tp) :
1350 xfs_inactive_symlink_local(ip, &tp);
1354 return VN_INACTIVE_CACHE;
1357 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1358 xfs_trans_ihold(tp, ip);
1360 error = xfs_trans_reserve(tp, 0,
1361 XFS_IFREE_LOG_RES(mp),
1362 0, XFS_TRANS_PERM_LOG_RES,
1363 XFS_INACTIVE_LOG_COUNT);
1365 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1366 xfs_trans_cancel(tp, 0);
1367 return VN_INACTIVE_CACHE;
1370 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1371 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1372 xfs_trans_ihold(tp, ip);
1376 * If there are attributes associated with the file
1377 * then blow them away now. The code calls a routine
1378 * that recursively deconstructs the attribute fork.
1379 * We need to just commit the current transaction
1380 * because we can't use it for xfs_attr_inactive().
1382 if (ip->i_d.di_anextents > 0) {
1383 error = xfs_inactive_attrs(ip, &tp);
1385 * If we got an error, the transaction is already
1386 * cancelled, and the inode is unlocked. Just get out.
1389 return VN_INACTIVE_CACHE;
1390 } else if (ip->i_afp) {
1391 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1397 XFS_BMAP_INIT(&free_list, &first_block);
1398 error = xfs_ifree(tp, ip, &free_list);
1401 * If we fail to free the inode, shut down. The cancel
1402 * might do that, we need to make sure. Otherwise the
1403 * inode might be lost for a long time or forever.
1405 if (!XFS_FORCED_SHUTDOWN(mp)) {
1407 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1408 error, mp->m_fsname);
1409 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1411 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1414 * Credit the quota account(s). The inode is gone.
1416 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1419 * Just ignore errors at this point. There is nothing we can
1420 * do except to try to keep going. Make sure it's not a silent
1423 error = xfs_bmap_finish(&tp, &free_list, &committed);
1425 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1426 "xfs_bmap_finish() returned error %d", error);
1427 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1429 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1430 "xfs_trans_commit() returned error %d", error);
1433 * Release the dquots held by inode, if any.
1435 XFS_QM_DQDETACH(mp, ip);
1437 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1440 return VN_INACTIVE_CACHE;
1444 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1445 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1446 * ci_name->name will point to a the actual name (caller must free) or
1447 * will be set to NULL if an exact match is found.
1452 struct xfs_name *name,
1454 struct xfs_name *ci_name)
1460 xfs_itrace_entry(dp);
1462 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1463 return XFS_ERROR(EIO);
1465 lock_mode = xfs_ilock_map_shared(dp);
1466 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
1467 xfs_iunlock_map_shared(dp, lock_mode);
1472 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
1476 xfs_itrace_ref(*ipp);
1481 kmem_free(ci_name->name);
1490 struct xfs_name *name,
1496 xfs_mount_t *mp = dp->i_mount;
1500 xfs_bmap_free_t free_list;
1501 xfs_fsblock_t first_block;
1502 boolean_t unlock_dp_on_error = B_FALSE;
1503 int dm_event_sent = 0;
1507 struct xfs_dquot *udqp, *gdqp;
1511 xfs_itrace_entry(dp);
1513 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
1514 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1515 dp, DM_RIGHT_NULL, NULL,
1516 DM_RIGHT_NULL, name->name, NULL,
1524 if (XFS_FORCED_SHUTDOWN(mp))
1525 return XFS_ERROR(EIO);
1527 /* Return through std_return after this point. */
1530 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1531 prid = dp->i_d.di_projid;
1533 prid = (xfs_prid_t)dfltprid;
1536 * Make sure that we have allocated dquot(s) on disk.
1538 error = XFS_QM_DQVOPALLOC(mp, dp,
1539 current_fsuid(credp), current_fsgid(credp), prid,
1540 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1546 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1547 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1548 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1550 * Initially assume that the file does not exist and
1551 * reserve the resources for that case. If that is not
1552 * the case we'll drop the one we have and get a more
1553 * appropriate transaction later.
1555 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1556 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1557 if (error == ENOSPC) {
1559 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1560 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1567 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1568 unlock_dp_on_error = B_TRUE;
1570 XFS_BMAP_INIT(&free_list, &first_block);
1575 * Reserve disk quota and the inode.
1577 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1581 error = xfs_dir_canenter(tp, dp, name, resblks);
1584 error = xfs_dir_ialloc(&tp, dp, mode, 1,
1585 rdev, credp, prid, resblks > 0,
1588 if (error == ENOSPC)
1595 * At this point, we've gotten a newly allocated inode.
1596 * It is locked (and joined to the transaction).
1599 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1602 * Now we join the directory inode to the transaction. We do not do it
1603 * earlier because xfs_dir_ialloc might commit the previous transaction
1604 * (and release all the locks). An error from here on will result in
1605 * the transaction cancel unlocking dp so don't do it explicitly in the
1609 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1610 unlock_dp_on_error = B_FALSE;
1612 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1613 &first_block, &free_list, resblks ?
1614 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1616 ASSERT(error != ENOSPC);
1619 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1620 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1623 * If this is a synchronous mount, make sure that the
1624 * create transaction goes to disk before returning to
1627 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1628 xfs_trans_set_sync(tp);
1634 * Attach the dquot(s) to the inodes and modify them incore.
1635 * These ids of the inode couldn't have changed since the new
1636 * inode has been locked ever since it was created.
1638 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
1641 * xfs_trans_commit normally decrements the vnode ref count
1642 * when it unlocks the inode. Since we want to return the
1643 * vnode to the caller, we bump the vnode ref count now.
1647 error = xfs_bmap_finish(&tp, &free_list, &committed);
1649 xfs_bmap_cancel(&free_list);
1653 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1660 XFS_QM_DQRELE(mp, udqp);
1661 XFS_QM_DQRELE(mp, gdqp);
1665 /* Fallthrough to std_return with error = 0 */
1668 if ((*ipp || (error != 0 && dm_event_sent != 0)) &&
1669 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
1670 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
1673 DM_RIGHT_NULL, name->name, NULL,
1679 cancel_flags |= XFS_TRANS_ABORT;
1684 xfs_trans_cancel(tp, cancel_flags);
1686 XFS_QM_DQRELE(mp, udqp);
1687 XFS_QM_DQRELE(mp, gdqp);
1689 if (unlock_dp_on_error)
1690 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1696 * Wait until after the current transaction is aborted to
1697 * release the inode. This prevents recursive transactions
1698 * and deadlocks from xfs_inactive.
1700 cancel_flags |= XFS_TRANS_ABORT;
1701 xfs_trans_cancel(tp, cancel_flags);
1704 XFS_QM_DQRELE(mp, udqp);
1705 XFS_QM_DQRELE(mp, gdqp);
1712 int xfs_small_retries;
1713 int xfs_middle_retries;
1714 int xfs_lots_retries;
1715 int xfs_lock_delays;
1719 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1723 xfs_lock_inumorder(int lock_mode, int subclass)
1725 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1726 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1727 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1728 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1734 * The following routine will lock n inodes in exclusive mode.
1735 * We assume the caller calls us with the inodes in i_ino order.
1737 * We need to detect deadlock where an inode that we lock
1738 * is in the AIL and we start waiting for another inode that is locked
1739 * by a thread in a long running transaction (such as truncate). This can
1740 * result in deadlock since the long running trans might need to wait
1741 * for the inode we just locked in order to push the tail and free space
1750 int attempts = 0, i, j, try_lock;
1753 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1759 for (; i < inodes; i++) {
1762 if (i && (ips[i] == ips[i-1])) /* Already locked */
1766 * If try_lock is not set yet, make sure all locked inodes
1767 * are not in the AIL.
1768 * If any are, set try_lock to be used later.
1772 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1773 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1774 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1781 * If any of the previous locks we have locked is in the AIL,
1782 * we must TRY to get the second and subsequent locks. If
1783 * we can't get any, we must release all we have
1788 /* try_lock must be 0 if i is 0. */
1790 * try_lock means we have an inode locked
1791 * that is in the AIL.
1794 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1798 * Unlock all previous guys and try again.
1799 * xfs_iunlock will try to push the tail
1800 * if the inode is in the AIL.
1803 for(j = i - 1; j >= 0; j--) {
1806 * Check to see if we've already
1807 * unlocked this one.
1808 * Not the first one going back,
1809 * and the inode ptr is the same.
1811 if ((j != (i - 1)) && ips[j] ==
1815 xfs_iunlock(ips[j], lock_mode);
1818 if ((attempts % 5) == 0) {
1819 delay(1); /* Don't just spin the CPU */
1829 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1835 if (attempts < 5) xfs_small_retries++;
1836 else if (attempts < 100) xfs_middle_retries++;
1837 else xfs_lots_retries++;
1845 xfs_lock_two_inodes(
1854 ASSERT(ip0->i_ino != ip1->i_ino);
1856 if (ip0->i_ino > ip1->i_ino) {
1863 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1866 * If the first lock we have locked is in the AIL, we must TRY to get
1867 * the second lock. If we can't get it, we must release the first one
1870 lp = (xfs_log_item_t *)ip0->i_itemp;
1871 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1872 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1873 xfs_iunlock(ip0, lock_mode);
1874 if ((++attempts % 5) == 0)
1875 delay(1); /* Don't just spin the CPU */
1879 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1886 struct xfs_name *name,
1889 xfs_mount_t *mp = dp->i_mount;
1890 xfs_trans_t *tp = NULL;
1891 int is_dir = S_ISDIR(ip->i_d.di_mode);
1893 xfs_bmap_free_t free_list;
1894 xfs_fsblock_t first_block;
1901 xfs_itrace_entry(dp);
1902 xfs_itrace_entry(ip);
1904 if (XFS_FORCED_SHUTDOWN(mp))
1905 return XFS_ERROR(EIO);
1907 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
1908 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL,
1909 NULL, DM_RIGHT_NULL, name->name, NULL,
1910 ip->i_d.di_mode, 0, 0);
1915 error = XFS_QM_DQATTACH(mp, dp, 0);
1919 error = XFS_QM_DQATTACH(mp, ip, 0);
1924 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1925 log_count = XFS_DEFAULT_LOG_COUNT;
1927 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1928 log_count = XFS_REMOVE_LOG_COUNT;
1930 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1933 * We try to get the real space reservation first,
1934 * allowing for directory btree deletion(s) implying
1935 * possible bmap insert(s). If we can't get the space
1936 * reservation then we use 0 instead, and avoid the bmap
1937 * btree insert(s) in the directory code by, if the bmap
1938 * insert tries to happen, instead trimming the LAST
1939 * block from the directory.
1941 resblks = XFS_REMOVE_SPACE_RES(mp);
1942 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1943 XFS_TRANS_PERM_LOG_RES, log_count);
1944 if (error == ENOSPC) {
1946 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1947 XFS_TRANS_PERM_LOG_RES, log_count);
1950 ASSERT(error != ENOSPC);
1952 goto out_trans_cancel;
1955 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1958 * At this point, we've gotten both the directory and the entry
1962 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1965 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1968 * If we're removing a directory perform some additional validation.
1971 ASSERT(ip->i_d.di_nlink >= 2);
1972 if (ip->i_d.di_nlink != 2) {
1973 error = XFS_ERROR(ENOTEMPTY);
1974 goto out_trans_cancel;
1976 if (!xfs_dir_isempty(ip)) {
1977 error = XFS_ERROR(ENOTEMPTY);
1978 goto out_trans_cancel;
1982 XFS_BMAP_INIT(&free_list, &first_block);
1983 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1984 &first_block, &free_list, resblks);
1986 ASSERT(error != ENOENT);
1987 goto out_bmap_cancel;
1989 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1992 * Bump the in memory generation count on the parent
1993 * directory so that other can know that it has changed.
1996 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2000 * Drop the link from ip's "..".
2002 error = xfs_droplink(tp, dp);
2004 goto out_bmap_cancel;
2007 * Drop the link from dp to ip.
2009 error = xfs_droplink(tp, ip);
2011 goto out_bmap_cancel;
2014 * When removing a non-directory we need to log the parent
2015 * inode here for the i_gen update. For a directory this is
2016 * done implicitly by the xfs_droplink call for the ".." entry.
2018 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2022 * Drop the "." link from ip to self.
2024 error = xfs_droplink(tp, ip);
2026 goto out_bmap_cancel;
2029 * Determine if this is the last link while
2030 * we are in the transaction.
2032 link_zero = (ip->i_d.di_nlink == 0);
2035 * If this is a synchronous mount, make sure that the
2036 * remove transaction goes to disk before returning to
2039 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2040 xfs_trans_set_sync(tp);
2042 error = xfs_bmap_finish(&tp, &free_list, &committed);
2044 goto out_bmap_cancel;
2046 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2051 * If we are using filestreams, kill the stream association.
2052 * If the file is still open it may get a new one but that
2053 * will get killed on last close in xfs_close() so we don't
2054 * have to worry about that.
2056 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2057 xfs_filestream_deassociate(ip);
2059 xfs_itrace_exit(ip);
2060 xfs_itrace_exit(dp);
2063 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
2064 XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL,
2065 NULL, DM_RIGHT_NULL, name->name, NULL,
2066 ip->i_d.di_mode, error, 0);
2072 xfs_bmap_cancel(&free_list);
2073 cancel_flags |= XFS_TRANS_ABORT;
2075 xfs_trans_cancel(tp, cancel_flags);
2083 struct xfs_name *target_name)
2085 xfs_mount_t *mp = tdp->i_mount;
2088 xfs_bmap_free_t free_list;
2089 xfs_fsblock_t first_block;
2094 xfs_itrace_entry(tdp);
2095 xfs_itrace_entry(sip);
2097 ASSERT(!S_ISDIR(sip->i_d.di_mode));
2099 if (XFS_FORCED_SHUTDOWN(mp))
2100 return XFS_ERROR(EIO);
2102 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
2103 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2106 target_name->name, NULL, 0, 0, 0);
2111 /* Return through std_return after this point. */
2113 error = XFS_QM_DQATTACH(mp, sip, 0);
2114 if (!error && sip != tdp)
2115 error = XFS_QM_DQATTACH(mp, tdp, 0);
2119 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2120 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2121 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
2122 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2123 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2124 if (error == ENOSPC) {
2126 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2127 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2134 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
2137 * Increment vnode ref counts since xfs_trans_commit &
2138 * xfs_trans_cancel will both unlock the inodes and
2139 * decrement the associated ref counts.
2143 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2144 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2147 * If the source has too many links, we can't make any more to it.
2149 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2150 error = XFS_ERROR(EMLINK);
2155 * If we are using project inheritance, we only allow hard link
2156 * creation in our tree when the project IDs are the same; else
2157 * the tree quota mechanism could be circumvented.
2159 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2160 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2161 error = XFS_ERROR(EXDEV);
2165 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
2169 XFS_BMAP_INIT(&free_list, &first_block);
2171 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
2172 &first_block, &free_list, resblks);
2175 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2177 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2179 error = xfs_bumplink(tp, sip);
2184 * If this is a synchronous mount, make sure that the
2185 * link transaction goes to disk before returning to
2188 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2189 xfs_trans_set_sync(tp);
2192 error = xfs_bmap_finish (&tp, &free_list, &committed);
2194 xfs_bmap_cancel(&free_list);
2198 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2202 /* Fall through to std_return with error = 0. */
2204 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
2205 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2208 target_name->name, NULL, 0, error, 0);
2213 cancel_flags |= XFS_TRANS_ABORT;
2217 xfs_trans_cancel(tp, cancel_flags);
2225 struct xfs_name *dir_name,
2230 xfs_mount_t *mp = dp->i_mount;
2231 xfs_inode_t *cdp; /* inode of created dir */
2236 xfs_bmap_free_t free_list;
2237 xfs_fsblock_t first_block;
2238 boolean_t unlock_dp_on_error = B_FALSE;
2239 boolean_t created = B_FALSE;
2240 int dm_event_sent = 0;
2242 struct xfs_dquot *udqp, *gdqp;
2245 if (XFS_FORCED_SHUTDOWN(mp))
2246 return XFS_ERROR(EIO);
2250 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
2251 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2252 dp, DM_RIGHT_NULL, NULL,
2253 DM_RIGHT_NULL, dir_name->name, NULL,
2260 /* Return through std_return after this point. */
2262 xfs_itrace_entry(dp);
2266 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2267 prid = dp->i_d.di_projid;
2269 prid = (xfs_prid_t)dfltprid;
2272 * Make sure that we have allocated dquot(s) on disk.
2274 error = XFS_QM_DQVOPALLOC(mp, dp,
2275 current_fsuid(credp), current_fsgid(credp), prid,
2276 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2280 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2281 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2282 resblks = XFS_MKDIR_SPACE_RES(mp, dir_name->len);
2283 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2284 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2285 if (error == ENOSPC) {
2287 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2288 XFS_TRANS_PERM_LOG_RES,
2289 XFS_MKDIR_LOG_COUNT);
2296 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2297 unlock_dp_on_error = B_TRUE;
2300 * Check for directory link count overflow.
2302 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2303 error = XFS_ERROR(EMLINK);
2308 * Reserve disk quota and the inode.
2310 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2314 error = xfs_dir_canenter(tp, dp, dir_name, resblks);
2318 * create the directory inode.
2320 error = xfs_dir_ialloc(&tp, dp, mode, 2,
2321 0, credp, prid, resblks > 0,
2324 if (error == ENOSPC)
2328 xfs_itrace_ref(cdp);
2331 * Now we add the directory inode to the transaction.
2332 * We waited until now since xfs_dir_ialloc might start
2333 * a new transaction. Had we joined the transaction
2334 * earlier, the locks might have gotten released. An error
2335 * from here on will result in the transaction cancel
2336 * unlocking dp so don't do it explicitly in the error path.
2339 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2340 unlock_dp_on_error = B_FALSE;
2342 XFS_BMAP_INIT(&free_list, &first_block);
2344 error = xfs_dir_createname(tp, dp, dir_name, cdp->i_ino,
2345 &first_block, &free_list, resblks ?
2346 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2348 ASSERT(error != ENOSPC);
2351 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2354 * Bump the in memory version number of the parent directory
2355 * so that other processes accessing it will recognize that
2356 * the directory has changed.
2360 error = xfs_dir_init(tp, cdp, dp);
2365 error = xfs_bumplink(tp, dp);
2375 * Attach the dquots to the new inode and modify the icount incore.
2377 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2380 * If this is a synchronous mount, make sure that the
2381 * mkdir transaction goes to disk before returning to
2384 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2385 xfs_trans_set_sync(tp);
2388 error = xfs_bmap_finish(&tp, &free_list, &committed);
2394 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2395 XFS_QM_DQRELE(mp, udqp);
2396 XFS_QM_DQRELE(mp, gdqp);
2401 /* Fall through to std_return with error = 0 or errno from
2402 * xfs_trans_commit. */
2405 if ((created || (error != 0 && dm_event_sent != 0)) &&
2406 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
2407 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2409 created ? cdp : NULL,
2411 dir_name->name, NULL,
2418 xfs_bmap_cancel(&free_list);
2420 cancel_flags |= XFS_TRANS_ABORT;
2422 xfs_trans_cancel(tp, cancel_flags);
2423 XFS_QM_DQRELE(mp, udqp);
2424 XFS_QM_DQRELE(mp, gdqp);
2426 if (unlock_dp_on_error)
2427 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2435 struct xfs_name *link_name,
2436 const char *target_path,
2441 xfs_mount_t *mp = dp->i_mount;
2446 xfs_bmap_free_t free_list;
2447 xfs_fsblock_t first_block;
2448 boolean_t unlock_dp_on_error = B_FALSE;
2451 xfs_fileoff_t first_fsb;
2452 xfs_filblks_t fs_blocks;
2454 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
2456 const char *cur_chunk;
2461 struct xfs_dquot *udqp, *gdqp;
2469 xfs_itrace_entry(dp);
2471 if (XFS_FORCED_SHUTDOWN(mp))
2472 return XFS_ERROR(EIO);
2475 * Check component lengths of the target path name.
2477 pathlen = strlen(target_path);
2478 if (pathlen >= MAXPATHLEN) /* total string too long */
2479 return XFS_ERROR(ENAMETOOLONG);
2481 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
2482 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
2483 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2484 link_name->name, target_path, 0, 0, 0);
2489 /* Return through std_return after this point. */
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_SYMLINK);
2507 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2509 * The symlink will fit into the inode data fork?
2510 * There can't be any attributes so we get the whole variable part.
2512 if (pathlen <= XFS_LITINO(mp))
2515 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
2516 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
2517 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2518 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2519 if (error == ENOSPC && fs_blocks == 0) {
2521 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2522 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2529 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2530 unlock_dp_on_error = B_TRUE;
2533 * Check whether the directory allows new symlinks or not.
2535 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2536 error = XFS_ERROR(EPERM);
2541 * Reserve disk quota : blocks and inode.
2543 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2548 * Check for ability to enter directory entry, if no space reserved.
2550 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2554 * Initialize the bmap freelist prior to calling either
2555 * bmapi or the directory create code.
2557 XFS_BMAP_INIT(&free_list, &first_block);
2560 * Allocate an inode for the symlink.
2562 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
2563 1, 0, credp, prid, resblks > 0, &ip, NULL);
2565 if (error == ENOSPC)
2572 * An error after we've joined dp to the transaction will result in the
2573 * transaction cancel unlocking dp so don't do it explicitly in the
2577 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2578 unlock_dp_on_error = B_FALSE;
2581 * Also attach the dquot(s) to it, if applicable.
2583 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2586 resblks -= XFS_IALLOC_SPACE_RES(mp);
2588 * If the symlink will fit into the inode, write it inline.
2590 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2591 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2592 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2593 ip->i_d.di_size = pathlen;
2596 * The inode was initially created in extent format.
2598 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2599 ip->i_df.if_flags |= XFS_IFINLINE;
2601 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2602 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2606 nmaps = SYMLINK_MAPS;
2608 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2609 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2610 &first_block, resblks, mval, &nmaps,
2617 resblks -= fs_blocks;
2618 ip->i_d.di_size = pathlen;
2619 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2621 cur_chunk = target_path;
2622 for (n = 0; n < nmaps; n++) {
2623 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2624 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2625 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2626 BTOBB(byte_cnt), 0);
2627 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2628 if (pathlen < byte_cnt) {
2631 pathlen -= byte_cnt;
2633 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2634 cur_chunk += byte_cnt;
2636 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2641 * Create the directory entry for the symlink.
2643 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2644 &first_block, &free_list, resblks);
2647 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2648 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2651 * Bump the in memory version number of the parent directory
2652 * so that other processes accessing it will recognize that
2653 * the directory has changed.
2658 * If this is a synchronous mount, make sure that the
2659 * symlink transaction goes to disk before returning to
2662 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2663 xfs_trans_set_sync(tp);
2667 * xfs_trans_commit normally decrements the vnode ref count
2668 * when it unlocks the inode. Since we want to return the
2669 * vnode to the caller, we bump the vnode ref count now.
2673 error = xfs_bmap_finish(&tp, &free_list, &committed);
2677 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2678 XFS_QM_DQRELE(mp, udqp);
2679 XFS_QM_DQRELE(mp, gdqp);
2681 /* Fall through to std_return with error = 0 or errno from
2682 * xfs_trans_commit */
2684 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
2685 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
2688 DM_RIGHT_NULL, link_name->name,
2689 target_path, 0, error, 0);
2699 xfs_bmap_cancel(&free_list);
2700 cancel_flags |= XFS_TRANS_ABORT;
2702 xfs_trans_cancel(tp, cancel_flags);
2703 XFS_QM_DQRELE(mp, udqp);
2704 XFS_QM_DQRELE(mp, gdqp);
2706 if (unlock_dp_on_error)
2707 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2717 xfs_mount_t *mp = ip->i_mount;
2720 if (XFS_FORCED_SHUTDOWN(mp))
2721 return XFS_ERROR(EIO);
2724 * Bypass inodes which have already been cleaned by
2725 * the inode flush clustering code inside xfs_iflush
2727 if (xfs_inode_clean(ip))
2731 * We make this non-blocking if the inode is contended,
2732 * return EAGAIN to indicate to the caller that they
2733 * did not succeed. This prevents the flush path from
2734 * blocking on inodes inside another operation right
2735 * now, they get caught later by xfs_sync.
2737 if (flags & FLUSH_SYNC) {
2738 xfs_ilock(ip, XFS_ILOCK_SHARED);
2740 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
2741 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
2742 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2749 error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC
2750 : XFS_IFLUSH_ASYNC_NOBLOCK);
2751 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2763 xfs_mount_t *mp = ip->i_mount;
2767 if (!capable(CAP_SYS_ADMIN))
2768 return XFS_ERROR(EPERM);
2770 if (XFS_FORCED_SHUTDOWN(mp))
2771 return XFS_ERROR(EIO);
2773 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2774 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2776 xfs_trans_cancel(tp, 0);
2779 xfs_ilock(ip, XFS_ILOCK_EXCL);
2780 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2782 ip->i_d.di_dmevmask = evmask;
2783 ip->i_d.di_dmstate = state;
2785 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2787 error = xfs_trans_commit(tp, 0);
2796 bhv_vnode_t *vp = VFS_I(ip);
2798 xfs_itrace_entry(ip);
2800 ASSERT(!VN_MAPPED(vp));
2802 /* bad inode, get out here ASAP */
2810 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
2813 * Make sure the atime in the XFS inode is correct before freeing the
2816 xfs_synchronize_atime(ip);
2819 * If we have nothing to flush with this inode then complete the
2820 * teardown now, otherwise break the link between the xfs inode and the
2821 * linux inode and clean up the xfs inode later. This avoids flushing
2822 * the inode to disk during the delete operation itself.
2824 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2825 * first to ensure that xfs_iunpin() will never see an xfs inode
2826 * that has a linux inode being reclaimed. Synchronisation is provided
2827 * by the i_flags_lock.
2829 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
2830 xfs_ilock(ip, XFS_ILOCK_EXCL);
2832 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
2834 xfs_mount_t *mp = ip->i_mount;
2836 /* Protect sync and unpin from us */
2837 XFS_MOUNT_ILOCK(mp);
2838 spin_lock(&ip->i_flags_lock);
2839 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
2840 vp->i_private = NULL;
2842 spin_unlock(&ip->i_flags_lock);
2843 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
2844 XFS_MOUNT_IUNLOCK(mp);
2855 xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino);
2856 bhv_vnode_t *vp = VFS_I(ip);
2858 if (vp && VN_BAD(vp))
2861 /* The hash lock here protects a thread in xfs_iget_core from
2862 * racing with us on linking the inode back with a vnode.
2863 * Once we have the XFS_IRECLAIM flag set it will not touch
2866 write_lock(&pag->pag_ici_lock);
2867 spin_lock(&ip->i_flags_lock);
2868 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
2869 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
2870 spin_unlock(&ip->i_flags_lock);
2871 write_unlock(&pag->pag_ici_lock);
2874 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2878 __xfs_iflags_set(ip, XFS_IRECLAIM);
2879 spin_unlock(&ip->i_flags_lock);
2880 write_unlock(&pag->pag_ici_lock);
2881 xfs_put_perag(ip->i_mount, pag);
2884 * If the inode is still dirty, then flush it out. If the inode
2885 * is not in the AIL, then it will be OK to flush it delwri as
2886 * long as xfs_iflush() does not keep any references to the inode.
2887 * We leave that decision up to xfs_iflush() since it has the
2888 * knowledge of whether it's OK to simply do a delwri flush of
2889 * the inode or whether we need to wait until the inode is
2890 * pulled from the AIL.
2891 * We get the flush lock regardless, though, just to make sure
2892 * we don't free it while it is being flushed.
2895 xfs_ilock(ip, XFS_ILOCK_EXCL);
2900 * In the case of a forced shutdown we rely on xfs_iflush() to
2901 * wait for the inode to be unpinned before returning an error.
2903 if (xfs_iflush(ip, sync_mode) == 0) {
2904 /* synchronize with xfs_iflush_done */
2909 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2917 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
2920 xfs_inode_t *ip, *n;
2925 XFS_MOUNT_ILOCK(mp);
2926 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
2928 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
2930 if (xfs_ipincount(ip) ||
2931 !xfs_iflock_nowait(ip)) {
2932 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2936 XFS_MOUNT_IUNLOCK(mp);
2937 if (xfs_finish_reclaim(ip, noblock,
2938 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
2947 XFS_MOUNT_IUNLOCK(mp);
2952 * xfs_alloc_file_space()
2953 * This routine allocates disk space for the given file.
2955 * If alloc_type == 0, this request is for an ALLOCSP type
2956 * request which will change the file size. In this case, no
2957 * DMAPI event will be generated by the call. A TRUNCATE event
2958 * will be generated later by xfs_setattr.
2960 * If alloc_type != 0, this request is for a RESVSP type
2961 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2962 * lower block boundary byte address is less than the file's
2971 xfs_alloc_file_space(
2978 xfs_mount_t *mp = ip->i_mount;
2980 xfs_filblks_t allocated_fsb;
2981 xfs_filblks_t allocatesize_fsb;
2982 xfs_extlen_t extsz, temp;
2983 xfs_fileoff_t startoffset_fsb;
2984 xfs_fsblock_t firstfsb;
2990 xfs_bmbt_irec_t imaps[1], *imapp;
2991 xfs_bmap_free_t free_list;
2992 uint qblocks, resblks, resrtextents;
2996 xfs_itrace_entry(ip);
2998 if (XFS_FORCED_SHUTDOWN(mp))
2999 return XFS_ERROR(EIO);
3001 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3005 return XFS_ERROR(EINVAL);
3007 rt = XFS_IS_REALTIME_INODE(ip);
3008 extsz = xfs_get_extsz_hint(ip);
3013 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
3014 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
3015 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
3017 /* Generate a DMAPI event if needed. */
3018 if (alloc_type != 0 && offset < ip->i_size &&
3019 (attr_flags & XFS_ATTR_DMI) == 0 &&
3020 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
3021 xfs_off_t end_dmi_offset;
3023 end_dmi_offset = offset+len;
3024 if (end_dmi_offset > ip->i_size)
3025 end_dmi_offset = ip->i_size;
3026 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
3027 end_dmi_offset - offset, 0, NULL);
3033 * Allocate file space until done or until there is an error
3036 while (allocatesize_fsb && !error) {
3040 * Determine space reservations for data/realtime.
3042 if (unlikely(extsz)) {
3043 s = startoffset_fsb;
3046 e = startoffset_fsb + allocatesize_fsb;
3047 if ((temp = do_mod(startoffset_fsb, extsz)))
3049 if ((temp = do_mod(e, extsz)))
3053 e = allocatesize_fsb;
3057 resrtextents = qblocks = (uint)(e - s);
3058 resrtextents /= mp->m_sb.sb_rextsize;
3059 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3060 quota_flag = XFS_QMOPT_RES_RTBLKS;
3063 resblks = qblocks = \
3064 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
3065 quota_flag = XFS_QMOPT_RES_REGBLKS;
3069 * Allocate and setup the transaction.
3071 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
3072 error = xfs_trans_reserve(tp, resblks,
3073 XFS_WRITE_LOG_RES(mp), resrtextents,
3074 XFS_TRANS_PERM_LOG_RES,
3075 XFS_WRITE_LOG_COUNT);
3077 * Check for running out of space
3081 * Free the transaction structure.
3083 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3084 xfs_trans_cancel(tp, 0);
3087 xfs_ilock(ip, XFS_ILOCK_EXCL);
3088 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
3089 qblocks, 0, quota_flag);
3093 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3094 xfs_trans_ihold(tp, ip);
3097 * Issue the xfs_bmapi() call to allocate the blocks
3099 XFS_BMAP_INIT(&free_list, &firstfsb);
3100 error = xfs_bmapi(tp, ip, startoffset_fsb,
3101 allocatesize_fsb, bmapi_flag,
3102 &firstfsb, 0, imapp, &nimaps,
3109 * Complete the transaction
3111 error = xfs_bmap_finish(&tp, &free_list, &committed);
3116 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3117 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3122 allocated_fsb = imapp->br_blockcount;
3125 error = XFS_ERROR(ENOSPC);
3129 startoffset_fsb += allocated_fsb;
3130 allocatesize_fsb -= allocated_fsb;
3133 if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 &&
3134 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
3135 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
3138 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
3140 goto retry; /* Maybe DMAPI app. has made space */
3141 /* else fall through with error from XFS_SEND_DATA */
3146 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
3147 xfs_bmap_cancel(&free_list);
3148 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
3150 error1: /* Just cancel transaction */
3151 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3152 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3153 goto dmapi_enospc_check;
3157 * Zero file bytes between startoff and endoff inclusive.
3158 * The iolock is held exclusive and no blocks are buffered.
3161 xfs_zero_remaining_bytes(
3166 xfs_bmbt_irec_t imap;
3167 xfs_fileoff_t offset_fsb;
3168 xfs_off_t lastoffset;
3171 xfs_mount_t *mp = ip->i_mount;
3175 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
3176 XFS_IS_REALTIME_INODE(ip) ?
3177 mp->m_rtdev_targp : mp->m_ddev_targp);
3179 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
3180 offset_fsb = XFS_B_TO_FSBT(mp, offset);
3182 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
3183 NULL, 0, &imap, &nimap, NULL, NULL);
3184 if (error || nimap < 1)
3186 ASSERT(imap.br_blockcount >= 1);
3187 ASSERT(imap.br_startoff == offset_fsb);
3188 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
3189 if (lastoffset > endoff)
3190 lastoffset = endoff;
3191 if (imap.br_startblock == HOLESTARTBLOCK)
3193 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3194 if (imap.br_state == XFS_EXT_UNWRITTEN)
3197 XFS_BUF_UNWRITE(bp);
3199 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
3201 error = xfs_iowait(bp);
3203 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
3204 mp, bp, XFS_BUF_ADDR(bp));
3207 memset(XFS_BUF_PTR(bp) +
3208 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
3209 0, lastoffset - offset + 1);
3214 error = xfs_iowait(bp);
3216 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
3217 mp, bp, XFS_BUF_ADDR(bp));
3226 * xfs_free_file_space()
3227 * This routine frees disk space for the given file.
3229 * This routine is only called by xfs_change_file_space
3230 * for an UNRESVSP type call.
3238 xfs_free_file_space(
3247 xfs_off_t end_dmi_offset;
3248 xfs_fileoff_t endoffset_fsb;
3250 xfs_fsblock_t firstfsb;
3251 xfs_bmap_free_t free_list;
3252 xfs_bmbt_irec_t imap;
3260 xfs_fileoff_t startoffset_fsb;
3262 int need_iolock = 1;
3267 xfs_itrace_entry(ip);
3269 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3273 if (len <= 0) /* if nothing being freed */
3275 rt = XFS_IS_REALTIME_INODE(ip);
3276 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
3277 end_dmi_offset = offset + len;
3278 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
3280 if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 &&
3281 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
3282 if (end_dmi_offset > ip->i_size)
3283 end_dmi_offset = ip->i_size;
3284 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
3285 offset, end_dmi_offset - offset,
3286 AT_DELAY_FLAG(attr_flags), NULL);
3291 if (attr_flags & XFS_ATTR_NOLOCK)
3294 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3295 vn_iowait(ip); /* wait for the completion of any pending DIOs */
3298 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
3299 ioffset = offset & ~(rounding - 1);
3301 if (VN_CACHED(vp) != 0) {
3302 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
3303 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
3305 goto out_unlock_iolock;
3309 * Need to zero the stuff we're not freeing, on disk.
3310 * If its a realtime file & can't use unwritten extents then we
3311 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3312 * will take care of it for us.
3314 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
3316 error = xfs_bmapi(NULL, ip, startoffset_fsb,
3317 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
3319 goto out_unlock_iolock;
3320 ASSERT(nimap == 0 || nimap == 1);
3321 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3324 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3325 block = imap.br_startblock;
3326 mod = do_div(block, mp->m_sb.sb_rextsize);
3328 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
3331 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
3332 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
3334 goto out_unlock_iolock;
3335 ASSERT(nimap == 0 || nimap == 1);
3336 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3337 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3339 if (mod && (mod != mp->m_sb.sb_rextsize))
3340 endoffset_fsb -= mod;
3343 if ((done = (endoffset_fsb <= startoffset_fsb)))
3345 * One contiguous piece to clear
3347 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
3350 * Some full blocks, possibly two pieces to clear
3352 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
3353 error = xfs_zero_remaining_bytes(ip, offset,
3354 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
3356 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
3357 error = xfs_zero_remaining_bytes(ip,
3358 XFS_FSB_TO_B(mp, endoffset_fsb),
3363 * free file space until done or until there is an error
3365 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3366 while (!error && !done) {
3369 * allocate and setup the transaction. Allow this
3370 * transaction to dip into the reserve blocks to ensure
3371 * the freeing of the space succeeds at ENOSPC.
3373 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
3374 tp->t_flags |= XFS_TRANS_RESERVE;
3375 error = xfs_trans_reserve(tp,
3377 XFS_WRITE_LOG_RES(mp),
3379 XFS_TRANS_PERM_LOG_RES,
3380 XFS_WRITE_LOG_COUNT);
3383 * check for running out of space
3387 * Free the transaction structure.
3389 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3390 xfs_trans_cancel(tp, 0);
3393 xfs_ilock(ip, XFS_ILOCK_EXCL);
3394 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
3395 ip->i_udquot, ip->i_gdquot, resblks, 0,
3396 XFS_QMOPT_RES_REGBLKS);
3400 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3401 xfs_trans_ihold(tp, ip);
3404 * issue the bunmapi() call to free the blocks
3406 XFS_BMAP_INIT(&free_list, &firstfsb);
3407 error = xfs_bunmapi(tp, ip, startoffset_fsb,
3408 endoffset_fsb - startoffset_fsb,
3409 0, 2, &firstfsb, &free_list, NULL, &done);
3415 * complete the transaction
3417 error = xfs_bmap_finish(&tp, &free_list, &committed);
3422 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3423 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3428 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3432 xfs_bmap_cancel(&free_list);
3434 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3435 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
3441 * xfs_change_file_space()
3442 * This routine allocates or frees disk space for the given file.
3443 * The user specified parameters are checked for alignment and size
3452 xfs_change_file_space(
3460 xfs_mount_t *mp = ip->i_mount;
3465 xfs_off_t startoffset;
3470 xfs_itrace_entry(ip);
3472 if (!S_ISREG(ip->i_d.di_mode))
3473 return XFS_ERROR(EINVAL);
3475 switch (bf->l_whence) {
3476 case 0: /*SEEK_SET*/
3478 case 1: /*SEEK_CUR*/
3479 bf->l_start += offset;
3481 case 2: /*SEEK_END*/
3482 bf->l_start += ip->i_size;
3485 return XFS_ERROR(EINVAL);
3488 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
3490 if ( (bf->l_start < 0)
3491 || (bf->l_start > XFS_MAXIOFFSET(mp))
3492 || (bf->l_start + llen < 0)
3493 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
3494 return XFS_ERROR(EINVAL);
3498 startoffset = bf->l_start;
3502 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3504 * These calls do NOT zero the data space allocated to the file,
3505 * nor do they change the file size.
3507 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3509 * These calls cause the new file data to be zeroed and the file
3510 * size to be changed.
3512 setprealloc = clrprealloc = 0;
3515 case XFS_IOC_RESVSP:
3516 case XFS_IOC_RESVSP64:
3517 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
3524 case XFS_IOC_UNRESVSP:
3525 case XFS_IOC_UNRESVSP64:
3526 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
3531 case XFS_IOC_ALLOCSP:
3532 case XFS_IOC_ALLOCSP64:
3533 case XFS_IOC_FREESP:
3534 case XFS_IOC_FREESP64:
3535 if (startoffset > fsize) {
3536 error = xfs_alloc_file_space(ip, fsize,
3537 startoffset - fsize, 0, attr_flags);
3542 iattr.ia_valid = ATTR_SIZE;
3543 iattr.ia_size = startoffset;
3545 error = xfs_setattr(ip, &iattr, attr_flags, credp);
3555 return XFS_ERROR(EINVAL);
3559 * update the inode timestamp, mode, and prealloc flag bits
3561 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
3563 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
3566 xfs_trans_cancel(tp, 0);
3570 xfs_ilock(ip, XFS_ILOCK_EXCL);
3572 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3573 xfs_trans_ihold(tp, ip);
3575 if ((attr_flags & XFS_ATTR_DMI) == 0) {
3576 ip->i_d.di_mode &= ~S_ISUID;
3579 * Note that we don't have to worry about mandatory
3580 * file locking being disabled here because we only
3581 * clear the S_ISGID bit if the Group execute bit is
3582 * on, but if it was on then mandatory locking wouldn't
3583 * have been enabled.
3585 if (ip->i_d.di_mode & S_IXGRP)
3586 ip->i_d.di_mode &= ~S_ISGID;
3588 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3591 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
3592 else if (clrprealloc)
3593 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
3595 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3596 xfs_trans_set_sync(tp);
3598 error = xfs_trans_commit(tp, 0);
3600 xfs_iunlock(ip, XFS_ILOCK_EXCL);