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"
48 #include "xfs_error.h"
49 #include "xfs_quota.h"
50 #include "xfs_utils.h"
51 #include "xfs_rtalloc.h"
52 #include "xfs_trans_space.h"
53 #include "xfs_log_priv.h"
54 #include "xfs_filestream.h"
55 #include "xfs_vnodeops.h"
63 xfs_mount_t *mp = ip->i_mount;
64 struct inode *inode = VFS_I(ip);
65 int mask = iattr->ia_valid;
72 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
77 if (mp->m_flags & XFS_MOUNT_RDONLY)
78 return XFS_ERROR(EROFS);
80 if (XFS_FORCED_SHUTDOWN(mp))
81 return XFS_ERROR(EIO);
83 code = -inode_change_ok(inode, iattr);
87 olddquot1 = olddquot2 = NULL;
91 * If disk quotas is on, we make sure that the dquots do exist on disk,
92 * before we start any other transactions. Trying to do this later
93 * is messy. We don't care to take a readlock to look at the ids
94 * in inode here, because we can't hold it across the trans_reserve.
95 * If the IDs do change before we take the ilock, we're covered
96 * because the i_*dquot fields will get updated anyway.
98 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
101 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
103 qflags |= XFS_QMOPT_UQUOTA;
105 uid = ip->i_d.di_uid;
107 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
109 qflags |= XFS_QMOPT_GQUOTA;
111 gid = ip->i_d.di_gid;
115 * We take a reference when we initialize udqp and gdqp,
116 * so it is important that we never blindly double trip on
117 * the same variable. See xfs_create() for an example.
119 ASSERT(udqp == NULL);
120 ASSERT(gdqp == NULL);
121 code = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
122 qflags, &udqp, &gdqp);
128 * For the other attributes, we acquire the inode lock and
129 * first do an error checking pass.
132 lock_flags = XFS_ILOCK_EXCL;
133 if (flags & XFS_ATTR_NOLOCK)
135 if (!(mask & ATTR_SIZE)) {
136 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
138 code = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp),
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);
162 * Change file ownership. Must be the owner or privileged.
164 if (mask & (ATTR_UID|ATTR_GID)) {
166 * These IDs could have changed since we last looked at them.
167 * But, we're assured that if the ownership did change
168 * while we didn't have the inode locked, inode's dquot(s)
169 * would have changed also.
171 iuid = ip->i_d.di_uid;
172 igid = ip->i_d.di_gid;
173 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
174 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
177 * Do a quota reservation only if uid/gid is actually
180 if (XFS_IS_QUOTA_RUNNING(mp) &&
181 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
182 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
184 code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
185 capable(CAP_FOWNER) ?
186 XFS_QMOPT_FORCE_RES : 0);
187 if (code) /* out of quota */
193 * Truncate file. Must have write permission and not be a directory.
195 if (mask & ATTR_SIZE) {
196 /* Short circuit the truncate case for zero length files */
197 if (iattr->ia_size == 0 &&
198 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
199 xfs_iunlock(ip, XFS_ILOCK_EXCL);
200 lock_flags &= ~XFS_ILOCK_EXCL;
201 if (mask & ATTR_CTIME)
202 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
207 if (S_ISDIR(ip->i_d.di_mode)) {
208 code = XFS_ERROR(EISDIR);
210 } else if (!S_ISREG(ip->i_d.di_mode)) {
211 code = XFS_ERROR(EINVAL);
216 * Make sure that the dquots are attached to the inode.
218 code = xfs_qm_dqattach_locked(ip, 0);
223 * Now we can make the changes. Before we join the inode
224 * to the transaction, if ATTR_SIZE is set then take care of
225 * the part of the truncation that must be done without the
226 * inode lock. This needs to be done before joining the inode
227 * to the transaction, because the inode cannot be unlocked
228 * once it is a part of the transaction.
230 if (iattr->ia_size > ip->i_size) {
232 * Do the first part of growing a file: zero any data
233 * in the last block that is beyond the old EOF. We
234 * need to do this before the inode is joined to the
235 * transaction to modify the i_size.
237 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
239 xfs_iunlock(ip, XFS_ILOCK_EXCL);
242 * We are going to log the inode size change in this
243 * transaction so any previous writes that are beyond the on
244 * disk EOF and the new EOF that have not been written out need
245 * to be written here. If we do not write the data out, we
246 * expose ourselves to the null files problem.
248 * Only flush from the on disk size to the smaller of the in
249 * memory file size or the new size as that's the range we
250 * really care about here and prevents waiting for other data
251 * not within the range we care about here.
254 ip->i_size != ip->i_d.di_size &&
255 iattr->ia_size > ip->i_d.di_size) {
256 code = xfs_flush_pages(ip,
257 ip->i_d.di_size, iattr->ia_size,
258 XFS_B_ASYNC, FI_NONE);
261 /* wait for all I/O to complete */
265 code = xfs_itruncate_data(ip, iattr->ia_size);
268 lock_flags &= ~XFS_ILOCK_EXCL;
269 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
272 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
273 if ((code = xfs_trans_reserve(tp, 0,
274 XFS_ITRUNCATE_LOG_RES(mp), 0,
275 XFS_TRANS_PERM_LOG_RES,
276 XFS_ITRUNCATE_LOG_COUNT))) {
277 xfs_trans_cancel(tp, 0);
279 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
282 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
283 xfs_ilock(ip, XFS_ILOCK_EXCL);
285 xfs_trans_ijoin(tp, ip, lock_flags);
286 xfs_trans_ihold(tp, ip);
289 * Only change the c/mtime if we are changing the size
290 * or we are explicitly asked to change it. This handles
291 * the semantic difference between truncate() and ftruncate()
292 * as implemented in the VFS.
294 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME
295 * is a special case where we need to update the times despite
296 * not having these flags set. For all other operations the
297 * VFS set these flags explicitly if it wants a timestamp
300 if (iattr->ia_size != ip->i_size &&
301 (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
302 iattr->ia_ctime = iattr->ia_mtime =
303 current_fs_time(inode->i_sb);
304 mask |= ATTR_CTIME | ATTR_MTIME;
307 if (iattr->ia_size > ip->i_size) {
308 ip->i_d.di_size = iattr->ia_size;
309 ip->i_size = iattr->ia_size;
310 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
311 } else if (iattr->ia_size <= ip->i_size ||
312 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
314 * signal a sync transaction unless
315 * we're truncating an already unlinked
316 * file on a wsync filesystem
318 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
320 ((ip->i_d.di_nlink != 0 ||
321 !(mp->m_flags & XFS_MOUNT_WSYNC))
326 * Truncated "down", so we're removing references
327 * to old data here - if we now delay flushing for
328 * a long time, we expose ourselves unduly to the
329 * notorious NULL files problem. So, we mark this
330 * vnode and flush it when the file is closed, and
331 * do not wait the usual (long) time for writeout.
333 xfs_iflags_set(ip, XFS_ITRUNCATED);
336 xfs_trans_ijoin(tp, ip, lock_flags);
337 xfs_trans_ihold(tp, ip);
341 * Change file ownership. Must be the owner or privileged.
343 if (mask & (ATTR_UID|ATTR_GID)) {
345 * CAP_FSETID overrides the following restrictions:
347 * The set-user-ID and set-group-ID bits of a file will be
348 * cleared upon successful return from chown()
350 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
351 !capable(CAP_FSETID)) {
352 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
356 * Change the ownerships and register quota modifications
357 * in the transaction.
360 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
361 ASSERT(mask & ATTR_UID);
363 olddquot1 = xfs_qm_vop_chown(tp, ip,
364 &ip->i_udquot, udqp);
366 ip->i_d.di_uid = uid;
370 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
371 ASSERT(!XFS_IS_PQUOTA_ON(mp));
372 ASSERT(mask & ATTR_GID);
374 olddquot2 = xfs_qm_vop_chown(tp, ip,
375 &ip->i_gdquot, gdqp);
377 ip->i_d.di_gid = gid;
383 * Change file access modes.
385 if (mask & ATTR_MODE) {
386 umode_t mode = iattr->ia_mode;
388 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
391 ip->i_d.di_mode &= S_IFMT;
392 ip->i_d.di_mode |= mode & ~S_IFMT;
394 inode->i_mode &= S_IFMT;
395 inode->i_mode |= mode & ~S_IFMT;
399 * Change file access or modified times.
401 if (mask & ATTR_ATIME) {
402 inode->i_atime = iattr->ia_atime;
403 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
404 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
405 ip->i_update_core = 1;
407 if (mask & ATTR_CTIME) {
408 inode->i_ctime = iattr->ia_ctime;
409 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
410 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
411 ip->i_update_core = 1;
413 if (mask & ATTR_MTIME) {
414 inode->i_mtime = iattr->ia_mtime;
415 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
416 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
417 ip->i_update_core = 1;
421 * And finally, log the inode core if any attribute in it
424 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE|
425 ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
426 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
428 XFS_STATS_INC(xs_ig_attrchg);
431 * If this is a synchronous mount, make sure that the
432 * transaction goes to disk before returning to the user.
433 * This is slightly sub-optimal in that truncates require
434 * two sync transactions instead of one for wsync filesystems.
435 * One for the truncate and one for the timestamps since we
436 * don't want to change the timestamps unless we're sure the
437 * truncate worked. Truncates are less than 1% of the laddis
438 * mix so this probably isn't worth the trouble to optimize.
441 if (mp->m_flags & XFS_MOUNT_WSYNC)
442 xfs_trans_set_sync(tp);
444 code = xfs_trans_commit(tp, commit_flags);
446 xfs_iunlock(ip, lock_flags);
449 * Release any dquot(s) the inode had kept before chown.
451 xfs_qm_dqrele(olddquot1);
452 xfs_qm_dqrele(olddquot2);
460 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
461 * update. We could avoid this with linked transactions
462 * and passing down the transaction pointer all the way
463 * to attr_set. No previous user of the generic
464 * Posix ACL code seems to care about this issue either.
466 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
467 code = -xfs_acl_chmod(inode);
469 return XFS_ERROR(code);
472 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
473 !(flags & XFS_ATTR_DMI)) {
474 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
475 NULL, DM_RIGHT_NULL, NULL, NULL,
476 0, 0, AT_DELAY_FLAG(flags));
481 commit_flags |= XFS_TRANS_ABORT;
487 xfs_trans_cancel(tp, commit_flags);
489 if (lock_flags != 0) {
490 xfs_iunlock(ip, lock_flags);
496 * The maximum pathlen is 1024 bytes. Since the minimum file system
497 * blocksize is 512 bytes, we can get a max of 2 extents back from
500 #define SYMLINK_MAPS 2
507 xfs_mount_t *mp = ip->i_mount;
508 int pathlen = ip->i_d.di_size;
509 int nmaps = SYMLINK_MAPS;
510 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
517 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
518 mval, &nmaps, NULL, NULL);
522 for (n = 0; n < nmaps; n++) {
523 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
524 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
526 bp = xfs_buf_read_flags(mp->m_ddev_targp, d, BTOBB(byte_cnt),
527 XBF_LOCK | XBF_MAPPED |
529 error = XFS_BUF_GETERROR(bp);
531 xfs_ioerror_alert("xfs_readlink",
532 ip->i_mount, bp, XFS_BUF_ADDR(bp));
536 if (pathlen < byte_cnt)
540 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
544 link[ip->i_d.di_size] = '\0';
556 xfs_mount_t *mp = ip->i_mount;
560 xfs_itrace_entry(ip);
562 if (XFS_FORCED_SHUTDOWN(mp))
563 return XFS_ERROR(EIO);
565 xfs_ilock(ip, XFS_ILOCK_SHARED);
567 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
568 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
570 pathlen = ip->i_d.di_size;
574 if (ip->i_df.if_flags & XFS_IFINLINE) {
575 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
576 link[pathlen] = '\0';
578 error = xfs_readlink_bmap(ip, link);
582 xfs_iunlock(ip, XFS_ILOCK_SHARED);
589 * This is called to sync the inode and its data out to disk. We need to hold
590 * the I/O lock while flushing the data, and the inode lock while flushing the
591 * inode. The inode lock CANNOT be held while flushing the data, so acquire
592 * after we're done with that.
602 xfs_itrace_entry(ip);
604 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
605 return XFS_ERROR(EIO);
608 * We always need to make sure that the required inode state is safe on
609 * disk. The inode might be clean but we still might need to force the
610 * log because of committed transactions that haven't hit the disk yet.
611 * Likewise, there could be unflushed non-transactional changes to the
612 * inode core that have to go to disk and this requires us to issue
613 * a synchronous transaction to capture these changes correctly.
615 * This code relies on the assumption that if the update_* fields
616 * of the inode are clear and the inode is unpinned then it is clean
617 * and no action is required.
619 xfs_ilock(ip, XFS_ILOCK_SHARED);
621 if (!ip->i_update_core) {
623 * Timestamps/size haven't changed since last inode flush or
624 * inode transaction commit. That means either nothing got
625 * written or a transaction committed which caught the updates.
626 * If the latter happened and the transaction hasn't hit the
627 * disk yet, the inode will be still be pinned. If it is,
630 xfs_iunlock(ip, XFS_ILOCK_SHARED);
631 if (xfs_ipincount(ip)) {
632 error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
633 XFS_LOG_FORCE | XFS_LOG_SYNC,
638 * Kick off a transaction to log the inode core to get the
639 * updates. The sync transaction will also force the log.
641 xfs_iunlock(ip, XFS_ILOCK_SHARED);
642 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
643 error = xfs_trans_reserve(tp, 0,
644 XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0);
646 xfs_trans_cancel(tp, 0);
649 xfs_ilock(ip, XFS_ILOCK_EXCL);
652 * Note - it's possible that we might have pushed ourselves out
653 * of the way during trans_reserve which would flush the inode.
654 * But there's no guarantee that the inode buffer has actually
655 * gone out yet (it's delwri). Plus the buffer could be pinned
656 * anyway if it's part of an inode in another recent
657 * transaction. So we play it safe and fire off the
658 * transaction anyway.
660 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
661 xfs_trans_ihold(tp, ip);
662 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
663 xfs_trans_set_sync(tp);
664 error = _xfs_trans_commit(tp, 0, &log_flushed);
666 xfs_iunlock(ip, XFS_ILOCK_EXCL);
669 if (ip->i_mount->m_flags & XFS_MOUNT_BARRIER) {
671 * If the log write didn't issue an ordered tag we need
672 * to flush the disk cache for the data device now.
675 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
678 * If this inode is on the RT dev we need to flush that
681 if (XFS_IS_REALTIME_INODE(ip))
682 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
689 * Flags for xfs_free_eofblocks
691 #define XFS_FREE_EOF_TRYLOCK (1<<0)
694 * This is called by xfs_inactive to free any blocks beyond eof
695 * when the link count isn't zero and by xfs_dm_punch_hole() when
696 * punching a hole to EOF.
706 xfs_fileoff_t end_fsb;
707 xfs_fileoff_t last_fsb;
708 xfs_filblks_t map_len;
710 xfs_bmbt_irec_t imap;
713 * Figure out if there are any blocks beyond the end
714 * of the file. If not, then there is nothing to do.
716 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
717 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
718 map_len = last_fsb - end_fsb;
723 xfs_ilock(ip, XFS_ILOCK_SHARED);
724 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
725 NULL, 0, &imap, &nimaps, NULL, NULL);
726 xfs_iunlock(ip, XFS_ILOCK_SHARED);
728 if (!error && (nimaps != 0) &&
729 (imap.br_startblock != HOLESTARTBLOCK ||
730 ip->i_delayed_blks)) {
732 * Attach the dquots to the inode up front.
734 error = xfs_qm_dqattach(ip, 0);
739 * There are blocks after the end of file.
740 * Free them up now by truncating the file to
743 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
746 * Do the xfs_itruncate_start() call before
747 * reserving any log space because
748 * itruncate_start will call into the buffer
750 * do that within a transaction.
752 if (flags & XFS_FREE_EOF_TRYLOCK) {
753 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
754 xfs_trans_cancel(tp, 0);
758 xfs_ilock(ip, XFS_IOLOCK_EXCL);
760 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
763 xfs_trans_cancel(tp, 0);
764 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
768 error = xfs_trans_reserve(tp, 0,
769 XFS_ITRUNCATE_LOG_RES(mp),
770 0, XFS_TRANS_PERM_LOG_RES,
771 XFS_ITRUNCATE_LOG_COUNT);
773 ASSERT(XFS_FORCED_SHUTDOWN(mp));
774 xfs_trans_cancel(tp, 0);
775 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
779 xfs_ilock(ip, XFS_ILOCK_EXCL);
780 xfs_trans_ijoin(tp, ip,
783 xfs_trans_ihold(tp, ip);
785 error = xfs_itruncate_finish(&tp, ip,
790 * If we get an error at this point we
791 * simply don't bother truncating the file.
795 (XFS_TRANS_RELEASE_LOG_RES |
798 error = xfs_trans_commit(tp,
799 XFS_TRANS_RELEASE_LOG_RES);
801 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
807 * Free a symlink that has blocks associated with it.
810 xfs_inactive_symlink_rmt(
818 xfs_fsblock_t first_block;
819 xfs_bmap_free_t free_list;
822 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
830 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
832 * We're freeing a symlink that has some
833 * blocks allocated to it. Free the
834 * blocks here. We know that we've got
835 * either 1 or 2 extents and that we can
836 * free them all in one bunmapi call.
838 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
839 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
840 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
841 ASSERT(XFS_FORCED_SHUTDOWN(mp));
842 xfs_trans_cancel(tp, 0);
847 * Lock the inode, fix the size, and join it to the transaction.
848 * Hold it so in the normal path, we still have it locked for
849 * the second transaction. In the error paths we need it
850 * held so the cancel won't rele it, see below.
852 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
853 size = (int)ip->i_d.di_size;
855 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
856 xfs_trans_ihold(tp, ip);
857 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
859 * Find the block(s) so we can inval and unmap them.
862 xfs_bmap_init(&free_list, &first_block);
863 nmaps = ARRAY_SIZE(mval);
864 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
865 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
869 * Invalidate the block(s).
871 for (i = 0; i < nmaps; i++) {
872 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
873 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
874 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
875 xfs_trans_binval(tp, bp);
878 * Unmap the dead block(s) to the free_list.
880 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
881 &first_block, &free_list, NULL, &done)))
885 * Commit the first transaction. This logs the EFI and the inode.
887 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
890 * The transaction must have been committed, since there were
891 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
892 * The new tp has the extent freeing and EFDs.
896 * The first xact was committed, so add the inode to the new one.
897 * Mark it dirty so it will be logged and moved forward in the log as
898 * part of every commit.
900 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
901 xfs_trans_ihold(tp, ip);
902 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
904 * Get a new, empty transaction to return to our caller.
906 ntp = xfs_trans_dup(tp);
908 * Commit the transaction containing extent freeing and EFDs.
909 * If we get an error on the commit here or on the reserve below,
910 * we need to unlock the inode since the new transaction doesn't
911 * have the inode attached.
913 error = xfs_trans_commit(tp, 0);
916 ASSERT(XFS_FORCED_SHUTDOWN(mp));
920 * transaction commit worked ok so we can drop the extra ticket
921 * reference that we gained in xfs_trans_dup()
923 xfs_log_ticket_put(tp->t_ticket);
926 * Remove the memory for extent descriptions (just bookkeeping).
928 if (ip->i_df.if_bytes)
929 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
930 ASSERT(ip->i_df.if_bytes == 0);
932 * Put an itruncate log reservation in the new transaction
935 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
936 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
937 ASSERT(XFS_FORCED_SHUTDOWN(mp));
941 * Return with the inode locked but not joined to the transaction.
947 xfs_bmap_cancel(&free_list);
950 * Have to come here with the inode locked and either
951 * (held and in the transaction) or (not in the transaction).
952 * If the inode isn't held then cancel would iput it, but
953 * that's wrong since this is inactive and the vnode ref
954 * count is 0 already.
955 * Cancel won't do anything to the inode if held, but it still
956 * needs to be locked until the cancel is done, if it was
957 * joined to the transaction.
959 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
960 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
967 xfs_inactive_symlink_local(
973 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
975 * We're freeing a symlink which fit into
976 * the inode. Just free the memory used
977 * to hold the old symlink.
979 error = xfs_trans_reserve(*tpp, 0,
980 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
981 0, XFS_TRANS_PERM_LOG_RES,
982 XFS_ITRUNCATE_LOG_COUNT);
985 xfs_trans_cancel(*tpp, 0);
989 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
992 * Zero length symlinks _can_ exist.
994 if (ip->i_df.if_bytes > 0) {
995 xfs_idata_realloc(ip,
996 -(ip->i_df.if_bytes),
998 ASSERT(ip->i_df.if_bytes == 0);
1012 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1015 ASSERT(ip->i_d.di_forkoff != 0);
1016 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1017 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1021 error = xfs_attr_inactive(ip);
1025 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1026 error = xfs_trans_reserve(tp, 0,
1027 XFS_IFREE_LOG_RES(mp),
1028 0, XFS_TRANS_PERM_LOG_RES,
1029 XFS_INACTIVE_LOG_COUNT);
1033 xfs_ilock(ip, XFS_ILOCK_EXCL);
1034 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1035 xfs_trans_ihold(tp, ip);
1036 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1038 ASSERT(ip->i_d.di_anextents == 0);
1044 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1045 xfs_trans_cancel(tp, 0);
1048 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1056 xfs_mount_t *mp = ip->i_mount;
1059 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1062 /* If this is a read-only mount, don't do this (would generate I/O) */
1063 if (mp->m_flags & XFS_MOUNT_RDONLY)
1066 if (!XFS_FORCED_SHUTDOWN(mp)) {
1070 * If we are using filestreams, and we have an unlinked
1071 * file that we are processing the last close on, then nothing
1072 * will be able to reopen and write to this file. Purge this
1073 * inode from the filestreams cache so that it doesn't delay
1074 * teardown of the inode.
1076 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1077 xfs_filestream_deassociate(ip);
1080 * If we previously truncated this file and removed old data
1081 * in the process, we want to initiate "early" writeout on
1082 * the last close. This is an attempt to combat the notorious
1083 * NULL files problem which is particularly noticable from a
1084 * truncate down, buffered (re-)write (delalloc), followed by
1085 * a crash. What we are effectively doing here is
1086 * significantly reducing the time window where we'd otherwise
1087 * be exposed to that problem.
1089 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1090 if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
1091 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
1094 if (ip->i_d.di_nlink != 0) {
1095 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1096 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
1097 ip->i_delayed_blks > 0)) &&
1098 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1099 (!(ip->i_d.di_flags &
1100 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1103 * If we can't get the iolock just skip truncating
1104 * the blocks past EOF because we could deadlock
1105 * with the mmap_sem otherwise. We'll get another
1106 * chance to drop them once the last reference to
1107 * the inode is dropped, so we'll never leak blocks
1110 error = xfs_free_eofblocks(mp, ip,
1111 XFS_FREE_EOF_TRYLOCK);
1123 * This is called when the vnode reference count for the vnode
1124 * goes to zero. If the file has been unlinked, then it must
1125 * now be truncated. Also, we clear all of the read-ahead state
1126 * kept for the inode here since the file is now closed.
1132 xfs_bmap_free_t free_list;
1133 xfs_fsblock_t first_block;
1140 xfs_itrace_entry(ip);
1143 * If the inode is already free, then there can be nothing
1146 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
1147 ASSERT(ip->i_df.if_real_bytes == 0);
1148 ASSERT(ip->i_df.if_broot_bytes == 0);
1149 return VN_INACTIVE_CACHE;
1153 * Only do a truncate if it's a regular file with
1154 * some actual space in it. It's OK to look at the
1155 * inode's fields without the lock because we're the
1156 * only one with a reference to the inode.
1158 truncate = ((ip->i_d.di_nlink == 0) &&
1159 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1160 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1161 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1165 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1166 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
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 (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 (ip->i_delayed_blks != 0)))) {
1182 error = xfs_free_eofblocks(mp, ip, 0);
1184 return VN_INACTIVE_CACHE;
1189 ASSERT(ip->i_d.di_nlink == 0);
1191 error = xfs_qm_dqattach(ip, 0);
1193 return VN_INACTIVE_CACHE;
1195 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1198 * Do the xfs_itruncate_start() call before
1199 * reserving any log space because itruncate_start
1200 * will call into the buffer cache and we can't
1201 * do that within a transaction.
1203 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1205 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1207 xfs_trans_cancel(tp, 0);
1208 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1209 return VN_INACTIVE_CACHE;
1212 error = xfs_trans_reserve(tp, 0,
1213 XFS_ITRUNCATE_LOG_RES(mp),
1214 0, XFS_TRANS_PERM_LOG_RES,
1215 XFS_ITRUNCATE_LOG_COUNT);
1217 /* Don't call itruncate_cleanup */
1218 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1219 xfs_trans_cancel(tp, 0);
1220 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1221 return VN_INACTIVE_CACHE;
1224 xfs_ilock(ip, XFS_ILOCK_EXCL);
1225 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1226 xfs_trans_ihold(tp, ip);
1229 * normally, we have to run xfs_itruncate_finish sync.
1230 * But if filesystem is wsync and we're in the inactive
1231 * path, then we know that nlink == 0, and that the
1232 * xaction that made nlink == 0 is permanently committed
1233 * since xfs_remove runs as a synchronous transaction.
1235 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1236 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1239 xfs_trans_cancel(tp,
1240 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1241 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1242 return VN_INACTIVE_CACHE;
1244 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1247 * If we get an error while cleaning up a
1248 * symlink we bail out.
1250 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1251 xfs_inactive_symlink_rmt(ip, &tp) :
1252 xfs_inactive_symlink_local(ip, &tp);
1256 return VN_INACTIVE_CACHE;
1259 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1260 xfs_trans_ihold(tp, ip);
1262 error = xfs_trans_reserve(tp, 0,
1263 XFS_IFREE_LOG_RES(mp),
1264 0, XFS_TRANS_PERM_LOG_RES,
1265 XFS_INACTIVE_LOG_COUNT);
1267 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1268 xfs_trans_cancel(tp, 0);
1269 return VN_INACTIVE_CACHE;
1272 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1273 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1274 xfs_trans_ihold(tp, ip);
1278 * If there are attributes associated with the file
1279 * then blow them away now. The code calls a routine
1280 * that recursively deconstructs the attribute fork.
1281 * We need to just commit the current transaction
1282 * because we can't use it for xfs_attr_inactive().
1284 if (ip->i_d.di_anextents > 0) {
1285 error = xfs_inactive_attrs(ip, &tp);
1287 * If we got an error, the transaction is already
1288 * cancelled, and the inode is unlocked. Just get out.
1291 return VN_INACTIVE_CACHE;
1292 } else if (ip->i_afp) {
1293 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1299 xfs_bmap_init(&free_list, &first_block);
1300 error = xfs_ifree(tp, ip, &free_list);
1303 * If we fail to free the inode, shut down. The cancel
1304 * might do that, we need to make sure. Otherwise the
1305 * inode might be lost for a long time or forever.
1307 if (!XFS_FORCED_SHUTDOWN(mp)) {
1309 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1310 error, mp->m_fsname);
1311 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1313 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1316 * Credit the quota account(s). The inode is gone.
1318 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1321 * Just ignore errors at this point. There is nothing we can
1322 * do except to try to keep going. Make sure it's not a silent
1325 error = xfs_bmap_finish(&tp, &free_list, &committed);
1327 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1328 "xfs_bmap_finish() returned error %d", error);
1329 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1331 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1332 "xfs_trans_commit() returned error %d", error);
1336 * Release the dquots held by inode, if any.
1338 xfs_qm_dqdetach(ip);
1339 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1342 return VN_INACTIVE_CACHE;
1346 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1347 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1348 * ci_name->name will point to a the actual name (caller must free) or
1349 * will be set to NULL if an exact match is found.
1354 struct xfs_name *name,
1356 struct xfs_name *ci_name)
1362 xfs_itrace_entry(dp);
1364 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1365 return XFS_ERROR(EIO);
1367 lock_mode = xfs_ilock_map_shared(dp);
1368 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
1369 xfs_iunlock_map_shared(dp, lock_mode);
1374 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
1378 xfs_itrace_ref(*ipp);
1383 kmem_free(ci_name->name);
1392 struct xfs_name *name,
1398 int is_dir = S_ISDIR(mode);
1399 struct xfs_mount *mp = dp->i_mount;
1400 struct xfs_inode *ip = NULL;
1401 struct xfs_trans *tp = NULL;
1403 xfs_bmap_free_t free_list;
1404 xfs_fsblock_t first_block;
1405 boolean_t unlock_dp_on_error = B_FALSE;
1409 struct xfs_dquot *udqp = NULL;
1410 struct xfs_dquot *gdqp = NULL;
1415 xfs_itrace_entry(dp);
1417 if (XFS_FORCED_SHUTDOWN(mp))
1418 return XFS_ERROR(EIO);
1420 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
1421 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1422 dp, DM_RIGHT_NULL, NULL,
1423 DM_RIGHT_NULL, name->name, NULL,
1430 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1431 prid = dp->i_d.di_projid;
1436 * Make sure that we have allocated dquot(s) on disk.
1438 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1439 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1445 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1446 log_res = XFS_MKDIR_LOG_RES(mp);
1447 log_count = XFS_MKDIR_LOG_COUNT;
1448 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1450 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1451 log_res = XFS_CREATE_LOG_RES(mp);
1452 log_count = XFS_CREATE_LOG_COUNT;
1453 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1456 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1459 * Initially assume that the file does not exist and
1460 * reserve the resources for that case. If that is not
1461 * the case we'll drop the one we have and get a more
1462 * appropriate transaction later.
1464 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1465 XFS_TRANS_PERM_LOG_RES, log_count);
1466 if (error == ENOSPC) {
1467 /* flush outstanding delalloc blocks and retry */
1468 xfs_flush_inodes(dp);
1469 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1470 XFS_TRANS_PERM_LOG_RES, log_count);
1472 if (error == ENOSPC) {
1473 /* No space at all so try a "no-allocation" reservation */
1475 error = xfs_trans_reserve(tp, 0, log_res, 0,
1476 XFS_TRANS_PERM_LOG_RES, log_count);
1480 goto out_trans_cancel;
1483 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1484 unlock_dp_on_error = B_TRUE;
1487 * Check for directory link count overflow.
1489 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
1490 error = XFS_ERROR(EMLINK);
1491 goto out_trans_cancel;
1494 xfs_bmap_init(&free_list, &first_block);
1497 * Reserve disk quota and the inode.
1499 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
1501 goto out_trans_cancel;
1503 error = xfs_dir_canenter(tp, dp, name, resblks);
1505 goto out_trans_cancel;
1508 * A newly created regular or special file just has one directory
1509 * entry pointing to them, but a directory also the "." entry
1510 * pointing to itself.
1512 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, credp,
1513 prid, resblks > 0, &ip, &committed);
1515 if (error == ENOSPC)
1516 goto out_trans_cancel;
1517 goto out_trans_abort;
1521 * At this point, we've gotten a newly allocated inode.
1522 * It is locked (and joined to the transaction).
1525 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1528 * Now we join the directory inode to the transaction. We do not do it
1529 * earlier because xfs_dir_ialloc might commit the previous transaction
1530 * (and release all the locks). An error from here on will result in
1531 * the transaction cancel unlocking dp so don't do it explicitly in the
1535 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1536 unlock_dp_on_error = B_FALSE;
1538 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1539 &first_block, &free_list, resblks ?
1540 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1542 ASSERT(error != ENOSPC);
1543 goto out_trans_abort;
1545 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1546 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1549 error = xfs_dir_init(tp, ip, dp);
1551 goto out_bmap_cancel;
1553 error = xfs_bumplink(tp, dp);
1555 goto out_bmap_cancel;
1559 * If this is a synchronous mount, make sure that the
1560 * create transaction goes to disk before returning to
1563 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1564 xfs_trans_set_sync(tp);
1567 * Attach the dquot(s) to the inodes and modify them incore.
1568 * These ids of the inode couldn't have changed since the new
1569 * inode has been locked ever since it was created.
1571 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
1574 * xfs_trans_commit normally decrements the vnode ref count
1575 * when it unlocks the inode. Since we want to return the
1576 * vnode to the caller, we bump the vnode ref count now.
1580 error = xfs_bmap_finish(&tp, &free_list, &committed);
1582 goto out_abort_rele;
1584 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1590 xfs_qm_dqrele(udqp);
1591 xfs_qm_dqrele(gdqp);
1595 /* Fallthrough to std_return with error = 0 */
1597 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
1598 XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, dp, DM_RIGHT_NULL,
1599 ip, DM_RIGHT_NULL, name->name, NULL, mode,
1606 xfs_bmap_cancel(&free_list);
1608 cancel_flags |= XFS_TRANS_ABORT;
1610 xfs_trans_cancel(tp, cancel_flags);
1612 xfs_qm_dqrele(udqp);
1613 xfs_qm_dqrele(gdqp);
1615 if (unlock_dp_on_error)
1616 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1622 * Wait until after the current transaction is aborted to
1623 * release the inode. This prevents recursive transactions
1624 * and deadlocks from xfs_inactive.
1626 xfs_bmap_cancel(&free_list);
1627 cancel_flags |= XFS_TRANS_ABORT;
1628 xfs_trans_cancel(tp, cancel_flags);
1630 unlock_dp_on_error = B_FALSE;
1636 int xfs_small_retries;
1637 int xfs_middle_retries;
1638 int xfs_lots_retries;
1639 int xfs_lock_delays;
1643 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1647 xfs_lock_inumorder(int lock_mode, int subclass)
1649 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1650 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1651 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1652 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1658 * The following routine will lock n inodes in exclusive mode.
1659 * We assume the caller calls us with the inodes in i_ino order.
1661 * We need to detect deadlock where an inode that we lock
1662 * is in the AIL and we start waiting for another inode that is locked
1663 * by a thread in a long running transaction (such as truncate). This can
1664 * result in deadlock since the long running trans might need to wait
1665 * for the inode we just locked in order to push the tail and free space
1674 int attempts = 0, i, j, try_lock;
1677 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1683 for (; i < inodes; i++) {
1686 if (i && (ips[i] == ips[i-1])) /* Already locked */
1690 * If try_lock is not set yet, make sure all locked inodes
1691 * are not in the AIL.
1692 * If any are, set try_lock to be used later.
1696 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1697 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1698 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1705 * If any of the previous locks we have locked is in the AIL,
1706 * we must TRY to get the second and subsequent locks. If
1707 * we can't get any, we must release all we have
1712 /* try_lock must be 0 if i is 0. */
1714 * try_lock means we have an inode locked
1715 * that is in the AIL.
1718 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1722 * Unlock all previous guys and try again.
1723 * xfs_iunlock will try to push the tail
1724 * if the inode is in the AIL.
1727 for(j = i - 1; j >= 0; j--) {
1730 * Check to see if we've already
1731 * unlocked this one.
1732 * Not the first one going back,
1733 * and the inode ptr is the same.
1735 if ((j != (i - 1)) && ips[j] ==
1739 xfs_iunlock(ips[j], lock_mode);
1742 if ((attempts % 5) == 0) {
1743 delay(1); /* Don't just spin the CPU */
1753 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1759 if (attempts < 5) xfs_small_retries++;
1760 else if (attempts < 100) xfs_middle_retries++;
1761 else xfs_lots_retries++;
1769 * xfs_lock_two_inodes() can only be used to lock one type of lock
1770 * at a time - the iolock or the ilock, but not both at once. If
1771 * we lock both at once, lockdep will report false positives saying
1772 * we have violated locking orders.
1775 xfs_lock_two_inodes(
1784 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1785 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1786 ASSERT(ip0->i_ino != ip1->i_ino);
1788 if (ip0->i_ino > ip1->i_ino) {
1795 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1798 * If the first lock we have locked is in the AIL, we must TRY to get
1799 * the second lock. If we can't get it, we must release the first one
1802 lp = (xfs_log_item_t *)ip0->i_itemp;
1803 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1804 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1805 xfs_iunlock(ip0, lock_mode);
1806 if ((++attempts % 5) == 0)
1807 delay(1); /* Don't just spin the CPU */
1811 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1818 struct xfs_name *name,
1821 xfs_mount_t *mp = dp->i_mount;
1822 xfs_trans_t *tp = NULL;
1823 int is_dir = S_ISDIR(ip->i_d.di_mode);
1825 xfs_bmap_free_t free_list;
1826 xfs_fsblock_t first_block;
1833 xfs_itrace_entry(dp);
1834 xfs_itrace_entry(ip);
1836 if (XFS_FORCED_SHUTDOWN(mp))
1837 return XFS_ERROR(EIO);
1839 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
1840 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL,
1841 NULL, DM_RIGHT_NULL, name->name, NULL,
1842 ip->i_d.di_mode, 0, 0);
1847 error = xfs_qm_dqattach(dp, 0);
1851 error = xfs_qm_dqattach(ip, 0);
1856 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1857 log_count = XFS_DEFAULT_LOG_COUNT;
1859 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1860 log_count = XFS_REMOVE_LOG_COUNT;
1862 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1865 * We try to get the real space reservation first,
1866 * allowing for directory btree deletion(s) implying
1867 * possible bmap insert(s). If we can't get the space
1868 * reservation then we use 0 instead, and avoid the bmap
1869 * btree insert(s) in the directory code by, if the bmap
1870 * insert tries to happen, instead trimming the LAST
1871 * block from the directory.
1873 resblks = XFS_REMOVE_SPACE_RES(mp);
1874 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1875 XFS_TRANS_PERM_LOG_RES, log_count);
1876 if (error == ENOSPC) {
1878 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1879 XFS_TRANS_PERM_LOG_RES, log_count);
1882 ASSERT(error != ENOSPC);
1884 goto out_trans_cancel;
1887 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1890 * At this point, we've gotten both the directory and the entry
1894 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1897 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1900 * If we're removing a directory perform some additional validation.
1903 ASSERT(ip->i_d.di_nlink >= 2);
1904 if (ip->i_d.di_nlink != 2) {
1905 error = XFS_ERROR(ENOTEMPTY);
1906 goto out_trans_cancel;
1908 if (!xfs_dir_isempty(ip)) {
1909 error = XFS_ERROR(ENOTEMPTY);
1910 goto out_trans_cancel;
1914 xfs_bmap_init(&free_list, &first_block);
1915 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1916 &first_block, &free_list, resblks);
1918 ASSERT(error != ENOENT);
1919 goto out_bmap_cancel;
1921 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1925 * Drop the link from ip's "..".
1927 error = xfs_droplink(tp, dp);
1929 goto out_bmap_cancel;
1932 * Drop the "." link from ip to self.
1934 error = xfs_droplink(tp, ip);
1936 goto out_bmap_cancel;
1939 * When removing a non-directory we need to log the parent
1940 * inode here. For a directory this is done implicitly
1941 * by the xfs_droplink call for the ".." entry.
1943 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1947 * Drop the link from dp to ip.
1949 error = xfs_droplink(tp, ip);
1951 goto out_bmap_cancel;
1954 * Determine if this is the last link while
1955 * we are in the transaction.
1957 link_zero = (ip->i_d.di_nlink == 0);
1960 * If this is a synchronous mount, make sure that the
1961 * remove transaction goes to disk before returning to
1964 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1965 xfs_trans_set_sync(tp);
1967 error = xfs_bmap_finish(&tp, &free_list, &committed);
1969 goto out_bmap_cancel;
1971 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1976 * If we are using filestreams, kill the stream association.
1977 * If the file is still open it may get a new one but that
1978 * will get killed on last close in xfs_close() so we don't
1979 * have to worry about that.
1981 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1982 xfs_filestream_deassociate(ip);
1984 xfs_itrace_exit(ip);
1985 xfs_itrace_exit(dp);
1988 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
1989 XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL,
1990 NULL, DM_RIGHT_NULL, name->name, NULL,
1991 ip->i_d.di_mode, error, 0);
1997 xfs_bmap_cancel(&free_list);
1998 cancel_flags |= XFS_TRANS_ABORT;
2000 xfs_trans_cancel(tp, cancel_flags);
2008 struct xfs_name *target_name)
2010 xfs_mount_t *mp = tdp->i_mount;
2013 xfs_bmap_free_t free_list;
2014 xfs_fsblock_t first_block;
2019 xfs_itrace_entry(tdp);
2020 xfs_itrace_entry(sip);
2022 ASSERT(!S_ISDIR(sip->i_d.di_mode));
2024 if (XFS_FORCED_SHUTDOWN(mp))
2025 return XFS_ERROR(EIO);
2027 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
2028 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2031 target_name->name, NULL, 0, 0, 0);
2036 /* Return through std_return after this point. */
2038 error = xfs_qm_dqattach(sip, 0);
2042 error = xfs_qm_dqattach(tdp, 0);
2046 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2047 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2048 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
2049 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2050 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2051 if (error == ENOSPC) {
2053 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2054 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2061 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
2064 * Increment vnode ref counts since xfs_trans_commit &
2065 * xfs_trans_cancel will both unlock the inodes and
2066 * decrement the associated ref counts.
2070 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2071 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2074 * If the source has too many links, we can't make any more to it.
2076 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2077 error = XFS_ERROR(EMLINK);
2082 * If we are using project inheritance, we only allow hard link
2083 * creation in our tree when the project IDs are the same; else
2084 * the tree quota mechanism could be circumvented.
2086 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2087 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2088 error = XFS_ERROR(EXDEV);
2092 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
2096 xfs_bmap_init(&free_list, &first_block);
2098 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
2099 &first_block, &free_list, resblks);
2102 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2103 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2105 error = xfs_bumplink(tp, sip);
2110 * If this is a synchronous mount, make sure that the
2111 * link transaction goes to disk before returning to
2114 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2115 xfs_trans_set_sync(tp);
2118 error = xfs_bmap_finish (&tp, &free_list, &committed);
2120 xfs_bmap_cancel(&free_list);
2124 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2128 /* Fall through to std_return with error = 0. */
2130 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
2131 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2134 target_name->name, NULL, 0, error, 0);
2139 cancel_flags |= XFS_TRANS_ABORT;
2143 xfs_trans_cancel(tp, cancel_flags);
2150 struct xfs_name *link_name,
2151 const char *target_path,
2156 xfs_mount_t *mp = dp->i_mount;
2161 xfs_bmap_free_t free_list;
2162 xfs_fsblock_t first_block;
2163 boolean_t unlock_dp_on_error = B_FALSE;
2166 xfs_fileoff_t first_fsb;
2167 xfs_filblks_t fs_blocks;
2169 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
2171 const char *cur_chunk;
2176 struct xfs_dquot *udqp, *gdqp;
2184 xfs_itrace_entry(dp);
2186 if (XFS_FORCED_SHUTDOWN(mp))
2187 return XFS_ERROR(EIO);
2190 * Check component lengths of the target path name.
2192 pathlen = strlen(target_path);
2193 if (pathlen >= MAXPATHLEN) /* total string too long */
2194 return XFS_ERROR(ENAMETOOLONG);
2196 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
2197 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
2198 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2199 link_name->name, target_path, 0, 0, 0);
2204 /* Return through std_return after this point. */
2207 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2208 prid = dp->i_d.di_projid;
2210 prid = (xfs_prid_t)dfltprid;
2213 * Make sure that we have allocated dquot(s) on disk.
2215 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
2216 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2220 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
2221 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2223 * The symlink will fit into the inode data fork?
2224 * There can't be any attributes so we get the whole variable part.
2226 if (pathlen <= XFS_LITINO(mp))
2229 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
2230 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
2231 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2232 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2233 if (error == ENOSPC && fs_blocks == 0) {
2235 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2236 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2243 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2244 unlock_dp_on_error = B_TRUE;
2247 * Check whether the directory allows new symlinks or not.
2249 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2250 error = XFS_ERROR(EPERM);
2255 * Reserve disk quota : blocks and inode.
2257 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
2262 * Check for ability to enter directory entry, if no space reserved.
2264 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2268 * Initialize the bmap freelist prior to calling either
2269 * bmapi or the directory create code.
2271 xfs_bmap_init(&free_list, &first_block);
2274 * Allocate an inode for the symlink.
2276 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
2277 1, 0, credp, prid, resblks > 0, &ip, NULL);
2279 if (error == ENOSPC)
2286 * An error after we've joined dp to the transaction will result in the
2287 * transaction cancel unlocking dp so don't do it explicitly in the
2291 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2292 unlock_dp_on_error = B_FALSE;
2295 * Also attach the dquot(s) to it, if applicable.
2297 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
2300 resblks -= XFS_IALLOC_SPACE_RES(mp);
2302 * If the symlink will fit into the inode, write it inline.
2304 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2305 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2306 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2307 ip->i_d.di_size = pathlen;
2310 * The inode was initially created in extent format.
2312 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2313 ip->i_df.if_flags |= XFS_IFINLINE;
2315 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2316 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2320 nmaps = SYMLINK_MAPS;
2322 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2323 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2324 &first_block, resblks, mval, &nmaps,
2331 resblks -= fs_blocks;
2332 ip->i_d.di_size = pathlen;
2333 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2335 cur_chunk = target_path;
2336 for (n = 0; n < nmaps; n++) {
2337 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2338 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2339 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2340 BTOBB(byte_cnt), 0);
2341 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2342 if (pathlen < byte_cnt) {
2345 pathlen -= byte_cnt;
2347 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2348 cur_chunk += byte_cnt;
2350 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2355 * Create the directory entry for the symlink.
2357 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2358 &first_block, &free_list, resblks);
2361 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2362 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2365 * If this is a synchronous mount, make sure that the
2366 * symlink transaction goes to disk before returning to
2369 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2370 xfs_trans_set_sync(tp);
2374 * xfs_trans_commit normally decrements the vnode ref count
2375 * when it unlocks the inode. Since we want to return the
2376 * vnode to the caller, we bump the vnode ref count now.
2380 error = xfs_bmap_finish(&tp, &free_list, &committed);
2384 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2385 xfs_qm_dqrele(udqp);
2386 xfs_qm_dqrele(gdqp);
2388 /* Fall through to std_return with error = 0 or errno from
2389 * xfs_trans_commit */
2391 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
2392 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
2395 DM_RIGHT_NULL, link_name->name,
2396 target_path, 0, error, 0);
2406 xfs_bmap_cancel(&free_list);
2407 cancel_flags |= XFS_TRANS_ABORT;
2409 xfs_trans_cancel(tp, cancel_flags);
2410 xfs_qm_dqrele(udqp);
2411 xfs_qm_dqrele(gdqp);
2413 if (unlock_dp_on_error)
2414 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2425 xfs_mount_t *mp = ip->i_mount;
2429 if (!capable(CAP_SYS_ADMIN))
2430 return XFS_ERROR(EPERM);
2432 if (XFS_FORCED_SHUTDOWN(mp))
2433 return XFS_ERROR(EIO);
2435 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2436 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2438 xfs_trans_cancel(tp, 0);
2441 xfs_ilock(ip, XFS_ILOCK_EXCL);
2442 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2444 ip->i_d.di_dmevmask = evmask;
2445 ip->i_d.di_dmstate = state;
2447 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2449 error = xfs_trans_commit(tp, 0);
2455 * xfs_alloc_file_space()
2456 * This routine allocates disk space for the given file.
2458 * If alloc_type == 0, this request is for an ALLOCSP type
2459 * request which will change the file size. In this case, no
2460 * DMAPI event will be generated by the call. A TRUNCATE event
2461 * will be generated later by xfs_setattr.
2463 * If alloc_type != 0, this request is for a RESVSP type
2464 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2465 * lower block boundary byte address is less than the file's
2474 xfs_alloc_file_space(
2481 xfs_mount_t *mp = ip->i_mount;
2483 xfs_filblks_t allocated_fsb;
2484 xfs_filblks_t allocatesize_fsb;
2485 xfs_extlen_t extsz, temp;
2486 xfs_fileoff_t startoffset_fsb;
2487 xfs_fsblock_t firstfsb;
2493 xfs_bmbt_irec_t imaps[1], *imapp;
2494 xfs_bmap_free_t free_list;
2495 uint qblocks, resblks, resrtextents;
2499 xfs_itrace_entry(ip);
2501 if (XFS_FORCED_SHUTDOWN(mp))
2502 return XFS_ERROR(EIO);
2504 error = xfs_qm_dqattach(ip, 0);
2509 return XFS_ERROR(EINVAL);
2511 rt = XFS_IS_REALTIME_INODE(ip);
2512 extsz = xfs_get_extsz_hint(ip);
2517 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
2518 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
2519 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
2521 /* Generate a DMAPI event if needed. */
2522 if (alloc_type != 0 && offset < ip->i_size &&
2523 (attr_flags & XFS_ATTR_DMI) == 0 &&
2524 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
2525 xfs_off_t end_dmi_offset;
2527 end_dmi_offset = offset+len;
2528 if (end_dmi_offset > ip->i_size)
2529 end_dmi_offset = ip->i_size;
2530 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
2531 end_dmi_offset - offset, 0, NULL);
2537 * Allocate file space until done or until there is an error
2540 while (allocatesize_fsb && !error) {
2544 * Determine space reservations for data/realtime.
2546 if (unlikely(extsz)) {
2547 s = startoffset_fsb;
2550 e = startoffset_fsb + allocatesize_fsb;
2551 if ((temp = do_mod(startoffset_fsb, extsz)))
2553 if ((temp = do_mod(e, extsz)))
2557 e = allocatesize_fsb;
2561 resrtextents = qblocks = (uint)(e - s);
2562 resrtextents /= mp->m_sb.sb_rextsize;
2563 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2564 quota_flag = XFS_QMOPT_RES_RTBLKS;
2567 resblks = qblocks = \
2568 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
2569 quota_flag = XFS_QMOPT_RES_REGBLKS;
2573 * Allocate and setup the transaction.
2575 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2576 error = xfs_trans_reserve(tp, resblks,
2577 XFS_WRITE_LOG_RES(mp), resrtextents,
2578 XFS_TRANS_PERM_LOG_RES,
2579 XFS_WRITE_LOG_COUNT);
2581 * Check for running out of space
2585 * Free the transaction structure.
2587 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2588 xfs_trans_cancel(tp, 0);
2591 xfs_ilock(ip, XFS_ILOCK_EXCL);
2592 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
2597 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2598 xfs_trans_ihold(tp, ip);
2601 * Issue the xfs_bmapi() call to allocate the blocks
2603 xfs_bmap_init(&free_list, &firstfsb);
2604 error = xfs_bmapi(tp, ip, startoffset_fsb,
2605 allocatesize_fsb, bmapi_flag,
2606 &firstfsb, 0, imapp, &nimaps,
2613 * Complete the transaction
2615 error = xfs_bmap_finish(&tp, &free_list, &committed);
2620 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2621 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2626 allocated_fsb = imapp->br_blockcount;
2629 error = XFS_ERROR(ENOSPC);
2633 startoffset_fsb += allocated_fsb;
2634 allocatesize_fsb -= allocated_fsb;
2637 if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 &&
2638 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
2639 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
2642 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
2644 goto retry; /* Maybe DMAPI app. has made space */
2645 /* else fall through with error from XFS_SEND_DATA */
2650 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
2651 xfs_bmap_cancel(&free_list);
2652 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
2654 error1: /* Just cancel transaction */
2655 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2656 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2657 goto dmapi_enospc_check;
2661 * Zero file bytes between startoff and endoff inclusive.
2662 * The iolock is held exclusive and no blocks are buffered.
2664 * This function is used by xfs_free_file_space() to zero
2665 * partial blocks when the range to free is not block aligned.
2666 * When unreserving space with boundaries that are not block
2667 * aligned we round up the start and round down the end
2668 * boundaries and then use this function to zero the parts of
2669 * the blocks that got dropped during the rounding.
2672 xfs_zero_remaining_bytes(
2677 xfs_bmbt_irec_t imap;
2678 xfs_fileoff_t offset_fsb;
2679 xfs_off_t lastoffset;
2682 xfs_mount_t *mp = ip->i_mount;
2687 * Avoid doing I/O beyond eof - it's not necessary
2688 * since nothing can read beyond eof. The space will
2689 * be zeroed when the file is extended anyway.
2691 if (startoff >= ip->i_size)
2694 if (endoff > ip->i_size)
2695 endoff = ip->i_size;
2697 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
2698 XFS_IS_REALTIME_INODE(ip) ?
2699 mp->m_rtdev_targp : mp->m_ddev_targp);
2701 return XFS_ERROR(ENOMEM);
2703 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
2704 offset_fsb = XFS_B_TO_FSBT(mp, offset);
2706 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
2707 NULL, 0, &imap, &nimap, NULL, NULL);
2708 if (error || nimap < 1)
2710 ASSERT(imap.br_blockcount >= 1);
2711 ASSERT(imap.br_startoff == offset_fsb);
2712 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
2713 if (lastoffset > endoff)
2714 lastoffset = endoff;
2715 if (imap.br_startblock == HOLESTARTBLOCK)
2717 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2718 if (imap.br_state == XFS_EXT_UNWRITTEN)
2721 XFS_BUF_UNWRITE(bp);
2723 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
2725 error = xfs_iowait(bp);
2727 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2728 mp, bp, XFS_BUF_ADDR(bp));
2731 memset(XFS_BUF_PTR(bp) +
2732 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2733 0, lastoffset - offset + 1);
2738 error = xfs_iowait(bp);
2740 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2741 mp, bp, XFS_BUF_ADDR(bp));
2750 * xfs_free_file_space()
2751 * This routine frees disk space for the given file.
2753 * This routine is only called by xfs_change_file_space
2754 * for an UNRESVSP type call.
2762 xfs_free_file_space(
2770 xfs_off_t end_dmi_offset;
2771 xfs_fileoff_t endoffset_fsb;
2773 xfs_fsblock_t firstfsb;
2774 xfs_bmap_free_t free_list;
2775 xfs_bmbt_irec_t imap;
2783 xfs_fileoff_t startoffset_fsb;
2785 int need_iolock = 1;
2789 xfs_itrace_entry(ip);
2791 error = xfs_qm_dqattach(ip, 0);
2796 if (len <= 0) /* if nothing being freed */
2798 rt = XFS_IS_REALTIME_INODE(ip);
2799 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
2800 end_dmi_offset = offset + len;
2801 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
2803 if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 &&
2804 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
2805 if (end_dmi_offset > ip->i_size)
2806 end_dmi_offset = ip->i_size;
2807 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
2808 offset, end_dmi_offset - offset,
2809 AT_DELAY_FLAG(attr_flags), NULL);
2814 if (attr_flags & XFS_ATTR_NOLOCK)
2817 xfs_ilock(ip, XFS_IOLOCK_EXCL);
2818 /* wait for the completion of any pending DIOs */
2822 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
2823 ioffset = offset & ~(rounding - 1);
2825 if (VN_CACHED(VFS_I(ip)) != 0) {
2826 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
2827 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
2829 goto out_unlock_iolock;
2833 * Need to zero the stuff we're not freeing, on disk.
2834 * If it's a realtime file & can't use unwritten extents then we
2835 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2836 * will take care of it for us.
2838 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
2840 error = xfs_bmapi(NULL, ip, startoffset_fsb,
2841 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
2843 goto out_unlock_iolock;
2844 ASSERT(nimap == 0 || nimap == 1);
2845 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2848 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2849 block = imap.br_startblock;
2850 mod = do_div(block, mp->m_sb.sb_rextsize);
2852 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2855 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
2856 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
2858 goto out_unlock_iolock;
2859 ASSERT(nimap == 0 || nimap == 1);
2860 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2861 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2863 if (mod && (mod != mp->m_sb.sb_rextsize))
2864 endoffset_fsb -= mod;
2867 if ((done = (endoffset_fsb <= startoffset_fsb)))
2869 * One contiguous piece to clear
2871 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2874 * Some full blocks, possibly two pieces to clear
2876 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2877 error = xfs_zero_remaining_bytes(ip, offset,
2878 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2880 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2881 error = xfs_zero_remaining_bytes(ip,
2882 XFS_FSB_TO_B(mp, endoffset_fsb),
2887 * free file space until done or until there is an error
2889 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2890 while (!error && !done) {
2893 * allocate and setup the transaction. Allow this
2894 * transaction to dip into the reserve blocks to ensure
2895 * the freeing of the space succeeds at ENOSPC.
2897 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2898 tp->t_flags |= XFS_TRANS_RESERVE;
2899 error = xfs_trans_reserve(tp,
2901 XFS_WRITE_LOG_RES(mp),
2903 XFS_TRANS_PERM_LOG_RES,
2904 XFS_WRITE_LOG_COUNT);
2907 * check for running out of space
2911 * Free the transaction structure.
2913 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2914 xfs_trans_cancel(tp, 0);
2917 xfs_ilock(ip, XFS_ILOCK_EXCL);
2918 error = xfs_trans_reserve_quota(tp, mp,
2919 ip->i_udquot, ip->i_gdquot,
2920 resblks, 0, XFS_QMOPT_RES_REGBLKS);
2924 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2925 xfs_trans_ihold(tp, ip);
2928 * issue the bunmapi() call to free the blocks
2930 xfs_bmap_init(&free_list, &firstfsb);
2931 error = xfs_bunmapi(tp, ip, startoffset_fsb,
2932 endoffset_fsb - startoffset_fsb,
2933 0, 2, &firstfsb, &free_list, NULL, &done);
2939 * complete the transaction
2941 error = xfs_bmap_finish(&tp, &free_list, &committed);
2946 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2947 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2952 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2956 xfs_bmap_cancel(&free_list);
2958 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2959 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2965 * xfs_change_file_space()
2966 * This routine allocates or frees disk space for the given file.
2967 * The user specified parameters are checked for alignment and size
2976 xfs_change_file_space(
2983 xfs_mount_t *mp = ip->i_mount;
2988 xfs_off_t startoffset;
2993 xfs_itrace_entry(ip);
2995 if (!S_ISREG(ip->i_d.di_mode))
2996 return XFS_ERROR(EINVAL);
2998 switch (bf->l_whence) {
2999 case 0: /*SEEK_SET*/
3001 case 1: /*SEEK_CUR*/
3002 bf->l_start += offset;
3004 case 2: /*SEEK_END*/
3005 bf->l_start += ip->i_size;
3008 return XFS_ERROR(EINVAL);
3011 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
3013 if ( (bf->l_start < 0)
3014 || (bf->l_start > XFS_MAXIOFFSET(mp))
3015 || (bf->l_start + llen < 0)
3016 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
3017 return XFS_ERROR(EINVAL);
3021 startoffset = bf->l_start;
3025 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3027 * These calls do NOT zero the data space allocated to the file,
3028 * nor do they change the file size.
3030 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3032 * These calls cause the new file data to be zeroed and the file
3033 * size to be changed.
3035 setprealloc = clrprealloc = 0;
3038 case XFS_IOC_RESVSP:
3039 case XFS_IOC_RESVSP64:
3040 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
3047 case XFS_IOC_UNRESVSP:
3048 case XFS_IOC_UNRESVSP64:
3049 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
3054 case XFS_IOC_ALLOCSP:
3055 case XFS_IOC_ALLOCSP64:
3056 case XFS_IOC_FREESP:
3057 case XFS_IOC_FREESP64:
3058 if (startoffset > fsize) {
3059 error = xfs_alloc_file_space(ip, fsize,
3060 startoffset - fsize, 0, attr_flags);
3065 iattr.ia_valid = ATTR_SIZE;
3066 iattr.ia_size = startoffset;
3068 error = xfs_setattr(ip, &iattr, attr_flags);
3078 return XFS_ERROR(EINVAL);
3082 * update the inode timestamp, mode, and prealloc flag bits
3084 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
3086 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
3089 xfs_trans_cancel(tp, 0);
3093 xfs_ilock(ip, XFS_ILOCK_EXCL);
3095 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3096 xfs_trans_ihold(tp, ip);
3098 if ((attr_flags & XFS_ATTR_DMI) == 0) {
3099 ip->i_d.di_mode &= ~S_ISUID;
3102 * Note that we don't have to worry about mandatory
3103 * file locking being disabled here because we only
3104 * clear the S_ISGID bit if the Group execute bit is
3105 * on, but if it was on then mandatory locking wouldn't
3106 * have been enabled.
3108 if (ip->i_d.di_mode & S_IXGRP)
3109 ip->i_d.di_mode &= ~S_ISGID;
3111 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3114 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
3115 else if (clrprealloc)
3116 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
3118 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3119 xfs_trans_set_sync(tp);
3121 error = xfs_trans_commit(tp, 0);
3123 xfs_iunlock(ip, XFS_ILOCK_EXCL);