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_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_itable.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_alloc.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_utils.h"
46 #include "xfs_rtalloc.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_log_priv.h"
49 #include "xfs_filestream.h"
50 #include "xfs_vnodeops.h"
51 #include "xfs_trace.h"
59 xfs_mount_t *mp = ip->i_mount;
60 struct inode *inode = VFS_I(ip);
61 int mask = iattr->ia_valid;
68 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
73 if (mp->m_flags & XFS_MOUNT_RDONLY)
74 return XFS_ERROR(EROFS);
76 if (XFS_FORCED_SHUTDOWN(mp))
77 return XFS_ERROR(EIO);
79 code = -inode_change_ok(inode, iattr);
83 olddquot1 = olddquot2 = NULL;
87 * If disk quotas is on, we make sure that the dquots do exist on disk,
88 * before we start any other transactions. Trying to do this later
89 * is messy. We don't care to take a readlock to look at the ids
90 * in inode here, because we can't hold it across the trans_reserve.
91 * If the IDs do change before we take the ilock, we're covered
92 * because the i_*dquot fields will get updated anyway.
94 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
97 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
99 qflags |= XFS_QMOPT_UQUOTA;
101 uid = ip->i_d.di_uid;
103 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
105 qflags |= XFS_QMOPT_GQUOTA;
107 gid = ip->i_d.di_gid;
111 * We take a reference when we initialize udqp and gdqp,
112 * so it is important that we never blindly double trip on
113 * the same variable. See xfs_create() for an example.
115 ASSERT(udqp == NULL);
116 ASSERT(gdqp == NULL);
117 code = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
118 qflags, &udqp, &gdqp);
124 * For the other attributes, we acquire the inode lock and
125 * first do an error checking pass.
128 lock_flags = XFS_ILOCK_EXCL;
129 if (flags & XFS_ATTR_NOLOCK)
131 if (!(mask & ATTR_SIZE)) {
132 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
134 code = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp),
142 lock_flags |= XFS_IOLOCK_EXCL;
145 xfs_ilock(ip, lock_flags);
148 * Change file ownership. Must be the owner or privileged.
150 if (mask & (ATTR_UID|ATTR_GID)) {
152 * These IDs could have changed since we last looked at them.
153 * But, we're assured that if the ownership did change
154 * while we didn't have the inode locked, inode's dquot(s)
155 * would have changed also.
157 iuid = ip->i_d.di_uid;
158 igid = ip->i_d.di_gid;
159 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
160 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
163 * Do a quota reservation only if uid/gid is actually
166 if (XFS_IS_QUOTA_RUNNING(mp) &&
167 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
168 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
170 code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
171 capable(CAP_FOWNER) ?
172 XFS_QMOPT_FORCE_RES : 0);
173 if (code) /* out of quota */
179 * Truncate file. Must have write permission and not be a directory.
181 if (mask & ATTR_SIZE) {
182 /* Short circuit the truncate case for zero length files */
183 if (iattr->ia_size == 0 &&
184 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
185 xfs_iunlock(ip, XFS_ILOCK_EXCL);
186 lock_flags &= ~XFS_ILOCK_EXCL;
187 if (mask & ATTR_CTIME)
188 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
193 if (S_ISDIR(ip->i_d.di_mode)) {
194 code = XFS_ERROR(EISDIR);
196 } else if (!S_ISREG(ip->i_d.di_mode)) {
197 code = XFS_ERROR(EINVAL);
202 * Make sure that the dquots are attached to the inode.
204 code = xfs_qm_dqattach_locked(ip, 0);
209 * Now we can make the changes. Before we join the inode
210 * to the transaction, if ATTR_SIZE is set then take care of
211 * the part of the truncation that must be done without the
212 * inode lock. This needs to be done before joining the inode
213 * to the transaction, because the inode cannot be unlocked
214 * once it is a part of the transaction.
216 if (iattr->ia_size > ip->i_size) {
218 * Do the first part of growing a file: zero any data
219 * in the last block that is beyond the old EOF. We
220 * need to do this before the inode is joined to the
221 * transaction to modify the i_size.
223 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
225 xfs_iunlock(ip, XFS_ILOCK_EXCL);
228 * We are going to log the inode size change in this
229 * transaction so any previous writes that are beyond the on
230 * disk EOF and the new EOF that have not been written out need
231 * to be written here. If we do not write the data out, we
232 * expose ourselves to the null files problem.
234 * Only flush from the on disk size to the smaller of the in
235 * memory file size or the new size as that's the range we
236 * really care about here and prevents waiting for other data
237 * not within the range we care about here.
240 ip->i_size != ip->i_d.di_size &&
241 iattr->ia_size > ip->i_d.di_size) {
242 code = xfs_flush_pages(ip,
243 ip->i_d.di_size, iattr->ia_size,
247 /* wait for all I/O to complete */
251 code = xfs_itruncate_data(ip, iattr->ia_size);
254 lock_flags &= ~XFS_ILOCK_EXCL;
255 ASSERT(lock_flags == XFS_IOLOCK_EXCL || !need_iolock);
258 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
259 if ((code = xfs_trans_reserve(tp, 0,
260 XFS_ITRUNCATE_LOG_RES(mp), 0,
261 XFS_TRANS_PERM_LOG_RES,
262 XFS_ITRUNCATE_LOG_COUNT))) {
263 xfs_trans_cancel(tp, 0);
265 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
268 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
269 xfs_ilock(ip, XFS_ILOCK_EXCL);
271 xfs_trans_ijoin(tp, ip, lock_flags);
272 xfs_trans_ihold(tp, ip);
275 * Only change the c/mtime if we are changing the size
276 * or we are explicitly asked to change it. This handles
277 * the semantic difference between truncate() and ftruncate()
278 * as implemented in the VFS.
280 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME
281 * is a special case where we need to update the times despite
282 * not having these flags set. For all other operations the
283 * VFS set these flags explicitly if it wants a timestamp
286 if (iattr->ia_size != ip->i_size &&
287 (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
288 iattr->ia_ctime = iattr->ia_mtime =
289 current_fs_time(inode->i_sb);
290 mask |= ATTR_CTIME | ATTR_MTIME;
293 if (iattr->ia_size > ip->i_size) {
294 ip->i_d.di_size = iattr->ia_size;
295 ip->i_size = iattr->ia_size;
296 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
297 } else if (iattr->ia_size <= ip->i_size ||
298 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
300 * signal a sync transaction unless
301 * we're truncating an already unlinked
302 * file on a wsync filesystem
304 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
306 ((ip->i_d.di_nlink != 0 ||
307 !(mp->m_flags & XFS_MOUNT_WSYNC))
312 * Truncated "down", so we're removing references
313 * to old data here - if we now delay flushing for
314 * a long time, we expose ourselves unduly to the
315 * notorious NULL files problem. So, we mark this
316 * vnode and flush it when the file is closed, and
317 * do not wait the usual (long) time for writeout.
319 xfs_iflags_set(ip, XFS_ITRUNCATED);
322 xfs_trans_ijoin(tp, ip, lock_flags);
323 xfs_trans_ihold(tp, ip);
327 * Change file ownership. Must be the owner or privileged.
329 if (mask & (ATTR_UID|ATTR_GID)) {
331 * CAP_FSETID overrides the following restrictions:
333 * The set-user-ID and set-group-ID bits of a file will be
334 * cleared upon successful return from chown()
336 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
337 !capable(CAP_FSETID)) {
338 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
342 * Change the ownerships and register quota modifications
343 * in the transaction.
346 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
347 ASSERT(mask & ATTR_UID);
349 olddquot1 = xfs_qm_vop_chown(tp, ip,
350 &ip->i_udquot, udqp);
352 ip->i_d.di_uid = uid;
356 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
357 ASSERT(!XFS_IS_PQUOTA_ON(mp));
358 ASSERT(mask & ATTR_GID);
360 olddquot2 = xfs_qm_vop_chown(tp, ip,
361 &ip->i_gdquot, gdqp);
363 ip->i_d.di_gid = gid;
369 * Change file access modes.
371 if (mask & ATTR_MODE) {
372 umode_t mode = iattr->ia_mode;
374 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
377 ip->i_d.di_mode &= S_IFMT;
378 ip->i_d.di_mode |= mode & ~S_IFMT;
380 inode->i_mode &= S_IFMT;
381 inode->i_mode |= mode & ~S_IFMT;
385 * Change file access or modified times.
387 if (mask & ATTR_ATIME) {
388 inode->i_atime = iattr->ia_atime;
389 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
390 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
391 ip->i_update_core = 1;
393 if (mask & ATTR_CTIME) {
394 inode->i_ctime = iattr->ia_ctime;
395 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
396 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
397 ip->i_update_core = 1;
399 if (mask & ATTR_MTIME) {
400 inode->i_mtime = iattr->ia_mtime;
401 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
402 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
403 ip->i_update_core = 1;
407 * And finally, log the inode core if any attribute in it
410 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE|
411 ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
412 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
414 XFS_STATS_INC(xs_ig_attrchg);
417 * If this is a synchronous mount, make sure that the
418 * transaction goes to disk before returning to the user.
419 * This is slightly sub-optimal in that truncates require
420 * two sync transactions instead of one for wsync filesystems.
421 * One for the truncate and one for the timestamps since we
422 * don't want to change the timestamps unless we're sure the
423 * truncate worked. Truncates are less than 1% of the laddis
424 * mix so this probably isn't worth the trouble to optimize.
427 if (mp->m_flags & XFS_MOUNT_WSYNC)
428 xfs_trans_set_sync(tp);
430 code = xfs_trans_commit(tp, commit_flags);
432 xfs_iunlock(ip, lock_flags);
435 * Release any dquot(s) the inode had kept before chown.
437 xfs_qm_dqrele(olddquot1);
438 xfs_qm_dqrele(olddquot2);
446 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
447 * update. We could avoid this with linked transactions
448 * and passing down the transaction pointer all the way
449 * to attr_set. No previous user of the generic
450 * Posix ACL code seems to care about this issue either.
452 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
453 code = -xfs_acl_chmod(inode);
455 return XFS_ERROR(code);
461 commit_flags |= XFS_TRANS_ABORT;
466 xfs_trans_cancel(tp, commit_flags);
468 if (lock_flags != 0) {
469 xfs_iunlock(ip, lock_flags);
475 * The maximum pathlen is 1024 bytes. Since the minimum file system
476 * blocksize is 512 bytes, we can get a max of 2 extents back from
479 #define SYMLINK_MAPS 2
486 xfs_mount_t *mp = ip->i_mount;
487 int pathlen = ip->i_d.di_size;
488 int nmaps = SYMLINK_MAPS;
489 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
496 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
497 mval, &nmaps, NULL, NULL);
501 for (n = 0; n < nmaps; n++) {
502 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
503 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
505 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt),
506 XBF_LOCK | XBF_MAPPED | XBF_DONT_BLOCK);
507 error = XFS_BUF_GETERROR(bp);
509 xfs_ioerror_alert("xfs_readlink",
510 ip->i_mount, bp, XFS_BUF_ADDR(bp));
514 if (pathlen < byte_cnt)
518 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
522 link[ip->i_d.di_size] = '\0';
534 xfs_mount_t *mp = ip->i_mount;
538 xfs_itrace_entry(ip);
540 if (XFS_FORCED_SHUTDOWN(mp))
541 return XFS_ERROR(EIO);
543 xfs_ilock(ip, XFS_ILOCK_SHARED);
545 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
546 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
548 pathlen = ip->i_d.di_size;
552 if (ip->i_df.if_flags & XFS_IFINLINE) {
553 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
554 link[pathlen] = '\0';
556 error = xfs_readlink_bmap(ip, link);
560 xfs_iunlock(ip, XFS_ILOCK_SHARED);
565 * Flags for xfs_free_eofblocks
567 #define XFS_FREE_EOF_TRYLOCK (1<<0)
570 * This is called by xfs_inactive to free any blocks beyond eof
571 * when the link count isn't zero and by xfs_dm_punch_hole() when
572 * punching a hole to EOF.
582 xfs_fileoff_t end_fsb;
583 xfs_fileoff_t last_fsb;
584 xfs_filblks_t map_len;
586 xfs_bmbt_irec_t imap;
589 * Figure out if there are any blocks beyond the end
590 * of the file. If not, then there is nothing to do.
592 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
593 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
594 map_len = last_fsb - end_fsb;
599 xfs_ilock(ip, XFS_ILOCK_SHARED);
600 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
601 NULL, 0, &imap, &nimaps, NULL, NULL);
602 xfs_iunlock(ip, XFS_ILOCK_SHARED);
604 if (!error && (nimaps != 0) &&
605 (imap.br_startblock != HOLESTARTBLOCK ||
606 ip->i_delayed_blks)) {
608 * Attach the dquots to the inode up front.
610 error = xfs_qm_dqattach(ip, 0);
615 * There are blocks after the end of file.
616 * Free them up now by truncating the file to
619 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
622 * Do the xfs_itruncate_start() call before
623 * reserving any log space because
624 * itruncate_start will call into the buffer
626 * do that within a transaction.
628 if (flags & XFS_FREE_EOF_TRYLOCK) {
629 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
630 xfs_trans_cancel(tp, 0);
634 xfs_ilock(ip, XFS_IOLOCK_EXCL);
636 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
639 xfs_trans_cancel(tp, 0);
640 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
644 error = xfs_trans_reserve(tp, 0,
645 XFS_ITRUNCATE_LOG_RES(mp),
646 0, XFS_TRANS_PERM_LOG_RES,
647 XFS_ITRUNCATE_LOG_COUNT);
649 ASSERT(XFS_FORCED_SHUTDOWN(mp));
650 xfs_trans_cancel(tp, 0);
651 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
655 xfs_ilock(ip, XFS_ILOCK_EXCL);
656 xfs_trans_ijoin(tp, ip,
659 xfs_trans_ihold(tp, ip);
661 error = xfs_itruncate_finish(&tp, ip,
666 * If we get an error at this point we
667 * simply don't bother truncating the file.
671 (XFS_TRANS_RELEASE_LOG_RES |
674 error = xfs_trans_commit(tp,
675 XFS_TRANS_RELEASE_LOG_RES);
677 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
683 * Free a symlink that has blocks associated with it.
686 xfs_inactive_symlink_rmt(
694 xfs_fsblock_t first_block;
695 xfs_bmap_free_t free_list;
698 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
706 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
708 * We're freeing a symlink that has some
709 * blocks allocated to it. Free the
710 * blocks here. We know that we've got
711 * either 1 or 2 extents and that we can
712 * free them all in one bunmapi call.
714 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
715 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
716 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
717 ASSERT(XFS_FORCED_SHUTDOWN(mp));
718 xfs_trans_cancel(tp, 0);
723 * Lock the inode, fix the size, and join it to the transaction.
724 * Hold it so in the normal path, we still have it locked for
725 * the second transaction. In the error paths we need it
726 * held so the cancel won't rele it, see below.
728 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
729 size = (int)ip->i_d.di_size;
731 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
732 xfs_trans_ihold(tp, ip);
733 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
735 * Find the block(s) so we can inval and unmap them.
738 xfs_bmap_init(&free_list, &first_block);
739 nmaps = ARRAY_SIZE(mval);
740 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
741 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
745 * Invalidate the block(s).
747 for (i = 0; i < nmaps; i++) {
748 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
749 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
750 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
751 xfs_trans_binval(tp, bp);
754 * Unmap the dead block(s) to the free_list.
756 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
757 &first_block, &free_list, NULL, &done)))
761 * Commit the first transaction. This logs the EFI and the inode.
763 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
766 * The transaction must have been committed, since there were
767 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
768 * The new tp has the extent freeing and EFDs.
772 * The first xact was committed, so add the inode to the new one.
773 * Mark it dirty so it will be logged and moved forward in the log as
774 * part of every commit.
776 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
777 xfs_trans_ihold(tp, ip);
778 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
780 * Get a new, empty transaction to return to our caller.
782 ntp = xfs_trans_dup(tp);
784 * Commit the transaction containing extent freeing and EFDs.
785 * If we get an error on the commit here or on the reserve below,
786 * we need to unlock the inode since the new transaction doesn't
787 * have the inode attached.
789 error = xfs_trans_commit(tp, 0);
792 ASSERT(XFS_FORCED_SHUTDOWN(mp));
796 * transaction commit worked ok so we can drop the extra ticket
797 * reference that we gained in xfs_trans_dup()
799 xfs_log_ticket_put(tp->t_ticket);
802 * Remove the memory for extent descriptions (just bookkeeping).
804 if (ip->i_df.if_bytes)
805 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
806 ASSERT(ip->i_df.if_bytes == 0);
808 * Put an itruncate log reservation in the new transaction
811 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
812 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
813 ASSERT(XFS_FORCED_SHUTDOWN(mp));
817 * Return with the inode locked but not joined to the transaction.
823 xfs_bmap_cancel(&free_list);
826 * Have to come here with the inode locked and either
827 * (held and in the transaction) or (not in the transaction).
828 * If the inode isn't held then cancel would iput it, but
829 * that's wrong since this is inactive and the vnode ref
830 * count is 0 already.
831 * Cancel won't do anything to the inode if held, but it still
832 * needs to be locked until the cancel is done, if it was
833 * joined to the transaction.
835 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
836 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
843 xfs_inactive_symlink_local(
849 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
851 * We're freeing a symlink which fit into
852 * the inode. Just free the memory used
853 * to hold the old symlink.
855 error = xfs_trans_reserve(*tpp, 0,
856 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
857 0, XFS_TRANS_PERM_LOG_RES,
858 XFS_ITRUNCATE_LOG_COUNT);
861 xfs_trans_cancel(*tpp, 0);
865 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
868 * Zero length symlinks _can_ exist.
870 if (ip->i_df.if_bytes > 0) {
871 xfs_idata_realloc(ip,
872 -(ip->i_df.if_bytes),
874 ASSERT(ip->i_df.if_bytes == 0);
888 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
891 ASSERT(ip->i_d.di_forkoff != 0);
892 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
893 xfs_iunlock(ip, XFS_ILOCK_EXCL);
897 error = xfs_attr_inactive(ip);
901 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
902 error = xfs_trans_reserve(tp, 0,
903 XFS_IFREE_LOG_RES(mp),
904 0, XFS_TRANS_PERM_LOG_RES,
905 XFS_INACTIVE_LOG_COUNT);
909 xfs_ilock(ip, XFS_ILOCK_EXCL);
910 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
911 xfs_trans_ihold(tp, ip);
912 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
914 ASSERT(ip->i_d.di_anextents == 0);
920 ASSERT(XFS_FORCED_SHUTDOWN(mp));
921 xfs_trans_cancel(tp, 0);
924 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
932 xfs_mount_t *mp = ip->i_mount;
935 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
938 /* If this is a read-only mount, don't do this (would generate I/O) */
939 if (mp->m_flags & XFS_MOUNT_RDONLY)
942 if (!XFS_FORCED_SHUTDOWN(mp)) {
946 * If we are using filestreams, and we have an unlinked
947 * file that we are processing the last close on, then nothing
948 * will be able to reopen and write to this file. Purge this
949 * inode from the filestreams cache so that it doesn't delay
950 * teardown of the inode.
952 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
953 xfs_filestream_deassociate(ip);
956 * If we previously truncated this file and removed old data
957 * in the process, we want to initiate "early" writeout on
958 * the last close. This is an attempt to combat the notorious
959 * NULL files problem which is particularly noticable from a
960 * truncate down, buffered (re-)write (delalloc), followed by
961 * a crash. What we are effectively doing here is
962 * significantly reducing the time window where we'd otherwise
963 * be exposed to that problem.
965 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
966 if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
967 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
970 if (ip->i_d.di_nlink != 0) {
971 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
972 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
973 ip->i_delayed_blks > 0)) &&
974 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
975 (!(ip->i_d.di_flags &
976 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
979 * If we can't get the iolock just skip truncating
980 * the blocks past EOF because we could deadlock
981 * with the mmap_sem otherwise. We'll get another
982 * chance to drop them once the last reference to
983 * the inode is dropped, so we'll never leak blocks
986 error = xfs_free_eofblocks(mp, ip,
987 XFS_FREE_EOF_TRYLOCK);
999 * This is called when the vnode reference count for the vnode
1000 * goes to zero. If the file has been unlinked, then it must
1001 * now be truncated. Also, we clear all of the read-ahead state
1002 * kept for the inode here since the file is now closed.
1008 xfs_bmap_free_t free_list;
1009 xfs_fsblock_t first_block;
1016 xfs_itrace_entry(ip);
1019 * If the inode is already free, then there can be nothing
1022 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
1023 ASSERT(ip->i_df.if_real_bytes == 0);
1024 ASSERT(ip->i_df.if_broot_bytes == 0);
1025 return VN_INACTIVE_CACHE;
1029 * Only do a truncate if it's a regular file with
1030 * some actual space in it. It's OK to look at the
1031 * inode's fields without the lock because we're the
1032 * only one with a reference to the inode.
1034 truncate = ((ip->i_d.di_nlink == 0) &&
1035 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1036 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1037 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1043 /* If this is a read-only mount, don't do this (would generate I/O) */
1044 if (mp->m_flags & XFS_MOUNT_RDONLY)
1047 if (ip->i_d.di_nlink != 0) {
1048 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1049 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
1050 ip->i_delayed_blks > 0)) &&
1051 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1052 (!(ip->i_d.di_flags &
1053 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1054 (ip->i_delayed_blks != 0)))) {
1055 error = xfs_free_eofblocks(mp, ip, 0);
1057 return VN_INACTIVE_CACHE;
1062 ASSERT(ip->i_d.di_nlink == 0);
1064 error = xfs_qm_dqattach(ip, 0);
1066 return VN_INACTIVE_CACHE;
1068 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1071 * Do the xfs_itruncate_start() call before
1072 * reserving any log space because itruncate_start
1073 * will call into the buffer cache and we can't
1074 * do that within a transaction.
1076 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1078 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1080 xfs_trans_cancel(tp, 0);
1081 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1082 return VN_INACTIVE_CACHE;
1085 error = xfs_trans_reserve(tp, 0,
1086 XFS_ITRUNCATE_LOG_RES(mp),
1087 0, XFS_TRANS_PERM_LOG_RES,
1088 XFS_ITRUNCATE_LOG_COUNT);
1090 /* Don't call itruncate_cleanup */
1091 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1092 xfs_trans_cancel(tp, 0);
1093 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1094 return VN_INACTIVE_CACHE;
1097 xfs_ilock(ip, XFS_ILOCK_EXCL);
1098 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1099 xfs_trans_ihold(tp, ip);
1102 * normally, we have to run xfs_itruncate_finish sync.
1103 * But if filesystem is wsync and we're in the inactive
1104 * path, then we know that nlink == 0, and that the
1105 * xaction that made nlink == 0 is permanently committed
1106 * since xfs_remove runs as a synchronous transaction.
1108 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1109 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1112 xfs_trans_cancel(tp,
1113 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1114 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1115 return VN_INACTIVE_CACHE;
1117 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1120 * If we get an error while cleaning up a
1121 * symlink we bail out.
1123 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1124 xfs_inactive_symlink_rmt(ip, &tp) :
1125 xfs_inactive_symlink_local(ip, &tp);
1129 return VN_INACTIVE_CACHE;
1132 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1133 xfs_trans_ihold(tp, ip);
1135 error = xfs_trans_reserve(tp, 0,
1136 XFS_IFREE_LOG_RES(mp),
1137 0, XFS_TRANS_PERM_LOG_RES,
1138 XFS_INACTIVE_LOG_COUNT);
1140 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1141 xfs_trans_cancel(tp, 0);
1142 return VN_INACTIVE_CACHE;
1145 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1146 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1147 xfs_trans_ihold(tp, ip);
1151 * If there are attributes associated with the file
1152 * then blow them away now. The code calls a routine
1153 * that recursively deconstructs the attribute fork.
1154 * We need to just commit the current transaction
1155 * because we can't use it for xfs_attr_inactive().
1157 if (ip->i_d.di_anextents > 0) {
1158 error = xfs_inactive_attrs(ip, &tp);
1160 * If we got an error, the transaction is already
1161 * cancelled, and the inode is unlocked. Just get out.
1164 return VN_INACTIVE_CACHE;
1165 } else if (ip->i_afp) {
1166 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1172 xfs_bmap_init(&free_list, &first_block);
1173 error = xfs_ifree(tp, ip, &free_list);
1176 * If we fail to free the inode, shut down. The cancel
1177 * might do that, we need to make sure. Otherwise the
1178 * inode might be lost for a long time or forever.
1180 if (!XFS_FORCED_SHUTDOWN(mp)) {
1182 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1183 error, mp->m_fsname);
1184 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1186 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1189 * Credit the quota account(s). The inode is gone.
1191 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1194 * Just ignore errors at this point. There is nothing we can
1195 * do except to try to keep going. Make sure it's not a silent
1198 error = xfs_bmap_finish(&tp, &free_list, &committed);
1200 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1201 "xfs_bmap_finish() returned error %d", error);
1202 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1204 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1205 "xfs_trans_commit() returned error %d", error);
1209 * Release the dquots held by inode, if any.
1211 xfs_qm_dqdetach(ip);
1212 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1215 return VN_INACTIVE_CACHE;
1219 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1220 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1221 * ci_name->name will point to a the actual name (caller must free) or
1222 * will be set to NULL if an exact match is found.
1227 struct xfs_name *name,
1229 struct xfs_name *ci_name)
1235 xfs_itrace_entry(dp);
1237 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1238 return XFS_ERROR(EIO);
1240 lock_mode = xfs_ilock_map_shared(dp);
1241 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
1242 xfs_iunlock_map_shared(dp, lock_mode);
1247 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
1255 kmem_free(ci_name->name);
1264 struct xfs_name *name,
1270 int is_dir = S_ISDIR(mode);
1271 struct xfs_mount *mp = dp->i_mount;
1272 struct xfs_inode *ip = NULL;
1273 struct xfs_trans *tp = NULL;
1275 xfs_bmap_free_t free_list;
1276 xfs_fsblock_t first_block;
1277 boolean_t unlock_dp_on_error = B_FALSE;
1281 struct xfs_dquot *udqp = NULL;
1282 struct xfs_dquot *gdqp = NULL;
1287 xfs_itrace_entry(dp);
1289 if (XFS_FORCED_SHUTDOWN(mp))
1290 return XFS_ERROR(EIO);
1292 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1293 prid = dp->i_d.di_projid;
1298 * Make sure that we have allocated dquot(s) on disk.
1300 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1301 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1307 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1308 log_res = XFS_MKDIR_LOG_RES(mp);
1309 log_count = XFS_MKDIR_LOG_COUNT;
1310 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1312 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1313 log_res = XFS_CREATE_LOG_RES(mp);
1314 log_count = XFS_CREATE_LOG_COUNT;
1315 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1318 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1321 * Initially assume that the file does not exist and
1322 * reserve the resources for that case. If that is not
1323 * the case we'll drop the one we have and get a more
1324 * appropriate transaction later.
1326 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1327 XFS_TRANS_PERM_LOG_RES, log_count);
1328 if (error == ENOSPC) {
1329 /* flush outstanding delalloc blocks and retry */
1330 xfs_flush_inodes(dp);
1331 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1332 XFS_TRANS_PERM_LOG_RES, log_count);
1334 if (error == ENOSPC) {
1335 /* No space at all so try a "no-allocation" reservation */
1337 error = xfs_trans_reserve(tp, 0, log_res, 0,
1338 XFS_TRANS_PERM_LOG_RES, log_count);
1342 goto out_trans_cancel;
1345 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1346 unlock_dp_on_error = B_TRUE;
1349 * Check for directory link count overflow.
1351 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
1352 error = XFS_ERROR(EMLINK);
1353 goto out_trans_cancel;
1356 xfs_bmap_init(&free_list, &first_block);
1359 * Reserve disk quota and the inode.
1361 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
1363 goto out_trans_cancel;
1365 error = xfs_dir_canenter(tp, dp, name, resblks);
1367 goto out_trans_cancel;
1370 * A newly created regular or special file just has one directory
1371 * entry pointing to them, but a directory also the "." entry
1372 * pointing to itself.
1374 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, credp,
1375 prid, resblks > 0, &ip, &committed);
1377 if (error == ENOSPC)
1378 goto out_trans_cancel;
1379 goto out_trans_abort;
1383 * At this point, we've gotten a newly allocated inode.
1384 * It is locked (and joined to the transaction).
1386 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1389 * Now we join the directory inode to the transaction. We do not do it
1390 * earlier because xfs_dir_ialloc might commit the previous transaction
1391 * (and release all the locks). An error from here on will result in
1392 * the transaction cancel unlocking dp so don't do it explicitly in the
1396 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1397 unlock_dp_on_error = B_FALSE;
1399 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1400 &first_block, &free_list, resblks ?
1401 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1403 ASSERT(error != ENOSPC);
1404 goto out_trans_abort;
1406 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1407 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1410 error = xfs_dir_init(tp, ip, dp);
1412 goto out_bmap_cancel;
1414 error = xfs_bumplink(tp, dp);
1416 goto out_bmap_cancel;
1420 * If this is a synchronous mount, make sure that the
1421 * create transaction goes to disk before returning to
1424 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1425 xfs_trans_set_sync(tp);
1428 * Attach the dquot(s) to the inodes and modify them incore.
1429 * These ids of the inode couldn't have changed since the new
1430 * inode has been locked ever since it was created.
1432 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
1435 * xfs_trans_commit normally decrements the vnode ref count
1436 * when it unlocks the inode. Since we want to return the
1437 * vnode to the caller, we bump the vnode ref count now.
1441 error = xfs_bmap_finish(&tp, &free_list, &committed);
1443 goto out_abort_rele;
1445 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1451 xfs_qm_dqrele(udqp);
1452 xfs_qm_dqrele(gdqp);
1458 xfs_bmap_cancel(&free_list);
1460 cancel_flags |= XFS_TRANS_ABORT;
1462 xfs_trans_cancel(tp, cancel_flags);
1464 xfs_qm_dqrele(udqp);
1465 xfs_qm_dqrele(gdqp);
1467 if (unlock_dp_on_error)
1468 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1474 * Wait until after the current transaction is aborted to
1475 * release the inode. This prevents recursive transactions
1476 * and deadlocks from xfs_inactive.
1478 xfs_bmap_cancel(&free_list);
1479 cancel_flags |= XFS_TRANS_ABORT;
1480 xfs_trans_cancel(tp, cancel_flags);
1482 unlock_dp_on_error = B_FALSE;
1488 int xfs_small_retries;
1489 int xfs_middle_retries;
1490 int xfs_lots_retries;
1491 int xfs_lock_delays;
1495 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1499 xfs_lock_inumorder(int lock_mode, int subclass)
1501 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1502 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1503 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1504 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1510 * The following routine will lock n inodes in exclusive mode.
1511 * We assume the caller calls us with the inodes in i_ino order.
1513 * We need to detect deadlock where an inode that we lock
1514 * is in the AIL and we start waiting for another inode that is locked
1515 * by a thread in a long running transaction (such as truncate). This can
1516 * result in deadlock since the long running trans might need to wait
1517 * for the inode we just locked in order to push the tail and free space
1526 int attempts = 0, i, j, try_lock;
1529 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1535 for (; i < inodes; i++) {
1538 if (i && (ips[i] == ips[i-1])) /* Already locked */
1542 * If try_lock is not set yet, make sure all locked inodes
1543 * are not in the AIL.
1544 * If any are, set try_lock to be used later.
1548 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1549 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1550 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1557 * If any of the previous locks we have locked is in the AIL,
1558 * we must TRY to get the second and subsequent locks. If
1559 * we can't get any, we must release all we have
1564 /* try_lock must be 0 if i is 0. */
1566 * try_lock means we have an inode locked
1567 * that is in the AIL.
1570 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1574 * Unlock all previous guys and try again.
1575 * xfs_iunlock will try to push the tail
1576 * if the inode is in the AIL.
1579 for(j = i - 1; j >= 0; j--) {
1582 * Check to see if we've already
1583 * unlocked this one.
1584 * Not the first one going back,
1585 * and the inode ptr is the same.
1587 if ((j != (i - 1)) && ips[j] ==
1591 xfs_iunlock(ips[j], lock_mode);
1594 if ((attempts % 5) == 0) {
1595 delay(1); /* Don't just spin the CPU */
1605 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1611 if (attempts < 5) xfs_small_retries++;
1612 else if (attempts < 100) xfs_middle_retries++;
1613 else xfs_lots_retries++;
1621 * xfs_lock_two_inodes() can only be used to lock one type of lock
1622 * at a time - the iolock or the ilock, but not both at once. If
1623 * we lock both at once, lockdep will report false positives saying
1624 * we have violated locking orders.
1627 xfs_lock_two_inodes(
1636 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1637 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1638 ASSERT(ip0->i_ino != ip1->i_ino);
1640 if (ip0->i_ino > ip1->i_ino) {
1647 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1650 * If the first lock we have locked is in the AIL, we must TRY to get
1651 * the second lock. If we can't get it, we must release the first one
1654 lp = (xfs_log_item_t *)ip0->i_itemp;
1655 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1656 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1657 xfs_iunlock(ip0, lock_mode);
1658 if ((++attempts % 5) == 0)
1659 delay(1); /* Don't just spin the CPU */
1663 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1670 struct xfs_name *name,
1673 xfs_mount_t *mp = dp->i_mount;
1674 xfs_trans_t *tp = NULL;
1675 int is_dir = S_ISDIR(ip->i_d.di_mode);
1677 xfs_bmap_free_t free_list;
1678 xfs_fsblock_t first_block;
1685 xfs_itrace_entry(dp);
1686 xfs_itrace_entry(ip);
1688 if (XFS_FORCED_SHUTDOWN(mp))
1689 return XFS_ERROR(EIO);
1691 error = xfs_qm_dqattach(dp, 0);
1695 error = xfs_qm_dqattach(ip, 0);
1700 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1701 log_count = XFS_DEFAULT_LOG_COUNT;
1703 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1704 log_count = XFS_REMOVE_LOG_COUNT;
1706 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1709 * We try to get the real space reservation first,
1710 * allowing for directory btree deletion(s) implying
1711 * possible bmap insert(s). If we can't get the space
1712 * reservation then we use 0 instead, and avoid the bmap
1713 * btree insert(s) in the directory code by, if the bmap
1714 * insert tries to happen, instead trimming the LAST
1715 * block from the directory.
1717 resblks = XFS_REMOVE_SPACE_RES(mp);
1718 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1719 XFS_TRANS_PERM_LOG_RES, log_count);
1720 if (error == ENOSPC) {
1722 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1723 XFS_TRANS_PERM_LOG_RES, log_count);
1726 ASSERT(error != ENOSPC);
1728 goto out_trans_cancel;
1731 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1734 * At this point, we've gotten both the directory and the entry
1738 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1741 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1744 * If we're removing a directory perform some additional validation.
1747 ASSERT(ip->i_d.di_nlink >= 2);
1748 if (ip->i_d.di_nlink != 2) {
1749 error = XFS_ERROR(ENOTEMPTY);
1750 goto out_trans_cancel;
1752 if (!xfs_dir_isempty(ip)) {
1753 error = XFS_ERROR(ENOTEMPTY);
1754 goto out_trans_cancel;
1758 xfs_bmap_init(&free_list, &first_block);
1759 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1760 &first_block, &free_list, resblks);
1762 ASSERT(error != ENOENT);
1763 goto out_bmap_cancel;
1765 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1769 * Drop the link from ip's "..".
1771 error = xfs_droplink(tp, dp);
1773 goto out_bmap_cancel;
1776 * Drop the "." link from ip to self.
1778 error = xfs_droplink(tp, ip);
1780 goto out_bmap_cancel;
1783 * When removing a non-directory we need to log the parent
1784 * inode here. For a directory this is done implicitly
1785 * by the xfs_droplink call for the ".." entry.
1787 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1791 * Drop the link from dp to ip.
1793 error = xfs_droplink(tp, ip);
1795 goto out_bmap_cancel;
1798 * Determine if this is the last link while
1799 * we are in the transaction.
1801 link_zero = (ip->i_d.di_nlink == 0);
1804 * If this is a synchronous mount, make sure that the
1805 * remove transaction goes to disk before returning to
1808 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1809 xfs_trans_set_sync(tp);
1811 error = xfs_bmap_finish(&tp, &free_list, &committed);
1813 goto out_bmap_cancel;
1815 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1820 * If we are using filestreams, kill the stream association.
1821 * If the file is still open it may get a new one but that
1822 * will get killed on last close in xfs_close() so we don't
1823 * have to worry about that.
1825 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1826 xfs_filestream_deassociate(ip);
1831 xfs_bmap_cancel(&free_list);
1832 cancel_flags |= XFS_TRANS_ABORT;
1834 xfs_trans_cancel(tp, cancel_flags);
1843 struct xfs_name *target_name)
1845 xfs_mount_t *mp = tdp->i_mount;
1848 xfs_bmap_free_t free_list;
1849 xfs_fsblock_t first_block;
1854 xfs_itrace_entry(tdp);
1855 xfs_itrace_entry(sip);
1857 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1859 if (XFS_FORCED_SHUTDOWN(mp))
1860 return XFS_ERROR(EIO);
1862 error = xfs_qm_dqattach(sip, 0);
1866 error = xfs_qm_dqattach(tdp, 0);
1870 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1871 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1872 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1873 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1874 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1875 if (error == ENOSPC) {
1877 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1878 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1885 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1888 * Increment vnode ref counts since xfs_trans_commit &
1889 * xfs_trans_cancel will both unlock the inodes and
1890 * decrement the associated ref counts.
1894 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1895 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1898 * If the source has too many links, we can't make any more to it.
1900 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
1901 error = XFS_ERROR(EMLINK);
1906 * If we are using project inheritance, we only allow hard link
1907 * creation in our tree when the project IDs are the same; else
1908 * the tree quota mechanism could be circumvented.
1910 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1911 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
1912 error = XFS_ERROR(EXDEV);
1916 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1920 xfs_bmap_init(&free_list, &first_block);
1922 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1923 &first_block, &free_list, resblks);
1926 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1927 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1929 error = xfs_bumplink(tp, sip);
1934 * If this is a synchronous mount, make sure that the
1935 * link transaction goes to disk before returning to
1938 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1939 xfs_trans_set_sync(tp);
1942 error = xfs_bmap_finish (&tp, &free_list, &committed);
1944 xfs_bmap_cancel(&free_list);
1948 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1951 cancel_flags |= XFS_TRANS_ABORT;
1953 xfs_trans_cancel(tp, cancel_flags);
1961 struct xfs_name *link_name,
1962 const char *target_path,
1967 xfs_mount_t *mp = dp->i_mount;
1972 xfs_bmap_free_t free_list;
1973 xfs_fsblock_t first_block;
1974 boolean_t unlock_dp_on_error = B_FALSE;
1977 xfs_fileoff_t first_fsb;
1978 xfs_filblks_t fs_blocks;
1980 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1982 const char *cur_chunk;
1987 struct xfs_dquot *udqp, *gdqp;
1995 xfs_itrace_entry(dp);
1997 if (XFS_FORCED_SHUTDOWN(mp))
1998 return XFS_ERROR(EIO);
2001 * Check component lengths of the target path name.
2003 pathlen = strlen(target_path);
2004 if (pathlen >= MAXPATHLEN) /* total string too long */
2005 return XFS_ERROR(ENAMETOOLONG);
2008 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2009 prid = dp->i_d.di_projid;
2011 prid = (xfs_prid_t)dfltprid;
2014 * Make sure that we have allocated dquot(s) on disk.
2016 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
2017 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2021 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
2022 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2024 * The symlink will fit into the inode data fork?
2025 * There can't be any attributes so we get the whole variable part.
2027 if (pathlen <= XFS_LITINO(mp))
2030 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
2031 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
2032 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2033 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2034 if (error == ENOSPC && fs_blocks == 0) {
2036 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2037 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2044 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2045 unlock_dp_on_error = B_TRUE;
2048 * Check whether the directory allows new symlinks or not.
2050 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2051 error = XFS_ERROR(EPERM);
2056 * Reserve disk quota : blocks and inode.
2058 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
2063 * Check for ability to enter directory entry, if no space reserved.
2065 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2069 * Initialize the bmap freelist prior to calling either
2070 * bmapi or the directory create code.
2072 xfs_bmap_init(&free_list, &first_block);
2075 * Allocate an inode for the symlink.
2077 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
2078 1, 0, credp, prid, resblks > 0, &ip, NULL);
2080 if (error == ENOSPC)
2086 * An error after we've joined dp to the transaction will result in the
2087 * transaction cancel unlocking dp so don't do it explicitly in the
2091 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2092 unlock_dp_on_error = B_FALSE;
2095 * Also attach the dquot(s) to it, if applicable.
2097 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
2100 resblks -= XFS_IALLOC_SPACE_RES(mp);
2102 * If the symlink will fit into the inode, write it inline.
2104 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2105 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2106 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2107 ip->i_d.di_size = pathlen;
2110 * The inode was initially created in extent format.
2112 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2113 ip->i_df.if_flags |= XFS_IFINLINE;
2115 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2116 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2120 nmaps = SYMLINK_MAPS;
2122 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2123 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2124 &first_block, resblks, mval, &nmaps,
2131 resblks -= fs_blocks;
2132 ip->i_d.di_size = pathlen;
2133 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2135 cur_chunk = target_path;
2136 for (n = 0; n < nmaps; n++) {
2137 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2138 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2139 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2140 BTOBB(byte_cnt), 0);
2141 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2142 if (pathlen < byte_cnt) {
2145 pathlen -= byte_cnt;
2147 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2148 cur_chunk += byte_cnt;
2150 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2155 * Create the directory entry for the symlink.
2157 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2158 &first_block, &free_list, resblks);
2161 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2162 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2165 * If this is a synchronous mount, make sure that the
2166 * symlink transaction goes to disk before returning to
2169 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2170 xfs_trans_set_sync(tp);
2174 * xfs_trans_commit normally decrements the vnode ref count
2175 * when it unlocks the inode. Since we want to return the
2176 * vnode to the caller, we bump the vnode ref count now.
2180 error = xfs_bmap_finish(&tp, &free_list, &committed);
2184 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2185 xfs_qm_dqrele(udqp);
2186 xfs_qm_dqrele(gdqp);
2194 xfs_bmap_cancel(&free_list);
2195 cancel_flags |= XFS_TRANS_ABORT;
2197 xfs_trans_cancel(tp, cancel_flags);
2198 xfs_qm_dqrele(udqp);
2199 xfs_qm_dqrele(gdqp);
2201 if (unlock_dp_on_error)
2202 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2213 xfs_mount_t *mp = ip->i_mount;
2217 if (!capable(CAP_SYS_ADMIN))
2218 return XFS_ERROR(EPERM);
2220 if (XFS_FORCED_SHUTDOWN(mp))
2221 return XFS_ERROR(EIO);
2223 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2224 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2226 xfs_trans_cancel(tp, 0);
2229 xfs_ilock(ip, XFS_ILOCK_EXCL);
2230 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2232 ip->i_d.di_dmevmask = evmask;
2233 ip->i_d.di_dmstate = state;
2235 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2237 error = xfs_trans_commit(tp, 0);
2243 * xfs_alloc_file_space()
2244 * This routine allocates disk space for the given file.
2246 * If alloc_type == 0, this request is for an ALLOCSP type
2247 * request which will change the file size. In this case, no
2248 * DMAPI event will be generated by the call. A TRUNCATE event
2249 * will be generated later by xfs_setattr.
2251 * If alloc_type != 0, this request is for a RESVSP type
2252 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2253 * lower block boundary byte address is less than the file's
2262 xfs_alloc_file_space(
2269 xfs_mount_t *mp = ip->i_mount;
2271 xfs_filblks_t allocated_fsb;
2272 xfs_filblks_t allocatesize_fsb;
2273 xfs_extlen_t extsz, temp;
2274 xfs_fileoff_t startoffset_fsb;
2275 xfs_fsblock_t firstfsb;
2281 xfs_bmbt_irec_t imaps[1], *imapp;
2282 xfs_bmap_free_t free_list;
2283 uint qblocks, resblks, resrtextents;
2287 xfs_itrace_entry(ip);
2289 if (XFS_FORCED_SHUTDOWN(mp))
2290 return XFS_ERROR(EIO);
2292 error = xfs_qm_dqattach(ip, 0);
2297 return XFS_ERROR(EINVAL);
2299 rt = XFS_IS_REALTIME_INODE(ip);
2300 extsz = xfs_get_extsz_hint(ip);
2305 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
2306 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
2307 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
2310 * Allocate file space until done or until there is an error
2312 while (allocatesize_fsb && !error) {
2316 * Determine space reservations for data/realtime.
2318 if (unlikely(extsz)) {
2319 s = startoffset_fsb;
2322 e = startoffset_fsb + allocatesize_fsb;
2323 if ((temp = do_mod(startoffset_fsb, extsz)))
2325 if ((temp = do_mod(e, extsz)))
2329 e = allocatesize_fsb;
2333 resrtextents = qblocks = (uint)(e - s);
2334 resrtextents /= mp->m_sb.sb_rextsize;
2335 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2336 quota_flag = XFS_QMOPT_RES_RTBLKS;
2339 resblks = qblocks = \
2340 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
2341 quota_flag = XFS_QMOPT_RES_REGBLKS;
2345 * Allocate and setup the transaction.
2347 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2348 error = xfs_trans_reserve(tp, resblks,
2349 XFS_WRITE_LOG_RES(mp), resrtextents,
2350 XFS_TRANS_PERM_LOG_RES,
2351 XFS_WRITE_LOG_COUNT);
2353 * Check for running out of space
2357 * Free the transaction structure.
2359 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2360 xfs_trans_cancel(tp, 0);
2363 xfs_ilock(ip, XFS_ILOCK_EXCL);
2364 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
2369 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2370 xfs_trans_ihold(tp, ip);
2373 * Issue the xfs_bmapi() call to allocate the blocks
2375 xfs_bmap_init(&free_list, &firstfsb);
2376 error = xfs_bmapi(tp, ip, startoffset_fsb,
2377 allocatesize_fsb, bmapi_flag,
2378 &firstfsb, 0, imapp, &nimaps,
2385 * Complete the transaction
2387 error = xfs_bmap_finish(&tp, &free_list, &committed);
2392 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2393 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2398 allocated_fsb = imapp->br_blockcount;
2401 error = XFS_ERROR(ENOSPC);
2405 startoffset_fsb += allocated_fsb;
2406 allocatesize_fsb -= allocated_fsb;
2411 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
2412 xfs_bmap_cancel(&free_list);
2413 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
2415 error1: /* Just cancel transaction */
2416 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2417 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2422 * Zero file bytes between startoff and endoff inclusive.
2423 * The iolock is held exclusive and no blocks are buffered.
2425 * This function is used by xfs_free_file_space() to zero
2426 * partial blocks when the range to free is not block aligned.
2427 * When unreserving space with boundaries that are not block
2428 * aligned we round up the start and round down the end
2429 * boundaries and then use this function to zero the parts of
2430 * the blocks that got dropped during the rounding.
2433 xfs_zero_remaining_bytes(
2438 xfs_bmbt_irec_t imap;
2439 xfs_fileoff_t offset_fsb;
2440 xfs_off_t lastoffset;
2443 xfs_mount_t *mp = ip->i_mount;
2448 * Avoid doing I/O beyond eof - it's not necessary
2449 * since nothing can read beyond eof. The space will
2450 * be zeroed when the file is extended anyway.
2452 if (startoff >= ip->i_size)
2455 if (endoff > ip->i_size)
2456 endoff = ip->i_size;
2458 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
2459 XFS_IS_REALTIME_INODE(ip) ?
2460 mp->m_rtdev_targp : mp->m_ddev_targp);
2462 return XFS_ERROR(ENOMEM);
2464 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
2465 offset_fsb = XFS_B_TO_FSBT(mp, offset);
2467 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
2468 NULL, 0, &imap, &nimap, NULL, NULL);
2469 if (error || nimap < 1)
2471 ASSERT(imap.br_blockcount >= 1);
2472 ASSERT(imap.br_startoff == offset_fsb);
2473 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
2474 if (lastoffset > endoff)
2475 lastoffset = endoff;
2476 if (imap.br_startblock == HOLESTARTBLOCK)
2478 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2479 if (imap.br_state == XFS_EXT_UNWRITTEN)
2482 XFS_BUF_UNWRITE(bp);
2484 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
2486 error = xfs_iowait(bp);
2488 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2489 mp, bp, XFS_BUF_ADDR(bp));
2492 memset(XFS_BUF_PTR(bp) +
2493 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2494 0, lastoffset - offset + 1);
2499 error = xfs_iowait(bp);
2501 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2502 mp, bp, XFS_BUF_ADDR(bp));
2511 * xfs_free_file_space()
2512 * This routine frees disk space for the given file.
2514 * This routine is only called by xfs_change_file_space
2515 * for an UNRESVSP type call.
2523 xfs_free_file_space(
2531 xfs_fileoff_t endoffset_fsb;
2533 xfs_fsblock_t firstfsb;
2534 xfs_bmap_free_t free_list;
2535 xfs_bmbt_irec_t imap;
2543 xfs_fileoff_t startoffset_fsb;
2545 int need_iolock = 1;
2549 xfs_itrace_entry(ip);
2551 error = xfs_qm_dqattach(ip, 0);
2556 if (len <= 0) /* if nothing being freed */
2558 rt = XFS_IS_REALTIME_INODE(ip);
2559 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
2560 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
2562 if (attr_flags & XFS_ATTR_NOLOCK)
2565 xfs_ilock(ip, XFS_IOLOCK_EXCL);
2566 /* wait for the completion of any pending DIOs */
2570 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
2571 ioffset = offset & ~(rounding - 1);
2573 if (VN_CACHED(VFS_I(ip)) != 0) {
2574 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
2576 goto out_unlock_iolock;
2580 * Need to zero the stuff we're not freeing, on disk.
2581 * If it's a realtime file & can't use unwritten extents then we
2582 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2583 * will take care of it for us.
2585 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
2587 error = xfs_bmapi(NULL, ip, startoffset_fsb,
2588 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
2590 goto out_unlock_iolock;
2591 ASSERT(nimap == 0 || nimap == 1);
2592 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2595 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2596 block = imap.br_startblock;
2597 mod = do_div(block, mp->m_sb.sb_rextsize);
2599 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2602 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
2603 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
2605 goto out_unlock_iolock;
2606 ASSERT(nimap == 0 || nimap == 1);
2607 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2608 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2610 if (mod && (mod != mp->m_sb.sb_rextsize))
2611 endoffset_fsb -= mod;
2614 if ((done = (endoffset_fsb <= startoffset_fsb)))
2616 * One contiguous piece to clear
2618 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2621 * Some full blocks, possibly two pieces to clear
2623 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2624 error = xfs_zero_remaining_bytes(ip, offset,
2625 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2627 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2628 error = xfs_zero_remaining_bytes(ip,
2629 XFS_FSB_TO_B(mp, endoffset_fsb),
2634 * free file space until done or until there is an error
2636 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2637 while (!error && !done) {
2640 * allocate and setup the transaction. Allow this
2641 * transaction to dip into the reserve blocks to ensure
2642 * the freeing of the space succeeds at ENOSPC.
2644 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2645 tp->t_flags |= XFS_TRANS_RESERVE;
2646 error = xfs_trans_reserve(tp,
2648 XFS_WRITE_LOG_RES(mp),
2650 XFS_TRANS_PERM_LOG_RES,
2651 XFS_WRITE_LOG_COUNT);
2654 * check for running out of space
2658 * Free the transaction structure.
2660 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2661 xfs_trans_cancel(tp, 0);
2664 xfs_ilock(ip, XFS_ILOCK_EXCL);
2665 error = xfs_trans_reserve_quota(tp, mp,
2666 ip->i_udquot, ip->i_gdquot,
2667 resblks, 0, XFS_QMOPT_RES_REGBLKS);
2671 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2672 xfs_trans_ihold(tp, ip);
2675 * issue the bunmapi() call to free the blocks
2677 xfs_bmap_init(&free_list, &firstfsb);
2678 error = xfs_bunmapi(tp, ip, startoffset_fsb,
2679 endoffset_fsb - startoffset_fsb,
2680 0, 2, &firstfsb, &free_list, NULL, &done);
2686 * complete the transaction
2688 error = xfs_bmap_finish(&tp, &free_list, &committed);
2693 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2694 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2699 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2703 xfs_bmap_cancel(&free_list);
2705 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2706 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2712 * xfs_change_file_space()
2713 * This routine allocates or frees disk space for the given file.
2714 * The user specified parameters are checked for alignment and size
2723 xfs_change_file_space(
2730 xfs_mount_t *mp = ip->i_mount;
2735 xfs_off_t startoffset;
2740 xfs_itrace_entry(ip);
2742 if (!S_ISREG(ip->i_d.di_mode))
2743 return XFS_ERROR(EINVAL);
2745 switch (bf->l_whence) {
2746 case 0: /*SEEK_SET*/
2748 case 1: /*SEEK_CUR*/
2749 bf->l_start += offset;
2751 case 2: /*SEEK_END*/
2752 bf->l_start += ip->i_size;
2755 return XFS_ERROR(EINVAL);
2758 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
2760 if ( (bf->l_start < 0)
2761 || (bf->l_start > XFS_MAXIOFFSET(mp))
2762 || (bf->l_start + llen < 0)
2763 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
2764 return XFS_ERROR(EINVAL);
2768 startoffset = bf->l_start;
2772 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2774 * These calls do NOT zero the data space allocated to the file,
2775 * nor do they change the file size.
2777 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2779 * These calls cause the new file data to be zeroed and the file
2780 * size to be changed.
2782 setprealloc = clrprealloc = 0;
2785 case XFS_IOC_RESVSP:
2786 case XFS_IOC_RESVSP64:
2787 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
2794 case XFS_IOC_UNRESVSP:
2795 case XFS_IOC_UNRESVSP64:
2796 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
2801 case XFS_IOC_ALLOCSP:
2802 case XFS_IOC_ALLOCSP64:
2803 case XFS_IOC_FREESP:
2804 case XFS_IOC_FREESP64:
2805 if (startoffset > fsize) {
2806 error = xfs_alloc_file_space(ip, fsize,
2807 startoffset - fsize, 0, attr_flags);
2812 iattr.ia_valid = ATTR_SIZE;
2813 iattr.ia_size = startoffset;
2815 error = xfs_setattr(ip, &iattr, attr_flags);
2825 return XFS_ERROR(EINVAL);
2829 * update the inode timestamp, mode, and prealloc flag bits
2831 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2833 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2836 xfs_trans_cancel(tp, 0);
2840 xfs_ilock(ip, XFS_ILOCK_EXCL);
2842 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2843 xfs_trans_ihold(tp, ip);
2845 if ((attr_flags & XFS_ATTR_DMI) == 0) {
2846 ip->i_d.di_mode &= ~S_ISUID;
2849 * Note that we don't have to worry about mandatory
2850 * file locking being disabled here because we only
2851 * clear the S_ISGID bit if the Group execute bit is
2852 * on, but if it was on then mandatory locking wouldn't
2853 * have been enabled.
2855 if (ip->i_d.di_mode & S_IXGRP)
2856 ip->i_d.di_mode &= ~S_ISGID;
2858 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2861 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2862 else if (clrprealloc)
2863 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2865 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2866 xfs_trans_set_sync(tp);
2868 error = xfs_trans_commit(tp, 0);
2870 xfs_iunlock(ip, XFS_ILOCK_EXCL);