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"
54 * The maximum pathlen is 1024 bytes. Since the minimum file system
55 * blocksize is 512 bytes, we can get a max of 2 extents back from
58 #define SYMLINK_MAPS 2
65 xfs_mount_t *mp = ip->i_mount;
66 int pathlen = ip->i_d.di_size;
67 int nmaps = SYMLINK_MAPS;
68 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
75 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, pathlen), mval, &nmaps,
80 for (n = 0; n < nmaps; n++) {
81 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
82 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
84 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt),
85 XBF_LOCK | XBF_MAPPED | XBF_DONT_BLOCK);
87 return XFS_ERROR(ENOMEM);
90 xfs_ioerror_alert("xfs_readlink",
91 ip->i_mount, bp, XFS_BUF_ADDR(bp));
95 if (pathlen < byte_cnt)
99 memcpy(link, bp->b_addr, byte_cnt);
103 link[ip->i_d.di_size] = '\0';
115 xfs_mount_t *mp = ip->i_mount;
119 trace_xfs_readlink(ip);
121 if (XFS_FORCED_SHUTDOWN(mp))
122 return XFS_ERROR(EIO);
124 xfs_ilock(ip, XFS_ILOCK_SHARED);
126 ASSERT(S_ISLNK(ip->i_d.di_mode));
127 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
129 pathlen = ip->i_d.di_size;
133 if (ip->i_df.if_flags & XFS_IFINLINE) {
134 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
135 link[pathlen] = '\0';
137 error = xfs_readlink_bmap(ip, link);
141 xfs_iunlock(ip, XFS_ILOCK_SHARED);
146 * Flags for xfs_free_eofblocks
148 #define XFS_FREE_EOF_TRYLOCK (1<<0)
151 * This is called by xfs_inactive to free any blocks beyond eof
152 * when the link count isn't zero and by xfs_dm_punch_hole() when
153 * punching a hole to EOF.
163 xfs_fileoff_t end_fsb;
164 xfs_fileoff_t last_fsb;
165 xfs_filblks_t map_len;
167 xfs_bmbt_irec_t imap;
170 * Figure out if there are any blocks beyond the end
171 * of the file. If not, then there is nothing to do.
173 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
174 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
175 if (last_fsb <= end_fsb)
177 map_len = last_fsb - end_fsb;
180 xfs_ilock(ip, XFS_ILOCK_SHARED);
181 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
182 xfs_iunlock(ip, XFS_ILOCK_SHARED);
184 if (!error && (nimaps != 0) &&
185 (imap.br_startblock != HOLESTARTBLOCK ||
186 ip->i_delayed_blks)) {
188 * Attach the dquots to the inode up front.
190 error = xfs_qm_dqattach(ip, 0);
195 * There are blocks after the end of file.
196 * Free them up now by truncating the file to
199 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
201 if (flags & XFS_FREE_EOF_TRYLOCK) {
202 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
203 xfs_trans_cancel(tp, 0);
207 xfs_ilock(ip, XFS_IOLOCK_EXCL);
210 error = xfs_trans_reserve(tp, 0,
211 XFS_ITRUNCATE_LOG_RES(mp),
212 0, XFS_TRANS_PERM_LOG_RES,
213 XFS_ITRUNCATE_LOG_COUNT);
215 ASSERT(XFS_FORCED_SHUTDOWN(mp));
216 xfs_trans_cancel(tp, 0);
217 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
221 xfs_ilock(ip, XFS_ILOCK_EXCL);
222 xfs_trans_ijoin(tp, ip);
224 error = xfs_itruncate_data(&tp, ip, ip->i_size);
227 * If we get an error at this point we simply don't
228 * bother truncating the file.
231 (XFS_TRANS_RELEASE_LOG_RES |
234 error = xfs_trans_commit(tp,
235 XFS_TRANS_RELEASE_LOG_RES);
237 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
243 * Free a symlink that has blocks associated with it.
246 xfs_inactive_symlink_rmt(
254 xfs_fsblock_t first_block;
255 xfs_bmap_free_t free_list;
258 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
266 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
268 * We're freeing a symlink that has some
269 * blocks allocated to it. Free the
270 * blocks here. We know that we've got
271 * either 1 or 2 extents and that we can
272 * free them all in one bunmapi call.
274 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
275 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
276 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
277 ASSERT(XFS_FORCED_SHUTDOWN(mp));
278 xfs_trans_cancel(tp, 0);
283 * Lock the inode, fix the size, and join it to the transaction.
284 * Hold it so in the normal path, we still have it locked for
285 * the second transaction. In the error paths we need it
286 * held so the cancel won't rele it, see below.
288 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
289 size = (int)ip->i_d.di_size;
291 xfs_trans_ijoin(tp, ip);
292 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
294 * Find the block(s) so we can inval and unmap them.
297 xfs_bmap_init(&free_list, &first_block);
298 nmaps = ARRAY_SIZE(mval);
299 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, size),
304 * Invalidate the block(s).
306 for (i = 0; i < nmaps; i++) {
307 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
308 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
309 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
314 xfs_trans_binval(tp, bp);
317 * Unmap the dead block(s) to the free_list.
319 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
320 &first_block, &free_list, &done)))
324 * Commit the first transaction. This logs the EFI and the inode.
326 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
329 * The transaction must have been committed, since there were
330 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
331 * The new tp has the extent freeing and EFDs.
335 * The first xact was committed, so add the inode to the new one.
336 * Mark it dirty so it will be logged and moved forward in the log as
337 * part of every commit.
339 xfs_trans_ijoin(tp, ip);
340 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
342 * Get a new, empty transaction to return to our caller.
344 ntp = xfs_trans_dup(tp);
346 * Commit the transaction containing extent freeing and EFDs.
347 * If we get an error on the commit here or on the reserve below,
348 * we need to unlock the inode since the new transaction doesn't
349 * have the inode attached.
351 error = xfs_trans_commit(tp, 0);
354 ASSERT(XFS_FORCED_SHUTDOWN(mp));
358 * transaction commit worked ok so we can drop the extra ticket
359 * reference that we gained in xfs_trans_dup()
361 xfs_log_ticket_put(tp->t_ticket);
364 * Remove the memory for extent descriptions (just bookkeeping).
366 if (ip->i_df.if_bytes)
367 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
368 ASSERT(ip->i_df.if_bytes == 0);
370 * Put an itruncate log reservation in the new transaction
373 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
374 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
375 ASSERT(XFS_FORCED_SHUTDOWN(mp));
379 * Return with the inode locked but not joined to the transaction.
385 xfs_bmap_cancel(&free_list);
388 * Have to come here with the inode locked and either
389 * (held and in the transaction) or (not in the transaction).
390 * If the inode isn't held then cancel would iput it, but
391 * that's wrong since this is inactive and the vnode ref
392 * count is 0 already.
393 * Cancel won't do anything to the inode if held, but it still
394 * needs to be locked until the cancel is done, if it was
395 * joined to the transaction.
397 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
398 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
405 xfs_inactive_symlink_local(
411 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
413 * We're freeing a symlink which fit into
414 * the inode. Just free the memory used
415 * to hold the old symlink.
417 error = xfs_trans_reserve(*tpp, 0,
418 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
419 0, XFS_TRANS_PERM_LOG_RES,
420 XFS_ITRUNCATE_LOG_COUNT);
423 xfs_trans_cancel(*tpp, 0);
427 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
430 * Zero length symlinks _can_ exist.
432 if (ip->i_df.if_bytes > 0) {
433 xfs_idata_realloc(ip,
434 -(ip->i_df.if_bytes),
436 ASSERT(ip->i_df.if_bytes == 0);
450 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
453 ASSERT(ip->i_d.di_forkoff != 0);
454 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
455 xfs_iunlock(ip, XFS_ILOCK_EXCL);
459 error = xfs_attr_inactive(ip);
463 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
464 error = xfs_trans_reserve(tp, 0,
465 XFS_IFREE_LOG_RES(mp),
466 0, XFS_TRANS_PERM_LOG_RES,
467 XFS_INACTIVE_LOG_COUNT);
471 xfs_ilock(ip, XFS_ILOCK_EXCL);
472 xfs_trans_ijoin(tp, ip);
473 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
475 ASSERT(ip->i_d.di_anextents == 0);
481 ASSERT(XFS_FORCED_SHUTDOWN(mp));
482 xfs_trans_cancel(tp, 0);
485 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
493 xfs_mount_t *mp = ip->i_mount;
496 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
499 /* If this is a read-only mount, don't do this (would generate I/O) */
500 if (mp->m_flags & XFS_MOUNT_RDONLY)
503 if (!XFS_FORCED_SHUTDOWN(mp)) {
507 * If we are using filestreams, and we have an unlinked
508 * file that we are processing the last close on, then nothing
509 * will be able to reopen and write to this file. Purge this
510 * inode from the filestreams cache so that it doesn't delay
511 * teardown of the inode.
513 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
514 xfs_filestream_deassociate(ip);
517 * If we previously truncated this file and removed old data
518 * in the process, we want to initiate "early" writeout on
519 * the last close. This is an attempt to combat the notorious
520 * NULL files problem which is particularly noticeable from a
521 * truncate down, buffered (re-)write (delalloc), followed by
522 * a crash. What we are effectively doing here is
523 * significantly reducing the time window where we'd otherwise
524 * be exposed to that problem.
526 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
528 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
529 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
530 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
534 if (ip->i_d.di_nlink == 0)
537 if ((S_ISREG(ip->i_d.di_mode) &&
538 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
539 ip->i_delayed_blks > 0)) &&
540 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
541 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
544 * If we can't get the iolock just skip truncating the blocks
545 * past EOF because we could deadlock with the mmap_sem
546 * otherwise. We'll get another chance to drop them once the
547 * last reference to the inode is dropped, so we'll never leak
548 * blocks permanently.
550 * Further, check if the inode is being opened, written and
551 * closed frequently and we have delayed allocation blocks
552 * outstanding (e.g. streaming writes from the NFS server),
553 * truncating the blocks past EOF will cause fragmentation to
556 * In this case don't do the truncation, either, but we have to
557 * be careful how we detect this case. Blocks beyond EOF show
558 * up as i_delayed_blks even when the inode is clean, so we
559 * need to truncate them away first before checking for a dirty
560 * release. Hence on the first dirty close we will still remove
561 * the speculative allocation, but after that we will leave it
564 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
567 error = xfs_free_eofblocks(mp, ip,
568 XFS_FREE_EOF_TRYLOCK);
572 /* delalloc blocks after truncation means it really is dirty */
573 if (ip->i_delayed_blks)
574 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
582 * This is called when the vnode reference count for the vnode
583 * goes to zero. If the file has been unlinked, then it must
584 * now be truncated. Also, we clear all of the read-ahead state
585 * kept for the inode here since the file is now closed.
591 xfs_bmap_free_t free_list;
592 xfs_fsblock_t first_block;
600 * If the inode is already free, then there can be nothing
603 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
604 ASSERT(ip->i_df.if_real_bytes == 0);
605 ASSERT(ip->i_df.if_broot_bytes == 0);
606 return VN_INACTIVE_CACHE;
610 * Only do a truncate if it's a regular file with
611 * some actual space in it. It's OK to look at the
612 * inode's fields without the lock because we're the
613 * only one with a reference to the inode.
615 truncate = ((ip->i_d.di_nlink == 0) &&
616 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
617 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
618 S_ISREG(ip->i_d.di_mode));
624 /* If this is a read-only mount, don't do this (would generate I/O) */
625 if (mp->m_flags & XFS_MOUNT_RDONLY)
628 if (ip->i_d.di_nlink != 0) {
629 if ((S_ISREG(ip->i_d.di_mode) &&
630 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
631 ip->i_delayed_blks > 0)) &&
632 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
633 (!(ip->i_d.di_flags &
634 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
635 (ip->i_delayed_blks != 0)))) {
636 error = xfs_free_eofblocks(mp, ip, 0);
638 return VN_INACTIVE_CACHE;
643 ASSERT(ip->i_d.di_nlink == 0);
645 error = xfs_qm_dqattach(ip, 0);
647 return VN_INACTIVE_CACHE;
649 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
651 xfs_ilock(ip, XFS_IOLOCK_EXCL);
653 error = xfs_trans_reserve(tp, 0,
654 XFS_ITRUNCATE_LOG_RES(mp),
655 0, XFS_TRANS_PERM_LOG_RES,
656 XFS_ITRUNCATE_LOG_COUNT);
658 /* Don't call itruncate_cleanup */
659 ASSERT(XFS_FORCED_SHUTDOWN(mp));
660 xfs_trans_cancel(tp, 0);
661 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
662 return VN_INACTIVE_CACHE;
665 xfs_ilock(ip, XFS_ILOCK_EXCL);
666 xfs_trans_ijoin(tp, ip);
668 error = xfs_itruncate_data(&tp, ip, 0);
671 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
672 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
673 return VN_INACTIVE_CACHE;
675 } else if (S_ISLNK(ip->i_d.di_mode)) {
678 * If we get an error while cleaning up a
679 * symlink we bail out.
681 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
682 xfs_inactive_symlink_rmt(ip, &tp) :
683 xfs_inactive_symlink_local(ip, &tp);
687 return VN_INACTIVE_CACHE;
690 xfs_trans_ijoin(tp, ip);
692 error = xfs_trans_reserve(tp, 0,
693 XFS_IFREE_LOG_RES(mp),
694 0, XFS_TRANS_PERM_LOG_RES,
695 XFS_INACTIVE_LOG_COUNT);
697 ASSERT(XFS_FORCED_SHUTDOWN(mp));
698 xfs_trans_cancel(tp, 0);
699 return VN_INACTIVE_CACHE;
702 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
703 xfs_trans_ijoin(tp, ip);
707 * If there are attributes associated with the file
708 * then blow them away now. The code calls a routine
709 * that recursively deconstructs the attribute fork.
710 * We need to just commit the current transaction
711 * because we can't use it for xfs_attr_inactive().
713 if (ip->i_d.di_anextents > 0) {
714 error = xfs_inactive_attrs(ip, &tp);
716 * If we got an error, the transaction is already
717 * cancelled, and the inode is unlocked. Just get out.
720 return VN_INACTIVE_CACHE;
721 } else if (ip->i_afp) {
722 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
728 xfs_bmap_init(&free_list, &first_block);
729 error = xfs_ifree(tp, ip, &free_list);
732 * If we fail to free the inode, shut down. The cancel
733 * might do that, we need to make sure. Otherwise the
734 * inode might be lost for a long time or forever.
736 if (!XFS_FORCED_SHUTDOWN(mp)) {
737 xfs_notice(mp, "%s: xfs_ifree returned error %d",
739 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
741 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
744 * Credit the quota account(s). The inode is gone.
746 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
749 * Just ignore errors at this point. There is nothing we can
750 * do except to try to keep going. Make sure it's not a silent
753 error = xfs_bmap_finish(&tp, &free_list, &committed);
755 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
757 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
759 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
764 * Release the dquots held by inode, if any.
767 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
770 return VN_INACTIVE_CACHE;
774 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
775 * is allowed, otherwise it has to be an exact match. If a CI match is found,
776 * ci_name->name will point to a the actual name (caller must free) or
777 * will be set to NULL if an exact match is found.
782 struct xfs_name *name,
784 struct xfs_name *ci_name)
790 trace_xfs_lookup(dp, name);
792 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
793 return XFS_ERROR(EIO);
795 lock_mode = xfs_ilock_map_shared(dp);
796 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
797 xfs_iunlock_map_shared(dp, lock_mode);
802 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
810 kmem_free(ci_name->name);
819 struct xfs_name *name,
824 int is_dir = S_ISDIR(mode);
825 struct xfs_mount *mp = dp->i_mount;
826 struct xfs_inode *ip = NULL;
827 struct xfs_trans *tp = NULL;
829 xfs_bmap_free_t free_list;
830 xfs_fsblock_t first_block;
831 boolean_t unlock_dp_on_error = B_FALSE;
835 struct xfs_dquot *udqp = NULL;
836 struct xfs_dquot *gdqp = NULL;
841 trace_xfs_create(dp, name);
843 if (XFS_FORCED_SHUTDOWN(mp))
844 return XFS_ERROR(EIO);
846 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
847 prid = xfs_get_projid(dp);
849 prid = XFS_PROJID_DEFAULT;
852 * Make sure that we have allocated dquot(s) on disk.
854 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
855 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
861 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
862 log_res = XFS_MKDIR_LOG_RES(mp);
863 log_count = XFS_MKDIR_LOG_COUNT;
864 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
866 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
867 log_res = XFS_CREATE_LOG_RES(mp);
868 log_count = XFS_CREATE_LOG_COUNT;
869 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
872 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
875 * Initially assume that the file does not exist and
876 * reserve the resources for that case. If that is not
877 * the case we'll drop the one we have and get a more
878 * appropriate transaction later.
880 error = xfs_trans_reserve(tp, resblks, log_res, 0,
881 XFS_TRANS_PERM_LOG_RES, log_count);
882 if (error == ENOSPC) {
883 /* flush outstanding delalloc blocks and retry */
884 xfs_flush_inodes(dp);
885 error = xfs_trans_reserve(tp, resblks, log_res, 0,
886 XFS_TRANS_PERM_LOG_RES, log_count);
888 if (error == ENOSPC) {
889 /* No space at all so try a "no-allocation" reservation */
891 error = xfs_trans_reserve(tp, 0, log_res, 0,
892 XFS_TRANS_PERM_LOG_RES, log_count);
896 goto out_trans_cancel;
899 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
900 unlock_dp_on_error = B_TRUE;
903 * Check for directory link count overflow.
905 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
906 error = XFS_ERROR(EMLINK);
907 goto out_trans_cancel;
910 xfs_bmap_init(&free_list, &first_block);
913 * Reserve disk quota and the inode.
915 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
917 goto out_trans_cancel;
919 error = xfs_dir_canenter(tp, dp, name, resblks);
921 goto out_trans_cancel;
924 * A newly created regular or special file just has one directory
925 * entry pointing to them, but a directory also the "." entry
926 * pointing to itself.
928 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
929 prid, resblks > 0, &ip, &committed);
932 goto out_trans_cancel;
933 goto out_trans_abort;
937 * Now we join the directory inode to the transaction. We do not do it
938 * earlier because xfs_dir_ialloc might commit the previous transaction
939 * (and release all the locks). An error from here on will result in
940 * the transaction cancel unlocking dp so don't do it explicitly in the
943 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
944 unlock_dp_on_error = B_FALSE;
946 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
947 &first_block, &free_list, resblks ?
948 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
950 ASSERT(error != ENOSPC);
951 goto out_trans_abort;
953 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
954 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
957 error = xfs_dir_init(tp, ip, dp);
959 goto out_bmap_cancel;
961 error = xfs_bumplink(tp, dp);
963 goto out_bmap_cancel;
967 * If this is a synchronous mount, make sure that the
968 * create transaction goes to disk before returning to
971 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
972 xfs_trans_set_sync(tp);
975 * Attach the dquot(s) to the inodes and modify them incore.
976 * These ids of the inode couldn't have changed since the new
977 * inode has been locked ever since it was created.
979 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
981 error = xfs_bmap_finish(&tp, &free_list, &committed);
983 goto out_bmap_cancel;
985 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
987 goto out_release_inode;
996 xfs_bmap_cancel(&free_list);
998 cancel_flags |= XFS_TRANS_ABORT;
1000 xfs_trans_cancel(tp, cancel_flags);
1003 * Wait until after the current transaction is aborted to
1004 * release the inode. This prevents recursive transactions
1005 * and deadlocks from xfs_inactive.
1010 xfs_qm_dqrele(udqp);
1011 xfs_qm_dqrele(gdqp);
1013 if (unlock_dp_on_error)
1014 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1020 int xfs_small_retries;
1021 int xfs_middle_retries;
1022 int xfs_lots_retries;
1023 int xfs_lock_delays;
1027 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1031 xfs_lock_inumorder(int lock_mode, int subclass)
1033 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1034 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1035 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1036 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1042 * The following routine will lock n inodes in exclusive mode.
1043 * We assume the caller calls us with the inodes in i_ino order.
1045 * We need to detect deadlock where an inode that we lock
1046 * is in the AIL and we start waiting for another inode that is locked
1047 * by a thread in a long running transaction (such as truncate). This can
1048 * result in deadlock since the long running trans might need to wait
1049 * for the inode we just locked in order to push the tail and free space
1058 int attempts = 0, i, j, try_lock;
1061 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1067 for (; i < inodes; i++) {
1070 if (i && (ips[i] == ips[i-1])) /* Already locked */
1074 * If try_lock is not set yet, make sure all locked inodes
1075 * are not in the AIL.
1076 * If any are, set try_lock to be used later.
1080 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1081 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1082 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1089 * If any of the previous locks we have locked is in the AIL,
1090 * we must TRY to get the second and subsequent locks. If
1091 * we can't get any, we must release all we have
1096 /* try_lock must be 0 if i is 0. */
1098 * try_lock means we have an inode locked
1099 * that is in the AIL.
1102 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1106 * Unlock all previous guys and try again.
1107 * xfs_iunlock will try to push the tail
1108 * if the inode is in the AIL.
1111 for(j = i - 1; j >= 0; j--) {
1114 * Check to see if we've already
1115 * unlocked this one.
1116 * Not the first one going back,
1117 * and the inode ptr is the same.
1119 if ((j != (i - 1)) && ips[j] ==
1123 xfs_iunlock(ips[j], lock_mode);
1126 if ((attempts % 5) == 0) {
1127 delay(1); /* Don't just spin the CPU */
1137 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1143 if (attempts < 5) xfs_small_retries++;
1144 else if (attempts < 100) xfs_middle_retries++;
1145 else xfs_lots_retries++;
1153 * xfs_lock_two_inodes() can only be used to lock one type of lock
1154 * at a time - the iolock or the ilock, but not both at once. If
1155 * we lock both at once, lockdep will report false positives saying
1156 * we have violated locking orders.
1159 xfs_lock_two_inodes(
1168 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1169 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1170 ASSERT(ip0->i_ino != ip1->i_ino);
1172 if (ip0->i_ino > ip1->i_ino) {
1179 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1182 * If the first lock we have locked is in the AIL, we must TRY to get
1183 * the second lock. If we can't get it, we must release the first one
1186 lp = (xfs_log_item_t *)ip0->i_itemp;
1187 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1188 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1189 xfs_iunlock(ip0, lock_mode);
1190 if ((++attempts % 5) == 0)
1191 delay(1); /* Don't just spin the CPU */
1195 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1202 struct xfs_name *name,
1205 xfs_mount_t *mp = dp->i_mount;
1206 xfs_trans_t *tp = NULL;
1207 int is_dir = S_ISDIR(ip->i_d.di_mode);
1209 xfs_bmap_free_t free_list;
1210 xfs_fsblock_t first_block;
1217 trace_xfs_remove(dp, name);
1219 if (XFS_FORCED_SHUTDOWN(mp))
1220 return XFS_ERROR(EIO);
1222 error = xfs_qm_dqattach(dp, 0);
1226 error = xfs_qm_dqattach(ip, 0);
1231 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1232 log_count = XFS_DEFAULT_LOG_COUNT;
1234 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1235 log_count = XFS_REMOVE_LOG_COUNT;
1237 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1240 * We try to get the real space reservation first,
1241 * allowing for directory btree deletion(s) implying
1242 * possible bmap insert(s). If we can't get the space
1243 * reservation then we use 0 instead, and avoid the bmap
1244 * btree insert(s) in the directory code by, if the bmap
1245 * insert tries to happen, instead trimming the LAST
1246 * block from the directory.
1248 resblks = XFS_REMOVE_SPACE_RES(mp);
1249 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1250 XFS_TRANS_PERM_LOG_RES, log_count);
1251 if (error == ENOSPC) {
1253 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1254 XFS_TRANS_PERM_LOG_RES, log_count);
1257 ASSERT(error != ENOSPC);
1259 goto out_trans_cancel;
1262 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1264 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
1265 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1268 * If we're removing a directory perform some additional validation.
1271 ASSERT(ip->i_d.di_nlink >= 2);
1272 if (ip->i_d.di_nlink != 2) {
1273 error = XFS_ERROR(ENOTEMPTY);
1274 goto out_trans_cancel;
1276 if (!xfs_dir_isempty(ip)) {
1277 error = XFS_ERROR(ENOTEMPTY);
1278 goto out_trans_cancel;
1282 xfs_bmap_init(&free_list, &first_block);
1283 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1284 &first_block, &free_list, resblks);
1286 ASSERT(error != ENOENT);
1287 goto out_bmap_cancel;
1289 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1293 * Drop the link from ip's "..".
1295 error = xfs_droplink(tp, dp);
1297 goto out_bmap_cancel;
1300 * Drop the "." link from ip to self.
1302 error = xfs_droplink(tp, ip);
1304 goto out_bmap_cancel;
1307 * When removing a non-directory we need to log the parent
1308 * inode here. For a directory this is done implicitly
1309 * by the xfs_droplink call for the ".." entry.
1311 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1315 * Drop the link from dp to ip.
1317 error = xfs_droplink(tp, ip);
1319 goto out_bmap_cancel;
1322 * Determine if this is the last link while
1323 * we are in the transaction.
1325 link_zero = (ip->i_d.di_nlink == 0);
1328 * If this is a synchronous mount, make sure that the
1329 * remove transaction goes to disk before returning to
1332 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1333 xfs_trans_set_sync(tp);
1335 error = xfs_bmap_finish(&tp, &free_list, &committed);
1337 goto out_bmap_cancel;
1339 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1344 * If we are using filestreams, kill the stream association.
1345 * If the file is still open it may get a new one but that
1346 * will get killed on last close in xfs_close() so we don't
1347 * have to worry about that.
1349 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1350 xfs_filestream_deassociate(ip);
1355 xfs_bmap_cancel(&free_list);
1356 cancel_flags |= XFS_TRANS_ABORT;
1358 xfs_trans_cancel(tp, cancel_flags);
1367 struct xfs_name *target_name)
1369 xfs_mount_t *mp = tdp->i_mount;
1372 xfs_bmap_free_t free_list;
1373 xfs_fsblock_t first_block;
1378 trace_xfs_link(tdp, target_name);
1380 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1382 if (XFS_FORCED_SHUTDOWN(mp))
1383 return XFS_ERROR(EIO);
1385 error = xfs_qm_dqattach(sip, 0);
1389 error = xfs_qm_dqattach(tdp, 0);
1393 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1394 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1395 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1396 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1397 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1398 if (error == ENOSPC) {
1400 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1401 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1408 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1410 xfs_trans_ijoin_ref(tp, sip, XFS_ILOCK_EXCL);
1411 xfs_trans_ijoin_ref(tp, tdp, XFS_ILOCK_EXCL);
1414 * If the source has too many links, we can't make any more to it.
1416 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
1417 error = XFS_ERROR(EMLINK);
1422 * If we are using project inheritance, we only allow hard link
1423 * creation in our tree when the project IDs are the same; else
1424 * the tree quota mechanism could be circumvented.
1426 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1427 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1428 error = XFS_ERROR(EXDEV);
1432 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1436 xfs_bmap_init(&free_list, &first_block);
1438 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1439 &first_block, &free_list, resblks);
1442 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1443 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1445 error = xfs_bumplink(tp, sip);
1450 * If this is a synchronous mount, make sure that the
1451 * link transaction goes to disk before returning to
1454 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1455 xfs_trans_set_sync(tp);
1458 error = xfs_bmap_finish (&tp, &free_list, &committed);
1460 xfs_bmap_cancel(&free_list);
1464 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1467 cancel_flags |= XFS_TRANS_ABORT;
1469 xfs_trans_cancel(tp, cancel_flags);
1477 struct xfs_name *link_name,
1478 const char *target_path,
1482 xfs_mount_t *mp = dp->i_mount;
1487 xfs_bmap_free_t free_list;
1488 xfs_fsblock_t first_block;
1489 boolean_t unlock_dp_on_error = B_FALSE;
1492 xfs_fileoff_t first_fsb;
1493 xfs_filblks_t fs_blocks;
1495 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1497 const char *cur_chunk;
1502 struct xfs_dquot *udqp, *gdqp;
1510 trace_xfs_symlink(dp, link_name);
1512 if (XFS_FORCED_SHUTDOWN(mp))
1513 return XFS_ERROR(EIO);
1516 * Check component lengths of the target path name.
1518 pathlen = strlen(target_path);
1519 if (pathlen >= MAXPATHLEN) /* total string too long */
1520 return XFS_ERROR(ENAMETOOLONG);
1523 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1524 prid = xfs_get_projid(dp);
1526 prid = XFS_PROJID_DEFAULT;
1529 * Make sure that we have allocated dquot(s) on disk.
1531 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1532 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1536 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
1537 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1539 * The symlink will fit into the inode data fork?
1540 * There can't be any attributes so we get the whole variable part.
1542 if (pathlen <= XFS_LITINO(mp))
1545 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
1546 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
1547 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
1548 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1549 if (error == ENOSPC && fs_blocks == 0) {
1551 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
1552 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1559 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1560 unlock_dp_on_error = B_TRUE;
1563 * Check whether the directory allows new symlinks or not.
1565 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
1566 error = XFS_ERROR(EPERM);
1571 * Reserve disk quota : blocks and inode.
1573 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
1578 * Check for ability to enter directory entry, if no space reserved.
1580 error = xfs_dir_canenter(tp, dp, link_name, resblks);
1584 * Initialize the bmap freelist prior to calling either
1585 * bmapi or the directory create code.
1587 xfs_bmap_init(&free_list, &first_block);
1590 * Allocate an inode for the symlink.
1592 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
1593 prid, resblks > 0, &ip, NULL);
1595 if (error == ENOSPC)
1601 * An error after we've joined dp to the transaction will result in the
1602 * transaction cancel unlocking dp so don't do it explicitly in the
1605 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
1606 unlock_dp_on_error = B_FALSE;
1609 * Also attach the dquot(s) to it, if applicable.
1611 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
1614 resblks -= XFS_IALLOC_SPACE_RES(mp);
1616 * If the symlink will fit into the inode, write it inline.
1618 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
1619 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
1620 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
1621 ip->i_d.di_size = pathlen;
1624 * The inode was initially created in extent format.
1626 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
1627 ip->i_df.if_flags |= XFS_IFINLINE;
1629 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
1630 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
1634 nmaps = SYMLINK_MAPS;
1636 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks,
1637 XFS_BMAPI_METADATA, &first_block, resblks,
1638 mval, &nmaps, &free_list);
1643 resblks -= fs_blocks;
1644 ip->i_d.di_size = pathlen;
1645 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1647 cur_chunk = target_path;
1648 for (n = 0; n < nmaps; n++) {
1649 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1650 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1651 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
1652 BTOBB(byte_cnt), 0);
1657 if (pathlen < byte_cnt) {
1660 pathlen -= byte_cnt;
1662 memcpy(bp->b_addr, cur_chunk, byte_cnt);
1663 cur_chunk += byte_cnt;
1665 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
1670 * Create the directory entry for the symlink.
1672 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
1673 &first_block, &free_list, resblks);
1676 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1677 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1680 * If this is a synchronous mount, make sure that the
1681 * symlink transaction goes to disk before returning to
1684 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1685 xfs_trans_set_sync(tp);
1688 error = xfs_bmap_finish(&tp, &free_list, &committed);
1692 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1693 xfs_qm_dqrele(udqp);
1694 xfs_qm_dqrele(gdqp);
1702 xfs_bmap_cancel(&free_list);
1703 cancel_flags |= XFS_TRANS_ABORT;
1705 xfs_trans_cancel(tp, cancel_flags);
1706 xfs_qm_dqrele(udqp);
1707 xfs_qm_dqrele(gdqp);
1709 if (unlock_dp_on_error)
1710 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1721 xfs_mount_t *mp = ip->i_mount;
1725 if (!capable(CAP_SYS_ADMIN))
1726 return XFS_ERROR(EPERM);
1728 if (XFS_FORCED_SHUTDOWN(mp))
1729 return XFS_ERROR(EIO);
1731 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
1732 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
1734 xfs_trans_cancel(tp, 0);
1737 xfs_ilock(ip, XFS_ILOCK_EXCL);
1738 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1740 ip->i_d.di_dmevmask = evmask;
1741 ip->i_d.di_dmstate = state;
1743 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1744 error = xfs_trans_commit(tp, 0);
1750 * xfs_alloc_file_space()
1751 * This routine allocates disk space for the given file.
1753 * If alloc_type == 0, this request is for an ALLOCSP type
1754 * request which will change the file size. In this case, no
1755 * DMAPI event will be generated by the call. A TRUNCATE event
1756 * will be generated later by xfs_setattr.
1758 * If alloc_type != 0, this request is for a RESVSP type
1759 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1760 * lower block boundary byte address is less than the file's
1769 xfs_alloc_file_space(
1776 xfs_mount_t *mp = ip->i_mount;
1778 xfs_filblks_t allocated_fsb;
1779 xfs_filblks_t allocatesize_fsb;
1780 xfs_extlen_t extsz, temp;
1781 xfs_fileoff_t startoffset_fsb;
1782 xfs_fsblock_t firstfsb;
1787 xfs_bmbt_irec_t imaps[1], *imapp;
1788 xfs_bmap_free_t free_list;
1789 uint qblocks, resblks, resrtextents;
1793 trace_xfs_alloc_file_space(ip);
1795 if (XFS_FORCED_SHUTDOWN(mp))
1796 return XFS_ERROR(EIO);
1798 error = xfs_qm_dqattach(ip, 0);
1803 return XFS_ERROR(EINVAL);
1805 rt = XFS_IS_REALTIME_INODE(ip);
1806 extsz = xfs_get_extsz_hint(ip);
1811 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1812 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1815 * Allocate file space until done or until there is an error
1817 while (allocatesize_fsb && !error) {
1821 * Determine space reservations for data/realtime.
1823 if (unlikely(extsz)) {
1824 s = startoffset_fsb;
1827 e = startoffset_fsb + allocatesize_fsb;
1828 if ((temp = do_mod(startoffset_fsb, extsz)))
1830 if ((temp = do_mod(e, extsz)))
1834 e = allocatesize_fsb;
1838 * The transaction reservation is limited to a 32-bit block
1839 * count, hence we need to limit the number of blocks we are
1840 * trying to reserve to avoid an overflow. We can't allocate
1841 * more than @nimaps extents, and an extent is limited on disk
1842 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1844 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1846 resrtextents = qblocks = resblks;
1847 resrtextents /= mp->m_sb.sb_rextsize;
1848 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1849 quota_flag = XFS_QMOPT_RES_RTBLKS;
1852 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1853 quota_flag = XFS_QMOPT_RES_REGBLKS;
1857 * Allocate and setup the transaction.
1859 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1860 error = xfs_trans_reserve(tp, resblks,
1861 XFS_WRITE_LOG_RES(mp), resrtextents,
1862 XFS_TRANS_PERM_LOG_RES,
1863 XFS_WRITE_LOG_COUNT);
1865 * Check for running out of space
1869 * Free the transaction structure.
1871 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1872 xfs_trans_cancel(tp, 0);
1875 xfs_ilock(ip, XFS_ILOCK_EXCL);
1876 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1881 xfs_trans_ijoin(tp, ip);
1883 xfs_bmap_init(&free_list, &firstfsb);
1884 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1885 allocatesize_fsb, alloc_type, &firstfsb,
1886 0, imapp, &nimaps, &free_list);
1892 * Complete the transaction
1894 error = xfs_bmap_finish(&tp, &free_list, &committed);
1899 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1900 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1905 allocated_fsb = imapp->br_blockcount;
1908 error = XFS_ERROR(ENOSPC);
1912 startoffset_fsb += allocated_fsb;
1913 allocatesize_fsb -= allocated_fsb;
1918 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1919 xfs_bmap_cancel(&free_list);
1920 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
1922 error1: /* Just cancel transaction */
1923 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1924 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1929 * Zero file bytes between startoff and endoff inclusive.
1930 * The iolock is held exclusive and no blocks are buffered.
1932 * This function is used by xfs_free_file_space() to zero
1933 * partial blocks when the range to free is not block aligned.
1934 * When unreserving space with boundaries that are not block
1935 * aligned we round up the start and round down the end
1936 * boundaries and then use this function to zero the parts of
1937 * the blocks that got dropped during the rounding.
1940 xfs_zero_remaining_bytes(
1945 xfs_bmbt_irec_t imap;
1946 xfs_fileoff_t offset_fsb;
1947 xfs_off_t lastoffset;
1950 xfs_mount_t *mp = ip->i_mount;
1955 * Avoid doing I/O beyond eof - it's not necessary
1956 * since nothing can read beyond eof. The space will
1957 * be zeroed when the file is extended anyway.
1959 if (startoff >= ip->i_size)
1962 if (endoff > ip->i_size)
1963 endoff = ip->i_size;
1965 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1966 mp->m_rtdev_targp : mp->m_ddev_targp,
1967 mp->m_sb.sb_blocksize, XBF_DONT_BLOCK);
1969 return XFS_ERROR(ENOMEM);
1973 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1974 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1976 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1977 if (error || nimap < 1)
1979 ASSERT(imap.br_blockcount >= 1);
1980 ASSERT(imap.br_startoff == offset_fsb);
1981 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1982 if (lastoffset > endoff)
1983 lastoffset = endoff;
1984 if (imap.br_startblock == HOLESTARTBLOCK)
1986 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1987 if (imap.br_state == XFS_EXT_UNWRITTEN)
1990 XFS_BUF_UNWRITE(bp);
1992 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1994 error = xfs_buf_iowait(bp);
1996 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
1997 mp, bp, XFS_BUF_ADDR(bp));
2001 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2002 0, lastoffset - offset + 1);
2007 error = xfs_buf_iowait(bp);
2009 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2010 mp, bp, XFS_BUF_ADDR(bp));
2019 * xfs_free_file_space()
2020 * This routine frees disk space for the given file.
2022 * This routine is only called by xfs_change_file_space
2023 * for an UNRESVSP type call.
2031 xfs_free_file_space(
2039 xfs_fileoff_t endoffset_fsb;
2041 xfs_fsblock_t firstfsb;
2042 xfs_bmap_free_t free_list;
2043 xfs_bmbt_irec_t imap;
2051 xfs_fileoff_t startoffset_fsb;
2053 int need_iolock = 1;
2057 trace_xfs_free_file_space(ip);
2059 error = xfs_qm_dqattach(ip, 0);
2064 if (len <= 0) /* if nothing being freed */
2066 rt = XFS_IS_REALTIME_INODE(ip);
2067 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
2068 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
2070 if (attr_flags & XFS_ATTR_NOLOCK)
2073 xfs_ilock(ip, XFS_IOLOCK_EXCL);
2074 /* wait for the completion of any pending DIOs */
2075 inode_dio_wait(VFS_I(ip));
2078 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
2079 ioffset = offset & ~(rounding - 1);
2081 if (VN_CACHED(VFS_I(ip)) != 0) {
2082 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
2084 goto out_unlock_iolock;
2088 * Need to zero the stuff we're not freeing, on disk.
2089 * If it's a realtime file & can't use unwritten extents then we
2090 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2091 * will take care of it for us.
2093 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
2095 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
2098 goto out_unlock_iolock;
2099 ASSERT(nimap == 0 || nimap == 1);
2100 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2103 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2104 block = imap.br_startblock;
2105 mod = do_div(block, mp->m_sb.sb_rextsize);
2107 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2110 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
2113 goto out_unlock_iolock;
2114 ASSERT(nimap == 0 || nimap == 1);
2115 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2116 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2118 if (mod && (mod != mp->m_sb.sb_rextsize))
2119 endoffset_fsb -= mod;
2122 if ((done = (endoffset_fsb <= startoffset_fsb)))
2124 * One contiguous piece to clear
2126 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2129 * Some full blocks, possibly two pieces to clear
2131 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2132 error = xfs_zero_remaining_bytes(ip, offset,
2133 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2135 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2136 error = xfs_zero_remaining_bytes(ip,
2137 XFS_FSB_TO_B(mp, endoffset_fsb),
2142 * free file space until done or until there is an error
2144 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2145 while (!error && !done) {
2148 * allocate and setup the transaction. Allow this
2149 * transaction to dip into the reserve blocks to ensure
2150 * the freeing of the space succeeds at ENOSPC.
2152 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2153 tp->t_flags |= XFS_TRANS_RESERVE;
2154 error = xfs_trans_reserve(tp,
2156 XFS_WRITE_LOG_RES(mp),
2158 XFS_TRANS_PERM_LOG_RES,
2159 XFS_WRITE_LOG_COUNT);
2162 * check for running out of space
2166 * Free the transaction structure.
2168 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2169 xfs_trans_cancel(tp, 0);
2172 xfs_ilock(ip, XFS_ILOCK_EXCL);
2173 error = xfs_trans_reserve_quota(tp, mp,
2174 ip->i_udquot, ip->i_gdquot,
2175 resblks, 0, XFS_QMOPT_RES_REGBLKS);
2179 xfs_trans_ijoin(tp, ip);
2182 * issue the bunmapi() call to free the blocks
2184 xfs_bmap_init(&free_list, &firstfsb);
2185 error = xfs_bunmapi(tp, ip, startoffset_fsb,
2186 endoffset_fsb - startoffset_fsb,
2187 0, 2, &firstfsb, &free_list, &done);
2193 * complete the transaction
2195 error = xfs_bmap_finish(&tp, &free_list, &committed);
2200 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2201 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2206 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2210 xfs_bmap_cancel(&free_list);
2212 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2213 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2219 * xfs_change_file_space()
2220 * This routine allocates or frees disk space for the given file.
2221 * The user specified parameters are checked for alignment and size
2230 xfs_change_file_space(
2237 xfs_mount_t *mp = ip->i_mount;
2242 xfs_off_t startoffset;
2248 if (!S_ISREG(ip->i_d.di_mode))
2249 return XFS_ERROR(EINVAL);
2251 switch (bf->l_whence) {
2252 case 0: /*SEEK_SET*/
2254 case 1: /*SEEK_CUR*/
2255 bf->l_start += offset;
2257 case 2: /*SEEK_END*/
2258 bf->l_start += ip->i_size;
2261 return XFS_ERROR(EINVAL);
2264 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
2266 if ( (bf->l_start < 0)
2267 || (bf->l_start > XFS_MAXIOFFSET(mp))
2268 || (bf->l_start + llen < 0)
2269 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
2270 return XFS_ERROR(EINVAL);
2274 startoffset = bf->l_start;
2278 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2280 * These calls do NOT zero the data space allocated to the file,
2281 * nor do they change the file size.
2283 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2285 * These calls cause the new file data to be zeroed and the file
2286 * size to be changed.
2288 setprealloc = clrprealloc = 0;
2289 prealloc_type = XFS_BMAPI_PREALLOC;
2292 case XFS_IOC_ZERO_RANGE:
2293 prealloc_type |= XFS_BMAPI_CONVERT;
2294 xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0);
2296 case XFS_IOC_RESVSP:
2297 case XFS_IOC_RESVSP64:
2298 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
2299 prealloc_type, attr_flags);
2305 case XFS_IOC_UNRESVSP:
2306 case XFS_IOC_UNRESVSP64:
2307 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
2312 case XFS_IOC_ALLOCSP:
2313 case XFS_IOC_ALLOCSP64:
2314 case XFS_IOC_FREESP:
2315 case XFS_IOC_FREESP64:
2316 if (startoffset > fsize) {
2317 error = xfs_alloc_file_space(ip, fsize,
2318 startoffset - fsize, 0, attr_flags);
2323 iattr.ia_valid = ATTR_SIZE;
2324 iattr.ia_size = startoffset;
2326 error = xfs_setattr_size(ip, &iattr, attr_flags);
2336 return XFS_ERROR(EINVAL);
2340 * update the inode timestamp, mode, and prealloc flag bits
2342 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2344 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2347 xfs_trans_cancel(tp, 0);
2351 xfs_ilock(ip, XFS_ILOCK_EXCL);
2352 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
2354 if ((attr_flags & XFS_ATTR_DMI) == 0) {
2355 ip->i_d.di_mode &= ~S_ISUID;
2358 * Note that we don't have to worry about mandatory
2359 * file locking being disabled here because we only
2360 * clear the S_ISGID bit if the Group execute bit is
2361 * on, but if it was on then mandatory locking wouldn't
2362 * have been enabled.
2364 if (ip->i_d.di_mode & S_IXGRP)
2365 ip->i_d.di_mode &= ~S_ISGID;
2367 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2370 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2371 else if (clrprealloc)
2372 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2374 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2375 if (attr_flags & XFS_ATTR_SYNC)
2376 xfs_trans_set_sync(tp);
2377 return xfs_trans_commit(tp, 0);