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"
24 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_itable.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_alloc.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_utils.h"
44 #include "xfs_rtalloc.h"
45 #include "xfs_trans_space.h"
46 #include "xfs_log_priv.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
52 * The maximum pathlen is 1024 bytes. Since the minimum file system
53 * blocksize is 512 bytes, we can get a max of 2 extents back from
56 #define SYMLINK_MAPS 2
63 xfs_mount_t *mp = ip->i_mount;
64 int pathlen = ip->i_d.di_size;
65 int nmaps = SYMLINK_MAPS;
66 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
73 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, pathlen), mval, &nmaps,
78 for (n = 0; n < nmaps; n++) {
79 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
80 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
82 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt),
85 return XFS_ERROR(ENOMEM);
88 xfs_buf_ioerror_alert(bp, __func__);
92 if (pathlen < byte_cnt)
96 memcpy(link, bp->b_addr, byte_cnt);
100 link[ip->i_d.di_size] = '\0';
112 xfs_mount_t *mp = ip->i_mount;
116 trace_xfs_readlink(ip);
118 if (XFS_FORCED_SHUTDOWN(mp))
119 return XFS_ERROR(EIO);
121 xfs_ilock(ip, XFS_ILOCK_SHARED);
123 pathlen = ip->i_d.di_size;
127 if (pathlen < 0 || pathlen > MAXPATHLEN) {
128 xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)",
129 __func__, (unsigned long long) ip->i_ino,
130 (long long) pathlen);
132 error = XFS_ERROR(EFSCORRUPTED);
137 if (ip->i_df.if_flags & XFS_IFINLINE) {
138 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
139 link[pathlen] = '\0';
141 error = xfs_readlink_bmap(ip, link);
145 xfs_iunlock(ip, XFS_ILOCK_SHARED);
150 * Flags for xfs_free_eofblocks
152 #define XFS_FREE_EOF_TRYLOCK (1<<0)
155 * This is called by xfs_inactive to free any blocks beyond eof
156 * when the link count isn't zero and by xfs_dm_punch_hole() when
157 * punching a hole to EOF.
167 xfs_fileoff_t end_fsb;
168 xfs_fileoff_t last_fsb;
169 xfs_filblks_t map_len;
171 xfs_bmbt_irec_t imap;
174 * Figure out if there are any blocks beyond the end
175 * of the file. If not, then there is nothing to do.
177 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
178 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
179 if (last_fsb <= end_fsb)
181 map_len = last_fsb - end_fsb;
184 xfs_ilock(ip, XFS_ILOCK_SHARED);
185 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
186 xfs_iunlock(ip, XFS_ILOCK_SHARED);
188 if (!error && (nimaps != 0) &&
189 (imap.br_startblock != HOLESTARTBLOCK ||
190 ip->i_delayed_blks)) {
192 * Attach the dquots to the inode up front.
194 error = xfs_qm_dqattach(ip, 0);
199 * There are blocks after the end of file.
200 * Free them up now by truncating the file to
203 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
205 if (flags & XFS_FREE_EOF_TRYLOCK) {
206 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
207 xfs_trans_cancel(tp, 0);
211 xfs_ilock(ip, XFS_IOLOCK_EXCL);
214 error = xfs_trans_reserve(tp, 0,
215 XFS_ITRUNCATE_LOG_RES(mp),
216 0, XFS_TRANS_PERM_LOG_RES,
217 XFS_ITRUNCATE_LOG_COUNT);
219 ASSERT(XFS_FORCED_SHUTDOWN(mp));
220 xfs_trans_cancel(tp, 0);
221 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
225 xfs_ilock(ip, XFS_ILOCK_EXCL);
226 xfs_trans_ijoin(tp, ip, 0);
229 * Do not update the on-disk file size. If we update the
230 * on-disk file size and then the system crashes before the
231 * contents of the file are flushed to disk then the files
232 * may be full of holes (ie NULL files bug).
234 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
238 * If we get an error at this point we simply don't
239 * bother truncating the file.
242 (XFS_TRANS_RELEASE_LOG_RES |
245 error = xfs_trans_commit(tp,
246 XFS_TRANS_RELEASE_LOG_RES);
248 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
254 * Free a symlink that has blocks associated with it.
257 xfs_inactive_symlink_rmt(
265 xfs_fsblock_t first_block;
266 xfs_bmap_free_t free_list;
269 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
277 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
279 * We're freeing a symlink that has some
280 * blocks allocated to it. Free the
281 * blocks here. We know that we've got
282 * either 1 or 2 extents and that we can
283 * free them all in one bunmapi call.
285 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
286 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
287 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
288 ASSERT(XFS_FORCED_SHUTDOWN(mp));
289 xfs_trans_cancel(tp, 0);
294 * Lock the inode, fix the size, and join it to the transaction.
295 * Hold it so in the normal path, we still have it locked for
296 * the second transaction. In the error paths we need it
297 * held so the cancel won't rele it, see below.
299 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
300 size = (int)ip->i_d.di_size;
302 xfs_trans_ijoin(tp, ip, 0);
303 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
305 * Find the block(s) so we can inval and unmap them.
308 xfs_bmap_init(&free_list, &first_block);
309 nmaps = ARRAY_SIZE(mval);
310 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, size),
315 * Invalidate the block(s).
317 for (i = 0; i < nmaps; i++) {
318 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
319 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
320 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
325 xfs_trans_binval(tp, bp);
328 * Unmap the dead block(s) to the free_list.
330 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
331 &first_block, &free_list, &done)))
335 * Commit the first transaction. This logs the EFI and the inode.
337 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
340 * The transaction must have been committed, since there were
341 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
342 * The new tp has the extent freeing and EFDs.
346 * The first xact was committed, so add the inode to the new one.
347 * Mark it dirty so it will be logged and moved forward in the log as
348 * part of every commit.
350 xfs_trans_ijoin(tp, ip, 0);
351 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
353 * Get a new, empty transaction to return to our caller.
355 ntp = xfs_trans_dup(tp);
357 * Commit the transaction containing extent freeing and EFDs.
358 * If we get an error on the commit here or on the reserve below,
359 * we need to unlock the inode since the new transaction doesn't
360 * have the inode attached.
362 error = xfs_trans_commit(tp, 0);
365 ASSERT(XFS_FORCED_SHUTDOWN(mp));
369 * transaction commit worked ok so we can drop the extra ticket
370 * reference that we gained in xfs_trans_dup()
372 xfs_log_ticket_put(tp->t_ticket);
375 * Remove the memory for extent descriptions (just bookkeeping).
377 if (ip->i_df.if_bytes)
378 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
379 ASSERT(ip->i_df.if_bytes == 0);
381 * Put an itruncate log reservation in the new transaction
384 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
385 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
386 ASSERT(XFS_FORCED_SHUTDOWN(mp));
390 * Return with the inode locked but not joined to the transaction.
396 xfs_bmap_cancel(&free_list);
399 * Have to come here with the inode locked and either
400 * (held and in the transaction) or (not in the transaction).
401 * If the inode isn't held then cancel would iput it, but
402 * that's wrong since this is inactive and the vnode ref
403 * count is 0 already.
404 * Cancel won't do anything to the inode if held, but it still
405 * needs to be locked until the cancel is done, if it was
406 * joined to the transaction.
408 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
409 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
416 xfs_inactive_symlink_local(
422 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
424 * We're freeing a symlink which fit into
425 * the inode. Just free the memory used
426 * to hold the old symlink.
428 error = xfs_trans_reserve(*tpp, 0,
429 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
430 0, XFS_TRANS_PERM_LOG_RES,
431 XFS_ITRUNCATE_LOG_COUNT);
434 xfs_trans_cancel(*tpp, 0);
438 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
441 * Zero length symlinks _can_ exist.
443 if (ip->i_df.if_bytes > 0) {
444 xfs_idata_realloc(ip,
445 -(ip->i_df.if_bytes),
447 ASSERT(ip->i_df.if_bytes == 0);
461 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
464 ASSERT(ip->i_d.di_forkoff != 0);
465 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
466 xfs_iunlock(ip, XFS_ILOCK_EXCL);
470 error = xfs_attr_inactive(ip);
474 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
475 error = xfs_trans_reserve(tp, 0,
476 XFS_IFREE_LOG_RES(mp),
477 0, XFS_TRANS_PERM_LOG_RES,
478 XFS_INACTIVE_LOG_COUNT);
482 xfs_ilock(ip, XFS_ILOCK_EXCL);
483 xfs_trans_ijoin(tp, ip, 0);
484 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
486 ASSERT(ip->i_d.di_anextents == 0);
492 ASSERT(XFS_FORCED_SHUTDOWN(mp));
493 xfs_trans_cancel(tp, 0);
496 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
504 xfs_mount_t *mp = ip->i_mount;
507 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
510 /* If this is a read-only mount, don't do this (would generate I/O) */
511 if (mp->m_flags & XFS_MOUNT_RDONLY)
514 if (!XFS_FORCED_SHUTDOWN(mp)) {
518 * If we are using filestreams, and we have an unlinked
519 * file that we are processing the last close on, then nothing
520 * will be able to reopen and write to this file. Purge this
521 * inode from the filestreams cache so that it doesn't delay
522 * teardown of the inode.
524 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
525 xfs_filestream_deassociate(ip);
528 * If we previously truncated this file and removed old data
529 * in the process, we want to initiate "early" writeout on
530 * the last close. This is an attempt to combat the notorious
531 * NULL files problem which is particularly noticeable from a
532 * truncate down, buffered (re-)write (delalloc), followed by
533 * a crash. What we are effectively doing here is
534 * significantly reducing the time window where we'd otherwise
535 * be exposed to that problem.
537 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
539 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
540 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
541 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
545 if (ip->i_d.di_nlink == 0)
548 if ((S_ISREG(ip->i_d.di_mode) &&
549 (VFS_I(ip)->i_size > 0 ||
550 (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) &&
551 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
552 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
555 * If we can't get the iolock just skip truncating the blocks
556 * past EOF because we could deadlock with the mmap_sem
557 * otherwise. We'll get another chance to drop them once the
558 * last reference to the inode is dropped, so we'll never leak
559 * blocks permanently.
561 * Further, check if the inode is being opened, written and
562 * closed frequently and we have delayed allocation blocks
563 * outstanding (e.g. streaming writes from the NFS server),
564 * truncating the blocks past EOF will cause fragmentation to
567 * In this case don't do the truncation, either, but we have to
568 * be careful how we detect this case. Blocks beyond EOF show
569 * up as i_delayed_blks even when the inode is clean, so we
570 * need to truncate them away first before checking for a dirty
571 * release. Hence on the first dirty close we will still remove
572 * the speculative allocation, but after that we will leave it
575 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
578 error = xfs_free_eofblocks(mp, ip,
579 XFS_FREE_EOF_TRYLOCK);
583 /* delalloc blocks after truncation means it really is dirty */
584 if (ip->i_delayed_blks)
585 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
593 * This is called when the vnode reference count for the vnode
594 * goes to zero. If the file has been unlinked, then it must
595 * now be truncated. Also, we clear all of the read-ahead state
596 * kept for the inode here since the file is now closed.
602 xfs_bmap_free_t free_list;
603 xfs_fsblock_t first_block;
611 * If the inode is already free, then there can be nothing
614 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
615 ASSERT(ip->i_df.if_real_bytes == 0);
616 ASSERT(ip->i_df.if_broot_bytes == 0);
617 return VN_INACTIVE_CACHE;
621 * Only do a truncate if it's a regular file with
622 * some actual space in it. It's OK to look at the
623 * inode's fields without the lock because we're the
624 * only one with a reference to the inode.
626 truncate = ((ip->i_d.di_nlink == 0) &&
627 ((ip->i_d.di_size != 0) || XFS_ISIZE(ip) != 0 ||
628 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
629 S_ISREG(ip->i_d.di_mode));
635 /* If this is a read-only mount, don't do this (would generate I/O) */
636 if (mp->m_flags & XFS_MOUNT_RDONLY)
639 if (ip->i_d.di_nlink != 0) {
640 if ((S_ISREG(ip->i_d.di_mode) &&
641 (VFS_I(ip)->i_size > 0 ||
642 (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) &&
643 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
644 (!(ip->i_d.di_flags &
645 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
646 ip->i_delayed_blks != 0))) {
647 error = xfs_free_eofblocks(mp, ip, 0);
649 return VN_INACTIVE_CACHE;
654 ASSERT(ip->i_d.di_nlink == 0);
656 error = xfs_qm_dqattach(ip, 0);
658 return VN_INACTIVE_CACHE;
660 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
662 xfs_ilock(ip, XFS_IOLOCK_EXCL);
664 error = xfs_trans_reserve(tp, 0,
665 XFS_ITRUNCATE_LOG_RES(mp),
666 0, XFS_TRANS_PERM_LOG_RES,
667 XFS_ITRUNCATE_LOG_COUNT);
669 /* Don't call itruncate_cleanup */
670 ASSERT(XFS_FORCED_SHUTDOWN(mp));
671 xfs_trans_cancel(tp, 0);
672 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
673 return VN_INACTIVE_CACHE;
676 xfs_ilock(ip, XFS_ILOCK_EXCL);
677 xfs_trans_ijoin(tp, ip, 0);
680 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
682 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
685 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
686 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
687 return VN_INACTIVE_CACHE;
690 ASSERT(ip->i_d.di_nextents == 0);
691 } else if (S_ISLNK(ip->i_d.di_mode)) {
694 * If we get an error while cleaning up a
695 * symlink we bail out.
697 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
698 xfs_inactive_symlink_rmt(ip, &tp) :
699 xfs_inactive_symlink_local(ip, &tp);
703 return VN_INACTIVE_CACHE;
706 xfs_trans_ijoin(tp, ip, 0);
708 error = xfs_trans_reserve(tp, 0,
709 XFS_IFREE_LOG_RES(mp),
710 0, XFS_TRANS_PERM_LOG_RES,
711 XFS_INACTIVE_LOG_COUNT);
713 ASSERT(XFS_FORCED_SHUTDOWN(mp));
714 xfs_trans_cancel(tp, 0);
715 return VN_INACTIVE_CACHE;
718 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
719 xfs_trans_ijoin(tp, ip, 0);
723 * If there are attributes associated with the file
724 * then blow them away now. The code calls a routine
725 * that recursively deconstructs the attribute fork.
726 * We need to just commit the current transaction
727 * because we can't use it for xfs_attr_inactive().
729 if (ip->i_d.di_anextents > 0) {
730 error = xfs_inactive_attrs(ip, &tp);
732 * If we got an error, the transaction is already
733 * cancelled, and the inode is unlocked. Just get out.
736 return VN_INACTIVE_CACHE;
737 } else if (ip->i_afp) {
738 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
744 xfs_bmap_init(&free_list, &first_block);
745 error = xfs_ifree(tp, ip, &free_list);
748 * If we fail to free the inode, shut down. The cancel
749 * might do that, we need to make sure. Otherwise the
750 * inode might be lost for a long time or forever.
752 if (!XFS_FORCED_SHUTDOWN(mp)) {
753 xfs_notice(mp, "%s: xfs_ifree returned error %d",
755 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
757 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
760 * Credit the quota account(s). The inode is gone.
762 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
765 * Just ignore errors at this point. There is nothing we can
766 * do except to try to keep going. Make sure it's not a silent
769 error = xfs_bmap_finish(&tp, &free_list, &committed);
771 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
773 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
775 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
780 * Release the dquots held by inode, if any.
783 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
786 return VN_INACTIVE_CACHE;
790 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
791 * is allowed, otherwise it has to be an exact match. If a CI match is found,
792 * ci_name->name will point to a the actual name (caller must free) or
793 * will be set to NULL if an exact match is found.
798 struct xfs_name *name,
800 struct xfs_name *ci_name)
806 trace_xfs_lookup(dp, name);
808 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
809 return XFS_ERROR(EIO);
811 lock_mode = xfs_ilock_map_shared(dp);
812 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
813 xfs_iunlock_map_shared(dp, lock_mode);
818 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
826 kmem_free(ci_name->name);
835 struct xfs_name *name,
840 int is_dir = S_ISDIR(mode);
841 struct xfs_mount *mp = dp->i_mount;
842 struct xfs_inode *ip = NULL;
843 struct xfs_trans *tp = NULL;
845 xfs_bmap_free_t free_list;
846 xfs_fsblock_t first_block;
847 boolean_t unlock_dp_on_error = B_FALSE;
851 struct xfs_dquot *udqp = NULL;
852 struct xfs_dquot *gdqp = NULL;
857 trace_xfs_create(dp, name);
859 if (XFS_FORCED_SHUTDOWN(mp))
860 return XFS_ERROR(EIO);
862 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
863 prid = xfs_get_projid(dp);
865 prid = XFS_PROJID_DEFAULT;
868 * Make sure that we have allocated dquot(s) on disk.
870 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
871 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
877 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
878 log_res = XFS_MKDIR_LOG_RES(mp);
879 log_count = XFS_MKDIR_LOG_COUNT;
880 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
882 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
883 log_res = XFS_CREATE_LOG_RES(mp);
884 log_count = XFS_CREATE_LOG_COUNT;
885 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
888 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
891 * Initially assume that the file does not exist and
892 * reserve the resources for that case. If that is not
893 * the case we'll drop the one we have and get a more
894 * appropriate transaction later.
896 error = xfs_trans_reserve(tp, resblks, log_res, 0,
897 XFS_TRANS_PERM_LOG_RES, log_count);
898 if (error == ENOSPC) {
899 /* flush outstanding delalloc blocks and retry */
900 xfs_flush_inodes(dp);
901 error = xfs_trans_reserve(tp, resblks, log_res, 0,
902 XFS_TRANS_PERM_LOG_RES, log_count);
904 if (error == ENOSPC) {
905 /* No space at all so try a "no-allocation" reservation */
907 error = xfs_trans_reserve(tp, 0, log_res, 0,
908 XFS_TRANS_PERM_LOG_RES, log_count);
912 goto out_trans_cancel;
915 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
916 unlock_dp_on_error = B_TRUE;
918 xfs_bmap_init(&free_list, &first_block);
921 * Reserve disk quota and the inode.
923 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
925 goto out_trans_cancel;
927 error = xfs_dir_canenter(tp, dp, name, resblks);
929 goto out_trans_cancel;
932 * A newly created regular or special file just has one directory
933 * entry pointing to them, but a directory also the "." entry
934 * pointing to itself.
936 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
937 prid, resblks > 0, &ip, &committed);
940 goto out_trans_cancel;
941 goto out_trans_abort;
945 * Now we join the directory inode to the transaction. We do not do it
946 * earlier because xfs_dir_ialloc might commit the previous transaction
947 * (and release all the locks). An error from here on will result in
948 * the transaction cancel unlocking dp so don't do it explicitly in the
951 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
952 unlock_dp_on_error = B_FALSE;
954 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
955 &first_block, &free_list, resblks ?
956 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
958 ASSERT(error != ENOSPC);
959 goto out_trans_abort;
961 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
962 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
965 error = xfs_dir_init(tp, ip, dp);
967 goto out_bmap_cancel;
969 error = xfs_bumplink(tp, dp);
971 goto out_bmap_cancel;
975 * If this is a synchronous mount, make sure that the
976 * create transaction goes to disk before returning to
979 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
980 xfs_trans_set_sync(tp);
983 * Attach the dquot(s) to the inodes and modify them incore.
984 * These ids of the inode couldn't have changed since the new
985 * inode has been locked ever since it was created.
987 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
989 error = xfs_bmap_finish(&tp, &free_list, &committed);
991 goto out_bmap_cancel;
993 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
995 goto out_release_inode;
1004 xfs_bmap_cancel(&free_list);
1006 cancel_flags |= XFS_TRANS_ABORT;
1008 xfs_trans_cancel(tp, cancel_flags);
1011 * Wait until after the current transaction is aborted to
1012 * release the inode. This prevents recursive transactions
1013 * and deadlocks from xfs_inactive.
1018 xfs_qm_dqrele(udqp);
1019 xfs_qm_dqrele(gdqp);
1021 if (unlock_dp_on_error)
1022 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1028 int xfs_small_retries;
1029 int xfs_middle_retries;
1030 int xfs_lots_retries;
1031 int xfs_lock_delays;
1035 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1039 xfs_lock_inumorder(int lock_mode, int subclass)
1041 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1042 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1043 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1044 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1050 * The following routine will lock n inodes in exclusive mode.
1051 * We assume the caller calls us with the inodes in i_ino order.
1053 * We need to detect deadlock where an inode that we lock
1054 * is in the AIL and we start waiting for another inode that is locked
1055 * by a thread in a long running transaction (such as truncate). This can
1056 * result in deadlock since the long running trans might need to wait
1057 * for the inode we just locked in order to push the tail and free space
1066 int attempts = 0, i, j, try_lock;
1069 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1075 for (; i < inodes; i++) {
1078 if (i && (ips[i] == ips[i-1])) /* Already locked */
1082 * If try_lock is not set yet, make sure all locked inodes
1083 * are not in the AIL.
1084 * If any are, set try_lock to be used later.
1088 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1089 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1090 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1097 * If any of the previous locks we have locked is in the AIL,
1098 * we must TRY to get the second and subsequent locks. If
1099 * we can't get any, we must release all we have
1104 /* try_lock must be 0 if i is 0. */
1106 * try_lock means we have an inode locked
1107 * that is in the AIL.
1110 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1114 * Unlock all previous guys and try again.
1115 * xfs_iunlock will try to push the tail
1116 * if the inode is in the AIL.
1119 for(j = i - 1; j >= 0; j--) {
1122 * Check to see if we've already
1123 * unlocked this one.
1124 * Not the first one going back,
1125 * and the inode ptr is the same.
1127 if ((j != (i - 1)) && ips[j] ==
1131 xfs_iunlock(ips[j], lock_mode);
1134 if ((attempts % 5) == 0) {
1135 delay(1); /* Don't just spin the CPU */
1145 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1151 if (attempts < 5) xfs_small_retries++;
1152 else if (attempts < 100) xfs_middle_retries++;
1153 else xfs_lots_retries++;
1161 * xfs_lock_two_inodes() can only be used to lock one type of lock
1162 * at a time - the iolock or the ilock, but not both at once. If
1163 * we lock both at once, lockdep will report false positives saying
1164 * we have violated locking orders.
1167 xfs_lock_two_inodes(
1176 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1177 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1178 ASSERT(ip0->i_ino != ip1->i_ino);
1180 if (ip0->i_ino > ip1->i_ino) {
1187 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1190 * If the first lock we have locked is in the AIL, we must TRY to get
1191 * the second lock. If we can't get it, we must release the first one
1194 lp = (xfs_log_item_t *)ip0->i_itemp;
1195 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1196 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1197 xfs_iunlock(ip0, lock_mode);
1198 if ((++attempts % 5) == 0)
1199 delay(1); /* Don't just spin the CPU */
1203 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1210 struct xfs_name *name,
1213 xfs_mount_t *mp = dp->i_mount;
1214 xfs_trans_t *tp = NULL;
1215 int is_dir = S_ISDIR(ip->i_d.di_mode);
1217 xfs_bmap_free_t free_list;
1218 xfs_fsblock_t first_block;
1225 trace_xfs_remove(dp, name);
1227 if (XFS_FORCED_SHUTDOWN(mp))
1228 return XFS_ERROR(EIO);
1230 error = xfs_qm_dqattach(dp, 0);
1234 error = xfs_qm_dqattach(ip, 0);
1239 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1240 log_count = XFS_DEFAULT_LOG_COUNT;
1242 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1243 log_count = XFS_REMOVE_LOG_COUNT;
1245 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1248 * We try to get the real space reservation first,
1249 * allowing for directory btree deletion(s) implying
1250 * possible bmap insert(s). If we can't get the space
1251 * reservation then we use 0 instead, and avoid the bmap
1252 * btree insert(s) in the directory code by, if the bmap
1253 * insert tries to happen, instead trimming the LAST
1254 * block from the directory.
1256 resblks = XFS_REMOVE_SPACE_RES(mp);
1257 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1258 XFS_TRANS_PERM_LOG_RES, log_count);
1259 if (error == ENOSPC) {
1261 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1262 XFS_TRANS_PERM_LOG_RES, log_count);
1265 ASSERT(error != ENOSPC);
1267 goto out_trans_cancel;
1270 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1272 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1273 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1276 * If we're removing a directory perform some additional validation.
1279 ASSERT(ip->i_d.di_nlink >= 2);
1280 if (ip->i_d.di_nlink != 2) {
1281 error = XFS_ERROR(ENOTEMPTY);
1282 goto out_trans_cancel;
1284 if (!xfs_dir_isempty(ip)) {
1285 error = XFS_ERROR(ENOTEMPTY);
1286 goto out_trans_cancel;
1290 xfs_bmap_init(&free_list, &first_block);
1291 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1292 &first_block, &free_list, resblks);
1294 ASSERT(error != ENOENT);
1295 goto out_bmap_cancel;
1297 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1301 * Drop the link from ip's "..".
1303 error = xfs_droplink(tp, dp);
1305 goto out_bmap_cancel;
1308 * Drop the "." link from ip to self.
1310 error = xfs_droplink(tp, ip);
1312 goto out_bmap_cancel;
1315 * When removing a non-directory we need to log the parent
1316 * inode here. For a directory this is done implicitly
1317 * by the xfs_droplink call for the ".." entry.
1319 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1323 * Drop the link from dp to ip.
1325 error = xfs_droplink(tp, ip);
1327 goto out_bmap_cancel;
1330 * Determine if this is the last link while
1331 * we are in the transaction.
1333 link_zero = (ip->i_d.di_nlink == 0);
1336 * If this is a synchronous mount, make sure that the
1337 * remove transaction goes to disk before returning to
1340 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1341 xfs_trans_set_sync(tp);
1343 error = xfs_bmap_finish(&tp, &free_list, &committed);
1345 goto out_bmap_cancel;
1347 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1352 * If we are using filestreams, kill the stream association.
1353 * If the file is still open it may get a new one but that
1354 * will get killed on last close in xfs_close() so we don't
1355 * have to worry about that.
1357 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1358 xfs_filestream_deassociate(ip);
1363 xfs_bmap_cancel(&free_list);
1364 cancel_flags |= XFS_TRANS_ABORT;
1366 xfs_trans_cancel(tp, cancel_flags);
1375 struct xfs_name *target_name)
1377 xfs_mount_t *mp = tdp->i_mount;
1380 xfs_bmap_free_t free_list;
1381 xfs_fsblock_t first_block;
1386 trace_xfs_link(tdp, target_name);
1388 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1390 if (XFS_FORCED_SHUTDOWN(mp))
1391 return XFS_ERROR(EIO);
1393 error = xfs_qm_dqattach(sip, 0);
1397 error = xfs_qm_dqattach(tdp, 0);
1401 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1402 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1403 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1404 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1405 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1406 if (error == ENOSPC) {
1408 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1409 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1416 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1418 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1419 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
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(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(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, 0);
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 >= XFS_ISIZE(ip))
1962 if (endoff > XFS_ISIZE(ip))
1963 endoff = XFS_ISIZE(ip);
1965 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1966 mp->m_rtdev_targp : mp->m_ddev_targp,
1967 BTOBB(mp->m_sb.sb_blocksize), 0);
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_buf_ioerror_alert(bp,
1997 "xfs_zero_remaining_bytes(read)");
2001 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2002 0, lastoffset - offset + 1);
2007 error = xfs_buf_iowait(bp);
2009 xfs_buf_ioerror_alert(bp,
2010 "xfs_zero_remaining_bytes(write)");
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, 0);
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 += XFS_ISIZE(ip);
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;
2275 fsize = XFS_ISIZE(ip);
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:
2317 * These operations actually do IO when extending the file, but
2318 * the allocation is done seperately to the zeroing that is
2319 * done. This set of operations need to be serialised against
2320 * other IO operations, such as truncate and buffered IO. We
2321 * need to take the IOLOCK here to serialise the allocation and
2322 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
2323 * truncate, direct IO) from racing against the transient
2324 * allocated but not written state we can have here.
2326 xfs_ilock(ip, XFS_IOLOCK_EXCL);
2327 if (startoffset > fsize) {
2328 error = xfs_alloc_file_space(ip, fsize,
2329 startoffset - fsize, 0,
2330 attr_flags | XFS_ATTR_NOLOCK);
2332 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2337 iattr.ia_valid = ATTR_SIZE;
2338 iattr.ia_size = startoffset;
2340 error = xfs_setattr_size(ip, &iattr,
2341 attr_flags | XFS_ATTR_NOLOCK);
2342 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2352 return XFS_ERROR(EINVAL);
2356 * update the inode timestamp, mode, and prealloc flag bits
2358 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2360 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2363 xfs_trans_cancel(tp, 0);
2367 xfs_ilock(ip, XFS_ILOCK_EXCL);
2368 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2370 if ((attr_flags & XFS_ATTR_DMI) == 0) {
2371 ip->i_d.di_mode &= ~S_ISUID;
2374 * Note that we don't have to worry about mandatory
2375 * file locking being disabled here because we only
2376 * clear the S_ISGID bit if the Group execute bit is
2377 * on, but if it was on then mandatory locking wouldn't
2378 * have been enabled.
2380 if (ip->i_d.di_mode & S_IXGRP)
2381 ip->i_d.di_mode &= ~S_ISGID;
2383 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2386 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2387 else if (clrprealloc)
2388 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2390 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2391 if (attr_flags & XFS_ATTR_SYNC)
2392 xfs_trans_set_sync(tp);
2393 return xfs_trans_commit(tp, 0);