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"
50 #include "xfs_icache.h"
53 * The maximum pathlen is 1024 bytes. Since the minimum file system
54 * blocksize is 512 bytes, we can get a max of 2 extents back from
57 #define SYMLINK_MAPS 2
64 xfs_mount_t *mp = ip->i_mount;
65 int pathlen = ip->i_d.di_size;
66 int nmaps = SYMLINK_MAPS;
67 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
74 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, pathlen), mval, &nmaps,
79 for (n = 0; n < nmaps; n++) {
80 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
81 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
83 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
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 * This is called by xfs_inactive to free any blocks beyond eof
151 * when the link count isn't zero and by xfs_dm_punch_hole() when
152 * punching a hole to EOF.
162 xfs_fileoff_t end_fsb;
163 xfs_fileoff_t last_fsb;
164 xfs_filblks_t map_len;
166 xfs_bmbt_irec_t imap;
169 * Figure out if there are any blocks beyond the end
170 * of the file. If not, then there is nothing to do.
172 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
173 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
174 if (last_fsb <= end_fsb)
176 map_len = last_fsb - end_fsb;
179 xfs_ilock(ip, XFS_ILOCK_SHARED);
180 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
181 xfs_iunlock(ip, XFS_ILOCK_SHARED);
183 if (!error && (nimaps != 0) &&
184 (imap.br_startblock != HOLESTARTBLOCK ||
185 ip->i_delayed_blks)) {
187 * Attach the dquots to the inode up front.
189 error = xfs_qm_dqattach(ip, 0);
194 * There are blocks after the end of file.
195 * Free them up now by truncating the file to
198 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
201 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
202 xfs_trans_cancel(tp, 0);
207 error = xfs_trans_reserve(tp, 0,
208 XFS_ITRUNCATE_LOG_RES(mp),
209 0, XFS_TRANS_PERM_LOG_RES,
210 XFS_ITRUNCATE_LOG_COUNT);
212 ASSERT(XFS_FORCED_SHUTDOWN(mp));
213 xfs_trans_cancel(tp, 0);
215 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
219 xfs_ilock(ip, XFS_ILOCK_EXCL);
220 xfs_trans_ijoin(tp, ip, 0);
223 * Do not update the on-disk file size. If we update the
224 * on-disk file size and then the system crashes before the
225 * contents of the file are flushed to disk then the files
226 * may be full of holes (ie NULL files bug).
228 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
232 * If we get an error at this point we simply don't
233 * bother truncating the file.
236 (XFS_TRANS_RELEASE_LOG_RES |
239 error = xfs_trans_commit(tp,
240 XFS_TRANS_RELEASE_LOG_RES);
242 xfs_inode_clear_eofblocks_tag(ip);
245 xfs_iunlock(ip, XFS_ILOCK_EXCL);
247 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
253 * Free a symlink that has blocks associated with it.
256 xfs_inactive_symlink_rmt(
264 xfs_fsblock_t first_block;
265 xfs_bmap_free_t free_list;
268 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
276 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
278 * We're freeing a symlink that has some
279 * blocks allocated to it. Free the
280 * blocks here. We know that we've got
281 * either 1 or 2 extents and that we can
282 * free them all in one bunmapi call.
284 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
287 * Lock the inode, fix the size, and join it to the transaction.
288 * Hold it so in the normal path, we still have it locked for
289 * the second transaction. In the error paths we need it
290 * held so the cancel won't rele it, see below.
292 size = (int)ip->i_d.di_size;
294 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
296 * Find the block(s) so we can inval and unmap them.
299 xfs_bmap_init(&free_list, &first_block);
300 nmaps = ARRAY_SIZE(mval);
301 error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, size),
306 * Invalidate the block(s).
308 for (i = 0; i < nmaps; i++) {
309 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
310 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
311 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
316 xfs_trans_binval(tp, bp);
319 * Unmap the dead block(s) to the free_list.
321 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
322 &first_block, &free_list, &done)))
326 * Commit the first transaction. This logs the EFI and the inode.
328 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
331 * The transaction must have been committed, since there were
332 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
333 * The new tp has the extent freeing and EFDs.
337 * The first xact was committed, so add the inode to the new one.
338 * Mark it dirty so it will be logged and moved forward in the log as
339 * part of every commit.
341 xfs_trans_ijoin(tp, ip, 0);
342 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
344 * Get a new, empty transaction to return to our caller.
346 ntp = xfs_trans_dup(tp);
348 * Commit the transaction containing extent freeing and EFDs.
349 * If we get an error on the commit here or on the reserve below,
350 * we need to unlock the inode since the new transaction doesn't
351 * have the inode attached.
353 error = xfs_trans_commit(tp, 0);
356 ASSERT(XFS_FORCED_SHUTDOWN(mp));
360 * transaction commit worked ok so we can drop the extra ticket
361 * reference that we gained in xfs_trans_dup()
363 xfs_log_ticket_put(tp->t_ticket);
366 * Remove the memory for extent descriptions (just bookkeeping).
368 if (ip->i_df.if_bytes)
369 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
370 ASSERT(ip->i_df.if_bytes == 0);
372 * Put an itruncate log reservation in the new transaction
375 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
376 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
377 ASSERT(XFS_FORCED_SHUTDOWN(mp));
381 xfs_trans_ijoin(tp, ip, 0);
386 xfs_bmap_cancel(&free_list);
395 xfs_mount_t *mp = ip->i_mount;
398 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
401 /* If this is a read-only mount, don't do this (would generate I/O) */
402 if (mp->m_flags & XFS_MOUNT_RDONLY)
405 if (!XFS_FORCED_SHUTDOWN(mp)) {
409 * If we are using filestreams, and we have an unlinked
410 * file that we are processing the last close on, then nothing
411 * will be able to reopen and write to this file. Purge this
412 * inode from the filestreams cache so that it doesn't delay
413 * teardown of the inode.
415 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
416 xfs_filestream_deassociate(ip);
419 * If we previously truncated this file and removed old data
420 * in the process, we want to initiate "early" writeout on
421 * the last close. This is an attempt to combat the notorious
422 * NULL files problem which is particularly noticeable from a
423 * truncate down, buffered (re-)write (delalloc), followed by
424 * a crash. What we are effectively doing here is
425 * significantly reducing the time window where we'd otherwise
426 * be exposed to that problem.
428 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
430 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
431 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
432 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
436 if (ip->i_d.di_nlink == 0)
439 if (xfs_can_free_eofblocks(ip, false)) {
442 * If we can't get the iolock just skip truncating the blocks
443 * past EOF because we could deadlock with the mmap_sem
444 * otherwise. We'll get another chance to drop them once the
445 * last reference to the inode is dropped, so we'll never leak
446 * blocks permanently.
448 * Further, check if the inode is being opened, written and
449 * closed frequently and we have delayed allocation blocks
450 * outstanding (e.g. streaming writes from the NFS server),
451 * truncating the blocks past EOF will cause fragmentation to
454 * In this case don't do the truncation, either, but we have to
455 * be careful how we detect this case. Blocks beyond EOF show
456 * up as i_delayed_blks even when the inode is clean, so we
457 * need to truncate them away first before checking for a dirty
458 * release. Hence on the first dirty close we will still remove
459 * the speculative allocation, but after that we will leave it
462 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
465 error = xfs_free_eofblocks(mp, ip, true);
469 /* delalloc blocks after truncation means it really is dirty */
470 if (ip->i_delayed_blks)
471 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
479 * This is called when the vnode reference count for the vnode
480 * goes to zero. If the file has been unlinked, then it must
481 * now be truncated. Also, we clear all of the read-ahead state
482 * kept for the inode here since the file is now closed.
488 xfs_bmap_free_t free_list;
489 xfs_fsblock_t first_block;
497 * If the inode is already free, then there can be nothing
500 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
501 ASSERT(ip->i_df.if_real_bytes == 0);
502 ASSERT(ip->i_df.if_broot_bytes == 0);
503 return VN_INACTIVE_CACHE;
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 (ip->i_d.di_nlink != 0) {
516 * force is true because we are evicting an inode from the
517 * cache. Post-eof blocks must be freed, lest we end up with
518 * broken free space accounting.
520 if (xfs_can_free_eofblocks(ip, true)) {
521 error = xfs_free_eofblocks(mp, ip, false);
523 return VN_INACTIVE_CACHE;
528 if (S_ISREG(ip->i_d.di_mode) &&
529 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
530 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
533 error = xfs_qm_dqattach(ip, 0);
535 return VN_INACTIVE_CACHE;
537 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
538 error = xfs_trans_reserve(tp, 0,
539 (truncate || S_ISLNK(ip->i_d.di_mode)) ?
540 XFS_ITRUNCATE_LOG_RES(mp) :
541 XFS_IFREE_LOG_RES(mp),
543 XFS_TRANS_PERM_LOG_RES,
544 XFS_ITRUNCATE_LOG_COUNT);
546 ASSERT(XFS_FORCED_SHUTDOWN(mp));
547 xfs_trans_cancel(tp, 0);
548 return VN_INACTIVE_CACHE;
551 xfs_ilock(ip, XFS_ILOCK_EXCL);
552 xfs_trans_ijoin(tp, ip, 0);
554 if (S_ISLNK(ip->i_d.di_mode)) {
556 * Zero length symlinks _can_ exist.
558 if (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) {
559 error = xfs_inactive_symlink_rmt(ip, &tp);
562 } else if (ip->i_df.if_bytes > 0) {
563 xfs_idata_realloc(ip, -(ip->i_df.if_bytes),
565 ASSERT(ip->i_df.if_bytes == 0);
567 } else if (truncate) {
569 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
571 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
575 ASSERT(ip->i_d.di_nextents == 0);
579 * If there are attributes associated with the file then blow them away
580 * now. The code calls a routine that recursively deconstructs the
581 * attribute fork. We need to just commit the current transaction
582 * because we can't use it for xfs_attr_inactive().
584 if (ip->i_d.di_anextents > 0) {
585 ASSERT(ip->i_d.di_forkoff != 0);
587 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
591 xfs_iunlock(ip, XFS_ILOCK_EXCL);
593 error = xfs_attr_inactive(ip);
597 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
598 error = xfs_trans_reserve(tp, 0,
599 XFS_IFREE_LOG_RES(mp),
600 0, XFS_TRANS_PERM_LOG_RES,
601 XFS_INACTIVE_LOG_COUNT);
603 xfs_trans_cancel(tp, 0);
607 xfs_ilock(ip, XFS_ILOCK_EXCL);
608 xfs_trans_ijoin(tp, ip, 0);
612 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
614 ASSERT(ip->i_d.di_anextents == 0);
619 xfs_bmap_init(&free_list, &first_block);
620 error = xfs_ifree(tp, ip, &free_list);
623 * If we fail to free the inode, shut down. The cancel
624 * might do that, we need to make sure. Otherwise the
625 * inode might be lost for a long time or forever.
627 if (!XFS_FORCED_SHUTDOWN(mp)) {
628 xfs_notice(mp, "%s: xfs_ifree returned error %d",
630 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
632 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
635 * Credit the quota account(s). The inode is gone.
637 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
640 * Just ignore errors at this point. There is nothing we can
641 * do except to try to keep going. Make sure it's not a silent
644 error = xfs_bmap_finish(&tp, &free_list, &committed);
646 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
648 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
650 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
655 * Release the dquots held by inode, if any.
659 xfs_iunlock(ip, XFS_ILOCK_EXCL);
661 return VN_INACTIVE_CACHE;
663 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
668 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
669 * is allowed, otherwise it has to be an exact match. If a CI match is found,
670 * ci_name->name will point to a the actual name (caller must free) or
671 * will be set to NULL if an exact match is found.
676 struct xfs_name *name,
678 struct xfs_name *ci_name)
684 trace_xfs_lookup(dp, name);
686 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
687 return XFS_ERROR(EIO);
689 lock_mode = xfs_ilock_map_shared(dp);
690 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
691 xfs_iunlock_map_shared(dp, lock_mode);
696 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
704 kmem_free(ci_name->name);
713 struct xfs_name *name,
718 int is_dir = S_ISDIR(mode);
719 struct xfs_mount *mp = dp->i_mount;
720 struct xfs_inode *ip = NULL;
721 struct xfs_trans *tp = NULL;
723 xfs_bmap_free_t free_list;
724 xfs_fsblock_t first_block;
725 boolean_t unlock_dp_on_error = B_FALSE;
729 struct xfs_dquot *udqp = NULL;
730 struct xfs_dquot *gdqp = NULL;
735 trace_xfs_create(dp, name);
737 if (XFS_FORCED_SHUTDOWN(mp))
738 return XFS_ERROR(EIO);
740 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
741 prid = xfs_get_projid(dp);
743 prid = XFS_PROJID_DEFAULT;
746 * Make sure that we have allocated dquot(s) on disk.
748 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
749 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
755 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
756 log_res = XFS_MKDIR_LOG_RES(mp);
757 log_count = XFS_MKDIR_LOG_COUNT;
758 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
760 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
761 log_res = XFS_CREATE_LOG_RES(mp);
762 log_count = XFS_CREATE_LOG_COUNT;
763 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
766 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
769 * Initially assume that the file does not exist and
770 * reserve the resources for that case. If that is not
771 * the case we'll drop the one we have and get a more
772 * appropriate transaction later.
774 error = xfs_trans_reserve(tp, resblks, log_res, 0,
775 XFS_TRANS_PERM_LOG_RES, log_count);
776 if (error == ENOSPC) {
777 /* flush outstanding delalloc blocks and retry */
778 xfs_flush_inodes(mp);
779 error = xfs_trans_reserve(tp, resblks, log_res, 0,
780 XFS_TRANS_PERM_LOG_RES, log_count);
782 if (error == ENOSPC) {
783 /* No space at all so try a "no-allocation" reservation */
785 error = xfs_trans_reserve(tp, 0, log_res, 0,
786 XFS_TRANS_PERM_LOG_RES, log_count);
790 goto out_trans_cancel;
793 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
794 unlock_dp_on_error = B_TRUE;
796 xfs_bmap_init(&free_list, &first_block);
799 * Reserve disk quota and the inode.
801 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
803 goto out_trans_cancel;
805 error = xfs_dir_canenter(tp, dp, name, resblks);
807 goto out_trans_cancel;
810 * A newly created regular or special file just has one directory
811 * entry pointing to them, but a directory also the "." entry
812 * pointing to itself.
814 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
815 prid, resblks > 0, &ip, &committed);
818 goto out_trans_cancel;
819 goto out_trans_abort;
823 * Now we join the directory inode to the transaction. We do not do it
824 * earlier because xfs_dir_ialloc might commit the previous transaction
825 * (and release all the locks). An error from here on will result in
826 * the transaction cancel unlocking dp so don't do it explicitly in the
829 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
830 unlock_dp_on_error = B_FALSE;
832 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
833 &first_block, &free_list, resblks ?
834 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
836 ASSERT(error != ENOSPC);
837 goto out_trans_abort;
839 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
840 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
843 error = xfs_dir_init(tp, ip, dp);
845 goto out_bmap_cancel;
847 error = xfs_bumplink(tp, dp);
849 goto out_bmap_cancel;
853 * If this is a synchronous mount, make sure that the
854 * create transaction goes to disk before returning to
857 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
858 xfs_trans_set_sync(tp);
861 * Attach the dquot(s) to the inodes and modify them incore.
862 * These ids of the inode couldn't have changed since the new
863 * inode has been locked ever since it was created.
865 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
867 error = xfs_bmap_finish(&tp, &free_list, &committed);
869 goto out_bmap_cancel;
871 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
873 goto out_release_inode;
882 xfs_bmap_cancel(&free_list);
884 cancel_flags |= XFS_TRANS_ABORT;
886 xfs_trans_cancel(tp, cancel_flags);
889 * Wait until after the current transaction is aborted to
890 * release the inode. This prevents recursive transactions
891 * and deadlocks from xfs_inactive.
899 if (unlock_dp_on_error)
900 xfs_iunlock(dp, XFS_ILOCK_EXCL);
906 int xfs_small_retries;
907 int xfs_middle_retries;
908 int xfs_lots_retries;
913 * Bump the subclass so xfs_lock_inodes() acquires each lock with
917 xfs_lock_inumorder(int lock_mode, int subclass)
919 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
920 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
921 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
922 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
928 * The following routine will lock n inodes in exclusive mode.
929 * We assume the caller calls us with the inodes in i_ino order.
931 * We need to detect deadlock where an inode that we lock
932 * is in the AIL and we start waiting for another inode that is locked
933 * by a thread in a long running transaction (such as truncate). This can
934 * result in deadlock since the long running trans might need to wait
935 * for the inode we just locked in order to push the tail and free space
944 int attempts = 0, i, j, try_lock;
947 ASSERT(ips && (inodes >= 2)); /* we need at least two */
953 for (; i < inodes; i++) {
956 if (i && (ips[i] == ips[i-1])) /* Already locked */
960 * If try_lock is not set yet, make sure all locked inodes
961 * are not in the AIL.
962 * If any are, set try_lock to be used later.
966 for (j = (i - 1); j >= 0 && !try_lock; j--) {
967 lp = (xfs_log_item_t *)ips[j]->i_itemp;
968 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
975 * If any of the previous locks we have locked is in the AIL,
976 * we must TRY to get the second and subsequent locks. If
977 * we can't get any, we must release all we have
982 /* try_lock must be 0 if i is 0. */
984 * try_lock means we have an inode locked
985 * that is in the AIL.
988 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
992 * Unlock all previous guys and try again.
993 * xfs_iunlock will try to push the tail
994 * if the inode is in the AIL.
997 for(j = i - 1; j >= 0; j--) {
1000 * Check to see if we've already
1001 * unlocked this one.
1002 * Not the first one going back,
1003 * and the inode ptr is the same.
1005 if ((j != (i - 1)) && ips[j] ==
1009 xfs_iunlock(ips[j], lock_mode);
1012 if ((attempts % 5) == 0) {
1013 delay(1); /* Don't just spin the CPU */
1023 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1029 if (attempts < 5) xfs_small_retries++;
1030 else if (attempts < 100) xfs_middle_retries++;
1031 else xfs_lots_retries++;
1039 * xfs_lock_two_inodes() can only be used to lock one type of lock
1040 * at a time - the iolock or the ilock, but not both at once. If
1041 * we lock both at once, lockdep will report false positives saying
1042 * we have violated locking orders.
1045 xfs_lock_two_inodes(
1054 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1055 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1056 ASSERT(ip0->i_ino != ip1->i_ino);
1058 if (ip0->i_ino > ip1->i_ino) {
1065 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1068 * If the first lock we have locked is in the AIL, we must TRY to get
1069 * the second lock. If we can't get it, we must release the first one
1072 lp = (xfs_log_item_t *)ip0->i_itemp;
1073 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1074 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1075 xfs_iunlock(ip0, lock_mode);
1076 if ((++attempts % 5) == 0)
1077 delay(1); /* Don't just spin the CPU */
1081 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1088 struct xfs_name *name,
1091 xfs_mount_t *mp = dp->i_mount;
1092 xfs_trans_t *tp = NULL;
1093 int is_dir = S_ISDIR(ip->i_d.di_mode);
1095 xfs_bmap_free_t free_list;
1096 xfs_fsblock_t first_block;
1103 trace_xfs_remove(dp, name);
1105 if (XFS_FORCED_SHUTDOWN(mp))
1106 return XFS_ERROR(EIO);
1108 error = xfs_qm_dqattach(dp, 0);
1112 error = xfs_qm_dqattach(ip, 0);
1117 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1118 log_count = XFS_DEFAULT_LOG_COUNT;
1120 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1121 log_count = XFS_REMOVE_LOG_COUNT;
1123 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1126 * We try to get the real space reservation first,
1127 * allowing for directory btree deletion(s) implying
1128 * possible bmap insert(s). If we can't get the space
1129 * reservation then we use 0 instead, and avoid the bmap
1130 * btree insert(s) in the directory code by, if the bmap
1131 * insert tries to happen, instead trimming the LAST
1132 * block from the directory.
1134 resblks = XFS_REMOVE_SPACE_RES(mp);
1135 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1136 XFS_TRANS_PERM_LOG_RES, log_count);
1137 if (error == ENOSPC) {
1139 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1140 XFS_TRANS_PERM_LOG_RES, log_count);
1143 ASSERT(error != ENOSPC);
1145 goto out_trans_cancel;
1148 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1150 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1151 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1154 * If we're removing a directory perform some additional validation.
1157 ASSERT(ip->i_d.di_nlink >= 2);
1158 if (ip->i_d.di_nlink != 2) {
1159 error = XFS_ERROR(ENOTEMPTY);
1160 goto out_trans_cancel;
1162 if (!xfs_dir_isempty(ip)) {
1163 error = XFS_ERROR(ENOTEMPTY);
1164 goto out_trans_cancel;
1168 xfs_bmap_init(&free_list, &first_block);
1169 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1170 &first_block, &free_list, resblks);
1172 ASSERT(error != ENOENT);
1173 goto out_bmap_cancel;
1175 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1179 * Drop the link from ip's "..".
1181 error = xfs_droplink(tp, dp);
1183 goto out_bmap_cancel;
1186 * Drop the "." link from ip to self.
1188 error = xfs_droplink(tp, ip);
1190 goto out_bmap_cancel;
1193 * When removing a non-directory we need to log the parent
1194 * inode here. For a directory this is done implicitly
1195 * by the xfs_droplink call for the ".." entry.
1197 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1201 * Drop the link from dp to ip.
1203 error = xfs_droplink(tp, ip);
1205 goto out_bmap_cancel;
1208 * Determine if this is the last link while
1209 * we are in the transaction.
1211 link_zero = (ip->i_d.di_nlink == 0);
1214 * If this is a synchronous mount, make sure that the
1215 * remove transaction goes to disk before returning to
1218 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1219 xfs_trans_set_sync(tp);
1221 error = xfs_bmap_finish(&tp, &free_list, &committed);
1223 goto out_bmap_cancel;
1225 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1230 * If we are using filestreams, kill the stream association.
1231 * If the file is still open it may get a new one but that
1232 * will get killed on last close in xfs_close() so we don't
1233 * have to worry about that.
1235 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1236 xfs_filestream_deassociate(ip);
1241 xfs_bmap_cancel(&free_list);
1242 cancel_flags |= XFS_TRANS_ABORT;
1244 xfs_trans_cancel(tp, cancel_flags);
1253 struct xfs_name *target_name)
1255 xfs_mount_t *mp = tdp->i_mount;
1258 xfs_bmap_free_t free_list;
1259 xfs_fsblock_t first_block;
1264 trace_xfs_link(tdp, target_name);
1266 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1268 if (XFS_FORCED_SHUTDOWN(mp))
1269 return XFS_ERROR(EIO);
1271 error = xfs_qm_dqattach(sip, 0);
1275 error = xfs_qm_dqattach(tdp, 0);
1279 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1280 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1281 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1282 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1283 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1284 if (error == ENOSPC) {
1286 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1287 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1294 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1296 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1297 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1300 * If we are using project inheritance, we only allow hard link
1301 * creation in our tree when the project IDs are the same; else
1302 * the tree quota mechanism could be circumvented.
1304 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1305 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1306 error = XFS_ERROR(EXDEV);
1310 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1314 xfs_bmap_init(&free_list, &first_block);
1316 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1317 &first_block, &free_list, resblks);
1320 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1321 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1323 error = xfs_bumplink(tp, sip);
1328 * If this is a synchronous mount, make sure that the
1329 * link transaction goes to disk before returning to
1332 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1333 xfs_trans_set_sync(tp);
1336 error = xfs_bmap_finish (&tp, &free_list, &committed);
1338 xfs_bmap_cancel(&free_list);
1342 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1345 cancel_flags |= XFS_TRANS_ABORT;
1347 xfs_trans_cancel(tp, cancel_flags);
1355 struct xfs_name *link_name,
1356 const char *target_path,
1360 xfs_mount_t *mp = dp->i_mount;
1365 xfs_bmap_free_t free_list;
1366 xfs_fsblock_t first_block;
1367 boolean_t unlock_dp_on_error = B_FALSE;
1370 xfs_fileoff_t first_fsb;
1371 xfs_filblks_t fs_blocks;
1373 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1375 const char *cur_chunk;
1380 struct xfs_dquot *udqp, *gdqp;
1388 trace_xfs_symlink(dp, link_name);
1390 if (XFS_FORCED_SHUTDOWN(mp))
1391 return XFS_ERROR(EIO);
1394 * Check component lengths of the target path name.
1396 pathlen = strlen(target_path);
1397 if (pathlen >= MAXPATHLEN) /* total string too long */
1398 return XFS_ERROR(ENAMETOOLONG);
1401 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1402 prid = xfs_get_projid(dp);
1404 prid = XFS_PROJID_DEFAULT;
1407 * Make sure that we have allocated dquot(s) on disk.
1409 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1410 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1414 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
1415 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1417 * The symlink will fit into the inode data fork?
1418 * There can't be any attributes so we get the whole variable part.
1420 if (pathlen <= XFS_LITINO(mp))
1423 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
1424 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
1425 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
1426 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1427 if (error == ENOSPC && fs_blocks == 0) {
1429 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
1430 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1437 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1438 unlock_dp_on_error = B_TRUE;
1441 * Check whether the directory allows new symlinks or not.
1443 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
1444 error = XFS_ERROR(EPERM);
1449 * Reserve disk quota : blocks and inode.
1451 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
1456 * Check for ability to enter directory entry, if no space reserved.
1458 error = xfs_dir_canenter(tp, dp, link_name, resblks);
1462 * Initialize the bmap freelist prior to calling either
1463 * bmapi or the directory create code.
1465 xfs_bmap_init(&free_list, &first_block);
1468 * Allocate an inode for the symlink.
1470 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
1471 prid, resblks > 0, &ip, NULL);
1473 if (error == ENOSPC)
1479 * An error after we've joined dp to the transaction will result in the
1480 * transaction cancel unlocking dp so don't do it explicitly in the
1483 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1484 unlock_dp_on_error = B_FALSE;
1487 * Also attach the dquot(s) to it, if applicable.
1489 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
1492 resblks -= XFS_IALLOC_SPACE_RES(mp);
1494 * If the symlink will fit into the inode, write it inline.
1496 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
1497 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
1498 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
1499 ip->i_d.di_size = pathlen;
1502 * The inode was initially created in extent format.
1504 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
1505 ip->i_df.if_flags |= XFS_IFINLINE;
1507 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
1508 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
1512 nmaps = SYMLINK_MAPS;
1514 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks,
1515 XFS_BMAPI_METADATA, &first_block, resblks,
1516 mval, &nmaps, &free_list);
1521 resblks -= fs_blocks;
1522 ip->i_d.di_size = pathlen;
1523 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1525 cur_chunk = target_path;
1526 for (n = 0; n < nmaps; n++) {
1527 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1528 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1529 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
1530 BTOBB(byte_cnt), 0);
1535 if (pathlen < byte_cnt) {
1538 pathlen -= byte_cnt;
1540 memcpy(bp->b_addr, cur_chunk, byte_cnt);
1541 cur_chunk += byte_cnt;
1543 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
1548 * Create the directory entry for the symlink.
1550 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
1551 &first_block, &free_list, resblks);
1554 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1555 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1558 * If this is a synchronous mount, make sure that the
1559 * symlink transaction goes to disk before returning to
1562 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1563 xfs_trans_set_sync(tp);
1566 error = xfs_bmap_finish(&tp, &free_list, &committed);
1570 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1571 xfs_qm_dqrele(udqp);
1572 xfs_qm_dqrele(gdqp);
1580 xfs_bmap_cancel(&free_list);
1581 cancel_flags |= XFS_TRANS_ABORT;
1583 xfs_trans_cancel(tp, cancel_flags);
1584 xfs_qm_dqrele(udqp);
1585 xfs_qm_dqrele(gdqp);
1587 if (unlock_dp_on_error)
1588 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1599 xfs_mount_t *mp = ip->i_mount;
1603 if (!capable(CAP_SYS_ADMIN))
1604 return XFS_ERROR(EPERM);
1606 if (XFS_FORCED_SHUTDOWN(mp))
1607 return XFS_ERROR(EIO);
1609 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
1610 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
1612 xfs_trans_cancel(tp, 0);
1615 xfs_ilock(ip, XFS_ILOCK_EXCL);
1616 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1618 ip->i_d.di_dmevmask = evmask;
1619 ip->i_d.di_dmstate = state;
1621 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1622 error = xfs_trans_commit(tp, 0);
1628 * xfs_alloc_file_space()
1629 * This routine allocates disk space for the given file.
1631 * If alloc_type == 0, this request is for an ALLOCSP type
1632 * request which will change the file size. In this case, no
1633 * DMAPI event will be generated by the call. A TRUNCATE event
1634 * will be generated later by xfs_setattr.
1636 * If alloc_type != 0, this request is for a RESVSP type
1637 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1638 * lower block boundary byte address is less than the file's
1647 xfs_alloc_file_space(
1654 xfs_mount_t *mp = ip->i_mount;
1656 xfs_filblks_t allocated_fsb;
1657 xfs_filblks_t allocatesize_fsb;
1658 xfs_extlen_t extsz, temp;
1659 xfs_fileoff_t startoffset_fsb;
1660 xfs_fsblock_t firstfsb;
1665 xfs_bmbt_irec_t imaps[1], *imapp;
1666 xfs_bmap_free_t free_list;
1667 uint qblocks, resblks, resrtextents;
1671 trace_xfs_alloc_file_space(ip);
1673 if (XFS_FORCED_SHUTDOWN(mp))
1674 return XFS_ERROR(EIO);
1676 error = xfs_qm_dqattach(ip, 0);
1681 return XFS_ERROR(EINVAL);
1683 rt = XFS_IS_REALTIME_INODE(ip);
1684 extsz = xfs_get_extsz_hint(ip);
1689 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1690 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1693 * Allocate file space until done or until there is an error
1695 while (allocatesize_fsb && !error) {
1699 * Determine space reservations for data/realtime.
1701 if (unlikely(extsz)) {
1702 s = startoffset_fsb;
1705 e = startoffset_fsb + allocatesize_fsb;
1706 if ((temp = do_mod(startoffset_fsb, extsz)))
1708 if ((temp = do_mod(e, extsz)))
1712 e = allocatesize_fsb;
1716 * The transaction reservation is limited to a 32-bit block
1717 * count, hence we need to limit the number of blocks we are
1718 * trying to reserve to avoid an overflow. We can't allocate
1719 * more than @nimaps extents, and an extent is limited on disk
1720 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1722 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1724 resrtextents = qblocks = resblks;
1725 resrtextents /= mp->m_sb.sb_rextsize;
1726 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1727 quota_flag = XFS_QMOPT_RES_RTBLKS;
1730 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1731 quota_flag = XFS_QMOPT_RES_REGBLKS;
1735 * Allocate and setup the transaction.
1737 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1738 error = xfs_trans_reserve(tp, resblks,
1739 XFS_WRITE_LOG_RES(mp), resrtextents,
1740 XFS_TRANS_PERM_LOG_RES,
1741 XFS_WRITE_LOG_COUNT);
1743 * Check for running out of space
1747 * Free the transaction structure.
1749 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1750 xfs_trans_cancel(tp, 0);
1753 xfs_ilock(ip, XFS_ILOCK_EXCL);
1754 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1759 xfs_trans_ijoin(tp, ip, 0);
1761 xfs_bmap_init(&free_list, &firstfsb);
1762 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1763 allocatesize_fsb, alloc_type, &firstfsb,
1764 0, imapp, &nimaps, &free_list);
1770 * Complete the transaction
1772 error = xfs_bmap_finish(&tp, &free_list, &committed);
1777 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1778 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1783 allocated_fsb = imapp->br_blockcount;
1786 error = XFS_ERROR(ENOSPC);
1790 startoffset_fsb += allocated_fsb;
1791 allocatesize_fsb -= allocated_fsb;
1796 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1797 xfs_bmap_cancel(&free_list);
1798 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1800 error1: /* Just cancel transaction */
1801 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1802 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1807 * Zero file bytes between startoff and endoff inclusive.
1808 * The iolock is held exclusive and no blocks are buffered.
1810 * This function is used by xfs_free_file_space() to zero
1811 * partial blocks when the range to free is not block aligned.
1812 * When unreserving space with boundaries that are not block
1813 * aligned we round up the start and round down the end
1814 * boundaries and then use this function to zero the parts of
1815 * the blocks that got dropped during the rounding.
1818 xfs_zero_remaining_bytes(
1823 xfs_bmbt_irec_t imap;
1824 xfs_fileoff_t offset_fsb;
1825 xfs_off_t lastoffset;
1828 xfs_mount_t *mp = ip->i_mount;
1833 * Avoid doing I/O beyond eof - it's not necessary
1834 * since nothing can read beyond eof. The space will
1835 * be zeroed when the file is extended anyway.
1837 if (startoff >= XFS_ISIZE(ip))
1840 if (endoff > XFS_ISIZE(ip))
1841 endoff = XFS_ISIZE(ip);
1843 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1844 mp->m_rtdev_targp : mp->m_ddev_targp,
1845 BTOBB(mp->m_sb.sb_blocksize), 0);
1847 return XFS_ERROR(ENOMEM);
1851 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1852 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1854 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1855 if (error || nimap < 1)
1857 ASSERT(imap.br_blockcount >= 1);
1858 ASSERT(imap.br_startoff == offset_fsb);
1859 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1860 if (lastoffset > endoff)
1861 lastoffset = endoff;
1862 if (imap.br_startblock == HOLESTARTBLOCK)
1864 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1865 if (imap.br_state == XFS_EXT_UNWRITTEN)
1868 XFS_BUF_UNWRITE(bp);
1870 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1872 error = xfs_buf_iowait(bp);
1874 xfs_buf_ioerror_alert(bp,
1875 "xfs_zero_remaining_bytes(read)");
1879 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1880 0, lastoffset - offset + 1);
1885 error = xfs_buf_iowait(bp);
1887 xfs_buf_ioerror_alert(bp,
1888 "xfs_zero_remaining_bytes(write)");
1897 * xfs_free_file_space()
1898 * This routine frees disk space for the given file.
1900 * This routine is only called by xfs_change_file_space
1901 * for an UNRESVSP type call.
1909 xfs_free_file_space(
1917 xfs_fileoff_t endoffset_fsb;
1919 xfs_fsblock_t firstfsb;
1920 xfs_bmap_free_t free_list;
1921 xfs_bmbt_irec_t imap;
1929 xfs_fileoff_t startoffset_fsb;
1931 int need_iolock = 1;
1935 trace_xfs_free_file_space(ip);
1937 error = xfs_qm_dqattach(ip, 0);
1942 if (len <= 0) /* if nothing being freed */
1944 rt = XFS_IS_REALTIME_INODE(ip);
1945 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1946 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1948 if (attr_flags & XFS_ATTR_NOLOCK)
1951 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1952 /* wait for the completion of any pending DIOs */
1953 inode_dio_wait(VFS_I(ip));
1956 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1957 ioffset = offset & ~(rounding - 1);
1959 if (VN_CACHED(VFS_I(ip)) != 0) {
1960 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
1962 goto out_unlock_iolock;
1966 * Need to zero the stuff we're not freeing, on disk.
1967 * If it's a realtime file & can't use unwritten extents then we
1968 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1969 * will take care of it for us.
1971 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1973 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1976 goto out_unlock_iolock;
1977 ASSERT(nimap == 0 || nimap == 1);
1978 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1981 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1982 block = imap.br_startblock;
1983 mod = do_div(block, mp->m_sb.sb_rextsize);
1985 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1988 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1991 goto out_unlock_iolock;
1992 ASSERT(nimap == 0 || nimap == 1);
1993 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1994 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1996 if (mod && (mod != mp->m_sb.sb_rextsize))
1997 endoffset_fsb -= mod;
2000 if ((done = (endoffset_fsb <= startoffset_fsb)))
2002 * One contiguous piece to clear
2004 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2007 * Some full blocks, possibly two pieces to clear
2009 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2010 error = xfs_zero_remaining_bytes(ip, offset,
2011 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2013 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2014 error = xfs_zero_remaining_bytes(ip,
2015 XFS_FSB_TO_B(mp, endoffset_fsb),
2020 * free file space until done or until there is an error
2022 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2023 while (!error && !done) {
2026 * allocate and setup the transaction. Allow this
2027 * transaction to dip into the reserve blocks to ensure
2028 * the freeing of the space succeeds at ENOSPC.
2030 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2031 tp->t_flags |= XFS_TRANS_RESERVE;
2032 error = xfs_trans_reserve(tp,
2034 XFS_WRITE_LOG_RES(mp),
2036 XFS_TRANS_PERM_LOG_RES,
2037 XFS_WRITE_LOG_COUNT);
2040 * check for running out of space
2044 * Free the transaction structure.
2046 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2047 xfs_trans_cancel(tp, 0);
2050 xfs_ilock(ip, XFS_ILOCK_EXCL);
2051 error = xfs_trans_reserve_quota(tp, mp,
2052 ip->i_udquot, ip->i_gdquot,
2053 resblks, 0, XFS_QMOPT_RES_REGBLKS);
2057 xfs_trans_ijoin(tp, ip, 0);
2060 * issue the bunmapi() call to free the blocks
2062 xfs_bmap_init(&free_list, &firstfsb);
2063 error = xfs_bunmapi(tp, ip, startoffset_fsb,
2064 endoffset_fsb - startoffset_fsb,
2065 0, 2, &firstfsb, &free_list, &done);
2071 * complete the transaction
2073 error = xfs_bmap_finish(&tp, &free_list, &committed);
2078 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2079 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2084 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2088 xfs_bmap_cancel(&free_list);
2090 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2091 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2097 * xfs_change_file_space()
2098 * This routine allocates or frees disk space for the given file.
2099 * The user specified parameters are checked for alignment and size
2108 xfs_change_file_space(
2115 xfs_mount_t *mp = ip->i_mount;
2120 xfs_off_t startoffset;
2126 if (!S_ISREG(ip->i_d.di_mode))
2127 return XFS_ERROR(EINVAL);
2129 switch (bf->l_whence) {
2130 case 0: /*SEEK_SET*/
2132 case 1: /*SEEK_CUR*/
2133 bf->l_start += offset;
2135 case 2: /*SEEK_END*/
2136 bf->l_start += XFS_ISIZE(ip);
2139 return XFS_ERROR(EINVAL);
2142 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
2144 if (bf->l_start < 0 ||
2145 bf->l_start > mp->m_super->s_maxbytes ||
2146 bf->l_start + llen < 0 ||
2147 bf->l_start + llen > mp->m_super->s_maxbytes)
2148 return XFS_ERROR(EINVAL);
2152 startoffset = bf->l_start;
2153 fsize = XFS_ISIZE(ip);
2156 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2158 * These calls do NOT zero the data space allocated to the file,
2159 * nor do they change the file size.
2161 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2163 * These calls cause the new file data to be zeroed and the file
2164 * size to be changed.
2166 setprealloc = clrprealloc = 0;
2167 prealloc_type = XFS_BMAPI_PREALLOC;
2170 case XFS_IOC_ZERO_RANGE:
2171 prealloc_type |= XFS_BMAPI_CONVERT;
2172 xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0);
2174 case XFS_IOC_RESVSP:
2175 case XFS_IOC_RESVSP64:
2176 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
2177 prealloc_type, attr_flags);
2183 case XFS_IOC_UNRESVSP:
2184 case XFS_IOC_UNRESVSP64:
2185 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
2190 case XFS_IOC_ALLOCSP:
2191 case XFS_IOC_ALLOCSP64:
2192 case XFS_IOC_FREESP:
2193 case XFS_IOC_FREESP64:
2195 * These operations actually do IO when extending the file, but
2196 * the allocation is done seperately to the zeroing that is
2197 * done. This set of operations need to be serialised against
2198 * other IO operations, such as truncate and buffered IO. We
2199 * need to take the IOLOCK here to serialise the allocation and
2200 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
2201 * truncate, direct IO) from racing against the transient
2202 * allocated but not written state we can have here.
2204 xfs_ilock(ip, XFS_IOLOCK_EXCL);
2205 if (startoffset > fsize) {
2206 error = xfs_alloc_file_space(ip, fsize,
2207 startoffset - fsize, 0,
2208 attr_flags | XFS_ATTR_NOLOCK);
2210 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2215 iattr.ia_valid = ATTR_SIZE;
2216 iattr.ia_size = startoffset;
2218 error = xfs_setattr_size(ip, &iattr,
2219 attr_flags | XFS_ATTR_NOLOCK);
2220 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2230 return XFS_ERROR(EINVAL);
2234 * update the inode timestamp, mode, and prealloc flag bits
2236 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2238 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2241 xfs_trans_cancel(tp, 0);
2245 xfs_ilock(ip, XFS_ILOCK_EXCL);
2246 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2248 if ((attr_flags & XFS_ATTR_DMI) == 0) {
2249 ip->i_d.di_mode &= ~S_ISUID;
2252 * Note that we don't have to worry about mandatory
2253 * file locking being disabled here because we only
2254 * clear the S_ISGID bit if the Group execute bit is
2255 * on, but if it was on then mandatory locking wouldn't
2256 * have been enabled.
2258 if (ip->i_d.di_mode & S_IXGRP)
2259 ip->i_d.di_mode &= ~S_ISGID;
2261 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2264 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2265 else if (clrprealloc)
2266 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2268 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2269 if (attr_flags & XFS_ATTR_SYNC)
2270 xfs_trans_set_sync(tp);
2271 return xfs_trans_commit(tp, 0);