2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
25 #include "xfs_trans.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_itable.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_alloc.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_utils.h"
46 #include "xfs_rtalloc.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_log_priv.h"
49 #include "xfs_filestream.h"
50 #include "xfs_vnodeops.h"
51 #include "xfs_trace.h"
54 * The maximum pathlen is 1024 bytes. Since the minimum file system
55 * blocksize is 512 bytes, we can get a max of 2 extents back from
58 #define SYMLINK_MAPS 2
65 xfs_mount_t *mp = ip->i_mount;
66 int pathlen = ip->i_d.di_size;
67 int nmaps = SYMLINK_MAPS;
68 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
75 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
80 for (n = 0; n < nmaps; n++) {
81 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
82 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
84 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt),
85 XBF_LOCK | XBF_MAPPED | XBF_DONT_BLOCK);
87 return XFS_ERROR(ENOMEM);
90 xfs_ioerror_alert("xfs_readlink",
91 ip->i_mount, bp, XFS_BUF_ADDR(bp));
95 if (pathlen < byte_cnt)
99 memcpy(link, bp->b_addr, byte_cnt);
103 link[ip->i_d.di_size] = '\0';
115 xfs_mount_t *mp = ip->i_mount;
119 trace_xfs_readlink(ip);
121 if (XFS_FORCED_SHUTDOWN(mp))
122 return XFS_ERROR(EIO);
124 xfs_ilock(ip, XFS_ILOCK_SHARED);
126 ASSERT(S_ISLNK(ip->i_d.di_mode));
127 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
129 pathlen = ip->i_d.di_size;
133 if (ip->i_df.if_flags & XFS_IFINLINE) {
134 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
135 link[pathlen] = '\0';
137 error = xfs_readlink_bmap(ip, link);
141 xfs_iunlock(ip, XFS_ILOCK_SHARED);
146 * Flags for xfs_free_eofblocks
148 #define XFS_FREE_EOF_TRYLOCK (1<<0)
151 * This is called by xfs_inactive to free any blocks beyond eof
152 * when the link count isn't zero and by xfs_dm_punch_hole() when
153 * punching a hole to EOF.
163 xfs_fileoff_t end_fsb;
164 xfs_fileoff_t last_fsb;
165 xfs_filblks_t map_len;
167 xfs_bmbt_irec_t imap;
170 * Figure out if there are any blocks beyond the end
171 * of the file. If not, then there is nothing to do.
173 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
174 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
175 if (last_fsb <= end_fsb)
177 map_len = last_fsb - end_fsb;
180 xfs_ilock(ip, XFS_ILOCK_SHARED);
181 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
182 NULL, 0, &imap, &nimaps, NULL);
183 xfs_iunlock(ip, XFS_ILOCK_SHARED);
185 if (!error && (nimaps != 0) &&
186 (imap.br_startblock != HOLESTARTBLOCK ||
187 ip->i_delayed_blks)) {
189 * Attach the dquots to the inode up front.
191 error = xfs_qm_dqattach(ip, 0);
196 * There are blocks after the end of file.
197 * Free them up now by truncating the file to
200 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
202 if (flags & XFS_FREE_EOF_TRYLOCK) {
203 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
204 xfs_trans_cancel(tp, 0);
208 xfs_ilock(ip, XFS_IOLOCK_EXCL);
211 error = xfs_trans_reserve(tp, 0,
212 XFS_ITRUNCATE_LOG_RES(mp),
213 0, XFS_TRANS_PERM_LOG_RES,
214 XFS_ITRUNCATE_LOG_COUNT);
216 ASSERT(XFS_FORCED_SHUTDOWN(mp));
217 xfs_trans_cancel(tp, 0);
218 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
222 xfs_ilock(ip, XFS_ILOCK_EXCL);
223 xfs_trans_ijoin(tp, ip);
225 error = xfs_itruncate_data(&tp, ip, ip->i_size);
228 * If we get an error at this point we simply don't
229 * bother truncating the file.
232 (XFS_TRANS_RELEASE_LOG_RES |
235 error = xfs_trans_commit(tp,
236 XFS_TRANS_RELEASE_LOG_RES);
238 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
244 * Free a symlink that has blocks associated with it.
247 xfs_inactive_symlink_rmt(
255 xfs_fsblock_t first_block;
256 xfs_bmap_free_t free_list;
259 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
267 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
269 * We're freeing a symlink that has some
270 * blocks allocated to it. Free the
271 * blocks here. We know that we've got
272 * either 1 or 2 extents and that we can
273 * free them all in one bunmapi call.
275 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
276 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
277 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
278 ASSERT(XFS_FORCED_SHUTDOWN(mp));
279 xfs_trans_cancel(tp, 0);
284 * Lock the inode, fix the size, and join it to the transaction.
285 * Hold it so in the normal path, we still have it locked for
286 * the second transaction. In the error paths we need it
287 * held so the cancel won't rele it, see below.
289 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
290 size = (int)ip->i_d.di_size;
292 xfs_trans_ijoin(tp, ip);
293 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
295 * Find the block(s) so we can inval and unmap them.
298 xfs_bmap_init(&free_list, &first_block);
299 nmaps = ARRAY_SIZE(mval);
300 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
301 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
305 * Invalidate the block(s).
307 for (i = 0; i < nmaps; i++) {
308 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
309 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
310 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
311 xfs_trans_binval(tp, bp);
314 * Unmap the dead block(s) to the free_list.
316 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
317 &first_block, &free_list, &done)))
321 * Commit the first transaction. This logs the EFI and the inode.
323 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
326 * The transaction must have been committed, since there were
327 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
328 * The new tp has the extent freeing and EFDs.
332 * The first xact was committed, so add the inode to the new one.
333 * Mark it dirty so it will be logged and moved forward in the log as
334 * part of every commit.
336 xfs_trans_ijoin(tp, ip);
337 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
339 * Get a new, empty transaction to return to our caller.
341 ntp = xfs_trans_dup(tp);
343 * Commit the transaction containing extent freeing and EFDs.
344 * If we get an error on the commit here or on the reserve below,
345 * we need to unlock the inode since the new transaction doesn't
346 * have the inode attached.
348 error = xfs_trans_commit(tp, 0);
351 ASSERT(XFS_FORCED_SHUTDOWN(mp));
355 * transaction commit worked ok so we can drop the extra ticket
356 * reference that we gained in xfs_trans_dup()
358 xfs_log_ticket_put(tp->t_ticket);
361 * Remove the memory for extent descriptions (just bookkeeping).
363 if (ip->i_df.if_bytes)
364 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
365 ASSERT(ip->i_df.if_bytes == 0);
367 * Put an itruncate log reservation in the new transaction
370 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
371 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
372 ASSERT(XFS_FORCED_SHUTDOWN(mp));
376 * Return with the inode locked but not joined to the transaction.
382 xfs_bmap_cancel(&free_list);
385 * Have to come here with the inode locked and either
386 * (held and in the transaction) or (not in the transaction).
387 * If the inode isn't held then cancel would iput it, but
388 * that's wrong since this is inactive and the vnode ref
389 * count is 0 already.
390 * Cancel won't do anything to the inode if held, but it still
391 * needs to be locked until the cancel is done, if it was
392 * joined to the transaction.
394 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
395 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
402 xfs_inactive_symlink_local(
408 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
410 * We're freeing a symlink which fit into
411 * the inode. Just free the memory used
412 * to hold the old symlink.
414 error = xfs_trans_reserve(*tpp, 0,
415 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
416 0, XFS_TRANS_PERM_LOG_RES,
417 XFS_ITRUNCATE_LOG_COUNT);
420 xfs_trans_cancel(*tpp, 0);
424 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
427 * Zero length symlinks _can_ exist.
429 if (ip->i_df.if_bytes > 0) {
430 xfs_idata_realloc(ip,
431 -(ip->i_df.if_bytes),
433 ASSERT(ip->i_df.if_bytes == 0);
447 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
450 ASSERT(ip->i_d.di_forkoff != 0);
451 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
452 xfs_iunlock(ip, XFS_ILOCK_EXCL);
456 error = xfs_attr_inactive(ip);
460 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
461 error = xfs_trans_reserve(tp, 0,
462 XFS_IFREE_LOG_RES(mp),
463 0, XFS_TRANS_PERM_LOG_RES,
464 XFS_INACTIVE_LOG_COUNT);
468 xfs_ilock(ip, XFS_ILOCK_EXCL);
469 xfs_trans_ijoin(tp, ip);
470 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
472 ASSERT(ip->i_d.di_anextents == 0);
478 ASSERT(XFS_FORCED_SHUTDOWN(mp));
479 xfs_trans_cancel(tp, 0);
482 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
490 xfs_mount_t *mp = ip->i_mount;
493 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
496 /* If this is a read-only mount, don't do this (would generate I/O) */
497 if (mp->m_flags & XFS_MOUNT_RDONLY)
500 if (!XFS_FORCED_SHUTDOWN(mp)) {
504 * If we are using filestreams, and we have an unlinked
505 * file that we are processing the last close on, then nothing
506 * will be able to reopen and write to this file. Purge this
507 * inode from the filestreams cache so that it doesn't delay
508 * teardown of the inode.
510 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
511 xfs_filestream_deassociate(ip);
514 * If we previously truncated this file and removed old data
515 * in the process, we want to initiate "early" writeout on
516 * the last close. This is an attempt to combat the notorious
517 * NULL files problem which is particularly noticeable from a
518 * truncate down, buffered (re-)write (delalloc), followed by
519 * a crash. What we are effectively doing here is
520 * significantly reducing the time window where we'd otherwise
521 * be exposed to that problem.
523 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
525 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
526 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
527 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
531 if (ip->i_d.di_nlink == 0)
534 if ((S_ISREG(ip->i_d.di_mode) &&
535 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
536 ip->i_delayed_blks > 0)) &&
537 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
538 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
541 * If we can't get the iolock just skip truncating the blocks
542 * past EOF because we could deadlock with the mmap_sem
543 * otherwise. We'll get another chance to drop them once the
544 * last reference to the inode is dropped, so we'll never leak
545 * blocks permanently.
547 * Further, check if the inode is being opened, written and
548 * closed frequently and we have delayed allocation blocks
549 * outstanding (e.g. streaming writes from the NFS server),
550 * truncating the blocks past EOF will cause fragmentation to
553 * In this case don't do the truncation, either, but we have to
554 * be careful how we detect this case. Blocks beyond EOF show
555 * up as i_delayed_blks even when the inode is clean, so we
556 * need to truncate them away first before checking for a dirty
557 * release. Hence on the first dirty close we will still remove
558 * the speculative allocation, but after that we will leave it
561 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
564 error = xfs_free_eofblocks(mp, ip,
565 XFS_FREE_EOF_TRYLOCK);
569 /* delalloc blocks after truncation means it really is dirty */
570 if (ip->i_delayed_blks)
571 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
579 * This is called when the vnode reference count for the vnode
580 * goes to zero. If the file has been unlinked, then it must
581 * now be truncated. Also, we clear all of the read-ahead state
582 * kept for the inode here since the file is now closed.
588 xfs_bmap_free_t free_list;
589 xfs_fsblock_t first_block;
597 * If the inode is already free, then there can be nothing
600 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
601 ASSERT(ip->i_df.if_real_bytes == 0);
602 ASSERT(ip->i_df.if_broot_bytes == 0);
603 return VN_INACTIVE_CACHE;
607 * Only do a truncate if it's a regular file with
608 * some actual space in it. It's OK to look at the
609 * inode's fields without the lock because we're the
610 * only one with a reference to the inode.
612 truncate = ((ip->i_d.di_nlink == 0) &&
613 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
614 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
615 S_ISREG(ip->i_d.di_mode));
621 /* If this is a read-only mount, don't do this (would generate I/O) */
622 if (mp->m_flags & XFS_MOUNT_RDONLY)
625 if (ip->i_d.di_nlink != 0) {
626 if ((S_ISREG(ip->i_d.di_mode) &&
627 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
628 ip->i_delayed_blks > 0)) &&
629 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
630 (!(ip->i_d.di_flags &
631 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
632 (ip->i_delayed_blks != 0)))) {
633 error = xfs_free_eofblocks(mp, ip, 0);
635 return VN_INACTIVE_CACHE;
640 ASSERT(ip->i_d.di_nlink == 0);
642 error = xfs_qm_dqattach(ip, 0);
644 return VN_INACTIVE_CACHE;
646 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
648 xfs_ilock(ip, XFS_IOLOCK_EXCL);
650 error = xfs_trans_reserve(tp, 0,
651 XFS_ITRUNCATE_LOG_RES(mp),
652 0, XFS_TRANS_PERM_LOG_RES,
653 XFS_ITRUNCATE_LOG_COUNT);
655 /* Don't call itruncate_cleanup */
656 ASSERT(XFS_FORCED_SHUTDOWN(mp));
657 xfs_trans_cancel(tp, 0);
658 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
659 return VN_INACTIVE_CACHE;
662 xfs_ilock(ip, XFS_ILOCK_EXCL);
663 xfs_trans_ijoin(tp, ip);
665 error = xfs_itruncate_data(&tp, ip, 0);
668 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
669 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
670 return VN_INACTIVE_CACHE;
672 } else if (S_ISLNK(ip->i_d.di_mode)) {
675 * If we get an error while cleaning up a
676 * symlink we bail out.
678 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
679 xfs_inactive_symlink_rmt(ip, &tp) :
680 xfs_inactive_symlink_local(ip, &tp);
684 return VN_INACTIVE_CACHE;
687 xfs_trans_ijoin(tp, ip);
689 error = xfs_trans_reserve(tp, 0,
690 XFS_IFREE_LOG_RES(mp),
691 0, XFS_TRANS_PERM_LOG_RES,
692 XFS_INACTIVE_LOG_COUNT);
694 ASSERT(XFS_FORCED_SHUTDOWN(mp));
695 xfs_trans_cancel(tp, 0);
696 return VN_INACTIVE_CACHE;
699 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
700 xfs_trans_ijoin(tp, ip);
704 * If there are attributes associated with the file
705 * then blow them away now. The code calls a routine
706 * that recursively deconstructs the attribute fork.
707 * We need to just commit the current transaction
708 * because we can't use it for xfs_attr_inactive().
710 if (ip->i_d.di_anextents > 0) {
711 error = xfs_inactive_attrs(ip, &tp);
713 * If we got an error, the transaction is already
714 * cancelled, and the inode is unlocked. Just get out.
717 return VN_INACTIVE_CACHE;
718 } else if (ip->i_afp) {
719 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
725 xfs_bmap_init(&free_list, &first_block);
726 error = xfs_ifree(tp, ip, &free_list);
729 * If we fail to free the inode, shut down. The cancel
730 * might do that, we need to make sure. Otherwise the
731 * inode might be lost for a long time or forever.
733 if (!XFS_FORCED_SHUTDOWN(mp)) {
734 xfs_notice(mp, "%s: xfs_ifree returned error %d",
736 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
738 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
741 * Credit the quota account(s). The inode is gone.
743 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
746 * Just ignore errors at this point. There is nothing we can
747 * do except to try to keep going. Make sure it's not a silent
750 error = xfs_bmap_finish(&tp, &free_list, &committed);
752 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
754 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
756 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
761 * Release the dquots held by inode, if any.
764 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
767 return VN_INACTIVE_CACHE;
771 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
772 * is allowed, otherwise it has to be an exact match. If a CI match is found,
773 * ci_name->name will point to a the actual name (caller must free) or
774 * will be set to NULL if an exact match is found.
779 struct xfs_name *name,
781 struct xfs_name *ci_name)
787 trace_xfs_lookup(dp, name);
789 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
790 return XFS_ERROR(EIO);
792 lock_mode = xfs_ilock_map_shared(dp);
793 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
794 xfs_iunlock_map_shared(dp, lock_mode);
799 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
807 kmem_free(ci_name->name);
816 struct xfs_name *name,
821 int is_dir = S_ISDIR(mode);
822 struct xfs_mount *mp = dp->i_mount;
823 struct xfs_inode *ip = NULL;
824 struct xfs_trans *tp = NULL;
826 xfs_bmap_free_t free_list;
827 xfs_fsblock_t first_block;
828 boolean_t unlock_dp_on_error = B_FALSE;
832 struct xfs_dquot *udqp = NULL;
833 struct xfs_dquot *gdqp = NULL;
838 trace_xfs_create(dp, name);
840 if (XFS_FORCED_SHUTDOWN(mp))
841 return XFS_ERROR(EIO);
843 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
844 prid = xfs_get_projid(dp);
846 prid = XFS_PROJID_DEFAULT;
849 * Make sure that we have allocated dquot(s) on disk.
851 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
852 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
858 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
859 log_res = XFS_MKDIR_LOG_RES(mp);
860 log_count = XFS_MKDIR_LOG_COUNT;
861 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
863 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
864 log_res = XFS_CREATE_LOG_RES(mp);
865 log_count = XFS_CREATE_LOG_COUNT;
866 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
869 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
872 * Initially assume that the file does not exist and
873 * reserve the resources for that case. If that is not
874 * the case we'll drop the one we have and get a more
875 * appropriate transaction later.
877 error = xfs_trans_reserve(tp, resblks, log_res, 0,
878 XFS_TRANS_PERM_LOG_RES, log_count);
879 if (error == ENOSPC) {
880 /* flush outstanding delalloc blocks and retry */
881 xfs_flush_inodes(dp);
882 error = xfs_trans_reserve(tp, resblks, log_res, 0,
883 XFS_TRANS_PERM_LOG_RES, log_count);
885 if (error == ENOSPC) {
886 /* No space at all so try a "no-allocation" reservation */
888 error = xfs_trans_reserve(tp, 0, log_res, 0,
889 XFS_TRANS_PERM_LOG_RES, log_count);
893 goto out_trans_cancel;
896 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
897 unlock_dp_on_error = B_TRUE;
900 * Check for directory link count overflow.
902 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
903 error = XFS_ERROR(EMLINK);
904 goto out_trans_cancel;
907 xfs_bmap_init(&free_list, &first_block);
910 * Reserve disk quota and the inode.
912 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
914 goto out_trans_cancel;
916 error = xfs_dir_canenter(tp, dp, name, resblks);
918 goto out_trans_cancel;
921 * A newly created regular or special file just has one directory
922 * entry pointing to them, but a directory also the "." entry
923 * pointing to itself.
925 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
926 prid, resblks > 0, &ip, &committed);
929 goto out_trans_cancel;
930 goto out_trans_abort;
934 * Now we join the directory inode to the transaction. We do not do it
935 * earlier because xfs_dir_ialloc might commit the previous transaction
936 * (and release all the locks). An error from here on will result in
937 * the transaction cancel unlocking dp so don't do it explicitly in the
940 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
941 unlock_dp_on_error = B_FALSE;
943 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
944 &first_block, &free_list, resblks ?
945 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
947 ASSERT(error != ENOSPC);
948 goto out_trans_abort;
950 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
951 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
954 error = xfs_dir_init(tp, ip, dp);
956 goto out_bmap_cancel;
958 error = xfs_bumplink(tp, dp);
960 goto out_bmap_cancel;
964 * If this is a synchronous mount, make sure that the
965 * create transaction goes to disk before returning to
968 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
969 xfs_trans_set_sync(tp);
972 * Attach the dquot(s) to the inodes and modify them incore.
973 * These ids of the inode couldn't have changed since the new
974 * inode has been locked ever since it was created.
976 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
978 error = xfs_bmap_finish(&tp, &free_list, &committed);
980 goto out_bmap_cancel;
982 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
984 goto out_release_inode;
993 xfs_bmap_cancel(&free_list);
995 cancel_flags |= XFS_TRANS_ABORT;
997 xfs_trans_cancel(tp, cancel_flags);
1000 * Wait until after the current transaction is aborted to
1001 * release the inode. This prevents recursive transactions
1002 * and deadlocks from xfs_inactive.
1007 xfs_qm_dqrele(udqp);
1008 xfs_qm_dqrele(gdqp);
1010 if (unlock_dp_on_error)
1011 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1017 int xfs_small_retries;
1018 int xfs_middle_retries;
1019 int xfs_lots_retries;
1020 int xfs_lock_delays;
1024 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1028 xfs_lock_inumorder(int lock_mode, int subclass)
1030 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1031 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1032 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1033 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1039 * The following routine will lock n inodes in exclusive mode.
1040 * We assume the caller calls us with the inodes in i_ino order.
1042 * We need to detect deadlock where an inode that we lock
1043 * is in the AIL and we start waiting for another inode that is locked
1044 * by a thread in a long running transaction (such as truncate). This can
1045 * result in deadlock since the long running trans might need to wait
1046 * for the inode we just locked in order to push the tail and free space
1055 int attempts = 0, i, j, try_lock;
1058 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1064 for (; i < inodes; i++) {
1067 if (i && (ips[i] == ips[i-1])) /* Already locked */
1071 * If try_lock is not set yet, make sure all locked inodes
1072 * are not in the AIL.
1073 * If any are, set try_lock to be used later.
1077 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1078 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1079 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1086 * If any of the previous locks we have locked is in the AIL,
1087 * we must TRY to get the second and subsequent locks. If
1088 * we can't get any, we must release all we have
1093 /* try_lock must be 0 if i is 0. */
1095 * try_lock means we have an inode locked
1096 * that is in the AIL.
1099 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1103 * Unlock all previous guys and try again.
1104 * xfs_iunlock will try to push the tail
1105 * if the inode is in the AIL.
1108 for(j = i - 1; j >= 0; j--) {
1111 * Check to see if we've already
1112 * unlocked this one.
1113 * Not the first one going back,
1114 * and the inode ptr is the same.
1116 if ((j != (i - 1)) && ips[j] ==
1120 xfs_iunlock(ips[j], lock_mode);
1123 if ((attempts % 5) == 0) {
1124 delay(1); /* Don't just spin the CPU */
1134 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1140 if (attempts < 5) xfs_small_retries++;
1141 else if (attempts < 100) xfs_middle_retries++;
1142 else xfs_lots_retries++;
1150 * xfs_lock_two_inodes() can only be used to lock one type of lock
1151 * at a time - the iolock or the ilock, but not both at once. If
1152 * we lock both at once, lockdep will report false positives saying
1153 * we have violated locking orders.
1156 xfs_lock_two_inodes(
1165 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1166 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1167 ASSERT(ip0->i_ino != ip1->i_ino);
1169 if (ip0->i_ino > ip1->i_ino) {
1176 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1179 * If the first lock we have locked is in the AIL, we must TRY to get
1180 * the second lock. If we can't get it, we must release the first one
1183 lp = (xfs_log_item_t *)ip0->i_itemp;
1184 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1185 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1186 xfs_iunlock(ip0, lock_mode);
1187 if ((++attempts % 5) == 0)
1188 delay(1); /* Don't just spin the CPU */
1192 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1199 struct xfs_name *name,
1202 xfs_mount_t *mp = dp->i_mount;
1203 xfs_trans_t *tp = NULL;
1204 int is_dir = S_ISDIR(ip->i_d.di_mode);
1206 xfs_bmap_free_t free_list;
1207 xfs_fsblock_t first_block;
1214 trace_xfs_remove(dp, name);
1216 if (XFS_FORCED_SHUTDOWN(mp))
1217 return XFS_ERROR(EIO);
1219 error = xfs_qm_dqattach(dp, 0);
1223 error = xfs_qm_dqattach(ip, 0);
1228 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1229 log_count = XFS_DEFAULT_LOG_COUNT;
1231 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1232 log_count = XFS_REMOVE_LOG_COUNT;
1234 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1237 * We try to get the real space reservation first,
1238 * allowing for directory btree deletion(s) implying
1239 * possible bmap insert(s). If we can't get the space
1240 * reservation then we use 0 instead, and avoid the bmap
1241 * btree insert(s) in the directory code by, if the bmap
1242 * insert tries to happen, instead trimming the LAST
1243 * block from the directory.
1245 resblks = XFS_REMOVE_SPACE_RES(mp);
1246 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1247 XFS_TRANS_PERM_LOG_RES, log_count);
1248 if (error == ENOSPC) {
1250 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1251 XFS_TRANS_PERM_LOG_RES, log_count);
1254 ASSERT(error != ENOSPC);
1256 goto out_trans_cancel;
1259 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1261 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
1262 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1265 * If we're removing a directory perform some additional validation.
1268 ASSERT(ip->i_d.di_nlink >= 2);
1269 if (ip->i_d.di_nlink != 2) {
1270 error = XFS_ERROR(ENOTEMPTY);
1271 goto out_trans_cancel;
1273 if (!xfs_dir_isempty(ip)) {
1274 error = XFS_ERROR(ENOTEMPTY);
1275 goto out_trans_cancel;
1279 xfs_bmap_init(&free_list, &first_block);
1280 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1281 &first_block, &free_list, resblks);
1283 ASSERT(error != ENOENT);
1284 goto out_bmap_cancel;
1286 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1290 * Drop the link from ip's "..".
1292 error = xfs_droplink(tp, dp);
1294 goto out_bmap_cancel;
1297 * Drop the "." link from ip to self.
1299 error = xfs_droplink(tp, ip);
1301 goto out_bmap_cancel;
1304 * When removing a non-directory we need to log the parent
1305 * inode here. For a directory this is done implicitly
1306 * by the xfs_droplink call for the ".." entry.
1308 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1312 * Drop the link from dp to ip.
1314 error = xfs_droplink(tp, ip);
1316 goto out_bmap_cancel;
1319 * Determine if this is the last link while
1320 * we are in the transaction.
1322 link_zero = (ip->i_d.di_nlink == 0);
1325 * If this is a synchronous mount, make sure that the
1326 * remove transaction goes to disk before returning to
1329 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1330 xfs_trans_set_sync(tp);
1332 error = xfs_bmap_finish(&tp, &free_list, &committed);
1334 goto out_bmap_cancel;
1336 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1341 * If we are using filestreams, kill the stream association.
1342 * If the file is still open it may get a new one but that
1343 * will get killed on last close in xfs_close() so we don't
1344 * have to worry about that.
1346 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1347 xfs_filestream_deassociate(ip);
1352 xfs_bmap_cancel(&free_list);
1353 cancel_flags |= XFS_TRANS_ABORT;
1355 xfs_trans_cancel(tp, cancel_flags);
1364 struct xfs_name *target_name)
1366 xfs_mount_t *mp = tdp->i_mount;
1369 xfs_bmap_free_t free_list;
1370 xfs_fsblock_t first_block;
1375 trace_xfs_link(tdp, target_name);
1377 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1379 if (XFS_FORCED_SHUTDOWN(mp))
1380 return XFS_ERROR(EIO);
1382 error = xfs_qm_dqattach(sip, 0);
1386 error = xfs_qm_dqattach(tdp, 0);
1390 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1391 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1392 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1393 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1394 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1395 if (error == ENOSPC) {
1397 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1398 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1405 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1407 xfs_trans_ijoin_ref(tp, sip, XFS_ILOCK_EXCL);
1408 xfs_trans_ijoin_ref(tp, tdp, XFS_ILOCK_EXCL);
1411 * If the source has too many links, we can't make any more to it.
1413 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
1414 error = XFS_ERROR(EMLINK);
1419 * If we are using project inheritance, we only allow hard link
1420 * creation in our tree when the project IDs are the same; else
1421 * the tree quota mechanism could be circumvented.
1423 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1424 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1425 error = XFS_ERROR(EXDEV);
1429 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1433 xfs_bmap_init(&free_list, &first_block);
1435 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1436 &first_block, &free_list, resblks);
1439 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1440 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1442 error = xfs_bumplink(tp, sip);
1447 * If this is a synchronous mount, make sure that the
1448 * link transaction goes to disk before returning to
1451 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1452 xfs_trans_set_sync(tp);
1455 error = xfs_bmap_finish (&tp, &free_list, &committed);
1457 xfs_bmap_cancel(&free_list);
1461 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1464 cancel_flags |= XFS_TRANS_ABORT;
1466 xfs_trans_cancel(tp, cancel_flags);
1474 struct xfs_name *link_name,
1475 const char *target_path,
1479 xfs_mount_t *mp = dp->i_mount;
1484 xfs_bmap_free_t free_list;
1485 xfs_fsblock_t first_block;
1486 boolean_t unlock_dp_on_error = B_FALSE;
1489 xfs_fileoff_t first_fsb;
1490 xfs_filblks_t fs_blocks;
1492 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1494 const char *cur_chunk;
1499 struct xfs_dquot *udqp, *gdqp;
1507 trace_xfs_symlink(dp, link_name);
1509 if (XFS_FORCED_SHUTDOWN(mp))
1510 return XFS_ERROR(EIO);
1513 * Check component lengths of the target path name.
1515 pathlen = strlen(target_path);
1516 if (pathlen >= MAXPATHLEN) /* total string too long */
1517 return XFS_ERROR(ENAMETOOLONG);
1520 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1521 prid = xfs_get_projid(dp);
1523 prid = XFS_PROJID_DEFAULT;
1526 * Make sure that we have allocated dquot(s) on disk.
1528 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1529 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1533 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
1534 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1536 * The symlink will fit into the inode data fork?
1537 * There can't be any attributes so we get the whole variable part.
1539 if (pathlen <= XFS_LITINO(mp))
1542 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
1543 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
1544 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
1545 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1546 if (error == ENOSPC && fs_blocks == 0) {
1548 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
1549 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
1556 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1557 unlock_dp_on_error = B_TRUE;
1560 * Check whether the directory allows new symlinks or not.
1562 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
1563 error = XFS_ERROR(EPERM);
1568 * Reserve disk quota : blocks and inode.
1570 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
1575 * Check for ability to enter directory entry, if no space reserved.
1577 error = xfs_dir_canenter(tp, dp, link_name, resblks);
1581 * Initialize the bmap freelist prior to calling either
1582 * bmapi or the directory create code.
1584 xfs_bmap_init(&free_list, &first_block);
1587 * Allocate an inode for the symlink.
1589 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
1590 prid, resblks > 0, &ip, NULL);
1592 if (error == ENOSPC)
1598 * An error after we've joined dp to the transaction will result in the
1599 * transaction cancel unlocking dp so don't do it explicitly in the
1602 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
1603 unlock_dp_on_error = B_FALSE;
1606 * Also attach the dquot(s) to it, if applicable.
1608 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
1611 resblks -= XFS_IALLOC_SPACE_RES(mp);
1613 * If the symlink will fit into the inode, write it inline.
1615 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
1616 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
1617 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
1618 ip->i_d.di_size = pathlen;
1621 * The inode was initially created in extent format.
1623 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
1624 ip->i_df.if_flags |= XFS_IFINLINE;
1626 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
1627 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
1631 nmaps = SYMLINK_MAPS;
1633 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
1634 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
1635 &first_block, resblks, mval, &nmaps,
1641 resblks -= fs_blocks;
1642 ip->i_d.di_size = pathlen;
1643 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1645 cur_chunk = target_path;
1646 for (n = 0; n < nmaps; n++) {
1647 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1648 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1649 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
1650 BTOBB(byte_cnt), 0);
1651 ASSERT(!xfs_buf_geterror(bp));
1652 if (pathlen < byte_cnt) {
1655 pathlen -= byte_cnt;
1657 memcpy(bp->b_addr, cur_chunk, byte_cnt);
1658 cur_chunk += byte_cnt;
1660 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
1665 * Create the directory entry for the symlink.
1667 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
1668 &first_block, &free_list, resblks);
1671 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1672 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1675 * If this is a synchronous mount, make sure that the
1676 * symlink transaction goes to disk before returning to
1679 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1680 xfs_trans_set_sync(tp);
1683 error = xfs_bmap_finish(&tp, &free_list, &committed);
1687 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1688 xfs_qm_dqrele(udqp);
1689 xfs_qm_dqrele(gdqp);
1697 xfs_bmap_cancel(&free_list);
1698 cancel_flags |= XFS_TRANS_ABORT;
1700 xfs_trans_cancel(tp, cancel_flags);
1701 xfs_qm_dqrele(udqp);
1702 xfs_qm_dqrele(gdqp);
1704 if (unlock_dp_on_error)
1705 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1716 xfs_mount_t *mp = ip->i_mount;
1720 if (!capable(CAP_SYS_ADMIN))
1721 return XFS_ERROR(EPERM);
1723 if (XFS_FORCED_SHUTDOWN(mp))
1724 return XFS_ERROR(EIO);
1726 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
1727 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
1729 xfs_trans_cancel(tp, 0);
1732 xfs_ilock(ip, XFS_ILOCK_EXCL);
1733 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1735 ip->i_d.di_dmevmask = evmask;
1736 ip->i_d.di_dmstate = state;
1738 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1739 error = xfs_trans_commit(tp, 0);
1745 * xfs_alloc_file_space()
1746 * This routine allocates disk space for the given file.
1748 * If alloc_type == 0, this request is for an ALLOCSP type
1749 * request which will change the file size. In this case, no
1750 * DMAPI event will be generated by the call. A TRUNCATE event
1751 * will be generated later by xfs_setattr.
1753 * If alloc_type != 0, this request is for a RESVSP type
1754 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1755 * lower block boundary byte address is less than the file's
1764 xfs_alloc_file_space(
1771 xfs_mount_t *mp = ip->i_mount;
1773 xfs_filblks_t allocated_fsb;
1774 xfs_filblks_t allocatesize_fsb;
1775 xfs_extlen_t extsz, temp;
1776 xfs_fileoff_t startoffset_fsb;
1777 xfs_fsblock_t firstfsb;
1783 xfs_bmbt_irec_t imaps[1], *imapp;
1784 xfs_bmap_free_t free_list;
1785 uint qblocks, resblks, resrtextents;
1789 trace_xfs_alloc_file_space(ip);
1791 if (XFS_FORCED_SHUTDOWN(mp))
1792 return XFS_ERROR(EIO);
1794 error = xfs_qm_dqattach(ip, 0);
1799 return XFS_ERROR(EINVAL);
1801 rt = XFS_IS_REALTIME_INODE(ip);
1802 extsz = xfs_get_extsz_hint(ip);
1807 bmapi_flag = XFS_BMAPI_WRITE | alloc_type;
1808 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1809 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1812 * Allocate file space until done or until there is an error
1814 while (allocatesize_fsb && !error) {
1818 * Determine space reservations for data/realtime.
1820 if (unlikely(extsz)) {
1821 s = startoffset_fsb;
1824 e = startoffset_fsb + allocatesize_fsb;
1825 if ((temp = do_mod(startoffset_fsb, extsz)))
1827 if ((temp = do_mod(e, extsz)))
1831 e = allocatesize_fsb;
1835 * The transaction reservation is limited to a 32-bit block
1836 * count, hence we need to limit the number of blocks we are
1837 * trying to reserve to avoid an overflow. We can't allocate
1838 * more than @nimaps extents, and an extent is limited on disk
1839 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1841 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1843 resrtextents = qblocks = resblks;
1844 resrtextents /= mp->m_sb.sb_rextsize;
1845 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1846 quota_flag = XFS_QMOPT_RES_RTBLKS;
1849 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1850 quota_flag = XFS_QMOPT_RES_REGBLKS;
1854 * Allocate and setup the transaction.
1856 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1857 error = xfs_trans_reserve(tp, resblks,
1858 XFS_WRITE_LOG_RES(mp), resrtextents,
1859 XFS_TRANS_PERM_LOG_RES,
1860 XFS_WRITE_LOG_COUNT);
1862 * Check for running out of space
1866 * Free the transaction structure.
1868 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1869 xfs_trans_cancel(tp, 0);
1872 xfs_ilock(ip, XFS_ILOCK_EXCL);
1873 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1878 xfs_trans_ijoin(tp, ip);
1881 * Issue the xfs_bmapi() call to allocate the blocks
1883 xfs_bmap_init(&free_list, &firstfsb);
1884 error = xfs_bmapi(tp, ip, startoffset_fsb,
1885 allocatesize_fsb, bmapi_flag,
1886 &firstfsb, 0, imapp, &nimaps,
1893 * Complete the transaction
1895 error = xfs_bmap_finish(&tp, &free_list, &committed);
1900 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1901 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1906 allocated_fsb = imapp->br_blockcount;
1909 error = XFS_ERROR(ENOSPC);
1913 startoffset_fsb += allocated_fsb;
1914 allocatesize_fsb -= allocated_fsb;
1919 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1920 xfs_bmap_cancel(&free_list);
1921 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
1923 error1: /* Just cancel transaction */
1924 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1925 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1930 * Zero file bytes between startoff and endoff inclusive.
1931 * The iolock is held exclusive and no blocks are buffered.
1933 * This function is used by xfs_free_file_space() to zero
1934 * partial blocks when the range to free is not block aligned.
1935 * When unreserving space with boundaries that are not block
1936 * aligned we round up the start and round down the end
1937 * boundaries and then use this function to zero the parts of
1938 * the blocks that got dropped during the rounding.
1941 xfs_zero_remaining_bytes(
1946 xfs_bmbt_irec_t imap;
1947 xfs_fileoff_t offset_fsb;
1948 xfs_off_t lastoffset;
1951 xfs_mount_t *mp = ip->i_mount;
1956 * Avoid doing I/O beyond eof - it's not necessary
1957 * since nothing can read beyond eof. The space will
1958 * be zeroed when the file is extended anyway.
1960 if (startoff >= ip->i_size)
1963 if (endoff > ip->i_size)
1964 endoff = ip->i_size;
1966 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1967 mp->m_rtdev_targp : mp->m_ddev_targp,
1968 mp->m_sb.sb_blocksize, XBF_DONT_BLOCK);
1970 return XFS_ERROR(ENOMEM);
1974 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1975 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1977 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
1978 NULL, 0, &imap, &nimap, NULL);
1979 if (error || nimap < 1)
1981 ASSERT(imap.br_blockcount >= 1);
1982 ASSERT(imap.br_startoff == offset_fsb);
1983 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1984 if (lastoffset > endoff)
1985 lastoffset = endoff;
1986 if (imap.br_startblock == HOLESTARTBLOCK)
1988 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1989 if (imap.br_state == XFS_EXT_UNWRITTEN)
1992 XFS_BUF_UNWRITE(bp);
1994 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1996 error = xfs_buf_iowait(bp);
1998 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
1999 mp, bp, XFS_BUF_ADDR(bp));
2003 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2004 0, lastoffset - offset + 1);
2009 error = xfs_buf_iowait(bp);
2011 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2012 mp, bp, XFS_BUF_ADDR(bp));
2021 * xfs_free_file_space()
2022 * This routine frees disk space for the given file.
2024 * This routine is only called by xfs_change_file_space
2025 * for an UNRESVSP type call.
2033 xfs_free_file_space(
2041 xfs_fileoff_t endoffset_fsb;
2043 xfs_fsblock_t firstfsb;
2044 xfs_bmap_free_t free_list;
2045 xfs_bmbt_irec_t imap;
2053 xfs_fileoff_t startoffset_fsb;
2055 int need_iolock = 1;
2059 trace_xfs_free_file_space(ip);
2061 error = xfs_qm_dqattach(ip, 0);
2066 if (len <= 0) /* if nothing being freed */
2068 rt = XFS_IS_REALTIME_INODE(ip);
2069 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
2070 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
2072 if (attr_flags & XFS_ATTR_NOLOCK)
2075 xfs_ilock(ip, XFS_IOLOCK_EXCL);
2076 /* wait for the completion of any pending DIOs */
2077 inode_dio_wait(VFS_I(ip));
2080 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
2081 ioffset = offset & ~(rounding - 1);
2083 if (VN_CACHED(VFS_I(ip)) != 0) {
2084 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
2086 goto out_unlock_iolock;
2090 * Need to zero the stuff we're not freeing, on disk.
2091 * If it's a realtime file & can't use unwritten extents then we
2092 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2093 * will take care of it for us.
2095 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
2097 error = xfs_bmapi(NULL, ip, startoffset_fsb,
2098 1, 0, NULL, 0, &imap, &nimap, NULL);
2100 goto out_unlock_iolock;
2101 ASSERT(nimap == 0 || nimap == 1);
2102 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2105 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2106 block = imap.br_startblock;
2107 mod = do_div(block, mp->m_sb.sb_rextsize);
2109 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2112 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
2113 1, 0, NULL, 0, &imap, &nimap, NULL);
2115 goto out_unlock_iolock;
2116 ASSERT(nimap == 0 || nimap == 1);
2117 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2118 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2120 if (mod && (mod != mp->m_sb.sb_rextsize))
2121 endoffset_fsb -= mod;
2124 if ((done = (endoffset_fsb <= startoffset_fsb)))
2126 * One contiguous piece to clear
2128 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2131 * Some full blocks, possibly two pieces to clear
2133 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2134 error = xfs_zero_remaining_bytes(ip, offset,
2135 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2137 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2138 error = xfs_zero_remaining_bytes(ip,
2139 XFS_FSB_TO_B(mp, endoffset_fsb),
2144 * free file space until done or until there is an error
2146 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2147 while (!error && !done) {
2150 * allocate and setup the transaction. Allow this
2151 * transaction to dip into the reserve blocks to ensure
2152 * the freeing of the space succeeds at ENOSPC.
2154 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2155 tp->t_flags |= XFS_TRANS_RESERVE;
2156 error = xfs_trans_reserve(tp,
2158 XFS_WRITE_LOG_RES(mp),
2160 XFS_TRANS_PERM_LOG_RES,
2161 XFS_WRITE_LOG_COUNT);
2164 * check for running out of space
2168 * Free the transaction structure.
2170 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2171 xfs_trans_cancel(tp, 0);
2174 xfs_ilock(ip, XFS_ILOCK_EXCL);
2175 error = xfs_trans_reserve_quota(tp, mp,
2176 ip->i_udquot, ip->i_gdquot,
2177 resblks, 0, XFS_QMOPT_RES_REGBLKS);
2181 xfs_trans_ijoin(tp, ip);
2184 * issue the bunmapi() call to free the blocks
2186 xfs_bmap_init(&free_list, &firstfsb);
2187 error = xfs_bunmapi(tp, ip, startoffset_fsb,
2188 endoffset_fsb - startoffset_fsb,
2189 0, 2, &firstfsb, &free_list, &done);
2195 * complete the transaction
2197 error = xfs_bmap_finish(&tp, &free_list, &committed);
2202 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2203 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2208 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2212 xfs_bmap_cancel(&free_list);
2214 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2215 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2221 * xfs_change_file_space()
2222 * This routine allocates or frees disk space for the given file.
2223 * The user specified parameters are checked for alignment and size
2232 xfs_change_file_space(
2239 xfs_mount_t *mp = ip->i_mount;
2244 xfs_off_t startoffset;
2250 if (!S_ISREG(ip->i_d.di_mode))
2251 return XFS_ERROR(EINVAL);
2253 switch (bf->l_whence) {
2254 case 0: /*SEEK_SET*/
2256 case 1: /*SEEK_CUR*/
2257 bf->l_start += offset;
2259 case 2: /*SEEK_END*/
2260 bf->l_start += ip->i_size;
2263 return XFS_ERROR(EINVAL);
2266 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
2268 if ( (bf->l_start < 0)
2269 || (bf->l_start > XFS_MAXIOFFSET(mp))
2270 || (bf->l_start + llen < 0)
2271 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
2272 return XFS_ERROR(EINVAL);
2276 startoffset = bf->l_start;
2280 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2282 * These calls do NOT zero the data space allocated to the file,
2283 * nor do they change the file size.
2285 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2287 * These calls cause the new file data to be zeroed and the file
2288 * size to be changed.
2290 setprealloc = clrprealloc = 0;
2291 prealloc_type = XFS_BMAPI_PREALLOC;
2294 case XFS_IOC_ZERO_RANGE:
2295 prealloc_type |= XFS_BMAPI_CONVERT;
2296 xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0);
2298 case XFS_IOC_RESVSP:
2299 case XFS_IOC_RESVSP64:
2300 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
2301 prealloc_type, attr_flags);
2307 case XFS_IOC_UNRESVSP:
2308 case XFS_IOC_UNRESVSP64:
2309 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
2314 case XFS_IOC_ALLOCSP:
2315 case XFS_IOC_ALLOCSP64:
2316 case XFS_IOC_FREESP:
2317 case XFS_IOC_FREESP64:
2318 if (startoffset > fsize) {
2319 error = xfs_alloc_file_space(ip, fsize,
2320 startoffset - fsize, 0, attr_flags);
2325 iattr.ia_valid = ATTR_SIZE;
2326 iattr.ia_size = startoffset;
2328 error = xfs_setattr_size(ip, &iattr, attr_flags);
2338 return XFS_ERROR(EINVAL);
2342 * update the inode timestamp, mode, and prealloc flag bits
2344 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2346 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2349 xfs_trans_cancel(tp, 0);
2353 xfs_ilock(ip, XFS_ILOCK_EXCL);
2355 xfs_trans_ijoin(tp, ip);
2357 if ((attr_flags & XFS_ATTR_DMI) == 0) {
2358 ip->i_d.di_mode &= ~S_ISUID;
2361 * Note that we don't have to worry about mandatory
2362 * file locking being disabled here because we only
2363 * clear the S_ISGID bit if the Group execute bit is
2364 * on, but if it was on then mandatory locking wouldn't
2365 * have been enabled.
2367 if (ip->i_d.di_mode & S_IXGRP)
2368 ip->i_d.di_mode &= ~S_ISGID;
2370 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2373 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2374 else if (clrprealloc)
2375 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2377 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2378 if (attr_flags & XFS_ATTR_SYNC)
2379 xfs_trans_set_sync(tp);
2381 error = xfs_trans_commit(tp, 0);
2383 xfs_iunlock(ip, XFS_ILOCK_EXCL);