| blob | aba69da097ae283382d16894e61a90c1bcf4a027 |
1 /*
2 * Copyright (c) 2000-2002 Silicon Graphics, Inc. All Rights Reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11 *
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
74 /*
75 * The maximum pathlen is 1024 bytes. Since the minimum file system
76 * blocksize is 512 bytes, we can get a max of 2 extents back from
77 * bmapi.
78 */
79 #define SYMLINK_MAPS 2
85 /*
86 * For xfs, we check that the file isn't too big to be opened by this kernel.
87 * No other open action is required for regular files. Devices are handled
88 * through the specfs file system, pipes through fifofs. Device and
89 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
90 * when a new vnode is first looked up or created.
91 */
92 STATIC int
96 {
107 /*
108 * If it's a directory with any blocks, read-ahead block 0
109 * as we're almost certain to have the next operation be a read there.
110 */
116 }
118 }
121 /*
122 * xfs_getattr
123 */
124 STATIC int
130 {
152 }
156 #if XFS_BIG_FILESYSTEMS
158 #else
160 #endif
163 /*
164 * Quick exit for non-stat callers
165 */
172 }
174 /*
175 * Copy from in-core inode.
176 */
183 /*
184 * Check vnode type block/char vs. everything else.
185 * Do it with bitmask because that's faster than looking
186 * for multiple values individually.
187 */
193 #if 0
194 /* Large block sizes confuse various
195 * user space programs, so letting the
196 * stripe size through is not a good
197 * idea for now.
198 */
200 /*
201 * If the underlying volume is a stripe, then
202 * return the stripe width in bytes as the
203 * recommended I/O size.
204 */
206 /*
207 * Return the largest of the preferred buffer
208 * sizes since doing small I/Os into larger
209 * buffers causes buffers to be decommissioned.
210 * The value returned is in bytes.
211 */
215 #else
217 /*
218 * Return the largest of the preferred buffer
219 * sizes since doing small I/Os into larger
220 * buffers causes buffers to be decommissioned.
221 * The value returned is in bytes.
222 */
225 #endif
228 /*
229 * If the file blocks are being allocated from a
230 * realtime partition, then return the inode's
231 * realtime extent size or the realtime volume's
232 * extent size.
233 */
237 }
241 }
250 /*
251 * Exit for stat callers. See if any of the rest of the fields
252 * to be filled in are needed.
253 */
260 }
261 /*
262 * convert di_flags to xflags
263 */
281 else
289 }
292 /*
293 * xfs_setattr
294 */
295 STATIC int
301 {
307 uint lock_flags;
321 /*
322 * Cannot set certain attributes.
323 */
327 }
335 /*
336 * Timestamps do not need to be logged and hence do not
337 * need to be done within a transaction.
338 */
346 }
351 /*
352 * If disk quotas is on, we make sure that the dquots do exist on disk,
353 * before we start any other transactions. Trying to do this later
354 * is messy. We don't care to take a readlock to look at the ids
355 * in inode here, because we can't hold it across the trans_reserve.
356 * If the IDs do change before we take the ilock, we're covered
357 * because the i_*dquot fields will get updated anyway.
358 */
367 }
373 }
374 /*
375 * We take a reference when we initialize udqp and gdqp,
376 * so it is important that we never blindly double trip on
377 * the same variable. See xfs_create() for an example.
378 */
384 }
386 /*
387 * For the other attributes, we acquire the inode lock and
388 * first do an error checking pass.
389 */
402 }
403 }
412 }
413 }
415 }
422 }
424 /* boolean: are we the file owner? */
427 /*
428 * Change various properties of a file.
429 * Only the owner or users with CAP_FOWNER
430 * capability may do these things.
431 */
435 /*
436 * CAP_FOWNER overrides the following restrictions:
437 *
438 * The user ID of the calling process must be equal
439 * to the file owner ID, except in cases where the
440 * CAP_FSETID capability is applicable.
441 */
445 }
447 /*
448 * CAP_FSETID overrides the following restrictions:
449 *
450 * The effective user ID of the calling process shall match
451 * the file owner when setting the set-user-ID and
452 * set-group-ID bits on that file.
453 *
454 * The effective group ID or one of the supplementary group
455 * IDs of the calling process shall match the group owner of
456 * the file when setting the set-group-ID bit on that file
457 */
466 #if 0
467 /* Linux allows this, Irix doesn't. */
470 #endif
473 }
474 }
476 /*
477 * Change file ownership. Must be the owner or privileged.
478 * If the system was configured with the "restricted_chown"
479 * option, the owner is not permitted to give away the file,
480 * and can change the group id only to a group of which he
481 * or she is a member.
482 */
484 /*
485 * These IDs could have changed since we last looked at them.
486 * But, we're assured that if the ownership did change
487 * while we didn't have the inode locked, inode's dquot(s)
488 * would have changed also.
489 */
496 iprojid;
498 /*
499 * CAP_CHOWN overrides the following restrictions:
500 *
501 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
502 * shall override the restriction that a process cannot
503 * change the user ID of a file it owns and the restriction
504 * that the group ID supplied to the chown() function
505 * shall be equal to either the group ID or one of the
506 * supplementary group IDs of the calling process.
507 *
508 * XXX: How does restricted_chown affect projid?
509 */
516 }
517 /*
518 * Do a quota reservation only if uid or gid is actually
519 * going to change.
520 */
529 }
530 }
532 /*
533 * Truncate file. Must have write permission and not be a directory.
534 */
536 /* Short circuit the truncate case for zero length files */
545 }
553 }
554 /*
555 * Make sure that the dquots are attached to the inode.
556 */
559 }
561 /*
562 * Change file access or modified times.
563 */
570 }
571 }
572 }
574 /*
575 * Change extent size or realtime flag.
576 */
578 /*
579 * Can't change extent size if any extents are allocated.
580 */
587 }
589 /*
590 * Can't set extent size unless the file is marked, or
591 * about to be marked as a realtime file.
592 *
593 * This check will be removed when fixed size extents
594 * with buffered data writes is implemented.
595 *
596 */
605 }
607 /*
608 * Can't change realtime flag if any extents are allocated.
609 */
615 }
616 /*
617 * Extent size must be a multiple of the appropriate block
618 * size, if set at all.
619 */
621 xfs_extlen_t size;
630 }
634 }
635 }
636 /*
637 * If realtime flag is set then must have realtime data.
638 */
646 }
647 }
648 }
650 /*
651 * Now we can make the changes. Before we join the inode
652 * to the transaction, if XFS_AT_SIZE is set then take care of
653 * the part of the truncation that must be done without the
654 * inode lock. This needs to be done before joining the inode
655 * to the transaction, because the inode cannot be unlocked
656 * once it is a part of the transaction.
657 */
670 }
676 }
680 XFS_TRANS_PERM_LOG_RES,
681 XFS_ITRUNCATE_LOG_COUNT))) {
685 }
688 }
693 }
695 /* determine whether mandatory locking mode changes */
698 /*
699 * Truncate file. Must have write permission and not be a directory.
700 */
707 /*
708 * signal a sync transaction unless
709 * we're truncating an already unlinked
710 * file on a wsync filesystem
711 */
714 XFS_DATA_FORK,
720 }
721 }
722 /*
723 * Have to do this even if the file's size doesn't change.
724 */
726 }
728 /*
729 * Change file access modes.
730 */
737 }
739 /*
740 * Change file ownership. Must be the owner or privileged.
741 * If the system was configured with the "restricted_chown"
742 * option, the owner is not permitted to give away the file,
743 * and can change the group id only to a group of which he
744 * or she is a member.
745 */
747 /*
748 * CAP_FSETID overrides the following restrictions:
749 *
750 * The set-user-ID and set-group-ID bits of a file will be
751 * cleared upon successful return from chown()
752 */
756 }
758 /*
759 * Change the ownerships and register quota modifications
760 * in the transaction.
761 */
768 }
770 }
777 }
779 }
782 /*
783 * We may have to rev the inode as well as
784 * the superblock version number since projids didn't
785 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
786 */
789 }
793 }
796 /*
797 * Change file access or modified times.
798 */
805 }
811 }
814 }
816 /*
817 * Change XFS-added attributes.
818 */
821 /*
822 * Converting bytes to fs blocks.
823 */
826 }
832 }
833 /* can't set PREALLOC this way, just ignore it */
834 }
837 }
839 /*
840 * Change file inode change time only if XFS_AT_CTIME set
841 * AND we have been called by a DMI function.
842 */
849 }
851 /*
852 * Send out timestamp changes that need to be set to the
853 * current time. Not done when called by a DMI function.
854 */
860 /*
861 * If this is a synchronous mount, make sure that the
862 * transaction goes to disk before returning to the user.
863 * This is slightly sub-optimal in that truncates require
864 * two sync transactions instead of one for wsync filesytems.
865 * One for the truncate and one for the timestamps since we
866 * don't want to change the timestamps unless we're sure the
867 * truncate worked. Truncates are less than 1% of the laddis
868 * mix so this probably isn't worth the trouble to optimize.
869 */
876 }
878 /*
879 * If the (regular) file's mandatory locking mode changed, then
880 * notify the vnode. We do this under the inode lock to prevent
881 * racing calls to vop_vnode_change.
882 */
886 mandlock_after);
887 }
891 /*
892 * Release any dquot(s) the inode had kept before chown.
893 */
901 }
908 }
911 abort_return:
913 /* FALLTHROUGH */
914 error_return:
919 }
922 }
924 }
927 /*
928 * xfs_access
929 * Null conversion from vnode mode bits to inode mode bits, as in efs.
930 */
931 STATIC int
936 {
948 }
951 /*
952 * xfs_readlink
953 *
954 */
955 STATIC int
960 {
963 xfs_off_t offset;
970 xfs_daddr_t d;
994 }
998 }
1002 }
1004 /*
1005 * See if the symlink is stored inline.
1006 */
1011 }
1013 /*
1014 * Symlink not inline. Call bmap to get it in.
1015 */
1023 }
1036 }
1044 }
1046 }
1049 error_return:
1054 }
1057 /*
1058 * xfs_fsync
1059 *
1060 * This is called to sync the inode and its data out to disk.
1061 * We need to hold the I/O lock while flushing the data, and
1062 * the inode lock while flushing the inode. The inode lock CANNOT
1063 * be held while flushing the data, so acquire after we're done
1064 * with that.
1065 */
1066 STATIC int
1071 xfs_off_t start,
1072 xfs_off_t stop)
1073 {
1100 }
1102 /*
1103 * If we're invalidating, always flush since we want to
1104 * tear things down. Otherwise, don't flush anything if
1105 * we're not dirty.
1106 */
1111 }
1114 /*
1115 * In the non-invalidating case, calls to fsync() do not
1116 * flush all the dirty mmap'd pages. That requires a
1117 * call to msync().
1118 */
1122 }
1127 }
1129 /*
1130 * We always need to make sure that the required inode state
1131 * is safe on disk. The vnode might be clean but because
1132 * of committed transactions that haven't hit the disk yet.
1133 * Likewise, there could be unflushed non-transactional
1134 * changes to the inode core that have to go to disk.
1135 *
1136 * The following code depends on one assumption: that
1137 * any transaction that changes an inode logs the core
1138 * because it has to change some field in the inode core
1139 * (typically nextents or nblocks). That assumption
1140 * implies that any transactions against an inode will
1141 * catch any non-transactional updates. If inode-altering
1142 * transactions exist that violate this assumption, the
1143 * code breaks. Right now, it figures that if the involved
1144 * update_* field is clear and the inode is unpinned, the
1145 * inode is clean. Either it's been flushed or it's been
1146 * committed and the commit has hit the disk unpinning the inode.
1147 * (Note that xfs_inode_item_format() called at commit clears
1148 * the update_* fields.)
1149 */
1152 /* If we are flushing data then we care about update_size
1153 * being set, otherwise we care about update_core
1154 */
1158 /*
1159 * Timestamps/size haven't changed since last inode
1160 * flush or inode transaction commit. That means
1161 * either nothing got written or a transaction
1162 * committed which caught the updates. If the
1163 * latter happened and the transaction hasn't
1164 * hit the disk yet, the inode will be still
1165 * be pinned. If it is, force the log.
1166 */
1172 XFS_LOG_FORCE |
1175 }
1178 /*
1179 * Kick off a transaction to log the inode
1180 * core to get the updates. Make it
1181 * sync if FSYNC_WAIT is passed in (which
1182 * is done by everybody but specfs). The
1183 * sync transaction will also force the log.
1184 */
1193 }
1196 /*
1197 * Note - it's possible that we might have pushed
1198 * ourselves out of the way during trans_reserve
1199 * which would flush the inode. But there's no
1200 * guarantee that the inode buffer has actually
1201 * gone out yet (it's delwri). Plus the buffer
1202 * could be pinned anyway if it's part of an
1203 * inode in another recent transaction. So we
1204 * play it safe and fire off the transaction anyway.
1205 */
1214 }
1216 }
1219 #if 0
1220 /*
1221 * This is a utility routine for xfs_inactive. It is called when a
1222 * transaction attempting to free up the disk space for a file encounters
1223 * an error. It cancels the old transaction and starts up a new one
1224 * to be used to free up the inode. It also sets the inode size and extent
1225 * counts to 0 and frees up any memory being used to store inline data,
1226 * extents, or btree roots.
1227 */
1228 STATIC void
1234 {
1236 /* REFERENCED */
1242 }
1246 XFS_DEFAULT_LOG_COUNT);
1249 }
1263 }
1265 }
1266 #endif
1268 /*
1269 * This is called by xfs_inactive to free any blocks beyond eof,
1270 * when the link count isn't zero.
1271 */
1272 STATIC int
1276 {
1279 xfs_fileoff_t end_fsb;
1280 xfs_fileoff_t last_fsb;
1281 xfs_filblks_t map_len;
1283 xfs_bmbt_irec_t imap;
1285 /*
1286 * Figure out if there are any blocks beyond the end
1287 * of the file. If not, then there is nothing to do.
1288 */
1303 /*
1304 * Attach the dquots to the inode up front.
1305 */
1309 /*
1310 * There are blocks after the end of file.
1311 * Free them up now by truncating the file to
1312 * its current size.
1313 */
1316 /*
1317 * Do the xfs_itruncate_start() call before
1318 * reserving any log space because
1319 * itruncate_start will call into the buffer
1320 * cache and we can't
1321 * do that within a transaction.
1322 */
1330 XFS_ITRUNCATE_LOG_COUNT);
1336 }
1340 XFS_IOLOCK_EXCL |
1341 XFS_ILOCK_EXCL);
1346 XFS_DATA_FORK,
1348 /*
1349 * If we get an error at this point we
1350 * simply don't bother truncating the file.
1351 */
1355 XFS_TRANS_ABORT));
1358 XFS_TRANS_RELEASE_LOG_RES,
1359 NULL);
1360 }
1362 }
1364 }
1366 /*
1367 * Free a symlink that has blocks associated with it.
1368 */
1369 STATIC int
1373 {
1378 xfs_fsblock_t first_block;
1379 xfs_bmap_free_t free_list;
1391 /*
1392 * We're freeing a symlink that has some
1393 * blocks allocated to it. Free the
1394 * blocks here. We know that we've got
1395 * either 1 or 2 extents and that we can
1396 * free them all in one bunmapi call.
1397 */
1405 }
1406 /*
1407 * Lock the inode, fix the size, and join it to the transaction.
1408 * Hold it so in the normal path, we still have it locked for
1409 * the second transaction. In the error paths we need it
1410 * held so the cancel won't rele it, see below.
1411 */
1418 /*
1419 * Find the block(s) so we can inval and unmap them.
1420 */
1428 /*
1429 * Invalidate the block(s).
1430 */
1436 }
1437 /*
1438 * Unmap the dead block(s) to the free_list.
1439 */
1444 /*
1445 * Commit the first transaction. This logs the EFI and the inode.
1446 */
1449 /*
1450 * The transaction must have been committed, since there were
1451 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1452 * The new tp has the extent freeing and EFDs.
1453 */
1455 /*
1456 * The first xact was committed, so add the inode to the new one.
1457 * Mark it dirty so it will be logged and moved forward in the log as
1458 * part of every commit.
1459 */
1463 /*
1464 * Get a new, empty transaction to return to our caller.
1465 */
1467 /*
1468 * Commit the transaction containing extent freeing and EFD's.
1469 * If we get an error on the commit here or on the reserve below,
1470 * we need to unlock the inode since the new transaction doesn't
1471 * have the inode attached.
1472 */
1478 }
1479 /*
1480 * Remove the memory for extent descriptions (just bookkeeping).
1481 */
1485 /*
1486 * Put an itruncate log reservation in the new transaction
1487 * for our caller.
1488 */
1493 }
1494 /*
1495 * Return with the inode locked but not joined to the transaction.
1496 */
1500 error1:
1502 error0:
1503 /*
1504 * Have to come here with the inode locked and either
1505 * (held and in the transaction) or (not in the transaction).
1506 * If the inode isn't held then cancel would iput it, but
1507 * that's wrong since this is inactive and the vnode ref
1508 * count is 0 already.
1509 * Cancel won't do anything to the inode if held, but it still
1510 * needs to be locked until the cancel is done, if it was
1511 * joined to the transaction.
1512 */
1518 }
1520 STATIC int
1524 {
1528 /*
1529 * We're freeing a symlink which fit into
1530 * the inode. Just free the memory used
1531 * to hold the old symlink.
1532 */
1536 XFS_ITRUNCATE_LOG_COUNT);
1542 }
1545 /*
1546 * Zero length symlinks _can_ exist.
1547 */
1551 XFS_DATA_FORK);
1553 }
1555 }
1557 /*
1558 *
1559 */
1560 STATIC int
1565 {
1583 }
1589 XFS_DEFAULT_LOG_COUNT);
1596 }
1607 }
1609 STATIC int
1612 {
1623 }
1625 /* If this is a read-only mount, don't do this (would generate I/O) */
1638 /* Update linux inode block count after free above */
1641 }
1642 }
1645 }
1647 /*
1648 * xfs_inactive
1649 *
1650 * This is called when the vnode reference count for the vnode
1651 * goes to zero. If the file has been unlinked, then it must
1652 * now be truncated. Also, we clear all of the read-ahead state
1653 * kept for the inode here since the file is now closed.
1654 */
1655 STATIC int
1659 {
1673 /*
1674 * If the inode is already free, then there can be nothing
1675 * to clean up here.
1676 */
1681 }
1683 /*
1684 * Only do a truncate if it's a regular file with
1685 * some actual space in it. It's OK to look at the
1686 * inode's fields without the lock because we're the
1687 * only one with a reference to the inode.
1688 */
1698 }
1702 /* If this is a read-only mount, don't do this (would generate I/O) */
1714 /* Update linux inode block count after free above */
1717 }
1719 }
1728 /*
1729 * Do the xfs_itruncate_start() call before
1730 * reserving any log space because itruncate_start
1731 * will call into the buffer cache and we can't
1732 * do that within a transaction.
1733 */
1741 XFS_ITRUNCATE_LOG_COUNT);
1743 /* Don't call itruncate_cleanup */
1748 }
1754 /*
1755 * normally, we have to run xfs_itruncate_finish sync.
1756 * But if filesystem is wsync and we're in the inactive
1757 * path, then we know that nlink == 0, and that the
1758 * xaction that made nlink == 0 is permanently committed
1759 * since xfs_remove runs as a synchronous transaction.
1760 */
1769 }
1772 /*
1773 * If we get an error while cleaning up a
1774 * symlink we bail out.
1775 */
1783 }
1793 XFS_DEFAULT_LOG_COUNT);
1798 }
1804 }
1806 /*
1807 * If there are attributes associated with the file
1808 * then blow them away now. The code calls a routine
1809 * that recursively deconstructs the attribute fork.
1810 * We need to just commit the current transaction
1811 * because we can't use it for xfs_attr_inactive().
1812 */
1815 /*
1816 * If we got an error, the transaction is already
1817 * cancelled, and the inode is unlocked. Just get out.
1818 */
1823 }
1825 /*
1826 * Free the inode.
1827 */
1830 /*
1831 * If we fail to free the inode, shut down. The cancel
1832 * might do that, we need to make sure. Otherwise the
1833 * inode might be lost for a long time or forever.
1834 */
1840 }
1843 /*
1844 * Credit the quota account(s). The inode is gone.
1845 */
1848 /*
1849 * Just ignore errors at this point. There is
1850 * nothing we can do except to try to keep going.
1851 */
1853 }
1854 /*
1855 * Release the dquots held by inode, if any.
1856 */
1861 out:
1863 }
1866 /*
1867 * xfs_lookup
1868 */
1869 STATIC int
1877 {
1879 xfs_ino_t e_inum;
1881 uint lock_mode;
1897 }
1900 }
1903 #define XFS_CREATE_NEW_MAXTRIES 10000
1905 /*
1906 * xfs_create (create a new file).
1907 */
1908 STATIC int
1915 {
1922 xfs_dev_t rdev;
1924 xfs_bmap_free_t free_list;
1925 xfs_fsblock_t first_block;
1926 boolean_t dp_joined_to_trans;
1928 uint cancel_flags;
1930 xfs_prid_t prid;
1932 uint resblks;
1955 }
1960 /* Return through std_return after this point. */
1965 else
1968 /*
1969 * Make sure that we have allocated dquot(s) on disk.
1970 */
1982 /*
1983 * Initially assume that the file does not exist and
1984 * reserve the resources for that case. If that is not
1985 * the case we'll drop the one we have and get a more
1986 * appropriate transaction later.
1987 */
1994 }
1999 }
2007 /*
2008 * Reserve disk quota and the inode.
2009 */
2026 }
2029 /*
2030 * At this point, we've gotten a newly allocated inode.
2031 * It is locked (and joined to the transaction).
2032 */
2036 /*
2037 * Now we join the directory inode to the transaction.
2038 * We do not do it earlier because xfs_dir_ialloc
2039 * might commit the previous transaction (and release
2040 * all the locks).
2041 */
2053 }
2057 /*
2058 * If this is a synchronous mount, make sure that the
2059 * create transaction goes to disk before returning to
2060 * the user.
2061 */
2064 }
2068 /*
2069 * Attach the dquot(s) to the inodes and modify them incore.
2070 * These ids of the inode couldn't have changed since the new
2071 * inode has been locked ever since it was created.
2072 */
2075 /*
2076 * xfs_trans_commit normally decrements the vnode ref count
2077 * when it unlocks the inode. Since we want to return the
2078 * vnode to the caller, we bump the vnode ref count now.
2079 */
2087 }
2094 }
2099 /*
2100 * Propogate the fact that the vnode changed after the
2101 * xfs_inode locks have been released.
2102 */
2107 /* Fallthrough to std_return with error = 0 */
2109 std_return:
2112 DM_EVENT_POSTCREATE)) {
2118 }
2121 abort_return:
2123 /* FALLTHROUGH */
2124 error_return:
2136 abort_rele:
2137 /*
2138 * Wait until after the current transaction is aborted to
2139 * release the inode. This prevents recursive transactions
2140 * and deadlocks from xfs_inactive.
2141 */
2150 }
2152 #ifdef DEBUG
2153 /*
2154 * Some counters to see if (and how often) we are hitting some deadlock
2155 * prevention code paths.
2156 */
2161 #endif
2163 /*
2164 * The following routine will lock the inodes associated with the
2165 * directory and the named entry in the directory. The locks are
2166 * acquired in increasing inode number.
2167 *
2168 * If the entry is "..", then only the directory is locked. The
2169 * vnode ref count will still include that from the .. entry in
2170 * this case.
2171 *
2172 * There is a deadlock we need to worry about. If the locked directory is
2173 * in the AIL, it might be blocking up the log. The next inode we lock
2174 * could be already locked by another thread waiting for log space (e.g
2175 * a permanent log reservation with a long running transaction (see
2176 * xfs_itruncate_finish)). To solve this, we must check if the directory
2177 * is in the ail and use lock_nowait. If we can't lock, we need to
2178 * drop the inode lock on the directory and try again. xfs_iunlock will
2179 * potentially push the tail if we were holding up the log.
2180 */
2181 STATIC int
2186 {
2188 xfs_ino_t e_inum;
2192 #ifdef DEBUG
2193 xfs_rm_locks++;
2194 #endif
2197 again:
2204 /*
2205 * We want to lock in increasing inum. Since we've already
2206 * acquired the lock on the directory, we may need to release
2207 * if if the inum of the entry turns out to be less.
2208 */
2210 /*
2211 * We are already in the right order, so just
2212 * lock on the inode of the entry.
2213 * We need to use nowait if dp is in the AIL.
2214 */
2219 attempts++;
2220 #ifdef DEBUG
2221 xfs_rm_attempts++;
2222 #endif
2224 /*
2225 * Unlock dp and try again.
2226 * xfs_iunlock will try to push the tail
2227 * if the inode is in the AIL.
2228 */
2234 #ifdef DEBUG
2235 xfs_rm_lock_delays++;
2236 #endif
2237 }
2239 }
2242 }
2249 }
2250 /* else e_inum == dp->i_ino */
2251 /* This can happen if we're asked to lock /x/..
2252 * the entry is "..", which is also the parent directory.
2253 */
2256 }
2258 #ifdef DEBUG
2264 #endif
2266 /*
2267 * The following routine will lock n inodes in exclusive mode.
2268 * We assume the caller calls us with the inodes in i_ino order.
2269 *
2270 * We need to detect deadlock where an inode that we lock
2271 * is in the AIL and we start waiting for another inode that is locked
2272 * by a thread in a long running transaction (such as truncate). This can
2273 * result in deadlock since the long running trans might need to wait
2274 * for the inode we just locked in order to push the tail and free space
2275 * in the log.
2276 */
2277 void
2282 uint lock_mode)
2283 {
2295 }
2297 again:
2304 /*
2305 * If try_lock is not set yet, make sure all locked inodes
2306 * are not in the AIL.
2307 * If any are, set try_lock to be used later.
2308 */
2314 try_lock++;
2315 }
2316 }
2317 }
2319 /*
2320 * If any of the previous locks we have locked is in the AIL,
2321 * we must TRY to get the second and subsequent locks. If
2322 * we can't get any, we must release all we have
2323 * and try again.
2324 */
2327 /* try_lock must be 0 if i is 0. */
2328 /*
2329 * try_lock means we have an inode locked
2330 * that is in the AIL.
2331 */
2334 attempts++;
2336 /*
2337 * Unlock all previous guys and try again.
2338 * xfs_iunlock will try to push the tail
2339 * if the inode is in the AIL.
2340 */
2344 /*
2345 * Check to see if we've already
2346 * unlocked this one.
2347 * Not the first one going back,
2348 * and the inode ptr is the same.
2349 */
2355 }
2359 #ifdef DEBUG
2360 xfs_lock_delays++;
2361 #endif
2362 }
2366 }
2369 }
2370 }
2372 #ifdef DEBUG
2378 xfs_locked_n++;
2379 }
2380 #endif
2381 }
2383 #ifdef DEBUG
2384 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2387 #define REMOVE_DEBUG_TRACE(x)
2391 /*
2392 * xfs_remove
2393 *
2394 */
2395 STATIC int
2400 {
2407 xfs_bmap_free_t free_list;
2408 xfs_fsblock_t first_block;
2413 uint resblks;
2433 }
2435 /* From this point on, return through std_return */
2438 /*
2439 * We need to get a reference to ip before we get our log
2440 * reservation. The reason for this is that we cannot call
2441 * xfs_iget for an inode for which we do not have a reference
2442 * once we've acquired a log reservation. This is because the
2443 * inode we are trying to get might be in xfs_inactive going
2444 * for a log reservation. Since we'll have to wait for the
2445 * inactive code to complete before returning from xfs_iget,
2446 * we need to make sure that we don't have log space reserved
2447 * when we call xfs_iget. Instead we get an unlocked referece
2448 * to the inode before getting our log reservation.
2449 */
2454 }
2469 }
2473 /*
2474 * We try to get the real space reservation first,
2475 * allowing for directory btree deletion(s) implying
2476 * possible bmap insert(s). If we can't get the space
2477 * reservation then we use 0 instead, and avoid the bmap
2478 * btree insert(s) in the directory code by, if the bmap
2479 * insert tries to happen, instead trimming the LAST
2480 * block from the directory.
2481 */
2489 }
2496 }
2504 }
2506 /*
2507 * At this point, we've gotten both the directory and the entry
2508 * inodes locked.
2509 */
2512 /*
2513 * Increment vnode ref count only in this case since
2514 * there's an extra vnode reference in the case where
2515 * dp == ip.
2516 */
2519 }
2524 }
2526 /*
2527 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2528 */
2536 }
2546 }
2548 /* Determine if this is the last link while
2549 * we are in the transaction.
2550 */
2553 /*
2554 * Take an extra ref on the inode so that it doesn't
2555 * go to xfs_inactive() from within the commit.
2556 */
2559 /*
2560 * If this is a synchronous mount, make sure that the
2561 * remove transaction goes to disk before returning to
2562 * the user.
2563 */
2566 }
2572 }
2578 }
2582 /*
2583 * Let interposed file systems know about removed links.
2584 */
2589 /* Fall through to std_return with error = 0 */
2590 std_return:
2592 DM_EVENT_POSTREMOVE)) {
2597 }
2600 error1:
2604 error_return:
2608 error_rele:
2609 /*
2610 * In this case make sure to not release the inode until after
2611 * the current transaction is aborted. Releasing it beforehand
2612 * can cause us to go to xfs_inactive and start a recursive
2613 * transaction which can easily deadlock with the current one.
2614 */
2622 }
2625 /*
2626 * xfs_link
2627 *
2628 */
2629 STATIC int
2635 {
2641 xfs_bmap_free_t free_list;
2642 xfs_fsblock_t first_block;
2659 /*
2660 * For now, manually find the XFS behavior descriptor for
2661 * the source vnode. If it doesn't exist then something
2662 * is wrong and we should just return an error.
2663 * Eventually we need to figure out how link is going to
2664 * work in the face of stacked vnodes.
2665 */
2669 }
2683 }
2685 /* Return through std_return after this point. */
2702 }
2706 }
2714 }
2718 /*
2719 * Increment vnode ref counts since xfs_trans_commit &
2720 * xfs_trans_cancel will both unlock the inodes and
2721 * decrement the associated ref counts.
2722 */
2728 /*
2729 * If the source has too many links, we can't make any more to it.
2730 */
2734 }
2738 target_namelen)))
2745 resblks);
2755 }
2757 /*
2758 * If this is a synchronous mount, make sure that the
2759 * link transaction goes to disk before returning to
2760 * the user.
2761 */
2764 }
2770 }
2775 }
2777 /* Fall through to std_return with error = 0. */
2778 std_return:
2780 DM_EVENT_POSTLINK)) {
2785 }
2788 abort_return:
2790 /* FALLTHROUGH */
2791 error_return:
2795 }
2796 /*
2797 * xfs_mkdir
2798 *
2799 */
2800 STATIC int
2807 {
2813 xfs_dev_t rdev;
2818 xfs_bmap_free_t free_list;
2819 xfs_fsblock_t first_block;
2821 boolean_t dp_joined_to_trans;
2824 xfs_prid_t prid;
2826 uint resblks;
2851 }
2853 /* Return through std_return after this point. */
2861 else
2864 /*
2865 * Make sure that we have allocated dquot(s) on disk.
2866 */
2880 XFS_TRANS_PERM_LOG_RES,
2881 XFS_MKDIR_LOG_COUNT);
2882 }
2887 }
2891 /*
2892 * Check for directory link count overflow.
2893 */
2897 }
2899 /*
2900 * Reserve disk quota and the inode.
2901 */
2909 /*
2910 * create the directory inode.
2911 */
2921 }
2924 /*
2925 * Now we add the directory inode to the transaction.
2926 * We waited until now since xfs_dir_ialloc might start
2927 * a new transaction. Had we joined the transaction
2928 * earlier, the locks might have gotten released.
2929 */
2942 }
2945 /*
2946 * Bump the in memory version number of the parent directory
2947 * so that other processes accessing it will recognize that
2948 * the directory has changed.
2949 */
2955 }
2961 }
2970 /*
2971 * Attach the dquots to the new inode and modify the icount incore.
2972 */
2975 /*
2976 * If this is a synchronous mount, make sure that the
2977 * mkdir transaction goes to disk before returning to
2978 * the user.
2979 */
2982 }
2988 }
2995 }
2997 /* Fall through to std_return with error = 0 or errno from
2998 * xfs_trans_commit. */
3000 std_return:
3003 DM_EVENT_POSTCREATE)) {
3007 DM_RIGHT_NULL,
3010 }
3013 error2:
3014 error1:
3016 abort_return:
3018 error_return:
3025 }
3028 }
3031 /*
3032 * xfs_rmdir
3033 *
3034 */
3035 STATIC int
3040 {
3047 xfs_bmap_free_t free_list;
3048 xfs_fsblock_t first_block;
3055 uint resblks;
3074 }
3076 /* Return through std_return after this point. */
3080 /*
3081 * We need to get a reference to cdp before we get our log
3082 * reservation. The reason for this is that we cannot call
3083 * xfs_iget for an inode for which we do not have a reference
3084 * once we've acquired a log reservation. This is because the
3085 * inode we are trying to get might be in xfs_inactive going
3086 * for a log reservation. Since we'll have to wait for the
3087 * inactive code to complete before returning from xfs_iget,
3088 * we need to make sure that we don't have log space reserved
3089 * when we call xfs_iget. Instead we get an unlocked referece
3090 * to the inode before getting our log reservation.
3091 */
3096 }
3100 /*
3101 * Get the dquots for the inodes.
3102 */
3110 }
3114 /*
3115 * We try to get the real space reservation first,
3116 * allowing for directory btree deletion(s) implying
3117 * possible bmap insert(s). If we can't get the space
3118 * reservation then we use 0 instead, and avoid the bmap
3119 * btree insert(s) in the directory code by, if the bmap
3120 * insert tries to happen, instead trimming the LAST
3121 * block from the directory.
3122 */
3130 }
3136 }
3139 /*
3140 * Now lock the child directory inode and the parent directory
3141 * inode in the proper order. This will take care of validating
3142 * that the directory entry for the child directory inode has
3143 * not changed while we were obtaining a log reservation.
3144 */
3150 }
3154 /*
3155 * Only increment the parent directory vnode count if
3156 * we didn't bump it in looking up cdp. The only time
3157 * we don't bump it is when we're looking up ".".
3158 */
3160 }
3167 }
3173 }
3177 }
3183 }
3187 /*
3188 * Bump the in memory generation count on the parent
3189 * directory so that other can know that it has changed.
3190 */
3193 /*
3194 * Drop the link from cdp's "..".
3195 */
3199 }
3201 /*
3202 * Drop the link from dp to cdp.
3203 */
3207 }
3209 /*
3210 * Drop the "." link from cdp to self.
3211 */
3215 }
3217 /* Determine these before committing transaction */
3220 /*
3221 * Take an extra ref on the child vnode so that it
3222 * does not go to xfs_inactive() from within the commit.
3223 */
3226 /*
3227 * If this is a synchronous mount, make sure that the
3228 * rmdir transaction goes to disk before returning to
3229 * the user.
3230 */
3233 }
3239 XFS_TRANS_ABORT));
3242 }
3248 }
3251 /*
3252 * Let interposed file systems know about removed links.
3253 */
3258 /* Fall through to std_return with error = 0 or the errno
3259 * from xfs_trans_commit. */
3260 std_return:
3267 }
3270 error1:
3273 error_return:
3276 }
3279 /*
3280 * xfs_readdir
3281 *
3282 * Read dp's entries starting at uiop->uio_offset and translate them into
3283 * bufsize bytes worth of struct dirents starting at bufbase.
3284 */
3285 STATIC int
3291 {
3295 uint lock_mode;
3296 xfs_off_t start_offset;
3304 }
3311 }
3314 }
3317 /*
3318 * xfs_symlink
3319 *
3320 */
3321 STATIC int
3329 {
3336 xfs_dev_t rdev;
3337 xfs_bmap_free_t free_list;
3338 xfs_fsblock_t first_block;
3339 boolean_t dp_joined_to_trans;
3341 uint cancel_flags;
3343 xfs_fileoff_t first_fsb;
3344 xfs_filblks_t fs_blocks;
3347 xfs_daddr_t d;
3352 xfs_prid_t prid;
3354 uint resblks;
3375 /*
3376 * Check component lengths of the target path name.
3377 */
3386 /*
3387 * Skip any slashes.
3388 */
3390 total++;
3391 path++;
3392 }
3394 /*
3395 * Count up to the next slash or end of path.
3396 * Error out if the component is bigger than MAXNAMELEN.
3397 */
3402 }
3403 }
3404 }
3405 }
3413 }
3415 /* Return through std_return after this point. */
3420 else
3423 /*
3424 * Make sure that we have allocated dquot(s) on disk.
3425 */
3433 /*
3434 * The symlink will fit into the inode data fork?
3435 * There can't be any attributes so we get the whole variable part.
3436 */
3439 else
3448 }
3453 }
3457 /*
3458 * Reserve disk quota : blocks and inode.
3459 */
3464 /*
3465 * Check for ability to enter directory entry, if no space reserved.
3466 */
3470 /*
3471 * Initialize the bmap freelist prior to calling either
3472 * bmapi or the directory create code.
3473 */
3476 /*
3477 * Allocate an inode for the symlink.
3478 */
3487 }
3494 /*
3495 * Also attach the dquot(s) to it, if applicable.
3496 */
3501 /*
3502 * If the symlink will fit into the inode, write it inline.
3503 */
3509 /*
3510 * The inode was initially created in extent format.
3511 */
3528 }
3544 }
3551 }
3552 }
3554 /*
3555 * Create the directory entry for the symlink.
3556 */
3561 }
3565 /*
3566 * Bump the in memory version number of the parent directory
3567 * so that other processes accessing it will recognize that
3568 * the directory has changed.
3569 */
3572 /*
3573 * If this is a synchronous mount, make sure that the
3574 * symlink transaction goes to disk before returning to
3575 * the user.
3576 */
3579 }
3581 /*
3582 * xfs_trans_commit normally decrements the vnode ref count
3583 * when it unlocks the inode. Since we want to return the
3584 * vnode to the caller, we bump the vnode ref count now.
3585 */
3591 }
3596 /* Fall through to std_return with error = 0 or errno from
3597 * xfs_trans_commit */
3598 std_return:
3600 DM_EVENT_POSTSYMLINK)) {
3606 }
3614 }
3617 error2:
3619 error1:
3622 error_return:
3629 }
3632 }
3635 /*
3636 * xfs_fid2
3637 *
3638 * A fid routine that takes a pointer to a previously allocated
3639 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3640 */
3641 STATIC int
3645 {
3657 /*
3658 * use memcpy because the inode is a long long and there's no
3659 * assurance that xfid->fid_ino is properly aligned.
3660 */
3665 }
3668 /*
3669 * xfs_rwlock
3670 */
3671 int
3674 vrwlock_t locktype)
3675 {
3693 }
3696 }
3699 /*
3700 * xfs_rwunlock
3701 */
3702 void
3705 vrwlock_t locktype)
3706 {
3720 }
3722 }
3724 STATIC int
3728 {
3741 /* Bypass inodes which have already been cleaned by
3742 * the inode flush clustering code inside xfs_iflush
3743 */
3754 xfs_lsn_t sync_lsn;
3767 log_flags);
3768 }
3769 }
3771 /* We make this non-blocking if the inode is contended,
3772 * return EAGAIN to indicate to the caller that they
3773 * did not succeed. This prevents the flush path from
3774 * blocking on inodes inside another operation right
3775 * now, they get caught later by xfs_sync.
3776 */
3782 #if 0
3783 /* not turning this on until some
3784 * performance analysis is done
3785 */
3788 else
3789 #endif
3806 }
3811 }
3815 }
3816 }
3819 }
3822 int
3825 u_int evmask,
3826 u_int16_t state,
3828 {
3848 }
3860 }
3863 /*
3864 * xfs_reclaim
3865 */
3866 STATIC int
3869 {
3882 /*
3883 * Flush and invalidate any data left around that is
3884 * a part of this file.
3885 *
3886 * Get the inode's i/o lock so that buffers are pushed
3887 * out while holding the proper lock. We can't hold
3888 * the inode lock here since flushing out buffers may
3889 * cause us to try to get the lock in xfs_strategy().
3890 *
3891 * We don't have to call remapf() here, because there
3892 * cannot be any mapped file references to this vnode
3893 * since it is being reclaimed.
3894 */
3897 /*
3898 * If we hit an IO error, we need to make sure that the
3899 * buffer and page caches of file data for
3900 * the file are tossed away. We don't want to use
3901 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3902 * pages to stay attached to the vnode, but be
3903 * marked P_BAD. pdflush/vnode_pagebad
3904 * hates that.
3905 */
3910 }
3917 /*
3918 * di_size field may not be quite accurate if we're
3919 * shutting down.
3920 */
3923 }
3924 }
3926 /* If we have nothing to flush with this inode then complete the
3927 * teardown now, otherwise break the link between the xfs inode
3928 * and the linux inode and clean up the xfs inode later. This
3929 * avoids flushing the inode to disk during the delete operation
3930 * itself.
3931 */
3939 /* Protect sync from us */
3945 }
3947 }
3949 int
3954 {
3958 /* The hash lock here protects a thread in xfs_iget_core from
3959 * racing with us on linking the inode back with a vnode.
3960 * Once we have the XFS_IRECLAIM flag set it will not touch
3961 * us.
3962 */
3971 }
3973 }
3977 /*
3978 * If the inode is still dirty, then flush it out. If the inode
3979 * is not in the AIL, then it will be OK to flush it delwri as
3980 * long as xfs_iflush() does not keep any references to the inode.
3981 * We leave that decision up to xfs_iflush() since it has the
3982 * knowledge of whether it's OK to simply do a delwri flush of
3983 * the inode or whether we need to wait until the inode is
3984 * pulled from the AIL.
3985 * We get the flush lock regardless, though, just to make sure
3986 * we don't free it while it is being flushed.
3987 */
3992 }
3998 /*
3999 * If we hit an error, typically because of filesystem
4000 * shutdown, we don't need to let vn_reclaim to know
4001 * because we're gonna reclaim the inode anyway.
4002 */
4007 }
4009 }
4016 /*
4017 * We are not interested in doing an iflush if we're
4018 * in the process of shutting down the filesystem forcibly.
4019 * So, just reclaim the inode.
4020 */
4023 }
4027 }
4029 int
4031 {
4049 }
4050 }
4053 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
4056 }
4059 }
4063 }
4065 /*
4066 * xfs_alloc_file_space()
4067 * This routine allocates disk space for the given file.
4068 *
4069 * If alloc_type == 0, this request is for an ALLOCSP type
4070 * request which will change the file size. In this case, no
4071 * DMAPI event will be generated by the call. A TRUNCATE event
4072 * will be generated later by xfs_setattr.
4073 *
4074 * If alloc_type != 0, this request is for a RESVSP type
4075 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4076 * lower block boundary byte address is less than the file's
4077 * length.
4078 *
4079 * RETURNS:
4080 * 0 on success
4081 * errno on error
4082 *
4083 */
4084 int
4087 xfs_off_t offset,
4088 xfs_off_t len,
4091 {
4092 xfs_filblks_t allocated_fsb;
4093 xfs_filblks_t allocatesize_fsb;
4095 xfs_off_t count;
4096 xfs_filblks_t datablocks;
4098 xfs_fsblock_t firstfsb;
4099 xfs_bmap_free_t free_list;
4105 uint resblks;
4108 xfs_fileoff_t startoffset_fsb;
4118 /*
4119 * determine if this is a realtime file
4120 */
4124 else
4143 /* Generate a DMAPI event if needed. */
4147 xfs_off_t end_dmi_offset;
4157 }
4159 /*
4160 * allocate file space until done or until there is an error
4161 */
4162 retry:
4164 /*
4165 * determine if reserving space on
4166 * the data or realtime partition.
4167 */
4175 rtextsize);
4181 }
4183 /*
4184 * allocate and setup the transaction
4185 */
4189 resblks,
4191 numrtextents,
4192 XFS_TRANS_PERM_LOG_RES,
4193 XFS_WRITE_LOG_COUNT);
4195 /*
4196 * check for running out of space
4197 */
4199 /*
4200 * Free the transaction structure.
4201 */
4205 }
4216 /*
4217 * issue the bmapi() call to allocate the blocks
4218 */
4226 }
4228 /*
4229 * complete the transaction
4230 */
4234 }
4240 }
4247 }
4251 }
4252 dmapi_enospc_check:
4262 /* else fall through with error from XFS_SEND_DATA */
4263 }
4267 error0:
4269 error1:
4273 }
4275 /*
4276 * Zero file bytes between startoff and endoff inclusive.
4277 * The iolock is held exclusive and no blocks are buffered.
4278 */
4279 STATIC int
4282 xfs_off_t startoff,
4283 xfs_off_t endoff)
4284 {
4287 xfs_bmbt_irec_t imap;
4288 xfs_off_t lastoffset;
4291 xfs_off_t offset;
4292 xfs_fileoff_t offset_fsb;
4302 }
4330 }
4342 }
4343 }
4346 }
4348 /*
4349 * xfs_free_file_space()
4350 * This routine frees disk space for the given file.
4351 *
4352 * This routine is only called by xfs_change_file_space
4353 * for an UNRESVSP type call.
4354 *
4355 * RETURNS:
4356 * 0 on success
4357 * errno on error
4358 *
4359 */
4360 STATIC int
4363 xfs_off_t offset,
4364 xfs_off_t len,
4366 {
4369 xfs_off_t end_dmi_offset;
4370 xfs_fileoff_t endoffset_fsb;
4372 xfs_fsblock_t firstfsb;
4373 xfs_bmap_free_t free_list;
4374 xfs_off_t ilen;
4375 xfs_bmbt_irec_t imap;
4376 xfs_off_t ioffset;
4380 uint resblks;
4383 xfs_fileoff_t startoffset_fsb;
4410 }
4420 /*
4421 * Need to zero the stuff we're not freeing, on disk.
4422 * If its a realtime file & can't use unwritten extents then we
4423 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4424 * will take care of it for us.
4425 */
4434 xfs_daddr_t block;
4441 }
4450 mod++;
4453 }
4454 }
4456 /*
4457 * One contiguous piece to clear
4458 */
4461 /*
4462 * Some full blocks, possibly two pieces to clear
4463 */
4472 }
4474 /*
4475 * free file space until done or until there is an error
4476 */
4480 /*
4481 * allocate and setup the transaction
4482 */
4485 resblks,
4488 XFS_TRANS_PERM_LOG_RES,
4489 XFS_WRITE_LOG_COUNT);
4491 /*
4492 * check for running out of space
4493 */
4495 /*
4496 * Free the transaction structure.
4497 */
4501 }
4512 /*
4513 * issue the bunmapi() call to free the blocks
4514 */
4521 }
4523 /*
4524 * complete the transaction
4525 */
4529 }
4533 }
4538 error0:
4540 error1:
4544 }
4546 /*
4547 * xfs_change_file_space()
4548 * This routine allocates or frees disk space for the given file.
4549 * The user specified parameters are checked for alignment and size
4550 * limitations.
4551 *
4552 * RETURNS:
4553 * 0 on success
4554 * errno on error
4555 *
4556 */
4557 int
4562 xfs_off_t offset,
4565 {
4568 xfs_fsize_t fsize;
4572 xfs_off_t startoffset;
4573 xfs_off_t llen;
4575 vattr_t va;
4584 /*
4585 * must be a regular file and have write permission
4586 */
4595 }
4610 }
4625 /*
4626 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4627 * file space.
4628 * These calls do NOT zero the data space allocated to the file,
4629 * nor do they change the file size.
4630 *
4631 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4632 * space.
4633 * These calls cause the new file data to be zeroed and the file
4634 * size to be changed.
4635 */
4651 attr_flags)))
4664 }
4680 }
4682 /*
4683 * update the inode timestamp, mode, and prealloc flag bits
4684 */
4689 /* ASSERT(0); */
4692 }
4701 /*
4702 * Note that we don't have to worry about mandatory
4703 * file locking being disabled here because we only
4704 * clear the ISGID bit if the Group execute bit is
4705 * on, but if it was on then mandatory locking wouldn't
4706 * have been enabled.
4707 */
4726 }
4728 vnodeops_t xfs_vnodeops = {
4766 };