| blob | 00a6b7dc24a0a6ed047999b85f4eeca4f1494d1f |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
56 STATIC int
60 {
68 /*
69 * If it's a directory with any blocks, read-ahead block 0
70 * as we're almost certain to have the next operation be a read there.
71 */
77 }
79 }
81 STATIC int
85 lastclose_t lastclose,
87 {
97 /*
98 * If we previously truncated this file and removed old data in
99 * the process, we want to initiate "early" writeout on the last
100 * close. This is an attempt to combat the notorious NULL files
101 * problem which is particularly noticable from a truncate down,
102 * buffered (re-)write (delalloc), followed by a crash. What we
103 * are effectively doing here is significantly reducing the time
104 * window where we'd otherwise be exposed to that problem.
105 */
109 }
111 /*
112 * xfs_getattr
113 */
114 STATIC int
120 {
144 #if XFS_BIG_INUMS
146 #endif
149 /*
150 * Quick exit for non-stat callers
151 */
157 /*
158 * Copy from in-core inode.
159 */
165 /*
166 * Check vnode type block/char vs. everything else.
167 */
181 /*
182 * If the file blocks are being allocated from a
183 * realtime partition, then return the inode's
184 * realtime extent size or the realtime volume's
185 * extent size.
186 */
190 }
192 }
200 /*
201 * Exit for stat callers. See if any of the rest of the fields
202 * to be filled in are needed.
203 */
209 /*
210 * Convert di_flags to xflags.
211 */
214 /*
215 * Exit for inode revalidate. See if any of the rest of
216 * the fields to be filled in are needed.
217 */
233 else
237 all_done:
241 }
244 /*
245 * xfs_setattr
246 */
247 int
253 {
259 uint lock_flags;
277 /*
278 * Cannot set certain attributes.
279 */
283 }
291 /*
292 * Timestamps do not need to be logged and hence do not
293 * need to be done within a transaction.
294 */
302 }
307 /*
308 * If disk quotas is on, we make sure that the dquots do exist on disk,
309 * before we start any other transactions. Trying to do this later
310 * is messy. We don't care to take a readlock to look at the ids
311 * in inode here, because we can't hold it across the trans_reserve.
312 * If the IDs do change before we take the ilock, we're covered
313 * because the i_*dquot fields will get updated anyway.
314 */
324 }
330 }
336 }
337 /*
338 * We take a reference when we initialize udqp and gdqp,
339 * so it is important that we never blindly double trip on
340 * the same variable. See xfs_create() for an example.
341 */
348 }
350 /*
351 * For the other attributes, we acquire the inode lock and
352 * first do an error checking pass.
353 */
368 }
369 }
379 }
380 }
383 }
387 /* boolean: are we the file owner? */
390 /*
391 * Change various properties of a file.
392 * Only the owner or users with CAP_FOWNER
393 * capability may do these things.
394 */
398 /*
399 * CAP_FOWNER overrides the following restrictions:
400 *
401 * The user ID of the calling process must be equal
402 * to the file owner ID, except in cases where the
403 * CAP_FSETID capability is applicable.
404 */
408 }
410 /*
411 * CAP_FSETID overrides the following restrictions:
412 *
413 * The effective user ID of the calling process shall match
414 * the file owner when setting the set-user-ID and
415 * set-group-ID bits on that file.
416 *
417 * The effective group ID or one of the supplementary group
418 * IDs of the calling process shall match the group owner of
419 * the file when setting the set-group-ID bit on that file
420 */
429 #if 0
430 /* Linux allows this, Irix doesn't. */
433 #endif
436 }
437 }
439 /*
440 * Change file ownership. Must be the owner or privileged.
441 * If the system was configured with the "restricted_chown"
442 * option, the owner is not permitted to give away the file,
443 * and can change the group id only to a group of which he
444 * or she is a member.
445 */
447 /*
448 * These IDs could have changed since we last looked at them.
449 * But, we're assured that if the ownership did change
450 * while we didn't have the inode locked, inode's dquot(s)
451 * would have changed also.
452 */
459 iprojid;
461 /*
462 * CAP_CHOWN overrides the following restrictions:
463 *
464 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
465 * shall override the restriction that a process cannot
466 * change the user ID of a file it owns and the restriction
467 * that the group ID supplied to the chown() function
468 * shall be equal to either the group ID or one of the
469 * supplementary group IDs of the calling process.
470 */
477 }
478 /*
479 * Do a quota reservation only if uid/projid/gid is actually
480 * going to change.
481 */
491 }
492 }
494 /*
495 * Truncate file. Must have write permission and not be a directory.
496 */
498 /* Short circuit the truncate case for zero length files */
507 }
515 }
516 /*
517 * Make sure that the dquots are attached to the inode.
518 */
521 }
523 /*
524 * Change file access or modified times.
525 */
532 }
533 }
534 }
536 /*
537 * Change extent size or realtime flag.
538 */
540 /*
541 * Can't change extent size if any extents are allocated.
542 */
548 }
550 /*
551 * Can't change realtime flag if any extents are allocated.
552 */
559 }
560 /*
561 * Extent size must be a multiple of the appropriate block
562 * size, if set at all.
563 */
565 xfs_extlen_t size;
574 }
578 }
579 }
580 /*
581 * If realtime flag is set then must have realtime data.
582 */
590 }
591 }
593 /*
594 * Can't modify an immutable/append-only file unless
595 * we have appropriate permission.
596 */
605 }
606 }
608 /*
609 * Now we can make the changes. Before we join the inode
610 * to the transaction, if XFS_AT_SIZE is set then take care of
611 * the part of the truncation that must be done without the
612 * inode lock. This needs to be done before joining the inode
613 * to the transaction, because the inode cannot be unlocked
614 * once it is a part of the transaction.
615 */
621 }
631 }
635 XFS_TRANS_PERM_LOG_RES,
636 XFS_ITRUNCATE_LOG_COUNT))) {
641 }
644 }
649 }
651 /* determine whether mandatory locking mode changes */
654 /*
655 * Truncate file. Must have write permission and not be a directory.
656 */
663 /*
664 * signal a sync transaction unless
665 * we're truncating an already unlinked
666 * file on a wsync filesystem
667 */
670 XFS_DATA_FORK,
676 /*
677 * Truncated "down", so we're removing references
678 * to old data here - if we now delay flushing for
679 * a long time, we expose ourselves unduly to the
680 * notorious NULL files problem. So, we mark this
681 * vnode and flush it when the file is closed, and
682 * do not wait the usual (long) time for writeout.
683 */
685 }
686 /*
687 * Have to do this even if the file's size doesn't change.
688 */
690 }
692 /*
693 * Change file access modes.
694 */
701 }
703 /*
704 * Change file ownership. Must be the owner or privileged.
705 * If the system was configured with the "restricted_chown"
706 * option, the owner is not permitted to give away the file,
707 * and can change the group id only to a group of which he
708 * or she is a member.
709 */
711 /*
712 * CAP_FSETID overrides the following restrictions:
713 *
714 * The set-user-ID and set-group-ID bits of a file will be
715 * cleared upon successful return from chown()
716 */
720 }
722 /*
723 * Change the ownerships and register quota modifications
724 * in the transaction.
725 */
732 }
734 }
742 }
744 }
752 }
754 /*
755 * We may have to rev the inode as well as
756 * the superblock version number since projids didn't
757 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
758 */
761 }
765 }
768 /*
769 * Change file access or modified times.
770 */
777 }
783 }
786 }
788 /*
789 * Change XFS-added attributes.
790 */
793 /*
794 * Converting bytes to fs blocks.
795 */
798 }
800 uint di_flags;
802 /* can't set PREALLOC this way, just preserve it */
831 }
834 }
836 }
839 }
841 /*
842 * Change file inode change time only if XFS_AT_CTIME set
843 * AND we have been called by a DMI function.
844 */
851 }
853 /*
854 * Send out timestamp changes that need to be set to the
855 * current time. Not done when called by a DMI function.
856 */
862 /*
863 * If this is a synchronous mount, make sure that the
864 * transaction goes to disk before returning to the user.
865 * This is slightly sub-optimal in that truncates require
866 * two sync transactions instead of one for wsync filesystems.
867 * One for the truncate and one for the timestamps since we
868 * don't want to change the timestamps unless we're sure the
869 * truncate worked. Truncates are less than 1% of the laddis
870 * mix so this probably isn't worth the trouble to optimize.
871 */
878 }
880 /*
881 * If the (regular) file's mandatory locking mode changed, then
882 * notify the vnode. We do this under the inode lock to prevent
883 * racing calls to vop_vnode_change.
884 */
888 mandlock_after);
889 }
893 /*
894 * Release any dquot(s) the inode had kept before chown.
895 */
903 }
910 }
913 abort_return:
915 /* FALLTHROUGH */
916 error_return:
921 }
924 }
926 }
929 /*
930 * xfs_access
931 * Null conversion from vnode mode bits to inode mode bits, as in efs.
932 */
933 STATIC int
938 {
950 }
953 /*
954 * The maximum pathlen is 1024 bytes. Since the minimum file system
955 * blocksize is 512 bytes, we can get a max of 2 extents back from
956 * bmapi.
957 */
958 #define SYMLINK_MAPS 2
960 /*
961 * xfs_readlink
962 *
963 */
964 STATIC int
970 {
973 xfs_off_t offset;
980 xfs_daddr_t d;
1004 }
1008 }
1010 /*
1011 * See if the symlink is stored inline.
1012 */
1017 }
1019 /*
1020 * Symlink not inline. Call bmap to get it in.
1021 */
1029 }
1042 }
1049 }
1051 }
1053 error_return:
1056 }
1059 /*
1060 * xfs_fsync
1061 *
1062 * This is called to sync the inode and its data out to disk.
1063 * We need to hold the I/O lock while flushing the data, and
1064 * the inode lock while flushing the inode. The inode lock CANNOT
1065 * be held while flushing the data, so acquire after we're done
1066 * with that.
1067 */
1068 STATIC int
1073 xfs_off_t start,
1074 xfs_off_t stop)
1075 {
1091 /*
1092 * We always need to make sure that the required inode state
1093 * is safe on disk. The vnode might be clean but because
1094 * of committed transactions that haven't hit the disk yet.
1095 * Likewise, there could be unflushed non-transactional
1096 * changes to the inode core that have to go to disk.
1097 *
1098 * The following code depends on one assumption: that
1099 * any transaction that changes an inode logs the core
1100 * because it has to change some field in the inode core
1101 * (typically nextents or nblocks). That assumption
1102 * implies that any transactions against an inode will
1103 * catch any non-transactional updates. If inode-altering
1104 * transactions exist that violate this assumption, the
1105 * code breaks. Right now, it figures that if the involved
1106 * update_* field is clear and the inode is unpinned, the
1107 * inode is clean. Either it's been flushed or it's been
1108 * committed and the commit has hit the disk unpinning the inode.
1109 * (Note that xfs_inode_item_format() called at commit clears
1110 * the update_* fields.)
1111 */
1114 /* If we are flushing data then we care about update_size
1115 * being set, otherwise we care about update_core
1116 */
1120 /*
1121 * Timestamps/size haven't changed since last inode
1122 * flush or inode transaction commit. That means
1123 * either nothing got written or a transaction
1124 * committed which caught the updates. If the
1125 * latter happened and the transaction hasn't
1126 * hit the disk yet, the inode will be still
1127 * be pinned. If it is, force the log.
1128 */
1134 XFS_LOG_FORCE |
1139 /*
1140 * If the inode is not pinned and nothing
1141 * has changed we don't need to flush the
1142 * cache.
1143 */
1145 }
1148 /*
1149 * Kick off a transaction to log the inode
1150 * core to get the updates. Make it
1151 * sync if FSYNC_WAIT is passed in (which
1152 * is done by everybody but specfs). The
1153 * sync transaction will also force the log.
1154 */
1162 }
1165 /*
1166 * Note - it's possible that we might have pushed
1167 * ourselves out of the way during trans_reserve
1168 * which would flush the inode. But there's no
1169 * guarantee that the inode buffer has actually
1170 * gone out yet (it's delwri). Plus the buffer
1171 * could be pinned anyway if it's part of an
1172 * inode in another recent transaction. So we
1173 * play it safe and fire off the transaction anyway.
1174 */
1183 }
1186 /*
1187 * If the log write didn't issue an ordered tag we need
1188 * to flush the disk cache for the data device now.
1189 */
1193 /*
1194 * If this inode is on the RT dev we need to flush that
1195 * cache as well.
1196 */
1199 }
1202 }
1204 /*
1205 * This is called by xfs_inactive to free any blocks beyond eof,
1206 * when the link count isn't zero.
1207 */
1208 STATIC int
1212 {
1215 xfs_fileoff_t end_fsb;
1216 xfs_fileoff_t last_fsb;
1217 xfs_filblks_t map_len;
1219 xfs_bmbt_irec_t imap;
1221 /*
1222 * Figure out if there are any blocks beyond the end
1223 * of the file. If not, then there is nothing to do.
1224 */
1240 /*
1241 * Attach the dquots to the inode up front.
1242 */
1246 /*
1247 * There are blocks after the end of file.
1248 * Free them up now by truncating the file to
1249 * its current size.
1250 */
1253 /*
1254 * Do the xfs_itruncate_start() call before
1255 * reserving any log space because
1256 * itruncate_start will call into the buffer
1257 * cache and we can't
1258 * do that within a transaction.
1259 */
1267 XFS_ITRUNCATE_LOG_COUNT);
1273 }
1277 XFS_IOLOCK_EXCL |
1278 XFS_ILOCK_EXCL);
1283 XFS_DATA_FORK,
1285 /*
1286 * If we get an error at this point we
1287 * simply don't bother truncating the file.
1288 */
1292 XFS_TRANS_ABORT));
1295 XFS_TRANS_RELEASE_LOG_RES,
1296 NULL);
1297 }
1299 }
1301 }
1303 /*
1304 * Free a symlink that has blocks associated with it.
1305 */
1306 STATIC int
1310 {
1315 xfs_fsblock_t first_block;
1316 xfs_bmap_free_t free_list;
1328 /*
1329 * We're freeing a symlink that has some
1330 * blocks allocated to it. Free the
1331 * blocks here. We know that we've got
1332 * either 1 or 2 extents and that we can
1333 * free them all in one bunmapi call.
1334 */
1342 }
1343 /*
1344 * Lock the inode, fix the size, and join it to the transaction.
1345 * Hold it so in the normal path, we still have it locked for
1346 * the second transaction. In the error paths we need it
1347 * held so the cancel won't rele it, see below.
1348 */
1355 /*
1356 * Find the block(s) so we can inval and unmap them.
1357 */
1365 /*
1366 * Invalidate the block(s).
1367 */
1373 }
1374 /*
1375 * Unmap the dead block(s) to the free_list.
1376 */
1381 /*
1382 * Commit the first transaction. This logs the EFI and the inode.
1383 */
1386 /*
1387 * The transaction must have been committed, since there were
1388 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1389 * The new tp has the extent freeing and EFDs.
1390 */
1392 /*
1393 * The first xact was committed, so add the inode to the new one.
1394 * Mark it dirty so it will be logged and moved forward in the log as
1395 * part of every commit.
1396 */
1400 /*
1401 * Get a new, empty transaction to return to our caller.
1402 */
1404 /*
1405 * Commit the transaction containing extent freeing and EFDs.
1406 * If we get an error on the commit here or on the reserve below,
1407 * we need to unlock the inode since the new transaction doesn't
1408 * have the inode attached.
1409 */
1415 }
1416 /*
1417 * Remove the memory for extent descriptions (just bookkeeping).
1418 */
1422 /*
1423 * Put an itruncate log reservation in the new transaction
1424 * for our caller.
1425 */
1430 }
1431 /*
1432 * Return with the inode locked but not joined to the transaction.
1433 */
1437 error1:
1439 error0:
1440 /*
1441 * Have to come here with the inode locked and either
1442 * (held and in the transaction) or (not in the transaction).
1443 * If the inode isn't held then cancel would iput it, but
1444 * that's wrong since this is inactive and the vnode ref
1445 * count is 0 already.
1446 * Cancel won't do anything to the inode if held, but it still
1447 * needs to be locked until the cancel is done, if it was
1448 * joined to the transaction.
1449 */
1455 }
1457 STATIC int
1461 {
1465 /*
1466 * We're freeing a symlink which fit into
1467 * the inode. Just free the memory used
1468 * to hold the old symlink.
1469 */
1473 XFS_ITRUNCATE_LOG_COUNT);
1479 }
1482 /*
1483 * Zero length symlinks _can_ exist.
1484 */
1488 XFS_DATA_FORK);
1490 }
1492 }
1494 STATIC int
1498 {
1515 }
1521 XFS_INACTIVE_LOG_COUNT);
1528 }
1539 }
1541 STATIC int
1544 {
1557 /* If this is a read-only mount, don't do this (would generate I/O) */
1561 #ifdef HAVE_REFCACHE
1562 /* If we are in the NFS reference cache then don't do this now */
1565 #endif
1576 /* Update linux inode block count after free above */
1579 }
1580 }
1583 }
1585 /*
1586 * xfs_inactive
1587 *
1588 * This is called when the vnode reference count for the vnode
1589 * goes to zero. If the file has been unlinked, then it must
1590 * now be truncated. Also, we clear all of the read-ahead state
1591 * kept for the inode here since the file is now closed.
1592 */
1593 STATIC int
1597 {
1600 xfs_bmap_free_t free_list;
1601 xfs_fsblock_t first_block;
1613 /*
1614 * If the inode is already free, then there can be nothing
1615 * to clean up here.
1616 */
1621 }
1623 /*
1624 * Only do a truncate if it's a regular file with
1625 * some actual space in it. It's OK to look at the
1626 * inode's fields without the lock because we're the
1627 * only one with a reference to the inode.
1628 */
1639 }
1643 /* If this is a read-only mount, don't do this (would generate I/O) */
1657 /* Update linux inode block count after free above */
1660 }
1662 }
1671 /*
1672 * Do the xfs_itruncate_start() call before
1673 * reserving any log space because itruncate_start
1674 * will call into the buffer cache and we can't
1675 * do that within a transaction.
1676 */
1684 XFS_ITRUNCATE_LOG_COUNT);
1686 /* Don't call itruncate_cleanup */
1691 }
1697 /*
1698 * normally, we have to run xfs_itruncate_finish sync.
1699 * But if filesystem is wsync and we're in the inactive
1700 * path, then we know that nlink == 0, and that the
1701 * xaction that made nlink == 0 is permanently committed
1702 * since xfs_remove runs as a synchronous transaction.
1703 */
1712 }
1715 /*
1716 * If we get an error while cleaning up a
1717 * symlink we bail out.
1718 */
1726 }
1734 XFS_INACTIVE_LOG_COUNT);
1739 }
1744 }
1746 /*
1747 * If there are attributes associated with the file
1748 * then blow them away now. The code calls a routine
1749 * that recursively deconstructs the attribute fork.
1750 * We need to just commit the current transaction
1751 * because we can't use it for xfs_attr_inactive().
1752 */
1755 /*
1756 * If we got an error, the transaction is already
1757 * cancelled, and the inode is unlocked. Just get out.
1758 */
1763 }
1765 /*
1766 * Free the inode.
1767 */
1771 /*
1772 * If we fail to free the inode, shut down. The cancel
1773 * might do that, we need to make sure. Otherwise the
1774 * inode might be lost for a long time or forever.
1775 */
1781 }
1784 /*
1785 * Credit the quota account(s). The inode is gone.
1786 */
1789 /*
1790 * Just ignore errors at this point. There is
1791 * nothing we can do except to try to keep going.
1792 */
1796 }
1797 /*
1798 * Release the dquots held by inode, if any.
1799 */
1804 out:
1806 }
1809 /*
1810 * xfs_lookup
1811 */
1812 STATIC int
1820 {
1822 xfs_ino_t e_inum;
1824 uint lock_mode;
1840 }
1843 }
1846 /*
1847 * xfs_create (create a new file).
1848 */
1849 STATIC int
1856 {
1863 xfs_dev_t rdev;
1865 xfs_bmap_free_t free_list;
1866 xfs_fsblock_t first_block;
1867 boolean_t dp_joined_to_trans;
1869 uint cancel_flags;
1871 xfs_prid_t prid;
1873 uint resblks;
1896 }
1901 /* Return through std_return after this point. */
1908 else
1911 /*
1912 * Make sure that we have allocated dquot(s) on disk.
1913 */
1926 /*
1927 * Initially assume that the file does not exist and
1928 * reserve the resources for that case. If that is not
1929 * the case we'll drop the one we have and get a more
1930 * appropriate transaction later.
1931 */
1938 }
1943 }
1951 /*
1952 * Reserve disk quota and the inode.
1953 */
1968 }
1971 /*
1972 * At this point, we've gotten a newly allocated inode.
1973 * It is locked (and joined to the transaction).
1974 */
1978 /*
1979 * Now we join the directory inode to the transaction.
1980 * We do not do it earlier because xfs_dir_ialloc
1981 * might commit the previous transaction (and release
1982 * all the locks).
1983 */
1995 }
1999 /*
2000 * If this is a synchronous mount, make sure that the
2001 * create transaction goes to disk before returning to
2002 * the user.
2003 */
2006 }
2010 /*
2011 * Attach the dquot(s) to the inodes and modify them incore.
2012 * These ids of the inode couldn't have changed since the new
2013 * inode has been locked ever since it was created.
2014 */
2017 /*
2018 * xfs_trans_commit normally decrements the vnode ref count
2019 * when it unlocks the inode. Since we want to return the
2020 * vnode to the caller, we bump the vnode ref count now.
2021 */
2029 }
2036 }
2041 /*
2042 * Propagate the fact that the vnode changed after the
2043 * xfs_inode locks have been released.
2044 */
2049 /* Fallthrough to std_return with error = 0 */
2051 std_return:
2054 DM_EVENT_POSTCREATE)) {
2060 }
2063 abort_return:
2065 /* FALLTHROUGH */
2067 error_return:
2078 abort_rele:
2079 /*
2080 * Wait until after the current transaction is aborted to
2081 * release the inode. This prevents recursive transactions
2082 * and deadlocks from xfs_inactive.
2083 */
2092 }
2094 #ifdef DEBUG
2095 /*
2096 * Some counters to see if (and how often) we are hitting some deadlock
2097 * prevention code paths.
2098 */
2103 #endif
2105 /*
2106 * The following routine will lock the inodes associated with the
2107 * directory and the named entry in the directory. The locks are
2108 * acquired in increasing inode number.
2109 *
2110 * If the entry is "..", then only the directory is locked. The
2111 * vnode ref count will still include that from the .. entry in
2112 * this case.
2113 *
2114 * There is a deadlock we need to worry about. If the locked directory is
2115 * in the AIL, it might be blocking up the log. The next inode we lock
2116 * could be already locked by another thread waiting for log space (e.g
2117 * a permanent log reservation with a long running transaction (see
2118 * xfs_itruncate_finish)). To solve this, we must check if the directory
2119 * is in the ail and use lock_nowait. If we can't lock, we need to
2120 * drop the inode lock on the directory and try again. xfs_iunlock will
2121 * potentially push the tail if we were holding up the log.
2122 */
2123 STATIC int
2128 {
2130 xfs_ino_t e_inum;
2134 #ifdef DEBUG
2135 xfs_rm_locks++;
2136 #endif
2139 again:
2146 /*
2147 * We want to lock in increasing inum. Since we've already
2148 * acquired the lock on the directory, we may need to release
2149 * if if the inum of the entry turns out to be less.
2150 */
2152 /*
2153 * We are already in the right order, so just
2154 * lock on the inode of the entry.
2155 * We need to use nowait if dp is in the AIL.
2156 */
2161 attempts++;
2162 #ifdef DEBUG
2163 xfs_rm_attempts++;
2164 #endif
2166 /*
2167 * Unlock dp and try again.
2168 * xfs_iunlock will try to push the tail
2169 * if the inode is in the AIL.
2170 */
2176 #ifdef DEBUG
2177 xfs_rm_lock_delays++;
2178 #endif
2179 }
2181 }
2184 }
2191 }
2192 /* else e_inum == dp->i_ino */
2193 /* This can happen if we're asked to lock /x/..
2194 * the entry is "..", which is also the parent directory.
2195 */
2198 }
2200 #ifdef DEBUG
2206 #endif
2208 /*
2209 * The following routine will lock n inodes in exclusive mode.
2210 * We assume the caller calls us with the inodes in i_ino order.
2211 *
2212 * We need to detect deadlock where an inode that we lock
2213 * is in the AIL and we start waiting for another inode that is locked
2214 * by a thread in a long running transaction (such as truncate). This can
2215 * result in deadlock since the long running trans might need to wait
2216 * for the inode we just locked in order to push the tail and free space
2217 * in the log.
2218 */
2219 void
2224 uint lock_mode)
2225 {
2237 }
2239 again:
2246 /*
2247 * If try_lock is not set yet, make sure all locked inodes
2248 * are not in the AIL.
2249 * If any are, set try_lock to be used later.
2250 */
2256 try_lock++;
2257 }
2258 }
2259 }
2261 /*
2262 * If any of the previous locks we have locked is in the AIL,
2263 * we must TRY to get the second and subsequent locks. If
2264 * we can't get any, we must release all we have
2265 * and try again.
2266 */
2269 /* try_lock must be 0 if i is 0. */
2270 /*
2271 * try_lock means we have an inode locked
2272 * that is in the AIL.
2273 */
2276 attempts++;
2278 /*
2279 * Unlock all previous guys and try again.
2280 * xfs_iunlock will try to push the tail
2281 * if the inode is in the AIL.
2282 */
2286 /*
2287 * Check to see if we've already
2288 * unlocked this one.
2289 * Not the first one going back,
2290 * and the inode ptr is the same.
2291 */
2297 }
2301 #ifdef DEBUG
2302 xfs_lock_delays++;
2303 #endif
2304 }
2308 }
2311 }
2312 }
2314 #ifdef DEBUG
2320 xfs_locked_n++;
2321 }
2322 #endif
2323 }
2325 #ifdef DEBUG
2326 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2329 #define REMOVE_DEBUG_TRACE(x)
2333 /*
2334 * xfs_remove
2335 *
2336 */
2337 STATIC int
2342 {
2349 xfs_bmap_free_t free_list;
2350 xfs_fsblock_t first_block;
2355 uint resblks;
2375 }
2377 /* From this point on, return through std_return */
2380 /*
2381 * We need to get a reference to ip before we get our log
2382 * reservation. The reason for this is that we cannot call
2383 * xfs_iget for an inode for which we do not have a reference
2384 * once we've acquired a log reservation. This is because the
2385 * inode we are trying to get might be in xfs_inactive going
2386 * for a log reservation. Since we'll have to wait for the
2387 * inactive code to complete before returning from xfs_iget,
2388 * we need to make sure that we don't have log space reserved
2389 * when we call xfs_iget. Instead we get an unlocked reference
2390 * to the inode before getting our log reservation.
2391 */
2396 }
2411 }
2415 /*
2416 * We try to get the real space reservation first,
2417 * allowing for directory btree deletion(s) implying
2418 * possible bmap insert(s). If we can't get the space
2419 * reservation then we use 0 instead, and avoid the bmap
2420 * btree insert(s) in the directory code by, if the bmap
2421 * insert tries to happen, instead trimming the LAST
2422 * block from the directory.
2423 */
2431 }
2438 }
2446 }
2448 /*
2449 * At this point, we've gotten both the directory and the entry
2450 * inodes locked.
2451 */
2454 /*
2455 * Increment vnode ref count only in this case since
2456 * there's an extra vnode reference in the case where
2457 * dp == ip.
2458 */
2461 }
2463 /*
2464 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2465 */
2473 }
2483 }
2485 /* Determine if this is the last link while
2486 * we are in the transaction.
2487 */
2490 /*
2491 * Take an extra ref on the inode so that it doesn't
2492 * go to xfs_inactive() from within the commit.
2493 */
2496 /*
2497 * If this is a synchronous mount, make sure that the
2498 * remove transaction goes to disk before returning to
2499 * the user.
2500 */
2503 }
2509 }
2515 }
2517 /*
2518 * Before we drop our extra reference to the inode, purge it
2519 * from the refcache if it is there. By waiting until afterwards
2520 * to do the IRELE, we ensure that we won't go inactive in the
2521 * xfs_refcache_purge_ip routine (although that would be OK).
2522 */
2527 /*
2528 * Let interposed file systems know about removed links.
2529 */
2534 /* Fall through to std_return with error = 0 */
2535 std_return:
2537 DM_EVENT_POSTREMOVE)) {
2542 }
2545 error1:
2551 error_rele:
2552 /*
2553 * In this case make sure to not release the inode until after
2554 * the current transaction is aborted. Releasing it beforehand
2555 * can cause us to go to xfs_inactive and start a recursive
2556 * transaction which can easily deadlock with the current one.
2557 */
2562 /*
2563 * Before we drop our extra reference to the inode, purge it
2564 * from the refcache if it is there. By waiting until afterwards
2565 * to do the IRELE, we ensure that we won't go inactive in the
2566 * xfs_refcache_purge_ip routine (although that would be OK).
2567 */
2573 }
2576 /*
2577 * xfs_link
2578 *
2579 */
2580 STATIC int
2586 {
2592 xfs_bmap_free_t free_list;
2593 xfs_fsblock_t first_block;
2622 }
2624 /* Return through std_return after this point. */
2641 }
2645 }
2653 }
2657 /*
2658 * Increment vnode ref counts since xfs_trans_commit &
2659 * xfs_trans_cancel will both unlock the inodes and
2660 * decrement the associated ref counts.
2661 */
2667 /*
2668 * If the source has too many links, we can't make any more to it.
2669 */
2673 }
2675 /*
2676 * If we are using project inheritance, we only allow hard link
2677 * creation in our tree when the project IDs are the same; else
2678 * the tree quota mechanism could be circumvented.
2679 */
2684 }
2694 resblks);
2704 }
2706 /*
2707 * If this is a synchronous mount, make sure that the
2708 * link transaction goes to disk before returning to
2709 * the user.
2710 */
2713 }
2719 }
2724 }
2726 /* Fall through to std_return with error = 0. */
2727 std_return:
2729 DM_EVENT_POSTLINK)) {
2734 }
2737 abort_return:
2739 /* FALLTHROUGH */
2741 error_return:
2744 }
2745 /*
2746 * xfs_mkdir
2747 *
2748 */
2749 STATIC int
2756 {
2766 xfs_bmap_free_t free_list;
2767 xfs_fsblock_t first_block;
2769 boolean_t dp_joined_to_trans;
2772 xfs_prid_t prid;
2774 uint resblks;
2799 }
2801 /* Return through std_return after this point. */
2811 else
2814 /*
2815 * Make sure that we have allocated dquot(s) on disk.
2816 */
2831 XFS_TRANS_PERM_LOG_RES,
2832 XFS_MKDIR_LOG_COUNT);
2833 }
2838 }
2842 /*
2843 * Check for directory link count overflow.
2844 */
2848 }
2850 /*
2851 * Reserve disk quota and the inode.
2852 */
2860 /*
2861 * create the directory inode.
2862 */
2870 }
2873 /*
2874 * Now we add the directory inode to the transaction.
2875 * We waited until now since xfs_dir_ialloc might start
2876 * a new transaction. Had we joined the transaction
2877 * earlier, the locks might have gotten released.
2878 */
2891 }
2894 /*
2895 * Bump the in memory version number of the parent directory
2896 * so that other processes accessing it will recognize that
2897 * the directory has changed.
2898 */
2917 /*
2918 * Attach the dquots to the new inode and modify the icount incore.
2919 */
2922 /*
2923 * If this is a synchronous mount, make sure that the
2924 * mkdir transaction goes to disk before returning to
2925 * the user.
2926 */
2929 }
2935 }
2942 }
2944 /* Fall through to std_return with error = 0 or errno from
2945 * xfs_trans_commit. */
2947 std_return:
2950 DM_EVENT_POSTCREATE)) {
2954 DM_RIGHT_NULL,
2957 }
2960 error2:
2961 error1:
2963 abort_return:
2965 error_return:
2972 }
2975 }
2978 /*
2979 * xfs_rmdir
2980 *
2981 */
2982 STATIC int
2987 {
2994 xfs_bmap_free_t free_list;
2995 xfs_fsblock_t first_block;
3002 uint resblks;
3021 }
3023 /* Return through std_return after this point. */
3027 /*
3028 * We need to get a reference to cdp before we get our log
3029 * reservation. The reason for this is that we cannot call
3030 * xfs_iget for an inode for which we do not have a reference
3031 * once we've acquired a log reservation. This is because the
3032 * inode we are trying to get might be in xfs_inactive going
3033 * for a log reservation. Since we'll have to wait for the
3034 * inactive code to complete before returning from xfs_iget,
3035 * we need to make sure that we don't have log space reserved
3036 * when we call xfs_iget. Instead we get an unlocked reference
3037 * to the inode before getting our log reservation.
3038 */
3043 }
3047 /*
3048 * Get the dquots for the inodes.
3049 */
3057 }
3061 /*
3062 * We try to get the real space reservation first,
3063 * allowing for directory btree deletion(s) implying
3064 * possible bmap insert(s). If we can't get the space
3065 * reservation then we use 0 instead, and avoid the bmap
3066 * btree insert(s) in the directory code by, if the bmap
3067 * insert tries to happen, instead trimming the LAST
3068 * block from the directory.
3069 */
3077 }
3083 }
3086 /*
3087 * Now lock the child directory inode and the parent directory
3088 * inode in the proper order. This will take care of validating
3089 * that the directory entry for the child directory inode has
3090 * not changed while we were obtaining a log reservation.
3091 */
3097 }
3101 /*
3102 * Only increment the parent directory vnode count if
3103 * we didn't bump it in looking up cdp. The only time
3104 * we don't bump it is when we're looking up ".".
3105 */
3107 }
3116 }
3120 }
3129 /*
3130 * Bump the in memory generation count on the parent
3131 * directory so that other can know that it has changed.
3132 */
3135 /*
3136 * Drop the link from cdp's "..".
3137 */
3141 }
3143 /*
3144 * Drop the link from dp to cdp.
3145 */
3149 }
3151 /*
3152 * Drop the "." link from cdp to self.
3153 */
3157 }
3159 /* Determine these before committing transaction */
3162 /*
3163 * Take an extra ref on the child vnode so that it
3164 * does not go to xfs_inactive() from within the commit.
3165 */
3168 /*
3169 * If this is a synchronous mount, make sure that the
3170 * rmdir transaction goes to disk before returning to
3171 * the user.
3172 */
3175 }
3181 XFS_TRANS_ABORT));
3184 }
3190 }
3193 /*
3194 * Let interposed file systems know about removed links.
3195 */
3200 /* Fall through to std_return with error = 0 or the errno
3201 * from xfs_trans_commit. */
3202 std_return:
3209 }
3212 error1:
3215 /* FALLTHROUGH */
3217 error_return:
3220 }
3223 /*
3224 * Read dp's entries starting at uiop->uio_offset and translate them into
3225 * bufsize bytes worth of struct dirents starting at bufbase.
3226 */
3227 STATIC int
3233 {
3237 uint lock_mode;
3250 }
3253 STATIC int
3261 {
3268 xfs_bmap_free_t free_list;
3269 xfs_fsblock_t first_block;
3270 boolean_t dp_joined_to_trans;
3272 uint cancel_flags;
3274 xfs_fileoff_t first_fsb;
3275 xfs_filblks_t fs_blocks;
3278 xfs_daddr_t d;
3283 xfs_prid_t prid;
3285 uint resblks;
3306 /*
3307 * Check component lengths of the target path name.
3308 */
3317 /*
3318 * Skip any slashes.
3319 */
3321 total++;
3322 path++;
3323 }
3325 /*
3326 * Count up to the next slash or end of path.
3327 * Error out if the component is bigger than MAXNAMELEN.
3328 */
3333 }
3334 }
3335 }
3336 }
3344 }
3346 /* Return through std_return after this point. */
3353 else
3356 /*
3357 * Make sure that we have allocated dquot(s) on disk.
3358 */
3367 /*
3368 * The symlink will fit into the inode data fork?
3369 * There can't be any attributes so we get the whole variable part.
3370 */
3373 else
3382 }
3387 }
3391 /*
3392 * Check whether the directory allows new symlinks or not.
3393 */
3397 }
3399 /*
3400 * Reserve disk quota : blocks and inode.
3401 */
3406 /*
3407 * Check for ability to enter directory entry, if no space reserved.
3408 */
3412 /*
3413 * Initialize the bmap freelist prior to calling either
3414 * bmapi or the directory create code.
3415 */
3418 /*
3419 * Allocate an inode for the symlink.
3420 */
3427 }
3434 /*
3435 * Also attach the dquot(s) to it, if applicable.
3436 */
3441 /*
3442 * If the symlink will fit into the inode, write it inline.
3443 */
3449 /*
3450 * The inode was initially created in extent format.
3451 */
3468 }
3484 }
3491 }
3492 }
3494 /*
3495 * Create the directory entry for the symlink.
3496 */
3504 /*
3505 * Bump the in memory version number of the parent directory
3506 * so that other processes accessing it will recognize that
3507 * the directory has changed.
3508 */
3511 /*
3512 * If this is a synchronous mount, make sure that the
3513 * symlink transaction goes to disk before returning to
3514 * the user.
3515 */
3518 }
3520 /*
3521 * xfs_trans_commit normally decrements the vnode ref count
3522 * when it unlocks the inode. Since we want to return the
3523 * vnode to the caller, we bump the vnode ref count now.
3524 */
3530 }
3535 /* Fall through to std_return with error = 0 or errno from
3536 * xfs_trans_commit */
3537 std_return:
3539 DM_EVENT_POSTSYMLINK)) {
3545 }
3553 }
3556 error2:
3558 error1:
3561 error_return:
3568 }
3571 }
3574 /*
3575 * xfs_fid2
3576 *
3577 * A fid routine that takes a pointer to a previously allocated
3578 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3579 */
3580 STATIC int
3584 {
3596 /*
3597 * use memcpy because the inode is a long long and there's no
3598 * assurance that xfid->fid_ino is properly aligned.
3599 */
3604 }
3607 /*
3608 * xfs_rwlock
3609 */
3610 int
3613 bhv_vrwlock_t locktype)
3614 {
3632 }
3635 }
3638 /*
3639 * xfs_rwunlock
3640 */
3641 void
3644 bhv_vrwlock_t locktype)
3645 {
3654 /*
3655 * In the write case, we may have added a new entry to
3656 * the reference cache. This might store a pointer to
3657 * an inode to be released in this inode. If it is there,
3658 * clear the pointer and release the inode after unlocking
3659 * this one.
3660 */
3666 }
3668 }
3670 STATIC int
3674 {
3687 /*
3688 * Bypass inodes which have already been cleaned by
3689 * the inode flush clustering code inside xfs_iflush
3690 */
3698 xfs_lsn_t sync_lsn;
3711 }
3712 }
3714 /*
3715 * We make this non-blocking if the inode is contended,
3716 * return EAGAIN to indicate to the caller that they
3717 * did not succeed. This prevents the flush path from
3718 * blocking on inodes inside another operation right
3719 * now, they get caught later by xfs_sync.
3720 */
3734 }
3737 }
3741 else
3746 }
3749 }
3751 int
3754 u_int evmask,
3755 u_int16_t state,
3757 {
3777 }
3789 }
3791 STATIC int
3794 {
3805 /* bad inode, get out here ASAP */
3809 }
3815 /*
3816 * Make sure the atime in the XFS inode is correct before freeing the
3817 * Linux inode.
3818 */
3821 /* If we have nothing to flush with this inode then complete the
3822 * teardown now, otherwise break the link between the xfs inode
3823 * and the linux inode and clean up the xfs inode later. This
3824 * avoids flushing the inode to disk during the delete operation
3825 * itself.
3826 */
3834 /* Protect sync from us */
3840 }
3842 }
3844 int
3849 {
3857 /* The hash lock here protects a thread in xfs_iget_core from
3858 * racing with us on linking the inode back with a vnode.
3859 * Once we have the XFS_IRECLAIM flag set it will not touch
3860 * us.
3861 */
3869 }
3871 }
3875 /*
3876 * If the inode is still dirty, then flush it out. If the inode
3877 * is not in the AIL, then it will be OK to flush it delwri as
3878 * long as xfs_iflush() does not keep any references to the inode.
3879 * We leave that decision up to xfs_iflush() since it has the
3880 * knowledge of whether it's OK to simply do a delwri flush of
3881 * the inode or whether we need to wait until the inode is
3882 * pulled from the AIL.
3883 * We get the flush lock regardless, though, just to make sure
3884 * we don't free it while it is being flushed.
3885 */
3890 }
3896 /*
3897 * If we hit an error, typically because of filesystem
3898 * shutdown, we don't need to let vn_reclaim to know
3899 * because we're gonna reclaim the inode anyway.
3900 */
3904 }
3906 }
3913 /*
3914 * We are not interested in doing an iflush if we're
3915 * in the process of shutting down the filesystem forcibly.
3916 * So, just reclaim the inode.
3917 */
3920 }
3922 reclaim:
3925 }
3927 int
3929 {
3945 }
3946 }
3949 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3953 }
3956 }
3960 }
3962 /*
3963 * xfs_alloc_file_space()
3964 * This routine allocates disk space for the given file.
3965 *
3966 * If alloc_type == 0, this request is for an ALLOCSP type
3967 * request which will change the file size. In this case, no
3968 * DMAPI event will be generated by the call. A TRUNCATE event
3969 * will be generated later by xfs_setattr.
3970 *
3971 * If alloc_type != 0, this request is for a RESVSP type
3972 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3973 * lower block boundary byte address is less than the file's
3974 * length.
3975 *
3976 * RETURNS:
3977 * 0 on success
3978 * errno on error
3979 *
3980 */
3981 STATIC int
3984 xfs_off_t offset,
3985 xfs_off_t len,
3988 {
3990 xfs_off_t count;
3991 xfs_filblks_t allocated_fsb;
3992 xfs_filblks_t allocatesize_fsb;
3994 xfs_fileoff_t startoffset_fsb;
3995 xfs_fsblock_t firstfsb;
4002 xfs_bmap_free_t free_list;
4018 }
4034 /* Generate a DMAPI event if needed. */
4038 xfs_off_t end_dmi_offset;
4048 }
4050 /*
4051 * Allocate file space until done or until there is an error
4052 */
4053 retry:
4057 /*
4058 * Determine space reservations for data/realtime.
4059 */
4072 }
4084 }
4086 /*
4087 * Allocate and setup the transaction.
4088 */
4092 XFS_TRANS_PERM_LOG_RES,
4093 XFS_WRITE_LOG_COUNT);
4094 /*
4095 * Check for running out of space
4096 */
4098 /*
4099 * Free the transaction structure.
4100 */
4104 }
4114 /*
4115 * Issue the xfs_bmapi() call to allocate the blocks
4116 */
4124 }
4126 /*
4127 * Complete the transaction
4128 */
4132 }
4138 }
4145 }
4149 }
4150 dmapi_enospc_check:
4160 /* else fall through with error from XFS_SEND_DATA */
4161 }
4173 }
4175 /*
4176 * Zero file bytes between startoff and endoff inclusive.
4177 * The iolock is held exclusive and no blocks are buffered.
4178 */
4179 STATIC int
4182 xfs_off_t startoff,
4183 xfs_off_t endoff)
4184 {
4185 xfs_bmbt_irec_t imap;
4186 xfs_fileoff_t offset_fsb;
4187 xfs_off_t lastoffset;
4188 xfs_off_t offset;
4224 }
4236 }
4237 }
4240 }
4242 /*
4243 * xfs_free_file_space()
4244 * This routine frees disk space for the given file.
4245 *
4246 * This routine is only called by xfs_change_file_space
4247 * for an UNRESVSP type call.
4248 *
4249 * RETURNS:
4250 * 0 on success
4251 * errno on error
4252 *
4253 */
4254 STATIC int
4257 xfs_off_t offset,
4258 xfs_off_t len,
4260 {
4264 xfs_off_t end_dmi_offset;
4265 xfs_fileoff_t endoffset_fsb;
4267 xfs_fsblock_t firstfsb;
4268 xfs_bmap_free_t free_list;
4269 xfs_off_t ilen;
4270 xfs_bmbt_irec_t imap;
4271 xfs_off_t ioffset;
4275 uint resblks;
4278 xfs_fileoff_t startoffset_fsb;
4308 }
4315 }
4329 }
4331 /*
4332 * Need to zero the stuff we're not freeing, on disk.
4333 * If its a realtime file & can't use unwritten extents then we
4334 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4335 * will take care of it for us.
4336 */
4345 xfs_daddr_t block;
4352 }
4361 mod++;
4364 }
4365 }
4367 /*
4368 * One contiguous piece to clear
4369 */
4372 /*
4373 * Some full blocks, possibly two pieces to clear
4374 */
4383 }
4385 /*
4386 * free file space until done or until there is an error
4387 */
4391 /*
4392 * allocate and setup the transaction
4393 */
4396 resblks,
4399 XFS_TRANS_PERM_LOG_RES,
4400 XFS_WRITE_LOG_COUNT);
4402 /*
4403 * check for running out of space
4404 */
4406 /*
4407 * Free the transaction structure.
4408 */
4412 }
4416 XFS_QMOPT_RES_REGBLKS);
4423 /*
4424 * issue the bunmapi() call to free the blocks
4425 */
4432 }
4434 /*
4435 * complete the transaction
4436 */
4440 }
4444 }
4446 out_unlock_iolock:
4451 error0:
4453 error1:
4456 XFS_ILOCK_EXCL);
4458 }
4460 /*
4461 * xfs_change_file_space()
4462 * This routine allocates or frees disk space for the given file.
4463 * The user specified parameters are checked for alignment and size
4464 * limitations.
4465 *
4466 * RETURNS:
4467 * 0 on success
4468 * errno on error
4469 *
4470 */
4471 int
4476 xfs_off_t offset,
4479 {
4482 xfs_fsize_t fsize;
4486 xfs_off_t startoffset;
4487 xfs_off_t llen;
4489 bhv_vattr_t va;
4498 /*
4499 * must be a regular file and have write permission
4500 */
4509 }
4524 }
4539 /*
4540 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4541 * file space.
4542 * These calls do NOT zero the data space allocated to the file,
4543 * nor do they change the file size.
4544 *
4545 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4546 * space.
4547 * These calls cause the new file data to be zeroed and the file
4548 * size to be changed.
4549 */
4565 attr_flags)))
4578 }
4594 }
4596 /*
4597 * update the inode timestamp, mode, and prealloc flag bits
4598 */
4603 /* ASSERT(0); */
4606 }
4616 /*
4617 * Note that we don't have to worry about mandatory
4618 * file locking being disabled here because we only
4619 * clear the S_ISGID bit if the Group execute bit is
4620 * on, but if it was on then mandatory locking wouldn't
4621 * have been enabled.
4622 */
4627 }
4641 }
4643 bhv_vnodeops_t xfs_vnodeops = {
4648 #ifdef HAVE_SENDFILE
4650 #endif
4651 #ifdef HAVE_SPLICE
4654 #endif
4688 };