| blob | 7027ae68ee38e6491897eb88033886b7c6fb0ea5 |
1 /*
2 * Copyright (c) 2000-2005 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 */
19 #include <linux/capability.h>
62 /*
63 * The maximum pathlen is 1024 bytes. Since the minimum file system
64 * blocksize is 512 bytes, we can get a max of 2 extents back from
65 * bmapi.
66 */
67 #define SYMLINK_MAPS 2
69 /*
70 * For xfs, we check that the file isn't too big to be opened by this kernel.
71 * No other open action is required for regular files. Devices are handled
72 * through the specfs file system, pipes through fifofs. Device and
73 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
74 * when a new vnode is first looked up or created.
75 */
76 STATIC int
80 {
91 /*
92 * If it's a directory with any blocks, read-ahead block 0
93 * as we're almost certain to have the next operation be a read there.
94 */
100 }
102 }
105 /*
106 * xfs_getattr
107 */
108 STATIC int
114 {
138 #if XFS_BIG_INUMS
140 #endif
143 /*
144 * Quick exit for non-stat callers
145 */
151 /*
152 * Copy from in-core inode.
153 */
159 /*
160 * Check vnode type block/char vs. everything else.
161 */
175 /*
176 * If the file blocks are being allocated from a
177 * realtime partition, then return the inode's
178 * realtime extent size or the realtime volume's
179 * extent size.
180 */
184 }
186 }
194 /*
195 * Exit for stat callers. See if any of the rest of the fields
196 * to be filled in are needed.
197 */
203 /*
204 * Convert di_flags to xflags.
205 */
208 /*
209 * Exit for inode revalidate. See if any of the rest of
210 * the fields to be filled in are needed.
211 */
227 else
231 all_done:
235 }
238 /*
239 * xfs_setattr
240 */
241 int
247 {
253 uint lock_flags;
271 /*
272 * Cannot set certain attributes.
273 */
277 }
285 /*
286 * Timestamps do not need to be logged and hence do not
287 * need to be done within a transaction.
288 */
296 }
301 /*
302 * If disk quotas is on, we make sure that the dquots do exist on disk,
303 * before we start any other transactions. Trying to do this later
304 * is messy. We don't care to take a readlock to look at the ids
305 * in inode here, because we can't hold it across the trans_reserve.
306 * If the IDs do change before we take the ilock, we're covered
307 * because the i_*dquot fields will get updated anyway.
308 */
318 }
324 }
330 }
331 /*
332 * We take a reference when we initialize udqp and gdqp,
333 * so it is important that we never blindly double trip on
334 * the same variable. See xfs_create() for an example.
335 */
342 }
344 /*
345 * For the other attributes, we acquire the inode lock and
346 * first do an error checking pass.
347 */
363 }
364 }
374 }
375 }
378 }
382 /* boolean: are we the file owner? */
385 /*
386 * Change various properties of a file.
387 * Only the owner or users with CAP_FOWNER
388 * capability may do these things.
389 */
393 /*
394 * CAP_FOWNER overrides the following restrictions:
395 *
396 * The user ID of the calling process must be equal
397 * to the file owner ID, except in cases where the
398 * CAP_FSETID capability is applicable.
399 */
403 }
405 /*
406 * CAP_FSETID overrides the following restrictions:
407 *
408 * The effective user ID of the calling process shall match
409 * the file owner when setting the set-user-ID and
410 * set-group-ID bits on that file.
411 *
412 * The effective group ID or one of the supplementary group
413 * IDs of the calling process shall match the group owner of
414 * the file when setting the set-group-ID bit on that file
415 */
424 #if 0
425 /* Linux allows this, Irix doesn't. */
428 #endif
431 }
432 }
434 /*
435 * Change file ownership. Must be the owner or privileged.
436 * If the system was configured with the "restricted_chown"
437 * option, the owner is not permitted to give away the file,
438 * and can change the group id only to a group of which he
439 * or she is a member.
440 */
442 /*
443 * These IDs could have changed since we last looked at them.
444 * But, we're assured that if the ownership did change
445 * while we didn't have the inode locked, inode's dquot(s)
446 * would have changed also.
447 */
454 iprojid;
456 /*
457 * CAP_CHOWN overrides the following restrictions:
458 *
459 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
460 * shall override the restriction that a process cannot
461 * change the user ID of a file it owns and the restriction
462 * that the group ID supplied to the chown() function
463 * shall be equal to either the group ID or one of the
464 * supplementary group IDs of the calling process.
465 */
472 }
473 /*
474 * Do a quota reservation only if uid/projid/gid is actually
475 * going to change.
476 */
486 }
487 }
489 /*
490 * Truncate file. Must have write permission and not be a directory.
491 */
493 /* Short circuit the truncate case for zero length files */
502 }
510 }
511 /*
512 * Make sure that the dquots are attached to the inode.
513 */
516 }
518 /*
519 * Change file access or modified times.
520 */
527 }
528 }
529 }
531 /*
532 * Change extent size or realtime flag.
533 */
535 /*
536 * Can't change extent size if any extents are allocated.
537 */
543 }
545 /*
546 * Can't change realtime flag if any extents are allocated.
547 */
554 }
555 /*
556 * Extent size must be a multiple of the appropriate block
557 * size, if set at all.
558 */
560 xfs_extlen_t size;
569 }
573 }
574 }
575 /*
576 * If realtime flag is set then must have realtime data.
577 */
585 }
586 }
588 /*
589 * Can't modify an immutable/append-only file unless
590 * we have appropriate permission.
591 */
600 }
601 }
603 /*
604 * Now we can make the changes. Before we join the inode
605 * to the transaction, if XFS_AT_SIZE is set then take care of
606 * the part of the truncation that must be done without the
607 * inode lock. This needs to be done before joining the inode
608 * to the transaction, because the inode cannot be unlocked
609 * once it is a part of the transaction.
610 */
616 }
626 }
630 XFS_TRANS_PERM_LOG_RES,
631 XFS_ITRUNCATE_LOG_COUNT))) {
636 }
639 }
644 }
646 /* determine whether mandatory locking mode changes */
649 /*
650 * Truncate file. Must have write permission and not be a directory.
651 */
658 /*
659 * signal a sync transaction unless
660 * we're truncating an already unlinked
661 * file on a wsync filesystem
662 */
665 XFS_DATA_FORK,
671 }
672 }
673 /*
674 * Have to do this even if the file's size doesn't change.
675 */
677 }
679 /*
680 * Change file access modes.
681 */
688 }
690 /*
691 * Change file ownership. Must be the owner or privileged.
692 * If the system was configured with the "restricted_chown"
693 * option, the owner is not permitted to give away the file,
694 * and can change the group id only to a group of which he
695 * or she is a member.
696 */
698 /*
699 * CAP_FSETID overrides the following restrictions:
700 *
701 * The set-user-ID and set-group-ID bits of a file will be
702 * cleared upon successful return from chown()
703 */
707 }
709 /*
710 * Change the ownerships and register quota modifications
711 * in the transaction.
712 */
719 }
721 }
729 }
731 }
739 }
741 /*
742 * We may have to rev the inode as well as
743 * the superblock version number since projids didn't
744 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
745 */
748 }
752 }
755 /*
756 * Change file access or modified times.
757 */
764 }
770 }
773 }
775 /*
776 * Change XFS-added attributes.
777 */
780 /*
781 * Converting bytes to fs blocks.
782 */
785 }
787 uint di_flags;
789 /* can't set PREALLOC this way, just preserve it */
816 }
819 }
821 }
824 }
826 /*
827 * Change file inode change time only if XFS_AT_CTIME set
828 * AND we have been called by a DMI function.
829 */
836 }
838 /*
839 * Send out timestamp changes that need to be set to the
840 * current time. Not done when called by a DMI function.
841 */
847 /*
848 * If this is a synchronous mount, make sure that the
849 * transaction goes to disk before returning to the user.
850 * This is slightly sub-optimal in that truncates require
851 * two sync transactions instead of one for wsync filesystems.
852 * One for the truncate and one for the timestamps since we
853 * don't want to change the timestamps unless we're sure the
854 * truncate worked. Truncates are less than 1% of the laddis
855 * mix so this probably isn't worth the trouble to optimize.
856 */
863 }
865 /*
866 * If the (regular) file's mandatory locking mode changed, then
867 * notify the vnode. We do this under the inode lock to prevent
868 * racing calls to vop_vnode_change.
869 */
873 mandlock_after);
874 }
878 /*
879 * Release any dquot(s) the inode had kept before chown.
880 */
888 }
895 }
898 abort_return:
900 /* FALLTHROUGH */
901 error_return:
906 }
909 }
911 }
914 /*
915 * xfs_access
916 * Null conversion from vnode mode bits to inode mode bits, as in efs.
917 */
918 STATIC int
923 {
935 }
938 /*
939 * xfs_readlink
940 *
941 */
942 STATIC int
948 {
951 xfs_off_t offset;
958 xfs_daddr_t d;
982 }
986 }
988 /*
989 * See if the symlink is stored inline.
990 */
995 }
997 /*
998 * Symlink not inline. Call bmap to get it in.
999 */
1007 }
1020 }
1027 }
1029 }
1031 error_return:
1034 }
1037 /*
1038 * xfs_fsync
1039 *
1040 * This is called to sync the inode and its data out to disk.
1041 * We need to hold the I/O lock while flushing the data, and
1042 * the inode lock while flushing the inode. The inode lock CANNOT
1043 * be held while flushing the data, so acquire after we're done
1044 * with that.
1045 */
1046 STATIC int
1051 xfs_off_t start,
1052 xfs_off_t stop)
1053 {
1069 /*
1070 * We always need to make sure that the required inode state
1071 * is safe on disk. The vnode might be clean but because
1072 * of committed transactions that haven't hit the disk yet.
1073 * Likewise, there could be unflushed non-transactional
1074 * changes to the inode core that have to go to disk.
1075 *
1076 * The following code depends on one assumption: that
1077 * any transaction that changes an inode logs the core
1078 * because it has to change some field in the inode core
1079 * (typically nextents or nblocks). That assumption
1080 * implies that any transactions against an inode will
1081 * catch any non-transactional updates. If inode-altering
1082 * transactions exist that violate this assumption, the
1083 * code breaks. Right now, it figures that if the involved
1084 * update_* field is clear and the inode is unpinned, the
1085 * inode is clean. Either it's been flushed or it's been
1086 * committed and the commit has hit the disk unpinning the inode.
1087 * (Note that xfs_inode_item_format() called at commit clears
1088 * the update_* fields.)
1089 */
1092 /* If we are flushing data then we care about update_size
1093 * being set, otherwise we care about update_core
1094 */
1098 /*
1099 * Timestamps/size haven't changed since last inode
1100 * flush or inode transaction commit. That means
1101 * either nothing got written or a transaction
1102 * committed which caught the updates. If the
1103 * latter happened and the transaction hasn't
1104 * hit the disk yet, the inode will be still
1105 * be pinned. If it is, force the log.
1106 */
1112 XFS_LOG_FORCE |
1117 /*
1118 * If the inode is not pinned and nothing
1119 * has changed we don't need to flush the
1120 * cache.
1121 */
1123 }
1126 /*
1127 * Kick off a transaction to log the inode
1128 * core to get the updates. Make it
1129 * sync if FSYNC_WAIT is passed in (which
1130 * is done by everybody but specfs). The
1131 * sync transaction will also force the log.
1132 */
1140 }
1143 /*
1144 * Note - it's possible that we might have pushed
1145 * ourselves out of the way during trans_reserve
1146 * which would flush the inode. But there's no
1147 * guarantee that the inode buffer has actually
1148 * gone out yet (it's delwri). Plus the buffer
1149 * could be pinned anyway if it's part of an
1150 * inode in another recent transaction. So we
1151 * play it safe and fire off the transaction anyway.
1152 */
1161 }
1164 /*
1165 * If the log write didn't issue an ordered tag we need
1166 * to flush the disk cache for the data device now.
1167 */
1171 /*
1172 * If this inode is on the RT dev we need to flush that
1173 * cache as well.
1174 */
1177 }
1180 }
1182 /*
1183 * This is called by xfs_inactive to free any blocks beyond eof,
1184 * when the link count isn't zero.
1185 */
1186 STATIC int
1190 {
1193 xfs_fileoff_t end_fsb;
1194 xfs_fileoff_t last_fsb;
1195 xfs_filblks_t map_len;
1197 xfs_bmbt_irec_t imap;
1199 /*
1200 * Figure out if there are any blocks beyond the end
1201 * of the file. If not, then there is nothing to do.
1202 */
1218 /*
1219 * Attach the dquots to the inode up front.
1220 */
1224 /*
1225 * There are blocks after the end of file.
1226 * Free them up now by truncating the file to
1227 * its current size.
1228 */
1231 /*
1232 * Do the xfs_itruncate_start() call before
1233 * reserving any log space because
1234 * itruncate_start will call into the buffer
1235 * cache and we can't
1236 * do that within a transaction.
1237 */
1245 XFS_ITRUNCATE_LOG_COUNT);
1251 }
1255 XFS_IOLOCK_EXCL |
1256 XFS_ILOCK_EXCL);
1261 XFS_DATA_FORK,
1263 /*
1264 * If we get an error at this point we
1265 * simply don't bother truncating the file.
1266 */
1270 XFS_TRANS_ABORT));
1273 XFS_TRANS_RELEASE_LOG_RES,
1274 NULL);
1275 }
1277 }
1279 }
1281 /*
1282 * Free a symlink that has blocks associated with it.
1283 */
1284 STATIC int
1288 {
1293 xfs_fsblock_t first_block;
1294 xfs_bmap_free_t free_list;
1306 /*
1307 * We're freeing a symlink that has some
1308 * blocks allocated to it. Free the
1309 * blocks here. We know that we've got
1310 * either 1 or 2 extents and that we can
1311 * free them all in one bunmapi call.
1312 */
1320 }
1321 /*
1322 * Lock the inode, fix the size, and join it to the transaction.
1323 * Hold it so in the normal path, we still have it locked for
1324 * the second transaction. In the error paths we need it
1325 * held so the cancel won't rele it, see below.
1326 */
1333 /*
1334 * Find the block(s) so we can inval and unmap them.
1335 */
1343 /*
1344 * Invalidate the block(s).
1345 */
1351 }
1352 /*
1353 * Unmap the dead block(s) to the free_list.
1354 */
1359 /*
1360 * Commit the first transaction. This logs the EFI and the inode.
1361 */
1364 /*
1365 * The transaction must have been committed, since there were
1366 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1367 * The new tp has the extent freeing and EFDs.
1368 */
1370 /*
1371 * The first xact was committed, so add the inode to the new one.
1372 * Mark it dirty so it will be logged and moved forward in the log as
1373 * part of every commit.
1374 */
1378 /*
1379 * Get a new, empty transaction to return to our caller.
1380 */
1382 /*
1383 * Commit the transaction containing extent freeing and EFDs.
1384 * If we get an error on the commit here or on the reserve below,
1385 * we need to unlock the inode since the new transaction doesn't
1386 * have the inode attached.
1387 */
1393 }
1394 /*
1395 * Remove the memory for extent descriptions (just bookkeeping).
1396 */
1400 /*
1401 * Put an itruncate log reservation in the new transaction
1402 * for our caller.
1403 */
1408 }
1409 /*
1410 * Return with the inode locked but not joined to the transaction.
1411 */
1415 error1:
1417 error0:
1418 /*
1419 * Have to come here with the inode locked and either
1420 * (held and in the transaction) or (not in the transaction).
1421 * If the inode isn't held then cancel would iput it, but
1422 * that's wrong since this is inactive and the vnode ref
1423 * count is 0 already.
1424 * Cancel won't do anything to the inode if held, but it still
1425 * needs to be locked until the cancel is done, if it was
1426 * joined to the transaction.
1427 */
1433 }
1435 STATIC int
1439 {
1443 /*
1444 * We're freeing a symlink which fit into
1445 * the inode. Just free the memory used
1446 * to hold the old symlink.
1447 */
1451 XFS_ITRUNCATE_LOG_COUNT);
1457 }
1460 /*
1461 * Zero length symlinks _can_ exist.
1462 */
1466 XFS_DATA_FORK);
1468 }
1470 }
1472 /*
1473 *
1474 */
1475 STATIC int
1479 {
1496 }
1502 XFS_INACTIVE_LOG_COUNT);
1509 }
1520 }
1522 STATIC int
1525 {
1536 }
1538 /* If this is a read-only mount, don't do this (would generate I/O) */
1542 #ifdef HAVE_REFCACHE
1543 /* If we are in the NFS reference cache then don't do this now */
1546 #endif
1559 /* Update linux inode block count after free above */
1562 }
1563 }
1566 }
1568 /*
1569 * xfs_inactive
1570 *
1571 * This is called when the vnode reference count for the vnode
1572 * goes to zero. If the file has been unlinked, then it must
1573 * now be truncated. Also, we clear all of the read-ahead state
1574 * kept for the inode here since the file is now closed.
1575 */
1576 STATIC int
1580 {
1583 xfs_bmap_free_t free_list;
1584 xfs_fsblock_t first_block;
1596 /*
1597 * If the inode is already free, then there can be nothing
1598 * to clean up here.
1599 */
1604 }
1606 /*
1607 * Only do a truncate if it's a regular file with
1608 * some actual space in it. It's OK to look at the
1609 * inode's fields without the lock because we're the
1610 * only one with a reference to the inode.
1611 */
1622 }
1626 /* If this is a read-only mount, don't do this (would generate I/O) */
1640 /* Update linux inode block count after free above */
1643 }
1645 }
1654 /*
1655 * Do the xfs_itruncate_start() call before
1656 * reserving any log space because itruncate_start
1657 * will call into the buffer cache and we can't
1658 * do that within a transaction.
1659 */
1667 XFS_ITRUNCATE_LOG_COUNT);
1669 /* Don't call itruncate_cleanup */
1674 }
1680 /*
1681 * normally, we have to run xfs_itruncate_finish sync.
1682 * But if filesystem is wsync and we're in the inactive
1683 * path, then we know that nlink == 0, and that the
1684 * xaction that made nlink == 0 is permanently committed
1685 * since xfs_remove runs as a synchronous transaction.
1686 */
1695 }
1698 /*
1699 * If we get an error while cleaning up a
1700 * symlink we bail out.
1701 */
1709 }
1717 XFS_INACTIVE_LOG_COUNT);
1722 }
1727 }
1729 /*
1730 * If there are attributes associated with the file
1731 * then blow them away now. The code calls a routine
1732 * that recursively deconstructs the attribute fork.
1733 * We need to just commit the current transaction
1734 * because we can't use it for xfs_attr_inactive().
1735 */
1738 /*
1739 * If we got an error, the transaction is already
1740 * cancelled, and the inode is unlocked. Just get out.
1741 */
1746 }
1748 /*
1749 * Free the inode.
1750 */
1754 /*
1755 * If we fail to free the inode, shut down. The cancel
1756 * might do that, we need to make sure. Otherwise the
1757 * inode might be lost for a long time or forever.
1758 */
1764 }
1767 /*
1768 * Credit the quota account(s). The inode is gone.
1769 */
1772 /*
1773 * Just ignore errors at this point. There is
1774 * nothing we can do except to try to keep going.
1775 */
1779 }
1780 /*
1781 * Release the dquots held by inode, if any.
1782 */
1787 out:
1789 }
1792 /*
1793 * xfs_lookup
1794 */
1795 STATIC int
1803 {
1805 xfs_ino_t e_inum;
1807 uint lock_mode;
1823 }
1826 }
1829 /*
1830 * xfs_create (create a new file).
1831 */
1832 STATIC int
1839 {
1846 xfs_dev_t rdev;
1848 xfs_bmap_free_t free_list;
1849 xfs_fsblock_t first_block;
1850 boolean_t dp_joined_to_trans;
1852 uint cancel_flags;
1854 xfs_prid_t prid;
1856 uint resblks;
1879 }
1884 /* Return through std_return after this point. */
1891 else
1894 /*
1895 * Make sure that we have allocated dquot(s) on disk.
1896 */
1909 /*
1910 * Initially assume that the file does not exist and
1911 * reserve the resources for that case. If that is not
1912 * the case we'll drop the one we have and get a more
1913 * appropriate transaction later.
1914 */
1921 }
1926 }
1934 /*
1935 * Reserve disk quota and the inode.
1936 */
1952 }
1955 /*
1956 * At this point, we've gotten a newly allocated inode.
1957 * It is locked (and joined to the transaction).
1958 */
1962 /*
1963 * Now we join the directory inode to the transaction.
1964 * We do not do it earlier because xfs_dir_ialloc
1965 * might commit the previous transaction (and release
1966 * all the locks).
1967 */
1979 }
1983 /*
1984 * If this is a synchronous mount, make sure that the
1985 * create transaction goes to disk before returning to
1986 * the user.
1987 */
1990 }
1994 /*
1995 * Attach the dquot(s) to the inodes and modify them incore.
1996 * These ids of the inode couldn't have changed since the new
1997 * inode has been locked ever since it was created.
1998 */
2001 /*
2002 * xfs_trans_commit normally decrements the vnode ref count
2003 * when it unlocks the inode. Since we want to return the
2004 * vnode to the caller, we bump the vnode ref count now.
2005 */
2013 }
2020 }
2025 /*
2026 * Propagate the fact that the vnode changed after the
2027 * xfs_inode locks have been released.
2028 */
2033 /* Fallthrough to std_return with error = 0 */
2035 std_return:
2038 DM_EVENT_POSTCREATE)) {
2044 }
2047 abort_return:
2049 /* FALLTHROUGH */
2051 error_return:
2062 abort_rele:
2063 /*
2064 * Wait until after the current transaction is aborted to
2065 * release the inode. This prevents recursive transactions
2066 * and deadlocks from xfs_inactive.
2067 */
2076 }
2078 #ifdef DEBUG
2079 /*
2080 * Some counters to see if (and how often) we are hitting some deadlock
2081 * prevention code paths.
2082 */
2087 #endif
2089 /*
2090 * The following routine will lock the inodes associated with the
2091 * directory and the named entry in the directory. The locks are
2092 * acquired in increasing inode number.
2093 *
2094 * If the entry is "..", then only the directory is locked. The
2095 * vnode ref count will still include that from the .. entry in
2096 * this case.
2097 *
2098 * There is a deadlock we need to worry about. If the locked directory is
2099 * in the AIL, it might be blocking up the log. The next inode we lock
2100 * could be already locked by another thread waiting for log space (e.g
2101 * a permanent log reservation with a long running transaction (see
2102 * xfs_itruncate_finish)). To solve this, we must check if the directory
2103 * is in the ail and use lock_nowait. If we can't lock, we need to
2104 * drop the inode lock on the directory and try again. xfs_iunlock will
2105 * potentially push the tail if we were holding up the log.
2106 */
2107 STATIC int
2112 {
2114 xfs_ino_t e_inum;
2118 #ifdef DEBUG
2119 xfs_rm_locks++;
2120 #endif
2123 again:
2130 /*
2131 * We want to lock in increasing inum. Since we've already
2132 * acquired the lock on the directory, we may need to release
2133 * if if the inum of the entry turns out to be less.
2134 */
2136 /*
2137 * We are already in the right order, so just
2138 * lock on the inode of the entry.
2139 * We need to use nowait if dp is in the AIL.
2140 */
2145 attempts++;
2146 #ifdef DEBUG
2147 xfs_rm_attempts++;
2148 #endif
2150 /*
2151 * Unlock dp and try again.
2152 * xfs_iunlock will try to push the tail
2153 * if the inode is in the AIL.
2154 */
2160 #ifdef DEBUG
2161 xfs_rm_lock_delays++;
2162 #endif
2163 }
2165 }
2168 }
2175 }
2176 /* else e_inum == dp->i_ino */
2177 /* This can happen if we're asked to lock /x/..
2178 * the entry is "..", which is also the parent directory.
2179 */
2182 }
2184 #ifdef DEBUG
2190 #endif
2192 /*
2193 * The following routine will lock n inodes in exclusive mode.
2194 * We assume the caller calls us with the inodes in i_ino order.
2195 *
2196 * We need to detect deadlock where an inode that we lock
2197 * is in the AIL and we start waiting for another inode that is locked
2198 * by a thread in a long running transaction (such as truncate). This can
2199 * result in deadlock since the long running trans might need to wait
2200 * for the inode we just locked in order to push the tail and free space
2201 * in the log.
2202 */
2203 void
2208 uint lock_mode)
2209 {
2221 }
2223 again:
2230 /*
2231 * If try_lock is not set yet, make sure all locked inodes
2232 * are not in the AIL.
2233 * If any are, set try_lock to be used later.
2234 */
2240 try_lock++;
2241 }
2242 }
2243 }
2245 /*
2246 * If any of the previous locks we have locked is in the AIL,
2247 * we must TRY to get the second and subsequent locks. If
2248 * we can't get any, we must release all we have
2249 * and try again.
2250 */
2253 /* try_lock must be 0 if i is 0. */
2254 /*
2255 * try_lock means we have an inode locked
2256 * that is in the AIL.
2257 */
2260 attempts++;
2262 /*
2263 * Unlock all previous guys and try again.
2264 * xfs_iunlock will try to push the tail
2265 * if the inode is in the AIL.
2266 */
2270 /*
2271 * Check to see if we've already
2272 * unlocked this one.
2273 * Not the first one going back,
2274 * and the inode ptr is the same.
2275 */
2281 }
2285 #ifdef DEBUG
2286 xfs_lock_delays++;
2287 #endif
2288 }
2292 }
2295 }
2296 }
2298 #ifdef DEBUG
2304 xfs_locked_n++;
2305 }
2306 #endif
2307 }
2309 #ifdef DEBUG
2310 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2313 #define REMOVE_DEBUG_TRACE(x)
2317 /*
2318 * xfs_remove
2319 *
2320 */
2321 STATIC int
2326 {
2333 xfs_bmap_free_t free_list;
2334 xfs_fsblock_t first_block;
2339 uint resblks;
2359 }
2361 /* From this point on, return through std_return */
2364 /*
2365 * We need to get a reference to ip before we get our log
2366 * reservation. The reason for this is that we cannot call
2367 * xfs_iget for an inode for which we do not have a reference
2368 * once we've acquired a log reservation. This is because the
2369 * inode we are trying to get might be in xfs_inactive going
2370 * for a log reservation. Since we'll have to wait for the
2371 * inactive code to complete before returning from xfs_iget,
2372 * we need to make sure that we don't have log space reserved
2373 * when we call xfs_iget. Instead we get an unlocked reference
2374 * to the inode before getting our log reservation.
2375 */
2380 }
2395 }
2399 /*
2400 * We try to get the real space reservation first,
2401 * allowing for directory btree deletion(s) implying
2402 * possible bmap insert(s). If we can't get the space
2403 * reservation then we use 0 instead, and avoid the bmap
2404 * btree insert(s) in the directory code by, if the bmap
2405 * insert tries to happen, instead trimming the LAST
2406 * block from the directory.
2407 */
2415 }
2422 }
2430 }
2432 /*
2433 * At this point, we've gotten both the directory and the entry
2434 * inodes locked.
2435 */
2438 /*
2439 * Increment vnode ref count only in this case since
2440 * there's an extra vnode reference in the case where
2441 * dp == ip.
2442 */
2445 }
2447 /*
2448 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2449 */
2457 }
2467 }
2469 /* Determine if this is the last link while
2470 * we are in the transaction.
2471 */
2474 /*
2475 * Take an extra ref on the inode so that it doesn't
2476 * go to xfs_inactive() from within the commit.
2477 */
2480 /*
2481 * If this is a synchronous mount, make sure that the
2482 * remove transaction goes to disk before returning to
2483 * the user.
2484 */
2487 }
2493 }
2499 }
2501 /*
2502 * Before we drop our extra reference to the inode, purge it
2503 * from the refcache if it is there. By waiting until afterwards
2504 * to do the IRELE, we ensure that we won't go inactive in the
2505 * xfs_refcache_purge_ip routine (although that would be OK).
2506 */
2511 /*
2512 * Let interposed file systems know about removed links.
2513 */
2518 /* Fall through to std_return with error = 0 */
2519 std_return:
2521 DM_EVENT_POSTREMOVE)) {
2526 }
2529 error1:
2535 error_rele:
2536 /*
2537 * In this case make sure to not release the inode until after
2538 * the current transaction is aborted. Releasing it beforehand
2539 * can cause us to go to xfs_inactive and start a recursive
2540 * transaction which can easily deadlock with the current one.
2541 */
2546 /*
2547 * Before we drop our extra reference to the inode, purge it
2548 * from the refcache if it is there. By waiting until afterwards
2549 * to do the IRELE, we ensure that we won't go inactive in the
2550 * xfs_refcache_purge_ip routine (although that would be OK).
2551 */
2557 }
2560 /*
2561 * xfs_link
2562 *
2563 */
2564 STATIC int
2570 {
2576 xfs_bmap_free_t free_list;
2577 xfs_fsblock_t first_block;
2606 }
2608 /* Return through std_return after this point. */
2625 }
2629 }
2637 }
2641 /*
2642 * Increment vnode ref counts since xfs_trans_commit &
2643 * xfs_trans_cancel will both unlock the inodes and
2644 * decrement the associated ref counts.
2645 */
2651 /*
2652 * If the source has too many links, we can't make any more to it.
2653 */
2657 }
2659 /*
2660 * If we are using project inheritance, we only allow hard link
2661 * creation in our tree when the project IDs are the same; else
2662 * the tree quota mechanism could be circumvented.
2663 */
2668 }
2672 target_namelen)))
2679 resblks);
2689 }
2691 /*
2692 * If this is a synchronous mount, make sure that the
2693 * link transaction goes to disk before returning to
2694 * the user.
2695 */
2698 }
2704 }
2709 }
2711 /* Fall through to std_return with error = 0. */
2712 std_return:
2714 DM_EVENT_POSTLINK)) {
2719 }
2722 abort_return:
2724 /* FALLTHROUGH */
2726 error_return:
2729 }
2730 /*
2731 * xfs_mkdir
2732 *
2733 */
2734 STATIC int
2741 {
2751 xfs_bmap_free_t free_list;
2752 xfs_fsblock_t first_block;
2754 boolean_t dp_joined_to_trans;
2757 xfs_prid_t prid;
2759 uint resblks;
2784 }
2786 /* Return through std_return after this point. */
2796 else
2799 /*
2800 * Make sure that we have allocated dquot(s) on disk.
2801 */
2816 XFS_TRANS_PERM_LOG_RES,
2817 XFS_MKDIR_LOG_COUNT);
2818 }
2823 }
2827 /*
2828 * Check for directory link count overflow.
2829 */
2833 }
2835 /*
2836 * Reserve disk quota and the inode.
2837 */
2845 /*
2846 * create the directory inode.
2847 */
2855 }
2858 /*
2859 * Now we add the directory inode to the transaction.
2860 * We waited until now since xfs_dir_ialloc might start
2861 * a new transaction. Had we joined the transaction
2862 * earlier, the locks might have gotten released.
2863 */
2876 }
2879 /*
2880 * Bump the in memory version number of the parent directory
2881 * so that other processes accessing it will recognize that
2882 * the directory has changed.
2883 */
2889 }
2895 }
2904 /*
2905 * Attach the dquots to the new inode and modify the icount incore.
2906 */
2909 /*
2910 * If this is a synchronous mount, make sure that the
2911 * mkdir transaction goes to disk before returning to
2912 * the user.
2913 */
2916 }
2922 }
2929 }
2931 /* Fall through to std_return with error = 0 or errno from
2932 * xfs_trans_commit. */
2934 std_return:
2937 DM_EVENT_POSTCREATE)) {
2941 DM_RIGHT_NULL,
2944 }
2947 error2:
2948 error1:
2950 abort_return:
2952 error_return:
2959 }
2962 }
2965 /*
2966 * xfs_rmdir
2967 *
2968 */
2969 STATIC int
2974 {
2981 xfs_bmap_free_t free_list;
2982 xfs_fsblock_t first_block;
2989 uint resblks;
3008 }
3010 /* Return through std_return after this point. */
3014 /*
3015 * We need to get a reference to cdp before we get our log
3016 * reservation. The reason for this is that we cannot call
3017 * xfs_iget for an inode for which we do not have a reference
3018 * once we've acquired a log reservation. This is because the
3019 * inode we are trying to get might be in xfs_inactive going
3020 * for a log reservation. Since we'll have to wait for the
3021 * inactive code to complete before returning from xfs_iget,
3022 * we need to make sure that we don't have log space reserved
3023 * when we call xfs_iget. Instead we get an unlocked reference
3024 * to the inode before getting our log reservation.
3025 */
3030 }
3034 /*
3035 * Get the dquots for the inodes.
3036 */
3044 }
3048 /*
3049 * We try to get the real space reservation first,
3050 * allowing for directory btree deletion(s) implying
3051 * possible bmap insert(s). If we can't get the space
3052 * reservation then we use 0 instead, and avoid the bmap
3053 * btree insert(s) in the directory code by, if the bmap
3054 * insert tries to happen, instead trimming the LAST
3055 * block from the directory.
3056 */
3064 }
3070 }
3073 /*
3074 * Now lock the child directory inode and the parent directory
3075 * inode in the proper order. This will take care of validating
3076 * that the directory entry for the child directory inode has
3077 * not changed while we were obtaining a log reservation.
3078 */
3084 }
3088 /*
3089 * Only increment the parent directory vnode count if
3090 * we didn't bump it in looking up cdp. The only time
3091 * we don't bump it is when we're looking up ".".
3092 */
3094 }
3103 }
3107 }
3113 }
3117 /*
3118 * Bump the in memory generation count on the parent
3119 * directory so that other can know that it has changed.
3120 */
3123 /*
3124 * Drop the link from cdp's "..".
3125 */
3129 }
3131 /*
3132 * Drop the link from dp to cdp.
3133 */
3137 }
3139 /*
3140 * Drop the "." link from cdp to self.
3141 */
3145 }
3147 /* Determine these before committing transaction */
3150 /*
3151 * Take an extra ref on the child vnode so that it
3152 * does not go to xfs_inactive() from within the commit.
3153 */
3156 /*
3157 * If this is a synchronous mount, make sure that the
3158 * rmdir transaction goes to disk before returning to
3159 * the user.
3160 */
3163 }
3169 XFS_TRANS_ABORT));
3172 }
3178 }
3181 /*
3182 * Let interposed file systems know about removed links.
3183 */
3188 /* Fall through to std_return with error = 0 or the errno
3189 * from xfs_trans_commit. */
3190 std_return:
3197 }
3200 error1:
3203 /* FALLTHROUGH */
3205 error_return:
3208 }
3211 /*
3212 * xfs_readdir
3213 *
3214 * Read dp's entries starting at uiop->uio_offset and translate them into
3215 * bufsize bytes worth of struct dirents starting at bufbase.
3216 */
3217 STATIC int
3223 {
3227 uint lock_mode;
3235 }
3241 }
3244 /*
3245 * xfs_symlink
3246 *
3247 */
3248 STATIC int
3256 {
3263 xfs_bmap_free_t free_list;
3264 xfs_fsblock_t first_block;
3265 boolean_t dp_joined_to_trans;
3267 uint cancel_flags;
3269 xfs_fileoff_t first_fsb;
3270 xfs_filblks_t fs_blocks;
3273 xfs_daddr_t d;
3278 xfs_prid_t prid;
3280 uint resblks;
3301 /*
3302 * Check component lengths of the target path name.
3303 */
3312 /*
3313 * Skip any slashes.
3314 */
3316 total++;
3317 path++;
3318 }
3320 /*
3321 * Count up to the next slash or end of path.
3322 * Error out if the component is bigger than MAXNAMELEN.
3323 */
3328 }
3329 }
3330 }
3331 }
3339 }
3341 /* Return through std_return after this point. */
3348 else
3351 /*
3352 * Make sure that we have allocated dquot(s) on disk.
3353 */
3362 /*
3363 * The symlink will fit into the inode data fork?
3364 * There can't be any attributes so we get the whole variable part.
3365 */
3368 else
3377 }
3382 }
3386 /*
3387 * Check whether the directory allows new symlinks or not.
3388 */
3392 }
3394 /*
3395 * Reserve disk quota : blocks and inode.
3396 */
3401 /*
3402 * Check for ability to enter directory entry, if no space reserved.
3403 */
3407 /*
3408 * Initialize the bmap freelist prior to calling either
3409 * bmapi or the directory create code.
3410 */
3413 /*
3414 * Allocate an inode for the symlink.
3415 */
3422 }
3429 /*
3430 * Also attach the dquot(s) to it, if applicable.
3431 */
3436 /*
3437 * If the symlink will fit into the inode, write it inline.
3438 */
3444 /*
3445 * The inode was initially created in extent format.
3446 */
3463 }
3479 }
3486 }
3487 }
3489 /*
3490 * Create the directory entry for the symlink.
3491 */
3496 }
3500 /*
3501 * Bump the in memory version number of the parent directory
3502 * so that other processes accessing it will recognize that
3503 * the directory has changed.
3504 */
3507 /*
3508 * If this is a synchronous mount, make sure that the
3509 * symlink transaction goes to disk before returning to
3510 * the user.
3511 */
3514 }
3516 /*
3517 * xfs_trans_commit normally decrements the vnode ref count
3518 * when it unlocks the inode. Since we want to return the
3519 * vnode to the caller, we bump the vnode ref count now.
3520 */
3526 }
3531 /* Fall through to std_return with error = 0 or errno from
3532 * xfs_trans_commit */
3533 std_return:
3535 DM_EVENT_POSTSYMLINK)) {
3541 }
3549 }
3552 error2:
3554 error1:
3557 error_return:
3564 }
3567 }
3570 /*
3571 * xfs_fid2
3572 *
3573 * A fid routine that takes a pointer to a previously allocated
3574 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3575 */
3576 STATIC int
3580 {
3592 /*
3593 * use memcpy because the inode is a long long and there's no
3594 * assurance that xfid->fid_ino is properly aligned.
3595 */
3600 }
3603 /*
3604 * xfs_rwlock
3605 */
3606 int
3609 vrwlock_t locktype)
3610 {
3628 }
3631 }
3634 /*
3635 * xfs_rwunlock
3636 */
3637 void
3640 vrwlock_t locktype)
3641 {
3650 /*
3651 * In the write case, we may have added a new entry to
3652 * the reference cache. This might store a pointer to
3653 * an inode to be released in this inode. If it is there,
3654 * clear the pointer and release the inode after unlocking
3655 * this one.
3656 */
3662 }
3664 }
3666 STATIC int
3670 {
3683 /*
3684 * Bypass inodes which have already been cleaned by
3685 * the inode flush clustering code inside xfs_iflush
3686 */
3694 xfs_lsn_t sync_lsn;
3707 }
3708 }
3710 /*
3711 * We make this non-blocking if the inode is contended,
3712 * return EAGAIN to indicate to the caller that they
3713 * did not succeed. This prevents the flush path from
3714 * blocking on inodes inside another operation right
3715 * now, they get caught later by xfs_sync.
3716 */
3730 }
3733 }
3737 else
3742 }
3745 }
3748 int
3751 u_int evmask,
3752 u_int16_t state,
3754 {
3774 }
3786 }
3789 /*
3790 * xfs_reclaim
3791 */
3792 STATIC int
3795 {
3806 /* bad inode, get out here ASAP */
3810 }
3816 /*
3817 * Make sure the atime in the XFS inode is correct before freeing the
3818 * Linux inode.
3819 */
3822 /* If we have nothing to flush with this inode then complete the
3823 * teardown now, otherwise break the link between the xfs inode
3824 * and the linux inode and clean up the xfs inode later. This
3825 * avoids flushing the inode to disk during the delete operation
3826 * itself.
3827 */
3835 /* Protect sync from us */
3841 }
3843 }
3845 int
3850 {
3858 /* The hash lock here protects a thread in xfs_iget_core from
3859 * racing with us on linking the inode back with a vnode.
3860 * Once we have the XFS_IRECLAIM flag set it will not touch
3861 * us.
3862 */
3870 }
3872 }
3876 /*
3877 * If the inode is still dirty, then flush it out. If the inode
3878 * is not in the AIL, then it will be OK to flush it delwri as
3879 * long as xfs_iflush() does not keep any references to the inode.
3880 * We leave that decision up to xfs_iflush() since it has the
3881 * knowledge of whether it's OK to simply do a delwri flush of
3882 * the inode or whether we need to wait until the inode is
3883 * pulled from the AIL.
3884 * We get the flush lock regardless, though, just to make sure
3885 * we don't free it while it is being flushed.
3886 */
3891 }
3897 /*
3898 * If we hit an error, typically because of filesystem
3899 * shutdown, we don't need to let vn_reclaim to know
3900 * because we're gonna reclaim the inode anyway.
3901 */
3905 }
3907 }
3914 /*
3915 * We are not interested in doing an iflush if we're
3916 * in the process of shutting down the filesystem forcibly.
3917 * So, just reclaim the inode.
3918 */
3921 }
3923 reclaim:
3926 }
3928 int
3930 {
3946 }
3947 }
3950 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3954 }
3957 }
3961 }
3963 /*
3964 * xfs_alloc_file_space()
3965 * This routine allocates disk space for the given file.
3966 *
3967 * If alloc_type == 0, this request is for an ALLOCSP type
3968 * request which will change the file size. In this case, no
3969 * DMAPI event will be generated by the call. A TRUNCATE event
3970 * will be generated later by xfs_setattr.
3971 *
3972 * If alloc_type != 0, this request is for a RESVSP type
3973 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3974 * lower block boundary byte address is less than the file's
3975 * length.
3976 *
3977 * RETURNS:
3978 * 0 on success
3979 * errno on error
3980 *
3981 */
3982 STATIC int
3985 xfs_off_t offset,
3986 xfs_off_t len,
3989 {
3991 xfs_off_t count;
3992 xfs_filblks_t allocated_fsb;
3993 xfs_filblks_t allocatesize_fsb;
3995 xfs_fileoff_t startoffset_fsb;
3996 xfs_fsblock_t firstfsb;
4003 xfs_bmap_free_t free_list;
4019 }
4035 /* Generate a DMAPI event if needed. */
4039 xfs_off_t end_dmi_offset;
4049 }
4051 /*
4052 * Allocate file space until done or until there is an error
4053 */
4054 retry:
4058 /*
4059 * Determine space reservations for data/realtime.
4060 */
4073 }
4085 }
4087 /*
4088 * Allocate and setup the transaction.
4089 */
4093 XFS_TRANS_PERM_LOG_RES,
4094 XFS_WRITE_LOG_COUNT);
4095 /*
4096 * Check for running out of space
4097 */
4099 /*
4100 * Free the transaction structure.
4101 */
4105 }
4115 /*
4116 * Issue the xfs_bmapi() call to allocate the blocks
4117 */
4125 }
4127 /*
4128 * Complete the transaction
4129 */
4133 }
4139 }
4146 }
4150 }
4151 dmapi_enospc_check:
4161 /* else fall through with error from XFS_SEND_DATA */
4162 }
4174 }
4176 /*
4177 * Zero file bytes between startoff and endoff inclusive.
4178 * The iolock is held exclusive and no blocks are buffered.
4179 */
4180 STATIC int
4183 xfs_off_t startoff,
4184 xfs_off_t endoff)
4185 {
4186 xfs_bmbt_irec_t imap;
4187 xfs_fileoff_t offset_fsb;
4188 xfs_off_t lastoffset;
4189 xfs_off_t offset;
4225 }
4237 }
4238 }
4241 }
4243 /*
4244 * xfs_free_file_space()
4245 * This routine frees disk space for the given file.
4246 *
4247 * This routine is only called by xfs_change_file_space
4248 * for an UNRESVSP type call.
4249 *
4250 * RETURNS:
4251 * 0 on success
4252 * errno on error
4253 *
4254 */
4255 STATIC int
4258 xfs_off_t offset,
4259 xfs_off_t len,
4261 {
4265 xfs_off_t end_dmi_offset;
4266 xfs_fileoff_t endoffset_fsb;
4268 xfs_fsblock_t firstfsb;
4269 xfs_bmap_free_t free_list;
4270 xfs_off_t ilen;
4271 xfs_bmbt_irec_t imap;
4272 xfs_off_t ioffset;
4276 uint resblks;
4279 xfs_fileoff_t startoffset_fsb;
4309 }
4317 }
4331 }
4333 /*
4334 * Need to zero the stuff we're not freeing, on disk.
4335 * If its a realtime file & can't use unwritten extents then we
4336 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4337 * will take care of it for us.
4338 */
4347 xfs_daddr_t block;
4354 }
4363 mod++;
4366 }
4367 }
4369 /*
4370 * One contiguous piece to clear
4371 */
4374 /*
4375 * Some full blocks, possibly two pieces to clear
4376 */
4385 }
4387 /*
4388 * free file space until done or until there is an error
4389 */
4393 /*
4394 * allocate and setup the transaction
4395 */
4398 resblks,
4401 XFS_TRANS_PERM_LOG_RES,
4402 XFS_WRITE_LOG_COUNT);
4404 /*
4405 * check for running out of space
4406 */
4408 /*
4409 * Free the transaction structure.
4410 */
4414 }
4418 XFS_QMOPT_RES_REGBLKS);
4425 /*
4426 * issue the bunmapi() call to free the blocks
4427 */
4434 }
4436 /*
4437 * complete the transaction
4438 */
4442 }
4446 }
4448 out_unlock_iolock:
4453 error0:
4455 error1:
4458 XFS_ILOCK_EXCL);
4460 }
4462 /*
4463 * xfs_change_file_space()
4464 * This routine allocates or frees disk space for the given file.
4465 * The user specified parameters are checked for alignment and size
4466 * limitations.
4467 *
4468 * RETURNS:
4469 * 0 on success
4470 * errno on error
4471 *
4472 */
4473 int
4478 xfs_off_t offset,
4481 {
4484 xfs_fsize_t fsize;
4488 xfs_off_t startoffset;
4489 xfs_off_t llen;
4491 vattr_t va;
4500 /*
4501 * must be a regular file and have write permission
4502 */
4511 }
4526 }
4541 /*
4542 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4543 * file space.
4544 * These calls do NOT zero the data space allocated to the file,
4545 * nor do they change the file size.
4546 *
4547 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4548 * space.
4549 * These calls cause the new file data to be zeroed and the file
4550 * size to be changed.
4551 */
4567 attr_flags)))
4580 }
4596 }
4598 /*
4599 * update the inode timestamp, mode, and prealloc flag bits
4600 */
4605 /* ASSERT(0); */
4608 }
4618 /*
4619 * Note that we don't have to worry about mandatory
4620 * file locking being disabled here because we only
4621 * clear the S_ISGID bit if the Group execute bit is
4622 * on, but if it was on then mandatory locking wouldn't
4623 * have been enabled.
4624 */
4629 }
4643 }
4645 vnodeops_t xfs_vnodeops = {
4649 #ifdef HAVE_SENDFILE
4651 #endif
4652 #ifdef HAVE_SPLICE
4655 #endif
4689 };