| blob | 5292d550d8e37711a9982496116346f11da28f2d |
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 */
55 STATIC int
59 {
67 /*
68 * If it's a directory with any blocks, read-ahead block 0
69 * as we're almost certain to have the next operation be a read there.
70 */
76 }
78 }
80 STATIC int
84 lastclose_t lastclose,
86 {
96 /*
97 * If we previously truncated this file and removed old data in
98 * the process, we want to initiate "early" writeout on the last
99 * close. This is an attempt to combat the notorious NULL files
100 * problem which is particularly noticable from a truncate down,
101 * buffered (re-)write (delalloc), followed by a crash. What we
102 * are effectively doing here is significantly reducing the time
103 * window where we'd otherwise be exposed to that problem.
104 */
108 }
110 /*
111 * xfs_getattr
112 */
113 STATIC int
119 {
143 #if XFS_BIG_INUMS
145 #endif
148 /*
149 * Quick exit for non-stat callers
150 */
156 /*
157 * Copy from in-core inode.
158 */
164 /*
165 * Check vnode type block/char vs. everything else.
166 */
180 /*
181 * If the file blocks are being allocated from a
182 * realtime partition, then return the inode's
183 * realtime extent size or the realtime volume's
184 * extent size.
185 */
189 }
191 }
199 /*
200 * Exit for stat callers. See if any of the rest of the fields
201 * to be filled in are needed.
202 */
208 /*
209 * Convert di_flags to xflags.
210 */
213 /*
214 * Exit for inode revalidate. See if any of the rest of
215 * the fields to be filled in are needed.
216 */
232 else
236 all_done:
240 }
243 /*
244 * xfs_setattr
245 */
246 int
252 {
258 uint lock_flags;
276 /*
277 * Cannot set certain attributes.
278 */
282 }
290 /*
291 * Timestamps do not need to be logged and hence do not
292 * need to be done within a transaction.
293 */
301 }
306 /*
307 * If disk quotas is on, we make sure that the dquots do exist on disk,
308 * before we start any other transactions. Trying to do this later
309 * is messy. We don't care to take a readlock to look at the ids
310 * in inode here, because we can't hold it across the trans_reserve.
311 * If the IDs do change before we take the ilock, we're covered
312 * because the i_*dquot fields will get updated anyway.
313 */
323 }
329 }
335 }
336 /*
337 * We take a reference when we initialize udqp and gdqp,
338 * so it is important that we never blindly double trip on
339 * the same variable. See xfs_create() for an example.
340 */
347 }
349 /*
350 * For the other attributes, we acquire the inode lock and
351 * first do an error checking pass.
352 */
367 }
368 }
378 }
379 }
382 }
386 /* boolean: are we the file owner? */
389 /*
390 * Change various properties of a file.
391 * Only the owner or users with CAP_FOWNER
392 * capability may do these things.
393 */
397 /*
398 * CAP_FOWNER overrides the following restrictions:
399 *
400 * The user ID of the calling process must be equal
401 * to the file owner ID, except in cases where the
402 * CAP_FSETID capability is applicable.
403 */
407 }
409 /*
410 * CAP_FSETID overrides the following restrictions:
411 *
412 * The effective user ID of the calling process shall match
413 * the file owner when setting the set-user-ID and
414 * set-group-ID bits on that file.
415 *
416 * The effective group ID or one of the supplementary group
417 * IDs of the calling process shall match the group owner of
418 * the file when setting the set-group-ID bit on that file
419 */
428 #if 0
429 /* Linux allows this, Irix doesn't. */
432 #endif
435 }
436 }
438 /*
439 * Change file ownership. Must be the owner or privileged.
440 * If the system was configured with the "restricted_chown"
441 * option, the owner is not permitted to give away the file,
442 * and can change the group id only to a group of which he
443 * or she is a member.
444 */
446 /*
447 * These IDs could have changed since we last looked at them.
448 * But, we're assured that if the ownership did change
449 * while we didn't have the inode locked, inode's dquot(s)
450 * would have changed also.
451 */
458 iprojid;
460 /*
461 * CAP_CHOWN overrides the following restrictions:
462 *
463 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
464 * shall override the restriction that a process cannot
465 * change the user ID of a file it owns and the restriction
466 * that the group ID supplied to the chown() function
467 * shall be equal to either the group ID or one of the
468 * supplementary group IDs of the calling process.
469 */
476 }
477 /*
478 * Do a quota reservation only if uid/projid/gid is actually
479 * going to change.
480 */
490 }
491 }
493 /*
494 * Truncate file. Must have write permission and not be a directory.
495 */
497 /* Short circuit the truncate case for zero length files */
506 }
514 }
515 /*
516 * Make sure that the dquots are attached to the inode.
517 */
520 }
522 /*
523 * Change file access or modified times.
524 */
531 }
532 }
533 }
535 /*
536 * Change extent size or realtime flag.
537 */
539 /*
540 * Can't change extent size if any extents are allocated.
541 */
547 }
549 /*
550 * Can't change realtime flag if any extents are allocated.
551 */
558 }
559 /*
560 * Extent size must be a multiple of the appropriate block
561 * size, if set at all.
562 */
564 xfs_extlen_t size;
573 }
577 }
578 }
579 /*
580 * If realtime flag is set then must have realtime data.
581 */
589 }
590 }
592 /*
593 * Can't modify an immutable/append-only file unless
594 * we have appropriate permission.
595 */
604 }
605 }
607 /*
608 * Now we can make the changes. Before we join the inode
609 * to the transaction, if XFS_AT_SIZE is set then take care of
610 * the part of the truncation that must be done without the
611 * inode lock. This needs to be done before joining the inode
612 * to the transaction, because the inode cannot be unlocked
613 * once it is a part of the transaction.
614 */
620 }
630 }
634 XFS_TRANS_PERM_LOG_RES,
635 XFS_ITRUNCATE_LOG_COUNT))) {
640 }
643 }
648 }
650 /* determine whether mandatory locking mode changes */
653 /*
654 * Truncate file. Must have write permission and not be a directory.
655 */
662 /*
663 * signal a sync transaction unless
664 * we're truncating an already unlinked
665 * file on a wsync filesystem
666 */
669 XFS_DATA_FORK,
675 /*
676 * Truncated "down", so we're removing references
677 * to old data here - if we now delay flushing for
678 * a long time, we expose ourselves unduly to the
679 * notorious NULL files problem. So, we mark this
680 * vnode and flush it when the file is closed, and
681 * do not wait the usual (long) time for writeout.
682 */
684 }
685 /*
686 * Have to do this even if the file's size doesn't change.
687 */
689 }
691 /*
692 * Change file access modes.
693 */
700 }
702 /*
703 * Change file ownership. Must be the owner or privileged.
704 * If the system was configured with the "restricted_chown"
705 * option, the owner is not permitted to give away the file,
706 * and can change the group id only to a group of which he
707 * or she is a member.
708 */
710 /*
711 * CAP_FSETID overrides the following restrictions:
712 *
713 * The set-user-ID and set-group-ID bits of a file will be
714 * cleared upon successful return from chown()
715 */
719 }
721 /*
722 * Change the ownerships and register quota modifications
723 * in the transaction.
724 */
731 }
733 }
741 }
743 }
751 }
753 /*
754 * We may have to rev the inode as well as
755 * the superblock version number since projids didn't
756 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
757 */
760 }
764 }
767 /*
768 * Change file access or modified times.
769 */
776 }
782 }
785 }
787 /*
788 * Change XFS-added attributes.
789 */
792 /*
793 * Converting bytes to fs blocks.
794 */
797 }
799 uint di_flags;
801 /* can't set PREALLOC this way, just preserve it */
830 }
833 }
835 }
838 }
840 /*
841 * Change file inode change time only if XFS_AT_CTIME set
842 * AND we have been called by a DMI function.
843 */
850 }
852 /*
853 * Send out timestamp changes that need to be set to the
854 * current time. Not done when called by a DMI function.
855 */
861 /*
862 * If this is a synchronous mount, make sure that the
863 * transaction goes to disk before returning to the user.
864 * This is slightly sub-optimal in that truncates require
865 * two sync transactions instead of one for wsync filesystems.
866 * One for the truncate and one for the timestamps since we
867 * don't want to change the timestamps unless we're sure the
868 * truncate worked. Truncates are less than 1% of the laddis
869 * mix so this probably isn't worth the trouble to optimize.
870 */
877 }
879 /*
880 * If the (regular) file's mandatory locking mode changed, then
881 * notify the vnode. We do this under the inode lock to prevent
882 * racing calls to vop_vnode_change.
883 */
887 mandlock_after);
888 }
892 /*
893 * Release any dquot(s) the inode had kept before chown.
894 */
902 }
909 }
912 abort_return:
914 /* FALLTHROUGH */
915 error_return:
920 }
923 }
925 }
928 /*
929 * xfs_access
930 * Null conversion from vnode mode bits to inode mode bits, as in efs.
931 */
932 STATIC int
937 {
949 }
952 /*
953 * The maximum pathlen is 1024 bytes. Since the minimum file system
954 * blocksize is 512 bytes, we can get a max of 2 extents back from
955 * bmapi.
956 */
957 #define SYMLINK_MAPS 2
959 /*
960 * xfs_readlink
961 *
962 */
963 STATIC int
969 {
972 xfs_off_t offset;
979 xfs_daddr_t d;
1003 }
1007 }
1009 /*
1010 * See if the symlink is stored inline.
1011 */
1016 }
1018 /*
1019 * Symlink not inline. Call bmap to get it in.
1020 */
1028 }
1041 }
1048 }
1050 }
1052 error_return:
1055 }
1058 /*
1059 * xfs_fsync
1060 *
1061 * This is called to sync the inode and its data out to disk.
1062 * We need to hold the I/O lock while flushing the data, and
1063 * the inode lock while flushing the inode. The inode lock CANNOT
1064 * be held while flushing the data, so acquire after we're done
1065 * with that.
1066 */
1067 STATIC int
1072 xfs_off_t start,
1073 xfs_off_t stop)
1074 {
1090 /*
1091 * We always need to make sure that the required inode state
1092 * is safe on disk. The vnode might be clean but because
1093 * of committed transactions that haven't hit the disk yet.
1094 * Likewise, there could be unflushed non-transactional
1095 * changes to the inode core that have to go to disk.
1096 *
1097 * The following code depends on one assumption: that
1098 * any transaction that changes an inode logs the core
1099 * because it has to change some field in the inode core
1100 * (typically nextents or nblocks). That assumption
1101 * implies that any transactions against an inode will
1102 * catch any non-transactional updates. If inode-altering
1103 * transactions exist that violate this assumption, the
1104 * code breaks. Right now, it figures that if the involved
1105 * update_* field is clear and the inode is unpinned, the
1106 * inode is clean. Either it's been flushed or it's been
1107 * committed and the commit has hit the disk unpinning the inode.
1108 * (Note that xfs_inode_item_format() called at commit clears
1109 * the update_* fields.)
1110 */
1113 /* If we are flushing data then we care about update_size
1114 * being set, otherwise we care about update_core
1115 */
1119 /*
1120 * Timestamps/size haven't changed since last inode
1121 * flush or inode transaction commit. That means
1122 * either nothing got written or a transaction
1123 * committed which caught the updates. If the
1124 * latter happened and the transaction hasn't
1125 * hit the disk yet, the inode will be still
1126 * be pinned. If it is, force the log.
1127 */
1133 XFS_LOG_FORCE |
1138 /*
1139 * If the inode is not pinned and nothing
1140 * has changed we don't need to flush the
1141 * cache.
1142 */
1144 }
1147 /*
1148 * Kick off a transaction to log the inode
1149 * core to get the updates. Make it
1150 * sync if FSYNC_WAIT is passed in (which
1151 * is done by everybody but specfs). The
1152 * sync transaction will also force the log.
1153 */
1161 }
1164 /*
1165 * Note - it's possible that we might have pushed
1166 * ourselves out of the way during trans_reserve
1167 * which would flush the inode. But there's no
1168 * guarantee that the inode buffer has actually
1169 * gone out yet (it's delwri). Plus the buffer
1170 * could be pinned anyway if it's part of an
1171 * inode in another recent transaction. So we
1172 * play it safe and fire off the transaction anyway.
1173 */
1182 }
1185 /*
1186 * If the log write didn't issue an ordered tag we need
1187 * to flush the disk cache for the data device now.
1188 */
1192 /*
1193 * If this inode is on the RT dev we need to flush that
1194 * cache as well.
1195 */
1198 }
1201 }
1203 /*
1204 * This is called by xfs_inactive to free any blocks beyond eof,
1205 * when the link count isn't zero.
1206 */
1207 STATIC int
1211 {
1214 xfs_fileoff_t end_fsb;
1215 xfs_fileoff_t last_fsb;
1216 xfs_filblks_t map_len;
1218 xfs_bmbt_irec_t imap;
1220 /*
1221 * Figure out if there are any blocks beyond the end
1222 * of the file. If not, then there is nothing to do.
1223 */
1239 /*
1240 * Attach the dquots to the inode up front.
1241 */
1245 /*
1246 * There are blocks after the end of file.
1247 * Free them up now by truncating the file to
1248 * its current size.
1249 */
1252 /*
1253 * Do the xfs_itruncate_start() call before
1254 * reserving any log space because
1255 * itruncate_start will call into the buffer
1256 * cache and we can't
1257 * do that within a transaction.
1258 */
1265 }
1270 XFS_ITRUNCATE_LOG_COUNT);
1276 }
1280 XFS_IOLOCK_EXCL |
1281 XFS_ILOCK_EXCL);
1286 XFS_DATA_FORK,
1288 /*
1289 * If we get an error at this point we
1290 * simply don't bother truncating the file.
1291 */
1295 XFS_TRANS_ABORT));
1298 XFS_TRANS_RELEASE_LOG_RES);
1299 }
1301 }
1303 }
1305 /*
1306 * Free a symlink that has blocks associated with it.
1307 */
1308 STATIC int
1312 {
1317 xfs_fsblock_t first_block;
1318 xfs_bmap_free_t free_list;
1330 /*
1331 * We're freeing a symlink that has some
1332 * blocks allocated to it. Free the
1333 * blocks here. We know that we've got
1334 * either 1 or 2 extents and that we can
1335 * free them all in one bunmapi call.
1336 */
1344 }
1345 /*
1346 * Lock the inode, fix the size, and join it to the transaction.
1347 * Hold it so in the normal path, we still have it locked for
1348 * the second transaction. In the error paths we need it
1349 * held so the cancel won't rele it, see below.
1350 */
1357 /*
1358 * Find the block(s) so we can inval and unmap them.
1359 */
1367 /*
1368 * Invalidate the block(s).
1369 */
1375 }
1376 /*
1377 * Unmap the dead block(s) to the free_list.
1378 */
1383 /*
1384 * Commit the first transaction. This logs the EFI and the inode.
1385 */
1388 /*
1389 * The transaction must have been committed, since there were
1390 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1391 * The new tp has the extent freeing and EFDs.
1392 */
1394 /*
1395 * The first xact was committed, so add the inode to the new one.
1396 * Mark it dirty so it will be logged and moved forward in the log as
1397 * part of every commit.
1398 */
1402 /*
1403 * Get a new, empty transaction to return to our caller.
1404 */
1406 /*
1407 * Commit the transaction containing extent freeing and EFDs.
1408 * If we get an error on the commit here or on the reserve below,
1409 * we need to unlock the inode since the new transaction doesn't
1410 * have the inode attached.
1411 */
1417 }
1418 /*
1419 * Remove the memory for extent descriptions (just bookkeeping).
1420 */
1424 /*
1425 * Put an itruncate log reservation in the new transaction
1426 * for our caller.
1427 */
1432 }
1433 /*
1434 * Return with the inode locked but not joined to the transaction.
1435 */
1439 error1:
1441 error0:
1442 /*
1443 * Have to come here with the inode locked and either
1444 * (held and in the transaction) or (not in the transaction).
1445 * If the inode isn't held then cancel would iput it, but
1446 * that's wrong since this is inactive and the vnode ref
1447 * count is 0 already.
1448 * Cancel won't do anything to the inode if held, but it still
1449 * needs to be locked until the cancel is done, if it was
1450 * joined to the transaction.
1451 */
1457 }
1459 STATIC int
1463 {
1467 /*
1468 * We're freeing a symlink which fit into
1469 * the inode. Just free the memory used
1470 * to hold the old symlink.
1471 */
1475 XFS_ITRUNCATE_LOG_COUNT);
1481 }
1484 /*
1485 * Zero length symlinks _can_ exist.
1486 */
1490 XFS_DATA_FORK);
1492 }
1494 }
1496 STATIC int
1500 {
1517 }
1523 XFS_INACTIVE_LOG_COUNT);
1530 }
1541 }
1543 STATIC int
1546 {
1559 /* If this is a read-only mount, don't do this (would generate I/O) */
1563 #ifdef HAVE_REFCACHE
1564 /* If we are in the NFS reference cache then don't do this now */
1567 #endif
1578 /* Update linux inode block count after free above */
1581 }
1582 }
1585 }
1587 /*
1588 * xfs_inactive
1589 *
1590 * This is called when the vnode reference count for the vnode
1591 * goes to zero. If the file has been unlinked, then it must
1592 * now be truncated. Also, we clear all of the read-ahead state
1593 * kept for the inode here since the file is now closed.
1594 */
1595 STATIC int
1599 {
1602 xfs_bmap_free_t free_list;
1603 xfs_fsblock_t first_block;
1615 /*
1616 * If the inode is already free, then there can be nothing
1617 * to clean up here.
1618 */
1623 }
1625 /*
1626 * Only do a truncate if it's a regular file with
1627 * some actual space in it. It's OK to look at the
1628 * inode's fields without the lock because we're the
1629 * only one with a reference to the inode.
1630 */
1641 }
1645 /* If this is a read-only mount, don't do this (would generate I/O) */
1659 /* Update linux inode block count after free above */
1662 }
1664 }
1673 /*
1674 * Do the xfs_itruncate_start() call before
1675 * reserving any log space because itruncate_start
1676 * will call into the buffer cache and we can't
1677 * do that within a transaction.
1678 */
1685 }
1690 XFS_ITRUNCATE_LOG_COUNT);
1692 /* Don't call itruncate_cleanup */
1697 }
1703 /*
1704 * normally, we have to run xfs_itruncate_finish sync.
1705 * But if filesystem is wsync and we're in the inactive
1706 * path, then we know that nlink == 0, and that the
1707 * xaction that made nlink == 0 is permanently committed
1708 * since xfs_remove runs as a synchronous transaction.
1709 */
1718 }
1721 /*
1722 * If we get an error while cleaning up a
1723 * symlink we bail out.
1724 */
1732 }
1740 XFS_INACTIVE_LOG_COUNT);
1745 }
1750 }
1752 /*
1753 * If there are attributes associated with the file
1754 * then blow them away now. The code calls a routine
1755 * that recursively deconstructs the attribute fork.
1756 * We need to just commit the current transaction
1757 * because we can't use it for xfs_attr_inactive().
1758 */
1761 /*
1762 * If we got an error, the transaction is already
1763 * cancelled, and the inode is unlocked. Just get out.
1764 */
1769 }
1771 /*
1772 * Free the inode.
1773 */
1777 /*
1778 * If we fail to free the inode, shut down. The cancel
1779 * might do that, we need to make sure. Otherwise the
1780 * inode might be lost for a long time or forever.
1781 */
1787 }
1790 /*
1791 * Credit the quota account(s). The inode is gone.
1792 */
1795 /*
1796 * Just ignore errors at this point. There is
1797 * nothing we can do except to try to keep going.
1798 */
1801 }
1802 /*
1803 * Release the dquots held by inode, if any.
1804 */
1809 out:
1811 }
1814 /*
1815 * xfs_lookup
1816 */
1817 STATIC int
1825 {
1827 xfs_ino_t e_inum;
1829 uint lock_mode;
1845 }
1848 }
1851 /*
1852 * xfs_create (create a new file).
1853 */
1854 STATIC int
1861 {
1868 xfs_dev_t rdev;
1870 xfs_bmap_free_t free_list;
1871 xfs_fsblock_t first_block;
1872 boolean_t dp_joined_to_trans;
1874 uint cancel_flags;
1876 xfs_prid_t prid;
1878 uint resblks;
1901 }
1906 /* Return through std_return after this point. */
1913 else
1916 /*
1917 * Make sure that we have allocated dquot(s) on disk.
1918 */
1931 /*
1932 * Initially assume that the file does not exist and
1933 * reserve the resources for that case. If that is not
1934 * the case we'll drop the one we have and get a more
1935 * appropriate transaction later.
1936 */
1943 }
1948 }
1956 /*
1957 * Reserve disk quota and the inode.
1958 */
1973 }
1976 /*
1977 * At this point, we've gotten a newly allocated inode.
1978 * It is locked (and joined to the transaction).
1979 */
1983 /*
1984 * Now we join the directory inode to the transaction.
1985 * We do not do it earlier because xfs_dir_ialloc
1986 * might commit the previous transaction (and release
1987 * all the locks).
1988 */
2000 }
2004 /*
2005 * If this is a synchronous mount, make sure that the
2006 * create transaction goes to disk before returning to
2007 * the user.
2008 */
2011 }
2015 /*
2016 * Attach the dquot(s) to the inodes and modify them incore.
2017 * These ids of the inode couldn't have changed since the new
2018 * inode has been locked ever since it was created.
2019 */
2022 /*
2023 * xfs_trans_commit normally decrements the vnode ref count
2024 * when it unlocks the inode. Since we want to return the
2025 * vnode to the caller, we bump the vnode ref count now.
2026 */
2034 }
2041 }
2046 /*
2047 * Propagate the fact that the vnode changed after the
2048 * xfs_inode locks have been released.
2049 */
2054 /* Fallthrough to std_return with error = 0 */
2056 std_return:
2059 DM_EVENT_POSTCREATE)) {
2065 }
2068 abort_return:
2070 /* FALLTHROUGH */
2072 error_return:
2083 abort_rele:
2084 /*
2085 * Wait until after the current transaction is aborted to
2086 * release the inode. This prevents recursive transactions
2087 * and deadlocks from xfs_inactive.
2088 */
2097 }
2099 #ifdef DEBUG
2100 /*
2101 * Some counters to see if (and how often) we are hitting some deadlock
2102 * prevention code paths.
2103 */
2108 #endif
2110 /*
2111 * The following routine will lock the inodes associated with the
2112 * directory and the named entry in the directory. The locks are
2113 * acquired in increasing inode number.
2114 *
2115 * If the entry is "..", then only the directory is locked. The
2116 * vnode ref count will still include that from the .. entry in
2117 * this case.
2118 *
2119 * There is a deadlock we need to worry about. If the locked directory is
2120 * in the AIL, it might be blocking up the log. The next inode we lock
2121 * could be already locked by another thread waiting for log space (e.g
2122 * a permanent log reservation with a long running transaction (see
2123 * xfs_itruncate_finish)). To solve this, we must check if the directory
2124 * is in the ail and use lock_nowait. If we can't lock, we need to
2125 * drop the inode lock on the directory and try again. xfs_iunlock will
2126 * potentially push the tail if we were holding up the log.
2127 */
2128 STATIC int
2132 {
2134 xfs_ino_t e_inum;
2138 #ifdef DEBUG
2139 xfs_rm_locks++;
2140 #endif
2143 again:
2150 /*
2151 * We want to lock in increasing inum. Since we've already
2152 * acquired the lock on the directory, we may need to release
2153 * if if the inum of the entry turns out to be less.
2154 */
2156 /*
2157 * We are already in the right order, so just
2158 * lock on the inode of the entry.
2159 * We need to use nowait if dp is in the AIL.
2160 */
2165 attempts++;
2166 #ifdef DEBUG
2167 xfs_rm_attempts++;
2168 #endif
2170 /*
2171 * Unlock dp and try again.
2172 * xfs_iunlock will try to push the tail
2173 * if the inode is in the AIL.
2174 */
2180 #ifdef DEBUG
2181 xfs_rm_lock_delays++;
2182 #endif
2183 }
2185 }
2188 }
2195 }
2196 /* else e_inum == dp->i_ino */
2197 /* This can happen if we're asked to lock /x/..
2198 * the entry is "..", which is also the parent directory.
2199 */
2202 }
2204 #ifdef DEBUG
2210 #endif
2212 /*
2213 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2214 * a different value
2215 */
2218 {
2225 }
2227 /*
2228 * The following routine will lock n inodes in exclusive mode.
2229 * We assume the caller calls us with the inodes in i_ino order.
2230 *
2231 * We need to detect deadlock where an inode that we lock
2232 * is in the AIL and we start waiting for another inode that is locked
2233 * by a thread in a long running transaction (such as truncate). This can
2234 * result in deadlock since the long running trans might need to wait
2235 * for the inode we just locked in order to push the tail and free space
2236 * in the log.
2237 */
2238 void
2243 uint lock_mode)
2244 {
2256 }
2258 again:
2265 /*
2266 * If try_lock is not set yet, make sure all locked inodes
2267 * are not in the AIL.
2268 * If any are, set try_lock to be used later.
2269 */
2275 try_lock++;
2276 }
2277 }
2278 }
2280 /*
2281 * If any of the previous locks we have locked is in the AIL,
2282 * we must TRY to get the second and subsequent locks. If
2283 * we can't get any, we must release all we have
2284 * and try again.
2285 */
2288 /* try_lock must be 0 if i is 0. */
2289 /*
2290 * try_lock means we have an inode locked
2291 * that is in the AIL.
2292 */
2295 attempts++;
2297 /*
2298 * Unlock all previous guys and try again.
2299 * xfs_iunlock will try to push the tail
2300 * if the inode is in the AIL.
2301 */
2305 /*
2306 * Check to see if we've already
2307 * unlocked this one.
2308 * Not the first one going back,
2309 * and the inode ptr is the same.
2310 */
2316 }
2320 #ifdef DEBUG
2321 xfs_lock_delays++;
2322 #endif
2323 }
2327 }
2330 }
2331 }
2333 #ifdef DEBUG
2339 xfs_locked_n++;
2340 }
2341 #endif
2342 }
2344 #ifdef DEBUG
2345 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2348 #define REMOVE_DEBUG_TRACE(x)
2352 /*
2353 * xfs_remove
2354 *
2355 */
2356 STATIC int
2361 {
2368 xfs_bmap_free_t free_list;
2369 xfs_fsblock_t first_block;
2374 uint resblks;
2391 }
2399 }
2401 /* From this point on, return through std_return */
2404 /*
2405 * We need to get a reference to ip before we get our log
2406 * reservation. The reason for this is that we cannot call
2407 * xfs_iget for an inode for which we do not have a reference
2408 * once we've acquired a log reservation. This is because the
2409 * inode we are trying to get might be in xfs_inactive going
2410 * for a log reservation. Since we'll have to wait for the
2411 * inactive code to complete before returning from xfs_iget,
2412 * we need to make sure that we don't have log space reserved
2413 * when we call xfs_iget. Instead we get an unlocked reference
2414 * to the inode before getting our log reservation.
2415 */
2420 }
2435 }
2439 /*
2440 * We try to get the real space reservation first,
2441 * allowing for directory btree deletion(s) implying
2442 * possible bmap insert(s). If we can't get the space
2443 * reservation then we use 0 instead, and avoid the bmap
2444 * btree insert(s) in the directory code by, if the bmap
2445 * insert tries to happen, instead trimming the LAST
2446 * block from the directory.
2447 */
2455 }
2462 }
2470 }
2472 /*
2473 * At this point, we've gotten both the directory and the entry
2474 * inodes locked.
2475 */
2478 /*
2479 * Increment vnode ref count only in this case since
2480 * there's an extra vnode reference in the case where
2481 * dp == ip.
2482 */
2485 }
2487 /*
2488 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2489 */
2497 }
2507 }
2509 /* Determine if this is the last link while
2510 * we are in the transaction.
2511 */
2514 /*
2515 * Take an extra ref on the inode so that it doesn't
2516 * go to xfs_inactive() from within the commit.
2517 */
2520 /*
2521 * If this is a synchronous mount, make sure that the
2522 * remove transaction goes to disk before returning to
2523 * the user.
2524 */
2527 }
2533 }
2539 }
2541 /*
2542 * Before we drop our extra reference to the inode, purge it
2543 * from the refcache if it is there. By waiting until afterwards
2544 * to do the IRELE, we ensure that we won't go inactive in the
2545 * xfs_refcache_purge_ip routine (although that would be OK).
2546 */
2551 /*
2552 * Let interposed file systems know about removed links.
2553 */
2558 /* Fall through to std_return with error = 0 */
2559 std_return:
2561 DM_EVENT_POSTREMOVE)) {
2566 }
2569 error1:
2575 error_rele:
2576 /*
2577 * In this case make sure to not release the inode until after
2578 * the current transaction is aborted. Releasing it beforehand
2579 * can cause us to go to xfs_inactive and start a recursive
2580 * transaction which can easily deadlock with the current one.
2581 */
2586 /*
2587 * Before we drop our extra reference to the inode, purge it
2588 * from the refcache if it is there. By waiting until afterwards
2589 * to do the IRELE, we ensure that we won't go inactive in the
2590 * xfs_refcache_purge_ip routine (although that would be OK).
2591 */
2597 }
2600 /*
2601 * xfs_link
2602 *
2603 */
2604 STATIC int
2610 {
2616 xfs_bmap_free_t free_list;
2617 xfs_fsblock_t first_block;
2645 }
2647 /* Return through std_return after this point. */
2664 }
2668 }
2676 }
2680 /*
2681 * Increment vnode ref counts since xfs_trans_commit &
2682 * xfs_trans_cancel will both unlock the inodes and
2683 * decrement the associated ref counts.
2684 */
2690 /*
2691 * If the source has too many links, we can't make any more to it.
2692 */
2696 }
2698 /*
2699 * If we are using project inheritance, we only allow hard link
2700 * creation in our tree when the project IDs are the same; else
2701 * the tree quota mechanism could be circumvented.
2702 */
2707 }
2717 resblks);
2728 /*
2729 * If this is a synchronous mount, make sure that the
2730 * link transaction goes to disk before returning to
2731 * the user.
2732 */
2735 }
2741 }
2747 /* Fall through to std_return with error = 0. */
2748 std_return:
2750 DM_EVENT_POSTLINK)) {
2755 }
2758 abort_return:
2760 /* FALLTHROUGH */
2762 error_return:
2765 }
2768 /*
2769 * xfs_mkdir
2770 *
2771 */
2772 STATIC int
2779 {
2789 xfs_bmap_free_t free_list;
2790 xfs_fsblock_t first_block;
2792 boolean_t dp_joined_to_trans;
2795 xfs_prid_t prid;
2797 uint resblks;
2822 }
2824 /* Return through std_return after this point. */
2834 else
2837 /*
2838 * Make sure that we have allocated dquot(s) on disk.
2839 */
2854 XFS_TRANS_PERM_LOG_RES,
2855 XFS_MKDIR_LOG_COUNT);
2856 }
2861 }
2865 /*
2866 * Check for directory link count overflow.
2867 */
2871 }
2873 /*
2874 * Reserve disk quota and the inode.
2875 */
2883 /*
2884 * create the directory inode.
2885 */
2893 }
2896 /*
2897 * Now we add the directory inode to the transaction.
2898 * We waited until now since xfs_dir_ialloc might start
2899 * a new transaction. Had we joined the transaction
2900 * earlier, the locks might have gotten released.
2901 */
2914 }
2917 /*
2918 * Bump the in memory version number of the parent directory
2919 * so that other processes accessing it will recognize that
2920 * the directory has changed.
2921 */
2940 /*
2941 * Attach the dquots to the new inode and modify the icount incore.
2942 */
2945 /*
2946 * If this is a synchronous mount, make sure that the
2947 * mkdir transaction goes to disk before returning to
2948 * the user.
2949 */
2952 }
2958 }
2965 }
2967 /* Fall through to std_return with error = 0 or errno from
2968 * xfs_trans_commit. */
2970 std_return:
2973 DM_EVENT_POSTCREATE)) {
2977 DM_RIGHT_NULL,
2980 }
2983 error2:
2984 error1:
2986 abort_return:
2988 error_return:
2995 }
2998 }
3001 /*
3002 * xfs_rmdir
3003 *
3004 */
3005 STATIC int
3010 {
3017 xfs_bmap_free_t free_list;
3018 xfs_fsblock_t first_block;
3025 uint resblks;
3040 }
3049 }
3051 /* Return through std_return after this point. */
3055 /*
3056 * We need to get a reference to cdp before we get our log
3057 * reservation. The reason for this is that we cannot call
3058 * xfs_iget for an inode for which we do not have a reference
3059 * once we've acquired a log reservation. This is because the
3060 * inode we are trying to get might be in xfs_inactive going
3061 * for a log reservation. Since we'll have to wait for the
3062 * inactive code to complete before returning from xfs_iget,
3063 * we need to make sure that we don't have log space reserved
3064 * when we call xfs_iget. Instead we get an unlocked reference
3065 * to the inode before getting our log reservation.
3066 */
3071 }
3075 /*
3076 * Get the dquots for the inodes.
3077 */
3085 }
3089 /*
3090 * We try to get the real space reservation first,
3091 * allowing for directory btree deletion(s) implying
3092 * possible bmap insert(s). If we can't get the space
3093 * reservation then we use 0 instead, and avoid the bmap
3094 * btree insert(s) in the directory code by, if the bmap
3095 * insert tries to happen, instead trimming the LAST
3096 * block from the directory.
3097 */
3105 }
3111 }
3114 /*
3115 * Now lock the child directory inode and the parent directory
3116 * inode in the proper order. This will take care of validating
3117 * that the directory entry for the child directory inode has
3118 * not changed while we were obtaining a log reservation.
3119 */
3125 }
3129 /*
3130 * Only increment the parent directory vnode count if
3131 * we didn't bump it in looking up cdp. The only time
3132 * we don't bump it is when we're looking up ".".
3133 */
3135 }
3144 }
3148 }
3157 /*
3158 * Bump the in memory generation count on the parent
3159 * directory so that other can know that it has changed.
3160 */
3163 /*
3164 * Drop the link from cdp's "..".
3165 */
3169 }
3171 /*
3172 * Drop the link from dp to cdp.
3173 */
3177 }
3179 /*
3180 * Drop the "." link from cdp to self.
3181 */
3185 }
3187 /* Determine these before committing transaction */
3190 /*
3191 * Take an extra ref on the child vnode so that it
3192 * does not go to xfs_inactive() from within the commit.
3193 */
3196 /*
3197 * If this is a synchronous mount, make sure that the
3198 * rmdir transaction goes to disk before returning to
3199 * the user.
3200 */
3203 }
3209 XFS_TRANS_ABORT));
3212 }
3218 }
3221 /*
3222 * Let interposed file systems know about removed links.
3223 */
3228 /* Fall through to std_return with error = 0 or the errno
3229 * from xfs_trans_commit. */
3230 std_return:
3237 }
3240 error1:
3243 /* FALLTHROUGH */
3245 error_return:
3248 }
3251 /*
3252 * Read dp's entries starting at uiop->uio_offset and translate them into
3253 * bufsize bytes worth of struct dirents starting at bufbase.
3254 */
3255 STATIC int
3261 {
3265 uint lock_mode;
3278 }
3281 STATIC int
3289 {
3296 xfs_bmap_free_t free_list;
3297 xfs_fsblock_t first_block;
3298 boolean_t dp_joined_to_trans;
3300 uint cancel_flags;
3302 xfs_fileoff_t first_fsb;
3303 xfs_filblks_t fs_blocks;
3306 xfs_daddr_t d;
3311 xfs_prid_t prid;
3313 uint resblks;
3334 /*
3335 * Check component lengths of the target path name.
3336 */
3345 /*
3346 * Skip any slashes.
3347 */
3349 total++;
3350 path++;
3351 }
3353 /*
3354 * Count up to the next slash or end of path.
3355 * Error out if the component is bigger than MAXNAMELEN.
3356 */
3361 }
3362 }
3363 }
3364 }
3372 }
3374 /* Return through std_return after this point. */
3381 else
3384 /*
3385 * Make sure that we have allocated dquot(s) on disk.
3386 */
3395 /*
3396 * The symlink will fit into the inode data fork?
3397 * There can't be any attributes so we get the whole variable part.
3398 */
3401 else
3410 }
3415 }
3419 /*
3420 * Check whether the directory allows new symlinks or not.
3421 */
3425 }
3427 /*
3428 * Reserve disk quota : blocks and inode.
3429 */
3434 /*
3435 * Check for ability to enter directory entry, if no space reserved.
3436 */
3440 /*
3441 * Initialize the bmap freelist prior to calling either
3442 * bmapi or the directory create code.
3443 */
3446 /*
3447 * Allocate an inode for the symlink.
3448 */
3455 }
3462 /*
3463 * Also attach the dquot(s) to it, if applicable.
3464 */
3469 /*
3470 * If the symlink will fit into the inode, write it inline.
3471 */
3477 /*
3478 * The inode was initially created in extent format.
3479 */
3496 }
3512 }
3519 }
3520 }
3522 /*
3523 * Create the directory entry for the symlink.
3524 */
3532 /*
3533 * Bump the in memory version number of the parent directory
3534 * so that other processes accessing it will recognize that
3535 * the directory has changed.
3536 */
3539 /*
3540 * If this is a synchronous mount, make sure that the
3541 * symlink transaction goes to disk before returning to
3542 * the user.
3543 */
3546 }
3548 /*
3549 * xfs_trans_commit normally decrements the vnode ref count
3550 * when it unlocks the inode. Since we want to return the
3551 * vnode to the caller, we bump the vnode ref count now.
3552 */
3558 }
3563 /* Fall through to std_return with error = 0 or errno from
3564 * xfs_trans_commit */
3565 std_return:
3567 DM_EVENT_POSTSYMLINK)) {
3573 }
3581 }
3584 error2:
3586 error1:
3589 error_return:
3596 }
3599 }
3602 STATIC int
3606 {
3615 /*
3616 * use memcpy because the inode is a long long and there's no
3617 * assurance that xfid->fid_ino is properly aligned.
3618 */
3623 }
3626 /*
3627 * xfs_rwlock
3628 */
3629 int
3632 bhv_vrwlock_t locktype)
3633 {
3651 }
3654 }
3657 /*
3658 * xfs_rwunlock
3659 */
3660 void
3663 bhv_vrwlock_t locktype)
3664 {
3673 /*
3674 * In the write case, we may have added a new entry to
3675 * the reference cache. This might store a pointer to
3676 * an inode to be released in this inode. If it is there,
3677 * clear the pointer and release the inode after unlocking
3678 * this one.
3679 */
3685 }
3687 }
3689 STATIC int
3693 {
3706 /*
3707 * Bypass inodes which have already been cleaned by
3708 * the inode flush clustering code inside xfs_iflush
3709 */
3717 xfs_lsn_t sync_lsn;
3730 }
3731 }
3733 /*
3734 * We make this non-blocking if the inode is contended,
3735 * return EAGAIN to indicate to the caller that they
3736 * did not succeed. This prevents the flush path from
3737 * blocking on inodes inside another operation right
3738 * now, they get caught later by xfs_sync.
3739 */
3753 }
3756 }
3760 else
3765 }
3768 }
3770 int
3773 u_int evmask,
3774 u_int16_t state,
3776 {
3796 }
3808 }
3810 STATIC int
3813 {
3824 /* bad inode, get out here ASAP */
3828 }
3834 /*
3835 * Make sure the atime in the XFS inode is correct before freeing the
3836 * Linux inode.
3837 */
3840 /*
3841 * If we have nothing to flush with this inode then complete the
3842 * teardown now, otherwise break the link between the xfs inode and the
3843 * linux inode and clean up the xfs inode later. This avoids flushing
3844 * the inode to disk during the delete operation itself.
3845 *
3846 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3847 * first to ensure that xfs_iunpin() will never see an xfs inode
3848 * that has a linux inode being reclaimed. Synchronisation is provided
3849 * by the i_flags_lock.
3850 */
3858 /* Protect sync and unpin from us */
3866 }
3868 }
3870 int
3875 {
3883 /* The hash lock here protects a thread in xfs_iget_core from
3884 * racing with us on linking the inode back with a vnode.
3885 * Once we have the XFS_IRECLAIM flag set it will not touch
3886 * us.
3887 */
3897 }
3899 }
3904 /*
3905 * If the inode is still dirty, then flush it out. If the inode
3906 * is not in the AIL, then it will be OK to flush it delwri as
3907 * long as xfs_iflush() does not keep any references to the inode.
3908 * We leave that decision up to xfs_iflush() since it has the
3909 * knowledge of whether it's OK to simply do a delwri flush of
3910 * the inode or whether we need to wait until the inode is
3911 * pulled from the AIL.
3912 * We get the flush lock regardless, though, just to make sure
3913 * we don't free it while it is being flushed.
3914 */
3919 }
3925 /*
3926 * If we hit an error, typically because of filesystem
3927 * shutdown, we don't need to let vn_reclaim to know
3928 * because we're gonna reclaim the inode anyway.
3929 */
3933 }
3935 }
3942 /*
3943 * We are not interested in doing an iflush if we're
3944 * in the process of shutting down the filesystem forcibly.
3945 * So, just reclaim the inode.
3946 */
3949 }
3951 reclaim:
3954 }
3956 int
3958 {
3974 }
3975 }
3978 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3982 }
3985 }
3989 }
3991 /*
3992 * xfs_alloc_file_space()
3993 * This routine allocates disk space for the given file.
3994 *
3995 * If alloc_type == 0, this request is for an ALLOCSP type
3996 * request which will change the file size. In this case, no
3997 * DMAPI event will be generated by the call. A TRUNCATE event
3998 * will be generated later by xfs_setattr.
3999 *
4000 * If alloc_type != 0, this request is for a RESVSP type
4001 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4002 * lower block boundary byte address is less than the file's
4003 * length.
4004 *
4005 * RETURNS:
4006 * 0 on success
4007 * errno on error
4008 *
4009 */
4010 STATIC int
4013 xfs_off_t offset,
4014 xfs_off_t len,
4017 {
4019 xfs_off_t count;
4020 xfs_filblks_t allocated_fsb;
4021 xfs_filblks_t allocatesize_fsb;
4023 xfs_fileoff_t startoffset_fsb;
4024 xfs_fsblock_t firstfsb;
4031 xfs_bmap_free_t free_list;
4047 }
4063 /* Generate a DMAPI event if needed. */
4067 xfs_off_t end_dmi_offset;
4077 }
4079 /*
4080 * Allocate file space until done or until there is an error
4081 */
4082 retry:
4086 /*
4087 * Determine space reservations for data/realtime.
4088 */
4101 }
4113 }
4115 /*
4116 * Allocate and setup the transaction.
4117 */
4121 XFS_TRANS_PERM_LOG_RES,
4122 XFS_WRITE_LOG_COUNT);
4123 /*
4124 * Check for running out of space
4125 */
4127 /*
4128 * Free the transaction structure.
4129 */
4133 }
4143 /*
4144 * Issue the xfs_bmapi() call to allocate the blocks
4145 */
4153 }
4155 /*
4156 * Complete the transaction
4157 */
4161 }
4167 }
4174 }
4178 }
4179 dmapi_enospc_check:
4189 /* else fall through with error from XFS_SEND_DATA */
4190 }
4202 }
4204 /*
4205 * Zero file bytes between startoff and endoff inclusive.
4206 * The iolock is held exclusive and no blocks are buffered.
4207 */
4208 STATIC int
4211 xfs_off_t startoff,
4212 xfs_off_t endoff)
4213 {
4214 xfs_bmbt_irec_t imap;
4215 xfs_fileoff_t offset_fsb;
4216 xfs_off_t lastoffset;
4217 xfs_off_t offset;
4253 }
4265 }
4266 }
4269 }
4271 /*
4272 * xfs_free_file_space()
4273 * This routine frees disk space for the given file.
4274 *
4275 * This routine is only called by xfs_change_file_space
4276 * for an UNRESVSP type call.
4277 *
4278 * RETURNS:
4279 * 0 on success
4280 * errno on error
4281 *
4282 */
4283 STATIC int
4286 xfs_off_t offset,
4287 xfs_off_t len,
4289 {
4293 xfs_off_t end_dmi_offset;
4294 xfs_fileoff_t endoffset_fsb;
4296 xfs_fsblock_t firstfsb;
4297 xfs_bmap_free_t free_list;
4298 xfs_bmbt_irec_t imap;
4299 xfs_off_t ioffset;
4303 uint resblks;
4304 uint rounding;
4306 xfs_fileoff_t startoffset_fsb;
4336 }
4343 }
4355 }
4357 /*
4358 * Need to zero the stuff we're not freeing, on disk.
4359 * If its a realtime file & can't use unwritten extents then we
4360 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4361 * will take care of it for us.
4362 */
4371 xfs_daddr_t block;
4378 }
4387 mod++;
4390 }
4391 }
4393 /*
4394 * One contiguous piece to clear
4395 */
4398 /*
4399 * Some full blocks, possibly two pieces to clear
4400 */
4409 }
4411 /*
4412 * free file space until done or until there is an error
4413 */
4417 /*
4418 * allocate and setup the transaction
4419 */
4422 resblks,
4425 XFS_TRANS_PERM_LOG_RES,
4426 XFS_WRITE_LOG_COUNT);
4428 /*
4429 * check for running out of space
4430 */
4432 /*
4433 * Free the transaction structure.
4434 */
4438 }
4442 XFS_QMOPT_RES_REGBLKS);
4449 /*
4450 * issue the bunmapi() call to free the blocks
4451 */
4458 }
4460 /*
4461 * complete the transaction
4462 */
4466 }
4470 }
4472 out_unlock_iolock:
4477 error0:
4479 error1:
4482 XFS_ILOCK_EXCL);
4484 }
4486 /*
4487 * xfs_change_file_space()
4488 * This routine allocates or frees disk space for the given file.
4489 * The user specified parameters are checked for alignment and size
4490 * limitations.
4491 *
4492 * RETURNS:
4493 * 0 on success
4494 * errno on error
4495 *
4496 */
4497 int
4502 xfs_off_t offset,
4505 {
4508 xfs_fsize_t fsize;
4512 xfs_off_t startoffset;
4513 xfs_off_t llen;
4515 bhv_vattr_t va;
4524 /*
4525 * must be a regular file and have write permission
4526 */
4535 }
4550 }
4565 /*
4566 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4567 * file space.
4568 * These calls do NOT zero the data space allocated to the file,
4569 * nor do they change the file size.
4570 *
4571 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4572 * space.
4573 * These calls cause the new file data to be zeroed and the file
4574 * size to be changed.
4575 */
4591 attr_flags)))
4604 }
4620 }
4622 /*
4623 * update the inode timestamp, mode, and prealloc flag bits
4624 */
4629 /* ASSERT(0); */
4632 }
4642 /*
4643 * Note that we don't have to worry about mandatory
4644 * file locking being disabled here because we only
4645 * clear the S_ISGID bit if the Group execute bit is
4646 * on, but if it was on then mandatory locking wouldn't
4647 * have been enabled.
4648 */
4653 }
4667 }
4669 bhv_vnodeops_t xfs_vnodeops = {
4674 #ifdef HAVE_SPLICE
4677 #endif
4711 };