| blob | 64c5953feca4b15513d23164b66fb46175b2833d |
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 */
57 int
60 {
66 /*
67 * If it's a directory with any blocks, read-ahead block 0
68 * as we're almost certain to have the next operation be a read there.
69 */
75 }
77 }
79 /*
80 * xfs_getattr
81 */
82 int
87 {
106 #if XFS_BIG_INUMS
108 #endif
111 /*
112 * Quick exit for non-stat callers
113 */
119 /*
120 * Copy from in-core inode.
121 */
127 /*
128 * Check vnode type block/char vs. everything else.
129 */
143 /*
144 * If the file blocks are being allocated from a
145 * realtime partition, then return the inode's
146 * realtime extent size or the realtime volume's
147 * extent size.
148 */
151 }
153 }
161 /*
162 * Exit for stat callers. See if any of the rest of the fields
163 * to be filled in are needed.
164 */
170 /*
171 * Convert di_flags to xflags.
172 */
175 /*
176 * Exit for inode revalidate. See if any of the rest of
177 * the fields to be filled in are needed.
178 */
194 else
198 all_done:
202 }
205 /*
206 * xfs_setattr
207 */
208 int
214 {
220 uint lock_flags;
236 /*
237 * Cannot set certain attributes.
238 */
242 }
247 /*
248 * Timestamps do not need to be logged and hence do not
249 * need to be done within a transaction.
250 */
258 }
263 /*
264 * If disk quotas is on, we make sure that the dquots do exist on disk,
265 * before we start any other transactions. Trying to do this later
266 * is messy. We don't care to take a readlock to look at the ids
267 * in inode here, because we can't hold it across the trans_reserve.
268 * If the IDs do change before we take the ilock, we're covered
269 * because the i_*dquot fields will get updated anyway.
270 */
280 }
286 }
292 }
293 /*
294 * We take a reference when we initialize udqp and gdqp,
295 * so it is important that we never blindly double trip on
296 * the same variable. See xfs_create() for an example.
297 */
304 }
306 /*
307 * For the other attributes, we acquire the inode lock and
308 * first do an error checking pass.
309 */
324 }
325 }
335 }
336 }
339 }
343 /* boolean: are we the file owner? */
346 /*
347 * Change various properties of a file.
348 * Only the owner or users with CAP_FOWNER
349 * capability may do these things.
350 */
354 /*
355 * CAP_FOWNER overrides the following restrictions:
356 *
357 * The user ID of the calling process must be equal
358 * to the file owner ID, except in cases where the
359 * CAP_FSETID capability is applicable.
360 */
364 }
366 /*
367 * CAP_FSETID overrides the following restrictions:
368 *
369 * The effective user ID of the calling process shall match
370 * the file owner when setting the set-user-ID and
371 * set-group-ID bits on that file.
372 *
373 * The effective group ID or one of the supplementary group
374 * IDs of the calling process shall match the group owner of
375 * the file when setting the set-group-ID bit on that file
376 */
385 #if 0
386 /* Linux allows this, Irix doesn't. */
389 #endif
392 }
393 }
395 /*
396 * Change file ownership. Must be the owner or privileged.
397 * If the system was configured with the "restricted_chown"
398 * option, the owner is not permitted to give away the file,
399 * and can change the group id only to a group of which he
400 * or she is a member.
401 */
403 /*
404 * These IDs could have changed since we last looked at them.
405 * But, we're assured that if the ownership did change
406 * while we didn't have the inode locked, inode's dquot(s)
407 * would have changed also.
408 */
415 iprojid;
417 /*
418 * CAP_CHOWN overrides the following restrictions:
419 *
420 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
421 * shall override the restriction that a process cannot
422 * change the user ID of a file it owns and the restriction
423 * that the group ID supplied to the chown() function
424 * shall be equal to either the group ID or one of the
425 * supplementary group IDs of the calling process.
426 */
433 }
434 /*
435 * Do a quota reservation only if uid/projid/gid is actually
436 * going to change.
437 */
447 }
448 }
450 /*
451 * Truncate file. Must have write permission and not be a directory.
452 */
454 /* Short circuit the truncate case for zero length files */
463 }
471 }
472 /*
473 * Make sure that the dquots are attached to the inode.
474 */
477 }
479 /*
480 * Change file access or modified times.
481 */
488 }
489 }
490 }
492 /*
493 * Change extent size or realtime flag.
494 */
496 /*
497 * Can't change extent size if any extents are allocated.
498 */
504 }
506 /*
507 * Can't change realtime flag if any extents are allocated.
508 */
515 }
516 /*
517 * Extent size must be a multiple of the appropriate block
518 * size, if set at all.
519 */
521 xfs_extlen_t size;
530 }
534 }
535 }
536 /*
537 * If realtime flag is set then must have realtime data.
538 */
546 }
547 }
549 /*
550 * Can't modify an immutable/append-only file unless
551 * we have appropriate permission.
552 */
561 }
562 }
564 /*
565 * Now we can make the changes. Before we join the inode
566 * to the transaction, if XFS_AT_SIZE is set then take care of
567 * the part of the truncation that must be done without the
568 * inode lock. This needs to be done before joining the inode
569 * to the transaction, because the inode cannot be unlocked
570 * once it is a part of the transaction.
571 */
577 }
580 /*
581 * We are going to log the inode size change in this
582 * transaction so any previous writes that are beyond the on
583 * disk EOF and the new EOF that have not been written out need
584 * to be written here. If we do not write the data out, we
585 * expose ourselves to the null files problem.
586 *
587 * Only flush from the on disk size to the smaller of the in
588 * memory file size or the new size as that's the range we
589 * really care about here and prevents waiting for other data
590 * not within the range we care about here.
591 */
598 }
600 /* wait for all I/O to complete */
610 }
614 XFS_TRANS_PERM_LOG_RES,
615 XFS_ITRUNCATE_LOG_COUNT))) {
620 }
623 }
628 }
630 /* determine whether mandatory locking mode changes */
633 /*
634 * Truncate file. Must have write permission and not be a directory.
635 */
642 /*
643 * signal a sync transaction unless
644 * we're truncating an already unlinked
645 * file on a wsync filesystem
646 */
649 XFS_DATA_FORK,
655 /*
656 * Truncated "down", so we're removing references
657 * to old data here - if we now delay flushing for
658 * a long time, we expose ourselves unduly to the
659 * notorious NULL files problem. So, we mark this
660 * vnode and flush it when the file is closed, and
661 * do not wait the usual (long) time for writeout.
662 */
664 }
665 /*
666 * Have to do this even if the file's size doesn't change.
667 */
669 }
671 /*
672 * Change file access modes.
673 */
680 }
682 /*
683 * Change file ownership. Must be the owner or privileged.
684 * If the system was configured with the "restricted_chown"
685 * option, the owner is not permitted to give away the file,
686 * and can change the group id only to a group of which he
687 * or she is a member.
688 */
690 /*
691 * CAP_FSETID overrides the following restrictions:
692 *
693 * The set-user-ID and set-group-ID bits of a file will be
694 * cleared upon successful return from chown()
695 */
699 }
701 /*
702 * Change the ownerships and register quota modifications
703 * in the transaction.
704 */
711 }
713 }
721 }
723 }
731 }
733 /*
734 * We may have to rev the inode as well as
735 * the superblock version number since projids didn't
736 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
737 */
740 }
744 }
747 /*
748 * Change file access or modified times.
749 */
756 }
762 }
765 }
767 /*
768 * Change XFS-added attributes.
769 */
772 /*
773 * Converting bytes to fs blocks.
774 */
777 }
779 uint di_flags;
781 /* can't set PREALLOC this way, just preserve it */
811 }
813 }
816 }
818 /*
819 * Change file inode change time only if XFS_AT_CTIME set
820 * AND we have been called by a DMI function.
821 */
828 }
830 /*
831 * Send out timestamp changes that need to be set to the
832 * current time. Not done when called by a DMI function.
833 */
839 /*
840 * If this is a synchronous mount, make sure that the
841 * transaction goes to disk before returning to the user.
842 * This is slightly sub-optimal in that truncates require
843 * two sync transactions instead of one for wsync filesystems.
844 * One for the truncate and one for the timestamps since we
845 * don't want to change the timestamps unless we're sure the
846 * truncate worked. Truncates are less than 1% of the laddis
847 * mix so this probably isn't worth the trouble to optimize.
848 */
855 }
857 /*
858 * If the (regular) file's mandatory locking mode changed, then
859 * notify the vnode. We do this under the inode lock to prevent
860 * racing calls to vop_vnode_change.
861 */
866 /*
867 * Release any dquot(s) the inode had kept before chown.
868 */
876 }
883 }
886 abort_return:
888 /* FALLTHROUGH */
889 error_return:
894 }
897 }
899 }
901 /*
902 * The maximum pathlen is 1024 bytes. Since the minimum file system
903 * blocksize is 512 bytes, we can get a max of 2 extents back from
904 * bmapi.
905 */
906 #define SYMLINK_MAPS 2
908 STATIC int
912 {
917 xfs_daddr_t d;
939 }
946 }
951 out:
953 }
955 int
959 {
983 }
985 out:
988 }
990 /*
991 * xfs_fsync
992 *
993 * This is called to sync the inode and its data out to disk.
994 * We need to hold the I/O lock while flushing the data, and
995 * the inode lock while flushing the inode. The inode lock CANNOT
996 * be held while flushing the data, so acquire after we're done
997 * with that.
998 */
999 int
1003 xfs_off_t start,
1004 xfs_off_t stop)
1005 {
1020 /*
1021 * We always need to make sure that the required inode state
1022 * is safe on disk. The vnode might be clean but because
1023 * of committed transactions that haven't hit the disk yet.
1024 * Likewise, there could be unflushed non-transactional
1025 * changes to the inode core that have to go to disk.
1026 *
1027 * The following code depends on one assumption: that
1028 * any transaction that changes an inode logs the core
1029 * because it has to change some field in the inode core
1030 * (typically nextents or nblocks). That assumption
1031 * implies that any transactions against an inode will
1032 * catch any non-transactional updates. If inode-altering
1033 * transactions exist that violate this assumption, the
1034 * code breaks. Right now, it figures that if the involved
1035 * update_* field is clear and the inode is unpinned, the
1036 * inode is clean. Either it's been flushed or it's been
1037 * committed and the commit has hit the disk unpinning the inode.
1038 * (Note that xfs_inode_item_format() called at commit clears
1039 * the update_* fields.)
1040 */
1043 /* If we are flushing data then we care about update_size
1044 * being set, otherwise we care about update_core
1045 */
1049 /*
1050 * Timestamps/size haven't changed since last inode
1051 * flush or inode transaction commit. That means
1052 * either nothing got written or a transaction
1053 * committed which caught the updates. If the
1054 * latter happened and the transaction hasn't
1055 * hit the disk yet, the inode will be still
1056 * be pinned. If it is, force the log.
1057 */
1063 XFS_LOG_FORCE |
1068 /*
1069 * If the inode is not pinned and nothing
1070 * has changed we don't need to flush the
1071 * cache.
1072 */
1074 }
1077 /*
1078 * Kick off a transaction to log the inode
1079 * core to get the updates. Make it
1080 * sync if FSYNC_WAIT is passed in (which
1081 * is done by everybody but specfs). The
1082 * sync transaction will also force the log.
1083 */
1091 }
1094 /*
1095 * Note - it's possible that we might have pushed
1096 * ourselves out of the way during trans_reserve
1097 * which would flush the inode. But there's no
1098 * guarantee that the inode buffer has actually
1099 * gone out yet (it's delwri). Plus the buffer
1100 * could be pinned anyway if it's part of an
1101 * inode in another recent transaction. So we
1102 * play it safe and fire off the transaction anyway.
1103 */
1112 }
1115 /*
1116 * If the log write didn't issue an ordered tag we need
1117 * to flush the disk cache for the data device now.
1118 */
1122 /*
1123 * If this inode is on the RT dev we need to flush that
1124 * cache as well.
1125 */
1128 }
1131 }
1133 /*
1134 * This is called by xfs_inactive to free any blocks beyond eof
1135 * when the link count isn't zero and by xfs_dm_punch_hole() when
1136 * punching a hole to EOF.
1137 */
1138 int
1143 {
1146 xfs_fileoff_t end_fsb;
1147 xfs_fileoff_t last_fsb;
1148 xfs_filblks_t map_len;
1150 xfs_bmbt_irec_t imap;
1153 /*
1154 * Figure out if there are any blocks beyond the end
1155 * of the file. If not, then there is nothing to do.
1156 */
1172 /*
1173 * Attach the dquots to the inode up front.
1174 */
1178 /*
1179 * There are blocks after the end of file.
1180 * Free them up now by truncating the file to
1181 * its current size.
1182 */
1185 /*
1186 * Do the xfs_itruncate_start() call before
1187 * reserving any log space because
1188 * itruncate_start will call into the buffer
1189 * cache and we can't
1190 * do that within a transaction.
1191 */
1201 }
1206 XFS_ITRUNCATE_LOG_COUNT);
1212 }
1216 XFS_IOLOCK_EXCL |
1217 XFS_ILOCK_EXCL);
1222 XFS_DATA_FORK,
1224 /*
1225 * If we get an error at this point we
1226 * simply don't bother truncating the file.
1227 */
1231 XFS_TRANS_ABORT));
1234 XFS_TRANS_RELEASE_LOG_RES);
1235 }
1238 }
1240 }
1242 /*
1243 * Free a symlink that has blocks associated with it.
1244 */
1245 STATIC int
1249 {
1254 xfs_fsblock_t first_block;
1255 xfs_bmap_free_t free_list;
1267 /*
1268 * We're freeing a symlink that has some
1269 * blocks allocated to it. Free the
1270 * blocks here. We know that we've got
1271 * either 1 or 2 extents and that we can
1272 * free them all in one bunmapi call.
1273 */
1281 }
1282 /*
1283 * Lock the inode, fix the size, and join it to the transaction.
1284 * Hold it so in the normal path, we still have it locked for
1285 * the second transaction. In the error paths we need it
1286 * held so the cancel won't rele it, see below.
1287 */
1294 /*
1295 * Find the block(s) so we can inval and unmap them.
1296 */
1304 /*
1305 * Invalidate the block(s).
1306 */
1312 }
1313 /*
1314 * Unmap the dead block(s) to the free_list.
1315 */
1320 /*
1321 * Commit the first transaction. This logs the EFI and the inode.
1322 */
1325 /*
1326 * The transaction must have been committed, since there were
1327 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1328 * The new tp has the extent freeing and EFDs.
1329 */
1331 /*
1332 * The first xact was committed, so add the inode to the new one.
1333 * Mark it dirty so it will be logged and moved forward in the log as
1334 * part of every commit.
1335 */
1339 /*
1340 * Get a new, empty transaction to return to our caller.
1341 */
1343 /*
1344 * Commit the transaction containing extent freeing and EFDs.
1345 * If we get an error on the commit here or on the reserve below,
1346 * we need to unlock the inode since the new transaction doesn't
1347 * have the inode attached.
1348 */
1354 }
1355 /*
1356 * Remove the memory for extent descriptions (just bookkeeping).
1357 */
1361 /*
1362 * Put an itruncate log reservation in the new transaction
1363 * for our caller.
1364 */
1369 }
1370 /*
1371 * Return with the inode locked but not joined to the transaction.
1372 */
1376 error1:
1378 error0:
1379 /*
1380 * Have to come here with the inode locked and either
1381 * (held and in the transaction) or (not in the transaction).
1382 * If the inode isn't held then cancel would iput it, but
1383 * that's wrong since this is inactive and the vnode ref
1384 * count is 0 already.
1385 * Cancel won't do anything to the inode if held, but it still
1386 * needs to be locked until the cancel is done, if it was
1387 * joined to the transaction.
1388 */
1394 }
1396 STATIC int
1400 {
1404 /*
1405 * We're freeing a symlink which fit into
1406 * the inode. Just free the memory used
1407 * to hold the old symlink.
1408 */
1412 XFS_ITRUNCATE_LOG_COUNT);
1418 }
1421 /*
1422 * Zero length symlinks _can_ exist.
1423 */
1427 XFS_DATA_FORK);
1429 }
1431 }
1433 STATIC int
1437 {
1454 }
1460 XFS_INACTIVE_LOG_COUNT);
1467 }
1478 }
1480 int
1483 {
1491 /* If this is a read-only mount, don't do this (would generate I/O) */
1498 /*
1499 * If we are using filestreams, and we have an unlinked
1500 * file that we are processing the last close on, then nothing
1501 * will be able to reopen and write to this file. Purge this
1502 * inode from the filestreams cache so that it doesn't delay
1503 * teardown of the inode.
1504 */
1508 /*
1509 * If we previously truncated this file and removed old data
1510 * in the process, we want to initiate "early" writeout on
1511 * the last close. This is an attempt to combat the notorious
1512 * NULL files problem which is particularly noticable from a
1513 * truncate down, buffered (re-)write (delalloc), followed by
1514 * a crash. What we are effectively doing here is
1515 * significantly reducing the time window where we'd otherwise
1516 * be exposed to that problem.
1517 */
1521 }
1523 #ifdef HAVE_REFCACHE
1524 /* If we are in the NFS reference cache then don't do this now */
1527 #endif
1539 }
1540 }
1543 }
1545 /*
1546 * xfs_inactive
1547 *
1548 * This is called when the vnode reference count for the vnode
1549 * goes to zero. If the file has been unlinked, then it must
1550 * now be truncated. Also, we clear all of the read-ahead state
1551 * kept for the inode here since the file is now closed.
1552 */
1553 int
1556 {
1558 xfs_bmap_free_t free_list;
1559 xfs_fsblock_t first_block;
1568 /*
1569 * If the inode is already free, then there can be nothing
1570 * to clean up here.
1571 */
1576 }
1578 /*
1579 * Only do a truncate if it's a regular file with
1580 * some actual space in it. It's OK to look at the
1581 * inode's fields without the lock because we're the
1582 * only one with a reference to the inode.
1583 */
1593 }
1597 /* If this is a read-only mount, don't do this (would generate I/O) */
1612 }
1614 }
1623 /*
1624 * Do the xfs_itruncate_start() call before
1625 * reserving any log space because itruncate_start
1626 * will call into the buffer cache and we can't
1627 * do that within a transaction.
1628 */
1636 }
1641 XFS_ITRUNCATE_LOG_COUNT);
1643 /* Don't call itruncate_cleanup */
1648 }
1654 /*
1655 * normally, we have to run xfs_itruncate_finish sync.
1656 * But if filesystem is wsync and we're in the inactive
1657 * path, then we know that nlink == 0, and that the
1658 * xaction that made nlink == 0 is permanently committed
1659 * since xfs_remove runs as a synchronous transaction.
1660 */
1669 }
1672 /*
1673 * If we get an error while cleaning up a
1674 * symlink we bail out.
1675 */
1683 }
1691 XFS_INACTIVE_LOG_COUNT);
1696 }
1701 }
1703 /*
1704 * If there are attributes associated with the file
1705 * then blow them away now. The code calls a routine
1706 * that recursively deconstructs the attribute fork.
1707 * We need to just commit the current transaction
1708 * because we can't use it for xfs_attr_inactive().
1709 */
1712 /*
1713 * If we got an error, the transaction is already
1714 * cancelled, and the inode is unlocked. Just get out.
1715 */
1720 }
1722 /*
1723 * Free the inode.
1724 */
1728 /*
1729 * If we fail to free the inode, shut down. The cancel
1730 * might do that, we need to make sure. Otherwise the
1731 * inode might be lost for a long time or forever.
1732 */
1738 }
1741 /*
1742 * Credit the quota account(s). The inode is gone.
1743 */
1746 /*
1747 * Just ignore errors at this point. There is
1748 * nothing we can do except to try to keep going.
1749 */
1752 }
1753 /*
1754 * Release the dquots held by inode, if any.
1755 */
1760 out:
1762 }
1765 int
1770 {
1772 xfs_ino_t e_inum;
1774 uint lock_mode;
1786 }
1789 }
1791 int
1795 mode_t mode,
1796 xfs_dev_t rdev,
1799 {
1807 xfs_bmap_free_t free_list;
1808 xfs_fsblock_t first_block;
1811 uint cancel_flags;
1813 xfs_prid_t prid;
1815 uint resblks;
1832 }
1837 /* Return through std_return after this point. */
1842 else
1845 /*
1846 * Make sure that we have allocated dquot(s) on disk.
1847 */
1859 /*
1860 * Initially assume that the file does not exist and
1861 * reserve the resources for that case. If that is not
1862 * the case we'll drop the one we have and get a more
1863 * appropriate transaction later.
1864 */
1871 }
1875 }
1884 /*
1885 * Reserve disk quota and the inode.
1886 */
1900 }
1903 /*
1904 * At this point, we've gotten a newly allocated inode.
1905 * It is locked (and joined to the transaction).
1906 */
1910 /*
1911 * Now we join the directory inode to the transaction. We do not do it
1912 * earlier because xfs_dir_ialloc might commit the previous transaction
1913 * (and release all the locks). An error from here on will result in
1914 * the transaction cancel unlocking dp so don't do it explicitly in the
1915 * error path.
1916 */
1927 }
1931 /*
1932 * If this is a synchronous mount, make sure that the
1933 * create transaction goes to disk before returning to
1934 * the user.
1935 */
1938 }
1942 /*
1943 * Attach the dquot(s) to the inodes and modify them incore.
1944 * These ids of the inode couldn't have changed since the new
1945 * inode has been locked ever since it was created.
1946 */
1949 /*
1950 * xfs_trans_commit normally decrements the vnode ref count
1951 * when it unlocks the inode. Since we want to return the
1952 * vnode to the caller, we bump the vnode ref count now.
1953 */
1961 }
1968 }
1975 /* Fallthrough to std_return with error = 0 */
1977 std_return:
1985 }
1988 abort_return:
1990 /* FALLTHROUGH */
1992 error_return:
2004 abort_rele:
2005 /*
2006 * Wait until after the current transaction is aborted to
2007 * release the inode. This prevents recursive transactions
2008 * and deadlocks from xfs_inactive.
2009 */
2018 }
2020 #ifdef DEBUG
2021 /*
2022 * Some counters to see if (and how often) we are hitting some deadlock
2023 * prevention code paths.
2024 */
2029 #endif
2031 /*
2032 * The following routine will lock the inodes associated with the
2033 * directory and the named entry in the directory. The locks are
2034 * acquired in increasing inode number.
2035 *
2036 * If the entry is "..", then only the directory is locked. The
2037 * vnode ref count will still include that from the .. entry in
2038 * this case.
2039 *
2040 * There is a deadlock we need to worry about. If the locked directory is
2041 * in the AIL, it might be blocking up the log. The next inode we lock
2042 * could be already locked by another thread waiting for log space (e.g
2043 * a permanent log reservation with a long running transaction (see
2044 * xfs_itruncate_finish)). To solve this, we must check if the directory
2045 * is in the ail and use lock_nowait. If we can't lock, we need to
2046 * drop the inode lock on the directory and try again. xfs_iunlock will
2047 * potentially push the tail if we were holding up the log.
2048 */
2049 STATIC int
2053 {
2055 xfs_ino_t e_inum;
2059 #ifdef DEBUG
2060 xfs_rm_locks++;
2061 #endif
2064 again:
2071 /*
2072 * We want to lock in increasing inum. Since we've already
2073 * acquired the lock on the directory, we may need to release
2074 * if if the inum of the entry turns out to be less.
2075 */
2077 /*
2078 * We are already in the right order, so just
2079 * lock on the inode of the entry.
2080 * We need to use nowait if dp is in the AIL.
2081 */
2086 attempts++;
2087 #ifdef DEBUG
2088 xfs_rm_attempts++;
2089 #endif
2091 /*
2092 * Unlock dp and try again.
2093 * xfs_iunlock will try to push the tail
2094 * if the inode is in the AIL.
2095 */
2101 #ifdef DEBUG
2102 xfs_rm_lock_delays++;
2103 #endif
2104 }
2106 }
2109 }
2116 }
2117 /* else e_inum == dp->i_ino */
2118 /* This can happen if we're asked to lock /x/..
2119 * the entry is "..", which is also the parent directory.
2120 */
2123 }
2125 #ifdef DEBUG
2131 #endif
2133 /*
2134 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2135 * a different value
2136 */
2139 {
2146 }
2148 /*
2149 * The following routine will lock n inodes in exclusive mode.
2150 * We assume the caller calls us with the inodes in i_ino order.
2151 *
2152 * We need to detect deadlock where an inode that we lock
2153 * is in the AIL and we start waiting for another inode that is locked
2154 * by a thread in a long running transaction (such as truncate). This can
2155 * result in deadlock since the long running trans might need to wait
2156 * for the inode we just locked in order to push the tail and free space
2157 * in the log.
2158 */
2159 void
2164 uint lock_mode)
2165 {
2177 }
2179 again:
2186 /*
2187 * If try_lock is not set yet, make sure all locked inodes
2188 * are not in the AIL.
2189 * If any are, set try_lock to be used later.
2190 */
2196 try_lock++;
2197 }
2198 }
2199 }
2201 /*
2202 * If any of the previous locks we have locked is in the AIL,
2203 * we must TRY to get the second and subsequent locks. If
2204 * we can't get any, we must release all we have
2205 * and try again.
2206 */
2209 /* try_lock must be 0 if i is 0. */
2210 /*
2211 * try_lock means we have an inode locked
2212 * that is in the AIL.
2213 */
2216 attempts++;
2218 /*
2219 * Unlock all previous guys and try again.
2220 * xfs_iunlock will try to push the tail
2221 * if the inode is in the AIL.
2222 */
2226 /*
2227 * Check to see if we've already
2228 * unlocked this one.
2229 * Not the first one going back,
2230 * and the inode ptr is the same.
2231 */
2237 }
2241 #ifdef DEBUG
2242 xfs_lock_delays++;
2243 #endif
2244 }
2248 }
2251 }
2252 }
2254 #ifdef DEBUG
2260 xfs_locked_n++;
2261 }
2262 #endif
2263 }
2265 #ifdef DEBUG
2266 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2269 #define REMOVE_DEBUG_TRACE(x)
2272 int
2276 {
2283 xfs_bmap_free_t free_list;
2284 xfs_fsblock_t first_block;
2289 uint resblks;
2302 }
2310 }
2312 /* From this point on, return through std_return */
2315 /*
2316 * We need to get a reference to ip before we get our log
2317 * reservation. The reason for this is that we cannot call
2318 * xfs_iget for an inode for which we do not have a reference
2319 * once we've acquired a log reservation. This is because the
2320 * inode we are trying to get might be in xfs_inactive going
2321 * for a log reservation. Since we'll have to wait for the
2322 * inactive code to complete before returning from xfs_iget,
2323 * we need to make sure that we don't have log space reserved
2324 * when we call xfs_iget. Instead we get an unlocked reference
2325 * to the inode before getting our log reservation.
2326 */
2331 }
2345 }
2349 /*
2350 * We try to get the real space reservation first,
2351 * allowing for directory btree deletion(s) implying
2352 * possible bmap insert(s). If we can't get the space
2353 * reservation then we use 0 instead, and avoid the bmap
2354 * btree insert(s) in the directory code by, if the bmap
2355 * insert tries to happen, instead trimming the LAST
2356 * block from the directory.
2357 */
2365 }
2372 }
2380 }
2382 /*
2383 * At this point, we've gotten both the directory and the entry
2384 * inodes locked.
2385 */
2388 /*
2389 * Increment vnode ref count only in this case since
2390 * there's an extra vnode reference in the case where
2391 * dp == ip.
2392 */
2395 }
2397 /*
2398 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2399 */
2407 }
2417 }
2419 /* Determine if this is the last link while
2420 * we are in the transaction.
2421 */
2424 /*
2425 * Take an extra ref on the inode so that it doesn't
2426 * go to xfs_inactive() from within the commit.
2427 */
2430 /*
2431 * If this is a synchronous mount, make sure that the
2432 * remove transaction goes to disk before returning to
2433 * the user.
2434 */
2437 }
2443 }
2449 }
2451 /*
2452 * Before we drop our extra reference to the inode, purge it
2453 * from the refcache if it is there. By waiting until afterwards
2454 * to do the IRELE, we ensure that we won't go inactive in the
2455 * xfs_refcache_purge_ip routine (although that would be OK).
2456 */
2459 /*
2460 * If we are using filestreams, kill the stream association.
2461 * If the file is still open it may get a new one but that
2462 * will get killed on last close in xfs_close() so we don't
2463 * have to worry about that.
2464 */
2471 /* Fall through to std_return with error = 0 */
2472 std_return:
2478 }
2481 error1:
2487 error_rele:
2488 /*
2489 * In this case make sure to not release the inode until after
2490 * the current transaction is aborted. Releasing it beforehand
2491 * can cause us to go to xfs_inactive and start a recursive
2492 * transaction which can easily deadlock with the current one.
2493 */
2498 /*
2499 * Before we drop our extra reference to the inode, purge it
2500 * from the refcache if it is there. By waiting until afterwards
2501 * to do the IRELE, we ensure that we won't go inactive in the
2502 * xfs_refcache_purge_ip routine (although that would be OK).
2503 */
2509 }
2511 int
2516 {
2523 xfs_bmap_free_t free_list;
2524 xfs_fsblock_t first_block;
2547 }
2549 /* Return through std_return after this point. */
2566 }
2570 }
2578 }
2582 /*
2583 * Increment vnode ref counts since xfs_trans_commit &
2584 * xfs_trans_cancel will both unlock the inodes and
2585 * decrement the associated ref counts.
2586 */
2592 /*
2593 * If the source has too many links, we can't make any more to it.
2594 */
2598 }
2600 /*
2601 * If we are using project inheritance, we only allow hard link
2602 * creation in our tree when the project IDs are the same; else
2603 * the tree quota mechanism could be circumvented.
2604 */
2609 }
2619 resblks);
2630 /*
2631 * If this is a synchronous mount, make sure that the
2632 * link transaction goes to disk before returning to
2633 * the user.
2634 */
2637 }
2643 }
2649 /* Fall through to std_return with error = 0. */
2650 std_return:
2656 }
2659 abort_return:
2661 /* FALLTHROUGH */
2663 error_return:
2666 }
2669 int
2673 mode_t mode,
2676 {
2687 xfs_bmap_free_t free_list;
2688 xfs_fsblock_t first_block;
2692 xfs_prid_t prid;
2694 uint resblks;
2709 }
2711 /* Return through std_return after this point. */
2719 else
2722 /*
2723 * Make sure that we have allocated dquot(s) on disk.
2724 */
2739 XFS_TRANS_PERM_LOG_RES,
2740 XFS_MKDIR_LOG_COUNT);
2741 }
2745 }
2750 /*
2751 * Check for directory link count overflow.
2752 */
2756 }
2758 /*
2759 * Reserve disk quota and the inode.
2760 */
2768 /*
2769 * create the directory inode.
2770 */
2778 }
2781 /*
2782 * Now we add the directory inode to the transaction.
2783 * We waited until now since xfs_dir_ialloc might start
2784 * a new transaction. Had we joined the transaction
2785 * earlier, the locks might have gotten released. An error
2786 * from here on will result in the transaction cancel
2787 * unlocking dp so don't do it explicitly in the error path.
2788 */
2801 }
2804 /*
2805 * Bump the in memory version number of the parent directory
2806 * so that other processes accessing it will recognize that
2807 * the directory has changed.
2808 */
2827 /*
2828 * Attach the dquots to the new inode and modify the icount incore.
2829 */
2832 /*
2833 * If this is a synchronous mount, make sure that the
2834 * mkdir transaction goes to disk before returning to
2835 * the user.
2836 */
2839 }
2845 }
2852 }
2854 /* Fall through to std_return with error = 0 or errno from
2855 * xfs_trans_commit. */
2857 std_return:
2863 DM_RIGHT_NULL,
2866 }
2869 error2:
2870 error1:
2872 abort_return:
2874 error_return:
2883 }
2885 int
2889 {
2897 xfs_bmap_free_t free_list;
2898 xfs_fsblock_t first_block;
2903 uint resblks;
2913 }
2922 }
2924 /* Return through std_return after this point. */
2928 /*
2929 * We need to get a reference to cdp before we get our log
2930 * reservation. The reason for this is that we cannot call
2931 * xfs_iget for an inode for which we do not have a reference
2932 * once we've acquired a log reservation. This is because the
2933 * inode we are trying to get might be in xfs_inactive going
2934 * for a log reservation. Since we'll have to wait for the
2935 * inactive code to complete before returning from xfs_iget,
2936 * we need to make sure that we don't have log space reserved
2937 * when we call xfs_iget. Instead we get an unlocked reference
2938 * to the inode before getting our log reservation.
2939 */
2944 }
2948 /*
2949 * Get the dquots for the inodes.
2950 */
2958 }
2962 /*
2963 * We try to get the real space reservation first,
2964 * allowing for directory btree deletion(s) implying
2965 * possible bmap insert(s). If we can't get the space
2966 * reservation then we use 0 instead, and avoid the bmap
2967 * btree insert(s) in the directory code by, if the bmap
2968 * insert tries to happen, instead trimming the LAST
2969 * block from the directory.
2970 */
2978 }
2984 }
2987 /*
2988 * Now lock the child directory inode and the parent directory
2989 * inode in the proper order. This will take care of validating
2990 * that the directory entry for the child directory inode has
2991 * not changed while we were obtaining a log reservation.
2992 */
2998 }
3002 /*
3003 * Only increment the parent directory vnode count if
3004 * we didn't bump it in looking up cdp. The only time
3005 * we don't bump it is when we're looking up ".".
3006 */
3008 }
3017 }
3021 }
3030 /*
3031 * Bump the in memory generation count on the parent
3032 * directory so that other can know that it has changed.
3033 */
3036 /*
3037 * Drop the link from cdp's "..".
3038 */
3042 }
3044 /*
3045 * Drop the link from dp to cdp.
3046 */
3050 }
3052 /*
3053 * Drop the "." link from cdp to self.
3054 */
3058 }
3060 /* Determine these before committing transaction */
3063 /*
3064 * Take an extra ref on the child vnode so that it
3065 * does not go to xfs_inactive() from within the commit.
3066 */
3069 /*
3070 * If this is a synchronous mount, make sure that the
3071 * rmdir transaction goes to disk before returning to
3072 * the user.
3073 */
3076 }
3082 XFS_TRANS_ABORT));
3085 }
3091 }
3096 /* Fall through to std_return with error = 0 or the errno
3097 * from xfs_trans_commit. */
3098 std_return:
3105 }
3108 error1:
3111 /* FALLTHROUGH */
3113 error_return:
3116 }
3118 int
3123 mode_t mode,
3126 {
3133 xfs_bmap_free_t free_list;
3134 xfs_fsblock_t first_block;
3136 uint cancel_flags;
3138 xfs_fileoff_t first_fsb;
3139 xfs_filblks_t fs_blocks;
3142 xfs_daddr_t d;
3147 xfs_prid_t prid;
3149 uint resblks;
3165 /*
3166 * Check component lengths of the target path name.
3167 */
3176 /*
3177 * Skip any slashes.
3178 */
3180 total++;
3181 path++;
3182 }
3184 /*
3185 * Count up to the next slash or end of path.
3186 * Error out if the component is bigger than MAXNAMELEN.
3187 */
3192 }
3193 }
3194 }
3195 }
3203 }
3205 /* Return through std_return after this point. */
3210 else
3213 /*
3214 * Make sure that we have allocated dquot(s) on disk.
3215 */
3224 /*
3225 * The symlink will fit into the inode data fork?
3226 * There can't be any attributes so we get the whole variable part.
3227 */
3230 else
3239 }
3243 }
3248 /*
3249 * Check whether the directory allows new symlinks or not.
3250 */
3254 }
3256 /*
3257 * Reserve disk quota : blocks and inode.
3258 */
3263 /*
3264 * Check for ability to enter directory entry, if no space reserved.
3265 */
3269 /*
3270 * Initialize the bmap freelist prior to calling either
3271 * bmapi or the directory create code.
3272 */
3275 /*
3276 * Allocate an inode for the symlink.
3277 */
3284 }
3287 /*
3288 * An error after we've joined dp to the transaction will result in the
3289 * transaction cancel unlocking dp so don't do it explicitly in the
3290 * error path.
3291 */
3296 /*
3297 * Also attach the dquot(s) to it, if applicable.
3298 */
3303 /*
3304 * If the symlink will fit into the inode, write it inline.
3305 */
3311 /*
3312 * The inode was initially created in extent format.
3313 */
3330 }
3346 }
3353 }
3354 }
3356 /*
3357 * Create the directory entry for the symlink.
3358 */
3366 /*
3367 * Bump the in memory version number of the parent directory
3368 * so that other processes accessing it will recognize that
3369 * the directory has changed.
3370 */
3373 /*
3374 * If this is a synchronous mount, make sure that the
3375 * symlink transaction goes to disk before returning to
3376 * the user.
3377 */
3380 }
3382 /*
3383 * xfs_trans_commit normally decrements the vnode ref count
3384 * when it unlocks the inode. Since we want to return the
3385 * vnode to the caller, we bump the vnode ref count now.
3386 */
3392 }
3397 /* Fall through to std_return with error = 0 or errno from
3398 * xfs_trans_commit */
3399 std_return:
3406 }
3414 }
3417 error2:
3419 error1:
3422 error_return:
3431 }
3433 int
3436 bhv_vrwlock_t locktype)
3437 {
3450 }
3453 }
3456 void
3459 bhv_vrwlock_t locktype)
3460 {
3464 /*
3465 * In the write case, we may have added a new entry to
3466 * the reference cache. This might store a pointer to
3467 * an inode to be released in this inode. If it is there,
3468 * clear the pointer and release the inode after unlocking
3469 * this one.
3470 */
3476 }
3478 }
3481 int
3485 {
3493 /*
3494 * Bypass inodes which have already been cleaned by
3495 * the inode flush clustering code inside xfs_iflush
3496 */
3504 xfs_lsn_t sync_lsn;
3517 }
3521 }
3522 }
3524 /*
3525 * We make this non-blocking if the inode is contended,
3526 * return EAGAIN to indicate to the caller that they
3527 * did not succeed. This prevents the flush path from
3528 * blocking on inodes inside another operation right
3529 * now, they get caught later by xfs_sync.
3530 */
3541 }
3544 }
3548 else
3553 }
3556 }
3559 int
3562 u_int evmask,
3563 u_int16_t state)
3564 {
3580 }
3592 }
3594 int
3597 {
3604 /* bad inode, get out here ASAP */
3608 }
3614 /*
3615 * Make sure the atime in the XFS inode is correct before freeing the
3616 * Linux inode.
3617 */
3620 /*
3621 * If we have nothing to flush with this inode then complete the
3622 * teardown now, otherwise break the link between the xfs inode and the
3623 * linux inode and clean up the xfs inode later. This avoids flushing
3624 * the inode to disk during the delete operation itself.
3625 *
3626 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3627 * first to ensure that xfs_iunpin() will never see an xfs inode
3628 * that has a linux inode being reclaimed. Synchronisation is provided
3629 * by the i_flags_lock.
3630 */
3638 /* Protect sync and unpin from us */
3647 }
3649 }
3651 int
3656 {
3664 /* The hash lock here protects a thread in xfs_iget_core from
3665 * racing with us on linking the inode back with a vnode.
3666 * Once we have the XFS_IRECLAIM flag set it will not touch
3667 * us.
3668 */
3678 }
3680 }
3686 /*
3687 * If the inode is still dirty, then flush it out. If the inode
3688 * is not in the AIL, then it will be OK to flush it delwri as
3689 * long as xfs_iflush() does not keep any references to the inode.
3690 * We leave that decision up to xfs_iflush() since it has the
3691 * knowledge of whether it's OK to simply do a delwri flush of
3692 * the inode or whether we need to wait until the inode is
3693 * pulled from the AIL.
3694 * We get the flush lock regardless, though, just to make sure
3695 * we don't free it while it is being flushed.
3696 */
3701 }
3707 /*
3708 * If we hit an error, typically because of filesystem
3709 * shutdown, we don't need to let vn_reclaim to know
3710 * because we're gonna reclaim the inode anyway.
3711 */
3715 }
3717 }
3724 /*
3725 * We are not interested in doing an iflush if we're
3726 * in the process of shutting down the filesystem forcibly.
3727 * So, just reclaim the inode.
3728 */
3731 }
3733 reclaim:
3736 }
3738 int
3740 {
3756 }
3757 }
3760 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3764 }
3767 }
3771 }
3773 /*
3774 * xfs_alloc_file_space()
3775 * This routine allocates disk space for the given file.
3776 *
3777 * If alloc_type == 0, this request is for an ALLOCSP type
3778 * request which will change the file size. In this case, no
3779 * DMAPI event will be generated by the call. A TRUNCATE event
3780 * will be generated later by xfs_setattr.
3781 *
3782 * If alloc_type != 0, this request is for a RESVSP type
3783 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3784 * lower block boundary byte address is less than the file's
3785 * length.
3786 *
3787 * RETURNS:
3788 * 0 on success
3789 * errno on error
3790 *
3791 */
3792 STATIC int
3795 xfs_off_t offset,
3796 xfs_off_t len,
3799 {
3801 xfs_off_t count;
3802 xfs_filblks_t allocated_fsb;
3803 xfs_filblks_t allocatesize_fsb;
3805 xfs_fileoff_t startoffset_fsb;
3806 xfs_fsblock_t firstfsb;
3813 xfs_bmap_free_t free_list;
3839 /* Generate a DMAPI event if needed. */
3843 xfs_off_t end_dmi_offset;
3853 }
3855 /*
3856 * Allocate file space until done or until there is an error
3857 */
3858 retry:
3862 /*
3863 * Determine space reservations for data/realtime.
3864 */
3877 }
3889 }
3891 /*
3892 * Allocate and setup the transaction.
3893 */
3897 XFS_TRANS_PERM_LOG_RES,
3898 XFS_WRITE_LOG_COUNT);
3899 /*
3900 * Check for running out of space
3901 */
3903 /*
3904 * Free the transaction structure.
3905 */
3909 }
3919 /*
3920 * Issue the xfs_bmapi() call to allocate the blocks
3921 */
3929 }
3931 /*
3932 * Complete the transaction
3933 */
3937 }
3943 }
3950 }
3954 }
3955 dmapi_enospc_check:
3964 /* else fall through with error from XFS_SEND_DATA */
3965 }
3977 }
3979 /*
3980 * Zero file bytes between startoff and endoff inclusive.
3981 * The iolock is held exclusive and no blocks are buffered.
3982 */
3983 STATIC int
3986 xfs_off_t startoff,
3987 xfs_off_t endoff)
3988 {
3989 xfs_bmbt_irec_t imap;
3990 xfs_fileoff_t offset_fsb;
3991 xfs_off_t lastoffset;
3992 xfs_off_t offset;
4028 }
4040 }
4041 }
4044 }
4046 /*
4047 * xfs_free_file_space()
4048 * This routine frees disk space for the given file.
4049 *
4050 * This routine is only called by xfs_change_file_space
4051 * for an UNRESVSP type call.
4052 *
4053 * RETURNS:
4054 * 0 on success
4055 * errno on error
4056 *
4057 */
4058 STATIC int
4061 xfs_off_t offset,
4062 xfs_off_t len,
4064 {
4068 xfs_off_t end_dmi_offset;
4069 xfs_fileoff_t endoffset_fsb;
4071 xfs_fsblock_t firstfsb;
4072 xfs_bmap_free_t free_list;
4073 xfs_bmbt_irec_t imap;
4074 xfs_off_t ioffset;
4078 uint resblks;
4079 uint rounding;
4081 xfs_fileoff_t startoffset_fsb;
4110 }
4117 }
4127 }
4129 /*
4130 * Need to zero the stuff we're not freeing, on disk.
4131 * If its a realtime file & can't use unwritten extents then we
4132 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4133 * will take care of it for us.
4134 */
4143 xfs_daddr_t block;
4150 }
4159 mod++;
4162 }
4163 }
4165 /*
4166 * One contiguous piece to clear
4167 */
4170 /*
4171 * Some full blocks, possibly two pieces to clear
4172 */
4181 }
4183 /*
4184 * free file space until done or until there is an error
4185 */
4189 /*
4190 * allocate and setup the transaction. Allow this
4191 * transaction to dip into the reserve blocks to ensure
4192 * the freeing of the space succeeds at ENOSPC.
4193 */
4197 resblks,
4200 XFS_TRANS_PERM_LOG_RES,
4201 XFS_WRITE_LOG_COUNT);
4203 /*
4204 * check for running out of space
4205 */
4207 /*
4208 * Free the transaction structure.
4209 */
4213 }
4217 XFS_QMOPT_RES_REGBLKS);
4224 /*
4225 * issue the bunmapi() call to free the blocks
4226 */
4233 }
4235 /*
4236 * complete the transaction
4237 */
4241 }
4245 }
4247 out_unlock_iolock:
4252 error0:
4254 error1:
4257 XFS_ILOCK_EXCL);
4259 }
4261 /*
4262 * xfs_change_file_space()
4263 * This routine allocates or frees disk space for the given file.
4264 * The user specified parameters are checked for alignment and size
4265 * limitations.
4266 *
4267 * RETURNS:
4268 * 0 on success
4269 * errno on error
4270 *
4271 */
4272 int
4277 xfs_off_t offset,
4280 {
4284 xfs_fsize_t fsize;
4286 xfs_off_t startoffset;
4287 xfs_off_t llen;
4289 bhv_vattr_t va;
4307 }
4322 /*
4323 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4324 * file space.
4325 * These calls do NOT zero the data space allocated to the file,
4326 * nor do they change the file size.
4327 *
4328 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4329 * space.
4330 * These calls cause the new file data to be zeroed and the file
4331 * size to be changed.
4332 */
4348 attr_flags)))
4361 }
4377 }
4379 /*
4380 * update the inode timestamp, mode, and prealloc flag bits
4381 */
4386 /* ASSERT(0); */
4389 }
4399 /*
4400 * Note that we don't have to worry about mandatory
4401 * file locking being disabled here because we only
4402 * clear the S_ISGID bit if the Group execute bit is
4403 * on, but if it was on then mandatory locking wouldn't
4404 * have been enabled.
4405 */
4410 }
4424 }