| blob | 5eb392d062983555ddb819aa7df8290cbd467e30 |
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 */
812 }
815 }
817 }
820 }
822 /*
823 * Change file inode change time only if XFS_AT_CTIME set
824 * AND we have been called by a DMI function.
825 */
832 }
834 /*
835 * Send out timestamp changes that need to be set to the
836 * current time. Not done when called by a DMI function.
837 */
843 /*
844 * If this is a synchronous mount, make sure that the
845 * transaction goes to disk before returning to the user.
846 * This is slightly sub-optimal in that truncates require
847 * two sync transactions instead of one for wsync filesystems.
848 * One for the truncate and one for the timestamps since we
849 * don't want to change the timestamps unless we're sure the
850 * truncate worked. Truncates are less than 1% of the laddis
851 * mix so this probably isn't worth the trouble to optimize.
852 */
859 }
861 /*
862 * If the (regular) file's mandatory locking mode changed, then
863 * notify the vnode. We do this under the inode lock to prevent
864 * racing calls to vop_vnode_change.
865 */
870 /*
871 * Release any dquot(s) the inode had kept before chown.
872 */
880 }
887 }
890 abort_return:
892 /* FALLTHROUGH */
893 error_return:
898 }
901 }
903 }
906 /*
907 * xfs_access
908 * Null conversion from vnode mode bits to inode mode bits, as in efs.
909 */
910 int
915 {
924 }
927 /*
928 * The maximum pathlen is 1024 bytes. Since the minimum file system
929 * blocksize is 512 bytes, we can get a max of 2 extents back from
930 * bmapi.
931 */
932 #define SYMLINK_MAPS 2
934 STATIC int
938 {
943 xfs_daddr_t d;
965 }
972 }
977 out:
979 }
981 int
985 {
1009 }
1011 out:
1014 }
1016 /*
1017 * xfs_fsync
1018 *
1019 * This is called to sync the inode and its data out to disk.
1020 * We need to hold the I/O lock while flushing the data, and
1021 * the inode lock while flushing the inode. The inode lock CANNOT
1022 * be held while flushing the data, so acquire after we're done
1023 * with that.
1024 */
1025 int
1029 xfs_off_t start,
1030 xfs_off_t stop)
1031 {
1046 /*
1047 * We always need to make sure that the required inode state
1048 * is safe on disk. The vnode might be clean but because
1049 * of committed transactions that haven't hit the disk yet.
1050 * Likewise, there could be unflushed non-transactional
1051 * changes to the inode core that have to go to disk.
1052 *
1053 * The following code depends on one assumption: that
1054 * any transaction that changes an inode logs the core
1055 * because it has to change some field in the inode core
1056 * (typically nextents or nblocks). That assumption
1057 * implies that any transactions against an inode will
1058 * catch any non-transactional updates. If inode-altering
1059 * transactions exist that violate this assumption, the
1060 * code breaks. Right now, it figures that if the involved
1061 * update_* field is clear and the inode is unpinned, the
1062 * inode is clean. Either it's been flushed or it's been
1063 * committed and the commit has hit the disk unpinning the inode.
1064 * (Note that xfs_inode_item_format() called at commit clears
1065 * the update_* fields.)
1066 */
1069 /* If we are flushing data then we care about update_size
1070 * being set, otherwise we care about update_core
1071 */
1075 /*
1076 * Timestamps/size haven't changed since last inode
1077 * flush or inode transaction commit. That means
1078 * either nothing got written or a transaction
1079 * committed which caught the updates. If the
1080 * latter happened and the transaction hasn't
1081 * hit the disk yet, the inode will be still
1082 * be pinned. If it is, force the log.
1083 */
1089 XFS_LOG_FORCE |
1094 /*
1095 * If the inode is not pinned and nothing
1096 * has changed we don't need to flush the
1097 * cache.
1098 */
1100 }
1103 /*
1104 * Kick off a transaction to log the inode
1105 * core to get the updates. Make it
1106 * sync if FSYNC_WAIT is passed in (which
1107 * is done by everybody but specfs). The
1108 * sync transaction will also force the log.
1109 */
1117 }
1120 /*
1121 * Note - it's possible that we might have pushed
1122 * ourselves out of the way during trans_reserve
1123 * which would flush the inode. But there's no
1124 * guarantee that the inode buffer has actually
1125 * gone out yet (it's delwri). Plus the buffer
1126 * could be pinned anyway if it's part of an
1127 * inode in another recent transaction. So we
1128 * play it safe and fire off the transaction anyway.
1129 */
1138 }
1141 /*
1142 * If the log write didn't issue an ordered tag we need
1143 * to flush the disk cache for the data device now.
1144 */
1148 /*
1149 * If this inode is on the RT dev we need to flush that
1150 * cache as well.
1151 */
1154 }
1157 }
1159 /*
1160 * This is called by xfs_inactive to free any blocks beyond eof
1161 * when the link count isn't zero and by xfs_dm_punch_hole() when
1162 * punching a hole to EOF.
1163 */
1164 int
1169 {
1172 xfs_fileoff_t end_fsb;
1173 xfs_fileoff_t last_fsb;
1174 xfs_filblks_t map_len;
1176 xfs_bmbt_irec_t imap;
1179 /*
1180 * Figure out if there are any blocks beyond the end
1181 * of the file. If not, then there is nothing to do.
1182 */
1198 /*
1199 * Attach the dquots to the inode up front.
1200 */
1204 /*
1205 * There are blocks after the end of file.
1206 * Free them up now by truncating the file to
1207 * its current size.
1208 */
1211 /*
1212 * Do the xfs_itruncate_start() call before
1213 * reserving any log space because
1214 * itruncate_start will call into the buffer
1215 * cache and we can't
1216 * do that within a transaction.
1217 */
1227 }
1232 XFS_ITRUNCATE_LOG_COUNT);
1238 }
1242 XFS_IOLOCK_EXCL |
1243 XFS_ILOCK_EXCL);
1248 XFS_DATA_FORK,
1250 /*
1251 * If we get an error at this point we
1252 * simply don't bother truncating the file.
1253 */
1257 XFS_TRANS_ABORT));
1260 XFS_TRANS_RELEASE_LOG_RES);
1261 }
1264 }
1266 }
1268 /*
1269 * Free a symlink that has blocks associated with it.
1270 */
1271 STATIC int
1275 {
1280 xfs_fsblock_t first_block;
1281 xfs_bmap_free_t free_list;
1293 /*
1294 * We're freeing a symlink that has some
1295 * blocks allocated to it. Free the
1296 * blocks here. We know that we've got
1297 * either 1 or 2 extents and that we can
1298 * free them all in one bunmapi call.
1299 */
1307 }
1308 /*
1309 * Lock the inode, fix the size, and join it to the transaction.
1310 * Hold it so in the normal path, we still have it locked for
1311 * the second transaction. In the error paths we need it
1312 * held so the cancel won't rele it, see below.
1313 */
1320 /*
1321 * Find the block(s) so we can inval and unmap them.
1322 */
1330 /*
1331 * Invalidate the block(s).
1332 */
1338 }
1339 /*
1340 * Unmap the dead block(s) to the free_list.
1341 */
1346 /*
1347 * Commit the first transaction. This logs the EFI and the inode.
1348 */
1351 /*
1352 * The transaction must have been committed, since there were
1353 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1354 * The new tp has the extent freeing and EFDs.
1355 */
1357 /*
1358 * The first xact was committed, so add the inode to the new one.
1359 * Mark it dirty so it will be logged and moved forward in the log as
1360 * part of every commit.
1361 */
1365 /*
1366 * Get a new, empty transaction to return to our caller.
1367 */
1369 /*
1370 * Commit the transaction containing extent freeing and EFDs.
1371 * If we get an error on the commit here or on the reserve below,
1372 * we need to unlock the inode since the new transaction doesn't
1373 * have the inode attached.
1374 */
1380 }
1381 /*
1382 * Remove the memory for extent descriptions (just bookkeeping).
1383 */
1387 /*
1388 * Put an itruncate log reservation in the new transaction
1389 * for our caller.
1390 */
1395 }
1396 /*
1397 * Return with the inode locked but not joined to the transaction.
1398 */
1402 error1:
1404 error0:
1405 /*
1406 * Have to come here with the inode locked and either
1407 * (held and in the transaction) or (not in the transaction).
1408 * If the inode isn't held then cancel would iput it, but
1409 * that's wrong since this is inactive and the vnode ref
1410 * count is 0 already.
1411 * Cancel won't do anything to the inode if held, but it still
1412 * needs to be locked until the cancel is done, if it was
1413 * joined to the transaction.
1414 */
1420 }
1422 STATIC int
1426 {
1430 /*
1431 * We're freeing a symlink which fit into
1432 * the inode. Just free the memory used
1433 * to hold the old symlink.
1434 */
1438 XFS_ITRUNCATE_LOG_COUNT);
1444 }
1447 /*
1448 * Zero length symlinks _can_ exist.
1449 */
1453 XFS_DATA_FORK);
1455 }
1457 }
1459 STATIC int
1463 {
1480 }
1486 XFS_INACTIVE_LOG_COUNT);
1493 }
1504 }
1506 int
1509 {
1517 /* If this is a read-only mount, don't do this (would generate I/O) */
1522 /*
1523 * If we are using filestreams, and we have an unlinked
1524 * file that we are processing the last close on, then nothing
1525 * will be able to reopen and write to this file. Purge this
1526 * inode from the filestreams cache so that it doesn't delay
1527 * teardown of the inode.
1528 */
1532 /*
1533 * If we previously truncated this file and removed old data
1534 * in the process, we want to initiate "early" writeout on
1535 * the last close. This is an attempt to combat the notorious
1536 * NULL files problem which is particularly noticable from a
1537 * truncate down, buffered (re-)write (delalloc), followed by
1538 * a crash. What we are effectively doing here is
1539 * significantly reducing the time window where we'd otherwise
1540 * be exposed to that problem.
1541 */
1544 }
1546 #ifdef HAVE_REFCACHE
1547 /* If we are in the NFS reference cache then don't do this now */
1550 #endif
1562 /* Update linux inode block count after free above */
1565 }
1566 }
1569 }
1571 /*
1572 * xfs_inactive
1573 *
1574 * This is called when the vnode reference count for the vnode
1575 * goes to zero. If the file has been unlinked, then it must
1576 * now be truncated. Also, we clear all of the read-ahead state
1577 * kept for the inode here since the file is now closed.
1578 */
1579 int
1582 {
1584 xfs_bmap_free_t free_list;
1585 xfs_fsblock_t first_block;
1594 /*
1595 * If the inode is already free, then there can be nothing
1596 * to clean up here.
1597 */
1602 }
1604 /*
1605 * Only do a truncate if it's a regular file with
1606 * some actual space in it. It's OK to look at the
1607 * inode's fields without the lock because we're the
1608 * only one with a reference to the inode.
1609 */
1619 }
1623 /* If this is a read-only mount, don't do this (would generate I/O) */
1638 /* Update linux inode block count after free above */
1641 }
1643 }
1652 /*
1653 * Do the xfs_itruncate_start() call before
1654 * reserving any log space because itruncate_start
1655 * will call into the buffer cache and we can't
1656 * do that within a transaction.
1657 */
1665 }
1670 XFS_ITRUNCATE_LOG_COUNT);
1672 /* Don't call itruncate_cleanup */
1677 }
1683 /*
1684 * normally, we have to run xfs_itruncate_finish sync.
1685 * But if filesystem is wsync and we're in the inactive
1686 * path, then we know that nlink == 0, and that the
1687 * xaction that made nlink == 0 is permanently committed
1688 * since xfs_remove runs as a synchronous transaction.
1689 */
1698 }
1701 /*
1702 * If we get an error while cleaning up a
1703 * symlink we bail out.
1704 */
1712 }
1720 XFS_INACTIVE_LOG_COUNT);
1725 }
1730 }
1732 /*
1733 * If there are attributes associated with the file
1734 * then blow them away now. The code calls a routine
1735 * that recursively deconstructs the attribute fork.
1736 * We need to just commit the current transaction
1737 * because we can't use it for xfs_attr_inactive().
1738 */
1741 /*
1742 * If we got an error, the transaction is already
1743 * cancelled, and the inode is unlocked. Just get out.
1744 */
1749 }
1751 /*
1752 * Free the inode.
1753 */
1757 /*
1758 * If we fail to free the inode, shut down. The cancel
1759 * might do that, we need to make sure. Otherwise the
1760 * inode might be lost for a long time or forever.
1761 */
1767 }
1770 /*
1771 * Credit the quota account(s). The inode is gone.
1772 */
1775 /*
1776 * Just ignore errors at this point. There is
1777 * nothing we can do except to try to keep going.
1778 */
1781 }
1782 /*
1783 * Release the dquots held by inode, if any.
1784 */
1789 out:
1791 }
1794 int
1799 {
1801 xfs_ino_t e_inum;
1803 uint lock_mode;
1815 }
1818 }
1820 int
1827 {
1834 xfs_dev_t rdev;
1836 xfs_bmap_free_t free_list;
1837 xfs_fsblock_t first_block;
1840 uint cancel_flags;
1842 xfs_prid_t prid;
1844 uint resblks;
1863 }
1868 /* Return through std_return after this point. */
1875 else
1878 /*
1879 * Make sure that we have allocated dquot(s) on disk.
1880 */
1892 /*
1893 * Initially assume that the file does not exist and
1894 * reserve the resources for that case. If that is not
1895 * the case we'll drop the one we have and get a more
1896 * appropriate transaction later.
1897 */
1904 }
1908 }
1917 /*
1918 * Reserve disk quota and the inode.
1919 */
1934 }
1937 /*
1938 * At this point, we've gotten a newly allocated inode.
1939 * It is locked (and joined to the transaction).
1940 */
1944 /*
1945 * Now we join the directory inode to the transaction. We do not do it
1946 * earlier because xfs_dir_ialloc might commit the previous transaction
1947 * (and release all the locks). An error from here on will result in
1948 * the transaction cancel unlocking dp so don't do it explicitly in the
1949 * error path.
1950 */
1961 }
1965 /*
1966 * If this is a synchronous mount, make sure that the
1967 * create transaction goes to disk before returning to
1968 * the user.
1969 */
1972 }
1976 /*
1977 * Attach the dquot(s) to the inodes and modify them incore.
1978 * These ids of the inode couldn't have changed since the new
1979 * inode has been locked ever since it was created.
1980 */
1983 /*
1984 * xfs_trans_commit normally decrements the vnode ref count
1985 * when it unlocks the inode. Since we want to return the
1986 * vnode to the caller, we bump the vnode ref count now.
1987 */
1995 }
2002 }
2009 /* Fallthrough to std_return with error = 0 */
2011 std_return:
2019 }
2022 abort_return:
2024 /* FALLTHROUGH */
2026 error_return:
2038 abort_rele:
2039 /*
2040 * Wait until after the current transaction is aborted to
2041 * release the inode. This prevents recursive transactions
2042 * and deadlocks from xfs_inactive.
2043 */
2052 }
2054 #ifdef DEBUG
2055 /*
2056 * Some counters to see if (and how often) we are hitting some deadlock
2057 * prevention code paths.
2058 */
2063 #endif
2065 /*
2066 * The following routine will lock the inodes associated with the
2067 * directory and the named entry in the directory. The locks are
2068 * acquired in increasing inode number.
2069 *
2070 * If the entry is "..", then only the directory is locked. The
2071 * vnode ref count will still include that from the .. entry in
2072 * this case.
2073 *
2074 * There is a deadlock we need to worry about. If the locked directory is
2075 * in the AIL, it might be blocking up the log. The next inode we lock
2076 * could be already locked by another thread waiting for log space (e.g
2077 * a permanent log reservation with a long running transaction (see
2078 * xfs_itruncate_finish)). To solve this, we must check if the directory
2079 * is in the ail and use lock_nowait. If we can't lock, we need to
2080 * drop the inode lock on the directory and try again. xfs_iunlock will
2081 * potentially push the tail if we were holding up the log.
2082 */
2083 STATIC int
2087 {
2089 xfs_ino_t e_inum;
2093 #ifdef DEBUG
2094 xfs_rm_locks++;
2095 #endif
2098 again:
2105 /*
2106 * We want to lock in increasing inum. Since we've already
2107 * acquired the lock on the directory, we may need to release
2108 * if if the inum of the entry turns out to be less.
2109 */
2111 /*
2112 * We are already in the right order, so just
2113 * lock on the inode of the entry.
2114 * We need to use nowait if dp is in the AIL.
2115 */
2120 attempts++;
2121 #ifdef DEBUG
2122 xfs_rm_attempts++;
2123 #endif
2125 /*
2126 * Unlock dp and try again.
2127 * xfs_iunlock will try to push the tail
2128 * if the inode is in the AIL.
2129 */
2135 #ifdef DEBUG
2136 xfs_rm_lock_delays++;
2137 #endif
2138 }
2140 }
2143 }
2150 }
2151 /* else e_inum == dp->i_ino */
2152 /* This can happen if we're asked to lock /x/..
2153 * the entry is "..", which is also the parent directory.
2154 */
2157 }
2159 #ifdef DEBUG
2165 #endif
2167 /*
2168 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2169 * a different value
2170 */
2173 {
2180 }
2182 /*
2183 * The following routine will lock n inodes in exclusive mode.
2184 * We assume the caller calls us with the inodes in i_ino order.
2185 *
2186 * We need to detect deadlock where an inode that we lock
2187 * is in the AIL and we start waiting for another inode that is locked
2188 * by a thread in a long running transaction (such as truncate). This can
2189 * result in deadlock since the long running trans might need to wait
2190 * for the inode we just locked in order to push the tail and free space
2191 * in the log.
2192 */
2193 void
2198 uint lock_mode)
2199 {
2211 }
2213 again:
2220 /*
2221 * If try_lock is not set yet, make sure all locked inodes
2222 * are not in the AIL.
2223 * If any are, set try_lock to be used later.
2224 */
2230 try_lock++;
2231 }
2232 }
2233 }
2235 /*
2236 * If any of the previous locks we have locked is in the AIL,
2237 * we must TRY to get the second and subsequent locks. If
2238 * we can't get any, we must release all we have
2239 * and try again.
2240 */
2243 /* try_lock must be 0 if i is 0. */
2244 /*
2245 * try_lock means we have an inode locked
2246 * that is in the AIL.
2247 */
2250 attempts++;
2252 /*
2253 * Unlock all previous guys and try again.
2254 * xfs_iunlock will try to push the tail
2255 * if the inode is in the AIL.
2256 */
2260 /*
2261 * Check to see if we've already
2262 * unlocked this one.
2263 * Not the first one going back,
2264 * and the inode ptr is the same.
2265 */
2271 }
2275 #ifdef DEBUG
2276 xfs_lock_delays++;
2277 #endif
2278 }
2282 }
2285 }
2286 }
2288 #ifdef DEBUG
2294 xfs_locked_n++;
2295 }
2296 #endif
2297 }
2299 #ifdef DEBUG
2300 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2303 #define REMOVE_DEBUG_TRACE(x)
2306 int
2310 {
2317 xfs_bmap_free_t free_list;
2318 xfs_fsblock_t first_block;
2323 uint resblks;
2336 }
2344 }
2346 /* From this point on, return through std_return */
2349 /*
2350 * We need to get a reference to ip before we get our log
2351 * reservation. The reason for this is that we cannot call
2352 * xfs_iget for an inode for which we do not have a reference
2353 * once we've acquired a log reservation. This is because the
2354 * inode we are trying to get might be in xfs_inactive going
2355 * for a log reservation. Since we'll have to wait for the
2356 * inactive code to complete before returning from xfs_iget,
2357 * we need to make sure that we don't have log space reserved
2358 * when we call xfs_iget. Instead we get an unlocked reference
2359 * to the inode before getting our log reservation.
2360 */
2365 }
2380 }
2384 /*
2385 * We try to get the real space reservation first,
2386 * allowing for directory btree deletion(s) implying
2387 * possible bmap insert(s). If we can't get the space
2388 * reservation then we use 0 instead, and avoid the bmap
2389 * btree insert(s) in the directory code by, if the bmap
2390 * insert tries to happen, instead trimming the LAST
2391 * block from the directory.
2392 */
2400 }
2407 }
2415 }
2417 /*
2418 * At this point, we've gotten both the directory and the entry
2419 * inodes locked.
2420 */
2423 /*
2424 * Increment vnode ref count only in this case since
2425 * there's an extra vnode reference in the case where
2426 * dp == ip.
2427 */
2430 }
2432 /*
2433 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2434 */
2442 }
2452 }
2454 /* Determine if this is the last link while
2455 * we are in the transaction.
2456 */
2459 /*
2460 * Take an extra ref on the inode so that it doesn't
2461 * go to xfs_inactive() from within the commit.
2462 */
2465 /*
2466 * If this is a synchronous mount, make sure that the
2467 * remove transaction goes to disk before returning to
2468 * the user.
2469 */
2472 }
2478 }
2484 }
2486 /*
2487 * Before we drop our extra reference to the inode, purge it
2488 * from the refcache if it is there. By waiting until afterwards
2489 * to do the IRELE, we ensure that we won't go inactive in the
2490 * xfs_refcache_purge_ip routine (although that would be OK).
2491 */
2494 /*
2495 * If we are using filestreams, kill the stream association.
2496 * If the file is still open it may get a new one but that
2497 * will get killed on last close in xfs_close() so we don't
2498 * have to worry about that.
2499 */
2507 /* Fall through to std_return with error = 0 */
2508 std_return:
2514 }
2517 error1:
2523 error_rele:
2524 /*
2525 * In this case make sure to not release the inode until after
2526 * the current transaction is aborted. Releasing it beforehand
2527 * can cause us to go to xfs_inactive and start a recursive
2528 * transaction which can easily deadlock with the current one.
2529 */
2534 /*
2535 * Before we drop our extra reference to the inode, purge it
2536 * from the refcache if it is there. By waiting until afterwards
2537 * to do the IRELE, we ensure that we won't go inactive in the
2538 * xfs_refcache_purge_ip routine (although that would be OK).
2539 */
2545 }
2547 int
2552 {
2559 xfs_bmap_free_t free_list;
2560 xfs_fsblock_t first_block;
2583 }
2585 /* Return through std_return after this point. */
2602 }
2606 }
2614 }
2618 /*
2619 * Increment vnode ref counts since xfs_trans_commit &
2620 * xfs_trans_cancel will both unlock the inodes and
2621 * decrement the associated ref counts.
2622 */
2628 /*
2629 * If the source has too many links, we can't make any more to it.
2630 */
2634 }
2636 /*
2637 * If we are using project inheritance, we only allow hard link
2638 * creation in our tree when the project IDs are the same; else
2639 * the tree quota mechanism could be circumvented.
2640 */
2645 }
2655 resblks);
2666 /*
2667 * If this is a synchronous mount, make sure that the
2668 * link transaction goes to disk before returning to
2669 * the user.
2670 */
2673 }
2679 }
2685 /* Fall through to std_return with error = 0. */
2686 std_return:
2692 }
2695 abort_return:
2697 /* FALLTHROUGH */
2699 error_return:
2702 }
2705 int
2712 {
2723 xfs_bmap_free_t free_list;
2724 xfs_fsblock_t first_block;
2728 xfs_prid_t prid;
2730 uint resblks;
2747 }
2749 /* Return through std_return after this point. */
2759 else
2762 /*
2763 * Make sure that we have allocated dquot(s) on disk.
2764 */
2779 XFS_TRANS_PERM_LOG_RES,
2780 XFS_MKDIR_LOG_COUNT);
2781 }
2785 }
2790 /*
2791 * Check for directory link count overflow.
2792 */
2796 }
2798 /*
2799 * Reserve disk quota and the inode.
2800 */
2808 /*
2809 * create the directory inode.
2810 */
2818 }
2821 /*
2822 * Now we add the directory inode to the transaction.
2823 * We waited until now since xfs_dir_ialloc might start
2824 * a new transaction. Had we joined the transaction
2825 * earlier, the locks might have gotten released. An error
2826 * from here on will result in the transaction cancel
2827 * unlocking dp so don't do it explicitly in the error path.
2828 */
2841 }
2844 /*
2845 * Bump the in memory version number of the parent directory
2846 * so that other processes accessing it will recognize that
2847 * the directory has changed.
2848 */
2867 /*
2868 * Attach the dquots to the new inode and modify the icount incore.
2869 */
2872 /*
2873 * If this is a synchronous mount, make sure that the
2874 * mkdir transaction goes to disk before returning to
2875 * the user.
2876 */
2879 }
2885 }
2892 }
2894 /* Fall through to std_return with error = 0 or errno from
2895 * xfs_trans_commit. */
2897 std_return:
2903 DM_RIGHT_NULL,
2906 }
2909 error2:
2910 error1:
2912 abort_return:
2914 error_return:
2923 }
2925 int
2929 {
2937 xfs_bmap_free_t free_list;
2938 xfs_fsblock_t first_block;
2943 uint resblks;
2953 }
2962 }
2964 /* Return through std_return after this point. */
2968 /*
2969 * We need to get a reference to cdp before we get our log
2970 * reservation. The reason for this is that we cannot call
2971 * xfs_iget for an inode for which we do not have a reference
2972 * once we've acquired a log reservation. This is because the
2973 * inode we are trying to get might be in xfs_inactive going
2974 * for a log reservation. Since we'll have to wait for the
2975 * inactive code to complete before returning from xfs_iget,
2976 * we need to make sure that we don't have log space reserved
2977 * when we call xfs_iget. Instead we get an unlocked reference
2978 * to the inode before getting our log reservation.
2979 */
2984 }
2988 /*
2989 * Get the dquots for the inodes.
2990 */
2998 }
3002 /*
3003 * We try to get the real space reservation first,
3004 * allowing for directory btree deletion(s) implying
3005 * possible bmap insert(s). If we can't get the space
3006 * reservation then we use 0 instead, and avoid the bmap
3007 * btree insert(s) in the directory code by, if the bmap
3008 * insert tries to happen, instead trimming the LAST
3009 * block from the directory.
3010 */
3018 }
3024 }
3027 /*
3028 * Now lock the child directory inode and the parent directory
3029 * inode in the proper order. This will take care of validating
3030 * that the directory entry for the child directory inode has
3031 * not changed while we were obtaining a log reservation.
3032 */
3038 }
3042 /*
3043 * Only increment the parent directory vnode count if
3044 * we didn't bump it in looking up cdp. The only time
3045 * we don't bump it is when we're looking up ".".
3046 */
3048 }
3057 }
3061 }
3070 /*
3071 * Bump the in memory generation count on the parent
3072 * directory so that other can know that it has changed.
3073 */
3076 /*
3077 * Drop the link from cdp's "..".
3078 */
3082 }
3084 /*
3085 * Drop the link from dp to cdp.
3086 */
3090 }
3092 /*
3093 * Drop the "." link from cdp to self.
3094 */
3098 }
3100 /* Determine these before committing transaction */
3103 /*
3104 * Take an extra ref on the child vnode so that it
3105 * does not go to xfs_inactive() from within the commit.
3106 */
3109 /*
3110 * If this is a synchronous mount, make sure that the
3111 * rmdir transaction goes to disk before returning to
3112 * the user.
3113 */
3116 }
3122 XFS_TRANS_ABORT));
3125 }
3131 }
3136 /* Fall through to std_return with error = 0 or the errno
3137 * from xfs_trans_commit. */
3138 std_return:
3145 }
3148 error1:
3151 /* FALLTHROUGH */
3153 error_return:
3156 }
3158 int
3166 {
3173 xfs_bmap_free_t free_list;
3174 xfs_fsblock_t first_block;
3176 uint cancel_flags;
3178 xfs_fileoff_t first_fsb;
3179 xfs_filblks_t fs_blocks;
3182 xfs_daddr_t d;
3187 xfs_prid_t prid;
3189 uint resblks;
3206 /*
3207 * Check component lengths of the target path name.
3208 */
3217 /*
3218 * Skip any slashes.
3219 */
3221 total++;
3222 path++;
3223 }
3225 /*
3226 * Count up to the next slash or end of path.
3227 * Error out if the component is bigger than MAXNAMELEN.
3228 */
3233 }
3234 }
3235 }
3236 }
3244 }
3246 /* Return through std_return after this point. */
3253 else
3256 /*
3257 * Make sure that we have allocated dquot(s) on disk.
3258 */
3267 /*
3268 * The symlink will fit into the inode data fork?
3269 * There can't be any attributes so we get the whole variable part.
3270 */
3273 else
3282 }
3286 }
3291 /*
3292 * Check whether the directory allows new symlinks or not.
3293 */
3297 }
3299 /*
3300 * Reserve disk quota : blocks and inode.
3301 */
3306 /*
3307 * Check for ability to enter directory entry, if no space reserved.
3308 */
3312 /*
3313 * Initialize the bmap freelist prior to calling either
3314 * bmapi or the directory create code.
3315 */
3318 /*
3319 * Allocate an inode for the symlink.
3320 */
3327 }
3330 /*
3331 * An error after we've joined dp to the transaction will result in the
3332 * transaction cancel unlocking dp so don't do it explicitly in the
3333 * error path.
3334 */
3339 /*
3340 * Also attach the dquot(s) to it, if applicable.
3341 */
3346 /*
3347 * If the symlink will fit into the inode, write it inline.
3348 */
3354 /*
3355 * The inode was initially created in extent format.
3356 */
3373 }
3389 }
3396 }
3397 }
3399 /*
3400 * Create the directory entry for the symlink.
3401 */
3409 /*
3410 * Bump the in memory version number of the parent directory
3411 * so that other processes accessing it will recognize that
3412 * the directory has changed.
3413 */
3416 /*
3417 * If this is a synchronous mount, make sure that the
3418 * symlink transaction goes to disk before returning to
3419 * the user.
3420 */
3423 }
3425 /*
3426 * xfs_trans_commit normally decrements the vnode ref count
3427 * when it unlocks the inode. Since we want to return the
3428 * vnode to the caller, we bump the vnode ref count now.
3429 */
3435 }
3440 /* Fall through to std_return with error = 0 or errno from
3441 * xfs_trans_commit */
3442 std_return:
3449 }
3457 }
3460 error2:
3462 error1:
3465 error_return:
3474 }
3477 /*
3478 * xfs_fid2
3479 *
3480 * A fid routine that takes a pointer to a previously allocated
3481 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3482 */
3483 int
3487 {
3495 /*
3496 * use memcpy because the inode is a long long and there's no
3497 * assurance that xfid->fid_ino is properly aligned.
3498 */
3503 }
3506 int
3509 bhv_vrwlock_t locktype)
3510 {
3523 }
3526 }
3529 void
3532 bhv_vrwlock_t locktype)
3533 {
3537 /*
3538 * In the write case, we may have added a new entry to
3539 * the reference cache. This might store a pointer to
3540 * an inode to be released in this inode. If it is there,
3541 * clear the pointer and release the inode after unlocking
3542 * this one.
3543 */
3549 }
3551 }
3554 int
3558 {
3566 /*
3567 * Bypass inodes which have already been cleaned by
3568 * the inode flush clustering code inside xfs_iflush
3569 */
3577 xfs_lsn_t sync_lsn;
3590 }
3594 }
3595 }
3597 /*
3598 * We make this non-blocking if the inode is contended,
3599 * return EAGAIN to indicate to the caller that they
3600 * did not succeed. This prevents the flush path from
3601 * blocking on inodes inside another operation right
3602 * now, they get caught later by xfs_sync.
3603 */
3614 }
3617 }
3621 else
3626 }
3629 }
3632 int
3635 u_int evmask,
3636 u_int16_t state)
3637 {
3653 }
3665 }
3667 int
3670 {
3677 /* bad inode, get out here ASAP */
3681 }
3687 /*
3688 * Make sure the atime in the XFS inode is correct before freeing the
3689 * Linux inode.
3690 */
3693 /*
3694 * If we have nothing to flush with this inode then complete the
3695 * teardown now, otherwise break the link between the xfs inode and the
3696 * linux inode and clean up the xfs inode later. This avoids flushing
3697 * the inode to disk during the delete operation itself.
3698 *
3699 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3700 * first to ensure that xfs_iunpin() will never see an xfs inode
3701 * that has a linux inode being reclaimed. Synchronisation is provided
3702 * by the i_flags_lock.
3703 */
3711 /* Protect sync and unpin from us */
3720 }
3722 }
3724 int
3729 {
3737 /* The hash lock here protects a thread in xfs_iget_core from
3738 * racing with us on linking the inode back with a vnode.
3739 * Once we have the XFS_IRECLAIM flag set it will not touch
3740 * us.
3741 */
3751 }
3753 }
3759 /*
3760 * If the inode is still dirty, then flush it out. If the inode
3761 * is not in the AIL, then it will be OK to flush it delwri as
3762 * long as xfs_iflush() does not keep any references to the inode.
3763 * We leave that decision up to xfs_iflush() since it has the
3764 * knowledge of whether it's OK to simply do a delwri flush of
3765 * the inode or whether we need to wait until the inode is
3766 * pulled from the AIL.
3767 * We get the flush lock regardless, though, just to make sure
3768 * we don't free it while it is being flushed.
3769 */
3774 }
3780 /*
3781 * If we hit an error, typically because of filesystem
3782 * shutdown, we don't need to let vn_reclaim to know
3783 * because we're gonna reclaim the inode anyway.
3784 */
3788 }
3790 }
3797 /*
3798 * We are not interested in doing an iflush if we're
3799 * in the process of shutting down the filesystem forcibly.
3800 * So, just reclaim the inode.
3801 */
3804 }
3806 reclaim:
3809 }
3811 int
3813 {
3829 }
3830 }
3833 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3837 }
3840 }
3844 }
3846 /*
3847 * xfs_alloc_file_space()
3848 * This routine allocates disk space for the given file.
3849 *
3850 * If alloc_type == 0, this request is for an ALLOCSP type
3851 * request which will change the file size. In this case, no
3852 * DMAPI event will be generated by the call. A TRUNCATE event
3853 * will be generated later by xfs_setattr.
3854 *
3855 * If alloc_type != 0, this request is for a RESVSP type
3856 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3857 * lower block boundary byte address is less than the file's
3858 * length.
3859 *
3860 * RETURNS:
3861 * 0 on success
3862 * errno on error
3863 *
3864 */
3865 STATIC int
3868 xfs_off_t offset,
3869 xfs_off_t len,
3872 {
3874 xfs_off_t count;
3875 xfs_filblks_t allocated_fsb;
3876 xfs_filblks_t allocatesize_fsb;
3878 xfs_fileoff_t startoffset_fsb;
3879 xfs_fsblock_t firstfsb;
3886 xfs_bmap_free_t free_list;
3912 /* Generate a DMAPI event if needed. */
3916 xfs_off_t end_dmi_offset;
3926 }
3928 /*
3929 * Allocate file space until done or until there is an error
3930 */
3931 retry:
3935 /*
3936 * Determine space reservations for data/realtime.
3937 */
3950 }
3962 }
3964 /*
3965 * Allocate and setup the transaction.
3966 */
3970 XFS_TRANS_PERM_LOG_RES,
3971 XFS_WRITE_LOG_COUNT);
3972 /*
3973 * Check for running out of space
3974 */
3976 /*
3977 * Free the transaction structure.
3978 */
3982 }
3992 /*
3993 * Issue the xfs_bmapi() call to allocate the blocks
3994 */
4002 }
4004 /*
4005 * Complete the transaction
4006 */
4010 }
4016 }
4023 }
4027 }
4028 dmapi_enospc_check:
4037 /* else fall through with error from XFS_SEND_DATA */
4038 }
4050 }
4052 /*
4053 * Zero file bytes between startoff and endoff inclusive.
4054 * The iolock is held exclusive and no blocks are buffered.
4055 */
4056 STATIC int
4059 xfs_off_t startoff,
4060 xfs_off_t endoff)
4061 {
4062 xfs_bmbt_irec_t imap;
4063 xfs_fileoff_t offset_fsb;
4064 xfs_off_t lastoffset;
4065 xfs_off_t offset;
4101 }
4113 }
4114 }
4117 }
4119 /*
4120 * xfs_free_file_space()
4121 * This routine frees disk space for the given file.
4122 *
4123 * This routine is only called by xfs_change_file_space
4124 * for an UNRESVSP type call.
4125 *
4126 * RETURNS:
4127 * 0 on success
4128 * errno on error
4129 *
4130 */
4131 STATIC int
4134 xfs_off_t offset,
4135 xfs_off_t len,
4137 {
4141 xfs_off_t end_dmi_offset;
4142 xfs_fileoff_t endoffset_fsb;
4144 xfs_fsblock_t firstfsb;
4145 xfs_bmap_free_t free_list;
4146 xfs_bmbt_irec_t imap;
4147 xfs_off_t ioffset;
4151 uint resblks;
4152 uint rounding;
4154 xfs_fileoff_t startoffset_fsb;
4183 }
4190 }
4203 }
4205 /*
4206 * Need to zero the stuff we're not freeing, on disk.
4207 * If its a realtime file & can't use unwritten extents then we
4208 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4209 * will take care of it for us.
4210 */
4219 xfs_daddr_t block;
4226 }
4235 mod++;
4238 }
4239 }
4241 /*
4242 * One contiguous piece to clear
4243 */
4246 /*
4247 * Some full blocks, possibly two pieces to clear
4248 */
4257 }
4259 /*
4260 * free file space until done or until there is an error
4261 */
4265 /*
4266 * allocate and setup the transaction. Allow this
4267 * transaction to dip into the reserve blocks to ensure
4268 * the freeing of the space succeeds at ENOSPC.
4269 */
4273 resblks,
4276 XFS_TRANS_PERM_LOG_RES,
4277 XFS_WRITE_LOG_COUNT);
4279 /*
4280 * check for running out of space
4281 */
4283 /*
4284 * Free the transaction structure.
4285 */
4289 }
4293 XFS_QMOPT_RES_REGBLKS);
4300 /*
4301 * issue the bunmapi() call to free the blocks
4302 */
4309 }
4311 /*
4312 * complete the transaction
4313 */
4317 }
4321 }
4323 out_unlock_iolock:
4328 error0:
4330 error1:
4333 XFS_ILOCK_EXCL);
4335 }
4337 /*
4338 * xfs_change_file_space()
4339 * This routine allocates or frees disk space for the given file.
4340 * The user specified parameters are checked for alignment and size
4341 * limitations.
4342 *
4343 * RETURNS:
4344 * 0 on success
4345 * errno on error
4346 *
4347 */
4348 int
4353 xfs_off_t offset,
4356 {
4360 xfs_fsize_t fsize;
4362 xfs_off_t startoffset;
4363 xfs_off_t llen;
4365 bhv_vattr_t va;
4369 /*
4370 * must be a regular file and have write permission
4371 */
4380 }
4395 }
4410 /*
4411 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4412 * file space.
4413 * These calls do NOT zero the data space allocated to the file,
4414 * nor do they change the file size.
4415 *
4416 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4417 * space.
4418 * These calls cause the new file data to be zeroed and the file
4419 * size to be changed.
4420 */
4436 attr_flags)))
4449 }
4465 }
4467 /*
4468 * update the inode timestamp, mode, and prealloc flag bits
4469 */
4474 /* ASSERT(0); */
4477 }
4487 /*
4488 * Note that we don't have to worry about mandatory
4489 * file locking being disabled here because we only
4490 * clear the S_ISGID bit if the Group execute bit is
4491 * on, but if it was on then mandatory locking wouldn't
4492 * have been enabled.
4493 */
4498 }
4512 }