| blob | 98653793b34a02139766eb0705f6c8fd0531d0ce |
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) */
1524 /*
1525 * If we are using filestreams, and we have an unlinked
1526 * file that we are processing the last close on, then nothing
1527 * will be able to reopen and write to this file. Purge this
1528 * inode from the filestreams cache so that it doesn't delay
1529 * teardown of the inode.
1530 */
1534 /*
1535 * If we previously truncated this file and removed old data
1536 * in the process, we want to initiate "early" writeout on
1537 * the last close. This is an attempt to combat the notorious
1538 * NULL files problem which is particularly noticable from a
1539 * truncate down, buffered (re-)write (delalloc), followed by
1540 * a crash. What we are effectively doing here is
1541 * significantly reducing the time window where we'd otherwise
1542 * be exposed to that problem.
1543 */
1547 }
1549 #ifdef HAVE_REFCACHE
1550 /* If we are in the NFS reference cache then don't do this now */
1553 #endif
1565 /* Update linux inode block count after free above */
1568 }
1569 }
1572 }
1574 /*
1575 * xfs_inactive
1576 *
1577 * This is called when the vnode reference count for the vnode
1578 * goes to zero. If the file has been unlinked, then it must
1579 * now be truncated. Also, we clear all of the read-ahead state
1580 * kept for the inode here since the file is now closed.
1581 */
1582 int
1585 {
1587 xfs_bmap_free_t free_list;
1588 xfs_fsblock_t first_block;
1597 /*
1598 * If the inode is already free, then there can be nothing
1599 * to clean up here.
1600 */
1605 }
1607 /*
1608 * Only do a truncate if it's a regular file with
1609 * some actual space in it. It's OK to look at the
1610 * inode's fields without the lock because we're the
1611 * only one with a reference to the inode.
1612 */
1622 }
1626 /* If this is a read-only mount, don't do this (would generate I/O) */
1641 /* Update linux inode block count after free above */
1644 }
1646 }
1655 /*
1656 * Do the xfs_itruncate_start() call before
1657 * reserving any log space because itruncate_start
1658 * will call into the buffer cache and we can't
1659 * do that within a transaction.
1660 */
1668 }
1673 XFS_ITRUNCATE_LOG_COUNT);
1675 /* Don't call itruncate_cleanup */
1680 }
1686 /*
1687 * normally, we have to run xfs_itruncate_finish sync.
1688 * But if filesystem is wsync and we're in the inactive
1689 * path, then we know that nlink == 0, and that the
1690 * xaction that made nlink == 0 is permanently committed
1691 * since xfs_remove runs as a synchronous transaction.
1692 */
1701 }
1704 /*
1705 * If we get an error while cleaning up a
1706 * symlink we bail out.
1707 */
1715 }
1723 XFS_INACTIVE_LOG_COUNT);
1728 }
1733 }
1735 /*
1736 * If there are attributes associated with the file
1737 * then blow them away now. The code calls a routine
1738 * that recursively deconstructs the attribute fork.
1739 * We need to just commit the current transaction
1740 * because we can't use it for xfs_attr_inactive().
1741 */
1744 /*
1745 * If we got an error, the transaction is already
1746 * cancelled, and the inode is unlocked. Just get out.
1747 */
1752 }
1754 /*
1755 * Free the inode.
1756 */
1760 /*
1761 * If we fail to free the inode, shut down. The cancel
1762 * might do that, we need to make sure. Otherwise the
1763 * inode might be lost for a long time or forever.
1764 */
1770 }
1773 /*
1774 * Credit the quota account(s). The inode is gone.
1775 */
1778 /*
1779 * Just ignore errors at this point. There is
1780 * nothing we can do except to try to keep going.
1781 */
1784 }
1785 /*
1786 * Release the dquots held by inode, if any.
1787 */
1792 out:
1794 }
1797 int
1802 {
1804 xfs_ino_t e_inum;
1806 uint lock_mode;
1818 }
1821 }
1823 int
1830 {
1837 xfs_dev_t rdev;
1839 xfs_bmap_free_t free_list;
1840 xfs_fsblock_t first_block;
1843 uint cancel_flags;
1845 xfs_prid_t prid;
1847 uint resblks;
1866 }
1871 /* Return through std_return after this point. */
1878 else
1881 /*
1882 * Make sure that we have allocated dquot(s) on disk.
1883 */
1895 /*
1896 * Initially assume that the file does not exist and
1897 * reserve the resources for that case. If that is not
1898 * the case we'll drop the one we have and get a more
1899 * appropriate transaction later.
1900 */
1907 }
1911 }
1920 /*
1921 * Reserve disk quota and the inode.
1922 */
1937 }
1940 /*
1941 * At this point, we've gotten a newly allocated inode.
1942 * It is locked (and joined to the transaction).
1943 */
1947 /*
1948 * Now we join the directory inode to the transaction. We do not do it
1949 * earlier because xfs_dir_ialloc might commit the previous transaction
1950 * (and release all the locks). An error from here on will result in
1951 * the transaction cancel unlocking dp so don't do it explicitly in the
1952 * error path.
1953 */
1964 }
1968 /*
1969 * If this is a synchronous mount, make sure that the
1970 * create transaction goes to disk before returning to
1971 * the user.
1972 */
1975 }
1979 /*
1980 * Attach the dquot(s) to the inodes and modify them incore.
1981 * These ids of the inode couldn't have changed since the new
1982 * inode has been locked ever since it was created.
1983 */
1986 /*
1987 * xfs_trans_commit normally decrements the vnode ref count
1988 * when it unlocks the inode. Since we want to return the
1989 * vnode to the caller, we bump the vnode ref count now.
1990 */
1998 }
2005 }
2012 /* Fallthrough to std_return with error = 0 */
2014 std_return:
2022 }
2025 abort_return:
2027 /* FALLTHROUGH */
2029 error_return:
2041 abort_rele:
2042 /*
2043 * Wait until after the current transaction is aborted to
2044 * release the inode. This prevents recursive transactions
2045 * and deadlocks from xfs_inactive.
2046 */
2055 }
2057 #ifdef DEBUG
2058 /*
2059 * Some counters to see if (and how often) we are hitting some deadlock
2060 * prevention code paths.
2061 */
2066 #endif
2068 /*
2069 * The following routine will lock the inodes associated with the
2070 * directory and the named entry in the directory. The locks are
2071 * acquired in increasing inode number.
2072 *
2073 * If the entry is "..", then only the directory is locked. The
2074 * vnode ref count will still include that from the .. entry in
2075 * this case.
2076 *
2077 * There is a deadlock we need to worry about. If the locked directory is
2078 * in the AIL, it might be blocking up the log. The next inode we lock
2079 * could be already locked by another thread waiting for log space (e.g
2080 * a permanent log reservation with a long running transaction (see
2081 * xfs_itruncate_finish)). To solve this, we must check if the directory
2082 * is in the ail and use lock_nowait. If we can't lock, we need to
2083 * drop the inode lock on the directory and try again. xfs_iunlock will
2084 * potentially push the tail if we were holding up the log.
2085 */
2086 STATIC int
2090 {
2092 xfs_ino_t e_inum;
2096 #ifdef DEBUG
2097 xfs_rm_locks++;
2098 #endif
2101 again:
2108 /*
2109 * We want to lock in increasing inum. Since we've already
2110 * acquired the lock on the directory, we may need to release
2111 * if if the inum of the entry turns out to be less.
2112 */
2114 /*
2115 * We are already in the right order, so just
2116 * lock on the inode of the entry.
2117 * We need to use nowait if dp is in the AIL.
2118 */
2123 attempts++;
2124 #ifdef DEBUG
2125 xfs_rm_attempts++;
2126 #endif
2128 /*
2129 * Unlock dp and try again.
2130 * xfs_iunlock will try to push the tail
2131 * if the inode is in the AIL.
2132 */
2138 #ifdef DEBUG
2139 xfs_rm_lock_delays++;
2140 #endif
2141 }
2143 }
2146 }
2153 }
2154 /* else e_inum == dp->i_ino */
2155 /* This can happen if we're asked to lock /x/..
2156 * the entry is "..", which is also the parent directory.
2157 */
2160 }
2162 #ifdef DEBUG
2168 #endif
2170 /*
2171 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2172 * a different value
2173 */
2176 {
2183 }
2185 /*
2186 * The following routine will lock n inodes in exclusive mode.
2187 * We assume the caller calls us with the inodes in i_ino order.
2188 *
2189 * We need to detect deadlock where an inode that we lock
2190 * is in the AIL and we start waiting for another inode that is locked
2191 * by a thread in a long running transaction (such as truncate). This can
2192 * result in deadlock since the long running trans might need to wait
2193 * for the inode we just locked in order to push the tail and free space
2194 * in the log.
2195 */
2196 void
2201 uint lock_mode)
2202 {
2214 }
2216 again:
2223 /*
2224 * If try_lock is not set yet, make sure all locked inodes
2225 * are not in the AIL.
2226 * If any are, set try_lock to be used later.
2227 */
2233 try_lock++;
2234 }
2235 }
2236 }
2238 /*
2239 * If any of the previous locks we have locked is in the AIL,
2240 * we must TRY to get the second and subsequent locks. If
2241 * we can't get any, we must release all we have
2242 * and try again.
2243 */
2246 /* try_lock must be 0 if i is 0. */
2247 /*
2248 * try_lock means we have an inode locked
2249 * that is in the AIL.
2250 */
2253 attempts++;
2255 /*
2256 * Unlock all previous guys and try again.
2257 * xfs_iunlock will try to push the tail
2258 * if the inode is in the AIL.
2259 */
2263 /*
2264 * Check to see if we've already
2265 * unlocked this one.
2266 * Not the first one going back,
2267 * and the inode ptr is the same.
2268 */
2274 }
2278 #ifdef DEBUG
2279 xfs_lock_delays++;
2280 #endif
2281 }
2285 }
2288 }
2289 }
2291 #ifdef DEBUG
2297 xfs_locked_n++;
2298 }
2299 #endif
2300 }
2302 #ifdef DEBUG
2303 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2306 #define REMOVE_DEBUG_TRACE(x)
2309 int
2313 {
2320 xfs_bmap_free_t free_list;
2321 xfs_fsblock_t first_block;
2326 uint resblks;
2339 }
2347 }
2349 /* From this point on, return through std_return */
2352 /*
2353 * We need to get a reference to ip before we get our log
2354 * reservation. The reason for this is that we cannot call
2355 * xfs_iget for an inode for which we do not have a reference
2356 * once we've acquired a log reservation. This is because the
2357 * inode we are trying to get might be in xfs_inactive going
2358 * for a log reservation. Since we'll have to wait for the
2359 * inactive code to complete before returning from xfs_iget,
2360 * we need to make sure that we don't have log space reserved
2361 * when we call xfs_iget. Instead we get an unlocked reference
2362 * to the inode before getting our log reservation.
2363 */
2368 }
2383 }
2387 /*
2388 * We try to get the real space reservation first,
2389 * allowing for directory btree deletion(s) implying
2390 * possible bmap insert(s). If we can't get the space
2391 * reservation then we use 0 instead, and avoid the bmap
2392 * btree insert(s) in the directory code by, if the bmap
2393 * insert tries to happen, instead trimming the LAST
2394 * block from the directory.
2395 */
2403 }
2410 }
2418 }
2420 /*
2421 * At this point, we've gotten both the directory and the entry
2422 * inodes locked.
2423 */
2426 /*
2427 * Increment vnode ref count only in this case since
2428 * there's an extra vnode reference in the case where
2429 * dp == ip.
2430 */
2433 }
2435 /*
2436 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2437 */
2445 }
2455 }
2457 /* Determine if this is the last link while
2458 * we are in the transaction.
2459 */
2462 /*
2463 * Take an extra ref on the inode so that it doesn't
2464 * go to xfs_inactive() from within the commit.
2465 */
2468 /*
2469 * If this is a synchronous mount, make sure that the
2470 * remove transaction goes to disk before returning to
2471 * the user.
2472 */
2475 }
2481 }
2487 }
2489 /*
2490 * Before we drop our extra reference to the inode, purge it
2491 * from the refcache if it is there. By waiting until afterwards
2492 * to do the IRELE, we ensure that we won't go inactive in the
2493 * xfs_refcache_purge_ip routine (although that would be OK).
2494 */
2497 /*
2498 * If we are using filestreams, kill the stream association.
2499 * If the file is still open it may get a new one but that
2500 * will get killed on last close in xfs_close() so we don't
2501 * have to worry about that.
2502 */
2510 /* Fall through to std_return with error = 0 */
2511 std_return:
2517 }
2520 error1:
2526 error_rele:
2527 /*
2528 * In this case make sure to not release the inode until after
2529 * the current transaction is aborted. Releasing it beforehand
2530 * can cause us to go to xfs_inactive and start a recursive
2531 * transaction which can easily deadlock with the current one.
2532 */
2537 /*
2538 * Before we drop our extra reference to the inode, purge it
2539 * from the refcache if it is there. By waiting until afterwards
2540 * to do the IRELE, we ensure that we won't go inactive in the
2541 * xfs_refcache_purge_ip routine (although that would be OK).
2542 */
2548 }
2550 int
2555 {
2562 xfs_bmap_free_t free_list;
2563 xfs_fsblock_t first_block;
2586 }
2588 /* Return through std_return after this point. */
2605 }
2609 }
2617 }
2621 /*
2622 * Increment vnode ref counts since xfs_trans_commit &
2623 * xfs_trans_cancel will both unlock the inodes and
2624 * decrement the associated ref counts.
2625 */
2631 /*
2632 * If the source has too many links, we can't make any more to it.
2633 */
2637 }
2639 /*
2640 * If we are using project inheritance, we only allow hard link
2641 * creation in our tree when the project IDs are the same; else
2642 * the tree quota mechanism could be circumvented.
2643 */
2648 }
2658 resblks);
2669 /*
2670 * If this is a synchronous mount, make sure that the
2671 * link transaction goes to disk before returning to
2672 * the user.
2673 */
2676 }
2682 }
2688 /* Fall through to std_return with error = 0. */
2689 std_return:
2695 }
2698 abort_return:
2700 /* FALLTHROUGH */
2702 error_return:
2705 }
2708 int
2715 {
2726 xfs_bmap_free_t free_list;
2727 xfs_fsblock_t first_block;
2731 xfs_prid_t prid;
2733 uint resblks;
2750 }
2752 /* Return through std_return after this point. */
2762 else
2765 /*
2766 * Make sure that we have allocated dquot(s) on disk.
2767 */
2782 XFS_TRANS_PERM_LOG_RES,
2783 XFS_MKDIR_LOG_COUNT);
2784 }
2788 }
2793 /*
2794 * Check for directory link count overflow.
2795 */
2799 }
2801 /*
2802 * Reserve disk quota and the inode.
2803 */
2811 /*
2812 * create the directory inode.
2813 */
2821 }
2824 /*
2825 * Now we add the directory inode to the transaction.
2826 * We waited until now since xfs_dir_ialloc might start
2827 * a new transaction. Had we joined the transaction
2828 * earlier, the locks might have gotten released. An error
2829 * from here on will result in the transaction cancel
2830 * unlocking dp so don't do it explicitly in the error path.
2831 */
2844 }
2847 /*
2848 * Bump the in memory version number of the parent directory
2849 * so that other processes accessing it will recognize that
2850 * the directory has changed.
2851 */
2870 /*
2871 * Attach the dquots to the new inode and modify the icount incore.
2872 */
2875 /*
2876 * If this is a synchronous mount, make sure that the
2877 * mkdir transaction goes to disk before returning to
2878 * the user.
2879 */
2882 }
2888 }
2895 }
2897 /* Fall through to std_return with error = 0 or errno from
2898 * xfs_trans_commit. */
2900 std_return:
2906 DM_RIGHT_NULL,
2909 }
2912 error2:
2913 error1:
2915 abort_return:
2917 error_return:
2926 }
2928 int
2932 {
2940 xfs_bmap_free_t free_list;
2941 xfs_fsblock_t first_block;
2946 uint resblks;
2956 }
2965 }
2967 /* Return through std_return after this point. */
2971 /*
2972 * We need to get a reference to cdp before we get our log
2973 * reservation. The reason for this is that we cannot call
2974 * xfs_iget for an inode for which we do not have a reference
2975 * once we've acquired a log reservation. This is because the
2976 * inode we are trying to get might be in xfs_inactive going
2977 * for a log reservation. Since we'll have to wait for the
2978 * inactive code to complete before returning from xfs_iget,
2979 * we need to make sure that we don't have log space reserved
2980 * when we call xfs_iget. Instead we get an unlocked reference
2981 * to the inode before getting our log reservation.
2982 */
2987 }
2991 /*
2992 * Get the dquots for the inodes.
2993 */
3001 }
3005 /*
3006 * We try to get the real space reservation first,
3007 * allowing for directory btree deletion(s) implying
3008 * possible bmap insert(s). If we can't get the space
3009 * reservation then we use 0 instead, and avoid the bmap
3010 * btree insert(s) in the directory code by, if the bmap
3011 * insert tries to happen, instead trimming the LAST
3012 * block from the directory.
3013 */
3021 }
3027 }
3030 /*
3031 * Now lock the child directory inode and the parent directory
3032 * inode in the proper order. This will take care of validating
3033 * that the directory entry for the child directory inode has
3034 * not changed while we were obtaining a log reservation.
3035 */
3041 }
3045 /*
3046 * Only increment the parent directory vnode count if
3047 * we didn't bump it in looking up cdp. The only time
3048 * we don't bump it is when we're looking up ".".
3049 */
3051 }
3060 }
3064 }
3073 /*
3074 * Bump the in memory generation count on the parent
3075 * directory so that other can know that it has changed.
3076 */
3079 /*
3080 * Drop the link from cdp's "..".
3081 */
3085 }
3087 /*
3088 * Drop the link from dp to cdp.
3089 */
3093 }
3095 /*
3096 * Drop the "." link from cdp to self.
3097 */
3101 }
3103 /* Determine these before committing transaction */
3106 /*
3107 * Take an extra ref on the child vnode so that it
3108 * does not go to xfs_inactive() from within the commit.
3109 */
3112 /*
3113 * If this is a synchronous mount, make sure that the
3114 * rmdir transaction goes to disk before returning to
3115 * the user.
3116 */
3119 }
3125 XFS_TRANS_ABORT));
3128 }
3134 }
3139 /* Fall through to std_return with error = 0 or the errno
3140 * from xfs_trans_commit. */
3141 std_return:
3148 }
3151 error1:
3154 /* FALLTHROUGH */
3156 error_return:
3159 }
3161 int
3169 {
3176 xfs_bmap_free_t free_list;
3177 xfs_fsblock_t first_block;
3179 uint cancel_flags;
3181 xfs_fileoff_t first_fsb;
3182 xfs_filblks_t fs_blocks;
3185 xfs_daddr_t d;
3190 xfs_prid_t prid;
3192 uint resblks;
3209 /*
3210 * Check component lengths of the target path name.
3211 */
3220 /*
3221 * Skip any slashes.
3222 */
3224 total++;
3225 path++;
3226 }
3228 /*
3229 * Count up to the next slash or end of path.
3230 * Error out if the component is bigger than MAXNAMELEN.
3231 */
3236 }
3237 }
3238 }
3239 }
3247 }
3249 /* Return through std_return after this point. */
3256 else
3259 /*
3260 * Make sure that we have allocated dquot(s) on disk.
3261 */
3270 /*
3271 * The symlink will fit into the inode data fork?
3272 * There can't be any attributes so we get the whole variable part.
3273 */
3276 else
3285 }
3289 }
3294 /*
3295 * Check whether the directory allows new symlinks or not.
3296 */
3300 }
3302 /*
3303 * Reserve disk quota : blocks and inode.
3304 */
3309 /*
3310 * Check for ability to enter directory entry, if no space reserved.
3311 */
3315 /*
3316 * Initialize the bmap freelist prior to calling either
3317 * bmapi or the directory create code.
3318 */
3321 /*
3322 * Allocate an inode for the symlink.
3323 */
3330 }
3333 /*
3334 * An error after we've joined dp to the transaction will result in the
3335 * transaction cancel unlocking dp so don't do it explicitly in the
3336 * error path.
3337 */
3342 /*
3343 * Also attach the dquot(s) to it, if applicable.
3344 */
3349 /*
3350 * If the symlink will fit into the inode, write it inline.
3351 */
3357 /*
3358 * The inode was initially created in extent format.
3359 */
3376 }
3392 }
3399 }
3400 }
3402 /*
3403 * Create the directory entry for the symlink.
3404 */
3412 /*
3413 * Bump the in memory version number of the parent directory
3414 * so that other processes accessing it will recognize that
3415 * the directory has changed.
3416 */
3419 /*
3420 * If this is a synchronous mount, make sure that the
3421 * symlink transaction goes to disk before returning to
3422 * the user.
3423 */
3426 }
3428 /*
3429 * xfs_trans_commit normally decrements the vnode ref count
3430 * when it unlocks the inode. Since we want to return the
3431 * vnode to the caller, we bump the vnode ref count now.
3432 */
3438 }
3443 /* Fall through to std_return with error = 0 or errno from
3444 * xfs_trans_commit */
3445 std_return:
3452 }
3460 }
3463 error2:
3465 error1:
3468 error_return:
3477 }
3480 /*
3481 * xfs_fid2
3482 *
3483 * A fid routine that takes a pointer to a previously allocated
3484 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3485 */
3486 int
3490 {
3498 /*
3499 * use memcpy because the inode is a long long and there's no
3500 * assurance that xfid->fid_ino is properly aligned.
3501 */
3506 }
3509 int
3512 bhv_vrwlock_t locktype)
3513 {
3526 }
3529 }
3532 void
3535 bhv_vrwlock_t locktype)
3536 {
3540 /*
3541 * In the write case, we may have added a new entry to
3542 * the reference cache. This might store a pointer to
3543 * an inode to be released in this inode. If it is there,
3544 * clear the pointer and release the inode after unlocking
3545 * this one.
3546 */
3552 }
3554 }
3557 int
3561 {
3569 /*
3570 * Bypass inodes which have already been cleaned by
3571 * the inode flush clustering code inside xfs_iflush
3572 */
3580 xfs_lsn_t sync_lsn;
3593 }
3597 }
3598 }
3600 /*
3601 * We make this non-blocking if the inode is contended,
3602 * return EAGAIN to indicate to the caller that they
3603 * did not succeed. This prevents the flush path from
3604 * blocking on inodes inside another operation right
3605 * now, they get caught later by xfs_sync.
3606 */
3617 }
3620 }
3624 else
3629 }
3632 }
3635 int
3638 u_int evmask,
3639 u_int16_t state)
3640 {
3656 }
3668 }
3670 int
3673 {
3680 /* bad inode, get out here ASAP */
3684 }
3690 /*
3691 * Make sure the atime in the XFS inode is correct before freeing the
3692 * Linux inode.
3693 */
3696 /*
3697 * If we have nothing to flush with this inode then complete the
3698 * teardown now, otherwise break the link between the xfs inode and the
3699 * linux inode and clean up the xfs inode later. This avoids flushing
3700 * the inode to disk during the delete operation itself.
3701 *
3702 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3703 * first to ensure that xfs_iunpin() will never see an xfs inode
3704 * that has a linux inode being reclaimed. Synchronisation is provided
3705 * by the i_flags_lock.
3706 */
3714 /* Protect sync and unpin from us */
3723 }
3725 }
3727 int
3732 {
3740 /* The hash lock here protects a thread in xfs_iget_core from
3741 * racing with us on linking the inode back with a vnode.
3742 * Once we have the XFS_IRECLAIM flag set it will not touch
3743 * us.
3744 */
3754 }
3756 }
3762 /*
3763 * If the inode is still dirty, then flush it out. If the inode
3764 * is not in the AIL, then it will be OK to flush it delwri as
3765 * long as xfs_iflush() does not keep any references to the inode.
3766 * We leave that decision up to xfs_iflush() since it has the
3767 * knowledge of whether it's OK to simply do a delwri flush of
3768 * the inode or whether we need to wait until the inode is
3769 * pulled from the AIL.
3770 * We get the flush lock regardless, though, just to make sure
3771 * we don't free it while it is being flushed.
3772 */
3777 }
3783 /*
3784 * If we hit an error, typically because of filesystem
3785 * shutdown, we don't need to let vn_reclaim to know
3786 * because we're gonna reclaim the inode anyway.
3787 */
3791 }
3793 }
3800 /*
3801 * We are not interested in doing an iflush if we're
3802 * in the process of shutting down the filesystem forcibly.
3803 * So, just reclaim the inode.
3804 */
3807 }
3809 reclaim:
3812 }
3814 int
3816 {
3832 }
3833 }
3836 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3840 }
3843 }
3847 }
3849 /*
3850 * xfs_alloc_file_space()
3851 * This routine allocates disk space for the given file.
3852 *
3853 * If alloc_type == 0, this request is for an ALLOCSP type
3854 * request which will change the file size. In this case, no
3855 * DMAPI event will be generated by the call. A TRUNCATE event
3856 * will be generated later by xfs_setattr.
3857 *
3858 * If alloc_type != 0, this request is for a RESVSP type
3859 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3860 * lower block boundary byte address is less than the file's
3861 * length.
3862 *
3863 * RETURNS:
3864 * 0 on success
3865 * errno on error
3866 *
3867 */
3868 STATIC int
3871 xfs_off_t offset,
3872 xfs_off_t len,
3875 {
3877 xfs_off_t count;
3878 xfs_filblks_t allocated_fsb;
3879 xfs_filblks_t allocatesize_fsb;
3881 xfs_fileoff_t startoffset_fsb;
3882 xfs_fsblock_t firstfsb;
3889 xfs_bmap_free_t free_list;
3915 /* Generate a DMAPI event if needed. */
3919 xfs_off_t end_dmi_offset;
3929 }
3931 /*
3932 * Allocate file space until done or until there is an error
3933 */
3934 retry:
3938 /*
3939 * Determine space reservations for data/realtime.
3940 */
3953 }
3965 }
3967 /*
3968 * Allocate and setup the transaction.
3969 */
3973 XFS_TRANS_PERM_LOG_RES,
3974 XFS_WRITE_LOG_COUNT);
3975 /*
3976 * Check for running out of space
3977 */
3979 /*
3980 * Free the transaction structure.
3981 */
3985 }
3995 /*
3996 * Issue the xfs_bmapi() call to allocate the blocks
3997 */
4005 }
4007 /*
4008 * Complete the transaction
4009 */
4013 }
4019 }
4026 }
4030 }
4031 dmapi_enospc_check:
4040 /* else fall through with error from XFS_SEND_DATA */
4041 }
4053 }
4055 /*
4056 * Zero file bytes between startoff and endoff inclusive.
4057 * The iolock is held exclusive and no blocks are buffered.
4058 */
4059 STATIC int
4062 xfs_off_t startoff,
4063 xfs_off_t endoff)
4064 {
4065 xfs_bmbt_irec_t imap;
4066 xfs_fileoff_t offset_fsb;
4067 xfs_off_t lastoffset;
4068 xfs_off_t offset;
4104 }
4116 }
4117 }
4120 }
4122 /*
4123 * xfs_free_file_space()
4124 * This routine frees disk space for the given file.
4125 *
4126 * This routine is only called by xfs_change_file_space
4127 * for an UNRESVSP type call.
4128 *
4129 * RETURNS:
4130 * 0 on success
4131 * errno on error
4132 *
4133 */
4134 STATIC int
4137 xfs_off_t offset,
4138 xfs_off_t len,
4140 {
4144 xfs_off_t end_dmi_offset;
4145 xfs_fileoff_t endoffset_fsb;
4147 xfs_fsblock_t firstfsb;
4148 xfs_bmap_free_t free_list;
4149 xfs_bmbt_irec_t imap;
4150 xfs_off_t ioffset;
4154 uint resblks;
4155 uint rounding;
4157 xfs_fileoff_t startoffset_fsb;
4186 }
4193 }
4206 }
4208 /*
4209 * Need to zero the stuff we're not freeing, on disk.
4210 * If its a realtime file & can't use unwritten extents then we
4211 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4212 * will take care of it for us.
4213 */
4222 xfs_daddr_t block;
4229 }
4238 mod++;
4241 }
4242 }
4244 /*
4245 * One contiguous piece to clear
4246 */
4249 /*
4250 * Some full blocks, possibly two pieces to clear
4251 */
4260 }
4262 /*
4263 * free file space until done or until there is an error
4264 */
4268 /*
4269 * allocate and setup the transaction. Allow this
4270 * transaction to dip into the reserve blocks to ensure
4271 * the freeing of the space succeeds at ENOSPC.
4272 */
4276 resblks,
4279 XFS_TRANS_PERM_LOG_RES,
4280 XFS_WRITE_LOG_COUNT);
4282 /*
4283 * check for running out of space
4284 */
4286 /*
4287 * Free the transaction structure.
4288 */
4292 }
4296 XFS_QMOPT_RES_REGBLKS);
4303 /*
4304 * issue the bunmapi() call to free the blocks
4305 */
4312 }
4314 /*
4315 * complete the transaction
4316 */
4320 }
4324 }
4326 out_unlock_iolock:
4331 error0:
4333 error1:
4336 XFS_ILOCK_EXCL);
4338 }
4340 /*
4341 * xfs_change_file_space()
4342 * This routine allocates or frees disk space for the given file.
4343 * The user specified parameters are checked for alignment and size
4344 * limitations.
4345 *
4346 * RETURNS:
4347 * 0 on success
4348 * errno on error
4349 *
4350 */
4351 int
4356 xfs_off_t offset,
4359 {
4363 xfs_fsize_t fsize;
4365 xfs_off_t startoffset;
4366 xfs_off_t llen;
4368 bhv_vattr_t va;
4372 /*
4373 * must be a regular file and have write permission
4374 */
4383 }
4398 }
4413 /*
4414 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4415 * file space.
4416 * These calls do NOT zero the data space allocated to the file,
4417 * nor do they change the file size.
4418 *
4419 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4420 * space.
4421 * These calls cause the new file data to be zeroed and the file
4422 * size to be changed.
4423 */
4439 attr_flags)))
4452 }
4468 }
4470 /*
4471 * update the inode timestamp, mode, and prealloc flag bits
4472 */
4477 /* ASSERT(0); */
4480 }
4490 /*
4491 * Note that we don't have to worry about mandatory
4492 * file locking being disabled here because we only
4493 * clear the S_ISGID bit if the Group execute bit is
4494 * on, but if it was on then mandatory locking wouldn't
4495 * have been enabled.
4496 */
4501 }
4515 }