| blob | efd5aff9eaf6aa1d149fa6431ca6355e944c5b1a |
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
1827 mode_t mode,
1828 xfs_dev_t rdev,
1831 {
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;
1864 }
1869 /* Return through std_return after this point. */
1874 else
1877 /*
1878 * Make sure that we have allocated dquot(s) on disk.
1879 */
1891 /*
1892 * Initially assume that the file does not exist and
1893 * reserve the resources for that case. If that is not
1894 * the case we'll drop the one we have and get a more
1895 * appropriate transaction later.
1896 */
1903 }
1907 }
1916 /*
1917 * Reserve disk quota and the inode.
1918 */
1932 }
1935 /*
1936 * At this point, we've gotten a newly allocated inode.
1937 * It is locked (and joined to the transaction).
1938 */
1942 /*
1943 * Now we join the directory inode to the transaction. We do not do it
1944 * earlier because xfs_dir_ialloc might commit the previous transaction
1945 * (and release all the locks). An error from here on will result in
1946 * the transaction cancel unlocking dp so don't do it explicitly in the
1947 * error path.
1948 */
1959 }
1963 /*
1964 * If this is a synchronous mount, make sure that the
1965 * create transaction goes to disk before returning to
1966 * the user.
1967 */
1970 }
1974 /*
1975 * Attach the dquot(s) to the inodes and modify them incore.
1976 * These ids of the inode couldn't have changed since the new
1977 * inode has been locked ever since it was created.
1978 */
1981 /*
1982 * xfs_trans_commit normally decrements the vnode ref count
1983 * when it unlocks the inode. Since we want to return the
1984 * vnode to the caller, we bump the vnode ref count now.
1985 */
1993 }
2000 }
2007 /* Fallthrough to std_return with error = 0 */
2009 std_return:
2017 }
2020 abort_return:
2022 /* FALLTHROUGH */
2024 error_return:
2036 abort_rele:
2037 /*
2038 * Wait until after the current transaction is aborted to
2039 * release the inode. This prevents recursive transactions
2040 * and deadlocks from xfs_inactive.
2041 */
2050 }
2052 #ifdef DEBUG
2053 /*
2054 * Some counters to see if (and how often) we are hitting some deadlock
2055 * prevention code paths.
2056 */
2061 #endif
2063 /*
2064 * The following routine will lock the inodes associated with the
2065 * directory and the named entry in the directory. The locks are
2066 * acquired in increasing inode number.
2067 *
2068 * If the entry is "..", then only the directory is locked. The
2069 * vnode ref count will still include that from the .. entry in
2070 * this case.
2071 *
2072 * There is a deadlock we need to worry about. If the locked directory is
2073 * in the AIL, it might be blocking up the log. The next inode we lock
2074 * could be already locked by another thread waiting for log space (e.g
2075 * a permanent log reservation with a long running transaction (see
2076 * xfs_itruncate_finish)). To solve this, we must check if the directory
2077 * is in the ail and use lock_nowait. If we can't lock, we need to
2078 * drop the inode lock on the directory and try again. xfs_iunlock will
2079 * potentially push the tail if we were holding up the log.
2080 */
2081 STATIC int
2085 {
2087 xfs_ino_t e_inum;
2091 #ifdef DEBUG
2092 xfs_rm_locks++;
2093 #endif
2096 again:
2103 /*
2104 * We want to lock in increasing inum. Since we've already
2105 * acquired the lock on the directory, we may need to release
2106 * if if the inum of the entry turns out to be less.
2107 */
2109 /*
2110 * We are already in the right order, so just
2111 * lock on the inode of the entry.
2112 * We need to use nowait if dp is in the AIL.
2113 */
2118 attempts++;
2119 #ifdef DEBUG
2120 xfs_rm_attempts++;
2121 #endif
2123 /*
2124 * Unlock dp and try again.
2125 * xfs_iunlock will try to push the tail
2126 * if the inode is in the AIL.
2127 */
2133 #ifdef DEBUG
2134 xfs_rm_lock_delays++;
2135 #endif
2136 }
2138 }
2141 }
2148 }
2149 /* else e_inum == dp->i_ino */
2150 /* This can happen if we're asked to lock /x/..
2151 * the entry is "..", which is also the parent directory.
2152 */
2155 }
2157 #ifdef DEBUG
2163 #endif
2165 /*
2166 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2167 * a different value
2168 */
2171 {
2178 }
2180 /*
2181 * The following routine will lock n inodes in exclusive mode.
2182 * We assume the caller calls us with the inodes in i_ino order.
2183 *
2184 * We need to detect deadlock where an inode that we lock
2185 * is in the AIL and we start waiting for another inode that is locked
2186 * by a thread in a long running transaction (such as truncate). This can
2187 * result in deadlock since the long running trans might need to wait
2188 * for the inode we just locked in order to push the tail and free space
2189 * in the log.
2190 */
2191 void
2196 uint lock_mode)
2197 {
2209 }
2211 again:
2218 /*
2219 * If try_lock is not set yet, make sure all locked inodes
2220 * are not in the AIL.
2221 * If any are, set try_lock to be used later.
2222 */
2228 try_lock++;
2229 }
2230 }
2231 }
2233 /*
2234 * If any of the previous locks we have locked is in the AIL,
2235 * we must TRY to get the second and subsequent locks. If
2236 * we can't get any, we must release all we have
2237 * and try again.
2238 */
2241 /* try_lock must be 0 if i is 0. */
2242 /*
2243 * try_lock means we have an inode locked
2244 * that is in the AIL.
2245 */
2248 attempts++;
2250 /*
2251 * Unlock all previous guys and try again.
2252 * xfs_iunlock will try to push the tail
2253 * if the inode is in the AIL.
2254 */
2258 /*
2259 * Check to see if we've already
2260 * unlocked this one.
2261 * Not the first one going back,
2262 * and the inode ptr is the same.
2263 */
2269 }
2273 #ifdef DEBUG
2274 xfs_lock_delays++;
2275 #endif
2276 }
2280 }
2283 }
2284 }
2286 #ifdef DEBUG
2292 xfs_locked_n++;
2293 }
2294 #endif
2295 }
2297 #ifdef DEBUG
2298 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2301 #define REMOVE_DEBUG_TRACE(x)
2304 int
2308 {
2315 xfs_bmap_free_t free_list;
2316 xfs_fsblock_t first_block;
2321 uint resblks;
2334 }
2342 }
2344 /* From this point on, return through std_return */
2347 /*
2348 * We need to get a reference to ip before we get our log
2349 * reservation. The reason for this is that we cannot call
2350 * xfs_iget for an inode for which we do not have a reference
2351 * once we've acquired a log reservation. This is because the
2352 * inode we are trying to get might be in xfs_inactive going
2353 * for a log reservation. Since we'll have to wait for the
2354 * inactive code to complete before returning from xfs_iget,
2355 * we need to make sure that we don't have log space reserved
2356 * when we call xfs_iget. Instead we get an unlocked reference
2357 * to the inode before getting our log reservation.
2358 */
2363 }
2378 }
2382 /*
2383 * We try to get the real space reservation first,
2384 * allowing for directory btree deletion(s) implying
2385 * possible bmap insert(s). If we can't get the space
2386 * reservation then we use 0 instead, and avoid the bmap
2387 * btree insert(s) in the directory code by, if the bmap
2388 * insert tries to happen, instead trimming the LAST
2389 * block from the directory.
2390 */
2398 }
2405 }
2413 }
2415 /*
2416 * At this point, we've gotten both the directory and the entry
2417 * inodes locked.
2418 */
2421 /*
2422 * Increment vnode ref count only in this case since
2423 * there's an extra vnode reference in the case where
2424 * dp == ip.
2425 */
2428 }
2430 /*
2431 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2432 */
2440 }
2450 }
2452 /* Determine if this is the last link while
2453 * we are in the transaction.
2454 */
2457 /*
2458 * Take an extra ref on the inode so that it doesn't
2459 * go to xfs_inactive() from within the commit.
2460 */
2463 /*
2464 * If this is a synchronous mount, make sure that the
2465 * remove transaction goes to disk before returning to
2466 * the user.
2467 */
2470 }
2476 }
2482 }
2484 /*
2485 * Before we drop our extra reference to the inode, purge it
2486 * from the refcache if it is there. By waiting until afterwards
2487 * to do the IRELE, we ensure that we won't go inactive in the
2488 * xfs_refcache_purge_ip routine (although that would be OK).
2489 */
2492 /*
2493 * If we are using filestreams, kill the stream association.
2494 * If the file is still open it may get a new one but that
2495 * will get killed on last close in xfs_close() so we don't
2496 * have to worry about that.
2497 */
2505 /* Fall through to std_return with error = 0 */
2506 std_return:
2512 }
2515 error1:
2521 error_rele:
2522 /*
2523 * In this case make sure to not release the inode until after
2524 * the current transaction is aborted. Releasing it beforehand
2525 * can cause us to go to xfs_inactive and start a recursive
2526 * transaction which can easily deadlock with the current one.
2527 */
2532 /*
2533 * Before we drop our extra reference to the inode, purge it
2534 * from the refcache if it is there. By waiting until afterwards
2535 * to do the IRELE, we ensure that we won't go inactive in the
2536 * xfs_refcache_purge_ip routine (although that would be OK).
2537 */
2543 }
2545 int
2550 {
2557 xfs_bmap_free_t free_list;
2558 xfs_fsblock_t first_block;
2581 }
2583 /* Return through std_return after this point. */
2600 }
2604 }
2612 }
2616 /*
2617 * Increment vnode ref counts since xfs_trans_commit &
2618 * xfs_trans_cancel will both unlock the inodes and
2619 * decrement the associated ref counts.
2620 */
2626 /*
2627 * If the source has too many links, we can't make any more to it.
2628 */
2632 }
2634 /*
2635 * If we are using project inheritance, we only allow hard link
2636 * creation in our tree when the project IDs are the same; else
2637 * the tree quota mechanism could be circumvented.
2638 */
2643 }
2653 resblks);
2664 /*
2665 * If this is a synchronous mount, make sure that the
2666 * link transaction goes to disk before returning to
2667 * the user.
2668 */
2671 }
2677 }
2683 /* Fall through to std_return with error = 0. */
2684 std_return:
2690 }
2693 abort_return:
2695 /* FALLTHROUGH */
2697 error_return:
2700 }
2703 int
2707 mode_t mode,
2710 {
2721 xfs_bmap_free_t free_list;
2722 xfs_fsblock_t first_block;
2726 xfs_prid_t prid;
2728 uint resblks;
2743 }
2745 /* Return through std_return after this point. */
2753 else
2756 /*
2757 * Make sure that we have allocated dquot(s) on disk.
2758 */
2773 XFS_TRANS_PERM_LOG_RES,
2774 XFS_MKDIR_LOG_COUNT);
2775 }
2779 }
2784 /*
2785 * Check for directory link count overflow.
2786 */
2790 }
2792 /*
2793 * Reserve disk quota and the inode.
2794 */
2802 /*
2803 * create the directory inode.
2804 */
2812 }
2815 /*
2816 * Now we add the directory inode to the transaction.
2817 * We waited until now since xfs_dir_ialloc might start
2818 * a new transaction. Had we joined the transaction
2819 * earlier, the locks might have gotten released. An error
2820 * from here on will result in the transaction cancel
2821 * unlocking dp so don't do it explicitly in the error path.
2822 */
2835 }
2838 /*
2839 * Bump the in memory version number of the parent directory
2840 * so that other processes accessing it will recognize that
2841 * the directory has changed.
2842 */
2861 /*
2862 * Attach the dquots to the new inode and modify the icount incore.
2863 */
2866 /*
2867 * If this is a synchronous mount, make sure that the
2868 * mkdir transaction goes to disk before returning to
2869 * the user.
2870 */
2873 }
2879 }
2886 }
2888 /* Fall through to std_return with error = 0 or errno from
2889 * xfs_trans_commit. */
2891 std_return:
2897 DM_RIGHT_NULL,
2900 }
2903 error2:
2904 error1:
2906 abort_return:
2908 error_return:
2917 }
2919 int
2923 {
2931 xfs_bmap_free_t free_list;
2932 xfs_fsblock_t first_block;
2937 uint resblks;
2947 }
2956 }
2958 /* Return through std_return after this point. */
2962 /*
2963 * We need to get a reference to cdp before we get our log
2964 * reservation. The reason for this is that we cannot call
2965 * xfs_iget for an inode for which we do not have a reference
2966 * once we've acquired a log reservation. This is because the
2967 * inode we are trying to get might be in xfs_inactive going
2968 * for a log reservation. Since we'll have to wait for the
2969 * inactive code to complete before returning from xfs_iget,
2970 * we need to make sure that we don't have log space reserved
2971 * when we call xfs_iget. Instead we get an unlocked reference
2972 * to the inode before getting our log reservation.
2973 */
2978 }
2982 /*
2983 * Get the dquots for the inodes.
2984 */
2992 }
2996 /*
2997 * We try to get the real space reservation first,
2998 * allowing for directory btree deletion(s) implying
2999 * possible bmap insert(s). If we can't get the space
3000 * reservation then we use 0 instead, and avoid the bmap
3001 * btree insert(s) in the directory code by, if the bmap
3002 * insert tries to happen, instead trimming the LAST
3003 * block from the directory.
3004 */
3012 }
3018 }
3021 /*
3022 * Now lock the child directory inode and the parent directory
3023 * inode in the proper order. This will take care of validating
3024 * that the directory entry for the child directory inode has
3025 * not changed while we were obtaining a log reservation.
3026 */
3032 }
3036 /*
3037 * Only increment the parent directory vnode count if
3038 * we didn't bump it in looking up cdp. The only time
3039 * we don't bump it is when we're looking up ".".
3040 */
3042 }
3051 }
3055 }
3064 /*
3065 * Bump the in memory generation count on the parent
3066 * directory so that other can know that it has changed.
3067 */
3070 /*
3071 * Drop the link from cdp's "..".
3072 */
3076 }
3078 /*
3079 * Drop the link from dp to cdp.
3080 */
3084 }
3086 /*
3087 * Drop the "." link from cdp to self.
3088 */
3092 }
3094 /* Determine these before committing transaction */
3097 /*
3098 * Take an extra ref on the child vnode so that it
3099 * does not go to xfs_inactive() from within the commit.
3100 */
3103 /*
3104 * If this is a synchronous mount, make sure that the
3105 * rmdir transaction goes to disk before returning to
3106 * the user.
3107 */
3110 }
3116 XFS_TRANS_ABORT));
3119 }
3125 }
3130 /* Fall through to std_return with error = 0 or the errno
3131 * from xfs_trans_commit. */
3132 std_return:
3139 }
3142 error1:
3145 /* FALLTHROUGH */
3147 error_return:
3150 }
3152 int
3157 mode_t mode,
3160 {
3167 xfs_bmap_free_t free_list;
3168 xfs_fsblock_t first_block;
3170 uint cancel_flags;
3172 xfs_fileoff_t first_fsb;
3173 xfs_filblks_t fs_blocks;
3176 xfs_daddr_t d;
3181 xfs_prid_t prid;
3183 uint resblks;
3200 /*
3201 * Check component lengths of the target path name.
3202 */
3211 /*
3212 * Skip any slashes.
3213 */
3215 total++;
3216 path++;
3217 }
3219 /*
3220 * Count up to the next slash or end of path.
3221 * Error out if the component is bigger than MAXNAMELEN.
3222 */
3227 }
3228 }
3229 }
3230 }
3238 }
3240 /* Return through std_return after this point. */
3245 else
3248 /*
3249 * Make sure that we have allocated dquot(s) on disk.
3250 */
3259 /*
3260 * The symlink will fit into the inode data fork?
3261 * There can't be any attributes so we get the whole variable part.
3262 */
3265 else
3274 }
3278 }
3283 /*
3284 * Check whether the directory allows new symlinks or not.
3285 */
3289 }
3291 /*
3292 * Reserve disk quota : blocks and inode.
3293 */
3298 /*
3299 * Check for ability to enter directory entry, if no space reserved.
3300 */
3304 /*
3305 * Initialize the bmap freelist prior to calling either
3306 * bmapi or the directory create code.
3307 */
3310 /*
3311 * Allocate an inode for the symlink.
3312 */
3319 }
3322 /*
3323 * An error after we've joined dp to the transaction will result in the
3324 * transaction cancel unlocking dp so don't do it explicitly in the
3325 * error path.
3326 */
3331 /*
3332 * Also attach the dquot(s) to it, if applicable.
3333 */
3338 /*
3339 * If the symlink will fit into the inode, write it inline.
3340 */
3346 /*
3347 * The inode was initially created in extent format.
3348 */
3365 }
3381 }
3388 }
3389 }
3391 /*
3392 * Create the directory entry for the symlink.
3393 */
3401 /*
3402 * Bump the in memory version number of the parent directory
3403 * so that other processes accessing it will recognize that
3404 * the directory has changed.
3405 */
3408 /*
3409 * If this is a synchronous mount, make sure that the
3410 * symlink transaction goes to disk before returning to
3411 * the user.
3412 */
3415 }
3417 /*
3418 * xfs_trans_commit normally decrements the vnode ref count
3419 * when it unlocks the inode. Since we want to return the
3420 * vnode to the caller, we bump the vnode ref count now.
3421 */
3427 }
3432 /* Fall through to std_return with error = 0 or errno from
3433 * xfs_trans_commit */
3434 std_return:
3441 }
3449 }
3452 error2:
3454 error1:
3457 error_return:
3466 }
3469 int
3473 {
3478 /*
3479 * use memcpy because the inode is a long long and there's no
3480 * assurance that xfid->fid_ino is properly aligned.
3481 */
3486 }
3489 int
3492 bhv_vrwlock_t locktype)
3493 {
3506 }
3509 }
3512 void
3515 bhv_vrwlock_t locktype)
3516 {
3520 /*
3521 * In the write case, we may have added a new entry to
3522 * the reference cache. This might store a pointer to
3523 * an inode to be released in this inode. If it is there,
3524 * clear the pointer and release the inode after unlocking
3525 * this one.
3526 */
3532 }
3534 }
3537 int
3541 {
3549 /*
3550 * Bypass inodes which have already been cleaned by
3551 * the inode flush clustering code inside xfs_iflush
3552 */
3560 xfs_lsn_t sync_lsn;
3573 }
3577 }
3578 }
3580 /*
3581 * We make this non-blocking if the inode is contended,
3582 * return EAGAIN to indicate to the caller that they
3583 * did not succeed. This prevents the flush path from
3584 * blocking on inodes inside another operation right
3585 * now, they get caught later by xfs_sync.
3586 */
3597 }
3600 }
3604 else
3609 }
3612 }
3615 int
3618 u_int evmask,
3619 u_int16_t state)
3620 {
3636 }
3648 }
3650 int
3653 {
3660 /* bad inode, get out here ASAP */
3664 }
3670 /*
3671 * Make sure the atime in the XFS inode is correct before freeing the
3672 * Linux inode.
3673 */
3676 /*
3677 * If we have nothing to flush with this inode then complete the
3678 * teardown now, otherwise break the link between the xfs inode and the
3679 * linux inode and clean up the xfs inode later. This avoids flushing
3680 * the inode to disk during the delete operation itself.
3681 *
3682 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3683 * first to ensure that xfs_iunpin() will never see an xfs inode
3684 * that has a linux inode being reclaimed. Synchronisation is provided
3685 * by the i_flags_lock.
3686 */
3694 /* Protect sync and unpin from us */
3703 }
3705 }
3707 int
3712 {
3720 /* The hash lock here protects a thread in xfs_iget_core from
3721 * racing with us on linking the inode back with a vnode.
3722 * Once we have the XFS_IRECLAIM flag set it will not touch
3723 * us.
3724 */
3734 }
3736 }
3742 /*
3743 * If the inode is still dirty, then flush it out. If the inode
3744 * is not in the AIL, then it will be OK to flush it delwri as
3745 * long as xfs_iflush() does not keep any references to the inode.
3746 * We leave that decision up to xfs_iflush() since it has the
3747 * knowledge of whether it's OK to simply do a delwri flush of
3748 * the inode or whether we need to wait until the inode is
3749 * pulled from the AIL.
3750 * We get the flush lock regardless, though, just to make sure
3751 * we don't free it while it is being flushed.
3752 */
3757 }
3763 /*
3764 * If we hit an error, typically because of filesystem
3765 * shutdown, we don't need to let vn_reclaim to know
3766 * because we're gonna reclaim the inode anyway.
3767 */
3771 }
3773 }
3780 /*
3781 * We are not interested in doing an iflush if we're
3782 * in the process of shutting down the filesystem forcibly.
3783 * So, just reclaim the inode.
3784 */
3787 }
3789 reclaim:
3792 }
3794 int
3796 {
3812 }
3813 }
3816 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3820 }
3823 }
3827 }
3829 /*
3830 * xfs_alloc_file_space()
3831 * This routine allocates disk space for the given file.
3832 *
3833 * If alloc_type == 0, this request is for an ALLOCSP type
3834 * request which will change the file size. In this case, no
3835 * DMAPI event will be generated by the call. A TRUNCATE event
3836 * will be generated later by xfs_setattr.
3837 *
3838 * If alloc_type != 0, this request is for a RESVSP type
3839 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3840 * lower block boundary byte address is less than the file's
3841 * length.
3842 *
3843 * RETURNS:
3844 * 0 on success
3845 * errno on error
3846 *
3847 */
3848 STATIC int
3851 xfs_off_t offset,
3852 xfs_off_t len,
3855 {
3857 xfs_off_t count;
3858 xfs_filblks_t allocated_fsb;
3859 xfs_filblks_t allocatesize_fsb;
3861 xfs_fileoff_t startoffset_fsb;
3862 xfs_fsblock_t firstfsb;
3869 xfs_bmap_free_t free_list;
3895 /* Generate a DMAPI event if needed. */
3899 xfs_off_t end_dmi_offset;
3909 }
3911 /*
3912 * Allocate file space until done or until there is an error
3913 */
3914 retry:
3918 /*
3919 * Determine space reservations for data/realtime.
3920 */
3933 }
3945 }
3947 /*
3948 * Allocate and setup the transaction.
3949 */
3953 XFS_TRANS_PERM_LOG_RES,
3954 XFS_WRITE_LOG_COUNT);
3955 /*
3956 * Check for running out of space
3957 */
3959 /*
3960 * Free the transaction structure.
3961 */
3965 }
3975 /*
3976 * Issue the xfs_bmapi() call to allocate the blocks
3977 */
3985 }
3987 /*
3988 * Complete the transaction
3989 */
3993 }
3999 }
4006 }
4010 }
4011 dmapi_enospc_check:
4020 /* else fall through with error from XFS_SEND_DATA */
4021 }
4033 }
4035 /*
4036 * Zero file bytes between startoff and endoff inclusive.
4037 * The iolock is held exclusive and no blocks are buffered.
4038 */
4039 STATIC int
4042 xfs_off_t startoff,
4043 xfs_off_t endoff)
4044 {
4045 xfs_bmbt_irec_t imap;
4046 xfs_fileoff_t offset_fsb;
4047 xfs_off_t lastoffset;
4048 xfs_off_t offset;
4084 }
4096 }
4097 }
4100 }
4102 /*
4103 * xfs_free_file_space()
4104 * This routine frees disk space for the given file.
4105 *
4106 * This routine is only called by xfs_change_file_space
4107 * for an UNRESVSP type call.
4108 *
4109 * RETURNS:
4110 * 0 on success
4111 * errno on error
4112 *
4113 */
4114 STATIC int
4117 xfs_off_t offset,
4118 xfs_off_t len,
4120 {
4124 xfs_off_t end_dmi_offset;
4125 xfs_fileoff_t endoffset_fsb;
4127 xfs_fsblock_t firstfsb;
4128 xfs_bmap_free_t free_list;
4129 xfs_bmbt_irec_t imap;
4130 xfs_off_t ioffset;
4134 uint resblks;
4135 uint rounding;
4137 xfs_fileoff_t startoffset_fsb;
4166 }
4173 }
4186 }
4188 /*
4189 * Need to zero the stuff we're not freeing, on disk.
4190 * If its a realtime file & can't use unwritten extents then we
4191 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4192 * will take care of it for us.
4193 */
4202 xfs_daddr_t block;
4209 }
4218 mod++;
4221 }
4222 }
4224 /*
4225 * One contiguous piece to clear
4226 */
4229 /*
4230 * Some full blocks, possibly two pieces to clear
4231 */
4240 }
4242 /*
4243 * free file space until done or until there is an error
4244 */
4248 /*
4249 * allocate and setup the transaction. Allow this
4250 * transaction to dip into the reserve blocks to ensure
4251 * the freeing of the space succeeds at ENOSPC.
4252 */
4256 resblks,
4259 XFS_TRANS_PERM_LOG_RES,
4260 XFS_WRITE_LOG_COUNT);
4262 /*
4263 * check for running out of space
4264 */
4266 /*
4267 * Free the transaction structure.
4268 */
4272 }
4276 XFS_QMOPT_RES_REGBLKS);
4283 /*
4284 * issue the bunmapi() call to free the blocks
4285 */
4292 }
4294 /*
4295 * complete the transaction
4296 */
4300 }
4304 }
4306 out_unlock_iolock:
4311 error0:
4313 error1:
4316 XFS_ILOCK_EXCL);
4318 }
4320 /*
4321 * xfs_change_file_space()
4322 * This routine allocates or frees disk space for the given file.
4323 * The user specified parameters are checked for alignment and size
4324 * limitations.
4325 *
4326 * RETURNS:
4327 * 0 on success
4328 * errno on error
4329 *
4330 */
4331 int
4336 xfs_off_t offset,
4339 {
4343 xfs_fsize_t fsize;
4345 xfs_off_t startoffset;
4346 xfs_off_t llen;
4348 bhv_vattr_t va;
4352 /*
4353 * must be a regular file and have write permission
4354 */
4363 }
4378 }
4393 /*
4394 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4395 * file space.
4396 * These calls do NOT zero the data space allocated to the file,
4397 * nor do they change the file size.
4398 *
4399 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4400 * space.
4401 * These calls cause the new file data to be zeroed and the file
4402 * size to be changed.
4403 */
4419 attr_flags)))
4432 }
4448 }
4450 /*
4451 * update the inode timestamp, mode, and prealloc flag bits
4452 */
4457 /* ASSERT(0); */
4460 }
4470 /*
4471 * Note that we don't have to worry about mandatory
4472 * file locking being disabled here because we only
4473 * clear the S_ISGID bit if the Group execute bit is
4474 * on, but if it was on then mandatory locking wouldn't
4475 * have been enabled.
4476 */
4481 }
4495 }