| blob | 6b71d9f763c79014acb3956269be095cc14a2e62 |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
57 int
60 {
66 /*
67 * If it's a directory with any blocks, read-ahead block 0
68 * as we're almost certain to have the next operation be a read there.
69 */
75 }
77 }
79 /*
80 * xfs_getattr
81 */
82 int
87 {
106 #if XFS_BIG_INUMS
108 #endif
111 /*
112 * Quick exit for non-stat callers
113 */
119 /*
120 * Copy from in-core inode.
121 */
127 /*
128 * Check vnode type block/char vs. everything else.
129 */
143 /*
144 * If the file blocks are being allocated from a
145 * realtime partition, then return the inode's
146 * realtime extent size or the realtime volume's
147 * extent size.
148 */
151 }
153 }
161 /*
162 * Exit for stat callers. See if any of the rest of the fields
163 * to be filled in are needed.
164 */
170 /*
171 * Convert di_flags to xflags.
172 */
175 /*
176 * Exit for inode revalidate. See if any of the rest of
177 * the fields to be filled in are needed.
178 */
194 else
198 all_done:
202 }
205 /*
206 * xfs_setattr
207 */
208 int
214 {
220 uint lock_flags;
236 /*
237 * Cannot set certain attributes.
238 */
242 }
247 /*
248 * Timestamps do not need to be logged and hence do not
249 * need to be done within a transaction.
250 */
258 }
263 /*
264 * If disk quotas is on, we make sure that the dquots do exist on disk,
265 * before we start any other transactions. Trying to do this later
266 * is messy. We don't care to take a readlock to look at the ids
267 * in inode here, because we can't hold it across the trans_reserve.
268 * If the IDs do change before we take the ilock, we're covered
269 * because the i_*dquot fields will get updated anyway.
270 */
280 }
286 }
292 }
293 /*
294 * We take a reference when we initialize udqp and gdqp,
295 * so it is important that we never blindly double trip on
296 * the same variable. See xfs_create() for an example.
297 */
304 }
306 /*
307 * For the other attributes, we acquire the inode lock and
308 * first do an error checking pass.
309 */
324 }
325 }
335 }
336 }
339 }
343 /* boolean: are we the file owner? */
346 /*
347 * Change various properties of a file.
348 * Only the owner or users with CAP_FOWNER
349 * capability may do these things.
350 */
354 /*
355 * CAP_FOWNER overrides the following restrictions:
356 *
357 * The user ID of the calling process must be equal
358 * to the file owner ID, except in cases where the
359 * CAP_FSETID capability is applicable.
360 */
364 }
366 /*
367 * CAP_FSETID overrides the following restrictions:
368 *
369 * The effective user ID of the calling process shall match
370 * the file owner when setting the set-user-ID and
371 * set-group-ID bits on that file.
372 *
373 * The effective group ID or one of the supplementary group
374 * IDs of the calling process shall match the group owner of
375 * the file when setting the set-group-ID bit on that file
376 */
385 #if 0
386 /* Linux allows this, Irix doesn't. */
389 #endif
392 }
393 }
395 /*
396 * Change file ownership. Must be the owner or privileged.
397 * If the system was configured with the "restricted_chown"
398 * option, the owner is not permitted to give away the file,
399 * and can change the group id only to a group of which he
400 * or she is a member.
401 */
403 /*
404 * These IDs could have changed since we last looked at them.
405 * But, we're assured that if the ownership did change
406 * while we didn't have the inode locked, inode's dquot(s)
407 * would have changed also.
408 */
415 iprojid;
417 /*
418 * CAP_CHOWN overrides the following restrictions:
419 *
420 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
421 * shall override the restriction that a process cannot
422 * change the user ID of a file it owns and the restriction
423 * that the group ID supplied to the chown() function
424 * shall be equal to either the group ID or one of the
425 * supplementary group IDs of the calling process.
426 */
433 }
434 /*
435 * Do a quota reservation only if uid/projid/gid is actually
436 * going to change.
437 */
447 }
448 }
450 /*
451 * Truncate file. Must have write permission and not be a directory.
452 */
454 /* Short circuit the truncate case for zero length files */
463 }
471 }
472 /*
473 * Make sure that the dquots are attached to the inode.
474 */
477 }
479 /*
480 * Change file access or modified times.
481 */
488 }
489 }
490 }
492 /*
493 * Change extent size or realtime flag.
494 */
496 /*
497 * Can't change extent size if any extents are allocated.
498 */
504 }
506 /*
507 * Can't change realtime flag if any extents are allocated.
508 */
515 }
516 /*
517 * Extent size must be a multiple of the appropriate block
518 * size, if set at all.
519 */
521 xfs_extlen_t size;
530 }
534 }
535 }
536 /*
537 * If realtime flag is set then must have realtime data.
538 */
546 }
547 }
549 /*
550 * Can't modify an immutable/append-only file unless
551 * we have appropriate permission.
552 */
561 }
562 }
564 /*
565 * Now we can make the changes. Before we join the inode
566 * to the transaction, if XFS_AT_SIZE is set then take care of
567 * the part of the truncation that must be done without the
568 * inode lock. This needs to be done before joining the inode
569 * to the transaction, because the inode cannot be unlocked
570 * once it is a part of the transaction.
571 */
577 }
580 /*
581 * We are going to log the inode size change in this
582 * transaction so any previous writes that are beyond the on
583 * disk EOF and the new EOF that have not been written out need
584 * to be written here. If we do not write the data out, we
585 * expose ourselves to the null files problem.
586 *
587 * Only flush from the on disk size to the smaller of the in
588 * memory file size or the new size as that's the range we
589 * really care about here and prevents waiting for other data
590 * not within the range we care about here.
591 */
598 }
600 /* wait for all I/O to complete */
610 }
614 XFS_TRANS_PERM_LOG_RES,
615 XFS_ITRUNCATE_LOG_COUNT))) {
620 }
623 }
628 }
630 /* determine whether mandatory locking mode changes */
633 /*
634 * Truncate file. Must have write permission and not be a directory.
635 */
642 /*
643 * signal a sync transaction unless
644 * we're truncating an already unlinked
645 * file on a wsync filesystem
646 */
649 XFS_DATA_FORK,
655 /*
656 * Truncated "down", so we're removing references
657 * to old data here - if we now delay flushing for
658 * a long time, we expose ourselves unduly to the
659 * notorious NULL files problem. So, we mark this
660 * vnode and flush it when the file is closed, and
661 * do not wait the usual (long) time for writeout.
662 */
664 }
665 /*
666 * Have to do this even if the file's size doesn't change.
667 */
669 }
671 /*
672 * Change file access modes.
673 */
680 }
682 /*
683 * Change file ownership. Must be the owner or privileged.
684 * If the system was configured with the "restricted_chown"
685 * option, the owner is not permitted to give away the file,
686 * and can change the group id only to a group of which he
687 * or she is a member.
688 */
690 /*
691 * CAP_FSETID overrides the following restrictions:
692 *
693 * The set-user-ID and set-group-ID bits of a file will be
694 * cleared upon successful return from chown()
695 */
699 }
701 /*
702 * Change the ownerships and register quota modifications
703 * in the transaction.
704 */
711 }
713 }
721 }
723 }
731 }
733 /*
734 * We may have to rev the inode as well as
735 * the superblock version number since projids didn't
736 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
737 */
740 }
744 }
747 /*
748 * Change file access or modified times.
749 */
756 }
762 }
765 }
767 /*
768 * Change XFS-added attributes.
769 */
772 /*
773 * Converting bytes to fs blocks.
774 */
777 }
779 uint di_flags;
781 /* can't set PREALLOC this way, just preserve it */
811 }
813 }
816 }
818 /*
819 * Change file inode change time only if XFS_AT_CTIME set
820 * AND we have been called by a DMI function.
821 */
828 }
830 /*
831 * Send out timestamp changes that need to be set to the
832 * current time. Not done when called by a DMI function.
833 */
839 /*
840 * If this is a synchronous mount, make sure that the
841 * transaction goes to disk before returning to the user.
842 * This is slightly sub-optimal in that truncates require
843 * two sync transactions instead of one for wsync filesystems.
844 * One for the truncate and one for the timestamps since we
845 * don't want to change the timestamps unless we're sure the
846 * truncate worked. Truncates are less than 1% of the laddis
847 * mix so this probably isn't worth the trouble to optimize.
848 */
855 }
857 /*
858 * If the (regular) file's mandatory locking mode changed, then
859 * notify the vnode. We do this under the inode lock to prevent
860 * racing calls to vop_vnode_change.
861 */
866 /*
867 * Release any dquot(s) the inode had kept before chown.
868 */
876 }
883 }
886 abort_return:
888 /* FALLTHROUGH */
889 error_return:
894 }
897 }
899 }
902 /*
903 * xfs_access
904 * Null conversion from vnode mode bits to inode mode bits, as in efs.
905 */
906 int
911 {
919 }
922 /*
923 * The maximum pathlen is 1024 bytes. Since the minimum file system
924 * blocksize is 512 bytes, we can get a max of 2 extents back from
925 * bmapi.
926 */
927 #define SYMLINK_MAPS 2
929 STATIC int
933 {
938 xfs_daddr_t d;
960 }
967 }
972 out:
974 }
976 int
980 {
1004 }
1006 out:
1009 }
1011 /*
1012 * xfs_fsync
1013 *
1014 * This is called to sync the inode and its data out to disk.
1015 * We need to hold the I/O lock while flushing the data, and
1016 * the inode lock while flushing the inode. The inode lock CANNOT
1017 * be held while flushing the data, so acquire after we're done
1018 * with that.
1019 */
1020 int
1024 xfs_off_t start,
1025 xfs_off_t stop)
1026 {
1041 /*
1042 * We always need to make sure that the required inode state
1043 * is safe on disk. The vnode might be clean but because
1044 * of committed transactions that haven't hit the disk yet.
1045 * Likewise, there could be unflushed non-transactional
1046 * changes to the inode core that have to go to disk.
1047 *
1048 * The following code depends on one assumption: that
1049 * any transaction that changes an inode logs the core
1050 * because it has to change some field in the inode core
1051 * (typically nextents or nblocks). That assumption
1052 * implies that any transactions against an inode will
1053 * catch any non-transactional updates. If inode-altering
1054 * transactions exist that violate this assumption, the
1055 * code breaks. Right now, it figures that if the involved
1056 * update_* field is clear and the inode is unpinned, the
1057 * inode is clean. Either it's been flushed or it's been
1058 * committed and the commit has hit the disk unpinning the inode.
1059 * (Note that xfs_inode_item_format() called at commit clears
1060 * the update_* fields.)
1061 */
1064 /* If we are flushing data then we care about update_size
1065 * being set, otherwise we care about update_core
1066 */
1070 /*
1071 * Timestamps/size haven't changed since last inode
1072 * flush or inode transaction commit. That means
1073 * either nothing got written or a transaction
1074 * committed which caught the updates. If the
1075 * latter happened and the transaction hasn't
1076 * hit the disk yet, the inode will be still
1077 * be pinned. If it is, force the log.
1078 */
1084 XFS_LOG_FORCE |
1089 /*
1090 * If the inode is not pinned and nothing
1091 * has changed we don't need to flush the
1092 * cache.
1093 */
1095 }
1098 /*
1099 * Kick off a transaction to log the inode
1100 * core to get the updates. Make it
1101 * sync if FSYNC_WAIT is passed in (which
1102 * is done by everybody but specfs). The
1103 * sync transaction will also force the log.
1104 */
1112 }
1115 /*
1116 * Note - it's possible that we might have pushed
1117 * ourselves out of the way during trans_reserve
1118 * which would flush the inode. But there's no
1119 * guarantee that the inode buffer has actually
1120 * gone out yet (it's delwri). Plus the buffer
1121 * could be pinned anyway if it's part of an
1122 * inode in another recent transaction. So we
1123 * play it safe and fire off the transaction anyway.
1124 */
1133 }
1136 /*
1137 * If the log write didn't issue an ordered tag we need
1138 * to flush the disk cache for the data device now.
1139 */
1143 /*
1144 * If this inode is on the RT dev we need to flush that
1145 * cache as well.
1146 */
1149 }
1152 }
1154 /*
1155 * This is called by xfs_inactive to free any blocks beyond eof
1156 * when the link count isn't zero and by xfs_dm_punch_hole() when
1157 * punching a hole to EOF.
1158 */
1159 int
1164 {
1167 xfs_fileoff_t end_fsb;
1168 xfs_fileoff_t last_fsb;
1169 xfs_filblks_t map_len;
1171 xfs_bmbt_irec_t imap;
1174 /*
1175 * Figure out if there are any blocks beyond the end
1176 * of the file. If not, then there is nothing to do.
1177 */
1193 /*
1194 * Attach the dquots to the inode up front.
1195 */
1199 /*
1200 * There are blocks after the end of file.
1201 * Free them up now by truncating the file to
1202 * its current size.
1203 */
1206 /*
1207 * Do the xfs_itruncate_start() call before
1208 * reserving any log space because
1209 * itruncate_start will call into the buffer
1210 * cache and we can't
1211 * do that within a transaction.
1212 */
1222 }
1227 XFS_ITRUNCATE_LOG_COUNT);
1233 }
1237 XFS_IOLOCK_EXCL |
1238 XFS_ILOCK_EXCL);
1243 XFS_DATA_FORK,
1245 /*
1246 * If we get an error at this point we
1247 * simply don't bother truncating the file.
1248 */
1252 XFS_TRANS_ABORT));
1255 XFS_TRANS_RELEASE_LOG_RES);
1256 }
1259 }
1261 }
1263 /*
1264 * Free a symlink that has blocks associated with it.
1265 */
1266 STATIC int
1270 {
1275 xfs_fsblock_t first_block;
1276 xfs_bmap_free_t free_list;
1288 /*
1289 * We're freeing a symlink that has some
1290 * blocks allocated to it. Free the
1291 * blocks here. We know that we've got
1292 * either 1 or 2 extents and that we can
1293 * free them all in one bunmapi call.
1294 */
1302 }
1303 /*
1304 * Lock the inode, fix the size, and join it to the transaction.
1305 * Hold it so in the normal path, we still have it locked for
1306 * the second transaction. In the error paths we need it
1307 * held so the cancel won't rele it, see below.
1308 */
1315 /*
1316 * Find the block(s) so we can inval and unmap them.
1317 */
1325 /*
1326 * Invalidate the block(s).
1327 */
1333 }
1334 /*
1335 * Unmap the dead block(s) to the free_list.
1336 */
1341 /*
1342 * Commit the first transaction. This logs the EFI and the inode.
1343 */
1346 /*
1347 * The transaction must have been committed, since there were
1348 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1349 * The new tp has the extent freeing and EFDs.
1350 */
1352 /*
1353 * The first xact was committed, so add the inode to the new one.
1354 * Mark it dirty so it will be logged and moved forward in the log as
1355 * part of every commit.
1356 */
1360 /*
1361 * Get a new, empty transaction to return to our caller.
1362 */
1364 /*
1365 * Commit the transaction containing extent freeing and EFDs.
1366 * If we get an error on the commit here or on the reserve below,
1367 * we need to unlock the inode since the new transaction doesn't
1368 * have the inode attached.
1369 */
1375 }
1376 /*
1377 * Remove the memory for extent descriptions (just bookkeeping).
1378 */
1382 /*
1383 * Put an itruncate log reservation in the new transaction
1384 * for our caller.
1385 */
1390 }
1391 /*
1392 * Return with the inode locked but not joined to the transaction.
1393 */
1397 error1:
1399 error0:
1400 /*
1401 * Have to come here with the inode locked and either
1402 * (held and in the transaction) or (not in the transaction).
1403 * If the inode isn't held then cancel would iput it, but
1404 * that's wrong since this is inactive and the vnode ref
1405 * count is 0 already.
1406 * Cancel won't do anything to the inode if held, but it still
1407 * needs to be locked until the cancel is done, if it was
1408 * joined to the transaction.
1409 */
1415 }
1417 STATIC int
1421 {
1425 /*
1426 * We're freeing a symlink which fit into
1427 * the inode. Just free the memory used
1428 * to hold the old symlink.
1429 */
1433 XFS_ITRUNCATE_LOG_COUNT);
1439 }
1442 /*
1443 * Zero length symlinks _can_ exist.
1444 */
1448 XFS_DATA_FORK);
1450 }
1452 }
1454 STATIC int
1458 {
1475 }
1481 XFS_INACTIVE_LOG_COUNT);
1488 }
1499 }
1501 int
1504 {
1512 /* If this is a read-only mount, don't do this (would generate I/O) */
1519 /*
1520 * If we are using filestreams, and we have an unlinked
1521 * file that we are processing the last close on, then nothing
1522 * will be able to reopen and write to this file. Purge this
1523 * inode from the filestreams cache so that it doesn't delay
1524 * teardown of the inode.
1525 */
1529 /*
1530 * If we previously truncated this file and removed old data
1531 * in the process, we want to initiate "early" writeout on
1532 * the last close. This is an attempt to combat the notorious
1533 * NULL files problem which is particularly noticable from a
1534 * truncate down, buffered (re-)write (delalloc), followed by
1535 * a crash. What we are effectively doing here is
1536 * significantly reducing the time window where we'd otherwise
1537 * be exposed to that problem.
1538 */
1542 }
1544 #ifdef HAVE_REFCACHE
1545 /* If we are in the NFS reference cache then don't do this now */
1548 #endif
1560 /* Update linux inode block count after free above */
1563 }
1564 }
1567 }
1569 /*
1570 * xfs_inactive
1571 *
1572 * This is called when the vnode reference count for the vnode
1573 * goes to zero. If the file has been unlinked, then it must
1574 * now be truncated. Also, we clear all of the read-ahead state
1575 * kept for the inode here since the file is now closed.
1576 */
1577 int
1580 {
1582 xfs_bmap_free_t free_list;
1583 xfs_fsblock_t first_block;
1592 /*
1593 * If the inode is already free, then there can be nothing
1594 * to clean up here.
1595 */
1600 }
1602 /*
1603 * Only do a truncate if it's a regular file with
1604 * some actual space in it. It's OK to look at the
1605 * inode's fields without the lock because we're the
1606 * only one with a reference to the inode.
1607 */
1617 }
1621 /* If this is a read-only mount, don't do this (would generate I/O) */
1636 /* Update linux inode block count after free above */
1639 }
1641 }
1650 /*
1651 * Do the xfs_itruncate_start() call before
1652 * reserving any log space because itruncate_start
1653 * will call into the buffer cache and we can't
1654 * do that within a transaction.
1655 */
1663 }
1668 XFS_ITRUNCATE_LOG_COUNT);
1670 /* Don't call itruncate_cleanup */
1675 }
1681 /*
1682 * normally, we have to run xfs_itruncate_finish sync.
1683 * But if filesystem is wsync and we're in the inactive
1684 * path, then we know that nlink == 0, and that the
1685 * xaction that made nlink == 0 is permanently committed
1686 * since xfs_remove runs as a synchronous transaction.
1687 */
1696 }
1699 /*
1700 * If we get an error while cleaning up a
1701 * symlink we bail out.
1702 */
1710 }
1718 XFS_INACTIVE_LOG_COUNT);
1723 }
1728 }
1730 /*
1731 * If there are attributes associated with the file
1732 * then blow them away now. The code calls a routine
1733 * that recursively deconstructs the attribute fork.
1734 * We need to just commit the current transaction
1735 * because we can't use it for xfs_attr_inactive().
1736 */
1739 /*
1740 * If we got an error, the transaction is already
1741 * cancelled, and the inode is unlocked. Just get out.
1742 */
1747 }
1749 /*
1750 * Free the inode.
1751 */
1755 /*
1756 * If we fail to free the inode, shut down. The cancel
1757 * might do that, we need to make sure. Otherwise the
1758 * inode might be lost for a long time or forever.
1759 */
1765 }
1768 /*
1769 * Credit the quota account(s). The inode is gone.
1770 */
1773 /*
1774 * Just ignore errors at this point. There is
1775 * nothing we can do except to try to keep going.
1776 */
1779 }
1780 /*
1781 * Release the dquots held by inode, if any.
1782 */
1787 out:
1789 }
1792 int
1797 {
1799 xfs_ino_t e_inum;
1801 uint lock_mode;
1813 }
1816 }
1818 int
1822 mode_t mode,
1823 xfs_dev_t rdev,
1826 {
1834 xfs_bmap_free_t free_list;
1835 xfs_fsblock_t first_block;
1838 uint cancel_flags;
1840 xfs_prid_t prid;
1842 uint resblks;
1859 }
1864 /* Return through std_return after this point. */
1869 else
1872 /*
1873 * Make sure that we have allocated dquot(s) on disk.
1874 */
1886 /*
1887 * Initially assume that the file does not exist and
1888 * reserve the resources for that case. If that is not
1889 * the case we'll drop the one we have and get a more
1890 * appropriate transaction later.
1891 */
1898 }
1902 }
1911 /*
1912 * Reserve disk quota and the inode.
1913 */
1927 }
1930 /*
1931 * At this point, we've gotten a newly allocated inode.
1932 * It is locked (and joined to the transaction).
1933 */
1937 /*
1938 * Now we join the directory inode to the transaction. We do not do it
1939 * earlier because xfs_dir_ialloc might commit the previous transaction
1940 * (and release all the locks). An error from here on will result in
1941 * the transaction cancel unlocking dp so don't do it explicitly in the
1942 * error path.
1943 */
1954 }
1958 /*
1959 * If this is a synchronous mount, make sure that the
1960 * create transaction goes to disk before returning to
1961 * the user.
1962 */
1965 }
1969 /*
1970 * Attach the dquot(s) to the inodes and modify them incore.
1971 * These ids of the inode couldn't have changed since the new
1972 * inode has been locked ever since it was created.
1973 */
1976 /*
1977 * xfs_trans_commit normally decrements the vnode ref count
1978 * when it unlocks the inode. Since we want to return the
1979 * vnode to the caller, we bump the vnode ref count now.
1980 */
1988 }
1995 }
2002 /* Fallthrough to std_return with error = 0 */
2004 std_return:
2012 }
2015 abort_return:
2017 /* FALLTHROUGH */
2019 error_return:
2031 abort_rele:
2032 /*
2033 * Wait until after the current transaction is aborted to
2034 * release the inode. This prevents recursive transactions
2035 * and deadlocks from xfs_inactive.
2036 */
2045 }
2047 #ifdef DEBUG
2048 /*
2049 * Some counters to see if (and how often) we are hitting some deadlock
2050 * prevention code paths.
2051 */
2056 #endif
2058 /*
2059 * The following routine will lock the inodes associated with the
2060 * directory and the named entry in the directory. The locks are
2061 * acquired in increasing inode number.
2062 *
2063 * If the entry is "..", then only the directory is locked. The
2064 * vnode ref count will still include that from the .. entry in
2065 * this case.
2066 *
2067 * There is a deadlock we need to worry about. If the locked directory is
2068 * in the AIL, it might be blocking up the log. The next inode we lock
2069 * could be already locked by another thread waiting for log space (e.g
2070 * a permanent log reservation with a long running transaction (see
2071 * xfs_itruncate_finish)). To solve this, we must check if the directory
2072 * is in the ail and use lock_nowait. If we can't lock, we need to
2073 * drop the inode lock on the directory and try again. xfs_iunlock will
2074 * potentially push the tail if we were holding up the log.
2075 */
2076 STATIC int
2080 {
2082 xfs_ino_t e_inum;
2086 #ifdef DEBUG
2087 xfs_rm_locks++;
2088 #endif
2091 again:
2098 /*
2099 * We want to lock in increasing inum. Since we've already
2100 * acquired the lock on the directory, we may need to release
2101 * if if the inum of the entry turns out to be less.
2102 */
2104 /*
2105 * We are already in the right order, so just
2106 * lock on the inode of the entry.
2107 * We need to use nowait if dp is in the AIL.
2108 */
2113 attempts++;
2114 #ifdef DEBUG
2115 xfs_rm_attempts++;
2116 #endif
2118 /*
2119 * Unlock dp and try again.
2120 * xfs_iunlock will try to push the tail
2121 * if the inode is in the AIL.
2122 */
2128 #ifdef DEBUG
2129 xfs_rm_lock_delays++;
2130 #endif
2131 }
2133 }
2136 }
2143 }
2144 /* else e_inum == dp->i_ino */
2145 /* This can happen if we're asked to lock /x/..
2146 * the entry is "..", which is also the parent directory.
2147 */
2150 }
2152 #ifdef DEBUG
2158 #endif
2160 /*
2161 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2162 * a different value
2163 */
2166 {
2173 }
2175 /*
2176 * The following routine will lock n inodes in exclusive mode.
2177 * We assume the caller calls us with the inodes in i_ino order.
2178 *
2179 * We need to detect deadlock where an inode that we lock
2180 * is in the AIL and we start waiting for another inode that is locked
2181 * by a thread in a long running transaction (such as truncate). This can
2182 * result in deadlock since the long running trans might need to wait
2183 * for the inode we just locked in order to push the tail and free space
2184 * in the log.
2185 */
2186 void
2191 uint lock_mode)
2192 {
2204 }
2206 again:
2213 /*
2214 * If try_lock is not set yet, make sure all locked inodes
2215 * are not in the AIL.
2216 * If any are, set try_lock to be used later.
2217 */
2223 try_lock++;
2224 }
2225 }
2226 }
2228 /*
2229 * If any of the previous locks we have locked is in the AIL,
2230 * we must TRY to get the second and subsequent locks. If
2231 * we can't get any, we must release all we have
2232 * and try again.
2233 */
2236 /* try_lock must be 0 if i is 0. */
2237 /*
2238 * try_lock means we have an inode locked
2239 * that is in the AIL.
2240 */
2243 attempts++;
2245 /*
2246 * Unlock all previous guys and try again.
2247 * xfs_iunlock will try to push the tail
2248 * if the inode is in the AIL.
2249 */
2253 /*
2254 * Check to see if we've already
2255 * unlocked this one.
2256 * Not the first one going back,
2257 * and the inode ptr is the same.
2258 */
2264 }
2268 #ifdef DEBUG
2269 xfs_lock_delays++;
2270 #endif
2271 }
2275 }
2278 }
2279 }
2281 #ifdef DEBUG
2287 xfs_locked_n++;
2288 }
2289 #endif
2290 }
2292 #ifdef DEBUG
2293 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2296 #define REMOVE_DEBUG_TRACE(x)
2299 int
2303 {
2310 xfs_bmap_free_t free_list;
2311 xfs_fsblock_t first_block;
2316 uint resblks;
2329 }
2337 }
2339 /* From this point on, return through std_return */
2342 /*
2343 * We need to get a reference to ip before we get our log
2344 * reservation. The reason for this is that we cannot call
2345 * xfs_iget for an inode for which we do not have a reference
2346 * once we've acquired a log reservation. This is because the
2347 * inode we are trying to get might be in xfs_inactive going
2348 * for a log reservation. Since we'll have to wait for the
2349 * inactive code to complete before returning from xfs_iget,
2350 * we need to make sure that we don't have log space reserved
2351 * when we call xfs_iget. Instead we get an unlocked reference
2352 * to the inode before getting our log reservation.
2353 */
2358 }
2372 }
2376 /*
2377 * We try to get the real space reservation first,
2378 * allowing for directory btree deletion(s) implying
2379 * possible bmap insert(s). If we can't get the space
2380 * reservation then we use 0 instead, and avoid the bmap
2381 * btree insert(s) in the directory code by, if the bmap
2382 * insert tries to happen, instead trimming the LAST
2383 * block from the directory.
2384 */
2392 }
2399 }
2407 }
2409 /*
2410 * At this point, we've gotten both the directory and the entry
2411 * inodes locked.
2412 */
2415 /*
2416 * Increment vnode ref count only in this case since
2417 * there's an extra vnode reference in the case where
2418 * dp == ip.
2419 */
2422 }
2424 /*
2425 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2426 */
2434 }
2444 }
2446 /* Determine if this is the last link while
2447 * we are in the transaction.
2448 */
2451 /*
2452 * Take an extra ref on the inode so that it doesn't
2453 * go to xfs_inactive() from within the commit.
2454 */
2457 /*
2458 * If this is a synchronous mount, make sure that the
2459 * remove transaction goes to disk before returning to
2460 * the user.
2461 */
2464 }
2470 }
2476 }
2478 /*
2479 * Before we drop our extra reference to the inode, purge it
2480 * from the refcache if it is there. By waiting until afterwards
2481 * to do the IRELE, we ensure that we won't go inactive in the
2482 * xfs_refcache_purge_ip routine (although that would be OK).
2483 */
2486 /*
2487 * If we are using filestreams, kill the stream association.
2488 * If the file is still open it may get a new one but that
2489 * will get killed on last close in xfs_close() so we don't
2490 * have to worry about that.
2491 */
2498 /* Fall through to std_return with error = 0 */
2499 std_return:
2505 }
2508 error1:
2514 error_rele:
2515 /*
2516 * In this case make sure to not release the inode until after
2517 * the current transaction is aborted. Releasing it beforehand
2518 * can cause us to go to xfs_inactive and start a recursive
2519 * transaction which can easily deadlock with the current one.
2520 */
2525 /*
2526 * Before we drop our extra reference to the inode, purge it
2527 * from the refcache if it is there. By waiting until afterwards
2528 * to do the IRELE, we ensure that we won't go inactive in the
2529 * xfs_refcache_purge_ip routine (although that would be OK).
2530 */
2536 }
2538 int
2543 {
2550 xfs_bmap_free_t free_list;
2551 xfs_fsblock_t first_block;
2574 }
2576 /* Return through std_return after this point. */
2593 }
2597 }
2605 }
2609 /*
2610 * Increment vnode ref counts since xfs_trans_commit &
2611 * xfs_trans_cancel will both unlock the inodes and
2612 * decrement the associated ref counts.
2613 */
2619 /*
2620 * If the source has too many links, we can't make any more to it.
2621 */
2625 }
2627 /*
2628 * If we are using project inheritance, we only allow hard link
2629 * creation in our tree when the project IDs are the same; else
2630 * the tree quota mechanism could be circumvented.
2631 */
2636 }
2646 resblks);
2657 /*
2658 * If this is a synchronous mount, make sure that the
2659 * link transaction goes to disk before returning to
2660 * the user.
2661 */
2664 }
2670 }
2676 /* Fall through to std_return with error = 0. */
2677 std_return:
2683 }
2686 abort_return:
2688 /* FALLTHROUGH */
2690 error_return:
2693 }
2696 int
2700 mode_t mode,
2703 {
2714 xfs_bmap_free_t free_list;
2715 xfs_fsblock_t first_block;
2719 xfs_prid_t prid;
2721 uint resblks;
2736 }
2738 /* Return through std_return after this point. */
2746 else
2749 /*
2750 * Make sure that we have allocated dquot(s) on disk.
2751 */
2766 XFS_TRANS_PERM_LOG_RES,
2767 XFS_MKDIR_LOG_COUNT);
2768 }
2772 }
2777 /*
2778 * Check for directory link count overflow.
2779 */
2783 }
2785 /*
2786 * Reserve disk quota and the inode.
2787 */
2795 /*
2796 * create the directory inode.
2797 */
2805 }
2808 /*
2809 * Now we add the directory inode to the transaction.
2810 * We waited until now since xfs_dir_ialloc might start
2811 * a new transaction. Had we joined the transaction
2812 * earlier, the locks might have gotten released. An error
2813 * from here on will result in the transaction cancel
2814 * unlocking dp so don't do it explicitly in the error path.
2815 */
2828 }
2831 /*
2832 * Bump the in memory version number of the parent directory
2833 * so that other processes accessing it will recognize that
2834 * the directory has changed.
2835 */
2854 /*
2855 * Attach the dquots to the new inode and modify the icount incore.
2856 */
2859 /*
2860 * If this is a synchronous mount, make sure that the
2861 * mkdir transaction goes to disk before returning to
2862 * the user.
2863 */
2866 }
2872 }
2879 }
2881 /* Fall through to std_return with error = 0 or errno from
2882 * xfs_trans_commit. */
2884 std_return:
2890 DM_RIGHT_NULL,
2893 }
2896 error2:
2897 error1:
2899 abort_return:
2901 error_return:
2910 }
2912 int
2916 {
2924 xfs_bmap_free_t free_list;
2925 xfs_fsblock_t first_block;
2930 uint resblks;
2940 }
2949 }
2951 /* Return through std_return after this point. */
2955 /*
2956 * We need to get a reference to cdp before we get our log
2957 * reservation. The reason for this is that we cannot call
2958 * xfs_iget for an inode for which we do not have a reference
2959 * once we've acquired a log reservation. This is because the
2960 * inode we are trying to get might be in xfs_inactive going
2961 * for a log reservation. Since we'll have to wait for the
2962 * inactive code to complete before returning from xfs_iget,
2963 * we need to make sure that we don't have log space reserved
2964 * when we call xfs_iget. Instead we get an unlocked reference
2965 * to the inode before getting our log reservation.
2966 */
2971 }
2975 /*
2976 * Get the dquots for the inodes.
2977 */
2985 }
2989 /*
2990 * We try to get the real space reservation first,
2991 * allowing for directory btree deletion(s) implying
2992 * possible bmap insert(s). If we can't get the space
2993 * reservation then we use 0 instead, and avoid the bmap
2994 * btree insert(s) in the directory code by, if the bmap
2995 * insert tries to happen, instead trimming the LAST
2996 * block from the directory.
2997 */
3005 }
3011 }
3014 /*
3015 * Now lock the child directory inode and the parent directory
3016 * inode in the proper order. This will take care of validating
3017 * that the directory entry for the child directory inode has
3018 * not changed while we were obtaining a log reservation.
3019 */
3025 }
3029 /*
3030 * Only increment the parent directory vnode count if
3031 * we didn't bump it in looking up cdp. The only time
3032 * we don't bump it is when we're looking up ".".
3033 */
3035 }
3044 }
3048 }
3057 /*
3058 * Bump the in memory generation count on the parent
3059 * directory so that other can know that it has changed.
3060 */
3063 /*
3064 * Drop the link from cdp's "..".
3065 */
3069 }
3071 /*
3072 * Drop the link from dp to cdp.
3073 */
3077 }
3079 /*
3080 * Drop the "." link from cdp to self.
3081 */
3085 }
3087 /* Determine these before committing transaction */
3090 /*
3091 * Take an extra ref on the child vnode so that it
3092 * does not go to xfs_inactive() from within the commit.
3093 */
3096 /*
3097 * If this is a synchronous mount, make sure that the
3098 * rmdir transaction goes to disk before returning to
3099 * the user.
3100 */
3103 }
3109 XFS_TRANS_ABORT));
3112 }
3118 }
3123 /* Fall through to std_return with error = 0 or the errno
3124 * from xfs_trans_commit. */
3125 std_return:
3132 }
3135 error1:
3138 /* FALLTHROUGH */
3140 error_return:
3143 }
3145 int
3150 mode_t mode,
3153 {
3160 xfs_bmap_free_t free_list;
3161 xfs_fsblock_t first_block;
3163 uint cancel_flags;
3165 xfs_fileoff_t first_fsb;
3166 xfs_filblks_t fs_blocks;
3169 xfs_daddr_t d;
3174 xfs_prid_t prid;
3176 uint resblks;
3192 /*
3193 * Check component lengths of the target path name.
3194 */
3203 /*
3204 * Skip any slashes.
3205 */
3207 total++;
3208 path++;
3209 }
3211 /*
3212 * Count up to the next slash or end of path.
3213 * Error out if the component is bigger than MAXNAMELEN.
3214 */
3219 }
3220 }
3221 }
3222 }
3230 }
3232 /* Return through std_return after this point. */
3237 else
3240 /*
3241 * Make sure that we have allocated dquot(s) on disk.
3242 */
3251 /*
3252 * The symlink will fit into the inode data fork?
3253 * There can't be any attributes so we get the whole variable part.
3254 */
3257 else
3266 }
3270 }
3275 /*
3276 * Check whether the directory allows new symlinks or not.
3277 */
3281 }
3283 /*
3284 * Reserve disk quota : blocks and inode.
3285 */
3290 /*
3291 * Check for ability to enter directory entry, if no space reserved.
3292 */
3296 /*
3297 * Initialize the bmap freelist prior to calling either
3298 * bmapi or the directory create code.
3299 */
3302 /*
3303 * Allocate an inode for the symlink.
3304 */
3311 }
3314 /*
3315 * An error after we've joined dp to the transaction will result in the
3316 * transaction cancel unlocking dp so don't do it explicitly in the
3317 * error path.
3318 */
3323 /*
3324 * Also attach the dquot(s) to it, if applicable.
3325 */
3330 /*
3331 * If the symlink will fit into the inode, write it inline.
3332 */
3338 /*
3339 * The inode was initially created in extent format.
3340 */
3357 }
3373 }
3380 }
3381 }
3383 /*
3384 * Create the directory entry for the symlink.
3385 */
3393 /*
3394 * Bump the in memory version number of the parent directory
3395 * so that other processes accessing it will recognize that
3396 * the directory has changed.
3397 */
3400 /*
3401 * If this is a synchronous mount, make sure that the
3402 * symlink transaction goes to disk before returning to
3403 * the user.
3404 */
3407 }
3409 /*
3410 * xfs_trans_commit normally decrements the vnode ref count
3411 * when it unlocks the inode. Since we want to return the
3412 * vnode to the caller, we bump the vnode ref count now.
3413 */
3419 }
3424 /* Fall through to std_return with error = 0 or errno from
3425 * xfs_trans_commit */
3426 std_return:
3433 }
3441 }
3444 error2:
3446 error1:
3449 error_return:
3458 }
3460 int
3463 bhv_vrwlock_t locktype)
3464 {
3477 }
3480 }
3483 void
3486 bhv_vrwlock_t locktype)
3487 {
3491 /*
3492 * In the write case, we may have added a new entry to
3493 * the reference cache. This might store a pointer to
3494 * an inode to be released in this inode. If it is there,
3495 * clear the pointer and release the inode after unlocking
3496 * this one.
3497 */
3503 }
3505 }
3508 int
3512 {
3520 /*
3521 * Bypass inodes which have already been cleaned by
3522 * the inode flush clustering code inside xfs_iflush
3523 */
3531 xfs_lsn_t sync_lsn;
3544 }
3548 }
3549 }
3551 /*
3552 * We make this non-blocking if the inode is contended,
3553 * return EAGAIN to indicate to the caller that they
3554 * did not succeed. This prevents the flush path from
3555 * blocking on inodes inside another operation right
3556 * now, they get caught later by xfs_sync.
3557 */
3568 }
3571 }
3575 else
3580 }
3583 }
3586 int
3589 u_int evmask,
3590 u_int16_t state)
3591 {
3607 }
3619 }
3621 int
3624 {
3631 /* bad inode, get out here ASAP */
3635 }
3641 /*
3642 * Make sure the atime in the XFS inode is correct before freeing the
3643 * Linux inode.
3644 */
3647 /*
3648 * If we have nothing to flush with this inode then complete the
3649 * teardown now, otherwise break the link between the xfs inode and the
3650 * linux inode and clean up the xfs inode later. This avoids flushing
3651 * the inode to disk during the delete operation itself.
3652 *
3653 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3654 * first to ensure that xfs_iunpin() will never see an xfs inode
3655 * that has a linux inode being reclaimed. Synchronisation is provided
3656 * by the i_flags_lock.
3657 */
3665 /* Protect sync and unpin from us */
3674 }
3676 }
3678 int
3683 {
3691 /* The hash lock here protects a thread in xfs_iget_core from
3692 * racing with us on linking the inode back with a vnode.
3693 * Once we have the XFS_IRECLAIM flag set it will not touch
3694 * us.
3695 */
3705 }
3707 }
3713 /*
3714 * If the inode is still dirty, then flush it out. If the inode
3715 * is not in the AIL, then it will be OK to flush it delwri as
3716 * long as xfs_iflush() does not keep any references to the inode.
3717 * We leave that decision up to xfs_iflush() since it has the
3718 * knowledge of whether it's OK to simply do a delwri flush of
3719 * the inode or whether we need to wait until the inode is
3720 * pulled from the AIL.
3721 * We get the flush lock regardless, though, just to make sure
3722 * we don't free it while it is being flushed.
3723 */
3728 }
3734 /*
3735 * If we hit an error, typically because of filesystem
3736 * shutdown, we don't need to let vn_reclaim to know
3737 * because we're gonna reclaim the inode anyway.
3738 */
3742 }
3744 }
3751 /*
3752 * We are not interested in doing an iflush if we're
3753 * in the process of shutting down the filesystem forcibly.
3754 * So, just reclaim the inode.
3755 */
3758 }
3760 reclaim:
3763 }
3765 int
3767 {
3783 }
3784 }
3787 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3791 }
3794 }
3798 }
3800 /*
3801 * xfs_alloc_file_space()
3802 * This routine allocates disk space for the given file.
3803 *
3804 * If alloc_type == 0, this request is for an ALLOCSP type
3805 * request which will change the file size. In this case, no
3806 * DMAPI event will be generated by the call. A TRUNCATE event
3807 * will be generated later by xfs_setattr.
3808 *
3809 * If alloc_type != 0, this request is for a RESVSP type
3810 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3811 * lower block boundary byte address is less than the file's
3812 * length.
3813 *
3814 * RETURNS:
3815 * 0 on success
3816 * errno on error
3817 *
3818 */
3819 STATIC int
3822 xfs_off_t offset,
3823 xfs_off_t len,
3826 {
3828 xfs_off_t count;
3829 xfs_filblks_t allocated_fsb;
3830 xfs_filblks_t allocatesize_fsb;
3832 xfs_fileoff_t startoffset_fsb;
3833 xfs_fsblock_t firstfsb;
3840 xfs_bmap_free_t free_list;
3866 /* Generate a DMAPI event if needed. */
3870 xfs_off_t end_dmi_offset;
3880 }
3882 /*
3883 * Allocate file space until done or until there is an error
3884 */
3885 retry:
3889 /*
3890 * Determine space reservations for data/realtime.
3891 */
3904 }
3916 }
3918 /*
3919 * Allocate and setup the transaction.
3920 */
3924 XFS_TRANS_PERM_LOG_RES,
3925 XFS_WRITE_LOG_COUNT);
3926 /*
3927 * Check for running out of space
3928 */
3930 /*
3931 * Free the transaction structure.
3932 */
3936 }
3946 /*
3947 * Issue the xfs_bmapi() call to allocate the blocks
3948 */
3956 }
3958 /*
3959 * Complete the transaction
3960 */
3964 }
3970 }
3977 }
3981 }
3982 dmapi_enospc_check:
3991 /* else fall through with error from XFS_SEND_DATA */
3992 }
4004 }
4006 /*
4007 * Zero file bytes between startoff and endoff inclusive.
4008 * The iolock is held exclusive and no blocks are buffered.
4009 */
4010 STATIC int
4013 xfs_off_t startoff,
4014 xfs_off_t endoff)
4015 {
4016 xfs_bmbt_irec_t imap;
4017 xfs_fileoff_t offset_fsb;
4018 xfs_off_t lastoffset;
4019 xfs_off_t offset;
4055 }
4067 }
4068 }
4071 }
4073 /*
4074 * xfs_free_file_space()
4075 * This routine frees disk space for the given file.
4076 *
4077 * This routine is only called by xfs_change_file_space
4078 * for an UNRESVSP type call.
4079 *
4080 * RETURNS:
4081 * 0 on success
4082 * errno on error
4083 *
4084 */
4085 STATIC int
4088 xfs_off_t offset,
4089 xfs_off_t len,
4091 {
4095 xfs_off_t end_dmi_offset;
4096 xfs_fileoff_t endoffset_fsb;
4098 xfs_fsblock_t firstfsb;
4099 xfs_bmap_free_t free_list;
4100 xfs_bmbt_irec_t imap;
4101 xfs_off_t ioffset;
4105 uint resblks;
4106 uint rounding;
4108 xfs_fileoff_t startoffset_fsb;
4137 }
4144 }
4154 }
4156 /*
4157 * Need to zero the stuff we're not freeing, on disk.
4158 * If its a realtime file & can't use unwritten extents then we
4159 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4160 * will take care of it for us.
4161 */
4170 xfs_daddr_t block;
4177 }
4186 mod++;
4189 }
4190 }
4192 /*
4193 * One contiguous piece to clear
4194 */
4197 /*
4198 * Some full blocks, possibly two pieces to clear
4199 */
4208 }
4210 /*
4211 * free file space until done or until there is an error
4212 */
4216 /*
4217 * allocate and setup the transaction. Allow this
4218 * transaction to dip into the reserve blocks to ensure
4219 * the freeing of the space succeeds at ENOSPC.
4220 */
4224 resblks,
4227 XFS_TRANS_PERM_LOG_RES,
4228 XFS_WRITE_LOG_COUNT);
4230 /*
4231 * check for running out of space
4232 */
4234 /*
4235 * Free the transaction structure.
4236 */
4240 }
4244 XFS_QMOPT_RES_REGBLKS);
4251 /*
4252 * issue the bunmapi() call to free the blocks
4253 */
4260 }
4262 /*
4263 * complete the transaction
4264 */
4268 }
4272 }
4274 out_unlock_iolock:
4279 error0:
4281 error1:
4284 XFS_ILOCK_EXCL);
4286 }
4288 /*
4289 * xfs_change_file_space()
4290 * This routine allocates or frees disk space for the given file.
4291 * The user specified parameters are checked for alignment and size
4292 * limitations.
4293 *
4294 * RETURNS:
4295 * 0 on success
4296 * errno on error
4297 *
4298 */
4299 int
4304 xfs_off_t offset,
4307 {
4311 xfs_fsize_t fsize;
4313 xfs_off_t startoffset;
4314 xfs_off_t llen;
4316 bhv_vattr_t va;
4320 /*
4321 * must be a regular file and have write permission
4322 */
4331 }
4346 }
4361 /*
4362 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4363 * file space.
4364 * These calls do NOT zero the data space allocated to the file,
4365 * nor do they change the file size.
4366 *
4367 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4368 * space.
4369 * These calls cause the new file data to be zeroed and the file
4370 * size to be changed.
4371 */
4387 attr_flags)))
4400 }
4416 }
4418 /*
4419 * update the inode timestamp, mode, and prealloc flag bits
4420 */
4425 /* ASSERT(0); */
4428 }
4438 /*
4439 * Note that we don't have to worry about mandatory
4440 * file locking being disabled here because we only
4441 * clear the S_ISGID bit if the Group execute bit is
4442 * on, but if it was on then mandatory locking wouldn't
4443 * have been enabled.
4444 */
4449 }
4463 }