| blob | 4765e7c4b75d7155cdbfbd4077c79a22eb4674d4 |
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 */
56 int
59 {
65 /*
66 * If it's a directory with any blocks, read-ahead block 0
67 * as we're almost certain to have the next operation be a read there.
68 */
74 }
76 }
78 /*
79 * xfs_getattr
80 */
81 int
86 {
105 #if XFS_BIG_INUMS
107 #endif
110 /*
111 * Quick exit for non-stat callers
112 */
118 /*
119 * Copy from in-core inode.
120 */
126 /*
127 * Check vnode type block/char vs. everything else.
128 */
142 /*
143 * If the file blocks are being allocated from a
144 * realtime partition, then return the inode's
145 * realtime extent size or the realtime volume's
146 * extent size.
147 */
150 }
152 }
160 /*
161 * Exit for stat callers. See if any of the rest of the fields
162 * to be filled in are needed.
163 */
169 /*
170 * Convert di_flags to xflags.
171 */
174 /*
175 * Exit for inode revalidate. See if any of the rest of
176 * the fields to be filled in are needed.
177 */
193 else
197 all_done:
201 }
204 /*
205 * xfs_setattr
206 */
207 int
213 {
219 uint lock_flags;
235 /*
236 * Cannot set certain attributes.
237 */
241 }
246 /*
247 * Timestamps do not need to be logged and hence do not
248 * need to be done within a transaction.
249 */
257 }
262 /*
263 * If disk quotas is on, we make sure that the dquots do exist on disk,
264 * before we start any other transactions. Trying to do this later
265 * is messy. We don't care to take a readlock to look at the ids
266 * in inode here, because we can't hold it across the trans_reserve.
267 * If the IDs do change before we take the ilock, we're covered
268 * because the i_*dquot fields will get updated anyway.
269 */
279 }
285 }
291 }
292 /*
293 * We take a reference when we initialize udqp and gdqp,
294 * so it is important that we never blindly double trip on
295 * the same variable. See xfs_create() for an example.
296 */
303 }
305 /*
306 * For the other attributes, we acquire the inode lock and
307 * first do an error checking pass.
308 */
323 }
324 }
334 }
335 }
338 }
342 /* boolean: are we the file owner? */
345 /*
346 * Change various properties of a file.
347 * Only the owner or users with CAP_FOWNER
348 * capability may do these things.
349 */
353 /*
354 * CAP_FOWNER overrides the following restrictions:
355 *
356 * The user ID of the calling process must be equal
357 * to the file owner ID, except in cases where the
358 * CAP_FSETID capability is applicable.
359 */
363 }
365 /*
366 * CAP_FSETID overrides the following restrictions:
367 *
368 * The effective user ID of the calling process shall match
369 * the file owner when setting the set-user-ID and
370 * set-group-ID bits on that file.
371 *
372 * The effective group ID or one of the supplementary group
373 * IDs of the calling process shall match the group owner of
374 * the file when setting the set-group-ID bit on that file
375 */
384 #if 0
385 /* Linux allows this, Irix doesn't. */
388 #endif
391 }
392 }
394 /*
395 * Change file ownership. Must be the owner or privileged.
396 * If the system was configured with the "restricted_chown"
397 * option, the owner is not permitted to give away the file,
398 * and can change the group id only to a group of which he
399 * or she is a member.
400 */
402 /*
403 * These IDs could have changed since we last looked at them.
404 * But, we're assured that if the ownership did change
405 * while we didn't have the inode locked, inode's dquot(s)
406 * would have changed also.
407 */
414 iprojid;
416 /*
417 * CAP_CHOWN overrides the following restrictions:
418 *
419 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
420 * shall override the restriction that a process cannot
421 * change the user ID of a file it owns and the restriction
422 * that the group ID supplied to the chown() function
423 * shall be equal to either the group ID or one of the
424 * supplementary group IDs of the calling process.
425 */
432 }
433 /*
434 * Do a quota reservation only if uid/projid/gid is actually
435 * going to change.
436 */
446 }
447 }
449 /*
450 * Truncate file. Must have write permission and not be a directory.
451 */
453 /* Short circuit the truncate case for zero length files */
462 }
470 }
471 /*
472 * Make sure that the dquots are attached to the inode.
473 */
476 }
478 /*
479 * Change file access or modified times.
480 */
487 }
488 }
489 }
491 /*
492 * Change extent size or realtime flag.
493 */
495 /*
496 * Can't change extent size if any extents are allocated.
497 */
503 }
505 /*
506 * Can't change realtime flag if any extents are allocated.
507 */
514 }
515 /*
516 * Extent size must be a multiple of the appropriate block
517 * size, if set at all.
518 */
520 xfs_extlen_t size;
529 }
533 }
534 }
535 /*
536 * If realtime flag is set then must have realtime data.
537 */
545 }
546 }
548 /*
549 * Can't modify an immutable/append-only file unless
550 * we have appropriate permission.
551 */
560 }
561 }
563 /*
564 * Now we can make the changes. Before we join the inode
565 * to the transaction, if XFS_AT_SIZE is set then take care of
566 * the part of the truncation that must be done without the
567 * inode lock. This needs to be done before joining the inode
568 * to the transaction, because the inode cannot be unlocked
569 * once it is a part of the transaction.
570 */
576 }
579 /*
580 * We are going to log the inode size change in this
581 * transaction so any previous writes that are beyond the on
582 * disk EOF and the new EOF that have not been written out need
583 * to be written here. If we do not write the data out, we
584 * expose ourselves to the null files problem.
585 *
586 * Only flush from the on disk size to the smaller of the in
587 * memory file size or the new size as that's the range we
588 * really care about here and prevents waiting for other data
589 * not within the range we care about here.
590 */
597 }
599 /* wait for all I/O to complete */
609 }
613 XFS_TRANS_PERM_LOG_RES,
614 XFS_ITRUNCATE_LOG_COUNT))) {
619 }
622 }
627 }
629 /* determine whether mandatory locking mode changes */
632 /*
633 * Truncate file. Must have write permission and not be a directory.
634 */
641 /*
642 * signal a sync transaction unless
643 * we're truncating an already unlinked
644 * file on a wsync filesystem
645 */
648 XFS_DATA_FORK,
654 /*
655 * Truncated "down", so we're removing references
656 * to old data here - if we now delay flushing for
657 * a long time, we expose ourselves unduly to the
658 * notorious NULL files problem. So, we mark this
659 * vnode and flush it when the file is closed, and
660 * do not wait the usual (long) time for writeout.
661 */
663 }
664 /*
665 * Have to do this even if the file's size doesn't change.
666 */
668 }
670 /*
671 * Change file access modes.
672 */
679 }
681 /*
682 * Change file ownership. Must be the owner or privileged.
683 * If the system was configured with the "restricted_chown"
684 * option, the owner is not permitted to give away the file,
685 * and can change the group id only to a group of which he
686 * or she is a member.
687 */
689 /*
690 * CAP_FSETID overrides the following restrictions:
691 *
692 * The set-user-ID and set-group-ID bits of a file will be
693 * cleared upon successful return from chown()
694 */
698 }
700 /*
701 * Change the ownerships and register quota modifications
702 * in the transaction.
703 */
710 }
712 }
720 }
722 }
730 }
732 /*
733 * We may have to rev the inode as well as
734 * the superblock version number since projids didn't
735 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
736 */
739 }
743 }
746 /*
747 * Change file access or modified times.
748 */
755 }
761 }
764 }
766 /*
767 * Change XFS-added attributes.
768 */
771 /*
772 * Converting bytes to fs blocks.
773 */
776 }
778 uint di_flags;
780 /* can't set PREALLOC this way, just preserve it */
810 }
812 }
815 }
817 /*
818 * Change file inode change time only if XFS_AT_CTIME set
819 * AND we have been called by a DMI function.
820 */
827 }
829 /*
830 * Send out timestamp changes that need to be set to the
831 * current time. Not done when called by a DMI function.
832 */
838 /*
839 * If this is a synchronous mount, make sure that the
840 * transaction goes to disk before returning to the user.
841 * This is slightly sub-optimal in that truncates require
842 * two sync transactions instead of one for wsync filesystems.
843 * One for the truncate and one for the timestamps since we
844 * don't want to change the timestamps unless we're sure the
845 * truncate worked. Truncates are less than 1% of the laddis
846 * mix so this probably isn't worth the trouble to optimize.
847 */
854 }
856 /*
857 * If the (regular) file's mandatory locking mode changed, then
858 * notify the vnode. We do this under the inode lock to prevent
859 * racing calls to vop_vnode_change.
860 */
865 /*
866 * Release any dquot(s) the inode had kept before chown.
867 */
875 }
882 }
885 abort_return:
887 /* FALLTHROUGH */
888 error_return:
893 }
896 }
898 }
900 /*
901 * The maximum pathlen is 1024 bytes. Since the minimum file system
902 * blocksize is 512 bytes, we can get a max of 2 extents back from
903 * bmapi.
904 */
905 #define SYMLINK_MAPS 2
907 STATIC int
911 {
916 xfs_daddr_t d;
938 }
945 }
950 out:
952 }
954 int
958 {
982 }
984 out:
987 }
989 /*
990 * xfs_fsync
991 *
992 * This is called to sync the inode and its data out to disk.
993 * We need to hold the I/O lock while flushing the data, and
994 * the inode lock while flushing the inode. The inode lock CANNOT
995 * be held while flushing the data, so acquire after we're done
996 * with that.
997 */
998 int
1002 xfs_off_t start,
1003 xfs_off_t stop)
1004 {
1019 /*
1020 * We always need to make sure that the required inode state
1021 * is safe on disk. The vnode might be clean but because
1022 * of committed transactions that haven't hit the disk yet.
1023 * Likewise, there could be unflushed non-transactional
1024 * changes to the inode core that have to go to disk.
1025 *
1026 * The following code depends on one assumption: that
1027 * any transaction that changes an inode logs the core
1028 * because it has to change some field in the inode core
1029 * (typically nextents or nblocks). That assumption
1030 * implies that any transactions against an inode will
1031 * catch any non-transactional updates. If inode-altering
1032 * transactions exist that violate this assumption, the
1033 * code breaks. Right now, it figures that if the involved
1034 * update_* field is clear and the inode is unpinned, the
1035 * inode is clean. Either it's been flushed or it's been
1036 * committed and the commit has hit the disk unpinning the inode.
1037 * (Note that xfs_inode_item_format() called at commit clears
1038 * the update_* fields.)
1039 */
1042 /* If we are flushing data then we care about update_size
1043 * being set, otherwise we care about update_core
1044 */
1048 /*
1049 * Timestamps/size haven't changed since last inode
1050 * flush or inode transaction commit. That means
1051 * either nothing got written or a transaction
1052 * committed which caught the updates. If the
1053 * latter happened and the transaction hasn't
1054 * hit the disk yet, the inode will be still
1055 * be pinned. If it is, force the log.
1056 */
1062 XFS_LOG_FORCE |
1067 /*
1068 * If the inode is not pinned and nothing
1069 * has changed we don't need to flush the
1070 * cache.
1071 */
1073 }
1076 /*
1077 * Kick off a transaction to log the inode
1078 * core to get the updates. Make it
1079 * sync if FSYNC_WAIT is passed in (which
1080 * is done by everybody but specfs). The
1081 * sync transaction will also force the log.
1082 */
1090 }
1093 /*
1094 * Note - it's possible that we might have pushed
1095 * ourselves out of the way during trans_reserve
1096 * which would flush the inode. But there's no
1097 * guarantee that the inode buffer has actually
1098 * gone out yet (it's delwri). Plus the buffer
1099 * could be pinned anyway if it's part of an
1100 * inode in another recent transaction. So we
1101 * play it safe and fire off the transaction anyway.
1102 */
1111 }
1114 /*
1115 * If the log write didn't issue an ordered tag we need
1116 * to flush the disk cache for the data device now.
1117 */
1121 /*
1122 * If this inode is on the RT dev we need to flush that
1123 * cache as well.
1124 */
1127 }
1130 }
1132 /*
1133 * This is called by xfs_inactive to free any blocks beyond eof
1134 * when the link count isn't zero and by xfs_dm_punch_hole() when
1135 * punching a hole to EOF.
1136 */
1137 int
1142 {
1145 xfs_fileoff_t end_fsb;
1146 xfs_fileoff_t last_fsb;
1147 xfs_filblks_t map_len;
1149 xfs_bmbt_irec_t imap;
1152 /*
1153 * Figure out if there are any blocks beyond the end
1154 * of the file. If not, then there is nothing to do.
1155 */
1171 /*
1172 * Attach the dquots to the inode up front.
1173 */
1177 /*
1178 * There are blocks after the end of file.
1179 * Free them up now by truncating the file to
1180 * its current size.
1181 */
1184 /*
1185 * Do the xfs_itruncate_start() call before
1186 * reserving any log space because
1187 * itruncate_start will call into the buffer
1188 * cache and we can't
1189 * do that within a transaction.
1190 */
1200 }
1205 XFS_ITRUNCATE_LOG_COUNT);
1211 }
1215 XFS_IOLOCK_EXCL |
1216 XFS_ILOCK_EXCL);
1221 XFS_DATA_FORK,
1223 /*
1224 * If we get an error at this point we
1225 * simply don't bother truncating the file.
1226 */
1230 XFS_TRANS_ABORT));
1233 XFS_TRANS_RELEASE_LOG_RES);
1234 }
1237 }
1239 }
1241 /*
1242 * Free a symlink that has blocks associated with it.
1243 */
1244 STATIC int
1248 {
1253 xfs_fsblock_t first_block;
1254 xfs_bmap_free_t free_list;
1266 /*
1267 * We're freeing a symlink that has some
1268 * blocks allocated to it. Free the
1269 * blocks here. We know that we've got
1270 * either 1 or 2 extents and that we can
1271 * free them all in one bunmapi call.
1272 */
1280 }
1281 /*
1282 * Lock the inode, fix the size, and join it to the transaction.
1283 * Hold it so in the normal path, we still have it locked for
1284 * the second transaction. In the error paths we need it
1285 * held so the cancel won't rele it, see below.
1286 */
1293 /*
1294 * Find the block(s) so we can inval and unmap them.
1295 */
1303 /*
1304 * Invalidate the block(s).
1305 */
1311 }
1312 /*
1313 * Unmap the dead block(s) to the free_list.
1314 */
1319 /*
1320 * Commit the first transaction. This logs the EFI and the inode.
1321 */
1324 /*
1325 * The transaction must have been committed, since there were
1326 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1327 * The new tp has the extent freeing and EFDs.
1328 */
1330 /*
1331 * The first xact was committed, so add the inode to the new one.
1332 * Mark it dirty so it will be logged and moved forward in the log as
1333 * part of every commit.
1334 */
1338 /*
1339 * Get a new, empty transaction to return to our caller.
1340 */
1342 /*
1343 * Commit the transaction containing extent freeing and EFDs.
1344 * If we get an error on the commit here or on the reserve below,
1345 * we need to unlock the inode since the new transaction doesn't
1346 * have the inode attached.
1347 */
1353 }
1354 /*
1355 * Remove the memory for extent descriptions (just bookkeeping).
1356 */
1360 /*
1361 * Put an itruncate log reservation in the new transaction
1362 * for our caller.
1363 */
1368 }
1369 /*
1370 * Return with the inode locked but not joined to the transaction.
1371 */
1375 error1:
1377 error0:
1378 /*
1379 * Have to come here with the inode locked and either
1380 * (held and in the transaction) or (not in the transaction).
1381 * If the inode isn't held then cancel would iput it, but
1382 * that's wrong since this is inactive and the vnode ref
1383 * count is 0 already.
1384 * Cancel won't do anything to the inode if held, but it still
1385 * needs to be locked until the cancel is done, if it was
1386 * joined to the transaction.
1387 */
1393 }
1395 STATIC int
1399 {
1403 /*
1404 * We're freeing a symlink which fit into
1405 * the inode. Just free the memory used
1406 * to hold the old symlink.
1407 */
1411 XFS_ITRUNCATE_LOG_COUNT);
1417 }
1420 /*
1421 * Zero length symlinks _can_ exist.
1422 */
1426 XFS_DATA_FORK);
1428 }
1430 }
1432 STATIC int
1436 {
1453 }
1459 XFS_INACTIVE_LOG_COUNT);
1466 }
1477 }
1479 int
1482 {
1490 /* If this is a read-only mount, don't do this (would generate I/O) */
1497 /*
1498 * If we are using filestreams, and we have an unlinked
1499 * file that we are processing the last close on, then nothing
1500 * will be able to reopen and write to this file. Purge this
1501 * inode from the filestreams cache so that it doesn't delay
1502 * teardown of the inode.
1503 */
1507 /*
1508 * If we previously truncated this file and removed old data
1509 * in the process, we want to initiate "early" writeout on
1510 * the last close. This is an attempt to combat the notorious
1511 * NULL files problem which is particularly noticable from a
1512 * truncate down, buffered (re-)write (delalloc), followed by
1513 * a crash. What we are effectively doing here is
1514 * significantly reducing the time window where we'd otherwise
1515 * be exposed to that problem.
1516 */
1520 }
1532 }
1533 }
1536 }
1538 /*
1539 * xfs_inactive
1540 *
1541 * This is called when the vnode reference count for the vnode
1542 * goes to zero. If the file has been unlinked, then it must
1543 * now be truncated. Also, we clear all of the read-ahead state
1544 * kept for the inode here since the file is now closed.
1545 */
1546 int
1549 {
1551 xfs_bmap_free_t free_list;
1552 xfs_fsblock_t first_block;
1561 /*
1562 * If the inode is already free, then there can be nothing
1563 * to clean up here.
1564 */
1569 }
1571 /*
1572 * Only do a truncate if it's a regular file with
1573 * some actual space in it. It's OK to look at the
1574 * inode's fields without the lock because we're the
1575 * only one with a reference to the inode.
1576 */
1586 }
1590 /* If this is a read-only mount, don't do this (would generate I/O) */
1605 }
1607 }
1616 /*
1617 * Do the xfs_itruncate_start() call before
1618 * reserving any log space because itruncate_start
1619 * will call into the buffer cache and we can't
1620 * do that within a transaction.
1621 */
1629 }
1634 XFS_ITRUNCATE_LOG_COUNT);
1636 /* Don't call itruncate_cleanup */
1641 }
1647 /*
1648 * normally, we have to run xfs_itruncate_finish sync.
1649 * But if filesystem is wsync and we're in the inactive
1650 * path, then we know that nlink == 0, and that the
1651 * xaction that made nlink == 0 is permanently committed
1652 * since xfs_remove runs as a synchronous transaction.
1653 */
1662 }
1665 /*
1666 * If we get an error while cleaning up a
1667 * symlink we bail out.
1668 */
1676 }
1684 XFS_INACTIVE_LOG_COUNT);
1689 }
1694 }
1696 /*
1697 * If there are attributes associated with the file
1698 * then blow them away now. The code calls a routine
1699 * that recursively deconstructs the attribute fork.
1700 * We need to just commit the current transaction
1701 * because we can't use it for xfs_attr_inactive().
1702 */
1705 /*
1706 * If we got an error, the transaction is already
1707 * cancelled, and the inode is unlocked. Just get out.
1708 */
1713 }
1715 /*
1716 * Free the inode.
1717 */
1721 /*
1722 * If we fail to free the inode, shut down. The cancel
1723 * might do that, we need to make sure. Otherwise the
1724 * inode might be lost for a long time or forever.
1725 */
1731 }
1734 /*
1735 * Credit the quota account(s). The inode is gone.
1736 */
1739 /*
1740 * Just ignore errors at this point. There is
1741 * nothing we can do except to try to keep going.
1742 */
1745 }
1746 /*
1747 * Release the dquots held by inode, if any.
1748 */
1753 out:
1755 }
1758 int
1763 {
1765 xfs_ino_t e_inum;
1767 uint lock_mode;
1779 }
1782 }
1784 int
1788 mode_t mode,
1789 xfs_dev_t rdev,
1792 {
1800 xfs_bmap_free_t free_list;
1801 xfs_fsblock_t first_block;
1804 uint cancel_flags;
1806 xfs_prid_t prid;
1808 uint resblks;
1825 }
1830 /* Return through std_return after this point. */
1835 else
1838 /*
1839 * Make sure that we have allocated dquot(s) on disk.
1840 */
1852 /*
1853 * Initially assume that the file does not exist and
1854 * reserve the resources for that case. If that is not
1855 * the case we'll drop the one we have and get a more
1856 * appropriate transaction later.
1857 */
1864 }
1868 }
1877 /*
1878 * Reserve disk quota and the inode.
1879 */
1893 }
1896 /*
1897 * At this point, we've gotten a newly allocated inode.
1898 * It is locked (and joined to the transaction).
1899 */
1903 /*
1904 * Now we join the directory inode to the transaction. We do not do it
1905 * earlier because xfs_dir_ialloc might commit the previous transaction
1906 * (and release all the locks). An error from here on will result in
1907 * the transaction cancel unlocking dp so don't do it explicitly in the
1908 * error path.
1909 */
1920 }
1924 /*
1925 * If this is a synchronous mount, make sure that the
1926 * create transaction goes to disk before returning to
1927 * the user.
1928 */
1931 }
1935 /*
1936 * Attach the dquot(s) to the inodes and modify them incore.
1937 * These ids of the inode couldn't have changed since the new
1938 * inode has been locked ever since it was created.
1939 */
1942 /*
1943 * xfs_trans_commit normally decrements the vnode ref count
1944 * when it unlocks the inode. Since we want to return the
1945 * vnode to the caller, we bump the vnode ref count now.
1946 */
1954 }
1961 }
1968 /* Fallthrough to std_return with error = 0 */
1970 std_return:
1978 }
1981 abort_return:
1983 /* FALLTHROUGH */
1985 error_return:
1997 abort_rele:
1998 /*
1999 * Wait until after the current transaction is aborted to
2000 * release the inode. This prevents recursive transactions
2001 * and deadlocks from xfs_inactive.
2002 */
2011 }
2013 #ifdef DEBUG
2014 /*
2015 * Some counters to see if (and how often) we are hitting some deadlock
2016 * prevention code paths.
2017 */
2022 #endif
2024 /*
2025 * The following routine will lock the inodes associated with the
2026 * directory and the named entry in the directory. The locks are
2027 * acquired in increasing inode number.
2028 *
2029 * If the entry is "..", then only the directory is locked. The
2030 * vnode ref count will still include that from the .. entry in
2031 * this case.
2032 *
2033 * There is a deadlock we need to worry about. If the locked directory is
2034 * in the AIL, it might be blocking up the log. The next inode we lock
2035 * could be already locked by another thread waiting for log space (e.g
2036 * a permanent log reservation with a long running transaction (see
2037 * xfs_itruncate_finish)). To solve this, we must check if the directory
2038 * is in the ail and use lock_nowait. If we can't lock, we need to
2039 * drop the inode lock on the directory and try again. xfs_iunlock will
2040 * potentially push the tail if we were holding up the log.
2041 */
2042 STATIC int
2046 {
2048 xfs_ino_t e_inum;
2052 #ifdef DEBUG
2053 xfs_rm_locks++;
2054 #endif
2057 again:
2064 /*
2065 * We want to lock in increasing inum. Since we've already
2066 * acquired the lock on the directory, we may need to release
2067 * if if the inum of the entry turns out to be less.
2068 */
2070 /*
2071 * We are already in the right order, so just
2072 * lock on the inode of the entry.
2073 * We need to use nowait if dp is in the AIL.
2074 */
2079 attempts++;
2080 #ifdef DEBUG
2081 xfs_rm_attempts++;
2082 #endif
2084 /*
2085 * Unlock dp and try again.
2086 * xfs_iunlock will try to push the tail
2087 * if the inode is in the AIL.
2088 */
2094 #ifdef DEBUG
2095 xfs_rm_lock_delays++;
2096 #endif
2097 }
2099 }
2102 }
2109 }
2110 /* else e_inum == dp->i_ino */
2111 /* This can happen if we're asked to lock /x/..
2112 * the entry is "..", which is also the parent directory.
2113 */
2116 }
2118 #ifdef DEBUG
2124 #endif
2126 /*
2127 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2128 * a different value
2129 */
2132 {
2139 }
2141 /*
2142 * The following routine will lock n inodes in exclusive mode.
2143 * We assume the caller calls us with the inodes in i_ino order.
2144 *
2145 * We need to detect deadlock where an inode that we lock
2146 * is in the AIL and we start waiting for another inode that is locked
2147 * by a thread in a long running transaction (such as truncate). This can
2148 * result in deadlock since the long running trans might need to wait
2149 * for the inode we just locked in order to push the tail and free space
2150 * in the log.
2151 */
2152 void
2157 uint lock_mode)
2158 {
2170 }
2172 again:
2179 /*
2180 * If try_lock is not set yet, make sure all locked inodes
2181 * are not in the AIL.
2182 * If any are, set try_lock to be used later.
2183 */
2189 try_lock++;
2190 }
2191 }
2192 }
2194 /*
2195 * If any of the previous locks we have locked is in the AIL,
2196 * we must TRY to get the second and subsequent locks. If
2197 * we can't get any, we must release all we have
2198 * and try again.
2199 */
2202 /* try_lock must be 0 if i is 0. */
2203 /*
2204 * try_lock means we have an inode locked
2205 * that is in the AIL.
2206 */
2209 attempts++;
2211 /*
2212 * Unlock all previous guys and try again.
2213 * xfs_iunlock will try to push the tail
2214 * if the inode is in the AIL.
2215 */
2219 /*
2220 * Check to see if we've already
2221 * unlocked this one.
2222 * Not the first one going back,
2223 * and the inode ptr is the same.
2224 */
2230 }
2234 #ifdef DEBUG
2235 xfs_lock_delays++;
2236 #endif
2237 }
2241 }
2244 }
2245 }
2247 #ifdef DEBUG
2253 xfs_locked_n++;
2254 }
2255 #endif
2256 }
2258 #ifdef DEBUG
2259 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2262 #define REMOVE_DEBUG_TRACE(x)
2265 int
2269 {
2277 xfs_bmap_free_t free_list;
2278 xfs_fsblock_t first_block;
2282 uint resblks;
2295 }
2297 /*
2298 * We need to get a reference to ip before we get our log
2299 * reservation. The reason for this is that we cannot call
2300 * xfs_iget for an inode for which we do not have a reference
2301 * once we've acquired a log reservation. This is because the
2302 * inode we are trying to get might be in xfs_inactive going
2303 * for a log reservation. Since we'll have to wait for the
2304 * inactive code to complete before returning from xfs_iget,
2305 * we need to make sure that we don't have log space reserved
2306 * when we call xfs_iget. Instead we get an unlocked reference
2307 * to the inode before getting our log reservation.
2308 */
2321 }
2325 /*
2326 * We try to get the real space reservation first,
2327 * allowing for directory btree deletion(s) implying
2328 * possible bmap insert(s). If we can't get the space
2329 * reservation then we use 0 instead, and avoid the bmap
2330 * btree insert(s) in the directory code by, if the bmap
2331 * insert tries to happen, instead trimming the LAST
2332 * block from the directory.
2333 */
2341 }
2348 }
2356 }
2358 /*
2359 * At this point, we've gotten both the directory and the entry
2360 * inodes locked.
2361 */
2364 /*
2365 * Increment vnode ref count only in this case since
2366 * there's an extra vnode reference in the case where
2367 * dp == ip.
2368 */
2371 }
2373 /*
2374 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2375 */
2383 }
2393 }
2395 /* Determine if this is the last link while
2396 * we are in the transaction.
2397 */
2400 /*
2401 * Take an extra ref on the inode so that it doesn't
2402 * go to xfs_inactive() from within the commit.
2403 */
2406 /*
2407 * If this is a synchronous mount, make sure that the
2408 * remove transaction goes to disk before returning to
2409 * the user.
2410 */
2413 }
2419 }
2425 }
2427 /*
2428 * If we are using filestreams, kill the stream association.
2429 * If the file is still open it may get a new one but that
2430 * will get killed on last close in xfs_close() so we don't
2431 * have to worry about that.
2432 */
2439 /* Fall through to std_return with error = 0 */
2440 std_return:
2446 }
2449 error1:
2455 error_rele:
2456 /*
2457 * In this case make sure to not release the inode until after
2458 * the current transaction is aborted. Releasing it beforehand
2459 * can cause us to go to xfs_inactive and start a recursive
2460 * transaction which can easily deadlock with the current one.
2461 */
2469 }
2471 int
2476 {
2483 xfs_bmap_free_t free_list;
2484 xfs_fsblock_t first_block;
2507 }
2509 /* Return through std_return after this point. */
2526 }
2530 }
2538 }
2542 /*
2543 * Increment vnode ref counts since xfs_trans_commit &
2544 * xfs_trans_cancel will both unlock the inodes and
2545 * decrement the associated ref counts.
2546 */
2552 /*
2553 * If the source has too many links, we can't make any more to it.
2554 */
2558 }
2560 /*
2561 * If we are using project inheritance, we only allow hard link
2562 * creation in our tree when the project IDs are the same; else
2563 * the tree quota mechanism could be circumvented.
2564 */
2569 }
2579 resblks);
2590 /*
2591 * If this is a synchronous mount, make sure that the
2592 * link transaction goes to disk before returning to
2593 * the user.
2594 */
2597 }
2603 }
2609 /* Fall through to std_return with error = 0. */
2610 std_return:
2616 }
2619 abort_return:
2621 /* FALLTHROUGH */
2623 error_return:
2626 }
2629 int
2633 mode_t mode,
2636 {
2647 xfs_bmap_free_t free_list;
2648 xfs_fsblock_t first_block;
2652 xfs_prid_t prid;
2654 uint resblks;
2669 }
2671 /* Return through std_return after this point. */
2679 else
2682 /*
2683 * Make sure that we have allocated dquot(s) on disk.
2684 */
2699 XFS_TRANS_PERM_LOG_RES,
2700 XFS_MKDIR_LOG_COUNT);
2701 }
2705 }
2710 /*
2711 * Check for directory link count overflow.
2712 */
2716 }
2718 /*
2719 * Reserve disk quota and the inode.
2720 */
2728 /*
2729 * create the directory inode.
2730 */
2738 }
2741 /*
2742 * Now we add the directory inode to the transaction.
2743 * We waited until now since xfs_dir_ialloc might start
2744 * a new transaction. Had we joined the transaction
2745 * earlier, the locks might have gotten released. An error
2746 * from here on will result in the transaction cancel
2747 * unlocking dp so don't do it explicitly in the error path.
2748 */
2761 }
2764 /*
2765 * Bump the in memory version number of the parent directory
2766 * so that other processes accessing it will recognize that
2767 * the directory has changed.
2768 */
2787 /*
2788 * Attach the dquots to the new inode and modify the icount incore.
2789 */
2792 /*
2793 * If this is a synchronous mount, make sure that the
2794 * mkdir transaction goes to disk before returning to
2795 * the user.
2796 */
2799 }
2805 }
2812 }
2814 /* Fall through to std_return with error = 0 or errno from
2815 * xfs_trans_commit. */
2817 std_return:
2823 DM_RIGHT_NULL,
2826 }
2829 error2:
2830 error1:
2832 abort_return:
2834 error_return:
2843 }
2845 int
2849 {
2857 xfs_bmap_free_t free_list;
2858 xfs_fsblock_t first_block;
2862 uint resblks;
2876 }
2878 /*
2879 * We need to get a reference to cdp before we get our log
2880 * reservation. The reason for this is that we cannot call
2881 * xfs_iget for an inode for which we do not have a reference
2882 * once we've acquired a log reservation. This is because the
2883 * inode we are trying to get might be in xfs_inactive going
2884 * for a log reservation. Since we'll have to wait for the
2885 * inactive code to complete before returning from xfs_iget,
2886 * we need to make sure that we don't have log space reserved
2887 * when we call xfs_iget. Instead we get an unlocked reference
2888 * to the inode before getting our log reservation.
2889 */
2892 /*
2893 * Get the dquots for the inodes.
2894 */
2902 }
2906 /*
2907 * We try to get the real space reservation first,
2908 * allowing for directory btree deletion(s) implying
2909 * possible bmap insert(s). If we can't get the space
2910 * reservation then we use 0 instead, and avoid the bmap
2911 * btree insert(s) in the directory code by, if the bmap
2912 * insert tries to happen, instead trimming the LAST
2913 * block from the directory.
2914 */
2922 }
2928 }
2931 /*
2932 * Now lock the child directory inode and the parent directory
2933 * inode in the proper order. This will take care of validating
2934 * that the directory entry for the child directory inode has
2935 * not changed while we were obtaining a log reservation.
2936 */
2942 }
2946 /*
2947 * Only increment the parent directory vnode count if
2948 * we didn't bump it in looking up cdp. The only time
2949 * we don't bump it is when we're looking up ".".
2950 */
2952 }
2961 }
2965 }
2974 /*
2975 * Bump the in memory generation count on the parent
2976 * directory so that other can know that it has changed.
2977 */
2980 /*
2981 * Drop the link from cdp's "..".
2982 */
2986 }
2988 /*
2989 * Drop the link from dp to cdp.
2990 */
2994 }
2996 /*
2997 * Drop the "." link from cdp to self.
2998 */
3002 }
3004 /* Determine these before committing transaction */
3007 /*
3008 * Take an extra ref on the child vnode so that it
3009 * does not go to xfs_inactive() from within the commit.
3010 */
3013 /*
3014 * If this is a synchronous mount, make sure that the
3015 * rmdir transaction goes to disk before returning to
3016 * the user.
3017 */
3020 }
3026 XFS_TRANS_ABORT));
3029 }
3035 }
3040 /* Fall through to std_return with error = 0 or the errno
3041 * from xfs_trans_commit. */
3042 std_return:
3049 }
3052 error1:
3055 /* FALLTHROUGH */
3057 error_return:
3060 }
3062 int
3067 mode_t mode,
3070 {
3077 xfs_bmap_free_t free_list;
3078 xfs_fsblock_t first_block;
3080 uint cancel_flags;
3082 xfs_fileoff_t first_fsb;
3083 xfs_filblks_t fs_blocks;
3086 xfs_daddr_t d;
3091 xfs_prid_t prid;
3093 uint resblks;
3109 /*
3110 * Check component lengths of the target path name.
3111 */
3120 /*
3121 * Skip any slashes.
3122 */
3124 total++;
3125 path++;
3126 }
3128 /*
3129 * Count up to the next slash or end of path.
3130 * Error out if the component is bigger than MAXNAMELEN.
3131 */
3136 }
3137 }
3138 }
3139 }
3147 }
3149 /* Return through std_return after this point. */
3154 else
3157 /*
3158 * Make sure that we have allocated dquot(s) on disk.
3159 */
3168 /*
3169 * The symlink will fit into the inode data fork?
3170 * There can't be any attributes so we get the whole variable part.
3171 */
3174 else
3183 }
3187 }
3192 /*
3193 * Check whether the directory allows new symlinks or not.
3194 */
3198 }
3200 /*
3201 * Reserve disk quota : blocks and inode.
3202 */
3207 /*
3208 * Check for ability to enter directory entry, if no space reserved.
3209 */
3213 /*
3214 * Initialize the bmap freelist prior to calling either
3215 * bmapi or the directory create code.
3216 */
3219 /*
3220 * Allocate an inode for the symlink.
3221 */
3228 }
3231 /*
3232 * An error after we've joined dp to the transaction will result in the
3233 * transaction cancel unlocking dp so don't do it explicitly in the
3234 * error path.
3235 */
3240 /*
3241 * Also attach the dquot(s) to it, if applicable.
3242 */
3247 /*
3248 * If the symlink will fit into the inode, write it inline.
3249 */
3255 /*
3256 * The inode was initially created in extent format.
3257 */
3274 }
3290 }
3297 }
3298 }
3300 /*
3301 * Create the directory entry for the symlink.
3302 */
3310 /*
3311 * Bump the in memory version number of the parent directory
3312 * so that other processes accessing it will recognize that
3313 * the directory has changed.
3314 */
3317 /*
3318 * If this is a synchronous mount, make sure that the
3319 * symlink transaction goes to disk before returning to
3320 * the user.
3321 */
3324 }
3326 /*
3327 * xfs_trans_commit normally decrements the vnode ref count
3328 * when it unlocks the inode. Since we want to return the
3329 * vnode to the caller, we bump the vnode ref count now.
3330 */
3336 }
3341 /* Fall through to std_return with error = 0 or errno from
3342 * xfs_trans_commit */
3343 std_return:
3350 }
3358 }
3361 error2:
3363 error1:
3366 error_return:
3375 }
3377 int
3380 bhv_vrwlock_t locktype)
3381 {
3394 }
3397 }
3400 void
3403 bhv_vrwlock_t locktype)
3404 {
3413 }
3415 }
3418 int
3422 {
3429 /*
3430 * Bypass inodes which have already been cleaned by
3431 * the inode flush clustering code inside xfs_iflush
3432 */
3436 /*
3437 * We make this non-blocking if the inode is contended,
3438 * return EAGAIN to indicate to the caller that they
3439 * did not succeed. This prevents the flush path from
3440 * blocking on inodes inside another operation right
3441 * now, they get caught later by xfs_sync.
3442 */
3450 }
3453 }
3460 }
3463 int
3466 u_int evmask,
3467 u_int16_t state)
3468 {
3484 }
3496 }
3498 int
3501 {
3508 /* bad inode, get out here ASAP */
3512 }
3518 /*
3519 * Make sure the atime in the XFS inode is correct before freeing the
3520 * Linux inode.
3521 */
3524 /*
3525 * If we have nothing to flush with this inode then complete the
3526 * teardown now, otherwise break the link between the xfs inode and the
3527 * linux inode and clean up the xfs inode later. This avoids flushing
3528 * the inode to disk during the delete operation itself.
3529 *
3530 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3531 * first to ensure that xfs_iunpin() will never see an xfs inode
3532 * that has a linux inode being reclaimed. Synchronisation is provided
3533 * by the i_flags_lock.
3534 */
3542 /* Protect sync and unpin from us */
3551 }
3553 }
3555 int
3560 {
3568 /* The hash lock here protects a thread in xfs_iget_core from
3569 * racing with us on linking the inode back with a vnode.
3570 * Once we have the XFS_IRECLAIM flag set it will not touch
3571 * us.
3572 */
3582 }
3584 }
3590 /*
3591 * If the inode is still dirty, then flush it out. If the inode
3592 * is not in the AIL, then it will be OK to flush it delwri as
3593 * long as xfs_iflush() does not keep any references to the inode.
3594 * We leave that decision up to xfs_iflush() since it has the
3595 * knowledge of whether it's OK to simply do a delwri flush of
3596 * the inode or whether we need to wait until the inode is
3597 * pulled from the AIL.
3598 * We get the flush lock regardless, though, just to make sure
3599 * we don't free it while it is being flushed.
3600 */
3604 }
3611 /*
3612 * If we hit an error, typically because of filesystem
3613 * shutdown, we don't need to let vn_reclaim to know
3614 * because we're gonna reclaim the inode anyway.
3615 */
3619 }
3621 }
3626 }
3631 reclaim:
3634 }
3636 int
3638 {
3654 }
3655 }
3658 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3662 }
3665 }
3669 }
3671 /*
3672 * xfs_alloc_file_space()
3673 * This routine allocates disk space for the given file.
3674 *
3675 * If alloc_type == 0, this request is for an ALLOCSP type
3676 * request which will change the file size. In this case, no
3677 * DMAPI event will be generated by the call. A TRUNCATE event
3678 * will be generated later by xfs_setattr.
3679 *
3680 * If alloc_type != 0, this request is for a RESVSP type
3681 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3682 * lower block boundary byte address is less than the file's
3683 * length.
3684 *
3685 * RETURNS:
3686 * 0 on success
3687 * errno on error
3688 *
3689 */
3690 STATIC int
3693 xfs_off_t offset,
3694 xfs_off_t len,
3697 {
3699 xfs_off_t count;
3700 xfs_filblks_t allocated_fsb;
3701 xfs_filblks_t allocatesize_fsb;
3703 xfs_fileoff_t startoffset_fsb;
3704 xfs_fsblock_t firstfsb;
3711 xfs_bmap_free_t free_list;
3737 /* Generate a DMAPI event if needed. */
3741 xfs_off_t end_dmi_offset;
3751 }
3753 /*
3754 * Allocate file space until done or until there is an error
3755 */
3756 retry:
3760 /*
3761 * Determine space reservations for data/realtime.
3762 */
3775 }
3787 }
3789 /*
3790 * Allocate and setup the transaction.
3791 */
3795 XFS_TRANS_PERM_LOG_RES,
3796 XFS_WRITE_LOG_COUNT);
3797 /*
3798 * Check for running out of space
3799 */
3801 /*
3802 * Free the transaction structure.
3803 */
3807 }
3817 /*
3818 * Issue the xfs_bmapi() call to allocate the blocks
3819 */
3827 }
3829 /*
3830 * Complete the transaction
3831 */
3835 }
3841 }
3848 }
3852 }
3853 dmapi_enospc_check:
3862 /* else fall through with error from XFS_SEND_DATA */
3863 }
3875 }
3877 /*
3878 * Zero file bytes between startoff and endoff inclusive.
3879 * The iolock is held exclusive and no blocks are buffered.
3880 */
3881 STATIC int
3884 xfs_off_t startoff,
3885 xfs_off_t endoff)
3886 {
3887 xfs_bmbt_irec_t imap;
3888 xfs_fileoff_t offset_fsb;
3889 xfs_off_t lastoffset;
3890 xfs_off_t offset;
3926 }
3938 }
3939 }
3942 }
3944 /*
3945 * xfs_free_file_space()
3946 * This routine frees disk space for the given file.
3947 *
3948 * This routine is only called by xfs_change_file_space
3949 * for an UNRESVSP type call.
3950 *
3951 * RETURNS:
3952 * 0 on success
3953 * errno on error
3954 *
3955 */
3956 STATIC int
3959 xfs_off_t offset,
3960 xfs_off_t len,
3962 {
3966 xfs_off_t end_dmi_offset;
3967 xfs_fileoff_t endoffset_fsb;
3969 xfs_fsblock_t firstfsb;
3970 xfs_bmap_free_t free_list;
3971 xfs_bmbt_irec_t imap;
3972 xfs_off_t ioffset;
3976 uint resblks;
3977 uint rounding;
3979 xfs_fileoff_t startoffset_fsb;
4008 }
4015 }
4025 }
4027 /*
4028 * Need to zero the stuff we're not freeing, on disk.
4029 * If its a realtime file & can't use unwritten extents then we
4030 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4031 * will take care of it for us.
4032 */
4041 xfs_daddr_t block;
4048 }
4057 mod++;
4060 }
4061 }
4063 /*
4064 * One contiguous piece to clear
4065 */
4068 /*
4069 * Some full blocks, possibly two pieces to clear
4070 */
4079 }
4081 /*
4082 * free file space until done or until there is an error
4083 */
4087 /*
4088 * allocate and setup the transaction. Allow this
4089 * transaction to dip into the reserve blocks to ensure
4090 * the freeing of the space succeeds at ENOSPC.
4091 */
4095 resblks,
4098 XFS_TRANS_PERM_LOG_RES,
4099 XFS_WRITE_LOG_COUNT);
4101 /*
4102 * check for running out of space
4103 */
4105 /*
4106 * Free the transaction structure.
4107 */
4111 }
4115 XFS_QMOPT_RES_REGBLKS);
4122 /*
4123 * issue the bunmapi() call to free the blocks
4124 */
4131 }
4133 /*
4134 * complete the transaction
4135 */
4139 }
4143 }
4145 out_unlock_iolock:
4150 error0:
4152 error1:
4155 XFS_ILOCK_EXCL);
4157 }
4159 /*
4160 * xfs_change_file_space()
4161 * This routine allocates or frees disk space for the given file.
4162 * The user specified parameters are checked for alignment and size
4163 * limitations.
4164 *
4165 * RETURNS:
4166 * 0 on success
4167 * errno on error
4168 *
4169 */
4170 int
4175 xfs_off_t offset,
4178 {
4182 xfs_fsize_t fsize;
4184 xfs_off_t startoffset;
4185 xfs_off_t llen;
4187 bhv_vattr_t va;
4205 }
4220 /*
4221 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4222 * file space.
4223 * These calls do NOT zero the data space allocated to the file,
4224 * nor do they change the file size.
4225 *
4226 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4227 * space.
4228 * These calls cause the new file data to be zeroed and the file
4229 * size to be changed.
4230 */
4246 attr_flags)))
4259 }
4275 }
4277 /*
4278 * update the inode timestamp, mode, and prealloc flag bits
4279 */
4284 /* ASSERT(0); */
4287 }
4297 /*
4298 * Note that we don't have to worry about mandatory
4299 * file locking being disabled here because we only
4300 * clear the S_ISGID bit if the Group execute bit is
4301 * on, but if it was on then mandatory locking wouldn't
4302 * have been enabled.
4303 */
4308 }
4322 }