| blob | e03fa2a3d5ed4bb2af022b31c4bf0309c481d702 |
1 /*
2 * Copyright (c) 2000-2005 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 */
59 /*
60 * The maximum pathlen is 1024 bytes. Since the minimum file system
61 * blocksize is 512 bytes, we can get a max of 2 extents back from
62 * bmapi.
63 */
64 #define SYMLINK_MAPS 2
66 /*
67 * For xfs, we check that the file isn't too big to be opened by this kernel.
68 * No other open action is required for regular files. Devices are handled
69 * through the specfs file system, pipes through fifofs. Device and
70 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
71 * when a new vnode is first looked up or created.
72 */
73 STATIC int
77 {
88 /*
89 * If it's a directory with any blocks, read-ahead block 0
90 * as we're almost certain to have the next operation be a read there.
91 */
97 }
99 }
102 /*
103 * xfs_getattr
104 */
105 STATIC int
111 {
135 #if XFS_BIG_INUMS
137 #endif
140 /*
141 * Quick exit for non-stat callers
142 */
148 /*
149 * Copy from in-core inode.
150 */
156 /*
157 * Check vnode type block/char vs. everything else.
158 */
172 /*
173 * If the file blocks are being allocated from a
174 * realtime partition, then return the inode's
175 * realtime extent size or the realtime volume's
176 * extent size.
177 */
181 }
183 }
192 /*
193 * Exit for stat callers. See if any of the rest of the fields
194 * to be filled in are needed.
195 */
201 /*
202 * Convert di_flags to xflags.
203 */
206 /*
207 * Exit for inode revalidate. See if any of the rest of
208 * the fields to be filled in are needed.
209 */
225 else
229 all_done:
233 }
236 /*
237 * xfs_setattr
238 */
239 int
245 {
251 uint lock_flags;
269 /*
270 * Cannot set certain attributes.
271 */
275 }
283 /*
284 * Timestamps do not need to be logged and hence do not
285 * need to be done within a transaction.
286 */
294 }
299 /*
300 * If disk quotas is on, we make sure that the dquots do exist on disk,
301 * before we start any other transactions. Trying to do this later
302 * is messy. We don't care to take a readlock to look at the ids
303 * in inode here, because we can't hold it across the trans_reserve.
304 * If the IDs do change before we take the ilock, we're covered
305 * because the i_*dquot fields will get updated anyway.
306 */
316 }
322 }
328 }
329 /*
330 * We take a reference when we initialize udqp and gdqp,
331 * so it is important that we never blindly double trip on
332 * the same variable. See xfs_create() for an example.
333 */
340 }
342 /*
343 * For the other attributes, we acquire the inode lock and
344 * first do an error checking pass.
345 */
361 }
362 }
372 }
373 }
376 }
380 /* boolean: are we the file owner? */
383 /*
384 * Change various properties of a file.
385 * Only the owner or users with CAP_FOWNER
386 * capability may do these things.
387 */
391 /*
392 * CAP_FOWNER overrides the following restrictions:
393 *
394 * The user ID of the calling process must be equal
395 * to the file owner ID, except in cases where the
396 * CAP_FSETID capability is applicable.
397 */
401 }
403 /*
404 * CAP_FSETID overrides the following restrictions:
405 *
406 * The effective user ID of the calling process shall match
407 * the file owner when setting the set-user-ID and
408 * set-group-ID bits on that file.
409 *
410 * The effective group ID or one of the supplementary group
411 * IDs of the calling process shall match the group owner of
412 * the file when setting the set-group-ID bit on that file
413 */
422 #if 0
423 /* Linux allows this, Irix doesn't. */
426 #endif
429 }
430 }
432 /*
433 * Change file ownership. Must be the owner or privileged.
434 * If the system was configured with the "restricted_chown"
435 * option, the owner is not permitted to give away the file,
436 * and can change the group id only to a group of which he
437 * or she is a member.
438 */
440 /*
441 * These IDs could have changed since we last looked at them.
442 * But, we're assured that if the ownership did change
443 * while we didn't have the inode locked, inode's dquot(s)
444 * would have changed also.
445 */
452 iprojid;
454 /*
455 * CAP_CHOWN overrides the following restrictions:
456 *
457 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
458 * shall override the restriction that a process cannot
459 * change the user ID of a file it owns and the restriction
460 * that the group ID supplied to the chown() function
461 * shall be equal to either the group ID or one of the
462 * supplementary group IDs of the calling process.
463 */
470 }
471 /*
472 * Do a quota reservation only if uid/projid/gid is actually
473 * going to change.
474 */
484 }
485 }
487 /*
488 * Truncate file. Must have write permission and not be a directory.
489 */
491 /* Short circuit the truncate case for zero length files */
500 }
508 }
509 /*
510 * Make sure that the dquots are attached to the inode.
511 */
514 }
516 /*
517 * Change file access or modified times.
518 */
525 }
526 }
527 }
529 /*
530 * Change extent size or realtime flag.
531 */
533 /*
534 * Can't change extent size if any extents are allocated.
535 */
541 }
543 /*
544 * Can't set extent size unless the file is marked, or
545 * about to be marked as a realtime file.
546 *
547 * This check will be removed when fixed size extents
548 * with buffered data writes is implemented.
549 *
550 */
559 }
561 /*
562 * Can't change realtime flag if any extents are allocated.
563 */
570 }
571 /*
572 * Extent size must be a multiple of the appropriate block
573 * size, if set at all.
574 */
576 xfs_extlen_t size;
585 }
589 }
590 }
591 /*
592 * If realtime flag is set then must have realtime data.
593 */
601 }
602 }
604 /*
605 * Can't modify an immutable/append-only file unless
606 * we have appropriate permission.
607 */
616 }
617 }
619 /*
620 * Now we can make the changes. Before we join the inode
621 * to the transaction, if XFS_AT_SIZE is set then take care of
622 * the part of the truncation that must be done without the
623 * inode lock. This needs to be done before joining the inode
624 * to the transaction, because the inode cannot be unlocked
625 * once it is a part of the transaction.
626 */
632 }
641 }
645 XFS_TRANS_PERM_LOG_RES,
646 XFS_ITRUNCATE_LOG_COUNT))) {
651 }
654 }
659 }
661 /* determine whether mandatory locking mode changes */
664 /*
665 * Truncate file. Must have write permission and not be a directory.
666 */
673 /*
674 * signal a sync transaction unless
675 * we're truncating an already unlinked
676 * file on a wsync filesystem
677 */
680 XFS_DATA_FORK,
686 }
687 }
688 /*
689 * Have to do this even if the file's size doesn't change.
690 */
692 }
694 /*
695 * Change file access modes.
696 */
703 }
705 /*
706 * Change file ownership. Must be the owner or privileged.
707 * If the system was configured with the "restricted_chown"
708 * option, the owner is not permitted to give away the file,
709 * and can change the group id only to a group of which he
710 * or she is a member.
711 */
713 /*
714 * CAP_FSETID overrides the following restrictions:
715 *
716 * The set-user-ID and set-group-ID bits of a file will be
717 * cleared upon successful return from chown()
718 */
722 }
724 /*
725 * Change the ownerships and register quota modifications
726 * in the transaction.
727 */
734 }
736 }
744 }
746 }
754 }
756 /*
757 * We may have to rev the inode as well as
758 * the superblock version number since projids didn't
759 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
760 */
763 }
767 }
770 /*
771 * Change file access or modified times.
772 */
779 }
785 }
788 }
790 /*
791 * Change XFS-added attributes.
792 */
795 /*
796 * Converting bytes to fs blocks.
797 */
800 }
802 uint di_flags;
804 /* can't set PREALLOC this way, just preserve it */
829 }
830 }
832 }
835 }
837 /*
838 * Change file inode change time only if XFS_AT_CTIME set
839 * AND we have been called by a DMI function.
840 */
847 }
849 /*
850 * Send out timestamp changes that need to be set to the
851 * current time. Not done when called by a DMI function.
852 */
858 /*
859 * If this is a synchronous mount, make sure that the
860 * transaction goes to disk before returning to the user.
861 * This is slightly sub-optimal in that truncates require
862 * two sync transactions instead of one for wsync filesytems.
863 * One for the truncate and one for the timestamps since we
864 * don't want to change the timestamps unless we're sure the
865 * truncate worked. Truncates are less than 1% of the laddis
866 * mix so this probably isn't worth the trouble to optimize.
867 */
874 }
876 /*
877 * If the (regular) file's mandatory locking mode changed, then
878 * notify the vnode. We do this under the inode lock to prevent
879 * racing calls to vop_vnode_change.
880 */
884 mandlock_after);
885 }
889 /*
890 * Release any dquot(s) the inode had kept before chown.
891 */
899 }
906 }
909 abort_return:
911 /* FALLTHROUGH */
912 error_return:
917 }
920 }
922 }
925 /*
926 * xfs_access
927 * Null conversion from vnode mode bits to inode mode bits, as in efs.
928 */
929 STATIC int
934 {
946 }
949 /*
950 * xfs_readlink
951 *
952 */
953 STATIC int
959 {
962 xfs_off_t offset;
969 xfs_daddr_t d;
993 }
997 }
1001 }
1003 /*
1004 * See if the symlink is stored inline.
1005 */
1010 }
1012 /*
1013 * Symlink not inline. Call bmap to get it in.
1014 */
1022 }
1035 }
1042 }
1044 }
1047 error_return:
1052 }
1055 /*
1056 * xfs_fsync
1057 *
1058 * This is called to sync the inode and its data out to disk.
1059 * We need to hold the I/O lock while flushing the data, and
1060 * the inode lock while flushing the inode. The inode lock CANNOT
1061 * be held while flushing the data, so acquire after we're done
1062 * with that.
1063 */
1064 STATIC int
1069 xfs_off_t start,
1070 xfs_off_t stop)
1071 {
1087 /*
1088 * We always need to make sure that the required inode state
1089 * is safe on disk. The vnode might be clean but because
1090 * of committed transactions that haven't hit the disk yet.
1091 * Likewise, there could be unflushed non-transactional
1092 * changes to the inode core that have to go to disk.
1093 *
1094 * The following code depends on one assumption: that
1095 * any transaction that changes an inode logs the core
1096 * because it has to change some field in the inode core
1097 * (typically nextents or nblocks). That assumption
1098 * implies that any transactions against an inode will
1099 * catch any non-transactional updates. If inode-altering
1100 * transactions exist that violate this assumption, the
1101 * code breaks. Right now, it figures that if the involved
1102 * update_* field is clear and the inode is unpinned, the
1103 * inode is clean. Either it's been flushed or it's been
1104 * committed and the commit has hit the disk unpinning the inode.
1105 * (Note that xfs_inode_item_format() called at commit clears
1106 * the update_* fields.)
1107 */
1110 /* If we are flushing data then we care about update_size
1111 * being set, otherwise we care about update_core
1112 */
1116 /*
1117 * Timestamps/size haven't changed since last inode
1118 * flush or inode transaction commit. That means
1119 * either nothing got written or a transaction
1120 * committed which caught the updates. If the
1121 * latter happened and the transaction hasn't
1122 * hit the disk yet, the inode will be still
1123 * be pinned. If it is, force the log.
1124 */
1130 XFS_LOG_FORCE |
1135 /*
1136 * If the inode is not pinned and nothing
1137 * has changed we don't need to flush the
1138 * cache.
1139 */
1141 }
1144 /*
1145 * Kick off a transaction to log the inode
1146 * core to get the updates. Make it
1147 * sync if FSYNC_WAIT is passed in (which
1148 * is done by everybody but specfs). The
1149 * sync transaction will also force the log.
1150 */
1158 }
1161 /*
1162 * Note - it's possible that we might have pushed
1163 * ourselves out of the way during trans_reserve
1164 * which would flush the inode. But there's no
1165 * guarantee that the inode buffer has actually
1166 * gone out yet (it's delwri). Plus the buffer
1167 * could be pinned anyway if it's part of an
1168 * inode in another recent transaction. So we
1169 * play it safe and fire off the transaction anyway.
1170 */
1179 }
1182 /*
1183 * If the log write didn't issue an ordered tag we need
1184 * to flush the disk cache for the data device now.
1185 */
1189 /*
1190 * If this inode is on the RT dev we need to flush that
1191 * cache aswell.
1192 */
1195 }
1198 }
1200 /*
1201 * This is called by xfs_inactive to free any blocks beyond eof,
1202 * when the link count isn't zero.
1203 */
1204 STATIC int
1208 {
1211 xfs_fileoff_t end_fsb;
1212 xfs_fileoff_t last_fsb;
1213 xfs_filblks_t map_len;
1215 xfs_bmbt_irec_t imap;
1217 /*
1218 * Figure out if there are any blocks beyond the end
1219 * of the file. If not, then there is nothing to do.
1220 */
1235 /*
1236 * Attach the dquots to the inode up front.
1237 */
1241 /*
1242 * There are blocks after the end of file.
1243 * Free them up now by truncating the file to
1244 * its current size.
1245 */
1248 /*
1249 * Do the xfs_itruncate_start() call before
1250 * reserving any log space because
1251 * itruncate_start will call into the buffer
1252 * cache and we can't
1253 * do that within a transaction.
1254 */
1262 XFS_ITRUNCATE_LOG_COUNT);
1268 }
1272 XFS_IOLOCK_EXCL |
1273 XFS_ILOCK_EXCL);
1278 XFS_DATA_FORK,
1280 /*
1281 * If we get an error at this point we
1282 * simply don't bother truncating the file.
1283 */
1287 XFS_TRANS_ABORT));
1290 XFS_TRANS_RELEASE_LOG_RES,
1291 NULL);
1292 }
1294 }
1296 }
1298 /*
1299 * Free a symlink that has blocks associated with it.
1300 */
1301 STATIC int
1305 {
1310 xfs_fsblock_t first_block;
1311 xfs_bmap_free_t free_list;
1323 /*
1324 * We're freeing a symlink that has some
1325 * blocks allocated to it. Free the
1326 * blocks here. We know that we've got
1327 * either 1 or 2 extents and that we can
1328 * free them all in one bunmapi call.
1329 */
1337 }
1338 /*
1339 * Lock the inode, fix the size, and join it to the transaction.
1340 * Hold it so in the normal path, we still have it locked for
1341 * the second transaction. In the error paths we need it
1342 * held so the cancel won't rele it, see below.
1343 */
1350 /*
1351 * Find the block(s) so we can inval and unmap them.
1352 */
1360 /*
1361 * Invalidate the block(s).
1362 */
1368 }
1369 /*
1370 * Unmap the dead block(s) to the free_list.
1371 */
1376 /*
1377 * Commit the first transaction. This logs the EFI and the inode.
1378 */
1381 /*
1382 * The transaction must have been committed, since there were
1383 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1384 * The new tp has the extent freeing and EFDs.
1385 */
1387 /*
1388 * The first xact was committed, so add the inode to the new one.
1389 * Mark it dirty so it will be logged and moved forward in the log as
1390 * part of every commit.
1391 */
1395 /*
1396 * Get a new, empty transaction to return to our caller.
1397 */
1399 /*
1400 * Commit the transaction containing extent freeing and EFD's.
1401 * If we get an error on the commit here or on the reserve below,
1402 * we need to unlock the inode since the new transaction doesn't
1403 * have the inode attached.
1404 */
1410 }
1411 /*
1412 * Remove the memory for extent descriptions (just bookkeeping).
1413 */
1417 /*
1418 * Put an itruncate log reservation in the new transaction
1419 * for our caller.
1420 */
1425 }
1426 /*
1427 * Return with the inode locked but not joined to the transaction.
1428 */
1432 error1:
1434 error0:
1435 /*
1436 * Have to come here with the inode locked and either
1437 * (held and in the transaction) or (not in the transaction).
1438 * If the inode isn't held then cancel would iput it, but
1439 * that's wrong since this is inactive and the vnode ref
1440 * count is 0 already.
1441 * Cancel won't do anything to the inode if held, but it still
1442 * needs to be locked until the cancel is done, if it was
1443 * joined to the transaction.
1444 */
1450 }
1452 STATIC int
1456 {
1460 /*
1461 * We're freeing a symlink which fit into
1462 * the inode. Just free the memory used
1463 * to hold the old symlink.
1464 */
1468 XFS_ITRUNCATE_LOG_COUNT);
1474 }
1477 /*
1478 * Zero length symlinks _can_ exist.
1479 */
1483 XFS_DATA_FORK);
1485 }
1487 }
1489 /*
1490 *
1491 */
1492 STATIC int
1496 {
1513 }
1519 XFS_INACTIVE_LOG_COUNT);
1526 }
1537 }
1539 STATIC int
1542 {
1553 }
1555 /* If this is a read-only mount, don't do this (would generate I/O) */
1559 #ifdef HAVE_REFCACHE
1560 /* If we are in the NFS reference cache then don't do this now */
1563 #endif
1574 /* Update linux inode block count after free above */
1577 }
1578 }
1581 }
1583 /*
1584 * xfs_inactive
1585 *
1586 * This is called when the vnode reference count for the vnode
1587 * goes to zero. If the file has been unlinked, then it must
1588 * now be truncated. Also, we clear all of the read-ahead state
1589 * kept for the inode here since the file is now closed.
1590 */
1591 STATIC int
1595 {
1598 xfs_bmap_free_t free_list;
1599 xfs_fsblock_t first_block;
1611 /*
1612 * If the inode is already free, then there can be nothing
1613 * to clean up here.
1614 */
1619 }
1621 /*
1622 * Only do a truncate if it's a regular file with
1623 * some actual space in it. It's OK to look at the
1624 * inode's fields without the lock because we're the
1625 * only one with a reference to the inode.
1626 */
1636 }
1640 /* If this is a read-only mount, don't do this (would generate I/O) */
1652 /* Update linux inode block count after free above */
1655 }
1657 }
1666 /*
1667 * Do the xfs_itruncate_start() call before
1668 * reserving any log space because itruncate_start
1669 * will call into the buffer cache and we can't
1670 * do that within a transaction.
1671 */
1679 XFS_ITRUNCATE_LOG_COUNT);
1681 /* Don't call itruncate_cleanup */
1686 }
1692 /*
1693 * normally, we have to run xfs_itruncate_finish sync.
1694 * But if filesystem is wsync and we're in the inactive
1695 * path, then we know that nlink == 0, and that the
1696 * xaction that made nlink == 0 is permanently committed
1697 * since xfs_remove runs as a synchronous transaction.
1698 */
1707 }
1710 /*
1711 * If we get an error while cleaning up a
1712 * symlink we bail out.
1713 */
1721 }
1729 XFS_INACTIVE_LOG_COUNT);
1734 }
1739 }
1741 /*
1742 * If there are attributes associated with the file
1743 * then blow them away now. The code calls a routine
1744 * that recursively deconstructs the attribute fork.
1745 * We need to just commit the current transaction
1746 * because we can't use it for xfs_attr_inactive().
1747 */
1750 /*
1751 * If we got an error, the transaction is already
1752 * cancelled, and the inode is unlocked. Just get out.
1753 */
1758 }
1760 /*
1761 * Free the inode.
1762 */
1766 /*
1767 * If we fail to free the inode, shut down. The cancel
1768 * might do that, we need to make sure. Otherwise the
1769 * inode might be lost for a long time or forever.
1770 */
1776 }
1779 /*
1780 * Credit the quota account(s). The inode is gone.
1781 */
1784 /*
1785 * Just ignore errors at this point. There is
1786 * nothing we can do except to try to keep going.
1787 */
1791 }
1792 /*
1793 * Release the dquots held by inode, if any.
1794 */
1799 out:
1801 }
1804 /*
1805 * xfs_lookup
1806 */
1807 STATIC int
1815 {
1817 xfs_ino_t e_inum;
1819 uint lock_mode;
1835 }
1838 }
1841 /*
1842 * xfs_create (create a new file).
1843 */
1844 STATIC int
1851 {
1858 xfs_dev_t rdev;
1860 xfs_bmap_free_t free_list;
1861 xfs_fsblock_t first_block;
1862 boolean_t dp_joined_to_trans;
1864 uint cancel_flags;
1866 xfs_prid_t prid;
1868 uint resblks;
1891 }
1896 /* Return through std_return after this point. */
1903 else
1906 /*
1907 * Make sure that we have allocated dquot(s) on disk.
1908 */
1921 /*
1922 * Initially assume that the file does not exist and
1923 * reserve the resources for that case. If that is not
1924 * the case we'll drop the one we have and get a more
1925 * appropriate transaction later.
1926 */
1933 }
1938 }
1946 /*
1947 * Reserve disk quota and the inode.
1948 */
1964 }
1967 /*
1968 * At this point, we've gotten a newly allocated inode.
1969 * It is locked (and joined to the transaction).
1970 */
1974 /*
1975 * Now we join the directory inode to the transaction.
1976 * We do not do it earlier because xfs_dir_ialloc
1977 * might commit the previous transaction (and release
1978 * all the locks).
1979 */
1991 }
1995 /*
1996 * If this is a synchronous mount, make sure that the
1997 * create transaction goes to disk before returning to
1998 * the user.
1999 */
2002 }
2006 /*
2007 * Attach the dquot(s) to the inodes and modify them incore.
2008 * These ids of the inode couldn't have changed since the new
2009 * inode has been locked ever since it was created.
2010 */
2013 /*
2014 * xfs_trans_commit normally decrements the vnode ref count
2015 * when it unlocks the inode. Since we want to return the
2016 * vnode to the caller, we bump the vnode ref count now.
2017 */
2025 }
2032 }
2037 /*
2038 * Propogate the fact that the vnode changed after the
2039 * xfs_inode locks have been released.
2040 */
2045 /* Fallthrough to std_return with error = 0 */
2047 std_return:
2050 DM_EVENT_POSTCREATE)) {
2056 }
2059 abort_return:
2061 /* FALLTHROUGH */
2062 error_return:
2074 abort_rele:
2075 /*
2076 * Wait until after the current transaction is aborted to
2077 * release the inode. This prevents recursive transactions
2078 * and deadlocks from xfs_inactive.
2079 */
2088 }
2090 #ifdef DEBUG
2091 /*
2092 * Some counters to see if (and how often) we are hitting some deadlock
2093 * prevention code paths.
2094 */
2099 #endif
2101 /*
2102 * The following routine will lock the inodes associated with the
2103 * directory and the named entry in the directory. The locks are
2104 * acquired in increasing inode number.
2105 *
2106 * If the entry is "..", then only the directory is locked. The
2107 * vnode ref count will still include that from the .. entry in
2108 * this case.
2109 *
2110 * There is a deadlock we need to worry about. If the locked directory is
2111 * in the AIL, it might be blocking up the log. The next inode we lock
2112 * could be already locked by another thread waiting for log space (e.g
2113 * a permanent log reservation with a long running transaction (see
2114 * xfs_itruncate_finish)). To solve this, we must check if the directory
2115 * is in the ail and use lock_nowait. If we can't lock, we need to
2116 * drop the inode lock on the directory and try again. xfs_iunlock will
2117 * potentially push the tail if we were holding up the log.
2118 */
2119 STATIC int
2124 {
2126 xfs_ino_t e_inum;
2130 #ifdef DEBUG
2131 xfs_rm_locks++;
2132 #endif
2135 again:
2142 /*
2143 * We want to lock in increasing inum. Since we've already
2144 * acquired the lock on the directory, we may need to release
2145 * if if the inum of the entry turns out to be less.
2146 */
2148 /*
2149 * We are already in the right order, so just
2150 * lock on the inode of the entry.
2151 * We need to use nowait if dp is in the AIL.
2152 */
2157 attempts++;
2158 #ifdef DEBUG
2159 xfs_rm_attempts++;
2160 #endif
2162 /*
2163 * Unlock dp and try again.
2164 * xfs_iunlock will try to push the tail
2165 * if the inode is in the AIL.
2166 */
2172 #ifdef DEBUG
2173 xfs_rm_lock_delays++;
2174 #endif
2175 }
2177 }
2180 }
2187 }
2188 /* else e_inum == dp->i_ino */
2189 /* This can happen if we're asked to lock /x/..
2190 * the entry is "..", which is also the parent directory.
2191 */
2194 }
2196 #ifdef DEBUG
2202 #endif
2204 /*
2205 * The following routine will lock n inodes in exclusive mode.
2206 * We assume the caller calls us with the inodes in i_ino order.
2207 *
2208 * We need to detect deadlock where an inode that we lock
2209 * is in the AIL and we start waiting for another inode that is locked
2210 * by a thread in a long running transaction (such as truncate). This can
2211 * result in deadlock since the long running trans might need to wait
2212 * for the inode we just locked in order to push the tail and free space
2213 * in the log.
2214 */
2215 void
2220 uint lock_mode)
2221 {
2233 }
2235 again:
2242 /*
2243 * If try_lock is not set yet, make sure all locked inodes
2244 * are not in the AIL.
2245 * If any are, set try_lock to be used later.
2246 */
2252 try_lock++;
2253 }
2254 }
2255 }
2257 /*
2258 * If any of the previous locks we have locked is in the AIL,
2259 * we must TRY to get the second and subsequent locks. If
2260 * we can't get any, we must release all we have
2261 * and try again.
2262 */
2265 /* try_lock must be 0 if i is 0. */
2266 /*
2267 * try_lock means we have an inode locked
2268 * that is in the AIL.
2269 */
2272 attempts++;
2274 /*
2275 * Unlock all previous guys and try again.
2276 * xfs_iunlock will try to push the tail
2277 * if the inode is in the AIL.
2278 */
2282 /*
2283 * Check to see if we've already
2284 * unlocked this one.
2285 * Not the first one going back,
2286 * and the inode ptr is the same.
2287 */
2293 }
2297 #ifdef DEBUG
2298 xfs_lock_delays++;
2299 #endif
2300 }
2304 }
2307 }
2308 }
2310 #ifdef DEBUG
2316 xfs_locked_n++;
2317 }
2318 #endif
2319 }
2321 #ifdef DEBUG
2322 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2325 #define REMOVE_DEBUG_TRACE(x)
2329 /*
2330 * xfs_remove
2331 *
2332 */
2333 STATIC int
2338 {
2345 xfs_bmap_free_t free_list;
2346 xfs_fsblock_t first_block;
2351 uint resblks;
2371 }
2373 /* From this point on, return through std_return */
2376 /*
2377 * We need to get a reference to ip before we get our log
2378 * reservation. The reason for this is that we cannot call
2379 * xfs_iget for an inode for which we do not have a reference
2380 * once we've acquired a log reservation. This is because the
2381 * inode we are trying to get might be in xfs_inactive going
2382 * for a log reservation. Since we'll have to wait for the
2383 * inactive code to complete before returning from xfs_iget,
2384 * we need to make sure that we don't have log space reserved
2385 * when we call xfs_iget. Instead we get an unlocked referece
2386 * to the inode before getting our log reservation.
2387 */
2392 }
2407 }
2411 /*
2412 * We try to get the real space reservation first,
2413 * allowing for directory btree deletion(s) implying
2414 * possible bmap insert(s). If we can't get the space
2415 * reservation then we use 0 instead, and avoid the bmap
2416 * btree insert(s) in the directory code by, if the bmap
2417 * insert tries to happen, instead trimming the LAST
2418 * block from the directory.
2419 */
2427 }
2434 }
2442 }
2444 /*
2445 * At this point, we've gotten both the directory and the entry
2446 * inodes locked.
2447 */
2450 /*
2451 * Increment vnode ref count only in this case since
2452 * there's an extra vnode reference in the case where
2453 * dp == ip.
2454 */
2457 }
2459 /*
2460 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2461 */
2469 }
2479 }
2481 /* Determine if this is the last link while
2482 * we are in the transaction.
2483 */
2486 /*
2487 * Take an extra ref on the inode so that it doesn't
2488 * go to xfs_inactive() from within the commit.
2489 */
2492 /*
2493 * If this is a synchronous mount, make sure that the
2494 * remove transaction goes to disk before returning to
2495 * the user.
2496 */
2499 }
2505 }
2511 }
2513 /*
2514 * Before we drop our extra reference to the inode, purge it
2515 * from the refcache if it is there. By waiting until afterwards
2516 * to do the IRELE, we ensure that we won't go inactive in the
2517 * xfs_refcache_purge_ip routine (although that would be OK).
2518 */
2523 /*
2524 * Let interposed file systems know about removed links.
2525 */
2530 /* Fall through to std_return with error = 0 */
2531 std_return:
2533 DM_EVENT_POSTREMOVE)) {
2538 }
2541 error1:
2547 error_rele:
2548 /*
2549 * In this case make sure to not release the inode until after
2550 * the current transaction is aborted. Releasing it beforehand
2551 * can cause us to go to xfs_inactive and start a recursive
2552 * transaction which can easily deadlock with the current one.
2553 */
2558 /*
2559 * Before we drop our extra reference to the inode, purge it
2560 * from the refcache if it is there. By waiting until afterwards
2561 * to do the IRELE, we ensure that we won't go inactive in the
2562 * xfs_refcache_purge_ip routine (although that would be OK).
2563 */
2569 }
2572 /*
2573 * xfs_link
2574 *
2575 */
2576 STATIC int
2582 {
2588 xfs_bmap_free_t free_list;
2589 xfs_fsblock_t first_block;
2620 }
2622 /* Return through std_return after this point. */
2639 }
2643 }
2651 }
2655 /*
2656 * Increment vnode ref counts since xfs_trans_commit &
2657 * xfs_trans_cancel will both unlock the inodes and
2658 * decrement the associated ref counts.
2659 */
2665 /*
2666 * If the source has too many links, we can't make any more to it.
2667 */
2671 }
2673 /*
2674 * If we are using project inheritance, we only allow hard link
2675 * creation in our tree when the project IDs are the same; else
2676 * the tree quota mechanism could be circumvented.
2677 */
2682 }
2686 target_namelen)))
2693 resblks);
2703 }
2705 /*
2706 * If this is a synchronous mount, make sure that the
2707 * link transaction goes to disk before returning to
2708 * the user.
2709 */
2712 }
2718 }
2723 }
2725 /* Fall through to std_return with error = 0. */
2726 std_return:
2728 DM_EVENT_POSTLINK)) {
2733 }
2736 abort_return:
2738 /* FALLTHROUGH */
2739 error_return:
2743 }
2744 /*
2745 * xfs_mkdir
2746 *
2747 */
2748 STATIC int
2755 {
2765 xfs_bmap_free_t free_list;
2766 xfs_fsblock_t first_block;
2768 boolean_t dp_joined_to_trans;
2771 xfs_prid_t prid;
2773 uint resblks;
2798 }
2800 /* Return through std_return after this point. */
2810 else
2813 /*
2814 * Make sure that we have allocated dquot(s) on disk.
2815 */
2830 XFS_TRANS_PERM_LOG_RES,
2831 XFS_MKDIR_LOG_COUNT);
2832 }
2837 }
2841 /*
2842 * Check for directory link count overflow.
2843 */
2847 }
2849 /*
2850 * Reserve disk quota and the inode.
2851 */
2859 /*
2860 * create the directory inode.
2861 */
2869 }
2872 /*
2873 * Now we add the directory inode to the transaction.
2874 * We waited until now since xfs_dir_ialloc might start
2875 * a new transaction. Had we joined the transaction
2876 * earlier, the locks might have gotten released.
2877 */
2890 }
2893 /*
2894 * Bump the in memory version number of the parent directory
2895 * so that other processes accessing it will recognize that
2896 * the directory has changed.
2897 */
2903 }
2909 }
2918 /*
2919 * Attach the dquots to the new inode and modify the icount incore.
2920 */
2923 /*
2924 * If this is a synchronous mount, make sure that the
2925 * mkdir transaction goes to disk before returning to
2926 * the user.
2927 */
2930 }
2936 }
2943 }
2945 /* Fall through to std_return with error = 0 or errno from
2946 * xfs_trans_commit. */
2948 std_return:
2951 DM_EVENT_POSTCREATE)) {
2955 DM_RIGHT_NULL,
2958 }
2961 error2:
2962 error1:
2964 abort_return:
2966 error_return:
2973 }
2976 }
2979 /*
2980 * xfs_rmdir
2981 *
2982 */
2983 STATIC int
2988 {
2995 xfs_bmap_free_t free_list;
2996 xfs_fsblock_t first_block;
3003 uint resblks;
3022 }
3024 /* Return through std_return after this point. */
3028 /*
3029 * We need to get a reference to cdp before we get our log
3030 * reservation. The reason for this is that we cannot call
3031 * xfs_iget for an inode for which we do not have a reference
3032 * once we've acquired a log reservation. This is because the
3033 * inode we are trying to get might be in xfs_inactive going
3034 * for a log reservation. Since we'll have to wait for the
3035 * inactive code to complete before returning from xfs_iget,
3036 * we need to make sure that we don't have log space reserved
3037 * when we call xfs_iget. Instead we get an unlocked referece
3038 * to the inode before getting our log reservation.
3039 */
3044 }
3048 /*
3049 * Get the dquots for the inodes.
3050 */
3058 }
3062 /*
3063 * We try to get the real space reservation first,
3064 * allowing for directory btree deletion(s) implying
3065 * possible bmap insert(s). If we can't get the space
3066 * reservation then we use 0 instead, and avoid the bmap
3067 * btree insert(s) in the directory code by, if the bmap
3068 * insert tries to happen, instead trimming the LAST
3069 * block from the directory.
3070 */
3078 }
3084 }
3087 /*
3088 * Now lock the child directory inode and the parent directory
3089 * inode in the proper order. This will take care of validating
3090 * that the directory entry for the child directory inode has
3091 * not changed while we were obtaining a log reservation.
3092 */
3098 }
3102 /*
3103 * Only increment the parent directory vnode count if
3104 * we didn't bump it in looking up cdp. The only time
3105 * we don't bump it is when we're looking up ".".
3106 */
3108 }
3117 }
3121 }
3127 }
3131 /*
3132 * Bump the in memory generation count on the parent
3133 * directory so that other can know that it has changed.
3134 */
3137 /*
3138 * Drop the link from cdp's "..".
3139 */
3143 }
3145 /*
3146 * Drop the link from dp to cdp.
3147 */
3151 }
3153 /*
3154 * Drop the "." link from cdp to self.
3155 */
3159 }
3161 /* Determine these before committing transaction */
3164 /*
3165 * Take an extra ref on the child vnode so that it
3166 * does not go to xfs_inactive() from within the commit.
3167 */
3170 /*
3171 * If this is a synchronous mount, make sure that the
3172 * rmdir transaction goes to disk before returning to
3173 * the user.
3174 */
3177 }
3183 XFS_TRANS_ABORT));
3186 }
3192 }
3195 /*
3196 * Let interposed file systems know about removed links.
3197 */
3202 /* Fall through to std_return with error = 0 or the errno
3203 * from xfs_trans_commit. */
3204 std_return:
3211 }
3214 error1:
3217 error_return:
3220 }
3223 /*
3224 * xfs_readdir
3225 *
3226 * Read dp's entries starting at uiop->uio_offset and translate them into
3227 * bufsize bytes worth of struct dirents starting at bufbase.
3228 */
3229 STATIC int
3235 {
3239 uint lock_mode;
3240 xfs_off_t start_offset;
3248 }
3255 }
3258 }
3261 /*
3262 * xfs_symlink
3263 *
3264 */
3265 STATIC int
3273 {
3280 xfs_bmap_free_t free_list;
3281 xfs_fsblock_t first_block;
3282 boolean_t dp_joined_to_trans;
3284 uint cancel_flags;
3286 xfs_fileoff_t first_fsb;
3287 xfs_filblks_t fs_blocks;
3290 xfs_daddr_t d;
3295 xfs_prid_t prid;
3297 uint resblks;
3318 /*
3319 * Check component lengths of the target path name.
3320 */
3329 /*
3330 * Skip any slashes.
3331 */
3333 total++;
3334 path++;
3335 }
3337 /*
3338 * Count up to the next slash or end of path.
3339 * Error out if the component is bigger than MAXNAMELEN.
3340 */
3345 }
3346 }
3347 }
3348 }
3356 }
3358 /* Return through std_return after this point. */
3365 else
3368 /*
3369 * Make sure that we have allocated dquot(s) on disk.
3370 */
3379 /*
3380 * The symlink will fit into the inode data fork?
3381 * There can't be any attributes so we get the whole variable part.
3382 */
3385 else
3394 }
3399 }
3403 /*
3404 * Check whether the directory allows new symlinks or not.
3405 */
3409 }
3411 /*
3412 * Reserve disk quota : blocks and inode.
3413 */
3418 /*
3419 * Check for ability to enter directory entry, if no space reserved.
3420 */
3424 /*
3425 * Initialize the bmap freelist prior to calling either
3426 * bmapi or the directory create code.
3427 */
3430 /*
3431 * Allocate an inode for the symlink.
3432 */
3439 }
3446 /*
3447 * Also attach the dquot(s) to it, if applicable.
3448 */
3453 /*
3454 * If the symlink will fit into the inode, write it inline.
3455 */
3461 /*
3462 * The inode was initially created in extent format.
3463 */
3480 }
3496 }
3503 }
3504 }
3506 /*
3507 * Create the directory entry for the symlink.
3508 */
3513 }
3517 /*
3518 * Bump the in memory version number of the parent directory
3519 * so that other processes accessing it will recognize that
3520 * the directory has changed.
3521 */
3524 /*
3525 * If this is a synchronous mount, make sure that the
3526 * symlink transaction goes to disk before returning to
3527 * the user.
3528 */
3531 }
3533 /*
3534 * xfs_trans_commit normally decrements the vnode ref count
3535 * when it unlocks the inode. Since we want to return the
3536 * vnode to the caller, we bump the vnode ref count now.
3537 */
3543 }
3548 /* Fall through to std_return with error = 0 or errno from
3549 * xfs_trans_commit */
3550 std_return:
3552 DM_EVENT_POSTSYMLINK)) {
3558 }
3566 }
3569 error2:
3571 error1:
3574 error_return:
3581 }
3584 }
3587 /*
3588 * xfs_fid2
3589 *
3590 * A fid routine that takes a pointer to a previously allocated
3591 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3592 */
3593 STATIC int
3597 {
3609 /*
3610 * use memcpy because the inode is a long long and there's no
3611 * assurance that xfid->fid_ino is properly aligned.
3612 */
3617 }
3620 /*
3621 * xfs_rwlock
3622 */
3623 int
3626 vrwlock_t locktype)
3627 {
3645 }
3648 }
3651 /*
3652 * xfs_rwunlock
3653 */
3654 void
3657 vrwlock_t locktype)
3658 {
3667 /*
3668 * In the write case, we may have added a new entry to
3669 * the reference cache. This might store a pointer to
3670 * an inode to be released in this inode. If it is there,
3671 * clear the pointer and release the inode after unlocking
3672 * this one.
3673 */
3679 }
3681 }
3683 STATIC int
3687 {
3700 /*
3701 * Bypass inodes which have already been cleaned by
3702 * the inode flush clustering code inside xfs_iflush
3703 */
3711 xfs_lsn_t sync_lsn;
3724 }
3725 }
3727 /*
3728 * We make this non-blocking if the inode is contended,
3729 * return EAGAIN to indicate to the caller that they
3730 * did not succeed. This prevents the flush path from
3731 * blocking on inodes inside another operation right
3732 * now, they get caught later by xfs_sync.
3733 */
3747 }
3750 }
3754 else
3759 }
3762 }
3765 int
3768 u_int evmask,
3769 u_int16_t state,
3771 {
3791 }
3803 }
3806 /*
3807 * xfs_reclaim
3808 */
3809 STATIC int
3812 {
3823 /* bad inode, get out here ASAP */
3827 }
3834 /* If we have nothing to flush with this inode then complete the
3835 * teardown now, otherwise break the link between the xfs inode
3836 * and the linux inode and clean up the xfs inode later. This
3837 * avoids flushing the inode to disk during the delete operation
3838 * itself.
3839 */
3847 /* Protect sync from us */
3853 }
3855 }
3857 int
3862 {
3870 /* The hash lock here protects a thread in xfs_iget_core from
3871 * racing with us on linking the inode back with a vnode.
3872 * Once we have the XFS_IRECLAIM flag set it will not touch
3873 * us.
3874 */
3882 }
3884 }
3888 /*
3889 * If the inode is still dirty, then flush it out. If the inode
3890 * is not in the AIL, then it will be OK to flush it delwri as
3891 * long as xfs_iflush() does not keep any references to the inode.
3892 * We leave that decision up to xfs_iflush() since it has the
3893 * knowledge of whether it's OK to simply do a delwri flush of
3894 * the inode or whether we need to wait until the inode is
3895 * pulled from the AIL.
3896 * We get the flush lock regardless, though, just to make sure
3897 * we don't free it while it is being flushed.
3898 */
3903 }
3909 /*
3910 * If we hit an error, typically because of filesystem
3911 * shutdown, we don't need to let vn_reclaim to know
3912 * because we're gonna reclaim the inode anyway.
3913 */
3917 }
3919 }
3926 /*
3927 * We are not interested in doing an iflush if we're
3928 * in the process of shutting down the filesystem forcibly.
3929 * So, just reclaim the inode.
3930 */
3933 }
3935 reclaim:
3938 }
3940 int
3942 {
3958 }
3959 }
3962 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3966 }
3969 }
3973 }
3975 /*
3976 * xfs_alloc_file_space()
3977 * This routine allocates disk space for the given file.
3978 *
3979 * If alloc_type == 0, this request is for an ALLOCSP type
3980 * request which will change the file size. In this case, no
3981 * DMAPI event will be generated by the call. A TRUNCATE event
3982 * will be generated later by xfs_setattr.
3983 *
3984 * If alloc_type != 0, this request is for a RESVSP type
3985 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3986 * lower block boundary byte address is less than the file's
3987 * length.
3988 *
3989 * RETURNS:
3990 * 0 on success
3991 * errno on error
3992 *
3993 */
3994 STATIC int
3997 xfs_off_t offset,
3998 xfs_off_t len,
4001 {
4002 xfs_filblks_t allocated_fsb;
4003 xfs_filblks_t allocatesize_fsb;
4005 xfs_off_t count;
4006 xfs_filblks_t datablocks;
4008 xfs_fsblock_t firstfsb;
4009 xfs_bmap_free_t free_list;
4015 uint resblks;
4018 xfs_fileoff_t startoffset_fsb;
4028 /*
4029 * determine if this is a realtime file
4030 */
4034 else
4053 /* Generate a DMAPI event if needed. */
4057 xfs_off_t end_dmi_offset;
4067 }
4069 /*
4070 * allocate file space until done or until there is an error
4071 */
4072 retry:
4074 /*
4075 * determine if reserving space on
4076 * the data or realtime partition.
4077 */
4085 rtextsize);
4091 }
4093 /*
4094 * allocate and setup the transaction
4095 */
4099 resblks,
4101 numrtextents,
4102 XFS_TRANS_PERM_LOG_RES,
4103 XFS_WRITE_LOG_COUNT);
4105 /*
4106 * check for running out of space
4107 */
4109 /*
4110 * Free the transaction structure.
4111 */
4115 }
4125 /*
4126 * issue the bmapi() call to allocate the blocks
4127 */
4135 }
4137 /*
4138 * complete the transaction
4139 */
4143 }
4149 }
4156 }
4160 }
4161 dmapi_enospc_check:
4171 /* else fall through with error from XFS_SEND_DATA */
4172 }
4176 error0:
4178 error1:
4182 }
4184 /*
4185 * Zero file bytes between startoff and endoff inclusive.
4186 * The iolock is held exclusive and no blocks are buffered.
4187 */
4188 STATIC int
4191 xfs_off_t startoff,
4192 xfs_off_t endoff)
4193 {
4194 xfs_bmbt_irec_t imap;
4195 xfs_fileoff_t offset_fsb;
4196 xfs_off_t lastoffset;
4197 xfs_off_t offset;
4233 }
4245 }
4246 }
4249 }
4251 /*
4252 * xfs_free_file_space()
4253 * This routine frees disk space for the given file.
4254 *
4255 * This routine is only called by xfs_change_file_space
4256 * for an UNRESVSP type call.
4257 *
4258 * RETURNS:
4259 * 0 on success
4260 * errno on error
4261 *
4262 */
4263 STATIC int
4266 xfs_off_t offset,
4267 xfs_off_t len,
4269 {
4273 xfs_off_t end_dmi_offset;
4274 xfs_fileoff_t endoffset_fsb;
4276 xfs_fsblock_t firstfsb;
4277 xfs_bmap_free_t free_list;
4278 xfs_off_t ilen;
4279 xfs_bmbt_irec_t imap;
4280 xfs_off_t ioffset;
4284 uint resblks;
4287 xfs_fileoff_t startoffset_fsb;
4317 }
4337 }
4339 /*
4340 * Need to zero the stuff we're not freeing, on disk.
4341 * If its a realtime file & can't use unwritten extents then we
4342 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4343 * will take care of it for us.
4344 */
4353 xfs_daddr_t block;
4360 }
4369 mod++;
4372 }
4373 }
4375 /*
4376 * One contiguous piece to clear
4377 */
4380 /*
4381 * Some full blocks, possibly two pieces to clear
4382 */
4391 }
4393 /*
4394 * free file space until done or until there is an error
4395 */
4399 /*
4400 * allocate and setup the transaction
4401 */
4404 resblks,
4407 XFS_TRANS_PERM_LOG_RES,
4408 XFS_WRITE_LOG_COUNT);
4410 /*
4411 * check for running out of space
4412 */
4414 /*
4415 * Free the transaction structure.
4416 */
4420 }
4431 /*
4432 * issue the bunmapi() call to free the blocks
4433 */
4440 }
4442 /*
4443 * complete the transaction
4444 */
4448 }
4452 }
4454 out_unlock_iolock:
4459 error0:
4461 error1:
4464 XFS_ILOCK_EXCL);
4466 }
4468 /*
4469 * xfs_change_file_space()
4470 * This routine allocates or frees disk space for the given file.
4471 * The user specified parameters are checked for alignment and size
4472 * limitations.
4473 *
4474 * RETURNS:
4475 * 0 on success
4476 * errno on error
4477 *
4478 */
4479 int
4484 xfs_off_t offset,
4487 {
4490 xfs_fsize_t fsize;
4494 xfs_off_t startoffset;
4495 xfs_off_t llen;
4497 vattr_t va;
4506 /*
4507 * must be a regular file and have write permission
4508 */
4517 }
4532 }
4547 /*
4548 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4549 * file space.
4550 * These calls do NOT zero the data space allocated to the file,
4551 * nor do they change the file size.
4552 *
4553 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4554 * space.
4555 * These calls cause the new file data to be zeroed and the file
4556 * size to be changed.
4557 */
4573 attr_flags)))
4586 }
4602 }
4604 /*
4605 * update the inode timestamp, mode, and prealloc flag bits
4606 */
4611 /* ASSERT(0); */
4614 }
4624 /*
4625 * Note that we don't have to worry about mandatory
4626 * file locking being disabled here because we only
4627 * clear the S_ISGID bit if the Group execute bit is
4628 * on, but if it was on then mandatory locking wouldn't
4629 * have been enabled.
4630 */
4635 }
4649 }
4651 vnodeops_t xfs_vnodeops = {
4655 #ifdef HAVE_SENDFILE
4657 #endif
4691 };