| blob | 322ba094dcc85bfd41a911028625ed9a56bdedf1 |
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_setattr
80 */
81 int
87 {
92 uint lock_flags;
107 /*
108 * Cannot set certain attributes.
109 */
113 }
118 /*
119 * Timestamps do not need to be logged and hence do not
120 * need to be done within a transaction.
121 */
129 }
134 /*
135 * If disk quotas is on, we make sure that the dquots do exist on disk,
136 * before we start any other transactions. Trying to do this later
137 * is messy. We don't care to take a readlock to look at the ids
138 * in inode here, because we can't hold it across the trans_reserve.
139 * If the IDs do change before we take the ilock, we're covered
140 * because the i_*dquot fields will get updated anyway.
141 */
151 }
157 }
163 }
164 /*
165 * We take a reference when we initialize udqp and gdqp,
166 * so it is important that we never blindly double trip on
167 * the same variable. See xfs_create() for an example.
168 */
175 }
177 /*
178 * For the other attributes, we acquire the inode lock and
179 * first do an error checking pass.
180 */
195 }
196 }
206 }
207 }
210 }
214 /* boolean: are we the file owner? */
217 /*
218 * Change various properties of a file.
219 * Only the owner or users with CAP_FOWNER
220 * capability may do these things.
221 */
225 /*
226 * CAP_FOWNER overrides the following restrictions:
227 *
228 * The user ID of the calling process must be equal
229 * to the file owner ID, except in cases where the
230 * CAP_FSETID capability is applicable.
231 */
235 }
237 /*
238 * CAP_FSETID overrides the following restrictions:
239 *
240 * The effective user ID of the calling process shall match
241 * the file owner when setting the set-user-ID and
242 * set-group-ID bits on that file.
243 *
244 * The effective group ID or one of the supplementary group
245 * IDs of the calling process shall match the group owner of
246 * the file when setting the set-group-ID bit on that file
247 */
256 #if 0
257 /* Linux allows this, Irix doesn't. */
260 #endif
263 }
264 }
266 /*
267 * Change file ownership. Must be the owner or privileged.
268 * If the system was configured with the "restricted_chown"
269 * option, the owner is not permitted to give away the file,
270 * and can change the group id only to a group of which he
271 * or she is a member.
272 */
274 /*
275 * These IDs could have changed since we last looked at them.
276 * But, we're assured that if the ownership did change
277 * while we didn't have the inode locked, inode's dquot(s)
278 * would have changed also.
279 */
286 iprojid;
288 /*
289 * CAP_CHOWN overrides the following restrictions:
290 *
291 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
292 * shall override the restriction that a process cannot
293 * change the user ID of a file it owns and the restriction
294 * that the group ID supplied to the chown() function
295 * shall be equal to either the group ID or one of the
296 * supplementary group IDs of the calling process.
297 */
304 }
305 /*
306 * Do a quota reservation only if uid/projid/gid is actually
307 * going to change.
308 */
318 }
319 }
321 /*
322 * Truncate file. Must have write permission and not be a directory.
323 */
325 /* Short circuit the truncate case for zero length files */
334 }
342 }
343 /*
344 * Make sure that the dquots are attached to the inode.
345 */
348 }
350 /*
351 * Change file access or modified times.
352 */
359 }
360 }
361 }
363 /*
364 * Change extent size or realtime flag.
365 */
367 /*
368 * Can't change extent size if any extents are allocated.
369 */
375 }
377 /*
378 * Can't change realtime flag if any extents are allocated.
379 */
386 }
387 /*
388 * Extent size must be a multiple of the appropriate block
389 * size, if set at all.
390 */
392 xfs_extlen_t size;
401 }
405 }
406 }
407 /*
408 * If realtime flag is set then must have realtime data.
409 */
417 }
418 }
420 /*
421 * Can't modify an immutable/append-only file unless
422 * we have appropriate permission.
423 */
432 }
433 }
435 /*
436 * Now we can make the changes. Before we join the inode
437 * to the transaction, if XFS_AT_SIZE is set then take care of
438 * the part of the truncation that must be done without the
439 * inode lock. This needs to be done before joining the inode
440 * to the transaction, because the inode cannot be unlocked
441 * once it is a part of the transaction.
442 */
448 }
451 /*
452 * We are going to log the inode size change in this
453 * transaction so any previous writes that are beyond the on
454 * disk EOF and the new EOF that have not been written out need
455 * to be written here. If we do not write the data out, we
456 * expose ourselves to the null files problem.
457 *
458 * Only flush from the on disk size to the smaller of the in
459 * memory file size or the new size as that's the range we
460 * really care about here and prevents waiting for other data
461 * not within the range we care about here.
462 */
469 }
471 /* wait for all I/O to complete */
481 }
485 XFS_TRANS_PERM_LOG_RES,
486 XFS_ITRUNCATE_LOG_COUNT))) {
491 }
494 }
499 }
501 /*
502 * Truncate file. Must have write permission and not be a directory.
503 */
505 /*
506 * Only change the c/mtime if we are changing the size
507 * or we are explicitly asked to change it. This handles
508 * the semantic difference between truncate() and ftruncate()
509 * as implemented in the VFS.
510 */
519 /*
520 * signal a sync transaction unless
521 * we're truncating an already unlinked
522 * file on a wsync filesystem
523 */
526 XFS_DATA_FORK,
532 /*
533 * Truncated "down", so we're removing references
534 * to old data here - if we now delay flushing for
535 * a long time, we expose ourselves unduly to the
536 * notorious NULL files problem. So, we mark this
537 * vnode and flush it when the file is closed, and
538 * do not wait the usual (long) time for writeout.
539 */
541 }
542 }
544 /*
545 * Change file access modes.
546 */
553 }
555 /*
556 * Change file ownership. Must be the owner or privileged.
557 * If the system was configured with the "restricted_chown"
558 * option, the owner is not permitted to give away the file,
559 * and can change the group id only to a group of which he
560 * or she is a member.
561 */
563 /*
564 * CAP_FSETID overrides the following restrictions:
565 *
566 * The set-user-ID and set-group-ID bits of a file will be
567 * cleared upon successful return from chown()
568 */
572 }
574 /*
575 * Change the ownerships and register quota modifications
576 * in the transaction.
577 */
584 }
586 }
594 }
596 }
604 }
606 /*
607 * We may have to rev the inode as well as
608 * the superblock version number since projids didn't
609 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
610 */
613 }
617 }
620 /*
621 * Change file access or modified times.
622 */
629 }
635 }
638 }
640 /*
641 * Change XFS-added attributes.
642 */
645 /*
646 * Converting bytes to fs blocks.
647 */
650 }
652 uint di_flags;
654 /* can't set PREALLOC this way, just preserve it */
684 }
686 }
689 }
691 /*
692 * Change file inode change time only if XFS_AT_CTIME set
693 * AND we have been called by a DMI function.
694 */
701 }
703 /*
704 * Send out timestamp changes that need to be set to the
705 * current time. Not done when called by a DMI function.
706 */
712 /*
713 * If this is a synchronous mount, make sure that the
714 * transaction goes to disk before returning to the user.
715 * This is slightly sub-optimal in that truncates require
716 * two sync transactions instead of one for wsync filesystems.
717 * One for the truncate and one for the timestamps since we
718 * don't want to change the timestamps unless we're sure the
719 * truncate worked. Truncates are less than 1% of the laddis
720 * mix so this probably isn't worth the trouble to optimize.
721 */
728 }
732 /*
733 * Release any dquot(s) the inode had kept before chown.
734 */
742 }
749 }
752 abort_return:
754 /* FALLTHROUGH */
755 error_return:
760 }
763 }
765 }
767 /*
768 * The maximum pathlen is 1024 bytes. Since the minimum file system
769 * blocksize is 512 bytes, we can get a max of 2 extents back from
770 * bmapi.
771 */
772 #define SYMLINK_MAPS 2
774 STATIC int
778 {
783 xfs_daddr_t d;
805 }
812 }
817 out:
819 }
821 int
825 {
849 }
851 out:
854 }
856 /*
857 * xfs_fsync
858 *
859 * This is called to sync the inode and its data out to disk.
860 * We need to hold the I/O lock while flushing the data, and
861 * the inode lock while flushing the inode. The inode lock CANNOT
862 * be held while flushing the data, so acquire after we're done
863 * with that.
864 */
865 int
869 xfs_off_t start,
870 xfs_off_t stop)
871 {
886 /*
887 * We always need to make sure that the required inode state
888 * is safe on disk. The vnode might be clean but because
889 * of committed transactions that haven't hit the disk yet.
890 * Likewise, there could be unflushed non-transactional
891 * changes to the inode core that have to go to disk.
892 *
893 * The following code depends on one assumption: that
894 * any transaction that changes an inode logs the core
895 * because it has to change some field in the inode core
896 * (typically nextents or nblocks). That assumption
897 * implies that any transactions against an inode will
898 * catch any non-transactional updates. If inode-altering
899 * transactions exist that violate this assumption, the
900 * code breaks. Right now, it figures that if the involved
901 * update_* field is clear and the inode is unpinned, the
902 * inode is clean. Either it's been flushed or it's been
903 * committed and the commit has hit the disk unpinning the inode.
904 * (Note that xfs_inode_item_format() called at commit clears
905 * the update_* fields.)
906 */
909 /* If we are flushing data then we care about update_size
910 * being set, otherwise we care about update_core
911 */
915 /*
916 * Timestamps/size haven't changed since last inode
917 * flush or inode transaction commit. That means
918 * either nothing got written or a transaction
919 * committed which caught the updates. If the
920 * latter happened and the transaction hasn't
921 * hit the disk yet, the inode will be still
922 * be pinned. If it is, force the log.
923 */
929 XFS_LOG_FORCE |
934 /*
935 * If the inode is not pinned and nothing
936 * has changed we don't need to flush the
937 * cache.
938 */
940 }
943 /*
944 * Kick off a transaction to log the inode
945 * core to get the updates. Make it
946 * sync if FSYNC_WAIT is passed in (which
947 * is done by everybody but specfs). The
948 * sync transaction will also force the log.
949 */
957 }
960 /*
961 * Note - it's possible that we might have pushed
962 * ourselves out of the way during trans_reserve
963 * which would flush the inode. But there's no
964 * guarantee that the inode buffer has actually
965 * gone out yet (it's delwri). Plus the buffer
966 * could be pinned anyway if it's part of an
967 * inode in another recent transaction. So we
968 * play it safe and fire off the transaction anyway.
969 */
978 }
981 /*
982 * If the log write didn't issue an ordered tag we need
983 * to flush the disk cache for the data device now.
984 */
988 /*
989 * If this inode is on the RT dev we need to flush that
990 * cache as well.
991 */
994 }
997 }
999 /*
1000 * This is called by xfs_inactive to free any blocks beyond eof
1001 * when the link count isn't zero and by xfs_dm_punch_hole() when
1002 * punching a hole to EOF.
1003 */
1004 int
1009 {
1012 xfs_fileoff_t end_fsb;
1013 xfs_fileoff_t last_fsb;
1014 xfs_filblks_t map_len;
1016 xfs_bmbt_irec_t imap;
1019 /*
1020 * Figure out if there are any blocks beyond the end
1021 * of the file. If not, then there is nothing to do.
1022 */
1038 /*
1039 * Attach the dquots to the inode up front.
1040 */
1044 /*
1045 * There are blocks after the end of file.
1046 * Free them up now by truncating the file to
1047 * its current size.
1048 */
1051 /*
1052 * Do the xfs_itruncate_start() call before
1053 * reserving any log space because
1054 * itruncate_start will call into the buffer
1055 * cache and we can't
1056 * do that within a transaction.
1057 */
1067 }
1072 XFS_ITRUNCATE_LOG_COUNT);
1078 }
1082 XFS_IOLOCK_EXCL |
1083 XFS_ILOCK_EXCL);
1088 XFS_DATA_FORK,
1090 /*
1091 * If we get an error at this point we
1092 * simply don't bother truncating the file.
1093 */
1097 XFS_TRANS_ABORT));
1100 XFS_TRANS_RELEASE_LOG_RES);
1101 }
1104 }
1106 }
1108 /*
1109 * Free a symlink that has blocks associated with it.
1110 */
1111 STATIC int
1115 {
1120 xfs_fsblock_t first_block;
1121 xfs_bmap_free_t free_list;
1133 /*
1134 * We're freeing a symlink that has some
1135 * blocks allocated to it. Free the
1136 * blocks here. We know that we've got
1137 * either 1 or 2 extents and that we can
1138 * free them all in one bunmapi call.
1139 */
1147 }
1148 /*
1149 * Lock the inode, fix the size, and join it to the transaction.
1150 * Hold it so in the normal path, we still have it locked for
1151 * the second transaction. In the error paths we need it
1152 * held so the cancel won't rele it, see below.
1153 */
1160 /*
1161 * Find the block(s) so we can inval and unmap them.
1162 */
1170 /*
1171 * Invalidate the block(s).
1172 */
1178 }
1179 /*
1180 * Unmap the dead block(s) to the free_list.
1181 */
1186 /*
1187 * Commit the first transaction. This logs the EFI and the inode.
1188 */
1191 /*
1192 * The transaction must have been committed, since there were
1193 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1194 * The new tp has the extent freeing and EFDs.
1195 */
1197 /*
1198 * The first xact was committed, so add the inode to the new one.
1199 * Mark it dirty so it will be logged and moved forward in the log as
1200 * part of every commit.
1201 */
1205 /*
1206 * Get a new, empty transaction to return to our caller.
1207 */
1209 /*
1210 * Commit the transaction containing extent freeing and EFDs.
1211 * If we get an error on the commit here or on the reserve below,
1212 * we need to unlock the inode since the new transaction doesn't
1213 * have the inode attached.
1214 */
1220 }
1221 /*
1222 * Remove the memory for extent descriptions (just bookkeeping).
1223 */
1227 /*
1228 * Put an itruncate log reservation in the new transaction
1229 * for our caller.
1230 */
1235 }
1236 /*
1237 * Return with the inode locked but not joined to the transaction.
1238 */
1242 error1:
1244 error0:
1245 /*
1246 * Have to come here with the inode locked and either
1247 * (held and in the transaction) or (not in the transaction).
1248 * If the inode isn't held then cancel would iput it, but
1249 * that's wrong since this is inactive and the vnode ref
1250 * count is 0 already.
1251 * Cancel won't do anything to the inode if held, but it still
1252 * needs to be locked until the cancel is done, if it was
1253 * joined to the transaction.
1254 */
1260 }
1262 STATIC int
1266 {
1270 /*
1271 * We're freeing a symlink which fit into
1272 * the inode. Just free the memory used
1273 * to hold the old symlink.
1274 */
1278 XFS_ITRUNCATE_LOG_COUNT);
1284 }
1287 /*
1288 * Zero length symlinks _can_ exist.
1289 */
1293 XFS_DATA_FORK);
1295 }
1297 }
1299 STATIC int
1303 {
1325 XFS_INACTIVE_LOG_COUNT);
1339 error_cancel:
1342 error_unlock:
1346 }
1348 int
1351 {
1359 /* If this is a read-only mount, don't do this (would generate I/O) */
1366 /*
1367 * If we are using filestreams, and we have an unlinked
1368 * file that we are processing the last close on, then nothing
1369 * will be able to reopen and write to this file. Purge this
1370 * inode from the filestreams cache so that it doesn't delay
1371 * teardown of the inode.
1372 */
1376 /*
1377 * If we previously truncated this file and removed old data
1378 * in the process, we want to initiate "early" writeout on
1379 * the last close. This is an attempt to combat the notorious
1380 * NULL files problem which is particularly noticable from a
1381 * truncate down, buffered (re-)write (delalloc), followed by
1382 * a crash. What we are effectively doing here is
1383 * significantly reducing the time window where we'd otherwise
1384 * be exposed to that problem.
1385 */
1389 }
1401 }
1402 }
1405 }
1407 /*
1408 * xfs_inactive
1409 *
1410 * This is called when the vnode reference count for the vnode
1411 * goes to zero. If the file has been unlinked, then it must
1412 * now be truncated. Also, we clear all of the read-ahead state
1413 * kept for the inode here since the file is now closed.
1414 */
1415 int
1418 {
1420 xfs_bmap_free_t free_list;
1421 xfs_fsblock_t first_block;
1430 /*
1431 * If the inode is already free, then there can be nothing
1432 * to clean up here.
1433 */
1438 }
1440 /*
1441 * Only do a truncate if it's a regular file with
1442 * some actual space in it. It's OK to look at the
1443 * inode's fields without the lock because we're the
1444 * only one with a reference to the inode.
1445 */
1458 /* If this is a read-only mount, don't do this (would generate I/O) */
1473 }
1475 }
1484 /*
1485 * Do the xfs_itruncate_start() call before
1486 * reserving any log space because itruncate_start
1487 * will call into the buffer cache and we can't
1488 * do that within a transaction.
1489 */
1497 }
1502 XFS_ITRUNCATE_LOG_COUNT);
1504 /* Don't call itruncate_cleanup */
1509 }
1515 /*
1516 * normally, we have to run xfs_itruncate_finish sync.
1517 * But if filesystem is wsync and we're in the inactive
1518 * path, then we know that nlink == 0, and that the
1519 * xaction that made nlink == 0 is permanently committed
1520 * since xfs_remove runs as a synchronous transaction.
1521 */
1530 }
1533 /*
1534 * If we get an error while cleaning up a
1535 * symlink we bail out.
1536 */
1544 }
1552 XFS_INACTIVE_LOG_COUNT);
1557 }
1562 }
1564 /*
1565 * If there are attributes associated with the file
1566 * then blow them away now. The code calls a routine
1567 * that recursively deconstructs the attribute fork.
1568 * We need to just commit the current transaction
1569 * because we can't use it for xfs_attr_inactive().
1570 */
1573 /*
1574 * If we got an error, the transaction is already
1575 * cancelled, and the inode is unlocked. Just get out.
1576 */
1581 }
1583 /*
1584 * Free the inode.
1585 */
1589 /*
1590 * If we fail to free the inode, shut down. The cancel
1591 * might do that, we need to make sure. Otherwise the
1592 * inode might be lost for a long time or forever.
1593 */
1599 }
1602 /*
1603 * Credit the quota account(s). The inode is gone.
1604 */
1607 /*
1608 * Just ignore errors at this point. There is nothing we can
1609 * do except to try to keep going. Make sure it's not a silent
1610 * error.
1611 */
1620 }
1621 /*
1622 * Release the dquots held by inode, if any.
1623 */
1628 out:
1630 }
1633 int
1638 {
1639 xfs_ino_t inum;
1641 uint lock_mode;
1662 out:
1665 }
1667 int
1671 mode_t mode,
1672 xfs_dev_t rdev,
1675 {
1680 xfs_bmap_free_t free_list;
1681 xfs_fsblock_t first_block;
1684 uint cancel_flags;
1686 xfs_prid_t prid;
1688 uint resblks;
1702 }
1707 /* Return through std_return after this point. */
1712 else
1715 /*
1716 * Make sure that we have allocated dquot(s) on disk.
1717 */
1729 /*
1730 * Initially assume that the file does not exist and
1731 * reserve the resources for that case. If that is not
1732 * the case we'll drop the one we have and get a more
1733 * appropriate transaction later.
1734 */
1741 }
1745 }
1754 /*
1755 * Reserve disk quota and the inode.
1756 */
1771 }
1774 /*
1775 * At this point, we've gotten a newly allocated inode.
1776 * It is locked (and joined to the transaction).
1777 */
1781 /*
1782 * Now we join the directory inode to the transaction. We do not do it
1783 * earlier because xfs_dir_ialloc might commit the previous transaction
1784 * (and release all the locks). An error from here on will result in
1785 * the transaction cancel unlocking dp so don't do it explicitly in the
1786 * error path.
1787 */
1798 }
1802 /*
1803 * If this is a synchronous mount, make sure that the
1804 * create transaction goes to disk before returning to
1805 * the user.
1806 */
1809 }
1813 /*
1814 * Attach the dquot(s) to the inodes and modify them incore.
1815 * These ids of the inode couldn't have changed since the new
1816 * inode has been locked ever since it was created.
1817 */
1820 /*
1821 * xfs_trans_commit normally decrements the vnode ref count
1822 * when it unlocks the inode. Since we want to return the
1823 * vnode to the caller, we bump the vnode ref count now.
1824 */
1831 }
1838 }
1845 /* Fallthrough to std_return with error = 0 */
1847 std_return:
1855 }
1858 abort_return:
1860 /* FALLTHROUGH */
1862 error_return:
1874 abort_rele:
1875 /*
1876 * Wait until after the current transaction is aborted to
1877 * release the inode. This prevents recursive transactions
1878 * and deadlocks from xfs_inactive.
1879 */
1888 }
1890 #ifdef DEBUG
1891 /*
1892 * Some counters to see if (and how often) we are hitting some deadlock
1893 * prevention code paths.
1894 */
1899 #endif
1901 /*
1902 * The following routine will lock the inodes associated with the
1903 * directory and the named entry in the directory. The locks are
1904 * acquired in increasing inode number.
1905 *
1906 * If the entry is "..", then only the directory is locked. The
1907 * vnode ref count will still include that from the .. entry in
1908 * this case.
1909 *
1910 * There is a deadlock we need to worry about. If the locked directory is
1911 * in the AIL, it might be blocking up the log. The next inode we lock
1912 * could be already locked by another thread waiting for log space (e.g
1913 * a permanent log reservation with a long running transaction (see
1914 * xfs_itruncate_finish)). To solve this, we must check if the directory
1915 * is in the ail and use lock_nowait. If we can't lock, we need to
1916 * drop the inode lock on the directory and try again. xfs_iunlock will
1917 * potentially push the tail if we were holding up the log.
1918 */
1919 STATIC int
1923 {
1925 xfs_ino_t e_inum;
1929 #ifdef DEBUG
1930 xfs_rm_locks++;
1931 #endif
1934 again:
1941 /*
1942 * We want to lock in increasing inum. Since we've already
1943 * acquired the lock on the directory, we may need to release
1944 * if if the inum of the entry turns out to be less.
1945 */
1947 /*
1948 * We are already in the right order, so just
1949 * lock on the inode of the entry.
1950 * We need to use nowait if dp is in the AIL.
1951 */
1956 attempts++;
1957 #ifdef DEBUG
1958 xfs_rm_attempts++;
1959 #endif
1961 /*
1962 * Unlock dp and try again.
1963 * xfs_iunlock will try to push the tail
1964 * if the inode is in the AIL.
1965 */
1971 #ifdef DEBUG
1972 xfs_rm_lock_delays++;
1973 #endif
1974 }
1976 }
1979 }
1986 }
1987 /* else e_inum == dp->i_ino */
1988 /* This can happen if we're asked to lock /x/..
1989 * the entry is "..", which is also the parent directory.
1990 */
1993 }
1995 #ifdef DEBUG
2001 #endif
2003 /*
2004 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2005 * a different value
2006 */
2009 {
2016 }
2018 /*
2019 * The following routine will lock n inodes in exclusive mode.
2020 * We assume the caller calls us with the inodes in i_ino order.
2021 *
2022 * We need to detect deadlock where an inode that we lock
2023 * is in the AIL and we start waiting for another inode that is locked
2024 * by a thread in a long running transaction (such as truncate). This can
2025 * result in deadlock since the long running trans might need to wait
2026 * for the inode we just locked in order to push the tail and free space
2027 * in the log.
2028 */
2029 void
2034 uint lock_mode)
2035 {
2047 }
2049 again:
2056 /*
2057 * If try_lock is not set yet, make sure all locked inodes
2058 * are not in the AIL.
2059 * If any are, set try_lock to be used later.
2060 */
2066 try_lock++;
2067 }
2068 }
2069 }
2071 /*
2072 * If any of the previous locks we have locked is in the AIL,
2073 * we must TRY to get the second and subsequent locks. If
2074 * we can't get any, we must release all we have
2075 * and try again.
2076 */
2079 /* try_lock must be 0 if i is 0. */
2080 /*
2081 * try_lock means we have an inode locked
2082 * that is in the AIL.
2083 */
2086 attempts++;
2088 /*
2089 * Unlock all previous guys and try again.
2090 * xfs_iunlock will try to push the tail
2091 * if the inode is in the AIL.
2092 */
2096 /*
2097 * Check to see if we've already
2098 * unlocked this one.
2099 * Not the first one going back,
2100 * and the inode ptr is the same.
2101 */
2107 }
2111 #ifdef DEBUG
2112 xfs_lock_delays++;
2113 #endif
2114 }
2118 }
2121 }
2122 }
2124 #ifdef DEBUG
2130 xfs_locked_n++;
2131 }
2132 #endif
2133 }
2135 #ifdef DEBUG
2136 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2139 #define REMOVE_DEBUG_TRACE(x)
2142 int
2147 {
2151 xfs_bmap_free_t free_list;
2152 xfs_fsblock_t first_block;
2156 uint resblks;
2169 }
2171 /*
2172 * We need to get a reference to ip before we get our log
2173 * reservation. The reason for this is that we cannot call
2174 * xfs_iget for an inode for which we do not have a reference
2175 * once we've acquired a log reservation. This is because the
2176 * inode we are trying to get might be in xfs_inactive going
2177 * for a log reservation. Since we'll have to wait for the
2178 * inactive code to complete before returning from xfs_iget,
2179 * we need to make sure that we don't have log space reserved
2180 * when we call xfs_iget. Instead we get an unlocked reference
2181 * to the inode before getting our log reservation.
2182 */
2195 }
2199 /*
2200 * We try to get the real space reservation first,
2201 * allowing for directory btree deletion(s) implying
2202 * possible bmap insert(s). If we can't get the space
2203 * reservation then we use 0 instead, and avoid the bmap
2204 * btree insert(s) in the directory code by, if the bmap
2205 * insert tries to happen, instead trimming the LAST
2206 * block from the directory.
2207 */
2215 }
2222 }
2230 }
2232 /*
2233 * At this point, we've gotten both the directory and the entry
2234 * inodes locked.
2235 */
2238 /*
2239 * Increment vnode ref count only in this case since
2240 * there's an extra vnode reference in the case where
2241 * dp == ip.
2242 */
2245 }
2247 /*
2248 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2249 */
2257 }
2267 }
2269 /* Determine if this is the last link while
2270 * we are in the transaction.
2271 */
2274 /*
2275 * Take an extra ref on the inode so that it doesn't
2276 * go to xfs_inactive() from within the commit.
2277 */
2280 /*
2281 * If this is a synchronous mount, make sure that the
2282 * remove transaction goes to disk before returning to
2283 * the user.
2284 */
2287 }
2293 }
2299 }
2301 /*
2302 * If we are using filestreams, kill the stream association.
2303 * If the file is still open it may get a new one but that
2304 * will get killed on last close in xfs_close() so we don't
2305 * have to worry about that.
2306 */
2313 /* Fall through to std_return with error = 0 */
2314 std_return:
2320 }
2323 error1:
2329 error_rele:
2330 /*
2331 * In this case make sure to not release the inode until after
2332 * the current transaction is aborted. Releasing it beforehand
2333 * can cause us to go to xfs_inactive and start a recursive
2334 * transaction which can easily deadlock with the current one.
2335 */
2343 }
2345 int
2350 {
2355 xfs_bmap_free_t free_list;
2356 xfs_fsblock_t first_block;
2376 }
2378 /* Return through std_return after this point. */
2395 }
2399 }
2407 }
2411 /*
2412 * Increment vnode ref counts since xfs_trans_commit &
2413 * xfs_trans_cancel will both unlock the inodes and
2414 * decrement the associated ref counts.
2415 */
2421 /*
2422 * If the source has too many links, we can't make any more to it.
2423 */
2427 }
2429 /*
2430 * If we are using project inheritance, we only allow hard link
2431 * creation in our tree when the project IDs are the same; else
2432 * the tree quota mechanism could be circumvented.
2433 */
2438 }
2458 /*
2459 * If this is a synchronous mount, make sure that the
2460 * link transaction goes to disk before returning to
2461 * the user.
2462 */
2465 }
2471 }
2477 /* Fall through to std_return with error = 0. */
2478 std_return:
2484 }
2487 abort_return:
2489 /* FALLTHROUGH */
2491 error_return:
2494 }
2497 int
2501 mode_t mode,
2504 {
2511 xfs_bmap_free_t free_list;
2512 xfs_fsblock_t first_block;
2516 xfs_prid_t prid;
2518 uint resblks;
2533 }
2535 /* Return through std_return after this point. */
2543 else
2546 /*
2547 * Make sure that we have allocated dquot(s) on disk.
2548 */
2563 XFS_TRANS_PERM_LOG_RES,
2564 XFS_MKDIR_LOG_COUNT);
2565 }
2569 }
2574 /*
2575 * Check for directory link count overflow.
2576 */
2580 }
2582 /*
2583 * Reserve disk quota and the inode.
2584 */
2592 /*
2593 * create the directory inode.
2594 */
2602 }
2605 /*
2606 * Now we add the directory inode to the transaction.
2607 * We waited until now since xfs_dir_ialloc might start
2608 * a new transaction. Had we joined the transaction
2609 * earlier, the locks might have gotten released. An error
2610 * from here on will result in the transaction cancel
2611 * unlocking dp so don't do it explicitly in the error path.
2612 */
2625 }
2628 /*
2629 * Bump the in memory version number of the parent directory
2630 * so that other processes accessing it will recognize that
2631 * the directory has changed.
2632 */
2649 /*
2650 * Attach the dquots to the new inode and modify the icount incore.
2651 */
2654 /*
2655 * If this is a synchronous mount, make sure that the
2656 * mkdir transaction goes to disk before returning to
2657 * the user.
2658 */
2661 }
2667 }
2674 }
2676 /* Fall through to std_return with error = 0 or errno from
2677 * xfs_trans_commit. */
2679 std_return:
2685 DM_RIGHT_NULL,
2688 }
2691 error2:
2692 error1:
2694 abort_return:
2696 error_return:
2705 }
2707 int
2712 {
2717 xfs_bmap_free_t free_list;
2718 xfs_fsblock_t first_block;
2722 uint resblks;
2736 }
2738 /*
2739 * We need to get a reference to cdp before we get our log
2740 * reservation. The reason for this is that we cannot call
2741 * xfs_iget for an inode for which we do not have a reference
2742 * once we've acquired a log reservation. This is because the
2743 * inode we are trying to get might be in xfs_inactive going
2744 * for a log reservation. Since we'll have to wait for the
2745 * inactive code to complete before returning from xfs_iget,
2746 * we need to make sure that we don't have log space reserved
2747 * when we call xfs_iget. Instead we get an unlocked reference
2748 * to the inode before getting our log reservation.
2749 */
2752 /*
2753 * Get the dquots for the inodes.
2754 */
2762 }
2766 /*
2767 * We try to get the real space reservation first,
2768 * allowing for directory btree deletion(s) implying
2769 * possible bmap insert(s). If we can't get the space
2770 * reservation then we use 0 instead, and avoid the bmap
2771 * btree insert(s) in the directory code by, if the bmap
2772 * insert tries to happen, instead trimming the LAST
2773 * block from the directory.
2774 */
2782 }
2788 }
2791 /*
2792 * Now lock the child directory inode and the parent directory
2793 * inode in the proper order. This will take care of validating
2794 * that the directory entry for the child directory inode has
2795 * not changed while we were obtaining a log reservation.
2796 */
2802 }
2806 /*
2807 * Only increment the parent directory vnode count if
2808 * we didn't bump it in looking up cdp. The only time
2809 * we don't bump it is when we're looking up ".".
2810 */
2812 }
2821 }
2825 }
2834 /*
2835 * Bump the in memory generation count on the parent
2836 * directory so that other can know that it has changed.
2837 */
2840 /*
2841 * Drop the link from cdp's "..".
2842 */
2846 }
2848 /*
2849 * Drop the link from dp to cdp.
2850 */
2854 }
2856 /*
2857 * Drop the "." link from cdp to self.
2858 */
2862 }
2864 /* Determine these before committing transaction */
2867 /*
2868 * Take an extra ref on the child vnode so that it
2869 * does not go to xfs_inactive() from within the commit.
2870 */
2873 /*
2874 * If this is a synchronous mount, make sure that the
2875 * rmdir transaction goes to disk before returning to
2876 * the user.
2877 */
2880 }
2886 XFS_TRANS_ABORT));
2889 }
2895 }
2900 /* Fall through to std_return with error = 0 or the errno
2901 * from xfs_trans_commit. */
2902 std_return:
2909 }
2912 error1:
2915 /* FALLTHROUGH */
2917 error_return:
2920 }
2922 int
2927 mode_t mode,
2930 {
2936 xfs_bmap_free_t free_list;
2937 xfs_fsblock_t first_block;
2939 uint cancel_flags;
2941 xfs_fileoff_t first_fsb;
2942 xfs_filblks_t fs_blocks;
2945 xfs_daddr_t d;
2950 xfs_prid_t prid;
2952 uint resblks;
2964 /*
2965 * Check component lengths of the target path name.
2966 */
2977 }
2979 /* Return through std_return after this point. */
2984 else
2987 /*
2988 * Make sure that we have allocated dquot(s) on disk.
2989 */
2998 /*
2999 * The symlink will fit into the inode data fork?
3000 * There can't be any attributes so we get the whole variable part.
3001 */
3004 else
3013 }
3017 }
3022 /*
3023 * Check whether the directory allows new symlinks or not.
3024 */
3028 }
3030 /*
3031 * Reserve disk quota : blocks and inode.
3032 */
3037 /*
3038 * Check for ability to enter directory entry, if no space reserved.
3039 */
3043 /*
3044 * Initialize the bmap freelist prior to calling either
3045 * bmapi or the directory create code.
3046 */
3049 /*
3050 * Allocate an inode for the symlink.
3051 */
3058 }
3061 /*
3062 * An error after we've joined dp to the transaction will result in the
3063 * transaction cancel unlocking dp so don't do it explicitly in the
3064 * error path.
3065 */
3070 /*
3071 * Also attach the dquot(s) to it, if applicable.
3072 */
3077 /*
3078 * If the symlink will fit into the inode, write it inline.
3079 */
3085 /*
3086 * The inode was initially created in extent format.
3087 */
3104 }
3120 }
3127 }
3128 }
3130 /*
3131 * Create the directory entry for the symlink.
3132 */
3140 /*
3141 * Bump the in memory version number of the parent directory
3142 * so that other processes accessing it will recognize that
3143 * the directory has changed.
3144 */
3147 /*
3148 * If this is a synchronous mount, make sure that the
3149 * symlink transaction goes to disk before returning to
3150 * the user.
3151 */
3154 }
3156 /*
3157 * xfs_trans_commit normally decrements the vnode ref count
3158 * when it unlocks the inode. Since we want to return the
3159 * vnode to the caller, we bump the vnode ref count now.
3160 */
3166 }
3171 /* Fall through to std_return with error = 0 or errno from
3172 * xfs_trans_commit */
3173 std_return:
3180 }
3186 error2:
3188 error1:
3191 error_return:
3200 }
3202 int
3206 {
3213 /*
3214 * Bypass inodes which have already been cleaned by
3215 * the inode flush clustering code inside xfs_iflush
3216 */
3220 /*
3221 * We make this non-blocking if the inode is contended,
3222 * return EAGAIN to indicate to the caller that they
3223 * did not succeed. This prevents the flush path from
3224 * blocking on inodes inside another operation right
3225 * now, they get caught later by xfs_sync.
3226 */
3234 }
3237 }
3244 }
3247 int
3250 u_int evmask,
3251 u_int16_t state)
3252 {
3268 }
3280 }
3282 int
3285 {
3292 /* bad inode, get out here ASAP */
3296 }
3302 /*
3303 * Make sure the atime in the XFS inode is correct before freeing the
3304 * Linux inode.
3305 */
3308 /*
3309 * If we have nothing to flush with this inode then complete the
3310 * teardown now, otherwise break the link between the xfs inode and the
3311 * linux inode and clean up the xfs inode later. This avoids flushing
3312 * the inode to disk during the delete operation itself.
3313 *
3314 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3315 * first to ensure that xfs_iunpin() will never see an xfs inode
3316 * that has a linux inode being reclaimed. Synchronisation is provided
3317 * by the i_flags_lock.
3318 */
3326 /* Protect sync and unpin from us */
3335 }
3337 }
3339 int
3344 {
3352 /* The hash lock here protects a thread in xfs_iget_core from
3353 * racing with us on linking the inode back with a vnode.
3354 * Once we have the XFS_IRECLAIM flag set it will not touch
3355 * us.
3356 */
3366 }
3368 }
3374 /*
3375 * If the inode is still dirty, then flush it out. If the inode
3376 * is not in the AIL, then it will be OK to flush it delwri as
3377 * long as xfs_iflush() does not keep any references to the inode.
3378 * We leave that decision up to xfs_iflush() since it has the
3379 * knowledge of whether it's OK to simply do a delwri flush of
3380 * the inode or whether we need to wait until the inode is
3381 * pulled from the AIL.
3382 * We get the flush lock regardless, though, just to make sure
3383 * we don't free it while it is being flushed.
3384 */
3388 }
3395 /*
3396 * If we hit an error, typically because of filesystem
3397 * shutdown, we don't need to let vn_reclaim to know
3398 * because we're gonna reclaim the inode anyway.
3399 */
3403 }
3405 }
3410 }
3415 reclaim:
3418 }
3420 int
3422 {
3438 }
3439 }
3442 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3446 }
3449 }
3453 }
3455 /*
3456 * xfs_alloc_file_space()
3457 * This routine allocates disk space for the given file.
3458 *
3459 * If alloc_type == 0, this request is for an ALLOCSP type
3460 * request which will change the file size. In this case, no
3461 * DMAPI event will be generated by the call. A TRUNCATE event
3462 * will be generated later by xfs_setattr.
3463 *
3464 * If alloc_type != 0, this request is for a RESVSP type
3465 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3466 * lower block boundary byte address is less than the file's
3467 * length.
3468 *
3469 * RETURNS:
3470 * 0 on success
3471 * errno on error
3472 *
3473 */
3474 STATIC int
3477 xfs_off_t offset,
3478 xfs_off_t len,
3481 {
3483 xfs_off_t count;
3484 xfs_filblks_t allocated_fsb;
3485 xfs_filblks_t allocatesize_fsb;
3487 xfs_fileoff_t startoffset_fsb;
3488 xfs_fsblock_t firstfsb;
3495 xfs_bmap_free_t free_list;
3521 /* Generate a DMAPI event if needed. */
3525 xfs_off_t end_dmi_offset;
3534 }
3536 /*
3537 * Allocate file space until done or until there is an error
3538 */
3539 retry:
3543 /*
3544 * Determine space reservations for data/realtime.
3545 */
3558 }
3570 }
3572 /*
3573 * Allocate and setup the transaction.
3574 */
3578 XFS_TRANS_PERM_LOG_RES,
3579 XFS_WRITE_LOG_COUNT);
3580 /*
3581 * Check for running out of space
3582 */
3584 /*
3585 * Free the transaction structure.
3586 */
3590 }
3600 /*
3601 * Issue the xfs_bmapi() call to allocate the blocks
3602 */
3610 }
3612 /*
3613 * Complete the transaction
3614 */
3618 }
3624 }
3631 }
3635 }
3636 dmapi_enospc_check:
3645 /* else fall through with error from XFS_SEND_DATA */
3646 }
3658 }
3660 /*
3661 * Zero file bytes between startoff and endoff inclusive.
3662 * The iolock is held exclusive and no blocks are buffered.
3663 */
3664 STATIC int
3667 xfs_off_t startoff,
3668 xfs_off_t endoff)
3669 {
3670 xfs_bmbt_irec_t imap;
3671 xfs_fileoff_t offset_fsb;
3672 xfs_off_t lastoffset;
3673 xfs_off_t offset;
3710 }
3723 }
3724 }
3727 }
3729 /*
3730 * xfs_free_file_space()
3731 * This routine frees disk space for the given file.
3732 *
3733 * This routine is only called by xfs_change_file_space
3734 * for an UNRESVSP type call.
3735 *
3736 * RETURNS:
3737 * 0 on success
3738 * errno on error
3739 *
3740 */
3741 STATIC int
3744 xfs_off_t offset,
3745 xfs_off_t len,
3747 {
3751 xfs_off_t end_dmi_offset;
3752 xfs_fileoff_t endoffset_fsb;
3754 xfs_fsblock_t firstfsb;
3755 xfs_bmap_free_t free_list;
3756 xfs_bmbt_irec_t imap;
3757 xfs_off_t ioffset;
3761 uint resblks;
3762 uint rounding;
3764 xfs_fileoff_t startoffset_fsb;
3793 }
3800 }
3810 }
3812 /*
3813 * Need to zero the stuff we're not freeing, on disk.
3814 * If its a realtime file & can't use unwritten extents then we
3815 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3816 * will take care of it for us.
3817 */
3826 xfs_daddr_t block;
3833 }
3842 mod++;
3845 }
3846 }
3848 /*
3849 * One contiguous piece to clear
3850 */
3853 /*
3854 * Some full blocks, possibly two pieces to clear
3855 */
3864 }
3866 /*
3867 * free file space until done or until there is an error
3868 */
3872 /*
3873 * allocate and setup the transaction. Allow this
3874 * transaction to dip into the reserve blocks to ensure
3875 * the freeing of the space succeeds at ENOSPC.
3876 */
3880 resblks,
3883 XFS_TRANS_PERM_LOG_RES,
3884 XFS_WRITE_LOG_COUNT);
3886 /*
3887 * check for running out of space
3888 */
3890 /*
3891 * Free the transaction structure.
3892 */
3896 }
3900 XFS_QMOPT_RES_REGBLKS);
3907 /*
3908 * issue the bunmapi() call to free the blocks
3909 */
3916 }
3918 /*
3919 * complete the transaction
3920 */
3924 }
3928 }
3930 out_unlock_iolock:
3935 error0:
3937 error1:
3940 XFS_ILOCK_EXCL);
3942 }
3944 /*
3945 * xfs_change_file_space()
3946 * This routine allocates or frees disk space for the given file.
3947 * The user specified parameters are checked for alignment and size
3948 * limitations.
3949 *
3950 * RETURNS:
3951 * 0 on success
3952 * errno on error
3953 *
3954 */
3955 int
3960 xfs_off_t offset,
3963 {
3967 xfs_fsize_t fsize;
3969 xfs_off_t startoffset;
3970 xfs_off_t llen;
3972 bhv_vattr_t va;
3990 }
4005 /*
4006 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4007 * file space.
4008 * These calls do NOT zero the data space allocated to the file,
4009 * nor do they change the file size.
4010 *
4011 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4012 * space.
4013 * These calls cause the new file data to be zeroed and the file
4014 * size to be changed.
4015 */
4031 attr_flags)))
4044 }
4060 }
4062 /*
4063 * update the inode timestamp, mode, and prealloc flag bits
4064 */
4069 /* ASSERT(0); */
4072 }
4082 /*
4083 * Note that we don't have to worry about mandatory
4084 * file locking being disabled here because we only
4085 * clear the S_ISGID bit if the Group execute bit is
4086 * on, but if it was on then mandatory locking wouldn't
4087 * have been enabled.
4088 */
4093 }
4107 }