| blob | f26b038004a74ab433b1cefb27e00d7b9a712601 |
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
61 {
67 uint lock_flags;
87 /*
88 * If disk quotas is on, we make sure that the dquots do exist on disk,
89 * before we start any other transactions. Trying to do this later
90 * is messy. We don't care to take a readlock to look at the ids
91 * in inode here, because we can't hold it across the trans_reserve.
92 * If the IDs do change before we take the ilock, we're covered
93 * because the i_*dquot fields will get updated anyway.
94 */
103 }
109 }
111 /*
112 * We take a reference when we initialize udqp and gdqp,
113 * so it is important that we never blindly double trip on
114 * the same variable. See xfs_create() for an example.
115 */
122 }
124 /*
125 * For the other attributes, we acquire the inode lock and
126 * first do an error checking pass.
127 */
142 }
143 }
153 }
154 }
157 }
161 /* boolean: are we the file owner? */
164 /*
165 * Change various properties of a file.
166 * Only the owner or users with CAP_FOWNER
167 * capability may do these things.
168 */
170 /*
171 * CAP_FOWNER overrides the following restrictions:
172 *
173 * The user ID of the calling process must be equal
174 * to the file owner ID, except in cases where the
175 * CAP_FSETID capability is applicable.
176 */
180 }
182 /*
183 * CAP_FSETID overrides the following restrictions:
184 *
185 * The effective user ID of the calling process shall match
186 * the file owner when setting the set-user-ID and
187 * set-group-ID bits on that file.
188 *
189 * The effective group ID or one of the supplementary group
190 * IDs of the calling process shall match the group owner of
191 * the file when setting the set-group-ID bit on that file
192 */
201 #if 0
202 /* Linux allows this, Irix doesn't. */
205 #endif
208 }
209 }
211 /*
212 * Change file ownership. Must be the owner or privileged.
213 */
215 /*
216 * These IDs could have changed since we last looked at them.
217 * But, we're assured that if the ownership did change
218 * while we didn't have the inode locked, inode's dquot(s)
219 * would have changed also.
220 */
226 /*
227 * CAP_CHOWN overrides the following restrictions:
228 *
229 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
230 * shall override the restriction that a process cannot
231 * change the user ID of a file it owns and the restriction
232 * that the group ID supplied to the chown() function
233 * shall be equal to either the group ID or one of the
234 * supplementary group IDs of the calling process.
235 */
241 }
242 /*
243 * Do a quota reservation only if uid/gid is actually
244 * going to change.
245 */
254 }
255 }
257 /*
258 * Truncate file. Must have write permission and not be a directory.
259 */
261 /* Short circuit the truncate case for zero length files */
270 }
278 }
279 /*
280 * Make sure that the dquots are attached to the inode.
281 */
284 }
286 /*
287 * Change file access or modified times.
288 */
295 }
296 }
297 }
299 /*
300 * Now we can make the changes. Before we join the inode
301 * to the transaction, if ATTR_SIZE is set then take care of
302 * the part of the truncation that must be done without the
303 * inode lock. This needs to be done before joining the inode
304 * to the transaction, because the inode cannot be unlocked
305 * once it is a part of the transaction.
306 */
310 /*
311 * Do the first part of growing a file: zero any data
312 * in the last block that is beyond the old EOF. We
313 * need to do this before the inode is joined to the
314 * transaction to modify the i_size.
315 */
317 }
320 /*
321 * We are going to log the inode size change in this
322 * transaction so any previous writes that are beyond the on
323 * disk EOF and the new EOF that have not been written out need
324 * to be written here. If we do not write the data out, we
325 * expose ourselves to the null files problem.
326 *
327 * Only flush from the on disk size to the smaller of the in
328 * memory file size or the new size as that's the range we
329 * really care about here and prevents waiting for other data
330 * not within the range we care about here.
331 */
338 }
340 /* wait for all I/O to complete */
350 }
354 XFS_TRANS_PERM_LOG_RES,
355 XFS_ITRUNCATE_LOG_COUNT))) {
360 }
363 }
368 }
370 /*
371 * Truncate file. Must have write permission and not be a directory.
372 */
374 /*
375 * Only change the c/mtime if we are changing the size
376 * or we are explicitly asked to change it. This handles
377 * the semantic difference between truncate() and ftruncate()
378 * as implemented in the VFS.
379 */
391 /*
392 * signal a sync transaction unless
393 * we're truncating an already unlinked
394 * file on a wsync filesystem
395 */
397 XFS_DATA_FORK,
403 /*
404 * Truncated "down", so we're removing references
405 * to old data here - if we now delay flushing for
406 * a long time, we expose ourselves unduly to the
407 * notorious NULL files problem. So, we mark this
408 * vnode and flush it when the file is closed, and
409 * do not wait the usual (long) time for writeout.
410 */
412 }
413 }
415 /*
416 * Change file access modes.
417 */
427 }
429 /*
430 * Change file ownership. Must be the owner or privileged.
431 */
433 /*
434 * CAP_FSETID overrides the following restrictions:
435 *
436 * The set-user-ID and set-group-ID bits of a file will be
437 * cleared upon successful return from chown()
438 */
442 }
444 /*
445 * Change the ownerships and register quota modifications
446 * in the transaction.
447 */
454 }
457 }
465 }
468 }
472 }
475 /*
476 * Change file access or modified times.
477 */
484 }
491 }
494 }
496 /*
497 * Change file inode change time only if ATTR_CTIME set
498 * AND we have been called by a DMI function.
499 */
507 }
509 /*
510 * Send out timestamp changes that need to be set to the
511 * current time. Not done when called by a DMI function.
512 */
518 /*
519 * If this is a synchronous mount, make sure that the
520 * transaction goes to disk before returning to the user.
521 * This is slightly sub-optimal in that truncates require
522 * two sync transactions instead of one for wsync filesystems.
523 * One for the truncate and one for the timestamps since we
524 * don't want to change the timestamps unless we're sure the
525 * truncate worked. Truncates are less than 1% of the laddis
526 * mix so this probably isn't worth the trouble to optimize.
527 */
534 }
538 /*
539 * Release any dquot(s) the inode had kept before chown.
540 */
548 }
555 }
558 abort_return:
560 /* FALLTHROUGH */
561 error_return:
566 }
569 }
571 }
573 /*
574 * The maximum pathlen is 1024 bytes. Since the minimum file system
575 * blocksize is 512 bytes, we can get a max of 2 extents back from
576 * bmapi.
577 */
578 #define SYMLINK_MAPS 2
580 STATIC int
584 {
589 xfs_daddr_t d;
611 }
618 }
623 out:
625 }
627 int
631 {
655 }
657 out:
660 }
662 /*
663 * xfs_fsync
664 *
665 * This is called to sync the inode and its data out to disk. We need to hold
666 * the I/O lock while flushing the data, and the inode lock while flushing the
667 * inode. The inode lock CANNOT be held while flushing the data, so acquire
668 * after we're done with that.
669 */
670 int
673 {
683 /* capture size updates in I/O completion before writing the inode. */
688 /*
689 * We always need to make sure that the required inode state is safe on
690 * disk. The vnode might be clean but we still might need to force the
691 * log because of committed transactions that haven't hit the disk yet.
692 * Likewise, there could be unflushed non-transactional changes to the
693 * inode core that have to go to disk and this requires us to issue
694 * a synchronous transaction to capture these changes correctly.
695 *
696 * This code relies on the assumption that if the update_* fields
697 * of the inode are clear and the inode is unpinned then it is clean
698 * and no action is required.
699 */
703 /*
704 * Timestamps/size haven't changed since last inode flush or
705 * inode transaction commit. That means either nothing got
706 * written or a transaction committed which caught the updates.
707 * If the latter happened and the transaction hasn't hit the
708 * disk yet, the inode will be still be pinned. If it is,
709 * force the log.
710 */
719 /*
720 * If the inode is not pinned and nothing has changed
721 * we don't need to flush the cache.
722 */
724 }
726 /*
727 * Kick off a transaction to log the inode core to get the
728 * updates. The sync transaction will also force the log.
729 */
737 }
740 /*
741 * Note - it's possible that we might have pushed ourselves out
742 * of the way during trans_reserve which would flush the inode.
743 * But there's no guarantee that the inode buffer has actually
744 * gone out yet (it's delwri). Plus the buffer could be pinned
745 * anyway if it's part of an inode in another recent
746 * transaction. So we play it safe and fire off the
747 * transaction anyway.
748 */
756 }
759 /*
760 * If the log write didn't issue an ordered tag we need
761 * to flush the disk cache for the data device now.
762 */
766 /*
767 * If this inode is on the RT dev we need to flush that
768 * cache as well.
769 */
772 }
775 }
777 /*
778 * This is called by xfs_inactive to free any blocks beyond eof
779 * when the link count isn't zero and by xfs_dm_punch_hole() when
780 * punching a hole to EOF.
781 */
782 int
787 {
790 xfs_fileoff_t end_fsb;
791 xfs_fileoff_t last_fsb;
792 xfs_filblks_t map_len;
794 xfs_bmbt_irec_t imap;
797 /*
798 * Figure out if there are any blocks beyond the end
799 * of the file. If not, then there is nothing to do.
800 */
816 /*
817 * Attach the dquots to the inode up front.
818 */
822 /*
823 * There are blocks after the end of file.
824 * Free them up now by truncating the file to
825 * its current size.
826 */
829 /*
830 * Do the xfs_itruncate_start() call before
831 * reserving any log space because
832 * itruncate_start will call into the buffer
833 * cache and we can't
834 * do that within a transaction.
835 */
845 }
850 XFS_ITRUNCATE_LOG_COUNT);
856 }
860 XFS_IOLOCK_EXCL |
861 XFS_ILOCK_EXCL);
866 XFS_DATA_FORK,
868 /*
869 * If we get an error at this point we
870 * simply don't bother truncating the file.
871 */
875 XFS_TRANS_ABORT));
878 XFS_TRANS_RELEASE_LOG_RES);
879 }
882 }
884 }
886 /*
887 * Free a symlink that has blocks associated with it.
888 */
889 STATIC int
893 {
898 xfs_fsblock_t first_block;
899 xfs_bmap_free_t free_list;
911 /*
912 * We're freeing a symlink that has some
913 * blocks allocated to it. Free the
914 * blocks here. We know that we've got
915 * either 1 or 2 extents and that we can
916 * free them all in one bunmapi call.
917 */
925 }
926 /*
927 * Lock the inode, fix the size, and join it to the transaction.
928 * Hold it so in the normal path, we still have it locked for
929 * the second transaction. In the error paths we need it
930 * held so the cancel won't rele it, see below.
931 */
938 /*
939 * Find the block(s) so we can inval and unmap them.
940 */
948 /*
949 * Invalidate the block(s).
950 */
956 }
957 /*
958 * Unmap the dead block(s) to the free_list.
959 */
964 /*
965 * Commit the first transaction. This logs the EFI and the inode.
966 */
969 /*
970 * The transaction must have been committed, since there were
971 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
972 * The new tp has the extent freeing and EFDs.
973 */
975 /*
976 * The first xact was committed, so add the inode to the new one.
977 * Mark it dirty so it will be logged and moved forward in the log as
978 * part of every commit.
979 */
983 /*
984 * Get a new, empty transaction to return to our caller.
985 */
987 /*
988 * Commit the transaction containing extent freeing and EFDs.
989 * If we get an error on the commit here or on the reserve below,
990 * we need to unlock the inode since the new transaction doesn't
991 * have the inode attached.
992 */
998 }
999 /*
1000 * transaction commit worked ok so we can drop the extra ticket
1001 * reference that we gained in xfs_trans_dup()
1002 */
1005 /*
1006 * Remove the memory for extent descriptions (just bookkeeping).
1007 */
1011 /*
1012 * Put an itruncate log reservation in the new transaction
1013 * for our caller.
1014 */
1019 }
1020 /*
1021 * Return with the inode locked but not joined to the transaction.
1022 */
1026 error1:
1028 error0:
1029 /*
1030 * Have to come here with the inode locked and either
1031 * (held and in the transaction) or (not in the transaction).
1032 * If the inode isn't held then cancel would iput it, but
1033 * that's wrong since this is inactive and the vnode ref
1034 * count is 0 already.
1035 * Cancel won't do anything to the inode if held, but it still
1036 * needs to be locked until the cancel is done, if it was
1037 * joined to the transaction.
1038 */
1044 }
1046 STATIC int
1050 {
1054 /*
1055 * We're freeing a symlink which fit into
1056 * the inode. Just free the memory used
1057 * to hold the old symlink.
1058 */
1062 XFS_ITRUNCATE_LOG_COUNT);
1068 }
1071 /*
1072 * Zero length symlinks _can_ exist.
1073 */
1077 XFS_DATA_FORK);
1079 }
1081 }
1083 STATIC int
1087 {
1109 XFS_INACTIVE_LOG_COUNT);
1123 error_cancel:
1126 error_unlock:
1130 }
1132 int
1135 {
1142 /* If this is a read-only mount, don't do this (would generate I/O) */
1149 /*
1150 * If we are using filestreams, and we have an unlinked
1151 * file that we are processing the last close on, then nothing
1152 * will be able to reopen and write to this file. Purge this
1153 * inode from the filestreams cache so that it doesn't delay
1154 * teardown of the inode.
1155 */
1159 /*
1160 * If we previously truncated this file and removed old data
1161 * in the process, we want to initiate "early" writeout on
1162 * the last close. This is an attempt to combat the notorious
1163 * NULL files problem which is particularly noticable from a
1164 * truncate down, buffered (re-)write (delalloc), followed by
1165 * a crash. What we are effectively doing here is
1166 * significantly reducing the time window where we'd otherwise
1167 * be exposed to that problem.
1168 */
1172 }
1184 }
1185 }
1188 }
1190 /*
1191 * xfs_inactive
1192 *
1193 * This is called when the vnode reference count for the vnode
1194 * goes to zero. If the file has been unlinked, then it must
1195 * now be truncated. Also, we clear all of the read-ahead state
1196 * kept for the inode here since the file is now closed.
1197 */
1198 int
1201 {
1202 xfs_bmap_free_t free_list;
1203 xfs_fsblock_t first_block;
1212 /*
1213 * If the inode is already free, then there can be nothing
1214 * to clean up here.
1215 */
1220 }
1222 /*
1223 * Only do a truncate if it's a regular file with
1224 * some actual space in it. It's OK to look at the
1225 * inode's fields without the lock because we're the
1226 * only one with a reference to the inode.
1227 */
1240 /* If this is a read-only mount, don't do this (would generate I/O) */
1255 }
1257 }
1266 /*
1267 * Do the xfs_itruncate_start() call before
1268 * reserving any log space because itruncate_start
1269 * will call into the buffer cache and we can't
1270 * do that within a transaction.
1271 */
1279 }
1284 XFS_ITRUNCATE_LOG_COUNT);
1286 /* Don't call itruncate_cleanup */
1291 }
1297 /*
1298 * normally, we have to run xfs_itruncate_finish sync.
1299 * But if filesystem is wsync and we're in the inactive
1300 * path, then we know that nlink == 0, and that the
1301 * xaction that made nlink == 0 is permanently committed
1302 * since xfs_remove runs as a synchronous transaction.
1303 */
1312 }
1315 /*
1316 * If we get an error while cleaning up a
1317 * symlink we bail out.
1318 */
1326 }
1334 XFS_INACTIVE_LOG_COUNT);
1339 }
1344 }
1346 /*
1347 * If there are attributes associated with the file
1348 * then blow them away now. The code calls a routine
1349 * that recursively deconstructs the attribute fork.
1350 * We need to just commit the current transaction
1351 * because we can't use it for xfs_attr_inactive().
1352 */
1355 /*
1356 * If we got an error, the transaction is already
1357 * cancelled, and the inode is unlocked. Just get out.
1358 */
1363 }
1365 /*
1366 * Free the inode.
1367 */
1371 /*
1372 * If we fail to free the inode, shut down. The cancel
1373 * might do that, we need to make sure. Otherwise the
1374 * inode might be lost for a long time or forever.
1375 */
1381 }
1384 /*
1385 * Credit the quota account(s). The inode is gone.
1386 */
1389 /*
1390 * Just ignore errors at this point. There is nothing we can
1391 * do except to try to keep going. Make sure it's not a silent
1392 * error.
1393 */
1402 }
1403 /*
1404 * Release the dquots held by inode, if any.
1405 */
1410 out:
1412 }
1414 /*
1415 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1416 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1417 * ci_name->name will point to a the actual name (caller must free) or
1418 * will be set to NULL if an exact match is found.
1419 */
1420 int
1426 {
1427 xfs_ino_t inum;
1429 uint lock_mode;
1450 out_free_name:
1453 out:
1456 }
1458 int
1462 mode_t mode,
1463 xfs_dev_t rdev,
1466 {
1471 xfs_bmap_free_t free_list;
1472 xfs_fsblock_t first_block;
1475 uint cancel_flags;
1477 xfs_prid_t prid;
1479 uint resblks;
1493 }
1498 /* Return through std_return after this point. */
1503 else
1506 /*
1507 * Make sure that we have allocated dquot(s) on disk.
1508 */
1520 /*
1521 * Initially assume that the file does not exist and
1522 * reserve the resources for that case. If that is not
1523 * the case we'll drop the one we have and get a more
1524 * appropriate transaction later.
1525 */
1532 }
1536 }
1545 /*
1546 * Reserve disk quota and the inode.
1547 */
1562 }
1565 /*
1566 * At this point, we've gotten a newly allocated inode.
1567 * It is locked (and joined to the transaction).
1568 */
1572 /*
1573 * Now we join the directory inode to the transaction. We do not do it
1574 * earlier because xfs_dir_ialloc might commit the previous transaction
1575 * (and release all the locks). An error from here on will result in
1576 * the transaction cancel unlocking dp so don't do it explicitly in the
1577 * error path.
1578 */
1589 }
1593 /*
1594 * If this is a synchronous mount, make sure that the
1595 * create transaction goes to disk before returning to
1596 * the user.
1597 */
1600 }
1604 /*
1605 * Attach the dquot(s) to the inodes and modify them incore.
1606 * These ids of the inode couldn't have changed since the new
1607 * inode has been locked ever since it was created.
1608 */
1611 /*
1612 * xfs_trans_commit normally decrements the vnode ref count
1613 * when it unlocks the inode. Since we want to return the
1614 * vnode to the caller, we bump the vnode ref count now.
1615 */
1622 }
1629 }
1636 /* Fallthrough to std_return with error = 0 */
1638 std_return:
1646 }
1649 abort_return:
1651 /* FALLTHROUGH */
1653 error_return:
1665 abort_rele:
1666 /*
1667 * Wait until after the current transaction is aborted to
1668 * release the inode. This prevents recursive transactions
1669 * and deadlocks from xfs_inactive.
1670 */
1679 }
1681 #ifdef DEBUG
1687 #endif
1689 /*
1690 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1691 * a different value
1692 */
1695 {
1702 }
1704 /*
1705 * The following routine will lock n inodes in exclusive mode.
1706 * We assume the caller calls us with the inodes in i_ino order.
1707 *
1708 * We need to detect deadlock where an inode that we lock
1709 * is in the AIL and we start waiting for another inode that is locked
1710 * by a thread in a long running transaction (such as truncate). This can
1711 * result in deadlock since the long running trans might need to wait
1712 * for the inode we just locked in order to push the tail and free space
1713 * in the log.
1714 */
1715 void
1719 uint lock_mode)
1720 {
1729 again:
1736 /*
1737 * If try_lock is not set yet, make sure all locked inodes
1738 * are not in the AIL.
1739 * If any are, set try_lock to be used later.
1740 */
1746 try_lock++;
1747 }
1748 }
1749 }
1751 /*
1752 * If any of the previous locks we have locked is in the AIL,
1753 * we must TRY to get the second and subsequent locks. If
1754 * we can't get any, we must release all we have
1755 * and try again.
1756 */
1759 /* try_lock must be 0 if i is 0. */
1760 /*
1761 * try_lock means we have an inode locked
1762 * that is in the AIL.
1763 */
1766 attempts++;
1768 /*
1769 * Unlock all previous guys and try again.
1770 * xfs_iunlock will try to push the tail
1771 * if the inode is in the AIL.
1772 */
1776 /*
1777 * Check to see if we've already
1778 * unlocked this one.
1779 * Not the first one going back,
1780 * and the inode ptr is the same.
1781 */
1787 }
1791 #ifdef DEBUG
1792 xfs_lock_delays++;
1793 #endif
1794 }
1798 }
1801 }
1802 }
1804 #ifdef DEBUG
1810 xfs_locked_n++;
1811 }
1812 #endif
1813 }
1815 /*
1816 * xfs_lock_two_inodes() can only be used to lock one type of lock
1817 * at a time - the iolock or the ilock, but not both at once. If
1818 * we lock both at once, lockdep will report false positives saying
1819 * we have violated locking orders.
1820 */
1821 void
1825 uint lock_mode)
1826 {
1839 }
1841 again:
1844 /*
1845 * If the first lock we have locked is in the AIL, we must TRY to get
1846 * the second lock. If we can't get it, we must release the first one
1847 * and try again.
1848 */
1856 }
1859 }
1860 }
1862 int
1867 {
1872 xfs_bmap_free_t free_list;
1873 xfs_fsblock_t first_block;
1877 uint resblks;
1878 uint log_count;
1892 }
1908 }
1911 /*
1912 * We try to get the real space reservation first,
1913 * allowing for directory btree deletion(s) implying
1914 * possible bmap insert(s). If we can't get the space
1915 * reservation then we use 0 instead, and avoid the bmap
1916 * btree insert(s) in the directory code by, if the bmap
1917 * insert tries to happen, instead trimming the LAST
1918 * block from the directory.
1919 */
1927 }
1932 }
1936 /*
1937 * At this point, we've gotten both the directory and the entry
1938 * inodes locked.
1939 */
1946 /*
1947 * If we're removing a directory perform some additional validation.
1948 */
1954 }
1958 }
1959 }
1967 }
1970 /*
1971 * Bump the in memory generation count on the parent
1972 * directory so that other can know that it has changed.
1973 */
1978 /*
1979 * Drop the link from ip's "..".
1980 */
1985 /*
1986 * Drop the "." link from ip to self.
1987 */
1992 /*
1993 * When removing a non-directory we need to log the parent
1994 * inode here for the i_gen update. For a directory this is
1995 * done implicitly by the xfs_droplink call for the ".." entry.
1996 */
1998 }
2000 /*
2001 * Drop the link from dp to ip.
2002 */
2007 /*
2008 * Determine if this is the last link while
2009 * we are in the transaction.
2010 */
2013 /*
2014 * If this is a synchronous mount, make sure that the
2015 * remove transaction goes to disk before returning to
2016 * the user.
2017 */
2029 /*
2030 * If we are using filestreams, kill the stream association.
2031 * If the file is still open it may get a new one but that
2032 * will get killed on last close in xfs_close() so we don't
2033 * have to worry about that.
2034 */
2041 std_return:
2046 }
2050 out_bmap_cancel:
2053 out_trans_cancel:
2056 }
2058 int
2063 {
2067 xfs_bmap_free_t free_list;
2068 xfs_fsblock_t first_block;
2088 }
2090 /* Return through std_return after this point. */
2107 }
2111 }
2115 /*
2116 * Increment vnode ref counts since xfs_trans_commit &
2117 * xfs_trans_cancel will both unlock the inodes and
2118 * decrement the associated ref counts.
2119 */
2125 /*
2126 * If the source has too many links, we can't make any more to it.
2127 */
2131 }
2133 /*
2134 * If we are using project inheritance, we only allow hard link
2135 * creation in our tree when the project IDs are the same; else
2136 * the tree quota mechanism could be circumvented.
2137 */
2142 }
2162 /*
2163 * If this is a synchronous mount, make sure that the
2164 * link transaction goes to disk before returning to
2165 * the user.
2166 */
2169 }
2175 }
2181 /* Fall through to std_return with error = 0. */
2182 std_return:
2188 }
2191 abort_return:
2193 /* FALLTHROUGH */
2195 error_return:
2198 }
2201 int
2205 mode_t mode,
2208 {
2215 xfs_bmap_free_t free_list;
2216 xfs_fsblock_t first_block;
2220 xfs_prid_t prid;
2222 uint resblks;
2237 }
2239 /* Return through std_return after this point. */
2247 else
2250 /*
2251 * Make sure that we have allocated dquot(s) on disk.
2252 */
2267 XFS_TRANS_PERM_LOG_RES,
2268 XFS_MKDIR_LOG_COUNT);
2269 }
2273 }
2278 /*
2279 * Check for directory link count overflow.
2280 */
2284 }
2286 /*
2287 * Reserve disk quota and the inode.
2288 */
2296 /*
2297 * create the directory inode.
2298 */
2306 }
2309 /*
2310 * Now we add the directory inode to the transaction.
2311 * We waited until now since xfs_dir_ialloc might start
2312 * a new transaction. Had we joined the transaction
2313 * earlier, the locks might have gotten released. An error
2314 * from here on will result in the transaction cancel
2315 * unlocking dp so don't do it explicitly in the error path.
2316 */
2329 }
2332 /*
2333 * Bump the in memory version number of the parent directory
2334 * so that other processes accessing it will recognize that
2335 * the directory has changed.
2336 */
2353 /*
2354 * Attach the dquots to the new inode and modify the icount incore.
2355 */
2358 /*
2359 * If this is a synchronous mount, make sure that the
2360 * mkdir transaction goes to disk before returning to
2361 * the user.
2362 */
2365 }
2371 }
2378 }
2380 /* Fall through to std_return with error = 0 or errno from
2381 * xfs_trans_commit. */
2383 std_return:
2389 DM_RIGHT_NULL,
2392 }
2395 error2:
2396 error1:
2398 abort_return:
2400 error_return:
2409 }
2411 int
2416 mode_t mode,
2419 {
2425 xfs_bmap_free_t free_list;
2426 xfs_fsblock_t first_block;
2428 uint cancel_flags;
2430 xfs_fileoff_t first_fsb;
2431 xfs_filblks_t fs_blocks;
2434 xfs_daddr_t d;
2439 xfs_prid_t prid;
2441 uint resblks;
2453 /*
2454 * Check component lengths of the target path name.
2455 */
2466 }
2468 /* Return through std_return after this point. */
2473 else
2476 /*
2477 * Make sure that we have allocated dquot(s) on disk.
2478 */
2487 /*
2488 * The symlink will fit into the inode data fork?
2489 * There can't be any attributes so we get the whole variable part.
2490 */
2493 else
2502 }
2506 }
2511 /*
2512 * Check whether the directory allows new symlinks or not.
2513 */
2517 }
2519 /*
2520 * Reserve disk quota : blocks and inode.
2521 */
2526 /*
2527 * Check for ability to enter directory entry, if no space reserved.
2528 */
2532 /*
2533 * Initialize the bmap freelist prior to calling either
2534 * bmapi or the directory create code.
2535 */
2538 /*
2539 * Allocate an inode for the symlink.
2540 */
2547 }
2550 /*
2551 * An error after we've joined dp to the transaction will result in the
2552 * transaction cancel unlocking dp so don't do it explicitly in the
2553 * error path.
2554 */
2559 /*
2560 * Also attach the dquot(s) to it, if applicable.
2561 */
2566 /*
2567 * If the symlink will fit into the inode, write it inline.
2568 */
2574 /*
2575 * The inode was initially created in extent format.
2576 */
2593 }
2609 }
2616 }
2617 }
2619 /*
2620 * Create the directory entry for the symlink.
2621 */
2629 /*
2630 * Bump the in memory version number of the parent directory
2631 * so that other processes accessing it will recognize that
2632 * the directory has changed.
2633 */
2636 /*
2637 * If this is a synchronous mount, make sure that the
2638 * symlink transaction goes to disk before returning to
2639 * the user.
2640 */
2643 }
2645 /*
2646 * xfs_trans_commit normally decrements the vnode ref count
2647 * when it unlocks the inode. Since we want to return the
2648 * vnode to the caller, we bump the vnode ref count now.
2649 */
2655 }
2660 /* Fall through to std_return with error = 0 or errno from
2661 * xfs_trans_commit */
2662 std_return:
2669 }
2675 error2:
2677 error1:
2680 error_return:
2689 }
2691 int
2695 {
2702 /*
2703 * Bypass inodes which have already been cleaned by
2704 * the inode flush clustering code inside xfs_iflush
2705 */
2709 /*
2710 * We make this non-blocking if the inode is contended,
2711 * return EAGAIN to indicate to the caller that they
2712 * did not succeed. This prevents the flush path from
2713 * blocking on inodes inside another operation right
2714 * now, they get caught later by xfs_sync.
2715 */
2723 }
2726 }
2733 }
2736 int
2739 u_int evmask,
2740 u_int16_t state)
2741 {
2757 }
2769 }
2771 int
2774 {
2780 /* bad inode, get out here ASAP */
2784 }
2790 /*
2791 * Make sure the atime in the XFS inode is correct before freeing the
2792 * Linux inode.
2793 */
2796 /*
2797 * If we have nothing to flush with this inode then complete the
2798 * teardown now, otherwise break the link between the xfs inode and the
2799 * linux inode and clean up the xfs inode later. This avoids flushing
2800 * the inode to disk during the delete operation itself.
2801 *
2802 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2803 * first to ensure that xfs_iunpin() will never see an xfs inode
2804 * that has a linux inode being reclaimed. Synchronisation is provided
2805 * by the i_flags_lock.
2806 */
2812 }
2815 }
2817 /*
2818 * xfs_alloc_file_space()
2819 * This routine allocates disk space for the given file.
2820 *
2821 * If alloc_type == 0, this request is for an ALLOCSP type
2822 * request which will change the file size. In this case, no
2823 * DMAPI event will be generated by the call. A TRUNCATE event
2824 * will be generated later by xfs_setattr.
2825 *
2826 * If alloc_type != 0, this request is for a RESVSP type
2827 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2828 * lower block boundary byte address is less than the file's
2829 * length.
2830 *
2831 * RETURNS:
2832 * 0 on success
2833 * errno on error
2834 *
2835 */
2836 STATIC int
2839 xfs_off_t offset,
2840 xfs_off_t len,
2843 {
2845 xfs_off_t count;
2846 xfs_filblks_t allocated_fsb;
2847 xfs_filblks_t allocatesize_fsb;
2849 xfs_fileoff_t startoffset_fsb;
2850 xfs_fsblock_t firstfsb;
2857 xfs_bmap_free_t free_list;
2883 /* Generate a DMAPI event if needed. */
2887 xfs_off_t end_dmi_offset;
2896 }
2898 /*
2899 * Allocate file space until done or until there is an error
2900 */
2901 retry:
2905 /*
2906 * Determine space reservations for data/realtime.
2907 */
2920 }
2932 }
2934 /*
2935 * Allocate and setup the transaction.
2936 */
2940 XFS_TRANS_PERM_LOG_RES,
2941 XFS_WRITE_LOG_COUNT);
2942 /*
2943 * Check for running out of space
2944 */
2946 /*
2947 * Free the transaction structure.
2948 */
2952 }
2962 /*
2963 * Issue the xfs_bmapi() call to allocate the blocks
2964 */
2972 }
2974 /*
2975 * Complete the transaction
2976 */
2980 }
2986 }
2993 }
2997 }
2998 dmapi_enospc_check:
3007 /* else fall through with error from XFS_SEND_DATA */
3008 }
3020 }
3022 /*
3023 * Zero file bytes between startoff and endoff inclusive.
3024 * The iolock is held exclusive and no blocks are buffered.
3025 *
3026 * This function is used by xfs_free_file_space() to zero
3027 * partial blocks when the range to free is not block aligned.
3028 * When unreserving space with boundaries that are not block
3029 * aligned we round up the start and round down the end
3030 * boundaries and then use this function to zero the parts of
3031 * the blocks that got dropped during the rounding.
3032 */
3033 STATIC int
3036 xfs_off_t startoff,
3037 xfs_off_t endoff)
3038 {
3039 xfs_bmbt_irec_t imap;
3040 xfs_fileoff_t offset_fsb;
3041 xfs_off_t lastoffset;
3042 xfs_off_t offset;
3048 /*
3049 * Avoid doing I/O beyond eof - it's not necessary
3050 * since nothing can read beyond eof. The space will
3051 * be zeroed when the file is extended anyway.
3052 */
3090 }
3103 }
3104 }
3107 }
3109 /*
3110 * xfs_free_file_space()
3111 * This routine frees disk space for the given file.
3112 *
3113 * This routine is only called by xfs_change_file_space
3114 * for an UNRESVSP type call.
3115 *
3116 * RETURNS:
3117 * 0 on success
3118 * errno on error
3119 *
3120 */
3121 STATIC int
3124 xfs_off_t offset,
3125 xfs_off_t len,
3127 {
3130 xfs_off_t end_dmi_offset;
3131 xfs_fileoff_t endoffset_fsb;
3133 xfs_fsblock_t firstfsb;
3134 xfs_bmap_free_t free_list;
3135 xfs_bmbt_irec_t imap;
3136 xfs_off_t ioffset;
3140 uint resblks;
3141 uint rounding;
3143 xfs_fileoff_t startoffset_fsb;
3171 }
3178 }
3188 }
3190 /*
3191 * Need to zero the stuff we're not freeing, on disk.
3192 * If its a realtime file & can't use unwritten extents then we
3193 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3194 * will take care of it for us.
3195 */
3204 xfs_daddr_t block;
3211 }
3220 mod++;
3223 }
3224 }
3226 /*
3227 * One contiguous piece to clear
3228 */
3231 /*
3232 * Some full blocks, possibly two pieces to clear
3233 */
3242 }
3244 /*
3245 * free file space until done or until there is an error
3246 */
3250 /*
3251 * allocate and setup the transaction. Allow this
3252 * transaction to dip into the reserve blocks to ensure
3253 * the freeing of the space succeeds at ENOSPC.
3254 */
3258 resblks,
3261 XFS_TRANS_PERM_LOG_RES,
3262 XFS_WRITE_LOG_COUNT);
3264 /*
3265 * check for running out of space
3266 */
3268 /*
3269 * Free the transaction structure.
3270 */
3274 }
3278 XFS_QMOPT_RES_REGBLKS);
3285 /*
3286 * issue the bunmapi() call to free the blocks
3287 */
3294 }
3296 /*
3297 * complete the transaction
3298 */
3302 }
3306 }
3308 out_unlock_iolock:
3313 error0:
3315 error1:
3318 XFS_ILOCK_EXCL);
3320 }
3322 /*
3323 * xfs_change_file_space()
3324 * This routine allocates or frees disk space for the given file.
3325 * The user specified parameters are checked for alignment and size
3326 * limitations.
3327 *
3328 * RETURNS:
3329 * 0 on success
3330 * errno on error
3331 *
3332 */
3333 int
3338 xfs_off_t offset,
3340 {
3344 xfs_fsize_t fsize;
3346 xfs_off_t startoffset;
3347 xfs_off_t llen;
3367 }
3382 /*
3383 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3384 * file space.
3385 * These calls do NOT zero the data space allocated to the file,
3386 * nor do they change the file size.
3387 *
3388 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3389 * space.
3390 * These calls cause the new file data to be zeroed and the file
3391 * size to be changed.
3392 */
3408 attr_flags)))
3421 }
3437 }
3439 /*
3440 * update the inode timestamp, mode, and prealloc flag bits
3441 */
3446 /* ASSERT(0); */
3449 }
3459 /*
3460 * Note that we don't have to worry about mandatory
3461 * file locking being disabled here because we only
3462 * clear the S_ISGID bit if the Group execute bit is
3463 * on, but if it was on then mandatory locking wouldn't
3464 * have been enabled.
3465 */
3470 }
3484 }