| blob | 1dd70688715643cf940e15054aafe8466f35a276 |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
57 int
62 {
68 uint lock_flags;
91 /*
92 * If disk quotas is on, we make sure that the dquots do exist on disk,
93 * before we start any other transactions. Trying to do this later
94 * is messy. We don't care to take a readlock to look at the ids
95 * in inode here, because we can't hold it across the trans_reserve.
96 * If the IDs do change before we take the ilock, we're covered
97 * because the i_*dquot fields will get updated anyway.
98 */
107 }
113 }
115 /*
116 * We take a reference when we initialize udqp and gdqp,
117 * so it is important that we never blindly double trip on
118 * the same variable. See xfs_create() for an example.
119 */
126 }
128 /*
129 * For the other attributes, we acquire the inode lock and
130 * first do an error checking pass.
131 */
146 }
147 }
157 }
158 }
161 }
165 /*
166 * Change file ownership. Must be the owner or privileged.
167 */
169 /*
170 * These IDs could have changed since we last looked at them.
171 * But, we're assured that if the ownership did change
172 * while we didn't have the inode locked, inode's dquot(s)
173 * would have changed also.
174 */
180 /*
181 * Do a quota reservation only if uid/gid is actually
182 * going to change.
183 */
193 }
194 }
196 /*
197 * Truncate file. Must have write permission and not be a directory.
198 */
200 /* Short circuit the truncate case for zero length files */
209 }
217 }
219 /*
220 * Make sure that the dquots are attached to the inode.
221 */
226 /*
227 * Now we can make the changes. Before we join the inode
228 * to the transaction, if ATTR_SIZE is set then take care of
229 * the part of the truncation that must be done without the
230 * inode lock. This needs to be done before joining the inode
231 * to the transaction, because the inode cannot be unlocked
232 * once it is a part of the transaction.
233 */
235 /*
236 * Do the first part of growing a file: zero any data
237 * in the last block that is beyond the old EOF. We
238 * need to do this before the inode is joined to the
239 * transaction to modify the i_size.
240 */
242 }
245 /*
246 * We are going to log the inode size change in this
247 * transaction so any previous writes that are beyond the on
248 * disk EOF and the new EOF that have not been written out need
249 * to be written here. If we do not write the data out, we
250 * expose ourselves to the null files problem.
251 *
252 * Only flush from the on disk size to the smaller of the in
253 * memory file size or the new size as that's the range we
254 * really care about here and prevents waiting for other data
255 * not within the range we care about here.
256 */
263 }
265 /* wait for all I/O to complete */
275 }
279 XFS_TRANS_PERM_LOG_RES,
280 XFS_ITRUNCATE_LOG_COUNT))) {
285 }
292 /*
293 * Only change the c/mtime if we are changing the size
294 * or we are explicitly asked to change it. This handles
295 * the semantic difference between truncate() and ftruncate()
296 * as implemented in the VFS.
297 */
309 /*
310 * signal a sync transaction unless
311 * we're truncating an already unlinked
312 * file on a wsync filesystem
313 */
315 XFS_DATA_FORK,
321 /*
322 * Truncated "down", so we're removing references
323 * to old data here - if we now delay flushing for
324 * a long time, we expose ourselves unduly to the
325 * notorious NULL files problem. So, we mark this
326 * vnode and flush it when the file is closed, and
327 * do not wait the usual (long) time for writeout.
328 */
330 }
334 }
336 /*
337 * Change file ownership. Must be the owner or privileged.
338 */
340 /*
341 * CAP_FSETID overrides the following restrictions:
342 *
343 * The set-user-ID and set-group-ID bits of a file will be
344 * cleared upon successful return from chown()
345 */
349 }
351 /*
352 * Change the ownerships and register quota modifications
353 * in the transaction.
354 */
361 }
364 }
372 }
375 }
379 }
381 /*
382 * Change file access modes.
383 */
398 }
400 /*
401 * Change file access or modified times.
402 */
409 }
416 }
419 }
421 /*
422 * Change file inode change time only if ATTR_CTIME set
423 * AND we have been called by a DMI function.
424 */
432 }
434 /*
435 * Send out timestamp changes that need to be set to the
436 * current time. Not done when called by a DMI function.
437 */
443 /*
444 * If this is a synchronous mount, make sure that the
445 * transaction goes to disk before returning to the user.
446 * This is slightly sub-optimal in that truncates require
447 * two sync transactions instead of one for wsync filesystems.
448 * One for the truncate and one for the timestamps since we
449 * don't want to change the timestamps unless we're sure the
450 * truncate worked. Truncates are less than 1% of the laddis
451 * mix so this probably isn't worth the trouble to optimize.
452 */
459 }
463 /*
464 * Release any dquot(s) the inode had kept before chown.
465 */
474 /*
475 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
476 * update. We could avoid this with linked transactions
477 * and passing down the transaction pointer all the way
478 * to attr_set. No previous user of the generic
479 * Posix ACL code seems to care about this issue either.
480 */
485 }
492 }
495 abort_return:
497 /* FALLTHROUGH */
498 error_return:
503 }
506 }
508 }
510 /*
511 * The maximum pathlen is 1024 bytes. Since the minimum file system
512 * blocksize is 512 bytes, we can get a max of 2 extents back from
513 * bmapi.
514 */
515 #define SYMLINK_MAPS 2
517 STATIC int
521 {
526 xfs_daddr_t d;
548 }
555 }
560 out:
562 }
564 int
568 {
592 }
594 out:
597 }
599 /*
600 * xfs_fsync
601 *
602 * This is called to sync the inode and its data out to disk. We need to hold
603 * the I/O lock while flushing the data, and the inode lock while flushing the
604 * inode. The inode lock CANNOT be held while flushing the data, so acquire
605 * after we're done with that.
606 */
607 int
610 {
620 /* capture size updates in I/O completion before writing the inode. */
625 /*
626 * We always need to make sure that the required inode state is safe on
627 * disk. The vnode might be clean but we still might need to force the
628 * log because of committed transactions that haven't hit the disk yet.
629 * Likewise, there could be unflushed non-transactional changes to the
630 * inode core that have to go to disk and this requires us to issue
631 * a synchronous transaction to capture these changes correctly.
632 *
633 * This code relies on the assumption that if the update_* fields
634 * of the inode are clear and the inode is unpinned then it is clean
635 * and no action is required.
636 */
640 /*
641 * Timestamps/size haven't changed since last inode flush or
642 * inode transaction commit. That means either nothing got
643 * written or a transaction committed which caught the updates.
644 * If the latter happened and the transaction hasn't hit the
645 * disk yet, the inode will be still be pinned. If it is,
646 * force the log.
647 */
656 /*
657 * If the inode is not pinned and nothing has changed
658 * we don't need to flush the cache.
659 */
661 }
663 /*
664 * Kick off a transaction to log the inode core to get the
665 * updates. The sync transaction will also force the log.
666 */
674 }
677 /*
678 * Note - it's possible that we might have pushed ourselves out
679 * of the way during trans_reserve which would flush the inode.
680 * But there's no guarantee that the inode buffer has actually
681 * gone out yet (it's delwri). Plus the buffer could be pinned
682 * anyway if it's part of an inode in another recent
683 * transaction. So we play it safe and fire off the
684 * transaction anyway.
685 */
693 }
696 /*
697 * If the log write didn't issue an ordered tag we need
698 * to flush the disk cache for the data device now.
699 */
703 /*
704 * If this inode is on the RT dev we need to flush that
705 * cache as well.
706 */
709 }
712 }
714 /*
715 * This is called by xfs_inactive to free any blocks beyond eof
716 * when the link count isn't zero and by xfs_dm_punch_hole() when
717 * punching a hole to EOF.
718 */
719 STATIC int
724 {
727 xfs_fileoff_t end_fsb;
728 xfs_fileoff_t last_fsb;
729 xfs_filblks_t map_len;
731 xfs_bmbt_irec_t imap;
734 /*
735 * Figure out if there are any blocks beyond the end
736 * of the file. If not, then there is nothing to do.
737 */
753 /*
754 * Attach the dquots to the inode up front.
755 */
760 /*
761 * There are blocks after the end of file.
762 * Free them up now by truncating the file to
763 * its current size.
764 */
767 /*
768 * Do the xfs_itruncate_start() call before
769 * reserving any log space because
770 * itruncate_start will call into the buffer
771 * cache and we can't
772 * do that within a transaction.
773 */
783 }
788 XFS_ITRUNCATE_LOG_COUNT);
794 }
798 XFS_IOLOCK_EXCL |
799 XFS_ILOCK_EXCL);
804 XFS_DATA_FORK,
806 /*
807 * If we get an error at this point we
808 * simply don't bother truncating the file.
809 */
813 XFS_TRANS_ABORT));
816 XFS_TRANS_RELEASE_LOG_RES);
817 }
820 }
822 }
824 /*
825 * Free a symlink that has blocks associated with it.
826 */
827 STATIC int
831 {
836 xfs_fsblock_t first_block;
837 xfs_bmap_free_t free_list;
849 /*
850 * We're freeing a symlink that has some
851 * blocks allocated to it. Free the
852 * blocks here. We know that we've got
853 * either 1 or 2 extents and that we can
854 * free them all in one bunmapi call.
855 */
863 }
864 /*
865 * Lock the inode, fix the size, and join it to the transaction.
866 * Hold it so in the normal path, we still have it locked for
867 * the second transaction. In the error paths we need it
868 * held so the cancel won't rele it, see below.
869 */
876 /*
877 * Find the block(s) so we can inval and unmap them.
878 */
886 /*
887 * Invalidate the block(s).
888 */
894 }
895 /*
896 * Unmap the dead block(s) to the free_list.
897 */
902 /*
903 * Commit the first transaction. This logs the EFI and the inode.
904 */
907 /*
908 * The transaction must have been committed, since there were
909 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
910 * The new tp has the extent freeing and EFDs.
911 */
913 /*
914 * The first xact was committed, so add the inode to the new one.
915 * Mark it dirty so it will be logged and moved forward in the log as
916 * part of every commit.
917 */
921 /*
922 * Get a new, empty transaction to return to our caller.
923 */
925 /*
926 * Commit the transaction containing extent freeing and EFDs.
927 * If we get an error on the commit here or on the reserve below,
928 * we need to unlock the inode since the new transaction doesn't
929 * have the inode attached.
930 */
936 }
937 /*
938 * transaction commit worked ok so we can drop the extra ticket
939 * reference that we gained in xfs_trans_dup()
940 */
943 /*
944 * Remove the memory for extent descriptions (just bookkeeping).
945 */
949 /*
950 * Put an itruncate log reservation in the new transaction
951 * for our caller.
952 */
957 }
958 /*
959 * Return with the inode locked but not joined to the transaction.
960 */
964 error1:
966 error0:
967 /*
968 * Have to come here with the inode locked and either
969 * (held and in the transaction) or (not in the transaction).
970 * If the inode isn't held then cancel would iput it, but
971 * that's wrong since this is inactive and the vnode ref
972 * count is 0 already.
973 * Cancel won't do anything to the inode if held, but it still
974 * needs to be locked until the cancel is done, if it was
975 * joined to the transaction.
976 */
982 }
984 STATIC int
988 {
992 /*
993 * We're freeing a symlink which fit into
994 * the inode. Just free the memory used
995 * to hold the old symlink.
996 */
1000 XFS_ITRUNCATE_LOG_COUNT);
1006 }
1009 /*
1010 * Zero length symlinks _can_ exist.
1011 */
1015 XFS_DATA_FORK);
1017 }
1019 }
1021 STATIC int
1025 {
1047 XFS_INACTIVE_LOG_COUNT);
1061 error_cancel:
1064 error_unlock:
1068 }
1070 int
1073 {
1080 /* If this is a read-only mount, don't do this (would generate I/O) */
1087 /*
1088 * If we are using filestreams, and we have an unlinked
1089 * file that we are processing the last close on, then nothing
1090 * will be able to reopen and write to this file. Purge this
1091 * inode from the filestreams cache so that it doesn't delay
1092 * teardown of the inode.
1093 */
1097 /*
1098 * If we previously truncated this file and removed old data
1099 * in the process, we want to initiate "early" writeout on
1100 * the last close. This is an attempt to combat the notorious
1101 * NULL files problem which is particularly noticable from a
1102 * truncate down, buffered (re-)write (delalloc), followed by
1103 * a crash. What we are effectively doing here is
1104 * significantly reducing the time window where we'd otherwise
1105 * be exposed to that problem.
1106 */
1110 }
1122 }
1123 }
1126 }
1128 /*
1129 * xfs_inactive
1130 *
1131 * This is called when the vnode reference count for the vnode
1132 * goes to zero. If the file has been unlinked, then it must
1133 * now be truncated. Also, we clear all of the read-ahead state
1134 * kept for the inode here since the file is now closed.
1135 */
1136 int
1139 {
1140 xfs_bmap_free_t free_list;
1141 xfs_fsblock_t first_block;
1150 /*
1151 * If the inode is already free, then there can be nothing
1152 * to clean up here.
1153 */
1158 }
1160 /*
1161 * Only do a truncate if it's a regular file with
1162 * some actual space in it. It's OK to look at the
1163 * inode's fields without the lock because we're the
1164 * only one with a reference to the inode.
1165 */
1178 /* If this is a read-only mount, don't do this (would generate I/O) */
1193 }
1195 }
1205 /*
1206 * Do the xfs_itruncate_start() call before
1207 * reserving any log space because itruncate_start
1208 * will call into the buffer cache and we can't
1209 * do that within a transaction.
1210 */
1218 }
1223 XFS_ITRUNCATE_LOG_COUNT);
1225 /* Don't call itruncate_cleanup */
1230 }
1236 /*
1237 * normally, we have to run xfs_itruncate_finish sync.
1238 * But if filesystem is wsync and we're in the inactive
1239 * path, then we know that nlink == 0, and that the
1240 * xaction that made nlink == 0 is permanently committed
1241 * since xfs_remove runs as a synchronous transaction.
1242 */
1251 }
1254 /*
1255 * If we get an error while cleaning up a
1256 * symlink we bail out.
1257 */
1265 }
1273 XFS_INACTIVE_LOG_COUNT);
1278 }
1283 }
1285 /*
1286 * If there are attributes associated with the file
1287 * then blow them away now. The code calls a routine
1288 * that recursively deconstructs the attribute fork.
1289 * We need to just commit the current transaction
1290 * because we can't use it for xfs_attr_inactive().
1291 */
1294 /*
1295 * If we got an error, the transaction is already
1296 * cancelled, and the inode is unlocked. Just get out.
1297 */
1302 }
1304 /*
1305 * Free the inode.
1306 */
1310 /*
1311 * If we fail to free the inode, shut down. The cancel
1312 * might do that, we need to make sure. Otherwise the
1313 * inode might be lost for a long time or forever.
1314 */
1320 }
1323 /*
1324 * Credit the quota account(s). The inode is gone.
1325 */
1328 /*
1329 * Just ignore errors at this point. There is nothing we can
1330 * do except to try to keep going. Make sure it's not a silent
1331 * error.
1332 */
1341 }
1343 /*
1344 * Release the dquots held by inode, if any.
1345 */
1349 out:
1351 }
1353 /*
1354 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1355 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1356 * ci_name->name will point to a the actual name (caller must free) or
1357 * will be set to NULL if an exact match is found.
1358 */
1359 int
1365 {
1366 xfs_ino_t inum;
1368 uint lock_mode;
1389 out_free_name:
1392 out:
1395 }
1397 int
1401 mode_t mode,
1402 xfs_dev_t rdev,
1405 {
1411 xfs_bmap_free_t free_list;
1412 xfs_fsblock_t first_block;
1414 uint cancel_flags;
1416 xfs_prid_t prid;
1419 uint resblks;
1420 uint log_res;
1421 uint log_count;
1436 }
1440 else
1443 /*
1444 * Make sure that we have allocated dquot(s) on disk.
1445 */
1462 }
1466 /*
1467 * Initially assume that the file does not exist and
1468 * reserve the resources for that case. If that is not
1469 * the case we'll drop the one we have and get a more
1470 * appropriate transaction later.
1471 */
1475 /* flush outstanding delalloc blocks and retry */
1479 }
1481 /* No space at all so try a "no-allocation" reservation */
1485 }
1489 }
1494 /*
1495 * Check for directory link count overflow.
1496 */
1500 }
1504 /*
1505 * Reserve disk quota and the inode.
1506 */
1515 /*
1516 * A newly created regular or special file just has one directory
1517 * entry pointing to them, but a directory also the "." entry
1518 * pointing to itself.
1519 */
1526 }
1528 /*
1529 * At this point, we've gotten a newly allocated inode.
1530 * It is locked (and joined to the transaction).
1531 */
1535 /*
1536 * Now we join the directory inode to the transaction. We do not do it
1537 * earlier because xfs_dir_ialloc might commit the previous transaction
1538 * (and release all the locks). An error from here on will result in
1539 * the transaction cancel unlocking dp so don't do it explicitly in the
1540 * error path.
1541 */
1552 }
1564 }
1566 /*
1567 * If this is a synchronous mount, make sure that the
1568 * create transaction goes to disk before returning to
1569 * the user.
1570 */
1574 /*
1575 * Attach the dquot(s) to the inodes and modify them incore.
1576 * These ids of the inode couldn't have changed since the new
1577 * inode has been locked ever since it was created.
1578 */
1581 /*
1582 * xfs_trans_commit normally decrements the vnode ref count
1583 * when it unlocks the inode. Since we want to return the
1584 * vnode to the caller, we bump the vnode ref count now.
1585 */
1596 }
1603 /* Fallthrough to std_return with error = 0 */
1604 std_return:
1609 }
1613 out_bmap_cancel:
1615 out_trans_abort:
1617 out_trans_cancel:
1619 out_dqrele:
1628 out_abort_rele:
1629 /*
1630 * Wait until after the current transaction is aborted to
1631 * release the inode. This prevents recursive transactions
1632 * and deadlocks from xfs_inactive.
1633 */
1640 }
1642 #ifdef DEBUG
1648 #endif
1650 /*
1651 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1652 * a different value
1653 */
1656 {
1663 }
1665 /*
1666 * The following routine will lock n inodes in exclusive mode.
1667 * We assume the caller calls us with the inodes in i_ino order.
1668 *
1669 * We need to detect deadlock where an inode that we lock
1670 * is in the AIL and we start waiting for another inode that is locked
1671 * by a thread in a long running transaction (such as truncate). This can
1672 * result in deadlock since the long running trans might need to wait
1673 * for the inode we just locked in order to push the tail and free space
1674 * in the log.
1675 */
1676 void
1680 uint lock_mode)
1681 {
1690 again:
1697 /*
1698 * If try_lock is not set yet, make sure all locked inodes
1699 * are not in the AIL.
1700 * If any are, set try_lock to be used later.
1701 */
1707 try_lock++;
1708 }
1709 }
1710 }
1712 /*
1713 * If any of the previous locks we have locked is in the AIL,
1714 * we must TRY to get the second and subsequent locks. If
1715 * we can't get any, we must release all we have
1716 * and try again.
1717 */
1720 /* try_lock must be 0 if i is 0. */
1721 /*
1722 * try_lock means we have an inode locked
1723 * that is in the AIL.
1724 */
1727 attempts++;
1729 /*
1730 * Unlock all previous guys and try again.
1731 * xfs_iunlock will try to push the tail
1732 * if the inode is in the AIL.
1733 */
1737 /*
1738 * Check to see if we've already
1739 * unlocked this one.
1740 * Not the first one going back,
1741 * and the inode ptr is the same.
1742 */
1748 }
1752 #ifdef DEBUG
1753 xfs_lock_delays++;
1754 #endif
1755 }
1759 }
1762 }
1763 }
1765 #ifdef DEBUG
1771 xfs_locked_n++;
1772 }
1773 #endif
1774 }
1776 /*
1777 * xfs_lock_two_inodes() can only be used to lock one type of lock
1778 * at a time - the iolock or the ilock, but not both at once. If
1779 * we lock both at once, lockdep will report false positives saying
1780 * we have violated locking orders.
1781 */
1782 void
1786 uint lock_mode)
1787 {
1800 }
1802 again:
1805 /*
1806 * If the first lock we have locked is in the AIL, we must TRY to get
1807 * the second lock. If we can't get it, we must release the first one
1808 * and try again.
1809 */
1817 }
1820 }
1821 }
1823 int
1828 {
1833 xfs_bmap_free_t free_list;
1834 xfs_fsblock_t first_block;
1838 uint resblks;
1839 uint log_count;
1853 }
1869 }
1872 /*
1873 * We try to get the real space reservation first,
1874 * allowing for directory btree deletion(s) implying
1875 * possible bmap insert(s). If we can't get the space
1876 * reservation then we use 0 instead, and avoid the bmap
1877 * btree insert(s) in the directory code by, if the bmap
1878 * insert tries to happen, instead trimming the LAST
1879 * block from the directory.
1880 */
1888 }
1893 }
1897 /*
1898 * At this point, we've gotten both the directory and the entry
1899 * inodes locked.
1900 */
1907 /*
1908 * If we're removing a directory perform some additional validation.
1909 */
1915 }
1919 }
1920 }
1928 }
1932 /*
1933 * Drop the link from ip's "..".
1934 */
1939 /*
1940 * Drop the "." link from ip to self.
1941 */
1946 /*
1947 * When removing a non-directory we need to log the parent
1948 * inode here. For a directory this is done implicitly
1949 * by the xfs_droplink call for the ".." entry.
1950 */
1952 }
1954 /*
1955 * Drop the link from dp to ip.
1956 */
1961 /*
1962 * Determine if this is the last link while
1963 * we are in the transaction.
1964 */
1967 /*
1968 * If this is a synchronous mount, make sure that the
1969 * remove transaction goes to disk before returning to
1970 * the user.
1971 */
1983 /*
1984 * If we are using filestreams, kill the stream association.
1985 * If the file is still open it may get a new one but that
1986 * will get killed on last close in xfs_close() so we don't
1987 * have to worry about that.
1988 */
1995 std_return:
2000 }
2004 out_bmap_cancel:
2007 out_trans_cancel:
2010 }
2012 int
2017 {
2021 xfs_bmap_free_t free_list;
2022 xfs_fsblock_t first_block;
2042 }
2044 /* Return through std_return after this point. */
2063 }
2067 }
2071 /*
2072 * Increment vnode ref counts since xfs_trans_commit &
2073 * xfs_trans_cancel will both unlock the inodes and
2074 * decrement the associated ref counts.
2075 */
2081 /*
2082 * If the source has too many links, we can't make any more to it.
2083 */
2087 }
2089 /*
2090 * If we are using project inheritance, we only allow hard link
2091 * creation in our tree when the project IDs are the same; else
2092 * the tree quota mechanism could be circumvented.
2093 */
2098 }
2117 /*
2118 * If this is a synchronous mount, make sure that the
2119 * link transaction goes to disk before returning to
2120 * the user.
2121 */
2124 }
2130 }
2136 /* Fall through to std_return with error = 0. */
2137 std_return:
2143 }
2146 abort_return:
2148 /* FALLTHROUGH */
2150 error_return:
2153 }
2155 int
2160 mode_t mode,
2163 {
2169 xfs_bmap_free_t free_list;
2170 xfs_fsblock_t first_block;
2172 uint cancel_flags;
2174 xfs_fileoff_t first_fsb;
2175 xfs_filblks_t fs_blocks;
2178 xfs_daddr_t d;
2183 xfs_prid_t prid;
2185 uint resblks;
2197 /*
2198 * Check component lengths of the target path name.
2199 */
2210 }
2212 /* Return through std_return after this point. */
2217 else
2220 /*
2221 * Make sure that we have allocated dquot(s) on disk.
2222 */
2230 /*
2231 * The symlink will fit into the inode data fork?
2232 * There can't be any attributes so we get the whole variable part.
2233 */
2236 else
2245 }
2249 }
2254 /*
2255 * Check whether the directory allows new symlinks or not.
2256 */
2260 }
2262 /*
2263 * Reserve disk quota : blocks and inode.
2264 */
2269 /*
2270 * Check for ability to enter directory entry, if no space reserved.
2271 */
2275 /*
2276 * Initialize the bmap freelist prior to calling either
2277 * bmapi or the directory create code.
2278 */
2281 /*
2282 * Allocate an inode for the symlink.
2283 */
2290 }
2293 /*
2294 * An error after we've joined dp to the transaction will result in the
2295 * transaction cancel unlocking dp so don't do it explicitly in the
2296 * error path.
2297 */
2302 /*
2303 * Also attach the dquot(s) to it, if applicable.
2304 */
2309 /*
2310 * If the symlink will fit into the inode, write it inline.
2311 */
2317 /*
2318 * The inode was initially created in extent format.
2319 */
2336 }
2352 }
2359 }
2360 }
2362 /*
2363 * Create the directory entry for the symlink.
2364 */
2372 /*
2373 * If this is a synchronous mount, make sure that the
2374 * symlink transaction goes to disk before returning to
2375 * the user.
2376 */
2379 }
2381 /*
2382 * xfs_trans_commit normally decrements the vnode ref count
2383 * when it unlocks the inode. Since we want to return the
2384 * vnode to the caller, we bump the vnode ref count now.
2385 */
2391 }
2396 /* Fall through to std_return with error = 0 or errno from
2397 * xfs_trans_commit */
2398 std_return:
2405 }
2411 error2:
2413 error1:
2416 error_return:
2425 }
2427 int
2430 u_int evmask,
2431 u_int16_t state)
2432 {
2448 }
2460 }
2462 int
2465 {
2471 /* bad inode, get out here ASAP */
2475 }
2481 /*
2482 * Make sure the atime in the XFS inode is correct before freeing the
2483 * Linux inode.
2484 */
2487 /*
2488 * If we have nothing to flush with this inode then complete the
2489 * teardown now, otherwise break the link between the xfs inode and the
2490 * linux inode and clean up the xfs inode later. This avoids flushing
2491 * the inode to disk during the delete operation itself.
2492 *
2493 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2494 * first to ensure that xfs_iunpin() will never see an xfs inode
2495 * that has a linux inode being reclaimed. Synchronisation is provided
2496 * by the i_flags_lock.
2497 */
2503 }
2506 }
2508 /*
2509 * xfs_alloc_file_space()
2510 * This routine allocates disk space for the given file.
2511 *
2512 * If alloc_type == 0, this request is for an ALLOCSP type
2513 * request which will change the file size. In this case, no
2514 * DMAPI event will be generated by the call. A TRUNCATE event
2515 * will be generated later by xfs_setattr.
2516 *
2517 * If alloc_type != 0, this request is for a RESVSP type
2518 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2519 * lower block boundary byte address is less than the file's
2520 * length.
2521 *
2522 * RETURNS:
2523 * 0 on success
2524 * errno on error
2525 *
2526 */
2527 STATIC int
2530 xfs_off_t offset,
2531 xfs_off_t len,
2534 {
2536 xfs_off_t count;
2537 xfs_filblks_t allocated_fsb;
2538 xfs_filblks_t allocatesize_fsb;
2540 xfs_fileoff_t startoffset_fsb;
2541 xfs_fsblock_t firstfsb;
2548 xfs_bmap_free_t free_list;
2575 /* Generate a DMAPI event if needed. */
2579 xfs_off_t end_dmi_offset;
2588 }
2590 /*
2591 * Allocate file space until done or until there is an error
2592 */
2593 retry:
2597 /*
2598 * Determine space reservations for data/realtime.
2599 */
2612 }
2624 }
2626 /*
2627 * Allocate and setup the transaction.
2628 */
2632 XFS_TRANS_PERM_LOG_RES,
2633 XFS_WRITE_LOG_COUNT);
2634 /*
2635 * Check for running out of space
2636 */
2638 /*
2639 * Free the transaction structure.
2640 */
2644 }
2654 /*
2655 * Issue the xfs_bmapi() call to allocate the blocks
2656 */
2664 }
2666 /*
2667 * Complete the transaction
2668 */
2672 }
2678 }
2685 }
2689 }
2690 dmapi_enospc_check:
2699 /* else fall through with error from XFS_SEND_DATA */
2700 }
2712 }
2714 /*
2715 * Zero file bytes between startoff and endoff inclusive.
2716 * The iolock is held exclusive and no blocks are buffered.
2717 *
2718 * This function is used by xfs_free_file_space() to zero
2719 * partial blocks when the range to free is not block aligned.
2720 * When unreserving space with boundaries that are not block
2721 * aligned we round up the start and round down the end
2722 * boundaries and then use this function to zero the parts of
2723 * the blocks that got dropped during the rounding.
2724 */
2725 STATIC int
2728 xfs_off_t startoff,
2729 xfs_off_t endoff)
2730 {
2731 xfs_bmbt_irec_t imap;
2732 xfs_fileoff_t offset_fsb;
2733 xfs_off_t lastoffset;
2734 xfs_off_t offset;
2740 /*
2741 * Avoid doing I/O beyond eof - it's not necessary
2742 * since nothing can read beyond eof. The space will
2743 * be zeroed when the file is extended anyway.
2744 */
2784 }
2797 }
2798 }
2801 }
2803 /*
2804 * xfs_free_file_space()
2805 * This routine frees disk space for the given file.
2806 *
2807 * This routine is only called by xfs_change_file_space
2808 * for an UNRESVSP type call.
2809 *
2810 * RETURNS:
2811 * 0 on success
2812 * errno on error
2813 *
2814 */
2815 STATIC int
2818 xfs_off_t offset,
2819 xfs_off_t len,
2821 {
2824 xfs_off_t end_dmi_offset;
2825 xfs_fileoff_t endoffset_fsb;
2827 xfs_fsblock_t firstfsb;
2828 xfs_bmap_free_t free_list;
2829 xfs_bmbt_irec_t imap;
2830 xfs_off_t ioffset;
2834 uint resblks;
2835 uint rounding;
2837 xfs_fileoff_t startoffset_fsb;
2866 }
2872 /* wait for the completion of any pending DIOs */
2874 }
2884 }
2886 /*
2887 * Need to zero the stuff we're not freeing, on disk.
2888 * If it's a realtime file & can't use unwritten extents then we
2889 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2890 * will take care of it for us.
2891 */
2900 xfs_daddr_t block;
2907 }
2916 mod++;
2919 }
2920 }
2922 /*
2923 * One contiguous piece to clear
2924 */
2927 /*
2928 * Some full blocks, possibly two pieces to clear
2929 */
2938 }
2940 /*
2941 * free file space until done or until there is an error
2942 */
2946 /*
2947 * allocate and setup the transaction. Allow this
2948 * transaction to dip into the reserve blocks to ensure
2949 * the freeing of the space succeeds at ENOSPC.
2950 */
2954 resblks,
2957 XFS_TRANS_PERM_LOG_RES,
2958 XFS_WRITE_LOG_COUNT);
2960 /*
2961 * check for running out of space
2962 */
2964 /*
2965 * Free the transaction structure.
2966 */
2970 }
2981 /*
2982 * issue the bunmapi() call to free the blocks
2983 */
2990 }
2992 /*
2993 * complete the transaction
2994 */
2998 }
3002 }
3004 out_unlock_iolock:
3009 error0:
3011 error1:
3014 XFS_ILOCK_EXCL);
3016 }
3018 /*
3019 * xfs_change_file_space()
3020 * This routine allocates or frees disk space for the given file.
3021 * The user specified parameters are checked for alignment and size
3022 * limitations.
3023 *
3024 * RETURNS:
3025 * 0 on success
3026 * errno on error
3027 *
3028 */
3029 int
3034 xfs_off_t offset,
3036 {
3040 xfs_fsize_t fsize;
3042 xfs_off_t startoffset;
3043 xfs_off_t llen;
3063 }
3078 /*
3079 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3080 * file space.
3081 * These calls do NOT zero the data space allocated to the file,
3082 * nor do they change the file size.
3083 *
3084 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3085 * space.
3086 * These calls cause the new file data to be zeroed and the file
3087 * size to be changed.
3088 */
3104 attr_flags)))
3117 }
3133 }
3135 /*
3136 * update the inode timestamp, mode, and prealloc flag bits
3137 */
3142 /* ASSERT(0); */
3145 }
3155 /*
3156 * Note that we don't have to worry about mandatory
3157 * file locking being disabled here because we only
3158 * clear the S_ISGID bit if the Group execute bit is
3159 * on, but if it was on then mandatory locking wouldn't
3160 * have been enabled.
3161 */
3166 }
3180 }