| blob | 492d75bae2bf22f25ff52fac48953e9e95141a01 |
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;
543 XBF_DONT_BLOCK);
550 }
557 }
562 out:
564 }
566 int
570 {
594 }
596 out:
599 }
601 /*
602 * xfs_fsync
603 *
604 * This is called to sync the inode and its data out to disk. We need to hold
605 * the I/O lock while flushing the data, and the inode lock while flushing the
606 * inode. The inode lock CANNOT be held while flushing the data, so acquire
607 * after we're done with that.
608 */
609 int
612 {
622 /* capture size updates in I/O completion before writing the inode. */
627 /*
628 * We always need to make sure that the required inode state is safe on
629 * disk. The vnode might be clean but we still might need to force the
630 * log because of committed transactions that haven't hit the disk yet.
631 * Likewise, there could be unflushed non-transactional changes to the
632 * inode core that have to go to disk and this requires us to issue
633 * a synchronous transaction to capture these changes correctly.
634 *
635 * This code relies on the assumption that if the update_* fields
636 * of the inode are clear and the inode is unpinned then it is clean
637 * and no action is required.
638 */
642 /*
643 * Timestamps/size haven't changed since last inode flush or
644 * inode transaction commit. That means either nothing got
645 * written or a transaction committed which caught the updates.
646 * If the latter happened and the transaction hasn't hit the
647 * disk yet, the inode will be still be pinned. If it is,
648 * force the log.
649 */
658 /*
659 * If the inode is not pinned and nothing has changed
660 * we don't need to flush the cache.
661 */
663 }
665 /*
666 * Kick off a transaction to log the inode core to get the
667 * updates. The sync transaction will also force the log.
668 */
676 }
679 /*
680 * Note - it's possible that we might have pushed ourselves out
681 * of the way during trans_reserve which would flush the inode.
682 * But there's no guarantee that the inode buffer has actually
683 * gone out yet (it's delwri). Plus the buffer could be pinned
684 * anyway if it's part of an inode in another recent
685 * transaction. So we play it safe and fire off the
686 * transaction anyway.
687 */
695 }
698 /*
699 * If the log write didn't issue an ordered tag we need
700 * to flush the disk cache for the data device now.
701 */
705 /*
706 * If this inode is on the RT dev we need to flush that
707 * cache as well.
708 */
711 }
714 }
716 /*
717 * This is called by xfs_inactive to free any blocks beyond eof
718 * when the link count isn't zero and by xfs_dm_punch_hole() when
719 * punching a hole to EOF.
720 */
721 int
726 {
729 xfs_fileoff_t end_fsb;
730 xfs_fileoff_t last_fsb;
731 xfs_filblks_t map_len;
733 xfs_bmbt_irec_t imap;
736 /*
737 * Figure out if there are any blocks beyond the end
738 * of the file. If not, then there is nothing to do.
739 */
755 /*
756 * Attach the dquots to the inode up front.
757 */
762 /*
763 * There are blocks after the end of file.
764 * Free them up now by truncating the file to
765 * its current size.
766 */
769 /*
770 * Do the xfs_itruncate_start() call before
771 * reserving any log space because
772 * itruncate_start will call into the buffer
773 * cache and we can't
774 * do that within a transaction.
775 */
785 }
790 XFS_ITRUNCATE_LOG_COUNT);
796 }
800 XFS_IOLOCK_EXCL |
801 XFS_ILOCK_EXCL);
806 XFS_DATA_FORK,
808 /*
809 * If we get an error at this point we
810 * simply don't bother truncating the file.
811 */
815 XFS_TRANS_ABORT));
818 XFS_TRANS_RELEASE_LOG_RES);
819 }
822 }
824 }
826 /*
827 * Free a symlink that has blocks associated with it.
828 */
829 STATIC int
833 {
838 xfs_fsblock_t first_block;
839 xfs_bmap_free_t free_list;
851 /*
852 * We're freeing a symlink that has some
853 * blocks allocated to it. Free the
854 * blocks here. We know that we've got
855 * either 1 or 2 extents and that we can
856 * free them all in one bunmapi call.
857 */
865 }
866 /*
867 * Lock the inode, fix the size, and join it to the transaction.
868 * Hold it so in the normal path, we still have it locked for
869 * the second transaction. In the error paths we need it
870 * held so the cancel won't rele it, see below.
871 */
878 /*
879 * Find the block(s) so we can inval and unmap them.
880 */
888 /*
889 * Invalidate the block(s).
890 */
896 }
897 /*
898 * Unmap the dead block(s) to the free_list.
899 */
904 /*
905 * Commit the first transaction. This logs the EFI and the inode.
906 */
909 /*
910 * The transaction must have been committed, since there were
911 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
912 * The new tp has the extent freeing and EFDs.
913 */
915 /*
916 * The first xact was committed, so add the inode to the new one.
917 * Mark it dirty so it will be logged and moved forward in the log as
918 * part of every commit.
919 */
923 /*
924 * Get a new, empty transaction to return to our caller.
925 */
927 /*
928 * Commit the transaction containing extent freeing and EFDs.
929 * If we get an error on the commit here or on the reserve below,
930 * we need to unlock the inode since the new transaction doesn't
931 * have the inode attached.
932 */
938 }
939 /*
940 * transaction commit worked ok so we can drop the extra ticket
941 * reference that we gained in xfs_trans_dup()
942 */
945 /*
946 * Remove the memory for extent descriptions (just bookkeeping).
947 */
951 /*
952 * Put an itruncate log reservation in the new transaction
953 * for our caller.
954 */
959 }
960 /*
961 * Return with the inode locked but not joined to the transaction.
962 */
966 error1:
968 error0:
969 /*
970 * Have to come here with the inode locked and either
971 * (held and in the transaction) or (not in the transaction).
972 * If the inode isn't held then cancel would iput it, but
973 * that's wrong since this is inactive and the vnode ref
974 * count is 0 already.
975 * Cancel won't do anything to the inode if held, but it still
976 * needs to be locked until the cancel is done, if it was
977 * joined to the transaction.
978 */
984 }
986 STATIC int
990 {
994 /*
995 * We're freeing a symlink which fit into
996 * the inode. Just free the memory used
997 * to hold the old symlink.
998 */
1002 XFS_ITRUNCATE_LOG_COUNT);
1008 }
1011 /*
1012 * Zero length symlinks _can_ exist.
1013 */
1017 XFS_DATA_FORK);
1019 }
1021 }
1023 STATIC int
1027 {
1049 XFS_INACTIVE_LOG_COUNT);
1063 error_cancel:
1066 error_unlock:
1070 }
1072 int
1075 {
1082 /* If this is a read-only mount, don't do this (would generate I/O) */
1089 /*
1090 * If we are using filestreams, and we have an unlinked
1091 * file that we are processing the last close on, then nothing
1092 * will be able to reopen and write to this file. Purge this
1093 * inode from the filestreams cache so that it doesn't delay
1094 * teardown of the inode.
1095 */
1099 /*
1100 * If we previously truncated this file and removed old data
1101 * in the process, we want to initiate "early" writeout on
1102 * the last close. This is an attempt to combat the notorious
1103 * NULL files problem which is particularly noticable from a
1104 * truncate down, buffered (re-)write (delalloc), followed by
1105 * a crash. What we are effectively doing here is
1106 * significantly reducing the time window where we'd otherwise
1107 * be exposed to that problem.
1108 */
1112 }
1124 }
1125 }
1128 }
1130 /*
1131 * xfs_inactive
1132 *
1133 * This is called when the vnode reference count for the vnode
1134 * goes to zero. If the file has been unlinked, then it must
1135 * now be truncated. Also, we clear all of the read-ahead state
1136 * kept for the inode here since the file is now closed.
1137 */
1138 int
1141 {
1142 xfs_bmap_free_t free_list;
1143 xfs_fsblock_t first_block;
1152 /*
1153 * If the inode is already free, then there can be nothing
1154 * to clean up here.
1155 */
1160 }
1162 /*
1163 * Only do a truncate if it's a regular file with
1164 * some actual space in it. It's OK to look at the
1165 * inode's fields without the lock because we're the
1166 * only one with a reference to the inode.
1167 */
1180 /* If this is a read-only mount, don't do this (would generate I/O) */
1195 }
1197 }
1207 /*
1208 * Do the xfs_itruncate_start() call before
1209 * reserving any log space because itruncate_start
1210 * will call into the buffer cache and we can't
1211 * do that within a transaction.
1212 */
1220 }
1225 XFS_ITRUNCATE_LOG_COUNT);
1227 /* Don't call itruncate_cleanup */
1232 }
1238 /*
1239 * normally, we have to run xfs_itruncate_finish sync.
1240 * But if filesystem is wsync and we're in the inactive
1241 * path, then we know that nlink == 0, and that the
1242 * xaction that made nlink == 0 is permanently committed
1243 * since xfs_remove runs as a synchronous transaction.
1244 */
1253 }
1256 /*
1257 * If we get an error while cleaning up a
1258 * symlink we bail out.
1259 */
1267 }
1275 XFS_INACTIVE_LOG_COUNT);
1280 }
1285 }
1287 /*
1288 * If there are attributes associated with the file
1289 * then blow them away now. The code calls a routine
1290 * that recursively deconstructs the attribute fork.
1291 * We need to just commit the current transaction
1292 * because we can't use it for xfs_attr_inactive().
1293 */
1296 /*
1297 * If we got an error, the transaction is already
1298 * cancelled, and the inode is unlocked. Just get out.
1299 */
1304 }
1306 /*
1307 * Free the inode.
1308 */
1312 /*
1313 * If we fail to free the inode, shut down. The cancel
1314 * might do that, we need to make sure. Otherwise the
1315 * inode might be lost for a long time or forever.
1316 */
1322 }
1325 /*
1326 * Credit the quota account(s). The inode is gone.
1327 */
1330 /*
1331 * Just ignore errors at this point. There is nothing we can
1332 * do except to try to keep going. Make sure it's not a silent
1333 * error.
1334 */
1343 }
1345 /*
1346 * Release the dquots held by inode, if any.
1347 */
1351 out:
1353 }
1355 /*
1356 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1357 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1358 * ci_name->name will point to a the actual name (caller must free) or
1359 * will be set to NULL if an exact match is found.
1360 */
1361 int
1367 {
1368 xfs_ino_t inum;
1370 uint lock_mode;
1391 out_free_name:
1394 out:
1397 }
1399 int
1403 mode_t mode,
1404 xfs_dev_t rdev,
1407 {
1413 xfs_bmap_free_t free_list;
1414 xfs_fsblock_t first_block;
1416 uint cancel_flags;
1418 xfs_prid_t prid;
1421 uint resblks;
1422 uint log_res;
1423 uint log_count;
1438 }
1442 else
1445 /*
1446 * Make sure that we have allocated dquot(s) on disk.
1447 */
1464 }
1468 /*
1469 * Initially assume that the file does not exist and
1470 * reserve the resources for that case. If that is not
1471 * the case we'll drop the one we have and get a more
1472 * appropriate transaction later.
1473 */
1477 /* flush outstanding delalloc blocks and retry */
1481 }
1483 /* No space at all so try a "no-allocation" reservation */
1487 }
1491 }
1496 /*
1497 * Check for directory link count overflow.
1498 */
1502 }
1506 /*
1507 * Reserve disk quota and the inode.
1508 */
1517 /*
1518 * A newly created regular or special file just has one directory
1519 * entry pointing to them, but a directory also the "." entry
1520 * pointing to itself.
1521 */
1528 }
1530 /*
1531 * At this point, we've gotten a newly allocated inode.
1532 * It is locked (and joined to the transaction).
1533 */
1537 /*
1538 * Now we join the directory inode to the transaction. We do not do it
1539 * earlier because xfs_dir_ialloc might commit the previous transaction
1540 * (and release all the locks). An error from here on will result in
1541 * the transaction cancel unlocking dp so don't do it explicitly in the
1542 * error path.
1543 */
1554 }
1566 }
1568 /*
1569 * If this is a synchronous mount, make sure that the
1570 * create transaction goes to disk before returning to
1571 * the user.
1572 */
1576 /*
1577 * Attach the dquot(s) to the inodes and modify them incore.
1578 * These ids of the inode couldn't have changed since the new
1579 * inode has been locked ever since it was created.
1580 */
1583 /*
1584 * xfs_trans_commit normally decrements the vnode ref count
1585 * when it unlocks the inode. Since we want to return the
1586 * vnode to the caller, we bump the vnode ref count now.
1587 */
1598 }
1605 /* Fallthrough to std_return with error = 0 */
1606 std_return:
1611 }
1615 out_bmap_cancel:
1617 out_trans_abort:
1619 out_trans_cancel:
1621 out_dqrele:
1630 out_abort_rele:
1631 /*
1632 * Wait until after the current transaction is aborted to
1633 * release the inode. This prevents recursive transactions
1634 * and deadlocks from xfs_inactive.
1635 */
1642 }
1644 #ifdef DEBUG
1650 #endif
1652 /*
1653 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1654 * a different value
1655 */
1658 {
1665 }
1667 /*
1668 * The following routine will lock n inodes in exclusive mode.
1669 * We assume the caller calls us with the inodes in i_ino order.
1670 *
1671 * We need to detect deadlock where an inode that we lock
1672 * is in the AIL and we start waiting for another inode that is locked
1673 * by a thread in a long running transaction (such as truncate). This can
1674 * result in deadlock since the long running trans might need to wait
1675 * for the inode we just locked in order to push the tail and free space
1676 * in the log.
1677 */
1678 void
1682 uint lock_mode)
1683 {
1692 again:
1699 /*
1700 * If try_lock is not set yet, make sure all locked inodes
1701 * are not in the AIL.
1702 * If any are, set try_lock to be used later.
1703 */
1709 try_lock++;
1710 }
1711 }
1712 }
1714 /*
1715 * If any of the previous locks we have locked is in the AIL,
1716 * we must TRY to get the second and subsequent locks. If
1717 * we can't get any, we must release all we have
1718 * and try again.
1719 */
1722 /* try_lock must be 0 if i is 0. */
1723 /*
1724 * try_lock means we have an inode locked
1725 * that is in the AIL.
1726 */
1729 attempts++;
1731 /*
1732 * Unlock all previous guys and try again.
1733 * xfs_iunlock will try to push the tail
1734 * if the inode is in the AIL.
1735 */
1739 /*
1740 * Check to see if we've already
1741 * unlocked this one.
1742 * Not the first one going back,
1743 * and the inode ptr is the same.
1744 */
1750 }
1754 #ifdef DEBUG
1755 xfs_lock_delays++;
1756 #endif
1757 }
1761 }
1764 }
1765 }
1767 #ifdef DEBUG
1773 xfs_locked_n++;
1774 }
1775 #endif
1776 }
1778 /*
1779 * xfs_lock_two_inodes() can only be used to lock one type of lock
1780 * at a time - the iolock or the ilock, but not both at once. If
1781 * we lock both at once, lockdep will report false positives saying
1782 * we have violated locking orders.
1783 */
1784 void
1788 uint lock_mode)
1789 {
1802 }
1804 again:
1807 /*
1808 * If the first lock we have locked is in the AIL, we must TRY to get
1809 * the second lock. If we can't get it, we must release the first one
1810 * and try again.
1811 */
1819 }
1822 }
1823 }
1825 int
1830 {
1835 xfs_bmap_free_t free_list;
1836 xfs_fsblock_t first_block;
1840 uint resblks;
1841 uint log_count;
1855 }
1871 }
1874 /*
1875 * We try to get the real space reservation first,
1876 * allowing for directory btree deletion(s) implying
1877 * possible bmap insert(s). If we can't get the space
1878 * reservation then we use 0 instead, and avoid the bmap
1879 * btree insert(s) in the directory code by, if the bmap
1880 * insert tries to happen, instead trimming the LAST
1881 * block from the directory.
1882 */
1890 }
1895 }
1899 /*
1900 * At this point, we've gotten both the directory and the entry
1901 * inodes locked.
1902 */
1909 /*
1910 * If we're removing a directory perform some additional validation.
1911 */
1917 }
1921 }
1922 }
1930 }
1934 /*
1935 * Drop the link from ip's "..".
1936 */
1941 /*
1942 * Drop the "." link from ip to self.
1943 */
1948 /*
1949 * When removing a non-directory we need to log the parent
1950 * inode here. For a directory this is done implicitly
1951 * by the xfs_droplink call for the ".." entry.
1952 */
1954 }
1956 /*
1957 * Drop the link from dp to ip.
1958 */
1963 /*
1964 * Determine if this is the last link while
1965 * we are in the transaction.
1966 */
1969 /*
1970 * If this is a synchronous mount, make sure that the
1971 * remove transaction goes to disk before returning to
1972 * the user.
1973 */
1985 /*
1986 * If we are using filestreams, kill the stream association.
1987 * If the file is still open it may get a new one but that
1988 * will get killed on last close in xfs_close() so we don't
1989 * have to worry about that.
1990 */
1997 std_return:
2002 }
2006 out_bmap_cancel:
2009 out_trans_cancel:
2012 }
2014 int
2019 {
2023 xfs_bmap_free_t free_list;
2024 xfs_fsblock_t first_block;
2044 }
2046 /* Return through std_return after this point. */
2065 }
2069 }
2073 /*
2074 * Increment vnode ref counts since xfs_trans_commit &
2075 * xfs_trans_cancel will both unlock the inodes and
2076 * decrement the associated ref counts.
2077 */
2083 /*
2084 * If the source has too many links, we can't make any more to it.
2085 */
2089 }
2091 /*
2092 * If we are using project inheritance, we only allow hard link
2093 * creation in our tree when the project IDs are the same; else
2094 * the tree quota mechanism could be circumvented.
2095 */
2100 }
2119 /*
2120 * If this is a synchronous mount, make sure that the
2121 * link transaction goes to disk before returning to
2122 * the user.
2123 */
2126 }
2132 }
2138 /* Fall through to std_return with error = 0. */
2139 std_return:
2145 }
2148 abort_return:
2150 /* FALLTHROUGH */
2152 error_return:
2155 }
2157 int
2162 mode_t mode,
2165 {
2171 xfs_bmap_free_t free_list;
2172 xfs_fsblock_t first_block;
2174 uint cancel_flags;
2176 xfs_fileoff_t first_fsb;
2177 xfs_filblks_t fs_blocks;
2180 xfs_daddr_t d;
2185 xfs_prid_t prid;
2187 uint resblks;
2199 /*
2200 * Check component lengths of the target path name.
2201 */
2212 }
2214 /* Return through std_return after this point. */
2219 else
2222 /*
2223 * Make sure that we have allocated dquot(s) on disk.
2224 */
2232 /*
2233 * The symlink will fit into the inode data fork?
2234 * There can't be any attributes so we get the whole variable part.
2235 */
2238 else
2247 }
2251 }
2256 /*
2257 * Check whether the directory allows new symlinks or not.
2258 */
2262 }
2264 /*
2265 * Reserve disk quota : blocks and inode.
2266 */
2271 /*
2272 * Check for ability to enter directory entry, if no space reserved.
2273 */
2277 /*
2278 * Initialize the bmap freelist prior to calling either
2279 * bmapi or the directory create code.
2280 */
2283 /*
2284 * Allocate an inode for the symlink.
2285 */
2292 }
2295 /*
2296 * An error after we've joined dp to the transaction will result in the
2297 * transaction cancel unlocking dp so don't do it explicitly in the
2298 * error path.
2299 */
2304 /*
2305 * Also attach the dquot(s) to it, if applicable.
2306 */
2311 /*
2312 * If the symlink will fit into the inode, write it inline.
2313 */
2319 /*
2320 * The inode was initially created in extent format.
2321 */
2338 }
2354 }
2361 }
2362 }
2364 /*
2365 * Create the directory entry for the symlink.
2366 */
2374 /*
2375 * If this is a synchronous mount, make sure that the
2376 * symlink transaction goes to disk before returning to
2377 * the user.
2378 */
2381 }
2383 /*
2384 * xfs_trans_commit normally decrements the vnode ref count
2385 * when it unlocks the inode. Since we want to return the
2386 * vnode to the caller, we bump the vnode ref count now.
2387 */
2393 }
2398 /* Fall through to std_return with error = 0 or errno from
2399 * xfs_trans_commit */
2400 std_return:
2407 }
2413 error2:
2415 error1:
2418 error_return:
2427 }
2429 int
2432 u_int evmask,
2433 u_int16_t state)
2434 {
2450 }
2462 }
2464 int
2467 {
2473 /* bad inode, get out here ASAP */
2477 }
2483 /*
2484 * Make sure the atime in the XFS inode is correct before freeing the
2485 * Linux inode.
2486 */
2489 /*
2490 * If we have nothing to flush with this inode then complete the
2491 * teardown now, otherwise break the link between the xfs inode and the
2492 * linux inode and clean up the xfs inode later. This avoids flushing
2493 * the inode to disk during the delete operation itself.
2494 *
2495 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2496 * first to ensure that xfs_iunpin() will never see an xfs inode
2497 * that has a linux inode being reclaimed. Synchronisation is provided
2498 * by the i_flags_lock.
2499 */
2505 }
2508 }
2510 /*
2511 * xfs_alloc_file_space()
2512 * This routine allocates disk space for the given file.
2513 *
2514 * If alloc_type == 0, this request is for an ALLOCSP type
2515 * request which will change the file size. In this case, no
2516 * DMAPI event will be generated by the call. A TRUNCATE event
2517 * will be generated later by xfs_setattr.
2518 *
2519 * If alloc_type != 0, this request is for a RESVSP type
2520 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2521 * lower block boundary byte address is less than the file's
2522 * length.
2523 *
2524 * RETURNS:
2525 * 0 on success
2526 * errno on error
2527 *
2528 */
2529 STATIC int
2532 xfs_off_t offset,
2533 xfs_off_t len,
2536 {
2538 xfs_off_t count;
2539 xfs_filblks_t allocated_fsb;
2540 xfs_filblks_t allocatesize_fsb;
2542 xfs_fileoff_t startoffset_fsb;
2543 xfs_fsblock_t firstfsb;
2550 xfs_bmap_free_t free_list;
2577 /* Generate a DMAPI event if needed. */
2581 xfs_off_t end_dmi_offset;
2590 }
2592 /*
2593 * Allocate file space until done or until there is an error
2594 */
2595 retry:
2599 /*
2600 * Determine space reservations for data/realtime.
2601 */
2614 }
2626 }
2628 /*
2629 * Allocate and setup the transaction.
2630 */
2634 XFS_TRANS_PERM_LOG_RES,
2635 XFS_WRITE_LOG_COUNT);
2636 /*
2637 * Check for running out of space
2638 */
2640 /*
2641 * Free the transaction structure.
2642 */
2646 }
2656 /*
2657 * Issue the xfs_bmapi() call to allocate the blocks
2658 */
2666 }
2668 /*
2669 * Complete the transaction
2670 */
2674 }
2680 }
2687 }
2691 }
2692 dmapi_enospc_check:
2701 /* else fall through with error from XFS_SEND_DATA */
2702 }
2714 }
2716 /*
2717 * Zero file bytes between startoff and endoff inclusive.
2718 * The iolock is held exclusive and no blocks are buffered.
2719 *
2720 * This function is used by xfs_free_file_space() to zero
2721 * partial blocks when the range to free is not block aligned.
2722 * When unreserving space with boundaries that are not block
2723 * aligned we round up the start and round down the end
2724 * boundaries and then use this function to zero the parts of
2725 * the blocks that got dropped during the rounding.
2726 */
2727 STATIC int
2730 xfs_off_t startoff,
2731 xfs_off_t endoff)
2732 {
2733 xfs_bmbt_irec_t imap;
2734 xfs_fileoff_t offset_fsb;
2735 xfs_off_t lastoffset;
2736 xfs_off_t offset;
2742 /*
2743 * Avoid doing I/O beyond eof - it's not necessary
2744 * since nothing can read beyond eof. The space will
2745 * be zeroed when the file is extended anyway.
2746 */
2786 }
2799 }
2800 }
2803 }
2805 /*
2806 * xfs_free_file_space()
2807 * This routine frees disk space for the given file.
2808 *
2809 * This routine is only called by xfs_change_file_space
2810 * for an UNRESVSP type call.
2811 *
2812 * RETURNS:
2813 * 0 on success
2814 * errno on error
2815 *
2816 */
2817 STATIC int
2820 xfs_off_t offset,
2821 xfs_off_t len,
2823 {
2826 xfs_off_t end_dmi_offset;
2827 xfs_fileoff_t endoffset_fsb;
2829 xfs_fsblock_t firstfsb;
2830 xfs_bmap_free_t free_list;
2831 xfs_bmbt_irec_t imap;
2832 xfs_off_t ioffset;
2836 uint resblks;
2837 uint rounding;
2839 xfs_fileoff_t startoffset_fsb;
2868 }
2874 /* wait for the completion of any pending DIOs */
2876 }
2886 }
2888 /*
2889 * Need to zero the stuff we're not freeing, on disk.
2890 * If it's a realtime file & can't use unwritten extents then we
2891 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2892 * will take care of it for us.
2893 */
2902 xfs_daddr_t block;
2909 }
2918 mod++;
2921 }
2922 }
2924 /*
2925 * One contiguous piece to clear
2926 */
2929 /*
2930 * Some full blocks, possibly two pieces to clear
2931 */
2940 }
2942 /*
2943 * free file space until done or until there is an error
2944 */
2948 /*
2949 * allocate and setup the transaction. Allow this
2950 * transaction to dip into the reserve blocks to ensure
2951 * the freeing of the space succeeds at ENOSPC.
2952 */
2956 resblks,
2959 XFS_TRANS_PERM_LOG_RES,
2960 XFS_WRITE_LOG_COUNT);
2962 /*
2963 * check for running out of space
2964 */
2966 /*
2967 * Free the transaction structure.
2968 */
2972 }
2983 /*
2984 * issue the bunmapi() call to free the blocks
2985 */
2992 }
2994 /*
2995 * complete the transaction
2996 */
3000 }
3004 }
3006 out_unlock_iolock:
3011 error0:
3013 error1:
3016 XFS_ILOCK_EXCL);
3018 }
3020 /*
3021 * xfs_change_file_space()
3022 * This routine allocates or frees disk space for the given file.
3023 * The user specified parameters are checked for alignment and size
3024 * limitations.
3025 *
3026 * RETURNS:
3027 * 0 on success
3028 * errno on error
3029 *
3030 */
3031 int
3036 xfs_off_t offset,
3038 {
3042 xfs_fsize_t fsize;
3044 xfs_off_t startoffset;
3045 xfs_off_t llen;
3065 }
3080 /*
3081 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3082 * file space.
3083 * These calls do NOT zero the data space allocated to the file,
3084 * nor do they change the file size.
3085 *
3086 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3087 * space.
3088 * These calls cause the new file data to be zeroed and the file
3089 * size to be changed.
3090 */
3106 attr_flags)))
3119 }
3135 }
3137 /*
3138 * update the inode timestamp, mode, and prealloc flag bits
3139 */
3144 /* ASSERT(0); */
3147 }
3157 /*
3158 * Note that we don't have to worry about mandatory
3159 * file locking being disabled here because we only
3160 * clear the S_ISGID bit if the Group execute bit is
3161 * on, but if it was on then mandatory locking wouldn't
3162 * have been enabled.
3163 */
3168 }
3182 }