| blob | 0e55c5d7db5fd2bbcc13e88baa38a915c2db94b7 |
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;
90 /*
91 * If disk quotas is on, we make sure that the dquots do exist on disk,
92 * before we start any other transactions. Trying to do this later
93 * is messy. We don't care to take a readlock to look at the ids
94 * in inode here, because we can't hold it across the trans_reserve.
95 * If the IDs do change before we take the ilock, we're covered
96 * because the i_*dquot fields will get updated anyway.
97 */
106 }
112 }
114 /*
115 * We take a reference when we initialize udqp and gdqp,
116 * so it is important that we never blindly double trip on
117 * the same variable. See xfs_create() for an example.
118 */
125 }
127 /*
128 * For the other attributes, we acquire the inode lock and
129 * first do an error checking pass.
130 */
145 }
146 }
156 }
157 }
160 }
164 /*
165 * Change file ownership. Must be the owner or privileged.
166 */
168 /*
169 * These IDs could have changed since we last looked at them.
170 * But, we're assured that if the ownership did change
171 * while we didn't have the inode locked, inode's dquot(s)
172 * would have changed also.
173 */
179 /*
180 * Do a quota reservation only if uid/gid is actually
181 * going to change.
182 */
191 }
192 }
194 /*
195 * Truncate file. Must have write permission and not be a directory.
196 */
198 /* Short circuit the truncate case for zero length files */
207 }
215 }
217 /*
218 * Make sure that the dquots are attached to the inode.
219 */
224 /*
225 * Now we can make the changes. Before we join the inode
226 * to the transaction, if ATTR_SIZE is set then take care of
227 * the part of the truncation that must be done without the
228 * inode lock. This needs to be done before joining the inode
229 * to the transaction, because the inode cannot be unlocked
230 * once it is a part of the transaction.
231 */
233 /*
234 * Do the first part of growing a file: zero any data
235 * in the last block that is beyond the old EOF. We
236 * need to do this before the inode is joined to the
237 * transaction to modify the i_size.
238 */
240 }
243 /*
244 * We are going to log the inode size change in this
245 * transaction so any previous writes that are beyond the on
246 * disk EOF and the new EOF that have not been written out need
247 * to be written here. If we do not write the data out, we
248 * expose ourselves to the null files problem.
249 *
250 * Only flush from the on disk size to the smaller of the in
251 * memory file size or the new size as that's the range we
252 * really care about here and prevents waiting for other data
253 * not within the range we care about here.
254 */
261 }
263 /* wait for all I/O to complete */
273 }
277 XFS_TRANS_PERM_LOG_RES,
278 XFS_ITRUNCATE_LOG_COUNT))) {
283 }
290 /*
291 * Only change the c/mtime if we are changing the size
292 * or we are explicitly asked to change it. This handles
293 * the semantic difference between truncate() and ftruncate()
294 * as implemented in the VFS.
295 */
307 /*
308 * signal a sync transaction unless
309 * we're truncating an already unlinked
310 * file on a wsync filesystem
311 */
313 XFS_DATA_FORK,
319 /*
320 * Truncated "down", so we're removing references
321 * to old data here - if we now delay flushing for
322 * a long time, we expose ourselves unduly to the
323 * notorious NULL files problem. So, we mark this
324 * vnode and flush it when the file is closed, and
325 * do not wait the usual (long) time for writeout.
326 */
328 }
332 }
334 /*
335 * Change file ownership. Must be the owner or privileged.
336 */
338 /*
339 * CAP_FSETID overrides the following restrictions:
340 *
341 * The set-user-ID and set-group-ID bits of a file will be
342 * cleared upon successful return from chown()
343 */
347 }
349 /*
350 * Change the ownerships and register quota modifications
351 * in the transaction.
352 */
359 }
362 }
370 }
373 }
377 }
379 /*
380 * Change file access modes.
381 */
396 }
398 /*
399 * Change file access or modified times.
400 */
407 }
414 }
417 }
419 /*
420 * Change file inode change time only if ATTR_CTIME set
421 * AND we have been called by a DMI function.
422 */
430 }
432 /*
433 * Send out timestamp changes that need to be set to the
434 * current time. Not done when called by a DMI function.
435 */
441 /*
442 * If this is a synchronous mount, make sure that the
443 * transaction goes to disk before returning to the user.
444 * This is slightly sub-optimal in that truncates require
445 * two sync transactions instead of one for wsync filesystems.
446 * One for the truncate and one for the timestamps since we
447 * don't want to change the timestamps unless we're sure the
448 * truncate worked. Truncates are less than 1% of the laddis
449 * mix so this probably isn't worth the trouble to optimize.
450 */
457 }
461 /*
462 * Release any dquot(s) the inode had kept before chown.
463 */
471 }
478 }
481 abort_return:
483 /* FALLTHROUGH */
484 error_return:
489 }
492 }
494 }
496 /*
497 * The maximum pathlen is 1024 bytes. Since the minimum file system
498 * blocksize is 512 bytes, we can get a max of 2 extents back from
499 * bmapi.
500 */
501 #define SYMLINK_MAPS 2
503 STATIC int
507 {
512 xfs_daddr_t d;
534 }
541 }
546 out:
548 }
550 int
554 {
578 }
580 out:
583 }
585 /*
586 * xfs_fsync
587 *
588 * This is called to sync the inode and its data out to disk. We need to hold
589 * the I/O lock while flushing the data, and the inode lock while flushing the
590 * inode. The inode lock CANNOT be held while flushing the data, so acquire
591 * after we're done with that.
592 */
593 int
596 {
606 /* capture size updates in I/O completion before writing the inode. */
611 /*
612 * We always need to make sure that the required inode state is safe on
613 * disk. The vnode might be clean but we still might need to force the
614 * log because of committed transactions that haven't hit the disk yet.
615 * Likewise, there could be unflushed non-transactional changes to the
616 * inode core that have to go to disk and this requires us to issue
617 * a synchronous transaction to capture these changes correctly.
618 *
619 * This code relies on the assumption that if the update_* fields
620 * of the inode are clear and the inode is unpinned then it is clean
621 * and no action is required.
622 */
626 /*
627 * Timestamps/size haven't changed since last inode flush or
628 * inode transaction commit. That means either nothing got
629 * written or a transaction committed which caught the updates.
630 * If the latter happened and the transaction hasn't hit the
631 * disk yet, the inode will be still be pinned. If it is,
632 * force the log.
633 */
642 /*
643 * If the inode is not pinned and nothing has changed
644 * we don't need to flush the cache.
645 */
647 }
649 /*
650 * Kick off a transaction to log the inode core to get the
651 * updates. The sync transaction will also force the log.
652 */
660 }
663 /*
664 * Note - it's possible that we might have pushed ourselves out
665 * of the way during trans_reserve which would flush the inode.
666 * But there's no guarantee that the inode buffer has actually
667 * gone out yet (it's delwri). Plus the buffer could be pinned
668 * anyway if it's part of an inode in another recent
669 * transaction. So we play it safe and fire off the
670 * transaction anyway.
671 */
679 }
682 /*
683 * If the log write didn't issue an ordered tag we need
684 * to flush the disk cache for the data device now.
685 */
689 /*
690 * If this inode is on the RT dev we need to flush that
691 * cache as well.
692 */
695 }
698 }
700 /*
701 * This is called by xfs_inactive to free any blocks beyond eof
702 * when the link count isn't zero and by xfs_dm_punch_hole() when
703 * punching a hole to EOF.
704 */
705 int
710 {
713 xfs_fileoff_t end_fsb;
714 xfs_fileoff_t last_fsb;
715 xfs_filblks_t map_len;
717 xfs_bmbt_irec_t imap;
720 /*
721 * Figure out if there are any blocks beyond the end
722 * of the file. If not, then there is nothing to do.
723 */
739 /*
740 * Attach the dquots to the inode up front.
741 */
745 /*
746 * There are blocks after the end of file.
747 * Free them up now by truncating the file to
748 * its current size.
749 */
752 /*
753 * Do the xfs_itruncate_start() call before
754 * reserving any log space because
755 * itruncate_start will call into the buffer
756 * cache and we can't
757 * do that within a transaction.
758 */
768 }
773 XFS_ITRUNCATE_LOG_COUNT);
779 }
783 XFS_IOLOCK_EXCL |
784 XFS_ILOCK_EXCL);
789 XFS_DATA_FORK,
791 /*
792 * If we get an error at this point we
793 * simply don't bother truncating the file.
794 */
798 XFS_TRANS_ABORT));
801 XFS_TRANS_RELEASE_LOG_RES);
802 }
805 }
807 }
809 /*
810 * Free a symlink that has blocks associated with it.
811 */
812 STATIC int
816 {
821 xfs_fsblock_t first_block;
822 xfs_bmap_free_t free_list;
834 /*
835 * We're freeing a symlink that has some
836 * blocks allocated to it. Free the
837 * blocks here. We know that we've got
838 * either 1 or 2 extents and that we can
839 * free them all in one bunmapi call.
840 */
848 }
849 /*
850 * Lock the inode, fix the size, and join it to the transaction.
851 * Hold it so in the normal path, we still have it locked for
852 * the second transaction. In the error paths we need it
853 * held so the cancel won't rele it, see below.
854 */
861 /*
862 * Find the block(s) so we can inval and unmap them.
863 */
871 /*
872 * Invalidate the block(s).
873 */
879 }
880 /*
881 * Unmap the dead block(s) to the free_list.
882 */
887 /*
888 * Commit the first transaction. This logs the EFI and the inode.
889 */
892 /*
893 * The transaction must have been committed, since there were
894 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
895 * The new tp has the extent freeing and EFDs.
896 */
898 /*
899 * The first xact was committed, so add the inode to the new one.
900 * Mark it dirty so it will be logged and moved forward in the log as
901 * part of every commit.
902 */
906 /*
907 * Get a new, empty transaction to return to our caller.
908 */
910 /*
911 * Commit the transaction containing extent freeing and EFDs.
912 * If we get an error on the commit here or on the reserve below,
913 * we need to unlock the inode since the new transaction doesn't
914 * have the inode attached.
915 */
921 }
922 /*
923 * transaction commit worked ok so we can drop the extra ticket
924 * reference that we gained in xfs_trans_dup()
925 */
928 /*
929 * Remove the memory for extent descriptions (just bookkeeping).
930 */
934 /*
935 * Put an itruncate log reservation in the new transaction
936 * for our caller.
937 */
942 }
943 /*
944 * Return with the inode locked but not joined to the transaction.
945 */
949 error1:
951 error0:
952 /*
953 * Have to come here with the inode locked and either
954 * (held and in the transaction) or (not in the transaction).
955 * If the inode isn't held then cancel would iput it, but
956 * that's wrong since this is inactive and the vnode ref
957 * count is 0 already.
958 * Cancel won't do anything to the inode if held, but it still
959 * needs to be locked until the cancel is done, if it was
960 * joined to the transaction.
961 */
967 }
969 STATIC int
973 {
977 /*
978 * We're freeing a symlink which fit into
979 * the inode. Just free the memory used
980 * to hold the old symlink.
981 */
985 XFS_ITRUNCATE_LOG_COUNT);
991 }
994 /*
995 * Zero length symlinks _can_ exist.
996 */
1000 XFS_DATA_FORK);
1002 }
1004 }
1006 STATIC int
1010 {
1032 XFS_INACTIVE_LOG_COUNT);
1046 error_cancel:
1049 error_unlock:
1053 }
1055 int
1058 {
1065 /* If this is a read-only mount, don't do this (would generate I/O) */
1072 /*
1073 * If we are using filestreams, and we have an unlinked
1074 * file that we are processing the last close on, then nothing
1075 * will be able to reopen and write to this file. Purge this
1076 * inode from the filestreams cache so that it doesn't delay
1077 * teardown of the inode.
1078 */
1082 /*
1083 * If we previously truncated this file and removed old data
1084 * in the process, we want to initiate "early" writeout on
1085 * the last close. This is an attempt to combat the notorious
1086 * NULL files problem which is particularly noticable from a
1087 * truncate down, buffered (re-)write (delalloc), followed by
1088 * a crash. What we are effectively doing here is
1089 * significantly reducing the time window where we'd otherwise
1090 * be exposed to that problem.
1091 */
1095 }
1107 }
1108 }
1111 }
1113 /*
1114 * xfs_inactive
1115 *
1116 * This is called when the vnode reference count for the vnode
1117 * goes to zero. If the file has been unlinked, then it must
1118 * now be truncated. Also, we clear all of the read-ahead state
1119 * kept for the inode here since the file is now closed.
1120 */
1121 int
1124 {
1125 xfs_bmap_free_t free_list;
1126 xfs_fsblock_t first_block;
1135 /*
1136 * If the inode is already free, then there can be nothing
1137 * to clean up here.
1138 */
1143 }
1145 /*
1146 * Only do a truncate if it's a regular file with
1147 * some actual space in it. It's OK to look at the
1148 * inode's fields without the lock because we're the
1149 * only one with a reference to the inode.
1150 */
1163 /* If this is a read-only mount, don't do this (would generate I/O) */
1178 }
1180 }
1189 /*
1190 * Do the xfs_itruncate_start() call before
1191 * reserving any log space because itruncate_start
1192 * will call into the buffer cache and we can't
1193 * do that within a transaction.
1194 */
1202 }
1207 XFS_ITRUNCATE_LOG_COUNT);
1209 /* Don't call itruncate_cleanup */
1214 }
1220 /*
1221 * normally, we have to run xfs_itruncate_finish sync.
1222 * But if filesystem is wsync and we're in the inactive
1223 * path, then we know that nlink == 0, and that the
1224 * xaction that made nlink == 0 is permanently committed
1225 * since xfs_remove runs as a synchronous transaction.
1226 */
1235 }
1238 /*
1239 * If we get an error while cleaning up a
1240 * symlink we bail out.
1241 */
1249 }
1257 XFS_INACTIVE_LOG_COUNT);
1262 }
1267 }
1269 /*
1270 * If there are attributes associated with the file
1271 * then blow them away now. The code calls a routine
1272 * that recursively deconstructs the attribute fork.
1273 * We need to just commit the current transaction
1274 * because we can't use it for xfs_attr_inactive().
1275 */
1278 /*
1279 * If we got an error, the transaction is already
1280 * cancelled, and the inode is unlocked. Just get out.
1281 */
1286 }
1288 /*
1289 * Free the inode.
1290 */
1294 /*
1295 * If we fail to free the inode, shut down. The cancel
1296 * might do that, we need to make sure. Otherwise the
1297 * inode might be lost for a long time or forever.
1298 */
1304 }
1307 /*
1308 * Credit the quota account(s). The inode is gone.
1309 */
1312 /*
1313 * Just ignore errors at this point. There is nothing we can
1314 * do except to try to keep going. Make sure it's not a silent
1315 * error.
1316 */
1325 }
1326 /*
1327 * Release the dquots held by inode, if any.
1328 */
1333 out:
1335 }
1337 /*
1338 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1339 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1340 * ci_name->name will point to a the actual name (caller must free) or
1341 * will be set to NULL if an exact match is found.
1342 */
1343 int
1349 {
1350 xfs_ino_t inum;
1352 uint lock_mode;
1373 out_free_name:
1376 out:
1379 }
1381 int
1385 mode_t mode,
1386 xfs_dev_t rdev,
1389 {
1394 xfs_bmap_free_t free_list;
1395 xfs_fsblock_t first_block;
1398 uint cancel_flags;
1400 xfs_prid_t prid;
1402 uint resblks;
1416 }
1421 /* Return through std_return after this point. */
1426 else
1429 /*
1430 * Make sure that we have allocated dquot(s) on disk.
1431 */
1443 /*
1444 * Initially assume that the file does not exist and
1445 * reserve the resources for that case. If that is not
1446 * the case we'll drop the one we have and get a more
1447 * appropriate transaction later.
1448 */
1455 }
1459 }
1468 /*
1469 * Reserve disk quota and the inode.
1470 */
1485 }
1488 /*
1489 * At this point, we've gotten a newly allocated inode.
1490 * It is locked (and joined to the transaction).
1491 */
1495 /*
1496 * Now we join the directory inode to the transaction. We do not do it
1497 * earlier because xfs_dir_ialloc might commit the previous transaction
1498 * (and release all the locks). An error from here on will result in
1499 * the transaction cancel unlocking dp so don't do it explicitly in the
1500 * error path.
1501 */
1512 }
1516 /*
1517 * If this is a synchronous mount, make sure that the
1518 * create transaction goes to disk before returning to
1519 * the user.
1520 */
1523 }
1525 /*
1526 * Attach the dquot(s) to the inodes and modify them incore.
1527 * These ids of the inode couldn't have changed since the new
1528 * inode has been locked ever since it was created.
1529 */
1532 /*
1533 * xfs_trans_commit normally decrements the vnode ref count
1534 * when it unlocks the inode. Since we want to return the
1535 * vnode to the caller, we bump the vnode ref count now.
1536 */
1543 }
1550 }
1557 /* Fallthrough to std_return with error = 0 */
1559 std_return:
1567 }
1570 abort_return:
1572 /* FALLTHROUGH */
1574 error_return:
1586 abort_rele:
1587 /*
1588 * Wait until after the current transaction is aborted to
1589 * release the inode. This prevents recursive transactions
1590 * and deadlocks from xfs_inactive.
1591 */
1600 }
1602 #ifdef DEBUG
1608 #endif
1610 /*
1611 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1612 * a different value
1613 */
1616 {
1623 }
1625 /*
1626 * The following routine will lock n inodes in exclusive mode.
1627 * We assume the caller calls us with the inodes in i_ino order.
1628 *
1629 * We need to detect deadlock where an inode that we lock
1630 * is in the AIL and we start waiting for another inode that is locked
1631 * by a thread in a long running transaction (such as truncate). This can
1632 * result in deadlock since the long running trans might need to wait
1633 * for the inode we just locked in order to push the tail and free space
1634 * in the log.
1635 */
1636 void
1640 uint lock_mode)
1641 {
1650 again:
1657 /*
1658 * If try_lock is not set yet, make sure all locked inodes
1659 * are not in the AIL.
1660 * If any are, set try_lock to be used later.
1661 */
1667 try_lock++;
1668 }
1669 }
1670 }
1672 /*
1673 * If any of the previous locks we have locked is in the AIL,
1674 * we must TRY to get the second and subsequent locks. If
1675 * we can't get any, we must release all we have
1676 * and try again.
1677 */
1680 /* try_lock must be 0 if i is 0. */
1681 /*
1682 * try_lock means we have an inode locked
1683 * that is in the AIL.
1684 */
1687 attempts++;
1689 /*
1690 * Unlock all previous guys and try again.
1691 * xfs_iunlock will try to push the tail
1692 * if the inode is in the AIL.
1693 */
1697 /*
1698 * Check to see if we've already
1699 * unlocked this one.
1700 * Not the first one going back,
1701 * and the inode ptr is the same.
1702 */
1708 }
1712 #ifdef DEBUG
1713 xfs_lock_delays++;
1714 #endif
1715 }
1719 }
1722 }
1723 }
1725 #ifdef DEBUG
1731 xfs_locked_n++;
1732 }
1733 #endif
1734 }
1736 /*
1737 * xfs_lock_two_inodes() can only be used to lock one type of lock
1738 * at a time - the iolock or the ilock, but not both at once. If
1739 * we lock both at once, lockdep will report false positives saying
1740 * we have violated locking orders.
1741 */
1742 void
1746 uint lock_mode)
1747 {
1760 }
1762 again:
1765 /*
1766 * If the first lock we have locked is in the AIL, we must TRY to get
1767 * the second lock. If we can't get it, we must release the first one
1768 * and try again.
1769 */
1777 }
1780 }
1781 }
1783 int
1788 {
1793 xfs_bmap_free_t free_list;
1794 xfs_fsblock_t first_block;
1798 uint resblks;
1799 uint log_count;
1813 }
1829 }
1832 /*
1833 * We try to get the real space reservation first,
1834 * allowing for directory btree deletion(s) implying
1835 * possible bmap insert(s). If we can't get the space
1836 * reservation then we use 0 instead, and avoid the bmap
1837 * btree insert(s) in the directory code by, if the bmap
1838 * insert tries to happen, instead trimming the LAST
1839 * block from the directory.
1840 */
1848 }
1853 }
1857 /*
1858 * At this point, we've gotten both the directory and the entry
1859 * inodes locked.
1860 */
1867 /*
1868 * If we're removing a directory perform some additional validation.
1869 */
1875 }
1879 }
1880 }
1888 }
1892 /*
1893 * Drop the link from ip's "..".
1894 */
1899 /*
1900 * Drop the "." link from ip to self.
1901 */
1906 /*
1907 * When removing a non-directory we need to log the parent
1908 * inode here. For a directory this is done implicitly
1909 * by the xfs_droplink call for the ".." entry.
1910 */
1912 }
1914 /*
1915 * Drop the link from dp to ip.
1916 */
1921 /*
1922 * Determine if this is the last link while
1923 * we are in the transaction.
1924 */
1927 /*
1928 * If this is a synchronous mount, make sure that the
1929 * remove transaction goes to disk before returning to
1930 * the user.
1931 */
1943 /*
1944 * If we are using filestreams, kill the stream association.
1945 * If the file is still open it may get a new one but that
1946 * will get killed on last close in xfs_close() so we don't
1947 * have to worry about that.
1948 */
1955 std_return:
1960 }
1964 out_bmap_cancel:
1967 out_trans_cancel:
1970 }
1972 int
1977 {
1981 xfs_bmap_free_t free_list;
1982 xfs_fsblock_t first_block;
2002 }
2004 /* Return through std_return after this point. */
2021 }
2025 }
2029 /*
2030 * Increment vnode ref counts since xfs_trans_commit &
2031 * xfs_trans_cancel will both unlock the inodes and
2032 * decrement the associated ref counts.
2033 */
2039 /*
2040 * If the source has too many links, we can't make any more to it.
2041 */
2045 }
2047 /*
2048 * If we are using project inheritance, we only allow hard link
2049 * creation in our tree when the project IDs are the same; else
2050 * the tree quota mechanism could be circumvented.
2051 */
2056 }
2075 /*
2076 * If this is a synchronous mount, make sure that the
2077 * link transaction goes to disk before returning to
2078 * the user.
2079 */
2082 }
2088 }
2094 /* Fall through to std_return with error = 0. */
2095 std_return:
2101 }
2104 abort_return:
2106 /* FALLTHROUGH */
2108 error_return:
2111 }
2114 int
2118 mode_t mode,
2121 {
2128 xfs_bmap_free_t free_list;
2129 xfs_fsblock_t first_block;
2133 xfs_prid_t prid;
2135 uint resblks;
2150 }
2152 /* Return through std_return after this point. */
2160 else
2163 /*
2164 * Make sure that we have allocated dquot(s) on disk.
2165 */
2180 XFS_TRANS_PERM_LOG_RES,
2181 XFS_MKDIR_LOG_COUNT);
2182 }
2186 }
2191 /*
2192 * Check for directory link count overflow.
2193 */
2197 }
2199 /*
2200 * Reserve disk quota and the inode.
2201 */
2209 /*
2210 * create the directory inode.
2211 */
2219 }
2222 /*
2223 * Now we add the directory inode to the transaction.
2224 * We waited until now since xfs_dir_ialloc might start
2225 * a new transaction. Had we joined the transaction
2226 * earlier, the locks might have gotten released. An error
2227 * from here on will result in the transaction cancel
2228 * unlocking dp so don't do it explicitly in the error path.
2229 */
2242 }
2258 /*
2259 * Attach the dquots to the new inode and modify the icount incore.
2260 */
2263 /*
2264 * If this is a synchronous mount, make sure that the
2265 * mkdir transaction goes to disk before returning to
2266 * the user.
2267 */
2270 }
2276 }
2283 }
2285 /* Fall through to std_return with error = 0 or errno from
2286 * xfs_trans_commit. */
2288 std_return:
2294 DM_RIGHT_NULL,
2297 }
2300 error2:
2301 error1:
2303 abort_return:
2305 error_return:
2314 }
2316 int
2321 mode_t mode,
2324 {
2330 xfs_bmap_free_t free_list;
2331 xfs_fsblock_t first_block;
2333 uint cancel_flags;
2335 xfs_fileoff_t first_fsb;
2336 xfs_filblks_t fs_blocks;
2339 xfs_daddr_t d;
2344 xfs_prid_t prid;
2346 uint resblks;
2358 /*
2359 * Check component lengths of the target path name.
2360 */
2371 }
2373 /* Return through std_return after this point. */
2378 else
2381 /*
2382 * Make sure that we have allocated dquot(s) on disk.
2383 */
2392 /*
2393 * The symlink will fit into the inode data fork?
2394 * There can't be any attributes so we get the whole variable part.
2395 */
2398 else
2407 }
2411 }
2416 /*
2417 * Check whether the directory allows new symlinks or not.
2418 */
2422 }
2424 /*
2425 * Reserve disk quota : blocks and inode.
2426 */
2431 /*
2432 * Check for ability to enter directory entry, if no space reserved.
2433 */
2437 /*
2438 * Initialize the bmap freelist prior to calling either
2439 * bmapi or the directory create code.
2440 */
2443 /*
2444 * Allocate an inode for the symlink.
2445 */
2452 }
2455 /*
2456 * An error after we've joined dp to the transaction will result in the
2457 * transaction cancel unlocking dp so don't do it explicitly in the
2458 * error path.
2459 */
2464 /*
2465 * Also attach the dquot(s) to it, if applicable.
2466 */
2471 /*
2472 * If the symlink will fit into the inode, write it inline.
2473 */
2479 /*
2480 * The inode was initially created in extent format.
2481 */
2498 }
2514 }
2521 }
2522 }
2524 /*
2525 * Create the directory entry for the symlink.
2526 */
2534 /*
2535 * If this is a synchronous mount, make sure that the
2536 * symlink transaction goes to disk before returning to
2537 * the user.
2538 */
2541 }
2543 /*
2544 * xfs_trans_commit normally decrements the vnode ref count
2545 * when it unlocks the inode. Since we want to return the
2546 * vnode to the caller, we bump the vnode ref count now.
2547 */
2553 }
2558 /* Fall through to std_return with error = 0 or errno from
2559 * xfs_trans_commit */
2560 std_return:
2567 }
2573 error2:
2575 error1:
2578 error_return:
2587 }
2589 int
2593 {
2600 /*
2601 * Bypass inodes which have already been cleaned by
2602 * the inode flush clustering code inside xfs_iflush
2603 */
2607 /*
2608 * We make this non-blocking if the inode is contended,
2609 * return EAGAIN to indicate to the caller that they
2610 * did not succeed. This prevents the flush path from
2611 * blocking on inodes inside another operation right
2612 * now, they get caught later by xfs_sync.
2613 */
2621 }
2624 }
2631 }
2634 int
2637 u_int evmask,
2638 u_int16_t state)
2639 {
2655 }
2667 }
2669 int
2672 {
2678 /* bad inode, get out here ASAP */
2682 }
2688 /*
2689 * Make sure the atime in the XFS inode is correct before freeing the
2690 * Linux inode.
2691 */
2694 /*
2695 * If we have nothing to flush with this inode then complete the
2696 * teardown now, otherwise break the link between the xfs inode and the
2697 * linux inode and clean up the xfs inode later. This avoids flushing
2698 * the inode to disk during the delete operation itself.
2699 *
2700 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2701 * first to ensure that xfs_iunpin() will never see an xfs inode
2702 * that has a linux inode being reclaimed. Synchronisation is provided
2703 * by the i_flags_lock.
2704 */
2710 }
2713 }
2715 /*
2716 * xfs_alloc_file_space()
2717 * This routine allocates disk space for the given file.
2718 *
2719 * If alloc_type == 0, this request is for an ALLOCSP type
2720 * request which will change the file size. In this case, no
2721 * DMAPI event will be generated by the call. A TRUNCATE event
2722 * will be generated later by xfs_setattr.
2723 *
2724 * If alloc_type != 0, this request is for a RESVSP type
2725 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2726 * lower block boundary byte address is less than the file's
2727 * length.
2728 *
2729 * RETURNS:
2730 * 0 on success
2731 * errno on error
2732 *
2733 */
2734 STATIC int
2737 xfs_off_t offset,
2738 xfs_off_t len,
2741 {
2743 xfs_off_t count;
2744 xfs_filblks_t allocated_fsb;
2745 xfs_filblks_t allocatesize_fsb;
2747 xfs_fileoff_t startoffset_fsb;
2748 xfs_fsblock_t firstfsb;
2755 xfs_bmap_free_t free_list;
2781 /* Generate a DMAPI event if needed. */
2785 xfs_off_t end_dmi_offset;
2794 }
2796 /*
2797 * Allocate file space until done or until there is an error
2798 */
2799 retry:
2803 /*
2804 * Determine space reservations for data/realtime.
2805 */
2818 }
2830 }
2832 /*
2833 * Allocate and setup the transaction.
2834 */
2838 XFS_TRANS_PERM_LOG_RES,
2839 XFS_WRITE_LOG_COUNT);
2840 /*
2841 * Check for running out of space
2842 */
2844 /*
2845 * Free the transaction structure.
2846 */
2850 }
2860 /*
2861 * Issue the xfs_bmapi() call to allocate the blocks
2862 */
2870 }
2872 /*
2873 * Complete the transaction
2874 */
2878 }
2884 }
2891 }
2895 }
2896 dmapi_enospc_check:
2905 /* else fall through with error from XFS_SEND_DATA */
2906 }
2918 }
2920 /*
2921 * Zero file bytes between startoff and endoff inclusive.
2922 * The iolock is held exclusive and no blocks are buffered.
2923 *
2924 * This function is used by xfs_free_file_space() to zero
2925 * partial blocks when the range to free is not block aligned.
2926 * When unreserving space with boundaries that are not block
2927 * aligned we round up the start and round down the end
2928 * boundaries and then use this function to zero the parts of
2929 * the blocks that got dropped during the rounding.
2930 */
2931 STATIC int
2934 xfs_off_t startoff,
2935 xfs_off_t endoff)
2936 {
2937 xfs_bmbt_irec_t imap;
2938 xfs_fileoff_t offset_fsb;
2939 xfs_off_t lastoffset;
2940 xfs_off_t offset;
2946 /*
2947 * Avoid doing I/O beyond eof - it's not necessary
2948 * since nothing can read beyond eof. The space will
2949 * be zeroed when the file is extended anyway.
2950 */
2990 }
3003 }
3004 }
3007 }
3009 /*
3010 * xfs_free_file_space()
3011 * This routine frees disk space for the given file.
3012 *
3013 * This routine is only called by xfs_change_file_space
3014 * for an UNRESVSP type call.
3015 *
3016 * RETURNS:
3017 * 0 on success
3018 * errno on error
3019 *
3020 */
3021 STATIC int
3024 xfs_off_t offset,
3025 xfs_off_t len,
3027 {
3030 xfs_off_t end_dmi_offset;
3031 xfs_fileoff_t endoffset_fsb;
3033 xfs_fsblock_t firstfsb;
3034 xfs_bmap_free_t free_list;
3035 xfs_bmbt_irec_t imap;
3036 xfs_off_t ioffset;
3040 uint resblks;
3041 uint rounding;
3043 xfs_fileoff_t startoffset_fsb;
3071 }
3077 /* wait for the completion of any pending DIOs */
3079 }
3089 }
3091 /*
3092 * Need to zero the stuff we're not freeing, on disk.
3093 * If its a realtime file & can't use unwritten extents then we
3094 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3095 * will take care of it for us.
3096 */
3105 xfs_daddr_t block;
3112 }
3121 mod++;
3124 }
3125 }
3127 /*
3128 * One contiguous piece to clear
3129 */
3132 /*
3133 * Some full blocks, possibly two pieces to clear
3134 */
3143 }
3145 /*
3146 * free file space until done or until there is an error
3147 */
3151 /*
3152 * allocate and setup the transaction. Allow this
3153 * transaction to dip into the reserve blocks to ensure
3154 * the freeing of the space succeeds at ENOSPC.
3155 */
3159 resblks,
3162 XFS_TRANS_PERM_LOG_RES,
3163 XFS_WRITE_LOG_COUNT);
3165 /*
3166 * check for running out of space
3167 */
3169 /*
3170 * Free the transaction structure.
3171 */
3175 }
3179 XFS_QMOPT_RES_REGBLKS);
3186 /*
3187 * issue the bunmapi() call to free the blocks
3188 */
3195 }
3197 /*
3198 * complete the transaction
3199 */
3203 }
3207 }
3209 out_unlock_iolock:
3214 error0:
3216 error1:
3219 XFS_ILOCK_EXCL);
3221 }
3223 /*
3224 * xfs_change_file_space()
3225 * This routine allocates or frees disk space for the given file.
3226 * The user specified parameters are checked for alignment and size
3227 * limitations.
3228 *
3229 * RETURNS:
3230 * 0 on success
3231 * errno on error
3232 *
3233 */
3234 int
3239 xfs_off_t offset,
3241 {
3245 xfs_fsize_t fsize;
3247 xfs_off_t startoffset;
3248 xfs_off_t llen;
3268 }
3283 /*
3284 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3285 * file space.
3286 * These calls do NOT zero the data space allocated to the file,
3287 * nor do they change the file size.
3288 *
3289 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3290 * space.
3291 * These calls cause the new file data to be zeroed and the file
3292 * size to be changed.
3293 */
3309 attr_flags)))
3322 }
3338 }
3340 /*
3341 * update the inode timestamp, mode, and prealloc flag bits
3342 */
3347 /* ASSERT(0); */
3350 }
3360 /*
3361 * Note that we don't have to worry about mandatory
3362 * file locking being disabled here because we only
3363 * clear the S_ISGID bit if the Group execute bit is
3364 * on, but if it was on then mandatory locking wouldn't
3365 * have been enabled.
3366 */
3371 }
3385 }