| blob | 82a282ab63dc57ae44d04fbc7262affa5cb2d678 |
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 */
18 #include <linux/log2.h>
54 /*
55 * Used in xfs_itruncate(). This is the maximum number of extents
56 * freed from a file in a single transaction.
57 */
58 #define XFS_ITRUNC_MAX_EXTENTS 2
65 #ifdef DEBUG
66 /*
67 * Make sure that the extents in the given memory buffer
68 * are valid.
69 */
70 STATIC void
74 xfs_exntfmt_t fmt)
75 {
76 xfs_bmbt_irec_t irec;
77 xfs_bmbt_rec_host_t rec;
87 }
88 }
90 #define xfs_validate_extents(ifp, nrecs, fmt)
93 /*
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
96 */
97 #if defined(DEBUG)
98 void
102 {
115 bp);
117 }
118 }
119 }
120 #endif
122 /*
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
126 */
127 STATIC int
133 uint buf_flags,
134 uint iget_flags)
135 {
150 }
152 }
154 /*
155 * Validate the magic number and version of every inode in the buffer
156 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
157 */
158 #ifdef DEBUG
162 #endif
173 XFS_ERRTAG_ITOBP_INOTOBP,
174 XFS_RANDOM_ITOBP_INOTOBP))) {
178 }
181 #ifdef DEBUG
187 #endif
190 }
191 }
195 /*
196 * Mark the buffer as an inode buffer now that it looks good
197 */
202 }
204 /*
205 * This routine is called to map an inode number within a file
206 * system to the buffer containing the on-disk version of the
207 * inode. It returns a pointer to the buffer containing the
208 * on-disk inode in the bpp parameter, and in the dip parameter
209 * it returns a pointer to the on-disk inode within that buffer.
210 *
211 * If a non-zero error is returned, then the contents of bpp and
212 * dipp are undefined.
213 *
214 * Use xfs_imap() to determine the size and location of the
215 * buffer to read from disk.
216 */
217 int
221 xfs_ino_t ino,
225 uint imap_flags)
226 {
244 }
247 /*
248 * This routine is called to map an inode to the buffer containing
249 * the on-disk version of the inode. It returns a pointer to the
250 * buffer containing the on-disk inode in the bpp parameter, and in
251 * the dip parameter it returns a pointer to the on-disk inode within
252 * that buffer.
253 *
254 * If a non-zero error is returned, then the contents of bpp and
255 * dipp are undefined.
256 *
257 * The inode is expected to already been mapped to its buffer and read
258 * in once, thus we can use the mapping information stored in the inode
259 * rather than calling xfs_imap(). This allows us to avoid the overhead
260 * of looking at the inode btree for small block file systems
261 * (see xfs_imap()).
262 */
263 int
270 uint buf_flags)
271 {
286 }
291 }
293 /*
294 * Move inode type and inode format specific information from the
295 * on-disk inode to the in-core inode. For fifos, devs, and sockets
296 * this means set if_rdev to the proper value. For files, directories,
297 * and symlinks this means to bring in the in-line data or extent
298 * pointers. For a file in B-tree format, only the root is immediately
299 * brought in-core. The rest will be in-lined in if_extents when it
300 * is first referenced (see xfs_iread_extents()).
301 */
302 STATIC int
306 {
310 xfs_fsize_t di_size;
328 }
337 }
347 }
358 }
369 /*
370 * no local regular files yet
371 */
377 XFS_ERRLEVEL_LOW,
380 }
389 XFS_ERRLEVEL_LOW,
392 }
407 }
413 }
416 }
434 XFS_ERRLEVEL_LOW,
437 }
450 }
455 }
457 }
459 /*
460 * The file is in-lined in the on-disk inode.
461 * If it fits into if_inline_data, then copy
462 * it there, otherwise allocate a buffer for it
463 * and copy the data there. Either way, set
464 * if_data to point at the data.
465 * If we allocate a buffer for the data, make
466 * sure that its size is a multiple of 4 and
467 * record the real size in i_real_bytes.
468 */
469 STATIC int
475 {
479 /*
480 * If the size is unreasonable, then something
481 * is wrong and we just bail out rather than crash in
482 * kmem_alloc() or memcpy() below.
483 */
492 }
502 }
510 }
512 /*
513 * The file consists of a set of extents all
514 * of which fit into the on-disk inode.
515 * If there are few enough extents to fit into
516 * the if_inline_ext, then copy them there.
517 * Otherwise allocate a buffer for them and copy
518 * them into it. Either way, set if_extents
519 * to point at the extents.
520 */
521 STATIC int
526 {
537 /*
538 * If the number of extents is unreasonable, then something
539 * is wrong and we just bail out rather than crash in
540 * kmem_alloc() or memcpy() below.
541 */
548 }
555 else
566 }
573 XFS_ERRLEVEL_LOW,
576 }
577 }
580 }
582 /*
583 * The file has too many extents to fit into
584 * the inode, so they are in B-tree format.
585 * Allocate a buffer for the root of the B-tree
586 * and copy the root into it. The i_extents
587 * field will remain NULL until all of the
588 * extents are read in (when they are needed).
589 */
590 STATIC int
595 {
598 /* REFERENCED */
607 /*
608 * blow out if -- fork has less extents than can fit in
609 * fork (fork shouldn't be a btree format), root btree
610 * block has more records than can fit into the fork,
611 * or the number of extents is greater than the number of
612 * blocks.
613 */
623 }
628 /*
629 * Copy and convert from the on-disk structure
630 * to the in-memory structure.
631 */
639 }
641 STATIC void
645 {
675 }
677 void
681 {
711 }
713 STATIC uint
715 __uint16_t di_flags)
716 {
748 }
751 }
753 uint
756 {
761 }
763 uint
766 {
769 }
771 /*
772 * Read the disk inode attributes into the in-core inode structure.
773 */
774 int
779 uint iget_flags)
780 {
785 /*
786 * Fill in the location information in the in-core inode.
787 */
792 /*
793 * Get pointers to the on-disk inode and the buffer containing it.
794 */
801 /*
802 * If we got something that isn't an inode it means someone
803 * (nfs or dmi) has a stale handle.
804 */
806 #ifdef DEBUG
813 }
815 /*
816 * If the on-disk inode is already linked to a directory
817 * entry, copy all of the inode into the in-core inode.
818 * xfs_iformat() handles copying in the inode format
819 * specific information.
820 * Otherwise, just get the truly permanent information.
821 */
826 #ifdef DEBUG
831 }
837 /*
838 * Make sure to pull in the mode here as well in
839 * case the inode is released without being used.
840 * This ensures that xfs_inactive() will see that
841 * the inode is already free and not try to mess
842 * with the uninitialized part of it.
843 */
845 /*
846 * Initialize the per-fork minima and maxima for a new
847 * inode here. xfs_iformat will do it for old inodes.
848 */
851 }
853 /*
854 * The inode format changed when we moved the link count and
855 * made it 32 bits long. If this is an old format inode,
856 * convert it in memory to look like a new one. If it gets
857 * flushed to disk we will convert back before flushing or
858 * logging it. We zero out the new projid field and the old link
859 * count field. We'll handle clearing the pad field (the remains
860 * of the old uuid field) when we actually convert the inode to
861 * the new format. We don't change the version number so that we
862 * can distinguish this from a real new format inode.
863 */
868 }
873 /*
874 * Mark the buffer containing the inode as something to keep
875 * around for a while. This helps to keep recently accessed
876 * meta-data in-core longer.
877 */
880 /*
881 * Use xfs_trans_brelse() to release the buffer containing the
882 * on-disk inode, because it was acquired with xfs_trans_read_buf()
883 * in xfs_itobp() above. If tp is NULL, this is just a normal
884 * brelse(). If we're within a transaction, then xfs_trans_brelse()
885 * will only release the buffer if it is not dirty within the
886 * transaction. It will be OK to release the buffer in this case,
887 * because inodes on disk are never destroyed and we will be
888 * locking the new in-core inode before putting it in the hash
889 * table where other processes can find it. Thus we don't have
890 * to worry about the inode being changed just because we released
891 * the buffer.
892 */
893 out_brelse:
896 }
898 /*
899 * Read in extents from a btree-format inode.
900 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
901 */
902 int
907 {
910 xfs_extnum_t nextents;
916 }
920 /*
921 * We know that the size is valid (it's checked in iformat_btree)
922 */
931 }
934 }
936 /*
937 * Allocate an inode on disk and return a copy of its in-core version.
938 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
939 * appropriately within the inode. The uid and gid for the inode are
940 * set according to the contents of the given cred structure.
941 *
942 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
943 * has a free inode available, call xfs_iget()
944 * to obtain the in-core version of the allocated inode. Finally,
945 * fill in the inode and log its initial contents. In this case,
946 * ialloc_context would be set to NULL and call_again set to false.
947 *
948 * If xfs_dialloc() does not have an available inode,
949 * it will replenish its supply by doing an allocation. Since we can
950 * only do one allocation within a transaction without deadlocks, we
951 * must commit the current transaction before returning the inode itself.
952 * In this case, therefore, we will set call_again to true and return.
953 * The caller should then commit the current transaction, start a new
954 * transaction, and call xfs_ialloc() again to actually get the inode.
955 *
956 * To ensure that some other process does not grab the inode that
957 * was allocated during the first call to xfs_ialloc(), this routine
958 * also returns the [locked] bp pointing to the head of the freelist
959 * as ialloc_context. The caller should hold this buffer across
960 * the commit and pass it back into this routine on the second call.
961 *
962 * If we are allocating quota inodes, we do not have a parent inode
963 * to attach to or associate with (i.e. pip == NULL) because they
964 * are not linked into the directory structure - they are attached
965 * directly to the superblock - and so have no parent.
966 */
967 int
971 mode_t mode,
972 xfs_nlink_t nlink,
973 xfs_dev_t rdev,
974 prid_t prid,
979 {
980 xfs_ino_t ino;
982 uint flags;
984 timespec_t tv;
987 /*
988 * Call the space management code to pick
989 * the on-disk inode to be allocated.
990 */
998 }
1001 /*
1002 * Get the in-core inode with the lock held exclusively.
1003 * This is because we're setting fields here we need
1004 * to prevent others from looking at until we're done.
1005 */
1021 /*
1022 * If the superblock version is up to where we support new format
1023 * inodes and this is currently an old format inode, then change
1024 * the inode version number now. This way we only do the conversion
1025 * here rather than here and in the flush/logging code.
1026 */
1030 /*
1031 * We've already zeroed the old link count, the projid field,
1032 * and the pad field.
1033 */
1034 }
1036 /*
1037 * Project ids won't be stored on disk if we are using a version 1 inode.
1038 */
1046 }
1047 }
1049 /*
1050 * If the group ID of the new file does not match the effective group
1051 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1052 * (and only if the irix_sgid_inherit compatibility variable is set).
1053 */
1058 }
1071 /*
1072 * di_gen will have been taken care of in xfs_iread.
1073 */
1090 /*
1091 * we can't set up filestreams until after the VFS inode
1092 * is set up properly.
1093 */
1096 /* fall through */
1107 }
1114 }
1115 }
1117 xfs_inherit_noatime)
1120 xfs_inherit_nodump)
1123 xfs_inherit_sync)
1126 xfs_inherit_nosymlinks)
1131 xfs_inherit_nodefrag)
1136 }
1137 /* FALLTHROUGH */
1146 }
1147 /*
1148 * Attribute fork settings for new inode.
1149 */
1153 /*
1154 * Log the new values stuffed into the inode.
1155 */
1159 /* now that we have an i_mode we can setup inode ops and unlock */
1162 /* now we have set up the vfs inode we can associate the filestream */
1169 }
1173 }
1175 /*
1176 * Check to make sure that there are no blocks allocated to the
1177 * file beyond the size of the file. We don't check this for
1178 * files with fixed size extents or real time extents, but we
1179 * at least do it for regular files.
1180 */
1181 #ifdef DEBUG
1182 void
1186 xfs_fsize_t isize)
1187 {
1188 xfs_fileoff_t map_first;
1203 /*
1204 * The filesystem could be shutting down, so bmapi may return
1205 * an error.
1206 */
1210 map_first),
1212 NULL))
1216 }
1219 /*
1220 * Free up the underlying blocks past new_size. The new size must be
1221 * smaller than the current size.
1222 *
1223 * The transaction passed to this routine must have made a permanent log
1224 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1225 * given transaction and start new ones, so make sure everything involved in
1226 * the transaction is tidy before calling here. Some transaction will be
1227 * returned to the caller to be committed. The incoming transaction must
1228 * already include the inode, and both inode locks must be held exclusively.
1229 * The inode must also be "held" within the transaction. On return the inode
1230 * will be "held" within the returned transaction. This routine does NOT
1231 * require any disk space to be reserved for it within the transaction.
1232 *
1233 * The fork parameter must be either XFS_ATTR_FORK or XFS_DATA_FORK, and it
1234 * indicates the fork which is to be truncated. For the attribute fork we only
1235 * support truncation to size 0.
1236 *
1237 * We use the sync parameter to indicate whether or not the first transaction
1238 * we perform might have to be synchronous. For the attr fork, it needs to be
1239 * so if the unlink of the inode is not yet known to be permanent in the log.
1240 * This keeps us from freeing and reusing the blocks of the attribute fork
1241 * before the unlink of the inode becomes permanent.
1242 *
1243 * For the data fork, we normally have to run synchronously if we're being
1244 * called out of the inactive path or we're being called out of the create path
1245 * where we're truncating an existing file. Either way, the truncate needs to
1246 * be sync so blocks don't reappear in the file with altered data in case of a
1247 * crash. wsync filesystems can run the first case async because anything that
1248 * shrinks the inode has to run sync so by the time we're called here from
1249 * inactive, the inode size is permanently set to 0.
1250 *
1251 * Calls from the truncate path always need to be sync unless we're in a wsync
1252 * filesystem and the file has already been unlinked.
1253 *
1254 * The caller is responsible for correctly setting the sync parameter. It gets
1255 * too hard for us to guess here which path we're being called out of just
1256 * based on inode state.
1257 *
1258 * If we get an error, we must return with the inode locked and linked into the
1259 * current transaction. This keeps things simple for the higher level code,
1260 * because it always knows that the inode is locked and held in the transaction
1261 * that returns to it whether errors occur or not. We don't mark the inode
1262 * dirty on error so that transactions can be easily aborted if possible.
1263 */
1264 int
1268 xfs_fsize_t new_size,
1271 {
1272 xfs_fsblock_t first_block;
1273 xfs_fileoff_t first_unmap_block;
1274 xfs_fileoff_t last_block;
1280 xfs_bmap_free_t free_list;
1296 /*
1297 * We only support truncating the entire attribute fork.
1298 */
1301 }
1305 /*
1306 * The first thing we do is set the size to new_size permanently
1307 * on disk. This way we don't have to worry about anyone ever
1308 * being able to look at the data being freed even in the face
1309 * of a crash. What we're getting around here is the case where
1310 * we free a block, it is allocated to another file, it is written
1311 * to, and then we crash. If the new data gets written to the
1312 * file but the log buffers containing the free and reallocation
1313 * don't, then we'd end up with garbage in the blocks being freed.
1314 * As long as we make the new_size permanent before actually
1315 * freeing any blocks it doesn't matter if they get written to.
1316 *
1317 * The callers must signal into us whether or not the size
1318 * setting here must be synchronous. There are a few cases
1319 * where it doesn't have to be synchronous. Those cases
1320 * occur if the file is unlinked and we know the unlink is
1321 * permanent or if the blocks being truncated are guaranteed
1322 * to be beyond the inode eof (regardless of the link count)
1323 * and the eof value is permanent. Both of these cases occur
1324 * only on wsync-mounted filesystems. In those cases, we're
1325 * guaranteed that no user will ever see the data in the blocks
1326 * that are being truncated so the truncate can run async.
1327 * In the free beyond eof case, the file may wind up with
1328 * more blocks allocated to it than it needs if we crash
1329 * and that won't get fixed until the next time the file
1330 * is re-opened and closed but that's ok as that shouldn't
1331 * be too many blocks.
1332 *
1333 * However, we can't just make all wsync xactions run async
1334 * because there's one call out of the create path that needs
1335 * to run sync where it's truncating an existing file to size
1336 * 0 whose size is > 0.
1337 *
1338 * It's probably possible to come up with a test in this
1339 * routine that would correctly distinguish all the above
1340 * cases from the values of the function parameters and the
1341 * inode state but for sanity's sake, I've decided to let the
1342 * layers above just tell us. It's simpler to correctly figure
1343 * out in the layer above exactly under what conditions we
1344 * can run async and I think it's easier for others read and
1345 * follow the logic in case something has to be changed.
1346 * cscope is your friend -- rcc.
1347 *
1348 * The attribute fork is much simpler.
1349 *
1350 * For the attribute fork we allow the caller to tell us whether
1351 * the unlink of the inode that led to this call is yet permanent
1352 * in the on disk log. If it is not and we will be freeing extents
1353 * in this inode then we make the first transaction synchronous
1354 * to make sure that the unlink is permanent by the time we free
1355 * the blocks.
1356 */
1359 /*
1360 * If we are not changing the file size then do
1361 * not update the on-disk file size - we may be
1362 * called from xfs_inactive_free_eofblocks(). If we
1363 * update the on-disk file size and then the system
1364 * crashes before the contents of the file are
1365 * flushed to disk then the files may be full of
1366 * holes (ie NULL files bug).
1367 */
1372 }
1373 }
1378 }
1384 /*
1385 * Since it is possible for space to become allocated beyond
1386 * the end of the file (in a crash where the space is allocated
1387 * but the inode size is not yet updated), simply remove any
1388 * blocks which show up between the new EOF and the maximum
1389 * possible file size. If the first block to be removed is
1390 * beyond the maximum file size (ie it is the same as last_block),
1391 * then there is nothing to do.
1392 */
1400 }
1402 /*
1403 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1404 * will tell us whether it freed the entire range or
1405 * not. If this is a synchronous mount (wsync),
1406 * then we can tell bunmapi to keep all the
1407 * transactions asynchronous since the unlink
1408 * transaction that made this inode inactive has
1409 * already hit the disk. There's no danger of
1410 * the freed blocks being reused, there being a
1411 * crash, and the reused blocks suddenly reappearing
1412 * in this file with garbage in them once recovery
1413 * runs.
1414 */
1419 XFS_ITRUNC_MAX_EXTENTS,
1423 /*
1424 * If the bunmapi call encounters an error,
1425 * return to the caller where the transaction
1426 * can be properly aborted. We just need to
1427 * make sure we're not holding any resources
1428 * that we were not when we came in.
1429 */
1432 }
1434 /*
1435 * Duplicate the transaction that has the permanent
1436 * reservation and commit the old transaction.
1437 */
1444 /*
1445 * If the bmap finish call encounters an error, return
1446 * to the caller where the transaction can be properly
1447 * aborted. We just need to make sure we're not
1448 * holding any resources that we were not when we came
1449 * in.
1450 *
1451 * Aborting from this point might lose some blocks in
1452 * the file system, but oh well.
1453 */
1456 }
1459 /*
1460 * Mark the inode dirty so it will be logged and
1461 * moved forward in the log as part of every commit.
1462 */
1464 }
1474 /*
1475 * transaction commit worked ok so we can drop the extra ticket
1476 * reference that we gained in xfs_trans_dup()
1477 */
1481 XFS_TRANS_PERM_LOG_RES,
1482 XFS_ITRUNCATE_LOG_COUNT);
1485 }
1486 /*
1487 * Only update the size in the case of the data fork, but
1488 * always re-log the inode so that our permanent transaction
1489 * can keep on rolling it forward in the log.
1490 */
1493 /*
1494 * If we are not changing the file size then do
1495 * not update the on-disk file size - we may be
1496 * called from xfs_inactive_free_eofblocks(). If we
1497 * update the on-disk file size and then the system
1498 * crashes before the contents of the file are
1499 * flushed to disk then the files may be full of
1500 * holes (ie NULL files bug).
1501 */
1505 }
1506 }
1516 }
1518 /*
1519 * This is called when the inode's link count goes to 0.
1520 * We place the on-disk inode on a list in the AGI. It
1521 * will be pulled from this list when the inode is freed.
1522 */
1523 int
1527 {
1533 xfs_agino_t agino;
1544 /*
1545 * Get the agi buffer first. It ensures lock ordering
1546 * on the list.
1547 */
1553 /*
1554 * Get the index into the agi hash table for the
1555 * list this inode will go on.
1556 */
1564 /*
1565 * There is already another inode in the bucket we need
1566 * to add ourselves to. Add us at the front of the list.
1567 * Here we put the head pointer into our next pointer,
1568 * and then we fall through to point the head at us.
1569 */
1575 /* both on-disk, don't endian flip twice */
1583 }
1585 /*
1586 * Point the bucket head pointer at the inode being inserted.
1587 */
1595 }
1597 /*
1598 * Pull the on-disk inode from the AGI unlinked list.
1599 */
1600 STATIC int
1604 {
1605 xfs_ino_t next_ino;
1611 xfs_agnumber_t agno;
1612 xfs_agino_t agino;
1613 xfs_agino_t next_agino;
1623 /*
1624 * Get the agi buffer first. It ensures lock ordering
1625 * on the list.
1626 */
1633 /*
1634 * Get the index into the agi hash table for the
1635 * list this inode will go on.
1636 */
1644 /*
1645 * We're at the head of the list. Get the inode's
1646 * on-disk buffer to see if there is anyone after us
1647 * on the list. Only modify our next pointer if it
1648 * is not already NULLAGINO. This saves us the overhead
1649 * of dealing with the buffer when there is no need to
1650 * change it.
1651 */
1657 }
1670 }
1671 /*
1672 * Point the bucket head pointer at the next inode.
1673 */
1682 /*
1683 * We need to search the list for the inode being freed.
1684 */
1688 /*
1689 * If the last inode wasn't the one pointing to
1690 * us, then release its buffer since we're not
1691 * going to do anything with it.
1692 */
1695 }
1704 }
1708 }
1709 /*
1710 * Now last_ibp points to the buffer previous to us on
1711 * the unlinked list. Pull us from the list.
1712 */
1718 }
1732 }
1733 /*
1734 * Point the previous inode on the list to the next inode.
1735 */
1743 }
1745 }
1747 /*
1748 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1749 * inodes that are in memory - they all must be marked stale and attached to
1750 * the cluster buffer.
1751 */
1752 STATIC void
1756 xfs_ino_t inum)
1757 {
1763 xfs_daddr_t blkno;
1780 }
1786 /*
1787 * We obtain and lock the backing buffer first in the process
1788 * here, as we have to ensure that any dirty inode that we
1789 * can't get the flush lock on is attached to the buffer.
1790 * If we scan the in-memory inodes first, then buffer IO can
1791 * complete before we get a lock on it, and hence we may fail
1792 * to mark all the active inodes on the buffer stale.
1793 */
1796 XBF_LOCK);
1798 /*
1799 * Walk the inodes already attached to the buffer and mark them
1800 * stale. These will all have the flush locks held, so an
1801 * in-memory inode walk can't lock them. By marking them all
1802 * stale first, we will not attempt to lock them in the loop
1803 * below as the XFS_ISTALE flag will be set.
1804 */
1815 }
1817 }
1820 /*
1821 * For each inode in memory attempt to add it to the inode
1822 * buffer and set it up for being staled on buffer IO
1823 * completion. This is safe as we've locked out tail pushing
1824 * and flushing by locking the buffer.
1825 *
1826 * We have already marked every inode that was part of a
1827 * transaction stale above, which means there is no point in
1828 * even trying to lock them.
1829 */
1831 retry:
1836 /* Inode not in memory, nothing to do */
1840 }
1842 /*
1843 * because this is an RCU protected lookup, we could
1844 * find a recently freed or even reallocated inode
1845 * during the lookup. We need to check under the
1846 * i_flags_lock for a valid inode here. Skip it if it
1847 * is not valid, the wrong inode or stale.
1848 */
1855 }
1858 /*
1859 * Don't try to lock/unlock the current inode, but we
1860 * _cannot_ skip the other inodes that we did not find
1861 * in the list attached to the buffer and are not
1862 * already marked stale. If we can't lock it, back off
1863 * and retry.
1864 */
1870 }
1876 /*
1877 * we don't need to attach clean inodes or those only
1878 * with unlogged changes (which we throw away, anyway).
1879 */
1887 }
1900 }
1904 }
1907 }
1909 /*
1910 * This is called to return an inode to the inode free list.
1911 * The inode should already be truncated to 0 length and have
1912 * no pages associated with it. This routine also assumes that
1913 * the inode is already a part of the transaction.
1914 *
1915 * The on-disk copy of the inode will have been added to the list
1916 * of unlinked inodes in the AGI. We need to remove the inode from
1917 * that list atomically with respect to freeing it here.
1918 */
1919 int
1924 {
1927 xfs_ino_t first_ino;
1940 /*
1941 * Pull the on-disk inode from the AGI unlinked list.
1942 */
1946 }
1951 }
1960 /*
1961 * Bump the generation count so no one will be confused
1962 * by reincarnations of this inode.
1963 */
1972 /*
1973 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1974 * from picking up this inode when it is reclaimed (its incore state
1975 * initialzed but not flushed to disk yet). The in-core di_mode is
1976 * already cleared and a corresponding transaction logged.
1977 * The hack here just synchronizes the in-core to on-disk
1978 * di_mode value in advance before the actual inode sync to disk.
1979 * This is OK because the inode is already unlinked and would never
1980 * change its di_mode again for this inode generation.
1981 * This is a temporary hack that would require a proper fix
1982 * in the future.
1983 */
1988 }
1991 }
1993 /*
1994 * Reallocate the space for if_broot based on the number of records
1995 * being added or deleted as indicated in rec_diff. Move the records
1996 * and pointers in if_broot to fit the new size. When shrinking this
1997 * will eliminate holes between the records and pointers created by
1998 * the caller. When growing this will create holes to be filled in
1999 * by the caller.
2000 *
2001 * The caller must not request to add more records than would fit in
2002 * the on-disk inode root. If the if_broot is currently NULL, then
2003 * if we adding records one will be allocated. The caller must also
2004 * not request that the number of records go below zero, although
2005 * it can go to zero.
2006 *
2007 * ip -- the inode whose if_broot area is changing
2008 * ext_diff -- the change in the number of records, positive or negative,
2009 * requested for the if_broot array.
2010 */
2011 void
2016 {
2026 /*
2027 * Handle the degenerate case quietly.
2028 */
2031 }
2035 /*
2036 * If there wasn't any memory allocated before, just
2037 * allocate it now and get out.
2038 */
2044 }
2046 /*
2047 * If there is already an existing if_broot, then we need
2048 * to realloc() it and shift the pointers to their new
2049 * location. The records don't change location because
2050 * they are kept butted up against the btree block header.
2051 */
2067 }
2069 /*
2070 * rec_diff is less than 0. In this case, we are shrinking the
2071 * if_broot buffer. It must already exist. If we go to zero
2072 * records, just get rid of the root and clear the status bit.
2073 */
2080 else
2084 /*
2085 * First copy over the btree block header.
2086 */
2091 }
2093 /*
2094 * Only copy the records and pointers if there are any.
2095 */
2097 /*
2098 * First copy the records.
2099 */
2104 /*
2105 * Then copy the pointers.
2106 */
2112 }
2119 }
2122 /*
2123 * This is called when the amount of space needed for if_data
2124 * is increased or decreased. The change in size is indicated by
2125 * the number of bytes that need to be added or deleted in the
2126 * byte_diff parameter.
2127 *
2128 * If the amount of space needed has decreased below the size of the
2129 * inline buffer, then switch to using the inline buffer. Otherwise,
2130 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2131 * to what is needed.
2132 *
2133 * ip -- the inode whose if_data area is changing
2134 * byte_diff -- the change in the number of bytes, positive or negative,
2135 * requested for the if_data array.
2136 */
2137 void
2142 {
2149 }
2158 }
2162 /*
2163 * If the valid extents/data can fit in if_inline_ext/data,
2164 * copy them from the malloc'd vector and free it.
2165 */
2171 new_size);
2174 }
2177 /*
2178 * Stuck with malloc/realloc.
2179 * For inline data, the underlying buffer must be
2180 * a multiple of 4 bytes in size so that it can be
2181 * logged and stay on word boundaries. We enforce
2182 * that here.
2183 */
2190 /*
2191 * Only do the realloc if the underlying size
2192 * is really changing.
2193 */
2197 real_size,
2200 }
2207 }
2208 }
2212 }
2214 void
2218 {
2225 }
2227 /*
2228 * If the format is local, then we can't have an extents
2229 * array so just look for an inline data array. If we're
2230 * not local then we may or may not have an extents list,
2231 * so check and free it up if we do.
2232 */
2240 }
2247 }
2254 }
2255 }
2257 /*
2258 * This is called to unpin an inode. The caller must have the inode locked
2259 * in at least shared mode so that the buffer cannot be subsequently pinned
2260 * once someone is waiting for it to be unpinned.
2261 */
2262 static void
2265 {
2270 /* Give the log a push to start the unpinning I/O */
2273 }
2275 void
2278 {
2282 }
2283 }
2285 /*
2286 * xfs_iextents_copy()
2287 *
2288 * This is called to copy the REAL extents (as opposed to the delayed
2289 * allocation extents) from the inode into the given buffer. It
2290 * returns the number of bytes copied into the buffer.
2291 *
2292 * If there are no delayed allocation extents, then we can just
2293 * memcpy() the extents into the buffer. Otherwise, we need to
2294 * examine each extent in turn and skip those which are delayed.
2295 */
2296 int
2301 {
2306 xfs_fsblock_t start_block;
2316 /*
2317 * There are some delayed allocation extents in the
2318 * inode, so copy the extents one at a time and skip
2319 * the delayed ones. There must be at least one
2320 * non-delayed extent.
2321 */
2327 /*
2328 * It's a delayed allocation extent, so skip it.
2329 */
2331 }
2333 /* Translate to on disk format */
2336 dp++;
2337 copied++;
2338 }
2343 }
2345 /*
2346 * Each of the following cases stores data into the same region
2347 * of the on-disk inode, so only one of them can be valid at
2348 * any given time. While it is possible to have conflicting formats
2349 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2350 * in EXTENTS format, this can only happen when the fork has
2351 * changed formats after being modified but before being flushed.
2352 * In these cases, the format always takes precedence, because the
2353 * format indicates the current state of the fork.
2354 */
2355 /*ARGSUSED*/
2356 STATIC void
2363 {
2367 #ifdef XFS_TRANS_DEBUG
2369 #endif
2380 /*
2381 * This can happen if we gave up in iformat in an error path,
2382 * for the attribute fork.
2383 */
2387 }
2397 }
2408 whichfork);
2409 }
2418 XFS_BROOT_SIZE_ADJ));
2422 }
2429 }
2438 }
2444 }
2445 }
2447 STATIC int
2451 {
2475 /* really need a gang lookup range call here */
2486 /*
2487 * because this is an RCU protected lookup, we could find a
2488 * recently freed or even reallocated inode during the lookup.
2489 * We need to check under the i_flags_lock for a valid inode
2490 * here. Skip it if it is not valid or the wrong inode.
2491 */
2497 }
2500 /*
2501 * Do an un-protected check to see if the inode is dirty and
2502 * is a candidate for flushing. These checks will be repeated
2503 * later after the appropriate locks are acquired.
2504 */
2508 /*
2509 * Try to get locks. If any are unavailable or it is pinned,
2510 * then this inode cannot be flushed and is skipped.
2511 */
2518 }
2523 }
2525 /*
2526 * arriving here means that this inode can be flushed. First
2527 * re-check that it's dirty before flushing.
2528 */
2535 }
2536 clcount++;
2539 }
2541 }
2546 }
2548 out_free:
2551 out_put:
2556 cluster_corrupt_out:
2557 /*
2558 * Corruption detected in the clustering loop. Invalidate the
2559 * inode buffer and shut down the filesystem.
2560 */
2562 /*
2563 * Clean up the buffer. If it was B_DELWRI, just release it --
2564 * brelse can handle it with no problems. If not, shut down the
2565 * filesystem before releasing the buffer.
2566 */
2574 /*
2575 * Just like incore_relse: if we have b_iodone functions,
2576 * mark the buffer as an error and call them. Otherwise
2577 * mark it as stale and brelse.
2578 */
2587 }
2588 }
2590 /*
2591 * Unlocks the flush lock
2592 */
2597 }
2599 /*
2600 * xfs_iflush() will write a modified inode's changes out to the
2601 * inode's on disk home. The caller must have the inode lock held
2602 * in at least shared mode and the inode flush completion must be
2603 * active as well. The inode lock will still be held upon return from
2604 * the call and the caller is free to unlock it.
2605 * The inode flush will be completed when the inode reaches the disk.
2606 * The flags indicate how the inode's buffer should be written out.
2607 */
2608 int
2611 uint flags)
2612 {
2629 /*
2630 * We can't flush the inode until it is unpinned, so wait for it if we
2631 * are allowed to block. We know no one new can pin it, because we are
2632 * holding the inode lock shared and you need to hold it exclusively to
2633 * pin the inode.
2634 *
2635 * If we are not allowed to block, force the log out asynchronously so
2636 * that when we come back the inode will be unpinned. If other inodes
2637 * in the same cluster are dirty, they will probably write the inode
2638 * out for us if they occur after the log force completes.
2639 */
2644 }
2647 /*
2648 * For stale inodes we cannot rely on the backing buffer remaining
2649 * stale in cache for the remaining life of the stale inode and so
2650 * xfs_itobp() below may give us a buffer that no longer contains
2651 * inodes below. We have to check this after ensuring the inode is
2652 * unpinned so that it is safe to reclaim the stale inode after the
2653 * flush call.
2654 */
2658 }
2660 /*
2661 * This may have been unpinned because the filesystem is shutting
2662 * down forcibly. If that's the case we must not write this inode
2663 * to disk, because the log record didn't make it to disk!
2664 */
2671 }
2673 /*
2674 * Get the buffer containing the on-disk inode.
2675 */
2681 }
2683 /*
2684 * First flush out the inode that xfs_iflush was called with.
2685 */
2690 /*
2691 * If the buffer is pinned then push on the log now so we won't
2692 * get stuck waiting in the write for too long.
2693 */
2697 /*
2698 * inode clustering:
2699 * see if other inodes can be gathered into this write
2700 */
2707 else
2711 corrupt_out:
2714 cluster_corrupt_out:
2715 /*
2716 * Unlocks the flush lock
2717 */
2720 }
2723 STATIC int
2727 {
2731 #ifdef XFS_TRANS_DEBUG
2733 #endif
2743 /* set *dip = inode's place in the buffer */
2746 /*
2747 * Clear i_update_core before copying out the data.
2748 * This is for coordination with our timestamp updates
2749 * that don't hold the inode lock. They will always
2750 * update the timestamps BEFORE setting i_update_core,
2751 * so if we clear i_update_core after they set it we
2752 * are guaranteed to see their updates to the timestamps.
2753 * I believe that this depends on strongly ordered memory
2754 * semantics, but we have that. We use the SYNCHRONIZE
2755 * macro to make sure that the compiler does not reorder
2756 * the i_update_core access below the data copy below.
2757 */
2761 /*
2762 * Make sure to get the latest timestamps from the Linux inode.
2763 */
2772 }
2779 }
2789 }
2800 }
2801 }
2804 XFS_RANDOM_IFLUSH_5)) {
2806 "%s: detected corrupt incore inode %Lu, "
2812 }
2819 }
2820 /*
2821 * bump the flush iteration count, used to detect flushes which
2822 * postdate a log record during recovery.
2823 */
2827 /*
2828 * Copy the dirty parts of the inode into the on-disk
2829 * inode. We always copy out the core of the inode,
2830 * because if the inode is dirty at all the core must
2831 * be.
2832 */
2835 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2839 /*
2840 * If this is really an old format inode and the superblock version
2841 * has not been updated to support only new format inodes, then
2842 * convert back to the old inode format. If the superblock version
2843 * has been updated, then make the conversion permanent.
2844 */
2848 /*
2849 * Convert it back.
2850 */
2854 /*
2855 * The superblock version has already been bumped,
2856 * so just make the conversion to the new inode
2857 * format permanent.
2858 */
2867 }
2868 }
2875 /*
2876 * We've recorded everything logged in the inode, so we'd
2877 * like to clear the ilf_fields bits so we don't log and
2878 * flush things unnecessarily. However, we can't stop
2879 * logging all this information until the data we've copied
2880 * into the disk buffer is written to disk. If we did we might
2881 * overwrite the copy of the inode in the log with all the
2882 * data after re-logging only part of it, and in the face of
2883 * a crash we wouldn't have all the data we need to recover.
2884 *
2885 * What we do is move the bits to the ili_last_fields field.
2886 * When logging the inode, these bits are moved back to the
2887 * ilf_fields field. In the xfs_iflush_done() routine we
2888 * clear ili_last_fields, since we know that the information
2889 * those bits represent is permanently on disk. As long as
2890 * the flush completes before the inode is logged again, then
2891 * both ilf_fields and ili_last_fields will be cleared.
2892 *
2893 * We can play with the ilf_fields bits here, because the inode
2894 * lock must be held exclusively in order to set bits there
2895 * and the flush lock protects the ili_last_fields bits.
2896 * Set ili_logged so the flush done
2897 * routine can tell whether or not to look in the AIL.
2898 * Also, store the current LSN of the inode so that we can tell
2899 * whether the item has moved in the AIL from xfs_iflush_done().
2900 * In order to read the lsn we need the AIL lock, because
2901 * it is a 64 bit value that cannot be read atomically.
2902 */
2911 /*
2912 * Attach the function xfs_iflush_done to the inode's
2913 * buffer. This will remove the inode from the AIL
2914 * and unlock the inode's flush lock when the inode is
2915 * completely written to disk.
2916 */
2922 /*
2923 * We're flushing an inode which is not in the AIL and has
2924 * not been logged but has i_update_core set. For this
2925 * case we can use a B_DELWRI flush and immediately drop
2926 * the inode flush lock because we can avoid the whole
2927 * AIL state thing. It's OK to drop the flush lock now,
2928 * because we've already locked the buffer and to do anything
2929 * you really need both.
2930 */
2935 }
2937 }
2941 corrupt_out:
2943 }
2945 /*
2946 * Return a pointer to the extent record at file index idx.
2947 */
2948 xfs_bmbt_rec_host_t *
2952 {
2969 }
2970 }
2972 /*
2973 * Insert new item(s) into the extent records for incore inode
2974 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2975 */
2976 void
2983 {
2993 }
2995 /*
2996 * This is called when the amount of space required for incore file
2997 * extents needs to be increased. The ext_diff parameter stores the
2998 * number of new extents being added and the idx parameter contains
2999 * the extent index where the new extents will be added. If the new
3000 * extents are being appended, then we just need to (re)allocate and
3001 * initialize the space. Otherwise, if the new extents are being
3002 * inserted into the middle of the existing entries, a bit more work
3003 * is required to make room for the new extents to be inserted. The
3004 * caller is responsible for filling in the new extent entries upon
3005 * return.
3006 */
3007 void
3012 {
3021 /*
3022 * If the new number of extents (nextents + ext_diff)
3023 * fits inside the inode, then continue to use the inline
3024 * extent buffer.
3025 */
3032 }
3035 }
3036 /*
3037 * Otherwise use a linear (direct) extent list.
3038 * If the extents are currently inside the inode,
3039 * xfs_iext_realloc_direct will switch us from
3040 * inline to direct extent allocation mode.
3041 */
3049 }
3050 }
3051 /* Indirection array */
3064 }
3065 /* Extents fit in target extent page */
3073 }
3076 }
3077 /* Insert a new extent page */
3081 }
3082 /*
3083 * If extent(s) are being appended to the last page in
3084 * the indirection array and the new extent(s) don't fit
3085 * in the page, then erp is NULL and erp_idx is set to
3086 * the next index needed in the indirection array.
3087 */
3096 erp_idx++;
3097 }
3098 }
3099 }
3100 }
3102 }
3104 /*
3105 * This is called when incore extents are being added to the indirection
3106 * array and the new extents do not fit in the target extent list. The
3107 * erp_idx parameter contains the irec index for the target extent list
3108 * in the indirection array, and the idx parameter contains the extent
3109 * index within the list. The number of extents being added is stored
3110 * in the count parameter.
3111 *
3112 * |-------| |-------|
3113 * | | | | idx - number of extents before idx
3114 * | idx | | count |
3115 * | | | | count - number of extents being inserted at idx
3116 * |-------| |-------|
3117 * | count | | nex2 | nex2 - number of extents after idx + count
3118 * |-------| |-------|
3119 */
3120 void
3126 {
3140 /*
3141 * Save second part of target extent list
3142 * (all extents past */
3150 }
3152 /*
3153 * Add the new extents to the end of the target
3154 * list, then allocate new irec record(s) and
3155 * extent buffer(s) as needed to store the rest
3156 * of the new extents.
3157 */
3164 }
3166 erp_idx++;
3172 }
3174 /* Add nex2 extents back to indirection array */
3176 xfs_extnum_t ext_avail;
3182 /*
3183 * If nex2 extents fit in the current page, append
3184 * nex2_ep after the new extents.
3185 */
3188 }
3189 /*
3190 * Otherwise, check if space is available in the
3191 * next page.
3192 */
3196 erp_idx++;
3197 erp++;
3198 /* Create a hole for nex2 extents */
3201 }
3202 /*
3203 * Final choice, create a new extent page for
3204 * nex2 extents.
3205 */
3207 erp_idx++;
3209 }
3214 }
3215 }
3217 /*
3218 * This is called when the amount of space required for incore file
3219 * extents needs to be decreased. The ext_diff parameter stores the
3220 * number of extents to be removed and the idx parameter contains
3221 * the extent index where the extents will be removed from.
3222 *
3223 * If the amount of space needed has decreased below the linear
3224 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3225 * extent array. Otherwise, use kmem_realloc() to adjust the
3226 * size to what is needed.
3227 */
3228 void
3234 {
3253 }
3255 }
3257 /*
3258 * This removes ext_diff extents from the inline buffer, beginning
3259 * at extent index idx.
3260 */
3261 void
3266 {
3285 }
3286 }
3288 /*
3289 * This removes ext_diff extents from a linear (direct) extent list,
3290 * beginning at extent index idx. If the extents are being removed
3291 * from the end of the list (ie. truncate) then we just need to re-
3292 * allocate the list to remove the extra space. Otherwise, if the
3293 * extents are being removed from the middle of the existing extent
3294 * entries, then we first need to move the extent records beginning
3295 * at idx + ext_diff up in the list to overwrite the records being
3296 * removed, then remove the extra space via kmem_realloc.
3297 */
3298 void
3303 {
3315 }
3316 /* Move extents up in the list (if needed) */
3322 }
3325 /*
3326 * Reallocate the direct extent list. If the extents
3327 * will fit inside the inode then xfs_iext_realloc_direct
3328 * will switch from direct to inline extent allocation
3329 * mode for us.
3330 */
3333 }
3335 /*
3336 * This is called when incore extents are being removed from the
3337 * indirection array and the extents being removed span multiple extent
3338 * buffers. The idx parameter contains the file extent index where we
3339 * want to begin removing extents, and the count parameter contains
3340 * how many extents need to be removed.
3341 *
3342 * |-------| |-------|
3343 * | nex1 | | | nex1 - number of extents before idx
3344 * |-------| | count |
3345 * | | | | count - number of extents being removed at idx
3346 * | count | |-------|
3347 * | | | nex2 | nex2 - number of extents after idx + count
3348 * |-------| |-------|
3349 */
3350 void
3355 {
3372 /*
3373 * Check for deletion of entire list;
3374 * xfs_iext_irec_remove() updates extent offsets.
3375 */
3382 XFS_IEXT_BUFSZ);
3388 }
3389 }
3390 /* Move extents up (if needed) */
3395 }
3396 /* Zero out rest of page */
3399 /* Update remaining counters */
3404 erp_idx++;
3405 erp++;
3406 }
3409 }
3411 /*
3412 * Create, destroy, or resize a linear (direct) block of extents.
3413 */
3414 void
3418 {
3427 /* Free extent records */
3430 }
3431 /* Resize direct extent list and zero any new bytes */
3433 /* Check if extents will fit inside the inode */
3439 }
3442 }
3446 rnew_size,
3448 }
3453 }
3454 }
3455 /*
3456 * Switch from the inline extent buffer to a direct
3457 * extent list. Be sure to include the inline extent
3458 * bytes in new_size.
3459 */
3464 }
3466 }
3469 }
3471 /*
3472 * Switch from linear (direct) extent records to inline buffer.
3473 */
3474 void
3478 {
3481 /*
3482 * The inline buffer was zeroed when we switched
3483 * from inline to direct extent allocation mode,
3484 * so we don't need to clear it here.
3485 */
3491 }
3493 /*
3494 * Switch from inline buffer to linear (direct) extent records.
3495 * new_size should already be rounded up to the next power of 2
3496 * by the caller (when appropriate), so use new_size as it is.
3497 * However, since new_size may be rounded up, we can't update
3498 * if_bytes here. It is the caller's responsibility to update
3499 * if_bytes upon return.
3500 */
3501 void
3505 {
3513 }
3515 }
3517 /*
3518 * Resize an extent indirection array to new_size bytes.
3519 */
3520 STATIC void
3524 {
3539 }
3540 }
3542 /*
3543 * Switch from indirection array to linear (direct) extent allocations.
3544 */
3545 STATIC void
3548 {
3568 }
3569 }
3571 /*
3572 * Free incore file extents.
3573 */
3574 void
3577 {
3585 }
3592 }
3596 }
3598 /*
3599 * Return a pointer to the extent record for file system block bno.
3600 */
3606 {
3621 }
3624 /* Find target extent list */
3632 }
3633 /* Binary search extent records */
3644 /* Convert back to file-based extent index */
3647 }
3650 }
3651 }
3652 /* Convert back to file-based extent index */
3655 }
3661 }
3662 }
3665 }
3667 /*
3668 * Return a pointer to the indirection array entry containing the
3669 * extent record for filesystem block bno. Store the index of the
3670 * target irec in *erp_idxp.
3671 */
3677 {
3701 }
3702 }
3705 }
3707 /*
3708 * Return a pointer to the indirection array entry containing the
3709 * extent record at file extent index *idxp. Store the index of the
3710 * target irec in *erp_idxp and store the page index of the target
3711 * extent record in *idxp.
3712 */
3713 xfs_ext_irec_t *
3719 {
3738 /* Binary search extent irec's */
3754 erp_idx++;
3760 }
3761 }
3765 }
3767 /*
3768 * Allocate and initialize an indirection array once the space needed
3769 * for incore extents increases above XFS_IEXT_BUFSZ.
3770 */
3771 void
3774 {
3790 }
3801 }
3803 /*
3804 * Allocate and initialize a new entry in the indirection array.
3805 */
3806 xfs_ext_irec_t *
3810 {
3818 /* Resize indirection array */
3821 /*
3822 * Move records down in the array so the
3823 * new page can use erp_idx.
3824 */
3828 }
3831 /* Initialize new extent record */
3840 }
3842 /*
3843 * Remove a record from the indirection array.
3844 */
3845 void
3849 {
3861 }
3862 /* Compact extent records */
3866 }
3867 /*
3868 * Manually free the last extent record from the indirection
3869 * array. A call to xfs_iext_realloc_indirect() with a size
3870 * of zero would result in a call to xfs_iext_destroy() which
3871 * would in turn call this function again, creating a nasty
3872 * infinite loop.
3873 */
3879 }
3881 }
3883 /*
3884 * This is called to clean up large amounts of unused memory allocated
3885 * by the indirection array. Before compacting anything though, verify
3886 * that the indirection array is still needed and switch back to the
3887 * linear extent list (or even the inline buffer) if possible. The
3888 * compaction policy is as follows:
3889 *
3890 * Full Compaction: Extents fit into a single page (or inline buffer)
3891 * Partial Compaction: Extents occupy less than 50% of allocated space
3892 * No Compaction: Extents occupy at least 50% of allocated space
3893 */
3894 void
3897 {
3914 }
3915 }
3917 /*
3918 * Combine extents from neighboring extent pages.
3919 */
3920 void
3923 {
3939 /*
3940 * Free page before removing extent record
3941 * so er_extoffs don't get modified in
3942 * xfs_iext_irec_remove.
3943 */
3949 erp_idx++;
3950 }
3951 }
3952 }
3954 /*
3955 * This is called to update the er_extoff field in the indirection
3956 * array when extents have been added or removed from one of the
3957 * extent lists. erp_idx contains the irec index to begin updating
3958 * at and ext_diff contains the number of extents that were added
3959 * or removed.
3960 */
3961 void
3966 {
3974 }
3975 }