| blob | a098a20ca63e29bbd021a266e287d87ca796fd8c |
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 * Calculate the last possible buffered byte in a file. This must
1221 * include data that was buffered beyond the EOF by the write code.
1222 * This also needs to deal with overflowing the xfs_fsize_t type
1223 * which can happen for sizes near the limit.
1224 *
1225 * We also need to take into account any blocks beyond the EOF. It
1226 * may be the case that they were buffered by a write which failed.
1227 * In that case the pages will still be in memory, but the inode size
1228 * will never have been updated.
1229 */
1230 STATIC xfs_fsize_t
1233 {
1235 xfs_fsize_t last_byte;
1236 xfs_fileoff_t last_block;
1237 xfs_fileoff_t size_last_block;
1243 /*
1244 * Only check for blocks beyond the EOF if the extents have
1245 * been read in. This eliminates the need for the inode lock,
1246 * and it also saves us from looking when it really isn't
1247 * necessary.
1248 */
1252 XFS_DATA_FORK);
1256 }
1259 }
1266 }
1270 }
1272 }
1274 /*
1275 * Start the truncation of the file to new_size. The new size
1276 * must be smaller than the current size. This routine will
1277 * clear the buffer and page caches of file data in the removed
1278 * range, and xfs_itruncate_finish() will remove the underlying
1279 * disk blocks.
1280 *
1281 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1282 * must NOT have the inode lock held at all. This is because we're
1283 * calling into the buffer/page cache code and we can't hold the
1284 * inode lock when we do so.
1285 *
1286 * We need to wait for any direct I/Os in flight to complete before we
1287 * proceed with the truncate. This is needed to prevent the extents
1288 * being read or written by the direct I/Os from being removed while the
1289 * I/O is in flight as there is no other method of synchronising
1290 * direct I/O with the truncate operation. Also, because we hold
1291 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1292 * started until the truncate completes and drops the lock. Essentially,
1293 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1294 * ordering between direct I/Os and the truncate operation.
1295 *
1296 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1297 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1298 * in the case that the caller is locking things out of order and
1299 * may not be able to call xfs_itruncate_finish() with the inode lock
1300 * held without dropping the I/O lock. If the caller must drop the
1301 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1302 * must be called again with all the same restrictions as the initial
1303 * call.
1304 */
1305 int
1308 uint flags,
1309 xfs_fsize_t new_size)
1310 {
1311 xfs_fsize_t last_byte;
1312 xfs_off_t toss_start;
1323 /* wait for the completion of any pending DIOs */
1327 /*
1328 * Call toss_pages or flushinval_pages to get rid of pages
1329 * overlapping the region being removed. We have to use
1330 * the less efficient flushinval_pages in the case that the
1331 * caller may not be able to finish the truncate without
1332 * dropping the inode's I/O lock. Make sure
1333 * to catch any pages brought in by buffers overlapping
1334 * the EOF by searching out beyond the isize by our
1335 * block size. We round new_size up to a block boundary
1336 * so that we don't toss things on the same block as
1337 * new_size but before it.
1338 *
1339 * Before calling toss_page or flushinval_pages, make sure to
1340 * call remapf() over the same region if the file is mapped.
1341 * This frees up mapped file references to the pages in the
1342 * given range and for the flushinval_pages case it ensures
1343 * that we get the latest mapped changes flushed out.
1344 */
1348 /*
1349 * The place to start tossing is beyond our maximum
1350 * file size, so there is no way that the data extended
1351 * out there.
1352 */
1354 }
1364 }
1365 }
1367 #ifdef DEBUG
1370 }
1371 #endif
1373 }
1375 /*
1376 * Shrink the file to the given new_size. The new size must be smaller than
1377 * the current size. This will free up the underlying blocks in the removed
1378 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1379 *
1380 * The transaction passed to this routine must have made a permanent log
1381 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1382 * given transaction and start new ones, so make sure everything involved in
1383 * the transaction is tidy before calling here. Some transaction will be
1384 * returned to the caller to be committed. The incoming transaction must
1385 * already include the inode, and both inode locks must be held exclusively.
1386 * The inode must also be "held" within the transaction. On return the inode
1387 * will be "held" within the returned transaction. This routine does NOT
1388 * require any disk space to be reserved for it within the transaction.
1389 *
1390 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1391 * indicates the fork which is to be truncated. For the attribute fork we only
1392 * support truncation to size 0.
1393 *
1394 * We use the sync parameter to indicate whether or not the first transaction
1395 * we perform might have to be synchronous. For the attr fork, it needs to be
1396 * so if the unlink of the inode is not yet known to be permanent in the log.
1397 * This keeps us from freeing and reusing the blocks of the attribute fork
1398 * before the unlink of the inode becomes permanent.
1399 *
1400 * For the data fork, we normally have to run synchronously if we're being
1401 * called out of the inactive path or we're being called out of the create path
1402 * where we're truncating an existing file. Either way, the truncate needs to
1403 * be sync so blocks don't reappear in the file with altered data in case of a
1404 * crash. wsync filesystems can run the first case async because anything that
1405 * shrinks the inode has to run sync so by the time we're called here from
1406 * inactive, the inode size is permanently set to 0.
1407 *
1408 * Calls from the truncate path always need to be sync unless we're in a wsync
1409 * filesystem and the file has already been unlinked.
1410 *
1411 * The caller is responsible for correctly setting the sync parameter. It gets
1412 * too hard for us to guess here which path we're being called out of just
1413 * based on inode state.
1414 *
1415 * If we get an error, we must return with the inode locked and linked into the
1416 * current transaction. This keeps things simple for the higher level code,
1417 * because it always knows that the inode is locked and held in the transaction
1418 * that returns to it whether errors occur or not. We don't mark the inode
1419 * dirty on error so that transactions can be easily aborted if possible.
1420 */
1421 int
1425 xfs_fsize_t new_size,
1428 {
1429 xfs_fsblock_t first_block;
1430 xfs_fileoff_t first_unmap_block;
1431 xfs_fileoff_t last_block;
1437 xfs_bmap_free_t free_list;
1453 /*
1454 * We only support truncating the entire attribute fork.
1455 */
1458 }
1462 /*
1463 * The first thing we do is set the size to new_size permanently
1464 * on disk. This way we don't have to worry about anyone ever
1465 * being able to look at the data being freed even in the face
1466 * of a crash. What we're getting around here is the case where
1467 * we free a block, it is allocated to another file, it is written
1468 * to, and then we crash. If the new data gets written to the
1469 * file but the log buffers containing the free and reallocation
1470 * don't, then we'd end up with garbage in the blocks being freed.
1471 * As long as we make the new_size permanent before actually
1472 * freeing any blocks it doesn't matter if they get written to.
1473 *
1474 * The callers must signal into us whether or not the size
1475 * setting here must be synchronous. There are a few cases
1476 * where it doesn't have to be synchronous. Those cases
1477 * occur if the file is unlinked and we know the unlink is
1478 * permanent or if the blocks being truncated are guaranteed
1479 * to be beyond the inode eof (regardless of the link count)
1480 * and the eof value is permanent. Both of these cases occur
1481 * only on wsync-mounted filesystems. In those cases, we're
1482 * guaranteed that no user will ever see the data in the blocks
1483 * that are being truncated so the truncate can run async.
1484 * In the free beyond eof case, the file may wind up with
1485 * more blocks allocated to it than it needs if we crash
1486 * and that won't get fixed until the next time the file
1487 * is re-opened and closed but that's ok as that shouldn't
1488 * be too many blocks.
1489 *
1490 * However, we can't just make all wsync xactions run async
1491 * because there's one call out of the create path that needs
1492 * to run sync where it's truncating an existing file to size
1493 * 0 whose size is > 0.
1494 *
1495 * It's probably possible to come up with a test in this
1496 * routine that would correctly distinguish all the above
1497 * cases from the values of the function parameters and the
1498 * inode state but for sanity's sake, I've decided to let the
1499 * layers above just tell us. It's simpler to correctly figure
1500 * out in the layer above exactly under what conditions we
1501 * can run async and I think it's easier for others read and
1502 * follow the logic in case something has to be changed.
1503 * cscope is your friend -- rcc.
1504 *
1505 * The attribute fork is much simpler.
1506 *
1507 * For the attribute fork we allow the caller to tell us whether
1508 * the unlink of the inode that led to this call is yet permanent
1509 * in the on disk log. If it is not and we will be freeing extents
1510 * in this inode then we make the first transaction synchronous
1511 * to make sure that the unlink is permanent by the time we free
1512 * the blocks.
1513 */
1516 /*
1517 * If we are not changing the file size then do
1518 * not update the on-disk file size - we may be
1519 * called from xfs_inactive_free_eofblocks(). If we
1520 * update the on-disk file size and then the system
1521 * crashes before the contents of the file are
1522 * flushed to disk then the files may be full of
1523 * holes (ie NULL files bug).
1524 */
1529 }
1530 }
1535 }
1541 /*
1542 * Since it is possible for space to become allocated beyond
1543 * the end of the file (in a crash where the space is allocated
1544 * but the inode size is not yet updated), simply remove any
1545 * blocks which show up between the new EOF and the maximum
1546 * possible file size. If the first block to be removed is
1547 * beyond the maximum file size (ie it is the same as last_block),
1548 * then there is nothing to do.
1549 */
1557 }
1559 /*
1560 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1561 * will tell us whether it freed the entire range or
1562 * not. If this is a synchronous mount (wsync),
1563 * then we can tell bunmapi to keep all the
1564 * transactions asynchronous since the unlink
1565 * transaction that made this inode inactive has
1566 * already hit the disk. There's no danger of
1567 * the freed blocks being reused, there being a
1568 * crash, and the reused blocks suddenly reappearing
1569 * in this file with garbage in them once recovery
1570 * runs.
1571 */
1576 XFS_ITRUNC_MAX_EXTENTS,
1580 /*
1581 * If the bunmapi call encounters an error,
1582 * return to the caller where the transaction
1583 * can be properly aborted. We just need to
1584 * make sure we're not holding any resources
1585 * that we were not when we came in.
1586 */
1589 }
1591 /*
1592 * Duplicate the transaction that has the permanent
1593 * reservation and commit the old transaction.
1594 */
1601 /*
1602 * If the bmap finish call encounters an error, return
1603 * to the caller where the transaction can be properly
1604 * aborted. We just need to make sure we're not
1605 * holding any resources that we were not when we came
1606 * in.
1607 *
1608 * Aborting from this point might lose some blocks in
1609 * the file system, but oh well.
1610 */
1613 }
1616 /*
1617 * Mark the inode dirty so it will be logged and
1618 * moved forward in the log as part of every commit.
1619 */
1621 }
1631 /*
1632 * transaction commit worked ok so we can drop the extra ticket
1633 * reference that we gained in xfs_trans_dup()
1634 */
1638 XFS_TRANS_PERM_LOG_RES,
1639 XFS_ITRUNCATE_LOG_COUNT);
1642 }
1643 /*
1644 * Only update the size in the case of the data fork, but
1645 * always re-log the inode so that our permanent transaction
1646 * can keep on rolling it forward in the log.
1647 */
1650 /*
1651 * If we are not changing the file size then do
1652 * not update the on-disk file size - we may be
1653 * called from xfs_inactive_free_eofblocks(). If we
1654 * update the on-disk file size and then the system
1655 * crashes before the contents of the file are
1656 * flushed to disk then the files may be full of
1657 * holes (ie NULL files bug).
1658 */
1662 }
1663 }
1673 }
1675 /*
1676 * This is called when the inode's link count goes to 0.
1677 * We place the on-disk inode on a list in the AGI. It
1678 * will be pulled from this list when the inode is freed.
1679 */
1680 int
1684 {
1690 xfs_agino_t agino;
1701 /*
1702 * Get the agi buffer first. It ensures lock ordering
1703 * on the list.
1704 */
1710 /*
1711 * Get the index into the agi hash table for the
1712 * list this inode will go on.
1713 */
1721 /*
1722 * There is already another inode in the bucket we need
1723 * to add ourselves to. Add us at the front of the list.
1724 * Here we put the head pointer into our next pointer,
1725 * and then we fall through to point the head at us.
1726 */
1732 /* both on-disk, don't endian flip twice */
1740 }
1742 /*
1743 * Point the bucket head pointer at the inode being inserted.
1744 */
1752 }
1754 /*
1755 * Pull the on-disk inode from the AGI unlinked list.
1756 */
1757 STATIC int
1761 {
1762 xfs_ino_t next_ino;
1768 xfs_agnumber_t agno;
1769 xfs_agino_t agino;
1770 xfs_agino_t next_agino;
1780 /*
1781 * Get the agi buffer first. It ensures lock ordering
1782 * on the list.
1783 */
1790 /*
1791 * Get the index into the agi hash table for the
1792 * list this inode will go on.
1793 */
1801 /*
1802 * We're at the head of the list. Get the inode's
1803 * on-disk buffer to see if there is anyone after us
1804 * on the list. Only modify our next pointer if it
1805 * is not already NULLAGINO. This saves us the overhead
1806 * of dealing with the buffer when there is no need to
1807 * change it.
1808 */
1814 }
1827 }
1828 /*
1829 * Point the bucket head pointer at the next inode.
1830 */
1839 /*
1840 * We need to search the list for the inode being freed.
1841 */
1845 /*
1846 * If the last inode wasn't the one pointing to
1847 * us, then release its buffer since we're not
1848 * going to do anything with it.
1849 */
1852 }
1861 }
1865 }
1866 /*
1867 * Now last_ibp points to the buffer previous to us on
1868 * the unlinked list. Pull us from the list.
1869 */
1875 }
1889 }
1890 /*
1891 * Point the previous inode on the list to the next inode.
1892 */
1900 }
1902 }
1904 /*
1905 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1906 * inodes that are in memory - they all must be marked stale and attached to
1907 * the cluster buffer.
1908 */
1909 STATIC void
1913 xfs_ino_t inum)
1914 {
1920 xfs_daddr_t blkno;
1937 }
1943 /*
1944 * We obtain and lock the backing buffer first in the process
1945 * here, as we have to ensure that any dirty inode that we
1946 * can't get the flush lock on is attached to the buffer.
1947 * If we scan the in-memory inodes first, then buffer IO can
1948 * complete before we get a lock on it, and hence we may fail
1949 * to mark all the active inodes on the buffer stale.
1950 */
1953 XBF_LOCK);
1955 /*
1956 * Walk the inodes already attached to the buffer and mark them
1957 * stale. These will all have the flush locks held, so an
1958 * in-memory inode walk can't lock them. By marking them all
1959 * stale first, we will not attempt to lock them in the loop
1960 * below as the XFS_ISTALE flag will be set.
1961 */
1972 }
1974 }
1977 /*
1978 * For each inode in memory attempt to add it to the inode
1979 * buffer and set it up for being staled on buffer IO
1980 * completion. This is safe as we've locked out tail pushing
1981 * and flushing by locking the buffer.
1982 *
1983 * We have already marked every inode that was part of a
1984 * transaction stale above, which means there is no point in
1985 * even trying to lock them.
1986 */
1988 retry:
1993 /* Inode not in memory, nothing to do */
1997 }
1999 /*
2000 * because this is an RCU protected lookup, we could
2001 * find a recently freed or even reallocated inode
2002 * during the lookup. We need to check under the
2003 * i_flags_lock for a valid inode here. Skip it if it
2004 * is not valid, the wrong inode or stale.
2005 */
2012 }
2015 /*
2016 * Don't try to lock/unlock the current inode, but we
2017 * _cannot_ skip the other inodes that we did not find
2018 * in the list attached to the buffer and are not
2019 * already marked stale. If we can't lock it, back off
2020 * and retry.
2021 */
2027 }
2033 /*
2034 * we don't need to attach clean inodes or those only
2035 * with unlogged changes (which we throw away, anyway).
2036 */
2044 }
2057 }
2061 }
2064 }
2066 /*
2067 * This is called to return an inode to the inode free list.
2068 * The inode should already be truncated to 0 length and have
2069 * no pages associated with it. This routine also assumes that
2070 * the inode is already a part of the transaction.
2071 *
2072 * The on-disk copy of the inode will have been added to the list
2073 * of unlinked inodes in the AGI. We need to remove the inode from
2074 * that list atomically with respect to freeing it here.
2075 */
2076 int
2081 {
2084 xfs_ino_t first_ino;
2097 /*
2098 * Pull the on-disk inode from the AGI unlinked list.
2099 */
2103 }
2108 }
2117 /*
2118 * Bump the generation count so no one will be confused
2119 * by reincarnations of this inode.
2120 */
2129 /*
2130 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2131 * from picking up this inode when it is reclaimed (its incore state
2132 * initialzed but not flushed to disk yet). The in-core di_mode is
2133 * already cleared and a corresponding transaction logged.
2134 * The hack here just synchronizes the in-core to on-disk
2135 * di_mode value in advance before the actual inode sync to disk.
2136 * This is OK because the inode is already unlinked and would never
2137 * change its di_mode again for this inode generation.
2138 * This is a temporary hack that would require a proper fix
2139 * in the future.
2140 */
2145 }
2148 }
2150 /*
2151 * Reallocate the space for if_broot based on the number of records
2152 * being added or deleted as indicated in rec_diff. Move the records
2153 * and pointers in if_broot to fit the new size. When shrinking this
2154 * will eliminate holes between the records and pointers created by
2155 * the caller. When growing this will create holes to be filled in
2156 * by the caller.
2157 *
2158 * The caller must not request to add more records than would fit in
2159 * the on-disk inode root. If the if_broot is currently NULL, then
2160 * if we adding records one will be allocated. The caller must also
2161 * not request that the number of records go below zero, although
2162 * it can go to zero.
2163 *
2164 * ip -- the inode whose if_broot area is changing
2165 * ext_diff -- the change in the number of records, positive or negative,
2166 * requested for the if_broot array.
2167 */
2168 void
2173 {
2183 /*
2184 * Handle the degenerate case quietly.
2185 */
2188 }
2192 /*
2193 * If there wasn't any memory allocated before, just
2194 * allocate it now and get out.
2195 */
2201 }
2203 /*
2204 * If there is already an existing if_broot, then we need
2205 * to realloc() it and shift the pointers to their new
2206 * location. The records don't change location because
2207 * they are kept butted up against the btree block header.
2208 */
2224 }
2226 /*
2227 * rec_diff is less than 0. In this case, we are shrinking the
2228 * if_broot buffer. It must already exist. If we go to zero
2229 * records, just get rid of the root and clear the status bit.
2230 */
2237 else
2241 /*
2242 * First copy over the btree block header.
2243 */
2248 }
2250 /*
2251 * Only copy the records and pointers if there are any.
2252 */
2254 /*
2255 * First copy the records.
2256 */
2261 /*
2262 * Then copy the pointers.
2263 */
2269 }
2276 }
2279 /*
2280 * This is called when the amount of space needed for if_data
2281 * is increased or decreased. The change in size is indicated by
2282 * the number of bytes that need to be added or deleted in the
2283 * byte_diff parameter.
2284 *
2285 * If the amount of space needed has decreased below the size of the
2286 * inline buffer, then switch to using the inline buffer. Otherwise,
2287 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2288 * to what is needed.
2289 *
2290 * ip -- the inode whose if_data area is changing
2291 * byte_diff -- the change in the number of bytes, positive or negative,
2292 * requested for the if_data array.
2293 */
2294 void
2299 {
2306 }
2315 }
2319 /*
2320 * If the valid extents/data can fit in if_inline_ext/data,
2321 * copy them from the malloc'd vector and free it.
2322 */
2328 new_size);
2331 }
2334 /*
2335 * Stuck with malloc/realloc.
2336 * For inline data, the underlying buffer must be
2337 * a multiple of 4 bytes in size so that it can be
2338 * logged and stay on word boundaries. We enforce
2339 * that here.
2340 */
2347 /*
2348 * Only do the realloc if the underlying size
2349 * is really changing.
2350 */
2354 real_size,
2357 }
2364 }
2365 }
2369 }
2371 void
2375 {
2382 }
2384 /*
2385 * If the format is local, then we can't have an extents
2386 * array so just look for an inline data array. If we're
2387 * not local then we may or may not have an extents list,
2388 * so check and free it up if we do.
2389 */
2397 }
2404 }
2411 }
2412 }
2414 /*
2415 * This is called to unpin an inode. The caller must have the inode locked
2416 * in at least shared mode so that the buffer cannot be subsequently pinned
2417 * once someone is waiting for it to be unpinned.
2418 */
2419 static void
2422 {
2427 /* Give the log a push to start the unpinning I/O */
2430 }
2432 void
2435 {
2439 }
2440 }
2442 /*
2443 * xfs_iextents_copy()
2444 *
2445 * This is called to copy the REAL extents (as opposed to the delayed
2446 * allocation extents) from the inode into the given buffer. It
2447 * returns the number of bytes copied into the buffer.
2448 *
2449 * If there are no delayed allocation extents, then we can just
2450 * memcpy() the extents into the buffer. Otherwise, we need to
2451 * examine each extent in turn and skip those which are delayed.
2452 */
2453 int
2458 {
2463 xfs_fsblock_t start_block;
2473 /*
2474 * There are some delayed allocation extents in the
2475 * inode, so copy the extents one at a time and skip
2476 * the delayed ones. There must be at least one
2477 * non-delayed extent.
2478 */
2484 /*
2485 * It's a delayed allocation extent, so skip it.
2486 */
2488 }
2490 /* Translate to on disk format */
2493 dp++;
2494 copied++;
2495 }
2500 }
2502 /*
2503 * Each of the following cases stores data into the same region
2504 * of the on-disk inode, so only one of them can be valid at
2505 * any given time. While it is possible to have conflicting formats
2506 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2507 * in EXTENTS format, this can only happen when the fork has
2508 * changed formats after being modified but before being flushed.
2509 * In these cases, the format always takes precedence, because the
2510 * format indicates the current state of the fork.
2511 */
2512 /*ARGSUSED*/
2513 STATIC void
2520 {
2524 #ifdef XFS_TRANS_DEBUG
2526 #endif
2537 /*
2538 * This can happen if we gave up in iformat in an error path,
2539 * for the attribute fork.
2540 */
2544 }
2554 }
2565 whichfork);
2566 }
2575 XFS_BROOT_SIZE_ADJ));
2579 }
2586 }
2595 }
2601 }
2602 }
2604 STATIC int
2608 {
2632 /* really need a gang lookup range call here */
2643 /*
2644 * because this is an RCU protected lookup, we could find a
2645 * recently freed or even reallocated inode during the lookup.
2646 * We need to check under the i_flags_lock for a valid inode
2647 * here. Skip it if it is not valid or the wrong inode.
2648 */
2654 }
2657 /*
2658 * Do an un-protected check to see if the inode is dirty and
2659 * is a candidate for flushing. These checks will be repeated
2660 * later after the appropriate locks are acquired.
2661 */
2665 /*
2666 * Try to get locks. If any are unavailable or it is pinned,
2667 * then this inode cannot be flushed and is skipped.
2668 */
2675 }
2680 }
2682 /*
2683 * arriving here means that this inode can be flushed. First
2684 * re-check that it's dirty before flushing.
2685 */
2692 }
2693 clcount++;
2696 }
2698 }
2703 }
2705 out_free:
2708 out_put:
2713 cluster_corrupt_out:
2714 /*
2715 * Corruption detected in the clustering loop. Invalidate the
2716 * inode buffer and shut down the filesystem.
2717 */
2719 /*
2720 * Clean up the buffer. If it was B_DELWRI, just release it --
2721 * brelse can handle it with no problems. If not, shut down the
2722 * filesystem before releasing the buffer.
2723 */
2731 /*
2732 * Just like incore_relse: if we have b_iodone functions,
2733 * mark the buffer as an error and call them. Otherwise
2734 * mark it as stale and brelse.
2735 */
2744 }
2745 }
2747 /*
2748 * Unlocks the flush lock
2749 */
2754 }
2756 /*
2757 * xfs_iflush() will write a modified inode's changes out to the
2758 * inode's on disk home. The caller must have the inode lock held
2759 * in at least shared mode and the inode flush completion must be
2760 * active as well. The inode lock will still be held upon return from
2761 * the call and the caller is free to unlock it.
2762 * The inode flush will be completed when the inode reaches the disk.
2763 * The flags indicate how the inode's buffer should be written out.
2764 */
2765 int
2768 uint flags)
2769 {
2786 /*
2787 * We can't flush the inode until it is unpinned, so wait for it if we
2788 * are allowed to block. We know no one new can pin it, because we are
2789 * holding the inode lock shared and you need to hold it exclusively to
2790 * pin the inode.
2791 *
2792 * If we are not allowed to block, force the log out asynchronously so
2793 * that when we come back the inode will be unpinned. If other inodes
2794 * in the same cluster are dirty, they will probably write the inode
2795 * out for us if they occur after the log force completes.
2796 */
2801 }
2804 /*
2805 * For stale inodes we cannot rely on the backing buffer remaining
2806 * stale in cache for the remaining life of the stale inode and so
2807 * xfs_itobp() below may give us a buffer that no longer contains
2808 * inodes below. We have to check this after ensuring the inode is
2809 * unpinned so that it is safe to reclaim the stale inode after the
2810 * flush call.
2811 */
2815 }
2817 /*
2818 * This may have been unpinned because the filesystem is shutting
2819 * down forcibly. If that's the case we must not write this inode
2820 * to disk, because the log record didn't make it to disk!
2821 */
2828 }
2830 /*
2831 * Get the buffer containing the on-disk inode.
2832 */
2838 }
2840 /*
2841 * First flush out the inode that xfs_iflush was called with.
2842 */
2847 /*
2848 * If the buffer is pinned then push on the log now so we won't
2849 * get stuck waiting in the write for too long.
2850 */
2854 /*
2855 * inode clustering:
2856 * see if other inodes can be gathered into this write
2857 */
2864 else
2868 corrupt_out:
2871 cluster_corrupt_out:
2872 /*
2873 * Unlocks the flush lock
2874 */
2877 }
2880 STATIC int
2884 {
2888 #ifdef XFS_TRANS_DEBUG
2890 #endif
2900 /* set *dip = inode's place in the buffer */
2903 /*
2904 * Clear i_update_core before copying out the data.
2905 * This is for coordination with our timestamp updates
2906 * that don't hold the inode lock. They will always
2907 * update the timestamps BEFORE setting i_update_core,
2908 * so if we clear i_update_core after they set it we
2909 * are guaranteed to see their updates to the timestamps.
2910 * I believe that this depends on strongly ordered memory
2911 * semantics, but we have that. We use the SYNCHRONIZE
2912 * macro to make sure that the compiler does not reorder
2913 * the i_update_core access below the data copy below.
2914 */
2918 /*
2919 * Make sure to get the latest timestamps from the Linux inode.
2920 */
2929 }
2936 }
2946 }
2957 }
2958 }
2961 XFS_RANDOM_IFLUSH_5)) {
2963 "%s: detected corrupt incore inode %Lu, "
2969 }
2976 }
2977 /*
2978 * bump the flush iteration count, used to detect flushes which
2979 * postdate a log record during recovery.
2980 */
2984 /*
2985 * Copy the dirty parts of the inode into the on-disk
2986 * inode. We always copy out the core of the inode,
2987 * because if the inode is dirty at all the core must
2988 * be.
2989 */
2992 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2996 /*
2997 * If this is really an old format inode and the superblock version
2998 * has not been updated to support only new format inodes, then
2999 * convert back to the old inode format. If the superblock version
3000 * has been updated, then make the conversion permanent.
3001 */
3005 /*
3006 * Convert it back.
3007 */
3011 /*
3012 * The superblock version has already been bumped,
3013 * so just make the conversion to the new inode
3014 * format permanent.
3015 */
3024 }
3025 }
3032 /*
3033 * We've recorded everything logged in the inode, so we'd
3034 * like to clear the ilf_fields bits so we don't log and
3035 * flush things unnecessarily. However, we can't stop
3036 * logging all this information until the data we've copied
3037 * into the disk buffer is written to disk. If we did we might
3038 * overwrite the copy of the inode in the log with all the
3039 * data after re-logging only part of it, and in the face of
3040 * a crash we wouldn't have all the data we need to recover.
3041 *
3042 * What we do is move the bits to the ili_last_fields field.
3043 * When logging the inode, these bits are moved back to the
3044 * ilf_fields field. In the xfs_iflush_done() routine we
3045 * clear ili_last_fields, since we know that the information
3046 * those bits represent is permanently on disk. As long as
3047 * the flush completes before the inode is logged again, then
3048 * both ilf_fields and ili_last_fields will be cleared.
3049 *
3050 * We can play with the ilf_fields bits here, because the inode
3051 * lock must be held exclusively in order to set bits there
3052 * and the flush lock protects the ili_last_fields bits.
3053 * Set ili_logged so the flush done
3054 * routine can tell whether or not to look in the AIL.
3055 * Also, store the current LSN of the inode so that we can tell
3056 * whether the item has moved in the AIL from xfs_iflush_done().
3057 * In order to read the lsn we need the AIL lock, because
3058 * it is a 64 bit value that cannot be read atomically.
3059 */
3068 /*
3069 * Attach the function xfs_iflush_done to the inode's
3070 * buffer. This will remove the inode from the AIL
3071 * and unlock the inode's flush lock when the inode is
3072 * completely written to disk.
3073 */
3079 /*
3080 * We're flushing an inode which is not in the AIL and has
3081 * not been logged but has i_update_core set. For this
3082 * case we can use a B_DELWRI flush and immediately drop
3083 * the inode flush lock because we can avoid the whole
3084 * AIL state thing. It's OK to drop the flush lock now,
3085 * because we've already locked the buffer and to do anything
3086 * you really need both.
3087 */
3092 }
3094 }
3098 corrupt_out:
3100 }
3102 /*
3103 * Return a pointer to the extent record at file index idx.
3104 */
3105 xfs_bmbt_rec_host_t *
3109 {
3126 }
3127 }
3129 /*
3130 * Insert new item(s) into the extent records for incore inode
3131 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3132 */
3133 void
3140 {
3150 }
3152 /*
3153 * This is called when the amount of space required for incore file
3154 * extents needs to be increased. The ext_diff parameter stores the
3155 * number of new extents being added and the idx parameter contains
3156 * the extent index where the new extents will be added. If the new
3157 * extents are being appended, then we just need to (re)allocate and
3158 * initialize the space. Otherwise, if the new extents are being
3159 * inserted into the middle of the existing entries, a bit more work
3160 * is required to make room for the new extents to be inserted. The
3161 * caller is responsible for filling in the new extent entries upon
3162 * return.
3163 */
3164 void
3169 {
3178 /*
3179 * If the new number of extents (nextents + ext_diff)
3180 * fits inside the inode, then continue to use the inline
3181 * extent buffer.
3182 */
3189 }
3192 }
3193 /*
3194 * Otherwise use a linear (direct) extent list.
3195 * If the extents are currently inside the inode,
3196 * xfs_iext_realloc_direct will switch us from
3197 * inline to direct extent allocation mode.
3198 */
3206 }
3207 }
3208 /* Indirection array */
3221 }
3222 /* Extents fit in target extent page */
3230 }
3233 }
3234 /* Insert a new extent page */
3238 }
3239 /*
3240 * If extent(s) are being appended to the last page in
3241 * the indirection array and the new extent(s) don't fit
3242 * in the page, then erp is NULL and erp_idx is set to
3243 * the next index needed in the indirection array.
3244 */
3253 erp_idx++;
3254 }
3255 }
3256 }
3257 }
3259 }
3261 /*
3262 * This is called when incore extents are being added to the indirection
3263 * array and the new extents do not fit in the target extent list. The
3264 * erp_idx parameter contains the irec index for the target extent list
3265 * in the indirection array, and the idx parameter contains the extent
3266 * index within the list. The number of extents being added is stored
3267 * in the count parameter.
3268 *
3269 * |-------| |-------|
3270 * | | | | idx - number of extents before idx
3271 * | idx | | count |
3272 * | | | | count - number of extents being inserted at idx
3273 * |-------| |-------|
3274 * | count | | nex2 | nex2 - number of extents after idx + count
3275 * |-------| |-------|
3276 */
3277 void
3283 {
3297 /*
3298 * Save second part of target extent list
3299 * (all extents past */
3307 }
3309 /*
3310 * Add the new extents to the end of the target
3311 * list, then allocate new irec record(s) and
3312 * extent buffer(s) as needed to store the rest
3313 * of the new extents.
3314 */
3321 }
3323 erp_idx++;
3329 }
3331 /* Add nex2 extents back to indirection array */
3333 xfs_extnum_t ext_avail;
3339 /*
3340 * If nex2 extents fit in the current page, append
3341 * nex2_ep after the new extents.
3342 */
3345 }
3346 /*
3347 * Otherwise, check if space is available in the
3348 * next page.
3349 */
3353 erp_idx++;
3354 erp++;
3355 /* Create a hole for nex2 extents */
3358 }
3359 /*
3360 * Final choice, create a new extent page for
3361 * nex2 extents.
3362 */
3364 erp_idx++;
3366 }
3371 }
3372 }
3374 /*
3375 * This is called when the amount of space required for incore file
3376 * extents needs to be decreased. The ext_diff parameter stores the
3377 * number of extents to be removed and the idx parameter contains
3378 * the extent index where the extents will be removed from.
3379 *
3380 * If the amount of space needed has decreased below the linear
3381 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3382 * extent array. Otherwise, use kmem_realloc() to adjust the
3383 * size to what is needed.
3384 */
3385 void
3391 {
3410 }
3412 }
3414 /*
3415 * This removes ext_diff extents from the inline buffer, beginning
3416 * at extent index idx.
3417 */
3418 void
3423 {
3442 }
3443 }
3445 /*
3446 * This removes ext_diff extents from a linear (direct) extent list,
3447 * beginning at extent index idx. If the extents are being removed
3448 * from the end of the list (ie. truncate) then we just need to re-
3449 * allocate the list to remove the extra space. Otherwise, if the
3450 * extents are being removed from the middle of the existing extent
3451 * entries, then we first need to move the extent records beginning
3452 * at idx + ext_diff up in the list to overwrite the records being
3453 * removed, then remove the extra space via kmem_realloc.
3454 */
3455 void
3460 {
3472 }
3473 /* Move extents up in the list (if needed) */
3479 }
3482 /*
3483 * Reallocate the direct extent list. If the extents
3484 * will fit inside the inode then xfs_iext_realloc_direct
3485 * will switch from direct to inline extent allocation
3486 * mode for us.
3487 */
3490 }
3492 /*
3493 * This is called when incore extents are being removed from the
3494 * indirection array and the extents being removed span multiple extent
3495 * buffers. The idx parameter contains the file extent index where we
3496 * want to begin removing extents, and the count parameter contains
3497 * how many extents need to be removed.
3498 *
3499 * |-------| |-------|
3500 * | nex1 | | | nex1 - number of extents before idx
3501 * |-------| | count |
3502 * | | | | count - number of extents being removed at idx
3503 * | count | |-------|
3504 * | | | nex2 | nex2 - number of extents after idx + count
3505 * |-------| |-------|
3506 */
3507 void
3512 {
3529 /*
3530 * Check for deletion of entire list;
3531 * xfs_iext_irec_remove() updates extent offsets.
3532 */
3539 XFS_IEXT_BUFSZ);
3545 }
3546 }
3547 /* Move extents up (if needed) */
3552 }
3553 /* Zero out rest of page */
3556 /* Update remaining counters */
3561 erp_idx++;
3562 erp++;
3563 }
3566 }
3568 /*
3569 * Create, destroy, or resize a linear (direct) block of extents.
3570 */
3571 void
3575 {
3584 /* Free extent records */
3587 }
3588 /* Resize direct extent list and zero any new bytes */
3590 /* Check if extents will fit inside the inode */
3596 }
3599 }
3603 rnew_size,
3605 }
3610 }
3611 }
3612 /*
3613 * Switch from the inline extent buffer to a direct
3614 * extent list. Be sure to include the inline extent
3615 * bytes in new_size.
3616 */
3621 }
3623 }
3626 }
3628 /*
3629 * Switch from linear (direct) extent records to inline buffer.
3630 */
3631 void
3635 {
3638 /*
3639 * The inline buffer was zeroed when we switched
3640 * from inline to direct extent allocation mode,
3641 * so we don't need to clear it here.
3642 */
3648 }
3650 /*
3651 * Switch from inline buffer to linear (direct) extent records.
3652 * new_size should already be rounded up to the next power of 2
3653 * by the caller (when appropriate), so use new_size as it is.
3654 * However, since new_size may be rounded up, we can't update
3655 * if_bytes here. It is the caller's responsibility to update
3656 * if_bytes upon return.
3657 */
3658 void
3662 {
3670 }
3672 }
3674 /*
3675 * Resize an extent indirection array to new_size bytes.
3676 */
3677 STATIC void
3681 {
3696 }
3697 }
3699 /*
3700 * Switch from indirection array to linear (direct) extent allocations.
3701 */
3702 STATIC void
3705 {
3725 }
3726 }
3728 /*
3729 * Free incore file extents.
3730 */
3731 void
3734 {
3742 }
3749 }
3753 }
3755 /*
3756 * Return a pointer to the extent record for file system block bno.
3757 */
3763 {
3778 }
3781 /* Find target extent list */
3789 }
3790 /* Binary search extent records */
3801 /* Convert back to file-based extent index */
3804 }
3807 }
3808 }
3809 /* Convert back to file-based extent index */
3812 }
3818 }
3819 }
3822 }
3824 /*
3825 * Return a pointer to the indirection array entry containing the
3826 * extent record for filesystem block bno. Store the index of the
3827 * target irec in *erp_idxp.
3828 */
3834 {
3858 }
3859 }
3862 }
3864 /*
3865 * Return a pointer to the indirection array entry containing the
3866 * extent record at file extent index *idxp. Store the index of the
3867 * target irec in *erp_idxp and store the page index of the target
3868 * extent record in *idxp.
3869 */
3870 xfs_ext_irec_t *
3876 {
3895 /* Binary search extent irec's */
3911 erp_idx++;
3917 }
3918 }
3922 }
3924 /*
3925 * Allocate and initialize an indirection array once the space needed
3926 * for incore extents increases above XFS_IEXT_BUFSZ.
3927 */
3928 void
3931 {
3947 }
3958 }
3960 /*
3961 * Allocate and initialize a new entry in the indirection array.
3962 */
3963 xfs_ext_irec_t *
3967 {
3975 /* Resize indirection array */
3978 /*
3979 * Move records down in the array so the
3980 * new page can use erp_idx.
3981 */
3985 }
3988 /* Initialize new extent record */
3997 }
3999 /*
4000 * Remove a record from the indirection array.
4001 */
4002 void
4006 {
4018 }
4019 /* Compact extent records */
4023 }
4024 /*
4025 * Manually free the last extent record from the indirection
4026 * array. A call to xfs_iext_realloc_indirect() with a size
4027 * of zero would result in a call to xfs_iext_destroy() which
4028 * would in turn call this function again, creating a nasty
4029 * infinite loop.
4030 */
4036 }
4038 }
4040 /*
4041 * This is called to clean up large amounts of unused memory allocated
4042 * by the indirection array. Before compacting anything though, verify
4043 * that the indirection array is still needed and switch back to the
4044 * linear extent list (or even the inline buffer) if possible. The
4045 * compaction policy is as follows:
4046 *
4047 * Full Compaction: Extents fit into a single page (or inline buffer)
4048 * Partial Compaction: Extents occupy less than 50% of allocated space
4049 * No Compaction: Extents occupy at least 50% of allocated space
4050 */
4051 void
4054 {
4071 }
4072 }
4074 /*
4075 * Combine extents from neighboring extent pages.
4076 */
4077 void
4080 {
4096 /*
4097 * Free page before removing extent record
4098 * so er_extoffs don't get modified in
4099 * xfs_iext_irec_remove.
4100 */
4106 erp_idx++;
4107 }
4108 }
4109 }
4111 /*
4112 * This is called to update the er_extoff field in the indirection
4113 * array when extents have been added or removed from one of the
4114 * extent lists. erp_idx contains the irec index to begin updating
4115 * at and ext_diff contains the number of extents that were added
4116 * or removed.
4117 */
4118 void
4123 {
4131 }
4132 }