| blob | d53c39de7d05930413bfb66b82a33fe6e76c813e |
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>
58 /*
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
61 */
62 #define XFS_ITRUNC_MAX_EXTENTS 2
69 #ifdef DEBUG
70 /*
71 * Make sure that the extents in the given memory buffer
72 * are valid.
73 */
74 STATIC void
78 xfs_exntfmt_t fmt)
79 {
80 xfs_bmbt_irec_t irec;
81 xfs_bmbt_rec_host_t rec;
91 }
92 }
94 #define xfs_validate_extents(ifp, nrecs, fmt)
97 /*
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
100 */
101 #if defined(DEBUG)
102 void
106 {
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
119 bp);
121 }
122 }
123 }
124 #endif
126 /*
127 * Find the buffer associated with the given inode map
128 * We do basic validation checks on the buffer once it has been
129 * retrieved from disk.
130 */
131 STATIC int
137 uint buf_flags,
138 uint iget_flags)
139 {
150 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
155 }
157 }
159 /*
160 * Validate the magic number and version of every inode in the buffer
161 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
162 */
163 #ifdef DEBUG
167 #endif
178 XFS_ERRTAG_ITOBP_INOTOBP,
179 XFS_RANDOM_ITOBP_INOTOBP))) {
183 }
186 #ifdef DEBUG
188 "Device %s - bad inode magic/vsn "
193 #endif
196 }
197 }
201 /*
202 * Mark the buffer as an inode buffer now that it looks good
203 */
208 }
210 /*
211 * This routine is called to map an inode number within a file
212 * system to the buffer containing the on-disk version of the
213 * inode. It returns a pointer to the buffer containing the
214 * on-disk inode in the bpp parameter, and in the dip parameter
215 * it returns a pointer to the on-disk inode within that buffer.
216 *
217 * If a non-zero error is returned, then the contents of bpp and
218 * dipp are undefined.
219 *
220 * Use xfs_imap() to determine the size and location of the
221 * buffer to read from disk.
222 */
223 int
227 xfs_ino_t ino,
231 uint imap_flags)
232 {
250 }
253 /*
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
258 * that buffer.
259 *
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
262 *
263 * The inode is expected to already been mapped to its buffer and read
264 * in once, thus we can use the mapping information stored in the inode
265 * rather than calling xfs_imap(). This allows us to avoid the overhead
266 * of looking at the inode btree for small block file systems
267 * (see xfs_imap()).
268 */
269 int
276 uint buf_flags)
277 {
292 }
297 }
299 /*
300 * Move inode type and inode format specific information from the
301 * on-disk inode to the in-core inode. For fifos, devs, and sockets
302 * this means set if_rdev to the proper value. For files, directories,
303 * and symlinks this means to bring in the in-line data or extent
304 * pointers. For a file in B-tree format, only the root is immediately
305 * brought in-core. The rest will be in-lined in if_extents when it
306 * is first referenced (see xfs_iread_extents()).
307 */
308 STATIC int
312 {
316 xfs_fsize_t di_size;
334 }
344 }
354 }
365 }
376 /*
377 * no local regular files yet
378 */
381 "corrupt inode %Lu "
385 XFS_ERRLEVEL_LOW,
388 }
393 "corrupt inode %Lu "
398 XFS_ERRLEVEL_LOW,
401 }
416 }
422 }
425 }
439 "corrupt inode %Lu "
444 XFS_ERRLEVEL_LOW,
447 }
460 }
465 }
467 }
469 /*
470 * The file is in-lined in the on-disk inode.
471 * If it fits into if_inline_data, then copy
472 * it there, otherwise allocate a buffer for it
473 * and copy the data there. Either way, set
474 * if_data to point at the data.
475 * If we allocate a buffer for the data, make
476 * sure that its size is a multiple of 4 and
477 * record the real size in i_real_bytes.
478 */
479 STATIC int
485 {
489 /*
490 * If the size is unreasonable, then something
491 * is wrong and we just bail out rather than crash in
492 * kmem_alloc() or memcpy() below.
493 */
496 "corrupt inode %Lu "
503 }
513 }
521 }
523 /*
524 * The file consists of a set of extents all
525 * of which fit into the on-disk inode.
526 * If there are few enough extents to fit into
527 * the if_inline_ext, then copy them there.
528 * Otherwise allocate a buffer for them and copy
529 * them into it. Either way, set if_extents
530 * to point at the extents.
531 */
532 STATIC int
537 {
548 /*
549 * If the number of extents is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
552 */
560 }
567 else
578 }
585 XFS_ERRLEVEL_LOW,
588 }
589 }
592 }
594 /*
595 * The file has too many extents to fit into
596 * the inode, so they are in B-tree format.
597 * Allocate a buffer for the root of the B-tree
598 * and copy the root into it. The i_extents
599 * field will remain NULL until all of the
600 * extents are read in (when they are needed).
601 */
602 STATIC int
607 {
610 /* REFERENCED */
619 /*
620 * blow out if -- fork has less extents than can fit in
621 * fork (fork shouldn't be a btree format), root btree
622 * block has more records than can fit into the fork,
623 * or the number of extents is greater than the number of
624 * blocks.
625 */
636 }
641 /*
642 * Copy and convert from the on-disk structure
643 * to the in-memory structure.
644 */
652 }
654 STATIC void
658 {
687 }
689 void
693 {
722 }
724 STATIC uint
726 __uint16_t di_flags)
727 {
759 }
762 }
764 uint
767 {
772 }
774 uint
777 {
780 }
782 /*
783 * Read the disk inode attributes into the in-core inode structure.
784 */
785 int
790 xfs_daddr_t bno,
791 uint iget_flags)
792 {
797 /*
798 * Fill in the location information in the in-core inode.
799 */
806 /*
807 * Get pointers to the on-disk inode and the buffer containing it.
808 */
815 /*
816 * If we got something that isn't an inode it means someone
817 * (nfs or dmi) has a stale handle.
818 */
820 #ifdef DEBUG
822 "dip->di_magic (0x%x) != "
825 XFS_DINODE_MAGIC);
829 }
831 /*
832 * If the on-disk inode is already linked to a directory
833 * entry, copy all of the inode into the in-core inode.
834 * xfs_iformat() handles copying in the inode format
835 * specific information.
836 * Otherwise, just get the truly permanent information.
837 */
842 #ifdef DEBUG
845 error);
848 }
854 /*
855 * Make sure to pull in the mode here as well in
856 * case the inode is released without being used.
857 * This ensures that xfs_inactive() will see that
858 * the inode is already free and not try to mess
859 * with the uninitialized part of it.
860 */
862 /*
863 * Initialize the per-fork minima and maxima for a new
864 * inode here. xfs_iformat will do it for old inodes.
865 */
868 }
870 /*
871 * The inode format changed when we moved the link count and
872 * made it 32 bits long. If this is an old format inode,
873 * convert it in memory to look like a new one. If it gets
874 * flushed to disk we will convert back before flushing or
875 * logging it. We zero out the new projid field and the old link
876 * count field. We'll handle clearing the pad field (the remains
877 * of the old uuid field) when we actually convert the inode to
878 * the new format. We don't change the version number so that we
879 * can distinguish this from a real new format inode.
880 */
885 }
890 /*
891 * Mark the buffer containing the inode as something to keep
892 * around for a while. This helps to keep recently accessed
893 * meta-data in-core longer.
894 */
897 /*
898 * Use xfs_trans_brelse() to release the buffer containing the
899 * on-disk inode, because it was acquired with xfs_trans_read_buf()
900 * in xfs_itobp() above. If tp is NULL, this is just a normal
901 * brelse(). If we're within a transaction, then xfs_trans_brelse()
902 * will only release the buffer if it is not dirty within the
903 * transaction. It will be OK to release the buffer in this case,
904 * because inodes on disk are never destroyed and we will be
905 * locking the new in-core inode before putting it in the hash
906 * table where other processes can find it. Thus we don't have
907 * to worry about the inode being changed just because we released
908 * the buffer.
909 */
910 out_brelse:
913 }
915 /*
916 * Read in extents from a btree-format inode.
917 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
918 */
919 int
924 {
927 xfs_extnum_t nextents;
934 }
939 /*
940 * We know that the size is valid (it's checked in iformat_btree)
941 */
951 }
954 }
956 /*
957 * Allocate an inode on disk and return a copy of its in-core version.
958 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
959 * appropriately within the inode. The uid and gid for the inode are
960 * set according to the contents of the given cred structure.
961 *
962 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
963 * has a free inode available, call xfs_iget()
964 * to obtain the in-core version of the allocated inode. Finally,
965 * fill in the inode and log its initial contents. In this case,
966 * ialloc_context would be set to NULL and call_again set to false.
967 *
968 * If xfs_dialloc() does not have an available inode,
969 * it will replenish its supply by doing an allocation. Since we can
970 * only do one allocation within a transaction without deadlocks, we
971 * must commit the current transaction before returning the inode itself.
972 * In this case, therefore, we will set call_again to true and return.
973 * The caller should then commit the current transaction, start a new
974 * transaction, and call xfs_ialloc() again to actually get the inode.
975 *
976 * To ensure that some other process does not grab the inode that
977 * was allocated during the first call to xfs_ialloc(), this routine
978 * also returns the [locked] bp pointing to the head of the freelist
979 * as ialloc_context. The caller should hold this buffer across
980 * the commit and pass it back into this routine on the second call.
981 *
982 * If we are allocating quota inodes, we do not have a parent inode
983 * to attach to or associate with (i.e. pip == NULL) because they
984 * are not linked into the directory structure - they are attached
985 * directly to the superblock - and so have no parent.
986 */
987 int
991 mode_t mode,
992 xfs_nlink_t nlink,
993 xfs_dev_t rdev,
995 xfs_prid_t prid,
1000 {
1001 xfs_ino_t ino;
1003 uint flags;
1005 timespec_t tv;
1008 /*
1009 * Call the space management code to pick
1010 * the on-disk inode to be allocated.
1011 */
1019 }
1022 /*
1023 * Get the in-core inode with the lock held exclusively.
1024 * This is because we're setting fields here we need
1025 * to prevent others from looking at until we're done.
1026 */
1042 /*
1043 * If the superblock version is up to where we support new format
1044 * inodes and this is currently an old format inode, then change
1045 * the inode version number now. This way we only do the conversion
1046 * here rather than here and in the flush/logging code.
1047 */
1051 /*
1052 * We've already zeroed the old link count, the projid field,
1053 * and the pad field.
1054 */
1055 }
1057 /*
1058 * Project ids won't be stored on disk if we are using a version 1 inode.
1059 */
1067 }
1068 }
1070 /*
1071 * If the group ID of the new file does not match the effective group
1072 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1073 * (and only if the irix_sgid_inherit compatibility variable is set).
1074 */
1079 }
1092 /*
1093 * di_gen will have been taken care of in xfs_iread.
1094 */
1111 /*
1112 * we can't set up filestreams until after the VFS inode
1113 * is set up properly.
1114 */
1117 /* fall through */
1128 }
1135 }
1136 }
1138 xfs_inherit_noatime)
1141 xfs_inherit_nodump)
1144 xfs_inherit_sync)
1147 xfs_inherit_nosymlinks)
1152 xfs_inherit_nodefrag)
1157 }
1158 /* FALLTHROUGH */
1167 }
1168 /*
1169 * Attribute fork settings for new inode.
1170 */
1174 /*
1175 * Log the new values stuffed into the inode.
1176 */
1179 /* now that we have an i_mode we can setup inode ops and unlock */
1182 /* now we have set up the vfs inode we can associate the filestream */
1189 }
1193 }
1195 /*
1196 * Check to make sure that there are no blocks allocated to the
1197 * file beyond the size of the file. We don't check this for
1198 * files with fixed size extents or real time extents, but we
1199 * at least do it for regular files.
1200 */
1201 #ifdef DEBUG
1202 void
1206 xfs_fsize_t isize)
1207 {
1208 xfs_fileoff_t map_first;
1223 /*
1224 * The filesystem could be shutting down, so bmapi may return
1225 * an error.
1226 */
1230 map_first),
1236 }
1239 /*
1240 * Calculate the last possible buffered byte in a file. This must
1241 * include data that was buffered beyond the EOF by the write code.
1242 * This also needs to deal with overflowing the xfs_fsize_t type
1243 * which can happen for sizes near the limit.
1244 *
1245 * We also need to take into account any blocks beyond the EOF. It
1246 * may be the case that they were buffered by a write which failed.
1247 * In that case the pages will still be in memory, but the inode size
1248 * will never have been updated.
1249 */
1250 STATIC xfs_fsize_t
1253 {
1255 xfs_fsize_t last_byte;
1256 xfs_fileoff_t last_block;
1257 xfs_fileoff_t size_last_block;
1263 /*
1264 * Only check for blocks beyond the EOF if the extents have
1265 * been read in. This eliminates the need for the inode lock,
1266 * and it also saves us from looking when it really isn't
1267 * necessary.
1268 */
1272 XFS_DATA_FORK);
1276 }
1279 }
1286 }
1290 }
1292 }
1294 /*
1295 * Start the truncation of the file to new_size. The new size
1296 * must be smaller than the current size. This routine will
1297 * clear the buffer and page caches of file data in the removed
1298 * range, and xfs_itruncate_finish() will remove the underlying
1299 * disk blocks.
1300 *
1301 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1302 * must NOT have the inode lock held at all. This is because we're
1303 * calling into the buffer/page cache code and we can't hold the
1304 * inode lock when we do so.
1305 *
1306 * We need to wait for any direct I/Os in flight to complete before we
1307 * proceed with the truncate. This is needed to prevent the extents
1308 * being read or written by the direct I/Os from being removed while the
1309 * I/O is in flight as there is no other method of synchronising
1310 * direct I/O with the truncate operation. Also, because we hold
1311 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1312 * started until the truncate completes and drops the lock. Essentially,
1313 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1314 * ordering between direct I/Os and the truncate operation.
1315 *
1316 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1317 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1318 * in the case that the caller is locking things out of order and
1319 * may not be able to call xfs_itruncate_finish() with the inode lock
1320 * held without dropping the I/O lock. If the caller must drop the
1321 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1322 * must be called again with all the same restrictions as the initial
1323 * call.
1324 */
1325 int
1328 uint flags,
1329 xfs_fsize_t new_size)
1330 {
1331 xfs_fsize_t last_byte;
1332 xfs_off_t toss_start;
1343 /* wait for the completion of any pending DIOs */
1347 /*
1348 * Call toss_pages or flushinval_pages to get rid of pages
1349 * overlapping the region being removed. We have to use
1350 * the less efficient flushinval_pages in the case that the
1351 * caller may not be able to finish the truncate without
1352 * dropping the inode's I/O lock. Make sure
1353 * to catch any pages brought in by buffers overlapping
1354 * the EOF by searching out beyond the isize by our
1355 * block size. We round new_size up to a block boundary
1356 * so that we don't toss things on the same block as
1357 * new_size but before it.
1358 *
1359 * Before calling toss_page or flushinval_pages, make sure to
1360 * call remapf() over the same region if the file is mapped.
1361 * This frees up mapped file references to the pages in the
1362 * given range and for the flushinval_pages case it ensures
1363 * that we get the latest mapped changes flushed out.
1364 */
1368 /*
1369 * The place to start tossing is beyond our maximum
1370 * file size, so there is no way that the data extended
1371 * out there.
1372 */
1374 }
1384 }
1385 }
1387 #ifdef DEBUG
1390 }
1391 #endif
1393 }
1395 /*
1396 * Shrink the file to the given new_size. The new size must be smaller than
1397 * the current size. This will free up the underlying blocks in the removed
1398 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1399 *
1400 * The transaction passed to this routine must have made a permanent log
1401 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1402 * given transaction and start new ones, so make sure everything involved in
1403 * the transaction is tidy before calling here. Some transaction will be
1404 * returned to the caller to be committed. The incoming transaction must
1405 * already include the inode, and both inode locks must be held exclusively.
1406 * The inode must also be "held" within the transaction. On return the inode
1407 * will be "held" within the returned transaction. This routine does NOT
1408 * require any disk space to be reserved for it within the transaction.
1409 *
1410 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1411 * indicates the fork which is to be truncated. For the attribute fork we only
1412 * support truncation to size 0.
1413 *
1414 * We use the sync parameter to indicate whether or not the first transaction
1415 * we perform might have to be synchronous. For the attr fork, it needs to be
1416 * so if the unlink of the inode is not yet known to be permanent in the log.
1417 * This keeps us from freeing and reusing the blocks of the attribute fork
1418 * before the unlink of the inode becomes permanent.
1419 *
1420 * For the data fork, we normally have to run synchronously if we're being
1421 * called out of the inactive path or we're being called out of the create path
1422 * where we're truncating an existing file. Either way, the truncate needs to
1423 * be sync so blocks don't reappear in the file with altered data in case of a
1424 * crash. wsync filesystems can run the first case async because anything that
1425 * shrinks the inode has to run sync so by the time we're called here from
1426 * inactive, the inode size is permanently set to 0.
1427 *
1428 * Calls from the truncate path always need to be sync unless we're in a wsync
1429 * filesystem and the file has already been unlinked.
1430 *
1431 * The caller is responsible for correctly setting the sync parameter. It gets
1432 * too hard for us to guess here which path we're being called out of just
1433 * based on inode state.
1434 *
1435 * If we get an error, we must return with the inode locked and linked into the
1436 * current transaction. This keeps things simple for the higher level code,
1437 * because it always knows that the inode is locked and held in the transaction
1438 * that returns to it whether errors occur or not. We don't mark the inode
1439 * dirty on error so that transactions can be easily aborted if possible.
1440 */
1441 int
1445 xfs_fsize_t new_size,
1448 {
1449 xfs_fsblock_t first_block;
1450 xfs_fileoff_t first_unmap_block;
1451 xfs_fileoff_t last_block;
1457 xfs_bmap_free_t free_list;
1473 /*
1474 * We only support truncating the entire attribute fork.
1475 */
1478 }
1482 /*
1483 * The first thing we do is set the size to new_size permanently
1484 * on disk. This way we don't have to worry about anyone ever
1485 * being able to look at the data being freed even in the face
1486 * of a crash. What we're getting around here is the case where
1487 * we free a block, it is allocated to another file, it is written
1488 * to, and then we crash. If the new data gets written to the
1489 * file but the log buffers containing the free and reallocation
1490 * don't, then we'd end up with garbage in the blocks being freed.
1491 * As long as we make the new_size permanent before actually
1492 * freeing any blocks it doesn't matter if they get writtten to.
1493 *
1494 * The callers must signal into us whether or not the size
1495 * setting here must be synchronous. There are a few cases
1496 * where it doesn't have to be synchronous. Those cases
1497 * occur if the file is unlinked and we know the unlink is
1498 * permanent or if the blocks being truncated are guaranteed
1499 * to be beyond the inode eof (regardless of the link count)
1500 * and the eof value is permanent. Both of these cases occur
1501 * only on wsync-mounted filesystems. In those cases, we're
1502 * guaranteed that no user will ever see the data in the blocks
1503 * that are being truncated so the truncate can run async.
1504 * In the free beyond eof case, the file may wind up with
1505 * more blocks allocated to it than it needs if we crash
1506 * and that won't get fixed until the next time the file
1507 * is re-opened and closed but that's ok as that shouldn't
1508 * be too many blocks.
1509 *
1510 * However, we can't just make all wsync xactions run async
1511 * because there's one call out of the create path that needs
1512 * to run sync where it's truncating an existing file to size
1513 * 0 whose size is > 0.
1514 *
1515 * It's probably possible to come up with a test in this
1516 * routine that would correctly distinguish all the above
1517 * cases from the values of the function parameters and the
1518 * inode state but for sanity's sake, I've decided to let the
1519 * layers above just tell us. It's simpler to correctly figure
1520 * out in the layer above exactly under what conditions we
1521 * can run async and I think it's easier for others read and
1522 * follow the logic in case something has to be changed.
1523 * cscope is your friend -- rcc.
1524 *
1525 * The attribute fork is much simpler.
1526 *
1527 * For the attribute fork we allow the caller to tell us whether
1528 * the unlink of the inode that led to this call is yet permanent
1529 * in the on disk log. If it is not and we will be freeing extents
1530 * in this inode then we make the first transaction synchronous
1531 * to make sure that the unlink is permanent by the time we free
1532 * the blocks.
1533 */
1536 /*
1537 * If we are not changing the file size then do
1538 * not update the on-disk file size - we may be
1539 * called from xfs_inactive_free_eofblocks(). If we
1540 * update the on-disk file size and then the system
1541 * crashes before the contents of the file are
1542 * flushed to disk then the files may be full of
1543 * holes (ie NULL files bug).
1544 */
1549 }
1550 }
1555 }
1561 /*
1562 * Since it is possible for space to become allocated beyond
1563 * the end of the file (in a crash where the space is allocated
1564 * but the inode size is not yet updated), simply remove any
1565 * blocks which show up between the new EOF and the maximum
1566 * possible file size. If the first block to be removed is
1567 * beyond the maximum file size (ie it is the same as last_block),
1568 * then there is nothing to do.
1569 */
1577 }
1579 /*
1580 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1581 * will tell us whether it freed the entire range or
1582 * not. If this is a synchronous mount (wsync),
1583 * then we can tell bunmapi to keep all the
1584 * transactions asynchronous since the unlink
1585 * transaction that made this inode inactive has
1586 * already hit the disk. There's no danger of
1587 * the freed blocks being reused, there being a
1588 * crash, and the reused blocks suddenly reappearing
1589 * in this file with garbage in them once recovery
1590 * runs.
1591 */
1597 XFS_ITRUNC_MAX_EXTENTS,
1601 /*
1602 * If the bunmapi call encounters an error,
1603 * return to the caller where the transaction
1604 * can be properly aborted. We just need to
1605 * make sure we're not holding any resources
1606 * that we were not when we came in.
1607 */
1610 }
1612 /*
1613 * Duplicate the transaction that has the permanent
1614 * reservation and commit the old transaction.
1615 */
1619 /* link the inode into the next xact in the chain */
1623 }
1626 /*
1627 * If the bmap finish call encounters an error, return
1628 * to the caller where the transaction can be properly
1629 * aborted. We just need to make sure we're not
1630 * holding any resources that we were not when we came
1631 * in.
1632 *
1633 * Aborting from this point might lose some blocks in
1634 * the file system, but oh well.
1635 */
1638 }
1641 /*
1642 * Mark the inode dirty so it will be logged and
1643 * moved forward in the log as part of every commit.
1644 */
1646 }
1652 /* link the inode into the next transaction in the chain */
1658 /*
1659 * transaction commit worked ok so we can drop the extra ticket
1660 * reference that we gained in xfs_trans_dup()
1661 */
1665 XFS_TRANS_PERM_LOG_RES,
1666 XFS_ITRUNCATE_LOG_COUNT);
1669 }
1670 /*
1671 * Only update the size in the case of the data fork, but
1672 * always re-log the inode so that our permanent transaction
1673 * can keep on rolling it forward in the log.
1674 */
1677 /*
1678 * If we are not changing the file size then do
1679 * not update the on-disk file size - we may be
1680 * called from xfs_inactive_free_eofblocks(). If we
1681 * update the on-disk file size and then the system
1682 * crashes before the contents of the file are
1683 * flushed to disk then the files may be full of
1684 * holes (ie NULL files bug).
1685 */
1689 }
1690 }
1700 }
1702 /*
1703 * This is called when the inode's link count goes to 0.
1704 * We place the on-disk inode on a list in the AGI. It
1705 * will be pulled from this list when the inode is freed.
1706 */
1707 int
1711 {
1717 xfs_agino_t agino;
1728 /*
1729 * Get the agi buffer first. It ensures lock ordering
1730 * on the list.
1731 */
1737 /*
1738 * Get the index into the agi hash table for the
1739 * list this inode will go on.
1740 */
1748 /*
1749 * There is already another inode in the bucket we need
1750 * to add ourselves to. Add us at the front of the list.
1751 * Here we put the head pointer into our next pointer,
1752 * and then we fall through to point the head at us.
1753 */
1759 /* both on-disk, don't endian flip twice */
1767 }
1769 /*
1770 * Point the bucket head pointer at the inode being inserted.
1771 */
1779 }
1781 /*
1782 * Pull the on-disk inode from the AGI unlinked list.
1783 */
1784 STATIC int
1788 {
1789 xfs_ino_t next_ino;
1795 xfs_agnumber_t agno;
1796 xfs_agino_t agino;
1797 xfs_agino_t next_agino;
1807 /*
1808 * Get the agi buffer first. It ensures lock ordering
1809 * on the list.
1810 */
1817 /*
1818 * Get the index into the agi hash table for the
1819 * list this inode will go on.
1820 */
1828 /*
1829 * We're at the head of the list. Get the inode's
1830 * on-disk buffer to see if there is anyone after us
1831 * on the list. Only modify our next pointer if it
1832 * is not already NULLAGINO. This saves us the overhead
1833 * of dealing with the buffer when there is no need to
1834 * change it.
1835 */
1842 }
1855 }
1856 /*
1857 * Point the bucket head pointer at the next inode.
1858 */
1867 /*
1868 * We need to search the list for the inode being freed.
1869 */
1873 /*
1874 * If the last inode wasn't the one pointing to
1875 * us, then release its buffer since we're not
1876 * going to do anything with it.
1877 */
1880 }
1889 }
1893 }
1894 /*
1895 * Now last_ibp points to the buffer previous to us on
1896 * the unlinked list. Pull us from the list.
1897 */
1904 }
1918 }
1919 /*
1920 * Point the previous inode on the list to the next inode.
1921 */
1929 }
1931 }
1933 STATIC void
1937 xfs_ino_t inum)
1938 {
1944 xfs_daddr_t blkno;
1961 }
1969 /*
1970 * We obtain and lock the backing buffer first in the process
1971 * here, as we have to ensure that any dirty inode that we
1972 * can't get the flush lock on is attached to the buffer.
1973 * If we scan the in-memory inodes first, then buffer IO can
1974 * complete before we get a lock on it, and hence we may fail
1975 * to mark all the active inodes on the buffer stale.
1976 */
1979 XBF_LOCK);
1981 /*
1982 * Walk the inodes already attached to the buffer and mark them
1983 * stale. These will all have the flush locks held, so an
1984 * in-memory inode walk can't lock them.
1985 */
1996 found++;
1997 }
1999 }
2001 /*
2002 * For each inode in memory attempt to add it to the inode
2003 * buffer and set it up for being staled on buffer IO
2004 * completion. This is safe as we've locked out tail pushing
2005 * and flushing by locking the buffer.
2006 *
2007 * We have already marked every inode that was part of a
2008 * transaction stale above, which means there is no point in
2009 * even trying to lock them.
2010 */
2016 /* Inode not in memory or stale, nothing to do */
2020 }
2022 /* don't try to lock/unlock the current inode */
2027 }
2034 }
2042 }
2046 /* inode with unlogged changes only */
2052 }
2053 found++;
2067 }
2072 }
2075 }
2077 /*
2078 * This is called to return an inode to the inode free list.
2079 * The inode should already be truncated to 0 length and have
2080 * no pages associated with it. This routine also assumes that
2081 * the inode is already a part of the transaction.
2082 *
2083 * The on-disk copy of the inode will have been added to the list
2084 * of unlinked inodes in the AGI. We need to remove the inode from
2085 * that list atomically with respect to freeing it here.
2086 */
2087 int
2092 {
2095 xfs_ino_t first_ino;
2108 /*
2109 * Pull the on-disk inode from the AGI unlinked list.
2110 */
2114 }
2119 }
2128 /*
2129 * Bump the generation count so no one will be confused
2130 * by reincarnations of this inode.
2131 */
2140 /*
2141 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2142 * from picking up this inode when it is reclaimed (its incore state
2143 * initialzed but not flushed to disk yet). The in-core di_mode is
2144 * already cleared and a corresponding transaction logged.
2145 * The hack here just synchronizes the in-core to on-disk
2146 * di_mode value in advance before the actual inode sync to disk.
2147 * This is OK because the inode is already unlinked and would never
2148 * change its di_mode again for this inode generation.
2149 * This is a temporary hack that would require a proper fix
2150 * in the future.
2151 */
2156 }
2159 }
2161 /*
2162 * Reallocate the space for if_broot based on the number of records
2163 * being added or deleted as indicated in rec_diff. Move the records
2164 * and pointers in if_broot to fit the new size. When shrinking this
2165 * will eliminate holes between the records and pointers created by
2166 * the caller. When growing this will create holes to be filled in
2167 * by the caller.
2168 *
2169 * The caller must not request to add more records than would fit in
2170 * the on-disk inode root. If the if_broot is currently NULL, then
2171 * if we adding records one will be allocated. The caller must also
2172 * not request that the number of records go below zero, although
2173 * it can go to zero.
2174 *
2175 * ip -- the inode whose if_broot area is changing
2176 * ext_diff -- the change in the number of records, positive or negative,
2177 * requested for the if_broot array.
2178 */
2179 void
2184 {
2194 /*
2195 * Handle the degenerate case quietly.
2196 */
2199 }
2203 /*
2204 * If there wasn't any memory allocated before, just
2205 * allocate it now and get out.
2206 */
2212 }
2214 /*
2215 * If there is already an existing if_broot, then we need
2216 * to realloc() it and shift the pointers to their new
2217 * location. The records don't change location because
2218 * they are kept butted up against the btree block header.
2219 */
2225 KM_SLEEP);
2235 }
2237 /*
2238 * rec_diff is less than 0. In this case, we are shrinking the
2239 * if_broot buffer. It must already exist. If we go to zero
2240 * records, just get rid of the root and clear the status bit.
2241 */
2248 else
2252 /*
2253 * First copy over the btree block header.
2254 */
2259 }
2261 /*
2262 * Only copy the records and pointers if there are any.
2263 */
2265 /*
2266 * First copy the records.
2267 */
2272 /*
2273 * Then copy the pointers.
2274 */
2280 }
2287 }
2290 /*
2291 * This is called when the amount of space needed for if_data
2292 * is increased or decreased. The change in size is indicated by
2293 * the number of bytes that need to be added or deleted in the
2294 * byte_diff parameter.
2295 *
2296 * If the amount of space needed has decreased below the size of the
2297 * inline buffer, then switch to using the inline buffer. Otherwise,
2298 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2299 * to what is needed.
2300 *
2301 * ip -- the inode whose if_data area is changing
2302 * byte_diff -- the change in the number of bytes, positive or negative,
2303 * requested for the if_data array.
2304 */
2305 void
2310 {
2317 }
2326 }
2330 /*
2331 * If the valid extents/data can fit in if_inline_ext/data,
2332 * copy them from the malloc'd vector and free it.
2333 */
2339 new_size);
2342 }
2345 /*
2346 * Stuck with malloc/realloc.
2347 * For inline data, the underlying buffer must be
2348 * a multiple of 4 bytes in size so that it can be
2349 * logged and stay on word boundaries. We enforce
2350 * that here.
2351 */
2357 /*
2358 * Only do the realloc if the underlying size
2359 * is really changing.
2360 */
2364 real_size,
2366 KM_SLEEP);
2367 }
2373 }
2374 }
2378 }
2380 void
2384 {
2391 }
2393 /*
2394 * If the format is local, then we can't have an extents
2395 * array so just look for an inline data array. If we're
2396 * not local then we may or may not have an extents list,
2397 * so check and free it up if we do.
2398 */
2406 }
2413 }
2420 }
2421 }
2423 /*
2424 * This is called to unpin an inode. The caller must have the inode locked
2425 * in at least shared mode so that the buffer cannot be subsequently pinned
2426 * once someone is waiting for it to be unpinned.
2427 */
2428 static void
2431 {
2436 /* Give the log a push to start the unpinning I/O */
2439 }
2441 void
2444 {
2448 }
2449 }
2451 /*
2452 * xfs_iextents_copy()
2453 *
2454 * This is called to copy the REAL extents (as opposed to the delayed
2455 * allocation extents) from the inode into the given buffer. It
2456 * returns the number of bytes copied into the buffer.
2457 *
2458 * If there are no delayed allocation extents, then we can just
2459 * memcpy() the extents into the buffer. Otherwise, we need to
2460 * examine each extent in turn and skip those which are delayed.
2461 */
2462 int
2467 {
2472 xfs_fsblock_t start_block;
2482 /*
2483 * There are some delayed allocation extents in the
2484 * inode, so copy the extents one at a time and skip
2485 * the delayed ones. There must be at least one
2486 * non-delayed extent.
2487 */
2493 /*
2494 * It's a delayed allocation extent, so skip it.
2495 */
2497 }
2499 /* Translate to on disk format */
2502 dp++;
2503 copied++;
2504 }
2509 }
2511 /*
2512 * Each of the following cases stores data into the same region
2513 * of the on-disk inode, so only one of them can be valid at
2514 * any given time. While it is possible to have conflicting formats
2515 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2516 * in EXTENTS format, this can only happen when the fork has
2517 * changed formats after being modified but before being flushed.
2518 * In these cases, the format always takes precedence, because the
2519 * format indicates the current state of the fork.
2520 */
2521 /*ARGSUSED*/
2522 STATIC void
2529 {
2533 #ifdef XFS_TRANS_DEBUG
2535 #endif
2546 /*
2547 * This can happen if we gave up in iformat in an error path,
2548 * for the attribute fork.
2549 */
2553 }
2563 }
2577 whichfork);
2578 }
2587 XFS_BROOT_SIZE_ADJ));
2591 }
2598 }
2607 }
2613 }
2614 }
2616 STATIC int
2620 {
2644 /* really need a gang lookup range call here */
2654 /* if the inode lies outside this cluster, we're done. */
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 */
2745 }
2746 }
2748 /*
2749 * Unlocks the flush lock
2750 */
2755 }
2757 /*
2758 * xfs_iflush() will write a modified inode's changes out to the
2759 * inode's on disk home. The caller must have the inode lock held
2760 * in at least shared mode and the inode flush completion must be
2761 * active as well. The inode lock will still be held upon return from
2762 * the call and the caller is free to unlock it.
2763 * The inode flush will be completed when the inode reaches the disk.
2764 * The flags indicate how the inode's buffer should be written out.
2765 */
2766 int
2769 uint flags)
2770 {
2787 /*
2788 * We can't flush the inode until it is unpinned, so wait for it if we
2789 * are allowed to block. We know noone new can pin it, because we are
2790 * holding the inode lock shared and you need to hold it exclusively to
2791 * pin the inode.
2792 *
2793 * If we are not allowed to block, force the log out asynchronously so
2794 * that when we come back the inode will be unpinned. If other inodes
2795 * in the same cluster are dirty, they will probably write the inode
2796 * out for us if they occur after the log force completes.
2797 */
2802 }
2805 /*
2806 * For stale inodes we cannot rely on the backing buffer remaining
2807 * stale in cache for the remaining life of the stale inode and so
2808 * xfs_itobp() below may give us a buffer that no longer contains
2809 * inodes below. We have to check this after ensuring the inode is
2810 * unpinned so that it is safe to reclaim the stale inode after the
2811 * flush call.
2812 */
2816 }
2818 /*
2819 * This may have been unpinned because the filesystem is shutting
2820 * down forcibly. If that's the case we must not write this inode
2821 * to disk, because the log record didn't make it to disk!
2822 */
2829 }
2831 /*
2832 * Get the buffer containing the on-disk inode.
2833 */
2839 }
2841 /*
2842 * First flush out the inode that xfs_iflush was called with.
2843 */
2848 /*
2849 * If the buffer is pinned then push on the log now so we won't
2850 * get stuck waiting in the write for too long.
2851 */
2855 /*
2856 * inode clustering:
2857 * see if other inodes can be gathered into this write
2858 */
2865 else
2869 corrupt_out:
2872 cluster_corrupt_out:
2873 /*
2874 * Unlocks the flush lock
2875 */
2878 }
2881 STATIC int
2885 {
2889 #ifdef XFS_TRANS_DEBUG
2891 #endif
2901 /* set *dip = inode's place in the buffer */
2904 /*
2905 * Clear i_update_core before copying out the data.
2906 * This is for coordination with our timestamp updates
2907 * that don't hold the inode lock. They will always
2908 * update the timestamps BEFORE setting i_update_core,
2909 * so if we clear i_update_core after they set it we
2910 * are guaranteed to see their updates to the timestamps.
2911 * I believe that this depends on strongly ordered memory
2912 * semantics, but we have that. We use the SYNCHRONIZE
2913 * macro to make sure that the compiler does not reorder
2914 * the i_update_core access below the data copy below.
2915 */
2919 /*
2920 * Make sure to get the latest timestamps from the Linux inode.
2921 */
2930 }
2937 }
2947 }
2958 }
2959 }
2962 XFS_RANDOM_IFLUSH_5)) {
2968 ip);
2970 }
2977 }
2978 /*
2979 * bump the flush iteration count, used to detect flushes which
2980 * postdate a log record during recovery.
2981 */
2985 /*
2986 * Copy the dirty parts of the inode into the on-disk
2987 * inode. We always copy out the core of the inode,
2988 * because if the inode is dirty at all the core must
2989 * be.
2990 */
2993 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2997 /*
2998 * If this is really an old format inode and the superblock version
2999 * has not been updated to support only new format inodes, then
3000 * convert back to the old inode format. If the superblock version
3001 * has been updated, then make the conversion permanent.
3002 */
3006 /*
3007 * Convert it back.
3008 */
3012 /*
3013 * The superblock version has already been bumped,
3014 * so just make the conversion to the new inode
3015 * format permanent.
3016 */
3025 }
3026 }
3033 /*
3034 * We've recorded everything logged in the inode, so we'd
3035 * like to clear the ilf_fields bits so we don't log and
3036 * flush things unnecessarily. However, we can't stop
3037 * logging all this information until the data we've copied
3038 * into the disk buffer is written to disk. If we did we might
3039 * overwrite the copy of the inode in the log with all the
3040 * data after re-logging only part of it, and in the face of
3041 * a crash we wouldn't have all the data we need to recover.
3042 *
3043 * What we do is move the bits to the ili_last_fields field.
3044 * When logging the inode, these bits are moved back to the
3045 * ilf_fields field. In the xfs_iflush_done() routine we
3046 * clear ili_last_fields, since we know that the information
3047 * those bits represent is permanently on disk. As long as
3048 * the flush completes before the inode is logged again, then
3049 * both ilf_fields and ili_last_fields will be cleared.
3050 *
3051 * We can play with the ilf_fields bits here, because the inode
3052 * lock must be held exclusively in order to set bits there
3053 * and the flush lock protects the ili_last_fields bits.
3054 * Set ili_logged so the flush done
3055 * routine can tell whether or not to look in the AIL.
3056 * Also, store the current LSN of the inode so that we can tell
3057 * whether the item has moved in the AIL from xfs_iflush_done().
3058 * In order to read the lsn we need the AIL lock, because
3059 * it is a 64 bit value that cannot be read atomically.
3060 */
3069 /*
3070 * Attach the function xfs_iflush_done to the inode's
3071 * buffer. This will remove the inode from the AIL
3072 * and unlock the inode's flush lock when the inode is
3073 * completely written to disk.
3074 */
3081 /*
3082 * We're flushing an inode which is not in the AIL and has
3083 * not been logged but has i_update_core set. For this
3084 * case we can use a B_DELWRI flush and immediately drop
3085 * the inode flush lock because we can avoid the whole
3086 * AIL state thing. It's OK to drop the flush lock now,
3087 * because we've already locked the buffer and to do anything
3088 * you really need both.
3089 */
3094 }
3096 }
3100 corrupt_out:
3102 }
3104 /*
3105 * Return a pointer to the extent record at file index idx.
3106 */
3107 xfs_bmbt_rec_host_t *
3111 {
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 }
3193 }
3194 /*
3195 * Otherwise use a linear (direct) extent list.
3196 * If the extents are currently inside the inode,
3197 * xfs_iext_realloc_direct will switch us from
3198 * inline to direct extent allocation mode.
3199 */
3207 }
3208 }
3209 /* Indirection array */
3222 }
3223 /* Extents fit in target extent page */
3231 }
3234 }
3235 /* Insert a new extent page */
3239 }
3240 /*
3241 * If extent(s) are being appended to the last page in
3242 * the indirection array and the new extent(s) don't fit
3243 * in the page, then erp is NULL and erp_idx is set to
3244 * the next index needed in the indirection array.
3245 */
3254 erp_idx++;
3255 }
3256 }
3257 }
3258 }
3260 }
3262 /*
3263 * This is called when incore extents are being added to the indirection
3264 * array and the new extents do not fit in the target extent list. The
3265 * erp_idx parameter contains the irec index for the target extent list
3266 * in the indirection array, and the idx parameter contains the extent
3267 * index within the list. The number of extents being added is stored
3268 * in the count parameter.
3269 *
3270 * |-------| |-------|
3271 * | | | | idx - number of extents before idx
3272 * | idx | | count |
3273 * | | | | count - number of extents being inserted at idx
3274 * |-------| |-------|
3275 * | count | | nex2 | nex2 - number of extents after idx + count
3276 * |-------| |-------|
3277 */
3278 void
3284 {
3298 /*
3299 * Save second part of target extent list
3300 * (all extents past */
3308 }
3310 /*
3311 * Add the new extents to the end of the target
3312 * list, then allocate new irec record(s) and
3313 * extent buffer(s) as needed to store the rest
3314 * of the new extents.
3315 */
3322 }
3324 erp_idx++;
3330 }
3332 /* Add nex2 extents back to indirection array */
3334 xfs_extnum_t ext_avail;
3340 /*
3341 * If nex2 extents fit in the current page, append
3342 * nex2_ep after the new extents.
3343 */
3346 }
3347 /*
3348 * Otherwise, check if space is available in the
3349 * next page.
3350 */
3354 erp_idx++;
3355 erp++;
3356 /* Create a hole for nex2 extents */
3359 }
3360 /*
3361 * Final choice, create a new extent page for
3362 * nex2 extents.
3363 */
3365 erp_idx++;
3367 }
3372 }
3373 }
3375 /*
3376 * This is called when the amount of space required for incore file
3377 * extents needs to be decreased. The ext_diff parameter stores the
3378 * number of extents to be removed and the idx parameter contains
3379 * the extent index where the extents will be removed from.
3380 *
3381 * If the amount of space needed has decreased below the linear
3382 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3383 * extent array. Otherwise, use kmem_realloc() to adjust the
3384 * size to what is needed.
3385 */
3386 void
3392 {
3411 }
3413 }
3415 /*
3416 * This removes ext_diff extents from the inline buffer, beginning
3417 * at extent index idx.
3418 */
3419 void
3424 {
3443 }
3444 }
3446 /*
3447 * This removes ext_diff extents from a linear (direct) extent list,
3448 * beginning at extent index idx. If the extents are being removed
3449 * from the end of the list (ie. truncate) then we just need to re-
3450 * allocate the list to remove the extra space. Otherwise, if the
3451 * extents are being removed from the middle of the existing extent
3452 * entries, then we first need to move the extent records beginning
3453 * at idx + ext_diff up in the list to overwrite the records being
3454 * removed, then remove the extra space via kmem_realloc.
3455 */
3456 void
3461 {
3473 }
3474 /* Move extents up in the list (if needed) */
3480 }
3483 /*
3484 * Reallocate the direct extent list. If the extents
3485 * will fit inside the inode then xfs_iext_realloc_direct
3486 * will switch from direct to inline extent allocation
3487 * mode for us.
3488 */
3491 }
3493 /*
3494 * This is called when incore extents are being removed from the
3495 * indirection array and the extents being removed span multiple extent
3496 * buffers. The idx parameter contains the file extent index where we
3497 * want to begin removing extents, and the count parameter contains
3498 * how many extents need to be removed.
3499 *
3500 * |-------| |-------|
3501 * | nex1 | | | nex1 - number of extents before idx
3502 * |-------| | count |
3503 * | | | | count - number of extents being removed at idx
3504 * | count | |-------|
3505 * | | | nex2 | nex2 - number of extents after idx + count
3506 * |-------| |-------|
3507 */
3508 void
3513 {
3532 /*
3533 * Check for deletion of entire list;
3534 * xfs_iext_irec_remove() updates extent offsets.
3535 */
3542 XFS_IEXT_BUFSZ);
3548 }
3549 }
3550 /* Move extents up (if needed) */
3555 }
3556 /* Zero out rest of page */
3559 /* Update remaining counters */
3564 erp_idx++;
3565 erp++;
3566 }
3569 }
3571 /*
3572 * Create, destroy, or resize a linear (direct) block of extents.
3573 */
3574 void
3578 {
3587 /* Free extent records */
3590 }
3591 /* Resize direct extent list and zero any new bytes */
3593 /* Check if extents will fit inside the inode */
3599 }
3602 }
3606 rnew_size,
3608 }
3613 }
3614 }
3615 /*
3616 * Switch from the inline extent buffer to a direct
3617 * extent list. Be sure to include the inline extent
3618 * bytes in new_size.
3619 */
3624 }
3626 }
3629 }
3631 /*
3632 * Switch from linear (direct) extent records to inline buffer.
3633 */
3634 void
3638 {
3641 /*
3642 * The inline buffer was zeroed when we switched
3643 * from inline to direct extent allocation mode,
3644 * so we don't need to clear it here.
3645 */
3651 }
3653 /*
3654 * Switch from inline buffer to linear (direct) extent records.
3655 * new_size should already be rounded up to the next power of 2
3656 * by the caller (when appropriate), so use new_size as it is.
3657 * However, since new_size may be rounded up, we can't update
3658 * if_bytes here. It is the caller's responsibility to update
3659 * if_bytes upon return.
3660 */
3661 void
3665 {
3673 }
3675 }
3677 /*
3678 * Resize an extent indirection array to new_size bytes.
3679 */
3680 STATIC void
3684 {
3699 }
3700 }
3702 /*
3703 * Switch from indirection array to linear (direct) extent allocations.
3704 */
3705 STATIC void
3708 {
3728 }
3729 }
3731 /*
3732 * Free incore file extents.
3733 */
3734 void
3737 {
3745 }
3752 }
3756 }
3758 /*
3759 * Return a pointer to the extent record for file system block bno.
3760 */
3766 {
3781 }
3784 /* Find target extent list */
3792 }
3793 /* Binary search extent records */
3804 /* Convert back to file-based extent index */
3807 }
3810 }
3811 }
3812 /* Convert back to file-based extent index */
3815 }
3821 }
3822 }
3825 }
3827 /*
3828 * Return a pointer to the indirection array entry containing the
3829 * extent record for filesystem block bno. Store the index of the
3830 * target irec in *erp_idxp.
3831 */
3837 {
3861 }
3862 }
3865 }
3867 /*
3868 * Return a pointer to the indirection array entry containing the
3869 * extent record at file extent index *idxp. Store the index of the
3870 * target irec in *erp_idxp and store the page index of the target
3871 * extent record in *idxp.
3872 */
3873 xfs_ext_irec_t *
3879 {
3896 /* Binary search extent irec's */
3912 erp_idx++;
3918 }
3919 }
3923 }
3925 /*
3926 * Allocate and initialize an indirection array once the space needed
3927 * for incore extents increases above XFS_IEXT_BUFSZ.
3928 */
3929 void
3932 {
3948 }
3959 }
3961 /*
3962 * Allocate and initialize a new entry in the indirection array.
3963 */
3964 xfs_ext_irec_t *
3968 {
3976 /* Resize indirection array */
3979 /*
3980 * Move records down in the array so the
3981 * new page can use erp_idx.
3982 */
3986 }
3989 /* Initialize new extent record */
3998 }
4000 /*
4001 * Remove a record from the indirection array.
4002 */
4003 void
4007 {
4019 }
4020 /* Compact extent records */
4024 }
4025 /*
4026 * Manually free the last extent record from the indirection
4027 * array. A call to xfs_iext_realloc_indirect() with a size
4028 * of zero would result in a call to xfs_iext_destroy() which
4029 * would in turn call this function again, creating a nasty
4030 * infinite loop.
4031 */
4037 }
4039 }
4041 /*
4042 * This is called to clean up large amounts of unused memory allocated
4043 * by the indirection array. Before compacting anything though, verify
4044 * that the indirection array is still needed and switch back to the
4045 * linear extent list (or even the inline buffer) if possible. The
4046 * compaction policy is as follows:
4047 *
4048 * Full Compaction: Extents fit into a single page (or inline buffer)
4049 * Partial Compaction: Extents occupy less than 50% of allocated space
4050 * No Compaction: Extents occupy at least 50% of allocated space
4051 */
4052 void
4055 {
4072 }
4073 }
4075 /*
4076 * Combine extents from neighboring extent pages.
4077 */
4078 void
4081 {
4097 /*
4098 * Free page before removing extent record
4099 * so er_extoffs don't get modified in
4100 * xfs_iext_irec_remove.
4101 */
4107 erp_idx++;
4108 }
4109 }
4110 }
4112 /*
4113 * This is called to update the er_extoff field in the indirection
4114 * array when extents have been added or removed from one of the
4115 * extent lists. erp_idx contains the irec index to begin updating
4116 * at and ext_diff contains the number of extents that were added
4117 * or removed.
4118 */
4119 void
4124 {
4132 }
4133 }