| blob | ef77fd88c8e3f1b5ff766983035de5e209efe591 |
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;
1960 }
1969 /*
1970 * Look for each inode in memory and attempt to lock it,
1971 * we can be racing with flush and tail pushing here.
1972 * any inode we get the locks on, add to an array of
1973 * inode items to process later.
1974 *
1975 * The get the buffer lock, we could beat a flush
1976 * or tail pushing thread to the lock here, in which
1977 * case they will go looking for the inode buffer
1978 * and fail, we need some other form of interlock
1979 * here.
1980 */
1987 /* Inode not in memory or we found it already,
1988 * nothing to do
1989 */
1993 }
1998 }
2000 /* If we can get the locks then add it to the
2001 * list, otherwise by the time we get the bp lock
2002 * below it will already be attached to the
2003 * inode buffer.
2004 */
2006 /* This inode will already be locked - by us, lets
2007 * keep it that way.
2008 */
2017 }
2018 }
2021 }
2032 }
2035 }
2036 }
2038 }
2042 XFS_BUF_LOCK);
2055 pre_flushed++;
2056 }
2058 }
2069 }
2082 }
2083 }
2088 }
2092 }
2094 /*
2095 * This is called to return an inode to the inode free list.
2096 * The inode should already be truncated to 0 length and have
2097 * no pages associated with it. This routine also assumes that
2098 * the inode is already a part of the transaction.
2099 *
2100 * The on-disk copy of the inode will have been added to the list
2101 * of unlinked inodes in the AGI. We need to remove the inode from
2102 * that list atomically with respect to freeing it here.
2103 */
2104 int
2109 {
2112 xfs_ino_t first_ino;
2125 /*
2126 * Pull the on-disk inode from the AGI unlinked list.
2127 */
2131 }
2136 }
2145 /*
2146 * Bump the generation count so no one will be confused
2147 * by reincarnations of this inode.
2148 */
2157 /*
2158 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2159 * from picking up this inode when it is reclaimed (its incore state
2160 * initialzed but not flushed to disk yet). The in-core di_mode is
2161 * already cleared and a corresponding transaction logged.
2162 * The hack here just synchronizes the in-core to on-disk
2163 * di_mode value in advance before the actual inode sync to disk.
2164 * This is OK because the inode is already unlinked and would never
2165 * change its di_mode again for this inode generation.
2166 * This is a temporary hack that would require a proper fix
2167 * in the future.
2168 */
2173 }
2176 }
2178 /*
2179 * Reallocate the space for if_broot based on the number of records
2180 * being added or deleted as indicated in rec_diff. Move the records
2181 * and pointers in if_broot to fit the new size. When shrinking this
2182 * will eliminate holes between the records and pointers created by
2183 * the caller. When growing this will create holes to be filled in
2184 * by the caller.
2185 *
2186 * The caller must not request to add more records than would fit in
2187 * the on-disk inode root. If the if_broot is currently NULL, then
2188 * if we adding records one will be allocated. The caller must also
2189 * not request that the number of records go below zero, although
2190 * it can go to zero.
2191 *
2192 * ip -- the inode whose if_broot area is changing
2193 * ext_diff -- the change in the number of records, positive or negative,
2194 * requested for the if_broot array.
2195 */
2196 void
2201 {
2211 /*
2212 * Handle the degenerate case quietly.
2213 */
2216 }
2220 /*
2221 * If there wasn't any memory allocated before, just
2222 * allocate it now and get out.
2223 */
2229 }
2231 /*
2232 * If there is already an existing if_broot, then we need
2233 * to realloc() it and shift the pointers to their new
2234 * location. The records don't change location because
2235 * they are kept butted up against the btree block header.
2236 */
2242 KM_SLEEP);
2252 }
2254 /*
2255 * rec_diff is less than 0. In this case, we are shrinking the
2256 * if_broot buffer. It must already exist. If we go to zero
2257 * records, just get rid of the root and clear the status bit.
2258 */
2265 else
2269 /*
2270 * First copy over the btree block header.
2271 */
2276 }
2278 /*
2279 * Only copy the records and pointers if there are any.
2280 */
2282 /*
2283 * First copy the records.
2284 */
2289 /*
2290 * Then copy the pointers.
2291 */
2297 }
2304 }
2307 /*
2308 * This is called when the amount of space needed for if_data
2309 * is increased or decreased. The change in size is indicated by
2310 * the number of bytes that need to be added or deleted in the
2311 * byte_diff parameter.
2312 *
2313 * If the amount of space needed has decreased below the size of the
2314 * inline buffer, then switch to using the inline buffer. Otherwise,
2315 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2316 * to what is needed.
2317 *
2318 * ip -- the inode whose if_data area is changing
2319 * byte_diff -- the change in the number of bytes, positive or negative,
2320 * requested for the if_data array.
2321 */
2322 void
2327 {
2334 }
2343 }
2347 /*
2348 * If the valid extents/data can fit in if_inline_ext/data,
2349 * copy them from the malloc'd vector and free it.
2350 */
2356 new_size);
2359 }
2362 /*
2363 * Stuck with malloc/realloc.
2364 * For inline data, the underlying buffer must be
2365 * a multiple of 4 bytes in size so that it can be
2366 * logged and stay on word boundaries. We enforce
2367 * that here.
2368 */
2374 /*
2375 * Only do the realloc if the underlying size
2376 * is really changing.
2377 */
2381 real_size,
2383 KM_SLEEP);
2384 }
2390 }
2391 }
2395 }
2397 void
2401 {
2408 }
2410 /*
2411 * If the format is local, then we can't have an extents
2412 * array so just look for an inline data array. If we're
2413 * not local then we may or may not have an extents list,
2414 * so check and free it up if we do.
2415 */
2423 }
2430 }
2437 }
2438 }
2440 /*
2441 * Increment the pin count of the given buffer.
2442 * This value is protected by ipinlock spinlock in the mount structure.
2443 */
2444 void
2447 {
2451 }
2453 /*
2454 * Decrement the pin count of the given inode, and wake up
2455 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2456 * inode must have been previously pinned with a call to xfs_ipin().
2457 */
2458 void
2461 {
2466 }
2468 /*
2469 * This is called to unpin an inode. It can be directed to wait or to return
2470 * immediately without waiting for the inode to be unpinned. The caller must
2471 * have the inode locked in at least shared mode so that the buffer cannot be
2472 * subsequently pinned once someone is waiting for it to be unpinned.
2473 */
2474 STATIC void
2478 {
2485 /* Give the log a push to start the unpinning I/O */
2490 }
2495 {
2497 }
2502 {
2504 }
2507 /*
2508 * xfs_iextents_copy()
2509 *
2510 * This is called to copy the REAL extents (as opposed to the delayed
2511 * allocation extents) from the inode into the given buffer. It
2512 * returns the number of bytes copied into the buffer.
2513 *
2514 * If there are no delayed allocation extents, then we can just
2515 * memcpy() the extents into the buffer. Otherwise, we need to
2516 * examine each extent in turn and skip those which are delayed.
2517 */
2518 int
2523 {
2528 xfs_fsblock_t start_block;
2538 /*
2539 * There are some delayed allocation extents in the
2540 * inode, so copy the extents one at a time and skip
2541 * the delayed ones. There must be at least one
2542 * non-delayed extent.
2543 */
2549 /*
2550 * It's a delayed allocation extent, so skip it.
2551 */
2553 }
2555 /* Translate to on disk format */
2558 dp++;
2559 copied++;
2560 }
2565 }
2567 /*
2568 * Each of the following cases stores data into the same region
2569 * of the on-disk inode, so only one of them can be valid at
2570 * any given time. While it is possible to have conflicting formats
2571 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2572 * in EXTENTS format, this can only happen when the fork has
2573 * changed formats after being modified but before being flushed.
2574 * In these cases, the format always takes precedence, because the
2575 * format indicates the current state of the fork.
2576 */
2577 /*ARGSUSED*/
2578 STATIC void
2585 {
2589 #ifdef XFS_TRANS_DEBUG
2591 #endif
2602 /*
2603 * This can happen if we gave up in iformat in an error path,
2604 * for the attribute fork.
2605 */
2609 }
2619 }
2633 whichfork);
2634 }
2643 XFS_BROOT_SIZE_ADJ));
2647 }
2654 }
2663 }
2669 }
2670 }
2672 STATIC int
2676 {
2701 /* really need a gang lookup range call here */
2711 /* if the inode lies outside this cluster, we're done. */
2714 /*
2715 * Do an un-protected check to see if the inode is dirty and
2716 * is a candidate for flushing. These checks will be repeated
2717 * later after the appropriate locks are acquired.
2718 */
2722 /*
2723 * Try to get locks. If any are unavailable or it is pinned,
2724 * then this inode cannot be flushed and is skipped.
2725 */
2732 }
2737 }
2739 /*
2740 * arriving here means that this inode can be flushed. First
2741 * re-check that it's dirty before flushing.
2742 */
2749 }
2750 clcount++;
2753 }
2755 }
2760 }
2762 out_free:
2768 cluster_corrupt_out:
2769 /*
2770 * Corruption detected in the clustering loop. Invalidate the
2771 * inode buffer and shut down the filesystem.
2772 */
2774 /*
2775 * Clean up the buffer. If it was B_DELWRI, just release it --
2776 * brelse can handle it with no problems. If not, shut down the
2777 * filesystem before releasing the buffer.
2778 */
2786 /*
2787 * Just like incore_relse: if we have b_iodone functions,
2788 * mark the buffer as an error and call them. Otherwise
2789 * mark it as stale and brelse.
2790 */
2800 }
2801 }
2803 /*
2804 * Unlocks the flush lock
2805 */
2809 }
2811 /*
2812 * xfs_iflush() will write a modified inode's changes out to the
2813 * inode's on disk home. The caller must have the inode lock held
2814 * in at least shared mode and the inode flush completion must be
2815 * active as well. The inode lock will still be held upon return from
2816 * the call and the caller is free to unlock it.
2817 * The inode flush will be completed when the inode reaches the disk.
2818 * The flags indicate how the inode's buffer should be written out.
2819 */
2820 int
2823 uint flags)
2824 {
2843 /*
2844 * If the inode isn't dirty, then just release the inode flush lock and
2845 * do nothing.
2846 */
2850 }
2852 /*
2853 * We can't flush the inode until it is unpinned, so wait for it if we
2854 * are allowed to block. We know noone new can pin it, because we are
2855 * holding the inode lock shared and you need to hold it exclusively to
2856 * pin the inode.
2857 *
2858 * If we are not allowed to block, force the log out asynchronously so
2859 * that when we come back the inode will be unpinned. If other inodes
2860 * in the same cluster are dirty, they will probably write the inode
2861 * out for us if they occur after the log force completes.
2862 */
2867 }
2870 /*
2871 * For stale inodes we cannot rely on the backing buffer remaining
2872 * stale in cache for the remaining life of the stale inode and so
2873 * xfs_itobp() below may give us a buffer that no longer contains
2874 * inodes below. We have to check this after ensuring the inode is
2875 * unpinned so that it is safe to reclaim the stale inode after the
2876 * flush call.
2877 */
2881 }
2883 /*
2884 * This may have been unpinned because the filesystem is shutting
2885 * down forcibly. If that's the case we must not write this inode
2886 * to disk, because the log record didn't make it to disk!
2887 */
2894 }
2896 /*
2897 * Decide how buffer will be flushed out. This is done before
2898 * the call to xfs_iflush_int because this field is zeroed by it.
2899 */
2901 /*
2902 * Flush out the inode buffer according to the directions
2903 * of the caller. In the cases where the caller has given
2904 * us a choice choose the non-delwri case. This is because
2905 * the inode is in the AIL and we need to get it out soon.
2906 */
2924 }
2943 }
2944 }
2946 /*
2947 * Get the buffer containing the on-disk inode.
2948 */
2954 }
2956 /*
2957 * First flush out the inode that xfs_iflush was called with.
2958 */
2963 /*
2964 * If the buffer is pinned then push on the log now so we won't
2965 * get stuck waiting in the write for too long.
2966 */
2970 /*
2971 * inode clustering:
2972 * see if other inodes can be gathered into this write
2973 */
2984 }
2987 corrupt_out:
2990 cluster_corrupt_out:
2991 /*
2992 * Unlocks the flush lock
2993 */
2996 }
2999 STATIC int
3003 {
3007 #ifdef XFS_TRANS_DEBUG
3009 #endif
3020 /*
3021 * If the inode isn't dirty, then just release the inode
3022 * flush lock and do nothing.
3023 */
3027 }
3029 /* set *dip = inode's place in the buffer */
3032 /*
3033 * Clear i_update_core before copying out the data.
3034 * This is for coordination with our timestamp updates
3035 * that don't hold the inode lock. They will always
3036 * update the timestamps BEFORE setting i_update_core,
3037 * so if we clear i_update_core after they set it we
3038 * are guaranteed to see their updates to the timestamps.
3039 * I believe that this depends on strongly ordered memory
3040 * semantics, but we have that. We use the SYNCHRONIZE
3041 * macro to make sure that the compiler does not reorder
3042 * the i_update_core access below the data copy below.
3043 */
3047 /*
3048 * Make sure to get the latest timestamps from the Linux inode.
3049 */
3058 }
3065 }
3075 }
3086 }
3087 }
3090 XFS_RANDOM_IFLUSH_5)) {
3096 ip);
3098 }
3105 }
3106 /*
3107 * bump the flush iteration count, used to detect flushes which
3108 * postdate a log record during recovery.
3109 */
3113 /*
3114 * Copy the dirty parts of the inode into the on-disk
3115 * inode. We always copy out the core of the inode,
3116 * because if the inode is dirty at all the core must
3117 * be.
3118 */
3121 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3125 /*
3126 * If this is really an old format inode and the superblock version
3127 * has not been updated to support only new format inodes, then
3128 * convert back to the old inode format. If the superblock version
3129 * has been updated, then make the conversion permanent.
3130 */
3134 /*
3135 * Convert it back.
3136 */
3140 /*
3141 * The superblock version has already been bumped,
3142 * so just make the conversion to the new inode
3143 * format permanent.
3144 */
3153 }
3154 }
3161 /*
3162 * We've recorded everything logged in the inode, so we'd
3163 * like to clear the ilf_fields bits so we don't log and
3164 * flush things unnecessarily. However, we can't stop
3165 * logging all this information until the data we've copied
3166 * into the disk buffer is written to disk. If we did we might
3167 * overwrite the copy of the inode in the log with all the
3168 * data after re-logging only part of it, and in the face of
3169 * a crash we wouldn't have all the data we need to recover.
3170 *
3171 * What we do is move the bits to the ili_last_fields field.
3172 * When logging the inode, these bits are moved back to the
3173 * ilf_fields field. In the xfs_iflush_done() routine we
3174 * clear ili_last_fields, since we know that the information
3175 * those bits represent is permanently on disk. As long as
3176 * the flush completes before the inode is logged again, then
3177 * both ilf_fields and ili_last_fields will be cleared.
3178 *
3179 * We can play with the ilf_fields bits here, because the inode
3180 * lock must be held exclusively in order to set bits there
3181 * and the flush lock protects the ili_last_fields bits.
3182 * Set ili_logged so the flush done
3183 * routine can tell whether or not to look in the AIL.
3184 * Also, store the current LSN of the inode so that we can tell
3185 * whether the item has moved in the AIL from xfs_iflush_done().
3186 * In order to read the lsn we need the AIL lock, because
3187 * it is a 64 bit value that cannot be read atomically.
3188 */
3197 /*
3198 * Attach the function xfs_iflush_done to the inode's
3199 * buffer. This will remove the inode from the AIL
3200 * and unlock the inode's flush lock when the inode is
3201 * completely written to disk.
3202 */
3209 /*
3210 * We're flushing an inode which is not in the AIL and has
3211 * not been logged but has i_update_core set. For this
3212 * case we can use a B_DELWRI flush and immediately drop
3213 * the inode flush lock because we can avoid the whole
3214 * AIL state thing. It's OK to drop the flush lock now,
3215 * because we've already locked the buffer and to do anything
3216 * you really need both.
3217 */
3222 }
3224 }
3228 corrupt_out:
3230 }
3232 /*
3233 * Return a pointer to the extent record at file index idx.
3234 */
3235 xfs_bmbt_rec_host_t *
3239 {
3254 }
3255 }
3257 /*
3258 * Insert new item(s) into the extent records for incore inode
3259 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3260 */
3261 void
3268 {
3278 }
3280 /*
3281 * This is called when the amount of space required for incore file
3282 * extents needs to be increased. The ext_diff parameter stores the
3283 * number of new extents being added and the idx parameter contains
3284 * the extent index where the new extents will be added. If the new
3285 * extents are being appended, then we just need to (re)allocate and
3286 * initialize the space. Otherwise, if the new extents are being
3287 * inserted into the middle of the existing entries, a bit more work
3288 * is required to make room for the new extents to be inserted. The
3289 * caller is responsible for filling in the new extent entries upon
3290 * return.
3291 */
3292 void
3297 {
3306 /*
3307 * If the new number of extents (nextents + ext_diff)
3308 * fits inside the inode, then continue to use the inline
3309 * extent buffer.
3310 */
3317 }
3321 }
3322 /*
3323 * Otherwise use a linear (direct) extent list.
3324 * If the extents are currently inside the inode,
3325 * xfs_iext_realloc_direct will switch us from
3326 * inline to direct extent allocation mode.
3327 */
3335 }
3336 }
3337 /* Indirection array */
3350 }
3351 /* Extents fit in target extent page */
3359 }
3362 }
3363 /* Insert a new extent page */
3367 }
3368 /*
3369 * If extent(s) are being appended to the last page in
3370 * the indirection array and the new extent(s) don't fit
3371 * in the page, then erp is NULL and erp_idx is set to
3372 * the next index needed in the indirection array.
3373 */
3382 erp_idx++;
3383 }
3384 }
3385 }
3386 }
3388 }
3390 /*
3391 * This is called when incore extents are being added to the indirection
3392 * array and the new extents do not fit in the target extent list. The
3393 * erp_idx parameter contains the irec index for the target extent list
3394 * in the indirection array, and the idx parameter contains the extent
3395 * index within the list. The number of extents being added is stored
3396 * in the count parameter.
3397 *
3398 * |-------| |-------|
3399 * | | | | idx - number of extents before idx
3400 * | idx | | count |
3401 * | | | | count - number of extents being inserted at idx
3402 * |-------| |-------|
3403 * | count | | nex2 | nex2 - number of extents after idx + count
3404 * |-------| |-------|
3405 */
3406 void
3412 {
3426 /*
3427 * Save second part of target extent list
3428 * (all extents past */
3436 }
3438 /*
3439 * Add the new extents to the end of the target
3440 * list, then allocate new irec record(s) and
3441 * extent buffer(s) as needed to store the rest
3442 * of the new extents.
3443 */
3450 }
3452 erp_idx++;
3458 }
3460 /* Add nex2 extents back to indirection array */
3462 xfs_extnum_t ext_avail;
3468 /*
3469 * If nex2 extents fit in the current page, append
3470 * nex2_ep after the new extents.
3471 */
3474 }
3475 /*
3476 * Otherwise, check if space is available in the
3477 * next page.
3478 */
3482 erp_idx++;
3483 erp++;
3484 /* Create a hole for nex2 extents */
3487 }
3488 /*
3489 * Final choice, create a new extent page for
3490 * nex2 extents.
3491 */
3493 erp_idx++;
3495 }
3500 }
3501 }
3503 /*
3504 * This is called when the amount of space required for incore file
3505 * extents needs to be decreased. The ext_diff parameter stores the
3506 * number of extents to be removed and the idx parameter contains
3507 * the extent index where the extents will be removed from.
3508 *
3509 * If the amount of space needed has decreased below the linear
3510 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3511 * extent array. Otherwise, use kmem_realloc() to adjust the
3512 * size to what is needed.
3513 */
3514 void
3520 {
3539 }
3541 }
3543 /*
3544 * This removes ext_diff extents from the inline buffer, beginning
3545 * at extent index idx.
3546 */
3547 void
3552 {
3571 }
3572 }
3574 /*
3575 * This removes ext_diff extents from a linear (direct) extent list,
3576 * beginning at extent index idx. If the extents are being removed
3577 * from the end of the list (ie. truncate) then we just need to re-
3578 * allocate the list to remove the extra space. Otherwise, if the
3579 * extents are being removed from the middle of the existing extent
3580 * entries, then we first need to move the extent records beginning
3581 * at idx + ext_diff up in the list to overwrite the records being
3582 * removed, then remove the extra space via kmem_realloc.
3583 */
3584 void
3589 {
3601 }
3602 /* Move extents up in the list (if needed) */
3608 }
3611 /*
3612 * Reallocate the direct extent list. If the extents
3613 * will fit inside the inode then xfs_iext_realloc_direct
3614 * will switch from direct to inline extent allocation
3615 * mode for us.
3616 */
3619 }
3621 /*
3622 * This is called when incore extents are being removed from the
3623 * indirection array and the extents being removed span multiple extent
3624 * buffers. The idx parameter contains the file extent index where we
3625 * want to begin removing extents, and the count parameter contains
3626 * how many extents need to be removed.
3627 *
3628 * |-------| |-------|
3629 * | nex1 | | | nex1 - number of extents before idx
3630 * |-------| | count |
3631 * | | | | count - number of extents being removed at idx
3632 * | count | |-------|
3633 * | | | nex2 | nex2 - number of extents after idx + count
3634 * |-------| |-------|
3635 */
3636 void
3641 {
3660 /*
3661 * Check for deletion of entire list;
3662 * xfs_iext_irec_remove() updates extent offsets.
3663 */
3670 XFS_IEXT_BUFSZ);
3676 }
3677 }
3678 /* Move extents up (if needed) */
3683 }
3684 /* Zero out rest of page */
3687 /* Update remaining counters */
3692 erp_idx++;
3693 erp++;
3694 }
3697 }
3699 /*
3700 * Create, destroy, or resize a linear (direct) block of extents.
3701 */
3702 void
3706 {
3715 /* Free extent records */
3718 }
3719 /* Resize direct extent list and zero any new bytes */
3721 /* Check if extents will fit inside the inode */
3727 }
3730 }
3734 rnew_size,
3736 }
3741 }
3742 }
3743 /*
3744 * Switch from the inline extent buffer to a direct
3745 * extent list. Be sure to include the inline extent
3746 * bytes in new_size.
3747 */
3752 }
3754 }
3757 }
3759 /*
3760 * Switch from linear (direct) extent records to inline buffer.
3761 */
3762 void
3766 {
3769 /*
3770 * The inline buffer was zeroed when we switched
3771 * from inline to direct extent allocation mode,
3772 * so we don't need to clear it here.
3773 */
3779 }
3781 /*
3782 * Switch from inline buffer to linear (direct) extent records.
3783 * new_size should already be rounded up to the next power of 2
3784 * by the caller (when appropriate), so use new_size as it is.
3785 * However, since new_size may be rounded up, we can't update
3786 * if_bytes here. It is the caller's responsibility to update
3787 * if_bytes upon return.
3788 */
3789 void
3793 {
3801 }
3803 }
3805 /*
3806 * Resize an extent indirection array to new_size bytes.
3807 */
3808 STATIC void
3812 {
3827 }
3828 }
3830 /*
3831 * Switch from indirection array to linear (direct) extent allocations.
3832 */
3833 STATIC void
3836 {
3856 }
3857 }
3859 /*
3860 * Free incore file extents.
3861 */
3862 void
3865 {
3873 }
3880 }
3884 }
3886 /*
3887 * Return a pointer to the extent record for file system block bno.
3888 */
3894 {
3909 }
3912 /* Find target extent list */
3920 }
3921 /* Binary search extent records */
3932 /* Convert back to file-based extent index */
3935 }
3938 }
3939 }
3940 /* Convert back to file-based extent index */
3943 }
3949 }
3950 }
3953 }
3955 /*
3956 * Return a pointer to the indirection array entry containing the
3957 * extent record for filesystem block bno. Store the index of the
3958 * target irec in *erp_idxp.
3959 */
3965 {
3989 }
3990 }
3993 }
3995 /*
3996 * Return a pointer to the indirection array entry containing the
3997 * extent record at file extent index *idxp. Store the index of the
3998 * target irec in *erp_idxp and store the page index of the target
3999 * extent record in *idxp.
4000 */
4001 xfs_ext_irec_t *
4007 {
4024 /* Binary search extent irec's */
4040 erp_idx++;
4046 }
4047 }
4051 }
4053 /*
4054 * Allocate and initialize an indirection array once the space needed
4055 * for incore extents increases above XFS_IEXT_BUFSZ.
4056 */
4057 void
4060 {
4076 }
4087 }
4089 /*
4090 * Allocate and initialize a new entry in the indirection array.
4091 */
4092 xfs_ext_irec_t *
4096 {
4104 /* Resize indirection array */
4107 /*
4108 * Move records down in the array so the
4109 * new page can use erp_idx.
4110 */
4114 }
4117 /* Initialize new extent record */
4126 }
4128 /*
4129 * Remove a record from the indirection array.
4130 */
4131 void
4135 {
4147 }
4148 /* Compact extent records */
4152 }
4153 /*
4154 * Manually free the last extent record from the indirection
4155 * array. A call to xfs_iext_realloc_indirect() with a size
4156 * of zero would result in a call to xfs_iext_destroy() which
4157 * would in turn call this function again, creating a nasty
4158 * infinite loop.
4159 */
4165 }
4167 }
4169 /*
4170 * This is called to clean up large amounts of unused memory allocated
4171 * by the indirection array. Before compacting anything though, verify
4172 * that the indirection array is still needed and switch back to the
4173 * linear extent list (or even the inline buffer) if possible. The
4174 * compaction policy is as follows:
4175 *
4176 * Full Compaction: Extents fit into a single page (or inline buffer)
4177 * Partial Compaction: Extents occupy less than 50% of allocated space
4178 * No Compaction: Extents occupy at least 50% of allocated space
4179 */
4180 void
4183 {
4200 }
4201 }
4203 /*
4204 * Combine extents from neighboring extent pages.
4205 */
4206 void
4209 {
4225 /*
4226 * Free page before removing extent record
4227 * so er_extoffs don't get modified in
4228 * xfs_iext_irec_remove.
4229 */
4235 erp_idx++;
4236 }
4237 }
4238 }
4240 /*
4241 * This is called to update the er_extoff field in the indirection
4242 * array when extents have been added or removed from one of the
4243 * extent lists. erp_idx contains the irec index to begin updating
4244 * at and ext_diff contains the number of extents that were added
4245 * or removed.
4246 */
4247 void
4252 {
4260 }
4261 }