| blob | be7cf625421f15a50a739af9dfc1166981a3fd06 |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include <linux/log2.h>
54 /*
55 * Used in xfs_itruncate(). This is the maximum number of extents
56 * freed from a file in a single transaction.
57 */
58 #define XFS_ITRUNC_MAX_EXTENTS 2
65 #ifdef DEBUG
66 /*
67 * Make sure that the extents in the given memory buffer
68 * are valid.
69 */
70 STATIC void
74 xfs_exntfmt_t fmt)
75 {
76 xfs_bmbt_irec_t irec;
77 xfs_bmbt_rec_host_t rec;
87 }
88 }
90 #define xfs_validate_extents(ifp, nrecs, fmt)
93 /*
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
96 */
97 #if defined(DEBUG)
98 void
102 {
114 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
115 bp);
117 }
118 }
119 }
120 #endif
122 /*
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
126 */
127 STATIC int
133 uint buf_flags,
134 uint iget_flags)
135 {
146 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
151 }
153 }
155 /*
156 * Validate the magic number and version of every inode in the buffer
157 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
158 */
159 #ifdef DEBUG
163 #endif
174 XFS_ERRTAG_ITOBP_INOTOBP,
175 XFS_RANDOM_ITOBP_INOTOBP))) {
179 }
182 #ifdef DEBUG
184 "Device %s - bad inode magic/vsn "
189 #endif
192 }
193 }
197 /*
198 * Mark the buffer as an inode buffer now that it looks good
199 */
204 }
206 /*
207 * This routine is called to map an inode number within a file
208 * system to the buffer containing the on-disk version of the
209 * inode. It returns a pointer to the buffer containing the
210 * on-disk inode in the bpp parameter, and in the dip parameter
211 * it returns a pointer to the on-disk inode within that buffer.
212 *
213 * If a non-zero error is returned, then the contents of bpp and
214 * dipp are undefined.
215 *
216 * Use xfs_imap() to determine the size and location of the
217 * buffer to read from disk.
218 */
219 int
223 xfs_ino_t ino,
227 uint imap_flags)
228 {
246 }
249 /*
250 * This routine is called to map an inode to the buffer containing
251 * the on-disk version of the inode. It returns a pointer to the
252 * buffer containing the on-disk inode in the bpp parameter, and in
253 * the dip parameter it returns a pointer to the on-disk inode within
254 * that buffer.
255 *
256 * If a non-zero error is returned, then the contents of bpp and
257 * dipp are undefined.
258 *
259 * The inode is expected to already been mapped to its buffer and read
260 * in once, thus we can use the mapping information stored in the inode
261 * rather than calling xfs_imap(). This allows us to avoid the overhead
262 * of looking at the inode btree for small block file systems
263 * (see xfs_imap()).
264 */
265 int
272 uint buf_flags)
273 {
288 }
293 }
295 /*
296 * Move inode type and inode format specific information from the
297 * on-disk inode to the in-core inode. For fifos, devs, and sockets
298 * this means set if_rdev to the proper value. For files, directories,
299 * and symlinks this means to bring in the in-line data or extent
300 * pointers. For a file in B-tree format, only the root is immediately
301 * brought in-core. The rest will be in-lined in if_extents when it
302 * is first referenced (see xfs_iread_extents()).
303 */
304 STATIC int
308 {
312 xfs_fsize_t di_size;
330 }
340 }
350 }
361 }
372 /*
373 * no local regular files yet
374 */
377 "corrupt inode %Lu "
381 XFS_ERRLEVEL_LOW,
384 }
389 "corrupt inode %Lu "
394 XFS_ERRLEVEL_LOW,
397 }
412 }
418 }
421 }
435 "corrupt inode %Lu "
440 XFS_ERRLEVEL_LOW,
443 }
456 }
461 }
463 }
465 /*
466 * The file is in-lined in the on-disk inode.
467 * If it fits into if_inline_data, then copy
468 * it there, otherwise allocate a buffer for it
469 * and copy the data there. Either way, set
470 * if_data to point at the data.
471 * If we allocate a buffer for the data, make
472 * sure that its size is a multiple of 4 and
473 * record the real size in i_real_bytes.
474 */
475 STATIC int
481 {
485 /*
486 * If the size is unreasonable, then something
487 * is wrong and we just bail out rather than crash in
488 * kmem_alloc() or memcpy() below.
489 */
492 "corrupt inode %Lu "
499 }
509 }
517 }
519 /*
520 * The file consists of a set of extents all
521 * of which fit into the on-disk inode.
522 * If there are few enough extents to fit into
523 * the if_inline_ext, then copy them there.
524 * Otherwise allocate a buffer for them and copy
525 * them into it. Either way, set if_extents
526 * to point at the extents.
527 */
528 STATIC int
533 {
544 /*
545 * If the number of extents is unreasonable, then something
546 * is wrong and we just bail out rather than crash in
547 * kmem_alloc() or memcpy() below.
548 */
556 }
563 else
574 }
581 XFS_ERRLEVEL_LOW,
584 }
585 }
588 }
590 /*
591 * The file has too many extents to fit into
592 * the inode, so they are in B-tree format.
593 * Allocate a buffer for the root of the B-tree
594 * and copy the root into it. The i_extents
595 * field will remain NULL until all of the
596 * extents are read in (when they are needed).
597 */
598 STATIC int
603 {
606 /* REFERENCED */
615 /*
616 * blow out if -- fork has less extents than can fit in
617 * fork (fork shouldn't be a btree format), root btree
618 * block has more records than can fit into the fork,
619 * or the number of extents is greater than the number of
620 * blocks.
621 */
632 }
637 /*
638 * Copy and convert from the on-disk structure
639 * to the in-memory structure.
640 */
648 }
650 STATIC void
654 {
684 }
686 void
690 {
720 }
722 STATIC uint
724 __uint16_t di_flags)
725 {
757 }
760 }
762 uint
765 {
770 }
772 uint
775 {
778 }
780 /*
781 * Read the disk inode attributes into the in-core inode structure.
782 */
783 int
788 uint iget_flags)
789 {
794 /*
795 * Fill in the location information in the in-core inode.
796 */
801 /*
802 * Get pointers to the on-disk inode and the buffer containing it.
803 */
810 /*
811 * If we got something that isn't an inode it means someone
812 * (nfs or dmi) has a stale handle.
813 */
815 #ifdef DEBUG
817 "dip->di_magic (0x%x) != "
820 XFS_DINODE_MAGIC);
824 }
826 /*
827 * If the on-disk inode is already linked to a directory
828 * entry, copy all of the inode into the in-core inode.
829 * xfs_iformat() handles copying in the inode format
830 * specific information.
831 * Otherwise, just get the truly permanent information.
832 */
837 #ifdef DEBUG
840 error);
843 }
849 /*
850 * Make sure to pull in the mode here as well in
851 * case the inode is released without being used.
852 * This ensures that xfs_inactive() will see that
853 * the inode is already free and not try to mess
854 * with the uninitialized part of it.
855 */
857 /*
858 * Initialize the per-fork minima and maxima for a new
859 * inode here. xfs_iformat will do it for old inodes.
860 */
863 }
865 /*
866 * The inode format changed when we moved the link count and
867 * made it 32 bits long. If this is an old format inode,
868 * convert it in memory to look like a new one. If it gets
869 * flushed to disk we will convert back before flushing or
870 * logging it. We zero out the new projid field and the old link
871 * count field. We'll handle clearing the pad field (the remains
872 * of the old uuid field) when we actually convert the inode to
873 * the new format. We don't change the version number so that we
874 * can distinguish this from a real new format inode.
875 */
880 }
885 /*
886 * Mark the buffer containing the inode as something to keep
887 * around for a while. This helps to keep recently accessed
888 * meta-data in-core longer.
889 */
892 /*
893 * Use xfs_trans_brelse() to release the buffer containing the
894 * on-disk inode, because it was acquired with xfs_trans_read_buf()
895 * in xfs_itobp() above. If tp is NULL, this is just a normal
896 * brelse(). If we're within a transaction, then xfs_trans_brelse()
897 * will only release the buffer if it is not dirty within the
898 * transaction. It will be OK to release the buffer in this case,
899 * because inodes on disk are never destroyed and we will be
900 * locking the new in-core inode before putting it in the hash
901 * table where other processes can find it. Thus we don't have
902 * to worry about the inode being changed just because we released
903 * the buffer.
904 */
905 out_brelse:
908 }
910 /*
911 * Read in extents from a btree-format inode.
912 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
913 */
914 int
919 {
922 xfs_extnum_t nextents;
928 }
932 /*
933 * We know that the size is valid (it's checked in iformat_btree)
934 */
944 }
947 }
949 /*
950 * Allocate an inode on disk and return a copy of its in-core version.
951 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
952 * appropriately within the inode. The uid and gid for the inode are
953 * set according to the contents of the given cred structure.
954 *
955 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
956 * has a free inode available, call xfs_iget()
957 * to obtain the in-core version of the allocated inode. Finally,
958 * fill in the inode and log its initial contents. In this case,
959 * ialloc_context would be set to NULL and call_again set to false.
960 *
961 * If xfs_dialloc() does not have an available inode,
962 * it will replenish its supply by doing an allocation. Since we can
963 * only do one allocation within a transaction without deadlocks, we
964 * must commit the current transaction before returning the inode itself.
965 * In this case, therefore, we will set call_again to true and return.
966 * The caller should then commit the current transaction, start a new
967 * transaction, and call xfs_ialloc() again to actually get the inode.
968 *
969 * To ensure that some other process does not grab the inode that
970 * was allocated during the first call to xfs_ialloc(), this routine
971 * also returns the [locked] bp pointing to the head of the freelist
972 * as ialloc_context. The caller should hold this buffer across
973 * the commit and pass it back into this routine on the second call.
974 *
975 * If we are allocating quota inodes, we do not have a parent inode
976 * to attach to or associate with (i.e. pip == NULL) because they
977 * are not linked into the directory structure - they are attached
978 * directly to the superblock - and so have no parent.
979 */
980 int
984 mode_t mode,
985 xfs_nlink_t nlink,
986 xfs_dev_t rdev,
987 prid_t prid,
992 {
993 xfs_ino_t ino;
995 uint flags;
997 timespec_t tv;
1000 /*
1001 * Call the space management code to pick
1002 * the on-disk inode to be allocated.
1003 */
1011 }
1014 /*
1015 * Get the in-core inode with the lock held exclusively.
1016 * This is because we're setting fields here we need
1017 * to prevent others from looking at until we're done.
1018 */
1034 /*
1035 * If the superblock version is up to where we support new format
1036 * inodes and this is currently an old format inode, then change
1037 * the inode version number now. This way we only do the conversion
1038 * here rather than here and in the flush/logging code.
1039 */
1043 /*
1044 * We've already zeroed the old link count, the projid field,
1045 * and the pad field.
1046 */
1047 }
1049 /*
1050 * Project ids won't be stored on disk if we are using a version 1 inode.
1051 */
1059 }
1060 }
1062 /*
1063 * If the group ID of the new file does not match the effective group
1064 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1065 * (and only if the irix_sgid_inherit compatibility variable is set).
1066 */
1071 }
1084 /*
1085 * di_gen will have been taken care of in xfs_iread.
1086 */
1103 /*
1104 * we can't set up filestreams until after the VFS inode
1105 * is set up properly.
1106 */
1109 /* fall through */
1120 }
1127 }
1128 }
1130 xfs_inherit_noatime)
1133 xfs_inherit_nodump)
1136 xfs_inherit_sync)
1139 xfs_inherit_nosymlinks)
1144 xfs_inherit_nodefrag)
1149 }
1150 /* FALLTHROUGH */
1159 }
1160 /*
1161 * Attribute fork settings for new inode.
1162 */
1166 /*
1167 * Log the new values stuffed into the inode.
1168 */
1171 /* now that we have an i_mode we can setup inode ops and unlock */
1174 /* now we have set up the vfs inode we can associate the filestream */
1181 }
1185 }
1187 /*
1188 * Check to make sure that there are no blocks allocated to the
1189 * file beyond the size of the file. We don't check this for
1190 * files with fixed size extents or real time extents, but we
1191 * at least do it for regular files.
1192 */
1193 #ifdef DEBUG
1194 void
1198 xfs_fsize_t isize)
1199 {
1200 xfs_fileoff_t map_first;
1215 /*
1216 * The filesystem could be shutting down, so bmapi may return
1217 * an error.
1218 */
1222 map_first),
1224 NULL))
1228 }
1231 /*
1232 * Calculate the last possible buffered byte in a file. This must
1233 * include data that was buffered beyond the EOF by the write code.
1234 * This also needs to deal with overflowing the xfs_fsize_t type
1235 * which can happen for sizes near the limit.
1236 *
1237 * We also need to take into account any blocks beyond the EOF. It
1238 * may be the case that they were buffered by a write which failed.
1239 * In that case the pages will still be in memory, but the inode size
1240 * will never have been updated.
1241 */
1242 STATIC xfs_fsize_t
1245 {
1247 xfs_fsize_t last_byte;
1248 xfs_fileoff_t last_block;
1249 xfs_fileoff_t size_last_block;
1255 /*
1256 * Only check for blocks beyond the EOF if the extents have
1257 * been read in. This eliminates the need for the inode lock,
1258 * and it also saves us from looking when it really isn't
1259 * necessary.
1260 */
1264 XFS_DATA_FORK);
1268 }
1271 }
1278 }
1282 }
1284 }
1286 /*
1287 * Start the truncation of the file to new_size. The new size
1288 * must be smaller than the current size. This routine will
1289 * clear the buffer and page caches of file data in the removed
1290 * range, and xfs_itruncate_finish() will remove the underlying
1291 * disk blocks.
1292 *
1293 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1294 * must NOT have the inode lock held at all. This is because we're
1295 * calling into the buffer/page cache code and we can't hold the
1296 * inode lock when we do so.
1297 *
1298 * We need to wait for any direct I/Os in flight to complete before we
1299 * proceed with the truncate. This is needed to prevent the extents
1300 * being read or written by the direct I/Os from being removed while the
1301 * I/O is in flight as there is no other method of synchronising
1302 * direct I/O with the truncate operation. Also, because we hold
1303 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1304 * started until the truncate completes and drops the lock. Essentially,
1305 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1306 * ordering between direct I/Os and the truncate operation.
1307 *
1308 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1309 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1310 * in the case that the caller is locking things out of order and
1311 * may not be able to call xfs_itruncate_finish() with the inode lock
1312 * held without dropping the I/O lock. If the caller must drop the
1313 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1314 * must be called again with all the same restrictions as the initial
1315 * call.
1316 */
1317 int
1320 uint flags,
1321 xfs_fsize_t new_size)
1322 {
1323 xfs_fsize_t last_byte;
1324 xfs_off_t toss_start;
1335 /* wait for the completion of any pending DIOs */
1339 /*
1340 * Call toss_pages or flushinval_pages to get rid of pages
1341 * overlapping the region being removed. We have to use
1342 * the less efficient flushinval_pages in the case that the
1343 * caller may not be able to finish the truncate without
1344 * dropping the inode's I/O lock. Make sure
1345 * to catch any pages brought in by buffers overlapping
1346 * the EOF by searching out beyond the isize by our
1347 * block size. We round new_size up to a block boundary
1348 * so that we don't toss things on the same block as
1349 * new_size but before it.
1350 *
1351 * Before calling toss_page or flushinval_pages, make sure to
1352 * call remapf() over the same region if the file is mapped.
1353 * This frees up mapped file references to the pages in the
1354 * given range and for the flushinval_pages case it ensures
1355 * that we get the latest mapped changes flushed out.
1356 */
1360 /*
1361 * The place to start tossing is beyond our maximum
1362 * file size, so there is no way that the data extended
1363 * out there.
1364 */
1366 }
1376 }
1377 }
1379 #ifdef DEBUG
1382 }
1383 #endif
1385 }
1387 /*
1388 * Shrink the file to the given new_size. The new size must be smaller than
1389 * the current size. This will free up the underlying blocks in the removed
1390 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1391 *
1392 * The transaction passed to this routine must have made a permanent log
1393 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1394 * given transaction and start new ones, so make sure everything involved in
1395 * the transaction is tidy before calling here. Some transaction will be
1396 * returned to the caller to be committed. The incoming transaction must
1397 * already include the inode, and both inode locks must be held exclusively.
1398 * The inode must also be "held" within the transaction. On return the inode
1399 * will be "held" within the returned transaction. This routine does NOT
1400 * require any disk space to be reserved for it within the transaction.
1401 *
1402 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1403 * indicates the fork which is to be truncated. For the attribute fork we only
1404 * support truncation to size 0.
1405 *
1406 * We use the sync parameter to indicate whether or not the first transaction
1407 * we perform might have to be synchronous. For the attr fork, it needs to be
1408 * so if the unlink of the inode is not yet known to be permanent in the log.
1409 * This keeps us from freeing and reusing the blocks of the attribute fork
1410 * before the unlink of the inode becomes permanent.
1411 *
1412 * For the data fork, we normally have to run synchronously if we're being
1413 * called out of the inactive path or we're being called out of the create path
1414 * where we're truncating an existing file. Either way, the truncate needs to
1415 * be sync so blocks don't reappear in the file with altered data in case of a
1416 * crash. wsync filesystems can run the first case async because anything that
1417 * shrinks the inode has to run sync so by the time we're called here from
1418 * inactive, the inode size is permanently set to 0.
1419 *
1420 * Calls from the truncate path always need to be sync unless we're in a wsync
1421 * filesystem and the file has already been unlinked.
1422 *
1423 * The caller is responsible for correctly setting the sync parameter. It gets
1424 * too hard for us to guess here which path we're being called out of just
1425 * based on inode state.
1426 *
1427 * If we get an error, we must return with the inode locked and linked into the
1428 * current transaction. This keeps things simple for the higher level code,
1429 * because it always knows that the inode is locked and held in the transaction
1430 * that returns to it whether errors occur or not. We don't mark the inode
1431 * dirty on error so that transactions can be easily aborted if possible.
1432 */
1433 int
1437 xfs_fsize_t new_size,
1440 {
1441 xfs_fsblock_t first_block;
1442 xfs_fileoff_t first_unmap_block;
1443 xfs_fileoff_t last_block;
1449 xfs_bmap_free_t free_list;
1465 /*
1466 * We only support truncating the entire attribute fork.
1467 */
1470 }
1474 /*
1475 * The first thing we do is set the size to new_size permanently
1476 * on disk. This way we don't have to worry about anyone ever
1477 * being able to look at the data being freed even in the face
1478 * of a crash. What we're getting around here is the case where
1479 * we free a block, it is allocated to another file, it is written
1480 * to, and then we crash. If the new data gets written to the
1481 * file but the log buffers containing the free and reallocation
1482 * don't, then we'd end up with garbage in the blocks being freed.
1483 * As long as we make the new_size permanent before actually
1484 * freeing any blocks it doesn't matter if they get writtten to.
1485 *
1486 * The callers must signal into us whether or not the size
1487 * setting here must be synchronous. There are a few cases
1488 * where it doesn't have to be synchronous. Those cases
1489 * occur if the file is unlinked and we know the unlink is
1490 * permanent or if the blocks being truncated are guaranteed
1491 * to be beyond the inode eof (regardless of the link count)
1492 * and the eof value is permanent. Both of these cases occur
1493 * only on wsync-mounted filesystems. In those cases, we're
1494 * guaranteed that no user will ever see the data in the blocks
1495 * that are being truncated so the truncate can run async.
1496 * In the free beyond eof case, the file may wind up with
1497 * more blocks allocated to it than it needs if we crash
1498 * and that won't get fixed until the next time the file
1499 * is re-opened and closed but that's ok as that shouldn't
1500 * be too many blocks.
1501 *
1502 * However, we can't just make all wsync xactions run async
1503 * because there's one call out of the create path that needs
1504 * to run sync where it's truncating an existing file to size
1505 * 0 whose size is > 0.
1506 *
1507 * It's probably possible to come up with a test in this
1508 * routine that would correctly distinguish all the above
1509 * cases from the values of the function parameters and the
1510 * inode state but for sanity's sake, I've decided to let the
1511 * layers above just tell us. It's simpler to correctly figure
1512 * out in the layer above exactly under what conditions we
1513 * can run async and I think it's easier for others read and
1514 * follow the logic in case something has to be changed.
1515 * cscope is your friend -- rcc.
1516 *
1517 * The attribute fork is much simpler.
1518 *
1519 * For the attribute fork we allow the caller to tell us whether
1520 * the unlink of the inode that led to this call is yet permanent
1521 * in the on disk log. If it is not and we will be freeing extents
1522 * in this inode then we make the first transaction synchronous
1523 * to make sure that the unlink is permanent by the time we free
1524 * the blocks.
1525 */
1528 /*
1529 * If we are not changing the file size then do
1530 * not update the on-disk file size - we may be
1531 * called from xfs_inactive_free_eofblocks(). If we
1532 * update the on-disk file size and then the system
1533 * crashes before the contents of the file are
1534 * flushed to disk then the files may be full of
1535 * holes (ie NULL files bug).
1536 */
1541 }
1542 }
1547 }
1553 /*
1554 * Since it is possible for space to become allocated beyond
1555 * the end of the file (in a crash where the space is allocated
1556 * but the inode size is not yet updated), simply remove any
1557 * blocks which show up between the new EOF and the maximum
1558 * possible file size. If the first block to be removed is
1559 * beyond the maximum file size (ie it is the same as last_block),
1560 * then there is nothing to do.
1561 */
1569 }
1571 /*
1572 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1573 * will tell us whether it freed the entire range or
1574 * not. If this is a synchronous mount (wsync),
1575 * then we can tell bunmapi to keep all the
1576 * transactions asynchronous since the unlink
1577 * transaction that made this inode inactive has
1578 * already hit the disk. There's no danger of
1579 * the freed blocks being reused, there being a
1580 * crash, and the reused blocks suddenly reappearing
1581 * in this file with garbage in them once recovery
1582 * runs.
1583 */
1588 XFS_ITRUNC_MAX_EXTENTS,
1592 /*
1593 * If the bunmapi call encounters an error,
1594 * return to the caller where the transaction
1595 * can be properly aborted. We just need to
1596 * make sure we're not holding any resources
1597 * that we were not when we came in.
1598 */
1601 }
1603 /*
1604 * Duplicate the transaction that has the permanent
1605 * reservation and commit the old transaction.
1606 */
1613 /*
1614 * If the bmap finish call encounters an error, return
1615 * to the caller where the transaction can be properly
1616 * aborted. We just need to make sure we're not
1617 * holding any resources that we were not when we came
1618 * in.
1619 *
1620 * Aborting from this point might lose some blocks in
1621 * the file system, but oh well.
1622 */
1625 }
1628 /*
1629 * Mark the inode dirty so it will be logged and
1630 * moved forward in the log as part of every commit.
1631 */
1633 }
1643 /*
1644 * transaction commit worked ok so we can drop the extra ticket
1645 * reference that we gained in xfs_trans_dup()
1646 */
1650 XFS_TRANS_PERM_LOG_RES,
1651 XFS_ITRUNCATE_LOG_COUNT);
1654 }
1655 /*
1656 * Only update the size in the case of the data fork, but
1657 * always re-log the inode so that our permanent transaction
1658 * can keep on rolling it forward in the log.
1659 */
1662 /*
1663 * If we are not changing the file size then do
1664 * not update the on-disk file size - we may be
1665 * called from xfs_inactive_free_eofblocks(). If we
1666 * update the on-disk file size and then the system
1667 * crashes before the contents of the file are
1668 * flushed to disk then the files may be full of
1669 * holes (ie NULL files bug).
1670 */
1674 }
1675 }
1685 }
1687 /*
1688 * This is called when the inode's link count goes to 0.
1689 * We place the on-disk inode on a list in the AGI. It
1690 * will be pulled from this list when the inode is freed.
1691 */
1692 int
1696 {
1702 xfs_agino_t agino;
1713 /*
1714 * Get the agi buffer first. It ensures lock ordering
1715 * on the list.
1716 */
1722 /*
1723 * Get the index into the agi hash table for the
1724 * list this inode will go on.
1725 */
1733 /*
1734 * There is already another inode in the bucket we need
1735 * to add ourselves to. Add us at the front of the list.
1736 * Here we put the head pointer into our next pointer,
1737 * and then we fall through to point the head at us.
1738 */
1744 /* both on-disk, don't endian flip twice */
1752 }
1754 /*
1755 * Point the bucket head pointer at the inode being inserted.
1756 */
1764 }
1766 /*
1767 * Pull the on-disk inode from the AGI unlinked list.
1768 */
1769 STATIC int
1773 {
1774 xfs_ino_t next_ino;
1780 xfs_agnumber_t agno;
1781 xfs_agino_t agino;
1782 xfs_agino_t next_agino;
1792 /*
1793 * Get the agi buffer first. It ensures lock ordering
1794 * on the list.
1795 */
1802 /*
1803 * Get the index into the agi hash table for the
1804 * list this inode will go on.
1805 */
1813 /*
1814 * We're at the head of the list. Get the inode's
1815 * on-disk buffer to see if there is anyone after us
1816 * on the list. Only modify our next pointer if it
1817 * is not already NULLAGINO. This saves us the overhead
1818 * of dealing with the buffer when there is no need to
1819 * change it.
1820 */
1827 }
1840 }
1841 /*
1842 * Point the bucket head pointer at the next inode.
1843 */
1852 /*
1853 * We need to search the list for the inode being freed.
1854 */
1858 /*
1859 * If the last inode wasn't the one pointing to
1860 * us, then release its buffer since we're not
1861 * going to do anything with it.
1862 */
1865 }
1874 }
1878 }
1879 /*
1880 * Now last_ibp points to the buffer previous to us on
1881 * the unlinked list. Pull us from the list.
1882 */
1889 }
1903 }
1904 /*
1905 * Point the previous inode on the list to the next inode.
1906 */
1914 }
1916 }
1918 /*
1919 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1920 * inodes that are in memory - they all must be marked stale and attached to
1921 * the cluster buffer.
1922 */
1923 STATIC void
1927 xfs_ino_t inum)
1928 {
1934 xfs_daddr_t blkno;
1951 }
1957 /*
1958 * We obtain and lock the backing buffer first in the process
1959 * here, as we have to ensure that any dirty inode that we
1960 * can't get the flush lock on is attached to the buffer.
1961 * If we scan the in-memory inodes first, then buffer IO can
1962 * complete before we get a lock on it, and hence we may fail
1963 * to mark all the active inodes on the buffer stale.
1964 */
1967 XBF_LOCK);
1969 /*
1970 * Walk the inodes already attached to the buffer and mark them
1971 * stale. These will all have the flush locks held, so an
1972 * in-memory inode walk can't lock them. By marking them all
1973 * stale first, we will not attempt to lock them in the loop
1974 * below as the XFS_ISTALE flag will be set.
1975 */
1986 }
1988 }
1991 /*
1992 * For each inode in memory attempt to add it to the inode
1993 * buffer and set it up for being staled on buffer IO
1994 * completion. This is safe as we've locked out tail pushing
1995 * and flushing by locking the buffer.
1996 *
1997 * We have already marked every inode that was part of a
1998 * transaction stale above, which means there is no point in
1999 * even trying to lock them.
2000 */
2002 retry:
2007 /* Inode not in memory, nothing to do */
2011 }
2013 /*
2014 * because this is an RCU protected lookup, we could
2015 * find a recently freed or even reallocated inode
2016 * during the lookup. We need to check under the
2017 * i_flags_lock for a valid inode here. Skip it if it
2018 * is not valid, the wrong inode or stale.
2019 */
2026 }
2029 /*
2030 * Don't try to lock/unlock the current inode, but we
2031 * _cannot_ skip the other inodes that we did not find
2032 * in the list attached to the buffer and are not
2033 * already marked stale. If we can't lock it, back off
2034 * and retry.
2035 */
2041 }
2047 /*
2048 * we don't need to attach clean inodes or those only
2049 * with unlogged changes (which we throw away, anyway).
2050 */
2058 }
2071 }
2075 }
2078 }
2080 /*
2081 * This is called to return an inode to the inode free list.
2082 * The inode should already be truncated to 0 length and have
2083 * no pages associated with it. This routine also assumes that
2084 * the inode is already a part of the transaction.
2085 *
2086 * The on-disk copy of the inode will have been added to the list
2087 * of unlinked inodes in the AGI. We need to remove the inode from
2088 * that list atomically with respect to freeing it here.
2089 */
2090 int
2095 {
2098 xfs_ino_t first_ino;
2111 /*
2112 * Pull the on-disk inode from the AGI unlinked list.
2113 */
2117 }
2122 }
2131 /*
2132 * Bump the generation count so no one will be confused
2133 * by reincarnations of this inode.
2134 */
2143 /*
2144 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2145 * from picking up this inode when it is reclaimed (its incore state
2146 * initialzed but not flushed to disk yet). The in-core di_mode is
2147 * already cleared and a corresponding transaction logged.
2148 * The hack here just synchronizes the in-core to on-disk
2149 * di_mode value in advance before the actual inode sync to disk.
2150 * This is OK because the inode is already unlinked and would never
2151 * change its di_mode again for this inode generation.
2152 * This is a temporary hack that would require a proper fix
2153 * in the future.
2154 */
2159 }
2162 }
2164 /*
2165 * Reallocate the space for if_broot based on the number of records
2166 * being added or deleted as indicated in rec_diff. Move the records
2167 * and pointers in if_broot to fit the new size. When shrinking this
2168 * will eliminate holes between the records and pointers created by
2169 * the caller. When growing this will create holes to be filled in
2170 * by the caller.
2171 *
2172 * The caller must not request to add more records than would fit in
2173 * the on-disk inode root. If the if_broot is currently NULL, then
2174 * if we adding records one will be allocated. The caller must also
2175 * not request that the number of records go below zero, although
2176 * it can go to zero.
2177 *
2178 * ip -- the inode whose if_broot area is changing
2179 * ext_diff -- the change in the number of records, positive or negative,
2180 * requested for the if_broot array.
2181 */
2182 void
2187 {
2197 /*
2198 * Handle the degenerate case quietly.
2199 */
2202 }
2206 /*
2207 * If there wasn't any memory allocated before, just
2208 * allocate it now and get out.
2209 */
2215 }
2217 /*
2218 * If there is already an existing if_broot, then we need
2219 * to realloc() it and shift the pointers to their new
2220 * location. The records don't change location because
2221 * they are kept butted up against the btree block header.
2222 */
2238 }
2240 /*
2241 * rec_diff is less than 0. In this case, we are shrinking the
2242 * if_broot buffer. It must already exist. If we go to zero
2243 * records, just get rid of the root and clear the status bit.
2244 */
2251 else
2255 /*
2256 * First copy over the btree block header.
2257 */
2262 }
2264 /*
2265 * Only copy the records and pointers if there are any.
2266 */
2268 /*
2269 * First copy the records.
2270 */
2275 /*
2276 * Then copy the pointers.
2277 */
2283 }
2290 }
2293 /*
2294 * This is called when the amount of space needed for if_data
2295 * is increased or decreased. The change in size is indicated by
2296 * the number of bytes that need to be added or deleted in the
2297 * byte_diff parameter.
2298 *
2299 * If the amount of space needed has decreased below the size of the
2300 * inline buffer, then switch to using the inline buffer. Otherwise,
2301 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2302 * to what is needed.
2303 *
2304 * ip -- the inode whose if_data area is changing
2305 * byte_diff -- the change in the number of bytes, positive or negative,
2306 * requested for the if_data array.
2307 */
2308 void
2313 {
2320 }
2329 }
2333 /*
2334 * If the valid extents/data can fit in if_inline_ext/data,
2335 * copy them from the malloc'd vector and free it.
2336 */
2342 new_size);
2345 }
2348 /*
2349 * Stuck with malloc/realloc.
2350 * For inline data, the underlying buffer must be
2351 * a multiple of 4 bytes in size so that it can be
2352 * logged and stay on word boundaries. We enforce
2353 * that here.
2354 */
2361 /*
2362 * Only do the realloc if the underlying size
2363 * is really changing.
2364 */
2368 real_size,
2371 }
2378 }
2379 }
2383 }
2385 void
2389 {
2396 }
2398 /*
2399 * If the format is local, then we can't have an extents
2400 * array so just look for an inline data array. If we're
2401 * not local then we may or may not have an extents list,
2402 * so check and free it up if we do.
2403 */
2411 }
2418 }
2425 }
2426 }
2428 /*
2429 * This is called to unpin an inode. The caller must have the inode locked
2430 * in at least shared mode so that the buffer cannot be subsequently pinned
2431 * once someone is waiting for it to be unpinned.
2432 */
2433 static void
2436 {
2441 /* Give the log a push to start the unpinning I/O */
2444 }
2446 void
2449 {
2453 }
2454 }
2456 /*
2457 * xfs_iextents_copy()
2458 *
2459 * This is called to copy the REAL extents (as opposed to the delayed
2460 * allocation extents) from the inode into the given buffer. It
2461 * returns the number of bytes copied into the buffer.
2462 *
2463 * If there are no delayed allocation extents, then we can just
2464 * memcpy() the extents into the buffer. Otherwise, we need to
2465 * examine each extent in turn and skip those which are delayed.
2466 */
2467 int
2472 {
2477 xfs_fsblock_t start_block;
2487 /*
2488 * There are some delayed allocation extents in the
2489 * inode, so copy the extents one at a time and skip
2490 * the delayed ones. There must be at least one
2491 * non-delayed extent.
2492 */
2498 /*
2499 * It's a delayed allocation extent, so skip it.
2500 */
2502 }
2504 /* Translate to on disk format */
2507 dp++;
2508 copied++;
2509 }
2514 }
2516 /*
2517 * Each of the following cases stores data into the same region
2518 * of the on-disk inode, so only one of them can be valid at
2519 * any given time. While it is possible to have conflicting formats
2520 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2521 * in EXTENTS format, this can only happen when the fork has
2522 * changed formats after being modified but before being flushed.
2523 * In these cases, the format always takes precedence, because the
2524 * format indicates the current state of the fork.
2525 */
2526 /*ARGSUSED*/
2527 STATIC void
2534 {
2538 #ifdef XFS_TRANS_DEBUG
2540 #endif
2551 /*
2552 * This can happen if we gave up in iformat in an error path,
2553 * for the attribute fork.
2554 */
2558 }
2568 }
2582 whichfork);
2583 }
2592 XFS_BROOT_SIZE_ADJ));
2596 }
2603 }
2612 }
2618 }
2619 }
2621 STATIC int
2625 {
2649 /* really need a gang lookup range call here */
2660 /*
2661 * because this is an RCU protected lookup, we could find a
2662 * recently freed or even reallocated inode during the lookup.
2663 * We need to check under the i_flags_lock for a valid inode
2664 * here. Skip it if it is not valid or the wrong inode.
2665 */
2671 }
2674 /*
2675 * Do an un-protected check to see if the inode is dirty and
2676 * is a candidate for flushing. These checks will be repeated
2677 * later after the appropriate locks are acquired.
2678 */
2682 /*
2683 * Try to get locks. If any are unavailable or it is pinned,
2684 * then this inode cannot be flushed and is skipped.
2685 */
2692 }
2697 }
2699 /*
2700 * arriving here means that this inode can be flushed. First
2701 * re-check that it's dirty before flushing.
2702 */
2709 }
2710 clcount++;
2713 }
2715 }
2720 }
2722 out_free:
2725 out_put:
2730 cluster_corrupt_out:
2731 /*
2732 * Corruption detected in the clustering loop. Invalidate the
2733 * inode buffer and shut down the filesystem.
2734 */
2736 /*
2737 * Clean up the buffer. If it was B_DELWRI, just release it --
2738 * brelse can handle it with no problems. If not, shut down the
2739 * filesystem before releasing the buffer.
2740 */
2748 /*
2749 * Just like incore_relse: if we have b_iodone functions,
2750 * mark the buffer as an error and call them. Otherwise
2751 * mark it as stale and brelse.
2752 */
2761 }
2762 }
2764 /*
2765 * Unlocks the flush lock
2766 */
2771 }
2773 /*
2774 * xfs_iflush() will write a modified inode's changes out to the
2775 * inode's on disk home. The caller must have the inode lock held
2776 * in at least shared mode and the inode flush completion must be
2777 * active as well. The inode lock will still be held upon return from
2778 * the call and the caller is free to unlock it.
2779 * The inode flush will be completed when the inode reaches the disk.
2780 * The flags indicate how the inode's buffer should be written out.
2781 */
2782 int
2785 uint flags)
2786 {
2803 /*
2804 * We can't flush the inode until it is unpinned, so wait for it if we
2805 * are allowed to block. We know noone new can pin it, because we are
2806 * holding the inode lock shared and you need to hold it exclusively to
2807 * pin the inode.
2808 *
2809 * If we are not allowed to block, force the log out asynchronously so
2810 * that when we come back the inode will be unpinned. If other inodes
2811 * in the same cluster are dirty, they will probably write the inode
2812 * out for us if they occur after the log force completes.
2813 */
2818 }
2821 /*
2822 * For stale inodes we cannot rely on the backing buffer remaining
2823 * stale in cache for the remaining life of the stale inode and so
2824 * xfs_itobp() below may give us a buffer that no longer contains
2825 * inodes below. We have to check this after ensuring the inode is
2826 * unpinned so that it is safe to reclaim the stale inode after the
2827 * flush call.
2828 */
2832 }
2834 /*
2835 * This may have been unpinned because the filesystem is shutting
2836 * down forcibly. If that's the case we must not write this inode
2837 * to disk, because the log record didn't make it to disk!
2838 */
2845 }
2847 /*
2848 * Get the buffer containing the on-disk inode.
2849 */
2855 }
2857 /*
2858 * First flush out the inode that xfs_iflush was called with.
2859 */
2864 /*
2865 * If the buffer is pinned then push on the log now so we won't
2866 * get stuck waiting in the write for too long.
2867 */
2871 /*
2872 * inode clustering:
2873 * see if other inodes can be gathered into this write
2874 */
2881 else
2885 corrupt_out:
2888 cluster_corrupt_out:
2889 /*
2890 * Unlocks the flush lock
2891 */
2894 }
2897 STATIC int
2901 {
2905 #ifdef XFS_TRANS_DEBUG
2907 #endif
2917 /* set *dip = inode's place in the buffer */
2920 /*
2921 * Clear i_update_core before copying out the data.
2922 * This is for coordination with our timestamp updates
2923 * that don't hold the inode lock. They will always
2924 * update the timestamps BEFORE setting i_update_core,
2925 * so if we clear i_update_core after they set it we
2926 * are guaranteed to see their updates to the timestamps.
2927 * I believe that this depends on strongly ordered memory
2928 * semantics, but we have that. We use the SYNCHRONIZE
2929 * macro to make sure that the compiler does not reorder
2930 * the i_update_core access below the data copy below.
2931 */
2935 /*
2936 * Make sure to get the latest timestamps from the Linux inode.
2937 */
2946 }
2953 }
2963 }
2974 }
2975 }
2978 XFS_RANDOM_IFLUSH_5)) {
2984 ip);
2986 }
2993 }
2994 /*
2995 * bump the flush iteration count, used to detect flushes which
2996 * postdate a log record during recovery.
2997 */
3001 /*
3002 * Copy the dirty parts of the inode into the on-disk
3003 * inode. We always copy out the core of the inode,
3004 * because if the inode is dirty at all the core must
3005 * be.
3006 */
3009 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3013 /*
3014 * If this is really an old format inode and the superblock version
3015 * has not been updated to support only new format inodes, then
3016 * convert back to the old inode format. If the superblock version
3017 * has been updated, then make the conversion permanent.
3018 */
3022 /*
3023 * Convert it back.
3024 */
3028 /*
3029 * The superblock version has already been bumped,
3030 * so just make the conversion to the new inode
3031 * format permanent.
3032 */
3041 }
3042 }
3049 /*
3050 * We've recorded everything logged in the inode, so we'd
3051 * like to clear the ilf_fields bits so we don't log and
3052 * flush things unnecessarily. However, we can't stop
3053 * logging all this information until the data we've copied
3054 * into the disk buffer is written to disk. If we did we might
3055 * overwrite the copy of the inode in the log with all the
3056 * data after re-logging only part of it, and in the face of
3057 * a crash we wouldn't have all the data we need to recover.
3058 *
3059 * What we do is move the bits to the ili_last_fields field.
3060 * When logging the inode, these bits are moved back to the
3061 * ilf_fields field. In the xfs_iflush_done() routine we
3062 * clear ili_last_fields, since we know that the information
3063 * those bits represent is permanently on disk. As long as
3064 * the flush completes before the inode is logged again, then
3065 * both ilf_fields and ili_last_fields will be cleared.
3066 *
3067 * We can play with the ilf_fields bits here, because the inode
3068 * lock must be held exclusively in order to set bits there
3069 * and the flush lock protects the ili_last_fields bits.
3070 * Set ili_logged so the flush done
3071 * routine can tell whether or not to look in the AIL.
3072 * Also, store the current LSN of the inode so that we can tell
3073 * whether the item has moved in the AIL from xfs_iflush_done().
3074 * In order to read the lsn we need the AIL lock, because
3075 * it is a 64 bit value that cannot be read atomically.
3076 */
3085 /*
3086 * Attach the function xfs_iflush_done to the inode's
3087 * buffer. This will remove the inode from the AIL
3088 * and unlock the inode's flush lock when the inode is
3089 * completely written to disk.
3090 */
3096 /*
3097 * We're flushing an inode which is not in the AIL and has
3098 * not been logged but has i_update_core set. For this
3099 * case we can use a B_DELWRI flush and immediately drop
3100 * the inode flush lock because we can avoid the whole
3101 * AIL state thing. It's OK to drop the flush lock now,
3102 * because we've already locked the buffer and to do anything
3103 * you really need both.
3104 */
3109 }
3111 }
3115 corrupt_out:
3117 }
3119 /*
3120 * Return a pointer to the extent record at file index idx.
3121 */
3122 xfs_bmbt_rec_host_t *
3126 {
3141 }
3142 }
3144 /*
3145 * Insert new item(s) into the extent records for incore inode
3146 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3147 */
3148 void
3155 {
3165 }
3167 /*
3168 * This is called when the amount of space required for incore file
3169 * extents needs to be increased. The ext_diff parameter stores the
3170 * number of new extents being added and the idx parameter contains
3171 * the extent index where the new extents will be added. If the new
3172 * extents are being appended, then we just need to (re)allocate and
3173 * initialize the space. Otherwise, if the new extents are being
3174 * inserted into the middle of the existing entries, a bit more work
3175 * is required to make room for the new extents to be inserted. The
3176 * caller is responsible for filling in the new extent entries upon
3177 * return.
3178 */
3179 void
3184 {
3193 /*
3194 * If the new number of extents (nextents + ext_diff)
3195 * fits inside the inode, then continue to use the inline
3196 * extent buffer.
3197 */
3204 }
3208 }
3209 /*
3210 * Otherwise use a linear (direct) extent list.
3211 * If the extents are currently inside the inode,
3212 * xfs_iext_realloc_direct will switch us from
3213 * inline to direct extent allocation mode.
3214 */
3222 }
3223 }
3224 /* Indirection array */
3237 }
3238 /* Extents fit in target extent page */
3246 }
3249 }
3250 /* Insert a new extent page */
3254 }
3255 /*
3256 * If extent(s) are being appended to the last page in
3257 * the indirection array and the new extent(s) don't fit
3258 * in the page, then erp is NULL and erp_idx is set to
3259 * the next index needed in the indirection array.
3260 */
3269 erp_idx++;
3270 }
3271 }
3272 }
3273 }
3275 }
3277 /*
3278 * This is called when incore extents are being added to the indirection
3279 * array and the new extents do not fit in the target extent list. The
3280 * erp_idx parameter contains the irec index for the target extent list
3281 * in the indirection array, and the idx parameter contains the extent
3282 * index within the list. The number of extents being added is stored
3283 * in the count parameter.
3284 *
3285 * |-------| |-------|
3286 * | | | | idx - number of extents before idx
3287 * | idx | | count |
3288 * | | | | count - number of extents being inserted at idx
3289 * |-------| |-------|
3290 * | count | | nex2 | nex2 - number of extents after idx + count
3291 * |-------| |-------|
3292 */
3293 void
3299 {
3313 /*
3314 * Save second part of target extent list
3315 * (all extents past */
3323 }
3325 /*
3326 * Add the new extents to the end of the target
3327 * list, then allocate new irec record(s) and
3328 * extent buffer(s) as needed to store the rest
3329 * of the new extents.
3330 */
3337 }
3339 erp_idx++;
3345 }
3347 /* Add nex2 extents back to indirection array */
3349 xfs_extnum_t ext_avail;
3355 /*
3356 * If nex2 extents fit in the current page, append
3357 * nex2_ep after the new extents.
3358 */
3361 }
3362 /*
3363 * Otherwise, check if space is available in the
3364 * next page.
3365 */
3369 erp_idx++;
3370 erp++;
3371 /* Create a hole for nex2 extents */
3374 }
3375 /*
3376 * Final choice, create a new extent page for
3377 * nex2 extents.
3378 */
3380 erp_idx++;
3382 }
3387 }
3388 }
3390 /*
3391 * This is called when the amount of space required for incore file
3392 * extents needs to be decreased. The ext_diff parameter stores the
3393 * number of extents to be removed and the idx parameter contains
3394 * the extent index where the extents will be removed from.
3395 *
3396 * If the amount of space needed has decreased below the linear
3397 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3398 * extent array. Otherwise, use kmem_realloc() to adjust the
3399 * size to what is needed.
3400 */
3401 void
3407 {
3426 }
3428 }
3430 /*
3431 * This removes ext_diff extents from the inline buffer, beginning
3432 * at extent index idx.
3433 */
3434 void
3439 {
3458 }
3459 }
3461 /*
3462 * This removes ext_diff extents from a linear (direct) extent list,
3463 * beginning at extent index idx. If the extents are being removed
3464 * from the end of the list (ie. truncate) then we just need to re-
3465 * allocate the list to remove the extra space. Otherwise, if the
3466 * extents are being removed from the middle of the existing extent
3467 * entries, then we first need to move the extent records beginning
3468 * at idx + ext_diff up in the list to overwrite the records being
3469 * removed, then remove the extra space via kmem_realloc.
3470 */
3471 void
3476 {
3488 }
3489 /* Move extents up in the list (if needed) */
3495 }
3498 /*
3499 * Reallocate the direct extent list. If the extents
3500 * will fit inside the inode then xfs_iext_realloc_direct
3501 * will switch from direct to inline extent allocation
3502 * mode for us.
3503 */
3506 }
3508 /*
3509 * This is called when incore extents are being removed from the
3510 * indirection array and the extents being removed span multiple extent
3511 * buffers. The idx parameter contains the file extent index where we
3512 * want to begin removing extents, and the count parameter contains
3513 * how many extents need to be removed.
3514 *
3515 * |-------| |-------|
3516 * | nex1 | | | nex1 - number of extents before idx
3517 * |-------| | count |
3518 * | | | | count - number of extents being removed at idx
3519 * | count | |-------|
3520 * | | | nex2 | nex2 - number of extents after idx + count
3521 * |-------| |-------|
3522 */
3523 void
3528 {
3545 /*
3546 * Check for deletion of entire list;
3547 * xfs_iext_irec_remove() updates extent offsets.
3548 */
3555 XFS_IEXT_BUFSZ);
3561 }
3562 }
3563 /* Move extents up (if needed) */
3568 }
3569 /* Zero out rest of page */
3572 /* Update remaining counters */
3577 erp_idx++;
3578 erp++;
3579 }
3582 }
3584 /*
3585 * Create, destroy, or resize a linear (direct) block of extents.
3586 */
3587 void
3591 {
3600 /* Free extent records */
3603 }
3604 /* Resize direct extent list and zero any new bytes */
3606 /* Check if extents will fit inside the inode */
3612 }
3615 }
3619 rnew_size,
3621 }
3626 }
3627 }
3628 /*
3629 * Switch from the inline extent buffer to a direct
3630 * extent list. Be sure to include the inline extent
3631 * bytes in new_size.
3632 */
3637 }
3639 }
3642 }
3644 /*
3645 * Switch from linear (direct) extent records to inline buffer.
3646 */
3647 void
3651 {
3654 /*
3655 * The inline buffer was zeroed when we switched
3656 * from inline to direct extent allocation mode,
3657 * so we don't need to clear it here.
3658 */
3664 }
3666 /*
3667 * Switch from inline buffer to linear (direct) extent records.
3668 * new_size should already be rounded up to the next power of 2
3669 * by the caller (when appropriate), so use new_size as it is.
3670 * However, since new_size may be rounded up, we can't update
3671 * if_bytes here. It is the caller's responsibility to update
3672 * if_bytes upon return.
3673 */
3674 void
3678 {
3686 }
3688 }
3690 /*
3691 * Resize an extent indirection array to new_size bytes.
3692 */
3693 STATIC void
3697 {
3712 }
3713 }
3715 /*
3716 * Switch from indirection array to linear (direct) extent allocations.
3717 */
3718 STATIC void
3721 {
3741 }
3742 }
3744 /*
3745 * Free incore file extents.
3746 */
3747 void
3750 {
3758 }
3765 }
3769 }
3771 /*
3772 * Return a pointer to the extent record for file system block bno.
3773 */
3779 {
3794 }
3797 /* Find target extent list */
3805 }
3806 /* Binary search extent records */
3817 /* Convert back to file-based extent index */
3820 }
3823 }
3824 }
3825 /* Convert back to file-based extent index */
3828 }
3834 }
3835 }
3838 }
3840 /*
3841 * Return a pointer to the indirection array entry containing the
3842 * extent record for filesystem block bno. Store the index of the
3843 * target irec in *erp_idxp.
3844 */
3850 {
3874 }
3875 }
3878 }
3880 /*
3881 * Return a pointer to the indirection array entry containing the
3882 * extent record at file extent index *idxp. Store the index of the
3883 * target irec in *erp_idxp and store the page index of the target
3884 * extent record in *idxp.
3885 */
3886 xfs_ext_irec_t *
3892 {
3909 /* Binary search extent irec's */
3925 erp_idx++;
3931 }
3932 }
3936 }
3938 /*
3939 * Allocate and initialize an indirection array once the space needed
3940 * for incore extents increases above XFS_IEXT_BUFSZ.
3941 */
3942 void
3945 {
3961 }
3972 }
3974 /*
3975 * Allocate and initialize a new entry in the indirection array.
3976 */
3977 xfs_ext_irec_t *
3981 {
3989 /* Resize indirection array */
3992 /*
3993 * Move records down in the array so the
3994 * new page can use erp_idx.
3995 */
3999 }
4002 /* Initialize new extent record */
4011 }
4013 /*
4014 * Remove a record from the indirection array.
4015 */
4016 void
4020 {
4032 }
4033 /* Compact extent records */
4037 }
4038 /*
4039 * Manually free the last extent record from the indirection
4040 * array. A call to xfs_iext_realloc_indirect() with a size
4041 * of zero would result in a call to xfs_iext_destroy() which
4042 * would in turn call this function again, creating a nasty
4043 * infinite loop.
4044 */
4050 }
4052 }
4054 /*
4055 * This is called to clean up large amounts of unused memory allocated
4056 * by the indirection array. Before compacting anything though, verify
4057 * that the indirection array is still needed and switch back to the
4058 * linear extent list (or even the inline buffer) if possible. The
4059 * compaction policy is as follows:
4060 *
4061 * Full Compaction: Extents fit into a single page (or inline buffer)
4062 * Partial Compaction: Extents occupy less than 50% of allocated space
4063 * No Compaction: Extents occupy at least 50% of allocated space
4064 */
4065 void
4068 {
4085 }
4086 }
4088 /*
4089 * Combine extents from neighboring extent pages.
4090 */
4091 void
4094 {
4110 /*
4111 * Free page before removing extent record
4112 * so er_extoffs don't get modified in
4113 * xfs_iext_irec_remove.
4114 */
4120 erp_idx++;
4121 }
4122 }
4123 }
4125 /*
4126 * This is called to update the er_extoff field in the indirection
4127 * array when extents have been added or removed from one of the
4128 * extent lists. erp_idx contains the irec index to begin updating
4129 * at and ext_diff contains the number of extents that were added
4130 * or removed.
4131 */
4132 void
4137 {
4145 }
4146 }