| blob | 34798f391c49349018f04a47d625c6aafa035bea |
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 {
683 }
685 void
689 {
718 }
720 STATIC uint
722 __uint16_t di_flags)
723 {
755 }
758 }
760 uint
763 {
768 }
770 uint
773 {
776 }
778 /*
779 * Read the disk inode attributes into the in-core inode structure.
780 */
781 int
786 uint iget_flags)
787 {
792 /*
793 * Fill in the location information in the in-core inode.
794 */
799 /*
800 * Get pointers to the on-disk inode and the buffer containing it.
801 */
808 /*
809 * If we got something that isn't an inode it means someone
810 * (nfs or dmi) has a stale handle.
811 */
813 #ifdef DEBUG
815 "dip->di_magic (0x%x) != "
818 XFS_DINODE_MAGIC);
822 }
824 /*
825 * If the on-disk inode is already linked to a directory
826 * entry, copy all of the inode into the in-core inode.
827 * xfs_iformat() handles copying in the inode format
828 * specific information.
829 * Otherwise, just get the truly permanent information.
830 */
835 #ifdef DEBUG
838 error);
841 }
847 /*
848 * Make sure to pull in the mode here as well in
849 * case the inode is released without being used.
850 * This ensures that xfs_inactive() will see that
851 * the inode is already free and not try to mess
852 * with the uninitialized part of it.
853 */
855 /*
856 * Initialize the per-fork minima and maxima for a new
857 * inode here. xfs_iformat will do it for old inodes.
858 */
861 }
863 /*
864 * The inode format changed when we moved the link count and
865 * made it 32 bits long. If this is an old format inode,
866 * convert it in memory to look like a new one. If it gets
867 * flushed to disk we will convert back before flushing or
868 * logging it. We zero out the new projid field and the old link
869 * count field. We'll handle clearing the pad field (the remains
870 * of the old uuid field) when we actually convert the inode to
871 * the new format. We don't change the version number so that we
872 * can distinguish this from a real new format inode.
873 */
878 }
883 /*
884 * Mark the buffer containing the inode as something to keep
885 * around for a while. This helps to keep recently accessed
886 * meta-data in-core longer.
887 */
890 /*
891 * Use xfs_trans_brelse() to release the buffer containing the
892 * on-disk inode, because it was acquired with xfs_trans_read_buf()
893 * in xfs_itobp() above. If tp is NULL, this is just a normal
894 * brelse(). If we're within a transaction, then xfs_trans_brelse()
895 * will only release the buffer if it is not dirty within the
896 * transaction. It will be OK to release the buffer in this case,
897 * because inodes on disk are never destroyed and we will be
898 * locking the new in-core inode before putting it in the hash
899 * table where other processes can find it. Thus we don't have
900 * to worry about the inode being changed just because we released
901 * the buffer.
902 */
903 out_brelse:
906 }
908 /*
909 * Read in extents from a btree-format inode.
910 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
911 */
912 int
917 {
920 xfs_extnum_t nextents;
926 }
930 /*
931 * We know that the size is valid (it's checked in iformat_btree)
932 */
942 }
945 }
947 /*
948 * Allocate an inode on disk and return a copy of its in-core version.
949 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
950 * appropriately within the inode. The uid and gid for the inode are
951 * set according to the contents of the given cred structure.
952 *
953 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
954 * has a free inode available, call xfs_iget()
955 * to obtain the in-core version of the allocated inode. Finally,
956 * fill in the inode and log its initial contents. In this case,
957 * ialloc_context would be set to NULL and call_again set to false.
958 *
959 * If xfs_dialloc() does not have an available inode,
960 * it will replenish its supply by doing an allocation. Since we can
961 * only do one allocation within a transaction without deadlocks, we
962 * must commit the current transaction before returning the inode itself.
963 * In this case, therefore, we will set call_again to true and return.
964 * The caller should then commit the current transaction, start a new
965 * transaction, and call xfs_ialloc() again to actually get the inode.
966 *
967 * To ensure that some other process does not grab the inode that
968 * was allocated during the first call to xfs_ialloc(), this routine
969 * also returns the [locked] bp pointing to the head of the freelist
970 * as ialloc_context. The caller should hold this buffer across
971 * the commit and pass it back into this routine on the second call.
972 *
973 * If we are allocating quota inodes, we do not have a parent inode
974 * to attach to or associate with (i.e. pip == NULL) because they
975 * are not linked into the directory structure - they are attached
976 * directly to the superblock - and so have no parent.
977 */
978 int
982 mode_t mode,
983 xfs_nlink_t nlink,
984 xfs_dev_t rdev,
986 xfs_prid_t prid,
991 {
992 xfs_ino_t ino;
994 uint flags;
996 timespec_t tv;
999 /*
1000 * Call the space management code to pick
1001 * the on-disk inode to be allocated.
1002 */
1010 }
1013 /*
1014 * Get the in-core inode with the lock held exclusively.
1015 * This is because we're setting fields here we need
1016 * to prevent others from looking at until we're done.
1017 */
1033 /*
1034 * If the superblock version is up to where we support new format
1035 * inodes and this is currently an old format inode, then change
1036 * the inode version number now. This way we only do the conversion
1037 * here rather than here and in the flush/logging code.
1038 */
1042 /*
1043 * We've already zeroed the old link count, the projid field,
1044 * and the pad field.
1045 */
1046 }
1048 /*
1049 * Project ids won't be stored on disk if we are using a version 1 inode.
1050 */
1058 }
1059 }
1061 /*
1062 * If the group ID of the new file does not match the effective group
1063 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1064 * (and only if the irix_sgid_inherit compatibility variable is set).
1065 */
1070 }
1083 /*
1084 * di_gen will have been taken care of in xfs_iread.
1085 */
1102 /*
1103 * we can't set up filestreams until after the VFS inode
1104 * is set up properly.
1105 */
1108 /* fall through */
1119 }
1126 }
1127 }
1129 xfs_inherit_noatime)
1132 xfs_inherit_nodump)
1135 xfs_inherit_sync)
1138 xfs_inherit_nosymlinks)
1143 xfs_inherit_nodefrag)
1148 }
1149 /* FALLTHROUGH */
1158 }
1159 /*
1160 * Attribute fork settings for new inode.
1161 */
1165 /*
1166 * Log the new values stuffed into the inode.
1167 */
1170 /* now that we have an i_mode we can setup inode ops and unlock */
1173 /* now we have set up the vfs inode we can associate the filestream */
1180 }
1184 }
1186 /*
1187 * Check to make sure that there are no blocks allocated to the
1188 * file beyond the size of the file. We don't check this for
1189 * files with fixed size extents or real time extents, but we
1190 * at least do it for regular files.
1191 */
1192 #ifdef DEBUG
1193 void
1197 xfs_fsize_t isize)
1198 {
1199 xfs_fileoff_t map_first;
1214 /*
1215 * The filesystem could be shutting down, so bmapi may return
1216 * an error.
1217 */
1221 map_first),
1223 NULL))
1227 }
1230 /*
1231 * Calculate the last possible buffered byte in a file. This must
1232 * include data that was buffered beyond the EOF by the write code.
1233 * This also needs to deal with overflowing the xfs_fsize_t type
1234 * which can happen for sizes near the limit.
1235 *
1236 * We also need to take into account any blocks beyond the EOF. It
1237 * may be the case that they were buffered by a write which failed.
1238 * In that case the pages will still be in memory, but the inode size
1239 * will never have been updated.
1240 */
1241 STATIC xfs_fsize_t
1244 {
1246 xfs_fsize_t last_byte;
1247 xfs_fileoff_t last_block;
1248 xfs_fileoff_t size_last_block;
1254 /*
1255 * Only check for blocks beyond the EOF if the extents have
1256 * been read in. This eliminates the need for the inode lock,
1257 * and it also saves us from looking when it really isn't
1258 * necessary.
1259 */
1263 XFS_DATA_FORK);
1267 }
1270 }
1277 }
1281 }
1283 }
1285 /*
1286 * Start the truncation of the file to new_size. The new size
1287 * must be smaller than the current size. This routine will
1288 * clear the buffer and page caches of file data in the removed
1289 * range, and xfs_itruncate_finish() will remove the underlying
1290 * disk blocks.
1291 *
1292 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1293 * must NOT have the inode lock held at all. This is because we're
1294 * calling into the buffer/page cache code and we can't hold the
1295 * inode lock when we do so.
1296 *
1297 * We need to wait for any direct I/Os in flight to complete before we
1298 * proceed with the truncate. This is needed to prevent the extents
1299 * being read or written by the direct I/Os from being removed while the
1300 * I/O is in flight as there is no other method of synchronising
1301 * direct I/O with the truncate operation. Also, because we hold
1302 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1303 * started until the truncate completes and drops the lock. Essentially,
1304 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1305 * ordering between direct I/Os and the truncate operation.
1306 *
1307 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1308 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1309 * in the case that the caller is locking things out of order and
1310 * may not be able to call xfs_itruncate_finish() with the inode lock
1311 * held without dropping the I/O lock. If the caller must drop the
1312 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1313 * must be called again with all the same restrictions as the initial
1314 * call.
1315 */
1316 int
1319 uint flags,
1320 xfs_fsize_t new_size)
1321 {
1322 xfs_fsize_t last_byte;
1323 xfs_off_t toss_start;
1334 /* wait for the completion of any pending DIOs */
1338 /*
1339 * Call toss_pages or flushinval_pages to get rid of pages
1340 * overlapping the region being removed. We have to use
1341 * the less efficient flushinval_pages in the case that the
1342 * caller may not be able to finish the truncate without
1343 * dropping the inode's I/O lock. Make sure
1344 * to catch any pages brought in by buffers overlapping
1345 * the EOF by searching out beyond the isize by our
1346 * block size. We round new_size up to a block boundary
1347 * so that we don't toss things on the same block as
1348 * new_size but before it.
1349 *
1350 * Before calling toss_page or flushinval_pages, make sure to
1351 * call remapf() over the same region if the file is mapped.
1352 * This frees up mapped file references to the pages in the
1353 * given range and for the flushinval_pages case it ensures
1354 * that we get the latest mapped changes flushed out.
1355 */
1359 /*
1360 * The place to start tossing is beyond our maximum
1361 * file size, so there is no way that the data extended
1362 * out there.
1363 */
1365 }
1375 }
1376 }
1378 #ifdef DEBUG
1381 }
1382 #endif
1384 }
1386 /*
1387 * Shrink the file to the given new_size. The new size must be smaller than
1388 * the current size. This will free up the underlying blocks in the removed
1389 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1390 *
1391 * The transaction passed to this routine must have made a permanent log
1392 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1393 * given transaction and start new ones, so make sure everything involved in
1394 * the transaction is tidy before calling here. Some transaction will be
1395 * returned to the caller to be committed. The incoming transaction must
1396 * already include the inode, and both inode locks must be held exclusively.
1397 * The inode must also be "held" within the transaction. On return the inode
1398 * will be "held" within the returned transaction. This routine does NOT
1399 * require any disk space to be reserved for it within the transaction.
1400 *
1401 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1402 * indicates the fork which is to be truncated. For the attribute fork we only
1403 * support truncation to size 0.
1404 *
1405 * We use the sync parameter to indicate whether or not the first transaction
1406 * we perform might have to be synchronous. For the attr fork, it needs to be
1407 * so if the unlink of the inode is not yet known to be permanent in the log.
1408 * This keeps us from freeing and reusing the blocks of the attribute fork
1409 * before the unlink of the inode becomes permanent.
1410 *
1411 * For the data fork, we normally have to run synchronously if we're being
1412 * called out of the inactive path or we're being called out of the create path
1413 * where we're truncating an existing file. Either way, the truncate needs to
1414 * be sync so blocks don't reappear in the file with altered data in case of a
1415 * crash. wsync filesystems can run the first case async because anything that
1416 * shrinks the inode has to run sync so by the time we're called here from
1417 * inactive, the inode size is permanently set to 0.
1418 *
1419 * Calls from the truncate path always need to be sync unless we're in a wsync
1420 * filesystem and the file has already been unlinked.
1421 *
1422 * The caller is responsible for correctly setting the sync parameter. It gets
1423 * too hard for us to guess here which path we're being called out of just
1424 * based on inode state.
1425 *
1426 * If we get an error, we must return with the inode locked and linked into the
1427 * current transaction. This keeps things simple for the higher level code,
1428 * because it always knows that the inode is locked and held in the transaction
1429 * that returns to it whether errors occur or not. We don't mark the inode
1430 * dirty on error so that transactions can be easily aborted if possible.
1431 */
1432 int
1436 xfs_fsize_t new_size,
1439 {
1440 xfs_fsblock_t first_block;
1441 xfs_fileoff_t first_unmap_block;
1442 xfs_fileoff_t last_block;
1448 xfs_bmap_free_t free_list;
1464 /*
1465 * We only support truncating the entire attribute fork.
1466 */
1469 }
1473 /*
1474 * The first thing we do is set the size to new_size permanently
1475 * on disk. This way we don't have to worry about anyone ever
1476 * being able to look at the data being freed even in the face
1477 * of a crash. What we're getting around here is the case where
1478 * we free a block, it is allocated to another file, it is written
1479 * to, and then we crash. If the new data gets written to the
1480 * file but the log buffers containing the free and reallocation
1481 * don't, then we'd end up with garbage in the blocks being freed.
1482 * As long as we make the new_size permanent before actually
1483 * freeing any blocks it doesn't matter if they get writtten to.
1484 *
1485 * The callers must signal into us whether or not the size
1486 * setting here must be synchronous. There are a few cases
1487 * where it doesn't have to be synchronous. Those cases
1488 * occur if the file is unlinked and we know the unlink is
1489 * permanent or if the blocks being truncated are guaranteed
1490 * to be beyond the inode eof (regardless of the link count)
1491 * and the eof value is permanent. Both of these cases occur
1492 * only on wsync-mounted filesystems. In those cases, we're
1493 * guaranteed that no user will ever see the data in the blocks
1494 * that are being truncated so the truncate can run async.
1495 * In the free beyond eof case, the file may wind up with
1496 * more blocks allocated to it than it needs if we crash
1497 * and that won't get fixed until the next time the file
1498 * is re-opened and closed but that's ok as that shouldn't
1499 * be too many blocks.
1500 *
1501 * However, we can't just make all wsync xactions run async
1502 * because there's one call out of the create path that needs
1503 * to run sync where it's truncating an existing file to size
1504 * 0 whose size is > 0.
1505 *
1506 * It's probably possible to come up with a test in this
1507 * routine that would correctly distinguish all the above
1508 * cases from the values of the function parameters and the
1509 * inode state but for sanity's sake, I've decided to let the
1510 * layers above just tell us. It's simpler to correctly figure
1511 * out in the layer above exactly under what conditions we
1512 * can run async and I think it's easier for others read and
1513 * follow the logic in case something has to be changed.
1514 * cscope is your friend -- rcc.
1515 *
1516 * The attribute fork is much simpler.
1517 *
1518 * For the attribute fork we allow the caller to tell us whether
1519 * the unlink of the inode that led to this call is yet permanent
1520 * in the on disk log. If it is not and we will be freeing extents
1521 * in this inode then we make the first transaction synchronous
1522 * to make sure that the unlink is permanent by the time we free
1523 * the blocks.
1524 */
1527 /*
1528 * If we are not changing the file size then do
1529 * not update the on-disk file size - we may be
1530 * called from xfs_inactive_free_eofblocks(). If we
1531 * update the on-disk file size and then the system
1532 * crashes before the contents of the file are
1533 * flushed to disk then the files may be full of
1534 * holes (ie NULL files bug).
1535 */
1540 }
1541 }
1546 }
1552 /*
1553 * Since it is possible for space to become allocated beyond
1554 * the end of the file (in a crash where the space is allocated
1555 * but the inode size is not yet updated), simply remove any
1556 * blocks which show up between the new EOF and the maximum
1557 * possible file size. If the first block to be removed is
1558 * beyond the maximum file size (ie it is the same as last_block),
1559 * then there is nothing to do.
1560 */
1568 }
1570 /*
1571 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1572 * will tell us whether it freed the entire range or
1573 * not. If this is a synchronous mount (wsync),
1574 * then we can tell bunmapi to keep all the
1575 * transactions asynchronous since the unlink
1576 * transaction that made this inode inactive has
1577 * already hit the disk. There's no danger of
1578 * the freed blocks being reused, there being a
1579 * crash, and the reused blocks suddenly reappearing
1580 * in this file with garbage in them once recovery
1581 * runs.
1582 */
1587 XFS_ITRUNC_MAX_EXTENTS,
1591 /*
1592 * If the bunmapi call encounters an error,
1593 * return to the caller where the transaction
1594 * can be properly aborted. We just need to
1595 * make sure we're not holding any resources
1596 * that we were not when we came in.
1597 */
1600 }
1602 /*
1603 * Duplicate the transaction that has the permanent
1604 * reservation and commit the old transaction.
1605 */
1612 /*
1613 * If the bmap finish call encounters an error, return
1614 * to the caller where the transaction can be properly
1615 * aborted. We just need to make sure we're not
1616 * holding any resources that we were not when we came
1617 * in.
1618 *
1619 * Aborting from this point might lose some blocks in
1620 * the file system, but oh well.
1621 */
1624 }
1627 /*
1628 * Mark the inode dirty so it will be logged and
1629 * moved forward in the log as part of every commit.
1630 */
1632 }
1642 /*
1643 * transaction commit worked ok so we can drop the extra ticket
1644 * reference that we gained in xfs_trans_dup()
1645 */
1649 XFS_TRANS_PERM_LOG_RES,
1650 XFS_ITRUNCATE_LOG_COUNT);
1653 }
1654 /*
1655 * Only update the size in the case of the data fork, but
1656 * always re-log the inode so that our permanent transaction
1657 * can keep on rolling it forward in the log.
1658 */
1661 /*
1662 * If we are not changing the file size then do
1663 * not update the on-disk file size - we may be
1664 * called from xfs_inactive_free_eofblocks(). If we
1665 * update the on-disk file size and then the system
1666 * crashes before the contents of the file are
1667 * flushed to disk then the files may be full of
1668 * holes (ie NULL files bug).
1669 */
1673 }
1674 }
1684 }
1686 /*
1687 * This is called when the inode's link count goes to 0.
1688 * We place the on-disk inode on a list in the AGI. It
1689 * will be pulled from this list when the inode is freed.
1690 */
1691 int
1695 {
1701 xfs_agino_t agino;
1712 /*
1713 * Get the agi buffer first. It ensures lock ordering
1714 * on the list.
1715 */
1721 /*
1722 * Get the index into the agi hash table for the
1723 * list this inode will go on.
1724 */
1732 /*
1733 * There is already another inode in the bucket we need
1734 * to add ourselves to. Add us at the front of the list.
1735 * Here we put the head pointer into our next pointer,
1736 * and then we fall through to point the head at us.
1737 */
1743 /* both on-disk, don't endian flip twice */
1751 }
1753 /*
1754 * Point the bucket head pointer at the inode being inserted.
1755 */
1763 }
1765 /*
1766 * Pull the on-disk inode from the AGI unlinked list.
1767 */
1768 STATIC int
1772 {
1773 xfs_ino_t next_ino;
1779 xfs_agnumber_t agno;
1780 xfs_agino_t agino;
1781 xfs_agino_t next_agino;
1791 /*
1792 * Get the agi buffer first. It ensures lock ordering
1793 * on the list.
1794 */
1801 /*
1802 * Get the index into the agi hash table for the
1803 * list this inode will go on.
1804 */
1812 /*
1813 * We're at the head of the list. Get the inode's
1814 * on-disk buffer to see if there is anyone after us
1815 * on the list. Only modify our next pointer if it
1816 * is not already NULLAGINO. This saves us the overhead
1817 * of dealing with the buffer when there is no need to
1818 * change it.
1819 */
1826 }
1839 }
1840 /*
1841 * Point the bucket head pointer at the next inode.
1842 */
1851 /*
1852 * We need to search the list for the inode being freed.
1853 */
1857 /*
1858 * If the last inode wasn't the one pointing to
1859 * us, then release its buffer since we're not
1860 * going to do anything with it.
1861 */
1864 }
1873 }
1877 }
1878 /*
1879 * Now last_ibp points to the buffer previous to us on
1880 * the unlinked list. Pull us from the list.
1881 */
1888 }
1902 }
1903 /*
1904 * Point the previous inode on the list to the next inode.
1905 */
1913 }
1915 }
1917 /*
1918 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1919 * inodes that are in memory - they all must be marked stale and attached to
1920 * the cluster buffer.
1921 */
1922 STATIC void
1926 xfs_ino_t inum)
1927 {
1933 xfs_daddr_t blkno;
1950 }
1956 /*
1957 * We obtain and lock the backing buffer first in the process
1958 * here, as we have to ensure that any dirty inode that we
1959 * can't get the flush lock on is attached to the buffer.
1960 * If we scan the in-memory inodes first, then buffer IO can
1961 * complete before we get a lock on it, and hence we may fail
1962 * to mark all the active inodes on the buffer stale.
1963 */
1966 XBF_LOCK);
1968 /*
1969 * Walk the inodes already attached to the buffer and mark them
1970 * stale. These will all have the flush locks held, so an
1971 * in-memory inode walk can't lock them. By marking them all
1972 * stale first, we will not attempt to lock them in the loop
1973 * below as the XFS_ISTALE flag will be set.
1974 */
1985 }
1987 }
1990 /*
1991 * For each inode in memory attempt to add it to the inode
1992 * buffer and set it up for being staled on buffer IO
1993 * completion. This is safe as we've locked out tail pushing
1994 * and flushing by locking the buffer.
1995 *
1996 * We have already marked every inode that was part of a
1997 * transaction stale above, which means there is no point in
1998 * even trying to lock them.
1999 */
2001 retry:
2006 /* Inode not in memory or stale, nothing to do */
2010 }
2012 /*
2013 * Don't try to lock/unlock the current inode, but we
2014 * _cannot_ skip the other inodes that we did not find
2015 * in the list attached to the buffer and are not
2016 * already marked stale. If we can't lock it, back off
2017 * and retry.
2018 */
2024 }
2030 /*
2031 * we don't need to attach clean inodes or those only
2032 * with unlogged changes (which we throw away, anyway).
2033 */
2041 }
2054 }
2058 }
2061 }
2063 /*
2064 * This is called to return an inode to the inode free list.
2065 * The inode should already be truncated to 0 length and have
2066 * no pages associated with it. This routine also assumes that
2067 * the inode is already a part of the transaction.
2068 *
2069 * The on-disk copy of the inode will have been added to the list
2070 * of unlinked inodes in the AGI. We need to remove the inode from
2071 * that list atomically with respect to freeing it here.
2072 */
2073 int
2078 {
2081 xfs_ino_t first_ino;
2094 /*
2095 * Pull the on-disk inode from the AGI unlinked list.
2096 */
2100 }
2105 }
2114 /*
2115 * Bump the generation count so no one will be confused
2116 * by reincarnations of this inode.
2117 */
2126 /*
2127 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2128 * from picking up this inode when it is reclaimed (its incore state
2129 * initialzed but not flushed to disk yet). The in-core di_mode is
2130 * already cleared and a corresponding transaction logged.
2131 * The hack here just synchronizes the in-core to on-disk
2132 * di_mode value in advance before the actual inode sync to disk.
2133 * This is OK because the inode is already unlinked and would never
2134 * change its di_mode again for this inode generation.
2135 * This is a temporary hack that would require a proper fix
2136 * in the future.
2137 */
2142 }
2145 }
2147 /*
2148 * Reallocate the space for if_broot based on the number of records
2149 * being added or deleted as indicated in rec_diff. Move the records
2150 * and pointers in if_broot to fit the new size. When shrinking this
2151 * will eliminate holes between the records and pointers created by
2152 * the caller. When growing this will create holes to be filled in
2153 * by the caller.
2154 *
2155 * The caller must not request to add more records than would fit in
2156 * the on-disk inode root. If the if_broot is currently NULL, then
2157 * if we adding records one will be allocated. The caller must also
2158 * not request that the number of records go below zero, although
2159 * it can go to zero.
2160 *
2161 * ip -- the inode whose if_broot area is changing
2162 * ext_diff -- the change in the number of records, positive or negative,
2163 * requested for the if_broot array.
2164 */
2165 void
2170 {
2180 /*
2181 * Handle the degenerate case quietly.
2182 */
2185 }
2189 /*
2190 * If there wasn't any memory allocated before, just
2191 * allocate it now and get out.
2192 */
2198 }
2200 /*
2201 * If there is already an existing if_broot, then we need
2202 * to realloc() it and shift the pointers to their new
2203 * location. The records don't change location because
2204 * they are kept butted up against the btree block header.
2205 */
2221 }
2223 /*
2224 * rec_diff is less than 0. In this case, we are shrinking the
2225 * if_broot buffer. It must already exist. If we go to zero
2226 * records, just get rid of the root and clear the status bit.
2227 */
2234 else
2238 /*
2239 * First copy over the btree block header.
2240 */
2245 }
2247 /*
2248 * Only copy the records and pointers if there are any.
2249 */
2251 /*
2252 * First copy the records.
2253 */
2258 /*
2259 * Then copy the pointers.
2260 */
2266 }
2273 }
2276 /*
2277 * This is called when the amount of space needed for if_data
2278 * is increased or decreased. The change in size is indicated by
2279 * the number of bytes that need to be added or deleted in the
2280 * byte_diff parameter.
2281 *
2282 * If the amount of space needed has decreased below the size of the
2283 * inline buffer, then switch to using the inline buffer. Otherwise,
2284 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2285 * to what is needed.
2286 *
2287 * ip -- the inode whose if_data area is changing
2288 * byte_diff -- the change in the number of bytes, positive or negative,
2289 * requested for the if_data array.
2290 */
2291 void
2296 {
2303 }
2312 }
2316 /*
2317 * If the valid extents/data can fit in if_inline_ext/data,
2318 * copy them from the malloc'd vector and free it.
2319 */
2325 new_size);
2328 }
2331 /*
2332 * Stuck with malloc/realloc.
2333 * For inline data, the underlying buffer must be
2334 * a multiple of 4 bytes in size so that it can be
2335 * logged and stay on word boundaries. We enforce
2336 * that here.
2337 */
2344 /*
2345 * Only do the realloc if the underlying size
2346 * is really changing.
2347 */
2351 real_size,
2354 }
2361 }
2362 }
2366 }
2368 void
2372 {
2379 }
2381 /*
2382 * If the format is local, then we can't have an extents
2383 * array so just look for an inline data array. If we're
2384 * not local then we may or may not have an extents list,
2385 * so check and free it up if we do.
2386 */
2394 }
2401 }
2408 }
2409 }
2411 /*
2412 * This is called to unpin an inode. The caller must have the inode locked
2413 * in at least shared mode so that the buffer cannot be subsequently pinned
2414 * once someone is waiting for it to be unpinned.
2415 */
2416 static void
2419 {
2424 /* Give the log a push to start the unpinning I/O */
2427 }
2429 void
2432 {
2436 }
2437 }
2439 /*
2440 * xfs_iextents_copy()
2441 *
2442 * This is called to copy the REAL extents (as opposed to the delayed
2443 * allocation extents) from the inode into the given buffer. It
2444 * returns the number of bytes copied into the buffer.
2445 *
2446 * If there are no delayed allocation extents, then we can just
2447 * memcpy() the extents into the buffer. Otherwise, we need to
2448 * examine each extent in turn and skip those which are delayed.
2449 */
2450 int
2455 {
2460 xfs_fsblock_t start_block;
2470 /*
2471 * There are some delayed allocation extents in the
2472 * inode, so copy the extents one at a time and skip
2473 * the delayed ones. There must be at least one
2474 * non-delayed extent.
2475 */
2481 /*
2482 * It's a delayed allocation extent, so skip it.
2483 */
2485 }
2487 /* Translate to on disk format */
2490 dp++;
2491 copied++;
2492 }
2497 }
2499 /*
2500 * Each of the following cases stores data into the same region
2501 * of the on-disk inode, so only one of them can be valid at
2502 * any given time. While it is possible to have conflicting formats
2503 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2504 * in EXTENTS format, this can only happen when the fork has
2505 * changed formats after being modified but before being flushed.
2506 * In these cases, the format always takes precedence, because the
2507 * format indicates the current state of the fork.
2508 */
2509 /*ARGSUSED*/
2510 STATIC void
2517 {
2521 #ifdef XFS_TRANS_DEBUG
2523 #endif
2534 /*
2535 * This can happen if we gave up in iformat in an error path,
2536 * for the attribute fork.
2537 */
2541 }
2551 }
2565 whichfork);
2566 }
2575 XFS_BROOT_SIZE_ADJ));
2579 }
2586 }
2595 }
2601 }
2602 }
2604 STATIC int
2608 {
2632 /* really need a gang lookup range call here */
2642 /* if the inode lies outside this cluster, we're done. */
2645 /*
2646 * Do an un-protected check to see if the inode is dirty and
2647 * is a candidate for flushing. These checks will be repeated
2648 * later after the appropriate locks are acquired.
2649 */
2653 /*
2654 * Try to get locks. If any are unavailable or it is pinned,
2655 * then this inode cannot be flushed and is skipped.
2656 */
2663 }
2668 }
2670 /*
2671 * arriving here means that this inode can be flushed. First
2672 * re-check that it's dirty before flushing.
2673 */
2680 }
2681 clcount++;
2684 }
2686 }
2691 }
2693 out_free:
2696 out_put:
2701 cluster_corrupt_out:
2702 /*
2703 * Corruption detected in the clustering loop. Invalidate the
2704 * inode buffer and shut down the filesystem.
2705 */
2707 /*
2708 * Clean up the buffer. If it was B_DELWRI, just release it --
2709 * brelse can handle it with no problems. If not, shut down the
2710 * filesystem before releasing the buffer.
2711 */
2719 /*
2720 * Just like incore_relse: if we have b_iodone functions,
2721 * mark the buffer as an error and call them. Otherwise
2722 * mark it as stale and brelse.
2723 */
2732 }
2733 }
2735 /*
2736 * Unlocks the flush lock
2737 */
2742 }
2744 /*
2745 * xfs_iflush() will write a modified inode's changes out to the
2746 * inode's on disk home. The caller must have the inode lock held
2747 * in at least shared mode and the inode flush completion must be
2748 * active as well. The inode lock will still be held upon return from
2749 * the call and the caller is free to unlock it.
2750 * The inode flush will be completed when the inode reaches the disk.
2751 * The flags indicate how the inode's buffer should be written out.
2752 */
2753 int
2756 uint flags)
2757 {
2774 /*
2775 * We can't flush the inode until it is unpinned, so wait for it if we
2776 * are allowed to block. We know noone new can pin it, because we are
2777 * holding the inode lock shared and you need to hold it exclusively to
2778 * pin the inode.
2779 *
2780 * If we are not allowed to block, force the log out asynchronously so
2781 * that when we come back the inode will be unpinned. If other inodes
2782 * in the same cluster are dirty, they will probably write the inode
2783 * out for us if they occur after the log force completes.
2784 */
2789 }
2792 /*
2793 * For stale inodes we cannot rely on the backing buffer remaining
2794 * stale in cache for the remaining life of the stale inode and so
2795 * xfs_itobp() below may give us a buffer that no longer contains
2796 * inodes below. We have to check this after ensuring the inode is
2797 * unpinned so that it is safe to reclaim the stale inode after the
2798 * flush call.
2799 */
2803 }
2805 /*
2806 * This may have been unpinned because the filesystem is shutting
2807 * down forcibly. If that's the case we must not write this inode
2808 * to disk, because the log record didn't make it to disk!
2809 */
2816 }
2818 /*
2819 * Get the buffer containing the on-disk inode.
2820 */
2826 }
2828 /*
2829 * First flush out the inode that xfs_iflush was called with.
2830 */
2835 /*
2836 * If the buffer is pinned then push on the log now so we won't
2837 * get stuck waiting in the write for too long.
2838 */
2842 /*
2843 * inode clustering:
2844 * see if other inodes can be gathered into this write
2845 */
2852 else
2856 corrupt_out:
2859 cluster_corrupt_out:
2860 /*
2861 * Unlocks the flush lock
2862 */
2865 }
2868 STATIC int
2872 {
2876 #ifdef XFS_TRANS_DEBUG
2878 #endif
2888 /* set *dip = inode's place in the buffer */
2891 /*
2892 * Clear i_update_core before copying out the data.
2893 * This is for coordination with our timestamp updates
2894 * that don't hold the inode lock. They will always
2895 * update the timestamps BEFORE setting i_update_core,
2896 * so if we clear i_update_core after they set it we
2897 * are guaranteed to see their updates to the timestamps.
2898 * I believe that this depends on strongly ordered memory
2899 * semantics, but we have that. We use the SYNCHRONIZE
2900 * macro to make sure that the compiler does not reorder
2901 * the i_update_core access below the data copy below.
2902 */
2906 /*
2907 * Make sure to get the latest timestamps from the Linux inode.
2908 */
2917 }
2924 }
2934 }
2945 }
2946 }
2949 XFS_RANDOM_IFLUSH_5)) {
2955 ip);
2957 }
2964 }
2965 /*
2966 * bump the flush iteration count, used to detect flushes which
2967 * postdate a log record during recovery.
2968 */
2972 /*
2973 * Copy the dirty parts of the inode into the on-disk
2974 * inode. We always copy out the core of the inode,
2975 * because if the inode is dirty at all the core must
2976 * be.
2977 */
2980 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2984 /*
2985 * If this is really an old format inode and the superblock version
2986 * has not been updated to support only new format inodes, then
2987 * convert back to the old inode format. If the superblock version
2988 * has been updated, then make the conversion permanent.
2989 */
2993 /*
2994 * Convert it back.
2995 */
2999 /*
3000 * The superblock version has already been bumped,
3001 * so just make the conversion to the new inode
3002 * format permanent.
3003 */
3012 }
3013 }
3020 /*
3021 * We've recorded everything logged in the inode, so we'd
3022 * like to clear the ilf_fields bits so we don't log and
3023 * flush things unnecessarily. However, we can't stop
3024 * logging all this information until the data we've copied
3025 * into the disk buffer is written to disk. If we did we might
3026 * overwrite the copy of the inode in the log with all the
3027 * data after re-logging only part of it, and in the face of
3028 * a crash we wouldn't have all the data we need to recover.
3029 *
3030 * What we do is move the bits to the ili_last_fields field.
3031 * When logging the inode, these bits are moved back to the
3032 * ilf_fields field. In the xfs_iflush_done() routine we
3033 * clear ili_last_fields, since we know that the information
3034 * those bits represent is permanently on disk. As long as
3035 * the flush completes before the inode is logged again, then
3036 * both ilf_fields and ili_last_fields will be cleared.
3037 *
3038 * We can play with the ilf_fields bits here, because the inode
3039 * lock must be held exclusively in order to set bits there
3040 * and the flush lock protects the ili_last_fields bits.
3041 * Set ili_logged so the flush done
3042 * routine can tell whether or not to look in the AIL.
3043 * Also, store the current LSN of the inode so that we can tell
3044 * whether the item has moved in the AIL from xfs_iflush_done().
3045 * In order to read the lsn we need the AIL lock, because
3046 * it is a 64 bit value that cannot be read atomically.
3047 */
3056 /*
3057 * Attach the function xfs_iflush_done to the inode's
3058 * buffer. This will remove the inode from the AIL
3059 * and unlock the inode's flush lock when the inode is
3060 * completely written to disk.
3061 */
3067 /*
3068 * We're flushing an inode which is not in the AIL and has
3069 * not been logged but has i_update_core set. For this
3070 * case we can use a B_DELWRI flush and immediately drop
3071 * the inode flush lock because we can avoid the whole
3072 * AIL state thing. It's OK to drop the flush lock now,
3073 * because we've already locked the buffer and to do anything
3074 * you really need both.
3075 */
3080 }
3082 }
3086 corrupt_out:
3088 }
3090 /*
3091 * Return a pointer to the extent record at file index idx.
3092 */
3093 xfs_bmbt_rec_host_t *
3097 {
3112 }
3113 }
3115 /*
3116 * Insert new item(s) into the extent records for incore inode
3117 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3118 */
3119 void
3126 {
3136 }
3138 /*
3139 * This is called when the amount of space required for incore file
3140 * extents needs to be increased. The ext_diff parameter stores the
3141 * number of new extents being added and the idx parameter contains
3142 * the extent index where the new extents will be added. If the new
3143 * extents are being appended, then we just need to (re)allocate and
3144 * initialize the space. Otherwise, if the new extents are being
3145 * inserted into the middle of the existing entries, a bit more work
3146 * is required to make room for the new extents to be inserted. The
3147 * caller is responsible for filling in the new extent entries upon
3148 * return.
3149 */
3150 void
3155 {
3164 /*
3165 * If the new number of extents (nextents + ext_diff)
3166 * fits inside the inode, then continue to use the inline
3167 * extent buffer.
3168 */
3175 }
3179 }
3180 /*
3181 * Otherwise use a linear (direct) extent list.
3182 * If the extents are currently inside the inode,
3183 * xfs_iext_realloc_direct will switch us from
3184 * inline to direct extent allocation mode.
3185 */
3193 }
3194 }
3195 /* Indirection array */
3208 }
3209 /* Extents fit in target extent page */
3217 }
3220 }
3221 /* Insert a new extent page */
3225 }
3226 /*
3227 * If extent(s) are being appended to the last page in
3228 * the indirection array and the new extent(s) don't fit
3229 * in the page, then erp is NULL and erp_idx is set to
3230 * the next index needed in the indirection array.
3231 */
3240 erp_idx++;
3241 }
3242 }
3243 }
3244 }
3246 }
3248 /*
3249 * This is called when incore extents are being added to the indirection
3250 * array and the new extents do not fit in the target extent list. The
3251 * erp_idx parameter contains the irec index for the target extent list
3252 * in the indirection array, and the idx parameter contains the extent
3253 * index within the list. The number of extents being added is stored
3254 * in the count parameter.
3255 *
3256 * |-------| |-------|
3257 * | | | | idx - number of extents before idx
3258 * | idx | | count |
3259 * | | | | count - number of extents being inserted at idx
3260 * |-------| |-------|
3261 * | count | | nex2 | nex2 - number of extents after idx + count
3262 * |-------| |-------|
3263 */
3264 void
3270 {
3284 /*
3285 * Save second part of target extent list
3286 * (all extents past */
3294 }
3296 /*
3297 * Add the new extents to the end of the target
3298 * list, then allocate new irec record(s) and
3299 * extent buffer(s) as needed to store the rest
3300 * of the new extents.
3301 */
3308 }
3310 erp_idx++;
3316 }
3318 /* Add nex2 extents back to indirection array */
3320 xfs_extnum_t ext_avail;
3326 /*
3327 * If nex2 extents fit in the current page, append
3328 * nex2_ep after the new extents.
3329 */
3332 }
3333 /*
3334 * Otherwise, check if space is available in the
3335 * next page.
3336 */
3340 erp_idx++;
3341 erp++;
3342 /* Create a hole for nex2 extents */
3345 }
3346 /*
3347 * Final choice, create a new extent page for
3348 * nex2 extents.
3349 */
3351 erp_idx++;
3353 }
3358 }
3359 }
3361 /*
3362 * This is called when the amount of space required for incore file
3363 * extents needs to be decreased. The ext_diff parameter stores the
3364 * number of extents to be removed and the idx parameter contains
3365 * the extent index where the extents will be removed from.
3366 *
3367 * If the amount of space needed has decreased below the linear
3368 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3369 * extent array. Otherwise, use kmem_realloc() to adjust the
3370 * size to what is needed.
3371 */
3372 void
3378 {
3397 }
3399 }
3401 /*
3402 * This removes ext_diff extents from the inline buffer, beginning
3403 * at extent index idx.
3404 */
3405 void
3410 {
3429 }
3430 }
3432 /*
3433 * This removes ext_diff extents from a linear (direct) extent list,
3434 * beginning at extent index idx. If the extents are being removed
3435 * from the end of the list (ie. truncate) then we just need to re-
3436 * allocate the list to remove the extra space. Otherwise, if the
3437 * extents are being removed from the middle of the existing extent
3438 * entries, then we first need to move the extent records beginning
3439 * at idx + ext_diff up in the list to overwrite the records being
3440 * removed, then remove the extra space via kmem_realloc.
3441 */
3442 void
3447 {
3459 }
3460 /* Move extents up in the list (if needed) */
3466 }
3469 /*
3470 * Reallocate the direct extent list. If the extents
3471 * will fit inside the inode then xfs_iext_realloc_direct
3472 * will switch from direct to inline extent allocation
3473 * mode for us.
3474 */
3477 }
3479 /*
3480 * This is called when incore extents are being removed from the
3481 * indirection array and the extents being removed span multiple extent
3482 * buffers. The idx parameter contains the file extent index where we
3483 * want to begin removing extents, and the count parameter contains
3484 * how many extents need to be removed.
3485 *
3486 * |-------| |-------|
3487 * | nex1 | | | nex1 - number of extents before idx
3488 * |-------| | count |
3489 * | | | | count - number of extents being removed at idx
3490 * | count | |-------|
3491 * | | | nex2 | nex2 - number of extents after idx + count
3492 * |-------| |-------|
3493 */
3494 void
3499 {
3516 /*
3517 * Check for deletion of entire list;
3518 * xfs_iext_irec_remove() updates extent offsets.
3519 */
3526 XFS_IEXT_BUFSZ);
3532 }
3533 }
3534 /* Move extents up (if needed) */
3539 }
3540 /* Zero out rest of page */
3543 /* Update remaining counters */
3548 erp_idx++;
3549 erp++;
3550 }
3553 }
3555 /*
3556 * Create, destroy, or resize a linear (direct) block of extents.
3557 */
3558 void
3562 {
3571 /* Free extent records */
3574 }
3575 /* Resize direct extent list and zero any new bytes */
3577 /* Check if extents will fit inside the inode */
3583 }
3586 }
3590 rnew_size,
3592 }
3597 }
3598 }
3599 /*
3600 * Switch from the inline extent buffer to a direct
3601 * extent list. Be sure to include the inline extent
3602 * bytes in new_size.
3603 */
3608 }
3610 }
3613 }
3615 /*
3616 * Switch from linear (direct) extent records to inline buffer.
3617 */
3618 void
3622 {
3625 /*
3626 * The inline buffer was zeroed when we switched
3627 * from inline to direct extent allocation mode,
3628 * so we don't need to clear it here.
3629 */
3635 }
3637 /*
3638 * Switch from inline buffer to linear (direct) extent records.
3639 * new_size should already be rounded up to the next power of 2
3640 * by the caller (when appropriate), so use new_size as it is.
3641 * However, since new_size may be rounded up, we can't update
3642 * if_bytes here. It is the caller's responsibility to update
3643 * if_bytes upon return.
3644 */
3645 void
3649 {
3657 }
3659 }
3661 /*
3662 * Resize an extent indirection array to new_size bytes.
3663 */
3664 STATIC void
3668 {
3683 }
3684 }
3686 /*
3687 * Switch from indirection array to linear (direct) extent allocations.
3688 */
3689 STATIC void
3692 {
3712 }
3713 }
3715 /*
3716 * Free incore file extents.
3717 */
3718 void
3721 {
3729 }
3736 }
3740 }
3742 /*
3743 * Return a pointer to the extent record for file system block bno.
3744 */
3750 {
3765 }
3768 /* Find target extent list */
3776 }
3777 /* Binary search extent records */
3788 /* Convert back to file-based extent index */
3791 }
3794 }
3795 }
3796 /* Convert back to file-based extent index */
3799 }
3805 }
3806 }
3809 }
3811 /*
3812 * Return a pointer to the indirection array entry containing the
3813 * extent record for filesystem block bno. Store the index of the
3814 * target irec in *erp_idxp.
3815 */
3821 {
3845 }
3846 }
3849 }
3851 /*
3852 * Return a pointer to the indirection array entry containing the
3853 * extent record at file extent index *idxp. Store the index of the
3854 * target irec in *erp_idxp and store the page index of the target
3855 * extent record in *idxp.
3856 */
3857 xfs_ext_irec_t *
3863 {
3880 /* Binary search extent irec's */
3896 erp_idx++;
3902 }
3903 }
3907 }
3909 /*
3910 * Allocate and initialize an indirection array once the space needed
3911 * for incore extents increases above XFS_IEXT_BUFSZ.
3912 */
3913 void
3916 {
3932 }
3943 }
3945 /*
3946 * Allocate and initialize a new entry in the indirection array.
3947 */
3948 xfs_ext_irec_t *
3952 {
3960 /* Resize indirection array */
3963 /*
3964 * Move records down in the array so the
3965 * new page can use erp_idx.
3966 */
3970 }
3973 /* Initialize new extent record */
3982 }
3984 /*
3985 * Remove a record from the indirection array.
3986 */
3987 void
3991 {
4003 }
4004 /* Compact extent records */
4008 }
4009 /*
4010 * Manually free the last extent record from the indirection
4011 * array. A call to xfs_iext_realloc_indirect() with a size
4012 * of zero would result in a call to xfs_iext_destroy() which
4013 * would in turn call this function again, creating a nasty
4014 * infinite loop.
4015 */
4021 }
4023 }
4025 /*
4026 * This is called to clean up large amounts of unused memory allocated
4027 * by the indirection array. Before compacting anything though, verify
4028 * that the indirection array is still needed and switch back to the
4029 * linear extent list (or even the inline buffer) if possible. The
4030 * compaction policy is as follows:
4031 *
4032 * Full Compaction: Extents fit into a single page (or inline buffer)
4033 * Partial Compaction: Extents occupy less than 50% of allocated space
4034 * No Compaction: Extents occupy at least 50% of allocated space
4035 */
4036 void
4039 {
4056 }
4057 }
4059 /*
4060 * Combine extents from neighboring extent pages.
4061 */
4062 void
4065 {
4081 /*
4082 * Free page before removing extent record
4083 * so er_extoffs don't get modified in
4084 * xfs_iext_irec_remove.
4085 */
4091 erp_idx++;
4092 }
4093 }
4094 }
4096 /*
4097 * This is called to update the er_extoff field in the indirection
4098 * array when extents have been added or removed from one of the
4099 * extent lists. erp_idx contains the irec index to begin updating
4100 * at and ext_diff contains the number of extents that were added
4101 * or removed.
4102 */
4103 void
4108 {
4116 }
4117 }