| blob | da871f5322368dffac20f1514a5966a45f26061e |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include <linux/log2.h>
54 /*
55 * Used in xfs_itruncate(). This is the maximum number of extents
56 * freed from a file in a single transaction.
57 */
58 #define XFS_ITRUNC_MAX_EXTENTS 2
65 #ifdef DEBUG
66 /*
67 * Make sure that the extents in the given memory buffer
68 * are valid.
69 */
70 STATIC void
74 xfs_exntfmt_t fmt)
75 {
76 xfs_bmbt_irec_t irec;
77 xfs_bmbt_rec_host_t rec;
87 }
88 }
90 #define xfs_validate_extents(ifp, nrecs, fmt)
93 /*
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
96 */
97 #if defined(DEBUG)
98 void
102 {
115 bp);
117 }
118 }
119 }
120 #endif
122 /*
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
126 */
127 STATIC int
133 uint buf_flags,
134 uint iget_flags)
135 {
150 }
152 }
154 /*
155 * Validate the magic number and version of every inode in the buffer
156 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
157 */
158 #ifdef DEBUG
162 #endif
173 XFS_ERRTAG_ITOBP_INOTOBP,
174 XFS_RANDOM_ITOBP_INOTOBP))) {
178 }
181 #ifdef DEBUG
187 #endif
190 }
191 }
195 /*
196 * Mark the buffer as an inode buffer now that it looks good
197 */
202 }
204 /*
205 * This routine is called to map an inode number within a file
206 * system to the buffer containing the on-disk version of the
207 * inode. It returns a pointer to the buffer containing the
208 * on-disk inode in the bpp parameter, and in the dip parameter
209 * it returns a pointer to the on-disk inode within that buffer.
210 *
211 * If a non-zero error is returned, then the contents of bpp and
212 * dipp are undefined.
213 *
214 * Use xfs_imap() to determine the size and location of the
215 * buffer to read from disk.
216 */
217 int
221 xfs_ino_t ino,
225 uint imap_flags)
226 {
244 }
247 /*
248 * This routine is called to map an inode to the buffer containing
249 * the on-disk version of the inode. It returns a pointer to the
250 * buffer containing the on-disk inode in the bpp parameter, and in
251 * the dip parameter it returns a pointer to the on-disk inode within
252 * that buffer.
253 *
254 * If a non-zero error is returned, then the contents of bpp and
255 * dipp are undefined.
256 *
257 * The inode is expected to already been mapped to its buffer and read
258 * in once, thus we can use the mapping information stored in the inode
259 * rather than calling xfs_imap(). This allows us to avoid the overhead
260 * of looking at the inode btree for small block file systems
261 * (see xfs_imap()).
262 */
263 int
270 uint buf_flags)
271 {
286 }
291 }
293 /*
294 * Move inode type and inode format specific information from the
295 * on-disk inode to the in-core inode. For fifos, devs, and sockets
296 * this means set if_rdev to the proper value. For files, directories,
297 * and symlinks this means to bring in the in-line data or extent
298 * pointers. For a file in B-tree format, only the root is immediately
299 * brought in-core. The rest will be in-lined in if_extents when it
300 * is first referenced (see xfs_iread_extents()).
301 */
302 STATIC int
306 {
310 xfs_fsize_t di_size;
328 }
337 }
347 }
358 }
369 /*
370 * no local regular files yet
371 */
377 XFS_ERRLEVEL_LOW,
380 }
389 XFS_ERRLEVEL_LOW,
392 }
407 }
413 }
416 }
434 XFS_ERRLEVEL_LOW,
437 }
450 }
455 }
457 }
459 /*
460 * The file is in-lined in the on-disk inode.
461 * If it fits into if_inline_data, then copy
462 * it there, otherwise allocate a buffer for it
463 * and copy the data there. Either way, set
464 * if_data to point at the data.
465 * If we allocate a buffer for the data, make
466 * sure that its size is a multiple of 4 and
467 * record the real size in i_real_bytes.
468 */
469 STATIC int
475 {
479 /*
480 * If the size is unreasonable, then something
481 * is wrong and we just bail out rather than crash in
482 * kmem_alloc() or memcpy() below.
483 */
492 }
502 }
510 }
512 /*
513 * The file consists of a set of extents all
514 * of which fit into the on-disk inode.
515 * If there are few enough extents to fit into
516 * the if_inline_ext, then copy them there.
517 * Otherwise allocate a buffer for them and copy
518 * them into it. Either way, set if_extents
519 * to point at the extents.
520 */
521 STATIC int
526 {
537 /*
538 * If the number of extents is unreasonable, then something
539 * is wrong and we just bail out rather than crash in
540 * kmem_alloc() or memcpy() below.
541 */
548 }
555 else
566 }
573 XFS_ERRLEVEL_LOW,
576 }
577 }
580 }
582 /*
583 * The file has too many extents to fit into
584 * the inode, so they are in B-tree format.
585 * Allocate a buffer for the root of the B-tree
586 * and copy the root into it. The i_extents
587 * field will remain NULL until all of the
588 * extents are read in (when they are needed).
589 */
590 STATIC int
595 {
598 /* REFERENCED */
607 /*
608 * blow out if -- fork has less extents than can fit in
609 * fork (fork shouldn't be a btree format), root btree
610 * block has more records than can fit into the fork,
611 * or the number of extents is greater than the number of
612 * blocks.
613 */
623 }
628 /*
629 * Copy and convert from the on-disk structure
630 * to the in-memory structure.
631 */
639 }
641 STATIC void
645 {
675 }
677 void
681 {
711 }
713 STATIC uint
715 __uint16_t di_flags)
716 {
748 }
751 }
753 uint
756 {
761 }
763 uint
766 {
769 }
771 /*
772 * Read the disk inode attributes into the in-core inode structure.
773 */
774 int
779 uint iget_flags)
780 {
785 /*
786 * Fill in the location information in the in-core inode.
787 */
792 /*
793 * Get pointers to the on-disk inode and the buffer containing it.
794 */
801 /*
802 * If we got something that isn't an inode it means someone
803 * (nfs or dmi) has a stale handle.
804 */
806 #ifdef DEBUG
813 }
815 /*
816 * If the on-disk inode is already linked to a directory
817 * entry, copy all of the inode into the in-core inode.
818 * xfs_iformat() handles copying in the inode format
819 * specific information.
820 * Otherwise, just get the truly permanent information.
821 */
826 #ifdef DEBUG
831 }
837 /*
838 * Make sure to pull in the mode here as well in
839 * case the inode is released without being used.
840 * This ensures that xfs_inactive() will see that
841 * the inode is already free and not try to mess
842 * with the uninitialized part of it.
843 */
845 /*
846 * Initialize the per-fork minima and maxima for a new
847 * inode here. xfs_iformat will do it for old inodes.
848 */
851 }
853 /*
854 * The inode format changed when we moved the link count and
855 * made it 32 bits long. If this is an old format inode,
856 * convert it in memory to look like a new one. If it gets
857 * flushed to disk we will convert back before flushing or
858 * logging it. We zero out the new projid field and the old link
859 * count field. We'll handle clearing the pad field (the remains
860 * of the old uuid field) when we actually convert the inode to
861 * the new format. We don't change the version number so that we
862 * can distinguish this from a real new format inode.
863 */
868 }
873 /*
874 * Mark the buffer containing the inode as something to keep
875 * around for a while. This helps to keep recently accessed
876 * meta-data in-core longer.
877 */
880 /*
881 * Use xfs_trans_brelse() to release the buffer containing the
882 * on-disk inode, because it was acquired with xfs_trans_read_buf()
883 * in xfs_itobp() above. If tp is NULL, this is just a normal
884 * brelse(). If we're within a transaction, then xfs_trans_brelse()
885 * will only release the buffer if it is not dirty within the
886 * transaction. It will be OK to release the buffer in this case,
887 * because inodes on disk are never destroyed and we will be
888 * locking the new in-core inode before putting it in the hash
889 * table where other processes can find it. Thus we don't have
890 * to worry about the inode being changed just because we released
891 * the buffer.
892 */
893 out_brelse:
896 }
898 /*
899 * Read in extents from a btree-format inode.
900 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
901 */
902 int
907 {
910 xfs_extnum_t nextents;
916 }
920 /*
921 * We know that the size is valid (it's checked in iformat_btree)
922 */
932 }
935 }
937 /*
938 * Allocate an inode on disk and return a copy of its in-core version.
939 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
940 * appropriately within the inode. The uid and gid for the inode are
941 * set according to the contents of the given cred structure.
942 *
943 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
944 * has a free inode available, call xfs_iget()
945 * to obtain the in-core version of the allocated inode. Finally,
946 * fill in the inode and log its initial contents. In this case,
947 * ialloc_context would be set to NULL and call_again set to false.
948 *
949 * If xfs_dialloc() does not have an available inode,
950 * it will replenish its supply by doing an allocation. Since we can
951 * only do one allocation within a transaction without deadlocks, we
952 * must commit the current transaction before returning the inode itself.
953 * In this case, therefore, we will set call_again to true and return.
954 * The caller should then commit the current transaction, start a new
955 * transaction, and call xfs_ialloc() again to actually get the inode.
956 *
957 * To ensure that some other process does not grab the inode that
958 * was allocated during the first call to xfs_ialloc(), this routine
959 * also returns the [locked] bp pointing to the head of the freelist
960 * as ialloc_context. The caller should hold this buffer across
961 * the commit and pass it back into this routine on the second call.
962 *
963 * If we are allocating quota inodes, we do not have a parent inode
964 * to attach to or associate with (i.e. pip == NULL) because they
965 * are not linked into the directory structure - they are attached
966 * directly to the superblock - and so have no parent.
967 */
968 int
972 mode_t mode,
973 xfs_nlink_t nlink,
974 xfs_dev_t rdev,
975 prid_t prid,
980 {
981 xfs_ino_t ino;
983 uint flags;
985 timespec_t tv;
988 /*
989 * Call the space management code to pick
990 * the on-disk inode to be allocated.
991 */
999 }
1002 /*
1003 * Get the in-core inode with the lock held exclusively.
1004 * This is because we're setting fields here we need
1005 * to prevent others from looking at until we're done.
1006 */
1022 /*
1023 * If the superblock version is up to where we support new format
1024 * inodes and this is currently an old format inode, then change
1025 * the inode version number now. This way we only do the conversion
1026 * here rather than here and in the flush/logging code.
1027 */
1031 /*
1032 * We've already zeroed the old link count, the projid field,
1033 * and the pad field.
1034 */
1035 }
1037 /*
1038 * Project ids won't be stored on disk if we are using a version 1 inode.
1039 */
1047 }
1048 }
1050 /*
1051 * If the group ID of the new file does not match the effective group
1052 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1053 * (and only if the irix_sgid_inherit compatibility variable is set).
1054 */
1059 }
1072 /*
1073 * di_gen will have been taken care of in xfs_iread.
1074 */
1091 /*
1092 * we can't set up filestreams until after the VFS inode
1093 * is set up properly.
1094 */
1097 /* fall through */
1108 }
1115 }
1116 }
1118 xfs_inherit_noatime)
1121 xfs_inherit_nodump)
1124 xfs_inherit_sync)
1127 xfs_inherit_nosymlinks)
1132 xfs_inherit_nodefrag)
1137 }
1138 /* FALLTHROUGH */
1147 }
1148 /*
1149 * Attribute fork settings for new inode.
1150 */
1154 /*
1155 * Log the new values stuffed into the inode.
1156 */
1160 /* now that we have an i_mode we can setup inode ops and unlock */
1163 /* now we have set up the vfs inode we can associate the filestream */
1170 }
1174 }
1176 /*
1177 * Check to make sure that there are no blocks allocated to the
1178 * file beyond the size of the file. We don't check this for
1179 * files with fixed size extents or real time extents, but we
1180 * at least do it for regular files.
1181 */
1182 #ifdef DEBUG
1183 void
1187 xfs_fsize_t isize)
1188 {
1189 xfs_fileoff_t map_first;
1204 /*
1205 * The filesystem could be shutting down, so bmapi may return
1206 * an error.
1207 */
1211 map_first),
1213 NULL))
1217 }
1220 /*
1221 * Calculate the last possible buffered byte in a file. This must
1222 * include data that was buffered beyond the EOF by the write code.
1223 * This also needs to deal with overflowing the xfs_fsize_t type
1224 * which can happen for sizes near the limit.
1225 *
1226 * We also need to take into account any blocks beyond the EOF. It
1227 * may be the case that they were buffered by a write which failed.
1228 * In that case the pages will still be in memory, but the inode size
1229 * will never have been updated.
1230 */
1231 STATIC xfs_fsize_t
1234 {
1236 xfs_fsize_t last_byte;
1237 xfs_fileoff_t last_block;
1238 xfs_fileoff_t size_last_block;
1244 /*
1245 * Only check for blocks beyond the EOF if the extents have
1246 * been read in. This eliminates the need for the inode lock,
1247 * and it also saves us from looking when it really isn't
1248 * necessary.
1249 */
1253 XFS_DATA_FORK);
1257 }
1260 }
1267 }
1271 }
1273 }
1275 /*
1276 * Start the truncation of the file to new_size. The new size
1277 * must be smaller than the current size. This routine will
1278 * clear the buffer and page caches of file data in the removed
1279 * range, and xfs_itruncate_finish() will remove the underlying
1280 * disk blocks.
1281 *
1282 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1283 * must NOT have the inode lock held at all. This is because we're
1284 * calling into the buffer/page cache code and we can't hold the
1285 * inode lock when we do so.
1286 *
1287 * We need to wait for any direct I/Os in flight to complete before we
1288 * proceed with the truncate. This is needed to prevent the extents
1289 * being read or written by the direct I/Os from being removed while the
1290 * I/O is in flight as there is no other method of synchronising
1291 * direct I/O with the truncate operation. Also, because we hold
1292 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1293 * started until the truncate completes and drops the lock. Essentially,
1294 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1295 * ordering between direct I/Os and the truncate operation.
1296 *
1297 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1298 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1299 * in the case that the caller is locking things out of order and
1300 * may not be able to call xfs_itruncate_finish() with the inode lock
1301 * held without dropping the I/O lock. If the caller must drop the
1302 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1303 * must be called again with all the same restrictions as the initial
1304 * call.
1305 */
1306 int
1309 uint flags,
1310 xfs_fsize_t new_size)
1311 {
1312 xfs_fsize_t last_byte;
1313 xfs_off_t toss_start;
1324 /* wait for the completion of any pending DIOs */
1328 /*
1329 * Call toss_pages or flushinval_pages to get rid of pages
1330 * overlapping the region being removed. We have to use
1331 * the less efficient flushinval_pages in the case that the
1332 * caller may not be able to finish the truncate without
1333 * dropping the inode's I/O lock. Make sure
1334 * to catch any pages brought in by buffers overlapping
1335 * the EOF by searching out beyond the isize by our
1336 * block size. We round new_size up to a block boundary
1337 * so that we don't toss things on the same block as
1338 * new_size but before it.
1339 *
1340 * Before calling toss_page or flushinval_pages, make sure to
1341 * call remapf() over the same region if the file is mapped.
1342 * This frees up mapped file references to the pages in the
1343 * given range and for the flushinval_pages case it ensures
1344 * that we get the latest mapped changes flushed out.
1345 */
1349 /*
1350 * The place to start tossing is beyond our maximum
1351 * file size, so there is no way that the data extended
1352 * out there.
1353 */
1355 }
1365 }
1366 }
1368 #ifdef DEBUG
1371 }
1372 #endif
1374 }
1376 /*
1377 * Shrink the file to the given new_size. The new size must be smaller than
1378 * the current size. This will free up the underlying blocks in the removed
1379 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1380 *
1381 * The transaction passed to this routine must have made a permanent log
1382 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1383 * given transaction and start new ones, so make sure everything involved in
1384 * the transaction is tidy before calling here. Some transaction will be
1385 * returned to the caller to be committed. The incoming transaction must
1386 * already include the inode, and both inode locks must be held exclusively.
1387 * The inode must also be "held" within the transaction. On return the inode
1388 * will be "held" within the returned transaction. This routine does NOT
1389 * require any disk space to be reserved for it within the transaction.
1390 *
1391 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1392 * indicates the fork which is to be truncated. For the attribute fork we only
1393 * support truncation to size 0.
1394 *
1395 * We use the sync parameter to indicate whether or not the first transaction
1396 * we perform might have to be synchronous. For the attr fork, it needs to be
1397 * so if the unlink of the inode is not yet known to be permanent in the log.
1398 * This keeps us from freeing and reusing the blocks of the attribute fork
1399 * before the unlink of the inode becomes permanent.
1400 *
1401 * For the data fork, we normally have to run synchronously if we're being
1402 * called out of the inactive path or we're being called out of the create path
1403 * where we're truncating an existing file. Either way, the truncate needs to
1404 * be sync so blocks don't reappear in the file with altered data in case of a
1405 * crash. wsync filesystems can run the first case async because anything that
1406 * shrinks the inode has to run sync so by the time we're called here from
1407 * inactive, the inode size is permanently set to 0.
1408 *
1409 * Calls from the truncate path always need to be sync unless we're in a wsync
1410 * filesystem and the file has already been unlinked.
1411 *
1412 * The caller is responsible for correctly setting the sync parameter. It gets
1413 * too hard for us to guess here which path we're being called out of just
1414 * based on inode state.
1415 *
1416 * If we get an error, we must return with the inode locked and linked into the
1417 * current transaction. This keeps things simple for the higher level code,
1418 * because it always knows that the inode is locked and held in the transaction
1419 * that returns to it whether errors occur or not. We don't mark the inode
1420 * dirty on error so that transactions can be easily aborted if possible.
1421 */
1422 int
1426 xfs_fsize_t new_size,
1429 {
1430 xfs_fsblock_t first_block;
1431 xfs_fileoff_t first_unmap_block;
1432 xfs_fileoff_t last_block;
1438 xfs_bmap_free_t free_list;
1454 /*
1455 * We only support truncating the entire attribute fork.
1456 */
1459 }
1463 /*
1464 * The first thing we do is set the size to new_size permanently
1465 * on disk. This way we don't have to worry about anyone ever
1466 * being able to look at the data being freed even in the face
1467 * of a crash. What we're getting around here is the case where
1468 * we free a block, it is allocated to another file, it is written
1469 * to, and then we crash. If the new data gets written to the
1470 * file but the log buffers containing the free and reallocation
1471 * don't, then we'd end up with garbage in the blocks being freed.
1472 * As long as we make the new_size permanent before actually
1473 * freeing any blocks it doesn't matter if they get writtten to.
1474 *
1475 * The callers must signal into us whether or not the size
1476 * setting here must be synchronous. There are a few cases
1477 * where it doesn't have to be synchronous. Those cases
1478 * occur if the file is unlinked and we know the unlink is
1479 * permanent or if the blocks being truncated are guaranteed
1480 * to be beyond the inode eof (regardless of the link count)
1481 * and the eof value is permanent. Both of these cases occur
1482 * only on wsync-mounted filesystems. In those cases, we're
1483 * guaranteed that no user will ever see the data in the blocks
1484 * that are being truncated so the truncate can run async.
1485 * In the free beyond eof case, the file may wind up with
1486 * more blocks allocated to it than it needs if we crash
1487 * and that won't get fixed until the next time the file
1488 * is re-opened and closed but that's ok as that shouldn't
1489 * be too many blocks.
1490 *
1491 * However, we can't just make all wsync xactions run async
1492 * because there's one call out of the create path that needs
1493 * to run sync where it's truncating an existing file to size
1494 * 0 whose size is > 0.
1495 *
1496 * It's probably possible to come up with a test in this
1497 * routine that would correctly distinguish all the above
1498 * cases from the values of the function parameters and the
1499 * inode state but for sanity's sake, I've decided to let the
1500 * layers above just tell us. It's simpler to correctly figure
1501 * out in the layer above exactly under what conditions we
1502 * can run async and I think it's easier for others read and
1503 * follow the logic in case something has to be changed.
1504 * cscope is your friend -- rcc.
1505 *
1506 * The attribute fork is much simpler.
1507 *
1508 * For the attribute fork we allow the caller to tell us whether
1509 * the unlink of the inode that led to this call is yet permanent
1510 * in the on disk log. If it is not and we will be freeing extents
1511 * in this inode then we make the first transaction synchronous
1512 * to make sure that the unlink is permanent by the time we free
1513 * the blocks.
1514 */
1517 /*
1518 * If we are not changing the file size then do
1519 * not update the on-disk file size - we may be
1520 * called from xfs_inactive_free_eofblocks(). If we
1521 * update the on-disk file size and then the system
1522 * crashes before the contents of the file are
1523 * flushed to disk then the files may be full of
1524 * holes (ie NULL files bug).
1525 */
1530 }
1531 }
1536 }
1542 /*
1543 * Since it is possible for space to become allocated beyond
1544 * the end of the file (in a crash where the space is allocated
1545 * but the inode size is not yet updated), simply remove any
1546 * blocks which show up between the new EOF and the maximum
1547 * possible file size. If the first block to be removed is
1548 * beyond the maximum file size (ie it is the same as last_block),
1549 * then there is nothing to do.
1550 */
1558 }
1560 /*
1561 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1562 * will tell us whether it freed the entire range or
1563 * not. If this is a synchronous mount (wsync),
1564 * then we can tell bunmapi to keep all the
1565 * transactions asynchronous since the unlink
1566 * transaction that made this inode inactive has
1567 * already hit the disk. There's no danger of
1568 * the freed blocks being reused, there being a
1569 * crash, and the reused blocks suddenly reappearing
1570 * in this file with garbage in them once recovery
1571 * runs.
1572 */
1577 XFS_ITRUNC_MAX_EXTENTS,
1581 /*
1582 * If the bunmapi call encounters an error,
1583 * return to the caller where the transaction
1584 * can be properly aborted. We just need to
1585 * make sure we're not holding any resources
1586 * that we were not when we came in.
1587 */
1590 }
1592 /*
1593 * Duplicate the transaction that has the permanent
1594 * reservation and commit the old transaction.
1595 */
1602 /*
1603 * If the bmap finish call encounters an error, return
1604 * to the caller where the transaction can be properly
1605 * aborted. We just need to make sure we're not
1606 * holding any resources that we were not when we came
1607 * in.
1608 *
1609 * Aborting from this point might lose some blocks in
1610 * the file system, but oh well.
1611 */
1614 }
1617 /*
1618 * Mark the inode dirty so it will be logged and
1619 * moved forward in the log as part of every commit.
1620 */
1622 }
1632 /*
1633 * transaction commit worked ok so we can drop the extra ticket
1634 * reference that we gained in xfs_trans_dup()
1635 */
1639 XFS_TRANS_PERM_LOG_RES,
1640 XFS_ITRUNCATE_LOG_COUNT);
1643 }
1644 /*
1645 * Only update the size in the case of the data fork, but
1646 * always re-log the inode so that our permanent transaction
1647 * can keep on rolling it forward in the log.
1648 */
1651 /*
1652 * If we are not changing the file size then do
1653 * not update the on-disk file size - we may be
1654 * called from xfs_inactive_free_eofblocks(). If we
1655 * update the on-disk file size and then the system
1656 * crashes before the contents of the file are
1657 * flushed to disk then the files may be full of
1658 * holes (ie NULL files bug).
1659 */
1663 }
1664 }
1674 }
1676 /*
1677 * This is called when the inode's link count goes to 0.
1678 * We place the on-disk inode on a list in the AGI. It
1679 * will be pulled from this list when the inode is freed.
1680 */
1681 int
1685 {
1691 xfs_agino_t agino;
1702 /*
1703 * Get the agi buffer first. It ensures lock ordering
1704 * on the list.
1705 */
1711 /*
1712 * Get the index into the agi hash table for the
1713 * list this inode will go on.
1714 */
1722 /*
1723 * There is already another inode in the bucket we need
1724 * to add ourselves to. Add us at the front of the list.
1725 * Here we put the head pointer into our next pointer,
1726 * and then we fall through to point the head at us.
1727 */
1733 /* both on-disk, don't endian flip twice */
1741 }
1743 /*
1744 * Point the bucket head pointer at the inode being inserted.
1745 */
1753 }
1755 /*
1756 * Pull the on-disk inode from the AGI unlinked list.
1757 */
1758 STATIC int
1762 {
1763 xfs_ino_t next_ino;
1769 xfs_agnumber_t agno;
1770 xfs_agino_t agino;
1771 xfs_agino_t next_agino;
1781 /*
1782 * Get the agi buffer first. It ensures lock ordering
1783 * on the list.
1784 */
1791 /*
1792 * Get the index into the agi hash table for the
1793 * list this inode will go on.
1794 */
1802 /*
1803 * We're at the head of the list. Get the inode's
1804 * on-disk buffer to see if there is anyone after us
1805 * on the list. Only modify our next pointer if it
1806 * is not already NULLAGINO. This saves us the overhead
1807 * of dealing with the buffer when there is no need to
1808 * change it.
1809 */
1815 }
1828 }
1829 /*
1830 * Point the bucket head pointer at the next inode.
1831 */
1840 /*
1841 * We need to search the list for the inode being freed.
1842 */
1846 /*
1847 * If the last inode wasn't the one pointing to
1848 * us, then release its buffer since we're not
1849 * going to do anything with it.
1850 */
1853 }
1862 }
1866 }
1867 /*
1868 * Now last_ibp points to the buffer previous to us on
1869 * the unlinked list. Pull us from the list.
1870 */
1876 }
1890 }
1891 /*
1892 * Point the previous inode on the list to the next inode.
1893 */
1901 }
1903 }
1905 /*
1906 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1907 * inodes that are in memory - they all must be marked stale and attached to
1908 * the cluster buffer.
1909 */
1910 STATIC void
1914 xfs_ino_t inum)
1915 {
1921 xfs_daddr_t blkno;
1938 }
1944 /*
1945 * We obtain and lock the backing buffer first in the process
1946 * here, as we have to ensure that any dirty inode that we
1947 * can't get the flush lock on is attached to the buffer.
1948 * If we scan the in-memory inodes first, then buffer IO can
1949 * complete before we get a lock on it, and hence we may fail
1950 * to mark all the active inodes on the buffer stale.
1951 */
1954 XBF_LOCK);
1956 /*
1957 * Walk the inodes already attached to the buffer and mark them
1958 * stale. These will all have the flush locks held, so an
1959 * in-memory inode walk can't lock them. By marking them all
1960 * stale first, we will not attempt to lock them in the loop
1961 * below as the XFS_ISTALE flag will be set.
1962 */
1973 }
1975 }
1978 /*
1979 * For each inode in memory attempt to add it to the inode
1980 * buffer and set it up for being staled on buffer IO
1981 * completion. This is safe as we've locked out tail pushing
1982 * and flushing by locking the buffer.
1983 *
1984 * We have already marked every inode that was part of a
1985 * transaction stale above, which means there is no point in
1986 * even trying to lock them.
1987 */
1989 retry:
1994 /* Inode not in memory, nothing to do */
1998 }
2000 /*
2001 * because this is an RCU protected lookup, we could
2002 * find a recently freed or even reallocated inode
2003 * during the lookup. We need to check under the
2004 * i_flags_lock for a valid inode here. Skip it if it
2005 * is not valid, the wrong inode or stale.
2006 */
2013 }
2016 /*
2017 * Don't try to lock/unlock the current inode, but we
2018 * _cannot_ skip the other inodes that we did not find
2019 * in the list attached to the buffer and are not
2020 * already marked stale. If we can't lock it, back off
2021 * and retry.
2022 */
2028 }
2034 /*
2035 * we don't need to attach clean inodes or those only
2036 * with unlogged changes (which we throw away, anyway).
2037 */
2045 }
2058 }
2062 }
2065 }
2067 /*
2068 * This is called to return an inode to the inode free list.
2069 * The inode should already be truncated to 0 length and have
2070 * no pages associated with it. This routine also assumes that
2071 * the inode is already a part of the transaction.
2072 *
2073 * The on-disk copy of the inode will have been added to the list
2074 * of unlinked inodes in the AGI. We need to remove the inode from
2075 * that list atomically with respect to freeing it here.
2076 */
2077 int
2082 {
2085 xfs_ino_t first_ino;
2098 /*
2099 * Pull the on-disk inode from the AGI unlinked list.
2100 */
2104 }
2109 }
2118 /*
2119 * Bump the generation count so no one will be confused
2120 * by reincarnations of this inode.
2121 */
2130 /*
2131 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2132 * from picking up this inode when it is reclaimed (its incore state
2133 * initialzed but not flushed to disk yet). The in-core di_mode is
2134 * already cleared and a corresponding transaction logged.
2135 * The hack here just synchronizes the in-core to on-disk
2136 * di_mode value in advance before the actual inode sync to disk.
2137 * This is OK because the inode is already unlinked and would never
2138 * change its di_mode again for this inode generation.
2139 * This is a temporary hack that would require a proper fix
2140 * in the future.
2141 */
2146 }
2149 }
2151 /*
2152 * Reallocate the space for if_broot based on the number of records
2153 * being added or deleted as indicated in rec_diff. Move the records
2154 * and pointers in if_broot to fit the new size. When shrinking this
2155 * will eliminate holes between the records and pointers created by
2156 * the caller. When growing this will create holes to be filled in
2157 * by the caller.
2158 *
2159 * The caller must not request to add more records than would fit in
2160 * the on-disk inode root. If the if_broot is currently NULL, then
2161 * if we adding records one will be allocated. The caller must also
2162 * not request that the number of records go below zero, although
2163 * it can go to zero.
2164 *
2165 * ip -- the inode whose if_broot area is changing
2166 * ext_diff -- the change in the number of records, positive or negative,
2167 * requested for the if_broot array.
2168 */
2169 void
2174 {
2184 /*
2185 * Handle the degenerate case quietly.
2186 */
2189 }
2193 /*
2194 * If there wasn't any memory allocated before, just
2195 * allocate it now and get out.
2196 */
2202 }
2204 /*
2205 * If there is already an existing if_broot, then we need
2206 * to realloc() it and shift the pointers to their new
2207 * location. The records don't change location because
2208 * they are kept butted up against the btree block header.
2209 */
2225 }
2227 /*
2228 * rec_diff is less than 0. In this case, we are shrinking the
2229 * if_broot buffer. It must already exist. If we go to zero
2230 * records, just get rid of the root and clear the status bit.
2231 */
2238 else
2242 /*
2243 * First copy over the btree block header.
2244 */
2249 }
2251 /*
2252 * Only copy the records and pointers if there are any.
2253 */
2255 /*
2256 * First copy the records.
2257 */
2262 /*
2263 * Then copy the pointers.
2264 */
2270 }
2277 }
2280 /*
2281 * This is called when the amount of space needed for if_data
2282 * is increased or decreased. The change in size is indicated by
2283 * the number of bytes that need to be added or deleted in the
2284 * byte_diff parameter.
2285 *
2286 * If the amount of space needed has decreased below the size of the
2287 * inline buffer, then switch to using the inline buffer. Otherwise,
2288 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2289 * to what is needed.
2290 *
2291 * ip -- the inode whose if_data area is changing
2292 * byte_diff -- the change in the number of bytes, positive or negative,
2293 * requested for the if_data array.
2294 */
2295 void
2300 {
2307 }
2316 }
2320 /*
2321 * If the valid extents/data can fit in if_inline_ext/data,
2322 * copy them from the malloc'd vector and free it.
2323 */
2329 new_size);
2332 }
2335 /*
2336 * Stuck with malloc/realloc.
2337 * For inline data, the underlying buffer must be
2338 * a multiple of 4 bytes in size so that it can be
2339 * logged and stay on word boundaries. We enforce
2340 * that here.
2341 */
2348 /*
2349 * Only do the realloc if the underlying size
2350 * is really changing.
2351 */
2355 real_size,
2358 }
2365 }
2366 }
2370 }
2372 void
2376 {
2383 }
2385 /*
2386 * If the format is local, then we can't have an extents
2387 * array so just look for an inline data array. If we're
2388 * not local then we may or may not have an extents list,
2389 * so check and free it up if we do.
2390 */
2398 }
2405 }
2412 }
2413 }
2415 /*
2416 * This is called to unpin an inode. The caller must have the inode locked
2417 * in at least shared mode so that the buffer cannot be subsequently pinned
2418 * once someone is waiting for it to be unpinned.
2419 */
2420 static void
2423 {
2428 /* Give the log a push to start the unpinning I/O */
2431 }
2433 void
2436 {
2440 }
2441 }
2443 /*
2444 * xfs_iextents_copy()
2445 *
2446 * This is called to copy the REAL extents (as opposed to the delayed
2447 * allocation extents) from the inode into the given buffer. It
2448 * returns the number of bytes copied into the buffer.
2449 *
2450 * If there are no delayed allocation extents, then we can just
2451 * memcpy() the extents into the buffer. Otherwise, we need to
2452 * examine each extent in turn and skip those which are delayed.
2453 */
2454 int
2459 {
2464 xfs_fsblock_t start_block;
2474 /*
2475 * There are some delayed allocation extents in the
2476 * inode, so copy the extents one at a time and skip
2477 * the delayed ones. There must be at least one
2478 * non-delayed extent.
2479 */
2485 /*
2486 * It's a delayed allocation extent, so skip it.
2487 */
2489 }
2491 /* Translate to on disk format */
2494 dp++;
2495 copied++;
2496 }
2501 }
2503 /*
2504 * Each of the following cases stores data into the same region
2505 * of the on-disk inode, so only one of them can be valid at
2506 * any given time. While it is possible to have conflicting formats
2507 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2508 * in EXTENTS format, this can only happen when the fork has
2509 * changed formats after being modified but before being flushed.
2510 * In these cases, the format always takes precedence, because the
2511 * format indicates the current state of the fork.
2512 */
2513 /*ARGSUSED*/
2514 STATIC void
2521 {
2525 #ifdef XFS_TRANS_DEBUG
2527 #endif
2538 /*
2539 * This can happen if we gave up in iformat in an error path,
2540 * for the attribute fork.
2541 */
2545 }
2555 }
2569 whichfork);
2570 }
2579 XFS_BROOT_SIZE_ADJ));
2583 }
2590 }
2599 }
2605 }
2606 }
2608 STATIC int
2612 {
2636 /* really need a gang lookup range call here */
2647 /*
2648 * because this is an RCU protected lookup, we could find a
2649 * recently freed or even reallocated inode during the lookup.
2650 * We need to check under the i_flags_lock for a valid inode
2651 * here. Skip it if it is not valid or the wrong inode.
2652 */
2658 }
2661 /*
2662 * Do an un-protected check to see if the inode is dirty and
2663 * is a candidate for flushing. These checks will be repeated
2664 * later after the appropriate locks are acquired.
2665 */
2669 /*
2670 * Try to get locks. If any are unavailable or it is pinned,
2671 * then this inode cannot be flushed and is skipped.
2672 */
2679 }
2684 }
2686 /*
2687 * arriving here means that this inode can be flushed. First
2688 * re-check that it's dirty before flushing.
2689 */
2696 }
2697 clcount++;
2700 }
2702 }
2707 }
2709 out_free:
2712 out_put:
2717 cluster_corrupt_out:
2718 /*
2719 * Corruption detected in the clustering loop. Invalidate the
2720 * inode buffer and shut down the filesystem.
2721 */
2723 /*
2724 * Clean up the buffer. If it was B_DELWRI, just release it --
2725 * brelse can handle it with no problems. If not, shut down the
2726 * filesystem before releasing the buffer.
2727 */
2735 /*
2736 * Just like incore_relse: if we have b_iodone functions,
2737 * mark the buffer as an error and call them. Otherwise
2738 * mark it as stale and brelse.
2739 */
2748 }
2749 }
2751 /*
2752 * Unlocks the flush lock
2753 */
2758 }
2760 /*
2761 * xfs_iflush() will write a modified inode's changes out to the
2762 * inode's on disk home. The caller must have the inode lock held
2763 * in at least shared mode and the inode flush completion must be
2764 * active as well. The inode lock will still be held upon return from
2765 * the call and the caller is free to unlock it.
2766 * The inode flush will be completed when the inode reaches the disk.
2767 * The flags indicate how the inode's buffer should be written out.
2768 */
2769 int
2772 uint flags)
2773 {
2790 /*
2791 * We can't flush the inode until it is unpinned, so wait for it if we
2792 * are allowed to block. We know noone new can pin it, because we are
2793 * holding the inode lock shared and you need to hold it exclusively to
2794 * pin the inode.
2795 *
2796 * If we are not allowed to block, force the log out asynchronously so
2797 * that when we come back the inode will be unpinned. If other inodes
2798 * in the same cluster are dirty, they will probably write the inode
2799 * out for us if they occur after the log force completes.
2800 */
2805 }
2808 /*
2809 * For stale inodes we cannot rely on the backing buffer remaining
2810 * stale in cache for the remaining life of the stale inode and so
2811 * xfs_itobp() below may give us a buffer that no longer contains
2812 * inodes below. We have to check this after ensuring the inode is
2813 * unpinned so that it is safe to reclaim the stale inode after the
2814 * flush call.
2815 */
2819 }
2821 /*
2822 * This may have been unpinned because the filesystem is shutting
2823 * down forcibly. If that's the case we must not write this inode
2824 * to disk, because the log record didn't make it to disk!
2825 */
2832 }
2834 /*
2835 * Get the buffer containing the on-disk inode.
2836 */
2842 }
2844 /*
2845 * First flush out the inode that xfs_iflush was called with.
2846 */
2851 /*
2852 * If the buffer is pinned then push on the log now so we won't
2853 * get stuck waiting in the write for too long.
2854 */
2858 /*
2859 * inode clustering:
2860 * see if other inodes can be gathered into this write
2861 */
2868 else
2872 corrupt_out:
2875 cluster_corrupt_out:
2876 /*
2877 * Unlocks the flush lock
2878 */
2881 }
2884 STATIC int
2888 {
2892 #ifdef XFS_TRANS_DEBUG
2894 #endif
2904 /* set *dip = inode's place in the buffer */
2907 /*
2908 * Clear i_update_core before copying out the data.
2909 * This is for coordination with our timestamp updates
2910 * that don't hold the inode lock. They will always
2911 * update the timestamps BEFORE setting i_update_core,
2912 * so if we clear i_update_core after they set it we
2913 * are guaranteed to see their updates to the timestamps.
2914 * I believe that this depends on strongly ordered memory
2915 * semantics, but we have that. We use the SYNCHRONIZE
2916 * macro to make sure that the compiler does not reorder
2917 * the i_update_core access below the data copy below.
2918 */
2922 /*
2923 * Make sure to get the latest timestamps from the Linux inode.
2924 */
2933 }
2940 }
2950 }
2961 }
2962 }
2965 XFS_RANDOM_IFLUSH_5)) {
2967 "%s: detected corrupt incore inode %Lu, "
2973 }
2980 }
2981 /*
2982 * bump the flush iteration count, used to detect flushes which
2983 * postdate a log record during recovery.
2984 */
2988 /*
2989 * Copy the dirty parts of the inode into the on-disk
2990 * inode. We always copy out the core of the inode,
2991 * because if the inode is dirty at all the core must
2992 * be.
2993 */
2996 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3000 /*
3001 * If this is really an old format inode and the superblock version
3002 * has not been updated to support only new format inodes, then
3003 * convert back to the old inode format. If the superblock version
3004 * has been updated, then make the conversion permanent.
3005 */
3009 /*
3010 * Convert it back.
3011 */
3015 /*
3016 * The superblock version has already been bumped,
3017 * so just make the conversion to the new inode
3018 * format permanent.
3019 */
3028 }
3029 }
3036 /*
3037 * We've recorded everything logged in the inode, so we'd
3038 * like to clear the ilf_fields bits so we don't log and
3039 * flush things unnecessarily. However, we can't stop
3040 * logging all this information until the data we've copied
3041 * into the disk buffer is written to disk. If we did we might
3042 * overwrite the copy of the inode in the log with all the
3043 * data after re-logging only part of it, and in the face of
3044 * a crash we wouldn't have all the data we need to recover.
3045 *
3046 * What we do is move the bits to the ili_last_fields field.
3047 * When logging the inode, these bits are moved back to the
3048 * ilf_fields field. In the xfs_iflush_done() routine we
3049 * clear ili_last_fields, since we know that the information
3050 * those bits represent is permanently on disk. As long as
3051 * the flush completes before the inode is logged again, then
3052 * both ilf_fields and ili_last_fields will be cleared.
3053 *
3054 * We can play with the ilf_fields bits here, because the inode
3055 * lock must be held exclusively in order to set bits there
3056 * and the flush lock protects the ili_last_fields bits.
3057 * Set ili_logged so the flush done
3058 * routine can tell whether or not to look in the AIL.
3059 * Also, store the current LSN of the inode so that we can tell
3060 * whether the item has moved in the AIL from xfs_iflush_done().
3061 * In order to read the lsn we need the AIL lock, because
3062 * it is a 64 bit value that cannot be read atomically.
3063 */
3072 /*
3073 * Attach the function xfs_iflush_done to the inode's
3074 * buffer. This will remove the inode from the AIL
3075 * and unlock the inode's flush lock when the inode is
3076 * completely written to disk.
3077 */
3083 /*
3084 * We're flushing an inode which is not in the AIL and has
3085 * not been logged but has i_update_core set. For this
3086 * case we can use a B_DELWRI flush and immediately drop
3087 * the inode flush lock because we can avoid the whole
3088 * AIL state thing. It's OK to drop the flush lock now,
3089 * because we've already locked the buffer and to do anything
3090 * you really need both.
3091 */
3096 }
3098 }
3102 corrupt_out:
3104 }
3106 /*
3107 * Return a pointer to the extent record at file index idx.
3108 */
3109 xfs_bmbt_rec_host_t *
3113 {
3128 }
3129 }
3131 /*
3132 * Insert new item(s) into the extent records for incore inode
3133 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3134 */
3135 void
3142 {
3152 }
3154 /*
3155 * This is called when the amount of space required for incore file
3156 * extents needs to be increased. The ext_diff parameter stores the
3157 * number of new extents being added and the idx parameter contains
3158 * the extent index where the new extents will be added. If the new
3159 * extents are being appended, then we just need to (re)allocate and
3160 * initialize the space. Otherwise, if the new extents are being
3161 * inserted into the middle of the existing entries, a bit more work
3162 * is required to make room for the new extents to be inserted. The
3163 * caller is responsible for filling in the new extent entries upon
3164 * return.
3165 */
3166 void
3171 {
3180 /*
3181 * If the new number of extents (nextents + ext_diff)
3182 * fits inside the inode, then continue to use the inline
3183 * extent buffer.
3184 */
3191 }
3195 }
3196 /*
3197 * Otherwise use a linear (direct) extent list.
3198 * If the extents are currently inside the inode,
3199 * xfs_iext_realloc_direct will switch us from
3200 * inline to direct extent allocation mode.
3201 */
3209 }
3210 }
3211 /* Indirection array */
3224 }
3225 /* Extents fit in target extent page */
3233 }
3236 }
3237 /* Insert a new extent page */
3241 }
3242 /*
3243 * If extent(s) are being appended to the last page in
3244 * the indirection array and the new extent(s) don't fit
3245 * in the page, then erp is NULL and erp_idx is set to
3246 * the next index needed in the indirection array.
3247 */
3256 erp_idx++;
3257 }
3258 }
3259 }
3260 }
3262 }
3264 /*
3265 * This is called when incore extents are being added to the indirection
3266 * array and the new extents do not fit in the target extent list. The
3267 * erp_idx parameter contains the irec index for the target extent list
3268 * in the indirection array, and the idx parameter contains the extent
3269 * index within the list. The number of extents being added is stored
3270 * in the count parameter.
3271 *
3272 * |-------| |-------|
3273 * | | | | idx - number of extents before idx
3274 * | idx | | count |
3275 * | | | | count - number of extents being inserted at idx
3276 * |-------| |-------|
3277 * | count | | nex2 | nex2 - number of extents after idx + count
3278 * |-------| |-------|
3279 */
3280 void
3286 {
3300 /*
3301 * Save second part of target extent list
3302 * (all extents past */
3310 }
3312 /*
3313 * Add the new extents to the end of the target
3314 * list, then allocate new irec record(s) and
3315 * extent buffer(s) as needed to store the rest
3316 * of the new extents.
3317 */
3324 }
3326 erp_idx++;
3332 }
3334 /* Add nex2 extents back to indirection array */
3336 xfs_extnum_t ext_avail;
3342 /*
3343 * If nex2 extents fit in the current page, append
3344 * nex2_ep after the new extents.
3345 */
3348 }
3349 /*
3350 * Otherwise, check if space is available in the
3351 * next page.
3352 */
3356 erp_idx++;
3357 erp++;
3358 /* Create a hole for nex2 extents */
3361 }
3362 /*
3363 * Final choice, create a new extent page for
3364 * nex2 extents.
3365 */
3367 erp_idx++;
3369 }
3374 }
3375 }
3377 /*
3378 * This is called when the amount of space required for incore file
3379 * extents needs to be decreased. The ext_diff parameter stores the
3380 * number of extents to be removed and the idx parameter contains
3381 * the extent index where the extents will be removed from.
3382 *
3383 * If the amount of space needed has decreased below the linear
3384 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3385 * extent array. Otherwise, use kmem_realloc() to adjust the
3386 * size to what is needed.
3387 */
3388 void
3394 {
3413 }
3415 }
3417 /*
3418 * This removes ext_diff extents from the inline buffer, beginning
3419 * at extent index idx.
3420 */
3421 void
3426 {
3445 }
3446 }
3448 /*
3449 * This removes ext_diff extents from a linear (direct) extent list,
3450 * beginning at extent index idx. If the extents are being removed
3451 * from the end of the list (ie. truncate) then we just need to re-
3452 * allocate the list to remove the extra space. Otherwise, if the
3453 * extents are being removed from the middle of the existing extent
3454 * entries, then we first need to move the extent records beginning
3455 * at idx + ext_diff up in the list to overwrite the records being
3456 * removed, then remove the extra space via kmem_realloc.
3457 */
3458 void
3463 {
3475 }
3476 /* Move extents up in the list (if needed) */
3482 }
3485 /*
3486 * Reallocate the direct extent list. If the extents
3487 * will fit inside the inode then xfs_iext_realloc_direct
3488 * will switch from direct to inline extent allocation
3489 * mode for us.
3490 */
3493 }
3495 /*
3496 * This is called when incore extents are being removed from the
3497 * indirection array and the extents being removed span multiple extent
3498 * buffers. The idx parameter contains the file extent index where we
3499 * want to begin removing extents, and the count parameter contains
3500 * how many extents need to be removed.
3501 *
3502 * |-------| |-------|
3503 * | nex1 | | | nex1 - number of extents before idx
3504 * |-------| | count |
3505 * | | | | count - number of extents being removed at idx
3506 * | count | |-------|
3507 * | | | nex2 | nex2 - number of extents after idx + count
3508 * |-------| |-------|
3509 */
3510 void
3515 {
3532 /*
3533 * Check for deletion of entire list;
3534 * xfs_iext_irec_remove() updates extent offsets.
3535 */
3542 XFS_IEXT_BUFSZ);
3548 }
3549 }
3550 /* Move extents up (if needed) */
3555 }
3556 /* Zero out rest of page */
3559 /* Update remaining counters */
3564 erp_idx++;
3565 erp++;
3566 }
3569 }
3571 /*
3572 * Create, destroy, or resize a linear (direct) block of extents.
3573 */
3574 void
3578 {
3587 /* Free extent records */
3590 }
3591 /* Resize direct extent list and zero any new bytes */
3593 /* Check if extents will fit inside the inode */
3599 }
3602 }
3606 rnew_size,
3608 }
3613 }
3614 }
3615 /*
3616 * Switch from the inline extent buffer to a direct
3617 * extent list. Be sure to include the inline extent
3618 * bytes in new_size.
3619 */
3624 }
3626 }
3629 }
3631 /*
3632 * Switch from linear (direct) extent records to inline buffer.
3633 */
3634 void
3638 {
3641 /*
3642 * The inline buffer was zeroed when we switched
3643 * from inline to direct extent allocation mode,
3644 * so we don't need to clear it here.
3645 */
3651 }
3653 /*
3654 * Switch from inline buffer to linear (direct) extent records.
3655 * new_size should already be rounded up to the next power of 2
3656 * by the caller (when appropriate), so use new_size as it is.
3657 * However, since new_size may be rounded up, we can't update
3658 * if_bytes here. It is the caller's responsibility to update
3659 * if_bytes upon return.
3660 */
3661 void
3665 {
3673 }
3675 }
3677 /*
3678 * Resize an extent indirection array to new_size bytes.
3679 */
3680 STATIC void
3684 {
3699 }
3700 }
3702 /*
3703 * Switch from indirection array to linear (direct) extent allocations.
3704 */
3705 STATIC void
3708 {
3728 }
3729 }
3731 /*
3732 * Free incore file extents.
3733 */
3734 void
3737 {
3745 }
3752 }
3756 }
3758 /*
3759 * Return a pointer to the extent record for file system block bno.
3760 */
3766 {
3781 }
3784 /* Find target extent list */
3792 }
3793 /* Binary search extent records */
3804 /* Convert back to file-based extent index */
3807 }
3810 }
3811 }
3812 /* Convert back to file-based extent index */
3815 }
3821 }
3822 }
3825 }
3827 /*
3828 * Return a pointer to the indirection array entry containing the
3829 * extent record for filesystem block bno. Store the index of the
3830 * target irec in *erp_idxp.
3831 */
3837 {
3861 }
3862 }
3865 }
3867 /*
3868 * Return a pointer to the indirection array entry containing the
3869 * extent record at file extent index *idxp. Store the index of the
3870 * target irec in *erp_idxp and store the page index of the target
3871 * extent record in *idxp.
3872 */
3873 xfs_ext_irec_t *
3879 {
3896 /* Binary search extent irec's */
3912 erp_idx++;
3918 }
3919 }
3923 }
3925 /*
3926 * Allocate and initialize an indirection array once the space needed
3927 * for incore extents increases above XFS_IEXT_BUFSZ.
3928 */
3929 void
3932 {
3948 }
3959 }
3961 /*
3962 * Allocate and initialize a new entry in the indirection array.
3963 */
3964 xfs_ext_irec_t *
3968 {
3976 /* Resize indirection array */
3979 /*
3980 * Move records down in the array so the
3981 * new page can use erp_idx.
3982 */
3986 }
3989 /* Initialize new extent record */
3998 }
4000 /*
4001 * Remove a record from the indirection array.
4002 */
4003 void
4007 {
4019 }
4020 /* Compact extent records */
4024 }
4025 /*
4026 * Manually free the last extent record from the indirection
4027 * array. A call to xfs_iext_realloc_indirect() with a size
4028 * of zero would result in a call to xfs_iext_destroy() which
4029 * would in turn call this function again, creating a nasty
4030 * infinite loop.
4031 */
4037 }
4039 }
4041 /*
4042 * This is called to clean up large amounts of unused memory allocated
4043 * by the indirection array. Before compacting anything though, verify
4044 * that the indirection array is still needed and switch back to the
4045 * linear extent list (or even the inline buffer) if possible. The
4046 * compaction policy is as follows:
4047 *
4048 * Full Compaction: Extents fit into a single page (or inline buffer)
4049 * Partial Compaction: Extents occupy less than 50% of allocated space
4050 * No Compaction: Extents occupy at least 50% of allocated space
4051 */
4052 void
4055 {
4072 }
4073 }
4075 /*
4076 * Combine extents from neighboring extent pages.
4077 */
4078 void
4081 {
4097 /*
4098 * Free page before removing extent record
4099 * so er_extoffs don't get modified in
4100 * xfs_iext_irec_remove.
4101 */
4107 erp_idx++;
4108 }
4109 }
4110 }
4112 /*
4113 * This is called to update the er_extoff field in the indirection
4114 * array when extents have been added or removed from one of the
4115 * extent lists. erp_idx contains the irec index to begin updating
4116 * at and ext_diff contains the number of extents that were added
4117 * or removed.
4118 */
4119 void
4124 {
4132 }
4133 }