| blob | b92a4fa2a0a12eda161c2ff897bd820f5cbde1e8 |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include <linux/log2.h>
58 /*
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
61 */
62 #define XFS_ITRUNC_MAX_EXTENTS 2
69 #ifdef DEBUG
70 /*
71 * Make sure that the extents in the given memory buffer
72 * are valid.
73 */
74 STATIC void
78 xfs_exntfmt_t fmt)
79 {
80 xfs_bmbt_irec_t irec;
81 xfs_bmbt_rec_host_t rec;
91 }
92 }
94 #define xfs_validate_extents(ifp, nrecs, fmt)
97 /*
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
100 */
101 #if defined(DEBUG)
102 void
106 {
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
119 bp);
121 }
122 }
123 }
124 #endif
126 /*
127 * Find the buffer associated with the given inode map
128 * We do basic validation checks on the buffer once it has been
129 * retrieved from disk.
130 */
131 STATIC int
137 uint buf_flags,
138 uint iget_flags)
139 {
150 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
155 }
157 }
159 /*
160 * Validate the magic number and version of every inode in the buffer
161 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
162 */
163 #ifdef DEBUG
167 #endif
178 XFS_ERRTAG_ITOBP_INOTOBP,
179 XFS_RANDOM_ITOBP_INOTOBP))) {
183 }
186 #ifdef DEBUG
188 "Device %s - bad inode magic/vsn "
193 #endif
196 }
197 }
201 /*
202 * Mark the buffer as an inode buffer now that it looks good
203 */
208 }
210 /*
211 * This routine is called to map an inode number within a file
212 * system to the buffer containing the on-disk version of the
213 * inode. It returns a pointer to the buffer containing the
214 * on-disk inode in the bpp parameter, and in the dip parameter
215 * it returns a pointer to the on-disk inode within that buffer.
216 *
217 * If a non-zero error is returned, then the contents of bpp and
218 * dipp are undefined.
219 *
220 * Use xfs_imap() to determine the size and location of the
221 * buffer to read from disk.
222 */
223 int
227 xfs_ino_t ino,
231 uint imap_flags)
232 {
250 }
253 /*
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
258 * that buffer.
259 *
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
262 *
263 * The inode is expected to already been mapped to its buffer and read
264 * in once, thus we can use the mapping information stored in the inode
265 * rather than calling xfs_imap(). This allows us to avoid the overhead
266 * of looking at the inode btree for small block file systems
267 * (see xfs_imap()).
268 */
269 int
276 uint buf_flags)
277 {
292 }
297 }
299 /*
300 * Move inode type and inode format specific information from the
301 * on-disk inode to the in-core inode. For fifos, devs, and sockets
302 * this means set if_rdev to the proper value. For files, directories,
303 * and symlinks this means to bring in the in-line data or extent
304 * pointers. For a file in B-tree format, only the root is immediately
305 * brought in-core. The rest will be in-lined in if_extents when it
306 * is first referenced (see xfs_iread_extents()).
307 */
308 STATIC int
312 {
316 xfs_fsize_t di_size;
334 }
344 }
354 }
365 }
376 /*
377 * no local regular files yet
378 */
381 "corrupt inode %Lu "
385 XFS_ERRLEVEL_LOW,
388 }
393 "corrupt inode %Lu "
398 XFS_ERRLEVEL_LOW,
401 }
416 }
422 }
425 }
439 "corrupt inode %Lu "
444 XFS_ERRLEVEL_LOW,
447 }
460 }
465 }
467 }
469 /*
470 * The file is in-lined in the on-disk inode.
471 * If it fits into if_inline_data, then copy
472 * it there, otherwise allocate a buffer for it
473 * and copy the data there. Either way, set
474 * if_data to point at the data.
475 * If we allocate a buffer for the data, make
476 * sure that its size is a multiple of 4 and
477 * record the real size in i_real_bytes.
478 */
479 STATIC int
485 {
489 /*
490 * If the size is unreasonable, then something
491 * is wrong and we just bail out rather than crash in
492 * kmem_alloc() or memcpy() below.
493 */
496 "corrupt inode %Lu "
503 }
513 }
521 }
523 /*
524 * The file consists of a set of extents all
525 * of which fit into the on-disk inode.
526 * If there are few enough extents to fit into
527 * the if_inline_ext, then copy them there.
528 * Otherwise allocate a buffer for them and copy
529 * them into it. Either way, set if_extents
530 * to point at the extents.
531 */
532 STATIC int
537 {
548 /*
549 * If the number of extents is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
552 */
560 }
567 else
578 }
585 XFS_ERRLEVEL_LOW,
588 }
589 }
592 }
594 /*
595 * The file has too many extents to fit into
596 * the inode, so they are in B-tree format.
597 * Allocate a buffer for the root of the B-tree
598 * and copy the root into it. The i_extents
599 * field will remain NULL until all of the
600 * extents are read in (when they are needed).
601 */
602 STATIC int
607 {
610 /* REFERENCED */
619 /*
620 * blow out if -- fork has less extents than can fit in
621 * fork (fork shouldn't be a btree format), root btree
622 * block has more records than can fit into the fork,
623 * or the number of extents is greater than the number of
624 * blocks.
625 */
636 }
641 /*
642 * Copy and convert from the on-disk structure
643 * to the in-memory structure.
644 */
652 }
654 STATIC void
658 {
687 }
689 void
693 {
722 }
724 STATIC uint
726 __uint16_t di_flags)
727 {
759 }
762 }
764 uint
767 {
772 }
774 uint
777 {
780 }
782 /*
783 * Read the disk inode attributes into the in-core inode structure.
784 */
785 int
790 xfs_daddr_t bno,
791 uint iget_flags)
792 {
797 /*
798 * Fill in the location information in the in-core inode.
799 */
806 /*
807 * Get pointers to the on-disk inode and the buffer containing it.
808 */
815 /*
816 * If we got something that isn't an inode it means someone
817 * (nfs or dmi) has a stale handle.
818 */
820 #ifdef DEBUG
822 "dip->di_magic (0x%x) != "
825 XFS_DINODE_MAGIC);
829 }
831 /*
832 * If the on-disk inode is already linked to a directory
833 * entry, copy all of the inode into the in-core inode.
834 * xfs_iformat() handles copying in the inode format
835 * specific information.
836 * Otherwise, just get the truly permanent information.
837 */
842 #ifdef DEBUG
845 error);
848 }
854 /*
855 * Make sure to pull in the mode here as well in
856 * case the inode is released without being used.
857 * This ensures that xfs_inactive() will see that
858 * the inode is already free and not try to mess
859 * with the uninitialized part of it.
860 */
862 /*
863 * Initialize the per-fork minima and maxima for a new
864 * inode here. xfs_iformat will do it for old inodes.
865 */
868 }
870 /*
871 * The inode format changed when we moved the link count and
872 * made it 32 bits long. If this is an old format inode,
873 * convert it in memory to look like a new one. If it gets
874 * flushed to disk we will convert back before flushing or
875 * logging it. We zero out the new projid field and the old link
876 * count field. We'll handle clearing the pad field (the remains
877 * of the old uuid field) when we actually convert the inode to
878 * the new format. We don't change the version number so that we
879 * can distinguish this from a real new format inode.
880 */
885 }
890 /*
891 * Mark the buffer containing the inode as something to keep
892 * around for a while. This helps to keep recently accessed
893 * meta-data in-core longer.
894 */
897 /*
898 * Use xfs_trans_brelse() to release the buffer containing the
899 * on-disk inode, because it was acquired with xfs_trans_read_buf()
900 * in xfs_itobp() above. If tp is NULL, this is just a normal
901 * brelse(). If we're within a transaction, then xfs_trans_brelse()
902 * will only release the buffer if it is not dirty within the
903 * transaction. It will be OK to release the buffer in this case,
904 * because inodes on disk are never destroyed and we will be
905 * locking the new in-core inode before putting it in the hash
906 * table where other processes can find it. Thus we don't have
907 * to worry about the inode being changed just because we released
908 * the buffer.
909 */
910 out_brelse:
913 }
915 /*
916 * Read in extents from a btree-format inode.
917 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
918 */
919 int
924 {
927 xfs_extnum_t nextents;
934 }
939 /*
940 * We know that the size is valid (it's checked in iformat_btree)
941 */
951 }
954 }
956 /*
957 * Allocate an inode on disk and return a copy of its in-core version.
958 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
959 * appropriately within the inode. The uid and gid for the inode are
960 * set according to the contents of the given cred structure.
961 *
962 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
963 * has a free inode available, call xfs_iget()
964 * to obtain the in-core version of the allocated inode. Finally,
965 * fill in the inode and log its initial contents. In this case,
966 * ialloc_context would be set to NULL and call_again set to false.
967 *
968 * If xfs_dialloc() does not have an available inode,
969 * it will replenish its supply by doing an allocation. Since we can
970 * only do one allocation within a transaction without deadlocks, we
971 * must commit the current transaction before returning the inode itself.
972 * In this case, therefore, we will set call_again to true and return.
973 * The caller should then commit the current transaction, start a new
974 * transaction, and call xfs_ialloc() again to actually get the inode.
975 *
976 * To ensure that some other process does not grab the inode that
977 * was allocated during the first call to xfs_ialloc(), this routine
978 * also returns the [locked] bp pointing to the head of the freelist
979 * as ialloc_context. The caller should hold this buffer across
980 * the commit and pass it back into this routine on the second call.
981 *
982 * If we are allocating quota inodes, we do not have a parent inode
983 * to attach to or associate with (i.e. pip == NULL) because they
984 * are not linked into the directory structure - they are attached
985 * directly to the superblock - and so have no parent.
986 */
987 int
991 mode_t mode,
992 xfs_nlink_t nlink,
993 xfs_dev_t rdev,
995 xfs_prid_t prid,
1000 {
1001 xfs_ino_t ino;
1003 uint flags;
1005 timespec_t tv;
1008 /*
1009 * Call the space management code to pick
1010 * the on-disk inode to be allocated.
1011 */
1019 }
1022 /*
1023 * Get the in-core inode with the lock held exclusively.
1024 * This is because we're setting fields here we need
1025 * to prevent others from looking at until we're done.
1026 */
1042 /*
1043 * If the superblock version is up to where we support new format
1044 * inodes and this is currently an old format inode, then change
1045 * the inode version number now. This way we only do the conversion
1046 * here rather than here and in the flush/logging code.
1047 */
1051 /*
1052 * We've already zeroed the old link count, the projid field,
1053 * and the pad field.
1054 */
1055 }
1057 /*
1058 * Project ids won't be stored on disk if we are using a version 1 inode.
1059 */
1067 }
1068 }
1070 /*
1071 * If the group ID of the new file does not match the effective group
1072 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1073 * (and only if the irix_sgid_inherit compatibility variable is set).
1074 */
1079 }
1092 /*
1093 * di_gen will have been taken care of in xfs_iread.
1094 */
1111 /*
1112 * we can't set up filestreams until after the VFS inode
1113 * is set up properly.
1114 */
1117 /* fall through */
1128 }
1135 }
1136 }
1138 xfs_inherit_noatime)
1141 xfs_inherit_nodump)
1144 xfs_inherit_sync)
1147 xfs_inherit_nosymlinks)
1152 xfs_inherit_nodefrag)
1157 }
1158 /* FALLTHROUGH */
1167 }
1168 /*
1169 * Attribute fork settings for new inode.
1170 */
1174 /*
1175 * Log the new values stuffed into the inode.
1176 */
1179 /* now that we have an i_mode we can setup inode ops and unlock */
1182 /* now we have set up the vfs inode we can associate the filestream */
1189 }
1193 }
1195 /*
1196 * Check to make sure that there are no blocks allocated to the
1197 * file beyond the size of the file. We don't check this for
1198 * files with fixed size extents or real time extents, but we
1199 * at least do it for regular files.
1200 */
1201 #ifdef DEBUG
1202 void
1206 xfs_fsize_t isize)
1207 {
1208 xfs_fileoff_t map_first;
1223 /*
1224 * The filesystem could be shutting down, so bmapi may return
1225 * an error.
1226 */
1230 map_first),
1236 }
1239 /*
1240 * Calculate the last possible buffered byte in a file. This must
1241 * include data that was buffered beyond the EOF by the write code.
1242 * This also needs to deal with overflowing the xfs_fsize_t type
1243 * which can happen for sizes near the limit.
1244 *
1245 * We also need to take into account any blocks beyond the EOF. It
1246 * may be the case that they were buffered by a write which failed.
1247 * In that case the pages will still be in memory, but the inode size
1248 * will never have been updated.
1249 */
1250 STATIC xfs_fsize_t
1253 {
1255 xfs_fsize_t last_byte;
1256 xfs_fileoff_t last_block;
1257 xfs_fileoff_t size_last_block;
1263 /*
1264 * Only check for blocks beyond the EOF if the extents have
1265 * been read in. This eliminates the need for the inode lock,
1266 * and it also saves us from looking when it really isn't
1267 * necessary.
1268 */
1272 XFS_DATA_FORK);
1276 }
1279 }
1286 }
1290 }
1292 }
1294 #if defined(XFS_RW_TRACE)
1295 STATIC void
1300 xfs_fsize_t new_size,
1301 xfs_off_t toss_start,
1302 xfs_off_t toss_finish)
1303 {
1306 }
1325 }
1326 #else
1327 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1328 #endif
1330 /*
1331 * Start the truncation of the file to new_size. The new size
1332 * must be smaller than the current size. This routine will
1333 * clear the buffer and page caches of file data in the removed
1334 * range, and xfs_itruncate_finish() will remove the underlying
1335 * disk blocks.
1336 *
1337 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1338 * must NOT have the inode lock held at all. This is because we're
1339 * calling into the buffer/page cache code and we can't hold the
1340 * inode lock when we do so.
1341 *
1342 * We need to wait for any direct I/Os in flight to complete before we
1343 * proceed with the truncate. This is needed to prevent the extents
1344 * being read or written by the direct I/Os from being removed while the
1345 * I/O is in flight as there is no other method of synchronising
1346 * direct I/O with the truncate operation. Also, because we hold
1347 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1348 * started until the truncate completes and drops the lock. Essentially,
1349 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1350 * ordering between direct I/Os and the truncate operation.
1351 *
1352 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1353 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1354 * in the case that the caller is locking things out of order and
1355 * may not be able to call xfs_itruncate_finish() with the inode lock
1356 * held without dropping the I/O lock. If the caller must drop the
1357 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1358 * must be called again with all the same restrictions as the initial
1359 * call.
1360 */
1361 int
1364 uint flags,
1365 xfs_fsize_t new_size)
1366 {
1367 xfs_fsize_t last_byte;
1368 xfs_off_t toss_start;
1379 /* wait for the completion of any pending DIOs */
1383 /*
1384 * Call toss_pages or flushinval_pages to get rid of pages
1385 * overlapping the region being removed. We have to use
1386 * the less efficient flushinval_pages in the case that the
1387 * caller may not be able to finish the truncate without
1388 * dropping the inode's I/O lock. Make sure
1389 * to catch any pages brought in by buffers overlapping
1390 * the EOF by searching out beyond the isize by our
1391 * block size. We round new_size up to a block boundary
1392 * so that we don't toss things on the same block as
1393 * new_size but before it.
1394 *
1395 * Before calling toss_page or flushinval_pages, make sure to
1396 * call remapf() over the same region if the file is mapped.
1397 * This frees up mapped file references to the pages in the
1398 * given range and for the flushinval_pages case it ensures
1399 * that we get the latest mapped changes flushed out.
1400 */
1404 /*
1405 * The place to start tossing is beyond our maximum
1406 * file size, so there is no way that the data extended
1407 * out there.
1408 */
1410 }
1413 last_byte);
1421 }
1422 }
1424 #ifdef DEBUG
1427 }
1428 #endif
1430 }
1432 /*
1433 * Shrink the file to the given new_size. The new size must be smaller than
1434 * the current size. This will free up the underlying blocks in the removed
1435 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1436 *
1437 * The transaction passed to this routine must have made a permanent log
1438 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1439 * given transaction and start new ones, so make sure everything involved in
1440 * the transaction is tidy before calling here. Some transaction will be
1441 * returned to the caller to be committed. The incoming transaction must
1442 * already include the inode, and both inode locks must be held exclusively.
1443 * The inode must also be "held" within the transaction. On return the inode
1444 * will be "held" within the returned transaction. This routine does NOT
1445 * require any disk space to be reserved for it within the transaction.
1446 *
1447 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1448 * indicates the fork which is to be truncated. For the attribute fork we only
1449 * support truncation to size 0.
1450 *
1451 * We use the sync parameter to indicate whether or not the first transaction
1452 * we perform might have to be synchronous. For the attr fork, it needs to be
1453 * so if the unlink of the inode is not yet known to be permanent in the log.
1454 * This keeps us from freeing and reusing the blocks of the attribute fork
1455 * before the unlink of the inode becomes permanent.
1456 *
1457 * For the data fork, we normally have to run synchronously if we're being
1458 * called out of the inactive path or we're being called out of the create path
1459 * where we're truncating an existing file. Either way, the truncate needs to
1460 * be sync so blocks don't reappear in the file with altered data in case of a
1461 * crash. wsync filesystems can run the first case async because anything that
1462 * shrinks the inode has to run sync so by the time we're called here from
1463 * inactive, the inode size is permanently set to 0.
1464 *
1465 * Calls from the truncate path always need to be sync unless we're in a wsync
1466 * filesystem and the file has already been unlinked.
1467 *
1468 * The caller is responsible for correctly setting the sync parameter. It gets
1469 * too hard for us to guess here which path we're being called out of just
1470 * based on inode state.
1471 *
1472 * If we get an error, we must return with the inode locked and linked into the
1473 * current transaction. This keeps things simple for the higher level code,
1474 * because it always knows that the inode is locked and held in the transaction
1475 * that returns to it whether errors occur or not. We don't mark the inode
1476 * dirty on error so that transactions can be easily aborted if possible.
1477 */
1478 int
1482 xfs_fsize_t new_size,
1485 {
1486 xfs_fsblock_t first_block;
1487 xfs_fileoff_t first_unmap_block;
1488 xfs_fileoff_t last_block;
1494 xfs_bmap_free_t free_list;
1510 /*
1511 * We only support truncating the entire attribute fork.
1512 */
1515 }
1518 /*
1519 * The first thing we do is set the size to new_size permanently
1520 * on disk. This way we don't have to worry about anyone ever
1521 * being able to look at the data being freed even in the face
1522 * of a crash. What we're getting around here is the case where
1523 * we free a block, it is allocated to another file, it is written
1524 * to, and then we crash. If the new data gets written to the
1525 * file but the log buffers containing the free and reallocation
1526 * don't, then we'd end up with garbage in the blocks being freed.
1527 * As long as we make the new_size permanent before actually
1528 * freeing any blocks it doesn't matter if they get writtten to.
1529 *
1530 * The callers must signal into us whether or not the size
1531 * setting here must be synchronous. There are a few cases
1532 * where it doesn't have to be synchronous. Those cases
1533 * occur if the file is unlinked and we know the unlink is
1534 * permanent or if the blocks being truncated are guaranteed
1535 * to be beyond the inode eof (regardless of the link count)
1536 * and the eof value is permanent. Both of these cases occur
1537 * only on wsync-mounted filesystems. In those cases, we're
1538 * guaranteed that no user will ever see the data in the blocks
1539 * that are being truncated so the truncate can run async.
1540 * In the free beyond eof case, the file may wind up with
1541 * more blocks allocated to it than it needs if we crash
1542 * and that won't get fixed until the next time the file
1543 * is re-opened and closed but that's ok as that shouldn't
1544 * be too many blocks.
1545 *
1546 * However, we can't just make all wsync xactions run async
1547 * because there's one call out of the create path that needs
1548 * to run sync where it's truncating an existing file to size
1549 * 0 whose size is > 0.
1550 *
1551 * It's probably possible to come up with a test in this
1552 * routine that would correctly distinguish all the above
1553 * cases from the values of the function parameters and the
1554 * inode state but for sanity's sake, I've decided to let the
1555 * layers above just tell us. It's simpler to correctly figure
1556 * out in the layer above exactly under what conditions we
1557 * can run async and I think it's easier for others read and
1558 * follow the logic in case something has to be changed.
1559 * cscope is your friend -- rcc.
1560 *
1561 * The attribute fork is much simpler.
1562 *
1563 * For the attribute fork we allow the caller to tell us whether
1564 * the unlink of the inode that led to this call is yet permanent
1565 * in the on disk log. If it is not and we will be freeing extents
1566 * in this inode then we make the first transaction synchronous
1567 * to make sure that the unlink is permanent by the time we free
1568 * the blocks.
1569 */
1572 /*
1573 * If we are not changing the file size then do
1574 * not update the on-disk file size - we may be
1575 * called from xfs_inactive_free_eofblocks(). If we
1576 * update the on-disk file size and then the system
1577 * crashes before the contents of the file are
1578 * flushed to disk then the files may be full of
1579 * holes (ie NULL files bug).
1580 */
1585 }
1586 }
1591 }
1597 /*
1598 * Since it is possible for space to become allocated beyond
1599 * the end of the file (in a crash where the space is allocated
1600 * but the inode size is not yet updated), simply remove any
1601 * blocks which show up between the new EOF and the maximum
1602 * possible file size. If the first block to be removed is
1603 * beyond the maximum file size (ie it is the same as last_block),
1604 * then there is nothing to do.
1605 */
1613 }
1615 /*
1616 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1617 * will tell us whether it freed the entire range or
1618 * not. If this is a synchronous mount (wsync),
1619 * then we can tell bunmapi to keep all the
1620 * transactions asynchronous since the unlink
1621 * transaction that made this inode inactive has
1622 * already hit the disk. There's no danger of
1623 * the freed blocks being reused, there being a
1624 * crash, and the reused blocks suddenly reappearing
1625 * in this file with garbage in them once recovery
1626 * runs.
1627 */
1633 XFS_ITRUNC_MAX_EXTENTS,
1637 /*
1638 * If the bunmapi call encounters an error,
1639 * return to the caller where the transaction
1640 * can be properly aborted. We just need to
1641 * make sure we're not holding any resources
1642 * that we were not when we came in.
1643 */
1646 }
1648 /*
1649 * Duplicate the transaction that has the permanent
1650 * reservation and commit the old transaction.
1651 */
1655 /* link the inode into the next xact in the chain */
1659 }
1662 /*
1663 * If the bmap finish call encounters an error, return
1664 * to the caller where the transaction can be properly
1665 * aborted. We just need to make sure we're not
1666 * holding any resources that we were not when we came
1667 * in.
1668 *
1669 * Aborting from this point might lose some blocks in
1670 * the file system, but oh well.
1671 */
1674 }
1677 /*
1678 * Mark the inode dirty so it will be logged and
1679 * moved forward in the log as part of every commit.
1680 */
1682 }
1688 /* link the inode into the next transaction in the chain */
1694 /*
1695 * transaction commit worked ok so we can drop the extra ticket
1696 * reference that we gained in xfs_trans_dup()
1697 */
1701 XFS_TRANS_PERM_LOG_RES,
1702 XFS_ITRUNCATE_LOG_COUNT);
1705 }
1706 /*
1707 * Only update the size in the case of the data fork, but
1708 * always re-log the inode so that our permanent transaction
1709 * can keep on rolling it forward in the log.
1710 */
1713 /*
1714 * If we are not changing the file size then do
1715 * not update the on-disk file size - we may be
1716 * called from xfs_inactive_free_eofblocks(). If we
1717 * update the on-disk file size and then the system
1718 * crashes before the contents of the file are
1719 * flushed to disk then the files may be full of
1720 * holes (ie NULL files bug).
1721 */
1725 }
1726 }
1736 }
1738 /*
1739 * This is called when the inode's link count goes to 0.
1740 * We place the on-disk inode on a list in the AGI. It
1741 * will be pulled from this list when the inode is freed.
1742 */
1743 int
1747 {
1753 xfs_agino_t agino;
1764 /*
1765 * Get the agi buffer first. It ensures lock ordering
1766 * on the list.
1767 */
1773 /*
1774 * Get the index into the agi hash table for the
1775 * list this inode will go on.
1776 */
1784 /*
1785 * There is already another inode in the bucket we need
1786 * to add ourselves to. Add us at the front of the list.
1787 * Here we put the head pointer into our next pointer,
1788 * and then we fall through to point the head at us.
1789 */
1795 /* both on-disk, don't endian flip twice */
1803 }
1805 /*
1806 * Point the bucket head pointer at the inode being inserted.
1807 */
1815 }
1817 /*
1818 * Pull the on-disk inode from the AGI unlinked list.
1819 */
1820 STATIC int
1824 {
1825 xfs_ino_t next_ino;
1831 xfs_agnumber_t agno;
1832 xfs_agino_t agino;
1833 xfs_agino_t next_agino;
1843 /*
1844 * Get the agi buffer first. It ensures lock ordering
1845 * on the list.
1846 */
1853 /*
1854 * Get the index into the agi hash table for the
1855 * list this inode will go on.
1856 */
1864 /*
1865 * We're at the head of the list. Get the inode's
1866 * on-disk buffer to see if there is anyone after us
1867 * on the list. Only modify our next pointer if it
1868 * is not already NULLAGINO. This saves us the overhead
1869 * of dealing with the buffer when there is no need to
1870 * change it.
1871 */
1878 }
1891 }
1892 /*
1893 * Point the bucket head pointer at the next inode.
1894 */
1903 /*
1904 * We need to search the list for the inode being freed.
1905 */
1909 /*
1910 * If the last inode wasn't the one pointing to
1911 * us, then release its buffer since we're not
1912 * going to do anything with it.
1913 */
1916 }
1925 }
1929 }
1930 /*
1931 * Now last_ibp points to the buffer previous to us on
1932 * the unlinked list. Pull us from the list.
1933 */
1940 }
1954 }
1955 /*
1956 * Point the previous inode on the list to the next inode.
1957 */
1965 }
1967 }
1969 STATIC void
1973 xfs_ino_t inum)
1974 {
1980 xfs_daddr_t blkno;
1996 }
2005 /*
2006 * Look for each inode in memory and attempt to lock it,
2007 * we can be racing with flush and tail pushing here.
2008 * any inode we get the locks on, add to an array of
2009 * inode items to process later.
2010 *
2011 * The get the buffer lock, we could beat a flush
2012 * or tail pushing thread to the lock here, in which
2013 * case they will go looking for the inode buffer
2014 * and fail, we need some other form of interlock
2015 * here.
2016 */
2023 /* Inode not in memory or we found it already,
2024 * nothing to do
2025 */
2029 }
2034 }
2036 /* If we can get the locks then add it to the
2037 * list, otherwise by the time we get the bp lock
2038 * below it will already be attached to the
2039 * inode buffer.
2040 */
2042 /* This inode will already be locked - by us, lets
2043 * keep it that way.
2044 */
2053 }
2054 }
2057 }
2068 }
2071 }
2072 }
2074 }
2078 XFS_BUF_LOCK);
2091 pre_flushed++;
2092 }
2094 }
2105 }
2118 }
2119 }
2124 }
2128 }
2130 /*
2131 * This is called to return an inode to the inode free list.
2132 * The inode should already be truncated to 0 length and have
2133 * no pages associated with it. This routine also assumes that
2134 * the inode is already a part of the transaction.
2135 *
2136 * The on-disk copy of the inode will have been added to the list
2137 * of unlinked inodes in the AGI. We need to remove the inode from
2138 * that list atomically with respect to freeing it here.
2139 */
2140 int
2145 {
2148 xfs_ino_t first_ino;
2161 /*
2162 * Pull the on-disk inode from the AGI unlinked list.
2163 */
2167 }
2172 }
2181 /*
2182 * Bump the generation count so no one will be confused
2183 * by reincarnations of this inode.
2184 */
2193 /*
2194 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2195 * from picking up this inode when it is reclaimed (its incore state
2196 * initialzed but not flushed to disk yet). The in-core di_mode is
2197 * already cleared and a corresponding transaction logged.
2198 * The hack here just synchronizes the in-core to on-disk
2199 * di_mode value in advance before the actual inode sync to disk.
2200 * This is OK because the inode is already unlinked and would never
2201 * change its di_mode again for this inode generation.
2202 * This is a temporary hack that would require a proper fix
2203 * in the future.
2204 */
2209 }
2212 }
2214 /*
2215 * Reallocate the space for if_broot based on the number of records
2216 * being added or deleted as indicated in rec_diff. Move the records
2217 * and pointers in if_broot to fit the new size. When shrinking this
2218 * will eliminate holes between the records and pointers created by
2219 * the caller. When growing this will create holes to be filled in
2220 * by the caller.
2221 *
2222 * The caller must not request to add more records than would fit in
2223 * the on-disk inode root. If the if_broot is currently NULL, then
2224 * if we adding records one will be allocated. The caller must also
2225 * not request that the number of records go below zero, although
2226 * it can go to zero.
2227 *
2228 * ip -- the inode whose if_broot area is changing
2229 * ext_diff -- the change in the number of records, positive or negative,
2230 * requested for the if_broot array.
2231 */
2232 void
2237 {
2247 /*
2248 * Handle the degenerate case quietly.
2249 */
2252 }
2256 /*
2257 * If there wasn't any memory allocated before, just
2258 * allocate it now and get out.
2259 */
2265 }
2267 /*
2268 * If there is already an existing if_broot, then we need
2269 * to realloc() it and shift the pointers to their new
2270 * location. The records don't change location because
2271 * they are kept butted up against the btree block header.
2272 */
2278 KM_SLEEP);
2288 }
2290 /*
2291 * rec_diff is less than 0. In this case, we are shrinking the
2292 * if_broot buffer. It must already exist. If we go to zero
2293 * records, just get rid of the root and clear the status bit.
2294 */
2301 else
2305 /*
2306 * First copy over the btree block header.
2307 */
2312 }
2314 /*
2315 * Only copy the records and pointers if there are any.
2316 */
2318 /*
2319 * First copy the records.
2320 */
2325 /*
2326 * Then copy the pointers.
2327 */
2333 }
2340 }
2343 /*
2344 * This is called when the amount of space needed for if_data
2345 * is increased or decreased. The change in size is indicated by
2346 * the number of bytes that need to be added or deleted in the
2347 * byte_diff parameter.
2348 *
2349 * If the amount of space needed has decreased below the size of the
2350 * inline buffer, then switch to using the inline buffer. Otherwise,
2351 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2352 * to what is needed.
2353 *
2354 * ip -- the inode whose if_data area is changing
2355 * byte_diff -- the change in the number of bytes, positive or negative,
2356 * requested for the if_data array.
2357 */
2358 void
2363 {
2370 }
2379 }
2383 /*
2384 * If the valid extents/data can fit in if_inline_ext/data,
2385 * copy them from the malloc'd vector and free it.
2386 */
2392 new_size);
2395 }
2398 /*
2399 * Stuck with malloc/realloc.
2400 * For inline data, the underlying buffer must be
2401 * a multiple of 4 bytes in size so that it can be
2402 * logged and stay on word boundaries. We enforce
2403 * that here.
2404 */
2410 /*
2411 * Only do the realloc if the underlying size
2412 * is really changing.
2413 */
2417 real_size,
2419 KM_SLEEP);
2420 }
2426 }
2427 }
2431 }
2433 void
2437 {
2444 }
2446 /*
2447 * If the format is local, then we can't have an extents
2448 * array so just look for an inline data array. If we're
2449 * not local then we may or may not have an extents list,
2450 * so check and free it up if we do.
2451 */
2459 }
2466 }
2473 }
2474 }
2476 /*
2477 * Increment the pin count of the given buffer.
2478 * This value is protected by ipinlock spinlock in the mount structure.
2479 */
2480 void
2483 {
2487 }
2489 /*
2490 * Decrement the pin count of the given inode, and wake up
2491 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2492 * inode must have been previously pinned with a call to xfs_ipin().
2493 */
2494 void
2497 {
2502 }
2504 /*
2505 * This is called to unpin an inode. It can be directed to wait or to return
2506 * immediately without waiting for the inode to be unpinned. The caller must
2507 * have the inode locked in at least shared mode so that the buffer cannot be
2508 * subsequently pinned once someone is waiting for it to be unpinned.
2509 */
2510 STATIC void
2514 {
2521 /* Give the log a push to start the unpinning I/O */
2526 }
2531 {
2533 }
2538 {
2540 }
2543 /*
2544 * xfs_iextents_copy()
2545 *
2546 * This is called to copy the REAL extents (as opposed to the delayed
2547 * allocation extents) from the inode into the given buffer. It
2548 * returns the number of bytes copied into the buffer.
2549 *
2550 * If there are no delayed allocation extents, then we can just
2551 * memcpy() the extents into the buffer. Otherwise, we need to
2552 * examine each extent in turn and skip those which are delayed.
2553 */
2554 int
2559 {
2564 xfs_fsblock_t start_block;
2574 /*
2575 * There are some delayed allocation extents in the
2576 * inode, so copy the extents one at a time and skip
2577 * the delayed ones. There must be at least one
2578 * non-delayed extent.
2579 */
2585 /*
2586 * It's a delayed allocation extent, so skip it.
2587 */
2589 }
2591 /* Translate to on disk format */
2594 dp++;
2595 copied++;
2596 }
2601 }
2603 /*
2604 * Each of the following cases stores data into the same region
2605 * of the on-disk inode, so only one of them can be valid at
2606 * any given time. While it is possible to have conflicting formats
2607 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2608 * in EXTENTS format, this can only happen when the fork has
2609 * changed formats after being modified but before being flushed.
2610 * In these cases, the format always takes precedence, because the
2611 * format indicates the current state of the fork.
2612 */
2613 /*ARGSUSED*/
2614 STATIC void
2621 {
2625 #ifdef XFS_TRANS_DEBUG
2627 #endif
2638 /*
2639 * This can happen if we gave up in iformat in an error path,
2640 * for the attribute fork.
2641 */
2645 }
2655 }
2669 whichfork);
2670 }
2679 XFS_BROOT_SIZE_ADJ));
2683 }
2690 }
2699 }
2705 }
2706 }
2708 STATIC int
2712 {
2737 /* really need a gang lookup range call here */
2747 /* if the inode lies outside this cluster, we're done. */
2750 /*
2751 * Do an un-protected check to see if the inode is dirty and
2752 * is a candidate for flushing. These checks will be repeated
2753 * later after the appropriate locks are acquired.
2754 */
2758 /*
2759 * Try to get locks. If any are unavailable or it is pinned,
2760 * then this inode cannot be flushed and is skipped.
2761 */
2768 }
2773 }
2775 /*
2776 * arriving here means that this inode can be flushed. First
2777 * re-check that it's dirty before flushing.
2778 */
2785 }
2786 clcount++;
2789 }
2791 }
2796 }
2798 out_free:
2804 cluster_corrupt_out:
2805 /*
2806 * Corruption detected in the clustering loop. Invalidate the
2807 * inode buffer and shut down the filesystem.
2808 */
2810 /*
2811 * Clean up the buffer. If it was B_DELWRI, just release it --
2812 * brelse can handle it with no problems. If not, shut down the
2813 * filesystem before releasing the buffer.
2814 */
2822 /*
2823 * Just like incore_relse: if we have b_iodone functions,
2824 * mark the buffer as an error and call them. Otherwise
2825 * mark it as stale and brelse.
2826 */
2836 }
2837 }
2839 /*
2840 * Unlocks the flush lock
2841 */
2845 }
2847 /*
2848 * xfs_iflush() will write a modified inode's changes out to the
2849 * inode's on disk home. The caller must have the inode lock held
2850 * in at least shared mode and the inode flush completion must be
2851 * active as well. The inode lock will still be held upon return from
2852 * the call and the caller is free to unlock it.
2853 * The inode flush will be completed when the inode reaches the disk.
2854 * The flags indicate how the inode's buffer should be written out.
2855 */
2856 int
2859 uint flags)
2860 {
2879 /*
2880 * If the inode isn't dirty, then just release the inode
2881 * flush lock and do nothing.
2882 */
2886 }
2888 /*
2889 * We can't flush the inode until it is unpinned, so wait for it if we
2890 * are allowed to block. We know noone new can pin it, because we are
2891 * holding the inode lock shared and you need to hold it exclusively to
2892 * pin the inode.
2893 *
2894 * If we are not allowed to block, force the log out asynchronously so
2895 * that when we come back the inode will be unpinned. If other inodes
2896 * in the same cluster are dirty, they will probably write the inode
2897 * out for us if they occur after the log force completes.
2898 */
2903 }
2906 /*
2907 * This may have been unpinned because the filesystem is shutting
2908 * down forcibly. If that's the case we must not write this inode
2909 * to disk, because the log record didn't make it to disk!
2910 */
2917 }
2919 /*
2920 * Decide how buffer will be flushed out. This is done before
2921 * the call to xfs_iflush_int because this field is zeroed by it.
2922 */
2924 /*
2925 * Flush out the inode buffer according to the directions
2926 * of the caller. In the cases where the caller has given
2927 * us a choice choose the non-delwri case. This is because
2928 * the inode is in the AIL and we need to get it out soon.
2929 */
2947 }
2966 }
2967 }
2969 /*
2970 * Get the buffer containing the on-disk inode.
2971 */
2977 }
2979 /*
2980 * First flush out the inode that xfs_iflush was called with.
2981 */
2986 /*
2987 * If the buffer is pinned then push on the log now so we won't
2988 * get stuck waiting in the write for too long.
2989 */
2993 /*
2994 * inode clustering:
2995 * see if other inodes can be gathered into this write
2996 */
3007 }
3010 corrupt_out:
3013 cluster_corrupt_out:
3014 /*
3015 * Unlocks the flush lock
3016 */
3019 }
3022 STATIC int
3026 {
3030 #ifdef XFS_TRANS_DEBUG
3032 #endif
3043 /*
3044 * If the inode isn't dirty, then just release the inode
3045 * flush lock and do nothing.
3046 */
3050 }
3052 /* set *dip = inode's place in the buffer */
3055 /*
3056 * Clear i_update_core before copying out the data.
3057 * This is for coordination with our timestamp updates
3058 * that don't hold the inode lock. They will always
3059 * update the timestamps BEFORE setting i_update_core,
3060 * so if we clear i_update_core after they set it we
3061 * are guaranteed to see their updates to the timestamps.
3062 * I believe that this depends on strongly ordered memory
3063 * semantics, but we have that. We use the SYNCHRONIZE
3064 * macro to make sure that the compiler does not reorder
3065 * the i_update_core access below the data copy below.
3066 */
3070 /*
3071 * Make sure to get the latest timestamps from the Linux inode.
3072 */
3081 }
3088 }
3098 }
3109 }
3110 }
3113 XFS_RANDOM_IFLUSH_5)) {
3119 ip);
3121 }
3128 }
3129 /*
3130 * bump the flush iteration count, used to detect flushes which
3131 * postdate a log record during recovery.
3132 */
3136 /*
3137 * Copy the dirty parts of the inode into the on-disk
3138 * inode. We always copy out the core of the inode,
3139 * because if the inode is dirty at all the core must
3140 * be.
3141 */
3144 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3148 /*
3149 * If this is really an old format inode and the superblock version
3150 * has not been updated to support only new format inodes, then
3151 * convert back to the old inode format. If the superblock version
3152 * has been updated, then make the conversion permanent.
3153 */
3157 /*
3158 * Convert it back.
3159 */
3163 /*
3164 * The superblock version has already been bumped,
3165 * so just make the conversion to the new inode
3166 * format permanent.
3167 */
3176 }
3177 }
3184 /*
3185 * We've recorded everything logged in the inode, so we'd
3186 * like to clear the ilf_fields bits so we don't log and
3187 * flush things unnecessarily. However, we can't stop
3188 * logging all this information until the data we've copied
3189 * into the disk buffer is written to disk. If we did we might
3190 * overwrite the copy of the inode in the log with all the
3191 * data after re-logging only part of it, and in the face of
3192 * a crash we wouldn't have all the data we need to recover.
3193 *
3194 * What we do is move the bits to the ili_last_fields field.
3195 * When logging the inode, these bits are moved back to the
3196 * ilf_fields field. In the xfs_iflush_done() routine we
3197 * clear ili_last_fields, since we know that the information
3198 * those bits represent is permanently on disk. As long as
3199 * the flush completes before the inode is logged again, then
3200 * both ilf_fields and ili_last_fields will be cleared.
3201 *
3202 * We can play with the ilf_fields bits here, because the inode
3203 * lock must be held exclusively in order to set bits there
3204 * and the flush lock protects the ili_last_fields bits.
3205 * Set ili_logged so the flush done
3206 * routine can tell whether or not to look in the AIL.
3207 * Also, store the current LSN of the inode so that we can tell
3208 * whether the item has moved in the AIL from xfs_iflush_done().
3209 * In order to read the lsn we need the AIL lock, because
3210 * it is a 64 bit value that cannot be read atomically.
3211 */
3220 /*
3221 * Attach the function xfs_iflush_done to the inode's
3222 * buffer. This will remove the inode from the AIL
3223 * and unlock the inode's flush lock when the inode is
3224 * completely written to disk.
3225 */
3232 /*
3233 * We're flushing an inode which is not in the AIL and has
3234 * not been logged but has i_update_core set. For this
3235 * case we can use a B_DELWRI flush and immediately drop
3236 * the inode flush lock because we can avoid the whole
3237 * AIL state thing. It's OK to drop the flush lock now,
3238 * because we've already locked the buffer and to do anything
3239 * you really need both.
3240 */
3245 }
3247 }
3251 corrupt_out:
3253 }
3257 #ifdef XFS_ILOCK_TRACE
3258 void
3260 {
3269 }
3270 #endif
3272 /*
3273 * Return a pointer to the extent record at file index idx.
3274 */
3275 xfs_bmbt_rec_host_t *
3279 {
3294 }
3295 }
3297 /*
3298 * Insert new item(s) into the extent records for incore inode
3299 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3300 */
3301 void
3307 {
3314 }
3316 /*
3317 * This is called when the amount of space required for incore file
3318 * extents needs to be increased. The ext_diff parameter stores the
3319 * number of new extents being added and the idx parameter contains
3320 * the extent index where the new extents will be added. If the new
3321 * extents are being appended, then we just need to (re)allocate and
3322 * initialize the space. Otherwise, if the new extents are being
3323 * inserted into the middle of the existing entries, a bit more work
3324 * is required to make room for the new extents to be inserted. The
3325 * caller is responsible for filling in the new extent entries upon
3326 * return.
3327 */
3328 void
3333 {
3342 /*
3343 * If the new number of extents (nextents + ext_diff)
3344 * fits inside the inode, then continue to use the inline
3345 * extent buffer.
3346 */
3353 }
3357 }
3358 /*
3359 * Otherwise use a linear (direct) extent list.
3360 * If the extents are currently inside the inode,
3361 * xfs_iext_realloc_direct will switch us from
3362 * inline to direct extent allocation mode.
3363 */
3371 }
3372 }
3373 /* Indirection array */
3386 }
3387 /* Extents fit in target extent page */
3395 }
3398 }
3399 /* Insert a new extent page */
3403 }
3404 /*
3405 * If extent(s) are being appended to the last page in
3406 * the indirection array and the new extent(s) don't fit
3407 * in the page, then erp is NULL and erp_idx is set to
3408 * the next index needed in the indirection array.
3409 */
3418 erp_idx++;
3419 }
3420 }
3421 }
3422 }
3424 }
3426 /*
3427 * This is called when incore extents are being added to the indirection
3428 * array and the new extents do not fit in the target extent list. The
3429 * erp_idx parameter contains the irec index for the target extent list
3430 * in the indirection array, and the idx parameter contains the extent
3431 * index within the list. The number of extents being added is stored
3432 * in the count parameter.
3433 *
3434 * |-------| |-------|
3435 * | | | | idx - number of extents before idx
3436 * | idx | | count |
3437 * | | | | count - number of extents being inserted at idx
3438 * |-------| |-------|
3439 * | count | | nex2 | nex2 - number of extents after idx + count
3440 * |-------| |-------|
3441 */
3442 void
3448 {
3462 /*
3463 * Save second part of target extent list
3464 * (all extents past */
3472 }
3474 /*
3475 * Add the new extents to the end of the target
3476 * list, then allocate new irec record(s) and
3477 * extent buffer(s) as needed to store the rest
3478 * of the new extents.
3479 */
3486 }
3488 erp_idx++;
3494 }
3496 /* Add nex2 extents back to indirection array */
3498 xfs_extnum_t ext_avail;
3504 /*
3505 * If nex2 extents fit in the current page, append
3506 * nex2_ep after the new extents.
3507 */
3510 }
3511 /*
3512 * Otherwise, check if space is available in the
3513 * next page.
3514 */
3518 erp_idx++;
3519 erp++;
3520 /* Create a hole for nex2 extents */
3523 }
3524 /*
3525 * Final choice, create a new extent page for
3526 * nex2 extents.
3527 */
3529 erp_idx++;
3531 }
3536 }
3537 }
3539 /*
3540 * This is called when the amount of space required for incore file
3541 * extents needs to be decreased. The ext_diff parameter stores the
3542 * number of extents to be removed and the idx parameter contains
3543 * the extent index where the extents will be removed from.
3544 *
3545 * If the amount of space needed has decreased below the linear
3546 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3547 * extent array. Otherwise, use kmem_realloc() to adjust the
3548 * size to what is needed.
3549 */
3550 void
3555 {
3571 }
3573 }
3575 /*
3576 * This removes ext_diff extents from the inline buffer, beginning
3577 * at extent index idx.
3578 */
3579 void
3584 {
3603 }
3604 }
3606 /*
3607 * This removes ext_diff extents from a linear (direct) extent list,
3608 * beginning at extent index idx. If the extents are being removed
3609 * from the end of the list (ie. truncate) then we just need to re-
3610 * allocate the list to remove the extra space. Otherwise, if the
3611 * extents are being removed from the middle of the existing extent
3612 * entries, then we first need to move the extent records beginning
3613 * at idx + ext_diff up in the list to overwrite the records being
3614 * removed, then remove the extra space via kmem_realloc.
3615 */
3616 void
3621 {
3633 }
3634 /* Move extents up in the list (if needed) */
3640 }
3643 /*
3644 * Reallocate the direct extent list. If the extents
3645 * will fit inside the inode then xfs_iext_realloc_direct
3646 * will switch from direct to inline extent allocation
3647 * mode for us.
3648 */
3651 }
3653 /*
3654 * This is called when incore extents are being removed from the
3655 * indirection array and the extents being removed span multiple extent
3656 * buffers. The idx parameter contains the file extent index where we
3657 * want to begin removing extents, and the count parameter contains
3658 * how many extents need to be removed.
3659 *
3660 * |-------| |-------|
3661 * | nex1 | | | nex1 - number of extents before idx
3662 * |-------| | count |
3663 * | | | | count - number of extents being removed at idx
3664 * | count | |-------|
3665 * | | | nex2 | nex2 - number of extents after idx + count
3666 * |-------| |-------|
3667 */
3668 void
3673 {
3692 /*
3693 * Check for deletion of entire list;
3694 * xfs_iext_irec_remove() updates extent offsets.
3695 */
3702 XFS_IEXT_BUFSZ);
3708 }
3709 }
3710 /* Move extents up (if needed) */
3715 }
3716 /* Zero out rest of page */
3719 /* Update remaining counters */
3724 erp_idx++;
3725 erp++;
3726 }
3729 }
3731 /*
3732 * Create, destroy, or resize a linear (direct) block of extents.
3733 */
3734 void
3738 {
3747 /* Free extent records */
3750 }
3751 /* Resize direct extent list and zero any new bytes */
3753 /* Check if extents will fit inside the inode */
3759 }
3762 }
3766 rnew_size,
3768 }
3773 }
3774 }
3775 /*
3776 * Switch from the inline extent buffer to a direct
3777 * extent list. Be sure to include the inline extent
3778 * bytes in new_size.
3779 */
3784 }
3786 }
3789 }
3791 /*
3792 * Switch from linear (direct) extent records to inline buffer.
3793 */
3794 void
3798 {
3801 /*
3802 * The inline buffer was zeroed when we switched
3803 * from inline to direct extent allocation mode,
3804 * so we don't need to clear it here.
3805 */
3811 }
3813 /*
3814 * Switch from inline buffer to linear (direct) extent records.
3815 * new_size should already be rounded up to the next power of 2
3816 * by the caller (when appropriate), so use new_size as it is.
3817 * However, since new_size may be rounded up, we can't update
3818 * if_bytes here. It is the caller's responsibility to update
3819 * if_bytes upon return.
3820 */
3821 void
3825 {
3833 }
3835 }
3837 /*
3838 * Resize an extent indirection array to new_size bytes.
3839 */
3840 STATIC void
3844 {
3859 }
3860 }
3862 /*
3863 * Switch from indirection array to linear (direct) extent allocations.
3864 */
3865 STATIC void
3868 {
3888 }
3889 }
3891 /*
3892 * Free incore file extents.
3893 */
3894 void
3897 {
3905 }
3912 }
3916 }
3918 /*
3919 * Return a pointer to the extent record for file system block bno.
3920 */
3926 {
3941 }
3944 /* Find target extent list */
3952 }
3953 /* Binary search extent records */
3964 /* Convert back to file-based extent index */
3967 }
3970 }
3971 }
3972 /* Convert back to file-based extent index */
3975 }
3981 }
3982 }
3985 }
3987 /*
3988 * Return a pointer to the indirection array entry containing the
3989 * extent record for filesystem block bno. Store the index of the
3990 * target irec in *erp_idxp.
3991 */
3997 {
4021 }
4022 }
4025 }
4027 /*
4028 * Return a pointer to the indirection array entry containing the
4029 * extent record at file extent index *idxp. Store the index of the
4030 * target irec in *erp_idxp and store the page index of the target
4031 * extent record in *idxp.
4032 */
4033 xfs_ext_irec_t *
4039 {
4056 /* Binary search extent irec's */
4072 erp_idx++;
4078 }
4079 }
4083 }
4085 /*
4086 * Allocate and initialize an indirection array once the space needed
4087 * for incore extents increases above XFS_IEXT_BUFSZ.
4088 */
4089 void
4092 {
4108 }
4119 }
4121 /*
4122 * Allocate and initialize a new entry in the indirection array.
4123 */
4124 xfs_ext_irec_t *
4128 {
4136 /* Resize indirection array */
4139 /*
4140 * Move records down in the array so the
4141 * new page can use erp_idx.
4142 */
4146 }
4149 /* Initialize new extent record */
4158 }
4160 /*
4161 * Remove a record from the indirection array.
4162 */
4163 void
4167 {
4179 }
4180 /* Compact extent records */
4184 }
4185 /*
4186 * Manually free the last extent record from the indirection
4187 * array. A call to xfs_iext_realloc_indirect() with a size
4188 * of zero would result in a call to xfs_iext_destroy() which
4189 * would in turn call this function again, creating a nasty
4190 * infinite loop.
4191 */
4197 }
4199 }
4201 /*
4202 * This is called to clean up large amounts of unused memory allocated
4203 * by the indirection array. Before compacting anything though, verify
4204 * that the indirection array is still needed and switch back to the
4205 * linear extent list (or even the inline buffer) if possible. The
4206 * compaction policy is as follows:
4207 *
4208 * Full Compaction: Extents fit into a single page (or inline buffer)
4209 * Partial Compaction: Extents occupy less than 50% of allocated space
4210 * No Compaction: Extents occupy at least 50% of allocated space
4211 */
4212 void
4215 {
4232 }
4233 }
4235 /*
4236 * Combine extents from neighboring extent pages.
4237 */
4238 void
4241 {
4257 /*
4258 * Free page before removing extent record
4259 * so er_extoffs don't get modified in
4260 * xfs_iext_irec_remove.
4261 */
4267 erp_idx++;
4268 }
4269 }
4270 }
4272 /*
4273 * This is called to update the er_extoff field in the indirection
4274 * array when extents have been added or removed from one of the
4275 * extent lists. erp_idx contains the irec index to begin updating
4276 * at and ext_diff contains the number of extents that were added
4277 * or removed.
4278 */
4279 void
4284 {
4292 }
4293 }