| blob | 8f43babc3346224928ecff998a3b4d0ebf13e068 |
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 uint iget_flags)
791 {
796 /*
797 * Fill in the location information in the in-core inode.
798 */
803 /*
804 * Get pointers to the on-disk inode and the buffer containing it.
805 */
812 /*
813 * If we got something that isn't an inode it means someone
814 * (nfs or dmi) has a stale handle.
815 */
817 #ifdef DEBUG
819 "dip->di_magic (0x%x) != "
822 XFS_DINODE_MAGIC);
826 }
828 /*
829 * If the on-disk inode is already linked to a directory
830 * entry, copy all of the inode into the in-core inode.
831 * xfs_iformat() handles copying in the inode format
832 * specific information.
833 * Otherwise, just get the truly permanent information.
834 */
839 #ifdef DEBUG
842 error);
845 }
851 /*
852 * Make sure to pull in the mode here as well in
853 * case the inode is released without being used.
854 * This ensures that xfs_inactive() will see that
855 * the inode is already free and not try to mess
856 * with the uninitialized part of it.
857 */
859 /*
860 * Initialize the per-fork minima and maxima for a new
861 * inode here. xfs_iformat will do it for old inodes.
862 */
865 }
867 /*
868 * The inode format changed when we moved the link count and
869 * made it 32 bits long. If this is an old format inode,
870 * convert it in memory to look like a new one. If it gets
871 * flushed to disk we will convert back before flushing or
872 * logging it. We zero out the new projid field and the old link
873 * count field. We'll handle clearing the pad field (the remains
874 * of the old uuid field) when we actually convert the inode to
875 * the new format. We don't change the version number so that we
876 * can distinguish this from a real new format inode.
877 */
882 }
887 /*
888 * Mark the buffer containing the inode as something to keep
889 * around for a while. This helps to keep recently accessed
890 * meta-data in-core longer.
891 */
894 /*
895 * Use xfs_trans_brelse() to release the buffer containing the
896 * on-disk inode, because it was acquired with xfs_trans_read_buf()
897 * in xfs_itobp() above. If tp is NULL, this is just a normal
898 * brelse(). If we're within a transaction, then xfs_trans_brelse()
899 * will only release the buffer if it is not dirty within the
900 * transaction. It will be OK to release the buffer in this case,
901 * because inodes on disk are never destroyed and we will be
902 * locking the new in-core inode before putting it in the hash
903 * table where other processes can find it. Thus we don't have
904 * to worry about the inode being changed just because we released
905 * the buffer.
906 */
907 out_brelse:
910 }
912 /*
913 * Read in extents from a btree-format inode.
914 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
915 */
916 int
921 {
924 xfs_extnum_t nextents;
931 }
936 /*
937 * We know that the size is valid (it's checked in iformat_btree)
938 */
948 }
951 }
953 /*
954 * Allocate an inode on disk and return a copy of its in-core version.
955 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
956 * appropriately within the inode. The uid and gid for the inode are
957 * set according to the contents of the given cred structure.
958 *
959 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
960 * has a free inode available, call xfs_iget()
961 * to obtain the in-core version of the allocated inode. Finally,
962 * fill in the inode and log its initial contents. In this case,
963 * ialloc_context would be set to NULL and call_again set to false.
964 *
965 * If xfs_dialloc() does not have an available inode,
966 * it will replenish its supply by doing an allocation. Since we can
967 * only do one allocation within a transaction without deadlocks, we
968 * must commit the current transaction before returning the inode itself.
969 * In this case, therefore, we will set call_again to true and return.
970 * The caller should then commit the current transaction, start a new
971 * transaction, and call xfs_ialloc() again to actually get the inode.
972 *
973 * To ensure that some other process does not grab the inode that
974 * was allocated during the first call to xfs_ialloc(), this routine
975 * also returns the [locked] bp pointing to the head of the freelist
976 * as ialloc_context. The caller should hold this buffer across
977 * the commit and pass it back into this routine on the second call.
978 *
979 * If we are allocating quota inodes, we do not have a parent inode
980 * to attach to or associate with (i.e. pip == NULL) because they
981 * are not linked into the directory structure - they are attached
982 * directly to the superblock - and so have no parent.
983 */
984 int
988 mode_t mode,
989 xfs_nlink_t nlink,
990 xfs_dev_t rdev,
992 xfs_prid_t prid,
997 {
998 xfs_ino_t ino;
1000 uint flags;
1002 timespec_t tv;
1005 /*
1006 * Call the space management code to pick
1007 * the on-disk inode to be allocated.
1008 */
1016 }
1019 /*
1020 * Get the in-core inode with the lock held exclusively.
1021 * This is because we're setting fields here we need
1022 * to prevent others from looking at until we're done.
1023 */
1039 /*
1040 * If the superblock version is up to where we support new format
1041 * inodes and this is currently an old format inode, then change
1042 * the inode version number now. This way we only do the conversion
1043 * here rather than here and in the flush/logging code.
1044 */
1048 /*
1049 * We've already zeroed the old link count, the projid field,
1050 * and the pad field.
1051 */
1052 }
1054 /*
1055 * Project ids won't be stored on disk if we are using a version 1 inode.
1056 */
1064 }
1065 }
1067 /*
1068 * If the group ID of the new file does not match the effective group
1069 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1070 * (and only if the irix_sgid_inherit compatibility variable is set).
1071 */
1076 }
1089 /*
1090 * di_gen will have been taken care of in xfs_iread.
1091 */
1108 /*
1109 * we can't set up filestreams until after the VFS inode
1110 * is set up properly.
1111 */
1114 /* fall through */
1125 }
1132 }
1133 }
1135 xfs_inherit_noatime)
1138 xfs_inherit_nodump)
1141 xfs_inherit_sync)
1144 xfs_inherit_nosymlinks)
1149 xfs_inherit_nodefrag)
1154 }
1155 /* FALLTHROUGH */
1164 }
1165 /*
1166 * Attribute fork settings for new inode.
1167 */
1171 /*
1172 * Log the new values stuffed into the inode.
1173 */
1176 /* now that we have an i_mode we can setup inode ops and unlock */
1179 /* now we have set up the vfs inode we can associate the filestream */
1186 }
1190 }
1192 /*
1193 * Check to make sure that there are no blocks allocated to the
1194 * file beyond the size of the file. We don't check this for
1195 * files with fixed size extents or real time extents, but we
1196 * at least do it for regular files.
1197 */
1198 #ifdef DEBUG
1199 void
1203 xfs_fsize_t isize)
1204 {
1205 xfs_fileoff_t map_first;
1220 /*
1221 * The filesystem could be shutting down, so bmapi may return
1222 * an error.
1223 */
1227 map_first),
1233 }
1236 /*
1237 * Calculate the last possible buffered byte in a file. This must
1238 * include data that was buffered beyond the EOF by the write code.
1239 * This also needs to deal with overflowing the xfs_fsize_t type
1240 * which can happen for sizes near the limit.
1241 *
1242 * We also need to take into account any blocks beyond the EOF. It
1243 * may be the case that they were buffered by a write which failed.
1244 * In that case the pages will still be in memory, but the inode size
1245 * will never have been updated.
1246 */
1247 STATIC xfs_fsize_t
1250 {
1252 xfs_fsize_t last_byte;
1253 xfs_fileoff_t last_block;
1254 xfs_fileoff_t size_last_block;
1260 /*
1261 * Only check for blocks beyond the EOF if the extents have
1262 * been read in. This eliminates the need for the inode lock,
1263 * and it also saves us from looking when it really isn't
1264 * necessary.
1265 */
1269 XFS_DATA_FORK);
1273 }
1276 }
1283 }
1287 }
1289 }
1291 #if defined(XFS_RW_TRACE)
1292 STATIC void
1297 xfs_fsize_t new_size,
1298 xfs_off_t toss_start,
1299 xfs_off_t toss_finish)
1300 {
1303 }
1322 }
1323 #else
1324 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1325 #endif
1327 /*
1328 * Start the truncation of the file to new_size. The new size
1329 * must be smaller than the current size. This routine will
1330 * clear the buffer and page caches of file data in the removed
1331 * range, and xfs_itruncate_finish() will remove the underlying
1332 * disk blocks.
1333 *
1334 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1335 * must NOT have the inode lock held at all. This is because we're
1336 * calling into the buffer/page cache code and we can't hold the
1337 * inode lock when we do so.
1338 *
1339 * We need to wait for any direct I/Os in flight to complete before we
1340 * proceed with the truncate. This is needed to prevent the extents
1341 * being read or written by the direct I/Os from being removed while the
1342 * I/O is in flight as there is no other method of synchronising
1343 * direct I/O with the truncate operation. Also, because we hold
1344 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1345 * started until the truncate completes and drops the lock. Essentially,
1346 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1347 * ordering between direct I/Os and the truncate operation.
1348 *
1349 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1350 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1351 * in the case that the caller is locking things out of order and
1352 * may not be able to call xfs_itruncate_finish() with the inode lock
1353 * held without dropping the I/O lock. If the caller must drop the
1354 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1355 * must be called again with all the same restrictions as the initial
1356 * call.
1357 */
1358 int
1361 uint flags,
1362 xfs_fsize_t new_size)
1363 {
1364 xfs_fsize_t last_byte;
1365 xfs_off_t toss_start;
1376 /* wait for the completion of any pending DIOs */
1380 /*
1381 * Call toss_pages or flushinval_pages to get rid of pages
1382 * overlapping the region being removed. We have to use
1383 * the less efficient flushinval_pages in the case that the
1384 * caller may not be able to finish the truncate without
1385 * dropping the inode's I/O lock. Make sure
1386 * to catch any pages brought in by buffers overlapping
1387 * the EOF by searching out beyond the isize by our
1388 * block size. We round new_size up to a block boundary
1389 * so that we don't toss things on the same block as
1390 * new_size but before it.
1391 *
1392 * Before calling toss_page or flushinval_pages, make sure to
1393 * call remapf() over the same region if the file is mapped.
1394 * This frees up mapped file references to the pages in the
1395 * given range and for the flushinval_pages case it ensures
1396 * that we get the latest mapped changes flushed out.
1397 */
1401 /*
1402 * The place to start tossing is beyond our maximum
1403 * file size, so there is no way that the data extended
1404 * out there.
1405 */
1407 }
1410 last_byte);
1418 }
1419 }
1421 #ifdef DEBUG
1424 }
1425 #endif
1427 }
1429 /*
1430 * Shrink the file to the given new_size. The new size must be smaller than
1431 * the current size. This will free up the underlying blocks in the removed
1432 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1433 *
1434 * The transaction passed to this routine must have made a permanent log
1435 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1436 * given transaction and start new ones, so make sure everything involved in
1437 * the transaction is tidy before calling here. Some transaction will be
1438 * returned to the caller to be committed. The incoming transaction must
1439 * already include the inode, and both inode locks must be held exclusively.
1440 * The inode must also be "held" within the transaction. On return the inode
1441 * will be "held" within the returned transaction. This routine does NOT
1442 * require any disk space to be reserved for it within the transaction.
1443 *
1444 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1445 * indicates the fork which is to be truncated. For the attribute fork we only
1446 * support truncation to size 0.
1447 *
1448 * We use the sync parameter to indicate whether or not the first transaction
1449 * we perform might have to be synchronous. For the attr fork, it needs to be
1450 * so if the unlink of the inode is not yet known to be permanent in the log.
1451 * This keeps us from freeing and reusing the blocks of the attribute fork
1452 * before the unlink of the inode becomes permanent.
1453 *
1454 * For the data fork, we normally have to run synchronously if we're being
1455 * called out of the inactive path or we're being called out of the create path
1456 * where we're truncating an existing file. Either way, the truncate needs to
1457 * be sync so blocks don't reappear in the file with altered data in case of a
1458 * crash. wsync filesystems can run the first case async because anything that
1459 * shrinks the inode has to run sync so by the time we're called here from
1460 * inactive, the inode size is permanently set to 0.
1461 *
1462 * Calls from the truncate path always need to be sync unless we're in a wsync
1463 * filesystem and the file has already been unlinked.
1464 *
1465 * The caller is responsible for correctly setting the sync parameter. It gets
1466 * too hard for us to guess here which path we're being called out of just
1467 * based on inode state.
1468 *
1469 * If we get an error, we must return with the inode locked and linked into the
1470 * current transaction. This keeps things simple for the higher level code,
1471 * because it always knows that the inode is locked and held in the transaction
1472 * that returns to it whether errors occur or not. We don't mark the inode
1473 * dirty on error so that transactions can be easily aborted if possible.
1474 */
1475 int
1479 xfs_fsize_t new_size,
1482 {
1483 xfs_fsblock_t first_block;
1484 xfs_fileoff_t first_unmap_block;
1485 xfs_fileoff_t last_block;
1491 xfs_bmap_free_t free_list;
1507 /*
1508 * We only support truncating the entire attribute fork.
1509 */
1512 }
1515 /*
1516 * The first thing we do is set the size to new_size permanently
1517 * on disk. This way we don't have to worry about anyone ever
1518 * being able to look at the data being freed even in the face
1519 * of a crash. What we're getting around here is the case where
1520 * we free a block, it is allocated to another file, it is written
1521 * to, and then we crash. If the new data gets written to the
1522 * file but the log buffers containing the free and reallocation
1523 * don't, then we'd end up with garbage in the blocks being freed.
1524 * As long as we make the new_size permanent before actually
1525 * freeing any blocks it doesn't matter if they get writtten to.
1526 *
1527 * The callers must signal into us whether or not the size
1528 * setting here must be synchronous. There are a few cases
1529 * where it doesn't have to be synchronous. Those cases
1530 * occur if the file is unlinked and we know the unlink is
1531 * permanent or if the blocks being truncated are guaranteed
1532 * to be beyond the inode eof (regardless of the link count)
1533 * and the eof value is permanent. Both of these cases occur
1534 * only on wsync-mounted filesystems. In those cases, we're
1535 * guaranteed that no user will ever see the data in the blocks
1536 * that are being truncated so the truncate can run async.
1537 * In the free beyond eof case, the file may wind up with
1538 * more blocks allocated to it than it needs if we crash
1539 * and that won't get fixed until the next time the file
1540 * is re-opened and closed but that's ok as that shouldn't
1541 * be too many blocks.
1542 *
1543 * However, we can't just make all wsync xactions run async
1544 * because there's one call out of the create path that needs
1545 * to run sync where it's truncating an existing file to size
1546 * 0 whose size is > 0.
1547 *
1548 * It's probably possible to come up with a test in this
1549 * routine that would correctly distinguish all the above
1550 * cases from the values of the function parameters and the
1551 * inode state but for sanity's sake, I've decided to let the
1552 * layers above just tell us. It's simpler to correctly figure
1553 * out in the layer above exactly under what conditions we
1554 * can run async and I think it's easier for others read and
1555 * follow the logic in case something has to be changed.
1556 * cscope is your friend -- rcc.
1557 *
1558 * The attribute fork is much simpler.
1559 *
1560 * For the attribute fork we allow the caller to tell us whether
1561 * the unlink of the inode that led to this call is yet permanent
1562 * in the on disk log. If it is not and we will be freeing extents
1563 * in this inode then we make the first transaction synchronous
1564 * to make sure that the unlink is permanent by the time we free
1565 * the blocks.
1566 */
1569 /*
1570 * If we are not changing the file size then do
1571 * not update the on-disk file size - we may be
1572 * called from xfs_inactive_free_eofblocks(). If we
1573 * update the on-disk file size and then the system
1574 * crashes before the contents of the file are
1575 * flushed to disk then the files may be full of
1576 * holes (ie NULL files bug).
1577 */
1582 }
1583 }
1588 }
1594 /*
1595 * Since it is possible for space to become allocated beyond
1596 * the end of the file (in a crash where the space is allocated
1597 * but the inode size is not yet updated), simply remove any
1598 * blocks which show up between the new EOF and the maximum
1599 * possible file size. If the first block to be removed is
1600 * beyond the maximum file size (ie it is the same as last_block),
1601 * then there is nothing to do.
1602 */
1610 }
1612 /*
1613 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1614 * will tell us whether it freed the entire range or
1615 * not. If this is a synchronous mount (wsync),
1616 * then we can tell bunmapi to keep all the
1617 * transactions asynchronous since the unlink
1618 * transaction that made this inode inactive has
1619 * already hit the disk. There's no danger of
1620 * the freed blocks being reused, there being a
1621 * crash, and the reused blocks suddenly reappearing
1622 * in this file with garbage in them once recovery
1623 * runs.
1624 */
1630 XFS_ITRUNC_MAX_EXTENTS,
1634 /*
1635 * If the bunmapi call encounters an error,
1636 * return to the caller where the transaction
1637 * can be properly aborted. We just need to
1638 * make sure we're not holding any resources
1639 * that we were not when we came in.
1640 */
1643 }
1645 /*
1646 * Duplicate the transaction that has the permanent
1647 * reservation and commit the old transaction.
1648 */
1652 /* link the inode into the next xact in the chain */
1656 }
1659 /*
1660 * If the bmap finish call encounters an error, return
1661 * to the caller where the transaction can be properly
1662 * aborted. We just need to make sure we're not
1663 * holding any resources that we were not when we came
1664 * in.
1665 *
1666 * Aborting from this point might lose some blocks in
1667 * the file system, but oh well.
1668 */
1671 }
1674 /*
1675 * Mark the inode dirty so it will be logged and
1676 * moved forward in the log as part of every commit.
1677 */
1679 }
1685 /* link the inode into the next transaction in the chain */
1691 /*
1692 * transaction commit worked ok so we can drop the extra ticket
1693 * reference that we gained in xfs_trans_dup()
1694 */
1698 XFS_TRANS_PERM_LOG_RES,
1699 XFS_ITRUNCATE_LOG_COUNT);
1702 }
1703 /*
1704 * Only update the size in the case of the data fork, but
1705 * always re-log the inode so that our permanent transaction
1706 * can keep on rolling it forward in the log.
1707 */
1710 /*
1711 * If we are not changing the file size then do
1712 * not update the on-disk file size - we may be
1713 * called from xfs_inactive_free_eofblocks(). If we
1714 * update the on-disk file size and then the system
1715 * crashes before the contents of the file are
1716 * flushed to disk then the files may be full of
1717 * holes (ie NULL files bug).
1718 */
1722 }
1723 }
1733 }
1735 /*
1736 * This is called when the inode's link count goes to 0.
1737 * We place the on-disk inode on a list in the AGI. It
1738 * will be pulled from this list when the inode is freed.
1739 */
1740 int
1744 {
1750 xfs_agino_t agino;
1761 /*
1762 * Get the agi buffer first. It ensures lock ordering
1763 * on the list.
1764 */
1770 /*
1771 * Get the index into the agi hash table for the
1772 * list this inode will go on.
1773 */
1781 /*
1782 * There is already another inode in the bucket we need
1783 * to add ourselves to. Add us at the front of the list.
1784 * Here we put the head pointer into our next pointer,
1785 * and then we fall through to point the head at us.
1786 */
1792 /* both on-disk, don't endian flip twice */
1800 }
1802 /*
1803 * Point the bucket head pointer at the inode being inserted.
1804 */
1812 }
1814 /*
1815 * Pull the on-disk inode from the AGI unlinked list.
1816 */
1817 STATIC int
1821 {
1822 xfs_ino_t next_ino;
1828 xfs_agnumber_t agno;
1829 xfs_agino_t agino;
1830 xfs_agino_t next_agino;
1840 /*
1841 * Get the agi buffer first. It ensures lock ordering
1842 * on the list.
1843 */
1850 /*
1851 * Get the index into the agi hash table for the
1852 * list this inode will go on.
1853 */
1861 /*
1862 * We're at the head of the list. Get the inode's
1863 * on-disk buffer to see if there is anyone after us
1864 * on the list. Only modify our next pointer if it
1865 * is not already NULLAGINO. This saves us the overhead
1866 * of dealing with the buffer when there is no need to
1867 * change it.
1868 */
1875 }
1888 }
1889 /*
1890 * Point the bucket head pointer at the next inode.
1891 */
1900 /*
1901 * We need to search the list for the inode being freed.
1902 */
1906 /*
1907 * If the last inode wasn't the one pointing to
1908 * us, then release its buffer since we're not
1909 * going to do anything with it.
1910 */
1913 }
1922 }
1926 }
1927 /*
1928 * Now last_ibp points to the buffer previous to us on
1929 * the unlinked list. Pull us from the list.
1930 */
1937 }
1951 }
1952 /*
1953 * Point the previous inode on the list to the next inode.
1954 */
1962 }
1964 }
1966 STATIC void
1970 xfs_ino_t inum)
1971 {
1977 xfs_daddr_t blkno;
1993 }
2002 /*
2003 * Look for each inode in memory and attempt to lock it,
2004 * we can be racing with flush and tail pushing here.
2005 * any inode we get the locks on, add to an array of
2006 * inode items to process later.
2007 *
2008 * The get the buffer lock, we could beat a flush
2009 * or tail pushing thread to the lock here, in which
2010 * case they will go looking for the inode buffer
2011 * and fail, we need some other form of interlock
2012 * here.
2013 */
2020 /* Inode not in memory or we found it already,
2021 * nothing to do
2022 */
2026 }
2031 }
2033 /* If we can get the locks then add it to the
2034 * list, otherwise by the time we get the bp lock
2035 * below it will already be attached to the
2036 * inode buffer.
2037 */
2039 /* This inode will already be locked - by us, lets
2040 * keep it that way.
2041 */
2050 }
2051 }
2054 }
2065 }
2068 }
2069 }
2071 }
2075 XFS_BUF_LOCK);
2088 pre_flushed++;
2089 }
2091 }
2102 }
2115 }
2116 }
2121 }
2125 }
2127 /*
2128 * This is called to return an inode to the inode free list.
2129 * The inode should already be truncated to 0 length and have
2130 * no pages associated with it. This routine also assumes that
2131 * the inode is already a part of the transaction.
2132 *
2133 * The on-disk copy of the inode will have been added to the list
2134 * of unlinked inodes in the AGI. We need to remove the inode from
2135 * that list atomically with respect to freeing it here.
2136 */
2137 int
2142 {
2145 xfs_ino_t first_ino;
2158 /*
2159 * Pull the on-disk inode from the AGI unlinked list.
2160 */
2164 }
2169 }
2178 /*
2179 * Bump the generation count so no one will be confused
2180 * by reincarnations of this inode.
2181 */
2190 /*
2191 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2192 * from picking up this inode when it is reclaimed (its incore state
2193 * initialzed but not flushed to disk yet). The in-core di_mode is
2194 * already cleared and a corresponding transaction logged.
2195 * The hack here just synchronizes the in-core to on-disk
2196 * di_mode value in advance before the actual inode sync to disk.
2197 * This is OK because the inode is already unlinked and would never
2198 * change its di_mode again for this inode generation.
2199 * This is a temporary hack that would require a proper fix
2200 * in the future.
2201 */
2206 }
2209 }
2211 /*
2212 * Reallocate the space for if_broot based on the number of records
2213 * being added or deleted as indicated in rec_diff. Move the records
2214 * and pointers in if_broot to fit the new size. When shrinking this
2215 * will eliminate holes between the records and pointers created by
2216 * the caller. When growing this will create holes to be filled in
2217 * by the caller.
2218 *
2219 * The caller must not request to add more records than would fit in
2220 * the on-disk inode root. If the if_broot is currently NULL, then
2221 * if we adding records one will be allocated. The caller must also
2222 * not request that the number of records go below zero, although
2223 * it can go to zero.
2224 *
2225 * ip -- the inode whose if_broot area is changing
2226 * ext_diff -- the change in the number of records, positive or negative,
2227 * requested for the if_broot array.
2228 */
2229 void
2234 {
2244 /*
2245 * Handle the degenerate case quietly.
2246 */
2249 }
2253 /*
2254 * If there wasn't any memory allocated before, just
2255 * allocate it now and get out.
2256 */
2262 }
2264 /*
2265 * If there is already an existing if_broot, then we need
2266 * to realloc() it and shift the pointers to their new
2267 * location. The records don't change location because
2268 * they are kept butted up against the btree block header.
2269 */
2275 KM_SLEEP);
2285 }
2287 /*
2288 * rec_diff is less than 0. In this case, we are shrinking the
2289 * if_broot buffer. It must already exist. If we go to zero
2290 * records, just get rid of the root and clear the status bit.
2291 */
2298 else
2302 /*
2303 * First copy over the btree block header.
2304 */
2309 }
2311 /*
2312 * Only copy the records and pointers if there are any.
2313 */
2315 /*
2316 * First copy the records.
2317 */
2322 /*
2323 * Then copy the pointers.
2324 */
2330 }
2337 }
2340 /*
2341 * This is called when the amount of space needed for if_data
2342 * is increased or decreased. The change in size is indicated by
2343 * the number of bytes that need to be added or deleted in the
2344 * byte_diff parameter.
2345 *
2346 * If the amount of space needed has decreased below the size of the
2347 * inline buffer, then switch to using the inline buffer. Otherwise,
2348 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2349 * to what is needed.
2350 *
2351 * ip -- the inode whose if_data area is changing
2352 * byte_diff -- the change in the number of bytes, positive or negative,
2353 * requested for the if_data array.
2354 */
2355 void
2360 {
2367 }
2376 }
2380 /*
2381 * If the valid extents/data can fit in if_inline_ext/data,
2382 * copy them from the malloc'd vector and free it.
2383 */
2389 new_size);
2392 }
2395 /*
2396 * Stuck with malloc/realloc.
2397 * For inline data, the underlying buffer must be
2398 * a multiple of 4 bytes in size so that it can be
2399 * logged and stay on word boundaries. We enforce
2400 * that here.
2401 */
2407 /*
2408 * Only do the realloc if the underlying size
2409 * is really changing.
2410 */
2414 real_size,
2416 KM_SLEEP);
2417 }
2423 }
2424 }
2428 }
2430 void
2434 {
2441 }
2443 /*
2444 * If the format is local, then we can't have an extents
2445 * array so just look for an inline data array. If we're
2446 * not local then we may or may not have an extents list,
2447 * so check and free it up if we do.
2448 */
2456 }
2463 }
2470 }
2471 }
2473 /*
2474 * Increment the pin count of the given buffer.
2475 * This value is protected by ipinlock spinlock in the mount structure.
2476 */
2477 void
2480 {
2484 }
2486 /*
2487 * Decrement the pin count of the given inode, and wake up
2488 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2489 * inode must have been previously pinned with a call to xfs_ipin().
2490 */
2491 void
2494 {
2499 }
2501 /*
2502 * This is called to unpin an inode. It can be directed to wait or to return
2503 * immediately without waiting for the inode to be unpinned. The caller must
2504 * have the inode locked in at least shared mode so that the buffer cannot be
2505 * subsequently pinned once someone is waiting for it to be unpinned.
2506 */
2507 STATIC void
2511 {
2518 /* Give the log a push to start the unpinning I/O */
2523 }
2528 {
2530 }
2535 {
2537 }
2540 /*
2541 * xfs_iextents_copy()
2542 *
2543 * This is called to copy the REAL extents (as opposed to the delayed
2544 * allocation extents) from the inode into the given buffer. It
2545 * returns the number of bytes copied into the buffer.
2546 *
2547 * If there are no delayed allocation extents, then we can just
2548 * memcpy() the extents into the buffer. Otherwise, we need to
2549 * examine each extent in turn and skip those which are delayed.
2550 */
2551 int
2556 {
2561 xfs_fsblock_t start_block;
2571 /*
2572 * There are some delayed allocation extents in the
2573 * inode, so copy the extents one at a time and skip
2574 * the delayed ones. There must be at least one
2575 * non-delayed extent.
2576 */
2582 /*
2583 * It's a delayed allocation extent, so skip it.
2584 */
2586 }
2588 /* Translate to on disk format */
2591 dp++;
2592 copied++;
2593 }
2598 }
2600 /*
2601 * Each of the following cases stores data into the same region
2602 * of the on-disk inode, so only one of them can be valid at
2603 * any given time. While it is possible to have conflicting formats
2604 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2605 * in EXTENTS format, this can only happen when the fork has
2606 * changed formats after being modified but before being flushed.
2607 * In these cases, the format always takes precedence, because the
2608 * format indicates the current state of the fork.
2609 */
2610 /*ARGSUSED*/
2611 STATIC void
2618 {
2622 #ifdef XFS_TRANS_DEBUG
2624 #endif
2635 /*
2636 * This can happen if we gave up in iformat in an error path,
2637 * for the attribute fork.
2638 */
2642 }
2652 }
2666 whichfork);
2667 }
2676 XFS_BROOT_SIZE_ADJ));
2680 }
2687 }
2696 }
2702 }
2703 }
2705 STATIC int
2709 {
2734 /* really need a gang lookup range call here */
2744 /* if the inode lies outside this cluster, we're done. */
2747 /*
2748 * Do an un-protected check to see if the inode is dirty and
2749 * is a candidate for flushing. These checks will be repeated
2750 * later after the appropriate locks are acquired.
2751 */
2755 /*
2756 * Try to get locks. If any are unavailable or it is pinned,
2757 * then this inode cannot be flushed and is skipped.
2758 */
2765 }
2770 }
2772 /*
2773 * arriving here means that this inode can be flushed. First
2774 * re-check that it's dirty before flushing.
2775 */
2782 }
2783 clcount++;
2786 }
2788 }
2793 }
2795 out_free:
2801 cluster_corrupt_out:
2802 /*
2803 * Corruption detected in the clustering loop. Invalidate the
2804 * inode buffer and shut down the filesystem.
2805 */
2807 /*
2808 * Clean up the buffer. If it was B_DELWRI, just release it --
2809 * brelse can handle it with no problems. If not, shut down the
2810 * filesystem before releasing the buffer.
2811 */
2819 /*
2820 * Just like incore_relse: if we have b_iodone functions,
2821 * mark the buffer as an error and call them. Otherwise
2822 * mark it as stale and brelse.
2823 */
2833 }
2834 }
2836 /*
2837 * Unlocks the flush lock
2838 */
2842 }
2844 /*
2845 * xfs_iflush() will write a modified inode's changes out to the
2846 * inode's on disk home. The caller must have the inode lock held
2847 * in at least shared mode and the inode flush completion must be
2848 * active as well. The inode lock will still be held upon return from
2849 * the call and the caller is free to unlock it.
2850 * The inode flush will be completed when the inode reaches the disk.
2851 * The flags indicate how the inode's buffer should be written out.
2852 */
2853 int
2856 uint flags)
2857 {
2876 /*
2877 * If the inode isn't dirty, then just release the inode flush lock and
2878 * do nothing.
2879 */
2883 }
2885 /*
2886 * We can't flush the inode until it is unpinned, so wait for it if we
2887 * are allowed to block. We know noone new can pin it, because we are
2888 * holding the inode lock shared and you need to hold it exclusively to
2889 * pin the inode.
2890 *
2891 * If we are not allowed to block, force the log out asynchronously so
2892 * that when we come back the inode will be unpinned. If other inodes
2893 * in the same cluster are dirty, they will probably write the inode
2894 * out for us if they occur after the log force completes.
2895 */
2900 }
2903 /*
2904 * For stale inodes we cannot rely on the backing buffer remaining
2905 * stale in cache for the remaining life of the stale inode and so
2906 * xfs_itobp() below may give us a buffer that no longer contains
2907 * inodes below. We have to check this after ensuring the inode is
2908 * unpinned so that it is safe to reclaim the stale inode after the
2909 * flush call.
2910 */
2914 }
2916 /*
2917 * This may have been unpinned because the filesystem is shutting
2918 * down forcibly. If that's the case we must not write this inode
2919 * to disk, because the log record didn't make it to disk!
2920 */
2927 }
2929 /*
2930 * Decide how buffer will be flushed out. This is done before
2931 * the call to xfs_iflush_int because this field is zeroed by it.
2932 */
2934 /*
2935 * Flush out the inode buffer according to the directions
2936 * of the caller. In the cases where the caller has given
2937 * us a choice choose the non-delwri case. This is because
2938 * the inode is in the AIL and we need to get it out soon.
2939 */
2957 }
2976 }
2977 }
2979 /*
2980 * Get the buffer containing the on-disk inode.
2981 */
2987 }
2989 /*
2990 * First flush out the inode that xfs_iflush was called with.
2991 */
2996 /*
2997 * If the buffer is pinned then push on the log now so we won't
2998 * get stuck waiting in the write for too long.
2999 */
3003 /*
3004 * inode clustering:
3005 * see if other inodes can be gathered into this write
3006 */
3017 }
3020 corrupt_out:
3023 cluster_corrupt_out:
3024 /*
3025 * Unlocks the flush lock
3026 */
3029 }
3032 STATIC int
3036 {
3040 #ifdef XFS_TRANS_DEBUG
3042 #endif
3053 /*
3054 * If the inode isn't dirty, then just release the inode
3055 * flush lock and do nothing.
3056 */
3060 }
3062 /* set *dip = inode's place in the buffer */
3065 /*
3066 * Clear i_update_core before copying out the data.
3067 * This is for coordination with our timestamp updates
3068 * that don't hold the inode lock. They will always
3069 * update the timestamps BEFORE setting i_update_core,
3070 * so if we clear i_update_core after they set it we
3071 * are guaranteed to see their updates to the timestamps.
3072 * I believe that this depends on strongly ordered memory
3073 * semantics, but we have that. We use the SYNCHRONIZE
3074 * macro to make sure that the compiler does not reorder
3075 * the i_update_core access below the data copy below.
3076 */
3080 /*
3081 * Make sure to get the latest timestamps from the Linux inode.
3082 */
3091 }
3098 }
3108 }
3119 }
3120 }
3123 XFS_RANDOM_IFLUSH_5)) {
3129 ip);
3131 }
3138 }
3139 /*
3140 * bump the flush iteration count, used to detect flushes which
3141 * postdate a log record during recovery.
3142 */
3146 /*
3147 * Copy the dirty parts of the inode into the on-disk
3148 * inode. We always copy out the core of the inode,
3149 * because if the inode is dirty at all the core must
3150 * be.
3151 */
3154 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3158 /*
3159 * If this is really an old format inode and the superblock version
3160 * has not been updated to support only new format inodes, then
3161 * convert back to the old inode format. If the superblock version
3162 * has been updated, then make the conversion permanent.
3163 */
3167 /*
3168 * Convert it back.
3169 */
3173 /*
3174 * The superblock version has already been bumped,
3175 * so just make the conversion to the new inode
3176 * format permanent.
3177 */
3186 }
3187 }
3194 /*
3195 * We've recorded everything logged in the inode, so we'd
3196 * like to clear the ilf_fields bits so we don't log and
3197 * flush things unnecessarily. However, we can't stop
3198 * logging all this information until the data we've copied
3199 * into the disk buffer is written to disk. If we did we might
3200 * overwrite the copy of the inode in the log with all the
3201 * data after re-logging only part of it, and in the face of
3202 * a crash we wouldn't have all the data we need to recover.
3203 *
3204 * What we do is move the bits to the ili_last_fields field.
3205 * When logging the inode, these bits are moved back to the
3206 * ilf_fields field. In the xfs_iflush_done() routine we
3207 * clear ili_last_fields, since we know that the information
3208 * those bits represent is permanently on disk. As long as
3209 * the flush completes before the inode is logged again, then
3210 * both ilf_fields and ili_last_fields will be cleared.
3211 *
3212 * We can play with the ilf_fields bits here, because the inode
3213 * lock must be held exclusively in order to set bits there
3214 * and the flush lock protects the ili_last_fields bits.
3215 * Set ili_logged so the flush done
3216 * routine can tell whether or not to look in the AIL.
3217 * Also, store the current LSN of the inode so that we can tell
3218 * whether the item has moved in the AIL from xfs_iflush_done().
3219 * In order to read the lsn we need the AIL lock, because
3220 * it is a 64 bit value that cannot be read atomically.
3221 */
3230 /*
3231 * Attach the function xfs_iflush_done to the inode's
3232 * buffer. This will remove the inode from the AIL
3233 * and unlock the inode's flush lock when the inode is
3234 * completely written to disk.
3235 */
3242 /*
3243 * We're flushing an inode which is not in the AIL and has
3244 * not been logged but has i_update_core set. For this
3245 * case we can use a B_DELWRI flush and immediately drop
3246 * the inode flush lock because we can avoid the whole
3247 * AIL state thing. It's OK to drop the flush lock now,
3248 * because we've already locked the buffer and to do anything
3249 * you really need both.
3250 */
3255 }
3257 }
3261 corrupt_out:
3263 }
3267 #ifdef XFS_ILOCK_TRACE
3268 void
3270 {
3279 }
3280 #endif
3282 /*
3283 * Return a pointer to the extent record at file index idx.
3284 */
3285 xfs_bmbt_rec_host_t *
3289 {
3304 }
3305 }
3307 /*
3308 * Insert new item(s) into the extent records for incore inode
3309 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3310 */
3311 void
3317 {
3324 }
3326 /*
3327 * This is called when the amount of space required for incore file
3328 * extents needs to be increased. The ext_diff parameter stores the
3329 * number of new extents being added and the idx parameter contains
3330 * the extent index where the new extents will be added. If the new
3331 * extents are being appended, then we just need to (re)allocate and
3332 * initialize the space. Otherwise, if the new extents are being
3333 * inserted into the middle of the existing entries, a bit more work
3334 * is required to make room for the new extents to be inserted. The
3335 * caller is responsible for filling in the new extent entries upon
3336 * return.
3337 */
3338 void
3343 {
3352 /*
3353 * If the new number of extents (nextents + ext_diff)
3354 * fits inside the inode, then continue to use the inline
3355 * extent buffer.
3356 */
3363 }
3367 }
3368 /*
3369 * Otherwise use a linear (direct) extent list.
3370 * If the extents are currently inside the inode,
3371 * xfs_iext_realloc_direct will switch us from
3372 * inline to direct extent allocation mode.
3373 */
3381 }
3382 }
3383 /* Indirection array */
3396 }
3397 /* Extents fit in target extent page */
3405 }
3408 }
3409 /* Insert a new extent page */
3413 }
3414 /*
3415 * If extent(s) are being appended to the last page in
3416 * the indirection array and the new extent(s) don't fit
3417 * in the page, then erp is NULL and erp_idx is set to
3418 * the next index needed in the indirection array.
3419 */
3428 erp_idx++;
3429 }
3430 }
3431 }
3432 }
3434 }
3436 /*
3437 * This is called when incore extents are being added to the indirection
3438 * array and the new extents do not fit in the target extent list. The
3439 * erp_idx parameter contains the irec index for the target extent list
3440 * in the indirection array, and the idx parameter contains the extent
3441 * index within the list. The number of extents being added is stored
3442 * in the count parameter.
3443 *
3444 * |-------| |-------|
3445 * | | | | idx - number of extents before idx
3446 * | idx | | count |
3447 * | | | | count - number of extents being inserted at idx
3448 * |-------| |-------|
3449 * | count | | nex2 | nex2 - number of extents after idx + count
3450 * |-------| |-------|
3451 */
3452 void
3458 {
3472 /*
3473 * Save second part of target extent list
3474 * (all extents past */
3482 }
3484 /*
3485 * Add the new extents to the end of the target
3486 * list, then allocate new irec record(s) and
3487 * extent buffer(s) as needed to store the rest
3488 * of the new extents.
3489 */
3496 }
3498 erp_idx++;
3504 }
3506 /* Add nex2 extents back to indirection array */
3508 xfs_extnum_t ext_avail;
3514 /*
3515 * If nex2 extents fit in the current page, append
3516 * nex2_ep after the new extents.
3517 */
3520 }
3521 /*
3522 * Otherwise, check if space is available in the
3523 * next page.
3524 */
3528 erp_idx++;
3529 erp++;
3530 /* Create a hole for nex2 extents */
3533 }
3534 /*
3535 * Final choice, create a new extent page for
3536 * nex2 extents.
3537 */
3539 erp_idx++;
3541 }
3546 }
3547 }
3549 /*
3550 * This is called when the amount of space required for incore file
3551 * extents needs to be decreased. The ext_diff parameter stores the
3552 * number of extents to be removed and the idx parameter contains
3553 * the extent index where the extents will be removed from.
3554 *
3555 * If the amount of space needed has decreased below the linear
3556 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3557 * extent array. Otherwise, use kmem_realloc() to adjust the
3558 * size to what is needed.
3559 */
3560 void
3565 {
3581 }
3583 }
3585 /*
3586 * This removes ext_diff extents from the inline buffer, beginning
3587 * at extent index idx.
3588 */
3589 void
3594 {
3613 }
3614 }
3616 /*
3617 * This removes ext_diff extents from a linear (direct) extent list,
3618 * beginning at extent index idx. If the extents are being removed
3619 * from the end of the list (ie. truncate) then we just need to re-
3620 * allocate the list to remove the extra space. Otherwise, if the
3621 * extents are being removed from the middle of the existing extent
3622 * entries, then we first need to move the extent records beginning
3623 * at idx + ext_diff up in the list to overwrite the records being
3624 * removed, then remove the extra space via kmem_realloc.
3625 */
3626 void
3631 {
3643 }
3644 /* Move extents up in the list (if needed) */
3650 }
3653 /*
3654 * Reallocate the direct extent list. If the extents
3655 * will fit inside the inode then xfs_iext_realloc_direct
3656 * will switch from direct to inline extent allocation
3657 * mode for us.
3658 */
3661 }
3663 /*
3664 * This is called when incore extents are being removed from the
3665 * indirection array and the extents being removed span multiple extent
3666 * buffers. The idx parameter contains the file extent index where we
3667 * want to begin removing extents, and the count parameter contains
3668 * how many extents need to be removed.
3669 *
3670 * |-------| |-------|
3671 * | nex1 | | | nex1 - number of extents before idx
3672 * |-------| | count |
3673 * | | | | count - number of extents being removed at idx
3674 * | count | |-------|
3675 * | | | nex2 | nex2 - number of extents after idx + count
3676 * |-------| |-------|
3677 */
3678 void
3683 {
3702 /*
3703 * Check for deletion of entire list;
3704 * xfs_iext_irec_remove() updates extent offsets.
3705 */
3712 XFS_IEXT_BUFSZ);
3718 }
3719 }
3720 /* Move extents up (if needed) */
3725 }
3726 /* Zero out rest of page */
3729 /* Update remaining counters */
3734 erp_idx++;
3735 erp++;
3736 }
3739 }
3741 /*
3742 * Create, destroy, or resize a linear (direct) block of extents.
3743 */
3744 void
3748 {
3757 /* Free extent records */
3760 }
3761 /* Resize direct extent list and zero any new bytes */
3763 /* Check if extents will fit inside the inode */
3769 }
3772 }
3776 rnew_size,
3778 }
3783 }
3784 }
3785 /*
3786 * Switch from the inline extent buffer to a direct
3787 * extent list. Be sure to include the inline extent
3788 * bytes in new_size.
3789 */
3794 }
3796 }
3799 }
3801 /*
3802 * Switch from linear (direct) extent records to inline buffer.
3803 */
3804 void
3808 {
3811 /*
3812 * The inline buffer was zeroed when we switched
3813 * from inline to direct extent allocation mode,
3814 * so we don't need to clear it here.
3815 */
3821 }
3823 /*
3824 * Switch from inline buffer to linear (direct) extent records.
3825 * new_size should already be rounded up to the next power of 2
3826 * by the caller (when appropriate), so use new_size as it is.
3827 * However, since new_size may be rounded up, we can't update
3828 * if_bytes here. It is the caller's responsibility to update
3829 * if_bytes upon return.
3830 */
3831 void
3835 {
3843 }
3845 }
3847 /*
3848 * Resize an extent indirection array to new_size bytes.
3849 */
3850 STATIC void
3854 {
3869 }
3870 }
3872 /*
3873 * Switch from indirection array to linear (direct) extent allocations.
3874 */
3875 STATIC void
3878 {
3898 }
3899 }
3901 /*
3902 * Free incore file extents.
3903 */
3904 void
3907 {
3915 }
3922 }
3926 }
3928 /*
3929 * Return a pointer to the extent record for file system block bno.
3930 */
3936 {
3951 }
3954 /* Find target extent list */
3962 }
3963 /* Binary search extent records */
3974 /* Convert back to file-based extent index */
3977 }
3980 }
3981 }
3982 /* Convert back to file-based extent index */
3985 }
3991 }
3992 }
3995 }
3997 /*
3998 * Return a pointer to the indirection array entry containing the
3999 * extent record for filesystem block bno. Store the index of the
4000 * target irec in *erp_idxp.
4001 */
4007 {
4031 }
4032 }
4035 }
4037 /*
4038 * Return a pointer to the indirection array entry containing the
4039 * extent record at file extent index *idxp. Store the index of the
4040 * target irec in *erp_idxp and store the page index of the target
4041 * extent record in *idxp.
4042 */
4043 xfs_ext_irec_t *
4049 {
4066 /* Binary search extent irec's */
4082 erp_idx++;
4088 }
4089 }
4093 }
4095 /*
4096 * Allocate and initialize an indirection array once the space needed
4097 * for incore extents increases above XFS_IEXT_BUFSZ.
4098 */
4099 void
4102 {
4118 }
4129 }
4131 /*
4132 * Allocate and initialize a new entry in the indirection array.
4133 */
4134 xfs_ext_irec_t *
4138 {
4146 /* Resize indirection array */
4149 /*
4150 * Move records down in the array so the
4151 * new page can use erp_idx.
4152 */
4156 }
4159 /* Initialize new extent record */
4168 }
4170 /*
4171 * Remove a record from the indirection array.
4172 */
4173 void
4177 {
4189 }
4190 /* Compact extent records */
4194 }
4195 /*
4196 * Manually free the last extent record from the indirection
4197 * array. A call to xfs_iext_realloc_indirect() with a size
4198 * of zero would result in a call to xfs_iext_destroy() which
4199 * would in turn call this function again, creating a nasty
4200 * infinite loop.
4201 */
4207 }
4209 }
4211 /*
4212 * This is called to clean up large amounts of unused memory allocated
4213 * by the indirection array. Before compacting anything though, verify
4214 * that the indirection array is still needed and switch back to the
4215 * linear extent list (or even the inline buffer) if possible. The
4216 * compaction policy is as follows:
4217 *
4218 * Full Compaction: Extents fit into a single page (or inline buffer)
4219 * Partial Compaction: Extents occupy less than 50% of allocated space
4220 * No Compaction: Extents occupy at least 50% of allocated space
4221 */
4222 void
4225 {
4242 }
4243 }
4245 /*
4246 * Combine extents from neighboring extent pages.
4247 */
4248 void
4251 {
4267 /*
4268 * Free page before removing extent record
4269 * so er_extoffs don't get modified in
4270 * xfs_iext_irec_remove.
4271 */
4277 erp_idx++;
4278 }
4279 }
4280 }
4282 /*
4283 * This is called to update the er_extoff field in the indirection
4284 * array when extents have been added or removed from one of the
4285 * extent lists. erp_idx contains the irec index to begin updating
4286 * at and ext_diff contains the number of extents that were added
4287 * or removed.
4288 */
4289 void
4294 {
4302 }
4303 }