| blob | 2dcb3d781ae5d5d2f9fee742ed8e868b12e615cd |
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 }
355 }
366 /*
367 * no local regular files yet
368 */
371 "corrupt inode %Lu "
375 XFS_ERRLEVEL_LOW,
378 }
383 "corrupt inode %Lu "
388 XFS_ERRLEVEL_LOW,
391 }
406 }
412 }
415 }
429 "corrupt inode %Lu "
434 XFS_ERRLEVEL_LOW,
437 }
450 }
455 }
457 }
459 /*
460 * The file is in-lined in the on-disk inode.
461 * If it fits into if_inline_data, then copy
462 * it there, otherwise allocate a buffer for it
463 * and copy the data there. Either way, set
464 * if_data to point at the data.
465 * If we allocate a buffer for the data, make
466 * sure that its size is a multiple of 4 and
467 * record the real size in i_real_bytes.
468 */
469 STATIC int
475 {
479 /*
480 * If the size is unreasonable, then something
481 * is wrong and we just bail out rather than crash in
482 * kmem_alloc() or memcpy() below.
483 */
486 "corrupt inode %Lu "
493 }
503 }
511 }
513 /*
514 * The file consists of a set of extents all
515 * of which fit into the on-disk inode.
516 * If there are few enough extents to fit into
517 * the if_inline_ext, then copy them there.
518 * Otherwise allocate a buffer for them and copy
519 * them into it. Either way, set if_extents
520 * to point at the extents.
521 */
522 STATIC int
527 {
538 /*
539 * If the number of extents is unreasonable, then something
540 * is wrong and we just bail out rather than crash in
541 * kmem_alloc() or memcpy() below.
542 */
550 }
557 else
568 }
575 XFS_ERRLEVEL_LOW,
578 }
579 }
582 }
584 /*
585 * The file has too many extents to fit into
586 * the inode, so they are in B-tree format.
587 * Allocate a buffer for the root of the B-tree
588 * and copy the root into it. The i_extents
589 * field will remain NULL until all of the
590 * extents are read in (when they are needed).
591 */
592 STATIC int
597 {
600 /* REFERENCED */
609 /*
610 * blow out if -- fork has less extents than can fit in
611 * fork (fork shouldn't be a btree format), root btree
612 * block has more records than can fit into the fork,
613 * or the number of extents is greater than the number of
614 * blocks.
615 */
626 }
631 /*
632 * Copy and convert from the on-disk structure
633 * to the in-memory structure.
634 */
642 }
644 STATIC void
648 {
677 }
679 void
683 {
712 }
714 STATIC uint
716 __uint16_t di_flags)
717 {
749 }
752 }
754 uint
757 {
762 }
764 uint
767 {
770 }
772 /*
773 * Read the disk inode attributes into the in-core inode structure.
774 */
775 int
780 xfs_daddr_t bno,
781 uint iget_flags)
782 {
787 /*
788 * Fill in the location information in the in-core inode.
789 */
796 /*
797 * Get pointers to the on-disk inode and the buffer containing it.
798 */
805 /*
806 * If we got something that isn't an inode it means someone
807 * (nfs or dmi) has a stale handle.
808 */
810 #ifdef DEBUG
812 "dip->di_magic (0x%x) != "
815 XFS_DINODE_MAGIC);
819 }
821 /*
822 * If the on-disk inode is already linked to a directory
823 * entry, copy all of the inode into the in-core inode.
824 * xfs_iformat() handles copying in the inode format
825 * specific information.
826 * Otherwise, just get the truly permanent information.
827 */
832 #ifdef DEBUG
835 error);
838 }
844 /*
845 * Make sure to pull in the mode here as well in
846 * case the inode is released without being used.
847 * This ensures that xfs_inactive() will see that
848 * the inode is already free and not try to mess
849 * with the uninitialized part of it.
850 */
852 /*
853 * Initialize the per-fork minima and maxima for a new
854 * inode here. xfs_iformat will do it for old inodes.
855 */
858 }
860 /*
861 * The inode format changed when we moved the link count and
862 * made it 32 bits long. If this is an old format inode,
863 * convert it in memory to look like a new one. If it gets
864 * flushed to disk we will convert back before flushing or
865 * logging it. We zero out the new projid field and the old link
866 * count field. We'll handle clearing the pad field (the remains
867 * of the old uuid field) when we actually convert the inode to
868 * the new format. We don't change the version number so that we
869 * can distinguish this from a real new format inode.
870 */
875 }
880 /*
881 * Mark the buffer containing the inode as something to keep
882 * around for a while. This helps to keep recently accessed
883 * meta-data in-core longer.
884 */
887 /*
888 * Use xfs_trans_brelse() to release the buffer containing the
889 * on-disk inode, because it was acquired with xfs_trans_read_buf()
890 * in xfs_itobp() above. If tp is NULL, this is just a normal
891 * brelse(). If we're within a transaction, then xfs_trans_brelse()
892 * will only release the buffer if it is not dirty within the
893 * transaction. It will be OK to release the buffer in this case,
894 * because inodes on disk are never destroyed and we will be
895 * locking the new in-core inode before putting it in the hash
896 * table where other processes can find it. Thus we don't have
897 * to worry about the inode being changed just because we released
898 * the buffer.
899 */
900 out_brelse:
903 }
905 /*
906 * Read in extents from a btree-format inode.
907 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
908 */
909 int
914 {
917 xfs_extnum_t nextents;
924 }
929 /*
930 * We know that the size is valid (it's checked in iformat_btree)
931 */
941 }
944 }
946 /*
947 * Allocate an inode on disk and return a copy of its in-core version.
948 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
949 * appropriately within the inode. The uid and gid for the inode are
950 * set according to the contents of the given cred structure.
951 *
952 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
953 * has a free inode available, call xfs_iget()
954 * to obtain the in-core version of the allocated inode. Finally,
955 * fill in the inode and log its initial contents. In this case,
956 * ialloc_context would be set to NULL and call_again set to false.
957 *
958 * If xfs_dialloc() does not have an available inode,
959 * it will replenish its supply by doing an allocation. Since we can
960 * only do one allocation within a transaction without deadlocks, we
961 * must commit the current transaction before returning the inode itself.
962 * In this case, therefore, we will set call_again to true and return.
963 * The caller should then commit the current transaction, start a new
964 * transaction, and call xfs_ialloc() again to actually get the inode.
965 *
966 * To ensure that some other process does not grab the inode that
967 * was allocated during the first call to xfs_ialloc(), this routine
968 * also returns the [locked] bp pointing to the head of the freelist
969 * as ialloc_context. The caller should hold this buffer across
970 * the commit and pass it back into this routine on the second call.
971 *
972 * If we are allocating quota inodes, we do not have a parent inode
973 * to attach to or associate with (i.e. pip == NULL) because they
974 * are not linked into the directory structure - they are attached
975 * directly to the superblock - and so have no parent.
976 */
977 int
981 mode_t mode,
982 xfs_nlink_t nlink,
983 xfs_dev_t rdev,
985 xfs_prid_t prid,
990 {
991 xfs_ino_t ino;
993 uint flags;
995 timespec_t tv;
998 /*
999 * Call the space management code to pick
1000 * the on-disk inode to be allocated.
1001 */
1009 }
1012 /*
1013 * Get the in-core inode with the lock held exclusively.
1014 * This is because we're setting fields here we need
1015 * to prevent others from looking at until we're done.
1016 */
1032 /*
1033 * If the superblock version is up to where we support new format
1034 * inodes and this is currently an old format inode, then change
1035 * the inode version number now. This way we only do the conversion
1036 * here rather than here and in the flush/logging code.
1037 */
1041 /*
1042 * We've already zeroed the old link count, the projid field,
1043 * and the pad field.
1044 */
1045 }
1047 /*
1048 * Project ids won't be stored on disk if we are using a version 1 inode.
1049 */
1057 }
1058 }
1060 /*
1061 * If the group ID of the new file does not match the effective group
1062 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1063 * (and only if the irix_sgid_inherit compatibility variable is set).
1064 */
1069 }
1082 /*
1083 * di_gen will have been taken care of in xfs_iread.
1084 */
1101 /*
1102 * we can't set up filestreams until after the VFS inode
1103 * is set up properly.
1104 */
1107 /* fall through */
1118 }
1125 }
1126 }
1128 xfs_inherit_noatime)
1131 xfs_inherit_nodump)
1134 xfs_inherit_sync)
1137 xfs_inherit_nosymlinks)
1142 xfs_inherit_nodefrag)
1147 }
1148 /* FALLTHROUGH */
1157 }
1158 /*
1159 * Attribute fork settings for new inode.
1160 */
1164 /*
1165 * Log the new values stuffed into the inode.
1166 */
1169 /* now that we have an i_mode we can setup inode ops and unlock */
1172 /* now we have set up the vfs inode we can associate the filestream */
1179 }
1183 }
1185 /*
1186 * Check to make sure that there are no blocks allocated to the
1187 * file beyond the size of the file. We don't check this for
1188 * files with fixed size extents or real time extents, but we
1189 * at least do it for regular files.
1190 */
1191 #ifdef DEBUG
1192 void
1196 xfs_fsize_t isize)
1197 {
1198 xfs_fileoff_t map_first;
1213 /*
1214 * The filesystem could be shutting down, so bmapi may return
1215 * an error.
1216 */
1220 map_first),
1226 }
1229 /*
1230 * Calculate the last possible buffered byte in a file. This must
1231 * include data that was buffered beyond the EOF by the write code.
1232 * This also needs to deal with overflowing the xfs_fsize_t type
1233 * which can happen for sizes near the limit.
1234 *
1235 * We also need to take into account any blocks beyond the EOF. It
1236 * may be the case that they were buffered by a write which failed.
1237 * In that case the pages will still be in memory, but the inode size
1238 * will never have been updated.
1239 */
1240 STATIC xfs_fsize_t
1243 {
1245 xfs_fsize_t last_byte;
1246 xfs_fileoff_t last_block;
1247 xfs_fileoff_t size_last_block;
1253 /*
1254 * Only check for blocks beyond the EOF if the extents have
1255 * been read in. This eliminates the need for the inode lock,
1256 * and it also saves us from looking when it really isn't
1257 * necessary.
1258 */
1262 XFS_DATA_FORK);
1266 }
1269 }
1276 }
1280 }
1282 }
1284 #if defined(XFS_RW_TRACE)
1285 STATIC void
1290 xfs_fsize_t new_size,
1291 xfs_off_t toss_start,
1292 xfs_off_t toss_finish)
1293 {
1296 }
1315 }
1316 #else
1317 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1318 #endif
1320 /*
1321 * Start the truncation of the file to new_size. The new size
1322 * must be smaller than the current size. This routine will
1323 * clear the buffer and page caches of file data in the removed
1324 * range, and xfs_itruncate_finish() will remove the underlying
1325 * disk blocks.
1326 *
1327 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1328 * must NOT have the inode lock held at all. This is because we're
1329 * calling into the buffer/page cache code and we can't hold the
1330 * inode lock when we do so.
1331 *
1332 * We need to wait for any direct I/Os in flight to complete before we
1333 * proceed with the truncate. This is needed to prevent the extents
1334 * being read or written by the direct I/Os from being removed while the
1335 * I/O is in flight as there is no other method of synchronising
1336 * direct I/O with the truncate operation. Also, because we hold
1337 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1338 * started until the truncate completes and drops the lock. Essentially,
1339 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1340 * ordering between direct I/Os and the truncate operation.
1341 *
1342 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1343 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1344 * in the case that the caller is locking things out of order and
1345 * may not be able to call xfs_itruncate_finish() with the inode lock
1346 * held without dropping the I/O lock. If the caller must drop the
1347 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1348 * must be called again with all the same restrictions as the initial
1349 * call.
1350 */
1351 int
1354 uint flags,
1355 xfs_fsize_t new_size)
1356 {
1357 xfs_fsize_t last_byte;
1358 xfs_off_t toss_start;
1369 /* wait for the completion of any pending DIOs */
1373 /*
1374 * Call toss_pages or flushinval_pages to get rid of pages
1375 * overlapping the region being removed. We have to use
1376 * the less efficient flushinval_pages in the case that the
1377 * caller may not be able to finish the truncate without
1378 * dropping the inode's I/O lock. Make sure
1379 * to catch any pages brought in by buffers overlapping
1380 * the EOF by searching out beyond the isize by our
1381 * block size. We round new_size up to a block boundary
1382 * so that we don't toss things on the same block as
1383 * new_size but before it.
1384 *
1385 * Before calling toss_page or flushinval_pages, make sure to
1386 * call remapf() over the same region if the file is mapped.
1387 * This frees up mapped file references to the pages in the
1388 * given range and for the flushinval_pages case it ensures
1389 * that we get the latest mapped changes flushed out.
1390 */
1394 /*
1395 * The place to start tossing is beyond our maximum
1396 * file size, so there is no way that the data extended
1397 * out there.
1398 */
1400 }
1403 last_byte);
1411 }
1412 }
1414 #ifdef DEBUG
1417 }
1418 #endif
1420 }
1422 /*
1423 * Shrink the file to the given new_size. The new size must be smaller than
1424 * the current size. This will free up the underlying blocks in the removed
1425 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1426 *
1427 * The transaction passed to this routine must have made a permanent log
1428 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1429 * given transaction and start new ones, so make sure everything involved in
1430 * the transaction is tidy before calling here. Some transaction will be
1431 * returned to the caller to be committed. The incoming transaction must
1432 * already include the inode, and both inode locks must be held exclusively.
1433 * The inode must also be "held" within the transaction. On return the inode
1434 * will be "held" within the returned transaction. This routine does NOT
1435 * require any disk space to be reserved for it within the transaction.
1436 *
1437 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1438 * indicates the fork which is to be truncated. For the attribute fork we only
1439 * support truncation to size 0.
1440 *
1441 * We use the sync parameter to indicate whether or not the first transaction
1442 * we perform might have to be synchronous. For the attr fork, it needs to be
1443 * so if the unlink of the inode is not yet known to be permanent in the log.
1444 * This keeps us from freeing and reusing the blocks of the attribute fork
1445 * before the unlink of the inode becomes permanent.
1446 *
1447 * For the data fork, we normally have to run synchronously if we're being
1448 * called out of the inactive path or we're being called out of the create path
1449 * where we're truncating an existing file. Either way, the truncate needs to
1450 * be sync so blocks don't reappear in the file with altered data in case of a
1451 * crash. wsync filesystems can run the first case async because anything that
1452 * shrinks the inode has to run sync so by the time we're called here from
1453 * inactive, the inode size is permanently set to 0.
1454 *
1455 * Calls from the truncate path always need to be sync unless we're in a wsync
1456 * filesystem and the file has already been unlinked.
1457 *
1458 * The caller is responsible for correctly setting the sync parameter. It gets
1459 * too hard for us to guess here which path we're being called out of just
1460 * based on inode state.
1461 *
1462 * If we get an error, we must return with the inode locked and linked into the
1463 * current transaction. This keeps things simple for the higher level code,
1464 * because it always knows that the inode is locked and held in the transaction
1465 * that returns to it whether errors occur or not. We don't mark the inode
1466 * dirty on error so that transactions can be easily aborted if possible.
1467 */
1468 int
1472 xfs_fsize_t new_size,
1475 {
1476 xfs_fsblock_t first_block;
1477 xfs_fileoff_t first_unmap_block;
1478 xfs_fileoff_t last_block;
1484 xfs_bmap_free_t free_list;
1500 /*
1501 * We only support truncating the entire attribute fork.
1502 */
1505 }
1508 /*
1509 * The first thing we do is set the size to new_size permanently
1510 * on disk. This way we don't have to worry about anyone ever
1511 * being able to look at the data being freed even in the face
1512 * of a crash. What we're getting around here is the case where
1513 * we free a block, it is allocated to another file, it is written
1514 * to, and then we crash. If the new data gets written to the
1515 * file but the log buffers containing the free and reallocation
1516 * don't, then we'd end up with garbage in the blocks being freed.
1517 * As long as we make the new_size permanent before actually
1518 * freeing any blocks it doesn't matter if they get writtten to.
1519 *
1520 * The callers must signal into us whether or not the size
1521 * setting here must be synchronous. There are a few cases
1522 * where it doesn't have to be synchronous. Those cases
1523 * occur if the file is unlinked and we know the unlink is
1524 * permanent or if the blocks being truncated are guaranteed
1525 * to be beyond the inode eof (regardless of the link count)
1526 * and the eof value is permanent. Both of these cases occur
1527 * only on wsync-mounted filesystems. In those cases, we're
1528 * guaranteed that no user will ever see the data in the blocks
1529 * that are being truncated so the truncate can run async.
1530 * In the free beyond eof case, the file may wind up with
1531 * more blocks allocated to it than it needs if we crash
1532 * and that won't get fixed until the next time the file
1533 * is re-opened and closed but that's ok as that shouldn't
1534 * be too many blocks.
1535 *
1536 * However, we can't just make all wsync xactions run async
1537 * because there's one call out of the create path that needs
1538 * to run sync where it's truncating an existing file to size
1539 * 0 whose size is > 0.
1540 *
1541 * It's probably possible to come up with a test in this
1542 * routine that would correctly distinguish all the above
1543 * cases from the values of the function parameters and the
1544 * inode state but for sanity's sake, I've decided to let the
1545 * layers above just tell us. It's simpler to correctly figure
1546 * out in the layer above exactly under what conditions we
1547 * can run async and I think it's easier for others read and
1548 * follow the logic in case something has to be changed.
1549 * cscope is your friend -- rcc.
1550 *
1551 * The attribute fork is much simpler.
1552 *
1553 * For the attribute fork we allow the caller to tell us whether
1554 * the unlink of the inode that led to this call is yet permanent
1555 * in the on disk log. If it is not and we will be freeing extents
1556 * in this inode then we make the first transaction synchronous
1557 * to make sure that the unlink is permanent by the time we free
1558 * the blocks.
1559 */
1562 /*
1563 * If we are not changing the file size then do
1564 * not update the on-disk file size - we may be
1565 * called from xfs_inactive_free_eofblocks(). If we
1566 * update the on-disk file size and then the system
1567 * crashes before the contents of the file are
1568 * flushed to disk then the files may be full of
1569 * holes (ie NULL files bug).
1570 */
1575 }
1576 }
1581 }
1587 /*
1588 * Since it is possible for space to become allocated beyond
1589 * the end of the file (in a crash where the space is allocated
1590 * but the inode size is not yet updated), simply remove any
1591 * blocks which show up between the new EOF and the maximum
1592 * possible file size. If the first block to be removed is
1593 * beyond the maximum file size (ie it is the same as last_block),
1594 * then there is nothing to do.
1595 */
1603 }
1605 /*
1606 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1607 * will tell us whether it freed the entire range or
1608 * not. If this is a synchronous mount (wsync),
1609 * then we can tell bunmapi to keep all the
1610 * transactions asynchronous since the unlink
1611 * transaction that made this inode inactive has
1612 * already hit the disk. There's no danger of
1613 * the freed blocks being reused, there being a
1614 * crash, and the reused blocks suddenly reappearing
1615 * in this file with garbage in them once recovery
1616 * runs.
1617 */
1623 XFS_ITRUNC_MAX_EXTENTS,
1627 /*
1628 * If the bunmapi call encounters an error,
1629 * return to the caller where the transaction
1630 * can be properly aborted. We just need to
1631 * make sure we're not holding any resources
1632 * that we were not when we came in.
1633 */
1636 }
1638 /*
1639 * Duplicate the transaction that has the permanent
1640 * reservation and commit the old transaction.
1641 */
1645 /* link the inode into the next xact in the chain */
1649 }
1652 /*
1653 * If the bmap finish call encounters an error, return
1654 * to the caller where the transaction can be properly
1655 * aborted. We just need to make sure we're not
1656 * holding any resources that we were not when we came
1657 * in.
1658 *
1659 * Aborting from this point might lose some blocks in
1660 * the file system, but oh well.
1661 */
1664 }
1667 /*
1668 * Mark the inode dirty so it will be logged and
1669 * moved forward in the log as part of every commit.
1670 */
1672 }
1678 /* link the inode into the next transaction in the chain */
1684 /*
1685 * transaction commit worked ok so we can drop the extra ticket
1686 * reference that we gained in xfs_trans_dup()
1687 */
1691 XFS_TRANS_PERM_LOG_RES,
1692 XFS_ITRUNCATE_LOG_COUNT);
1695 }
1696 /*
1697 * Only update the size in the case of the data fork, but
1698 * always re-log the inode so that our permanent transaction
1699 * can keep on rolling it forward in the log.
1700 */
1703 /*
1704 * If we are not changing the file size then do
1705 * not update the on-disk file size - we may be
1706 * called from xfs_inactive_free_eofblocks(). If we
1707 * update the on-disk file size and then the system
1708 * crashes before the contents of the file are
1709 * flushed to disk then the files may be full of
1710 * holes (ie NULL files bug).
1711 */
1715 }
1716 }
1726 }
1728 /*
1729 * This is called when the inode's link count goes to 0.
1730 * We place the on-disk inode on a list in the AGI. It
1731 * will be pulled from this list when the inode is freed.
1732 */
1733 int
1737 {
1743 xfs_agino_t agino;
1754 /*
1755 * Get the agi buffer first. It ensures lock ordering
1756 * on the list.
1757 */
1763 /*
1764 * Get the index into the agi hash table for the
1765 * list this inode will go on.
1766 */
1774 /*
1775 * There is already another inode in the bucket we need
1776 * to add ourselves to. Add us at the front of the list.
1777 * Here we put the head pointer into our next pointer,
1778 * and then we fall through to point the head at us.
1779 */
1785 /* both on-disk, don't endian flip twice */
1793 }
1795 /*
1796 * Point the bucket head pointer at the inode being inserted.
1797 */
1805 }
1807 /*
1808 * Pull the on-disk inode from the AGI unlinked list.
1809 */
1810 STATIC int
1814 {
1815 xfs_ino_t next_ino;
1821 xfs_agnumber_t agno;
1822 xfs_agino_t agino;
1823 xfs_agino_t next_agino;
1833 /*
1834 * Get the agi buffer first. It ensures lock ordering
1835 * on the list.
1836 */
1843 /*
1844 * Get the index into the agi hash table for the
1845 * list this inode will go on.
1846 */
1854 /*
1855 * We're at the head of the list. Get the inode's
1856 * on-disk buffer to see if there is anyone after us
1857 * on the list. Only modify our next pointer if it
1858 * is not already NULLAGINO. This saves us the overhead
1859 * of dealing with the buffer when there is no need to
1860 * change it.
1861 */
1868 }
1881 }
1882 /*
1883 * Point the bucket head pointer at the next inode.
1884 */
1893 /*
1894 * We need to search the list for the inode being freed.
1895 */
1899 /*
1900 * If the last inode wasn't the one pointing to
1901 * us, then release its buffer since we're not
1902 * going to do anything with it.
1903 */
1906 }
1915 }
1919 }
1920 /*
1921 * Now last_ibp points to the buffer previous to us on
1922 * the unlinked list. Pull us from the list.
1923 */
1930 }
1944 }
1945 /*
1946 * Point the previous inode on the list to the next inode.
1947 */
1955 }
1957 }
1959 STATIC void
1963 xfs_ino_t inum)
1964 {
1970 xfs_daddr_t blkno;
1986 }
1995 /*
1996 * Look for each inode in memory and attempt to lock it,
1997 * we can be racing with flush and tail pushing here.
1998 * any inode we get the locks on, add to an array of
1999 * inode items to process later.
2000 *
2001 * The get the buffer lock, we could beat a flush
2002 * or tail pushing thread to the lock here, in which
2003 * case they will go looking for the inode buffer
2004 * and fail, we need some other form of interlock
2005 * here.
2006 */
2013 /* Inode not in memory or we found it already,
2014 * nothing to do
2015 */
2019 }
2024 }
2026 /* If we can get the locks then add it to the
2027 * list, otherwise by the time we get the bp lock
2028 * below it will already be attached to the
2029 * inode buffer.
2030 */
2032 /* This inode will already be locked - by us, lets
2033 * keep it that way.
2034 */
2043 }
2044 }
2047 }
2058 }
2061 }
2062 }
2064 }
2068 XFS_BUF_LOCK);
2081 pre_flushed++;
2082 }
2084 }
2095 }
2108 }
2109 }
2114 }
2118 }
2120 /*
2121 * This is called to return an inode to the inode free list.
2122 * The inode should already be truncated to 0 length and have
2123 * no pages associated with it. This routine also assumes that
2124 * the inode is already a part of the transaction.
2125 *
2126 * The on-disk copy of the inode will have been added to the list
2127 * of unlinked inodes in the AGI. We need to remove the inode from
2128 * that list atomically with respect to freeing it here.
2129 */
2130 int
2135 {
2138 xfs_ino_t first_ino;
2151 /*
2152 * Pull the on-disk inode from the AGI unlinked list.
2153 */
2157 }
2162 }
2171 /*
2172 * Bump the generation count so no one will be confused
2173 * by reincarnations of this inode.
2174 */
2183 /*
2184 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2185 * from picking up this inode when it is reclaimed (its incore state
2186 * initialzed but not flushed to disk yet). The in-core di_mode is
2187 * already cleared and a corresponding transaction logged.
2188 * The hack here just synchronizes the in-core to on-disk
2189 * di_mode value in advance before the actual inode sync to disk.
2190 * This is OK because the inode is already unlinked and would never
2191 * change its di_mode again for this inode generation.
2192 * This is a temporary hack that would require a proper fix
2193 * in the future.
2194 */
2199 }
2202 }
2204 /*
2205 * Reallocate the space for if_broot based on the number of records
2206 * being added or deleted as indicated in rec_diff. Move the records
2207 * and pointers in if_broot to fit the new size. When shrinking this
2208 * will eliminate holes between the records and pointers created by
2209 * the caller. When growing this will create holes to be filled in
2210 * by the caller.
2211 *
2212 * The caller must not request to add more records than would fit in
2213 * the on-disk inode root. If the if_broot is currently NULL, then
2214 * if we adding records one will be allocated. The caller must also
2215 * not request that the number of records go below zero, although
2216 * it can go to zero.
2217 *
2218 * ip -- the inode whose if_broot area is changing
2219 * ext_diff -- the change in the number of records, positive or negative,
2220 * requested for the if_broot array.
2221 */
2222 void
2227 {
2237 /*
2238 * Handle the degenerate case quietly.
2239 */
2242 }
2246 /*
2247 * If there wasn't any memory allocated before, just
2248 * allocate it now and get out.
2249 */
2255 }
2257 /*
2258 * If there is already an existing if_broot, then we need
2259 * to realloc() it and shift the pointers to their new
2260 * location. The records don't change location because
2261 * they are kept butted up against the btree block header.
2262 */
2268 KM_SLEEP);
2278 }
2280 /*
2281 * rec_diff is less than 0. In this case, we are shrinking the
2282 * if_broot buffer. It must already exist. If we go to zero
2283 * records, just get rid of the root and clear the status bit.
2284 */
2291 else
2295 /*
2296 * First copy over the btree block header.
2297 */
2302 }
2304 /*
2305 * Only copy the records and pointers if there are any.
2306 */
2308 /*
2309 * First copy the records.
2310 */
2315 /*
2316 * Then copy the pointers.
2317 */
2323 }
2330 }
2333 /*
2334 * This is called when the amount of space needed for if_data
2335 * is increased or decreased. The change in size is indicated by
2336 * the number of bytes that need to be added or deleted in the
2337 * byte_diff parameter.
2338 *
2339 * If the amount of space needed has decreased below the size of the
2340 * inline buffer, then switch to using the inline buffer. Otherwise,
2341 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2342 * to what is needed.
2343 *
2344 * ip -- the inode whose if_data area is changing
2345 * byte_diff -- the change in the number of bytes, positive or negative,
2346 * requested for the if_data array.
2347 */
2348 void
2353 {
2360 }
2369 }
2373 /*
2374 * If the valid extents/data can fit in if_inline_ext/data,
2375 * copy them from the malloc'd vector and free it.
2376 */
2382 new_size);
2385 }
2388 /*
2389 * Stuck with malloc/realloc.
2390 * For inline data, the underlying buffer must be
2391 * a multiple of 4 bytes in size so that it can be
2392 * logged and stay on word boundaries. We enforce
2393 * that here.
2394 */
2400 /*
2401 * Only do the realloc if the underlying size
2402 * is really changing.
2403 */
2407 real_size,
2409 KM_SLEEP);
2410 }
2416 }
2417 }
2421 }
2423 void
2427 {
2434 }
2436 /*
2437 * If the format is local, then we can't have an extents
2438 * array so just look for an inline data array. If we're
2439 * not local then we may or may not have an extents list,
2440 * so check and free it up if we do.
2441 */
2449 }
2456 }
2463 }
2464 }
2466 /*
2467 * Increment the pin count of the given buffer.
2468 * This value is protected by ipinlock spinlock in the mount structure.
2469 */
2470 void
2473 {
2477 }
2479 /*
2480 * Decrement the pin count of the given inode, and wake up
2481 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2482 * inode must have been previously pinned with a call to xfs_ipin().
2483 */
2484 void
2487 {
2492 }
2494 /*
2495 * This is called to unpin an inode. It can be directed to wait or to return
2496 * immediately without waiting for the inode to be unpinned. The caller must
2497 * have the inode locked in at least shared mode so that the buffer cannot be
2498 * subsequently pinned once someone is waiting for it to be unpinned.
2499 */
2500 STATIC void
2504 {
2511 /* Give the log a push to start the unpinning I/O */
2516 }
2521 {
2523 }
2528 {
2530 }
2533 /*
2534 * xfs_iextents_copy()
2535 *
2536 * This is called to copy the REAL extents (as opposed to the delayed
2537 * allocation extents) from the inode into the given buffer. It
2538 * returns the number of bytes copied into the buffer.
2539 *
2540 * If there are no delayed allocation extents, then we can just
2541 * memcpy() the extents into the buffer. Otherwise, we need to
2542 * examine each extent in turn and skip those which are delayed.
2543 */
2544 int
2549 {
2554 xfs_fsblock_t start_block;
2564 /*
2565 * There are some delayed allocation extents in the
2566 * inode, so copy the extents one at a time and skip
2567 * the delayed ones. There must be at least one
2568 * non-delayed extent.
2569 */
2575 /*
2576 * It's a delayed allocation extent, so skip it.
2577 */
2579 }
2581 /* Translate to on disk format */
2584 dp++;
2585 copied++;
2586 }
2591 }
2593 /*
2594 * Each of the following cases stores data into the same region
2595 * of the on-disk inode, so only one of them can be valid at
2596 * any given time. While it is possible to have conflicting formats
2597 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2598 * in EXTENTS format, this can only happen when the fork has
2599 * changed formats after being modified but before being flushed.
2600 * In these cases, the format always takes precedence, because the
2601 * format indicates the current state of the fork.
2602 */
2603 /*ARGSUSED*/
2604 STATIC void
2611 {
2615 #ifdef XFS_TRANS_DEBUG
2617 #endif
2628 /*
2629 * This can happen if we gave up in iformat in an error path,
2630 * for the attribute fork.
2631 */
2635 }
2645 }
2659 whichfork);
2660 }
2669 XFS_BROOT_SIZE_ADJ));
2673 }
2680 }
2689 }
2695 }
2696 }
2698 STATIC int
2702 {
2727 /* really need a gang lookup range call here */
2737 /* if the inode lies outside this cluster, we're done. */
2740 /*
2741 * Do an un-protected check to see if the inode is dirty and
2742 * is a candidate for flushing. These checks will be repeated
2743 * later after the appropriate locks are acquired.
2744 */
2748 /*
2749 * Try to get locks. If any are unavailable or it is pinned,
2750 * then this inode cannot be flushed and is skipped.
2751 */
2758 }
2763 }
2765 /*
2766 * arriving here means that this inode can be flushed. First
2767 * re-check that it's dirty before flushing.
2768 */
2775 }
2776 clcount++;
2779 }
2781 }
2786 }
2788 out_free:
2794 cluster_corrupt_out:
2795 /*
2796 * Corruption detected in the clustering loop. Invalidate the
2797 * inode buffer and shut down the filesystem.
2798 */
2800 /*
2801 * Clean up the buffer. If it was B_DELWRI, just release it --
2802 * brelse can handle it with no problems. If not, shut down the
2803 * filesystem before releasing the buffer.
2804 */
2812 /*
2813 * Just like incore_relse: if we have b_iodone functions,
2814 * mark the buffer as an error and call them. Otherwise
2815 * mark it as stale and brelse.
2816 */
2826 }
2827 }
2829 /*
2830 * Unlocks the flush lock
2831 */
2835 }
2837 /*
2838 * xfs_iflush() will write a modified inode's changes out to the
2839 * inode's on disk home. The caller must have the inode lock held
2840 * in at least shared mode and the inode flush completion must be
2841 * active as well. The inode lock will still be held upon return from
2842 * the call and the caller is free to unlock it.
2843 * The inode flush will be completed when the inode reaches the disk.
2844 * The flags indicate how the inode's buffer should be written out.
2845 */
2846 int
2849 uint flags)
2850 {
2869 /*
2870 * If the inode isn't dirty, then just release the inode
2871 * flush lock and do nothing.
2872 */
2876 }
2878 /*
2879 * We can't flush the inode until it is unpinned, so wait for it if we
2880 * are allowed to block. We know noone new can pin it, because we are
2881 * holding the inode lock shared and you need to hold it exclusively to
2882 * pin the inode.
2883 *
2884 * If we are not allowed to block, force the log out asynchronously so
2885 * that when we come back the inode will be unpinned. If other inodes
2886 * in the same cluster are dirty, they will probably write the inode
2887 * out for us if they occur after the log force completes.
2888 */
2893 }
2896 /*
2897 * This may have been unpinned because the filesystem is shutting
2898 * down forcibly. If that's the case we must not write this inode
2899 * to disk, because the log record didn't make it to disk!
2900 */
2907 }
2909 /*
2910 * Decide how buffer will be flushed out. This is done before
2911 * the call to xfs_iflush_int because this field is zeroed by it.
2912 */
2914 /*
2915 * Flush out the inode buffer according to the directions
2916 * of the caller. In the cases where the caller has given
2917 * us a choice choose the non-delwri case. This is because
2918 * the inode is in the AIL and we need to get it out soon.
2919 */
2937 }
2956 }
2957 }
2959 /*
2960 * Get the buffer containing the on-disk inode.
2961 */
2967 }
2969 /*
2970 * First flush out the inode that xfs_iflush was called with.
2971 */
2976 /*
2977 * If the buffer is pinned then push on the log now so we won't
2978 * get stuck waiting in the write for too long.
2979 */
2983 /*
2984 * inode clustering:
2985 * see if other inodes can be gathered into this write
2986 */
2997 }
3000 corrupt_out:
3003 cluster_corrupt_out:
3004 /*
3005 * Unlocks the flush lock
3006 */
3009 }
3012 STATIC int
3016 {
3020 #ifdef XFS_TRANS_DEBUG
3022 #endif
3033 /*
3034 * If the inode isn't dirty, then just release the inode
3035 * flush lock and do nothing.
3036 */
3040 }
3042 /* set *dip = inode's place in the buffer */
3045 /*
3046 * Clear i_update_core before copying out the data.
3047 * This is for coordination with our timestamp updates
3048 * that don't hold the inode lock. They will always
3049 * update the timestamps BEFORE setting i_update_core,
3050 * so if we clear i_update_core after they set it we
3051 * are guaranteed to see their updates to the timestamps.
3052 * I believe that this depends on strongly ordered memory
3053 * semantics, but we have that. We use the SYNCHRONIZE
3054 * macro to make sure that the compiler does not reorder
3055 * the i_update_core access below the data copy below.
3056 */
3060 /*
3061 * Make sure to get the latest atime from the Linux inode.
3062 */
3071 }
3078 }
3088 }
3099 }
3100 }
3103 XFS_RANDOM_IFLUSH_5)) {
3109 ip);
3111 }
3118 }
3119 /*
3120 * bump the flush iteration count, used to detect flushes which
3121 * postdate a log record during recovery.
3122 */
3126 /*
3127 * Copy the dirty parts of the inode into the on-disk
3128 * inode. We always copy out the core of the inode,
3129 * because if the inode is dirty at all the core must
3130 * be.
3131 */
3134 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3138 /*
3139 * If this is really an old format inode and the superblock version
3140 * has not been updated to support only new format inodes, then
3141 * convert back to the old inode format. If the superblock version
3142 * has been updated, then make the conversion permanent.
3143 */
3147 /*
3148 * Convert it back.
3149 */
3153 /*
3154 * The superblock version has already been bumped,
3155 * so just make the conversion to the new inode
3156 * format permanent.
3157 */
3166 }
3167 }
3174 /*
3175 * We've recorded everything logged in the inode, so we'd
3176 * like to clear the ilf_fields bits so we don't log and
3177 * flush things unnecessarily. However, we can't stop
3178 * logging all this information until the data we've copied
3179 * into the disk buffer is written to disk. If we did we might
3180 * overwrite the copy of the inode in the log with all the
3181 * data after re-logging only part of it, and in the face of
3182 * a crash we wouldn't have all the data we need to recover.
3183 *
3184 * What we do is move the bits to the ili_last_fields field.
3185 * When logging the inode, these bits are moved back to the
3186 * ilf_fields field. In the xfs_iflush_done() routine we
3187 * clear ili_last_fields, since we know that the information
3188 * those bits represent is permanently on disk. As long as
3189 * the flush completes before the inode is logged again, then
3190 * both ilf_fields and ili_last_fields will be cleared.
3191 *
3192 * We can play with the ilf_fields bits here, because the inode
3193 * lock must be held exclusively in order to set bits there
3194 * and the flush lock protects the ili_last_fields bits.
3195 * Set ili_logged so the flush done
3196 * routine can tell whether or not to look in the AIL.
3197 * Also, store the current LSN of the inode so that we can tell
3198 * whether the item has moved in the AIL from xfs_iflush_done().
3199 * In order to read the lsn we need the AIL lock, because
3200 * it is a 64 bit value that cannot be read atomically.
3201 */
3210 /*
3211 * Attach the function xfs_iflush_done to the inode's
3212 * buffer. This will remove the inode from the AIL
3213 * and unlock the inode's flush lock when the inode is
3214 * completely written to disk.
3215 */
3222 /*
3223 * We're flushing an inode which is not in the AIL and has
3224 * not been logged but has i_update_core set. For this
3225 * case we can use a B_DELWRI flush and immediately drop
3226 * the inode flush lock because we can avoid the whole
3227 * AIL state thing. It's OK to drop the flush lock now,
3228 * because we've already locked the buffer and to do anything
3229 * you really need both.
3230 */
3235 }
3237 }
3241 corrupt_out:
3243 }
3247 #ifdef XFS_ILOCK_TRACE
3248 void
3250 {
3259 }
3260 #endif
3262 /*
3263 * Return a pointer to the extent record at file index idx.
3264 */
3265 xfs_bmbt_rec_host_t *
3269 {
3284 }
3285 }
3287 /*
3288 * Insert new item(s) into the extent records for incore inode
3289 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3290 */
3291 void
3297 {
3304 }
3306 /*
3307 * This is called when the amount of space required for incore file
3308 * extents needs to be increased. The ext_diff parameter stores the
3309 * number of new extents being added and the idx parameter contains
3310 * the extent index where the new extents will be added. If the new
3311 * extents are being appended, then we just need to (re)allocate and
3312 * initialize the space. Otherwise, if the new extents are being
3313 * inserted into the middle of the existing entries, a bit more work
3314 * is required to make room for the new extents to be inserted. The
3315 * caller is responsible for filling in the new extent entries upon
3316 * return.
3317 */
3318 void
3323 {
3332 /*
3333 * If the new number of extents (nextents + ext_diff)
3334 * fits inside the inode, then continue to use the inline
3335 * extent buffer.
3336 */
3343 }
3347 }
3348 /*
3349 * Otherwise use a linear (direct) extent list.
3350 * If the extents are currently inside the inode,
3351 * xfs_iext_realloc_direct will switch us from
3352 * inline to direct extent allocation mode.
3353 */
3361 }
3362 }
3363 /* Indirection array */
3376 }
3377 /* Extents fit in target extent page */
3385 }
3388 }
3389 /* Insert a new extent page */
3393 }
3394 /*
3395 * If extent(s) are being appended to the last page in
3396 * the indirection array and the new extent(s) don't fit
3397 * in the page, then erp is NULL and erp_idx is set to
3398 * the next index needed in the indirection array.
3399 */
3408 erp_idx++;
3409 }
3410 }
3411 }
3412 }
3414 }
3416 /*
3417 * This is called when incore extents are being added to the indirection
3418 * array and the new extents do not fit in the target extent list. The
3419 * erp_idx parameter contains the irec index for the target extent list
3420 * in the indirection array, and the idx parameter contains the extent
3421 * index within the list. The number of extents being added is stored
3422 * in the count parameter.
3423 *
3424 * |-------| |-------|
3425 * | | | | idx - number of extents before idx
3426 * | idx | | count |
3427 * | | | | count - number of extents being inserted at idx
3428 * |-------| |-------|
3429 * | count | | nex2 | nex2 - number of extents after idx + count
3430 * |-------| |-------|
3431 */
3432 void
3438 {
3452 /*
3453 * Save second part of target extent list
3454 * (all extents past */
3462 }
3464 /*
3465 * Add the new extents to the end of the target
3466 * list, then allocate new irec record(s) and
3467 * extent buffer(s) as needed to store the rest
3468 * of the new extents.
3469 */
3476 }
3478 erp_idx++;
3484 }
3486 /* Add nex2 extents back to indirection array */
3488 xfs_extnum_t ext_avail;
3494 /*
3495 * If nex2 extents fit in the current page, append
3496 * nex2_ep after the new extents.
3497 */
3500 }
3501 /*
3502 * Otherwise, check if space is available in the
3503 * next page.
3504 */
3508 erp_idx++;
3509 erp++;
3510 /* Create a hole for nex2 extents */
3513 }
3514 /*
3515 * Final choice, create a new extent page for
3516 * nex2 extents.
3517 */
3519 erp_idx++;
3521 }
3526 }
3527 }
3529 /*
3530 * This is called when the amount of space required for incore file
3531 * extents needs to be decreased. The ext_diff parameter stores the
3532 * number of extents to be removed and the idx parameter contains
3533 * the extent index where the extents will be removed from.
3534 *
3535 * If the amount of space needed has decreased below the linear
3536 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3537 * extent array. Otherwise, use kmem_realloc() to adjust the
3538 * size to what is needed.
3539 */
3540 void
3545 {
3561 }
3563 }
3565 /*
3566 * This removes ext_diff extents from the inline buffer, beginning
3567 * at extent index idx.
3568 */
3569 void
3574 {
3593 }
3594 }
3596 /*
3597 * This removes ext_diff extents from a linear (direct) extent list,
3598 * beginning at extent index idx. If the extents are being removed
3599 * from the end of the list (ie. truncate) then we just need to re-
3600 * allocate the list to remove the extra space. Otherwise, if the
3601 * extents are being removed from the middle of the existing extent
3602 * entries, then we first need to move the extent records beginning
3603 * at idx + ext_diff up in the list to overwrite the records being
3604 * removed, then remove the extra space via kmem_realloc.
3605 */
3606 void
3611 {
3623 }
3624 /* Move extents up in the list (if needed) */
3630 }
3633 /*
3634 * Reallocate the direct extent list. If the extents
3635 * will fit inside the inode then xfs_iext_realloc_direct
3636 * will switch from direct to inline extent allocation
3637 * mode for us.
3638 */
3641 }
3643 /*
3644 * This is called when incore extents are being removed from the
3645 * indirection array and the extents being removed span multiple extent
3646 * buffers. The idx parameter contains the file extent index where we
3647 * want to begin removing extents, and the count parameter contains
3648 * how many extents need to be removed.
3649 *
3650 * |-------| |-------|
3651 * | nex1 | | | nex1 - number of extents before idx
3652 * |-------| | count |
3653 * | | | | count - number of extents being removed at idx
3654 * | count | |-------|
3655 * | | | nex2 | nex2 - number of extents after idx + count
3656 * |-------| |-------|
3657 */
3658 void
3663 {
3682 /*
3683 * Check for deletion of entire list;
3684 * xfs_iext_irec_remove() updates extent offsets.
3685 */
3692 XFS_IEXT_BUFSZ);
3698 }
3699 }
3700 /* Move extents up (if needed) */
3705 }
3706 /* Zero out rest of page */
3709 /* Update remaining counters */
3714 erp_idx++;
3715 erp++;
3716 }
3719 }
3721 /*
3722 * Create, destroy, or resize a linear (direct) block of extents.
3723 */
3724 void
3728 {
3737 /* Free extent records */
3740 }
3741 /* Resize direct extent list and zero any new bytes */
3743 /* Check if extents will fit inside the inode */
3749 }
3752 }
3756 rnew_size,
3758 }
3763 }
3764 }
3765 /*
3766 * Switch from the inline extent buffer to a direct
3767 * extent list. Be sure to include the inline extent
3768 * bytes in new_size.
3769 */
3774 }
3776 }
3779 }
3781 /*
3782 * Switch from linear (direct) extent records to inline buffer.
3783 */
3784 void
3788 {
3791 /*
3792 * The inline buffer was zeroed when we switched
3793 * from inline to direct extent allocation mode,
3794 * so we don't need to clear it here.
3795 */
3801 }
3803 /*
3804 * Switch from inline buffer to linear (direct) extent records.
3805 * new_size should already be rounded up to the next power of 2
3806 * by the caller (when appropriate), so use new_size as it is.
3807 * However, since new_size may be rounded up, we can't update
3808 * if_bytes here. It is the caller's responsibility to update
3809 * if_bytes upon return.
3810 */
3811 void
3815 {
3823 }
3825 }
3827 /*
3828 * Resize an extent indirection array to new_size bytes.
3829 */
3830 STATIC void
3834 {
3849 }
3850 }
3852 /*
3853 * Switch from indirection array to linear (direct) extent allocations.
3854 */
3855 STATIC void
3858 {
3878 }
3879 }
3881 /*
3882 * Free incore file extents.
3883 */
3884 void
3887 {
3895 }
3902 }
3906 }
3908 /*
3909 * Return a pointer to the extent record for file system block bno.
3910 */
3916 {
3931 }
3934 /* Find target extent list */
3942 }
3943 /* Binary search extent records */
3954 /* Convert back to file-based extent index */
3957 }
3960 }
3961 }
3962 /* Convert back to file-based extent index */
3965 }
3971 }
3972 }
3975 }
3977 /*
3978 * Return a pointer to the indirection array entry containing the
3979 * extent record for filesystem block bno. Store the index of the
3980 * target irec in *erp_idxp.
3981 */
3987 {
4011 }
4012 }
4015 }
4017 /*
4018 * Return a pointer to the indirection array entry containing the
4019 * extent record at file extent index *idxp. Store the index of the
4020 * target irec in *erp_idxp and store the page index of the target
4021 * extent record in *idxp.
4022 */
4023 xfs_ext_irec_t *
4029 {
4046 /* Binary search extent irec's */
4062 erp_idx++;
4068 }
4069 }
4073 }
4075 /*
4076 * Allocate and initialize an indirection array once the space needed
4077 * for incore extents increases above XFS_IEXT_BUFSZ.
4078 */
4079 void
4082 {
4098 }
4109 }
4111 /*
4112 * Allocate and initialize a new entry in the indirection array.
4113 */
4114 xfs_ext_irec_t *
4118 {
4126 /* Resize indirection array */
4129 /*
4130 * Move records down in the array so the
4131 * new page can use erp_idx.
4132 */
4136 }
4139 /* Initialize new extent record */
4148 }
4150 /*
4151 * Remove a record from the indirection array.
4152 */
4153 void
4157 {
4169 }
4170 /* Compact extent records */
4174 }
4175 /*
4176 * Manually free the last extent record from the indirection
4177 * array. A call to xfs_iext_realloc_indirect() with a size
4178 * of zero would result in a call to xfs_iext_destroy() which
4179 * would in turn call this function again, creating a nasty
4180 * infinite loop.
4181 */
4187 }
4189 }
4191 /*
4192 * This is called to clean up large amounts of unused memory allocated
4193 * by the indirection array. Before compacting anything though, verify
4194 * that the indirection array is still needed and switch back to the
4195 * linear extent list (or even the inline buffer) if possible. The
4196 * compaction policy is as follows:
4197 *
4198 * Full Compaction: Extents fit into a single page (or inline buffer)
4199 * Partial Compaction: Extents occupy less than 50% of allocated space
4200 * No Compaction: Extents occupy at least 50% of allocated space
4201 */
4202 void
4205 {
4222 }
4223 }
4225 /*
4226 * Combine extents from neighboring extent pages.
4227 */
4228 void
4231 {
4247 /*
4248 * Free page before removing extent record
4249 * so er_extoffs don't get modified in
4250 * xfs_iext_irec_remove.
4251 */
4257 erp_idx++;
4258 }
4259 }
4260 }
4262 /*
4263 * This is called to update the er_extoff field in the indirection
4264 * array when extents have been added or removed from one of the
4265 * extent lists. erp_idx contains the irec index to begin updating
4266 * at and ext_diff contains the number of extents that were added
4267 * or removed.
4268 */
4269 void
4274 {
4282 }
4283 }