| blob | 123b20c8cbf23ddd54e697d0f3a0d058e126f44b |
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>
59 /*
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
62 */
63 #define XFS_ITRUNC_MAX_EXTENTS 2
70 #ifdef DEBUG
71 /*
72 * Make sure that the extents in the given memory buffer
73 * are valid.
74 */
75 STATIC void
79 xfs_exntfmt_t fmt)
80 {
81 xfs_bmbt_irec_t irec;
82 xfs_bmbt_rec_host_t rec;
92 }
93 }
95 #define xfs_validate_extents(ifp, nrecs, fmt)
98 /*
99 * Check that none of the inode's in the buffer have a next
100 * unlinked field of 0.
101 */
102 #if defined(DEBUG)
103 void
107 {
119 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 bp);
122 }
123 }
124 }
125 #endif
127 /*
128 * Find the buffer associated with the given inode map
129 * We do basic validation checks on the buffer once it has been
130 * retrieved from disk.
131 */
132 STATIC int
138 uint buf_flags,
139 uint iget_flags)
140 {
151 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
156 }
158 }
160 /*
161 * Validate the magic number and version of every inode in the buffer
162 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
163 */
164 #ifdef DEBUG
168 #endif
179 XFS_ERRTAG_ITOBP_INOTOBP,
180 XFS_RANDOM_ITOBP_INOTOBP))) {
184 }
187 #ifdef DEBUG
189 "Device %s - bad inode magic/vsn "
194 #endif
197 }
198 }
202 /*
203 * Mark the buffer as an inode buffer now that it looks good
204 */
209 }
211 /*
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
217 *
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
220 *
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
223 */
224 int
228 xfs_ino_t ino,
232 uint imap_flags)
233 {
251 }
254 /*
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
259 * that buffer.
260 *
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
263 *
264 * The inode is expected to already been mapped to its buffer and read
265 * in once, thus we can use the mapping information stored in the inode
266 * rather than calling xfs_imap(). This allows us to avoid the overhead
267 * of looking at the inode btree for small block file systems
268 * (see xfs_imap()).
269 */
270 int
277 uint buf_flags)
278 {
293 }
298 }
300 /*
301 * Move inode type and inode format specific information from the
302 * on-disk inode to the in-core inode. For fifos, devs, and sockets
303 * this means set if_rdev to the proper value. For files, directories,
304 * and symlinks this means to bring in the in-line data or extent
305 * pointers. For a file in B-tree format, only the root is immediately
306 * brought in-core. The rest will be in-lined in if_extents when it
307 * is first referenced (see xfs_iread_extents()).
308 */
309 STATIC int
313 {
317 xfs_fsize_t di_size;
335 }
345 }
356 }
367 /*
368 * no local regular files yet
369 */
372 "corrupt inode %Lu "
376 XFS_ERRLEVEL_LOW,
379 }
384 "corrupt inode %Lu "
389 XFS_ERRLEVEL_LOW,
392 }
407 }
413 }
416 }
430 "corrupt inode %Lu "
435 XFS_ERRLEVEL_LOW,
438 }
451 }
456 }
458 }
460 /*
461 * The file is in-lined in the on-disk inode.
462 * If it fits into if_inline_data, then copy
463 * it there, otherwise allocate a buffer for it
464 * and copy the data there. Either way, set
465 * if_data to point at the data.
466 * If we allocate a buffer for the data, make
467 * sure that its size is a multiple of 4 and
468 * record the real size in i_real_bytes.
469 */
470 STATIC int
476 {
480 /*
481 * If the size is unreasonable, then something
482 * is wrong and we just bail out rather than crash in
483 * kmem_alloc() or memcpy() below.
484 */
487 "corrupt inode %Lu "
494 }
504 }
512 }
514 /*
515 * The file consists of a set of extents all
516 * of which fit into the on-disk inode.
517 * If there are few enough extents to fit into
518 * the if_inline_ext, then copy them there.
519 * Otherwise allocate a buffer for them and copy
520 * them into it. Either way, set if_extents
521 * to point at the extents.
522 */
523 STATIC int
528 {
539 /*
540 * If the number of extents is unreasonable, then something
541 * is wrong and we just bail out rather than crash in
542 * kmem_alloc() or memcpy() below.
543 */
551 }
558 else
569 }
576 XFS_ERRLEVEL_LOW,
579 }
580 }
583 }
585 /*
586 * The file has too many extents to fit into
587 * the inode, so they are in B-tree format.
588 * Allocate a buffer for the root of the B-tree
589 * and copy the root into it. The i_extents
590 * field will remain NULL until all of the
591 * extents are read in (when they are needed).
592 */
593 STATIC int
598 {
601 /* REFERENCED */
610 /*
611 * blow out if -- fork has less extents than can fit in
612 * fork (fork shouldn't be a btree format), root btree
613 * block has more records than can fit into the fork,
614 * or the number of extents is greater than the number of
615 * blocks.
616 */
627 }
632 /*
633 * Copy and convert from the on-disk structure
634 * to the in-memory structure.
635 */
643 }
645 void
649 {
678 }
680 void
684 {
713 }
715 STATIC uint
717 __uint16_t di_flags)
718 {
750 }
753 }
755 uint
758 {
763 }
765 uint
768 {
771 }
773 /*
774 * Read the disk inode attributes into the in-core inode structure.
775 */
776 int
781 xfs_daddr_t bno,
782 uint iget_flags)
783 {
788 /*
789 * Fill in the location information in the in-core inode.
790 */
797 /*
798 * Get pointers to the on-disk inode and the buffer containing it.
799 */
806 /*
807 * If we got something that isn't an inode it means someone
808 * (nfs or dmi) has a stale handle.
809 */
811 #ifdef DEBUG
813 "dip->di_magic (0x%x) != "
816 XFS_DINODE_MAGIC);
820 }
822 /*
823 * If the on-disk inode is already linked to a directory
824 * entry, copy all of the inode into the in-core inode.
825 * xfs_iformat() handles copying in the inode format
826 * specific information.
827 * Otherwise, just get the truly permanent information.
828 */
833 #ifdef DEBUG
836 error);
839 }
845 /*
846 * Make sure to pull in the mode here as well in
847 * case the inode is released without being used.
848 * This ensures that xfs_inactive() will see that
849 * the inode is already free and not try to mess
850 * with the uninitialized part of it.
851 */
853 /*
854 * Initialize the per-fork minima and maxima for a new
855 * inode here. xfs_iformat will do it for old inodes.
856 */
859 }
861 /*
862 * The inode format changed when we moved the link count and
863 * made it 32 bits long. If this is an old format inode,
864 * convert it in memory to look like a new one. If it gets
865 * flushed to disk we will convert back before flushing or
866 * logging it. We zero out the new projid field and the old link
867 * count field. We'll handle clearing the pad field (the remains
868 * of the old uuid field) when we actually convert the inode to
869 * the new format. We don't change the version number so that we
870 * can distinguish this from a real new format inode.
871 */
876 }
881 /*
882 * Mark the buffer containing the inode as something to keep
883 * around for a while. This helps to keep recently accessed
884 * meta-data in-core longer.
885 */
888 /*
889 * Use xfs_trans_brelse() to release the buffer containing the
890 * on-disk inode, because it was acquired with xfs_trans_read_buf()
891 * in xfs_itobp() above. If tp is NULL, this is just a normal
892 * brelse(). If we're within a transaction, then xfs_trans_brelse()
893 * will only release the buffer if it is not dirty within the
894 * transaction. It will be OK to release the buffer in this case,
895 * because inodes on disk are never destroyed and we will be
896 * locking the new in-core inode before putting it in the hash
897 * table where other processes can find it. Thus we don't have
898 * to worry about the inode being changed just because we released
899 * the buffer.
900 */
901 out_brelse:
904 }
906 /*
907 * Read in extents from a btree-format inode.
908 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
909 */
910 int
915 {
918 xfs_extnum_t nextents;
925 }
930 /*
931 * We know that the size is valid (it's checked in iformat_btree)
932 */
942 }
945 }
947 /*
948 * Allocate an inode on disk and return a copy of its in-core version.
949 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
950 * appropriately within the inode. The uid and gid for the inode are
951 * set according to the contents of the given cred structure.
952 *
953 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
954 * has a free inode available, call xfs_iget()
955 * to obtain the in-core version of the allocated inode. Finally,
956 * fill in the inode and log its initial contents. In this case,
957 * ialloc_context would be set to NULL and call_again set to false.
958 *
959 * If xfs_dialloc() does not have an available inode,
960 * it will replenish its supply by doing an allocation. Since we can
961 * only do one allocation within a transaction without deadlocks, we
962 * must commit the current transaction before returning the inode itself.
963 * In this case, therefore, we will set call_again to true and return.
964 * The caller should then commit the current transaction, start a new
965 * transaction, and call xfs_ialloc() again to actually get the inode.
966 *
967 * To ensure that some other process does not grab the inode that
968 * was allocated during the first call to xfs_ialloc(), this routine
969 * also returns the [locked] bp pointing to the head of the freelist
970 * as ialloc_context. The caller should hold this buffer across
971 * the commit and pass it back into this routine on the second call.
972 *
973 * If we are allocating quota inodes, we do not have a parent inode
974 * to attach to or associate with (i.e. pip == NULL) because they
975 * are not linked into the directory structure - they are attached
976 * directly to the superblock - and so have no parent.
977 */
978 int
982 mode_t mode,
983 xfs_nlink_t nlink,
984 xfs_dev_t rdev,
986 xfs_prid_t prid,
991 {
992 xfs_ino_t ino;
994 uint flags;
996 timespec_t tv;
999 /*
1000 * Call the space management code to pick
1001 * the on-disk inode to be allocated.
1002 */
1010 }
1013 /*
1014 * Get the in-core inode with the lock held exclusively.
1015 * This is because we're setting fields here we need
1016 * to prevent others from looking at until we're done.
1017 */
1033 /*
1034 * If the superblock version is up to where we support new format
1035 * inodes and this is currently an old format inode, then change
1036 * the inode version number now. This way we only do the conversion
1037 * here rather than here and in the flush/logging code.
1038 */
1042 /*
1043 * We've already zeroed the old link count, the projid field,
1044 * and the pad field.
1045 */
1046 }
1048 /*
1049 * Project ids won't be stored on disk if we are using a version 1 inode.
1050 */
1058 }
1059 }
1061 /*
1062 * If the group ID of the new file does not match the effective group
1063 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1064 * (and only if the irix_sgid_inherit compatibility variable is set).
1065 */
1070 }
1083 /*
1084 * di_gen will have been taken care of in xfs_iread.
1085 */
1102 /*
1103 * we can't set up filestreams until after the VFS inode
1104 * is set up properly.
1105 */
1108 /* fall through */
1119 }
1126 }
1127 }
1129 xfs_inherit_noatime)
1132 xfs_inherit_nodump)
1135 xfs_inherit_sync)
1138 xfs_inherit_nosymlinks)
1143 xfs_inherit_nodefrag)
1148 }
1149 /* FALLTHROUGH */
1158 }
1159 /*
1160 * Attribute fork settings for new inode.
1161 */
1165 /*
1166 * Log the new values stuffed into the inode.
1167 */
1170 /* now that we have an i_mode we can setup inode ops and unlock */
1173 /* now we have set up the vfs inode we can associate the filestream */
1180 }
1184 }
1186 /*
1187 * Check to make sure that there are no blocks allocated to the
1188 * file beyond the size of the file. We don't check this for
1189 * files with fixed size extents or real time extents, but we
1190 * at least do it for regular files.
1191 */
1192 #ifdef DEBUG
1193 void
1197 xfs_fsize_t isize)
1198 {
1199 xfs_fileoff_t map_first;
1214 /*
1215 * The filesystem could be shutting down, so bmapi may return
1216 * an error.
1217 */
1221 map_first),
1227 }
1230 /*
1231 * Calculate the last possible buffered byte in a file. This must
1232 * include data that was buffered beyond the EOF by the write code.
1233 * This also needs to deal with overflowing the xfs_fsize_t type
1234 * which can happen for sizes near the limit.
1235 *
1236 * We also need to take into account any blocks beyond the EOF. It
1237 * may be the case that they were buffered by a write which failed.
1238 * In that case the pages will still be in memory, but the inode size
1239 * will never have been updated.
1240 */
1241 xfs_fsize_t
1244 {
1246 xfs_fsize_t last_byte;
1247 xfs_fileoff_t last_block;
1248 xfs_fileoff_t size_last_block;
1254 /*
1255 * Only check for blocks beyond the EOF if the extents have
1256 * been read in. This eliminates the need for the inode lock,
1257 * and it also saves us from looking when it really isn't
1258 * necessary.
1259 */
1263 XFS_DATA_FORK);
1267 }
1270 }
1277 }
1281 }
1283 }
1285 #if defined(XFS_RW_TRACE)
1286 STATIC void
1291 xfs_fsize_t new_size,
1292 xfs_off_t toss_start,
1293 xfs_off_t toss_finish)
1294 {
1297 }
1316 }
1317 #else
1318 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1319 #endif
1321 /*
1322 * Start the truncation of the file to new_size. The new size
1323 * must be smaller than the current size. This routine will
1324 * clear the buffer and page caches of file data in the removed
1325 * range, and xfs_itruncate_finish() will remove the underlying
1326 * disk blocks.
1327 *
1328 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1329 * must NOT have the inode lock held at all. This is because we're
1330 * calling into the buffer/page cache code and we can't hold the
1331 * inode lock when we do so.
1332 *
1333 * We need to wait for any direct I/Os in flight to complete before we
1334 * proceed with the truncate. This is needed to prevent the extents
1335 * being read or written by the direct I/Os from being removed while the
1336 * I/O is in flight as there is no other method of synchronising
1337 * direct I/O with the truncate operation. Also, because we hold
1338 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1339 * started until the truncate completes and drops the lock. Essentially,
1340 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1341 * ordering between direct I/Os and the truncate operation.
1342 *
1343 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1344 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1345 * in the case that the caller is locking things out of order and
1346 * may not be able to call xfs_itruncate_finish() with the inode lock
1347 * held without dropping the I/O lock. If the caller must drop the
1348 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1349 * must be called again with all the same restrictions as the initial
1350 * call.
1351 */
1352 int
1355 uint flags,
1356 xfs_fsize_t new_size)
1357 {
1358 xfs_fsize_t last_byte;
1359 xfs_off_t toss_start;
1370 /* wait for the completion of any pending DIOs */
1374 /*
1375 * Call toss_pages or flushinval_pages to get rid of pages
1376 * overlapping the region being removed. We have to use
1377 * the less efficient flushinval_pages in the case that the
1378 * caller may not be able to finish the truncate without
1379 * dropping the inode's I/O lock. Make sure
1380 * to catch any pages brought in by buffers overlapping
1381 * the EOF by searching out beyond the isize by our
1382 * block size. We round new_size up to a block boundary
1383 * so that we don't toss things on the same block as
1384 * new_size but before it.
1385 *
1386 * Before calling toss_page or flushinval_pages, make sure to
1387 * call remapf() over the same region if the file is mapped.
1388 * This frees up mapped file references to the pages in the
1389 * given range and for the flushinval_pages case it ensures
1390 * that we get the latest mapped changes flushed out.
1391 */
1395 /*
1396 * The place to start tossing is beyond our maximum
1397 * file size, so there is no way that the data extended
1398 * out there.
1399 */
1401 }
1404 last_byte);
1412 }
1413 }
1415 #ifdef DEBUG
1418 }
1419 #endif
1421 }
1423 /*
1424 * Shrink the file to the given new_size. The new size must be smaller than
1425 * the current size. This will free up the underlying blocks in the removed
1426 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1427 *
1428 * The transaction passed to this routine must have made a permanent log
1429 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1430 * given transaction and start new ones, so make sure everything involved in
1431 * the transaction is tidy before calling here. Some transaction will be
1432 * returned to the caller to be committed. The incoming transaction must
1433 * already include the inode, and both inode locks must be held exclusively.
1434 * The inode must also be "held" within the transaction. On return the inode
1435 * will be "held" within the returned transaction. This routine does NOT
1436 * require any disk space to be reserved for it within the transaction.
1437 *
1438 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1439 * indicates the fork which is to be truncated. For the attribute fork we only
1440 * support truncation to size 0.
1441 *
1442 * We use the sync parameter to indicate whether or not the first transaction
1443 * we perform might have to be synchronous. For the attr fork, it needs to be
1444 * so if the unlink of the inode is not yet known to be permanent in the log.
1445 * This keeps us from freeing and reusing the blocks of the attribute fork
1446 * before the unlink of the inode becomes permanent.
1447 *
1448 * For the data fork, we normally have to run synchronously if we're being
1449 * called out of the inactive path or we're being called out of the create path
1450 * where we're truncating an existing file. Either way, the truncate needs to
1451 * be sync so blocks don't reappear in the file with altered data in case of a
1452 * crash. wsync filesystems can run the first case async because anything that
1453 * shrinks the inode has to run sync so by the time we're called here from
1454 * inactive, the inode size is permanently set to 0.
1455 *
1456 * Calls from the truncate path always need to be sync unless we're in a wsync
1457 * filesystem and the file has already been unlinked.
1458 *
1459 * The caller is responsible for correctly setting the sync parameter. It gets
1460 * too hard for us to guess here which path we're being called out of just
1461 * based on inode state.
1462 *
1463 * If we get an error, we must return with the inode locked and linked into the
1464 * current transaction. This keeps things simple for the higher level code,
1465 * because it always knows that the inode is locked and held in the transaction
1466 * that returns to it whether errors occur or not. We don't mark the inode
1467 * dirty on error so that transactions can be easily aborted if possible.
1468 */
1469 int
1473 xfs_fsize_t new_size,
1476 {
1477 xfs_fsblock_t first_block;
1478 xfs_fileoff_t first_unmap_block;
1479 xfs_fileoff_t last_block;
1485 xfs_bmap_free_t free_list;
1501 /*
1502 * We only support truncating the entire attribute fork.
1503 */
1506 }
1509 /*
1510 * The first thing we do is set the size to new_size permanently
1511 * on disk. This way we don't have to worry about anyone ever
1512 * being able to look at the data being freed even in the face
1513 * of a crash. What we're getting around here is the case where
1514 * we free a block, it is allocated to another file, it is written
1515 * to, and then we crash. If the new data gets written to the
1516 * file but the log buffers containing the free and reallocation
1517 * don't, then we'd end up with garbage in the blocks being freed.
1518 * As long as we make the new_size permanent before actually
1519 * freeing any blocks it doesn't matter if they get writtten to.
1520 *
1521 * The callers must signal into us whether or not the size
1522 * setting here must be synchronous. There are a few cases
1523 * where it doesn't have to be synchronous. Those cases
1524 * occur if the file is unlinked and we know the unlink is
1525 * permanent or if the blocks being truncated are guaranteed
1526 * to be beyond the inode eof (regardless of the link count)
1527 * and the eof value is permanent. Both of these cases occur
1528 * only on wsync-mounted filesystems. In those cases, we're
1529 * guaranteed that no user will ever see the data in the blocks
1530 * that are being truncated so the truncate can run async.
1531 * In the free beyond eof case, the file may wind up with
1532 * more blocks allocated to it than it needs if we crash
1533 * and that won't get fixed until the next time the file
1534 * is re-opened and closed but that's ok as that shouldn't
1535 * be too many blocks.
1536 *
1537 * However, we can't just make all wsync xactions run async
1538 * because there's one call out of the create path that needs
1539 * to run sync where it's truncating an existing file to size
1540 * 0 whose size is > 0.
1541 *
1542 * It's probably possible to come up with a test in this
1543 * routine that would correctly distinguish all the above
1544 * cases from the values of the function parameters and the
1545 * inode state but for sanity's sake, I've decided to let the
1546 * layers above just tell us. It's simpler to correctly figure
1547 * out in the layer above exactly under what conditions we
1548 * can run async and I think it's easier for others read and
1549 * follow the logic in case something has to be changed.
1550 * cscope is your friend -- rcc.
1551 *
1552 * The attribute fork is much simpler.
1553 *
1554 * For the attribute fork we allow the caller to tell us whether
1555 * the unlink of the inode that led to this call is yet permanent
1556 * in the on disk log. If it is not and we will be freeing extents
1557 * in this inode then we make the first transaction synchronous
1558 * to make sure that the unlink is permanent by the time we free
1559 * the blocks.
1560 */
1563 /*
1564 * If we are not changing the file size then do
1565 * not update the on-disk file size - we may be
1566 * called from xfs_inactive_free_eofblocks(). If we
1567 * update the on-disk file size and then the system
1568 * crashes before the contents of the file are
1569 * flushed to disk then the files may be full of
1570 * holes (ie NULL files bug).
1571 */
1576 }
1577 }
1582 }
1588 /*
1589 * Since it is possible for space to become allocated beyond
1590 * the end of the file (in a crash where the space is allocated
1591 * but the inode size is not yet updated), simply remove any
1592 * blocks which show up between the new EOF and the maximum
1593 * possible file size. If the first block to be removed is
1594 * beyond the maximum file size (ie it is the same as last_block),
1595 * then there is nothing to do.
1596 */
1604 }
1606 /*
1607 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1608 * will tell us whether it freed the entire range or
1609 * not. If this is a synchronous mount (wsync),
1610 * then we can tell bunmapi to keep all the
1611 * transactions asynchronous since the unlink
1612 * transaction that made this inode inactive has
1613 * already hit the disk. There's no danger of
1614 * the freed blocks being reused, there being a
1615 * crash, and the reused blocks suddenly reappearing
1616 * in this file with garbage in them once recovery
1617 * runs.
1618 */
1624 XFS_ITRUNC_MAX_EXTENTS,
1628 /*
1629 * If the bunmapi call encounters an error,
1630 * return to the caller where the transaction
1631 * can be properly aborted. We just need to
1632 * make sure we're not holding any resources
1633 * that we were not when we came in.
1634 */
1637 }
1639 /*
1640 * Duplicate the transaction that has the permanent
1641 * reservation and commit the old transaction.
1642 */
1646 /* link the inode into the next xact in the chain */
1650 }
1653 /*
1654 * If the bmap finish call encounters an error, return
1655 * to the caller where the transaction can be properly
1656 * aborted. We just need to make sure we're not
1657 * holding any resources that we were not when we came
1658 * in.
1659 *
1660 * Aborting from this point might lose some blocks in
1661 * the file system, but oh well.
1662 */
1665 }
1668 /*
1669 * Mark the inode dirty so it will be logged and
1670 * moved forward in the log as part of every commit.
1671 */
1673 }
1679 /* link the inode into the next transaction in the chain */
1685 /*
1686 * transaction commit worked ok so we can drop the extra ticket
1687 * reference that we gained in xfs_trans_dup()
1688 */
1692 XFS_TRANS_PERM_LOG_RES,
1693 XFS_ITRUNCATE_LOG_COUNT);
1696 }
1697 /*
1698 * Only update the size in the case of the data fork, but
1699 * always re-log the inode so that our permanent transaction
1700 * can keep on rolling it forward in the log.
1701 */
1704 /*
1705 * If we are not changing the file size then do
1706 * not update the on-disk file size - we may be
1707 * called from xfs_inactive_free_eofblocks(). If we
1708 * update the on-disk file size and then the system
1709 * crashes before the contents of the file are
1710 * flushed to disk then the files may be full of
1711 * holes (ie NULL files bug).
1712 */
1716 }
1717 }
1727 }
1729 /*
1730 * This is called when the inode's link count goes to 0.
1731 * We place the on-disk inode on a list in the AGI. It
1732 * will be pulled from this list when the inode is freed.
1733 */
1734 int
1738 {
1744 xfs_agino_t agino;
1755 /*
1756 * Get the agi buffer first. It ensures lock ordering
1757 * on the list.
1758 */
1764 /*
1765 * Get the index into the agi hash table for the
1766 * list this inode will go on.
1767 */
1775 /*
1776 * There is already another inode in the bucket we need
1777 * to add ourselves to. Add us at the front of the list.
1778 * Here we put the head pointer into our next pointer,
1779 * and then we fall through to point the head at us.
1780 */
1786 /* both on-disk, don't endian flip twice */
1794 }
1796 /*
1797 * Point the bucket head pointer at the inode being inserted.
1798 */
1806 }
1808 /*
1809 * Pull the on-disk inode from the AGI unlinked list.
1810 */
1811 STATIC int
1815 {
1816 xfs_ino_t next_ino;
1822 xfs_agnumber_t agno;
1823 xfs_agino_t agino;
1824 xfs_agino_t next_agino;
1834 /*
1835 * Get the agi buffer first. It ensures lock ordering
1836 * on the list.
1837 */
1844 /*
1845 * Get the index into the agi hash table for the
1846 * list this inode will go on.
1847 */
1855 /*
1856 * We're at the head of the list. Get the inode's
1857 * on-disk buffer to see if there is anyone after us
1858 * on the list. Only modify our next pointer if it
1859 * is not already NULLAGINO. This saves us the overhead
1860 * of dealing with the buffer when there is no need to
1861 * change it.
1862 */
1869 }
1882 }
1883 /*
1884 * Point the bucket head pointer at the next inode.
1885 */
1894 /*
1895 * We need to search the list for the inode being freed.
1896 */
1900 /*
1901 * If the last inode wasn't the one pointing to
1902 * us, then release its buffer since we're not
1903 * going to do anything with it.
1904 */
1907 }
1916 }
1920 }
1921 /*
1922 * Now last_ibp points to the buffer previous to us on
1923 * the unlinked list. Pull us from the list.
1924 */
1931 }
1945 }
1946 /*
1947 * Point the previous inode on the list to the next inode.
1948 */
1956 }
1958 }
1960 STATIC void
1964 xfs_ino_t inum)
1965 {
1971 xfs_daddr_t blkno;
1987 }
1996 /*
1997 * Look for each inode in memory and attempt to lock it,
1998 * we can be racing with flush and tail pushing here.
1999 * any inode we get the locks on, add to an array of
2000 * inode items to process later.
2001 *
2002 * The get the buffer lock, we could beat a flush
2003 * or tail pushing thread to the lock here, in which
2004 * case they will go looking for the inode buffer
2005 * and fail, we need some other form of interlock
2006 * here.
2007 */
2014 /* Inode not in memory or we found it already,
2015 * nothing to do
2016 */
2020 }
2025 }
2027 /* If we can get the locks then add it to the
2028 * list, otherwise by the time we get the bp lock
2029 * below it will already be attached to the
2030 * inode buffer.
2031 */
2033 /* This inode will already be locked - by us, lets
2034 * keep it that way.
2035 */
2044 }
2045 }
2048 }
2059 }
2062 }
2063 }
2065 }
2069 XFS_BUF_LOCK);
2082 pre_flushed++;
2083 }
2085 }
2096 }
2109 }
2110 }
2115 }
2119 }
2121 /*
2122 * This is called to return an inode to the inode free list.
2123 * The inode should already be truncated to 0 length and have
2124 * no pages associated with it. This routine also assumes that
2125 * the inode is already a part of the transaction.
2126 *
2127 * The on-disk copy of the inode will have been added to the list
2128 * of unlinked inodes in the AGI. We need to remove the inode from
2129 * that list atomically with respect to freeing it here.
2130 */
2131 int
2136 {
2139 xfs_ino_t first_ino;
2152 /*
2153 * Pull the on-disk inode from the AGI unlinked list.
2154 */
2158 }
2163 }
2172 /*
2173 * Bump the generation count so no one will be confused
2174 * by reincarnations of this inode.
2175 */
2184 /*
2185 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2186 * from picking up this inode when it is reclaimed (its incore state
2187 * initialzed but not flushed to disk yet). The in-core di_mode is
2188 * already cleared and a corresponding transaction logged.
2189 * The hack here just synchronizes the in-core to on-disk
2190 * di_mode value in advance before the actual inode sync to disk.
2191 * This is OK because the inode is already unlinked and would never
2192 * change its di_mode again for this inode generation.
2193 * This is a temporary hack that would require a proper fix
2194 * in the future.
2195 */
2200 }
2203 }
2205 /*
2206 * Reallocate the space for if_broot based on the number of records
2207 * being added or deleted as indicated in rec_diff. Move the records
2208 * and pointers in if_broot to fit the new size. When shrinking this
2209 * will eliminate holes between the records and pointers created by
2210 * the caller. When growing this will create holes to be filled in
2211 * by the caller.
2212 *
2213 * The caller must not request to add more records than would fit in
2214 * the on-disk inode root. If the if_broot is currently NULL, then
2215 * if we adding records one will be allocated. The caller must also
2216 * not request that the number of records go below zero, although
2217 * it can go to zero.
2218 *
2219 * ip -- the inode whose if_broot area is changing
2220 * ext_diff -- the change in the number of records, positive or negative,
2221 * requested for the if_broot array.
2222 */
2223 void
2228 {
2238 /*
2239 * Handle the degenerate case quietly.
2240 */
2243 }
2247 /*
2248 * If there wasn't any memory allocated before, just
2249 * allocate it now and get out.
2250 */
2256 }
2258 /*
2259 * If there is already an existing if_broot, then we need
2260 * to realloc() it and shift the pointers to their new
2261 * location. The records don't change location because
2262 * they are kept butted up against the btree block header.
2263 */
2269 KM_SLEEP);
2279 }
2281 /*
2282 * rec_diff is less than 0. In this case, we are shrinking the
2283 * if_broot buffer. It must already exist. If we go to zero
2284 * records, just get rid of the root and clear the status bit.
2285 */
2292 else
2296 /*
2297 * First copy over the btree block header.
2298 */
2303 }
2305 /*
2306 * Only copy the records and pointers if there are any.
2307 */
2309 /*
2310 * First copy the records.
2311 */
2316 /*
2317 * Then copy the pointers.
2318 */
2324 }
2331 }
2334 /*
2335 * This is called when the amount of space needed for if_data
2336 * is increased or decreased. The change in size is indicated by
2337 * the number of bytes that need to be added or deleted in the
2338 * byte_diff parameter.
2339 *
2340 * If the amount of space needed has decreased below the size of the
2341 * inline buffer, then switch to using the inline buffer. Otherwise,
2342 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2343 * to what is needed.
2344 *
2345 * ip -- the inode whose if_data area is changing
2346 * byte_diff -- the change in the number of bytes, positive or negative,
2347 * requested for the if_data array.
2348 */
2349 void
2354 {
2361 }
2370 }
2374 /*
2375 * If the valid extents/data can fit in if_inline_ext/data,
2376 * copy them from the malloc'd vector and free it.
2377 */
2383 new_size);
2386 }
2389 /*
2390 * Stuck with malloc/realloc.
2391 * For inline data, the underlying buffer must be
2392 * a multiple of 4 bytes in size so that it can be
2393 * logged and stay on word boundaries. We enforce
2394 * that here.
2395 */
2401 /*
2402 * Only do the realloc if the underlying size
2403 * is really changing.
2404 */
2408 real_size,
2410 KM_SLEEP);
2411 }
2417 }
2418 }
2422 }
2424 void
2428 {
2435 }
2437 /*
2438 * If the format is local, then we can't have an extents
2439 * array so just look for an inline data array. If we're
2440 * not local then we may or may not have an extents list,
2441 * so check and free it up if we do.
2442 */
2450 }
2457 }
2464 }
2465 }
2467 /*
2468 * Increment the pin count of the given buffer.
2469 * This value is protected by ipinlock spinlock in the mount structure.
2470 */
2471 void
2474 {
2478 }
2480 /*
2481 * Decrement the pin count of the given inode, and wake up
2482 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2483 * inode must have been previously pinned with a call to xfs_ipin().
2484 */
2485 void
2488 {
2493 }
2495 /*
2496 * This is called to unpin an inode. It can be directed to wait or to return
2497 * immediately without waiting for the inode to be unpinned. The caller must
2498 * have the inode locked in at least shared mode so that the buffer cannot be
2499 * subsequently pinned once someone is waiting for it to be unpinned.
2500 */
2501 STATIC void
2505 {
2512 /* Give the log a push to start the unpinning I/O */
2517 }
2522 {
2524 }
2529 {
2531 }
2534 /*
2535 * xfs_iextents_copy()
2536 *
2537 * This is called to copy the REAL extents (as opposed to the delayed
2538 * allocation extents) from the inode into the given buffer. It
2539 * returns the number of bytes copied into the buffer.
2540 *
2541 * If there are no delayed allocation extents, then we can just
2542 * memcpy() the extents into the buffer. Otherwise, we need to
2543 * examine each extent in turn and skip those which are delayed.
2544 */
2545 int
2550 {
2555 xfs_fsblock_t start_block;
2565 /*
2566 * There are some delayed allocation extents in the
2567 * inode, so copy the extents one at a time and skip
2568 * the delayed ones. There must be at least one
2569 * non-delayed extent.
2570 */
2576 /*
2577 * It's a delayed allocation extent, so skip it.
2578 */
2580 }
2582 /* Translate to on disk format */
2585 dp++;
2586 copied++;
2587 }
2592 }
2594 /*
2595 * Each of the following cases stores data into the same region
2596 * of the on-disk inode, so only one of them can be valid at
2597 * any given time. While it is possible to have conflicting formats
2598 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2599 * in EXTENTS format, this can only happen when the fork has
2600 * changed formats after being modified but before being flushed.
2601 * In these cases, the format always takes precedence, because the
2602 * format indicates the current state of the fork.
2603 */
2604 /*ARGSUSED*/
2605 STATIC void
2612 {
2616 #ifdef XFS_TRANS_DEBUG
2618 #endif
2629 /*
2630 * This can happen if we gave up in iformat in an error path,
2631 * for the attribute fork.
2632 */
2636 }
2646 }
2660 whichfork);
2661 }
2670 XFS_BROOT_SIZE_ADJ));
2674 }
2681 }
2690 }
2696 }
2697 }
2699 STATIC int
2703 {
2728 /* really need a gang lookup range call here */
2738 /* if the inode lies outside this cluster, we're done. */
2741 /*
2742 * Do an un-protected check to see if the inode is dirty and
2743 * is a candidate for flushing. These checks will be repeated
2744 * later after the appropriate locks are acquired.
2745 */
2749 /*
2750 * Try to get locks. If any are unavailable or it is pinned,
2751 * then this inode cannot be flushed and is skipped.
2752 */
2759 }
2764 }
2766 /*
2767 * arriving here means that this inode can be flushed. First
2768 * re-check that it's dirty before flushing.
2769 */
2776 }
2777 clcount++;
2780 }
2782 }
2787 }
2789 out_free:
2795 cluster_corrupt_out:
2796 /*
2797 * Corruption detected in the clustering loop. Invalidate the
2798 * inode buffer and shut down the filesystem.
2799 */
2801 /*
2802 * Clean up the buffer. If it was B_DELWRI, just release it --
2803 * brelse can handle it with no problems. If not, shut down the
2804 * filesystem before releasing the buffer.
2805 */
2813 /*
2814 * Just like incore_relse: if we have b_iodone functions,
2815 * mark the buffer as an error and call them. Otherwise
2816 * mark it as stale and brelse.
2817 */
2827 }
2828 }
2830 /*
2831 * Unlocks the flush lock
2832 */
2836 }
2838 /*
2839 * xfs_iflush() will write a modified inode's changes out to the
2840 * inode's on disk home. The caller must have the inode lock held
2841 * in at least shared mode and the inode flush completion must be
2842 * active as well. The inode lock will still be held upon return from
2843 * the call and the caller is free to unlock it.
2844 * The inode flush will be completed when the inode reaches the disk.
2845 * The flags indicate how the inode's buffer should be written out.
2846 */
2847 int
2850 uint flags)
2851 {
2870 /*
2871 * If the inode isn't dirty, then just release the inode
2872 * flush lock and do nothing.
2873 */
2877 }
2879 /*
2880 * We can't flush the inode until it is unpinned, so wait for it if we
2881 * are allowed to block. We know noone new can pin it, because we are
2882 * holding the inode lock shared and you need to hold it exclusively to
2883 * pin the inode.
2884 *
2885 * If we are not allowed to block, force the log out asynchronously so
2886 * that when we come back the inode will be unpinned. If other inodes
2887 * in the same cluster are dirty, they will probably write the inode
2888 * out for us if they occur after the log force completes.
2889 */
2894 }
2897 /*
2898 * This may have been unpinned because the filesystem is shutting
2899 * down forcibly. If that's the case we must not write this inode
2900 * to disk, because the log record didn't make it to disk!
2901 */
2908 }
2910 /*
2911 * Decide how buffer will be flushed out. This is done before
2912 * the call to xfs_iflush_int because this field is zeroed by it.
2913 */
2915 /*
2916 * Flush out the inode buffer according to the directions
2917 * of the caller. In the cases where the caller has given
2918 * us a choice choose the non-delwri case. This is because
2919 * the inode is in the AIL and we need to get it out soon.
2920 */
2938 }
2957 }
2958 }
2960 /*
2961 * Get the buffer containing the on-disk inode.
2962 */
2968 }
2970 /*
2971 * First flush out the inode that xfs_iflush was called with.
2972 */
2977 /*
2978 * If the buffer is pinned then push on the log now so we won't
2979 * get stuck waiting in the write for too long.
2980 */
2984 /*
2985 * inode clustering:
2986 * see if other inodes can be gathered into this write
2987 */
2998 }
3001 corrupt_out:
3004 cluster_corrupt_out:
3005 /*
3006 * Unlocks the flush lock
3007 */
3010 }
3013 STATIC int
3017 {
3021 #ifdef XFS_TRANS_DEBUG
3023 #endif
3034 /*
3035 * If the inode isn't dirty, then just release the inode
3036 * flush lock and do nothing.
3037 */
3041 }
3043 /* set *dip = inode's place in the buffer */
3046 /*
3047 * Clear i_update_core before copying out the data.
3048 * This is for coordination with our timestamp updates
3049 * that don't hold the inode lock. They will always
3050 * update the timestamps BEFORE setting i_update_core,
3051 * so if we clear i_update_core after they set it we
3052 * are guaranteed to see their updates to the timestamps.
3053 * I believe that this depends on strongly ordered memory
3054 * semantics, but we have that. We use the SYNCHRONIZE
3055 * macro to make sure that the compiler does not reorder
3056 * the i_update_core access below the data copy below.
3057 */
3061 /*
3062 * Make sure to get the latest atime from the Linux inode.
3063 */
3072 }
3079 }
3089 }
3100 }
3101 }
3104 XFS_RANDOM_IFLUSH_5)) {
3110 ip);
3112 }
3119 }
3120 /*
3121 * bump the flush iteration count, used to detect flushes which
3122 * postdate a log record during recovery.
3123 */
3127 /*
3128 * Copy the dirty parts of the inode into the on-disk
3129 * inode. We always copy out the core of the inode,
3130 * because if the inode is dirty at all the core must
3131 * be.
3132 */
3135 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3139 /*
3140 * If this is really an old format inode and the superblock version
3141 * has not been updated to support only new format inodes, then
3142 * convert back to the old inode format. If the superblock version
3143 * has been updated, then make the conversion permanent.
3144 */
3148 /*
3149 * Convert it back.
3150 */
3154 /*
3155 * The superblock version has already been bumped,
3156 * so just make the conversion to the new inode
3157 * format permanent.
3158 */
3167 }
3168 }
3175 /*
3176 * We've recorded everything logged in the inode, so we'd
3177 * like to clear the ilf_fields bits so we don't log and
3178 * flush things unnecessarily. However, we can't stop
3179 * logging all this information until the data we've copied
3180 * into the disk buffer is written to disk. If we did we might
3181 * overwrite the copy of the inode in the log with all the
3182 * data after re-logging only part of it, and in the face of
3183 * a crash we wouldn't have all the data we need to recover.
3184 *
3185 * What we do is move the bits to the ili_last_fields field.
3186 * When logging the inode, these bits are moved back to the
3187 * ilf_fields field. In the xfs_iflush_done() routine we
3188 * clear ili_last_fields, since we know that the information
3189 * those bits represent is permanently on disk. As long as
3190 * the flush completes before the inode is logged again, then
3191 * both ilf_fields and ili_last_fields will be cleared.
3192 *
3193 * We can play with the ilf_fields bits here, because the inode
3194 * lock must be held exclusively in order to set bits there
3195 * and the flush lock protects the ili_last_fields bits.
3196 * Set ili_logged so the flush done
3197 * routine can tell whether or not to look in the AIL.
3198 * Also, store the current LSN of the inode so that we can tell
3199 * whether the item has moved in the AIL from xfs_iflush_done().
3200 * In order to read the lsn we need the AIL lock, because
3201 * it is a 64 bit value that cannot be read atomically.
3202 */
3211 /*
3212 * Attach the function xfs_iflush_done to the inode's
3213 * buffer. This will remove the inode from the AIL
3214 * and unlock the inode's flush lock when the inode is
3215 * completely written to disk.
3216 */
3223 /*
3224 * We're flushing an inode which is not in the AIL and has
3225 * not been logged but has i_update_core set. For this
3226 * case we can use a B_DELWRI flush and immediately drop
3227 * the inode flush lock because we can avoid the whole
3228 * AIL state thing. It's OK to drop the flush lock now,
3229 * because we've already locked the buffer and to do anything
3230 * you really need both.
3231 */
3236 }
3238 }
3242 corrupt_out:
3244 }
3248 #ifdef XFS_ILOCK_TRACE
3249 void
3251 {
3260 }
3261 #endif
3263 /*
3264 * Return a pointer to the extent record at file index idx.
3265 */
3266 xfs_bmbt_rec_host_t *
3270 {
3285 }
3286 }
3288 /*
3289 * Insert new item(s) into the extent records for incore inode
3290 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3291 */
3292 void
3298 {
3305 }
3307 /*
3308 * This is called when the amount of space required for incore file
3309 * extents needs to be increased. The ext_diff parameter stores the
3310 * number of new extents being added and the idx parameter contains
3311 * the extent index where the new extents will be added. If the new
3312 * extents are being appended, then we just need to (re)allocate and
3313 * initialize the space. Otherwise, if the new extents are being
3314 * inserted into the middle of the existing entries, a bit more work
3315 * is required to make room for the new extents to be inserted. The
3316 * caller is responsible for filling in the new extent entries upon
3317 * return.
3318 */
3319 void
3324 {
3333 /*
3334 * If the new number of extents (nextents + ext_diff)
3335 * fits inside the inode, then continue to use the inline
3336 * extent buffer.
3337 */
3344 }
3348 }
3349 /*
3350 * Otherwise use a linear (direct) extent list.
3351 * If the extents are currently inside the inode,
3352 * xfs_iext_realloc_direct will switch us from
3353 * inline to direct extent allocation mode.
3354 */
3362 }
3363 }
3364 /* Indirection array */
3377 }
3378 /* Extents fit in target extent page */
3386 }
3389 }
3390 /* Insert a new extent page */
3394 }
3395 /*
3396 * If extent(s) are being appended to the last page in
3397 * the indirection array and the new extent(s) don't fit
3398 * in the page, then erp is NULL and erp_idx is set to
3399 * the next index needed in the indirection array.
3400 */
3409 erp_idx++;
3410 }
3411 }
3412 }
3413 }
3415 }
3417 /*
3418 * This is called when incore extents are being added to the indirection
3419 * array and the new extents do not fit in the target extent list. The
3420 * erp_idx parameter contains the irec index for the target extent list
3421 * in the indirection array, and the idx parameter contains the extent
3422 * index within the list. The number of extents being added is stored
3423 * in the count parameter.
3424 *
3425 * |-------| |-------|
3426 * | | | | idx - number of extents before idx
3427 * | idx | | count |
3428 * | | | | count - number of extents being inserted at idx
3429 * |-------| |-------|
3430 * | count | | nex2 | nex2 - number of extents after idx + count
3431 * |-------| |-------|
3432 */
3433 void
3439 {
3453 /*
3454 * Save second part of target extent list
3455 * (all extents past */
3463 }
3465 /*
3466 * Add the new extents to the end of the target
3467 * list, then allocate new irec record(s) and
3468 * extent buffer(s) as needed to store the rest
3469 * of the new extents.
3470 */
3477 }
3479 erp_idx++;
3485 }
3487 /* Add nex2 extents back to indirection array */
3489 xfs_extnum_t ext_avail;
3495 /*
3496 * If nex2 extents fit in the current page, append
3497 * nex2_ep after the new extents.
3498 */
3501 }
3502 /*
3503 * Otherwise, check if space is available in the
3504 * next page.
3505 */
3509 erp_idx++;
3510 erp++;
3511 /* Create a hole for nex2 extents */
3514 }
3515 /*
3516 * Final choice, create a new extent page for
3517 * nex2 extents.
3518 */
3520 erp_idx++;
3522 }
3527 }
3528 }
3530 /*
3531 * This is called when the amount of space required for incore file
3532 * extents needs to be decreased. The ext_diff parameter stores the
3533 * number of extents to be removed and the idx parameter contains
3534 * the extent index where the extents will be removed from.
3535 *
3536 * If the amount of space needed has decreased below the linear
3537 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3538 * extent array. Otherwise, use kmem_realloc() to adjust the
3539 * size to what is needed.
3540 */
3541 void
3546 {
3562 }
3564 }
3566 /*
3567 * This removes ext_diff extents from the inline buffer, beginning
3568 * at extent index idx.
3569 */
3570 void
3575 {
3594 }
3595 }
3597 /*
3598 * This removes ext_diff extents from a linear (direct) extent list,
3599 * beginning at extent index idx. If the extents are being removed
3600 * from the end of the list (ie. truncate) then we just need to re-
3601 * allocate the list to remove the extra space. Otherwise, if the
3602 * extents are being removed from the middle of the existing extent
3603 * entries, then we first need to move the extent records beginning
3604 * at idx + ext_diff up in the list to overwrite the records being
3605 * removed, then remove the extra space via kmem_realloc.
3606 */
3607 void
3612 {
3624 }
3625 /* Move extents up in the list (if needed) */
3631 }
3634 /*
3635 * Reallocate the direct extent list. If the extents
3636 * will fit inside the inode then xfs_iext_realloc_direct
3637 * will switch from direct to inline extent allocation
3638 * mode for us.
3639 */
3642 }
3644 /*
3645 * This is called when incore extents are being removed from the
3646 * indirection array and the extents being removed span multiple extent
3647 * buffers. The idx parameter contains the file extent index where we
3648 * want to begin removing extents, and the count parameter contains
3649 * how many extents need to be removed.
3650 *
3651 * |-------| |-------|
3652 * | nex1 | | | nex1 - number of extents before idx
3653 * |-------| | count |
3654 * | | | | count - number of extents being removed at idx
3655 * | count | |-------|
3656 * | | | nex2 | nex2 - number of extents after idx + count
3657 * |-------| |-------|
3658 */
3659 void
3664 {
3683 /*
3684 * Check for deletion of entire list;
3685 * xfs_iext_irec_remove() updates extent offsets.
3686 */
3693 XFS_IEXT_BUFSZ);
3699 }
3700 }
3701 /* Move extents up (if needed) */
3706 }
3707 /* Zero out rest of page */
3710 /* Update remaining counters */
3715 erp_idx++;
3716 erp++;
3717 }
3720 }
3722 /*
3723 * Create, destroy, or resize a linear (direct) block of extents.
3724 */
3725 void
3729 {
3738 /* Free extent records */
3741 }
3742 /* Resize direct extent list and zero any new bytes */
3744 /* Check if extents will fit inside the inode */
3750 }
3753 }
3757 rnew_size,
3759 }
3764 }
3765 }
3766 /*
3767 * Switch from the inline extent buffer to a direct
3768 * extent list. Be sure to include the inline extent
3769 * bytes in new_size.
3770 */
3775 }
3777 }
3780 }
3782 /*
3783 * Switch from linear (direct) extent records to inline buffer.
3784 */
3785 void
3789 {
3792 /*
3793 * The inline buffer was zeroed when we switched
3794 * from inline to direct extent allocation mode,
3795 * so we don't need to clear it here.
3796 */
3802 }
3804 /*
3805 * Switch from inline buffer to linear (direct) extent records.
3806 * new_size should already be rounded up to the next power of 2
3807 * by the caller (when appropriate), so use new_size as it is.
3808 * However, since new_size may be rounded up, we can't update
3809 * if_bytes here. It is the caller's responsibility to update
3810 * if_bytes upon return.
3811 */
3812 void
3816 {
3824 }
3826 }
3828 /*
3829 * Resize an extent indirection array to new_size bytes.
3830 */
3831 void
3835 {
3850 }
3851 }
3853 /*
3854 * Switch from indirection array to linear (direct) extent allocations.
3855 */
3856 void
3859 {
3879 }
3880 }
3882 /*
3883 * Free incore file extents.
3884 */
3885 void
3888 {
3896 }
3903 }
3907 }
3909 /*
3910 * Return a pointer to the extent record for file system block bno.
3911 */
3917 {
3932 }
3935 /* Find target extent list */
3943 }
3944 /* Binary search extent records */
3955 /* Convert back to file-based extent index */
3958 }
3961 }
3962 }
3963 /* Convert back to file-based extent index */
3966 }
3972 }
3973 }
3976 }
3978 /*
3979 * Return a pointer to the indirection array entry containing the
3980 * extent record for filesystem block bno. Store the index of the
3981 * target irec in *erp_idxp.
3982 */
3988 {
4012 }
4013 }
4016 }
4018 /*
4019 * Return a pointer to the indirection array entry containing the
4020 * extent record at file extent index *idxp. Store the index of the
4021 * target irec in *erp_idxp and store the page index of the target
4022 * extent record in *idxp.
4023 */
4024 xfs_ext_irec_t *
4030 {
4047 /* Binary search extent irec's */
4063 erp_idx++;
4069 }
4070 }
4074 }
4076 /*
4077 * Allocate and initialize an indirection array once the space needed
4078 * for incore extents increases above XFS_IEXT_BUFSZ.
4079 */
4080 void
4083 {
4099 }
4110 }
4112 /*
4113 * Allocate and initialize a new entry in the indirection array.
4114 */
4115 xfs_ext_irec_t *
4119 {
4127 /* Resize indirection array */
4130 /*
4131 * Move records down in the array so the
4132 * new page can use erp_idx.
4133 */
4137 }
4140 /* Initialize new extent record */
4149 }
4151 /*
4152 * Remove a record from the indirection array.
4153 */
4154 void
4158 {
4170 }
4171 /* Compact extent records */
4175 }
4176 /*
4177 * Manually free the last extent record from the indirection
4178 * array. A call to xfs_iext_realloc_indirect() with a size
4179 * of zero would result in a call to xfs_iext_destroy() which
4180 * would in turn call this function again, creating a nasty
4181 * infinite loop.
4182 */
4188 }
4190 }
4192 /*
4193 * This is called to clean up large amounts of unused memory allocated
4194 * by the indirection array. Before compacting anything though, verify
4195 * that the indirection array is still needed and switch back to the
4196 * linear extent list (or even the inline buffer) if possible. The
4197 * compaction policy is as follows:
4198 *
4199 * Full Compaction: Extents fit into a single page (or inline buffer)
4200 * Partial Compaction: Extents occupy less than 50% of allocated space
4201 * No Compaction: Extents occupy at least 50% of allocated space
4202 */
4203 void
4206 {
4223 }
4224 }
4226 /*
4227 * Combine extents from neighboring extent pages.
4228 */
4229 void
4232 {
4248 /*
4249 * Free page before removing extent record
4250 * so er_extoffs don't get modified in
4251 * xfs_iext_irec_remove.
4252 */
4258 erp_idx++;
4259 }
4260 }
4261 }
4263 /*
4264 * This is called to update the er_extoff field in the indirection
4265 * array when extents have been added or removed from one of the
4266 * extent lists. erp_idx contains the irec index to begin updating
4267 * at and ext_diff contains the number of extents that were added
4268 * or removed.
4269 */
4270 void
4275 {
4283 }
4284 }