| blob | d820ada49b18288e6e1d0f8d66528b62e66d6f99 |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include <linux/log2.h>
54 /*
55 * Used in xfs_itruncate(). This is the maximum number of extents
56 * freed from a file in a single transaction.
57 */
58 #define XFS_ITRUNC_MAX_EXTENTS 2
65 #ifdef DEBUG
66 /*
67 * Make sure that the extents in the given memory buffer
68 * are valid.
69 */
70 STATIC void
74 xfs_exntfmt_t fmt)
75 {
76 xfs_bmbt_irec_t irec;
77 xfs_bmbt_rec_host_t rec;
87 }
88 }
90 #define xfs_validate_extents(ifp, nrecs, fmt)
93 /*
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
96 */
97 #if defined(DEBUG)
98 void
102 {
115 bp);
117 }
118 }
119 }
120 #endif
122 /*
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
126 */
127 STATIC int
133 uint buf_flags,
134 uint iget_flags)
135 {
146 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
151 }
153 }
155 /*
156 * Validate the magic number and version of every inode in the buffer
157 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
158 */
159 #ifdef DEBUG
163 #endif
174 XFS_ERRTAG_ITOBP_INOTOBP,
175 XFS_RANDOM_ITOBP_INOTOBP))) {
179 }
182 #ifdef DEBUG
184 "Device %s - bad inode magic/vsn "
189 #endif
192 }
193 }
197 /*
198 * Mark the buffer as an inode buffer now that it looks good
199 */
204 }
206 /*
207 * This routine is called to map an inode number within a file
208 * system to the buffer containing the on-disk version of the
209 * inode. It returns a pointer to the buffer containing the
210 * on-disk inode in the bpp parameter, and in the dip parameter
211 * it returns a pointer to the on-disk inode within that buffer.
212 *
213 * If a non-zero error is returned, then the contents of bpp and
214 * dipp are undefined.
215 *
216 * Use xfs_imap() to determine the size and location of the
217 * buffer to read from disk.
218 */
219 int
223 xfs_ino_t ino,
227 uint imap_flags)
228 {
246 }
249 /*
250 * This routine is called to map an inode to the buffer containing
251 * the on-disk version of the inode. It returns a pointer to the
252 * buffer containing the on-disk inode in the bpp parameter, and in
253 * the dip parameter it returns a pointer to the on-disk inode within
254 * that buffer.
255 *
256 * If a non-zero error is returned, then the contents of bpp and
257 * dipp are undefined.
258 *
259 * The inode is expected to already been mapped to its buffer and read
260 * in once, thus we can use the mapping information stored in the inode
261 * rather than calling xfs_imap(). This allows us to avoid the overhead
262 * of looking at the inode btree for small block file systems
263 * (see xfs_imap()).
264 */
265 int
272 uint buf_flags)
273 {
288 }
293 }
295 /*
296 * Move inode type and inode format specific information from the
297 * on-disk inode to the in-core inode. For fifos, devs, and sockets
298 * this means set if_rdev to the proper value. For files, directories,
299 * and symlinks this means to bring in the in-line data or extent
300 * pointers. For a file in B-tree format, only the root is immediately
301 * brought in-core. The rest will be in-lined in if_extents when it
302 * is first referenced (see xfs_iread_extents()).
303 */
304 STATIC int
308 {
312 xfs_fsize_t di_size;
330 }
339 }
349 }
360 }
371 /*
372 * no local regular files yet
373 */
379 XFS_ERRLEVEL_LOW,
382 }
391 XFS_ERRLEVEL_LOW,
394 }
409 }
415 }
418 }
436 XFS_ERRLEVEL_LOW,
439 }
452 }
457 }
459 }
461 /*
462 * The file is in-lined in the on-disk inode.
463 * If it fits into if_inline_data, then copy
464 * it there, otherwise allocate a buffer for it
465 * and copy the data there. Either way, set
466 * if_data to point at the data.
467 * If we allocate a buffer for the data, make
468 * sure that its size is a multiple of 4 and
469 * record the real size in i_real_bytes.
470 */
471 STATIC int
477 {
481 /*
482 * If the size is unreasonable, then something
483 * is wrong and we just bail out rather than crash in
484 * kmem_alloc() or memcpy() below.
485 */
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 */
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 */
625 }
630 /*
631 * Copy and convert from the on-disk structure
632 * to the in-memory structure.
633 */
641 }
643 STATIC void
647 {
677 }
679 void
683 {
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 uint iget_flags)
782 {
787 /*
788 * Fill in the location information in the in-core inode.
789 */
794 /*
795 * Get pointers to the on-disk inode and the buffer containing it.
796 */
803 /*
804 * If we got something that isn't an inode it means someone
805 * (nfs or dmi) has a stale handle.
806 */
808 #ifdef DEBUG
815 }
817 /*
818 * If the on-disk inode is already linked to a directory
819 * entry, copy all of the inode into the in-core inode.
820 * xfs_iformat() handles copying in the inode format
821 * specific information.
822 * Otherwise, just get the truly permanent information.
823 */
828 #ifdef DEBUG
833 }
839 /*
840 * Make sure to pull in the mode here as well in
841 * case the inode is released without being used.
842 * This ensures that xfs_inactive() will see that
843 * the inode is already free and not try to mess
844 * with the uninitialized part of it.
845 */
847 /*
848 * Initialize the per-fork minima and maxima for a new
849 * inode here. xfs_iformat will do it for old inodes.
850 */
853 }
855 /*
856 * The inode format changed when we moved the link count and
857 * made it 32 bits long. If this is an old format inode,
858 * convert it in memory to look like a new one. If it gets
859 * flushed to disk we will convert back before flushing or
860 * logging it. We zero out the new projid field and the old link
861 * count field. We'll handle clearing the pad field (the remains
862 * of the old uuid field) when we actually convert the inode to
863 * the new format. We don't change the version number so that we
864 * can distinguish this from a real new format inode.
865 */
870 }
875 /*
876 * Mark the buffer containing the inode as something to keep
877 * around for a while. This helps to keep recently accessed
878 * meta-data in-core longer.
879 */
882 /*
883 * Use xfs_trans_brelse() to release the buffer containing the
884 * on-disk inode, because it was acquired with xfs_trans_read_buf()
885 * in xfs_itobp() above. If tp is NULL, this is just a normal
886 * brelse(). If we're within a transaction, then xfs_trans_brelse()
887 * will only release the buffer if it is not dirty within the
888 * transaction. It will be OK to release the buffer in this case,
889 * because inodes on disk are never destroyed and we will be
890 * locking the new in-core inode before putting it in the hash
891 * table where other processes can find it. Thus we don't have
892 * to worry about the inode being changed just because we released
893 * the buffer.
894 */
895 out_brelse:
898 }
900 /*
901 * Read in extents from a btree-format inode.
902 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
903 */
904 int
909 {
912 xfs_extnum_t nextents;
918 }
922 /*
923 * We know that the size is valid (it's checked in iformat_btree)
924 */
934 }
937 }
939 /*
940 * Allocate an inode on disk and return a copy of its in-core version.
941 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
942 * appropriately within the inode. The uid and gid for the inode are
943 * set according to the contents of the given cred structure.
944 *
945 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
946 * has a free inode available, call xfs_iget()
947 * to obtain the in-core version of the allocated inode. Finally,
948 * fill in the inode and log its initial contents. In this case,
949 * ialloc_context would be set to NULL and call_again set to false.
950 *
951 * If xfs_dialloc() does not have an available inode,
952 * it will replenish its supply by doing an allocation. Since we can
953 * only do one allocation within a transaction without deadlocks, we
954 * must commit the current transaction before returning the inode itself.
955 * In this case, therefore, we will set call_again to true and return.
956 * The caller should then commit the current transaction, start a new
957 * transaction, and call xfs_ialloc() again to actually get the inode.
958 *
959 * To ensure that some other process does not grab the inode that
960 * was allocated during the first call to xfs_ialloc(), this routine
961 * also returns the [locked] bp pointing to the head of the freelist
962 * as ialloc_context. The caller should hold this buffer across
963 * the commit and pass it back into this routine on the second call.
964 *
965 * If we are allocating quota inodes, we do not have a parent inode
966 * to attach to or associate with (i.e. pip == NULL) because they
967 * are not linked into the directory structure - they are attached
968 * directly to the superblock - and so have no parent.
969 */
970 int
974 mode_t mode,
975 xfs_nlink_t nlink,
976 xfs_dev_t rdev,
977 prid_t prid,
982 {
983 xfs_ino_t ino;
985 uint flags;
987 timespec_t tv;
990 /*
991 * Call the space management code to pick
992 * the on-disk inode to be allocated.
993 */
1001 }
1004 /*
1005 * Get the in-core inode with the lock held exclusively.
1006 * This is because we're setting fields here we need
1007 * to prevent others from looking at until we're done.
1008 */
1024 /*
1025 * If the superblock version is up to where we support new format
1026 * inodes and this is currently an old format inode, then change
1027 * the inode version number now. This way we only do the conversion
1028 * here rather than here and in the flush/logging code.
1029 */
1033 /*
1034 * We've already zeroed the old link count, the projid field,
1035 * and the pad field.
1036 */
1037 }
1039 /*
1040 * Project ids won't be stored on disk if we are using a version 1 inode.
1041 */
1049 }
1050 }
1052 /*
1053 * If the group ID of the new file does not match the effective group
1054 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1055 * (and only if the irix_sgid_inherit compatibility variable is set).
1056 */
1061 }
1074 /*
1075 * di_gen will have been taken care of in xfs_iread.
1076 */
1093 /*
1094 * we can't set up filestreams until after the VFS inode
1095 * is set up properly.
1096 */
1099 /* fall through */
1110 }
1117 }
1118 }
1120 xfs_inherit_noatime)
1123 xfs_inherit_nodump)
1126 xfs_inherit_sync)
1129 xfs_inherit_nosymlinks)
1134 xfs_inherit_nodefrag)
1139 }
1140 /* FALLTHROUGH */
1149 }
1150 /*
1151 * Attribute fork settings for new inode.
1152 */
1156 /*
1157 * Log the new values stuffed into the inode.
1158 */
1162 /* now that we have an i_mode we can setup inode ops and unlock */
1165 /* now we have set up the vfs inode we can associate the filestream */
1172 }
1176 }
1178 /*
1179 * Check to make sure that there are no blocks allocated to the
1180 * file beyond the size of the file. We don't check this for
1181 * files with fixed size extents or real time extents, but we
1182 * at least do it for regular files.
1183 */
1184 #ifdef DEBUG
1185 void
1189 xfs_fsize_t isize)
1190 {
1191 xfs_fileoff_t map_first;
1206 /*
1207 * The filesystem could be shutting down, so bmapi may return
1208 * an error.
1209 */
1213 map_first),
1215 NULL))
1219 }
1222 /*
1223 * Calculate the last possible buffered byte in a file. This must
1224 * include data that was buffered beyond the EOF by the write code.
1225 * This also needs to deal with overflowing the xfs_fsize_t type
1226 * which can happen for sizes near the limit.
1227 *
1228 * We also need to take into account any blocks beyond the EOF. It
1229 * may be the case that they were buffered by a write which failed.
1230 * In that case the pages will still be in memory, but the inode size
1231 * will never have been updated.
1232 */
1233 STATIC xfs_fsize_t
1236 {
1238 xfs_fsize_t last_byte;
1239 xfs_fileoff_t last_block;
1240 xfs_fileoff_t size_last_block;
1246 /*
1247 * Only check for blocks beyond the EOF if the extents have
1248 * been read in. This eliminates the need for the inode lock,
1249 * and it also saves us from looking when it really isn't
1250 * necessary.
1251 */
1255 XFS_DATA_FORK);
1259 }
1262 }
1269 }
1273 }
1275 }
1277 /*
1278 * Start the truncation of the file to new_size. The new size
1279 * must be smaller than the current size. This routine will
1280 * clear the buffer and page caches of file data in the removed
1281 * range, and xfs_itruncate_finish() will remove the underlying
1282 * disk blocks.
1283 *
1284 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1285 * must NOT have the inode lock held at all. This is because we're
1286 * calling into the buffer/page cache code and we can't hold the
1287 * inode lock when we do so.
1288 *
1289 * We need to wait for any direct I/Os in flight to complete before we
1290 * proceed with the truncate. This is needed to prevent the extents
1291 * being read or written by the direct I/Os from being removed while the
1292 * I/O is in flight as there is no other method of synchronising
1293 * direct I/O with the truncate operation. Also, because we hold
1294 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1295 * started until the truncate completes and drops the lock. Essentially,
1296 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1297 * ordering between direct I/Os and the truncate operation.
1298 *
1299 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1300 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1301 * in the case that the caller is locking things out of order and
1302 * may not be able to call xfs_itruncate_finish() with the inode lock
1303 * held without dropping the I/O lock. If the caller must drop the
1304 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1305 * must be called again with all the same restrictions as the initial
1306 * call.
1307 */
1308 int
1311 uint flags,
1312 xfs_fsize_t new_size)
1313 {
1314 xfs_fsize_t last_byte;
1315 xfs_off_t toss_start;
1326 /* wait for the completion of any pending DIOs */
1330 /*
1331 * Call toss_pages or flushinval_pages to get rid of pages
1332 * overlapping the region being removed. We have to use
1333 * the less efficient flushinval_pages in the case that the
1334 * caller may not be able to finish the truncate without
1335 * dropping the inode's I/O lock. Make sure
1336 * to catch any pages brought in by buffers overlapping
1337 * the EOF by searching out beyond the isize by our
1338 * block size. We round new_size up to a block boundary
1339 * so that we don't toss things on the same block as
1340 * new_size but before it.
1341 *
1342 * Before calling toss_page or flushinval_pages, make sure to
1343 * call remapf() over the same region if the file is mapped.
1344 * This frees up mapped file references to the pages in the
1345 * given range and for the flushinval_pages case it ensures
1346 * that we get the latest mapped changes flushed out.
1347 */
1351 /*
1352 * The place to start tossing is beyond our maximum
1353 * file size, so there is no way that the data extended
1354 * out there.
1355 */
1357 }
1367 }
1368 }
1370 #ifdef DEBUG
1373 }
1374 #endif
1376 }
1378 /*
1379 * Shrink the file to the given new_size. The new size must be smaller than
1380 * the current size. This will free up the underlying blocks in the removed
1381 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1382 *
1383 * The transaction passed to this routine must have made a permanent log
1384 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1385 * given transaction and start new ones, so make sure everything involved in
1386 * the transaction is tidy before calling here. Some transaction will be
1387 * returned to the caller to be committed. The incoming transaction must
1388 * already include the inode, and both inode locks must be held exclusively.
1389 * The inode must also be "held" within the transaction. On return the inode
1390 * will be "held" within the returned transaction. This routine does NOT
1391 * require any disk space to be reserved for it within the transaction.
1392 *
1393 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1394 * indicates the fork which is to be truncated. For the attribute fork we only
1395 * support truncation to size 0.
1396 *
1397 * We use the sync parameter to indicate whether or not the first transaction
1398 * we perform might have to be synchronous. For the attr fork, it needs to be
1399 * so if the unlink of the inode is not yet known to be permanent in the log.
1400 * This keeps us from freeing and reusing the blocks of the attribute fork
1401 * before the unlink of the inode becomes permanent.
1402 *
1403 * For the data fork, we normally have to run synchronously if we're being
1404 * called out of the inactive path or we're being called out of the create path
1405 * where we're truncating an existing file. Either way, the truncate needs to
1406 * be sync so blocks don't reappear in the file with altered data in case of a
1407 * crash. wsync filesystems can run the first case async because anything that
1408 * shrinks the inode has to run sync so by the time we're called here from
1409 * inactive, the inode size is permanently set to 0.
1410 *
1411 * Calls from the truncate path always need to be sync unless we're in a wsync
1412 * filesystem and the file has already been unlinked.
1413 *
1414 * The caller is responsible for correctly setting the sync parameter. It gets
1415 * too hard for us to guess here which path we're being called out of just
1416 * based on inode state.
1417 *
1418 * If we get an error, we must return with the inode locked and linked into the
1419 * current transaction. This keeps things simple for the higher level code,
1420 * because it always knows that the inode is locked and held in the transaction
1421 * that returns to it whether errors occur or not. We don't mark the inode
1422 * dirty on error so that transactions can be easily aborted if possible.
1423 */
1424 int
1428 xfs_fsize_t new_size,
1431 {
1432 xfs_fsblock_t first_block;
1433 xfs_fileoff_t first_unmap_block;
1434 xfs_fileoff_t last_block;
1440 xfs_bmap_free_t free_list;
1456 /*
1457 * We only support truncating the entire attribute fork.
1458 */
1461 }
1465 /*
1466 * The first thing we do is set the size to new_size permanently
1467 * on disk. This way we don't have to worry about anyone ever
1468 * being able to look at the data being freed even in the face
1469 * of a crash. What we're getting around here is the case where
1470 * we free a block, it is allocated to another file, it is written
1471 * to, and then we crash. If the new data gets written to the
1472 * file but the log buffers containing the free and reallocation
1473 * don't, then we'd end up with garbage in the blocks being freed.
1474 * As long as we make the new_size permanent before actually
1475 * freeing any blocks it doesn't matter if they get writtten to.
1476 *
1477 * The callers must signal into us whether or not the size
1478 * setting here must be synchronous. There are a few cases
1479 * where it doesn't have to be synchronous. Those cases
1480 * occur if the file is unlinked and we know the unlink is
1481 * permanent or if the blocks being truncated are guaranteed
1482 * to be beyond the inode eof (regardless of the link count)
1483 * and the eof value is permanent. Both of these cases occur
1484 * only on wsync-mounted filesystems. In those cases, we're
1485 * guaranteed that no user will ever see the data in the blocks
1486 * that are being truncated so the truncate can run async.
1487 * In the free beyond eof case, the file may wind up with
1488 * more blocks allocated to it than it needs if we crash
1489 * and that won't get fixed until the next time the file
1490 * is re-opened and closed but that's ok as that shouldn't
1491 * be too many blocks.
1492 *
1493 * However, we can't just make all wsync xactions run async
1494 * because there's one call out of the create path that needs
1495 * to run sync where it's truncating an existing file to size
1496 * 0 whose size is > 0.
1497 *
1498 * It's probably possible to come up with a test in this
1499 * routine that would correctly distinguish all the above
1500 * cases from the values of the function parameters and the
1501 * inode state but for sanity's sake, I've decided to let the
1502 * layers above just tell us. It's simpler to correctly figure
1503 * out in the layer above exactly under what conditions we
1504 * can run async and I think it's easier for others read and
1505 * follow the logic in case something has to be changed.
1506 * cscope is your friend -- rcc.
1507 *
1508 * The attribute fork is much simpler.
1509 *
1510 * For the attribute fork we allow the caller to tell us whether
1511 * the unlink of the inode that led to this call is yet permanent
1512 * in the on disk log. If it is not and we will be freeing extents
1513 * in this inode then we make the first transaction synchronous
1514 * to make sure that the unlink is permanent by the time we free
1515 * the blocks.
1516 */
1519 /*
1520 * If we are not changing the file size then do
1521 * not update the on-disk file size - we may be
1522 * called from xfs_inactive_free_eofblocks(). If we
1523 * update the on-disk file size and then the system
1524 * crashes before the contents of the file are
1525 * flushed to disk then the files may be full of
1526 * holes (ie NULL files bug).
1527 */
1532 }
1533 }
1538 }
1544 /*
1545 * Since it is possible for space to become allocated beyond
1546 * the end of the file (in a crash where the space is allocated
1547 * but the inode size is not yet updated), simply remove any
1548 * blocks which show up between the new EOF and the maximum
1549 * possible file size. If the first block to be removed is
1550 * beyond the maximum file size (ie it is the same as last_block),
1551 * then there is nothing to do.
1552 */
1560 }
1562 /*
1563 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1564 * will tell us whether it freed the entire range or
1565 * not. If this is a synchronous mount (wsync),
1566 * then we can tell bunmapi to keep all the
1567 * transactions asynchronous since the unlink
1568 * transaction that made this inode inactive has
1569 * already hit the disk. There's no danger of
1570 * the freed blocks being reused, there being a
1571 * crash, and the reused blocks suddenly reappearing
1572 * in this file with garbage in them once recovery
1573 * runs.
1574 */
1579 XFS_ITRUNC_MAX_EXTENTS,
1583 /*
1584 * If the bunmapi call encounters an error,
1585 * return to the caller where the transaction
1586 * can be properly aborted. We just need to
1587 * make sure we're not holding any resources
1588 * that we were not when we came in.
1589 */
1592 }
1594 /*
1595 * Duplicate the transaction that has the permanent
1596 * reservation and commit the old transaction.
1597 */
1604 /*
1605 * If the bmap finish call encounters an error, return
1606 * to the caller where the transaction can be properly
1607 * aborted. We just need to make sure we're not
1608 * holding any resources that we were not when we came
1609 * in.
1610 *
1611 * Aborting from this point might lose some blocks in
1612 * the file system, but oh well.
1613 */
1616 }
1619 /*
1620 * Mark the inode dirty so it will be logged and
1621 * moved forward in the log as part of every commit.
1622 */
1624 }
1634 /*
1635 * transaction commit worked ok so we can drop the extra ticket
1636 * reference that we gained in xfs_trans_dup()
1637 */
1641 XFS_TRANS_PERM_LOG_RES,
1642 XFS_ITRUNCATE_LOG_COUNT);
1645 }
1646 /*
1647 * Only update the size in the case of the data fork, but
1648 * always re-log the inode so that our permanent transaction
1649 * can keep on rolling it forward in the log.
1650 */
1653 /*
1654 * If we are not changing the file size then do
1655 * not update the on-disk file size - we may be
1656 * called from xfs_inactive_free_eofblocks(). If we
1657 * update the on-disk file size and then the system
1658 * crashes before the contents of the file are
1659 * flushed to disk then the files may be full of
1660 * holes (ie NULL files bug).
1661 */
1665 }
1666 }
1676 }
1678 /*
1679 * This is called when the inode's link count goes to 0.
1680 * We place the on-disk inode on a list in the AGI. It
1681 * will be pulled from this list when the inode is freed.
1682 */
1683 int
1687 {
1693 xfs_agino_t agino;
1704 /*
1705 * Get the agi buffer first. It ensures lock ordering
1706 * on the list.
1707 */
1713 /*
1714 * Get the index into the agi hash table for the
1715 * list this inode will go on.
1716 */
1724 /*
1725 * There is already another inode in the bucket we need
1726 * to add ourselves to. Add us at the front of the list.
1727 * Here we put the head pointer into our next pointer,
1728 * and then we fall through to point the head at us.
1729 */
1735 /* both on-disk, don't endian flip twice */
1743 }
1745 /*
1746 * Point the bucket head pointer at the inode being inserted.
1747 */
1755 }
1757 /*
1758 * Pull the on-disk inode from the AGI unlinked list.
1759 */
1760 STATIC int
1764 {
1765 xfs_ino_t next_ino;
1771 xfs_agnumber_t agno;
1772 xfs_agino_t agino;
1773 xfs_agino_t next_agino;
1783 /*
1784 * Get the agi buffer first. It ensures lock ordering
1785 * on the list.
1786 */
1793 /*
1794 * Get the index into the agi hash table for the
1795 * list this inode will go on.
1796 */
1804 /*
1805 * We're at the head of the list. Get the inode's
1806 * on-disk buffer to see if there is anyone after us
1807 * on the list. Only modify our next pointer if it
1808 * is not already NULLAGINO. This saves us the overhead
1809 * of dealing with the buffer when there is no need to
1810 * change it.
1811 */
1818 }
1831 }
1832 /*
1833 * Point the bucket head pointer at the next inode.
1834 */
1843 /*
1844 * We need to search the list for the inode being freed.
1845 */
1849 /*
1850 * If the last inode wasn't the one pointing to
1851 * us, then release its buffer since we're not
1852 * going to do anything with it.
1853 */
1856 }
1865 }
1869 }
1870 /*
1871 * Now last_ibp points to the buffer previous to us on
1872 * the unlinked list. Pull us from the list.
1873 */
1880 }
1894 }
1895 /*
1896 * Point the previous inode on the list to the next inode.
1897 */
1905 }
1907 }
1909 /*
1910 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1911 * inodes that are in memory - they all must be marked stale and attached to
1912 * the cluster buffer.
1913 */
1914 STATIC void
1918 xfs_ino_t inum)
1919 {
1925 xfs_daddr_t blkno;
1942 }
1948 /*
1949 * We obtain and lock the backing buffer first in the process
1950 * here, as we have to ensure that any dirty inode that we
1951 * can't get the flush lock on is attached to the buffer.
1952 * If we scan the in-memory inodes first, then buffer IO can
1953 * complete before we get a lock on it, and hence we may fail
1954 * to mark all the active inodes on the buffer stale.
1955 */
1958 XBF_LOCK);
1960 /*
1961 * Walk the inodes already attached to the buffer and mark them
1962 * stale. These will all have the flush locks held, so an
1963 * in-memory inode walk can't lock them. By marking them all
1964 * stale first, we will not attempt to lock them in the loop
1965 * below as the XFS_ISTALE flag will be set.
1966 */
1977 }
1979 }
1982 /*
1983 * For each inode in memory attempt to add it to the inode
1984 * buffer and set it up for being staled on buffer IO
1985 * completion. This is safe as we've locked out tail pushing
1986 * and flushing by locking the buffer.
1987 *
1988 * We have already marked every inode that was part of a
1989 * transaction stale above, which means there is no point in
1990 * even trying to lock them.
1991 */
1993 retry:
1998 /* Inode not in memory, nothing to do */
2002 }
2004 /*
2005 * because this is an RCU protected lookup, we could
2006 * find a recently freed or even reallocated inode
2007 * during the lookup. We need to check under the
2008 * i_flags_lock for a valid inode here. Skip it if it
2009 * is not valid, the wrong inode or stale.
2010 */
2017 }
2020 /*
2021 * Don't try to lock/unlock the current inode, but we
2022 * _cannot_ skip the other inodes that we did not find
2023 * in the list attached to the buffer and are not
2024 * already marked stale. If we can't lock it, back off
2025 * and retry.
2026 */
2032 }
2038 /*
2039 * we don't need to attach clean inodes or those only
2040 * with unlogged changes (which we throw away, anyway).
2041 */
2049 }
2062 }
2066 }
2069 }
2071 /*
2072 * This is called to return an inode to the inode free list.
2073 * The inode should already be truncated to 0 length and have
2074 * no pages associated with it. This routine also assumes that
2075 * the inode is already a part of the transaction.
2076 *
2077 * The on-disk copy of the inode will have been added to the list
2078 * of unlinked inodes in the AGI. We need to remove the inode from
2079 * that list atomically with respect to freeing it here.
2080 */
2081 int
2086 {
2089 xfs_ino_t first_ino;
2102 /*
2103 * Pull the on-disk inode from the AGI unlinked list.
2104 */
2108 }
2113 }
2122 /*
2123 * Bump the generation count so no one will be confused
2124 * by reincarnations of this inode.
2125 */
2134 /*
2135 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2136 * from picking up this inode when it is reclaimed (its incore state
2137 * initialzed but not flushed to disk yet). The in-core di_mode is
2138 * already cleared and a corresponding transaction logged.
2139 * The hack here just synchronizes the in-core to on-disk
2140 * di_mode value in advance before the actual inode sync to disk.
2141 * This is OK because the inode is already unlinked and would never
2142 * change its di_mode again for this inode generation.
2143 * This is a temporary hack that would require a proper fix
2144 * in the future.
2145 */
2150 }
2153 }
2155 /*
2156 * Reallocate the space for if_broot based on the number of records
2157 * being added or deleted as indicated in rec_diff. Move the records
2158 * and pointers in if_broot to fit the new size. When shrinking this
2159 * will eliminate holes between the records and pointers created by
2160 * the caller. When growing this will create holes to be filled in
2161 * by the caller.
2162 *
2163 * The caller must not request to add more records than would fit in
2164 * the on-disk inode root. If the if_broot is currently NULL, then
2165 * if we adding records one will be allocated. The caller must also
2166 * not request that the number of records go below zero, although
2167 * it can go to zero.
2168 *
2169 * ip -- the inode whose if_broot area is changing
2170 * ext_diff -- the change in the number of records, positive or negative,
2171 * requested for the if_broot array.
2172 */
2173 void
2178 {
2188 /*
2189 * Handle the degenerate case quietly.
2190 */
2193 }
2197 /*
2198 * If there wasn't any memory allocated before, just
2199 * allocate it now and get out.
2200 */
2206 }
2208 /*
2209 * If there is already an existing if_broot, then we need
2210 * to realloc() it and shift the pointers to their new
2211 * location. The records don't change location because
2212 * they are kept butted up against the btree block header.
2213 */
2229 }
2231 /*
2232 * rec_diff is less than 0. In this case, we are shrinking the
2233 * if_broot buffer. It must already exist. If we go to zero
2234 * records, just get rid of the root and clear the status bit.
2235 */
2242 else
2246 /*
2247 * First copy over the btree block header.
2248 */
2253 }
2255 /*
2256 * Only copy the records and pointers if there are any.
2257 */
2259 /*
2260 * First copy the records.
2261 */
2266 /*
2267 * Then copy the pointers.
2268 */
2274 }
2281 }
2284 /*
2285 * This is called when the amount of space needed for if_data
2286 * is increased or decreased. The change in size is indicated by
2287 * the number of bytes that need to be added or deleted in the
2288 * byte_diff parameter.
2289 *
2290 * If the amount of space needed has decreased below the size of the
2291 * inline buffer, then switch to using the inline buffer. Otherwise,
2292 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2293 * to what is needed.
2294 *
2295 * ip -- the inode whose if_data area is changing
2296 * byte_diff -- the change in the number of bytes, positive or negative,
2297 * requested for the if_data array.
2298 */
2299 void
2304 {
2311 }
2320 }
2324 /*
2325 * If the valid extents/data can fit in if_inline_ext/data,
2326 * copy them from the malloc'd vector and free it.
2327 */
2333 new_size);
2336 }
2339 /*
2340 * Stuck with malloc/realloc.
2341 * For inline data, the underlying buffer must be
2342 * a multiple of 4 bytes in size so that it can be
2343 * logged and stay on word boundaries. We enforce
2344 * that here.
2345 */
2352 /*
2353 * Only do the realloc if the underlying size
2354 * is really changing.
2355 */
2359 real_size,
2362 }
2369 }
2370 }
2374 }
2376 void
2380 {
2387 }
2389 /*
2390 * If the format is local, then we can't have an extents
2391 * array so just look for an inline data array. If we're
2392 * not local then we may or may not have an extents list,
2393 * so check and free it up if we do.
2394 */
2402 }
2409 }
2416 }
2417 }
2419 /*
2420 * This is called to unpin an inode. The caller must have the inode locked
2421 * in at least shared mode so that the buffer cannot be subsequently pinned
2422 * once someone is waiting for it to be unpinned.
2423 */
2424 static void
2427 {
2432 /* Give the log a push to start the unpinning I/O */
2435 }
2437 void
2440 {
2444 }
2445 }
2447 /*
2448 * xfs_iextents_copy()
2449 *
2450 * This is called to copy the REAL extents (as opposed to the delayed
2451 * allocation extents) from the inode into the given buffer. It
2452 * returns the number of bytes copied into the buffer.
2453 *
2454 * If there are no delayed allocation extents, then we can just
2455 * memcpy() the extents into the buffer. Otherwise, we need to
2456 * examine each extent in turn and skip those which are delayed.
2457 */
2458 int
2463 {
2468 xfs_fsblock_t start_block;
2478 /*
2479 * There are some delayed allocation extents in the
2480 * inode, so copy the extents one at a time and skip
2481 * the delayed ones. There must be at least one
2482 * non-delayed extent.
2483 */
2489 /*
2490 * It's a delayed allocation extent, so skip it.
2491 */
2493 }
2495 /* Translate to on disk format */
2498 dp++;
2499 copied++;
2500 }
2505 }
2507 /*
2508 * Each of the following cases stores data into the same region
2509 * of the on-disk inode, so only one of them can be valid at
2510 * any given time. While it is possible to have conflicting formats
2511 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2512 * in EXTENTS format, this can only happen when the fork has
2513 * changed formats after being modified but before being flushed.
2514 * In these cases, the format always takes precedence, because the
2515 * format indicates the current state of the fork.
2516 */
2517 /*ARGSUSED*/
2518 STATIC void
2525 {
2529 #ifdef XFS_TRANS_DEBUG
2531 #endif
2542 /*
2543 * This can happen if we gave up in iformat in an error path,
2544 * for the attribute fork.
2545 */
2549 }
2559 }
2573 whichfork);
2574 }
2583 XFS_BROOT_SIZE_ADJ));
2587 }
2594 }
2603 }
2609 }
2610 }
2612 STATIC int
2616 {
2640 /* really need a gang lookup range call here */
2651 /*
2652 * because this is an RCU protected lookup, we could find a
2653 * recently freed or even reallocated inode during the lookup.
2654 * We need to check under the i_flags_lock for a valid inode
2655 * here. Skip it if it is not valid or the wrong inode.
2656 */
2662 }
2665 /*
2666 * Do an un-protected check to see if the inode is dirty and
2667 * is a candidate for flushing. These checks will be repeated
2668 * later after the appropriate locks are acquired.
2669 */
2673 /*
2674 * Try to get locks. If any are unavailable or it is pinned,
2675 * then this inode cannot be flushed and is skipped.
2676 */
2683 }
2688 }
2690 /*
2691 * arriving here means that this inode can be flushed. First
2692 * re-check that it's dirty before flushing.
2693 */
2700 }
2701 clcount++;
2704 }
2706 }
2711 }
2713 out_free:
2716 out_put:
2721 cluster_corrupt_out:
2722 /*
2723 * Corruption detected in the clustering loop. Invalidate the
2724 * inode buffer and shut down the filesystem.
2725 */
2727 /*
2728 * Clean up the buffer. If it was B_DELWRI, just release it --
2729 * brelse can handle it with no problems. If not, shut down the
2730 * filesystem before releasing the buffer.
2731 */
2739 /*
2740 * Just like incore_relse: if we have b_iodone functions,
2741 * mark the buffer as an error and call them. Otherwise
2742 * mark it as stale and brelse.
2743 */
2752 }
2753 }
2755 /*
2756 * Unlocks the flush lock
2757 */
2762 }
2764 /*
2765 * xfs_iflush() will write a modified inode's changes out to the
2766 * inode's on disk home. The caller must have the inode lock held
2767 * in at least shared mode and the inode flush completion must be
2768 * active as well. The inode lock will still be held upon return from
2769 * the call and the caller is free to unlock it.
2770 * The inode flush will be completed when the inode reaches the disk.
2771 * The flags indicate how the inode's buffer should be written out.
2772 */
2773 int
2776 uint flags)
2777 {
2794 /*
2795 * We can't flush the inode until it is unpinned, so wait for it if we
2796 * are allowed to block. We know noone new can pin it, because we are
2797 * holding the inode lock shared and you need to hold it exclusively to
2798 * pin the inode.
2799 *
2800 * If we are not allowed to block, force the log out asynchronously so
2801 * that when we come back the inode will be unpinned. If other inodes
2802 * in the same cluster are dirty, they will probably write the inode
2803 * out for us if they occur after the log force completes.
2804 */
2809 }
2812 /*
2813 * For stale inodes we cannot rely on the backing buffer remaining
2814 * stale in cache for the remaining life of the stale inode and so
2815 * xfs_itobp() below may give us a buffer that no longer contains
2816 * inodes below. We have to check this after ensuring the inode is
2817 * unpinned so that it is safe to reclaim the stale inode after the
2818 * flush call.
2819 */
2823 }
2825 /*
2826 * This may have been unpinned because the filesystem is shutting
2827 * down forcibly. If that's the case we must not write this inode
2828 * to disk, because the log record didn't make it to disk!
2829 */
2836 }
2838 /*
2839 * Get the buffer containing the on-disk inode.
2840 */
2846 }
2848 /*
2849 * First flush out the inode that xfs_iflush was called with.
2850 */
2855 /*
2856 * If the buffer is pinned then push on the log now so we won't
2857 * get stuck waiting in the write for too long.
2858 */
2862 /*
2863 * inode clustering:
2864 * see if other inodes can be gathered into this write
2865 */
2872 else
2876 corrupt_out:
2879 cluster_corrupt_out:
2880 /*
2881 * Unlocks the flush lock
2882 */
2885 }
2888 STATIC int
2892 {
2896 #ifdef XFS_TRANS_DEBUG
2898 #endif
2908 /* set *dip = inode's place in the buffer */
2911 /*
2912 * Clear i_update_core before copying out the data.
2913 * This is for coordination with our timestamp updates
2914 * that don't hold the inode lock. They will always
2915 * update the timestamps BEFORE setting i_update_core,
2916 * so if we clear i_update_core after they set it we
2917 * are guaranteed to see their updates to the timestamps.
2918 * I believe that this depends on strongly ordered memory
2919 * semantics, but we have that. We use the SYNCHRONIZE
2920 * macro to make sure that the compiler does not reorder
2921 * the i_update_core access below the data copy below.
2922 */
2926 /*
2927 * Make sure to get the latest timestamps from the Linux inode.
2928 */
2937 }
2944 }
2954 }
2965 }
2966 }
2969 XFS_RANDOM_IFLUSH_5)) {
2971 "%s: detected corrupt incore inode %Lu, "
2977 }
2984 }
2985 /*
2986 * bump the flush iteration count, used to detect flushes which
2987 * postdate a log record during recovery.
2988 */
2992 /*
2993 * Copy the dirty parts of the inode into the on-disk
2994 * inode. We always copy out the core of the inode,
2995 * because if the inode is dirty at all the core must
2996 * be.
2997 */
3000 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3004 /*
3005 * If this is really an old format inode and the superblock version
3006 * has not been updated to support only new format inodes, then
3007 * convert back to the old inode format. If the superblock version
3008 * has been updated, then make the conversion permanent.
3009 */
3013 /*
3014 * Convert it back.
3015 */
3019 /*
3020 * The superblock version has already been bumped,
3021 * so just make the conversion to the new inode
3022 * format permanent.
3023 */
3032 }
3033 }
3040 /*
3041 * We've recorded everything logged in the inode, so we'd
3042 * like to clear the ilf_fields bits so we don't log and
3043 * flush things unnecessarily. However, we can't stop
3044 * logging all this information until the data we've copied
3045 * into the disk buffer is written to disk. If we did we might
3046 * overwrite the copy of the inode in the log with all the
3047 * data after re-logging only part of it, and in the face of
3048 * a crash we wouldn't have all the data we need to recover.
3049 *
3050 * What we do is move the bits to the ili_last_fields field.
3051 * When logging the inode, these bits are moved back to the
3052 * ilf_fields field. In the xfs_iflush_done() routine we
3053 * clear ili_last_fields, since we know that the information
3054 * those bits represent is permanently on disk. As long as
3055 * the flush completes before the inode is logged again, then
3056 * both ilf_fields and ili_last_fields will be cleared.
3057 *
3058 * We can play with the ilf_fields bits here, because the inode
3059 * lock must be held exclusively in order to set bits there
3060 * and the flush lock protects the ili_last_fields bits.
3061 * Set ili_logged so the flush done
3062 * routine can tell whether or not to look in the AIL.
3063 * Also, store the current LSN of the inode so that we can tell
3064 * whether the item has moved in the AIL from xfs_iflush_done().
3065 * In order to read the lsn we need the AIL lock, because
3066 * it is a 64 bit value that cannot be read atomically.
3067 */
3076 /*
3077 * Attach the function xfs_iflush_done to the inode's
3078 * buffer. This will remove the inode from the AIL
3079 * and unlock the inode's flush lock when the inode is
3080 * completely written to disk.
3081 */
3087 /*
3088 * We're flushing an inode which is not in the AIL and has
3089 * not been logged but has i_update_core set. For this
3090 * case we can use a B_DELWRI flush and immediately drop
3091 * the inode flush lock because we can avoid the whole
3092 * AIL state thing. It's OK to drop the flush lock now,
3093 * because we've already locked the buffer and to do anything
3094 * you really need both.
3095 */
3100 }
3102 }
3106 corrupt_out:
3108 }
3110 /*
3111 * Return a pointer to the extent record at file index idx.
3112 */
3113 xfs_bmbt_rec_host_t *
3117 {
3132 }
3133 }
3135 /*
3136 * Insert new item(s) into the extent records for incore inode
3137 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3138 */
3139 void
3146 {
3156 }
3158 /*
3159 * This is called when the amount of space required for incore file
3160 * extents needs to be increased. The ext_diff parameter stores the
3161 * number of new extents being added and the idx parameter contains
3162 * the extent index where the new extents will be added. If the new
3163 * extents are being appended, then we just need to (re)allocate and
3164 * initialize the space. Otherwise, if the new extents are being
3165 * inserted into the middle of the existing entries, a bit more work
3166 * is required to make room for the new extents to be inserted. The
3167 * caller is responsible for filling in the new extent entries upon
3168 * return.
3169 */
3170 void
3175 {
3184 /*
3185 * If the new number of extents (nextents + ext_diff)
3186 * fits inside the inode, then continue to use the inline
3187 * extent buffer.
3188 */
3195 }
3199 }
3200 /*
3201 * Otherwise use a linear (direct) extent list.
3202 * If the extents are currently inside the inode,
3203 * xfs_iext_realloc_direct will switch us from
3204 * inline to direct extent allocation mode.
3205 */
3213 }
3214 }
3215 /* Indirection array */
3228 }
3229 /* Extents fit in target extent page */
3237 }
3240 }
3241 /* Insert a new extent page */
3245 }
3246 /*
3247 * If extent(s) are being appended to the last page in
3248 * the indirection array and the new extent(s) don't fit
3249 * in the page, then erp is NULL and erp_idx is set to
3250 * the next index needed in the indirection array.
3251 */
3260 erp_idx++;
3261 }
3262 }
3263 }
3264 }
3266 }
3268 /*
3269 * This is called when incore extents are being added to the indirection
3270 * array and the new extents do not fit in the target extent list. The
3271 * erp_idx parameter contains the irec index for the target extent list
3272 * in the indirection array, and the idx parameter contains the extent
3273 * index within the list. The number of extents being added is stored
3274 * in the count parameter.
3275 *
3276 * |-------| |-------|
3277 * | | | | idx - number of extents before idx
3278 * | idx | | count |
3279 * | | | | count - number of extents being inserted at idx
3280 * |-------| |-------|
3281 * | count | | nex2 | nex2 - number of extents after idx + count
3282 * |-------| |-------|
3283 */
3284 void
3290 {
3304 /*
3305 * Save second part of target extent list
3306 * (all extents past */
3314 }
3316 /*
3317 * Add the new extents to the end of the target
3318 * list, then allocate new irec record(s) and
3319 * extent buffer(s) as needed to store the rest
3320 * of the new extents.
3321 */
3328 }
3330 erp_idx++;
3336 }
3338 /* Add nex2 extents back to indirection array */
3340 xfs_extnum_t ext_avail;
3346 /*
3347 * If nex2 extents fit in the current page, append
3348 * nex2_ep after the new extents.
3349 */
3352 }
3353 /*
3354 * Otherwise, check if space is available in the
3355 * next page.
3356 */
3360 erp_idx++;
3361 erp++;
3362 /* Create a hole for nex2 extents */
3365 }
3366 /*
3367 * Final choice, create a new extent page for
3368 * nex2 extents.
3369 */
3371 erp_idx++;
3373 }
3378 }
3379 }
3381 /*
3382 * This is called when the amount of space required for incore file
3383 * extents needs to be decreased. The ext_diff parameter stores the
3384 * number of extents to be removed and the idx parameter contains
3385 * the extent index where the extents will be removed from.
3386 *
3387 * If the amount of space needed has decreased below the linear
3388 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3389 * extent array. Otherwise, use kmem_realloc() to adjust the
3390 * size to what is needed.
3391 */
3392 void
3398 {
3417 }
3419 }
3421 /*
3422 * This removes ext_diff extents from the inline buffer, beginning
3423 * at extent index idx.
3424 */
3425 void
3430 {
3449 }
3450 }
3452 /*
3453 * This removes ext_diff extents from a linear (direct) extent list,
3454 * beginning at extent index idx. If the extents are being removed
3455 * from the end of the list (ie. truncate) then we just need to re-
3456 * allocate the list to remove the extra space. Otherwise, if the
3457 * extents are being removed from the middle of the existing extent
3458 * entries, then we first need to move the extent records beginning
3459 * at idx + ext_diff up in the list to overwrite the records being
3460 * removed, then remove the extra space via kmem_realloc.
3461 */
3462 void
3467 {
3479 }
3480 /* Move extents up in the list (if needed) */
3486 }
3489 /*
3490 * Reallocate the direct extent list. If the extents
3491 * will fit inside the inode then xfs_iext_realloc_direct
3492 * will switch from direct to inline extent allocation
3493 * mode for us.
3494 */
3497 }
3499 /*
3500 * This is called when incore extents are being removed from the
3501 * indirection array and the extents being removed span multiple extent
3502 * buffers. The idx parameter contains the file extent index where we
3503 * want to begin removing extents, and the count parameter contains
3504 * how many extents need to be removed.
3505 *
3506 * |-------| |-------|
3507 * | nex1 | | | nex1 - number of extents before idx
3508 * |-------| | count |
3509 * | | | | count - number of extents being removed at idx
3510 * | count | |-------|
3511 * | | | nex2 | nex2 - number of extents after idx + count
3512 * |-------| |-------|
3513 */
3514 void
3519 {
3536 /*
3537 * Check for deletion of entire list;
3538 * xfs_iext_irec_remove() updates extent offsets.
3539 */
3546 XFS_IEXT_BUFSZ);
3552 }
3553 }
3554 /* Move extents up (if needed) */
3559 }
3560 /* Zero out rest of page */
3563 /* Update remaining counters */
3568 erp_idx++;
3569 erp++;
3570 }
3573 }
3575 /*
3576 * Create, destroy, or resize a linear (direct) block of extents.
3577 */
3578 void
3582 {
3591 /* Free extent records */
3594 }
3595 /* Resize direct extent list and zero any new bytes */
3597 /* Check if extents will fit inside the inode */
3603 }
3606 }
3610 rnew_size,
3612 }
3617 }
3618 }
3619 /*
3620 * Switch from the inline extent buffer to a direct
3621 * extent list. Be sure to include the inline extent
3622 * bytes in new_size.
3623 */
3628 }
3630 }
3633 }
3635 /*
3636 * Switch from linear (direct) extent records to inline buffer.
3637 */
3638 void
3642 {
3645 /*
3646 * The inline buffer was zeroed when we switched
3647 * from inline to direct extent allocation mode,
3648 * so we don't need to clear it here.
3649 */
3655 }
3657 /*
3658 * Switch from inline buffer to linear (direct) extent records.
3659 * new_size should already be rounded up to the next power of 2
3660 * by the caller (when appropriate), so use new_size as it is.
3661 * However, since new_size may be rounded up, we can't update
3662 * if_bytes here. It is the caller's responsibility to update
3663 * if_bytes upon return.
3664 */
3665 void
3669 {
3677 }
3679 }
3681 /*
3682 * Resize an extent indirection array to new_size bytes.
3683 */
3684 STATIC void
3688 {
3703 }
3704 }
3706 /*
3707 * Switch from indirection array to linear (direct) extent allocations.
3708 */
3709 STATIC void
3712 {
3732 }
3733 }
3735 /*
3736 * Free incore file extents.
3737 */
3738 void
3741 {
3749 }
3756 }
3760 }
3762 /*
3763 * Return a pointer to the extent record for file system block bno.
3764 */
3770 {
3785 }
3788 /* Find target extent list */
3796 }
3797 /* Binary search extent records */
3808 /* Convert back to file-based extent index */
3811 }
3814 }
3815 }
3816 /* Convert back to file-based extent index */
3819 }
3825 }
3826 }
3829 }
3831 /*
3832 * Return a pointer to the indirection array entry containing the
3833 * extent record for filesystem block bno. Store the index of the
3834 * target irec in *erp_idxp.
3835 */
3841 {
3865 }
3866 }
3869 }
3871 /*
3872 * Return a pointer to the indirection array entry containing the
3873 * extent record at file extent index *idxp. Store the index of the
3874 * target irec in *erp_idxp and store the page index of the target
3875 * extent record in *idxp.
3876 */
3877 xfs_ext_irec_t *
3883 {
3900 /* Binary search extent irec's */
3916 erp_idx++;
3922 }
3923 }
3927 }
3929 /*
3930 * Allocate and initialize an indirection array once the space needed
3931 * for incore extents increases above XFS_IEXT_BUFSZ.
3932 */
3933 void
3936 {
3952 }
3963 }
3965 /*
3966 * Allocate and initialize a new entry in the indirection array.
3967 */
3968 xfs_ext_irec_t *
3972 {
3980 /* Resize indirection array */
3983 /*
3984 * Move records down in the array so the
3985 * new page can use erp_idx.
3986 */
3990 }
3993 /* Initialize new extent record */
4002 }
4004 /*
4005 * Remove a record from the indirection array.
4006 */
4007 void
4011 {
4023 }
4024 /* Compact extent records */
4028 }
4029 /*
4030 * Manually free the last extent record from the indirection
4031 * array. A call to xfs_iext_realloc_indirect() with a size
4032 * of zero would result in a call to xfs_iext_destroy() which
4033 * would in turn call this function again, creating a nasty
4034 * infinite loop.
4035 */
4041 }
4043 }
4045 /*
4046 * This is called to clean up large amounts of unused memory allocated
4047 * by the indirection array. Before compacting anything though, verify
4048 * that the indirection array is still needed and switch back to the
4049 * linear extent list (or even the inline buffer) if possible. The
4050 * compaction policy is as follows:
4051 *
4052 * Full Compaction: Extents fit into a single page (or inline buffer)
4053 * Partial Compaction: Extents occupy less than 50% of allocated space
4054 * No Compaction: Extents occupy at least 50% of allocated space
4055 */
4056 void
4059 {
4076 }
4077 }
4079 /*
4080 * Combine extents from neighboring extent pages.
4081 */
4082 void
4085 {
4101 /*
4102 * Free page before removing extent record
4103 * so er_extoffs don't get modified in
4104 * xfs_iext_irec_remove.
4105 */
4111 erp_idx++;
4112 }
4113 }
4114 }
4116 /*
4117 * This is called to update the er_extoff field in the indirection
4118 * array when extents have been added or removed from one of the
4119 * extent lists. erp_idx contains the irec index to begin updating
4120 * at and ext_diff contains the number of extents that were added
4121 * or removed.
4122 */
4123 void
4128 {
4136 }
4137 }