| blob | e7ae08d1df485105a10f89cd8e976a826e63af53 |
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 */
1262 XFS_DATA_FORK);
1265 }
1268 }
1275 }
1279 }
1281 }
1283 #if defined(XFS_RW_TRACE)
1284 STATIC void
1289 xfs_fsize_t new_size,
1290 xfs_off_t toss_start,
1291 xfs_off_t toss_finish)
1292 {
1295 }
1314 }
1315 #else
1316 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1317 #endif
1319 /*
1320 * Start the truncation of the file to new_size. The new size
1321 * must be smaller than the current size. This routine will
1322 * clear the buffer and page caches of file data in the removed
1323 * range, and xfs_itruncate_finish() will remove the underlying
1324 * disk blocks.
1325 *
1326 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1327 * must NOT have the inode lock held at all. This is because we're
1328 * calling into the buffer/page cache code and we can't hold the
1329 * inode lock when we do so.
1330 *
1331 * We need to wait for any direct I/Os in flight to complete before we
1332 * proceed with the truncate. This is needed to prevent the extents
1333 * being read or written by the direct I/Os from being removed while the
1334 * I/O is in flight as there is no other method of synchronising
1335 * direct I/O with the truncate operation. Also, because we hold
1336 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1337 * started until the truncate completes and drops the lock. Essentially,
1338 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1339 * ordering between direct I/Os and the truncate operation.
1340 *
1341 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1342 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1343 * in the case that the caller is locking things out of order and
1344 * may not be able to call xfs_itruncate_finish() with the inode lock
1345 * held without dropping the I/O lock. If the caller must drop the
1346 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1347 * must be called again with all the same restrictions as the initial
1348 * call.
1349 */
1350 int
1353 uint flags,
1354 xfs_fsize_t new_size)
1355 {
1356 xfs_fsize_t last_byte;
1357 xfs_off_t toss_start;
1368 /* wait for the completion of any pending DIOs */
1372 /*
1373 * Call toss_pages or flushinval_pages to get rid of pages
1374 * overlapping the region being removed. We have to use
1375 * the less efficient flushinval_pages in the case that the
1376 * caller may not be able to finish the truncate without
1377 * dropping the inode's I/O lock. Make sure
1378 * to catch any pages brought in by buffers overlapping
1379 * the EOF by searching out beyond the isize by our
1380 * block size. We round new_size up to a block boundary
1381 * so that we don't toss things on the same block as
1382 * new_size but before it.
1383 *
1384 * Before calling toss_page or flushinval_pages, make sure to
1385 * call remapf() over the same region if the file is mapped.
1386 * This frees up mapped file references to the pages in the
1387 * given range and for the flushinval_pages case it ensures
1388 * that we get the latest mapped changes flushed out.
1389 */
1393 /*
1394 * The place to start tossing is beyond our maximum
1395 * file size, so there is no way that the data extended
1396 * out there.
1397 */
1399 }
1402 last_byte);
1410 }
1411 }
1413 #ifdef DEBUG
1416 }
1417 #endif
1419 }
1421 /*
1422 * Shrink the file to the given new_size. The new size must be smaller than
1423 * the current size. This will free up the underlying blocks in the removed
1424 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1425 *
1426 * The transaction passed to this routine must have made a permanent log
1427 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1428 * given transaction and start new ones, so make sure everything involved in
1429 * the transaction is tidy before calling here. Some transaction will be
1430 * returned to the caller to be committed. The incoming transaction must
1431 * already include the inode, and both inode locks must be held exclusively.
1432 * The inode must also be "held" within the transaction. On return the inode
1433 * will be "held" within the returned transaction. This routine does NOT
1434 * require any disk space to be reserved for it within the transaction.
1435 *
1436 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1437 * indicates the fork which is to be truncated. For the attribute fork we only
1438 * support truncation to size 0.
1439 *
1440 * We use the sync parameter to indicate whether or not the first transaction
1441 * we perform might have to be synchronous. For the attr fork, it needs to be
1442 * so if the unlink of the inode is not yet known to be permanent in the log.
1443 * This keeps us from freeing and reusing the blocks of the attribute fork
1444 * before the unlink of the inode becomes permanent.
1445 *
1446 * For the data fork, we normally have to run synchronously if we're being
1447 * called out of the inactive path or we're being called out of the create path
1448 * where we're truncating an existing file. Either way, the truncate needs to
1449 * be sync so blocks don't reappear in the file with altered data in case of a
1450 * crash. wsync filesystems can run the first case async because anything that
1451 * shrinks the inode has to run sync so by the time we're called here from
1452 * inactive, the inode size is permanently set to 0.
1453 *
1454 * Calls from the truncate path always need to be sync unless we're in a wsync
1455 * filesystem and the file has already been unlinked.
1456 *
1457 * The caller is responsible for correctly setting the sync parameter. It gets
1458 * too hard for us to guess here which path we're being called out of just
1459 * based on inode state.
1460 *
1461 * If we get an error, we must return with the inode locked and linked into the
1462 * current transaction. This keeps things simple for the higher level code,
1463 * because it always knows that the inode is locked and held in the transaction
1464 * that returns to it whether errors occur or not. We don't mark the inode
1465 * dirty on error so that transactions can be easily aborted if possible.
1466 */
1467 int
1471 xfs_fsize_t new_size,
1474 {
1475 xfs_fsblock_t first_block;
1476 xfs_fileoff_t first_unmap_block;
1477 xfs_fileoff_t last_block;
1483 xfs_bmap_free_t free_list;
1499 /*
1500 * We only support truncating the entire attribute fork.
1501 */
1504 }
1507 /*
1508 * The first thing we do is set the size to new_size permanently
1509 * on disk. This way we don't have to worry about anyone ever
1510 * being able to look at the data being freed even in the face
1511 * of a crash. What we're getting around here is the case where
1512 * we free a block, it is allocated to another file, it is written
1513 * to, and then we crash. If the new data gets written to the
1514 * file but the log buffers containing the free and reallocation
1515 * don't, then we'd end up with garbage in the blocks being freed.
1516 * As long as we make the new_size permanent before actually
1517 * freeing any blocks it doesn't matter if they get writtten to.
1518 *
1519 * The callers must signal into us whether or not the size
1520 * setting here must be synchronous. There are a few cases
1521 * where it doesn't have to be synchronous. Those cases
1522 * occur if the file is unlinked and we know the unlink is
1523 * permanent or if the blocks being truncated are guaranteed
1524 * to be beyond the inode eof (regardless of the link count)
1525 * and the eof value is permanent. Both of these cases occur
1526 * only on wsync-mounted filesystems. In those cases, we're
1527 * guaranteed that no user will ever see the data in the blocks
1528 * that are being truncated so the truncate can run async.
1529 * In the free beyond eof case, the file may wind up with
1530 * more blocks allocated to it than it needs if we crash
1531 * and that won't get fixed until the next time the file
1532 * is re-opened and closed but that's ok as that shouldn't
1533 * be too many blocks.
1534 *
1535 * However, we can't just make all wsync xactions run async
1536 * because there's one call out of the create path that needs
1537 * to run sync where it's truncating an existing file to size
1538 * 0 whose size is > 0.
1539 *
1540 * It's probably possible to come up with a test in this
1541 * routine that would correctly distinguish all the above
1542 * cases from the values of the function parameters and the
1543 * inode state but for sanity's sake, I've decided to let the
1544 * layers above just tell us. It's simpler to correctly figure
1545 * out in the layer above exactly under what conditions we
1546 * can run async and I think it's easier for others read and
1547 * follow the logic in case something has to be changed.
1548 * cscope is your friend -- rcc.
1549 *
1550 * The attribute fork is much simpler.
1551 *
1552 * For the attribute fork we allow the caller to tell us whether
1553 * the unlink of the inode that led to this call is yet permanent
1554 * in the on disk log. If it is not and we will be freeing extents
1555 * in this inode then we make the first transaction synchronous
1556 * to make sure that the unlink is permanent by the time we free
1557 * the blocks.
1558 */
1561 /*
1562 * If we are not changing the file size then do
1563 * not update the on-disk file size - we may be
1564 * called from xfs_inactive_free_eofblocks(). If we
1565 * update the on-disk file size and then the system
1566 * crashes before the contents of the file are
1567 * flushed to disk then the files may be full of
1568 * holes (ie NULL files bug).
1569 */
1574 }
1575 }
1580 }
1586 /*
1587 * Since it is possible for space to become allocated beyond
1588 * the end of the file (in a crash where the space is allocated
1589 * but the inode size is not yet updated), simply remove any
1590 * blocks which show up between the new EOF and the maximum
1591 * possible file size. If the first block to be removed is
1592 * beyond the maximum file size (ie it is the same as last_block),
1593 * then there is nothing to do.
1594 */
1602 }
1604 /*
1605 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1606 * will tell us whether it freed the entire range or
1607 * not. If this is a synchronous mount (wsync),
1608 * then we can tell bunmapi to keep all the
1609 * transactions asynchronous since the unlink
1610 * transaction that made this inode inactive has
1611 * already hit the disk. There's no danger of
1612 * the freed blocks being reused, there being a
1613 * crash, and the reused blocks suddenly reappearing
1614 * in this file with garbage in them once recovery
1615 * runs.
1616 */
1622 XFS_ITRUNC_MAX_EXTENTS,
1626 /*
1627 * If the bunmapi call encounters an error,
1628 * return to the caller where the transaction
1629 * can be properly aborted. We just need to
1630 * make sure we're not holding any resources
1631 * that we were not when we came in.
1632 */
1635 }
1637 /*
1638 * Duplicate the transaction that has the permanent
1639 * reservation and commit the old transaction.
1640 */
1644 /* link the inode into the next xact in the chain */
1648 }
1651 /*
1652 * If the bmap finish call encounters an error, return
1653 * to the caller where the transaction can be properly
1654 * aborted. We just need to make sure we're not
1655 * holding any resources that we were not when we came
1656 * in.
1657 *
1658 * Aborting from this point might lose some blocks in
1659 * the file system, but oh well.
1660 */
1663 }
1666 /*
1667 * Mark the inode dirty so it will be logged and
1668 * moved forward in the log as part of every commit.
1669 */
1671 }
1677 /* link the inode into the next transaction in the chain */
1683 /*
1684 * transaction commit worked ok so we can drop the extra ticket
1685 * reference that we gained in xfs_trans_dup()
1686 */
1690 XFS_TRANS_PERM_LOG_RES,
1691 XFS_ITRUNCATE_LOG_COUNT);
1694 }
1695 /*
1696 * Only update the size in the case of the data fork, but
1697 * always re-log the inode so that our permanent transaction
1698 * can keep on rolling it forward in the log.
1699 */
1702 /*
1703 * If we are not changing the file size then do
1704 * not update the on-disk file size - we may be
1705 * called from xfs_inactive_free_eofblocks(). If we
1706 * update the on-disk file size and then the system
1707 * crashes before the contents of the file are
1708 * flushed to disk then the files may be full of
1709 * holes (ie NULL files bug).
1710 */
1714 }
1715 }
1725 }
1727 /*
1728 * This is called when the inode's link count goes to 0.
1729 * We place the on-disk inode on a list in the AGI. It
1730 * will be pulled from this list when the inode is freed.
1731 */
1732 int
1736 {
1742 xfs_agino_t agino;
1753 /*
1754 * Get the agi buffer first. It ensures lock ordering
1755 * on the list.
1756 */
1762 /*
1763 * Get the index into the agi hash table for the
1764 * list this inode will go on.
1765 */
1773 /*
1774 * There is already another inode in the bucket we need
1775 * to add ourselves to. Add us at the front of the list.
1776 * Here we put the head pointer into our next pointer,
1777 * and then we fall through to point the head at us.
1778 */
1784 /* both on-disk, don't endian flip twice */
1792 }
1794 /*
1795 * Point the bucket head pointer at the inode being inserted.
1796 */
1804 }
1806 /*
1807 * Pull the on-disk inode from the AGI unlinked list.
1808 */
1809 STATIC int
1813 {
1814 xfs_ino_t next_ino;
1820 xfs_agnumber_t agno;
1821 xfs_agino_t agino;
1822 xfs_agino_t next_agino;
1832 /*
1833 * Get the agi buffer first. It ensures lock ordering
1834 * on the list.
1835 */
1842 /*
1843 * Get the index into the agi hash table for the
1844 * list this inode will go on.
1845 */
1853 /*
1854 * We're at the head of the list. Get the inode's
1855 * on-disk buffer to see if there is anyone after us
1856 * on the list. Only modify our next pointer if it
1857 * is not already NULLAGINO. This saves us the overhead
1858 * of dealing with the buffer when there is no need to
1859 * change it.
1860 */
1867 }
1880 }
1881 /*
1882 * Point the bucket head pointer at the next inode.
1883 */
1892 /*
1893 * We need to search the list for the inode being freed.
1894 */
1898 /*
1899 * If the last inode wasn't the one pointing to
1900 * us, then release its buffer since we're not
1901 * going to do anything with it.
1902 */
1905 }
1914 }
1918 }
1919 /*
1920 * Now last_ibp points to the buffer previous to us on
1921 * the unlinked list. Pull us from the list.
1922 */
1929 }
1943 }
1944 /*
1945 * Point the previous inode on the list to the next inode.
1946 */
1954 }
1956 }
1958 STATIC void
1962 xfs_ino_t inum)
1963 {
1969 xfs_daddr_t blkno;
1985 }
1994 /*
1995 * Look for each inode in memory and attempt to lock it,
1996 * we can be racing with flush and tail pushing here.
1997 * any inode we get the locks on, add to an array of
1998 * inode items to process later.
1999 *
2000 * The get the buffer lock, we could beat a flush
2001 * or tail pushing thread to the lock here, in which
2002 * case they will go looking for the inode buffer
2003 * and fail, we need some other form of interlock
2004 * here.
2005 */
2012 /* Inode not in memory or we found it already,
2013 * nothing to do
2014 */
2018 }
2023 }
2025 /* If we can get the locks then add it to the
2026 * list, otherwise by the time we get the bp lock
2027 * below it will already be attached to the
2028 * inode buffer.
2029 */
2031 /* This inode will already be locked - by us, lets
2032 * keep it that way.
2033 */
2042 }
2043 }
2046 }
2057 }
2060 }
2061 }
2063 }
2067 XFS_BUF_LOCK);
2080 pre_flushed++;
2081 }
2083 }
2094 }
2107 }
2108 }
2113 }
2117 }
2119 /*
2120 * This is called to return an inode to the inode free list.
2121 * The inode should already be truncated to 0 length and have
2122 * no pages associated with it. This routine also assumes that
2123 * the inode is already a part of the transaction.
2124 *
2125 * The on-disk copy of the inode will have been added to the list
2126 * of unlinked inodes in the AGI. We need to remove the inode from
2127 * that list atomically with respect to freeing it here.
2128 */
2129 int
2134 {
2137 xfs_ino_t first_ino;
2150 /*
2151 * Pull the on-disk inode from the AGI unlinked list.
2152 */
2156 }
2161 }
2170 /*
2171 * Bump the generation count so no one will be confused
2172 * by reincarnations of this inode.
2173 */
2182 /*
2183 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2184 * from picking up this inode when it is reclaimed (its incore state
2185 * initialzed but not flushed to disk yet). The in-core di_mode is
2186 * already cleared and a corresponding transaction logged.
2187 * The hack here just synchronizes the in-core to on-disk
2188 * di_mode value in advance before the actual inode sync to disk.
2189 * This is OK because the inode is already unlinked and would never
2190 * change its di_mode again for this inode generation.
2191 * This is a temporary hack that would require a proper fix
2192 * in the future.
2193 */
2198 }
2201 }
2203 /*
2204 * Reallocate the space for if_broot based on the number of records
2205 * being added or deleted as indicated in rec_diff. Move the records
2206 * and pointers in if_broot to fit the new size. When shrinking this
2207 * will eliminate holes between the records and pointers created by
2208 * the caller. When growing this will create holes to be filled in
2209 * by the caller.
2210 *
2211 * The caller must not request to add more records than would fit in
2212 * the on-disk inode root. If the if_broot is currently NULL, then
2213 * if we adding records one will be allocated. The caller must also
2214 * not request that the number of records go below zero, although
2215 * it can go to zero.
2216 *
2217 * ip -- the inode whose if_broot area is changing
2218 * ext_diff -- the change in the number of records, positive or negative,
2219 * requested for the if_broot array.
2220 */
2221 void
2226 {
2236 /*
2237 * Handle the degenerate case quietly.
2238 */
2241 }
2245 /*
2246 * If there wasn't any memory allocated before, just
2247 * allocate it now and get out.
2248 */
2254 }
2256 /*
2257 * If there is already an existing if_broot, then we need
2258 * to realloc() it and shift the pointers to their new
2259 * location. The records don't change location because
2260 * they are kept butted up against the btree block header.
2261 */
2267 KM_SLEEP);
2277 }
2279 /*
2280 * rec_diff is less than 0. In this case, we are shrinking the
2281 * if_broot buffer. It must already exist. If we go to zero
2282 * records, just get rid of the root and clear the status bit.
2283 */
2290 else
2294 /*
2295 * First copy over the btree block header.
2296 */
2301 }
2303 /*
2304 * Only copy the records and pointers if there are any.
2305 */
2307 /*
2308 * First copy the records.
2309 */
2314 /*
2315 * Then copy the pointers.
2316 */
2322 }
2329 }
2332 /*
2333 * This is called when the amount of space needed for if_data
2334 * is increased or decreased. The change in size is indicated by
2335 * the number of bytes that need to be added or deleted in the
2336 * byte_diff parameter.
2337 *
2338 * If the amount of space needed has decreased below the size of the
2339 * inline buffer, then switch to using the inline buffer. Otherwise,
2340 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2341 * to what is needed.
2342 *
2343 * ip -- the inode whose if_data area is changing
2344 * byte_diff -- the change in the number of bytes, positive or negative,
2345 * requested for the if_data array.
2346 */
2347 void
2352 {
2359 }
2368 }
2372 /*
2373 * If the valid extents/data can fit in if_inline_ext/data,
2374 * copy them from the malloc'd vector and free it.
2375 */
2381 new_size);
2384 }
2387 /*
2388 * Stuck with malloc/realloc.
2389 * For inline data, the underlying buffer must be
2390 * a multiple of 4 bytes in size so that it can be
2391 * logged and stay on word boundaries. We enforce
2392 * that here.
2393 */
2399 /*
2400 * Only do the realloc if the underlying size
2401 * is really changing.
2402 */
2406 real_size,
2408 KM_SLEEP);
2409 }
2415 }
2416 }
2420 }
2422 void
2426 {
2433 }
2435 /*
2436 * If the format is local, then we can't have an extents
2437 * array so just look for an inline data array. If we're
2438 * not local then we may or may not have an extents list,
2439 * so check and free it up if we do.
2440 */
2448 }
2455 }
2462 }
2463 }
2465 /*
2466 * Increment the pin count of the given buffer.
2467 * This value is protected by ipinlock spinlock in the mount structure.
2468 */
2469 void
2472 {
2476 }
2478 /*
2479 * Decrement the pin count of the given inode, and wake up
2480 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2481 * inode must have been previously pinned with a call to xfs_ipin().
2482 */
2483 void
2486 {
2491 }
2493 /*
2494 * This is called to unpin an inode. It can be directed to wait or to return
2495 * immediately without waiting for the inode to be unpinned. The caller must
2496 * have the inode locked in at least shared mode so that the buffer cannot be
2497 * subsequently pinned once someone is waiting for it to be unpinned.
2498 */
2499 STATIC void
2503 {
2510 /* Give the log a push to start the unpinning I/O */
2515 }
2520 {
2522 }
2527 {
2529 }
2532 /*
2533 * xfs_iextents_copy()
2534 *
2535 * This is called to copy the REAL extents (as opposed to the delayed
2536 * allocation extents) from the inode into the given buffer. It
2537 * returns the number of bytes copied into the buffer.
2538 *
2539 * If there are no delayed allocation extents, then we can just
2540 * memcpy() the extents into the buffer. Otherwise, we need to
2541 * examine each extent in turn and skip those which are delayed.
2542 */
2543 int
2548 {
2553 xfs_fsblock_t start_block;
2563 /*
2564 * There are some delayed allocation extents in the
2565 * inode, so copy the extents one at a time and skip
2566 * the delayed ones. There must be at least one
2567 * non-delayed extent.
2568 */
2574 /*
2575 * It's a delayed allocation extent, so skip it.
2576 */
2578 }
2580 /* Translate to on disk format */
2583 dp++;
2584 copied++;
2585 }
2590 }
2592 /*
2593 * Each of the following cases stores data into the same region
2594 * of the on-disk inode, so only one of them can be valid at
2595 * any given time. While it is possible to have conflicting formats
2596 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2597 * in EXTENTS format, this can only happen when the fork has
2598 * changed formats after being modified but before being flushed.
2599 * In these cases, the format always takes precedence, because the
2600 * format indicates the current state of the fork.
2601 */
2602 /*ARGSUSED*/
2603 STATIC void
2610 {
2614 #ifdef XFS_TRANS_DEBUG
2616 #endif
2627 /*
2628 * This can happen if we gave up in iformat in an error path,
2629 * for the attribute fork.
2630 */
2634 }
2644 }
2658 whichfork);
2659 }
2668 XFS_BROOT_SIZE_ADJ));
2672 }
2679 }
2688 }
2694 }
2695 }
2697 STATIC int
2701 {
2726 /* really need a gang lookup range call here */
2736 /* if the inode lies outside this cluster, we're done. */
2739 /*
2740 * Do an un-protected check to see if the inode is dirty and
2741 * is a candidate for flushing. These checks will be repeated
2742 * later after the appropriate locks are acquired.
2743 */
2747 /*
2748 * Try to get locks. If any are unavailable or it is pinned,
2749 * then this inode cannot be flushed and is skipped.
2750 */
2757 }
2762 }
2764 /*
2765 * arriving here means that this inode can be flushed. First
2766 * re-check that it's dirty before flushing.
2767 */
2774 }
2775 clcount++;
2778 }
2780 }
2785 }
2787 out_free:
2793 cluster_corrupt_out:
2794 /*
2795 * Corruption detected in the clustering loop. Invalidate the
2796 * inode buffer and shut down the filesystem.
2797 */
2799 /*
2800 * Clean up the buffer. If it was B_DELWRI, just release it --
2801 * brelse can handle it with no problems. If not, shut down the
2802 * filesystem before releasing the buffer.
2803 */
2811 /*
2812 * Just like incore_relse: if we have b_iodone functions,
2813 * mark the buffer as an error and call them. Otherwise
2814 * mark it as stale and brelse.
2815 */
2825 }
2826 }
2828 /*
2829 * Unlocks the flush lock
2830 */
2834 }
2836 /*
2837 * xfs_iflush() will write a modified inode's changes out to the
2838 * inode's on disk home. The caller must have the inode lock held
2839 * in at least shared mode and the inode flush completion must be
2840 * active as well. The inode lock will still be held upon return from
2841 * the call and the caller is free to unlock it.
2842 * The inode flush will be completed when the inode reaches the disk.
2843 * The flags indicate how the inode's buffer should be written out.
2844 */
2845 int
2848 uint flags)
2849 {
2868 /*
2869 * If the inode isn't dirty, then just release the inode
2870 * flush lock and do nothing.
2871 */
2875 }
2877 /*
2878 * We can't flush the inode until it is unpinned, so wait for it if we
2879 * are allowed to block. We know noone new can pin it, because we are
2880 * holding the inode lock shared and you need to hold it exclusively to
2881 * pin the inode.
2882 *
2883 * If we are not allowed to block, force the log out asynchronously so
2884 * that when we come back the inode will be unpinned. If other inodes
2885 * in the same cluster are dirty, they will probably write the inode
2886 * out for us if they occur after the log force completes.
2887 */
2892 }
2895 /*
2896 * This may have been unpinned because the filesystem is shutting
2897 * down forcibly. If that's the case we must not write this inode
2898 * to disk, because the log record didn't make it to disk!
2899 */
2906 }
2908 /*
2909 * Decide how buffer will be flushed out. This is done before
2910 * the call to xfs_iflush_int because this field is zeroed by it.
2911 */
2913 /*
2914 * Flush out the inode buffer according to the directions
2915 * of the caller. In the cases where the caller has given
2916 * us a choice choose the non-delwri case. This is because
2917 * the inode is in the AIL and we need to get it out soon.
2918 */
2936 }
2955 }
2956 }
2958 /*
2959 * Get the buffer containing the on-disk inode.
2960 */
2966 }
2968 /*
2969 * First flush out the inode that xfs_iflush was called with.
2970 */
2975 /*
2976 * If the buffer is pinned then push on the log now so we won't
2977 * get stuck waiting in the write for too long.
2978 */
2982 /*
2983 * inode clustering:
2984 * see if other inodes can be gathered into this write
2985 */
2996 }
2999 corrupt_out:
3002 cluster_corrupt_out:
3003 /*
3004 * Unlocks the flush lock
3005 */
3008 }
3011 STATIC int
3015 {
3019 #ifdef XFS_TRANS_DEBUG
3021 #endif
3032 /*
3033 * If the inode isn't dirty, then just release the inode
3034 * flush lock and do nothing.
3035 */
3039 }
3041 /* set *dip = inode's place in the buffer */
3044 /*
3045 * Clear i_update_core before copying out the data.
3046 * This is for coordination with our timestamp updates
3047 * that don't hold the inode lock. They will always
3048 * update the timestamps BEFORE setting i_update_core,
3049 * so if we clear i_update_core after they set it we
3050 * are guaranteed to see their updates to the timestamps.
3051 * I believe that this depends on strongly ordered memory
3052 * semantics, but we have that. We use the SYNCHRONIZE
3053 * macro to make sure that the compiler does not reorder
3054 * the i_update_core access below the data copy below.
3055 */
3059 /*
3060 * Make sure to get the latest atime from the Linux inode.
3061 */
3070 }
3077 }
3087 }
3098 }
3099 }
3102 XFS_RANDOM_IFLUSH_5)) {
3108 ip);
3110 }
3117 }
3118 /*
3119 * bump the flush iteration count, used to detect flushes which
3120 * postdate a log record during recovery.
3121 */
3125 /*
3126 * Copy the dirty parts of the inode into the on-disk
3127 * inode. We always copy out the core of the inode,
3128 * because if the inode is dirty at all the core must
3129 * be.
3130 */
3133 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3137 /*
3138 * If this is really an old format inode and the superblock version
3139 * has not been updated to support only new format inodes, then
3140 * convert back to the old inode format. If the superblock version
3141 * has been updated, then make the conversion permanent.
3142 */
3146 /*
3147 * Convert it back.
3148 */
3152 /*
3153 * The superblock version has already been bumped,
3154 * so just make the conversion to the new inode
3155 * format permanent.
3156 */
3165 }
3166 }
3173 /*
3174 * We've recorded everything logged in the inode, so we'd
3175 * like to clear the ilf_fields bits so we don't log and
3176 * flush things unnecessarily. However, we can't stop
3177 * logging all this information until the data we've copied
3178 * into the disk buffer is written to disk. If we did we might
3179 * overwrite the copy of the inode in the log with all the
3180 * data after re-logging only part of it, and in the face of
3181 * a crash we wouldn't have all the data we need to recover.
3182 *
3183 * What we do is move the bits to the ili_last_fields field.
3184 * When logging the inode, these bits are moved back to the
3185 * ilf_fields field. In the xfs_iflush_done() routine we
3186 * clear ili_last_fields, since we know that the information
3187 * those bits represent is permanently on disk. As long as
3188 * the flush completes before the inode is logged again, then
3189 * both ilf_fields and ili_last_fields will be cleared.
3190 *
3191 * We can play with the ilf_fields bits here, because the inode
3192 * lock must be held exclusively in order to set bits there
3193 * and the flush lock protects the ili_last_fields bits.
3194 * Set ili_logged so the flush done
3195 * routine can tell whether or not to look in the AIL.
3196 * Also, store the current LSN of the inode so that we can tell
3197 * whether the item has moved in the AIL from xfs_iflush_done().
3198 * In order to read the lsn we need the AIL lock, because
3199 * it is a 64 bit value that cannot be read atomically.
3200 */
3209 /*
3210 * Attach the function xfs_iflush_done to the inode's
3211 * buffer. This will remove the inode from the AIL
3212 * and unlock the inode's flush lock when the inode is
3213 * completely written to disk.
3214 */
3221 /*
3222 * We're flushing an inode which is not in the AIL and has
3223 * not been logged but has i_update_core set. For this
3224 * case we can use a B_DELWRI flush and immediately drop
3225 * the inode flush lock because we can avoid the whole
3226 * AIL state thing. It's OK to drop the flush lock now,
3227 * because we've already locked the buffer and to do anything
3228 * you really need both.
3229 */
3234 }
3236 }
3240 corrupt_out:
3242 }
3246 #ifdef XFS_ILOCK_TRACE
3247 void
3249 {
3258 }
3259 #endif
3261 /*
3262 * Return a pointer to the extent record at file index idx.
3263 */
3264 xfs_bmbt_rec_host_t *
3268 {
3283 }
3284 }
3286 /*
3287 * Insert new item(s) into the extent records for incore inode
3288 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3289 */
3290 void
3296 {
3303 }
3305 /*
3306 * This is called when the amount of space required for incore file
3307 * extents needs to be increased. The ext_diff parameter stores the
3308 * number of new extents being added and the idx parameter contains
3309 * the extent index where the new extents will be added. If the new
3310 * extents are being appended, then we just need to (re)allocate and
3311 * initialize the space. Otherwise, if the new extents are being
3312 * inserted into the middle of the existing entries, a bit more work
3313 * is required to make room for the new extents to be inserted. The
3314 * caller is responsible for filling in the new extent entries upon
3315 * return.
3316 */
3317 void
3322 {
3331 /*
3332 * If the new number of extents (nextents + ext_diff)
3333 * fits inside the inode, then continue to use the inline
3334 * extent buffer.
3335 */
3342 }
3346 }
3347 /*
3348 * Otherwise use a linear (direct) extent list.
3349 * If the extents are currently inside the inode,
3350 * xfs_iext_realloc_direct will switch us from
3351 * inline to direct extent allocation mode.
3352 */
3360 }
3361 }
3362 /* Indirection array */
3375 }
3376 /* Extents fit in target extent page */
3384 }
3387 }
3388 /* Insert a new extent page */
3392 }
3393 /*
3394 * If extent(s) are being appended to the last page in
3395 * the indirection array and the new extent(s) don't fit
3396 * in the page, then erp is NULL and erp_idx is set to
3397 * the next index needed in the indirection array.
3398 */
3407 erp_idx++;
3408 }
3409 }
3410 }
3411 }
3413 }
3415 /*
3416 * This is called when incore extents are being added to the indirection
3417 * array and the new extents do not fit in the target extent list. The
3418 * erp_idx parameter contains the irec index for the target extent list
3419 * in the indirection array, and the idx parameter contains the extent
3420 * index within the list. The number of extents being added is stored
3421 * in the count parameter.
3422 *
3423 * |-------| |-------|
3424 * | | | | idx - number of extents before idx
3425 * | idx | | count |
3426 * | | | | count - number of extents being inserted at idx
3427 * |-------| |-------|
3428 * | count | | nex2 | nex2 - number of extents after idx + count
3429 * |-------| |-------|
3430 */
3431 void
3437 {
3451 /*
3452 * Save second part of target extent list
3453 * (all extents past */
3461 }
3463 /*
3464 * Add the new extents to the end of the target
3465 * list, then allocate new irec record(s) and
3466 * extent buffer(s) as needed to store the rest
3467 * of the new extents.
3468 */
3475 }
3477 erp_idx++;
3483 }
3485 /* Add nex2 extents back to indirection array */
3487 xfs_extnum_t ext_avail;
3493 /*
3494 * If nex2 extents fit in the current page, append
3495 * nex2_ep after the new extents.
3496 */
3499 }
3500 /*
3501 * Otherwise, check if space is available in the
3502 * next page.
3503 */
3507 erp_idx++;
3508 erp++;
3509 /* Create a hole for nex2 extents */
3512 }
3513 /*
3514 * Final choice, create a new extent page for
3515 * nex2 extents.
3516 */
3518 erp_idx++;
3520 }
3525 }
3526 }
3528 /*
3529 * This is called when the amount of space required for incore file
3530 * extents needs to be decreased. The ext_diff parameter stores the
3531 * number of extents to be removed and the idx parameter contains
3532 * the extent index where the extents will be removed from.
3533 *
3534 * If the amount of space needed has decreased below the linear
3535 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3536 * extent array. Otherwise, use kmem_realloc() to adjust the
3537 * size to what is needed.
3538 */
3539 void
3544 {
3560 }
3562 }
3564 /*
3565 * This removes ext_diff extents from the inline buffer, beginning
3566 * at extent index idx.
3567 */
3568 void
3573 {
3592 }
3593 }
3595 /*
3596 * This removes ext_diff extents from a linear (direct) extent list,
3597 * beginning at extent index idx. If the extents are being removed
3598 * from the end of the list (ie. truncate) then we just need to re-
3599 * allocate the list to remove the extra space. Otherwise, if the
3600 * extents are being removed from the middle of the existing extent
3601 * entries, then we first need to move the extent records beginning
3602 * at idx + ext_diff up in the list to overwrite the records being
3603 * removed, then remove the extra space via kmem_realloc.
3604 */
3605 void
3610 {
3622 }
3623 /* Move extents up in the list (if needed) */
3629 }
3632 /*
3633 * Reallocate the direct extent list. If the extents
3634 * will fit inside the inode then xfs_iext_realloc_direct
3635 * will switch from direct to inline extent allocation
3636 * mode for us.
3637 */
3640 }
3642 /*
3643 * This is called when incore extents are being removed from the
3644 * indirection array and the extents being removed span multiple extent
3645 * buffers. The idx parameter contains the file extent index where we
3646 * want to begin removing extents, and the count parameter contains
3647 * how many extents need to be removed.
3648 *
3649 * |-------| |-------|
3650 * | nex1 | | | nex1 - number of extents before idx
3651 * |-------| | count |
3652 * | | | | count - number of extents being removed at idx
3653 * | count | |-------|
3654 * | | | nex2 | nex2 - number of extents after idx + count
3655 * |-------| |-------|
3656 */
3657 void
3662 {
3681 /*
3682 * Check for deletion of entire list;
3683 * xfs_iext_irec_remove() updates extent offsets.
3684 */
3691 XFS_IEXT_BUFSZ);
3697 }
3698 }
3699 /* Move extents up (if needed) */
3704 }
3705 /* Zero out rest of page */
3708 /* Update remaining counters */
3713 erp_idx++;
3714 erp++;
3715 }
3718 }
3720 /*
3721 * Create, destroy, or resize a linear (direct) block of extents.
3722 */
3723 void
3727 {
3736 /* Free extent records */
3739 }
3740 /* Resize direct extent list and zero any new bytes */
3742 /* Check if extents will fit inside the inode */
3748 }
3751 }
3755 rnew_size,
3757 }
3762 }
3763 }
3764 /*
3765 * Switch from the inline extent buffer to a direct
3766 * extent list. Be sure to include the inline extent
3767 * bytes in new_size.
3768 */
3773 }
3775 }
3778 }
3780 /*
3781 * Switch from linear (direct) extent records to inline buffer.
3782 */
3783 void
3787 {
3790 /*
3791 * The inline buffer was zeroed when we switched
3792 * from inline to direct extent allocation mode,
3793 * so we don't need to clear it here.
3794 */
3800 }
3802 /*
3803 * Switch from inline buffer to linear (direct) extent records.
3804 * new_size should already be rounded up to the next power of 2
3805 * by the caller (when appropriate), so use new_size as it is.
3806 * However, since new_size may be rounded up, we can't update
3807 * if_bytes here. It is the caller's responsibility to update
3808 * if_bytes upon return.
3809 */
3810 void
3814 {
3822 }
3824 }
3826 /*
3827 * Resize an extent indirection array to new_size bytes.
3828 */
3829 void
3833 {
3848 }
3849 }
3851 /*
3852 * Switch from indirection array to linear (direct) extent allocations.
3853 */
3854 void
3857 {
3877 }
3878 }
3880 /*
3881 * Free incore file extents.
3882 */
3883 void
3886 {
3894 }
3901 }
3905 }
3907 /*
3908 * Return a pointer to the extent record for file system block bno.
3909 */
3915 {
3930 }
3933 /* Find target extent list */
3941 }
3942 /* Binary search extent records */
3953 /* Convert back to file-based extent index */
3956 }
3959 }
3960 }
3961 /* Convert back to file-based extent index */
3964 }
3970 }
3971 }
3974 }
3976 /*
3977 * Return a pointer to the indirection array entry containing the
3978 * extent record for filesystem block bno. Store the index of the
3979 * target irec in *erp_idxp.
3980 */
3986 {
4010 }
4011 }
4014 }
4016 /*
4017 * Return a pointer to the indirection array entry containing the
4018 * extent record at file extent index *idxp. Store the index of the
4019 * target irec in *erp_idxp and store the page index of the target
4020 * extent record in *idxp.
4021 */
4022 xfs_ext_irec_t *
4028 {
4045 /* Binary search extent irec's */
4061 erp_idx++;
4067 }
4068 }
4072 }
4074 /*
4075 * Allocate and initialize an indirection array once the space needed
4076 * for incore extents increases above XFS_IEXT_BUFSZ.
4077 */
4078 void
4081 {
4097 }
4108 }
4110 /*
4111 * Allocate and initialize a new entry in the indirection array.
4112 */
4113 xfs_ext_irec_t *
4117 {
4125 /* Resize indirection array */
4128 /*
4129 * Move records down in the array so the
4130 * new page can use erp_idx.
4131 */
4135 }
4138 /* Initialize new extent record */
4147 }
4149 /*
4150 * Remove a record from the indirection array.
4151 */
4152 void
4156 {
4168 }
4169 /* Compact extent records */
4173 }
4174 /*
4175 * Manually free the last extent record from the indirection
4176 * array. A call to xfs_iext_realloc_indirect() with a size
4177 * of zero would result in a call to xfs_iext_destroy() which
4178 * would in turn call this function again, creating a nasty
4179 * infinite loop.
4180 */
4186 }
4188 }
4190 /*
4191 * This is called to clean up large amounts of unused memory allocated
4192 * by the indirection array. Before compacting anything though, verify
4193 * that the indirection array is still needed and switch back to the
4194 * linear extent list (or even the inline buffer) if possible. The
4195 * compaction policy is as follows:
4196 *
4197 * Full Compaction: Extents fit into a single page (or inline buffer)
4198 * Partial Compaction: Extents occupy less than 50% of allocated space
4199 * No Compaction: Extents occupy at least 50% of allocated space
4200 */
4201 void
4204 {
4221 }
4222 }
4224 /*
4225 * Combine extents from neighboring extent pages.
4226 */
4227 void
4230 {
4246 /*
4247 * Free page before removing extent record
4248 * so er_extoffs don't get modified in
4249 * xfs_iext_irec_remove.
4250 */
4256 erp_idx++;
4257 }
4258 }
4259 }
4261 /*
4262 * This is called to update the er_extoff field in the indirection
4263 * array when extents have been added or removed from one of the
4264 * extent lists. erp_idx contains the irec index to begin updating
4265 * at and ext_diff contains the number of extents that were added
4266 * or removed.
4267 */
4268 void
4273 {
4281 }
4282 }