| blob | cced2fed9d0f9e9096fc102c8811f9795239d28e |
1 /*
2 * Copyright (C) International Business Machines Corp., 2000-2004
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the 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 to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 */
19 #include <linux/fs.h>
28 /*
29 * Debug code for double-checking block map
30 */
31 /* #define _JFS_DEBUG_DMAP 1 */
33 #ifdef _JFS_DEBUG_DMAP
34 #define DBINITMAP(size,ipbmap,results) \
35 DBinitmap(size,ipbmap,results)
36 #define DBALLOC(dbmap,mapsize,blkno,nblocks) \
37 DBAlloc(dbmap,mapsize,blkno,nblocks)
38 #define DBFREE(dbmap,mapsize,blkno,nblocks) \
39 DBFree(dbmap,mapsize,blkno,nblocks)
40 #define DBALLOCCK(dbmap,mapsize,blkno,nblocks) \
41 DBAllocCK(dbmap,mapsize,blkno,nblocks)
42 #define DBFREECK(dbmap,mapsize,blkno,nblocks) \
43 DBFreeCK(dbmap,mapsize,blkno,nblocks)
50 #else
51 #define DBINITMAP(size,ipbmap,results)
52 #define DBALLOC(dbmap, mapsize, blkno, nblocks)
53 #define DBFREE(dbmap, mapsize, blkno, nblocks)
54 #define DBALLOCCK(dbmap, mapsize, blkno, nblocks)
55 #define DBFREECK(dbmap, mapsize, blkno, nblocks)
58 /*
59 * SERIALIZATION of the Block Allocation Map.
60 *
61 * the working state of the block allocation map is accessed in
62 * two directions:
63 *
64 * 1) allocation and free requests that start at the dmap
65 * level and move up through the dmap control pages (i.e.
66 * the vast majority of requests).
67 *
68 * 2) allocation requests that start at dmap control page
69 * level and work down towards the dmaps.
70 *
71 * the serialization scheme used here is as follows.
72 *
73 * requests which start at the bottom are serialized against each
74 * other through buffers and each requests holds onto its buffers
75 * as it works it way up from a single dmap to the required level
76 * of dmap control page.
77 * requests that start at the top are serialized against each other
78 * and request that start from the bottom by the multiple read/single
79 * write inode lock of the bmap inode. requests starting at the top
80 * take this lock in write mode while request starting at the bottom
81 * take the lock in read mode. a single top-down request may proceed
82 * exclusively while multiple bottoms-up requests may proceed
83 * simultaneously (under the protection of busy buffers).
84 *
85 * in addition to information found in dmaps and dmap control pages,
86 * the working state of the block allocation map also includes read/
87 * write information maintained in the bmap descriptor (i.e. total
88 * free block count, allocation group level free block counts).
89 * a single exclusive lock (BMAP_LOCK) is used to guard this information
90 * in the face of multiple-bottoms up requests.
91 * (lock ordering: IREAD_LOCK, BMAP_LOCK);
92 *
93 * accesses to the persistent state of the block allocation map (limited
94 * to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
95 */
97 #define BMAP_LOCK_INIT(bmp) init_MUTEX(&bmp->db_bmaplock)
98 #define BMAP_LOCK(bmp) down(&bmp->db_bmaplock)
99 #define BMAP_UNLOCK(bmp) up(&bmp->db_bmaplock)
101 /*
102 * forward references
103 */
150 /*
151 * buddy table
152 *
153 * table used for determining buddy sizes within characters of
154 * dmap bitmap words. the characters themselves serve as indexes
155 * into the table, with the table elements yielding the maximum
156 * binary buddy of free bits within the character.
157 */
175 };
178 /*
179 * NAME: dbMount()
180 *
181 * FUNCTION: initializate the block allocation map.
182 *
183 * memory is allocated for the in-core bmap descriptor and
184 * the in-core descriptor is initialized from disk.
185 *
186 * PARAMETERS:
187 * ipbmap - pointer to in-core inode for the block map.
188 *
189 * RETURN VALUES:
190 * 0 - success
191 * -ENOMEM - insufficient memory
192 * -EIO - i/o error
193 */
195 {
201 /*
202 * allocate/initialize the in-memory bmap descriptor
203 */
204 /* allocate memory for the in-memory bmap descriptor */
209 /* read the on-disk bmap descriptor. */
216 }
218 /* copy the on-disk bmap descriptor to its in-memory version. */
237 /* release the buffer. */
240 /* bind the bmap inode and the bmap descriptor to each other. */
247 /*
248 * allocate/initialize the bmap lock
249 */
253 }
256 /*
257 * NAME: dbUnmount()
258 *
259 * FUNCTION: terminate the block allocation map in preparation for
260 * file system unmount.
261 *
262 * the in-core bmap descriptor is written to disk and
263 * the memory for this descriptor is freed.
264 *
265 * PARAMETERS:
266 * ipbmap - pointer to in-core inode for the block map.
267 *
268 * RETURN VALUES:
269 * 0 - success
270 * -EIO - i/o error
271 */
273 {
279 /*
280 * Invalidate the page cache buffers
281 */
284 /* free the memory for the in-memory bmap. */
288 }
290 /*
291 * dbSync()
292 */
294 {
300 /*
301 * write bmap global control page
302 */
303 /* get the buffer for the on-disk bmap descriptor. */
310 }
311 /* copy the in-memory version of the bmap to the on-disk version */
330 /* write the buffer */
333 /*
334 * write out dirty pages of bmap
335 */
343 }
346 /*
347 * NAME: dbFree()
348 *
349 * FUNCTION: free the specified block range from the working block
350 * allocation map.
351 *
352 * the blocks will be free from the working map one dmap
353 * at a time.
354 *
355 * PARAMETERS:
356 * ip - pointer to in-core inode;
357 * blkno - starting block number to be freed.
358 * nblocks - number of blocks to be freed.
359 *
360 * RETURN VALUES:
361 * 0 - success
362 * -EIO - i/o error
363 */
365 {
375 /* block to be freed better be within the mapsize. */
384 }
386 /*
387 * free the blocks a dmap at a time.
388 */
391 /* release previous dmap if any */
394 }
396 /* get the buffer for the current dmap. */
402 }
405 /* determine the number of blocks to be freed from
406 * this dmap.
407 */
412 /* free the blocks. */
417 }
420 }
422 /* write the last buffer. */
428 }
431 /*
432 * NAME: dbUpdatePMap()
433 *
434 * FUNCTION: update the allocation state (free or allocate) of the
435 * specified block range in the persistent block allocation map.
436 *
437 * the blocks will be updated in the persistent map one
438 * dmap at a time.
439 *
440 * PARAMETERS:
441 * ipbmap - pointer to in-core inode for the block map.
442 * free - TRUE if block range is to be freed from the persistent
443 * map; FALSE if it is to be allocated.
444 * blkno - starting block number of the range.
445 * nblocks - number of contiguous blocks in the range.
446 * tblk - transaction block;
447 *
448 * RETURN VALUES:
449 * 0 - success
450 * -EIO - i/o error
451 */
452 int
455 {
460 u32 mask;
467 /* the blocks better be within the mapsize. */
475 }
477 /* compute delta of transaction lsn from log syncpt */
482 /*
483 * update the block state a dmap at a time.
484 */
488 /* get the buffer for the current dmap. */
493 }
500 }
503 /* determine the bit number and word within the dmap of
504 * the starting block. also determine how many blocks
505 * are to be updated within this dmap.
506 */
511 /* update the bits of the dmap words. the first and last
512 * words may only have a subset of their bits updated. if
513 * this is the case, we'll work against that word (i.e.
514 * partial first and/or last) only in a single pass. a
515 * single pass will also be used to update all words that
516 * are to have all their bits updated.
517 */
520 /* determine the bit number within the word and
521 * the number of bits within the word.
522 */
526 /* check if only part of the word is to be updated. */
528 /* update (free or allocate) the bits
529 * in this word.
530 */
531 mask =
536 else
542 /* one or more words are to have all
543 * their bits updated. determine how
544 * many words and how many bits.
545 */
549 /* update (free or allocate) the bits
550 * in these words.
551 */
555 else
560 }
561 }
563 /*
564 * update dmap lsn
565 */
572 /* inherit older/smaller lsn */
578 /* move bp after tblock in logsync list */
580 }
582 /* inherit younger/larger clsn */
592 /* insert bp after tblock in logsync list */
600 }
601 }
603 /* write the last buffer. */
606 }
609 }
612 /*
613 * NAME: dbNextAG()
614 *
615 * FUNCTION: find the preferred allocation group for new allocations.
616 *
617 * Within the allocation groups, we maintain a preferred
618 * allocation group which consists of a group with at least
619 * average free space. It is the preferred group that we target
620 * new inode allocation towards. The tie-in between inode
621 * allocation and block allocation occurs as we allocate the
622 * first (data) block of an inode and specify the inode (block)
623 * as the allocation hint for this block.
624 *
625 * We try to avoid having more than one open file growing in
626 * an allocation group, as this will lead to fragmentation.
627 * This differs from the old OS/2 method of trying to keep
628 * empty ags around for large allocations.
629 *
630 * PARAMETERS:
631 * ipbmap - pointer to in-core inode for the block map.
632 *
633 * RETURN VALUES:
634 * the preferred allocation group number.
635 */
637 {
638 s64 avgfree;
647 /* determine the average number of free blocks within the ags. */
650 /*
651 * if the current preferred ag does not have an active allocator
652 * and has at least average freespace, return it
653 */
659 /* From the last preferred ag, find the next one with at least
660 * average free space.
661 */
667 /* open file is currently growing in this ag */
670 /* Return this one */
674 /* Less than avg. freespace, but best so far */
677 }
678 }
680 /*
681 * If no inactive ag was found with average freespace, use the
682 * next best
683 */
686 /* else leave db_agpref unchanged */
687 unlock:
690 /* return the preferred group.
691 */
693 }
695 /*
696 * NAME: dbAlloc()
697 *
698 * FUNCTION: attempt to allocate a specified number of contiguous free
699 * blocks from the working allocation block map.
700 *
701 * the block allocation policy uses hints and a multi-step
702 * approach.
703 *
704 * for allocation requests smaller than the number of blocks
705 * per dmap, we first try to allocate the new blocks
706 * immediately following the hint. if these blocks are not
707 * available, we try to allocate blocks near the hint. if
708 * no blocks near the hint are available, we next try to
709 * allocate within the same dmap as contains the hint.
710 *
711 * if no blocks are available in the dmap or the allocation
712 * request is larger than the dmap size, we try to allocate
713 * within the same allocation group as contains the hint. if
714 * this does not succeed, we finally try to allocate anywhere
715 * within the aggregate.
716 *
717 * we also try to allocate anywhere within the aggregate for
718 * for allocation requests larger than the allocation group
719 * size or requests that specify no hint value.
720 *
721 * PARAMETERS:
722 * ip - pointer to in-core inode;
723 * hint - allocation hint.
724 * nblocks - number of contiguous blocks in the range.
725 * results - on successful return, set to the starting block number
726 * of the newly allocated contiguous range.
727 *
728 * RETURN VALUES:
729 * 0 - success
730 * -ENOSPC - insufficient disk resources
731 * -EIO - i/o error
732 */
734 {
742 s64 mapSize;
745 /* assert that nblocks is valid */
748 #ifdef _STILL_TO_PORT
749 /* DASD limit check F226941 */
754 /* get the log2 number of blocks to be allocated.
755 * if the number of blocks is not a log2 multiple,
756 * it will be rounded up to the next log2 multiple.
757 */
762 //retry: /* serialize w.r.t.extendfs() */
765 /* the hint should be within the map */
769 }
771 /* if the number of blocks to be allocated is greater than the
772 * allocation group size, try to allocate anywhere.
773 */
780 nblocks);
781 }
784 }
786 /*
787 * If no hint, let dbNextAG recommend an allocation group
788 */
792 /* we would like to allocate close to the hint. adjust the
793 * hint to the block following the hint since the allocators
794 * will start looking for free space starting at this point.
795 */
803 /* check if blkno crosses over into a new allocation group.
804 * if so, check if we should allow allocations within this
805 * allocation group.
806 */
808 /* check if the AG is currenly being written to.
809 * if so, call dbNextAG() to find a non-busy
810 * AG with sufficient free space.
811 */
815 /* check if the allocation request size can be satisfied from a
816 * single dmap. if so, try to allocate from the dmap containing
817 * the hint using a tiered strategy.
818 */
822 /* get the buffer for the dmap containing the hint.
823 */
832 /* first, try to satisfy the allocation request with the
833 * blocks beginning at the hint.
834 */
842 }
846 }
851 /*
852 * Someone else is writing in this allocation
853 * group. To avoid fragmenting, try another ag
854 */
858 }
860 /* next, try to satisfy the allocation request with blocks
861 * near the hint.
862 */
870 }
874 }
876 /* try to satisfy the allocation request with blocks within
877 * the same dmap as the hint.
878 */
885 }
889 }
893 }
895 /* try to satisfy the allocation request with blocks within
896 * the same allocation group as the hint.
897 */
905 }
909 pref_ag:
910 /*
911 * Let dbNextAG recommend a preferred allocation group
912 */
916 /* Try to allocate within this allocation group. if that fails, try to
917 * allocate anywhere in the map.
918 */
923 }
925 write_unlock:
930 read_unlock:
934 }
936 #ifdef _NOTYET
937 /*
938 * NAME: dbAllocExact()
939 *
940 * FUNCTION: try to allocate the requested extent;
941 *
942 * PARAMETERS:
943 * ip - pointer to in-core inode;
944 * blkno - extent address;
945 * nblocks - extent length;
946 *
947 * RETURN VALUES:
948 * 0 - success
949 * -ENOSPC - insufficient disk resources
950 * -EIO - i/o error
951 */
953 {
958 s64 lblkno;
963 /*
964 * validate extent request:
965 *
966 * note: defragfs policy:
967 * max 64 blocks will be moved.
968 * allocation request size must be satisfied from a single dmap.
969 */
973 }
976 /* the free space is no longer available */
979 }
981 /* read in the dmap covering the extent */
987 }
990 /* try to allocate the requested extent */
998 }
1002 }
1005 /*
1006 * NAME: dbReAlloc()
1007 *
1008 * FUNCTION: attempt to extend a current allocation by a specified
1009 * number of blocks.
1010 *
1011 * this routine attempts to satisfy the allocation request
1012 * by first trying to extend the existing allocation in
1013 * place by allocating the additional blocks as the blocks
1014 * immediately following the current allocation. if these
1015 * blocks are not available, this routine will attempt to
1016 * allocate a new set of contiguous blocks large enough
1017 * to cover the existing allocation plus the additional
1018 * number of blocks required.
1019 *
1020 * PARAMETERS:
1021 * ip - pointer to in-core inode requiring allocation.
1022 * blkno - starting block of the current allocation.
1023 * nblocks - number of contiguous blocks within the current
1024 * allocation.
1025 * addnblocks - number of blocks to add to the allocation.
1026 * results - on successful return, set to the starting block number
1027 * of the existing allocation if the existing allocation
1028 * was extended in place or to a newly allocated contiguous
1029 * range if the existing allocation could not be extended
1030 * in place.
1031 *
1032 * RETURN VALUES:
1033 * 0 - success
1034 * -ENOSPC - insufficient disk resources
1035 * -EIO - i/o error
1036 */
1037 int
1040 {
1043 /* try to extend the allocation in place.
1044 */
1051 }
1053 /* could not extend the allocation in place, so allocate a
1054 * new set of blocks for the entire request (i.e. try to get
1055 * a range of contiguous blocks large enough to cover the
1056 * existing allocation plus the additional blocks.)
1057 */
1060 }
1063 /*
1064 * NAME: dbExtend()
1065 *
1066 * FUNCTION: attempt to extend a current allocation by a specified
1067 * number of blocks.
1068 *
1069 * this routine attempts to satisfy the allocation request
1070 * by first trying to extend the existing allocation in
1071 * place by allocating the additional blocks as the blocks
1072 * immediately following the current allocation.
1073 *
1074 * PARAMETERS:
1075 * ip - pointer to in-core inode requiring allocation.
1076 * blkno - starting block of the current allocation.
1077 * nblocks - number of contiguous blocks within the current
1078 * allocation.
1079 * addnblocks - number of blocks to add to the allocation.
1080 *
1081 * RETURN VALUES:
1082 * 0 - success
1083 * -ENOSPC - insufficient disk resources
1084 * -EIO - i/o error
1085 */
1087 {
1090 uint rel_block;
1097 /*
1098 * We don't want a non-aligned extent to cross a page boundary
1099 */
1104 /* get the last block of the current allocation */
1107 /* determine the block number of the block following
1108 * the existing allocation.
1109 */
1114 /* better be within the file system */
1121 }
1123 /* we'll attempt to extend the current allocation in place by
1124 * allocating the additional blocks as the blocks immediately
1125 * following the current allocation. we only try to extend the
1126 * current allocation in place if the number of additional blocks
1127 * can fit into a dmap, the last block of the current allocation
1128 * is not the last block of the file system, and the start of the
1129 * inplace extension is not on an allocation group boundary.
1130 */
1135 }
1137 /* get the buffer for the dmap containing the first block
1138 * of the extension.
1139 */
1145 }
1150 /* try to allocate the blocks immediately following the
1151 * current allocation.
1152 */
1157 /* were we successful ? */
1160 addnblocks);
1163 /* we were not successful */
1168 }
1171 /*
1172 * NAME: dbAllocNext()
1173 *
1174 * FUNCTION: attempt to allocate the blocks of the specified block
1175 * range within a dmap.
1176 *
1177 * PARAMETERS:
1178 * bmp - pointer to bmap descriptor
1179 * dp - pointer to dmap.
1180 * blkno - starting block number of the range.
1181 * nblocks - number of contiguous free blocks of the range.
1182 *
1183 * RETURN VALUES:
1184 * 0 - success
1185 * -ENOSPC - insufficient disk resources
1186 * -EIO - i/o error
1187 *
1188 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1189 */
1192 {
1196 u32 mask;
1202 }
1204 /* pick up a pointer to the leaves of the dmap tree.
1205 */
1208 /* determine the bit number and word within the dmap of the
1209 * starting block.
1210 */
1214 /* check if the specified block range is contained within
1215 * this dmap.
1216 */
1220 /* check if the starting leaf indicates that anything
1221 * is free.
1222 */
1226 /* check the dmaps words corresponding to block range to see
1227 * if the block range is free. not all bits of the first and
1228 * last words may be contained within the block range. if this
1229 * is the case, we'll work against those words (i.e. partial first
1230 * and/or last) on an individual basis (a single pass) and examine
1231 * the actual bits to determine if they are free. a single pass
1232 * will be used for all dmap words fully contained within the
1233 * specified range. within this pass, the leaves of the dmap
1234 * tree will be examined to determine if the blocks are free. a
1235 * single leaf may describe the free space of multiple dmap
1236 * words, so we may visit only a subset of the actual leaves
1237 * corresponding to the dmap words of the block range.
1238 */
1240 /* determine the bit number within the word and
1241 * the number of bits within the word.
1242 */
1246 /* check if only part of the word is to be examined.
1247 */
1249 /* check if the bits are free.
1250 */
1257 /* one or more dmap words are fully contained
1258 * within the block range. determine how many
1259 * words and how many bits.
1260 */
1264 /* now examine the appropriate leaves to determine
1265 * if the blocks are free.
1266 */
1268 /* does the leaf describe any free space ?
1269 */
1273 /* determine the l2 number of bits provided
1274 * by this leaf.
1275 */
1276 l2size =
1279 /* determine how many words were handled.
1280 */
1285 }
1286 }
1287 }
1289 /* allocate the blocks.
1290 */
1292 }
1295 /*
1296 * NAME: dbAllocNear()
1297 *
1298 * FUNCTION: attempt to allocate a number of contiguous free blocks near
1299 * a specified block (hint) within a dmap.
1300 *
1301 * starting with the dmap leaf that covers the hint, we'll
1302 * check the next four contiguous leaves for sufficient free
1303 * space. if sufficient free space is found, we'll allocate
1304 * the desired free space.
1305 *
1306 * PARAMETERS:
1307 * bmp - pointer to bmap descriptor
1308 * dp - pointer to dmap.
1309 * blkno - block number to allocate near.
1310 * nblocks - actual number of contiguous free blocks desired.
1311 * l2nb - log2 number of contiguous free blocks desired.
1312 * results - on successful return, set to the starting block number
1313 * of the newly allocated range.
1314 *
1315 * RETURN VALUES:
1316 * 0 - success
1317 * -ENOSPC - insufficient disk resources
1318 * -EIO - i/o error
1319 *
1320 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1321 */
1322 static int
1325 {
1333 }
1337 /* determine the word within the dmap that holds the hint
1338 * (i.e. blkno). also, determine the last word in the dmap
1339 * that we'll include in our examination.
1340 */
1344 /* examine the leaves for sufficient free space.
1345 */
1347 /* does the leaf describe sufficient free space ?
1348 */
1352 /* determine the block number within the file system
1353 * of the first block described by this dmap word.
1354 */
1357 /* if not all bits of the dmap word are free, get the
1358 * starting bit number within the dmap word of the required
1359 * string of free bits and adjust the block number with the
1360 * value.
1361 */
1363 blkno +=
1366 /* allocate the blocks.
1367 */
1372 }
1375 }
1378 /*
1379 * NAME: dbAllocAG()
1380 *
1381 * FUNCTION: attempt to allocate the specified number of contiguous
1382 * free blocks within the specified allocation group.
1383 *
1384 * unless the allocation group size is equal to the number
1385 * of blocks per dmap, the dmap control pages will be used to
1386 * find the required free space, if available. we start the
1387 * search at the highest dmap control page level which
1388 * distinctly describes the allocation group's free space
1389 * (i.e. the highest level at which the allocation group's
1390 * free space is not mixed in with that of any other group).
1391 * in addition, we start the search within this level at a
1392 * height of the dmapctl dmtree at which the nodes distinctly
1393 * describe the allocation group's free space. at this height,
1394 * the allocation group's free space may be represented by 1
1395 * or two sub-trees, depending on the allocation group size.
1396 * we search the top nodes of these subtrees left to right for
1397 * sufficient free space. if sufficient free space is found,
1398 * the subtree is searched to find the leftmost leaf that
1399 * has free space. once we have made it to the leaf, we
1400 * move the search to the next lower level dmap control page
1401 * corresponding to this leaf. we continue down the dmap control
1402 * pages until we find the dmap that contains or starts the
1403 * sufficient free space and we allocate at this dmap.
1404 *
1405 * if the allocation group size is equal to the dmap size,
1406 * we'll start at the dmap corresponding to the allocation
1407 * group and attempt the allocation at this level.
1408 *
1409 * the dmap control page search is also not performed if the
1410 * allocation group is completely free and we go to the first
1411 * dmap of the allocation group to do the allocation. this is
1412 * done because the allocation group may be part (not the first
1413 * part) of a larger binary buddy system, causing the dmap
1414 * control pages to indicate no free space (NOFREE) within
1415 * the allocation group.
1416 *
1417 * PARAMETERS:
1418 * bmp - pointer to bmap descriptor
1419 * agno - allocation group number.
1420 * nblocks - actual number of contiguous free blocks desired.
1421 * l2nb - log2 number of contiguous free blocks desired.
1422 * results - on successful return, set to the starting block number
1423 * of the newly allocated range.
1424 *
1425 * RETURN VALUES:
1426 * 0 - success
1427 * -ENOSPC - insufficient disk resources
1428 * -EIO - i/o error
1429 *
1430 * note: IWRITE_LOCK(ipmap) held on entry/exit;
1431 */
1432 static int
1434 {
1441 /* allocation request should not be for more than the
1442 * allocation group size.
1443 */
1446 "dbAllocAG: allocation request is larger than the "
1449 }
1451 /* determine the starting block number of the allocation
1452 * group.
1453 */
1456 /* check if the allocation group size is the minimum allocation
1457 * group size or if the allocation group is completely free. if
1458 * the allocation group size is the minimum size of BPERDMAP (i.e.
1459 * 1 dmap), there is no need to search the dmap control page (below)
1460 * that fully describes the allocation group since the allocation
1461 * group is already fully described by a dmap. in this case, we
1462 * just call dbAllocCtl() to search the dmap tree and allocate the
1463 * required space if available.
1464 *
1465 * if the allocation group is completely free, dbAllocCtl() is
1466 * also called to allocate the required space. this is done for
1467 * two reasons. first, it makes no sense searching the dmap control
1468 * pages for free space when we know that free space exists. second,
1469 * the dmap control pages may indicate that the allocation group
1470 * has no free space if the allocation group is part (not the first
1471 * part) of a larger binary buddy system.
1472 */
1483 }
1485 }
1487 /* the buffer for the dmap control page that fully describes the
1488 * allocation group.
1489 */
1502 }
1504 /* search the subtree(s) of the dmap control page that describes
1505 * the allocation group, looking for sufficient free space. to begin,
1506 * determine how many allocation groups are represented in a dmap
1507 * control page at the control page level (i.e. L0, L1, L2) that
1508 * fully describes an allocation group. next, determine the starting
1509 * tree index of this allocation group within the control page.
1510 */
1511 agperlev =
1515 /* dmap control page trees fan-out by 4 and a single allocation
1516 * group may be described by 1 or 2 subtrees within the ag level
1517 * dmap control page, depending upon the ag size. examine the ag's
1518 * subtrees for sufficient free space, starting with the leftmost
1519 * subtree.
1520 */
1522 /* is there sufficient free space ?
1523 */
1527 /* sufficient free space found in a subtree. now search down
1528 * the subtree to find the leftmost leaf that describes this
1529 * free space.
1530 */
1536 }
1537 }
1543 }
1544 }
1546 /* determine the block number within the file system
1547 * that corresponds to this leaf.
1548 */
1556 blkno +=
1559 /* release the buffer in preparation for going down
1560 * the next level of dmap control pages.
1561 */
1564 /* check if we need to continue to search down the lower
1565 * level dmap control pages. we need to if the number of
1566 * blocks required is less than maximum number of blocks
1567 * described at the next lower level.
1568 */
1571 /* search the lower level dmap control pages to get
1572 * the starting block number of the the dmap that
1573 * contains or starts off the free space.
1574 */
1580 "dbAllocAG: control page "
1583 }
1585 }
1586 }
1588 /* allocate the blocks.
1589 */
1595 }
1597 }
1599 /* no space in the allocation group. release the buffer and
1600 * return -ENOSPC.
1601 */
1605 }
1608 /*
1609 * NAME: dbAllocAny()
1610 *
1611 * FUNCTION: attempt to allocate the specified number of contiguous
1612 * free blocks anywhere in the file system.
1613 *
1614 * dbAllocAny() attempts to find the sufficient free space by
1615 * searching down the dmap control pages, starting with the
1616 * highest level (i.e. L0, L1, L2) control page. if free space
1617 * large enough to satisfy the desired free space is found, the
1618 * desired free space is allocated.
1619 *
1620 * PARAMETERS:
1621 * bmp - pointer to bmap descriptor
1622 * nblocks - actual number of contiguous free blocks desired.
1623 * l2nb - log2 number of contiguous free blocks desired.
1624 * results - on successful return, set to the starting block number
1625 * of the newly allocated range.
1626 *
1627 * RETURN VALUES:
1628 * 0 - success
1629 * -ENOSPC - insufficient disk resources
1630 * -EIO - i/o error
1631 *
1632 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1633 */
1635 {
1639 /* starting with the top level dmap control page, search
1640 * down the dmap control levels for sufficient free space.
1641 * if free space is found, dbFindCtl() returns the starting
1642 * block number of the dmap that contains or starts off the
1643 * range of free space.
1644 */
1648 /* allocate the blocks.
1649 */
1655 }
1657 }
1660 /*
1661 * NAME: dbFindCtl()
1662 *
1663 * FUNCTION: starting at a specified dmap control page level and block
1664 * number, search down the dmap control levels for a range of
1665 * contiguous free blocks large enough to satisfy an allocation
1666 * request for the specified number of free blocks.
1667 *
1668 * if sufficient contiguous free blocks are found, this routine
1669 * returns the starting block number within a dmap page that
1670 * contains or starts a range of contiqious free blocks that
1671 * is sufficient in size.
1672 *
1673 * PARAMETERS:
1674 * bmp - pointer to bmap descriptor
1675 * level - starting dmap control page level.
1676 * l2nb - log2 number of contiguous free blocks desired.
1677 * *blkno - on entry, starting block number for conducting the search.
1678 * on successful return, the first block within a dmap page
1679 * that contains or starts a range of contiguous free blocks.
1680 *
1681 * RETURN VALUES:
1682 * 0 - success
1683 * -ENOSPC - insufficient disk resources
1684 * -EIO - i/o error
1685 *
1686 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1687 */
1689 {
1696 /* starting at the specified dmap control page level and block
1697 * number, search down the dmap control levels for the starting
1698 * block number of a dmap page that contains or starts off
1699 * sufficient free blocks.
1700 */
1702 /* get the buffer of the dmap control page for the block
1703 * number and level (i.e. L0, L1, L2).
1704 */
1717 }
1719 /* search the tree within the dmap control page for
1720 * sufficent free space. if sufficient free space is found,
1721 * dbFindLeaf() returns the index of the leaf at which
1722 * free space was found.
1723 */
1726 /* release the buffer.
1727 */
1730 /* space found ?
1731 */
1737 }
1739 }
1741 /* adjust the block number to reflect the location within
1742 * the dmap control page (i.e. the leaf) at which free
1743 * space was found.
1744 */
1747 /* we stop the search at this dmap control page level if
1748 * the number of blocks required is greater than or equal
1749 * to the maximum number of blocks described at the next
1750 * (lower) level.
1751 */
1754 }
1758 }
1761 /*
1762 * NAME: dbAllocCtl()
1763 *
1764 * FUNCTION: attempt to allocate a specified number of contiguous
1765 * blocks starting within a specific dmap.
1766 *
1767 * this routine is called by higher level routines that search
1768 * the dmap control pages above the actual dmaps for contiguous
1769 * free space. the result of successful searches by these
1770 * routines are the starting block numbers within dmaps, with
1771 * the dmaps themselves containing the desired contiguous free
1772 * space or starting a contiguous free space of desired size
1773 * that is made up of the blocks of one or more dmaps. these
1774 * calls should not fail due to insufficent resources.
1775 *
1776 * this routine is called in some cases where it is not known
1777 * whether it will fail due to insufficient resources. more
1778 * specifically, this occurs when allocating from an allocation
1779 * group whose size is equal to the number of blocks per dmap.
1780 * in this case, the dmap control pages are not examined prior
1781 * to calling this routine (to save pathlength) and the call
1782 * might fail.
1783 *
1784 * for a request size that fits within a dmap, this routine relies
1785 * upon the dmap's dmtree to find the requested contiguous free
1786 * space. for request sizes that are larger than a dmap, the
1787 * requested free space will start at the first block of the
1788 * first dmap (i.e. blkno).
1789 *
1790 * PARAMETERS:
1791 * bmp - pointer to bmap descriptor
1792 * nblocks - actual number of contiguous free blocks to allocate.
1793 * l2nb - log2 number of contiguous free blocks to allocate.
1794 * blkno - starting block number of the dmap to start the allocation
1795 * from.
1796 * results - on successful return, set to the starting block number
1797 * of the newly allocated range.
1798 *
1799 * RETURN VALUES:
1800 * 0 - success
1801 * -ENOSPC - insufficient disk resources
1802 * -EIO - i/o error
1803 *
1804 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1805 */
1806 static int
1808 {
1814 /* check if the allocation request is confined to a single dmap.
1815 */
1817 /* get the buffer for the dmap.
1818 */
1825 /* try to allocate the blocks.
1826 */
1834 }
1836 /* allocation request involving multiple dmaps. it must start on
1837 * a dmap boundary.
1838 */
1841 /* allocate the blocks dmap by dmap.
1842 */
1844 /* get the buffer for the dmap.
1845 */
1851 }
1854 /* the dmap better be all free.
1855 */
1862 }
1864 /* determine how many blocks to allocate from this dmap.
1865 */
1868 /* allocate the blocks from the dmap.
1869 */
1873 }
1875 /* write the buffer.
1876 */
1878 }
1880 /* set the results (starting block number) and return.
1881 */
1885 /* something failed in handling an allocation request involving
1886 * multiple dmaps. we'll try to clean up by backing out any
1887 * allocation that has already happened for this request. if
1888 * we fail in backing out the allocation, we'll mark the file
1889 * system to indicate that blocks have been leaked.
1890 */
1891 backout:
1893 /* try to backout the allocations dmap by dmap.
1894 */
1897 /* get the buffer for this dmap.
1898 */
1902 /* could not back out. mark the file system
1903 * to indicate that we have leaked blocks.
1904 */
1908 }
1911 /* free the blocks is this dmap.
1912 */
1914 /* could not back out. mark the file system
1915 * to indicate that we have leaked blocks.
1916 */
1921 }
1923 /* write the buffer.
1924 */
1926 }
1929 }
1932 /*
1933 * NAME: dbAllocDmapLev()
1934 *
1935 * FUNCTION: attempt to allocate a specified number of contiguous blocks
1936 * from a specified dmap.
1937 *
1938 * this routine checks if the contiguous blocks are available.
1939 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1940 * returned.
1941 *
1942 * PARAMETERS:
1943 * mp - pointer to bmap descriptor
1944 * dp - pointer to dmap to attempt to allocate blocks from.
1945 * l2nb - log2 number of contiguous block desired.
1946 * nblocks - actual number of contiguous block desired.
1947 * results - on successful return, set to the starting block number
1948 * of the newly allocated range.
1949 *
1950 * RETURN VALUES:
1951 * 0 - success
1952 * -ENOSPC - insufficient disk resources
1953 * -EIO - i/o error
1954 *
1955 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1956 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1957 */
1958 static int
1961 {
1962 s64 blkno;
1965 /* can't be more than a dmaps worth of blocks */
1968 /* search the tree within the dmap page for sufficient
1969 * free space. if sufficient free space is found, dbFindLeaf()
1970 * returns the index of the leaf at which free space was found.
1971 */
1975 /* determine the block number within the file system corresponding
1976 * to the leaf at which free space was found.
1977 */
1980 /* if not all bits of the dmap word are free, get the starting
1981 * bit number within the dmap word of the required string of free
1982 * bits and adjust the block number with this value.
1983 */
1987 /* allocate the blocks */
1992 }
1995 /*
1996 * NAME: dbAllocDmap()
1997 *
1998 * FUNCTION: adjust the disk allocation map to reflect the allocation
1999 * of a specified block range within a dmap.
2000 *
2001 * this routine allocates the specified blocks from the dmap
2002 * through a call to dbAllocBits(). if the allocation of the
2003 * block range causes the maximum string of free blocks within
2004 * the dmap to change (i.e. the value of the root of the dmap's
2005 * dmtree), this routine will cause this change to be reflected
2006 * up through the appropriate levels of the dmap control pages
2007 * by a call to dbAdjCtl() for the L0 dmap control page that
2008 * covers this dmap.
2009 *
2010 * PARAMETERS:
2011 * bmp - pointer to bmap descriptor
2012 * dp - pointer to dmap to allocate the block range from.
2013 * blkno - starting block number of the block to be allocated.
2014 * nblocks - number of blocks to be allocated.
2015 *
2016 * RETURN VALUES:
2017 * 0 - success
2018 * -EIO - i/o error
2019 *
2020 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2021 */
2024 {
2025 s8 oldroot;
2028 /* save the current value of the root (i.e. maximum free string)
2029 * of the dmap tree.
2030 */
2033 /* allocate the specified (blocks) bits */
2036 /* if the root has not changed, done. */
2040 /* root changed. bubble the change up to the dmap control pages.
2041 * if the adjustment of the upper level control pages fails,
2042 * backout the bit allocation (thus making everything consistent).
2043 */
2048 }
2051 /*
2052 * NAME: dbFreeDmap()
2053 *
2054 * FUNCTION: adjust the disk allocation map to reflect the allocation
2055 * of a specified block range within a dmap.
2056 *
2057 * this routine frees the specified blocks from the dmap through
2058 * a call to dbFreeBits(). if the deallocation of the block range
2059 * causes the maximum string of free blocks within the dmap to
2060 * change (i.e. the value of the root of the dmap's dmtree), this
2061 * routine will cause this change to be reflected up through the
2062 * appropriate levels of the dmap control pages by a call to
2063 * dbAdjCtl() for the L0 dmap control page that covers this dmap.
2064 *
2065 * PARAMETERS:
2066 * bmp - pointer to bmap descriptor
2067 * dp - pointer to dmap to free the block range from.
2068 * blkno - starting block number of the block to be freed.
2069 * nblocks - number of blocks to be freed.
2070 *
2071 * RETURN VALUES:
2072 * 0 - success
2073 * -EIO - i/o error
2074 *
2075 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2076 */
2079 {
2080 s8 oldroot;
2083 /* save the current value of the root (i.e. maximum free string)
2084 * of the dmap tree.
2085 */
2088 /* free the specified (blocks) bits */
2091 /* if the root has not changed, done. */
2095 /* root changed. bubble the change up to the dmap control pages.
2096 * if the adjustment of the upper level control pages fails,
2097 * backout the deallocation.
2098 */
2102 /* as part of backing out the deallocation, we will have
2103 * to back split the dmap tree if the deallocation caused
2104 * the freed blocks to become part of a larger binary buddy
2105 * system.
2106 */
2111 }
2114 }
2117 /*
2118 * NAME: dbAllocBits()
2119 *
2120 * FUNCTION: allocate a specified block range from a dmap.
2121 *
2122 * this routine updates the dmap to reflect the working
2123 * state allocation of the specified block range. it directly
2124 * updates the bits of the working map and causes the adjustment
2125 * of the binary buddy system described by the dmap's dmtree
2126 * leaves to reflect the bits allocated. it also causes the
2127 * dmap's dmtree, as a whole, to reflect the allocated range.
2128 *
2129 * PARAMETERS:
2130 * bmp - pointer to bmap descriptor
2131 * dp - pointer to dmap to allocate bits from.
2132 * blkno - starting block number of the bits to be allocated.
2133 * nblocks - number of bits to be allocated.
2134 *
2135 * RETURN VALUES: none
2136 *
2137 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2138 */
2141 {
2147 /* pick up a pointer to the leaves of the dmap tree */
2150 /* determine the bit number and word within the dmap of the
2151 * starting block.
2152 */
2156 /* block range better be within the dmap */
2159 /* allocate the bits of the dmap's words corresponding to the block
2160 * range. not all bits of the first and last words may be contained
2161 * within the block range. if this is the case, we'll work against
2162 * those words (i.e. partial first and/or last) on an individual basis
2163 * (a single pass), allocating the bits of interest by hand and
2164 * updating the leaf corresponding to the dmap word. a single pass
2165 * will be used for all dmap words fully contained within the
2166 * specified range. within this pass, the bits of all fully contained
2167 * dmap words will be marked as free in a single shot and the leaves
2168 * will be updated. a single leaf may describe the free space of
2169 * multiple dmap words, so we may update only a subset of the actual
2170 * leaves corresponding to the dmap words of the block range.
2171 */
2173 /* determine the bit number within the word and
2174 * the number of bits within the word.
2175 */
2179 /* check if only part of a word is to be allocated.
2180 */
2182 /* allocate (set to 1) the appropriate bits within
2183 * this dmap word.
2184 */
2188 /* update the leaf for this dmap word. in addition
2189 * to setting the leaf value to the binary buddy max
2190 * of the updated dmap word, dbSplit() will split
2191 * the binary system of the leaves if need be.
2192 */
2198 /* one or more dmap words are fully contained
2199 * within the block range. determine how many
2200 * words and allocate (set to 1) the bits of these
2201 * words.
2202 */
2206 /* determine how many bits.
2207 */
2210 /* now update the appropriate leaves to reflect
2211 * the allocated words.
2212 */
2216 "dbAllocBits: leaf page "
2219 }
2221 /* determine what the leaf value should be
2222 * updated to as the minimum of the l2 number
2223 * of bits being allocated and the l2 number
2224 * of bits currently described by this leaf.
2225 */
2228 /* update the leaf to reflect the allocation.
2229 * in addition to setting the leaf value to
2230 * NOFREE, dbSplit() will split the binary
2231 * system of the leaves to reflect the current
2232 * allocation (size).
2233 */
2236 /* get the number of dmap words handled */
2239 }
2240 }
2241 }
2243 /* update the free count for this dmap */
2248 /* if this allocation group is completely free,
2249 * update the maximum allocation group number if this allocation
2250 * group is the new max.
2251 */
2256 /* update the free count for the allocation group and map */
2261 }
2264 /*
2265 * NAME: dbFreeBits()
2266 *
2267 * FUNCTION: free a specified block range from a dmap.
2268 *
2269 * this routine updates the dmap to reflect the working
2270 * state allocation of the specified block range. it directly
2271 * updates the bits of the working map and causes the adjustment
2272 * of the binary buddy system described by the dmap's dmtree
2273 * leaves to reflect the bits freed. it also causes the dmap's
2274 * dmtree, as a whole, to reflect the deallocated range.
2275 *
2276 * PARAMETERS:
2277 * bmp - pointer to bmap descriptor
2278 * dp - pointer to dmap to free bits from.
2279 * blkno - starting block number of the bits to be freed.
2280 * nblocks - number of bits to be freed.
2281 *
2282 * RETURN VALUES: none
2283 *
2284 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2285 */
2288 {
2293 /* determine the bit number and word within the dmap of the
2294 * starting block.
2295 */
2299 /* block range better be within the dmap.
2300 */
2303 /* free the bits of the dmaps words corresponding to the block range.
2304 * not all bits of the first and last words may be contained within
2305 * the block range. if this is the case, we'll work against those
2306 * words (i.e. partial first and/or last) on an individual basis
2307 * (a single pass), freeing the bits of interest by hand and updating
2308 * the leaf corresponding to the dmap word. a single pass will be used
2309 * for all dmap words fully contained within the specified range.
2310 * within this pass, the bits of all fully contained dmap words will
2311 * be marked as free in a single shot and the leaves will be updated. a
2312 * single leaf may describe the free space of multiple dmap words,
2313 * so we may update only a subset of the actual leaves corresponding
2314 * to the dmap words of the block range.
2315 *
2316 * dbJoin() is used to update leaf values and will join the binary
2317 * buddy system of the leaves if the new leaf values indicate this
2318 * should be done.
2319 */
2321 /* determine the bit number within the word and
2322 * the number of bits within the word.
2323 */
2327 /* check if only part of a word is to be freed.
2328 */
2330 /* free (zero) the appropriate bits within this
2331 * dmap word.
2332 */
2337 /* update the leaf for this dmap word.
2338 */
2344 /* one or more dmap words are fully contained
2345 * within the block range. determine how many
2346 * words and free (zero) the bits of these words.
2347 */
2351 /* determine how many bits.
2352 */
2355 /* now update the appropriate leaves to reflect
2356 * the freed words.
2357 */
2359 /* determine what the leaf value should be
2360 * updated to as the minimum of the l2 number
2361 * of bits being freed and the l2 (max) number
2362 * of bits that can be described by this leaf.
2363 */
2364 size =
2369 /* update the leaf.
2370 */
2373 /* get the number of dmap words handled.
2374 */
2377 }
2378 }
2379 }
2381 /* update the free count for this dmap.
2382 */
2387 /* update the free count for the allocation group and
2388 * map.
2389 */
2394 /* check if this allocation group is not completely free and
2395 * if it is currently the maximum (rightmost) allocation group.
2396 * if so, establish the new maximum allocation group number by
2397 * searching left for the first allocation group with allocation.
2398 */
2407 }
2409 /* re-establish the allocation group preference if the
2410 * current preference is right of the maximum allocation
2411 * group.
2412 */
2415 }
2418 }
2421 /*
2422 * NAME: dbAdjCtl()
2423 *
2424 * FUNCTION: adjust a dmap control page at a specified level to reflect
2425 * the change in a lower level dmap or dmap control page's
2426 * maximum string of free blocks (i.e. a change in the root
2427 * of the lower level object's dmtree) due to the allocation
2428 * or deallocation of a range of blocks with a single dmap.
2429 *
2430 * on entry, this routine is provided with the new value of
2431 * the lower level dmap or dmap control page root and the
2432 * starting block number of the block range whose allocation
2433 * or deallocation resulted in the root change. this range
2434 * is respresented by a single leaf of the current dmapctl
2435 * and the leaf will be updated with this value, possibly
2436 * causing a binary buddy system within the leaves to be
2437 * split or joined. the update may also cause the dmapctl's
2438 * dmtree to be updated.
2439 *
2440 * if the adjustment of the dmap control page, itself, causes its
2441 * root to change, this change will be bubbled up to the next dmap
2442 * control level by a recursive call to this routine, specifying
2443 * the new root value and the next dmap control page level to
2444 * be adjusted.
2445 * PARAMETERS:
2446 * bmp - pointer to bmap descriptor
2447 * blkno - the first block of a block range within a dmap. it is
2448 * the allocation or deallocation of this block range that
2449 * requires the dmap control page to be adjusted.
2450 * newval - the new value of the lower level dmap or dmap control
2451 * page root.
2452 * alloc - TRUE if adjustment is due to an allocation.
2453 * level - current level of dmap control page (i.e. L0, L1, L2) to
2454 * be adjusted.
2455 *
2456 * RETURN VALUES:
2457 * 0 - success
2458 * -EIO - i/o error
2459 *
2460 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2461 */
2462 static int
2464 {
2466 s8 oldroot;
2468 s64 lblkno;
2472 /* get the buffer for the dmap control page for the specified
2473 * block number and control page level.
2474 */
2486 }
2488 /* determine the leaf number corresponding to the block and
2489 * the index within the dmap control tree.
2490 */
2494 /* save the current leaf value and the current root level (i.e.
2495 * maximum l2 free string described by this dmapctl).
2496 */
2500 /* check if this is a control page update for an allocation.
2501 * if so, update the leaf to reflect the new leaf value using
2502 * dbSplit(); otherwise (deallocation), use dbJoin() to udpate
2503 * the leaf with the new value. in addition to updating the
2504 * leaf, dbSplit() will also split the binary buddy system of
2505 * the leaves, if required, and bubble new values within the
2506 * dmapctl tree, if required. similarly, dbJoin() will join
2507 * the binary buddy system of leaves and bubble new values up
2508 * the dmapctl tree as required by the new leaf value.
2509 */
2511 /* check if we are in the middle of a binary buddy
2512 * system. this happens when we are performing the
2513 * first allocation out of an allocation group that
2514 * is part (not the first part) of a larger binary
2515 * buddy system. if we are in the middle, back split
2516 * the system prior to calling dbSplit() which assumes
2517 * that it is at the front of a binary buddy system.
2518 */
2522 }
2526 }
2528 /* check if the root of the current dmap control page changed due
2529 * to the update and if the current dmap control page is not at
2530 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e.
2531 * root changed and this is not the top level), call this routine
2532 * again (recursion) for the next higher level of the mapping to
2533 * reflect the change in root for the current dmap control page.
2534 */
2536 /* are we below the top level of the map. if so,
2537 * bubble the root up to the next higher level.
2538 */
2540 /* bubble up the new root of this dmap control page to
2541 * the next level.
2542 */
2546 /* something went wrong in bubbling up the new
2547 * root value, so backout the changes to the
2548 * current dmap control page.
2549 */
2552 oldval);
2554 /* the dbJoin() above might have
2555 * caused a larger binary buddy system
2556 * to form and we may now be in the
2557 * middle of it. if this is the case,
2558 * back split the buddies.
2559 */
2565 }
2567 /* release the buffer and return the error.
2568 */
2571 }
2573 /* we're at the top level of the map. update
2574 * the bmap control page to reflect the size
2575 * of the maximum free buddy system.
2576 */
2580 "dbAdjCtl: the maximum free buddy is "
2582 }
2584 }
2585 }
2587 /* write the buffer.
2588 */
2592 }
2595 /*
2596 * NAME: dbSplit()
2597 *
2598 * FUNCTION: update the leaf of a dmtree with a new value, splitting
2599 * the leaf from the binary buddy system of the dmtree's
2600 * leaves, as required.
2601 *
2602 * PARAMETERS:
2603 * tp - pointer to the tree containing the leaf.
2604 * leafno - the number of the leaf to be updated.
2605 * splitsz - the size the binary buddy system starting at the leaf
2606 * must be split to, specified as the log2 number of blocks.
2607 * newval - the new value for the leaf.
2608 *
2609 * RETURN VALUES: none
2610 *
2611 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2612 */
2614 {
2619 /* check if the leaf needs to be split.
2620 */
2622 /* the split occurs by cutting the buddy system in half
2623 * at the specified leaf until we reach the specified
2624 * size. pick up the starting split size (current size
2625 * - 1 in l2) and the corresponding buddy size.
2626 */
2630 /* split until we reach the specified size.
2631 */
2633 /* update the buddy's leaf with its new value.
2634 */
2637 /* on to the next size and buddy.
2638 */
2641 }
2642 }
2644 /* adjust the dmap tree to reflect the specified leaf's new
2645 * value.
2646 */
2648 }
2651 /*
2652 * NAME: dbBackSplit()
2653 *
2654 * FUNCTION: back split the binary buddy system of dmtree leaves
2655 * that hold a specified leaf until the specified leaf
2656 * starts its own binary buddy system.
2657 *
2658 * the allocators typically perform allocations at the start
2659 * of binary buddy systems and dbSplit() is used to accomplish
2660 * any required splits. in some cases, however, allocation
2661 * may occur in the middle of a binary system and requires a
2662 * back split, with the split proceeding out from the middle of
2663 * the system (less efficient) rather than the start of the
2664 * system (more efficient). the cases in which a back split
2665 * is required are rare and are limited to the first allocation
2666 * within an allocation group which is a part (not first part)
2667 * of a larger binary buddy system and a few exception cases
2668 * in which a previous join operation must be backed out.
2669 *
2670 * PARAMETERS:
2671 * tp - pointer to the tree containing the leaf.
2672 * leafno - the number of the leaf to be updated.
2673 *
2674 * RETURN VALUES: none
2675 *
2676 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2677 */
2679 {
2684 /* leaf should be part (not first part) of a binary
2685 * buddy system.
2686 */
2689 /* the back split is accomplished by iteratively finding the leaf
2690 * that starts the buddy system that contains the specified leaf and
2691 * splitting that system in two. this iteration continues until
2692 * the specified leaf becomes the start of a buddy system.
2693 *
2694 * determine maximum possible l2 size for the specified leaf.
2695 */
2696 size =
2700 /* determine the number of leaves covered by this size. this
2701 * is the buddy size that we will start with as we search for
2702 * the buddy system that contains the specified leaf.
2703 */
2706 /* back split.
2707 */
2709 /* find the leftmost buddy leaf.
2710 */
2715 /* determine the buddy.
2716 */
2719 /* check if this buddy is the start of the system.
2720 */
2722 /* split the leaf at the start of the
2723 * system in two.
2724 */
2728 }
2729 }
2730 }
2733 }
2736 /*
2737 * NAME: dbJoin()
2738 *
2739 * FUNCTION: update the leaf of a dmtree with a new value, joining
2740 * the leaf with other leaves of the dmtree into a multi-leaf
2741 * binary buddy system, as required.
2742 *
2743 * PARAMETERS:
2744 * tp - pointer to the tree containing the leaf.
2745 * leafno - the number of the leaf to be updated.
2746 * newval - the new value for the leaf.
2747 *
2748 * RETURN VALUES: none
2749 */
2751 {
2755 /* can the new leaf value require a join with other leaves ?
2756 */
2758 /* pickup a pointer to the leaves of the tree.
2759 */
2762 /* try to join the specified leaf into a large binary
2763 * buddy system. the join proceeds by attempting to join
2764 * the specified leafno with its buddy (leaf) at new value.
2765 * if the join occurs, we attempt to join the left leaf
2766 * of the joined buddies with its buddy at new value + 1.
2767 * we continue to join until we find a buddy that cannot be
2768 * joined (does not have a value equal to the size of the
2769 * last join) or until all leaves have been joined into a
2770 * single system.
2771 *
2772 * get the buddy size (number of words covered) of
2773 * the new value.
2774 */
2777 /* try to join.
2778 */
2780 /* get the buddy leaf.
2781 */
2784 /* if the leaf's new value is greater than its
2785 * buddy's value, we join no more.
2786 */
2792 /* check which (leafno or buddy) is the left buddy.
2793 * the left buddy gets to claim the blocks resulting
2794 * from the join while the right gets to claim none.
2795 * the left buddy is also eligable to participate in
2796 * a join at the next higher level while the right
2797 * is not.
2798 *
2799 */
2801 /* leafno is the left buddy.
2802 */
2805 /* buddy is the left buddy and becomes
2806 * leafno.
2807 */
2810 }
2812 /* on to try the next join.
2813 */
2816 }
2817 }
2819 /* update the leaf value.
2820 */
2822 }
2825 /*
2826 * NAME: dbAdjTree()
2827 *
2828 * FUNCTION: update a leaf of a dmtree with a new value, adjusting
2829 * the dmtree, as required, to reflect the new leaf value.
2830 * the combination of any buddies must already be done before
2831 * this is called.
2832 *
2833 * PARAMETERS:
2834 * tp - pointer to the tree to be adjusted.
2835 * leafno - the number of the leaf to be updated.
2836 * newval - the new value for the leaf.
2837 *
2838 * RETURN VALUES: none
2839 */
2841 {
2845 /* pick up the index of the leaf for this leafno.
2846 */
2849 /* is the current value the same as the old value ? if so,
2850 * there is nothing to do.
2851 */
2855 /* set the new value.
2856 */
2859 /* bubble the new value up the tree as required.
2860 */
2862 /* get the index of the first leaf of the 4 leaf
2863 * group containing the specified leaf (leafno).
2864 */
2867 /* get the index of the parent of this 4 leaf group.
2868 */
2871 /* determine the maximum of the 4 leaves.
2872 */
2875 /* if the maximum of the 4 is the same as the
2876 * parent's value, we're done.
2877 */
2881 /* parent gets new value.
2882 */
2885 /* parent becomes leaf for next go-round.
2886 */
2888 }
2889 }
2892 /*
2893 * NAME: dbFindLeaf()
2894 *
2895 * FUNCTION: search a dmtree_t for sufficient free blocks, returning
2896 * the index of a leaf describing the free blocks if
2897 * sufficient free blocks are found.
2898 *
2899 * the search starts at the top of the dmtree_t tree and
2900 * proceeds down the tree to the leftmost leaf with sufficient
2901 * free space.
2902 *
2903 * PARAMETERS:
2904 * tp - pointer to the tree to be searched.
2905 * l2nb - log2 number of free blocks to search for.
2906 * leafidx - return pointer to be set to the index of the leaf
2907 * describing at least l2nb free blocks if sufficient
2908 * free blocks are found.
2909 *
2910 * RETURN VALUES:
2911 * 0 - success
2912 * -ENOSPC - insufficient free blocks.
2913 */
2915 {
2918 /* first check the root of the tree to see if there is
2919 * sufficient free space.
2920 */
2924 /* sufficient free space available. now search down the tree
2925 * starting at the next level for the leftmost leaf that
2926 * describes sufficient free space.
2927 */
2930 /* search the four nodes at this level, starting from
2931 * the left.
2932 */
2934 /* sufficient free space found. move to the next
2935 * level (or quit if this is the last level).
2936 */
2939 }
2941 /* better have found something since the higher
2942 * levels of the tree said it was here.
2943 */
2945 }
2947 /* set the return to the leftmost leaf describing sufficient
2948 * free space.
2949 */
2953 }
2956 /*
2957 * NAME: dbFindBits()
2958 *
2959 * FUNCTION: find a specified number of binary buddy free bits within a
2960 * dmap bitmap word value.
2961 *
2962 * this routine searches the bitmap value for (1 << l2nb) free
2963 * bits at (1 << l2nb) alignments within the value.
2964 *
2965 * PARAMETERS:
2966 * word - dmap bitmap word value.
2967 * l2nb - number of free bits specified as a log2 number.
2968 *
2969 * RETURN VALUES:
2970 * starting bit number of free bits.
2971 */
2973 {
2975 u32 mask;
2977 /* get the number of bits.
2978 */
2982 /* complement the word so we can use a mask (i.e. 0s represent
2983 * free bits) and compute the mask.
2984 */
2988 /* scan the word for nb free bits at nb alignments.
2989 */
2993 }
2997 /* return the bit number.
2998 */
3000 }
3003 /*
3004 * NAME: dbMaxBud(u8 *cp)
3005 *
3006 * FUNCTION: determine the largest binary buddy string of free
3007 * bits within 32-bits of the map.
3008 *
3009 * PARAMETERS:
3010 * cp - pointer to the 32-bit value.
3011 *
3012 * RETURN VALUES:
3013 * largest binary buddy of free bits within a dmap word.
3014 */
3016 {
3019 /* check if the wmap word is all free. if so, the
3020 * free buddy size is BUDMIN.
3021 */
3025 /* check if the wmap word is half free. if so, the
3026 * free buddy size is BUDMIN-1.
3027 */
3031 /* not all free or half free. determine the free buddy
3032 * size thru table lookup using quarters of the wmap word.
3033 */
3037 }
3040 /*
3041 * NAME: cnttz(uint word)
3042 *
3043 * FUNCTION: determine the number of trailing zeros within a 32-bit
3044 * value.
3045 *
3046 * PARAMETERS:
3047 * value - 32-bit value to be examined.
3048 *
3049 * RETURN VALUES:
3050 * count of trailing zeros
3051 */
3053 {
3059 }
3062 }
3065 /*
3066 * NAME: cntlz(u32 value)
3067 *
3068 * FUNCTION: determine the number of leading zeros within a 32-bit
3069 * value.
3070 *
3071 * PARAMETERS:
3072 * value - 32-bit value to be examined.
3073 *
3074 * RETURN VALUES:
3075 * count of leading zeros
3076 */
3078 {
3084 }
3086 }
3089 /*
3090 * NAME: blkstol2(s64 nb)
3091 *
3092 * FUNCTION: convert a block count to its log2 value. if the block
3093 * count is not a l2 multiple, it is rounded up to the next
3094 * larger l2 multiple.
3095 *
3096 * PARAMETERS:
3097 * nb - number of blocks
3098 *
3099 * RETURN VALUES:
3100 * log2 number of blocks
3101 */
3103 {
3109 /* count the leading bits.
3110 */
3112 /* leading bit found.
3113 */
3115 /* determine the l2 value.
3116 */
3119 /* check if we need to round up.
3120 */
3122 l2nb++;
3125 }
3126 }
3129 }
3132 /*
3133 * NAME: dbAllocBottomUp()
3134 *
3135 * FUNCTION: alloc the specified block range from the working block
3136 * allocation map.
3137 *
3138 * the blocks will be alloc from the working map one dmap
3139 * at a time.
3140 *
3141 * PARAMETERS:
3142 * ip - pointer to in-core inode;
3143 * blkno - starting block number to be freed.
3144 * nblocks - number of blocks to be freed.
3145 *
3146 * RETURN VALUES:
3147 * 0 - success
3148 * -EIO - i/o error
3149 */
3151 {
3161 /* block to be allocated better be within the mapsize. */
3164 /*
3165 * allocate the blocks a dmap at a time.
3166 */
3169 /* release previous dmap if any */
3172 }
3174 /* get the buffer for the current dmap. */
3180 }
3183 /* determine the number of blocks to be allocated from
3184 * this dmap.
3185 */
3190 /* allocate the blocks. */
3195 }
3198 }
3200 /* write the last buffer. */
3206 }
3211 {
3217 /* save the current value of the root (i.e. maximum free string)
3218 * of the dmap tree.
3219 */
3222 /* pick up a pointer to the leaves of the dmap tree */
3225 /* determine the bit number and word within the dmap of the
3226 * starting block.
3227 */
3231 /* block range better be within the dmap */
3234 /* allocate the bits of the dmap's words corresponding to the block
3235 * range. not all bits of the first and last words may be contained
3236 * within the block range. if this is the case, we'll work against
3237 * those words (i.e. partial first and/or last) on an individual basis
3238 * (a single pass), allocating the bits of interest by hand and
3239 * updating the leaf corresponding to the dmap word. a single pass
3240 * will be used for all dmap words fully contained within the
3241 * specified range. within this pass, the bits of all fully contained
3242 * dmap words will be marked as free in a single shot and the leaves
3243 * will be updated. a single leaf may describe the free space of
3244 * multiple dmap words, so we may update only a subset of the actual
3245 * leaves corresponding to the dmap words of the block range.
3246 */
3248 /* determine the bit number within the word and
3249 * the number of bits within the word.
3250 */
3254 /* check if only part of a word is to be allocated.
3255 */
3257 /* allocate (set to 1) the appropriate bits within
3258 * this dmap word.
3259 */
3263 word++;
3265 /* one or more dmap words are fully contained
3266 * within the block range. determine how many
3267 * words and allocate (set to 1) the bits of these
3268 * words.
3269 */
3273 /* determine how many bits */
3276 }
3277 }
3279 /* update the free count for this dmap */
3282 /* reconstruct summary tree */
3287 /* if this allocation group is completely free,
3288 * update the highest active allocation group number
3289 * if this allocation group is the new max.
3290 */
3295 /* update the free count for the allocation group and map */
3301 /* if the root has not changed, done. */
3305 /* root changed. bubble the change up to the dmap control pages.
3306 * if the adjustment of the upper level control pages fails,
3307 * backout the bit allocation (thus making everything consistent).
3308 */
3313 }
3316 /*
3317 * NAME: dbExtendFS()
3318 *
3319 * FUNCTION: extend bmap from blkno for nblocks;
3320 * dbExtendFS() updates bmap ready for dbAllocBottomUp();
3321 *
3322 * L2
3323 * |
3324 * L1---------------------------------L1
3325 * | |
3326 * L0---------L0---------L0 L0---------L0---------L0
3327 * | | | | | |
3328 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm;
3329 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3330 *
3331 * <---old---><----------------------------extend----------------------->
3332 */
3334 {
3338 s64 newsize;
3339 s64 p;
3346 s64 ag_rem;
3353 /*
3354 * initialize bmap control page.
3355 *
3356 * all the data in bmap control page should exclude
3357 * the mkfs hidden dmap page.
3358 */
3360 /* update mapsize */
3364 /* compute new AG size */
3371 /* compute new number of AG */
3376 /*
3377 * reconfigure db_agfree[]
3378 * from old AG configuration to new AG configuration;
3379 *
3380 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3381 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3382 * note: new AG size = old AG size * (2**x).
3383 */
3391 /* coalesce cotiguous k AGs; */
3393 /* merge AGi to AGn */
3395 }
3396 }
3402 /*
3403 * update highest active ag number
3404 */
3408 /*
3409 * extend bmap
3410 *
3411 * update bit maps and corresponding level control pages;
3412 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3413 */
3414 extend:
3415 /* get L2 page */
3421 }
3424 /* compute start L1 */
3429 /*
3430 * extend each L1 in L2
3431 */
3433 /* get L1 page */
3435 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3441 /* compute start L0 */
3447 /* assign/init L1 page */
3454 /* compute start L0 */
3458 }
3460 /*
3461 * extend each L0 in L1
3462 */
3464 /* get L0 page */
3466 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3473 /* compute start dmap */
3475 L2BPERDMAP;
3481 /* assign/init L0 page */
3488 /* compute start dmap */
3492 }
3494 /*
3495 * extend each dmap in L0
3496 */
3498 /*
3499 * reconstruct the dmap page, and
3500 * initialize corresponding parent L0 leaf
3501 */
3503 /* read in dmap page: */
3510 /* assign/init dmap page */
3517 }
3528 l0leaf++;
3537 /*
3538 * build current L0 page from its leaves, and
3539 * initialize corresponding parent L1 leaf
3540 */
3548 /* more than 1 L0 ? */
3552 /* summarize in global bmap page */
3557 }
3558 }
3561 /*
3562 * build current L1 page from its leaves, and
3563 * initialize corresponding parent L2 leaf
3564 */
3572 /* more than 1 L1 ? */
3576 /* summarize in global bmap page */
3580 }
3581 }
3586 errout:
3594 /*
3595 * finalize bmap control page
3596 */
3597 finalize:
3600 }
3603 /*
3604 * dbFinalizeBmap()
3605 */
3607 {
3613 /*
3614 * finalize bmap control page
3615 */
3616 //finalize:
3617 /*
3618 * compute db_agpref: preferred ag to allocate from
3619 * (the leftmost ag with average free space in it);
3620 */
3621 //agpref:
3622 /* get the number of active ags and inacitve ags */
3627 /* determine how many blocks are in the inactive allocation
3628 * groups. in doing this, we must account for the fact that
3629 * the rightmost group might be a partial group (i.e. file
3630 * system size is not a multiple of the group size).
3631 */
3636 /* determine how many free blocks are in the active
3637 * allocation groups plus the average number of free blocks
3638 * within the active ags.
3639 */
3643 /* if the preferred allocation group has not average free space.
3644 * re-establish the preferred group as the leftmost
3645 * group with average free space.
3646 */
3652 }
3656 }
3657 }
3659 /*
3660 * compute db_aglevel, db_agheigth, db_width, db_agstart:
3661 * an ag is covered in aglevel dmapctl summary tree,
3662 * at agheight level height (from leaf) with agwidth number of nodes
3663 * each, which starts at agstart index node of the smmary tree node
3664 * array;
3665 */
3667 l2nl =
3672 i--) {
3675 }
3677 }
3680 /*
3681 * NAME: dbInitDmap()/ujfs_idmap_page()
3682 *
3683 * FUNCTION: initialize working/persistent bitmap of the dmap page
3684 * for the specified number of blocks:
3685 *
3686 * at entry, the bitmaps had been initialized as free (ZEROS);
3687 * The number of blocks will only account for the actually
3688 * existing blocks. Blocks which don't actually exist in
3689 * the aggregate will be marked as allocated (ONES);
3690 *
3691 * PARAMETERS:
3692 * dp - pointer to page of map
3693 * nblocks - number of blocks this page
3694 *
3695 * RETURNS: NONE
3696 */
3698 {
3701 /* starting block number within the dmap */
3712 }
3717 }
3719 /* word number containing start block number */
3722 /*
3723 * free the bits corresponding to the block range (ZEROS):
3724 * note: not all bits of the first and last words may be contained
3725 * within the block range.
3726 */
3728 /* number of bits preceding range to be freed in the word */
3730 /* number of bits to free in the word */
3733 /* is partial word to be freed ? */
3735 /* free (set to 0) from the bitmap word */
3741 /* skip the word freed */
3742 w++;
3744 /* free (set to 0) contiguous bitmap words */
3749 /* skip the words freed */
3752 }
3753 }
3755 /*
3756 * mark bits following the range to be freed (non-existing
3757 * blocks) as allocated (ONES)
3758 */
3763 /* the first word beyond the end of existing blocks */
3766 /* does nblocks fall on a 32-bit boundary ? */
3769 /* mark a partial word allocated */
3771 w++;
3772 }
3774 /* set the rest of the words in the page to allocated (ONES) */
3778 /*
3779 * init tree
3780 */
3781 initTree:
3783 }
3786 /*
3787 * NAME: dbInitDmapTree()/ujfs_complete_dmap()
3788 *
3789 * FUNCTION: initialize summary tree of the specified dmap:
3790 *
3791 * at entry, bitmap of the dmap has been initialized;
3792 *
3793 * PARAMETERS:
3794 * dp - dmap to complete
3795 * blkno - starting block number for this dmap
3796 * treemax - will be filled in with max free for this dmap
3797 *
3798 * RETURNS: max free string at the root of the tree
3799 */
3801 {
3806 /* init fixed info of tree */
3814 /* init each leaf from corresponding wmap word:
3815 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3816 * bitmap word are allocated.
3817 */
3822 /* build the dmap's binary buddy summary tree */
3824 }
3827 /*
3828 * NAME: dbInitTree()/ujfs_adjtree()
3829 *
3830 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl.
3831 *
3832 * at entry, the leaves of the tree has been initialized
3833 * from corresponding bitmap word or root of summary tree
3834 * of the child control page;
3835 * configure binary buddy system at the leaf level, then
3836 * bubble up the values of the leaf nodes up the tree.
3837 *
3838 * PARAMETERS:
3839 * cp - Pointer to the root of the tree
3840 * l2leaves- Number of leaf nodes as a power of 2
3841 * l2min - Number of blocks that can be covered by a leaf
3842 * as a power of 2
3843 *
3844 * RETURNS: max free string at the root of the tree
3845 */
3847 {
3854 /* Determine the maximum free string possible for the leaves */
3857 /*
3858 * configure the leaf levevl into binary buddy system
3859 *
3860 * Try to combine buddies starting with a buddy size of 1
3861 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3862 * can be combined if both buddies have a maximum free of l2min;
3863 * the combination will result in the left-most buddy leaf having
3864 * a maximum free of l2min+1.
3865 * After processing all buddies for a given size, process buddies
3866 * at the next higher buddy size (i.e. current size * 2) and
3867 * the next maximum free (current free + 1).
3868 * This continues until the maximum possible buddy combination
3869 * yields maximum free.
3870 */
3873 /* get next buddy size == current buddy pair size */
3876 /* scan each adjacent buddy pair at current buddy size */
3880 /* coalesce if both adjacent buddies are max free */
3884 }
3885 }
3886 }
3888 /*
3889 * bubble summary information of leaves up the tree.
3890 *
3891 * Starting at the leaf node level, the four nodes described by
3892 * the higher level parent node are compared for a maximum free and
3893 * this maximum becomes the value of the parent node.
3894 * when all lower level nodes are processed in this fashion then
3895 * move up to the next level (parent becomes a lower level node) and
3896 * continue the process for that level.
3897 */
3901 /* get index of 1st node of parent level */
3904 /* set the value of the parent node as the maximum
3905 * of the four nodes of the current level.
3906 */
3910 }
3913 }
3916 /*
3917 * dbInitDmapCtl()
3918 *
3919 * function: initialize dmapctl page
3920 */
3931 /*
3932 * initialize the leaves of current level that were not covered
3933 * by the specified input block range (i.e. the leaves have no
3934 * low level dmapctl or dmap).
3935 */
3940 /* build the dmap's binary buddy summary tree */
3942 }
3945 /*
3946 * NAME: dbGetL2AGSize()/ujfs_getagl2size()
3947 *
3948 * FUNCTION: Determine log2(allocation group size) from aggregate size
3949 *
3950 * PARAMETERS:
3951 * nblocks - Number of blocks in aggregate
3952 *
3953 * RETURNS: log2(allocation group size) in aggregate blocks
3954 */
3956 {
3957 s64 sz;
3958 s64 m;
3964 /* round up aggregate size to power of 2 */
3969 }
3975 /* agsize = roundupSize/max_number_of_ag */
3977 }
3980 /*
3981 * NAME: dbMapFileSizeToMapSize()
3982 *
3983 * FUNCTION: compute number of blocks the block allocation map file
3984 * can cover from the map file size;
3985 *
3986 * RETURNS: Number of blocks which can be covered by this block map file;
3987 */
3989 /*
3990 * maximum number of map pages at each level including control pages
3991 */
3992 #define MAXL0PAGES (1 + LPERCTL)
3993 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES)
3994 #define MAXL2PAGES (1 + LPERCTL * MAXL1PAGES)
3996 /*
3997 * convert number of map pages to the zero origin top dmapctl level
3998 */
3999 #define BMAPPGTOLEV(npages) \
4000 (((npages) <= 3 + MAXL0PAGES) ? 0 \
4001 : ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
4004 {
4006 s64 nblocks;
4015 /* At each level, accumulate the number of dmap pages covered by
4016 * the number of full child levels below it;
4017 * repeat for the last incomplete child level.
4018 */
4021 /* skip higher level control pages above top level covered by map */
4025 factor =
4031 /* pages in last/incomplete child */
4033 /* skip incomplete child's level control page */
4034 npages--;
4035 }
4037 /* convert the number of dmaps into the number of blocks
4038 * which can be covered by the dmaps;
4039 */
4043 }
4046 #ifdef _JFS_DEBUG_DMAP
4047 /*
4048 * DBinitmap()
4049 */
4051 {
4069 /* Need to initialize from disk map pages
4070 */
4075 db_l2nbperpage);
4081 }
4088 }
4091 }
4094 /*
4095 * DBAlloc()
4096 */
4098 {
4100 u32 mask;
4116 dbmap++;
4119 }
4120 }
4123 /*
4124 * DBFree()
4125 */
4127 {
4129 u32 mask;
4145 dbmap++;
4148 }
4149 }
4152 /*
4153 * DBAllocCK()
4154 */
4156 {
4158 u32 mask;
4173 dbmap++;
4176 }
4177 }
4180 /*
4181 * DBFreeCK()
4182 */
4184 {
4186 u32 mask;
4201 dbmap++;
4204 }
4205 }
4208 /*
4209 * dbPrtMap()
4210 */
4212 {
4226 }
4229 /*
4230 * dbPrtCtl()
4231 */
4233 {
4250 }
4261 }
4262 }