| blob | 9a55f53be5ff6a38678a8d52253b14c7e8004941 |
1 /*
2 * Copyright (C) International Business Machines Corp., 2000-2004
3 * Portions Copyright (C) Tino Reichardt, 2012
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
20 #include <linux/fs.h>
21 #include <linux/slab.h>
31 /*
32 * SERIALIZATION of the Block Allocation Map.
33 *
34 * the working state of the block allocation map is accessed in
35 * two directions:
36 *
37 * 1) allocation and free requests that start at the dmap
38 * level and move up through the dmap control pages (i.e.
39 * the vast majority of requests).
40 *
41 * 2) allocation requests that start at dmap control page
42 * level and work down towards the dmaps.
43 *
44 * the serialization scheme used here is as follows.
45 *
46 * requests which start at the bottom are serialized against each
47 * other through buffers and each requests holds onto its buffers
48 * as it works it way up from a single dmap to the required level
49 * of dmap control page.
50 * requests that start at the top are serialized against each other
51 * and request that start from the bottom by the multiple read/single
52 * write inode lock of the bmap inode. requests starting at the top
53 * take this lock in write mode while request starting at the bottom
54 * take the lock in read mode. a single top-down request may proceed
55 * exclusively while multiple bottoms-up requests may proceed
56 * simultaneously (under the protection of busy buffers).
57 *
58 * in addition to information found in dmaps and dmap control pages,
59 * the working state of the block allocation map also includes read/
60 * write information maintained in the bmap descriptor (i.e. total
61 * free block count, allocation group level free block counts).
62 * a single exclusive lock (BMAP_LOCK) is used to guard this information
63 * in the face of multiple-bottoms up requests.
64 * (lock ordering: IREAD_LOCK, BMAP_LOCK);
65 *
66 * accesses to the persistent state of the block allocation map (limited
67 * to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
68 */
70 #define BMAP_LOCK_INIT(bmp) mutex_init(&bmp->db_bmaplock)
71 #define BMAP_LOCK(bmp) mutex_lock(&bmp->db_bmaplock)
72 #define BMAP_UNLOCK(bmp) mutex_unlock(&bmp->db_bmaplock)
74 /*
75 * forward references
76 */
122 /*
123 * buddy table
124 *
125 * table used for determining buddy sizes within characters of
126 * dmap bitmap words. the characters themselves serve as indexes
127 * into the table, with the table elements yielding the maximum
128 * binary buddy of free bits within the character.
129 */
147 };
149 /*
150 * NAME: dbMount()
151 *
152 * FUNCTION: initializate the block allocation map.
153 *
154 * memory is allocated for the in-core bmap descriptor and
155 * the in-core descriptor is initialized from disk.
156 *
157 * PARAMETERS:
158 * ipbmap - pointer to in-core inode for the block map.
159 *
160 * RETURN VALUES:
161 * 0 - success
162 * -ENOMEM - insufficient memory
163 * -EIO - i/o error
164 */
166 {
172 /*
173 * allocate/initialize the in-memory bmap descriptor
174 */
175 /* allocate memory for the in-memory bmap descriptor */
180 /* read the on-disk bmap descriptor. */
187 }
189 /* copy the on-disk bmap descriptor to its in-memory version. */
208 /* release the buffer. */
211 /* bind the bmap inode and the bmap descriptor to each other. */
217 /*
218 * allocate/initialize the bmap lock
219 */
223 }
226 /*
227 * NAME: dbUnmount()
228 *
229 * FUNCTION: terminate the block allocation map in preparation for
230 * file system unmount.
231 *
232 * the in-core bmap descriptor is written to disk and
233 * the memory for this descriptor is freed.
234 *
235 * PARAMETERS:
236 * ipbmap - pointer to in-core inode for the block map.
237 *
238 * RETURN VALUES:
239 * 0 - success
240 * -EIO - i/o error
241 */
243 {
249 /*
250 * Invalidate the page cache buffers
251 */
254 /* free the memory for the in-memory bmap. */
258 }
260 /*
261 * dbSync()
262 */
264 {
270 /*
271 * write bmap global control page
272 */
273 /* get the buffer for the on-disk bmap descriptor. */
280 }
281 /* copy the in-memory version of the bmap to the on-disk version */
300 /* write the buffer */
303 /*
304 * write out dirty pages of bmap
305 */
311 }
313 /*
314 * NAME: dbFree()
315 *
316 * FUNCTION: free the specified block range from the working block
317 * allocation map.
318 *
319 * the blocks will be free from the working map one dmap
320 * at a time.
321 *
322 * PARAMETERS:
323 * ip - pointer to in-core inode;
324 * blkno - starting block number to be freed.
325 * nblocks - number of blocks to be freed.
326 *
327 * RETURN VALUES:
328 * 0 - success
329 * -EIO - i/o error
330 */
332 {
343 /* block to be freed better be within the mapsize. */
352 }
354 /**
355 * TRIM the blocks, when mounted with discard option
356 */
361 /*
362 * free the blocks a dmap at a time.
363 */
366 /* release previous dmap if any */
369 }
371 /* get the buffer for the current dmap. */
377 }
380 /* determine the number of blocks to be freed from
381 * this dmap.
382 */
385 /* free the blocks. */
391 }
392 }
394 /* write the last buffer. */
400 }
403 /*
404 * NAME: dbUpdatePMap()
405 *
406 * FUNCTION: update the allocation state (free or allocate) of the
407 * specified block range in the persistent block allocation map.
408 *
409 * the blocks will be updated in the persistent map one
410 * dmap at a time.
411 *
412 * PARAMETERS:
413 * ipbmap - pointer to in-core inode for the block map.
414 * free - 'true' if block range is to be freed from the persistent
415 * map; 'false' if it is to be allocated.
416 * blkno - starting block number of the range.
417 * nblocks - number of contiguous blocks in the range.
418 * tblk - transaction block;
419 *
420 * RETURN VALUES:
421 * 0 - success
422 * -EIO - i/o error
423 */
424 int
427 {
432 u32 mask;
439 /* the blocks better be within the mapsize. */
447 }
449 /* compute delta of transaction lsn from log syncpt */
454 /*
455 * update the block state a dmap at a time.
456 */
460 /* get the buffer for the current dmap. */
465 }
472 }
475 /* determine the bit number and word within the dmap of
476 * the starting block. also determine how many blocks
477 * are to be updated within this dmap.
478 */
483 /* update the bits of the dmap words. the first and last
484 * words may only have a subset of their bits updated. if
485 * this is the case, we'll work against that word (i.e.
486 * partial first and/or last) only in a single pass. a
487 * single pass will also be used to update all words that
488 * are to have all their bits updated.
489 */
492 /* determine the bit number within the word and
493 * the number of bits within the word.
494 */
498 /* check if only part of the word is to be updated. */
500 /* update (free or allocate) the bits
501 * in this word.
502 */
503 mask =
508 else
514 /* one or more words are to have all
515 * their bits updated. determine how
516 * many words and how many bits.
517 */
521 /* update (free or allocate) the bits
522 * in these words.
523 */
527 else
532 }
533 }
535 /*
536 * update dmap lsn
537 */
545 /* inherit older/smaller lsn */
550 /* move bp after tblock in logsync list */
552 }
554 /* inherit younger/larger clsn */
563 /* insert bp after tblock in logsync list */
568 }
570 }
572 /* write the last buffer. */
575 }
578 }
581 /*
582 * NAME: dbNextAG()
583 *
584 * FUNCTION: find the preferred allocation group for new allocations.
585 *
586 * Within the allocation groups, we maintain a preferred
587 * allocation group which consists of a group with at least
588 * average free space. It is the preferred group that we target
589 * new inode allocation towards. The tie-in between inode
590 * allocation and block allocation occurs as we allocate the
591 * first (data) block of an inode and specify the inode (block)
592 * as the allocation hint for this block.
593 *
594 * We try to avoid having more than one open file growing in
595 * an allocation group, as this will lead to fragmentation.
596 * This differs from the old OS/2 method of trying to keep
597 * empty ags around for large allocations.
598 *
599 * PARAMETERS:
600 * ipbmap - pointer to in-core inode for the block map.
601 *
602 * RETURN VALUES:
603 * the preferred allocation group number.
604 */
606 {
607 s64 avgfree;
616 /* determine the average number of free blocks within the ags. */
619 /*
620 * if the current preferred ag does not have an active allocator
621 * and has at least average freespace, return it
622 */
628 /* From the last preferred ag, find the next one with at least
629 * average free space.
630 */
636 /* open file is currently growing in this ag */
639 /* Return this one */
643 /* Less than avg. freespace, but best so far */
646 }
647 }
649 /*
650 * If no inactive ag was found with average freespace, use the
651 * next best
652 */
655 /* else leave db_agpref unchanged */
656 unlock:
659 /* return the preferred group.
660 */
662 }
664 /*
665 * NAME: dbAlloc()
666 *
667 * FUNCTION: attempt to allocate a specified number of contiguous free
668 * blocks from the working allocation block map.
669 *
670 * the block allocation policy uses hints and a multi-step
671 * approach.
672 *
673 * for allocation requests smaller than the number of blocks
674 * per dmap, we first try to allocate the new blocks
675 * immediately following the hint. if these blocks are not
676 * available, we try to allocate blocks near the hint. if
677 * no blocks near the hint are available, we next try to
678 * allocate within the same dmap as contains the hint.
679 *
680 * if no blocks are available in the dmap or the allocation
681 * request is larger than the dmap size, we try to allocate
682 * within the same allocation group as contains the hint. if
683 * this does not succeed, we finally try to allocate anywhere
684 * within the aggregate.
685 *
686 * we also try to allocate anywhere within the aggregate for
687 * for allocation requests larger than the allocation group
688 * size or requests that specify no hint value.
689 *
690 * PARAMETERS:
691 * ip - pointer to in-core inode;
692 * hint - allocation hint.
693 * nblocks - number of contiguous blocks in the range.
694 * results - on successful return, set to the starting block number
695 * of the newly allocated contiguous range.
696 *
697 * RETURN VALUES:
698 * 0 - success
699 * -ENOSPC - insufficient disk resources
700 * -EIO - i/o error
701 */
703 {
711 s64 mapSize;
714 /* assert that nblocks is valid */
717 /* get the log2 number of blocks to be allocated.
718 * if the number of blocks is not a log2 multiple,
719 * it will be rounded up to the next log2 multiple.
720 */
727 /* the hint should be within the map */
731 }
733 /* if the number of blocks to be allocated is greater than the
734 * allocation group size, try to allocate anywhere.
735 */
742 }
744 /*
745 * If no hint, let dbNextAG recommend an allocation group
746 */
750 /* we would like to allocate close to the hint. adjust the
751 * hint to the block following the hint since the allocators
752 * will start looking for free space starting at this point.
753 */
761 /* check if blkno crosses over into a new allocation group.
762 * if so, check if we should allow allocations within this
763 * allocation group.
764 */
766 /* check if the AG is currently being written to.
767 * if so, call dbNextAG() to find a non-busy
768 * AG with sufficient free space.
769 */
773 /* check if the allocation request size can be satisfied from a
774 * single dmap. if so, try to allocate from the dmap containing
775 * the hint using a tiered strategy.
776 */
780 /* get the buffer for the dmap containing the hint.
781 */
790 /* first, try to satisfy the allocation request with the
791 * blocks beginning at the hint.
792 */
798 }
802 }
807 /*
808 * Someone else is writing in this allocation
809 * group. To avoid fragmenting, try another ag
810 */
814 }
816 /* next, try to satisfy the allocation request with blocks
817 * near the hint.
818 */
827 }
829 /* try to satisfy the allocation request with blocks within
830 * the same dmap as the hint.
831 */
839 }
843 }
845 /* try to satisfy the allocation request with blocks within
846 * the same allocation group as the hint.
847 */
855 pref_ag:
856 /*
857 * Let dbNextAG recommend a preferred allocation group
858 */
862 /* Try to allocate within this allocation group. if that fails, try to
863 * allocate anywhere in the map.
864 */
868 write_unlock:
873 read_unlock:
877 }
879 #ifdef _NOTYET
880 /*
881 * NAME: dbAllocExact()
882 *
883 * FUNCTION: try to allocate the requested extent;
884 *
885 * PARAMETERS:
886 * ip - pointer to in-core inode;
887 * blkno - extent address;
888 * nblocks - extent length;
889 *
890 * RETURN VALUES:
891 * 0 - success
892 * -ENOSPC - insufficient disk resources
893 * -EIO - i/o error
894 */
896 {
901 s64 lblkno;
906 /*
907 * validate extent request:
908 *
909 * note: defragfs policy:
910 * max 64 blocks will be moved.
911 * allocation request size must be satisfied from a single dmap.
912 */
916 }
919 /* the free space is no longer available */
922 }
924 /* read in the dmap covering the extent */
930 }
933 /* try to allocate the requested extent */
944 }
947 /*
948 * NAME: dbReAlloc()
949 *
950 * FUNCTION: attempt to extend a current allocation by a specified
951 * number of blocks.
952 *
953 * this routine attempts to satisfy the allocation request
954 * by first trying to extend the existing allocation in
955 * place by allocating the additional blocks as the blocks
956 * immediately following the current allocation. if these
957 * blocks are not available, this routine will attempt to
958 * allocate a new set of contiguous blocks large enough
959 * to cover the existing allocation plus the additional
960 * number of blocks required.
961 *
962 * PARAMETERS:
963 * ip - pointer to in-core inode requiring allocation.
964 * blkno - starting block of the current allocation.
965 * nblocks - number of contiguous blocks within the current
966 * allocation.
967 * addnblocks - number of blocks to add to the allocation.
968 * results - on successful return, set to the starting block number
969 * of the existing allocation if the existing allocation
970 * was extended in place or to a newly allocated contiguous
971 * range if the existing allocation could not be extended
972 * in place.
973 *
974 * RETURN VALUES:
975 * 0 - success
976 * -ENOSPC - insufficient disk resources
977 * -EIO - i/o error
978 */
979 int
982 {
985 /* try to extend the allocation in place.
986 */
993 }
995 /* could not extend the allocation in place, so allocate a
996 * new set of blocks for the entire request (i.e. try to get
997 * a range of contiguous blocks large enough to cover the
998 * existing allocation plus the additional blocks.)
999 */
1002 }
1005 /*
1006 * NAME: dbExtend()
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.
1015 *
1016 * PARAMETERS:
1017 * ip - pointer to in-core inode requiring allocation.
1018 * blkno - starting block of the current allocation.
1019 * nblocks - number of contiguous blocks within the current
1020 * allocation.
1021 * addnblocks - number of blocks to add to the allocation.
1022 *
1023 * RETURN VALUES:
1024 * 0 - success
1025 * -ENOSPC - insufficient disk resources
1026 * -EIO - i/o error
1027 */
1029 {
1032 uint rel_block;
1039 /*
1040 * We don't want a non-aligned extent to cross a page boundary
1041 */
1046 /* get the last block of the current allocation */
1049 /* determine the block number of the block following
1050 * the existing allocation.
1051 */
1056 /* better be within the file system */
1063 }
1065 /* we'll attempt to extend the current allocation in place by
1066 * allocating the additional blocks as the blocks immediately
1067 * following the current allocation. we only try to extend the
1068 * current allocation in place if the number of additional blocks
1069 * can fit into a dmap, the last block of the current allocation
1070 * is not the last block of the file system, and the start of the
1071 * inplace extension is not on an allocation group boundary.
1072 */
1077 }
1079 /* get the buffer for the dmap containing the first block
1080 * of the extension.
1081 */
1087 }
1091 /* try to allocate the blocks immediately following the
1092 * current allocation.
1093 */
1098 /* were we successful ? */
1101 else
1102 /* we were not successful */
1106 }
1109 /*
1110 * NAME: dbAllocNext()
1111 *
1112 * FUNCTION: attempt to allocate the blocks of the specified block
1113 * range within a dmap.
1114 *
1115 * PARAMETERS:
1116 * bmp - pointer to bmap descriptor
1117 * dp - pointer to dmap.
1118 * blkno - starting block number of the range.
1119 * nblocks - number of contiguous free blocks of the range.
1120 *
1121 * RETURN VALUES:
1122 * 0 - success
1123 * -ENOSPC - insufficient disk resources
1124 * -EIO - i/o error
1125 *
1126 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1127 */
1130 {
1134 u32 mask;
1140 }
1142 /* pick up a pointer to the leaves of the dmap tree.
1143 */
1146 /* determine the bit number and word within the dmap of the
1147 * starting block.
1148 */
1152 /* check if the specified block range is contained within
1153 * this dmap.
1154 */
1158 /* check if the starting leaf indicates that anything
1159 * is free.
1160 */
1164 /* check the dmaps words corresponding to block range to see
1165 * if the block range is free. not all bits of the first and
1166 * last words may be contained within the block range. if this
1167 * is the case, we'll work against those words (i.e. partial first
1168 * and/or last) on an individual basis (a single pass) and examine
1169 * the actual bits to determine if they are free. a single pass
1170 * will be used for all dmap words fully contained within the
1171 * specified range. within this pass, the leaves of the dmap
1172 * tree will be examined to determine if the blocks are free. a
1173 * single leaf may describe the free space of multiple dmap
1174 * words, so we may visit only a subset of the actual leaves
1175 * corresponding to the dmap words of the block range.
1176 */
1178 /* determine the bit number within the word and
1179 * the number of bits within the word.
1180 */
1184 /* check if only part of the word is to be examined.
1185 */
1187 /* check if the bits are free.
1188 */
1195 /* one or more dmap words are fully contained
1196 * within the block range. determine how many
1197 * words and how many bits.
1198 */
1202 /* now examine the appropriate leaves to determine
1203 * if the blocks are free.
1204 */
1206 /* does the leaf describe any free space ?
1207 */
1211 /* determine the l2 number of bits provided
1212 * by this leaf.
1213 */
1214 l2size =
1217 /* determine how many words were handled.
1218 */
1223 }
1224 }
1225 }
1227 /* allocate the blocks.
1228 */
1230 }
1233 /*
1234 * NAME: dbAllocNear()
1235 *
1236 * FUNCTION: attempt to allocate a number of contiguous free blocks near
1237 * a specified block (hint) within a dmap.
1238 *
1239 * starting with the dmap leaf that covers the hint, we'll
1240 * check the next four contiguous leaves for sufficient free
1241 * space. if sufficient free space is found, we'll allocate
1242 * the desired free space.
1243 *
1244 * PARAMETERS:
1245 * bmp - pointer to bmap descriptor
1246 * dp - pointer to dmap.
1247 * blkno - block number to allocate near.
1248 * nblocks - actual number of contiguous free blocks desired.
1249 * l2nb - log2 number of contiguous free blocks desired.
1250 * results - on successful return, set to the starting block number
1251 * of the newly allocated range.
1252 *
1253 * RETURN VALUES:
1254 * 0 - success
1255 * -ENOSPC - insufficient disk resources
1256 * -EIO - i/o error
1257 *
1258 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1259 */
1260 static int
1263 {
1271 }
1275 /* determine the word within the dmap that holds the hint
1276 * (i.e. blkno). also, determine the last word in the dmap
1277 * that we'll include in our examination.
1278 */
1282 /* examine the leaves for sufficient free space.
1283 */
1285 /* does the leaf describe sufficient free space ?
1286 */
1290 /* determine the block number within the file system
1291 * of the first block described by this dmap word.
1292 */
1295 /* if not all bits of the dmap word are free, get the
1296 * starting bit number within the dmap word of the required
1297 * string of free bits and adjust the block number with the
1298 * value.
1299 */
1301 blkno +=
1304 /* allocate the blocks.
1305 */
1310 }
1313 }
1316 /*
1317 * NAME: dbAllocAG()
1318 *
1319 * FUNCTION: attempt to allocate the specified number of contiguous
1320 * free blocks within the specified allocation group.
1321 *
1322 * unless the allocation group size is equal to the number
1323 * of blocks per dmap, the dmap control pages will be used to
1324 * find the required free space, if available. we start the
1325 * search at the highest dmap control page level which
1326 * distinctly describes the allocation group's free space
1327 * (i.e. the highest level at which the allocation group's
1328 * free space is not mixed in with that of any other group).
1329 * in addition, we start the search within this level at a
1330 * height of the dmapctl dmtree at which the nodes distinctly
1331 * describe the allocation group's free space. at this height,
1332 * the allocation group's free space may be represented by 1
1333 * or two sub-trees, depending on the allocation group size.
1334 * we search the top nodes of these subtrees left to right for
1335 * sufficient free space. if sufficient free space is found,
1336 * the subtree is searched to find the leftmost leaf that
1337 * has free space. once we have made it to the leaf, we
1338 * move the search to the next lower level dmap control page
1339 * corresponding to this leaf. we continue down the dmap control
1340 * pages until we find the dmap that contains or starts the
1341 * sufficient free space and we allocate at this dmap.
1342 *
1343 * if the allocation group size is equal to the dmap size,
1344 * we'll start at the dmap corresponding to the allocation
1345 * group and attempt the allocation at this level.
1346 *
1347 * the dmap control page search is also not performed if the
1348 * allocation group is completely free and we go to the first
1349 * dmap of the allocation group to do the allocation. this is
1350 * done because the allocation group may be part (not the first
1351 * part) of a larger binary buddy system, causing the dmap
1352 * control pages to indicate no free space (NOFREE) within
1353 * the allocation group.
1354 *
1355 * PARAMETERS:
1356 * bmp - pointer to bmap descriptor
1357 * agno - allocation group number.
1358 * nblocks - actual number of contiguous free blocks desired.
1359 * l2nb - log2 number of contiguous free blocks desired.
1360 * results - on successful return, set to the starting block number
1361 * of the newly allocated range.
1362 *
1363 * RETURN VALUES:
1364 * 0 - success
1365 * -ENOSPC - insufficient disk resources
1366 * -EIO - i/o error
1367 *
1368 * note: IWRITE_LOCK(ipmap) held on entry/exit;
1369 */
1370 static int
1372 {
1379 /* allocation request should not be for more than the
1380 * allocation group size.
1381 */
1384 "dbAllocAG: allocation request is larger than the "
1387 }
1389 /* determine the starting block number of the allocation
1390 * group.
1391 */
1394 /* check if the allocation group size is the minimum allocation
1395 * group size or if the allocation group is completely free. if
1396 * the allocation group size is the minimum size of BPERDMAP (i.e.
1397 * 1 dmap), there is no need to search the dmap control page (below)
1398 * that fully describes the allocation group since the allocation
1399 * group is already fully described by a dmap. in this case, we
1400 * just call dbAllocCtl() to search the dmap tree and allocate the
1401 * required space if available.
1402 *
1403 * if the allocation group is completely free, dbAllocCtl() is
1404 * also called to allocate the required space. this is done for
1405 * two reasons. first, it makes no sense searching the dmap control
1406 * pages for free space when we know that free space exists. second,
1407 * the dmap control pages may indicate that the allocation group
1408 * has no free space if the allocation group is part (not the first
1409 * part) of a larger binary buddy system.
1410 */
1421 }
1423 }
1425 /* the buffer for the dmap control page that fully describes the
1426 * allocation group.
1427 */
1440 }
1442 /* search the subtree(s) of the dmap control page that describes
1443 * the allocation group, looking for sufficient free space. to begin,
1444 * determine how many allocation groups are represented in a dmap
1445 * control page at the control page level (i.e. L0, L1, L2) that
1446 * fully describes an allocation group. next, determine the starting
1447 * tree index of this allocation group within the control page.
1448 */
1449 agperlev =
1453 /* dmap control page trees fan-out by 4 and a single allocation
1454 * group may be described by 1 or 2 subtrees within the ag level
1455 * dmap control page, depending upon the ag size. examine the ag's
1456 * subtrees for sufficient free space, starting with the leftmost
1457 * subtree.
1458 */
1460 /* is there sufficient free space ?
1461 */
1465 /* sufficient free space found in a subtree. now search down
1466 * the subtree to find the leftmost leaf that describes this
1467 * free space.
1468 */
1474 }
1475 }
1481 }
1482 }
1484 /* determine the block number within the file system
1485 * that corresponds to this leaf.
1486 */
1494 blkno +=
1497 /* release the buffer in preparation for going down
1498 * the next level of dmap control pages.
1499 */
1502 /* check if we need to continue to search down the lower
1503 * level dmap control pages. we need to if the number of
1504 * blocks required is less than maximum number of blocks
1505 * described at the next lower level.
1506 */
1509 /* search the lower level dmap control pages to get
1510 * the starting block number of the dmap that
1511 * contains or starts off the free space.
1512 */
1518 "dbAllocAG: control page "
1521 }
1523 }
1524 }
1526 /* allocate the blocks.
1527 */
1533 }
1535 }
1537 /* no space in the allocation group. release the buffer and
1538 * return -ENOSPC.
1539 */
1543 }
1546 /*
1547 * NAME: dbAllocAny()
1548 *
1549 * FUNCTION: attempt to allocate the specified number of contiguous
1550 * free blocks anywhere in the file system.
1551 *
1552 * dbAllocAny() attempts to find the sufficient free space by
1553 * searching down the dmap control pages, starting with the
1554 * highest level (i.e. L0, L1, L2) control page. if free space
1555 * large enough to satisfy the desired free space is found, the
1556 * desired free space is allocated.
1557 *
1558 * PARAMETERS:
1559 * bmp - pointer to bmap descriptor
1560 * nblocks - actual number of contiguous free blocks desired.
1561 * l2nb - log2 number of contiguous free blocks desired.
1562 * results - on successful return, set to the starting block number
1563 * of the newly allocated range.
1564 *
1565 * RETURN VALUES:
1566 * 0 - success
1567 * -ENOSPC - insufficient disk resources
1568 * -EIO - i/o error
1569 *
1570 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1571 */
1573 {
1577 /* starting with the top level dmap control page, search
1578 * down the dmap control levels for sufficient free space.
1579 * if free space is found, dbFindCtl() returns the starting
1580 * block number of the dmap that contains or starts off the
1581 * range of free space.
1582 */
1586 /* allocate the blocks.
1587 */
1593 }
1595 }
1598 /*
1599 * NAME: dbDiscardAG()
1600 *
1601 * FUNCTION: attempt to discard (TRIM) all free blocks of specific AG
1602 *
1603 * algorithm:
1604 * 1) allocate blocks, as large as possible and save them
1605 * while holding IWRITE_LOCK on ipbmap
1606 * 2) trim all these saved block/length values
1607 * 3) mark the blocks free again
1608 *
1609 * benefit:
1610 * - we work only on one ag at some time, minimizing how long we
1611 * need to lock ipbmap
1612 * - reading / writing the fs is possible most time, even on
1613 * trimming
1614 *
1615 * downside:
1616 * - we write two times to the dmapctl and dmap pages
1617 * - but for me, this seems the best way, better ideas?
1618 * /TR 2012
1619 *
1620 * PARAMETERS:
1621 * ip - pointer to in-core inode
1622 * agno - ag to trim
1623 * minlen - minimum value of contiguous blocks
1624 *
1625 * RETURN VALUES:
1626 * s64 - actual number of blocks trimmed
1627 */
1629 {
1638 u64 blkno;
1639 u64 nblocks;
1642 /* max blkno / nblocks pairs to trim */
1644 u64 max_ranges;
1646 /* prevent others from writing new stuff here, while trimming */
1659 }
1665 /* 0 = okay, -EIO = fatal, -ENOSPC -> try smaller block */
1672 /* the whole ag is free, trim now */
1676 /* give a hint for the next while */
1680 /* search for next smaller log2 block */
1684 /* Trim any already allocated blocks */
1688 }
1690 /* check, if our trim array is full */
1693 }
1698 /* when mounted with online discard, dbFree() will
1699 * call jfs_issue_discard() itself */
1704 }
1708 }
1710 /*
1711 * NAME: dbFindCtl()
1712 *
1713 * FUNCTION: starting at a specified dmap control page level and block
1714 * number, search down the dmap control levels for a range of
1715 * contiguous free blocks large enough to satisfy an allocation
1716 * request for the specified number of free blocks.
1717 *
1718 * if sufficient contiguous free blocks are found, this routine
1719 * returns the starting block number within a dmap page that
1720 * contains or starts a range of contiqious free blocks that
1721 * is sufficient in size.
1722 *
1723 * PARAMETERS:
1724 * bmp - pointer to bmap descriptor
1725 * level - starting dmap control page level.
1726 * l2nb - log2 number of contiguous free blocks desired.
1727 * *blkno - on entry, starting block number for conducting the search.
1728 * on successful return, the first block within a dmap page
1729 * that contains or starts a range of contiguous free blocks.
1730 *
1731 * RETURN VALUES:
1732 * 0 - success
1733 * -ENOSPC - insufficient disk resources
1734 * -EIO - i/o error
1735 *
1736 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1737 */
1739 {
1746 /* starting at the specified dmap control page level and block
1747 * number, search down the dmap control levels for the starting
1748 * block number of a dmap page that contains or starts off
1749 * sufficient free blocks.
1750 */
1752 /* get the buffer of the dmap control page for the block
1753 * number and level (i.e. L0, L1, L2).
1754 */
1767 }
1769 /* search the tree within the dmap control page for
1770 * sufficient free space. if sufficient free space is found,
1771 * dbFindLeaf() returns the index of the leaf at which
1772 * free space was found.
1773 */
1776 /* release the buffer.
1777 */
1780 /* space found ?
1781 */
1787 }
1789 }
1791 /* adjust the block number to reflect the location within
1792 * the dmap control page (i.e. the leaf) at which free
1793 * space was found.
1794 */
1797 /* we stop the search at this dmap control page level if
1798 * the number of blocks required is greater than or equal
1799 * to the maximum number of blocks described at the next
1800 * (lower) level.
1801 */
1804 }
1808 }
1811 /*
1812 * NAME: dbAllocCtl()
1813 *
1814 * FUNCTION: attempt to allocate a specified number of contiguous
1815 * blocks starting within a specific dmap.
1816 *
1817 * this routine is called by higher level routines that search
1818 * the dmap control pages above the actual dmaps for contiguous
1819 * free space. the result of successful searches by these
1820 * routines are the starting block numbers within dmaps, with
1821 * the dmaps themselves containing the desired contiguous free
1822 * space or starting a contiguous free space of desired size
1823 * that is made up of the blocks of one or more dmaps. these
1824 * calls should not fail due to insufficent resources.
1825 *
1826 * this routine is called in some cases where it is not known
1827 * whether it will fail due to insufficient resources. more
1828 * specifically, this occurs when allocating from an allocation
1829 * group whose size is equal to the number of blocks per dmap.
1830 * in this case, the dmap control pages are not examined prior
1831 * to calling this routine (to save pathlength) and the call
1832 * might fail.
1833 *
1834 * for a request size that fits within a dmap, this routine relies
1835 * upon the dmap's dmtree to find the requested contiguous free
1836 * space. for request sizes that are larger than a dmap, the
1837 * requested free space will start at the first block of the
1838 * first dmap (i.e. blkno).
1839 *
1840 * PARAMETERS:
1841 * bmp - pointer to bmap descriptor
1842 * nblocks - actual number of contiguous free blocks to allocate.
1843 * l2nb - log2 number of contiguous free blocks to allocate.
1844 * blkno - starting block number of the dmap to start the allocation
1845 * from.
1846 * results - on successful return, set to the starting block number
1847 * of the newly allocated range.
1848 *
1849 * RETURN VALUES:
1850 * 0 - success
1851 * -ENOSPC - insufficient disk resources
1852 * -EIO - i/o error
1853 *
1854 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1855 */
1856 static int
1858 {
1864 /* check if the allocation request is confined to a single dmap.
1865 */
1867 /* get the buffer for the dmap.
1868 */
1875 /* try to allocate the blocks.
1876 */
1884 }
1886 /* allocation request involving multiple dmaps. it must start on
1887 * a dmap boundary.
1888 */
1891 /* allocate the blocks dmap by dmap.
1892 */
1894 /* get the buffer for the dmap.
1895 */
1901 }
1904 /* the dmap better be all free.
1905 */
1912 }
1914 /* determine how many blocks to allocate from this dmap.
1915 */
1918 /* allocate the blocks from the dmap.
1919 */
1923 }
1925 /* write the buffer.
1926 */
1928 }
1930 /* set the results (starting block number) and return.
1931 */
1935 /* something failed in handling an allocation request involving
1936 * multiple dmaps. we'll try to clean up by backing out any
1937 * allocation that has already happened for this request. if
1938 * we fail in backing out the allocation, we'll mark the file
1939 * system to indicate that blocks have been leaked.
1940 */
1941 backout:
1943 /* try to backout the allocations dmap by dmap.
1944 */
1947 /* get the buffer for this dmap.
1948 */
1952 /* could not back out. mark the file system
1953 * to indicate that we have leaked blocks.
1954 */
1958 }
1961 /* free the blocks is this dmap.
1962 */
1964 /* could not back out. mark the file system
1965 * to indicate that we have leaked blocks.
1966 */
1971 }
1973 /* write the buffer.
1974 */
1976 }
1979 }
1982 /*
1983 * NAME: dbAllocDmapLev()
1984 *
1985 * FUNCTION: attempt to allocate a specified number of contiguous blocks
1986 * from a specified dmap.
1987 *
1988 * this routine checks if the contiguous blocks are available.
1989 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1990 * returned.
1991 *
1992 * PARAMETERS:
1993 * mp - pointer to bmap descriptor
1994 * dp - pointer to dmap to attempt to allocate blocks from.
1995 * l2nb - log2 number of contiguous block desired.
1996 * nblocks - actual number of contiguous block desired.
1997 * results - on successful return, set to the starting block number
1998 * of the newly allocated range.
1999 *
2000 * RETURN VALUES:
2001 * 0 - success
2002 * -ENOSPC - insufficient disk resources
2003 * -EIO - i/o error
2004 *
2005 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
2006 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
2007 */
2008 static int
2011 {
2012 s64 blkno;
2015 /* can't be more than a dmaps worth of blocks */
2018 /* search the tree within the dmap page for sufficient
2019 * free space. if sufficient free space is found, dbFindLeaf()
2020 * returns the index of the leaf at which free space was found.
2021 */
2025 /* determine the block number within the file system corresponding
2026 * to the leaf at which free space was found.
2027 */
2030 /* if not all bits of the dmap word are free, get the starting
2031 * bit number within the dmap word of the required string of free
2032 * bits and adjust the block number with this value.
2033 */
2037 /* allocate the blocks */
2042 }
2045 /*
2046 * NAME: dbAllocDmap()
2047 *
2048 * FUNCTION: adjust the disk allocation map to reflect the allocation
2049 * of a specified block range within a dmap.
2050 *
2051 * this routine allocates the specified blocks from the dmap
2052 * through a call to dbAllocBits(). if the allocation of the
2053 * block range causes the maximum string of free blocks within
2054 * the dmap to change (i.e. the value of the root of the dmap's
2055 * dmtree), this routine will cause this change to be reflected
2056 * up through the appropriate levels of the dmap control pages
2057 * by a call to dbAdjCtl() for the L0 dmap control page that
2058 * covers this dmap.
2059 *
2060 * PARAMETERS:
2061 * bmp - pointer to bmap descriptor
2062 * dp - pointer to dmap to allocate the block range from.
2063 * blkno - starting block number of the block to be allocated.
2064 * nblocks - number of blocks to be allocated.
2065 *
2066 * RETURN VALUES:
2067 * 0 - success
2068 * -EIO - i/o error
2069 *
2070 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2071 */
2074 {
2075 s8 oldroot;
2078 /* save the current value of the root (i.e. maximum free string)
2079 * of the dmap tree.
2080 */
2083 /* allocate the specified (blocks) bits */
2086 /* if the root has not changed, done. */
2090 /* root changed. bubble the change up to the dmap control pages.
2091 * if the adjustment of the upper level control pages fails,
2092 * backout the bit allocation (thus making everything consistent).
2093 */
2098 }
2101 /*
2102 * NAME: dbFreeDmap()
2103 *
2104 * FUNCTION: adjust the disk allocation map to reflect the allocation
2105 * of a specified block range within a dmap.
2106 *
2107 * this routine frees the specified blocks from the dmap through
2108 * a call to dbFreeBits(). if the deallocation of the block range
2109 * causes the maximum string of free blocks within the dmap to
2110 * change (i.e. the value of the root of the dmap's dmtree), this
2111 * routine will cause this change to be reflected up through the
2112 * appropriate levels of the dmap control pages by a call to
2113 * dbAdjCtl() for the L0 dmap control page that covers this dmap.
2114 *
2115 * PARAMETERS:
2116 * bmp - pointer to bmap descriptor
2117 * dp - pointer to dmap to free the block range from.
2118 * blkno - starting block number of the block to be freed.
2119 * nblocks - number of blocks to be freed.
2120 *
2121 * RETURN VALUES:
2122 * 0 - success
2123 * -EIO - i/o error
2124 *
2125 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2126 */
2129 {
2130 s8 oldroot;
2133 /* save the current value of the root (i.e. maximum free string)
2134 * of the dmap tree.
2135 */
2138 /* free the specified (blocks) bits */
2141 /* if error or the root has not changed, done. */
2145 /* root changed. bubble the change up to the dmap control pages.
2146 * if the adjustment of the upper level control pages fails,
2147 * backout the deallocation.
2148 */
2152 /* as part of backing out the deallocation, we will have
2153 * to back split the dmap tree if the deallocation caused
2154 * the freed blocks to become part of a larger binary buddy
2155 * system.
2156 */
2161 }
2164 }
2167 /*
2168 * NAME: dbAllocBits()
2169 *
2170 * FUNCTION: allocate a specified block range from a dmap.
2171 *
2172 * this routine updates the dmap to reflect the working
2173 * state allocation of the specified block range. it directly
2174 * updates the bits of the working map and causes the adjustment
2175 * of the binary buddy system described by the dmap's dmtree
2176 * leaves to reflect the bits allocated. it also causes the
2177 * dmap's dmtree, as a whole, to reflect the allocated range.
2178 *
2179 * PARAMETERS:
2180 * bmp - pointer to bmap descriptor
2181 * dp - pointer to dmap to allocate bits from.
2182 * blkno - starting block number of the bits to be allocated.
2183 * nblocks - number of bits to be allocated.
2184 *
2185 * RETURN VALUES: none
2186 *
2187 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2188 */
2191 {
2197 /* pick up a pointer to the leaves of the dmap tree */
2200 /* determine the bit number and word within the dmap of the
2201 * starting block.
2202 */
2206 /* block range better be within the dmap */
2209 /* allocate the bits of the dmap's words corresponding to the block
2210 * range. not all bits of the first and last words may be contained
2211 * within the block range. if this is the case, we'll work against
2212 * those words (i.e. partial first and/or last) on an individual basis
2213 * (a single pass), allocating the bits of interest by hand and
2214 * updating the leaf corresponding to the dmap word. a single pass
2215 * will be used for all dmap words fully contained within the
2216 * specified range. within this pass, the bits of all fully contained
2217 * dmap words will be marked as free in a single shot and the leaves
2218 * will be updated. a single leaf may describe the free space of
2219 * multiple dmap words, so we may update only a subset of the actual
2220 * leaves corresponding to the dmap words of the block range.
2221 */
2223 /* determine the bit number within the word and
2224 * the number of bits within the word.
2225 */
2229 /* check if only part of a word is to be allocated.
2230 */
2232 /* allocate (set to 1) the appropriate bits within
2233 * this dmap word.
2234 */
2238 /* update the leaf for this dmap word. in addition
2239 * to setting the leaf value to the binary buddy max
2240 * of the updated dmap word, dbSplit() will split
2241 * the binary system of the leaves if need be.
2242 */
2248 /* one or more dmap words are fully contained
2249 * within the block range. determine how many
2250 * words and allocate (set to 1) the bits of these
2251 * words.
2252 */
2256 /* determine how many bits.
2257 */
2260 /* now update the appropriate leaves to reflect
2261 * the allocated words.
2262 */
2266 "dbAllocBits: leaf page "
2269 }
2271 /* determine what the leaf value should be
2272 * updated to as the minimum of the l2 number
2273 * of bits being allocated and the l2 number
2274 * of bits currently described by this leaf.
2275 */
2278 /* update the leaf to reflect the allocation.
2279 * in addition to setting the leaf value to
2280 * NOFREE, dbSplit() will split the binary
2281 * system of the leaves to reflect the current
2282 * allocation (size).
2283 */
2286 /* get the number of dmap words handled */
2289 }
2290 }
2291 }
2293 /* update the free count for this dmap */
2298 /* if this allocation group is completely free,
2299 * update the maximum allocation group number if this allocation
2300 * group is the new max.
2301 */
2306 /* update the free count for the allocation group and map */
2311 }
2314 /*
2315 * NAME: dbFreeBits()
2316 *
2317 * FUNCTION: free a specified block range from a dmap.
2318 *
2319 * this routine updates the dmap to reflect the working
2320 * state allocation of the specified block range. it directly
2321 * updates the bits of the working map and causes the adjustment
2322 * of the binary buddy system described by the dmap's dmtree
2323 * leaves to reflect the bits freed. it also causes the dmap's
2324 * dmtree, as a whole, to reflect the deallocated range.
2325 *
2326 * PARAMETERS:
2327 * bmp - pointer to bmap descriptor
2328 * dp - pointer to dmap to free bits from.
2329 * blkno - starting block number of the bits to be freed.
2330 * nblocks - number of bits to be freed.
2331 *
2332 * RETURN VALUES: 0 for success
2333 *
2334 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2335 */
2338 {
2344 /* determine the bit number and word within the dmap of the
2345 * starting block.
2346 */
2350 /* block range better be within the dmap.
2351 */
2354 /* free the bits of the dmaps words corresponding to the block range.
2355 * not all bits of the first and last words may be contained within
2356 * the block range. if this is the case, we'll work against those
2357 * words (i.e. partial first and/or last) on an individual basis
2358 * (a single pass), freeing the bits of interest by hand and updating
2359 * the leaf corresponding to the dmap word. a single pass will be used
2360 * for all dmap words fully contained within the specified range.
2361 * within this pass, the bits of all fully contained dmap words will
2362 * be marked as free in a single shot and the leaves will be updated. a
2363 * single leaf may describe the free space of multiple dmap words,
2364 * so we may update only a subset of the actual leaves corresponding
2365 * to the dmap words of the block range.
2366 *
2367 * dbJoin() is used to update leaf values and will join the binary
2368 * buddy system of the leaves if the new leaf values indicate this
2369 * should be done.
2370 */
2372 /* determine the bit number within the word and
2373 * the number of bits within the word.
2374 */
2378 /* check if only part of a word is to be freed.
2379 */
2381 /* free (zero) the appropriate bits within this
2382 * dmap word.
2383 */
2388 /* update the leaf for this dmap word.
2389 */
2397 /* one or more dmap words are fully contained
2398 * within the block range. determine how many
2399 * words and free (zero) the bits of these words.
2400 */
2404 /* determine how many bits.
2405 */
2408 /* now update the appropriate leaves to reflect
2409 * the freed words.
2410 */
2412 /* determine what the leaf value should be
2413 * updated to as the minimum of the l2 number
2414 * of bits being freed and the l2 (max) number
2415 * of bits that can be described by this leaf.
2416 */
2417 size =
2422 /* update the leaf.
2423 */
2428 /* get the number of dmap words handled.
2429 */
2432 }
2433 }
2434 }
2436 /* update the free count for this dmap.
2437 */
2442 /* update the free count for the allocation group and
2443 * map.
2444 */
2449 /* check if this allocation group is not completely free and
2450 * if it is currently the maximum (rightmost) allocation group.
2451 * if so, establish the new maximum allocation group number by
2452 * searching left for the first allocation group with allocation.
2453 */
2462 }
2464 /* re-establish the allocation group preference if the
2465 * current preference is right of the maximum allocation
2466 * group.
2467 */
2470 }
2475 }
2478 /*
2479 * NAME: dbAdjCtl()
2480 *
2481 * FUNCTION: adjust a dmap control page at a specified level to reflect
2482 * the change in a lower level dmap or dmap control page's
2483 * maximum string of free blocks (i.e. a change in the root
2484 * of the lower level object's dmtree) due to the allocation
2485 * or deallocation of a range of blocks with a single dmap.
2486 *
2487 * on entry, this routine is provided with the new value of
2488 * the lower level dmap or dmap control page root and the
2489 * starting block number of the block range whose allocation
2490 * or deallocation resulted in the root change. this range
2491 * is respresented by a single leaf of the current dmapctl
2492 * and the leaf will be updated with this value, possibly
2493 * causing a binary buddy system within the leaves to be
2494 * split or joined. the update may also cause the dmapctl's
2495 * dmtree to be updated.
2496 *
2497 * if the adjustment of the dmap control page, itself, causes its
2498 * root to change, this change will be bubbled up to the next dmap
2499 * control level by a recursive call to this routine, specifying
2500 * the new root value and the next dmap control page level to
2501 * be adjusted.
2502 * PARAMETERS:
2503 * bmp - pointer to bmap descriptor
2504 * blkno - the first block of a block range within a dmap. it is
2505 * the allocation or deallocation of this block range that
2506 * requires the dmap control page to be adjusted.
2507 * newval - the new value of the lower level dmap or dmap control
2508 * page root.
2509 * alloc - 'true' if adjustment is due to an allocation.
2510 * level - current level of dmap control page (i.e. L0, L1, L2) to
2511 * be adjusted.
2512 *
2513 * RETURN VALUES:
2514 * 0 - success
2515 * -EIO - i/o error
2516 *
2517 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2518 */
2519 static int
2521 {
2523 s8 oldroot;
2525 s64 lblkno;
2529 /* get the buffer for the dmap control page for the specified
2530 * block number and control page level.
2531 */
2543 }
2545 /* determine the leaf number corresponding to the block and
2546 * the index within the dmap control tree.
2547 */
2551 /* save the current leaf value and the current root level (i.e.
2552 * maximum l2 free string described by this dmapctl).
2553 */
2557 /* check if this is a control page update for an allocation.
2558 * if so, update the leaf to reflect the new leaf value using
2559 * dbSplit(); otherwise (deallocation), use dbJoin() to update
2560 * the leaf with the new value. in addition to updating the
2561 * leaf, dbSplit() will also split the binary buddy system of
2562 * the leaves, if required, and bubble new values within the
2563 * dmapctl tree, if required. similarly, dbJoin() will join
2564 * the binary buddy system of leaves and bubble new values up
2565 * the dmapctl tree as required by the new leaf value.
2566 */
2568 /* check if we are in the middle of a binary buddy
2569 * system. this happens when we are performing the
2570 * first allocation out of an allocation group that
2571 * is part (not the first part) of a larger binary
2572 * buddy system. if we are in the middle, back split
2573 * the system prior to calling dbSplit() which assumes
2574 * that it is at the front of a binary buddy system.
2575 */
2581 }
2587 }
2589 /* check if the root of the current dmap control page changed due
2590 * to the update and if the current dmap control page is not at
2591 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e.
2592 * root changed and this is not the top level), call this routine
2593 * again (recursion) for the next higher level of the mapping to
2594 * reflect the change in root for the current dmap control page.
2595 */
2597 /* are we below the top level of the map. if so,
2598 * bubble the root up to the next higher level.
2599 */
2601 /* bubble up the new root of this dmap control page to
2602 * the next level.
2603 */
2607 /* something went wrong in bubbling up the new
2608 * root value, so backout the changes to the
2609 * current dmap control page.
2610 */
2613 oldval);
2615 /* the dbJoin() above might have
2616 * caused a larger binary buddy system
2617 * to form and we may now be in the
2618 * middle of it. if this is the case,
2619 * back split the buddies.
2620 */
2626 }
2628 /* release the buffer and return the error.
2629 */
2632 }
2634 /* we're at the top level of the map. update
2635 * the bmap control page to reflect the size
2636 * of the maximum free buddy system.
2637 */
2641 "dbAdjCtl: the maximum free buddy is "
2643 }
2645 }
2646 }
2648 /* write the buffer.
2649 */
2653 }
2656 /*
2657 * NAME: dbSplit()
2658 *
2659 * FUNCTION: update the leaf of a dmtree with a new value, splitting
2660 * the leaf from the binary buddy system of the dmtree's
2661 * leaves, as required.
2662 *
2663 * PARAMETERS:
2664 * tp - pointer to the tree containing the leaf.
2665 * leafno - the number of the leaf to be updated.
2666 * splitsz - the size the binary buddy system starting at the leaf
2667 * must be split to, specified as the log2 number of blocks.
2668 * newval - the new value for the leaf.
2669 *
2670 * RETURN VALUES: none
2671 *
2672 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2673 */
2675 {
2680 /* check if the leaf needs to be split.
2681 */
2683 /* the split occurs by cutting the buddy system in half
2684 * at the specified leaf until we reach the specified
2685 * size. pick up the starting split size (current size
2686 * - 1 in l2) and the corresponding buddy size.
2687 */
2691 /* split until we reach the specified size.
2692 */
2694 /* update the buddy's leaf with its new value.
2695 */
2698 /* on to the next size and buddy.
2699 */
2702 }
2703 }
2705 /* adjust the dmap tree to reflect the specified leaf's new
2706 * value.
2707 */
2709 }
2712 /*
2713 * NAME: dbBackSplit()
2714 *
2715 * FUNCTION: back split the binary buddy system of dmtree leaves
2716 * that hold a specified leaf until the specified leaf
2717 * starts its own binary buddy system.
2718 *
2719 * the allocators typically perform allocations at the start
2720 * of binary buddy systems and dbSplit() is used to accomplish
2721 * any required splits. in some cases, however, allocation
2722 * may occur in the middle of a binary system and requires a
2723 * back split, with the split proceeding out from the middle of
2724 * the system (less efficient) rather than the start of the
2725 * system (more efficient). the cases in which a back split
2726 * is required are rare and are limited to the first allocation
2727 * within an allocation group which is a part (not first part)
2728 * of a larger binary buddy system and a few exception cases
2729 * in which a previous join operation must be backed out.
2730 *
2731 * PARAMETERS:
2732 * tp - pointer to the tree containing the leaf.
2733 * leafno - the number of the leaf to be updated.
2734 *
2735 * RETURN VALUES: none
2736 *
2737 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2738 */
2740 {
2745 /* leaf should be part (not first part) of a binary
2746 * buddy system.
2747 */
2750 /* the back split is accomplished by iteratively finding the leaf
2751 * that starts the buddy system that contains the specified leaf and
2752 * splitting that system in two. this iteration continues until
2753 * the specified leaf becomes the start of a buddy system.
2754 *
2755 * determine maximum possible l2 size for the specified leaf.
2756 */
2757 size =
2761 /* determine the number of leaves covered by this size. this
2762 * is the buddy size that we will start with as we search for
2763 * the buddy system that contains the specified leaf.
2764 */
2767 /* back split.
2768 */
2770 /* find the leftmost buddy leaf.
2771 */
2777 }
2779 /* determine the buddy.
2780 */
2783 /* check if this buddy is the start of the system.
2784 */
2786 /* split the leaf at the start of the
2787 * system in two.
2788 */
2792 }
2793 }
2794 }
2799 }
2801 }
2804 /*
2805 * NAME: dbJoin()
2806 *
2807 * FUNCTION: update the leaf of a dmtree with a new value, joining
2808 * the leaf with other leaves of the dmtree into a multi-leaf
2809 * binary buddy system, as required.
2810 *
2811 * PARAMETERS:
2812 * tp - pointer to the tree containing the leaf.
2813 * leafno - the number of the leaf to be updated.
2814 * newval - the new value for the leaf.
2815 *
2816 * RETURN VALUES: none
2817 */
2819 {
2823 /* can the new leaf value require a join with other leaves ?
2824 */
2826 /* pickup a pointer to the leaves of the tree.
2827 */
2830 /* try to join the specified leaf into a large binary
2831 * buddy system. the join proceeds by attempting to join
2832 * the specified leafno with its buddy (leaf) at new value.
2833 * if the join occurs, we attempt to join the left leaf
2834 * of the joined buddies with its buddy at new value + 1.
2835 * we continue to join until we find a buddy that cannot be
2836 * joined (does not have a value equal to the size of the
2837 * last join) or until all leaves have been joined into a
2838 * single system.
2839 *
2840 * get the buddy size (number of words covered) of
2841 * the new value.
2842 */
2845 /* try to join.
2846 */
2848 /* get the buddy leaf.
2849 */
2852 /* if the leaf's new value is greater than its
2853 * buddy's value, we join no more.
2854 */
2858 /* It shouldn't be less */
2862 /* check which (leafno or buddy) is the left buddy.
2863 * the left buddy gets to claim the blocks resulting
2864 * from the join while the right gets to claim none.
2865 * the left buddy is also eligible to participate in
2866 * a join at the next higher level while the right
2867 * is not.
2868 *
2869 */
2871 /* leafno is the left buddy.
2872 */
2875 /* buddy is the left buddy and becomes
2876 * leafno.
2877 */
2880 }
2882 /* on to try the next join.
2883 */
2886 }
2887 }
2889 /* update the leaf value.
2890 */
2894 }
2897 /*
2898 * NAME: dbAdjTree()
2899 *
2900 * FUNCTION: update a leaf of a dmtree with a new value, adjusting
2901 * the dmtree, as required, to reflect the new leaf value.
2902 * the combination of any buddies must already be done before
2903 * this is called.
2904 *
2905 * PARAMETERS:
2906 * tp - pointer to the tree to be adjusted.
2907 * leafno - the number of the leaf to be updated.
2908 * newval - the new value for the leaf.
2909 *
2910 * RETURN VALUES: none
2911 */
2913 {
2917 /* pick up the index of the leaf for this leafno.
2918 */
2921 /* is the current value the same as the old value ? if so,
2922 * there is nothing to do.
2923 */
2927 /* set the new value.
2928 */
2931 /* bubble the new value up the tree as required.
2932 */
2934 /* get the index of the first leaf of the 4 leaf
2935 * group containing the specified leaf (leafno).
2936 */
2939 /* get the index of the parent of this 4 leaf group.
2940 */
2943 /* determine the maximum of the 4 leaves.
2944 */
2947 /* if the maximum of the 4 is the same as the
2948 * parent's value, we're done.
2949 */
2953 /* parent gets new value.
2954 */
2957 /* parent becomes leaf for next go-round.
2958 */
2960 }
2961 }
2964 /*
2965 * NAME: dbFindLeaf()
2966 *
2967 * FUNCTION: search a dmtree_t for sufficient free blocks, returning
2968 * the index of a leaf describing the free blocks if
2969 * sufficient free blocks are found.
2970 *
2971 * the search starts at the top of the dmtree_t tree and
2972 * proceeds down the tree to the leftmost leaf with sufficient
2973 * free space.
2974 *
2975 * PARAMETERS:
2976 * tp - pointer to the tree to be searched.
2977 * l2nb - log2 number of free blocks to search for.
2978 * leafidx - return pointer to be set to the index of the leaf
2979 * describing at least l2nb free blocks if sufficient
2980 * free blocks are found.
2981 *
2982 * RETURN VALUES:
2983 * 0 - success
2984 * -ENOSPC - insufficient free blocks.
2985 */
2987 {
2990 /* first check the root of the tree to see if there is
2991 * sufficient free space.
2992 */
2996 /* sufficient free space available. now search down the tree
2997 * starting at the next level for the leftmost leaf that
2998 * describes sufficient free space.
2999 */
3002 /* search the four nodes at this level, starting from
3003 * the left.
3004 */
3006 /* sufficient free space found. move to the next
3007 * level (or quit if this is the last level).
3008 */
3011 }
3013 /* better have found something since the higher
3014 * levels of the tree said it was here.
3015 */
3017 }
3019 /* set the return to the leftmost leaf describing sufficient
3020 * free space.
3021 */
3025 }
3028 /*
3029 * NAME: dbFindBits()
3030 *
3031 * FUNCTION: find a specified number of binary buddy free bits within a
3032 * dmap bitmap word value.
3033 *
3034 * this routine searches the bitmap value for (1 << l2nb) free
3035 * bits at (1 << l2nb) alignments within the value.
3036 *
3037 * PARAMETERS:
3038 * word - dmap bitmap word value.
3039 * l2nb - number of free bits specified as a log2 number.
3040 *
3041 * RETURN VALUES:
3042 * starting bit number of free bits.
3043 */
3045 {
3047 u32 mask;
3049 /* get the number of bits.
3050 */
3054 /* complement the word so we can use a mask (i.e. 0s represent
3055 * free bits) and compute the mask.
3056 */
3060 /* scan the word for nb free bits at nb alignments.
3061 */
3065 }
3069 /* return the bit number.
3070 */
3072 }
3075 /*
3076 * NAME: dbMaxBud(u8 *cp)
3077 *
3078 * FUNCTION: determine the largest binary buddy string of free
3079 * bits within 32-bits of the map.
3080 *
3081 * PARAMETERS:
3082 * cp - pointer to the 32-bit value.
3083 *
3084 * RETURN VALUES:
3085 * largest binary buddy of free bits within a dmap word.
3086 */
3088 {
3091 /* check if the wmap word is all free. if so, the
3092 * free buddy size is BUDMIN.
3093 */
3097 /* check if the wmap word is half free. if so, the
3098 * free buddy size is BUDMIN-1.
3099 */
3103 /* not all free or half free. determine the free buddy
3104 * size thru table lookup using quarters of the wmap word.
3105 */
3109 }
3112 /*
3113 * NAME: cnttz(uint word)
3114 *
3115 * FUNCTION: determine the number of trailing zeros within a 32-bit
3116 * value.
3117 *
3118 * PARAMETERS:
3119 * value - 32-bit value to be examined.
3120 *
3121 * RETURN VALUES:
3122 * count of trailing zeros
3123 */
3125 {
3131 }
3134 }
3137 /*
3138 * NAME: cntlz(u32 value)
3139 *
3140 * FUNCTION: determine the number of leading zeros within a 32-bit
3141 * value.
3142 *
3143 * PARAMETERS:
3144 * value - 32-bit value to be examined.
3145 *
3146 * RETURN VALUES:
3147 * count of leading zeros
3148 */
3150 {
3156 }
3158 }
3161 /*
3162 * NAME: blkstol2(s64 nb)
3163 *
3164 * FUNCTION: convert a block count to its log2 value. if the block
3165 * count is not a l2 multiple, it is rounded up to the next
3166 * larger l2 multiple.
3167 *
3168 * PARAMETERS:
3169 * nb - number of blocks
3170 *
3171 * RETURN VALUES:
3172 * log2 number of blocks
3173 */
3175 {
3181 /* count the leading bits.
3182 */
3184 /* leading bit found.
3185 */
3187 /* determine the l2 value.
3188 */
3191 /* check if we need to round up.
3192 */
3194 l2nb++;
3197 }
3198 }
3201 }
3204 /*
3205 * NAME: dbAllocBottomUp()
3206 *
3207 * FUNCTION: alloc the specified block range from the working block
3208 * allocation map.
3209 *
3210 * the blocks will be alloc from the working map one dmap
3211 * at a time.
3212 *
3213 * PARAMETERS:
3214 * ip - pointer to in-core inode;
3215 * blkno - starting block number to be freed.
3216 * nblocks - number of blocks to be freed.
3217 *
3218 * RETURN VALUES:
3219 * 0 - success
3220 * -EIO - i/o error
3221 */
3223 {
3233 /* block to be allocated better be within the mapsize. */
3236 /*
3237 * allocate the blocks a dmap at a time.
3238 */
3241 /* release previous dmap if any */
3244 }
3246 /* get the buffer for the current dmap. */
3252 }
3255 /* determine the number of blocks to be allocated from
3256 * this dmap.
3257 */
3260 /* allocate the blocks. */
3265 }
3266 }
3268 /* write the last buffer. */
3274 }
3279 {
3282 s8 oldroot;
3285 /* save the current value of the root (i.e. maximum free string)
3286 * of the dmap tree.
3287 */
3290 /* determine the bit number and word within the dmap of the
3291 * starting block.
3292 */
3296 /* block range better be within the dmap */
3299 /* allocate the bits of the dmap's words corresponding to the block
3300 * range. not all bits of the first and last words may be contained
3301 * within the block range. if this is the case, we'll work against
3302 * those words (i.e. partial first and/or last) on an individual basis
3303 * (a single pass), allocating the bits of interest by hand and
3304 * updating the leaf corresponding to the dmap word. a single pass
3305 * will be used for all dmap words fully contained within the
3306 * specified range. within this pass, the bits of all fully contained
3307 * dmap words will be marked as free in a single shot and the leaves
3308 * will be updated. a single leaf may describe the free space of
3309 * multiple dmap words, so we may update only a subset of the actual
3310 * leaves corresponding to the dmap words of the block range.
3311 */
3313 /* determine the bit number within the word and
3314 * the number of bits within the word.
3315 */
3319 /* check if only part of a word is to be allocated.
3320 */
3322 /* allocate (set to 1) the appropriate bits within
3323 * this dmap word.
3324 */
3328 word++;
3330 /* one or more dmap words are fully contained
3331 * within the block range. determine how many
3332 * words and allocate (set to 1) the bits of these
3333 * words.
3334 */
3338 /* determine how many bits */
3341 }
3342 }
3344 /* update the free count for this dmap */
3347 /* reconstruct summary tree */
3352 /* if this allocation group is completely free,
3353 * update the highest active allocation group number
3354 * if this allocation group is the new max.
3355 */
3360 /* update the free count for the allocation group and map */
3366 /* if the root has not changed, done. */
3370 /* root changed. bubble the change up to the dmap control pages.
3371 * if the adjustment of the upper level control pages fails,
3372 * backout the bit allocation (thus making everything consistent).
3373 */
3378 }
3381 /*
3382 * NAME: dbExtendFS()
3383 *
3384 * FUNCTION: extend bmap from blkno for nblocks;
3385 * dbExtendFS() updates bmap ready for dbAllocBottomUp();
3386 *
3387 * L2
3388 * |
3389 * L1---------------------------------L1
3390 * | |
3391 * L0---------L0---------L0 L0---------L0---------L0
3392 * | | | | | |
3393 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm;
3394 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3395 *
3396 * <---old---><----------------------------extend----------------------->
3397 */
3399 {
3403 s64 newsize;
3404 s64 p;
3411 s64 ag_rem;
3418 /*
3419 * initialize bmap control page.
3420 *
3421 * all the data in bmap control page should exclude
3422 * the mkfs hidden dmap page.
3423 */
3425 /* update mapsize */
3429 /* compute new AG size */
3436 /* compute new number of AG */
3441 /*
3442 * reconfigure db_agfree[]
3443 * from old AG configuration to new AG configuration;
3444 *
3445 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3446 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3447 * note: new AG size = old AG size * (2**x).
3448 */
3456 /* coalesce contiguous k AGs; */
3458 /* merge AGi to AGn */
3460 }
3461 }
3467 /*
3468 * update highest active ag number
3469 */
3473 /*
3474 * extend bmap
3475 *
3476 * update bit maps and corresponding level control pages;
3477 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3478 */
3479 extend:
3480 /* get L2 page */
3486 }
3489 /* compute start L1 */
3494 /*
3495 * extend each L1 in L2
3496 */
3498 /* get L1 page */
3500 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3506 /* compute start L0 */
3512 /* assign/init L1 page */
3519 /* compute start L0 */
3523 }
3525 /*
3526 * extend each L0 in L1
3527 */
3529 /* get L0 page */
3531 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3538 /* compute start dmap */
3540 L2BPERDMAP;
3546 /* assign/init L0 page */
3553 /* compute start dmap */
3557 }
3559 /*
3560 * extend each dmap in L0
3561 */
3563 /*
3564 * reconstruct the dmap page, and
3565 * initialize corresponding parent L0 leaf
3566 */
3568 /* read in dmap page: */
3575 /* assign/init dmap page */
3582 }
3593 l0leaf++;
3602 /*
3603 * build current L0 page from its leaves, and
3604 * initialize corresponding parent L1 leaf
3605 */
3613 /* more than 1 L0 ? */
3617 /* summarize in global bmap page */
3622 }
3623 }
3626 /*
3627 * build current L1 page from its leaves, and
3628 * initialize corresponding parent L2 leaf
3629 */
3637 /* more than 1 L1 ? */
3641 /* summarize in global bmap page */
3645 }
3646 }
3651 errout:
3659 /*
3660 * finalize bmap control page
3661 */
3662 finalize:
3665 }
3668 /*
3669 * dbFinalizeBmap()
3670 */
3672 {
3678 /*
3679 * finalize bmap control page
3680 */
3681 //finalize:
3682 /*
3683 * compute db_agpref: preferred ag to allocate from
3684 * (the leftmost ag with average free space in it);
3685 */
3686 //agpref:
3687 /* get the number of active ags and inacitve ags */
3692 /* determine how many blocks are in the inactive allocation
3693 * groups. in doing this, we must account for the fact that
3694 * the rightmost group might be a partial group (i.e. file
3695 * system size is not a multiple of the group size).
3696 */
3701 /* determine how many free blocks are in the active
3702 * allocation groups plus the average number of free blocks
3703 * within the active ags.
3704 */
3708 /* if the preferred allocation group has not average free space.
3709 * re-establish the preferred group as the leftmost
3710 * group with average free space.
3711 */
3717 }
3721 }
3722 }
3724 /*
3725 * compute db_aglevel, db_agheight, db_width, db_agstart:
3726 * an ag is covered in aglevel dmapctl summary tree,
3727 * at agheight level height (from leaf) with agwidth number of nodes
3728 * each, which starts at agstart index node of the smmary tree node
3729 * array;
3730 */
3732 l2nl =
3737 i--) {
3740 }
3742 }
3745 /*
3746 * NAME: dbInitDmap()/ujfs_idmap_page()
3747 *
3748 * FUNCTION: initialize working/persistent bitmap of the dmap page
3749 * for the specified number of blocks:
3750 *
3751 * at entry, the bitmaps had been initialized as free (ZEROS);
3752 * The number of blocks will only account for the actually
3753 * existing blocks. Blocks which don't actually exist in
3754 * the aggregate will be marked as allocated (ONES);
3755 *
3756 * PARAMETERS:
3757 * dp - pointer to page of map
3758 * nblocks - number of blocks this page
3759 *
3760 * RETURNS: NONE
3761 */
3763 {
3766 /* starting block number within the dmap */
3777 }
3781 }
3783 /* word number containing start block number */
3786 /*
3787 * free the bits corresponding to the block range (ZEROS):
3788 * note: not all bits of the first and last words may be contained
3789 * within the block range.
3790 */
3792 /* number of bits preceding range to be freed in the word */
3794 /* number of bits to free in the word */
3797 /* is partial word to be freed ? */
3799 /* free (set to 0) from the bitmap word */
3805 /* skip the word freed */
3806 w++;
3808 /* free (set to 0) contiguous bitmap words */
3813 /* skip the words freed */
3816 }
3817 }
3819 /*
3820 * mark bits following the range to be freed (non-existing
3821 * blocks) as allocated (ONES)
3822 */
3827 /* the first word beyond the end of existing blocks */
3830 /* does nblocks fall on a 32-bit boundary ? */
3833 /* mark a partial word allocated */
3835 w++;
3836 }
3838 /* set the rest of the words in the page to allocated (ONES) */
3842 /*
3843 * init tree
3844 */
3845 initTree:
3847 }
3850 /*
3851 * NAME: dbInitDmapTree()/ujfs_complete_dmap()
3852 *
3853 * FUNCTION: initialize summary tree of the specified dmap:
3854 *
3855 * at entry, bitmap of the dmap has been initialized;
3856 *
3857 * PARAMETERS:
3858 * dp - dmap to complete
3859 * blkno - starting block number for this dmap
3860 * treemax - will be filled in with max free for this dmap
3861 *
3862 * RETURNS: max free string at the root of the tree
3863 */
3865 {
3870 /* init fixed info of tree */
3878 /* init each leaf from corresponding wmap word:
3879 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3880 * bitmap word are allocated.
3881 */
3886 /* build the dmap's binary buddy summary tree */
3888 }
3891 /*
3892 * NAME: dbInitTree()/ujfs_adjtree()
3893 *
3894 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl.
3895 *
3896 * at entry, the leaves of the tree has been initialized
3897 * from corresponding bitmap word or root of summary tree
3898 * of the child control page;
3899 * configure binary buddy system at the leaf level, then
3900 * bubble up the values of the leaf nodes up the tree.
3901 *
3902 * PARAMETERS:
3903 * cp - Pointer to the root of the tree
3904 * l2leaves- Number of leaf nodes as a power of 2
3905 * l2min - Number of blocks that can be covered by a leaf
3906 * as a power of 2
3907 *
3908 * RETURNS: max free string at the root of the tree
3909 */
3911 {
3918 /* Determine the maximum free string possible for the leaves */
3921 /*
3922 * configure the leaf levevl into binary buddy system
3923 *
3924 * Try to combine buddies starting with a buddy size of 1
3925 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3926 * can be combined if both buddies have a maximum free of l2min;
3927 * the combination will result in the left-most buddy leaf having
3928 * a maximum free of l2min+1.
3929 * After processing all buddies for a given size, process buddies
3930 * at the next higher buddy size (i.e. current size * 2) and
3931 * the next maximum free (current free + 1).
3932 * This continues until the maximum possible buddy combination
3933 * yields maximum free.
3934 */
3937 /* get next buddy size == current buddy pair size */
3940 /* scan each adjacent buddy pair at current buddy size */
3944 /* coalesce if both adjacent buddies are max free */
3948 }
3949 }
3950 }
3952 /*
3953 * bubble summary information of leaves up the tree.
3954 *
3955 * Starting at the leaf node level, the four nodes described by
3956 * the higher level parent node are compared for a maximum free and
3957 * this maximum becomes the value of the parent node.
3958 * when all lower level nodes are processed in this fashion then
3959 * move up to the next level (parent becomes a lower level node) and
3960 * continue the process for that level.
3961 */
3965 /* get index of 1st node of parent level */
3968 /* set the value of the parent node as the maximum
3969 * of the four nodes of the current level.
3970 */
3974 }
3977 }
3980 /*
3981 * dbInitDmapCtl()
3982 *
3983 * function: initialize dmapctl page
3984 */
3995 /*
3996 * initialize the leaves of current level that were not covered
3997 * by the specified input block range (i.e. the leaves have no
3998 * low level dmapctl or dmap).
3999 */
4004 /* build the dmap's binary buddy summary tree */
4006 }
4009 /*
4010 * NAME: dbGetL2AGSize()/ujfs_getagl2size()
4011 *
4012 * FUNCTION: Determine log2(allocation group size) from aggregate size
4013 *
4014 * PARAMETERS:
4015 * nblocks - Number of blocks in aggregate
4016 *
4017 * RETURNS: log2(allocation group size) in aggregate blocks
4018 */
4020 {
4021 s64 sz;
4022 s64 m;
4028 /* round up aggregate size to power of 2 */
4033 }
4039 /* agsize = roundupSize/max_number_of_ag */
4041 }
4044 /*
4045 * NAME: dbMapFileSizeToMapSize()
4046 *
4047 * FUNCTION: compute number of blocks the block allocation map file
4048 * can cover from the map file size;
4049 *
4050 * RETURNS: Number of blocks which can be covered by this block map file;
4051 */
4053 /*
4054 * maximum number of map pages at each level including control pages
4055 */
4056 #define MAXL0PAGES (1 + LPERCTL)
4057 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES)
4058 #define MAXL2PAGES (1 + LPERCTL * MAXL1PAGES)
4060 /*
4061 * convert number of map pages to the zero origin top dmapctl level
4062 */
4063 #define BMAPPGTOLEV(npages) \
4064 (((npages) <= 3 + MAXL0PAGES) ? 0 : \
4065 ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
4068 {
4070 s64 nblocks;
4079 /* At each level, accumulate the number of dmap pages covered by
4080 * the number of full child levels below it;
4081 * repeat for the last incomplete child level.
4082 */
4085 /* skip higher level control pages above top level covered by map */
4089 factor =
4095 /* pages in last/incomplete child */
4097 /* skip incomplete child's level control page */
4098 npages--;
4099 }
4101 /* convert the number of dmaps into the number of blocks
4102 * which can be covered by the dmaps;
4103 */
4107 }