| blob | eadf319bee22a67eca765ba30c0fac74ca46d020 |
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 * SERIALIZATION of the Block Allocation Map.
30 *
31 * the working state of the block allocation map is accessed in
32 * two directions:
33 *
34 * 1) allocation and free requests that start at the dmap
35 * level and move up through the dmap control pages (i.e.
36 * the vast majority of requests).
37 *
38 * 2) allocation requests that start at dmap control page
39 * level and work down towards the dmaps.
40 *
41 * the serialization scheme used here is as follows.
42 *
43 * requests which start at the bottom are serialized against each
44 * other through buffers and each requests holds onto its buffers
45 * as it works it way up from a single dmap to the required level
46 * of dmap control page.
47 * requests that start at the top are serialized against each other
48 * and request that start from the bottom by the multiple read/single
49 * write inode lock of the bmap inode. requests starting at the top
50 * take this lock in write mode while request starting at the bottom
51 * take the lock in read mode. a single top-down request may proceed
52 * exclusively while multiple bottoms-up requests may proceed
53 * simultaneously (under the protection of busy buffers).
54 *
55 * in addition to information found in dmaps and dmap control pages,
56 * the working state of the block allocation map also includes read/
57 * write information maintained in the bmap descriptor (i.e. total
58 * free block count, allocation group level free block counts).
59 * a single exclusive lock (BMAP_LOCK) is used to guard this information
60 * in the face of multiple-bottoms up requests.
61 * (lock ordering: IREAD_LOCK, BMAP_LOCK);
62 *
63 * accesses to the persistent state of the block allocation map (limited
64 * to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
65 */
67 #define BMAP_LOCK_INIT(bmp) init_MUTEX(&bmp->db_bmaplock)
68 #define BMAP_LOCK(bmp) down(&bmp->db_bmaplock)
69 #define BMAP_UNLOCK(bmp) up(&bmp->db_bmaplock)
71 /*
72 * forward references
73 */
120 /*
121 * buddy table
122 *
123 * table used for determining buddy sizes within characters of
124 * dmap bitmap words. the characters themselves serve as indexes
125 * into the table, with the table elements yielding the maximum
126 * binary buddy of free bits within the character.
127 */
145 };
148 /*
149 * NAME: dbMount()
150 *
151 * FUNCTION: initializate the block allocation map.
152 *
153 * memory is allocated for the in-core bmap descriptor and
154 * the in-core descriptor is initialized from disk.
155 *
156 * PARAMETERS:
157 * ipbmap - pointer to in-core inode for the block map.
158 *
159 * RETURN VALUES:
160 * 0 - success
161 * -ENOMEM - insufficient memory
162 * -EIO - i/o error
163 */
165 {
171 /*
172 * allocate/initialize the in-memory bmap descriptor
173 */
174 /* allocate memory for the in-memory bmap descriptor */
179 /* read the on-disk bmap descriptor. */
186 }
188 /* copy the on-disk bmap descriptor to its in-memory version. */
207 /* release the buffer. */
210 /* bind the bmap inode and the bmap descriptor to each other. */
216 /*
217 * allocate/initialize the bmap lock
218 */
222 }
225 /*
226 * NAME: dbUnmount()
227 *
228 * FUNCTION: terminate the block allocation map in preparation for
229 * file system unmount.
230 *
231 * the in-core bmap descriptor is written to disk and
232 * the memory for this descriptor is freed.
233 *
234 * PARAMETERS:
235 * ipbmap - pointer to in-core inode for the block map.
236 *
237 * RETURN VALUES:
238 * 0 - success
239 * -EIO - i/o error
240 */
242 {
248 /*
249 * Invalidate the page cache buffers
250 */
253 /* free the memory for the in-memory bmap. */
257 }
259 /*
260 * dbSync()
261 */
263 {
269 /*
270 * write bmap global control page
271 */
272 /* get the buffer for the on-disk bmap descriptor. */
279 }
280 /* copy the in-memory version of the bmap to the on-disk version */
299 /* write the buffer */
302 /*
303 * write out dirty pages of bmap
304 */
312 }
315 /*
316 * NAME: dbFree()
317 *
318 * FUNCTION: free the specified block range from the working block
319 * allocation map.
320 *
321 * the blocks will be free from the working map one dmap
322 * at a time.
323 *
324 * PARAMETERS:
325 * ip - pointer to in-core inode;
326 * blkno - starting block number to be freed.
327 * nblocks - number of blocks to be freed.
328 *
329 * RETURN VALUES:
330 * 0 - success
331 * -EIO - i/o error
332 */
334 {
344 /* block to be freed better be within the mapsize. */
353 }
355 /*
356 * free the blocks a dmap at a time.
357 */
360 /* release previous dmap if any */
363 }
365 /* get the buffer for the current dmap. */
371 }
374 /* determine the number of blocks to be freed from
375 * this dmap.
376 */
379 /* free the blocks. */
385 }
386 }
388 /* write the last buffer. */
394 }
397 /*
398 * NAME: dbUpdatePMap()
399 *
400 * FUNCTION: update the allocation state (free or allocate) of the
401 * specified block range in the persistent block allocation map.
402 *
403 * the blocks will be updated in the persistent map one
404 * dmap at a time.
405 *
406 * PARAMETERS:
407 * ipbmap - pointer to in-core inode for the block map.
408 * free - TRUE if block range is to be freed from the persistent
409 * map; FALSE if it is to be allocated.
410 * blkno - starting block number of the range.
411 * nblocks - number of contiguous blocks in the range.
412 * tblk - transaction block;
413 *
414 * RETURN VALUES:
415 * 0 - success
416 * -EIO - i/o error
417 */
418 int
421 {
426 u32 mask;
433 /* the blocks better be within the mapsize. */
441 }
443 /* compute delta of transaction lsn from log syncpt */
448 /*
449 * update the block state a dmap at a time.
450 */
454 /* get the buffer for the current dmap. */
459 }
466 }
469 /* determine the bit number and word within the dmap of
470 * the starting block. also determine how many blocks
471 * are to be updated within this dmap.
472 */
477 /* update the bits of the dmap words. the first and last
478 * words may only have a subset of their bits updated. if
479 * this is the case, we'll work against that word (i.e.
480 * partial first and/or last) only in a single pass. a
481 * single pass will also be used to update all words that
482 * are to have all their bits updated.
483 */
486 /* determine the bit number within the word and
487 * the number of bits within the word.
488 */
492 /* check if only part of the word is to be updated. */
494 /* update (free or allocate) the bits
495 * in this word.
496 */
497 mask =
502 else
508 /* one or more words are to have all
509 * their bits updated. determine how
510 * many words and how many bits.
511 */
515 /* update (free or allocate) the bits
516 * in these words.
517 */
521 else
526 }
527 }
529 /*
530 * update dmap lsn
531 */
538 /* inherit older/smaller lsn */
544 /* move bp after tblock in logsync list */
546 }
548 /* inherit younger/larger clsn */
558 /* insert bp after tblock in logsync list */
566 }
567 }
569 /* write the last buffer. */
572 }
575 }
578 /*
579 * NAME: dbNextAG()
580 *
581 * FUNCTION: find the preferred allocation group for new allocations.
582 *
583 * Within the allocation groups, we maintain a preferred
584 * allocation group which consists of a group with at least
585 * average free space. It is the preferred group that we target
586 * new inode allocation towards. The tie-in between inode
587 * allocation and block allocation occurs as we allocate the
588 * first (data) block of an inode and specify the inode (block)
589 * as the allocation hint for this block.
590 *
591 * We try to avoid having more than one open file growing in
592 * an allocation group, as this will lead to fragmentation.
593 * This differs from the old OS/2 method of trying to keep
594 * empty ags around for large allocations.
595 *
596 * PARAMETERS:
597 * ipbmap - pointer to in-core inode for the block map.
598 *
599 * RETURN VALUES:
600 * the preferred allocation group number.
601 */
603 {
604 s64 avgfree;
613 /* determine the average number of free blocks within the ags. */
616 /*
617 * if the current preferred ag does not have an active allocator
618 * and has at least average freespace, return it
619 */
625 /* From the last preferred ag, find the next one with at least
626 * average free space.
627 */
633 /* open file is currently growing in this ag */
636 /* Return this one */
640 /* Less than avg. freespace, but best so far */
643 }
644 }
646 /*
647 * If no inactive ag was found with average freespace, use the
648 * next best
649 */
652 /* else leave db_agpref unchanged */
653 unlock:
656 /* return the preferred group.
657 */
659 }
661 /*
662 * NAME: dbAlloc()
663 *
664 * FUNCTION: attempt to allocate a specified number of contiguous free
665 * blocks from the working allocation block map.
666 *
667 * the block allocation policy uses hints and a multi-step
668 * approach.
669 *
670 * for allocation requests smaller than the number of blocks
671 * per dmap, we first try to allocate the new blocks
672 * immediately following the hint. if these blocks are not
673 * available, we try to allocate blocks near the hint. if
674 * no blocks near the hint are available, we next try to
675 * allocate within the same dmap as contains the hint.
676 *
677 * if no blocks are available in the dmap or the allocation
678 * request is larger than the dmap size, we try to allocate
679 * within the same allocation group as contains the hint. if
680 * this does not succeed, we finally try to allocate anywhere
681 * within the aggregate.
682 *
683 * we also try to allocate anywhere within the aggregate for
684 * for allocation requests larger than the allocation group
685 * size or requests that specify no hint value.
686 *
687 * PARAMETERS:
688 * ip - pointer to in-core inode;
689 * hint - allocation hint.
690 * nblocks - number of contiguous blocks in the range.
691 * results - on successful return, set to the starting block number
692 * of the newly allocated contiguous range.
693 *
694 * RETURN VALUES:
695 * 0 - success
696 * -ENOSPC - insufficient disk resources
697 * -EIO - i/o error
698 */
700 {
708 s64 mapSize;
711 /* assert that nblocks is valid */
714 #ifdef _STILL_TO_PORT
715 /* DASD limit check F226941 */
720 /* get the log2 number of blocks to be allocated.
721 * if the number of blocks is not a log2 multiple,
722 * it will be rounded up to the next log2 multiple.
723 */
728 //retry: /* serialize w.r.t.extendfs() */
731 /* the hint should be within the map */
735 }
737 /* if the number of blocks to be allocated is greater than the
738 * allocation group size, try to allocate anywhere.
739 */
746 }
748 /*
749 * If no hint, let dbNextAG recommend an allocation group
750 */
754 /* we would like to allocate close to the hint. adjust the
755 * hint to the block following the hint since the allocators
756 * will start looking for free space starting at this point.
757 */
765 /* check if blkno crosses over into a new allocation group.
766 * if so, check if we should allow allocations within this
767 * allocation group.
768 */
770 /* check if the AG is currenly being written to.
771 * if so, call dbNextAG() to find a non-busy
772 * AG with sufficient free space.
773 */
777 /* check if the allocation request size can be satisfied from a
778 * single dmap. if so, try to allocate from the dmap containing
779 * the hint using a tiered strategy.
780 */
784 /* get the buffer for the dmap containing the hint.
785 */
794 /* first, try to satisfy the allocation request with the
795 * blocks beginning at the hint.
796 */
802 }
806 }
811 /*
812 * Someone else is writing in this allocation
813 * group. To avoid fragmenting, try another ag
814 */
818 }
820 /* next, try to satisfy the allocation request with blocks
821 * near the hint.
822 */
831 }
833 /* try to satisfy the allocation request with blocks within
834 * the same dmap as the hint.
835 */
843 }
847 }
849 /* try to satisfy the allocation request with blocks within
850 * the same allocation group as the hint.
851 */
859 pref_ag:
860 /*
861 * Let dbNextAG recommend a preferred allocation group
862 */
866 /* Try to allocate within this allocation group. if that fails, try to
867 * allocate anywhere in the map.
868 */
872 write_unlock:
877 read_unlock:
881 }
883 #ifdef _NOTYET
884 /*
885 * NAME: dbAllocExact()
886 *
887 * FUNCTION: try to allocate the requested extent;
888 *
889 * PARAMETERS:
890 * ip - pointer to in-core inode;
891 * blkno - extent address;
892 * nblocks - extent length;
893 *
894 * RETURN VALUES:
895 * 0 - success
896 * -ENOSPC - insufficient disk resources
897 * -EIO - i/o error
898 */
900 {
905 s64 lblkno;
910 /*
911 * validate extent request:
912 *
913 * note: defragfs policy:
914 * max 64 blocks will be moved.
915 * allocation request size must be satisfied from a single dmap.
916 */
920 }
923 /* the free space is no longer available */
926 }
928 /* read in the dmap covering the extent */
934 }
937 /* try to allocate the requested extent */
948 }
951 /*
952 * NAME: dbReAlloc()
953 *
954 * FUNCTION: attempt to extend a current allocation by a specified
955 * number of blocks.
956 *
957 * this routine attempts to satisfy the allocation request
958 * by first trying to extend the existing allocation in
959 * place by allocating the additional blocks as the blocks
960 * immediately following the current allocation. if these
961 * blocks are not available, this routine will attempt to
962 * allocate a new set of contiguous blocks large enough
963 * to cover the existing allocation plus the additional
964 * number of blocks required.
965 *
966 * PARAMETERS:
967 * ip - pointer to in-core inode requiring allocation.
968 * blkno - starting block of the current allocation.
969 * nblocks - number of contiguous blocks within the current
970 * allocation.
971 * addnblocks - number of blocks to add to the allocation.
972 * results - on successful return, set to the starting block number
973 * of the existing allocation if the existing allocation
974 * was extended in place or to a newly allocated contiguous
975 * range if the existing allocation could not be extended
976 * in place.
977 *
978 * RETURN VALUES:
979 * 0 - success
980 * -ENOSPC - insufficient disk resources
981 * -EIO - i/o error
982 */
983 int
986 {
989 /* try to extend the allocation in place.
990 */
997 }
999 /* could not extend the allocation in place, so allocate a
1000 * new set of blocks for the entire request (i.e. try to get
1001 * a range of contiguous blocks large enough to cover the
1002 * existing allocation plus the additional blocks.)
1003 */
1006 }
1009 /*
1010 * NAME: dbExtend()
1011 *
1012 * FUNCTION: attempt to extend a current allocation by a specified
1013 * number of blocks.
1014 *
1015 * this routine attempts to satisfy the allocation request
1016 * by first trying to extend the existing allocation in
1017 * place by allocating the additional blocks as the blocks
1018 * immediately following the current allocation.
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 *
1027 * RETURN VALUES:
1028 * 0 - success
1029 * -ENOSPC - insufficient disk resources
1030 * -EIO - i/o error
1031 */
1033 {
1036 uint rel_block;
1043 /*
1044 * We don't want a non-aligned extent to cross a page boundary
1045 */
1050 /* get the last block of the current allocation */
1053 /* determine the block number of the block following
1054 * the existing allocation.
1055 */
1060 /* better be within the file system */
1067 }
1069 /* we'll attempt to extend the current allocation in place by
1070 * allocating the additional blocks as the blocks immediately
1071 * following the current allocation. we only try to extend the
1072 * current allocation in place if the number of additional blocks
1073 * can fit into a dmap, the last block of the current allocation
1074 * is not the last block of the file system, and the start of the
1075 * inplace extension is not on an allocation group boundary.
1076 */
1081 }
1083 /* get the buffer for the dmap containing the first block
1084 * of the extension.
1085 */
1091 }
1095 /* try to allocate the blocks immediately following the
1096 * current allocation.
1097 */
1102 /* were we successful ? */
1105 else
1106 /* we were not successful */
1111 }
1114 /*
1115 * NAME: dbAllocNext()
1116 *
1117 * FUNCTION: attempt to allocate the blocks of the specified block
1118 * range within a dmap.
1119 *
1120 * PARAMETERS:
1121 * bmp - pointer to bmap descriptor
1122 * dp - pointer to dmap.
1123 * blkno - starting block number of the range.
1124 * nblocks - number of contiguous free blocks of the range.
1125 *
1126 * RETURN VALUES:
1127 * 0 - success
1128 * -ENOSPC - insufficient disk resources
1129 * -EIO - i/o error
1130 *
1131 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1132 */
1135 {
1139 u32 mask;
1145 }
1147 /* pick up a pointer to the leaves of the dmap tree.
1148 */
1151 /* determine the bit number and word within the dmap of the
1152 * starting block.
1153 */
1157 /* check if the specified block range is contained within
1158 * this dmap.
1159 */
1163 /* check if the starting leaf indicates that anything
1164 * is free.
1165 */
1169 /* check the dmaps words corresponding to block range to see
1170 * if the block range is free. not all bits of the first and
1171 * last words may be contained within the block range. if this
1172 * is the case, we'll work against those words (i.e. partial first
1173 * and/or last) on an individual basis (a single pass) and examine
1174 * the actual bits to determine if they are free. a single pass
1175 * will be used for all dmap words fully contained within the
1176 * specified range. within this pass, the leaves of the dmap
1177 * tree will be examined to determine if the blocks are free. a
1178 * single leaf may describe the free space of multiple dmap
1179 * words, so we may visit only a subset of the actual leaves
1180 * corresponding to the dmap words of the block range.
1181 */
1183 /* determine the bit number within the word and
1184 * the number of bits within the word.
1185 */
1189 /* check if only part of the word is to be examined.
1190 */
1192 /* check if the bits are free.
1193 */
1200 /* one or more dmap words are fully contained
1201 * within the block range. determine how many
1202 * words and how many bits.
1203 */
1207 /* now examine the appropriate leaves to determine
1208 * if the blocks are free.
1209 */
1211 /* does the leaf describe any free space ?
1212 */
1216 /* determine the l2 number of bits provided
1217 * by this leaf.
1218 */
1219 l2size =
1222 /* determine how many words were handled.
1223 */
1228 }
1229 }
1230 }
1232 /* allocate the blocks.
1233 */
1235 }
1238 /*
1239 * NAME: dbAllocNear()
1240 *
1241 * FUNCTION: attempt to allocate a number of contiguous free blocks near
1242 * a specified block (hint) within a dmap.
1243 *
1244 * starting with the dmap leaf that covers the hint, we'll
1245 * check the next four contiguous leaves for sufficient free
1246 * space. if sufficient free space is found, we'll allocate
1247 * the desired free space.
1248 *
1249 * PARAMETERS:
1250 * bmp - pointer to bmap descriptor
1251 * dp - pointer to dmap.
1252 * blkno - block number to allocate near.
1253 * nblocks - actual number of contiguous free blocks desired.
1254 * l2nb - log2 number of contiguous free blocks desired.
1255 * results - on successful return, set to the starting block number
1256 * of the newly allocated range.
1257 *
1258 * RETURN VALUES:
1259 * 0 - success
1260 * -ENOSPC - insufficient disk resources
1261 * -EIO - i/o error
1262 *
1263 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1264 */
1265 static int
1268 {
1276 }
1280 /* determine the word within the dmap that holds the hint
1281 * (i.e. blkno). also, determine the last word in the dmap
1282 * that we'll include in our examination.
1283 */
1287 /* examine the leaves for sufficient free space.
1288 */
1290 /* does the leaf describe sufficient free space ?
1291 */
1295 /* determine the block number within the file system
1296 * of the first block described by this dmap word.
1297 */
1300 /* if not all bits of the dmap word are free, get the
1301 * starting bit number within the dmap word of the required
1302 * string of free bits and adjust the block number with the
1303 * value.
1304 */
1306 blkno +=
1309 /* allocate the blocks.
1310 */
1315 }
1318 }
1321 /*
1322 * NAME: dbAllocAG()
1323 *
1324 * FUNCTION: attempt to allocate the specified number of contiguous
1325 * free blocks within the specified allocation group.
1326 *
1327 * unless the allocation group size is equal to the number
1328 * of blocks per dmap, the dmap control pages will be used to
1329 * find the required free space, if available. we start the
1330 * search at the highest dmap control page level which
1331 * distinctly describes the allocation group's free space
1332 * (i.e. the highest level at which the allocation group's
1333 * free space is not mixed in with that of any other group).
1334 * in addition, we start the search within this level at a
1335 * height of the dmapctl dmtree at which the nodes distinctly
1336 * describe the allocation group's free space. at this height,
1337 * the allocation group's free space may be represented by 1
1338 * or two sub-trees, depending on the allocation group size.
1339 * we search the top nodes of these subtrees left to right for
1340 * sufficient free space. if sufficient free space is found,
1341 * the subtree is searched to find the leftmost leaf that
1342 * has free space. once we have made it to the leaf, we
1343 * move the search to the next lower level dmap control page
1344 * corresponding to this leaf. we continue down the dmap control
1345 * pages until we find the dmap that contains or starts the
1346 * sufficient free space and we allocate at this dmap.
1347 *
1348 * if the allocation group size is equal to the dmap size,
1349 * we'll start at the dmap corresponding to the allocation
1350 * group and attempt the allocation at this level.
1351 *
1352 * the dmap control page search is also not performed if the
1353 * allocation group is completely free and we go to the first
1354 * dmap of the allocation group to do the allocation. this is
1355 * done because the allocation group may be part (not the first
1356 * part) of a larger binary buddy system, causing the dmap
1357 * control pages to indicate no free space (NOFREE) within
1358 * the allocation group.
1359 *
1360 * PARAMETERS:
1361 * bmp - pointer to bmap descriptor
1362 * agno - allocation group number.
1363 * nblocks - actual number of contiguous free blocks desired.
1364 * l2nb - log2 number of contiguous free blocks desired.
1365 * results - on successful return, set to the starting block number
1366 * of the newly allocated range.
1367 *
1368 * RETURN VALUES:
1369 * 0 - success
1370 * -ENOSPC - insufficient disk resources
1371 * -EIO - i/o error
1372 *
1373 * note: IWRITE_LOCK(ipmap) held on entry/exit;
1374 */
1375 static int
1377 {
1384 /* allocation request should not be for more than the
1385 * allocation group size.
1386 */
1389 "dbAllocAG: allocation request is larger than the "
1392 }
1394 /* determine the starting block number of the allocation
1395 * group.
1396 */
1399 /* check if the allocation group size is the minimum allocation
1400 * group size or if the allocation group is completely free. if
1401 * the allocation group size is the minimum size of BPERDMAP (i.e.
1402 * 1 dmap), there is no need to search the dmap control page (below)
1403 * that fully describes the allocation group since the allocation
1404 * group is already fully described by a dmap. in this case, we
1405 * just call dbAllocCtl() to search the dmap tree and allocate the
1406 * required space if available.
1407 *
1408 * if the allocation group is completely free, dbAllocCtl() is
1409 * also called to allocate the required space. this is done for
1410 * two reasons. first, it makes no sense searching the dmap control
1411 * pages for free space when we know that free space exists. second,
1412 * the dmap control pages may indicate that the allocation group
1413 * has no free space if the allocation group is part (not the first
1414 * part) of a larger binary buddy system.
1415 */
1426 }
1428 }
1430 /* the buffer for the dmap control page that fully describes the
1431 * allocation group.
1432 */
1445 }
1447 /* search the subtree(s) of the dmap control page that describes
1448 * the allocation group, looking for sufficient free space. to begin,
1449 * determine how many allocation groups are represented in a dmap
1450 * control page at the control page level (i.e. L0, L1, L2) that
1451 * fully describes an allocation group. next, determine the starting
1452 * tree index of this allocation group within the control page.
1453 */
1454 agperlev =
1458 /* dmap control page trees fan-out by 4 and a single allocation
1459 * group may be described by 1 or 2 subtrees within the ag level
1460 * dmap control page, depending upon the ag size. examine the ag's
1461 * subtrees for sufficient free space, starting with the leftmost
1462 * subtree.
1463 */
1465 /* is there sufficient free space ?
1466 */
1470 /* sufficient free space found in a subtree. now search down
1471 * the subtree to find the leftmost leaf that describes this
1472 * free space.
1473 */
1479 }
1480 }
1486 }
1487 }
1489 /* determine the block number within the file system
1490 * that corresponds to this leaf.
1491 */
1499 blkno +=
1502 /* release the buffer in preparation for going down
1503 * the next level of dmap control pages.
1504 */
1507 /* check if we need to continue to search down the lower
1508 * level dmap control pages. we need to if the number of
1509 * blocks required is less than maximum number of blocks
1510 * described at the next lower level.
1511 */
1514 /* search the lower level dmap control pages to get
1515 * the starting block number of the the dmap that
1516 * contains or starts off the free space.
1517 */
1523 "dbAllocAG: control page "
1526 }
1528 }
1529 }
1531 /* allocate the blocks.
1532 */
1538 }
1540 }
1542 /* no space in the allocation group. release the buffer and
1543 * return -ENOSPC.
1544 */
1548 }
1551 /*
1552 * NAME: dbAllocAny()
1553 *
1554 * FUNCTION: attempt to allocate the specified number of contiguous
1555 * free blocks anywhere in the file system.
1556 *
1557 * dbAllocAny() attempts to find the sufficient free space by
1558 * searching down the dmap control pages, starting with the
1559 * highest level (i.e. L0, L1, L2) control page. if free space
1560 * large enough to satisfy the desired free space is found, the
1561 * desired free space is allocated.
1562 *
1563 * PARAMETERS:
1564 * bmp - pointer to bmap descriptor
1565 * nblocks - actual number of contiguous free blocks desired.
1566 * l2nb - log2 number of contiguous free blocks desired.
1567 * results - on successful return, set to the starting block number
1568 * of the newly allocated range.
1569 *
1570 * RETURN VALUES:
1571 * 0 - success
1572 * -ENOSPC - insufficient disk resources
1573 * -EIO - i/o error
1574 *
1575 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1576 */
1578 {
1582 /* starting with the top level dmap control page, search
1583 * down the dmap control levels for sufficient free space.
1584 * if free space is found, dbFindCtl() returns the starting
1585 * block number of the dmap that contains or starts off the
1586 * range of free space.
1587 */
1591 /* allocate the blocks.
1592 */
1598 }
1600 }
1603 /*
1604 * NAME: dbFindCtl()
1605 *
1606 * FUNCTION: starting at a specified dmap control page level and block
1607 * number, search down the dmap control levels for a range of
1608 * contiguous free blocks large enough to satisfy an allocation
1609 * request for the specified number of free blocks.
1610 *
1611 * if sufficient contiguous free blocks are found, this routine
1612 * returns the starting block number within a dmap page that
1613 * contains or starts a range of contiqious free blocks that
1614 * is sufficient in size.
1615 *
1616 * PARAMETERS:
1617 * bmp - pointer to bmap descriptor
1618 * level - starting dmap control page level.
1619 * l2nb - log2 number of contiguous free blocks desired.
1620 * *blkno - on entry, starting block number for conducting the search.
1621 * on successful return, the first block within a dmap page
1622 * that contains or starts a range of contiguous free blocks.
1623 *
1624 * RETURN VALUES:
1625 * 0 - success
1626 * -ENOSPC - insufficient disk resources
1627 * -EIO - i/o error
1628 *
1629 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1630 */
1632 {
1639 /* starting at the specified dmap control page level and block
1640 * number, search down the dmap control levels for the starting
1641 * block number of a dmap page that contains or starts off
1642 * sufficient free blocks.
1643 */
1645 /* get the buffer of the dmap control page for the block
1646 * number and level (i.e. L0, L1, L2).
1647 */
1660 }
1662 /* search the tree within the dmap control page for
1663 * sufficent free space. if sufficient free space is found,
1664 * dbFindLeaf() returns the index of the leaf at which
1665 * free space was found.
1666 */
1669 /* release the buffer.
1670 */
1673 /* space found ?
1674 */
1680 }
1682 }
1684 /* adjust the block number to reflect the location within
1685 * the dmap control page (i.e. the leaf) at which free
1686 * space was found.
1687 */
1690 /* we stop the search at this dmap control page level if
1691 * the number of blocks required is greater than or equal
1692 * to the maximum number of blocks described at the next
1693 * (lower) level.
1694 */
1697 }
1701 }
1704 /*
1705 * NAME: dbAllocCtl()
1706 *
1707 * FUNCTION: attempt to allocate a specified number of contiguous
1708 * blocks starting within a specific dmap.
1709 *
1710 * this routine is called by higher level routines that search
1711 * the dmap control pages above the actual dmaps for contiguous
1712 * free space. the result of successful searches by these
1713 * routines are the starting block numbers within dmaps, with
1714 * the dmaps themselves containing the desired contiguous free
1715 * space or starting a contiguous free space of desired size
1716 * that is made up of the blocks of one or more dmaps. these
1717 * calls should not fail due to insufficent resources.
1718 *
1719 * this routine is called in some cases where it is not known
1720 * whether it will fail due to insufficient resources. more
1721 * specifically, this occurs when allocating from an allocation
1722 * group whose size is equal to the number of blocks per dmap.
1723 * in this case, the dmap control pages are not examined prior
1724 * to calling this routine (to save pathlength) and the call
1725 * might fail.
1726 *
1727 * for a request size that fits within a dmap, this routine relies
1728 * upon the dmap's dmtree to find the requested contiguous free
1729 * space. for request sizes that are larger than a dmap, the
1730 * requested free space will start at the first block of the
1731 * first dmap (i.e. blkno).
1732 *
1733 * PARAMETERS:
1734 * bmp - pointer to bmap descriptor
1735 * nblocks - actual number of contiguous free blocks to allocate.
1736 * l2nb - log2 number of contiguous free blocks to allocate.
1737 * blkno - starting block number of the dmap to start the allocation
1738 * from.
1739 * results - on successful return, set to the starting block number
1740 * of the newly allocated range.
1741 *
1742 * RETURN VALUES:
1743 * 0 - success
1744 * -ENOSPC - insufficient disk resources
1745 * -EIO - i/o error
1746 *
1747 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1748 */
1749 static int
1751 {
1757 /* check if the allocation request is confined to a single dmap.
1758 */
1760 /* get the buffer for the dmap.
1761 */
1768 /* try to allocate the blocks.
1769 */
1777 }
1779 /* allocation request involving multiple dmaps. it must start on
1780 * a dmap boundary.
1781 */
1784 /* allocate the blocks dmap by dmap.
1785 */
1787 /* get the buffer for the dmap.
1788 */
1794 }
1797 /* the dmap better be all free.
1798 */
1805 }
1807 /* determine how many blocks to allocate from this dmap.
1808 */
1811 /* allocate the blocks from the dmap.
1812 */
1816 }
1818 /* write the buffer.
1819 */
1821 }
1823 /* set the results (starting block number) and return.
1824 */
1828 /* something failed in handling an allocation request involving
1829 * multiple dmaps. we'll try to clean up by backing out any
1830 * allocation that has already happened for this request. if
1831 * we fail in backing out the allocation, we'll mark the file
1832 * system to indicate that blocks have been leaked.
1833 */
1834 backout:
1836 /* try to backout the allocations dmap by dmap.
1837 */
1840 /* get the buffer for this dmap.
1841 */
1845 /* could not back out. mark the file system
1846 * to indicate that we have leaked blocks.
1847 */
1851 }
1854 /* free the blocks is this dmap.
1855 */
1857 /* could not back out. mark the file system
1858 * to indicate that we have leaked blocks.
1859 */
1864 }
1866 /* write the buffer.
1867 */
1869 }
1872 }
1875 /*
1876 * NAME: dbAllocDmapLev()
1877 *
1878 * FUNCTION: attempt to allocate a specified number of contiguous blocks
1879 * from a specified dmap.
1880 *
1881 * this routine checks if the contiguous blocks are available.
1882 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1883 * returned.
1884 *
1885 * PARAMETERS:
1886 * mp - pointer to bmap descriptor
1887 * dp - pointer to dmap to attempt to allocate blocks from.
1888 * l2nb - log2 number of contiguous block desired.
1889 * nblocks - actual number of contiguous block desired.
1890 * results - on successful return, set to the starting block number
1891 * of the newly allocated range.
1892 *
1893 * RETURN VALUES:
1894 * 0 - success
1895 * -ENOSPC - insufficient disk resources
1896 * -EIO - i/o error
1897 *
1898 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1899 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1900 */
1901 static int
1904 {
1905 s64 blkno;
1908 /* can't be more than a dmaps worth of blocks */
1911 /* search the tree within the dmap page for sufficient
1912 * free space. if sufficient free space is found, dbFindLeaf()
1913 * returns the index of the leaf at which free space was found.
1914 */
1918 /* determine the block number within the file system corresponding
1919 * to the leaf at which free space was found.
1920 */
1923 /* if not all bits of the dmap word are free, get the starting
1924 * bit number within the dmap word of the required string of free
1925 * bits and adjust the block number with this value.
1926 */
1930 /* allocate the blocks */
1935 }
1938 /*
1939 * NAME: dbAllocDmap()
1940 *
1941 * FUNCTION: adjust the disk allocation map to reflect the allocation
1942 * of a specified block range within a dmap.
1943 *
1944 * this routine allocates the specified blocks from the dmap
1945 * through a call to dbAllocBits(). if the allocation of the
1946 * block range causes the maximum string of free blocks within
1947 * the dmap to change (i.e. the value of the root of the dmap's
1948 * dmtree), this routine will cause this change to be reflected
1949 * up through the appropriate levels of the dmap control pages
1950 * by a call to dbAdjCtl() for the L0 dmap control page that
1951 * covers this dmap.
1952 *
1953 * PARAMETERS:
1954 * bmp - pointer to bmap descriptor
1955 * dp - pointer to dmap to allocate the block range from.
1956 * blkno - starting block number of the block to be allocated.
1957 * nblocks - number of blocks to be allocated.
1958 *
1959 * RETURN VALUES:
1960 * 0 - success
1961 * -EIO - i/o error
1962 *
1963 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
1964 */
1967 {
1968 s8 oldroot;
1971 /* save the current value of the root (i.e. maximum free string)
1972 * of the dmap tree.
1973 */
1976 /* allocate the specified (blocks) bits */
1979 /* if the root has not changed, done. */
1983 /* root changed. bubble the change up to the dmap control pages.
1984 * if the adjustment of the upper level control pages fails,
1985 * backout the bit allocation (thus making everything consistent).
1986 */
1991 }
1994 /*
1995 * NAME: dbFreeDmap()
1996 *
1997 * FUNCTION: adjust the disk allocation map to reflect the allocation
1998 * of a specified block range within a dmap.
1999 *
2000 * this routine frees the specified blocks from the dmap through
2001 * a call to dbFreeBits(). if the deallocation of the block range
2002 * causes the maximum string of free blocks within the dmap to
2003 * change (i.e. the value of the root of the dmap's dmtree), this
2004 * routine will cause this change to be reflected up through the
2005 * appropriate levels of the dmap control pages by a call to
2006 * dbAdjCtl() for the L0 dmap control page that covers this dmap.
2007 *
2008 * PARAMETERS:
2009 * bmp - pointer to bmap descriptor
2010 * dp - pointer to dmap to free the block range from.
2011 * blkno - starting block number of the block to be freed.
2012 * nblocks - number of blocks to be freed.
2013 *
2014 * RETURN VALUES:
2015 * 0 - success
2016 * -EIO - i/o error
2017 *
2018 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2019 */
2022 {
2023 s8 oldroot;
2026 /* save the current value of the root (i.e. maximum free string)
2027 * of the dmap tree.
2028 */
2031 /* free the specified (blocks) bits */
2034 /* if error or the root has not changed, done. */
2038 /* root changed. bubble the change up to the dmap control pages.
2039 * if the adjustment of the upper level control pages fails,
2040 * backout the deallocation.
2041 */
2045 /* as part of backing out the deallocation, we will have
2046 * to back split the dmap tree if the deallocation caused
2047 * the freed blocks to become part of a larger binary buddy
2048 * system.
2049 */
2054 }
2057 }
2060 /*
2061 * NAME: dbAllocBits()
2062 *
2063 * FUNCTION: allocate a specified block range from a dmap.
2064 *
2065 * this routine updates the dmap to reflect the working
2066 * state allocation of the specified block range. it directly
2067 * updates the bits of the working map and causes the adjustment
2068 * of the binary buddy system described by the dmap's dmtree
2069 * leaves to reflect the bits allocated. it also causes the
2070 * dmap's dmtree, as a whole, to reflect the allocated range.
2071 *
2072 * PARAMETERS:
2073 * bmp - pointer to bmap descriptor
2074 * dp - pointer to dmap to allocate bits from.
2075 * blkno - starting block number of the bits to be allocated.
2076 * nblocks - number of bits to be allocated.
2077 *
2078 * RETURN VALUES: none
2079 *
2080 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2081 */
2084 {
2090 /* pick up a pointer to the leaves of the dmap tree */
2093 /* determine the bit number and word within the dmap of the
2094 * starting block.
2095 */
2099 /* block range better be within the dmap */
2102 /* allocate the bits of the dmap's words corresponding to the block
2103 * range. not all bits of the first and last words may be contained
2104 * within the block range. if this is the case, we'll work against
2105 * those words (i.e. partial first and/or last) on an individual basis
2106 * (a single pass), allocating the bits of interest by hand and
2107 * updating the leaf corresponding to the dmap word. a single pass
2108 * will be used for all dmap words fully contained within the
2109 * specified range. within this pass, the bits of all fully contained
2110 * dmap words will be marked as free in a single shot and the leaves
2111 * will be updated. a single leaf may describe the free space of
2112 * multiple dmap words, so we may update only a subset of the actual
2113 * leaves corresponding to the dmap words of the block range.
2114 */
2116 /* determine the bit number within the word and
2117 * the number of bits within the word.
2118 */
2122 /* check if only part of a word is to be allocated.
2123 */
2125 /* allocate (set to 1) the appropriate bits within
2126 * this dmap word.
2127 */
2131 /* update the leaf for this dmap word. in addition
2132 * to setting the leaf value to the binary buddy max
2133 * of the updated dmap word, dbSplit() will split
2134 * the binary system of the leaves if need be.
2135 */
2141 /* one or more dmap words are fully contained
2142 * within the block range. determine how many
2143 * words and allocate (set to 1) the bits of these
2144 * words.
2145 */
2149 /* determine how many bits.
2150 */
2153 /* now update the appropriate leaves to reflect
2154 * the allocated words.
2155 */
2159 "dbAllocBits: leaf page "
2162 }
2164 /* determine what the leaf value should be
2165 * updated to as the minimum of the l2 number
2166 * of bits being allocated and the l2 number
2167 * of bits currently described by this leaf.
2168 */
2171 /* update the leaf to reflect the allocation.
2172 * in addition to setting the leaf value to
2173 * NOFREE, dbSplit() will split the binary
2174 * system of the leaves to reflect the current
2175 * allocation (size).
2176 */
2179 /* get the number of dmap words handled */
2182 }
2183 }
2184 }
2186 /* update the free count for this dmap */
2191 /* if this allocation group is completely free,
2192 * update the maximum allocation group number if this allocation
2193 * group is the new max.
2194 */
2199 /* update the free count for the allocation group and map */
2204 }
2207 /*
2208 * NAME: dbFreeBits()
2209 *
2210 * FUNCTION: free a specified block range from a dmap.
2211 *
2212 * this routine updates the dmap to reflect the working
2213 * state allocation of the specified block range. it directly
2214 * updates the bits of the working map and causes the adjustment
2215 * of the binary buddy system described by the dmap's dmtree
2216 * leaves to reflect the bits freed. it also causes the dmap's
2217 * dmtree, as a whole, to reflect the deallocated range.
2218 *
2219 * PARAMETERS:
2220 * bmp - pointer to bmap descriptor
2221 * dp - pointer to dmap to free bits from.
2222 * blkno - starting block number of the bits to be freed.
2223 * nblocks - number of bits to be freed.
2224 *
2225 * RETURN VALUES: 0 for success
2226 *
2227 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2228 */
2231 {
2237 /* determine the bit number and word within the dmap of the
2238 * starting block.
2239 */
2243 /* block range better be within the dmap.
2244 */
2247 /* free the bits of the dmaps words corresponding to the block range.
2248 * not all bits of the first and last words may be contained within
2249 * the block range. if this is the case, we'll work against those
2250 * words (i.e. partial first and/or last) on an individual basis
2251 * (a single pass), freeing the bits of interest by hand and updating
2252 * the leaf corresponding to the dmap word. a single pass will be used
2253 * for all dmap words fully contained within the specified range.
2254 * within this pass, the bits of all fully contained dmap words will
2255 * be marked as free in a single shot and the leaves will be updated. a
2256 * single leaf may describe the free space of multiple dmap words,
2257 * so we may update only a subset of the actual leaves corresponding
2258 * to the dmap words of the block range.
2259 *
2260 * dbJoin() is used to update leaf values and will join the binary
2261 * buddy system of the leaves if the new leaf values indicate this
2262 * should be done.
2263 */
2265 /* determine the bit number within the word and
2266 * the number of bits within the word.
2267 */
2271 /* check if only part of a word is to be freed.
2272 */
2274 /* free (zero) the appropriate bits within this
2275 * dmap word.
2276 */
2281 /* update the leaf for this dmap word.
2282 */
2290 /* one or more dmap words are fully contained
2291 * within the block range. determine how many
2292 * words and free (zero) the bits of these words.
2293 */
2297 /* determine how many bits.
2298 */
2301 /* now update the appropriate leaves to reflect
2302 * the freed words.
2303 */
2305 /* determine what the leaf value should be
2306 * updated to as the minimum of the l2 number
2307 * of bits being freed and the l2 (max) number
2308 * of bits that can be described by this leaf.
2309 */
2310 size =
2315 /* update the leaf.
2316 */
2321 /* get the number of dmap words handled.
2322 */
2325 }
2326 }
2327 }
2329 /* update the free count for this dmap.
2330 */
2335 /* update the free count for the allocation group and
2336 * map.
2337 */
2342 /* check if this allocation group is not completely free and
2343 * if it is currently the maximum (rightmost) allocation group.
2344 * if so, establish the new maximum allocation group number by
2345 * searching left for the first allocation group with allocation.
2346 */
2355 }
2357 /* re-establish the allocation group preference if the
2358 * current preference is right of the maximum allocation
2359 * group.
2360 */
2363 }
2368 }
2371 /*
2372 * NAME: dbAdjCtl()
2373 *
2374 * FUNCTION: adjust a dmap control page at a specified level to reflect
2375 * the change in a lower level dmap or dmap control page's
2376 * maximum string of free blocks (i.e. a change in the root
2377 * of the lower level object's dmtree) due to the allocation
2378 * or deallocation of a range of blocks with a single dmap.
2379 *
2380 * on entry, this routine is provided with the new value of
2381 * the lower level dmap or dmap control page root and the
2382 * starting block number of the block range whose allocation
2383 * or deallocation resulted in the root change. this range
2384 * is respresented by a single leaf of the current dmapctl
2385 * and the leaf will be updated with this value, possibly
2386 * causing a binary buddy system within the leaves to be
2387 * split or joined. the update may also cause the dmapctl's
2388 * dmtree to be updated.
2389 *
2390 * if the adjustment of the dmap control page, itself, causes its
2391 * root to change, this change will be bubbled up to the next dmap
2392 * control level by a recursive call to this routine, specifying
2393 * the new root value and the next dmap control page level to
2394 * be adjusted.
2395 * PARAMETERS:
2396 * bmp - pointer to bmap descriptor
2397 * blkno - the first block of a block range within a dmap. it is
2398 * the allocation or deallocation of this block range that
2399 * requires the dmap control page to be adjusted.
2400 * newval - the new value of the lower level dmap or dmap control
2401 * page root.
2402 * alloc - TRUE if adjustment is due to an allocation.
2403 * level - current level of dmap control page (i.e. L0, L1, L2) to
2404 * be adjusted.
2405 *
2406 * RETURN VALUES:
2407 * 0 - success
2408 * -EIO - i/o error
2409 *
2410 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2411 */
2412 static int
2414 {
2416 s8 oldroot;
2418 s64 lblkno;
2422 /* get the buffer for the dmap control page for the specified
2423 * block number and control page level.
2424 */
2436 }
2438 /* determine the leaf number corresponding to the block and
2439 * the index within the dmap control tree.
2440 */
2444 /* save the current leaf value and the current root level (i.e.
2445 * maximum l2 free string described by this dmapctl).
2446 */
2450 /* check if this is a control page update for an allocation.
2451 * if so, update the leaf to reflect the new leaf value using
2452 * dbSplit(); otherwise (deallocation), use dbJoin() to udpate
2453 * the leaf with the new value. in addition to updating the
2454 * leaf, dbSplit() will also split the binary buddy system of
2455 * the leaves, if required, and bubble new values within the
2456 * dmapctl tree, if required. similarly, dbJoin() will join
2457 * the binary buddy system of leaves and bubble new values up
2458 * the dmapctl tree as required by the new leaf value.
2459 */
2461 /* check if we are in the middle of a binary buddy
2462 * system. this happens when we are performing the
2463 * first allocation out of an allocation group that
2464 * is part (not the first part) of a larger binary
2465 * buddy system. if we are in the middle, back split
2466 * the system prior to calling dbSplit() which assumes
2467 * that it is at the front of a binary buddy system.
2468 */
2472 }
2478 }
2480 /* check if the root of the current dmap control page changed due
2481 * to the update and if the current dmap control page is not at
2482 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e.
2483 * root changed and this is not the top level), call this routine
2484 * again (recursion) for the next higher level of the mapping to
2485 * reflect the change in root for the current dmap control page.
2486 */
2488 /* are we below the top level of the map. if so,
2489 * bubble the root up to the next higher level.
2490 */
2492 /* bubble up the new root of this dmap control page to
2493 * the next level.
2494 */
2498 /* something went wrong in bubbling up the new
2499 * root value, so backout the changes to the
2500 * current dmap control page.
2501 */
2504 oldval);
2506 /* the dbJoin() above might have
2507 * caused a larger binary buddy system
2508 * to form and we may now be in the
2509 * middle of it. if this is the case,
2510 * back split the buddies.
2511 */
2517 }
2519 /* release the buffer and return the error.
2520 */
2523 }
2525 /* we're at the top level of the map. update
2526 * the bmap control page to reflect the size
2527 * of the maximum free buddy system.
2528 */
2532 "dbAdjCtl: the maximum free buddy is "
2534 }
2536 }
2537 }
2539 /* write the buffer.
2540 */
2544 }
2547 /*
2548 * NAME: dbSplit()
2549 *
2550 * FUNCTION: update the leaf of a dmtree with a new value, splitting
2551 * the leaf from the binary buddy system of the dmtree's
2552 * leaves, as required.
2553 *
2554 * PARAMETERS:
2555 * tp - pointer to the tree containing the leaf.
2556 * leafno - the number of the leaf to be updated.
2557 * splitsz - the size the binary buddy system starting at the leaf
2558 * must be split to, specified as the log2 number of blocks.
2559 * newval - the new value for the leaf.
2560 *
2561 * RETURN VALUES: none
2562 *
2563 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2564 */
2566 {
2571 /* check if the leaf needs to be split.
2572 */
2574 /* the split occurs by cutting the buddy system in half
2575 * at the specified leaf until we reach the specified
2576 * size. pick up the starting split size (current size
2577 * - 1 in l2) and the corresponding buddy size.
2578 */
2582 /* split until we reach the specified size.
2583 */
2585 /* update the buddy's leaf with its new value.
2586 */
2589 /* on to the next size and buddy.
2590 */
2593 }
2594 }
2596 /* adjust the dmap tree to reflect the specified leaf's new
2597 * value.
2598 */
2600 }
2603 /*
2604 * NAME: dbBackSplit()
2605 *
2606 * FUNCTION: back split the binary buddy system of dmtree leaves
2607 * that hold a specified leaf until the specified leaf
2608 * starts its own binary buddy system.
2609 *
2610 * the allocators typically perform allocations at the start
2611 * of binary buddy systems and dbSplit() is used to accomplish
2612 * any required splits. in some cases, however, allocation
2613 * may occur in the middle of a binary system and requires a
2614 * back split, with the split proceeding out from the middle of
2615 * the system (less efficient) rather than the start of the
2616 * system (more efficient). the cases in which a back split
2617 * is required are rare and are limited to the first allocation
2618 * within an allocation group which is a part (not first part)
2619 * of a larger binary buddy system and a few exception cases
2620 * in which a previous join operation must be backed out.
2621 *
2622 * PARAMETERS:
2623 * tp - pointer to the tree containing the leaf.
2624 * leafno - the number of the leaf to be updated.
2625 *
2626 * RETURN VALUES: none
2627 *
2628 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2629 */
2631 {
2636 /* leaf should be part (not first part) of a binary
2637 * buddy system.
2638 */
2641 /* the back split is accomplished by iteratively finding the leaf
2642 * that starts the buddy system that contains the specified leaf and
2643 * splitting that system in two. this iteration continues until
2644 * the specified leaf becomes the start of a buddy system.
2645 *
2646 * determine maximum possible l2 size for the specified leaf.
2647 */
2648 size =
2652 /* determine the number of leaves covered by this size. this
2653 * is the buddy size that we will start with as we search for
2654 * the buddy system that contains the specified leaf.
2655 */
2658 /* back split.
2659 */
2661 /* find the leftmost buddy leaf.
2662 */
2667 /* determine the buddy.
2668 */
2671 /* check if this buddy is the start of the system.
2672 */
2674 /* split the leaf at the start of the
2675 * system in two.
2676 */
2680 }
2681 }
2682 }
2685 }
2688 /*
2689 * NAME: dbJoin()
2690 *
2691 * FUNCTION: update the leaf of a dmtree with a new value, joining
2692 * the leaf with other leaves of the dmtree into a multi-leaf
2693 * binary buddy system, as required.
2694 *
2695 * PARAMETERS:
2696 * tp - pointer to the tree containing the leaf.
2697 * leafno - the number of the leaf to be updated.
2698 * newval - the new value for the leaf.
2699 *
2700 * RETURN VALUES: none
2701 */
2703 {
2707 /* can the new leaf value require a join with other leaves ?
2708 */
2710 /* pickup a pointer to the leaves of the tree.
2711 */
2714 /* try to join the specified leaf into a large binary
2715 * buddy system. the join proceeds by attempting to join
2716 * the specified leafno with its buddy (leaf) at new value.
2717 * if the join occurs, we attempt to join the left leaf
2718 * of the joined buddies with its buddy at new value + 1.
2719 * we continue to join until we find a buddy that cannot be
2720 * joined (does not have a value equal to the size of the
2721 * last join) or until all leaves have been joined into a
2722 * single system.
2723 *
2724 * get the buddy size (number of words covered) of
2725 * the new value.
2726 */
2729 /* try to join.
2730 */
2732 /* get the buddy leaf.
2733 */
2736 /* if the leaf's new value is greater than its
2737 * buddy's value, we join no more.
2738 */
2742 /* It shouldn't be less */
2746 /* check which (leafno or buddy) is the left buddy.
2747 * the left buddy gets to claim the blocks resulting
2748 * from the join while the right gets to claim none.
2749 * the left buddy is also eligable to participate in
2750 * a join at the next higher level while the right
2751 * is not.
2752 *
2753 */
2755 /* leafno is the left buddy.
2756 */
2759 /* buddy is the left buddy and becomes
2760 * leafno.
2761 */
2764 }
2766 /* on to try the next join.
2767 */
2770 }
2771 }
2773 /* update the leaf value.
2774 */
2778 }
2781 /*
2782 * NAME: dbAdjTree()
2783 *
2784 * FUNCTION: update a leaf of a dmtree with a new value, adjusting
2785 * the dmtree, as required, to reflect the new leaf value.
2786 * the combination of any buddies must already be done before
2787 * this is called.
2788 *
2789 * PARAMETERS:
2790 * tp - pointer to the tree to be adjusted.
2791 * leafno - the number of the leaf to be updated.
2792 * newval - the new value for the leaf.
2793 *
2794 * RETURN VALUES: none
2795 */
2797 {
2801 /* pick up the index of the leaf for this leafno.
2802 */
2805 /* is the current value the same as the old value ? if so,
2806 * there is nothing to do.
2807 */
2811 /* set the new value.
2812 */
2815 /* bubble the new value up the tree as required.
2816 */
2818 /* get the index of the first leaf of the 4 leaf
2819 * group containing the specified leaf (leafno).
2820 */
2823 /* get the index of the parent of this 4 leaf group.
2824 */
2827 /* determine the maximum of the 4 leaves.
2828 */
2831 /* if the maximum of the 4 is the same as the
2832 * parent's value, we're done.
2833 */
2837 /* parent gets new value.
2838 */
2841 /* parent becomes leaf for next go-round.
2842 */
2844 }
2845 }
2848 /*
2849 * NAME: dbFindLeaf()
2850 *
2851 * FUNCTION: search a dmtree_t for sufficient free blocks, returning
2852 * the index of a leaf describing the free blocks if
2853 * sufficient free blocks are found.
2854 *
2855 * the search starts at the top of the dmtree_t tree and
2856 * proceeds down the tree to the leftmost leaf with sufficient
2857 * free space.
2858 *
2859 * PARAMETERS:
2860 * tp - pointer to the tree to be searched.
2861 * l2nb - log2 number of free blocks to search for.
2862 * leafidx - return pointer to be set to the index of the leaf
2863 * describing at least l2nb free blocks if sufficient
2864 * free blocks are found.
2865 *
2866 * RETURN VALUES:
2867 * 0 - success
2868 * -ENOSPC - insufficient free blocks.
2869 */
2871 {
2874 /* first check the root of the tree to see if there is
2875 * sufficient free space.
2876 */
2880 /* sufficient free space available. now search down the tree
2881 * starting at the next level for the leftmost leaf that
2882 * describes sufficient free space.
2883 */
2886 /* search the four nodes at this level, starting from
2887 * the left.
2888 */
2890 /* sufficient free space found. move to the next
2891 * level (or quit if this is the last level).
2892 */
2895 }
2897 /* better have found something since the higher
2898 * levels of the tree said it was here.
2899 */
2901 }
2903 /* set the return to the leftmost leaf describing sufficient
2904 * free space.
2905 */
2909 }
2912 /*
2913 * NAME: dbFindBits()
2914 *
2915 * FUNCTION: find a specified number of binary buddy free bits within a
2916 * dmap bitmap word value.
2917 *
2918 * this routine searches the bitmap value for (1 << l2nb) free
2919 * bits at (1 << l2nb) alignments within the value.
2920 *
2921 * PARAMETERS:
2922 * word - dmap bitmap word value.
2923 * l2nb - number of free bits specified as a log2 number.
2924 *
2925 * RETURN VALUES:
2926 * starting bit number of free bits.
2927 */
2929 {
2931 u32 mask;
2933 /* get the number of bits.
2934 */
2938 /* complement the word so we can use a mask (i.e. 0s represent
2939 * free bits) and compute the mask.
2940 */
2944 /* scan the word for nb free bits at nb alignments.
2945 */
2949 }
2953 /* return the bit number.
2954 */
2956 }
2959 /*
2960 * NAME: dbMaxBud(u8 *cp)
2961 *
2962 * FUNCTION: determine the largest binary buddy string of free
2963 * bits within 32-bits of the map.
2964 *
2965 * PARAMETERS:
2966 * cp - pointer to the 32-bit value.
2967 *
2968 * RETURN VALUES:
2969 * largest binary buddy of free bits within a dmap word.
2970 */
2972 {
2975 /* check if the wmap word is all free. if so, the
2976 * free buddy size is BUDMIN.
2977 */
2981 /* check if the wmap word is half free. if so, the
2982 * free buddy size is BUDMIN-1.
2983 */
2987 /* not all free or half free. determine the free buddy
2988 * size thru table lookup using quarters of the wmap word.
2989 */
2993 }
2996 /*
2997 * NAME: cnttz(uint word)
2998 *
2999 * FUNCTION: determine the number of trailing zeros within a 32-bit
3000 * value.
3001 *
3002 * PARAMETERS:
3003 * value - 32-bit value to be examined.
3004 *
3005 * RETURN VALUES:
3006 * count of trailing zeros
3007 */
3009 {
3015 }
3018 }
3021 /*
3022 * NAME: cntlz(u32 value)
3023 *
3024 * FUNCTION: determine the number of leading zeros within a 32-bit
3025 * value.
3026 *
3027 * PARAMETERS:
3028 * value - 32-bit value to be examined.
3029 *
3030 * RETURN VALUES:
3031 * count of leading zeros
3032 */
3034 {
3040 }
3042 }
3045 /*
3046 * NAME: blkstol2(s64 nb)
3047 *
3048 * FUNCTION: convert a block count to its log2 value. if the block
3049 * count is not a l2 multiple, it is rounded up to the next
3050 * larger l2 multiple.
3051 *
3052 * PARAMETERS:
3053 * nb - number of blocks
3054 *
3055 * RETURN VALUES:
3056 * log2 number of blocks
3057 */
3059 {
3065 /* count the leading bits.
3066 */
3068 /* leading bit found.
3069 */
3071 /* determine the l2 value.
3072 */
3075 /* check if we need to round up.
3076 */
3078 l2nb++;
3081 }
3082 }
3085 }
3088 /*
3089 * NAME: dbAllocBottomUp()
3090 *
3091 * FUNCTION: alloc the specified block range from the working block
3092 * allocation map.
3093 *
3094 * the blocks will be alloc from the working map one dmap
3095 * at a time.
3096 *
3097 * PARAMETERS:
3098 * ip - pointer to in-core inode;
3099 * blkno - starting block number to be freed.
3100 * nblocks - number of blocks to be freed.
3101 *
3102 * RETURN VALUES:
3103 * 0 - success
3104 * -EIO - i/o error
3105 */
3107 {
3117 /* block to be allocated better be within the mapsize. */
3120 /*
3121 * allocate the blocks a dmap at a time.
3122 */
3125 /* release previous dmap if any */
3128 }
3130 /* get the buffer for the current dmap. */
3136 }
3139 /* determine the number of blocks to be allocated from
3140 * this dmap.
3141 */
3144 /* allocate the blocks. */
3149 }
3150 }
3152 /* write the last buffer. */
3158 }
3163 {
3169 /* save the current value of the root (i.e. maximum free string)
3170 * of the dmap tree.
3171 */
3174 /* pick up a pointer to the leaves of the dmap tree */
3177 /* determine the bit number and word within the dmap of the
3178 * starting block.
3179 */
3183 /* block range better be within the dmap */
3186 /* allocate the bits of the dmap's words corresponding to the block
3187 * range. not all bits of the first and last words may be contained
3188 * within the block range. if this is the case, we'll work against
3189 * those words (i.e. partial first and/or last) on an individual basis
3190 * (a single pass), allocating the bits of interest by hand and
3191 * updating the leaf corresponding to the dmap word. a single pass
3192 * will be used for all dmap words fully contained within the
3193 * specified range. within this pass, the bits of all fully contained
3194 * dmap words will be marked as free in a single shot and the leaves
3195 * will be updated. a single leaf may describe the free space of
3196 * multiple dmap words, so we may update only a subset of the actual
3197 * leaves corresponding to the dmap words of the block range.
3198 */
3200 /* determine the bit number within the word and
3201 * the number of bits within the word.
3202 */
3206 /* check if only part of a word is to be allocated.
3207 */
3209 /* allocate (set to 1) the appropriate bits within
3210 * this dmap word.
3211 */
3215 word++;
3217 /* one or more dmap words are fully contained
3218 * within the block range. determine how many
3219 * words and allocate (set to 1) the bits of these
3220 * words.
3221 */
3225 /* determine how many bits */
3228 }
3229 }
3231 /* update the free count for this dmap */
3234 /* reconstruct summary tree */
3239 /* if this allocation group is completely free,
3240 * update the highest active allocation group number
3241 * if this allocation group is the new max.
3242 */
3247 /* update the free count for the allocation group and map */
3253 /* if the root has not changed, done. */
3257 /* root changed. bubble the change up to the dmap control pages.
3258 * if the adjustment of the upper level control pages fails,
3259 * backout the bit allocation (thus making everything consistent).
3260 */
3265 }
3268 /*
3269 * NAME: dbExtendFS()
3270 *
3271 * FUNCTION: extend bmap from blkno for nblocks;
3272 * dbExtendFS() updates bmap ready for dbAllocBottomUp();
3273 *
3274 * L2
3275 * |
3276 * L1---------------------------------L1
3277 * | |
3278 * L0---------L0---------L0 L0---------L0---------L0
3279 * | | | | | |
3280 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm;
3281 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3282 *
3283 * <---old---><----------------------------extend----------------------->
3284 */
3286 {
3290 s64 newsize;
3291 s64 p;
3298 s64 ag_rem;
3305 /*
3306 * initialize bmap control page.
3307 *
3308 * all the data in bmap control page should exclude
3309 * the mkfs hidden dmap page.
3310 */
3312 /* update mapsize */
3316 /* compute new AG size */
3323 /* compute new number of AG */
3328 /*
3329 * reconfigure db_agfree[]
3330 * from old AG configuration to new AG configuration;
3331 *
3332 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3333 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3334 * note: new AG size = old AG size * (2**x).
3335 */
3343 /* coalesce cotiguous k AGs; */
3345 /* merge AGi to AGn */
3347 }
3348 }
3354 /*
3355 * update highest active ag number
3356 */
3360 /*
3361 * extend bmap
3362 *
3363 * update bit maps and corresponding level control pages;
3364 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3365 */
3366 extend:
3367 /* get L2 page */
3373 }
3376 /* compute start L1 */
3381 /*
3382 * extend each L1 in L2
3383 */
3385 /* get L1 page */
3387 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3393 /* compute start L0 */
3399 /* assign/init L1 page */
3406 /* compute start L0 */
3410 }
3412 /*
3413 * extend each L0 in L1
3414 */
3416 /* get L0 page */
3418 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3425 /* compute start dmap */
3427 L2BPERDMAP;
3433 /* assign/init L0 page */
3440 /* compute start dmap */
3444 }
3446 /*
3447 * extend each dmap in L0
3448 */
3450 /*
3451 * reconstruct the dmap page, and
3452 * initialize corresponding parent L0 leaf
3453 */
3455 /* read in dmap page: */
3462 /* assign/init dmap page */
3469 }
3480 l0leaf++;
3489 /*
3490 * build current L0 page from its leaves, and
3491 * initialize corresponding parent L1 leaf
3492 */
3500 /* more than 1 L0 ? */
3504 /* summarize in global bmap page */
3509 }
3510 }
3513 /*
3514 * build current L1 page from its leaves, and
3515 * initialize corresponding parent L2 leaf
3516 */
3524 /* more than 1 L1 ? */
3528 /* summarize in global bmap page */
3532 }
3533 }
3538 errout:
3546 /*
3547 * finalize bmap control page
3548 */
3549 finalize:
3552 }
3555 /*
3556 * dbFinalizeBmap()
3557 */
3559 {
3565 /*
3566 * finalize bmap control page
3567 */
3568 //finalize:
3569 /*
3570 * compute db_agpref: preferred ag to allocate from
3571 * (the leftmost ag with average free space in it);
3572 */
3573 //agpref:
3574 /* get the number of active ags and inacitve ags */
3579 /* determine how many blocks are in the inactive allocation
3580 * groups. in doing this, we must account for the fact that
3581 * the rightmost group might be a partial group (i.e. file
3582 * system size is not a multiple of the group size).
3583 */
3588 /* determine how many free blocks are in the active
3589 * allocation groups plus the average number of free blocks
3590 * within the active ags.
3591 */
3595 /* if the preferred allocation group has not average free space.
3596 * re-establish the preferred group as the leftmost
3597 * group with average free space.
3598 */
3604 }
3608 }
3609 }
3611 /*
3612 * compute db_aglevel, db_agheigth, db_width, db_agstart:
3613 * an ag is covered in aglevel dmapctl summary tree,
3614 * at agheight level height (from leaf) with agwidth number of nodes
3615 * each, which starts at agstart index node of the smmary tree node
3616 * array;
3617 */
3619 l2nl =
3624 i--) {
3627 }
3629 }
3632 /*
3633 * NAME: dbInitDmap()/ujfs_idmap_page()
3634 *
3635 * FUNCTION: initialize working/persistent bitmap of the dmap page
3636 * for the specified number of blocks:
3637 *
3638 * at entry, the bitmaps had been initialized as free (ZEROS);
3639 * The number of blocks will only account for the actually
3640 * existing blocks. Blocks which don't actually exist in
3641 * the aggregate will be marked as allocated (ONES);
3642 *
3643 * PARAMETERS:
3644 * dp - pointer to page of map
3645 * nblocks - number of blocks this page
3646 *
3647 * RETURNS: NONE
3648 */
3650 {
3653 /* starting block number within the dmap */
3664 }
3669 }
3671 /* word number containing start block number */
3674 /*
3675 * free the bits corresponding to the block range (ZEROS):
3676 * note: not all bits of the first and last words may be contained
3677 * within the block range.
3678 */
3680 /* number of bits preceding range to be freed in the word */
3682 /* number of bits to free in the word */
3685 /* is partial word to be freed ? */
3687 /* free (set to 0) from the bitmap word */
3693 /* skip the word freed */
3694 w++;
3696 /* free (set to 0) contiguous bitmap words */
3701 /* skip the words freed */
3704 }
3705 }
3707 /*
3708 * mark bits following the range to be freed (non-existing
3709 * blocks) as allocated (ONES)
3710 */
3715 /* the first word beyond the end of existing blocks */
3718 /* does nblocks fall on a 32-bit boundary ? */
3721 /* mark a partial word allocated */
3723 w++;
3724 }
3726 /* set the rest of the words in the page to allocated (ONES) */
3730 /*
3731 * init tree
3732 */
3733 initTree:
3735 }
3738 /*
3739 * NAME: dbInitDmapTree()/ujfs_complete_dmap()
3740 *
3741 * FUNCTION: initialize summary tree of the specified dmap:
3742 *
3743 * at entry, bitmap of the dmap has been initialized;
3744 *
3745 * PARAMETERS:
3746 * dp - dmap to complete
3747 * blkno - starting block number for this dmap
3748 * treemax - will be filled in with max free for this dmap
3749 *
3750 * RETURNS: max free string at the root of the tree
3751 */
3753 {
3758 /* init fixed info of tree */
3766 /* init each leaf from corresponding wmap word:
3767 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3768 * bitmap word are allocated.
3769 */
3774 /* build the dmap's binary buddy summary tree */
3776 }
3779 /*
3780 * NAME: dbInitTree()/ujfs_adjtree()
3781 *
3782 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl.
3783 *
3784 * at entry, the leaves of the tree has been initialized
3785 * from corresponding bitmap word or root of summary tree
3786 * of the child control page;
3787 * configure binary buddy system at the leaf level, then
3788 * bubble up the values of the leaf nodes up the tree.
3789 *
3790 * PARAMETERS:
3791 * cp - Pointer to the root of the tree
3792 * l2leaves- Number of leaf nodes as a power of 2
3793 * l2min - Number of blocks that can be covered by a leaf
3794 * as a power of 2
3795 *
3796 * RETURNS: max free string at the root of the tree
3797 */
3799 {
3806 /* Determine the maximum free string possible for the leaves */
3809 /*
3810 * configure the leaf levevl into binary buddy system
3811 *
3812 * Try to combine buddies starting with a buddy size of 1
3813 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3814 * can be combined if both buddies have a maximum free of l2min;
3815 * the combination will result in the left-most buddy leaf having
3816 * a maximum free of l2min+1.
3817 * After processing all buddies for a given size, process buddies
3818 * at the next higher buddy size (i.e. current size * 2) and
3819 * the next maximum free (current free + 1).
3820 * This continues until the maximum possible buddy combination
3821 * yields maximum free.
3822 */
3825 /* get next buddy size == current buddy pair size */
3828 /* scan each adjacent buddy pair at current buddy size */
3832 /* coalesce if both adjacent buddies are max free */
3836 }
3837 }
3838 }
3840 /*
3841 * bubble summary information of leaves up the tree.
3842 *
3843 * Starting at the leaf node level, the four nodes described by
3844 * the higher level parent node are compared for a maximum free and
3845 * this maximum becomes the value of the parent node.
3846 * when all lower level nodes are processed in this fashion then
3847 * move up to the next level (parent becomes a lower level node) and
3848 * continue the process for that level.
3849 */
3853 /* get index of 1st node of parent level */
3856 /* set the value of the parent node as the maximum
3857 * of the four nodes of the current level.
3858 */
3862 }
3865 }
3868 /*
3869 * dbInitDmapCtl()
3870 *
3871 * function: initialize dmapctl page
3872 */
3883 /*
3884 * initialize the leaves of current level that were not covered
3885 * by the specified input block range (i.e. the leaves have no
3886 * low level dmapctl or dmap).
3887 */
3892 /* build the dmap's binary buddy summary tree */
3894 }
3897 /*
3898 * NAME: dbGetL2AGSize()/ujfs_getagl2size()
3899 *
3900 * FUNCTION: Determine log2(allocation group size) from aggregate size
3901 *
3902 * PARAMETERS:
3903 * nblocks - Number of blocks in aggregate
3904 *
3905 * RETURNS: log2(allocation group size) in aggregate blocks
3906 */
3908 {
3909 s64 sz;
3910 s64 m;
3916 /* round up aggregate size to power of 2 */
3921 }
3927 /* agsize = roundupSize/max_number_of_ag */
3929 }
3932 /*
3933 * NAME: dbMapFileSizeToMapSize()
3934 *
3935 * FUNCTION: compute number of blocks the block allocation map file
3936 * can cover from the map file size;
3937 *
3938 * RETURNS: Number of blocks which can be covered by this block map file;
3939 */
3941 /*
3942 * maximum number of map pages at each level including control pages
3943 */
3944 #define MAXL0PAGES (1 + LPERCTL)
3945 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES)
3946 #define MAXL2PAGES (1 + LPERCTL * MAXL1PAGES)
3948 /*
3949 * convert number of map pages to the zero origin top dmapctl level
3950 */
3951 #define BMAPPGTOLEV(npages) \
3952 (((npages) <= 3 + MAXL0PAGES) ? 0 \
3953 : ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
3956 {
3958 s64 nblocks;
3967 /* At each level, accumulate the number of dmap pages covered by
3968 * the number of full child levels below it;
3969 * repeat for the last incomplete child level.
3970 */
3973 /* skip higher level control pages above top level covered by map */
3977 factor =
3983 /* pages in last/incomplete child */
3985 /* skip incomplete child's level control page */
3986 npages--;
3987 }
3989 /* convert the number of dmaps into the number of blocks
3990 * which can be covered by the dmaps;
3991 */
3995 }