| blob | c161c98954e08b17afb7e849ef57c7b47bbd1b6f |
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) mutex_init(&bmp->db_bmaplock)
68 #define BMAP_LOCK(bmp) mutex_lock(&bmp->db_bmaplock)
69 #define BMAP_UNLOCK(bmp) mutex_unlock(&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 */
310 }
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 {
342 /* block to be freed better be within the mapsize. */
351 }
353 /*
354 * free the blocks a dmap at a time.
355 */
358 /* release previous dmap if any */
361 }
363 /* get the buffer for the current dmap. */
369 }
372 /* determine the number of blocks to be freed from
373 * this dmap.
374 */
377 /* free the blocks. */
383 }
384 }
386 /* write the last buffer. */
392 }
395 /*
396 * NAME: dbUpdatePMap()
397 *
398 * FUNCTION: update the allocation state (free or allocate) of the
399 * specified block range in the persistent block allocation map.
400 *
401 * the blocks will be updated in the persistent map one
402 * dmap at a time.
403 *
404 * PARAMETERS:
405 * ipbmap - pointer to in-core inode for the block map.
406 * free - TRUE if block range is to be freed from the persistent
407 * map; FALSE if it is to be allocated.
408 * blkno - starting block number of the range.
409 * nblocks - number of contiguous blocks in the range.
410 * tblk - transaction block;
411 *
412 * RETURN VALUES:
413 * 0 - success
414 * -EIO - i/o error
415 */
416 int
419 {
424 u32 mask;
431 /* the blocks better be within the mapsize. */
439 }
441 /* compute delta of transaction lsn from log syncpt */
446 /*
447 * update the block state a dmap at a time.
448 */
452 /* get the buffer for the current dmap. */
457 }
464 }
467 /* determine the bit number and word within the dmap of
468 * the starting block. also determine how many blocks
469 * are to be updated within this dmap.
470 */
475 /* update the bits of the dmap words. the first and last
476 * words may only have a subset of their bits updated. if
477 * this is the case, we'll work against that word (i.e.
478 * partial first and/or last) only in a single pass. a
479 * single pass will also be used to update all words that
480 * are to have all their bits updated.
481 */
484 /* determine the bit number within the word and
485 * the number of bits within the word.
486 */
490 /* check if only part of the word is to be updated. */
492 /* update (free or allocate) the bits
493 * in this word.
494 */
495 mask =
500 else
506 /* one or more words are to have all
507 * their bits updated. determine how
508 * many words and how many bits.
509 */
513 /* update (free or allocate) the bits
514 * in these words.
515 */
519 else
524 }
525 }
527 /*
528 * update dmap lsn
529 */
537 /* inherit older/smaller lsn */
542 /* move bp after tblock in logsync list */
544 }
546 /* inherit younger/larger clsn */
555 /* insert bp after tblock in logsync list */
560 }
562 }
564 /* write the last buffer. */
567 }
570 }
573 /*
574 * NAME: dbNextAG()
575 *
576 * FUNCTION: find the preferred allocation group for new allocations.
577 *
578 * Within the allocation groups, we maintain a preferred
579 * allocation group which consists of a group with at least
580 * average free space. It is the preferred group that we target
581 * new inode allocation towards. The tie-in between inode
582 * allocation and block allocation occurs as we allocate the
583 * first (data) block of an inode and specify the inode (block)
584 * as the allocation hint for this block.
585 *
586 * We try to avoid having more than one open file growing in
587 * an allocation group, as this will lead to fragmentation.
588 * This differs from the old OS/2 method of trying to keep
589 * empty ags around for large allocations.
590 *
591 * PARAMETERS:
592 * ipbmap - pointer to in-core inode for the block map.
593 *
594 * RETURN VALUES:
595 * the preferred allocation group number.
596 */
598 {
599 s64 avgfree;
608 /* determine the average number of free blocks within the ags. */
611 /*
612 * if the current preferred ag does not have an active allocator
613 * and has at least average freespace, return it
614 */
620 /* From the last preferred ag, find the next one with at least
621 * average free space.
622 */
628 /* open file is currently growing in this ag */
631 /* Return this one */
635 /* Less than avg. freespace, but best so far */
638 }
639 }
641 /*
642 * If no inactive ag was found with average freespace, use the
643 * next best
644 */
647 /* else leave db_agpref unchanged */
648 unlock:
651 /* return the preferred group.
652 */
654 }
656 /*
657 * NAME: dbAlloc()
658 *
659 * FUNCTION: attempt to allocate a specified number of contiguous free
660 * blocks from the working allocation block map.
661 *
662 * the block allocation policy uses hints and a multi-step
663 * approach.
664 *
665 * for allocation requests smaller than the number of blocks
666 * per dmap, we first try to allocate the new blocks
667 * immediately following the hint. if these blocks are not
668 * available, we try to allocate blocks near the hint. if
669 * no blocks near the hint are available, we next try to
670 * allocate within the same dmap as contains the hint.
671 *
672 * if no blocks are available in the dmap or the allocation
673 * request is larger than the dmap size, we try to allocate
674 * within the same allocation group as contains the hint. if
675 * this does not succeed, we finally try to allocate anywhere
676 * within the aggregate.
677 *
678 * we also try to allocate anywhere within the aggregate for
679 * for allocation requests larger than the allocation group
680 * size or requests that specify no hint value.
681 *
682 * PARAMETERS:
683 * ip - pointer to in-core inode;
684 * hint - allocation hint.
685 * nblocks - number of contiguous blocks in the range.
686 * results - on successful return, set to the starting block number
687 * of the newly allocated contiguous range.
688 *
689 * RETURN VALUES:
690 * 0 - success
691 * -ENOSPC - insufficient disk resources
692 * -EIO - i/o error
693 */
695 {
703 s64 mapSize;
706 /* assert that nblocks is valid */
709 #ifdef _STILL_TO_PORT
710 /* DASD limit check F226941 */
715 /* get the log2 number of blocks to be allocated.
716 * if the number of blocks is not a log2 multiple,
717 * it will be rounded up to the next log2 multiple.
718 */
723 //retry: /* serialize w.r.t.extendfs() */
726 /* the hint should be within the map */
730 }
732 /* if the number of blocks to be allocated is greater than the
733 * allocation group size, try to allocate anywhere.
734 */
741 }
743 /*
744 * If no hint, let dbNextAG recommend an allocation group
745 */
749 /* we would like to allocate close to the hint. adjust the
750 * hint to the block following the hint since the allocators
751 * will start looking for free space starting at this point.
752 */
760 /* check if blkno crosses over into a new allocation group.
761 * if so, check if we should allow allocations within this
762 * allocation group.
763 */
765 /* check if the AG is currenly being written to.
766 * if so, call dbNextAG() to find a non-busy
767 * AG with sufficient free space.
768 */
772 /* check if the allocation request size can be satisfied from a
773 * single dmap. if so, try to allocate from the dmap containing
774 * the hint using a tiered strategy.
775 */
779 /* get the buffer for the dmap containing the hint.
780 */
789 /* first, try to satisfy the allocation request with the
790 * blocks beginning at the hint.
791 */
797 }
801 }
806 /*
807 * Someone else is writing in this allocation
808 * group. To avoid fragmenting, try another ag
809 */
813 }
815 /* next, try to satisfy the allocation request with blocks
816 * near the hint.
817 */
826 }
828 /* try to satisfy the allocation request with blocks within
829 * the same dmap as the hint.
830 */
838 }
842 }
844 /* try to satisfy the allocation request with blocks within
845 * the same allocation group as the hint.
846 */
854 pref_ag:
855 /*
856 * Let dbNextAG recommend a preferred allocation group
857 */
861 /* Try to allocate within this allocation group. if that fails, try to
862 * allocate anywhere in the map.
863 */
867 write_unlock:
872 read_unlock:
876 }
878 #ifdef _NOTYET
879 /*
880 * NAME: dbAllocExact()
881 *
882 * FUNCTION: try to allocate the requested extent;
883 *
884 * PARAMETERS:
885 * ip - pointer to in-core inode;
886 * blkno - extent address;
887 * nblocks - extent length;
888 *
889 * RETURN VALUES:
890 * 0 - success
891 * -ENOSPC - insufficient disk resources
892 * -EIO - i/o error
893 */
895 {
900 s64 lblkno;
905 /*
906 * validate extent request:
907 *
908 * note: defragfs policy:
909 * max 64 blocks will be moved.
910 * allocation request size must be satisfied from a single dmap.
911 */
915 }
918 /* the free space is no longer available */
921 }
923 /* read in the dmap covering the extent */
929 }
932 /* try to allocate the requested extent */
943 }
946 /*
947 * NAME: dbReAlloc()
948 *
949 * FUNCTION: attempt to extend a current allocation by a specified
950 * number of blocks.
951 *
952 * this routine attempts to satisfy the allocation request
953 * by first trying to extend the existing allocation in
954 * place by allocating the additional blocks as the blocks
955 * immediately following the current allocation. if these
956 * blocks are not available, this routine will attempt to
957 * allocate a new set of contiguous blocks large enough
958 * to cover the existing allocation plus the additional
959 * number of blocks required.
960 *
961 * PARAMETERS:
962 * ip - pointer to in-core inode requiring allocation.
963 * blkno - starting block of the current allocation.
964 * nblocks - number of contiguous blocks within the current
965 * allocation.
966 * addnblocks - number of blocks to add to the allocation.
967 * results - on successful return, set to the starting block number
968 * of the existing allocation if the existing allocation
969 * was extended in place or to a newly allocated contiguous
970 * range if the existing allocation could not be extended
971 * in place.
972 *
973 * RETURN VALUES:
974 * 0 - success
975 * -ENOSPC - insufficient disk resources
976 * -EIO - i/o error
977 */
978 int
981 {
984 /* try to extend the allocation in place.
985 */
992 }
994 /* could not extend the allocation in place, so allocate a
995 * new set of blocks for the entire request (i.e. try to get
996 * a range of contiguous blocks large enough to cover the
997 * existing allocation plus the additional blocks.)
998 */
1001 }
1004 /*
1005 * NAME: dbExtend()
1006 *
1007 * FUNCTION: attempt to extend a current allocation by a specified
1008 * number of blocks.
1009 *
1010 * this routine attempts to satisfy the allocation request
1011 * by first trying to extend the existing allocation in
1012 * place by allocating the additional blocks as the blocks
1013 * immediately following the current allocation.
1014 *
1015 * PARAMETERS:
1016 * ip - pointer to in-core inode requiring allocation.
1017 * blkno - starting block of the current allocation.
1018 * nblocks - number of contiguous blocks within the current
1019 * allocation.
1020 * addnblocks - number of blocks to add to the allocation.
1021 *
1022 * RETURN VALUES:
1023 * 0 - success
1024 * -ENOSPC - insufficient disk resources
1025 * -EIO - i/o error
1026 */
1028 {
1031 uint rel_block;
1038 /*
1039 * We don't want a non-aligned extent to cross a page boundary
1040 */
1045 /* get the last block of the current allocation */
1048 /* determine the block number of the block following
1049 * the existing allocation.
1050 */
1055 /* better be within the file system */
1062 }
1064 /* we'll attempt to extend the current allocation in place by
1065 * allocating the additional blocks as the blocks immediately
1066 * following the current allocation. we only try to extend the
1067 * current allocation in place if the number of additional blocks
1068 * can fit into a dmap, the last block of the current allocation
1069 * is not the last block of the file system, and the start of the
1070 * inplace extension is not on an allocation group boundary.
1071 */
1076 }
1078 /* get the buffer for the dmap containing the first block
1079 * of the extension.
1080 */
1086 }
1090 /* try to allocate the blocks immediately following the
1091 * current allocation.
1092 */
1097 /* were we successful ? */
1100 else
1101 /* 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 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: dbFindCtl()
1600 *
1601 * FUNCTION: starting at a specified dmap control page level and block
1602 * number, search down the dmap control levels for a range of
1603 * contiguous free blocks large enough to satisfy an allocation
1604 * request for the specified number of free blocks.
1605 *
1606 * if sufficient contiguous free blocks are found, this routine
1607 * returns the starting block number within a dmap page that
1608 * contains or starts a range of contiqious free blocks that
1609 * is sufficient in size.
1610 *
1611 * PARAMETERS:
1612 * bmp - pointer to bmap descriptor
1613 * level - starting dmap control page level.
1614 * l2nb - log2 number of contiguous free blocks desired.
1615 * *blkno - on entry, starting block number for conducting the search.
1616 * on successful return, the first block within a dmap page
1617 * that contains or starts a range of contiguous free blocks.
1618 *
1619 * RETURN VALUES:
1620 * 0 - success
1621 * -ENOSPC - insufficient disk resources
1622 * -EIO - i/o error
1623 *
1624 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1625 */
1627 {
1634 /* starting at the specified dmap control page level and block
1635 * number, search down the dmap control levels for the starting
1636 * block number of a dmap page that contains or starts off
1637 * sufficient free blocks.
1638 */
1640 /* get the buffer of the dmap control page for the block
1641 * number and level (i.e. L0, L1, L2).
1642 */
1655 }
1657 /* search the tree within the dmap control page for
1658 * sufficent free space. if sufficient free space is found,
1659 * dbFindLeaf() returns the index of the leaf at which
1660 * free space was found.
1661 */
1664 /* release the buffer.
1665 */
1668 /* space found ?
1669 */
1675 }
1677 }
1679 /* adjust the block number to reflect the location within
1680 * the dmap control page (i.e. the leaf) at which free
1681 * space was found.
1682 */
1685 /* we stop the search at this dmap control page level if
1686 * the number of blocks required is greater than or equal
1687 * to the maximum number of blocks described at the next
1688 * (lower) level.
1689 */
1692 }
1696 }
1699 /*
1700 * NAME: dbAllocCtl()
1701 *
1702 * FUNCTION: attempt to allocate a specified number of contiguous
1703 * blocks starting within a specific dmap.
1704 *
1705 * this routine is called by higher level routines that search
1706 * the dmap control pages above the actual dmaps for contiguous
1707 * free space. the result of successful searches by these
1708 * routines are the starting block numbers within dmaps, with
1709 * the dmaps themselves containing the desired contiguous free
1710 * space or starting a contiguous free space of desired size
1711 * that is made up of the blocks of one or more dmaps. these
1712 * calls should not fail due to insufficent resources.
1713 *
1714 * this routine is called in some cases where it is not known
1715 * whether it will fail due to insufficient resources. more
1716 * specifically, this occurs when allocating from an allocation
1717 * group whose size is equal to the number of blocks per dmap.
1718 * in this case, the dmap control pages are not examined prior
1719 * to calling this routine (to save pathlength) and the call
1720 * might fail.
1721 *
1722 * for a request size that fits within a dmap, this routine relies
1723 * upon the dmap's dmtree to find the requested contiguous free
1724 * space. for request sizes that are larger than a dmap, the
1725 * requested free space will start at the first block of the
1726 * first dmap (i.e. blkno).
1727 *
1728 * PARAMETERS:
1729 * bmp - pointer to bmap descriptor
1730 * nblocks - actual number of contiguous free blocks to allocate.
1731 * l2nb - log2 number of contiguous free blocks to allocate.
1732 * blkno - starting block number of the dmap to start the allocation
1733 * from.
1734 * results - on successful return, set to the starting block number
1735 * of the newly allocated range.
1736 *
1737 * RETURN VALUES:
1738 * 0 - success
1739 * -ENOSPC - insufficient disk resources
1740 * -EIO - i/o error
1741 *
1742 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1743 */
1744 static int
1746 {
1752 /* check if the allocation request is confined to a single dmap.
1753 */
1755 /* get the buffer for the dmap.
1756 */
1763 /* try to allocate the blocks.
1764 */
1772 }
1774 /* allocation request involving multiple dmaps. it must start on
1775 * a dmap boundary.
1776 */
1779 /* allocate the blocks dmap by dmap.
1780 */
1782 /* get the buffer for the dmap.
1783 */
1789 }
1792 /* the dmap better be all free.
1793 */
1800 }
1802 /* determine how many blocks to allocate from this dmap.
1803 */
1806 /* allocate the blocks from the dmap.
1807 */
1811 }
1813 /* write the buffer.
1814 */
1816 }
1818 /* set the results (starting block number) and return.
1819 */
1823 /* something failed in handling an allocation request involving
1824 * multiple dmaps. we'll try to clean up by backing out any
1825 * allocation that has already happened for this request. if
1826 * we fail in backing out the allocation, we'll mark the file
1827 * system to indicate that blocks have been leaked.
1828 */
1829 backout:
1831 /* try to backout the allocations dmap by dmap.
1832 */
1835 /* get the buffer for this dmap.
1836 */
1840 /* could not back out. mark the file system
1841 * to indicate that we have leaked blocks.
1842 */
1846 }
1849 /* free the blocks is this dmap.
1850 */
1852 /* could not back out. mark the file system
1853 * to indicate that we have leaked blocks.
1854 */
1859 }
1861 /* write the buffer.
1862 */
1864 }
1867 }
1870 /*
1871 * NAME: dbAllocDmapLev()
1872 *
1873 * FUNCTION: attempt to allocate a specified number of contiguous blocks
1874 * from a specified dmap.
1875 *
1876 * this routine checks if the contiguous blocks are available.
1877 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1878 * returned.
1879 *
1880 * PARAMETERS:
1881 * mp - pointer to bmap descriptor
1882 * dp - pointer to dmap to attempt to allocate blocks from.
1883 * l2nb - log2 number of contiguous block desired.
1884 * nblocks - actual number of contiguous block desired.
1885 * results - on successful return, set to the starting block number
1886 * of the newly allocated range.
1887 *
1888 * RETURN VALUES:
1889 * 0 - success
1890 * -ENOSPC - insufficient disk resources
1891 * -EIO - i/o error
1892 *
1893 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1894 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1895 */
1896 static int
1899 {
1900 s64 blkno;
1903 /* can't be more than a dmaps worth of blocks */
1906 /* search the tree within the dmap page for sufficient
1907 * free space. if sufficient free space is found, dbFindLeaf()
1908 * returns the index of the leaf at which free space was found.
1909 */
1913 /* determine the block number within the file system corresponding
1914 * to the leaf at which free space was found.
1915 */
1918 /* if not all bits of the dmap word are free, get the starting
1919 * bit number within the dmap word of the required string of free
1920 * bits and adjust the block number with this value.
1921 */
1925 /* allocate the blocks */
1930 }
1933 /*
1934 * NAME: dbAllocDmap()
1935 *
1936 * FUNCTION: adjust the disk allocation map to reflect the allocation
1937 * of a specified block range within a dmap.
1938 *
1939 * this routine allocates the specified blocks from the dmap
1940 * through a call to dbAllocBits(). if the allocation of the
1941 * block range causes the maximum string of free blocks within
1942 * the dmap to change (i.e. the value of the root of the dmap's
1943 * dmtree), this routine will cause this change to be reflected
1944 * up through the appropriate levels of the dmap control pages
1945 * by a call to dbAdjCtl() for the L0 dmap control page that
1946 * covers this dmap.
1947 *
1948 * PARAMETERS:
1949 * bmp - pointer to bmap descriptor
1950 * dp - pointer to dmap to allocate the block range from.
1951 * blkno - starting block number of the block to be allocated.
1952 * nblocks - number of blocks to be allocated.
1953 *
1954 * RETURN VALUES:
1955 * 0 - success
1956 * -EIO - i/o error
1957 *
1958 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
1959 */
1962 {
1963 s8 oldroot;
1966 /* save the current value of the root (i.e. maximum free string)
1967 * of the dmap tree.
1968 */
1971 /* allocate the specified (blocks) bits */
1974 /* if the root has not changed, done. */
1978 /* root changed. bubble the change up to the dmap control pages.
1979 * if the adjustment of the upper level control pages fails,
1980 * backout the bit allocation (thus making everything consistent).
1981 */
1986 }
1989 /*
1990 * NAME: dbFreeDmap()
1991 *
1992 * FUNCTION: adjust the disk allocation map to reflect the allocation
1993 * of a specified block range within a dmap.
1994 *
1995 * this routine frees the specified blocks from the dmap through
1996 * a call to dbFreeBits(). if the deallocation of the block range
1997 * causes the maximum string of free blocks within the dmap to
1998 * change (i.e. the value of the root of the dmap's dmtree), this
1999 * routine will cause this change to be reflected up through the
2000 * appropriate levels of the dmap control pages by a call to
2001 * dbAdjCtl() for the L0 dmap control page that covers this dmap.
2002 *
2003 * PARAMETERS:
2004 * bmp - pointer to bmap descriptor
2005 * dp - pointer to dmap to free the block range from.
2006 * blkno - starting block number of the block to be freed.
2007 * nblocks - number of blocks to be freed.
2008 *
2009 * RETURN VALUES:
2010 * 0 - success
2011 * -EIO - i/o error
2012 *
2013 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2014 */
2017 {
2018 s8 oldroot;
2021 /* save the current value of the root (i.e. maximum free string)
2022 * of the dmap tree.
2023 */
2026 /* free the specified (blocks) bits */
2029 /* if error or the root has not changed, done. */
2033 /* root changed. bubble the change up to the dmap control pages.
2034 * if the adjustment of the upper level control pages fails,
2035 * backout the deallocation.
2036 */
2040 /* as part of backing out the deallocation, we will have
2041 * to back split the dmap tree if the deallocation caused
2042 * the freed blocks to become part of a larger binary buddy
2043 * system.
2044 */
2049 }
2052 }
2055 /*
2056 * NAME: dbAllocBits()
2057 *
2058 * FUNCTION: allocate a specified block range from a dmap.
2059 *
2060 * this routine updates the dmap to reflect the working
2061 * state allocation of the specified block range. it directly
2062 * updates the bits of the working map and causes the adjustment
2063 * of the binary buddy system described by the dmap's dmtree
2064 * leaves to reflect the bits allocated. it also causes the
2065 * dmap's dmtree, as a whole, to reflect the allocated range.
2066 *
2067 * PARAMETERS:
2068 * bmp - pointer to bmap descriptor
2069 * dp - pointer to dmap to allocate bits from.
2070 * blkno - starting block number of the bits to be allocated.
2071 * nblocks - number of bits to be allocated.
2072 *
2073 * RETURN VALUES: none
2074 *
2075 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2076 */
2079 {
2085 /* pick up a pointer to the leaves of the dmap tree */
2088 /* determine the bit number and word within the dmap of the
2089 * starting block.
2090 */
2094 /* block range better be within the dmap */
2097 /* allocate the bits of the dmap's words corresponding to the block
2098 * range. not all bits of the first and last words may be contained
2099 * within the block range. if this is the case, we'll work against
2100 * those words (i.e. partial first and/or last) on an individual basis
2101 * (a single pass), allocating the bits of interest by hand and
2102 * updating the leaf corresponding to the dmap word. a single pass
2103 * will be used for all dmap words fully contained within the
2104 * specified range. within this pass, the bits of all fully contained
2105 * dmap words will be marked as free in a single shot and the leaves
2106 * will be updated. a single leaf may describe the free space of
2107 * multiple dmap words, so we may update only a subset of the actual
2108 * leaves corresponding to the dmap words of the block range.
2109 */
2111 /* determine the bit number within the word and
2112 * the number of bits within the word.
2113 */
2117 /* check if only part of a word is to be allocated.
2118 */
2120 /* allocate (set to 1) the appropriate bits within
2121 * this dmap word.
2122 */
2126 /* update the leaf for this dmap word. in addition
2127 * to setting the leaf value to the binary buddy max
2128 * of the updated dmap word, dbSplit() will split
2129 * the binary system of the leaves if need be.
2130 */
2136 /* one or more dmap words are fully contained
2137 * within the block range. determine how many
2138 * words and allocate (set to 1) the bits of these
2139 * words.
2140 */
2144 /* determine how many bits.
2145 */
2148 /* now update the appropriate leaves to reflect
2149 * the allocated words.
2150 */
2154 "dbAllocBits: leaf page "
2157 }
2159 /* determine what the leaf value should be
2160 * updated to as the minimum of the l2 number
2161 * of bits being allocated and the l2 number
2162 * of bits currently described by this leaf.
2163 */
2166 /* update the leaf to reflect the allocation.
2167 * in addition to setting the leaf value to
2168 * NOFREE, dbSplit() will split the binary
2169 * system of the leaves to reflect the current
2170 * allocation (size).
2171 */
2174 /* get the number of dmap words handled */
2177 }
2178 }
2179 }
2181 /* update the free count for this dmap */
2186 /* if this allocation group is completely free,
2187 * update the maximum allocation group number if this allocation
2188 * group is the new max.
2189 */
2194 /* update the free count for the allocation group and map */
2199 }
2202 /*
2203 * NAME: dbFreeBits()
2204 *
2205 * FUNCTION: free a specified block range from a dmap.
2206 *
2207 * this routine updates the dmap to reflect the working
2208 * state allocation of the specified block range. it directly
2209 * updates the bits of the working map and causes the adjustment
2210 * of the binary buddy system described by the dmap's dmtree
2211 * leaves to reflect the bits freed. it also causes the dmap's
2212 * dmtree, as a whole, to reflect the deallocated range.
2213 *
2214 * PARAMETERS:
2215 * bmp - pointer to bmap descriptor
2216 * dp - pointer to dmap to free bits from.
2217 * blkno - starting block number of the bits to be freed.
2218 * nblocks - number of bits to be freed.
2219 *
2220 * RETURN VALUES: 0 for success
2221 *
2222 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2223 */
2226 {
2232 /* determine the bit number and word within the dmap of the
2233 * starting block.
2234 */
2238 /* block range better be within the dmap.
2239 */
2242 /* free the bits of the dmaps words corresponding to the block range.
2243 * not all bits of the first and last words may be contained within
2244 * the block range. if this is the case, we'll work against those
2245 * words (i.e. partial first and/or last) on an individual basis
2246 * (a single pass), freeing the bits of interest by hand and updating
2247 * the leaf corresponding to the dmap word. a single pass will be used
2248 * for all dmap words fully contained within the specified range.
2249 * within this pass, the bits of all fully contained dmap words will
2250 * be marked as free in a single shot and the leaves will be updated. a
2251 * single leaf may describe the free space of multiple dmap words,
2252 * so we may update only a subset of the actual leaves corresponding
2253 * to the dmap words of the block range.
2254 *
2255 * dbJoin() is used to update leaf values and will join the binary
2256 * buddy system of the leaves if the new leaf values indicate this
2257 * should be done.
2258 */
2260 /* determine the bit number within the word and
2261 * the number of bits within the word.
2262 */
2266 /* check if only part of a word is to be freed.
2267 */
2269 /* free (zero) the appropriate bits within this
2270 * dmap word.
2271 */
2276 /* update the leaf for this dmap word.
2277 */
2285 /* one or more dmap words are fully contained
2286 * within the block range. determine how many
2287 * words and free (zero) the bits of these words.
2288 */
2292 /* determine how many bits.
2293 */
2296 /* now update the appropriate leaves to reflect
2297 * the freed words.
2298 */
2300 /* determine what the leaf value should be
2301 * updated to as the minimum of the l2 number
2302 * of bits being freed and the l2 (max) number
2303 * of bits that can be described by this leaf.
2304 */
2305 size =
2310 /* update the leaf.
2311 */
2316 /* get the number of dmap words handled.
2317 */
2320 }
2321 }
2322 }
2324 /* update the free count for this dmap.
2325 */
2330 /* update the free count for the allocation group and
2331 * map.
2332 */
2337 /* check if this allocation group is not completely free and
2338 * if it is currently the maximum (rightmost) allocation group.
2339 * if so, establish the new maximum allocation group number by
2340 * searching left for the first allocation group with allocation.
2341 */
2350 }
2352 /* re-establish the allocation group preference if the
2353 * current preference is right of the maximum allocation
2354 * group.
2355 */
2358 }
2363 }
2366 /*
2367 * NAME: dbAdjCtl()
2368 *
2369 * FUNCTION: adjust a dmap control page at a specified level to reflect
2370 * the change in a lower level dmap or dmap control page's
2371 * maximum string of free blocks (i.e. a change in the root
2372 * of the lower level object's dmtree) due to the allocation
2373 * or deallocation of a range of blocks with a single dmap.
2374 *
2375 * on entry, this routine is provided with the new value of
2376 * the lower level dmap or dmap control page root and the
2377 * starting block number of the block range whose allocation
2378 * or deallocation resulted in the root change. this range
2379 * is respresented by a single leaf of the current dmapctl
2380 * and the leaf will be updated with this value, possibly
2381 * causing a binary buddy system within the leaves to be
2382 * split or joined. the update may also cause the dmapctl's
2383 * dmtree to be updated.
2384 *
2385 * if the adjustment of the dmap control page, itself, causes its
2386 * root to change, this change will be bubbled up to the next dmap
2387 * control level by a recursive call to this routine, specifying
2388 * the new root value and the next dmap control page level to
2389 * be adjusted.
2390 * PARAMETERS:
2391 * bmp - pointer to bmap descriptor
2392 * blkno - the first block of a block range within a dmap. it is
2393 * the allocation or deallocation of this block range that
2394 * requires the dmap control page to be adjusted.
2395 * newval - the new value of the lower level dmap or dmap control
2396 * page root.
2397 * alloc - TRUE if adjustment is due to an allocation.
2398 * level - current level of dmap control page (i.e. L0, L1, L2) to
2399 * be adjusted.
2400 *
2401 * RETURN VALUES:
2402 * 0 - success
2403 * -EIO - i/o error
2404 *
2405 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2406 */
2407 static int
2409 {
2411 s8 oldroot;
2413 s64 lblkno;
2417 /* get the buffer for the dmap control page for the specified
2418 * block number and control page level.
2419 */
2431 }
2433 /* determine the leaf number corresponding to the block and
2434 * the index within the dmap control tree.
2435 */
2439 /* save the current leaf value and the current root level (i.e.
2440 * maximum l2 free string described by this dmapctl).
2441 */
2445 /* check if this is a control page update for an allocation.
2446 * if so, update the leaf to reflect the new leaf value using
2447 * dbSplit(); otherwise (deallocation), use dbJoin() to udpate
2448 * the leaf with the new value. in addition to updating the
2449 * leaf, dbSplit() will also split the binary buddy system of
2450 * the leaves, if required, and bubble new values within the
2451 * dmapctl tree, if required. similarly, dbJoin() will join
2452 * the binary buddy system of leaves and bubble new values up
2453 * the dmapctl tree as required by the new leaf value.
2454 */
2456 /* check if we are in the middle of a binary buddy
2457 * system. this happens when we are performing the
2458 * first allocation out of an allocation group that
2459 * is part (not the first part) of a larger binary
2460 * buddy system. if we are in the middle, back split
2461 * the system prior to calling dbSplit() which assumes
2462 * that it is at the front of a binary buddy system.
2463 */
2469 }
2475 }
2477 /* check if the root of the current dmap control page changed due
2478 * to the update and if the current dmap control page is not at
2479 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e.
2480 * root changed and this is not the top level), call this routine
2481 * again (recursion) for the next higher level of the mapping to
2482 * reflect the change in root for the current dmap control page.
2483 */
2485 /* are we below the top level of the map. if so,
2486 * bubble the root up to the next higher level.
2487 */
2489 /* bubble up the new root of this dmap control page to
2490 * the next level.
2491 */
2495 /* something went wrong in bubbling up the new
2496 * root value, so backout the changes to the
2497 * current dmap control page.
2498 */
2501 oldval);
2503 /* the dbJoin() above might have
2504 * caused a larger binary buddy system
2505 * to form and we may now be in the
2506 * middle of it. if this is the case,
2507 * back split the buddies.
2508 */
2514 }
2516 /* release the buffer and return the error.
2517 */
2520 }
2522 /* we're at the top level of the map. update
2523 * the bmap control page to reflect the size
2524 * of the maximum free buddy system.
2525 */
2529 "dbAdjCtl: the maximum free buddy is "
2531 }
2533 }
2534 }
2536 /* write the buffer.
2537 */
2541 }
2544 /*
2545 * NAME: dbSplit()
2546 *
2547 * FUNCTION: update the leaf of a dmtree with a new value, splitting
2548 * the leaf from the binary buddy system of the dmtree's
2549 * leaves, as required.
2550 *
2551 * PARAMETERS:
2552 * tp - pointer to the tree containing the leaf.
2553 * leafno - the number of the leaf to be updated.
2554 * splitsz - the size the binary buddy system starting at the leaf
2555 * must be split to, specified as the log2 number of blocks.
2556 * newval - the new value for the leaf.
2557 *
2558 * RETURN VALUES: none
2559 *
2560 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2561 */
2563 {
2568 /* check if the leaf needs to be split.
2569 */
2571 /* the split occurs by cutting the buddy system in half
2572 * at the specified leaf until we reach the specified
2573 * size. pick up the starting split size (current size
2574 * - 1 in l2) and the corresponding buddy size.
2575 */
2579 /* split until we reach the specified size.
2580 */
2582 /* update the buddy's leaf with its new value.
2583 */
2586 /* on to the next size and buddy.
2587 */
2590 }
2591 }
2593 /* adjust the dmap tree to reflect the specified leaf's new
2594 * value.
2595 */
2597 }
2600 /*
2601 * NAME: dbBackSplit()
2602 *
2603 * FUNCTION: back split the binary buddy system of dmtree leaves
2604 * that hold a specified leaf until the specified leaf
2605 * starts its own binary buddy system.
2606 *
2607 * the allocators typically perform allocations at the start
2608 * of binary buddy systems and dbSplit() is used to accomplish
2609 * any required splits. in some cases, however, allocation
2610 * may occur in the middle of a binary system and requires a
2611 * back split, with the split proceeding out from the middle of
2612 * the system (less efficient) rather than the start of the
2613 * system (more efficient). the cases in which a back split
2614 * is required are rare and are limited to the first allocation
2615 * within an allocation group which is a part (not first part)
2616 * of a larger binary buddy system and a few exception cases
2617 * in which a previous join operation must be backed out.
2618 *
2619 * PARAMETERS:
2620 * tp - pointer to the tree containing the leaf.
2621 * leafno - the number of the leaf to be updated.
2622 *
2623 * RETURN VALUES: none
2624 *
2625 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2626 */
2628 {
2633 /* leaf should be part (not first part) of a binary
2634 * buddy system.
2635 */
2638 /* the back split is accomplished by iteratively finding the leaf
2639 * that starts the buddy system that contains the specified leaf and
2640 * splitting that system in two. this iteration continues until
2641 * the specified leaf becomes the start of a buddy system.
2642 *
2643 * determine maximum possible l2 size for the specified leaf.
2644 */
2645 size =
2649 /* determine the number of leaves covered by this size. this
2650 * is the buddy size that we will start with as we search for
2651 * the buddy system that contains the specified leaf.
2652 */
2655 /* back split.
2656 */
2658 /* find the leftmost buddy leaf.
2659 */
2665 }
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 }
2687 }
2689 }
2692 /*
2693 * NAME: dbJoin()
2694 *
2695 * FUNCTION: update the leaf of a dmtree with a new value, joining
2696 * the leaf with other leaves of the dmtree into a multi-leaf
2697 * binary buddy system, as required.
2698 *
2699 * PARAMETERS:
2700 * tp - pointer to the tree containing the leaf.
2701 * leafno - the number of the leaf to be updated.
2702 * newval - the new value for the leaf.
2703 *
2704 * RETURN VALUES: none
2705 */
2707 {
2711 /* can the new leaf value require a join with other leaves ?
2712 */
2714 /* pickup a pointer to the leaves of the tree.
2715 */
2718 /* try to join the specified leaf into a large binary
2719 * buddy system. the join proceeds by attempting to join
2720 * the specified leafno with its buddy (leaf) at new value.
2721 * if the join occurs, we attempt to join the left leaf
2722 * of the joined buddies with its buddy at new value + 1.
2723 * we continue to join until we find a buddy that cannot be
2724 * joined (does not have a value equal to the size of the
2725 * last join) or until all leaves have been joined into a
2726 * single system.
2727 *
2728 * get the buddy size (number of words covered) of
2729 * the new value.
2730 */
2733 /* try to join.
2734 */
2736 /* get the buddy leaf.
2737 */
2740 /* if the leaf's new value is greater than its
2741 * buddy's value, we join no more.
2742 */
2746 /* It shouldn't be less */
2750 /* check which (leafno or buddy) is the left buddy.
2751 * the left buddy gets to claim the blocks resulting
2752 * from the join while the right gets to claim none.
2753 * the left buddy is also eligable to participate in
2754 * a join at the next higher level while the right
2755 * is not.
2756 *
2757 */
2759 /* leafno is the left buddy.
2760 */
2763 /* buddy is the left buddy and becomes
2764 * leafno.
2765 */
2768 }
2770 /* on to try the next join.
2771 */
2774 }
2775 }
2777 /* update the leaf value.
2778 */
2782 }
2785 /*
2786 * NAME: dbAdjTree()
2787 *
2788 * FUNCTION: update a leaf of a dmtree with a new value, adjusting
2789 * the dmtree, as required, to reflect the new leaf value.
2790 * the combination of any buddies must already be done before
2791 * this is called.
2792 *
2793 * PARAMETERS:
2794 * tp - pointer to the tree to be adjusted.
2795 * leafno - the number of the leaf to be updated.
2796 * newval - the new value for the leaf.
2797 *
2798 * RETURN VALUES: none
2799 */
2801 {
2805 /* pick up the index of the leaf for this leafno.
2806 */
2809 /* is the current value the same as the old value ? if so,
2810 * there is nothing to do.
2811 */
2815 /* set the new value.
2816 */
2819 /* bubble the new value up the tree as required.
2820 */
2822 /* get the index of the first leaf of the 4 leaf
2823 * group containing the specified leaf (leafno).
2824 */
2827 /* get the index of the parent of this 4 leaf group.
2828 */
2831 /* determine the maximum of the 4 leaves.
2832 */
2835 /* if the maximum of the 4 is the same as the
2836 * parent's value, we're done.
2837 */
2841 /* parent gets new value.
2842 */
2845 /* parent becomes leaf for next go-round.
2846 */
2848 }
2849 }
2852 /*
2853 * NAME: dbFindLeaf()
2854 *
2855 * FUNCTION: search a dmtree_t for sufficient free blocks, returning
2856 * the index of a leaf describing the free blocks if
2857 * sufficient free blocks are found.
2858 *
2859 * the search starts at the top of the dmtree_t tree and
2860 * proceeds down the tree to the leftmost leaf with sufficient
2861 * free space.
2862 *
2863 * PARAMETERS:
2864 * tp - pointer to the tree to be searched.
2865 * l2nb - log2 number of free blocks to search for.
2866 * leafidx - return pointer to be set to the index of the leaf
2867 * describing at least l2nb free blocks if sufficient
2868 * free blocks are found.
2869 *
2870 * RETURN VALUES:
2871 * 0 - success
2872 * -ENOSPC - insufficient free blocks.
2873 */
2875 {
2878 /* first check the root of the tree to see if there is
2879 * sufficient free space.
2880 */
2884 /* sufficient free space available. now search down the tree
2885 * starting at the next level for the leftmost leaf that
2886 * describes sufficient free space.
2887 */
2890 /* search the four nodes at this level, starting from
2891 * the left.
2892 */
2894 /* sufficient free space found. move to the next
2895 * level (or quit if this is the last level).
2896 */
2899 }
2901 /* better have found something since the higher
2902 * levels of the tree said it was here.
2903 */
2905 }
2907 /* set the return to the leftmost leaf describing sufficient
2908 * free space.
2909 */
2913 }
2916 /*
2917 * NAME: dbFindBits()
2918 *
2919 * FUNCTION: find a specified number of binary buddy free bits within a
2920 * dmap bitmap word value.
2921 *
2922 * this routine searches the bitmap value for (1 << l2nb) free
2923 * bits at (1 << l2nb) alignments within the value.
2924 *
2925 * PARAMETERS:
2926 * word - dmap bitmap word value.
2927 * l2nb - number of free bits specified as a log2 number.
2928 *
2929 * RETURN VALUES:
2930 * starting bit number of free bits.
2931 */
2933 {
2935 u32 mask;
2937 /* get the number of bits.
2938 */
2942 /* complement the word so we can use a mask (i.e. 0s represent
2943 * free bits) and compute the mask.
2944 */
2948 /* scan the word for nb free bits at nb alignments.
2949 */
2953 }
2957 /* return the bit number.
2958 */
2960 }
2963 /*
2964 * NAME: dbMaxBud(u8 *cp)
2965 *
2966 * FUNCTION: determine the largest binary buddy string of free
2967 * bits within 32-bits of the map.
2968 *
2969 * PARAMETERS:
2970 * cp - pointer to the 32-bit value.
2971 *
2972 * RETURN VALUES:
2973 * largest binary buddy of free bits within a dmap word.
2974 */
2976 {
2979 /* check if the wmap word is all free. if so, the
2980 * free buddy size is BUDMIN.
2981 */
2985 /* check if the wmap word is half free. if so, the
2986 * free buddy size is BUDMIN-1.
2987 */
2991 /* not all free or half free. determine the free buddy
2992 * size thru table lookup using quarters of the wmap word.
2993 */
2997 }
3000 /*
3001 * NAME: cnttz(uint word)
3002 *
3003 * FUNCTION: determine the number of trailing zeros within a 32-bit
3004 * value.
3005 *
3006 * PARAMETERS:
3007 * value - 32-bit value to be examined.
3008 *
3009 * RETURN VALUES:
3010 * count of trailing zeros
3011 */
3013 {
3019 }
3022 }
3025 /*
3026 * NAME: cntlz(u32 value)
3027 *
3028 * FUNCTION: determine the number of leading zeros within a 32-bit
3029 * value.
3030 *
3031 * PARAMETERS:
3032 * value - 32-bit value to be examined.
3033 *
3034 * RETURN VALUES:
3035 * count of leading zeros
3036 */
3038 {
3044 }
3046 }
3049 /*
3050 * NAME: blkstol2(s64 nb)
3051 *
3052 * FUNCTION: convert a block count to its log2 value. if the block
3053 * count is not a l2 multiple, it is rounded up to the next
3054 * larger l2 multiple.
3055 *
3056 * PARAMETERS:
3057 * nb - number of blocks
3058 *
3059 * RETURN VALUES:
3060 * log2 number of blocks
3061 */
3063 {
3069 /* count the leading bits.
3070 */
3072 /* leading bit found.
3073 */
3075 /* determine the l2 value.
3076 */
3079 /* check if we need to round up.
3080 */
3082 l2nb++;
3085 }
3086 }
3089 }
3092 /*
3093 * NAME: dbAllocBottomUp()
3094 *
3095 * FUNCTION: alloc the specified block range from the working block
3096 * allocation map.
3097 *
3098 * the blocks will be alloc from the working map one dmap
3099 * at a time.
3100 *
3101 * PARAMETERS:
3102 * ip - pointer to in-core inode;
3103 * blkno - starting block number to be freed.
3104 * nblocks - number of blocks to be freed.
3105 *
3106 * RETURN VALUES:
3107 * 0 - success
3108 * -EIO - i/o error
3109 */
3111 {
3121 /* block to be allocated better be within the mapsize. */
3124 /*
3125 * allocate the blocks a dmap at a time.
3126 */
3129 /* release previous dmap if any */
3132 }
3134 /* get the buffer for the current dmap. */
3140 }
3143 /* determine the number of blocks to be allocated from
3144 * this dmap.
3145 */
3148 /* allocate the blocks. */
3153 }
3154 }
3156 /* write the last buffer. */
3162 }
3167 {
3173 /* save the current value of the root (i.e. maximum free string)
3174 * of the dmap tree.
3175 */
3178 /* pick up a pointer to the leaves of the dmap tree */
3181 /* determine the bit number and word within the dmap of the
3182 * starting block.
3183 */
3187 /* block range better be within the dmap */
3190 /* allocate the bits of the dmap's words corresponding to the block
3191 * range. not all bits of the first and last words may be contained
3192 * within the block range. if this is the case, we'll work against
3193 * those words (i.e. partial first and/or last) on an individual basis
3194 * (a single pass), allocating the bits of interest by hand and
3195 * updating the leaf corresponding to the dmap word. a single pass
3196 * will be used for all dmap words fully contained within the
3197 * specified range. within this pass, the bits of all fully contained
3198 * dmap words will be marked as free in a single shot and the leaves
3199 * will be updated. a single leaf may describe the free space of
3200 * multiple dmap words, so we may update only a subset of the actual
3201 * leaves corresponding to the dmap words of the block range.
3202 */
3204 /* determine the bit number within the word and
3205 * the number of bits within the word.
3206 */
3210 /* check if only part of a word is to be allocated.
3211 */
3213 /* allocate (set to 1) the appropriate bits within
3214 * this dmap word.
3215 */
3219 word++;
3221 /* one or more dmap words are fully contained
3222 * within the block range. determine how many
3223 * words and allocate (set to 1) the bits of these
3224 * words.
3225 */
3229 /* determine how many bits */
3232 }
3233 }
3235 /* update the free count for this dmap */
3238 /* reconstruct summary tree */
3243 /* if this allocation group is completely free,
3244 * update the highest active allocation group number
3245 * if this allocation group is the new max.
3246 */
3251 /* update the free count for the allocation group and map */
3257 /* if the root has not changed, done. */
3261 /* root changed. bubble the change up to the dmap control pages.
3262 * if the adjustment of the upper level control pages fails,
3263 * backout the bit allocation (thus making everything consistent).
3264 */
3269 }
3272 /*
3273 * NAME: dbExtendFS()
3274 *
3275 * FUNCTION: extend bmap from blkno for nblocks;
3276 * dbExtendFS() updates bmap ready for dbAllocBottomUp();
3277 *
3278 * L2
3279 * |
3280 * L1---------------------------------L1
3281 * | |
3282 * L0---------L0---------L0 L0---------L0---------L0
3283 * | | | | | |
3284 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm;
3285 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3286 *
3287 * <---old---><----------------------------extend----------------------->
3288 */
3290 {
3294 s64 newsize;
3295 s64 p;
3302 s64 ag_rem;
3309 /*
3310 * initialize bmap control page.
3311 *
3312 * all the data in bmap control page should exclude
3313 * the mkfs hidden dmap page.
3314 */
3316 /* update mapsize */
3320 /* compute new AG size */
3327 /* compute new number of AG */
3332 /*
3333 * reconfigure db_agfree[]
3334 * from old AG configuration to new AG configuration;
3335 *
3336 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3337 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3338 * note: new AG size = old AG size * (2**x).
3339 */
3347 /* coalesce cotiguous k AGs; */
3349 /* merge AGi to AGn */
3351 }
3352 }
3358 /*
3359 * update highest active ag number
3360 */
3364 /*
3365 * extend bmap
3366 *
3367 * update bit maps and corresponding level control pages;
3368 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3369 */
3370 extend:
3371 /* get L2 page */
3377 }
3380 /* compute start L1 */
3385 /*
3386 * extend each L1 in L2
3387 */
3389 /* get L1 page */
3391 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3397 /* compute start L0 */
3403 /* assign/init L1 page */
3410 /* compute start L0 */
3414 }
3416 /*
3417 * extend each L0 in L1
3418 */
3420 /* get L0 page */
3422 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3429 /* compute start dmap */
3431 L2BPERDMAP;
3437 /* assign/init L0 page */
3444 /* compute start dmap */
3448 }
3450 /*
3451 * extend each dmap in L0
3452 */
3454 /*
3455 * reconstruct the dmap page, and
3456 * initialize corresponding parent L0 leaf
3457 */
3459 /* read in dmap page: */
3466 /* assign/init dmap page */
3473 }
3484 l0leaf++;
3493 /*
3494 * build current L0 page from its leaves, and
3495 * initialize corresponding parent L1 leaf
3496 */
3504 /* more than 1 L0 ? */
3508 /* summarize in global bmap page */
3513 }
3514 }
3517 /*
3518 * build current L1 page from its leaves, and
3519 * initialize corresponding parent L2 leaf
3520 */
3528 /* more than 1 L1 ? */
3532 /* summarize in global bmap page */
3536 }
3537 }
3542 errout:
3550 /*
3551 * finalize bmap control page
3552 */
3553 finalize:
3556 }
3559 /*
3560 * dbFinalizeBmap()
3561 */
3563 {
3569 /*
3570 * finalize bmap control page
3571 */
3572 //finalize:
3573 /*
3574 * compute db_agpref: preferred ag to allocate from
3575 * (the leftmost ag with average free space in it);
3576 */
3577 //agpref:
3578 /* get the number of active ags and inacitve ags */
3583 /* determine how many blocks are in the inactive allocation
3584 * groups. in doing this, we must account for the fact that
3585 * the rightmost group might be a partial group (i.e. file
3586 * system size is not a multiple of the group size).
3587 */
3592 /* determine how many free blocks are in the active
3593 * allocation groups plus the average number of free blocks
3594 * within the active ags.
3595 */
3599 /* if the preferred allocation group has not average free space.
3600 * re-establish the preferred group as the leftmost
3601 * group with average free space.
3602 */
3608 }
3612 }
3613 }
3615 /*
3616 * compute db_aglevel, db_agheigth, db_width, db_agstart:
3617 * an ag is covered in aglevel dmapctl summary tree,
3618 * at agheight level height (from leaf) with agwidth number of nodes
3619 * each, which starts at agstart index node of the smmary tree node
3620 * array;
3621 */
3623 l2nl =
3628 i--) {
3631 }
3633 }
3636 /*
3637 * NAME: dbInitDmap()/ujfs_idmap_page()
3638 *
3639 * FUNCTION: initialize working/persistent bitmap of the dmap page
3640 * for the specified number of blocks:
3641 *
3642 * at entry, the bitmaps had been initialized as free (ZEROS);
3643 * The number of blocks will only account for the actually
3644 * existing blocks. Blocks which don't actually exist in
3645 * the aggregate will be marked as allocated (ONES);
3646 *
3647 * PARAMETERS:
3648 * dp - pointer to page of map
3649 * nblocks - number of blocks this page
3650 *
3651 * RETURNS: NONE
3652 */
3654 {
3657 /* starting block number within the dmap */
3668 }
3673 }
3675 /* word number containing start block number */
3678 /*
3679 * free the bits corresponding to the block range (ZEROS):
3680 * note: not all bits of the first and last words may be contained
3681 * within the block range.
3682 */
3684 /* number of bits preceding range to be freed in the word */
3686 /* number of bits to free in the word */
3689 /* is partial word to be freed ? */
3691 /* free (set to 0) from the bitmap word */
3697 /* skip the word freed */
3698 w++;
3700 /* free (set to 0) contiguous bitmap words */
3705 /* skip the words freed */
3708 }
3709 }
3711 /*
3712 * mark bits following the range to be freed (non-existing
3713 * blocks) as allocated (ONES)
3714 */
3719 /* the first word beyond the end of existing blocks */
3722 /* does nblocks fall on a 32-bit boundary ? */
3725 /* mark a partial word allocated */
3727 w++;
3728 }
3730 /* set the rest of the words in the page to allocated (ONES) */
3734 /*
3735 * init tree
3736 */
3737 initTree:
3739 }
3742 /*
3743 * NAME: dbInitDmapTree()/ujfs_complete_dmap()
3744 *
3745 * FUNCTION: initialize summary tree of the specified dmap:
3746 *
3747 * at entry, bitmap of the dmap has been initialized;
3748 *
3749 * PARAMETERS:
3750 * dp - dmap to complete
3751 * blkno - starting block number for this dmap
3752 * treemax - will be filled in with max free for this dmap
3753 *
3754 * RETURNS: max free string at the root of the tree
3755 */
3757 {
3762 /* init fixed info of tree */
3770 /* init each leaf from corresponding wmap word:
3771 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3772 * bitmap word are allocated.
3773 */
3778 /* build the dmap's binary buddy summary tree */
3780 }
3783 /*
3784 * NAME: dbInitTree()/ujfs_adjtree()
3785 *
3786 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl.
3787 *
3788 * at entry, the leaves of the tree has been initialized
3789 * from corresponding bitmap word or root of summary tree
3790 * of the child control page;
3791 * configure binary buddy system at the leaf level, then
3792 * bubble up the values of the leaf nodes up the tree.
3793 *
3794 * PARAMETERS:
3795 * cp - Pointer to the root of the tree
3796 * l2leaves- Number of leaf nodes as a power of 2
3797 * l2min - Number of blocks that can be covered by a leaf
3798 * as a power of 2
3799 *
3800 * RETURNS: max free string at the root of the tree
3801 */
3803 {
3810 /* Determine the maximum free string possible for the leaves */
3813 /*
3814 * configure the leaf levevl into binary buddy system
3815 *
3816 * Try to combine buddies starting with a buddy size of 1
3817 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3818 * can be combined if both buddies have a maximum free of l2min;
3819 * the combination will result in the left-most buddy leaf having
3820 * a maximum free of l2min+1.
3821 * After processing all buddies for a given size, process buddies
3822 * at the next higher buddy size (i.e. current size * 2) and
3823 * the next maximum free (current free + 1).
3824 * This continues until the maximum possible buddy combination
3825 * yields maximum free.
3826 */
3829 /* get next buddy size == current buddy pair size */
3832 /* scan each adjacent buddy pair at current buddy size */
3836 /* coalesce if both adjacent buddies are max free */
3840 }
3841 }
3842 }
3844 /*
3845 * bubble summary information of leaves up the tree.
3846 *
3847 * Starting at the leaf node level, the four nodes described by
3848 * the higher level parent node are compared for a maximum free and
3849 * this maximum becomes the value of the parent node.
3850 * when all lower level nodes are processed in this fashion then
3851 * move up to the next level (parent becomes a lower level node) and
3852 * continue the process for that level.
3853 */
3857 /* get index of 1st node of parent level */
3860 /* set the value of the parent node as the maximum
3861 * of the four nodes of the current level.
3862 */
3866 }
3869 }
3872 /*
3873 * dbInitDmapCtl()
3874 *
3875 * function: initialize dmapctl page
3876 */
3887 /*
3888 * initialize the leaves of current level that were not covered
3889 * by the specified input block range (i.e. the leaves have no
3890 * low level dmapctl or dmap).
3891 */
3896 /* build the dmap's binary buddy summary tree */
3898 }
3901 /*
3902 * NAME: dbGetL2AGSize()/ujfs_getagl2size()
3903 *
3904 * FUNCTION: Determine log2(allocation group size) from aggregate size
3905 *
3906 * PARAMETERS:
3907 * nblocks - Number of blocks in aggregate
3908 *
3909 * RETURNS: log2(allocation group size) in aggregate blocks
3910 */
3912 {
3913 s64 sz;
3914 s64 m;
3920 /* round up aggregate size to power of 2 */
3925 }
3931 /* agsize = roundupSize/max_number_of_ag */
3933 }
3936 /*
3937 * NAME: dbMapFileSizeToMapSize()
3938 *
3939 * FUNCTION: compute number of blocks the block allocation map file
3940 * can cover from the map file size;
3941 *
3942 * RETURNS: Number of blocks which can be covered by this block map file;
3943 */
3945 /*
3946 * maximum number of map pages at each level including control pages
3947 */
3948 #define MAXL0PAGES (1 + LPERCTL)
3949 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES)
3950 #define MAXL2PAGES (1 + LPERCTL * MAXL1PAGES)
3952 /*
3953 * convert number of map pages to the zero origin top dmapctl level
3954 */
3955 #define BMAPPGTOLEV(npages) \
3956 (((npages) <= 3 + MAXL0PAGES) ? 0 \
3957 : ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
3960 {
3962 s64 nblocks;
3971 /* At each level, accumulate the number of dmap pages covered by
3972 * the number of full child levels below it;
3973 * repeat for the last incomplete child level.
3974 */
3977 /* skip higher level control pages above top level covered by map */
3981 factor =
3987 /* pages in last/incomplete child */
3989 /* skip incomplete child's level control page */
3990 npages--;
3991 }
3993 /* convert the number of dmaps into the number of blocks
3994 * which can be covered by the dmaps;
3995 */
3999 }