2 * Copyright (C) International Business Machines Corp., 2000-2004
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.
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.
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
20 #include "jfs_incore.h"
21 #include "jfs_superblock.h"
25 #include "jfs_metapage.h"
26 #include "jfs_debug.h"
29 * SERIALIZATION of the Block Allocation Map.
31 * the working state of the block allocation map is accessed in
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).
38 * 2) allocation requests that start at dmap control page
39 * level and work down towards the dmaps.
41 * the serialization scheme used here is as follows.
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).
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);
63 * accesses to the persistent state of the block allocation map (limited
64 * to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
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)
74 static void dbAllocBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
76 static void dbSplit(dmtree_t
* tp
, int leafno
, int splitsz
, int newval
);
77 static int dbBackSplit(dmtree_t
* tp
, int leafno
);
78 static int dbJoin(dmtree_t
* tp
, int leafno
, int newval
);
79 static void dbAdjTree(dmtree_t
* tp
, int leafno
, int newval
);
80 static int dbAdjCtl(struct bmap
* bmp
, s64 blkno
, int newval
, int alloc
,
82 static int dbAllocAny(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64
* results
);
83 static int dbAllocNext(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
85 static int dbAllocNear(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
87 int l2nb
, s64
* results
);
88 static int dbAllocDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
90 static int dbAllocDmapLev(struct bmap
* bmp
, struct dmap
* dp
, int nblocks
,
93 static int dbAllocAG(struct bmap
* bmp
, int agno
, s64 nblocks
, int l2nb
,
95 static int dbAllocCtl(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64 blkno
,
97 static int dbExtend(struct inode
*ip
, s64 blkno
, s64 nblocks
, s64 addnblocks
);
98 static int dbFindBits(u32 word
, int l2nb
);
99 static int dbFindCtl(struct bmap
* bmp
, int l2nb
, int level
, s64
* blkno
);
100 static int dbFindLeaf(dmtree_t
* tp
, int l2nb
, int *leafidx
);
101 static int dbFreeBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
103 static int dbFreeDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
105 static int dbMaxBud(u8
* cp
);
106 s64
dbMapFileSizeToMapSize(struct inode
*ipbmap
);
107 static int blkstol2(s64 nb
);
109 static int cntlz(u32 value
);
110 static int cnttz(u32 word
);
112 static int dbAllocDmapBU(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
114 static int dbInitDmap(struct dmap
* dp
, s64 blkno
, int nblocks
);
115 static int dbInitDmapTree(struct dmap
* dp
);
116 static int dbInitTree(struct dmaptree
* dtp
);
117 static int dbInitDmapCtl(struct dmapctl
* dcp
, int level
, int i
);
118 static int dbGetL2AGSize(s64 nblocks
);
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.
128 static const s8 budtab
[256] = {
129 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
130 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
131 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
132 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
133 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
134 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
135 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
136 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
137 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
138 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
139 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
140 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
141 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
142 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
143 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
144 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1
151 * FUNCTION: initializate the block allocation map.
153 * memory is allocated for the in-core bmap descriptor and
154 * the in-core descriptor is initialized from disk.
157 * ipbmap - pointer to in-core inode for the block map.
161 * -ENOMEM - insufficient memory
164 int dbMount(struct inode
*ipbmap
)
167 struct dbmap_disk
*dbmp_le
;
172 * allocate/initialize the in-memory bmap descriptor
174 /* allocate memory for the in-memory bmap descriptor */
175 bmp
= kmalloc(sizeof(struct bmap
), GFP_KERNEL
);
179 /* read the on-disk bmap descriptor. */
180 mp
= read_metapage(ipbmap
,
181 BMAPBLKNO
<< JFS_SBI(ipbmap
->i_sb
)->l2nbperpage
,
188 /* copy the on-disk bmap descriptor to its in-memory version. */
189 dbmp_le
= (struct dbmap_disk
*) mp
->data
;
190 bmp
->db_mapsize
= le64_to_cpu(dbmp_le
->dn_mapsize
);
191 bmp
->db_nfree
= le64_to_cpu(dbmp_le
->dn_nfree
);
192 bmp
->db_l2nbperpage
= le32_to_cpu(dbmp_le
->dn_l2nbperpage
);
193 bmp
->db_numag
= le32_to_cpu(dbmp_le
->dn_numag
);
194 bmp
->db_maxlevel
= le32_to_cpu(dbmp_le
->dn_maxlevel
);
195 bmp
->db_maxag
= le32_to_cpu(dbmp_le
->dn_maxag
);
196 bmp
->db_agpref
= le32_to_cpu(dbmp_le
->dn_agpref
);
197 bmp
->db_aglevel
= le32_to_cpu(dbmp_le
->dn_aglevel
);
198 bmp
->db_agheigth
= le32_to_cpu(dbmp_le
->dn_agheigth
);
199 bmp
->db_agwidth
= le32_to_cpu(dbmp_le
->dn_agwidth
);
200 bmp
->db_agstart
= le32_to_cpu(dbmp_le
->dn_agstart
);
201 bmp
->db_agl2size
= le32_to_cpu(dbmp_le
->dn_agl2size
);
202 for (i
= 0; i
< MAXAG
; i
++)
203 bmp
->db_agfree
[i
] = le64_to_cpu(dbmp_le
->dn_agfree
[i
]);
204 bmp
->db_agsize
= le64_to_cpu(dbmp_le
->dn_agsize
);
205 bmp
->db_maxfreebud
= dbmp_le
->dn_maxfreebud
;
207 /* release the buffer. */
208 release_metapage(mp
);
210 /* bind the bmap inode and the bmap descriptor to each other. */
211 bmp
->db_ipbmap
= ipbmap
;
212 JFS_SBI(ipbmap
->i_sb
)->bmap
= bmp
;
214 memset(bmp
->db_active
, 0, sizeof(bmp
->db_active
));
217 * allocate/initialize the bmap lock
228 * FUNCTION: terminate the block allocation map in preparation for
229 * file system unmount.
231 * the in-core bmap descriptor is written to disk and
232 * the memory for this descriptor is freed.
235 * ipbmap - pointer to in-core inode for the block map.
241 int dbUnmount(struct inode
*ipbmap
, int mounterror
)
243 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
245 if (!(mounterror
|| isReadOnly(ipbmap
)))
249 * Invalidate the page cache buffers
251 truncate_inode_pages(ipbmap
->i_mapping
, 0);
253 /* free the memory for the in-memory bmap. */
262 int dbSync(struct inode
*ipbmap
)
264 struct dbmap_disk
*dbmp_le
;
265 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
270 * write bmap global control page
272 /* get the buffer for the on-disk bmap descriptor. */
273 mp
= read_metapage(ipbmap
,
274 BMAPBLKNO
<< JFS_SBI(ipbmap
->i_sb
)->l2nbperpage
,
277 jfs_err("dbSync: read_metapage failed!");
280 /* copy the in-memory version of the bmap to the on-disk version */
281 dbmp_le
= (struct dbmap_disk
*) mp
->data
;
282 dbmp_le
->dn_mapsize
= cpu_to_le64(bmp
->db_mapsize
);
283 dbmp_le
->dn_nfree
= cpu_to_le64(bmp
->db_nfree
);
284 dbmp_le
->dn_l2nbperpage
= cpu_to_le32(bmp
->db_l2nbperpage
);
285 dbmp_le
->dn_numag
= cpu_to_le32(bmp
->db_numag
);
286 dbmp_le
->dn_maxlevel
= cpu_to_le32(bmp
->db_maxlevel
);
287 dbmp_le
->dn_maxag
= cpu_to_le32(bmp
->db_maxag
);
288 dbmp_le
->dn_agpref
= cpu_to_le32(bmp
->db_agpref
);
289 dbmp_le
->dn_aglevel
= cpu_to_le32(bmp
->db_aglevel
);
290 dbmp_le
->dn_agheigth
= cpu_to_le32(bmp
->db_agheigth
);
291 dbmp_le
->dn_agwidth
= cpu_to_le32(bmp
->db_agwidth
);
292 dbmp_le
->dn_agstart
= cpu_to_le32(bmp
->db_agstart
);
293 dbmp_le
->dn_agl2size
= cpu_to_le32(bmp
->db_agl2size
);
294 for (i
= 0; i
< MAXAG
; i
++)
295 dbmp_le
->dn_agfree
[i
] = cpu_to_le64(bmp
->db_agfree
[i
]);
296 dbmp_le
->dn_agsize
= cpu_to_le64(bmp
->db_agsize
);
297 dbmp_le
->dn_maxfreebud
= bmp
->db_maxfreebud
;
299 /* write the buffer */
303 * write out dirty pages of bmap
305 filemap_write_and_wait(ipbmap
->i_mapping
);
307 diWriteSpecial(ipbmap
, 0);
316 * FUNCTION: free the specified block range from the working block
319 * the blocks will be free from the working map one dmap
323 * ip - pointer to in-core inode;
324 * blkno - starting block number to be freed.
325 * nblocks - number of blocks to be freed.
331 int dbFree(struct inode
*ip
, s64 blkno
, s64 nblocks
)
337 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
338 struct bmap
*bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
342 /* block to be freed better be within the mapsize. */
343 if (unlikely((blkno
== 0) || (blkno
+ nblocks
> bmp
->db_mapsize
))) {
344 IREAD_UNLOCK(ipbmap
);
345 printk(KERN_ERR
"blkno = %Lx, nblocks = %Lx\n",
346 (unsigned long long) blkno
,
347 (unsigned long long) nblocks
);
349 "dbFree: block to be freed is outside the map");
354 * free the blocks a dmap at a time.
357 for (rem
= nblocks
; rem
> 0; rem
-= nb
, blkno
+= nb
) {
358 /* release previous dmap if any */
363 /* get the buffer for the current dmap. */
364 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
365 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
367 IREAD_UNLOCK(ipbmap
);
370 dp
= (struct dmap
*) mp
->data
;
372 /* determine the number of blocks to be freed from
375 nb
= min(rem
, BPERDMAP
- (blkno
& (BPERDMAP
- 1)));
377 /* free the blocks. */
378 if ((rc
= dbFreeDmap(bmp
, dp
, blkno
, nb
))) {
379 jfs_error(ip
->i_sb
, "dbFree: error in block map\n");
380 release_metapage(mp
);
381 IREAD_UNLOCK(ipbmap
);
386 /* write the last buffer. */
389 IREAD_UNLOCK(ipbmap
);
396 * NAME: dbUpdatePMap()
398 * FUNCTION: update the allocation state (free or allocate) of the
399 * specified block range in the persistent block allocation map.
401 * the blocks will be updated in the persistent map one
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;
417 dbUpdatePMap(struct inode
*ipbmap
,
418 int free
, s64 blkno
, s64 nblocks
, struct tblock
* tblk
)
420 int nblks
, dbitno
, wbitno
, rbits
;
421 int word
, nbits
, nwords
;
422 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
423 s64 lblkno
, rem
, lastlblkno
;
428 int lsn
, difft
, diffp
;
431 /* the blocks better be within the mapsize. */
432 if (blkno
+ nblocks
> bmp
->db_mapsize
) {
433 printk(KERN_ERR
"blkno = %Lx, nblocks = %Lx\n",
434 (unsigned long long) blkno
,
435 (unsigned long long) nblocks
);
436 jfs_error(ipbmap
->i_sb
,
437 "dbUpdatePMap: blocks are outside the map");
441 /* compute delta of transaction lsn from log syncpt */
443 log
= (struct jfs_log
*) JFS_SBI(tblk
->sb
)->log
;
444 logdiff(difft
, lsn
, log
);
447 * update the block state a dmap at a time.
451 for (rem
= nblocks
; rem
> 0; rem
-= nblks
, blkno
+= nblks
) {
452 /* get the buffer for the current dmap. */
453 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
454 if (lblkno
!= lastlblkno
) {
459 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
,
463 metapage_wait_for_io(mp
);
465 dp
= (struct dmap
*) mp
->data
;
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.
471 dbitno
= blkno
& (BPERDMAP
- 1);
472 word
= dbitno
>> L2DBWORD
;
473 nblks
= min(rem
, (s64
)BPERDMAP
- dbitno
);
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.
482 for (rbits
= nblks
; rbits
> 0;
483 rbits
-= nbits
, dbitno
+= nbits
) {
484 /* determine the bit number within the word and
485 * the number of bits within the word.
487 wbitno
= dbitno
& (DBWORD
- 1);
488 nbits
= min(rbits
, DBWORD
- wbitno
);
490 /* check if only part of the word is to be updated. */
491 if (nbits
< DBWORD
) {
492 /* update (free or allocate) the bits
496 (ONES
<< (DBWORD
- nbits
) >> wbitno
);
506 /* one or more words are to have all
507 * their bits updated. determine how
508 * many words and how many bits.
510 nwords
= rbits
>> L2DBWORD
;
511 nbits
= nwords
<< L2DBWORD
;
513 /* update (free or allocate) the bits
517 memset(&dp
->pmap
[word
], 0,
520 memset(&dp
->pmap
[word
], (int) ONES
,
530 if (lblkno
== lastlblkno
)
535 LOGSYNC_LOCK(log
, flags
);
537 /* inherit older/smaller lsn */
538 logdiff(diffp
, mp
->lsn
, log
);
542 /* move bp after tblock in logsync list */
543 list_move(&mp
->synclist
, &tblk
->synclist
);
546 /* inherit younger/larger clsn */
547 logdiff(difft
, tblk
->clsn
, log
);
548 logdiff(diffp
, mp
->clsn
, log
);
550 mp
->clsn
= tblk
->clsn
;
555 /* insert bp after tblock in logsync list */
557 list_add(&mp
->synclist
, &tblk
->synclist
);
559 mp
->clsn
= tblk
->clsn
;
561 LOGSYNC_UNLOCK(log
, flags
);
564 /* write the last buffer. */
576 * FUNCTION: find the preferred allocation group for new allocations.
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.
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.
592 * ipbmap - pointer to in-core inode for the block map.
595 * the preferred allocation group number.
597 int dbNextAG(struct inode
*ipbmap
)
604 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
608 /* determine the average number of free blocks within the ags. */
609 avgfree
= (u32
)bmp
->db_nfree
/ bmp
->db_numag
;
612 * if the current preferred ag does not have an active allocator
613 * and has at least average freespace, return it
615 agpref
= bmp
->db_agpref
;
616 if ((atomic_read(&bmp
->db_active
[agpref
]) == 0) &&
617 (bmp
->db_agfree
[agpref
] >= avgfree
))
620 /* From the last preferred ag, find the next one with at least
621 * average free space.
623 for (i
= 0 ; i
< bmp
->db_numag
; i
++, agpref
++) {
624 if (agpref
== bmp
->db_numag
)
627 if (atomic_read(&bmp
->db_active
[agpref
]))
628 /* open file is currently growing in this ag */
630 if (bmp
->db_agfree
[agpref
] >= avgfree
) {
631 /* Return this one */
632 bmp
->db_agpref
= agpref
;
634 } else if (bmp
->db_agfree
[agpref
] > hwm
) {
635 /* Less than avg. freespace, but best so far */
636 hwm
= bmp
->db_agfree
[agpref
];
642 * If no inactive ag was found with average freespace, use the
646 bmp
->db_agpref
= next_best
;
647 /* else leave db_agpref unchanged */
651 /* return the preferred group.
653 return (bmp
->db_agpref
);
659 * FUNCTION: attempt to allocate a specified number of contiguous free
660 * blocks from the working allocation block map.
662 * the block allocation policy uses hints and a multi-step
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.
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.
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.
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.
691 * -ENOSPC - insufficient disk resources
694 int dbAlloc(struct inode
*ip
, s64 hint
, s64 nblocks
, s64
* results
)
697 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
706 /* assert that nblocks is valid */
709 #ifdef _STILL_TO_PORT
710 /* DASD limit check F226941 */
711 if (OVER_LIMIT(ip
, nblocks
))
713 #endif /* _STILL_TO_PORT */
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.
719 l2nb
= BLKSTOL2(nblocks
);
721 bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
723 //retry: /* serialize w.r.t.extendfs() */
724 mapSize
= bmp
->db_mapsize
;
726 /* the hint should be within the map */
727 if (hint
>= mapSize
) {
728 jfs_error(ip
->i_sb
, "dbAlloc: the hint is outside the map");
732 /* if the number of blocks to be allocated is greater than the
733 * allocation group size, try to allocate anywhere.
735 if (l2nb
> bmp
->db_agl2size
) {
738 rc
= dbAllocAny(bmp
, nblocks
, l2nb
, results
);
744 * If no hint, let dbNextAG recommend an allocation group
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.
755 if (blkno
>= bmp
->db_mapsize
)
758 agno
= blkno
>> bmp
->db_agl2size
;
760 /* check if blkno crosses over into a new allocation group.
761 * if so, check if we should allow allocations within this
764 if ((blkno
& (bmp
->db_agsize
- 1)) == 0)
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.
769 if (atomic_read(&bmp
->db_active
[agno
]))
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.
776 if (nblocks
<= BPERDMAP
) {
779 /* get the buffer for the dmap containing the hint.
782 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
783 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
787 dp
= (struct dmap
*) mp
->data
;
789 /* first, try to satisfy the allocation request with the
790 * blocks beginning at the hint.
792 if ((rc
= dbAllocNext(bmp
, dp
, blkno
, (int) nblocks
))
796 mark_metapage_dirty(mp
);
799 release_metapage(mp
);
803 writers
= atomic_read(&bmp
->db_active
[agno
]);
805 ((writers
== 1) && (JFS_IP(ip
)->active_ag
!= agno
))) {
807 * Someone else is writing in this allocation
808 * group. To avoid fragmenting, try another ag
810 release_metapage(mp
);
811 IREAD_UNLOCK(ipbmap
);
815 /* next, try to satisfy the allocation request with blocks
819 dbAllocNear(bmp
, dp
, blkno
, (int) nblocks
, l2nb
, results
))
822 mark_metapage_dirty(mp
);
824 release_metapage(mp
);
828 /* try to satisfy the allocation request with blocks within
829 * the same dmap as the hint.
831 if ((rc
= dbAllocDmapLev(bmp
, dp
, (int) nblocks
, l2nb
, results
))
834 mark_metapage_dirty(mp
);
836 release_metapage(mp
);
840 release_metapage(mp
);
841 IREAD_UNLOCK(ipbmap
);
844 /* try to satisfy the allocation request with blocks within
845 * the same allocation group as the hint.
848 if ((rc
= dbAllocAG(bmp
, agno
, nblocks
, l2nb
, results
)) != -ENOSPC
)
851 IWRITE_UNLOCK(ipbmap
);
856 * Let dbNextAG recommend a preferred allocation group
858 agno
= dbNextAG(ipbmap
);
861 /* Try to allocate within this allocation group. if that fails, try to
862 * allocate anywhere in the map.
864 if ((rc
= dbAllocAG(bmp
, agno
, nblocks
, l2nb
, results
)) == -ENOSPC
)
865 rc
= dbAllocAny(bmp
, nblocks
, l2nb
, results
);
868 IWRITE_UNLOCK(ipbmap
);
873 IREAD_UNLOCK(ipbmap
);
880 * NAME: dbAllocExact()
882 * FUNCTION: try to allocate the requested extent;
885 * ip - pointer to in-core inode;
886 * blkno - extent address;
887 * nblocks - extent length;
891 * -ENOSPC - insufficient disk resources
894 int dbAllocExact(struct inode
*ip
, s64 blkno
, int nblocks
)
897 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
898 struct bmap
*bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
906 * validate extent request:
908 * note: defragfs policy:
909 * max 64 blocks will be moved.
910 * allocation request size must be satisfied from a single dmap.
912 if (nblocks
<= 0 || nblocks
> BPERDMAP
|| blkno
>= bmp
->db_mapsize
) {
913 IREAD_UNLOCK(ipbmap
);
917 if (nblocks
> ((s64
) 1 << bmp
->db_maxfreebud
)) {
918 /* the free space is no longer available */
919 IREAD_UNLOCK(ipbmap
);
923 /* read in the dmap covering the extent */
924 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
925 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
927 IREAD_UNLOCK(ipbmap
);
930 dp
= (struct dmap
*) mp
->data
;
932 /* try to allocate the requested extent */
933 rc
= dbAllocNext(bmp
, dp
, blkno
, nblocks
);
935 IREAD_UNLOCK(ipbmap
);
938 mark_metapage_dirty(mp
);
940 release_metapage(mp
);
949 * FUNCTION: attempt to extend a current allocation by a specified
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.
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
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
975 * -ENOSPC - insufficient disk resources
979 dbReAlloc(struct inode
*ip
,
980 s64 blkno
, s64 nblocks
, s64 addnblocks
, s64
* results
)
984 /* try to extend the allocation in place.
986 if ((rc
= dbExtend(ip
, blkno
, nblocks
, addnblocks
)) == 0) {
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.)
1000 (ip
, blkno
+ nblocks
- 1, addnblocks
+ nblocks
, results
));
1007 * FUNCTION: attempt to extend a current allocation by a specified
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.
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
1020 * addnblocks - number of blocks to add to the allocation.
1024 * -ENOSPC - insufficient disk resources
1027 static int dbExtend(struct inode
*ip
, s64 blkno
, s64 nblocks
, s64 addnblocks
)
1029 struct jfs_sb_info
*sbi
= JFS_SBI(ip
->i_sb
);
1030 s64 lblkno
, lastblkno
, extblkno
;
1032 struct metapage
*mp
;
1035 struct inode
*ipbmap
= sbi
->ipbmap
;
1039 * We don't want a non-aligned extent to cross a page boundary
1041 if (((rel_block
= blkno
& (sbi
->nbperpage
- 1))) &&
1042 (rel_block
+ nblocks
+ addnblocks
> sbi
->nbperpage
))
1045 /* get the last block of the current allocation */
1046 lastblkno
= blkno
+ nblocks
- 1;
1048 /* determine the block number of the block following
1049 * the existing allocation.
1051 extblkno
= lastblkno
+ 1;
1055 /* better be within the file system */
1057 if (lastblkno
< 0 || lastblkno
>= bmp
->db_mapsize
) {
1058 IREAD_UNLOCK(ipbmap
);
1060 "dbExtend: the block is outside the filesystem");
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.
1072 if (addnblocks
> BPERDMAP
|| extblkno
>= bmp
->db_mapsize
||
1073 (extblkno
& (bmp
->db_agsize
- 1)) == 0) {
1074 IREAD_UNLOCK(ipbmap
);
1078 /* get the buffer for the dmap containing the first block
1081 lblkno
= BLKTODMAP(extblkno
, bmp
->db_l2nbperpage
);
1082 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
1084 IREAD_UNLOCK(ipbmap
);
1088 dp
= (struct dmap
*) mp
->data
;
1090 /* try to allocate the blocks immediately following the
1091 * current allocation.
1093 rc
= dbAllocNext(bmp
, dp
, extblkno
, (int) addnblocks
);
1095 IREAD_UNLOCK(ipbmap
);
1097 /* were we successful ? */
1101 /* we were not successful */
1102 release_metapage(mp
);
1110 * NAME: dbAllocNext()
1112 * FUNCTION: attempt to allocate the blocks of the specified block
1113 * range within a dmap.
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.
1123 * -ENOSPC - insufficient disk resources
1126 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1128 static int dbAllocNext(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
1131 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, nw
;
1136 if (dp
->tree
.leafidx
!= cpu_to_le32(LEAFIND
)) {
1137 jfs_error(bmp
->db_ipbmap
->i_sb
,
1138 "dbAllocNext: Corrupt dmap page");
1142 /* pick up a pointer to the leaves of the dmap tree.
1144 leaf
= dp
->tree
.stree
+ le32_to_cpu(dp
->tree
.leafidx
);
1146 /* determine the bit number and word within the dmap of the
1149 dbitno
= blkno
& (BPERDMAP
- 1);
1150 word
= dbitno
>> L2DBWORD
;
1152 /* check if the specified block range is contained within
1155 if (dbitno
+ nblocks
> BPERDMAP
)
1158 /* check if the starting leaf indicates that anything
1161 if (leaf
[word
] == NOFREE
)
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.
1177 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
1178 /* determine the bit number within the word and
1179 * the number of bits within the word.
1181 wbitno
= dbitno
& (DBWORD
- 1);
1182 nb
= min(rembits
, DBWORD
- wbitno
);
1184 /* check if only part of the word is to be examined.
1187 /* check if the bits are free.
1189 mask
= (ONES
<< (DBWORD
- nb
) >> wbitno
);
1190 if ((mask
& ~le32_to_cpu(dp
->wmap
[word
])) != mask
)
1195 /* one or more dmap words are fully contained
1196 * within the block range. determine how many
1197 * words and how many bits.
1199 nwords
= rembits
>> L2DBWORD
;
1200 nb
= nwords
<< L2DBWORD
;
1202 /* now examine the appropriate leaves to determine
1203 * if the blocks are free.
1205 while (nwords
> 0) {
1206 /* does the leaf describe any free space ?
1208 if (leaf
[word
] < BUDMIN
)
1211 /* determine the l2 number of bits provided
1215 min((int)leaf
[word
], NLSTOL2BSZ(nwords
));
1217 /* determine how many words were handled.
1219 nw
= BUDSIZE(l2size
, BUDMIN
);
1227 /* allocate the blocks.
1229 return (dbAllocDmap(bmp
, dp
, blkno
, nblocks
));
1234 * NAME: dbAllocNear()
1236 * FUNCTION: attempt to allocate a number of contiguous free blocks near
1237 * a specified block (hint) within a dmap.
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.
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.
1255 * -ENOSPC - insufficient disk resources
1258 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1261 dbAllocNear(struct bmap
* bmp
,
1262 struct dmap
* dp
, s64 blkno
, int nblocks
, int l2nb
, s64
* results
)
1264 int word
, lword
, rc
;
1267 if (dp
->tree
.leafidx
!= cpu_to_le32(LEAFIND
)) {
1268 jfs_error(bmp
->db_ipbmap
->i_sb
,
1269 "dbAllocNear: Corrupt dmap page");
1273 leaf
= dp
->tree
.stree
+ le32_to_cpu(dp
->tree
.leafidx
);
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.
1279 word
= (blkno
& (BPERDMAP
- 1)) >> L2DBWORD
;
1280 lword
= min(word
+ 4, LPERDMAP
);
1282 /* examine the leaves for sufficient free space.
1284 for (; word
< lword
; word
++) {
1285 /* does the leaf describe sufficient free space ?
1287 if (leaf
[word
] < l2nb
)
1290 /* determine the block number within the file system
1291 * of the first block described by this dmap word.
1293 blkno
= le64_to_cpu(dp
->start
) + (word
<< L2DBWORD
);
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
1300 if (leaf
[word
] < BUDMIN
)
1302 dbFindBits(le32_to_cpu(dp
->wmap
[word
]), l2nb
);
1304 /* allocate the blocks.
1306 if ((rc
= dbAllocDmap(bmp
, dp
, blkno
, nblocks
)) == 0)
1319 * FUNCTION: attempt to allocate the specified number of contiguous
1320 * free blocks within the specified allocation group.
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.
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.
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.
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.
1365 * -ENOSPC - insufficient disk resources
1368 * note: IWRITE_LOCK(ipmap) held on entry/exit;
1371 dbAllocAG(struct bmap
* bmp
, int agno
, s64 nblocks
, int l2nb
, s64
* results
)
1373 struct metapage
*mp
;
1374 struct dmapctl
*dcp
;
1375 int rc
, ti
, i
, k
, m
, n
, agperlev
;
1379 /* allocation request should not be for more than the
1380 * allocation group size.
1382 if (l2nb
> bmp
->db_agl2size
) {
1383 jfs_error(bmp
->db_ipbmap
->i_sb
,
1384 "dbAllocAG: allocation request is larger than the "
1385 "allocation group size");
1389 /* determine the starting block number of the allocation
1392 blkno
= (s64
) agno
<< bmp
->db_agl2size
;
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.
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.
1411 if (bmp
->db_agsize
== BPERDMAP
1412 || bmp
->db_agfree
[agno
] == bmp
->db_agsize
) {
1413 rc
= dbAllocCtl(bmp
, nblocks
, l2nb
, blkno
, results
);
1414 if ((rc
== -ENOSPC
) &&
1415 (bmp
->db_agfree
[agno
] == bmp
->db_agsize
)) {
1416 printk(KERN_ERR
"blkno = %Lx, blocks = %Lx\n",
1417 (unsigned long long) blkno
,
1418 (unsigned long long) nblocks
);
1419 jfs_error(bmp
->db_ipbmap
->i_sb
,
1420 "dbAllocAG: dbAllocCtl failed in free AG");
1425 /* the buffer for the dmap control page that fully describes the
1428 lblkno
= BLKTOCTL(blkno
, bmp
->db_l2nbperpage
, bmp
->db_aglevel
);
1429 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1432 dcp
= (struct dmapctl
*) mp
->data
;
1433 budmin
= dcp
->budmin
;
1435 if (dcp
->leafidx
!= cpu_to_le32(CTLLEAFIND
)) {
1436 jfs_error(bmp
->db_ipbmap
->i_sb
,
1437 "dbAllocAG: Corrupt dmapctl page");
1438 release_metapage(mp
);
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.
1450 (1 << (L2LPERCTL
- (bmp
->db_agheigth
<< 1))) / bmp
->db_agwidth
;
1451 ti
= bmp
->db_agstart
+ bmp
->db_agwidth
* (agno
& (agperlev
- 1));
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
1459 for (i
= 0; i
< bmp
->db_agwidth
; i
++, ti
++) {
1460 /* is there sufficient free space ?
1462 if (l2nb
> dcp
->stree
[ti
])
1465 /* sufficient free space found in a subtree. now search down
1466 * the subtree to find the leftmost leaf that describes this
1469 for (k
= bmp
->db_agheigth
; k
> 0; k
--) {
1470 for (n
= 0, m
= (ti
<< 2) + 1; n
< 4; n
++) {
1471 if (l2nb
<= dcp
->stree
[m
+ n
]) {
1477 jfs_error(bmp
->db_ipbmap
->i_sb
,
1478 "dbAllocAG: failed descending stree");
1479 release_metapage(mp
);
1484 /* determine the block number within the file system
1485 * that corresponds to this leaf.
1487 if (bmp
->db_aglevel
== 2)
1489 else if (bmp
->db_aglevel
== 1)
1490 blkno
&= ~(MAXL1SIZE
- 1);
1491 else /* bmp->db_aglevel == 0 */
1492 blkno
&= ~(MAXL0SIZE
- 1);
1495 ((s64
) (ti
- le32_to_cpu(dcp
->leafidx
))) << budmin
;
1497 /* release the buffer in preparation for going down
1498 * the next level of dmap control pages.
1500 release_metapage(mp
);
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.
1507 if (l2nb
< budmin
) {
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.
1514 dbFindCtl(bmp
, l2nb
, bmp
->db_aglevel
- 1,
1516 if (rc
== -ENOSPC
) {
1517 jfs_error(bmp
->db_ipbmap
->i_sb
,
1518 "dbAllocAG: control page "
1526 /* allocate the blocks.
1528 rc
= dbAllocCtl(bmp
, nblocks
, l2nb
, blkno
, results
);
1529 if (rc
== -ENOSPC
) {
1530 jfs_error(bmp
->db_ipbmap
->i_sb
,
1531 "dbAllocAG: unable to allocate blocks");
1537 /* no space in the allocation group. release the buffer and
1540 release_metapage(mp
);
1547 * NAME: dbAllocAny()
1549 * FUNCTION: attempt to allocate the specified number of contiguous
1550 * free blocks anywhere in the file system.
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.
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.
1567 * -ENOSPC - insufficient disk resources
1570 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1572 static int dbAllocAny(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64
* results
)
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.
1583 if ((rc
= dbFindCtl(bmp
, l2nb
, bmp
->db_maxlevel
, &blkno
)))
1586 /* allocate the blocks.
1588 rc
= dbAllocCtl(bmp
, nblocks
, l2nb
, blkno
, results
);
1589 if (rc
== -ENOSPC
) {
1590 jfs_error(bmp
->db_ipbmap
->i_sb
,
1591 "dbAllocAny: unable to allocate blocks");
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.
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.
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.
1621 * -ENOSPC - insufficient disk resources
1624 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1626 static int dbFindCtl(struct bmap
* bmp
, int l2nb
, int level
, s64
* blkno
)
1628 int rc
, leafidx
, lev
;
1630 struct dmapctl
*dcp
;
1632 struct metapage
*mp
;
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.
1639 for (lev
= level
, b
= *blkno
; lev
>= 0; lev
--) {
1640 /* get the buffer of the dmap control page for the block
1641 * number and level (i.e. L0, L1, L2).
1643 lblkno
= BLKTOCTL(b
, bmp
->db_l2nbperpage
, lev
);
1644 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1647 dcp
= (struct dmapctl
*) mp
->data
;
1648 budmin
= dcp
->budmin
;
1650 if (dcp
->leafidx
!= cpu_to_le32(CTLLEAFIND
)) {
1651 jfs_error(bmp
->db_ipbmap
->i_sb
,
1652 "dbFindCtl: Corrupt dmapctl page");
1653 release_metapage(mp
);
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.
1662 rc
= dbFindLeaf((dmtree_t
*) dcp
, l2nb
, &leafidx
);
1664 /* release the buffer.
1666 release_metapage(mp
);
1672 jfs_error(bmp
->db_ipbmap
->i_sb
,
1673 "dbFindCtl: dmap inconsistent");
1679 /* adjust the block number to reflect the location within
1680 * the dmap control page (i.e. the leaf) at which free
1683 b
+= (((s64
) leafidx
) << budmin
);
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
1700 * NAME: dbAllocCtl()
1702 * FUNCTION: attempt to allocate a specified number of contiguous
1703 * blocks starting within a specific dmap.
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.
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
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).
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
1734 * results - on successful return, set to the starting block number
1735 * of the newly allocated range.
1739 * -ENOSPC - insufficient disk resources
1742 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1745 dbAllocCtl(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64 blkno
, s64
* results
)
1749 struct metapage
*mp
;
1752 /* check if the allocation request is confined to a single dmap.
1754 if (l2nb
<= L2BPERDMAP
) {
1755 /* get the buffer for the dmap.
1757 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
1758 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1761 dp
= (struct dmap
*) mp
->data
;
1763 /* try to allocate the blocks.
1765 rc
= dbAllocDmapLev(bmp
, dp
, (int) nblocks
, l2nb
, results
);
1767 mark_metapage_dirty(mp
);
1769 release_metapage(mp
);
1774 /* allocation request involving multiple dmaps. it must start on
1777 assert((blkno
& (BPERDMAP
- 1)) == 0);
1779 /* allocate the blocks dmap by dmap.
1781 for (n
= nblocks
, b
= blkno
; n
> 0; n
-= nb
, b
+= nb
) {
1782 /* get the buffer for the dmap.
1784 lblkno
= BLKTODMAP(b
, bmp
->db_l2nbperpage
);
1785 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1790 dp
= (struct dmap
*) mp
->data
;
1792 /* the dmap better be all free.
1794 if (dp
->tree
.stree
[ROOT
] != L2BPERDMAP
) {
1795 release_metapage(mp
);
1796 jfs_error(bmp
->db_ipbmap
->i_sb
,
1797 "dbAllocCtl: the dmap is not all free");
1802 /* determine how many blocks to allocate from this dmap.
1804 nb
= min(n
, (s64
)BPERDMAP
);
1806 /* allocate the blocks from the dmap.
1808 if ((rc
= dbAllocDmap(bmp
, dp
, b
, nb
))) {
1809 release_metapage(mp
);
1813 /* write the buffer.
1818 /* set the results (starting block number) and return.
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.
1831 /* try to backout the allocations dmap by dmap.
1833 for (n
= nblocks
- n
, b
= blkno
; n
> 0;
1834 n
-= BPERDMAP
, b
+= BPERDMAP
) {
1835 /* get the buffer for this dmap.
1837 lblkno
= BLKTODMAP(b
, bmp
->db_l2nbperpage
);
1838 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1840 /* could not back out. mark the file system
1841 * to indicate that we have leaked blocks.
1843 jfs_error(bmp
->db_ipbmap
->i_sb
,
1844 "dbAllocCtl: I/O Error: Block Leakage.");
1847 dp
= (struct dmap
*) mp
->data
;
1849 /* free the blocks is this dmap.
1851 if (dbFreeDmap(bmp
, dp
, b
, BPERDMAP
)) {
1852 /* could not back out. mark the file system
1853 * to indicate that we have leaked blocks.
1855 release_metapage(mp
);
1856 jfs_error(bmp
->db_ipbmap
->i_sb
,
1857 "dbAllocCtl: Block Leakage.");
1861 /* write the buffer.
1871 * NAME: dbAllocDmapLev()
1873 * FUNCTION: attempt to allocate a specified number of contiguous blocks
1874 * from a specified dmap.
1876 * this routine checks if the contiguous blocks are available.
1877 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is
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.
1890 * -ENOSPC - insufficient disk resources
1893 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1894 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1897 dbAllocDmapLev(struct bmap
* bmp
,
1898 struct dmap
* dp
, int nblocks
, int l2nb
, s64
* results
)
1903 /* can't be more than a dmaps worth of blocks */
1904 assert(l2nb
<= L2BPERDMAP
);
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.
1910 if (dbFindLeaf((dmtree_t
*) & dp
->tree
, l2nb
, &leafidx
))
1913 /* determine the block number within the file system corresponding
1914 * to the leaf at which free space was found.
1916 blkno
= le64_to_cpu(dp
->start
) + (leafidx
<< L2DBWORD
);
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.
1922 if (dp
->tree
.stree
[leafidx
+ LEAFIND
] < BUDMIN
)
1923 blkno
+= dbFindBits(le32_to_cpu(dp
->wmap
[leafidx
]), l2nb
);
1925 /* allocate the blocks */
1926 if ((rc
= dbAllocDmap(bmp
, dp
, blkno
, nblocks
)) == 0)
1934 * NAME: dbAllocDmap()
1936 * FUNCTION: adjust the disk allocation map to reflect the allocation
1937 * of a specified block range within a dmap.
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
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.
1958 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
1960 static int dbAllocDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
1966 /* save the current value of the root (i.e. maximum free string)
1969 oldroot
= dp
->tree
.stree
[ROOT
];
1971 /* allocate the specified (blocks) bits */
1972 dbAllocBits(bmp
, dp
, blkno
, nblocks
);
1974 /* if the root has not changed, done. */
1975 if (dp
->tree
.stree
[ROOT
] == oldroot
)
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).
1982 if ((rc
= dbAdjCtl(bmp
, blkno
, dp
->tree
.stree
[ROOT
], 1, 0)))
1983 dbFreeBits(bmp
, dp
, blkno
, nblocks
);
1990 * NAME: dbFreeDmap()
1992 * FUNCTION: adjust the disk allocation map to reflect the allocation
1993 * of a specified block range within a dmap.
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.
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.
2013 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2015 static int dbFreeDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
2021 /* save the current value of the root (i.e. maximum free string)
2024 oldroot
= dp
->tree
.stree
[ROOT
];
2026 /* free the specified (blocks) bits */
2027 rc
= dbFreeBits(bmp
, dp
, blkno
, nblocks
);
2029 /* if error or the root has not changed, done. */
2030 if (rc
|| (dp
->tree
.stree
[ROOT
] == oldroot
))
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.
2037 if ((rc
= dbAdjCtl(bmp
, blkno
, dp
->tree
.stree
[ROOT
], 0, 0))) {
2038 word
= (blkno
& (BPERDMAP
- 1)) >> L2DBWORD
;
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
2045 if (dp
->tree
.stree
[word
] == NOFREE
)
2046 dbBackSplit((dmtree_t
*) & dp
->tree
, word
);
2048 dbAllocBits(bmp
, dp
, blkno
, nblocks
);
2056 * NAME: dbAllocBits()
2058 * FUNCTION: allocate a specified block range from a dmap.
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.
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.
2073 * RETURN VALUES: none
2075 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2077 static void dbAllocBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
2080 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, nw
, agno
;
2081 dmtree_t
*tp
= (dmtree_t
*) & dp
->tree
;
2085 /* pick up a pointer to the leaves of the dmap tree */
2086 leaf
= dp
->tree
.stree
+ LEAFIND
;
2088 /* determine the bit number and word within the dmap of the
2091 dbitno
= blkno
& (BPERDMAP
- 1);
2092 word
= dbitno
>> L2DBWORD
;
2094 /* block range better be within the dmap */
2095 assert(dbitno
+ nblocks
<= BPERDMAP
);
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.
2110 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
2111 /* determine the bit number within the word and
2112 * the number of bits within the word.
2114 wbitno
= dbitno
& (DBWORD
- 1);
2115 nb
= min(rembits
, DBWORD
- wbitno
);
2117 /* check if only part of a word is to be allocated.
2120 /* allocate (set to 1) the appropriate bits within
2123 dp
->wmap
[word
] |= cpu_to_le32(ONES
<< (DBWORD
- nb
)
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.
2131 dbSplit(tp
, word
, BUDMIN
,
2132 dbMaxBud((u8
*) & dp
->wmap
[word
]));
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
2141 nwords
= rembits
>> L2DBWORD
;
2142 memset(&dp
->wmap
[word
], (int) ONES
, nwords
* 4);
2144 /* determine how many bits.
2146 nb
= nwords
<< L2DBWORD
;
2148 /* now update the appropriate leaves to reflect
2149 * the allocated words.
2151 for (; nwords
> 0; nwords
-= nw
) {
2152 if (leaf
[word
] < BUDMIN
) {
2153 jfs_error(bmp
->db_ipbmap
->i_sb
,
2154 "dbAllocBits: leaf page "
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.
2164 size
= min((int)leaf
[word
], NLSTOL2BSZ(nwords
));
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).
2172 dbSplit(tp
, word
, size
, NOFREE
);
2174 /* get the number of dmap words handled */
2175 nw
= BUDSIZE(size
, BUDMIN
);
2181 /* update the free count for this dmap */
2182 dp
->nfree
= cpu_to_le32(le32_to_cpu(dp
->nfree
) - nblocks
);
2186 /* if this allocation group is completely free,
2187 * update the maximum allocation group number if this allocation
2188 * group is the new max.
2190 agno
= blkno
>> bmp
->db_agl2size
;
2191 if (agno
> bmp
->db_maxag
)
2192 bmp
->db_maxag
= agno
;
2194 /* update the free count for the allocation group and map */
2195 bmp
->db_agfree
[agno
] -= nblocks
;
2196 bmp
->db_nfree
-= nblocks
;
2203 * NAME: dbFreeBits()
2205 * FUNCTION: free a specified block range from a dmap.
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.
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.
2220 * RETURN VALUES: 0 for success
2222 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2224 static int dbFreeBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
2227 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, nw
, agno
;
2228 dmtree_t
*tp
= (dmtree_t
*) & dp
->tree
;
2232 /* determine the bit number and word within the dmap of the
2235 dbitno
= blkno
& (BPERDMAP
- 1);
2236 word
= dbitno
>> L2DBWORD
;
2238 /* block range better be within the dmap.
2240 assert(dbitno
+ nblocks
<= BPERDMAP
);
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.
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
2259 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
2260 /* determine the bit number within the word and
2261 * the number of bits within the word.
2263 wbitno
= dbitno
& (DBWORD
- 1);
2264 nb
= min(rembits
, DBWORD
- wbitno
);
2266 /* check if only part of a word is to be freed.
2269 /* free (zero) the appropriate bits within this
2273 cpu_to_le32(~(ONES
<< (DBWORD
- nb
)
2276 /* update the leaf for this dmap word.
2278 rc
= dbJoin(tp
, word
,
2279 dbMaxBud((u8
*) & dp
->wmap
[word
]));
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.
2289 nwords
= rembits
>> L2DBWORD
;
2290 memset(&dp
->wmap
[word
], 0, nwords
* 4);
2292 /* determine how many bits.
2294 nb
= nwords
<< L2DBWORD
;
2296 /* now update the appropriate leaves to reflect
2299 for (; nwords
> 0; nwords
-= nw
) {
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.
2307 (word
, L2LPERDMAP
, BUDMIN
),
2308 NLSTOL2BSZ(nwords
));
2312 rc
= dbJoin(tp
, word
, size
);
2316 /* get the number of dmap words handled.
2318 nw
= BUDSIZE(size
, BUDMIN
);
2324 /* update the free count for this dmap.
2326 dp
->nfree
= cpu_to_le32(le32_to_cpu(dp
->nfree
) + nblocks
);
2330 /* update the free count for the allocation group and
2333 agno
= blkno
>> bmp
->db_agl2size
;
2334 bmp
->db_nfree
+= nblocks
;
2335 bmp
->db_agfree
[agno
] += nblocks
;
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.
2342 if ((bmp
->db_agfree
[agno
] == bmp
->db_agsize
&& agno
== bmp
->db_maxag
) ||
2343 (agno
== bmp
->db_numag
- 1 &&
2344 bmp
->db_agfree
[agno
] == (bmp
-> db_mapsize
& (BPERDMAP
- 1)))) {
2345 while (bmp
->db_maxag
> 0) {
2347 if (bmp
->db_agfree
[bmp
->db_maxag
] !=
2352 /* re-establish the allocation group preference if the
2353 * current preference is right of the maximum allocation
2356 if (bmp
->db_agpref
> bmp
->db_maxag
)
2357 bmp
->db_agpref
= bmp
->db_maxag
;
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.
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.
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
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
2397 * alloc - 'true' if adjustment is due to an allocation.
2398 * level - current level of dmap control page (i.e. L0, L1, L2) to
2405 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2408 dbAdjCtl(struct bmap
* bmp
, s64 blkno
, int newval
, int alloc
, int level
)
2410 struct metapage
*mp
;
2414 struct dmapctl
*dcp
;
2417 /* get the buffer for the dmap control page for the specified
2418 * block number and control page level.
2420 lblkno
= BLKTOCTL(blkno
, bmp
->db_l2nbperpage
, level
);
2421 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
2424 dcp
= (struct dmapctl
*) mp
->data
;
2426 if (dcp
->leafidx
!= cpu_to_le32(CTLLEAFIND
)) {
2427 jfs_error(bmp
->db_ipbmap
->i_sb
,
2428 "dbAdjCtl: Corrupt dmapctl page");
2429 release_metapage(mp
);
2433 /* determine the leaf number corresponding to the block and
2434 * the index within the dmap control tree.
2436 leafno
= BLKTOCTLLEAF(blkno
, dcp
->budmin
);
2437 ti
= leafno
+ le32_to_cpu(dcp
->leafidx
);
2439 /* save the current leaf value and the current root level (i.e.
2440 * maximum l2 free string described by this dmapctl).
2442 oldval
= dcp
->stree
[ti
];
2443 oldroot
= dcp
->stree
[ROOT
];
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.
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.
2464 if (oldval
== NOFREE
) {
2465 rc
= dbBackSplit((dmtree_t
*) dcp
, leafno
);
2468 oldval
= dcp
->stree
[ti
];
2470 dbSplit((dmtree_t
*) dcp
, leafno
, dcp
->budmin
, newval
);
2472 rc
= dbJoin((dmtree_t
*) dcp
, leafno
, newval
);
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.
2484 if (dcp
->stree
[ROOT
] != oldroot
) {
2485 /* are we below the top level of the map. if so,
2486 * bubble the root up to the next higher level.
2488 if (level
< bmp
->db_maxlevel
) {
2489 /* bubble up the new root of this dmap control page to
2493 dbAdjCtl(bmp
, blkno
, dcp
->stree
[ROOT
], alloc
,
2495 /* something went wrong in bubbling up the new
2496 * root value, so backout the changes to the
2497 * current dmap control page.
2500 dbJoin((dmtree_t
*) dcp
, leafno
,
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.
2509 if (dcp
->stree
[ti
] == NOFREE
)
2510 dbBackSplit((dmtree_t
*)
2512 dbSplit((dmtree_t
*) dcp
, leafno
,
2513 dcp
->budmin
, oldval
);
2516 /* release the buffer and return the error.
2518 release_metapage(mp
);
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.
2526 assert(level
== bmp
->db_maxlevel
);
2527 if (bmp
->db_maxfreebud
!= oldroot
) {
2528 jfs_error(bmp
->db_ipbmap
->i_sb
,
2529 "dbAdjCtl: the maximum free buddy is "
2530 "not the old root");
2532 bmp
->db_maxfreebud
= dcp
->stree
[ROOT
];
2536 /* write the buffer.
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.
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.
2558 * RETURN VALUES: none
2560 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2562 static void dbSplit(dmtree_t
* tp
, int leafno
, int splitsz
, int newval
)
2566 s8
*leaf
= tp
->dmt_stree
+ le32_to_cpu(tp
->dmt_leafidx
);
2568 /* check if the leaf needs to be split.
2570 if (leaf
[leafno
] > tp
->dmt_budmin
) {
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.
2576 cursz
= leaf
[leafno
] - 1;
2577 budsz
= BUDSIZE(cursz
, tp
->dmt_budmin
);
2579 /* split until we reach the specified size.
2581 while (cursz
>= splitsz
) {
2582 /* update the buddy's leaf with its new value.
2584 dbAdjTree(tp
, leafno
^ budsz
, cursz
);
2586 /* on to the next size and buddy.
2593 /* adjust the dmap tree to reflect the specified leaf's new
2596 dbAdjTree(tp
, leafno
, newval
);
2601 * NAME: dbBackSplit()
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.
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.
2620 * tp - pointer to the tree containing the leaf.
2621 * leafno - the number of the leaf to be updated.
2623 * RETURN VALUES: none
2625 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2627 static int dbBackSplit(dmtree_t
* tp
, int leafno
)
2629 int budsz
, bud
, w
, bsz
, size
;
2631 s8
*leaf
= tp
->dmt_stree
+ le32_to_cpu(tp
->dmt_leafidx
);
2633 /* leaf should be part (not first part) of a binary
2636 assert(leaf
[leafno
] == NOFREE
);
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.
2643 * determine maximum possible l2 size for the specified leaf.
2646 LITOL2BSZ(leafno
, le32_to_cpu(tp
->dmt_l2nleafs
),
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.
2653 budsz
= BUDSIZE(size
, tp
->dmt_budmin
);
2657 while (leaf
[leafno
] == NOFREE
) {
2658 /* find the leftmost buddy leaf.
2660 for (w
= leafno
, bsz
= budsz
;; bsz
<<= 1,
2661 w
= (w
< bud
) ? w
: bud
) {
2662 if (bsz
>= le32_to_cpu(tp
->dmt_nleafs
)) {
2663 jfs_err("JFS: block map error in dbBackSplit");
2667 /* determine the buddy.
2671 /* check if this buddy is the start of the system.
2673 if (leaf
[bud
] != NOFREE
) {
2674 /* split the leaf at the start of the
2677 cursz
= leaf
[bud
] - 1;
2678 dbSplit(tp
, bud
, cursz
, cursz
);
2684 if (leaf
[leafno
] != size
) {
2685 jfs_err("JFS: wrong leaf value in dbBackSplit");
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.
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.
2704 * RETURN VALUES: none
2706 static int dbJoin(dmtree_t
* tp
, int leafno
, int newval
)
2711 /* can the new leaf value require a join with other leaves ?
2713 if (newval
>= tp
->dmt_budmin
) {
2714 /* pickup a pointer to the leaves of the tree.
2716 leaf
= tp
->dmt_stree
+ le32_to_cpu(tp
->dmt_leafidx
);
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
2728 * get the buddy size (number of words covered) of
2731 budsz
= BUDSIZE(newval
, tp
->dmt_budmin
);
2735 while (budsz
< le32_to_cpu(tp
->dmt_nleafs
)) {
2736 /* get the buddy leaf.
2738 buddy
= leafno
^ budsz
;
2740 /* if the leaf's new value is greater than its
2741 * buddy's value, we join no more.
2743 if (newval
> leaf
[buddy
])
2746 /* It shouldn't be less */
2747 if (newval
< leaf
[buddy
])
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
2758 if (leafno
< buddy
) {
2759 /* leafno is the left buddy.
2761 dbAdjTree(tp
, buddy
, NOFREE
);
2763 /* buddy is the left buddy and becomes
2766 dbAdjTree(tp
, leafno
, NOFREE
);
2770 /* on to try the next join.
2777 /* update the leaf value.
2779 dbAdjTree(tp
, leafno
, newval
);
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
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.
2798 * RETURN VALUES: none
2800 static void dbAdjTree(dmtree_t
* tp
, int leafno
, int newval
)
2805 /* pick up the index of the leaf for this leafno.
2807 lp
= leafno
+ le32_to_cpu(tp
->dmt_leafidx
);
2809 /* is the current value the same as the old value ? if so,
2810 * there is nothing to do.
2812 if (tp
->dmt_stree
[lp
] == newval
)
2815 /* set the new value.
2817 tp
->dmt_stree
[lp
] = newval
;
2819 /* bubble the new value up the tree as required.
2821 for (k
= 0; k
< le32_to_cpu(tp
->dmt_height
); k
++) {
2822 /* get the index of the first leaf of the 4 leaf
2823 * group containing the specified leaf (leafno).
2825 lp
= ((lp
- 1) & ~0x03) + 1;
2827 /* get the index of the parent of this 4 leaf group.
2831 /* determine the maximum of the 4 leaves.
2833 max
= TREEMAX(&tp
->dmt_stree
[lp
]);
2835 /* if the maximum of the 4 is the same as the
2836 * parent's value, we're done.
2838 if (tp
->dmt_stree
[pp
] == max
)
2841 /* parent gets new value.
2843 tp
->dmt_stree
[pp
] = max
;
2845 /* parent becomes leaf for next go-round.
2853 * NAME: dbFindLeaf()
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.
2859 * the search starts at the top of the dmtree_t tree and
2860 * proceeds down the tree to the leftmost leaf with sufficient
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.
2872 * -ENOSPC - insufficient free blocks.
2874 static int dbFindLeaf(dmtree_t
* tp
, int l2nb
, int *leafidx
)
2876 int ti
, n
= 0, k
, x
= 0;
2878 /* first check the root of the tree to see if there is
2879 * sufficient free space.
2881 if (l2nb
> tp
->dmt_stree
[ROOT
])
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.
2888 for (k
= le32_to_cpu(tp
->dmt_height
), ti
= 1;
2889 k
> 0; k
--, ti
= ((ti
+ n
) << 2) + 1) {
2890 /* search the four nodes at this level, starting from
2893 for (x
= ti
, n
= 0; n
< 4; n
++) {
2894 /* sufficient free space found. move to the next
2895 * level (or quit if this is the last level).
2897 if (l2nb
<= tp
->dmt_stree
[x
+ n
])
2901 /* better have found something since the higher
2902 * levels of the tree said it was here.
2907 /* set the return to the leftmost leaf describing sufficient
2910 *leafidx
= x
+ n
- le32_to_cpu(tp
->dmt_leafidx
);
2917 * NAME: dbFindBits()
2919 * FUNCTION: find a specified number of binary buddy free bits within a
2920 * dmap bitmap word value.
2922 * this routine searches the bitmap value for (1 << l2nb) free
2923 * bits at (1 << l2nb) alignments within the value.
2926 * word - dmap bitmap word value.
2927 * l2nb - number of free bits specified as a log2 number.
2930 * starting bit number of free bits.
2932 static int dbFindBits(u32 word
, int l2nb
)
2937 /* get the number of bits.
2940 assert(nb
<= DBWORD
);
2942 /* complement the word so we can use a mask (i.e. 0s represent
2943 * free bits) and compute the mask.
2946 mask
= ONES
<< (DBWORD
- nb
);
2948 /* scan the word for nb free bits at nb alignments.
2950 for (bitno
= 0; mask
!= 0; bitno
+= nb
, mask
>>= nb
) {
2951 if ((mask
& word
) == mask
)
2957 /* return the bit number.
2964 * NAME: dbMaxBud(u8 *cp)
2966 * FUNCTION: determine the largest binary buddy string of free
2967 * bits within 32-bits of the map.
2970 * cp - pointer to the 32-bit value.
2973 * largest binary buddy of free bits within a dmap word.
2975 static int dbMaxBud(u8
* cp
)
2977 signed char tmp1
, tmp2
;
2979 /* check if the wmap word is all free. if so, the
2980 * free buddy size is BUDMIN.
2982 if (*((uint
*) cp
) == 0)
2985 /* check if the wmap word is half free. if so, the
2986 * free buddy size is BUDMIN-1.
2988 if (*((u16
*) cp
) == 0 || *((u16
*) cp
+ 1) == 0)
2989 return (BUDMIN
- 1);
2991 /* not all free or half free. determine the free buddy
2992 * size thru table lookup using quarters of the wmap word.
2994 tmp1
= max(budtab
[cp
[2]], budtab
[cp
[3]]);
2995 tmp2
= max(budtab
[cp
[0]], budtab
[cp
[1]]);
2996 return (max(tmp1
, tmp2
));
3001 * NAME: cnttz(uint word)
3003 * FUNCTION: determine the number of trailing zeros within a 32-bit
3007 * value - 32-bit value to be examined.
3010 * count of trailing zeros
3012 static int cnttz(u32 word
)
3016 for (n
= 0; n
< 32; n
++, word
>>= 1) {
3026 * NAME: cntlz(u32 value)
3028 * FUNCTION: determine the number of leading zeros within a 32-bit
3032 * value - 32-bit value to be examined.
3035 * count of leading zeros
3037 static int cntlz(u32 value
)
3041 for (n
= 0; n
< 32; n
++, value
<<= 1) {
3042 if (value
& HIGHORDER
)
3050 * NAME: blkstol2(s64 nb)
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.
3057 * nb - number of blocks
3060 * log2 number of blocks
3062 static int blkstol2(s64 nb
)
3065 s64 mask
; /* meant to be signed */
3067 mask
= (s64
) 1 << (64 - 1);
3069 /* count the leading bits.
3071 for (l2nb
= 0; l2nb
< 64; l2nb
++, mask
>>= 1) {
3072 /* leading bit found.
3075 /* determine the l2 value.
3077 l2nb
= (64 - 1) - l2nb
;
3079 /* check if we need to round up.
3088 return 0; /* fix compiler warning */
3093 * NAME: dbAllocBottomUp()
3095 * FUNCTION: alloc the specified block range from the working block
3098 * the blocks will be alloc from the working map one dmap
3102 * ip - pointer to in-core inode;
3103 * blkno - starting block number to be freed.
3104 * nblocks - number of blocks to be freed.
3110 int dbAllocBottomUp(struct inode
*ip
, s64 blkno
, s64 nblocks
)
3112 struct metapage
*mp
;
3116 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
3117 struct bmap
*bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
3121 /* block to be allocated better be within the mapsize. */
3122 ASSERT(nblocks
<= bmp
->db_mapsize
- blkno
);
3125 * allocate the blocks a dmap at a time.
3128 for (rem
= nblocks
; rem
> 0; rem
-= nb
, blkno
+= nb
) {
3129 /* release previous dmap if any */
3134 /* get the buffer for the current dmap. */
3135 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
3136 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
3138 IREAD_UNLOCK(ipbmap
);
3141 dp
= (struct dmap
*) mp
->data
;
3143 /* determine the number of blocks to be allocated from
3146 nb
= min(rem
, BPERDMAP
- (blkno
& (BPERDMAP
- 1)));
3148 /* allocate the blocks. */
3149 if ((rc
= dbAllocDmapBU(bmp
, dp
, blkno
, nb
))) {
3150 release_metapage(mp
);
3151 IREAD_UNLOCK(ipbmap
);
3156 /* write the last buffer. */
3159 IREAD_UNLOCK(ipbmap
);
3165 static int dbAllocDmapBU(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
3169 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, agno
;
3171 struct dmaptree
*tp
= (struct dmaptree
*) & dp
->tree
;
3173 /* save the current value of the root (i.e. maximum free string)
3176 oldroot
= tp
->stree
[ROOT
];
3178 /* pick up a pointer to the leaves of the dmap tree */
3179 leaf
= tp
->stree
+ LEAFIND
;
3181 /* determine the bit number and word within the dmap of the
3184 dbitno
= blkno
& (BPERDMAP
- 1);
3185 word
= dbitno
>> L2DBWORD
;
3187 /* block range better be within the dmap */
3188 assert(dbitno
+ nblocks
<= BPERDMAP
);
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.
3203 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
3204 /* determine the bit number within the word and
3205 * the number of bits within the word.
3207 wbitno
= dbitno
& (DBWORD
- 1);
3208 nb
= min(rembits
, DBWORD
- wbitno
);
3210 /* check if only part of a word is to be allocated.
3213 /* allocate (set to 1) the appropriate bits within
3216 dp
->wmap
[word
] |= cpu_to_le32(ONES
<< (DBWORD
- nb
)
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
3226 nwords
= rembits
>> L2DBWORD
;
3227 memset(&dp
->wmap
[word
], (int) ONES
, nwords
* 4);
3229 /* determine how many bits */
3230 nb
= nwords
<< L2DBWORD
;
3235 /* update the free count for this dmap */
3236 dp
->nfree
= cpu_to_le32(le32_to_cpu(dp
->nfree
) - nblocks
);
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.
3247 agno
= blkno
>> bmp
->db_agl2size
;
3248 if (agno
> bmp
->db_maxag
)
3249 bmp
->db_maxag
= agno
;
3251 /* update the free count for the allocation group and map */
3252 bmp
->db_agfree
[agno
] -= nblocks
;
3253 bmp
->db_nfree
-= nblocks
;
3257 /* if the root has not changed, done. */
3258 if (tp
->stree
[ROOT
] == oldroot
)
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).
3265 if ((rc
= dbAdjCtl(bmp
, blkno
, tp
->stree
[ROOT
], 1, 0)))
3266 dbFreeBits(bmp
, dp
, blkno
, nblocks
);
3273 * NAME: dbExtendFS()
3275 * FUNCTION: extend bmap from blkno for nblocks;
3276 * dbExtendFS() updates bmap ready for dbAllocBottomUp();
3280 * L1---------------------------------L1
3282 * L0---------L0---------L0 L0---------L0---------L0
3284 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm;
3285 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3287 * <---old---><----------------------------extend----------------------->
3289 int dbExtendFS(struct inode
*ipbmap
, s64 blkno
, s64 nblocks
)
3291 struct jfs_sb_info
*sbi
= JFS_SBI(ipbmap
->i_sb
);
3292 int nbperpage
= sbi
->nbperpage
;
3293 int i
, i0
= true, j
, j0
= true, k
, n
;
3296 struct metapage
*mp
, *l2mp
, *l1mp
= NULL
, *l0mp
= NULL
;
3297 struct dmapctl
*l2dcp
, *l1dcp
, *l0dcp
;
3299 s8
*l0leaf
, *l1leaf
, *l2leaf
;
3300 struct bmap
*bmp
= sbi
->bmap
;
3301 int agno
, l2agsize
, oldl2agsize
;
3304 newsize
= blkno
+ nblocks
;
3306 jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld",
3307 (long long) blkno
, (long long) nblocks
, (long long) newsize
);
3310 * initialize bmap control page.
3312 * all the data in bmap control page should exclude
3313 * the mkfs hidden dmap page.
3316 /* update mapsize */
3317 bmp
->db_mapsize
= newsize
;
3318 bmp
->db_maxlevel
= BMAPSZTOLEV(bmp
->db_mapsize
);
3320 /* compute new AG size */
3321 l2agsize
= dbGetL2AGSize(newsize
);
3322 oldl2agsize
= bmp
->db_agl2size
;
3324 bmp
->db_agl2size
= l2agsize
;
3325 bmp
->db_agsize
= 1 << l2agsize
;
3327 /* compute new number of AG */
3328 agno
= bmp
->db_numag
;
3329 bmp
->db_numag
= newsize
>> l2agsize
;
3330 bmp
->db_numag
+= ((u32
) newsize
% (u32
) bmp
->db_agsize
) ? 1 : 0;
3333 * reconfigure db_agfree[]
3334 * from old AG configuration to new AG configuration;
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).
3340 if (l2agsize
== oldl2agsize
)
3342 k
= 1 << (l2agsize
- oldl2agsize
);
3343 ag_rem
= bmp
->db_agfree
[0]; /* save agfree[0] */
3344 for (i
= 0, n
= 0; i
< agno
; n
++) {
3345 bmp
->db_agfree
[n
] = 0; /* init collection point */
3347 /* coalesce cotiguous k AGs; */
3348 for (j
= 0; j
< k
&& i
< agno
; j
++, i
++) {
3349 /* merge AGi to AGn */
3350 bmp
->db_agfree
[n
] += bmp
->db_agfree
[i
];
3353 bmp
->db_agfree
[0] += ag_rem
; /* restore agfree[0] */
3355 for (; n
< MAXAG
; n
++)
3356 bmp
->db_agfree
[n
] = 0;
3359 * update highest active ag number
3362 bmp
->db_maxag
= bmp
->db_maxag
/ k
;
3367 * update bit maps and corresponding level control pages;
3368 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3372 p
= BMAPBLKNO
+ nbperpage
; /* L2 page */
3373 l2mp
= read_metapage(ipbmap
, p
, PSIZE
, 0);
3375 jfs_error(ipbmap
->i_sb
, "dbExtendFS: L2 page could not be read");
3378 l2dcp
= (struct dmapctl
*) l2mp
->data
;
3380 /* compute start L1 */
3381 k
= blkno
>> L2MAXL1SIZE
;
3382 l2leaf
= l2dcp
->stree
+ CTLLEAFIND
+ k
;
3383 p
= BLKTOL1(blkno
, sbi
->l2nbperpage
); /* L1 page */
3386 * extend each L1 in L2
3388 for (; k
< LPERCTL
; k
++, p
+= nbperpage
) {
3391 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3392 l1mp
= read_metapage(ipbmap
, p
, PSIZE
, 0);
3395 l1dcp
= (struct dmapctl
*) l1mp
->data
;
3397 /* compute start L0 */
3398 j
= (blkno
& (MAXL1SIZE
- 1)) >> L2MAXL0SIZE
;
3399 l1leaf
= l1dcp
->stree
+ CTLLEAFIND
+ j
;
3400 p
= BLKTOL0(blkno
, sbi
->l2nbperpage
);
3403 /* assign/init L1 page */
3404 l1mp
= get_metapage(ipbmap
, p
, PSIZE
, 0);
3408 l1dcp
= (struct dmapctl
*) l1mp
->data
;
3410 /* compute start L0 */
3412 l1leaf
= l1dcp
->stree
+ CTLLEAFIND
;
3413 p
+= nbperpage
; /* 1st L0 of L1.k */
3417 * extend each L0 in L1
3419 for (; j
< LPERCTL
; j
++) {
3422 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3424 l0mp
= read_metapage(ipbmap
, p
, PSIZE
, 0);
3427 l0dcp
= (struct dmapctl
*) l0mp
->data
;
3429 /* compute start dmap */
3430 i
= (blkno
& (MAXL0SIZE
- 1)) >>
3432 l0leaf
= l0dcp
->stree
+ CTLLEAFIND
+ i
;
3433 p
= BLKTODMAP(blkno
,
3437 /* assign/init L0 page */
3438 l0mp
= get_metapage(ipbmap
, p
, PSIZE
, 0);
3442 l0dcp
= (struct dmapctl
*) l0mp
->data
;
3444 /* compute start dmap */
3446 l0leaf
= l0dcp
->stree
+ CTLLEAFIND
;
3447 p
+= nbperpage
; /* 1st dmap of L0.j */
3451 * extend each dmap in L0
3453 for (; i
< LPERCTL
; i
++) {
3455 * reconstruct the dmap page, and
3456 * initialize corresponding parent L0 leaf
3458 if ((n
= blkno
& (BPERDMAP
- 1))) {
3459 /* read in dmap page: */
3460 mp
= read_metapage(ipbmap
, p
,
3464 n
= min(nblocks
, (s64
)BPERDMAP
- n
);
3466 /* assign/init dmap page */
3467 mp
= read_metapage(ipbmap
, p
,
3472 n
= min(nblocks
, (s64
)BPERDMAP
);
3475 dp
= (struct dmap
*) mp
->data
;
3476 *l0leaf
= dbInitDmap(dp
, blkno
, n
);
3479 agno
= le64_to_cpu(dp
->start
) >> l2agsize
;
3480 bmp
->db_agfree
[agno
] += n
;
3491 } /* for each dmap in a L0 */
3494 * build current L0 page from its leaves, and
3495 * initialize corresponding parent L1 leaf
3497 *l1leaf
= dbInitDmapCtl(l0dcp
, 0, ++i
);
3498 write_metapage(l0mp
);
3502 l1leaf
++; /* continue for next L0 */
3504 /* more than 1 L0 ? */
3506 break; /* build L1 page */
3508 /* summarize in global bmap page */
3509 bmp
->db_maxfreebud
= *l1leaf
;
3510 release_metapage(l1mp
);
3511 release_metapage(l2mp
);
3515 } /* for each L0 in a L1 */
3518 * build current L1 page from its leaves, and
3519 * initialize corresponding parent L2 leaf
3521 *l2leaf
= dbInitDmapCtl(l1dcp
, 1, ++j
);
3522 write_metapage(l1mp
);
3526 l2leaf
++; /* continue for next L1 */
3528 /* more than 1 L1 ? */
3530 break; /* build L2 page */
3532 /* summarize in global bmap page */
3533 bmp
->db_maxfreebud
= *l2leaf
;
3534 release_metapage(l2mp
);
3538 } /* for each L1 in a L2 */
3540 jfs_error(ipbmap
->i_sb
,
3541 "dbExtendFS: function has not returned as expected");
3544 release_metapage(l0mp
);
3546 release_metapage(l1mp
);
3547 release_metapage(l2mp
);
3551 * finalize bmap control page
3562 void dbFinalizeBmap(struct inode
*ipbmap
)
3564 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
3565 int actags
, inactags
, l2nl
;
3566 s64 ag_rem
, actfree
, inactfree
, avgfree
;
3570 * finalize bmap control page
3574 * compute db_agpref: preferred ag to allocate from
3575 * (the leftmost ag with average free space in it);
3578 /* get the number of active ags and inacitve ags */
3579 actags
= bmp
->db_maxag
+ 1;
3580 inactags
= bmp
->db_numag
- actags
;
3581 ag_rem
= bmp
->db_mapsize
& (bmp
->db_agsize
- 1); /* ??? */
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).
3588 inactfree
= (inactags
&& ag_rem
) ?
3589 ((inactags
- 1) << bmp
->db_agl2size
) + ag_rem
3590 : inactags
<< bmp
->db_agl2size
;
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.
3596 actfree
= bmp
->db_nfree
- inactfree
;
3597 avgfree
= (u32
) actfree
/ (u32
) actags
;
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.
3603 if (bmp
->db_agfree
[bmp
->db_agpref
] < avgfree
) {
3604 for (bmp
->db_agpref
= 0; bmp
->db_agpref
< actags
;
3606 if (bmp
->db_agfree
[bmp
->db_agpref
] >= avgfree
)
3609 if (bmp
->db_agpref
>= bmp
->db_numag
) {
3610 jfs_error(ipbmap
->i_sb
,
3611 "cannot find ag with average freespace");
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
3622 bmp
->db_aglevel
= BMAPSZTOLEV(bmp
->db_agsize
);
3624 bmp
->db_agl2size
- (L2BPERDMAP
+ bmp
->db_aglevel
* L2LPERCTL
);
3625 bmp
->db_agheigth
= l2nl
>> 1;
3626 bmp
->db_agwidth
= 1 << (l2nl
- (bmp
->db_agheigth
<< 1));
3627 for (i
= 5 - bmp
->db_agheigth
, bmp
->db_agstart
= 0, n
= 1; i
> 0;
3629 bmp
->db_agstart
+= n
;
3637 * NAME: dbInitDmap()/ujfs_idmap_page()
3639 * FUNCTION: initialize working/persistent bitmap of the dmap page
3640 * for the specified number of blocks:
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);
3648 * dp - pointer to page of map
3649 * nblocks - number of blocks this page
3653 static int dbInitDmap(struct dmap
* dp
, s64 Blkno
, int nblocks
)
3655 int blkno
, w
, b
, r
, nw
, nb
, i
;
3657 /* starting block number within the dmap */
3658 blkno
= Blkno
& (BPERDMAP
- 1);
3661 dp
->nblocks
= dp
->nfree
= cpu_to_le32(nblocks
);
3662 dp
->start
= cpu_to_le64(Blkno
);
3664 if (nblocks
== BPERDMAP
) {
3665 memset(&dp
->wmap
[0], 0, LPERDMAP
* 4);
3666 memset(&dp
->pmap
[0], 0, LPERDMAP
* 4);
3671 cpu_to_le32(le32_to_cpu(dp
->nblocks
) + nblocks
);
3672 dp
->nfree
= cpu_to_le32(le32_to_cpu(dp
->nfree
) + nblocks
);
3675 /* word number containing start block number */
3676 w
= blkno
>> L2DBWORD
;
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.
3683 for (r
= nblocks
; r
> 0; r
-= nb
, blkno
+= nb
) {
3684 /* number of bits preceding range to be freed in the word */
3685 b
= blkno
& (DBWORD
- 1);
3686 /* number of bits to free in the word */
3687 nb
= min(r
, DBWORD
- b
);
3689 /* is partial word to be freed ? */
3691 /* free (set to 0) from the bitmap word */
3692 dp
->wmap
[w
] &= cpu_to_le32(~(ONES
<< (DBWORD
- nb
)
3694 dp
->pmap
[w
] &= cpu_to_le32(~(ONES
<< (DBWORD
- nb
)
3697 /* skip the word freed */
3700 /* free (set to 0) contiguous bitmap words */
3702 memset(&dp
->wmap
[w
], 0, nw
* 4);
3703 memset(&dp
->pmap
[w
], 0, nw
* 4);
3705 /* skip the words freed */
3706 nb
= nw
<< L2DBWORD
;
3712 * mark bits following the range to be freed (non-existing
3713 * blocks) as allocated (ONES)
3716 if (blkno
== BPERDMAP
)
3719 /* the first word beyond the end of existing blocks */
3720 w
= blkno
>> L2DBWORD
;
3722 /* does nblocks fall on a 32-bit boundary ? */
3723 b
= blkno
& (DBWORD
- 1);
3725 /* mark a partial word allocated */
3726 dp
->wmap
[w
] = dp
->pmap
[w
] = cpu_to_le32(ONES
>> b
);
3730 /* set the rest of the words in the page to allocated (ONES) */
3731 for (i
= w
; i
< LPERDMAP
; i
++)
3732 dp
->pmap
[i
] = dp
->wmap
[i
] = cpu_to_le32(ONES
);
3738 return (dbInitDmapTree(dp
));
3743 * NAME: dbInitDmapTree()/ujfs_complete_dmap()
3745 * FUNCTION: initialize summary tree of the specified dmap:
3747 * at entry, bitmap of the dmap has been initialized;
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
3754 * RETURNS: max free string at the root of the tree
3756 static int dbInitDmapTree(struct dmap
* dp
)
3758 struct dmaptree
*tp
;
3762 /* init fixed info of tree */
3764 tp
->nleafs
= cpu_to_le32(LPERDMAP
);
3765 tp
->l2nleafs
= cpu_to_le32(L2LPERDMAP
);
3766 tp
->leafidx
= cpu_to_le32(LEAFIND
);
3767 tp
->height
= cpu_to_le32(4);
3768 tp
->budmin
= BUDMIN
;
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.
3774 cp
= tp
->stree
+ le32_to_cpu(tp
->leafidx
);
3775 for (i
= 0; i
< LPERDMAP
; i
++)
3776 *cp
++ = dbMaxBud((u8
*) & dp
->wmap
[i
]);
3778 /* build the dmap's binary buddy summary tree */
3779 return (dbInitTree(tp
));
3784 * NAME: dbInitTree()/ujfs_adjtree()
3786 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl.
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.
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
3800 * RETURNS: max free string at the root of the tree
3802 static int dbInitTree(struct dmaptree
* dtp
)
3804 int l2max
, l2free
, bsize
, nextb
, i
;
3805 int child
, parent
, nparent
;
3810 /* Determine the maximum free string possible for the leaves */
3811 l2max
= le32_to_cpu(dtp
->l2nleafs
) + dtp
->budmin
;
3814 * configure the leaf levevl into binary buddy system
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.
3827 for (l2free
= dtp
->budmin
, bsize
= 1; l2free
< l2max
;
3828 l2free
++, bsize
= nextb
) {
3829 /* get next buddy size == current buddy pair size */
3832 /* scan each adjacent buddy pair at current buddy size */
3833 for (i
= 0, cp
= tp
+ le32_to_cpu(dtp
->leafidx
);
3834 i
< le32_to_cpu(dtp
->nleafs
);
3835 i
+= nextb
, cp
+= nextb
) {
3836 /* coalesce if both adjacent buddies are max free */
3837 if (*cp
== l2free
&& *(cp
+ bsize
) == l2free
) {
3838 *cp
= l2free
+ 1; /* left take right */
3839 *(cp
+ bsize
) = -1; /* right give left */
3845 * bubble summary information of leaves up the tree.
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.
3854 for (child
= le32_to_cpu(dtp
->leafidx
),
3855 nparent
= le32_to_cpu(dtp
->nleafs
) >> 2;
3856 nparent
> 0; nparent
>>= 2, child
= parent
) {
3857 /* get index of 1st node of parent level */
3858 parent
= (child
- 1) >> 2;
3860 /* set the value of the parent node as the maximum
3861 * of the four nodes of the current level.
3863 for (i
= 0, cp
= tp
+ child
, cp1
= tp
+ parent
;
3864 i
< nparent
; i
++, cp
+= 4, cp1
++)
3875 * function: initialize dmapctl page
3877 static int dbInitDmapCtl(struct dmapctl
* dcp
, int level
, int i
)
3878 { /* start leaf index not covered by range */
3881 dcp
->nleafs
= cpu_to_le32(LPERCTL
);
3882 dcp
->l2nleafs
= cpu_to_le32(L2LPERCTL
);
3883 dcp
->leafidx
= cpu_to_le32(CTLLEAFIND
);
3884 dcp
->height
= cpu_to_le32(5);
3885 dcp
->budmin
= L2BPERDMAP
+ L2LPERCTL
* level
;
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).
3892 cp
= &dcp
->stree
[CTLLEAFIND
+ i
];
3893 for (; i
< LPERCTL
; i
++)
3896 /* build the dmap's binary buddy summary tree */
3897 return (dbInitTree((struct dmaptree
*) dcp
));
3902 * NAME: dbGetL2AGSize()/ujfs_getagl2size()
3904 * FUNCTION: Determine log2(allocation group size) from aggregate size
3907 * nblocks - Number of blocks in aggregate
3909 * RETURNS: log2(allocation group size) in aggregate blocks
3911 static int dbGetL2AGSize(s64 nblocks
)
3917 if (nblocks
< BPERDMAP
* MAXAG
)
3918 return (L2BPERDMAP
);
3920 /* round up aggregate size to power of 2 */
3921 m
= ((u64
) 1 << (64 - 1));
3922 for (l2sz
= 64; l2sz
>= 0; l2sz
--, m
>>= 1) {
3927 sz
= (s64
) 1 << l2sz
;
3931 /* agsize = roundupSize/max_number_of_ag */
3932 return (l2sz
- L2MAXAG
);
3937 * NAME: dbMapFileSizeToMapSize()
3939 * FUNCTION: compute number of blocks the block allocation map file
3940 * can cover from the map file size;
3942 * RETURNS: Number of blocks which can be covered by this block map file;
3946 * maximum number of map pages at each level including control pages
3948 #define MAXL0PAGES (1 + LPERCTL)
3949 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES)
3950 #define MAXL2PAGES (1 + LPERCTL * MAXL1PAGES)
3953 * convert number of map pages to the zero origin top dmapctl level
3955 #define BMAPPGTOLEV(npages) \
3956 (((npages) <= 3 + MAXL0PAGES) ? 0 \
3957 : ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
3959 s64
dbMapFileSizeToMapSize(struct inode
* ipbmap
)
3961 struct super_block
*sb
= ipbmap
->i_sb
;
3965 int complete
, factor
;
3967 nblocks
= ipbmap
->i_size
>> JFS_SBI(sb
)->l2bsize
;
3968 npages
= nblocks
>> JFS_SBI(sb
)->l2nbperpage
;
3969 level
= BMAPPGTOLEV(npages
);
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.
3976 npages
--; /* skip the first global control page */
3977 /* skip higher level control pages above top level covered by map */
3978 npages
-= (2 - level
);
3979 npages
--; /* skip top level's control page */
3980 for (i
= level
; i
>= 0; i
--) {
3982 (i
== 2) ? MAXL1PAGES
: ((i
== 1) ? MAXL0PAGES
: 1);
3983 complete
= (u32
) npages
/ factor
;
3984 ndmaps
+= complete
* ((i
== 2) ? LPERCTL
* LPERCTL
3985 : ((i
== 1) ? LPERCTL
: 1));
3987 /* pages in last/incomplete child */
3988 npages
= (u32
) npages
% factor
;
3989 /* skip incomplete child's level control page */
3993 /* convert the number of dmaps into the number of blocks
3994 * which can be covered by the dmaps;
3996 nblocks
= ndmaps
<< L2BPERDMAP
;
This page took 0.114983 seconds and 5 git commands to generate.