]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/jfs/jfs_dmap.c
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 <linux/slab.h>
21 #include "jfs_incore.h"
22 #include "jfs_superblock.h"
26 #include "jfs_metapage.h"
27 #include "jfs_debug.h"
30 * SERIALIZATION of the Block Allocation Map.
32 * the working state of the block allocation map is accessed in
35 * 1) allocation and free requests that start at the dmap
36 * level and move up through the dmap control pages (i.e.
37 * the vast majority of requests).
39 * 2) allocation requests that start at dmap control page
40 * level and work down towards the dmaps.
42 * the serialization scheme used here is as follows.
44 * requests which start at the bottom are serialized against each
45 * other through buffers and each requests holds onto its buffers
46 * as it works it way up from a single dmap to the required level
47 * of dmap control page.
48 * requests that start at the top are serialized against each other
49 * and request that start from the bottom by the multiple read/single
50 * write inode lock of the bmap inode. requests starting at the top
51 * take this lock in write mode while request starting at the bottom
52 * take the lock in read mode. a single top-down request may proceed
53 * exclusively while multiple bottoms-up requests may proceed
54 * simultaneously (under the protection of busy buffers).
56 * in addition to information found in dmaps and dmap control pages,
57 * the working state of the block allocation map also includes read/
58 * write information maintained in the bmap descriptor (i.e. total
59 * free block count, allocation group level free block counts).
60 * a single exclusive lock (BMAP_LOCK) is used to guard this information
61 * in the face of multiple-bottoms up requests.
62 * (lock ordering: IREAD_LOCK, BMAP_LOCK);
64 * accesses to the persistent state of the block allocation map (limited
65 * to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
68 #define BMAP_LOCK_INIT(bmp) mutex_init(&bmp->db_bmaplock)
69 #define BMAP_LOCK(bmp) mutex_lock(&bmp->db_bmaplock)
70 #define BMAP_UNLOCK(bmp) mutex_unlock(&bmp->db_bmaplock)
75 static void dbAllocBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
77 static void dbSplit(dmtree_t
* tp
, int leafno
, int splitsz
, int newval
);
78 static int dbBackSplit(dmtree_t
* tp
, int leafno
);
79 static int dbJoin(dmtree_t
* tp
, int leafno
, int newval
);
80 static void dbAdjTree(dmtree_t
* tp
, int leafno
, int newval
);
81 static int dbAdjCtl(struct bmap
* bmp
, s64 blkno
, int newval
, int alloc
,
83 static int dbAllocAny(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64
* results
);
84 static int dbAllocNext(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
86 static int dbAllocNear(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
88 int l2nb
, s64
* results
);
89 static int dbAllocDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
91 static int dbAllocDmapLev(struct bmap
* bmp
, struct dmap
* dp
, int nblocks
,
94 static int dbAllocAG(struct bmap
* bmp
, int agno
, s64 nblocks
, int l2nb
,
96 static int dbAllocCtl(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64 blkno
,
98 static int dbExtend(struct inode
*ip
, s64 blkno
, s64 nblocks
, s64 addnblocks
);
99 static int dbFindBits(u32 word
, int l2nb
);
100 static int dbFindCtl(struct bmap
* bmp
, int l2nb
, int level
, s64
* blkno
);
101 static int dbFindLeaf(dmtree_t
* tp
, int l2nb
, int *leafidx
);
102 static int dbFreeBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
104 static int dbFreeDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
106 static int dbMaxBud(u8
* cp
);
107 s64
dbMapFileSizeToMapSize(struct inode
*ipbmap
);
108 static int blkstol2(s64 nb
);
110 static int cntlz(u32 value
);
111 static int cnttz(u32 word
);
113 static int dbAllocDmapBU(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
115 static int dbInitDmap(struct dmap
* dp
, s64 blkno
, int nblocks
);
116 static int dbInitDmapTree(struct dmap
* dp
);
117 static int dbInitTree(struct dmaptree
* dtp
);
118 static int dbInitDmapCtl(struct dmapctl
* dcp
, int level
, int i
);
119 static int dbGetL2AGSize(s64 nblocks
);
124 * table used for determining buddy sizes within characters of
125 * dmap bitmap words. the characters themselves serve as indexes
126 * into the table, with the table elements yielding the maximum
127 * binary buddy of free bits within the character.
129 static const s8 budtab
[256] = {
130 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
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, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
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, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
138 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
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, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
142 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
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, 0,
145 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1
152 * FUNCTION: initializate the block allocation map.
154 * memory is allocated for the in-core bmap descriptor and
155 * the in-core descriptor is initialized from disk.
158 * ipbmap - pointer to in-core inode for the block map.
162 * -ENOMEM - insufficient memory
165 int dbMount(struct inode
*ipbmap
)
168 struct dbmap_disk
*dbmp_le
;
173 * allocate/initialize the in-memory bmap descriptor
175 /* allocate memory for the in-memory bmap descriptor */
176 bmp
= kmalloc(sizeof(struct bmap
), GFP_KERNEL
);
180 /* read the on-disk bmap descriptor. */
181 mp
= read_metapage(ipbmap
,
182 BMAPBLKNO
<< JFS_SBI(ipbmap
->i_sb
)->l2nbperpage
,
189 /* copy the on-disk bmap descriptor to its in-memory version. */
190 dbmp_le
= (struct dbmap_disk
*) mp
->data
;
191 bmp
->db_mapsize
= le64_to_cpu(dbmp_le
->dn_mapsize
);
192 bmp
->db_nfree
= le64_to_cpu(dbmp_le
->dn_nfree
);
193 bmp
->db_l2nbperpage
= le32_to_cpu(dbmp_le
->dn_l2nbperpage
);
194 bmp
->db_numag
= le32_to_cpu(dbmp_le
->dn_numag
);
195 bmp
->db_maxlevel
= le32_to_cpu(dbmp_le
->dn_maxlevel
);
196 bmp
->db_maxag
= le32_to_cpu(dbmp_le
->dn_maxag
);
197 bmp
->db_agpref
= le32_to_cpu(dbmp_le
->dn_agpref
);
198 bmp
->db_aglevel
= le32_to_cpu(dbmp_le
->dn_aglevel
);
199 bmp
->db_agheight
= le32_to_cpu(dbmp_le
->dn_agheight
);
200 bmp
->db_agwidth
= le32_to_cpu(dbmp_le
->dn_agwidth
);
201 bmp
->db_agstart
= le32_to_cpu(dbmp_le
->dn_agstart
);
202 bmp
->db_agl2size
= le32_to_cpu(dbmp_le
->dn_agl2size
);
203 for (i
= 0; i
< MAXAG
; i
++)
204 bmp
->db_agfree
[i
] = le64_to_cpu(dbmp_le
->dn_agfree
[i
]);
205 bmp
->db_agsize
= le64_to_cpu(dbmp_le
->dn_agsize
);
206 bmp
->db_maxfreebud
= dbmp_le
->dn_maxfreebud
;
208 /* release the buffer. */
209 release_metapage(mp
);
211 /* bind the bmap inode and the bmap descriptor to each other. */
212 bmp
->db_ipbmap
= ipbmap
;
213 JFS_SBI(ipbmap
->i_sb
)->bmap
= bmp
;
215 memset(bmp
->db_active
, 0, sizeof(bmp
->db_active
));
218 * allocate/initialize the bmap lock
229 * FUNCTION: terminate the block allocation map in preparation for
230 * file system unmount.
232 * the in-core bmap descriptor is written to disk and
233 * the memory for this descriptor is freed.
236 * ipbmap - pointer to in-core inode for the block map.
242 int dbUnmount(struct inode
*ipbmap
, int mounterror
)
244 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
246 if (!(mounterror
|| isReadOnly(ipbmap
)))
250 * Invalidate the page cache buffers
252 truncate_inode_pages(ipbmap
->i_mapping
, 0);
254 /* free the memory for the in-memory bmap. */
263 int dbSync(struct inode
*ipbmap
)
265 struct dbmap_disk
*dbmp_le
;
266 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
271 * write bmap global control page
273 /* get the buffer for the on-disk bmap descriptor. */
274 mp
= read_metapage(ipbmap
,
275 BMAPBLKNO
<< JFS_SBI(ipbmap
->i_sb
)->l2nbperpage
,
278 jfs_err("dbSync: read_metapage failed!");
281 /* copy the in-memory version of the bmap to the on-disk version */
282 dbmp_le
= (struct dbmap_disk
*) mp
->data
;
283 dbmp_le
->dn_mapsize
= cpu_to_le64(bmp
->db_mapsize
);
284 dbmp_le
->dn_nfree
= cpu_to_le64(bmp
->db_nfree
);
285 dbmp_le
->dn_l2nbperpage
= cpu_to_le32(bmp
->db_l2nbperpage
);
286 dbmp_le
->dn_numag
= cpu_to_le32(bmp
->db_numag
);
287 dbmp_le
->dn_maxlevel
= cpu_to_le32(bmp
->db_maxlevel
);
288 dbmp_le
->dn_maxag
= cpu_to_le32(bmp
->db_maxag
);
289 dbmp_le
->dn_agpref
= cpu_to_le32(bmp
->db_agpref
);
290 dbmp_le
->dn_aglevel
= cpu_to_le32(bmp
->db_aglevel
);
291 dbmp_le
->dn_agheight
= cpu_to_le32(bmp
->db_agheight
);
292 dbmp_le
->dn_agwidth
= cpu_to_le32(bmp
->db_agwidth
);
293 dbmp_le
->dn_agstart
= cpu_to_le32(bmp
->db_agstart
);
294 dbmp_le
->dn_agl2size
= cpu_to_le32(bmp
->db_agl2size
);
295 for (i
= 0; i
< MAXAG
; i
++)
296 dbmp_le
->dn_agfree
[i
] = cpu_to_le64(bmp
->db_agfree
[i
]);
297 dbmp_le
->dn_agsize
= cpu_to_le64(bmp
->db_agsize
);
298 dbmp_le
->dn_maxfreebud
= bmp
->db_maxfreebud
;
300 /* write the buffer */
304 * write out dirty pages of bmap
306 filemap_write_and_wait(ipbmap
->i_mapping
);
308 diWriteSpecial(ipbmap
, 0);
317 * FUNCTION: free the specified block range from the working block
320 * the blocks will be free from the working map one dmap
324 * ip - pointer to in-core inode;
325 * blkno - starting block number to be freed.
326 * nblocks - number of blocks to be freed.
332 int dbFree(struct inode
*ip
, s64 blkno
, s64 nblocks
)
338 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
339 struct bmap
*bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
341 IREAD_LOCK(ipbmap
, RDWRLOCK_DMAP
);
343 /* block to be freed better be within the mapsize. */
344 if (unlikely((blkno
== 0) || (blkno
+ nblocks
> bmp
->db_mapsize
))) {
345 IREAD_UNLOCK(ipbmap
);
346 printk(KERN_ERR
"blkno = %Lx, nblocks = %Lx\n",
347 (unsigned long long) blkno
,
348 (unsigned long long) nblocks
);
350 "dbFree: block to be freed is outside the map");
355 * free the blocks a dmap at a time.
358 for (rem
= nblocks
; rem
> 0; rem
-= nb
, blkno
+= nb
) {
359 /* release previous dmap if any */
364 /* get the buffer for the current dmap. */
365 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
366 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
368 IREAD_UNLOCK(ipbmap
);
371 dp
= (struct dmap
*) mp
->data
;
373 /* determine the number of blocks to be freed from
376 nb
= min(rem
, BPERDMAP
- (blkno
& (BPERDMAP
- 1)));
378 /* free the blocks. */
379 if ((rc
= dbFreeDmap(bmp
, dp
, blkno
, nb
))) {
380 jfs_error(ip
->i_sb
, "dbFree: error in block map\n");
381 release_metapage(mp
);
382 IREAD_UNLOCK(ipbmap
);
387 /* write the last buffer. */
390 IREAD_UNLOCK(ipbmap
);
397 * NAME: dbUpdatePMap()
399 * FUNCTION: update the allocation state (free or allocate) of the
400 * specified block range in the persistent block allocation map.
402 * the blocks will be updated in the persistent map one
406 * ipbmap - pointer to in-core inode for the block map.
407 * free - 'true' if block range is to be freed from the persistent
408 * map; 'false' if it is to be allocated.
409 * blkno - starting block number of the range.
410 * nblocks - number of contiguous blocks in the range.
411 * tblk - transaction block;
418 dbUpdatePMap(struct inode
*ipbmap
,
419 int free
, s64 blkno
, s64 nblocks
, struct tblock
* tblk
)
421 int nblks
, dbitno
, wbitno
, rbits
;
422 int word
, nbits
, nwords
;
423 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
424 s64 lblkno
, rem
, lastlblkno
;
429 int lsn
, difft
, diffp
;
432 /* the blocks better be within the mapsize. */
433 if (blkno
+ nblocks
> bmp
->db_mapsize
) {
434 printk(KERN_ERR
"blkno = %Lx, nblocks = %Lx\n",
435 (unsigned long long) blkno
,
436 (unsigned long long) nblocks
);
437 jfs_error(ipbmap
->i_sb
,
438 "dbUpdatePMap: blocks are outside the map");
442 /* compute delta of transaction lsn from log syncpt */
444 log
= (struct jfs_log
*) JFS_SBI(tblk
->sb
)->log
;
445 logdiff(difft
, lsn
, log
);
448 * update the block state a dmap at a time.
452 for (rem
= nblocks
; rem
> 0; rem
-= nblks
, blkno
+= nblks
) {
453 /* get the buffer for the current dmap. */
454 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
455 if (lblkno
!= lastlblkno
) {
460 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
,
464 metapage_wait_for_io(mp
);
466 dp
= (struct dmap
*) mp
->data
;
468 /* determine the bit number and word within the dmap of
469 * the starting block. also determine how many blocks
470 * are to be updated within this dmap.
472 dbitno
= blkno
& (BPERDMAP
- 1);
473 word
= dbitno
>> L2DBWORD
;
474 nblks
= min(rem
, (s64
)BPERDMAP
- dbitno
);
476 /* update the bits of the dmap words. the first and last
477 * words may only have a subset of their bits updated. if
478 * this is the case, we'll work against that word (i.e.
479 * partial first and/or last) only in a single pass. a
480 * single pass will also be used to update all words that
481 * are to have all their bits updated.
483 for (rbits
= nblks
; rbits
> 0;
484 rbits
-= nbits
, dbitno
+= nbits
) {
485 /* determine the bit number within the word and
486 * the number of bits within the word.
488 wbitno
= dbitno
& (DBWORD
- 1);
489 nbits
= min(rbits
, DBWORD
- wbitno
);
491 /* check if only part of the word is to be updated. */
492 if (nbits
< DBWORD
) {
493 /* update (free or allocate) the bits
497 (ONES
<< (DBWORD
- nbits
) >> wbitno
);
507 /* one or more words are to have all
508 * their bits updated. determine how
509 * many words and how many bits.
511 nwords
= rbits
>> L2DBWORD
;
512 nbits
= nwords
<< L2DBWORD
;
514 /* update (free or allocate) the bits
518 memset(&dp
->pmap
[word
], 0,
521 memset(&dp
->pmap
[word
], (int) ONES
,
531 if (lblkno
== lastlblkno
)
536 LOGSYNC_LOCK(log
, flags
);
538 /* inherit older/smaller lsn */
539 logdiff(diffp
, mp
->lsn
, log
);
543 /* move bp after tblock in logsync list */
544 list_move(&mp
->synclist
, &tblk
->synclist
);
547 /* inherit younger/larger clsn */
548 logdiff(difft
, tblk
->clsn
, log
);
549 logdiff(diffp
, mp
->clsn
, log
);
551 mp
->clsn
= tblk
->clsn
;
556 /* insert bp after tblock in logsync list */
558 list_add(&mp
->synclist
, &tblk
->synclist
);
560 mp
->clsn
= tblk
->clsn
;
562 LOGSYNC_UNLOCK(log
, flags
);
565 /* write the last buffer. */
577 * FUNCTION: find the preferred allocation group for new allocations.
579 * Within the allocation groups, we maintain a preferred
580 * allocation group which consists of a group with at least
581 * average free space. It is the preferred group that we target
582 * new inode allocation towards. The tie-in between inode
583 * allocation and block allocation occurs as we allocate the
584 * first (data) block of an inode and specify the inode (block)
585 * as the allocation hint for this block.
587 * We try to avoid having more than one open file growing in
588 * an allocation group, as this will lead to fragmentation.
589 * This differs from the old OS/2 method of trying to keep
590 * empty ags around for large allocations.
593 * ipbmap - pointer to in-core inode for the block map.
596 * the preferred allocation group number.
598 int dbNextAG(struct inode
*ipbmap
)
605 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
609 /* determine the average number of free blocks within the ags. */
610 avgfree
= (u32
)bmp
->db_nfree
/ bmp
->db_numag
;
613 * if the current preferred ag does not have an active allocator
614 * and has at least average freespace, return it
616 agpref
= bmp
->db_agpref
;
617 if ((atomic_read(&bmp
->db_active
[agpref
]) == 0) &&
618 (bmp
->db_agfree
[agpref
] >= avgfree
))
621 /* From the last preferred ag, find the next one with at least
622 * average free space.
624 for (i
= 0 ; i
< bmp
->db_numag
; i
++, agpref
++) {
625 if (agpref
== bmp
->db_numag
)
628 if (atomic_read(&bmp
->db_active
[agpref
]))
629 /* open file is currently growing in this ag */
631 if (bmp
->db_agfree
[agpref
] >= avgfree
) {
632 /* Return this one */
633 bmp
->db_agpref
= agpref
;
635 } else if (bmp
->db_agfree
[agpref
] > hwm
) {
636 /* Less than avg. freespace, but best so far */
637 hwm
= bmp
->db_agfree
[agpref
];
643 * If no inactive ag was found with average freespace, use the
647 bmp
->db_agpref
= next_best
;
648 /* else leave db_agpref unchanged */
652 /* return the preferred group.
654 return (bmp
->db_agpref
);
660 * FUNCTION: attempt to allocate a specified number of contiguous free
661 * blocks from the working allocation block map.
663 * the block allocation policy uses hints and a multi-step
666 * for allocation requests smaller than the number of blocks
667 * per dmap, we first try to allocate the new blocks
668 * immediately following the hint. if these blocks are not
669 * available, we try to allocate blocks near the hint. if
670 * no blocks near the hint are available, we next try to
671 * allocate within the same dmap as contains the hint.
673 * if no blocks are available in the dmap or the allocation
674 * request is larger than the dmap size, we try to allocate
675 * within the same allocation group as contains the hint. if
676 * this does not succeed, we finally try to allocate anywhere
677 * within the aggregate.
679 * we also try to allocate anywhere within the aggregate for
680 * for allocation requests larger than the allocation group
681 * size or requests that specify no hint value.
684 * ip - pointer to in-core inode;
685 * hint - allocation hint.
686 * nblocks - number of contiguous blocks in the range.
687 * results - on successful return, set to the starting block number
688 * of the newly allocated contiguous range.
692 * -ENOSPC - insufficient disk resources
695 int dbAlloc(struct inode
*ip
, s64 hint
, s64 nblocks
, s64
* results
)
698 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
707 /* assert that nblocks is valid */
710 /* get the log2 number of blocks to be allocated.
711 * if the number of blocks is not a log2 multiple,
712 * it will be rounded up to the next log2 multiple.
714 l2nb
= BLKSTOL2(nblocks
);
716 bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
718 mapSize
= bmp
->db_mapsize
;
720 /* the hint should be within the map */
721 if (hint
>= mapSize
) {
722 jfs_error(ip
->i_sb
, "dbAlloc: the hint is outside the map");
726 /* if the number of blocks to be allocated is greater than the
727 * allocation group size, try to allocate anywhere.
729 if (l2nb
> bmp
->db_agl2size
) {
730 IWRITE_LOCK(ipbmap
, RDWRLOCK_DMAP
);
732 rc
= dbAllocAny(bmp
, nblocks
, l2nb
, results
);
738 * If no hint, let dbNextAG recommend an allocation group
743 /* we would like to allocate close to the hint. adjust the
744 * hint to the block following the hint since the allocators
745 * will start looking for free space starting at this point.
749 if (blkno
>= bmp
->db_mapsize
)
752 agno
= blkno
>> bmp
->db_agl2size
;
754 /* check if blkno crosses over into a new allocation group.
755 * if so, check if we should allow allocations within this
758 if ((blkno
& (bmp
->db_agsize
- 1)) == 0)
759 /* check if the AG is currently being written to.
760 * if so, call dbNextAG() to find a non-busy
761 * AG with sufficient free space.
763 if (atomic_read(&bmp
->db_active
[agno
]))
766 /* check if the allocation request size can be satisfied from a
767 * single dmap. if so, try to allocate from the dmap containing
768 * the hint using a tiered strategy.
770 if (nblocks
<= BPERDMAP
) {
771 IREAD_LOCK(ipbmap
, RDWRLOCK_DMAP
);
773 /* get the buffer for the dmap containing the hint.
776 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
777 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
781 dp
= (struct dmap
*) mp
->data
;
783 /* first, try to satisfy the allocation request with the
784 * blocks beginning at the hint.
786 if ((rc
= dbAllocNext(bmp
, dp
, blkno
, (int) nblocks
))
790 mark_metapage_dirty(mp
);
793 release_metapage(mp
);
797 writers
= atomic_read(&bmp
->db_active
[agno
]);
799 ((writers
== 1) && (JFS_IP(ip
)->active_ag
!= agno
))) {
801 * Someone else is writing in this allocation
802 * group. To avoid fragmenting, try another ag
804 release_metapage(mp
);
805 IREAD_UNLOCK(ipbmap
);
809 /* next, try to satisfy the allocation request with blocks
813 dbAllocNear(bmp
, dp
, blkno
, (int) nblocks
, l2nb
, results
))
816 mark_metapage_dirty(mp
);
818 release_metapage(mp
);
822 /* try to satisfy the allocation request with blocks within
823 * the same dmap as the hint.
825 if ((rc
= dbAllocDmapLev(bmp
, dp
, (int) nblocks
, l2nb
, results
))
828 mark_metapage_dirty(mp
);
830 release_metapage(mp
);
834 release_metapage(mp
);
835 IREAD_UNLOCK(ipbmap
);
838 /* try to satisfy the allocation request with blocks within
839 * the same allocation group as the hint.
841 IWRITE_LOCK(ipbmap
, RDWRLOCK_DMAP
);
842 if ((rc
= dbAllocAG(bmp
, agno
, nblocks
, l2nb
, results
)) != -ENOSPC
)
845 IWRITE_UNLOCK(ipbmap
);
850 * Let dbNextAG recommend a preferred allocation group
852 agno
= dbNextAG(ipbmap
);
853 IWRITE_LOCK(ipbmap
, RDWRLOCK_DMAP
);
855 /* Try to allocate within this allocation group. if that fails, try to
856 * allocate anywhere in the map.
858 if ((rc
= dbAllocAG(bmp
, agno
, nblocks
, l2nb
, results
)) == -ENOSPC
)
859 rc
= dbAllocAny(bmp
, nblocks
, l2nb
, results
);
862 IWRITE_UNLOCK(ipbmap
);
867 IREAD_UNLOCK(ipbmap
);
874 * NAME: dbAllocExact()
876 * FUNCTION: try to allocate the requested extent;
879 * ip - pointer to in-core inode;
880 * blkno - extent address;
881 * nblocks - extent length;
885 * -ENOSPC - insufficient disk resources
888 int dbAllocExact(struct inode
*ip
, s64 blkno
, int nblocks
)
891 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
892 struct bmap
*bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
897 IREAD_LOCK(ipbmap
, RDWRLOCK_DMAP
);
900 * validate extent request:
902 * note: defragfs policy:
903 * max 64 blocks will be moved.
904 * allocation request size must be satisfied from a single dmap.
906 if (nblocks
<= 0 || nblocks
> BPERDMAP
|| blkno
>= bmp
->db_mapsize
) {
907 IREAD_UNLOCK(ipbmap
);
911 if (nblocks
> ((s64
) 1 << bmp
->db_maxfreebud
)) {
912 /* the free space is no longer available */
913 IREAD_UNLOCK(ipbmap
);
917 /* read in the dmap covering the extent */
918 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
919 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
921 IREAD_UNLOCK(ipbmap
);
924 dp
= (struct dmap
*) mp
->data
;
926 /* try to allocate the requested extent */
927 rc
= dbAllocNext(bmp
, dp
, blkno
, nblocks
);
929 IREAD_UNLOCK(ipbmap
);
932 mark_metapage_dirty(mp
);
934 release_metapage(mp
);
943 * FUNCTION: attempt to extend a current allocation by a specified
946 * this routine attempts to satisfy the allocation request
947 * by first trying to extend the existing allocation in
948 * place by allocating the additional blocks as the blocks
949 * immediately following the current allocation. if these
950 * blocks are not available, this routine will attempt to
951 * allocate a new set of contiguous blocks large enough
952 * to cover the existing allocation plus the additional
953 * number of blocks required.
956 * ip - pointer to in-core inode requiring allocation.
957 * blkno - starting block of the current allocation.
958 * nblocks - number of contiguous blocks within the current
960 * addnblocks - number of blocks to add to the allocation.
961 * results - on successful return, set to the starting block number
962 * of the existing allocation if the existing allocation
963 * was extended in place or to a newly allocated contiguous
964 * range if the existing allocation could not be extended
969 * -ENOSPC - insufficient disk resources
973 dbReAlloc(struct inode
*ip
,
974 s64 blkno
, s64 nblocks
, s64 addnblocks
, s64
* results
)
978 /* try to extend the allocation in place.
980 if ((rc
= dbExtend(ip
, blkno
, nblocks
, addnblocks
)) == 0) {
988 /* could not extend the allocation in place, so allocate a
989 * new set of blocks for the entire request (i.e. try to get
990 * a range of contiguous blocks large enough to cover the
991 * existing allocation plus the additional blocks.)
994 (ip
, blkno
+ nblocks
- 1, addnblocks
+ nblocks
, results
));
1001 * FUNCTION: attempt to extend a current allocation by a specified
1004 * this routine attempts to satisfy the allocation request
1005 * by first trying to extend the existing allocation in
1006 * place by allocating the additional blocks as the blocks
1007 * immediately following the current allocation.
1010 * ip - pointer to in-core inode requiring allocation.
1011 * blkno - starting block of the current allocation.
1012 * nblocks - number of contiguous blocks within the current
1014 * addnblocks - number of blocks to add to the allocation.
1018 * -ENOSPC - insufficient disk resources
1021 static int dbExtend(struct inode
*ip
, s64 blkno
, s64 nblocks
, s64 addnblocks
)
1023 struct jfs_sb_info
*sbi
= JFS_SBI(ip
->i_sb
);
1024 s64 lblkno
, lastblkno
, extblkno
;
1026 struct metapage
*mp
;
1029 struct inode
*ipbmap
= sbi
->ipbmap
;
1033 * We don't want a non-aligned extent to cross a page boundary
1035 if (((rel_block
= blkno
& (sbi
->nbperpage
- 1))) &&
1036 (rel_block
+ nblocks
+ addnblocks
> sbi
->nbperpage
))
1039 /* get the last block of the current allocation */
1040 lastblkno
= blkno
+ nblocks
- 1;
1042 /* determine the block number of the block following
1043 * the existing allocation.
1045 extblkno
= lastblkno
+ 1;
1047 IREAD_LOCK(ipbmap
, RDWRLOCK_DMAP
);
1049 /* better be within the file system */
1051 if (lastblkno
< 0 || lastblkno
>= bmp
->db_mapsize
) {
1052 IREAD_UNLOCK(ipbmap
);
1054 "dbExtend: the block is outside the filesystem");
1058 /* we'll attempt to extend the current allocation in place by
1059 * allocating the additional blocks as the blocks immediately
1060 * following the current allocation. we only try to extend the
1061 * current allocation in place if the number of additional blocks
1062 * can fit into a dmap, the last block of the current allocation
1063 * is not the last block of the file system, and the start of the
1064 * inplace extension is not on an allocation group boundary.
1066 if (addnblocks
> BPERDMAP
|| extblkno
>= bmp
->db_mapsize
||
1067 (extblkno
& (bmp
->db_agsize
- 1)) == 0) {
1068 IREAD_UNLOCK(ipbmap
);
1072 /* get the buffer for the dmap containing the first block
1075 lblkno
= BLKTODMAP(extblkno
, bmp
->db_l2nbperpage
);
1076 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
1078 IREAD_UNLOCK(ipbmap
);
1082 dp
= (struct dmap
*) mp
->data
;
1084 /* try to allocate the blocks immediately following the
1085 * current allocation.
1087 rc
= dbAllocNext(bmp
, dp
, extblkno
, (int) addnblocks
);
1089 IREAD_UNLOCK(ipbmap
);
1091 /* were we successful ? */
1095 /* we were not successful */
1096 release_metapage(mp
);
1104 * NAME: dbAllocNext()
1106 * FUNCTION: attempt to allocate the blocks of the specified block
1107 * range within a dmap.
1110 * bmp - pointer to bmap descriptor
1111 * dp - pointer to dmap.
1112 * blkno - starting block number of the range.
1113 * nblocks - number of contiguous free blocks of the range.
1117 * -ENOSPC - insufficient disk resources
1120 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1122 static int dbAllocNext(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
1125 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, nw
;
1130 if (dp
->tree
.leafidx
!= cpu_to_le32(LEAFIND
)) {
1131 jfs_error(bmp
->db_ipbmap
->i_sb
,
1132 "dbAllocNext: Corrupt dmap page");
1136 /* pick up a pointer to the leaves of the dmap tree.
1138 leaf
= dp
->tree
.stree
+ le32_to_cpu(dp
->tree
.leafidx
);
1140 /* determine the bit number and word within the dmap of the
1143 dbitno
= blkno
& (BPERDMAP
- 1);
1144 word
= dbitno
>> L2DBWORD
;
1146 /* check if the specified block range is contained within
1149 if (dbitno
+ nblocks
> BPERDMAP
)
1152 /* check if the starting leaf indicates that anything
1155 if (leaf
[word
] == NOFREE
)
1158 /* check the dmaps words corresponding to block range to see
1159 * if the block range is free. not all bits of the first and
1160 * last words may be contained within the block range. if this
1161 * is the case, we'll work against those words (i.e. partial first
1162 * and/or last) on an individual basis (a single pass) and examine
1163 * the actual bits to determine if they are free. a single pass
1164 * will be used for all dmap words fully contained within the
1165 * specified range. within this pass, the leaves of the dmap
1166 * tree will be examined to determine if the blocks are free. a
1167 * single leaf may describe the free space of multiple dmap
1168 * words, so we may visit only a subset of the actual leaves
1169 * corresponding to the dmap words of the block range.
1171 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
1172 /* determine the bit number within the word and
1173 * the number of bits within the word.
1175 wbitno
= dbitno
& (DBWORD
- 1);
1176 nb
= min(rembits
, DBWORD
- wbitno
);
1178 /* check if only part of the word is to be examined.
1181 /* check if the bits are free.
1183 mask
= (ONES
<< (DBWORD
- nb
) >> wbitno
);
1184 if ((mask
& ~le32_to_cpu(dp
->wmap
[word
])) != mask
)
1189 /* one or more dmap words are fully contained
1190 * within the block range. determine how many
1191 * words and how many bits.
1193 nwords
= rembits
>> L2DBWORD
;
1194 nb
= nwords
<< L2DBWORD
;
1196 /* now examine the appropriate leaves to determine
1197 * if the blocks are free.
1199 while (nwords
> 0) {
1200 /* does the leaf describe any free space ?
1202 if (leaf
[word
] < BUDMIN
)
1205 /* determine the l2 number of bits provided
1209 min((int)leaf
[word
], NLSTOL2BSZ(nwords
));
1211 /* determine how many words were handled.
1213 nw
= BUDSIZE(l2size
, BUDMIN
);
1221 /* allocate the blocks.
1223 return (dbAllocDmap(bmp
, dp
, blkno
, nblocks
));
1228 * NAME: dbAllocNear()
1230 * FUNCTION: attempt to allocate a number of contiguous free blocks near
1231 * a specified block (hint) within a dmap.
1233 * starting with the dmap leaf that covers the hint, we'll
1234 * check the next four contiguous leaves for sufficient free
1235 * space. if sufficient free space is found, we'll allocate
1236 * the desired free space.
1239 * bmp - pointer to bmap descriptor
1240 * dp - pointer to dmap.
1241 * blkno - block number to allocate near.
1242 * nblocks - actual number of contiguous free blocks desired.
1243 * l2nb - log2 number of contiguous free blocks desired.
1244 * results - on successful return, set to the starting block number
1245 * of the newly allocated range.
1249 * -ENOSPC - insufficient disk resources
1252 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1255 dbAllocNear(struct bmap
* bmp
,
1256 struct dmap
* dp
, s64 blkno
, int nblocks
, int l2nb
, s64
* results
)
1258 int word
, lword
, rc
;
1261 if (dp
->tree
.leafidx
!= cpu_to_le32(LEAFIND
)) {
1262 jfs_error(bmp
->db_ipbmap
->i_sb
,
1263 "dbAllocNear: Corrupt dmap page");
1267 leaf
= dp
->tree
.stree
+ le32_to_cpu(dp
->tree
.leafidx
);
1269 /* determine the word within the dmap that holds the hint
1270 * (i.e. blkno). also, determine the last word in the dmap
1271 * that we'll include in our examination.
1273 word
= (blkno
& (BPERDMAP
- 1)) >> L2DBWORD
;
1274 lword
= min(word
+ 4, LPERDMAP
);
1276 /* examine the leaves for sufficient free space.
1278 for (; word
< lword
; word
++) {
1279 /* does the leaf describe sufficient free space ?
1281 if (leaf
[word
] < l2nb
)
1284 /* determine the block number within the file system
1285 * of the first block described by this dmap word.
1287 blkno
= le64_to_cpu(dp
->start
) + (word
<< L2DBWORD
);
1289 /* if not all bits of the dmap word are free, get the
1290 * starting bit number within the dmap word of the required
1291 * string of free bits and adjust the block number with the
1294 if (leaf
[word
] < BUDMIN
)
1296 dbFindBits(le32_to_cpu(dp
->wmap
[word
]), l2nb
);
1298 /* allocate the blocks.
1300 if ((rc
= dbAllocDmap(bmp
, dp
, blkno
, nblocks
)) == 0)
1313 * FUNCTION: attempt to allocate the specified number of contiguous
1314 * free blocks within the specified allocation group.
1316 * unless the allocation group size is equal to the number
1317 * of blocks per dmap, the dmap control pages will be used to
1318 * find the required free space, if available. we start the
1319 * search at the highest dmap control page level which
1320 * distinctly describes the allocation group's free space
1321 * (i.e. the highest level at which the allocation group's
1322 * free space is not mixed in with that of any other group).
1323 * in addition, we start the search within this level at a
1324 * height of the dmapctl dmtree at which the nodes distinctly
1325 * describe the allocation group's free space. at this height,
1326 * the allocation group's free space may be represented by 1
1327 * or two sub-trees, depending on the allocation group size.
1328 * we search the top nodes of these subtrees left to right for
1329 * sufficient free space. if sufficient free space is found,
1330 * the subtree is searched to find the leftmost leaf that
1331 * has free space. once we have made it to the leaf, we
1332 * move the search to the next lower level dmap control page
1333 * corresponding to this leaf. we continue down the dmap control
1334 * pages until we find the dmap that contains or starts the
1335 * sufficient free space and we allocate at this dmap.
1337 * if the allocation group size is equal to the dmap size,
1338 * we'll start at the dmap corresponding to the allocation
1339 * group and attempt the allocation at this level.
1341 * the dmap control page search is also not performed if the
1342 * allocation group is completely free and we go to the first
1343 * dmap of the allocation group to do the allocation. this is
1344 * done because the allocation group may be part (not the first
1345 * part) of a larger binary buddy system, causing the dmap
1346 * control pages to indicate no free space (NOFREE) within
1347 * the allocation group.
1350 * bmp - pointer to bmap descriptor
1351 * agno - allocation group number.
1352 * nblocks - actual number of contiguous free blocks desired.
1353 * l2nb - log2 number of contiguous free blocks desired.
1354 * results - on successful return, set to the starting block number
1355 * of the newly allocated range.
1359 * -ENOSPC - insufficient disk resources
1362 * note: IWRITE_LOCK(ipmap) held on entry/exit;
1365 dbAllocAG(struct bmap
* bmp
, int agno
, s64 nblocks
, int l2nb
, s64
* results
)
1367 struct metapage
*mp
;
1368 struct dmapctl
*dcp
;
1369 int rc
, ti
, i
, k
, m
, n
, agperlev
;
1373 /* allocation request should not be for more than the
1374 * allocation group size.
1376 if (l2nb
> bmp
->db_agl2size
) {
1377 jfs_error(bmp
->db_ipbmap
->i_sb
,
1378 "dbAllocAG: allocation request is larger than the "
1379 "allocation group size");
1383 /* determine the starting block number of the allocation
1386 blkno
= (s64
) agno
<< bmp
->db_agl2size
;
1388 /* check if the allocation group size is the minimum allocation
1389 * group size or if the allocation group is completely free. if
1390 * the allocation group size is the minimum size of BPERDMAP (i.e.
1391 * 1 dmap), there is no need to search the dmap control page (below)
1392 * that fully describes the allocation group since the allocation
1393 * group is already fully described by a dmap. in this case, we
1394 * just call dbAllocCtl() to search the dmap tree and allocate the
1395 * required space if available.
1397 * if the allocation group is completely free, dbAllocCtl() is
1398 * also called to allocate the required space. this is done for
1399 * two reasons. first, it makes no sense searching the dmap control
1400 * pages for free space when we know that free space exists. second,
1401 * the dmap control pages may indicate that the allocation group
1402 * has no free space if the allocation group is part (not the first
1403 * part) of a larger binary buddy system.
1405 if (bmp
->db_agsize
== BPERDMAP
1406 || bmp
->db_agfree
[agno
] == bmp
->db_agsize
) {
1407 rc
= dbAllocCtl(bmp
, nblocks
, l2nb
, blkno
, results
);
1408 if ((rc
== -ENOSPC
) &&
1409 (bmp
->db_agfree
[agno
] == bmp
->db_agsize
)) {
1410 printk(KERN_ERR
"blkno = %Lx, blocks = %Lx\n",
1411 (unsigned long long) blkno
,
1412 (unsigned long long) nblocks
);
1413 jfs_error(bmp
->db_ipbmap
->i_sb
,
1414 "dbAllocAG: dbAllocCtl failed in free AG");
1419 /* the buffer for the dmap control page that fully describes the
1422 lblkno
= BLKTOCTL(blkno
, bmp
->db_l2nbperpage
, bmp
->db_aglevel
);
1423 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1426 dcp
= (struct dmapctl
*) mp
->data
;
1427 budmin
= dcp
->budmin
;
1429 if (dcp
->leafidx
!= cpu_to_le32(CTLLEAFIND
)) {
1430 jfs_error(bmp
->db_ipbmap
->i_sb
,
1431 "dbAllocAG: Corrupt dmapctl page");
1432 release_metapage(mp
);
1436 /* search the subtree(s) of the dmap control page that describes
1437 * the allocation group, looking for sufficient free space. to begin,
1438 * determine how many allocation groups are represented in a dmap
1439 * control page at the control page level (i.e. L0, L1, L2) that
1440 * fully describes an allocation group. next, determine the starting
1441 * tree index of this allocation group within the control page.
1444 (1 << (L2LPERCTL
- (bmp
->db_agheight
<< 1))) / bmp
->db_agwidth
;
1445 ti
= bmp
->db_agstart
+ bmp
->db_agwidth
* (agno
& (agperlev
- 1));
1447 /* dmap control page trees fan-out by 4 and a single allocation
1448 * group may be described by 1 or 2 subtrees within the ag level
1449 * dmap control page, depending upon the ag size. examine the ag's
1450 * subtrees for sufficient free space, starting with the leftmost
1453 for (i
= 0; i
< bmp
->db_agwidth
; i
++, ti
++) {
1454 /* is there sufficient free space ?
1456 if (l2nb
> dcp
->stree
[ti
])
1459 /* sufficient free space found in a subtree. now search down
1460 * the subtree to find the leftmost leaf that describes this
1463 for (k
= bmp
->db_agheight
; k
> 0; k
--) {
1464 for (n
= 0, m
= (ti
<< 2) + 1; n
< 4; n
++) {
1465 if (l2nb
<= dcp
->stree
[m
+ n
]) {
1471 jfs_error(bmp
->db_ipbmap
->i_sb
,
1472 "dbAllocAG: failed descending stree");
1473 release_metapage(mp
);
1478 /* determine the block number within the file system
1479 * that corresponds to this leaf.
1481 if (bmp
->db_aglevel
== 2)
1483 else if (bmp
->db_aglevel
== 1)
1484 blkno
&= ~(MAXL1SIZE
- 1);
1485 else /* bmp->db_aglevel == 0 */
1486 blkno
&= ~(MAXL0SIZE
- 1);
1489 ((s64
) (ti
- le32_to_cpu(dcp
->leafidx
))) << budmin
;
1491 /* release the buffer in preparation for going down
1492 * the next level of dmap control pages.
1494 release_metapage(mp
);
1496 /* check if we need to continue to search down the lower
1497 * level dmap control pages. we need to if the number of
1498 * blocks required is less than maximum number of blocks
1499 * described at the next lower level.
1501 if (l2nb
< budmin
) {
1503 /* search the lower level dmap control pages to get
1504 * the starting block number of the dmap that
1505 * contains or starts off the free space.
1508 dbFindCtl(bmp
, l2nb
, bmp
->db_aglevel
- 1,
1510 if (rc
== -ENOSPC
) {
1511 jfs_error(bmp
->db_ipbmap
->i_sb
,
1512 "dbAllocAG: control page "
1520 /* allocate the blocks.
1522 rc
= dbAllocCtl(bmp
, nblocks
, l2nb
, blkno
, results
);
1523 if (rc
== -ENOSPC
) {
1524 jfs_error(bmp
->db_ipbmap
->i_sb
,
1525 "dbAllocAG: unable to allocate blocks");
1531 /* no space in the allocation group. release the buffer and
1534 release_metapage(mp
);
1541 * NAME: dbAllocAny()
1543 * FUNCTION: attempt to allocate the specified number of contiguous
1544 * free blocks anywhere in the file system.
1546 * dbAllocAny() attempts to find the sufficient free space by
1547 * searching down the dmap control pages, starting with the
1548 * highest level (i.e. L0, L1, L2) control page. if free space
1549 * large enough to satisfy the desired free space is found, the
1550 * desired free space is allocated.
1553 * bmp - pointer to bmap descriptor
1554 * nblocks - actual number of contiguous free blocks desired.
1555 * l2nb - log2 number of contiguous free blocks desired.
1556 * results - on successful return, set to the starting block number
1557 * of the newly allocated range.
1561 * -ENOSPC - insufficient disk resources
1564 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1566 static int dbAllocAny(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64
* results
)
1571 /* starting with the top level dmap control page, search
1572 * down the dmap control levels for sufficient free space.
1573 * if free space is found, dbFindCtl() returns the starting
1574 * block number of the dmap that contains or starts off the
1575 * range of free space.
1577 if ((rc
= dbFindCtl(bmp
, l2nb
, bmp
->db_maxlevel
, &blkno
)))
1580 /* allocate the blocks.
1582 rc
= dbAllocCtl(bmp
, nblocks
, l2nb
, blkno
, results
);
1583 if (rc
== -ENOSPC
) {
1584 jfs_error(bmp
->db_ipbmap
->i_sb
,
1585 "dbAllocAny: unable to allocate blocks");
1595 * FUNCTION: starting at a specified dmap control page level and block
1596 * number, search down the dmap control levels for a range of
1597 * contiguous free blocks large enough to satisfy an allocation
1598 * request for the specified number of free blocks.
1600 * if sufficient contiguous free blocks are found, this routine
1601 * returns the starting block number within a dmap page that
1602 * contains or starts a range of contiqious free blocks that
1603 * is sufficient in size.
1606 * bmp - pointer to bmap descriptor
1607 * level - starting dmap control page level.
1608 * l2nb - log2 number of contiguous free blocks desired.
1609 * *blkno - on entry, starting block number for conducting the search.
1610 * on successful return, the first block within a dmap page
1611 * that contains or starts a range of contiguous free blocks.
1615 * -ENOSPC - insufficient disk resources
1618 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1620 static int dbFindCtl(struct bmap
* bmp
, int l2nb
, int level
, s64
* blkno
)
1622 int rc
, leafidx
, lev
;
1624 struct dmapctl
*dcp
;
1626 struct metapage
*mp
;
1628 /* starting at the specified dmap control page level and block
1629 * number, search down the dmap control levels for the starting
1630 * block number of a dmap page that contains or starts off
1631 * sufficient free blocks.
1633 for (lev
= level
, b
= *blkno
; lev
>= 0; lev
--) {
1634 /* get the buffer of the dmap control page for the block
1635 * number and level (i.e. L0, L1, L2).
1637 lblkno
= BLKTOCTL(b
, bmp
->db_l2nbperpage
, lev
);
1638 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1641 dcp
= (struct dmapctl
*) mp
->data
;
1642 budmin
= dcp
->budmin
;
1644 if (dcp
->leafidx
!= cpu_to_le32(CTLLEAFIND
)) {
1645 jfs_error(bmp
->db_ipbmap
->i_sb
,
1646 "dbFindCtl: Corrupt dmapctl page");
1647 release_metapage(mp
);
1651 /* search the tree within the dmap control page for
1652 * sufficent free space. if sufficient free space is found,
1653 * dbFindLeaf() returns the index of the leaf at which
1654 * free space was found.
1656 rc
= dbFindLeaf((dmtree_t
*) dcp
, l2nb
, &leafidx
);
1658 /* release the buffer.
1660 release_metapage(mp
);
1666 jfs_error(bmp
->db_ipbmap
->i_sb
,
1667 "dbFindCtl: dmap inconsistent");
1673 /* adjust the block number to reflect the location within
1674 * the dmap control page (i.e. the leaf) at which free
1677 b
+= (((s64
) leafidx
) << budmin
);
1679 /* we stop the search at this dmap control page level if
1680 * the number of blocks required is greater than or equal
1681 * to the maximum number of blocks described at the next
1694 * NAME: dbAllocCtl()
1696 * FUNCTION: attempt to allocate a specified number of contiguous
1697 * blocks starting within a specific dmap.
1699 * this routine is called by higher level routines that search
1700 * the dmap control pages above the actual dmaps for contiguous
1701 * free space. the result of successful searches by these
1702 * routines are the starting block numbers within dmaps, with
1703 * the dmaps themselves containing the desired contiguous free
1704 * space or starting a contiguous free space of desired size
1705 * that is made up of the blocks of one or more dmaps. these
1706 * calls should not fail due to insufficent resources.
1708 * this routine is called in some cases where it is not known
1709 * whether it will fail due to insufficient resources. more
1710 * specifically, this occurs when allocating from an allocation
1711 * group whose size is equal to the number of blocks per dmap.
1712 * in this case, the dmap control pages are not examined prior
1713 * to calling this routine (to save pathlength) and the call
1716 * for a request size that fits within a dmap, this routine relies
1717 * upon the dmap's dmtree to find the requested contiguous free
1718 * space. for request sizes that are larger than a dmap, the
1719 * requested free space will start at the first block of the
1720 * first dmap (i.e. blkno).
1723 * bmp - pointer to bmap descriptor
1724 * nblocks - actual number of contiguous free blocks to allocate.
1725 * l2nb - log2 number of contiguous free blocks to allocate.
1726 * blkno - starting block number of the dmap to start the allocation
1728 * results - on successful return, set to the starting block number
1729 * of the newly allocated range.
1733 * -ENOSPC - insufficient disk resources
1736 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1739 dbAllocCtl(struct bmap
* bmp
, s64 nblocks
, int l2nb
, s64 blkno
, s64
* results
)
1743 struct metapage
*mp
;
1746 /* check if the allocation request is confined to a single dmap.
1748 if (l2nb
<= L2BPERDMAP
) {
1749 /* get the buffer for the dmap.
1751 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
1752 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1755 dp
= (struct dmap
*) mp
->data
;
1757 /* try to allocate the blocks.
1759 rc
= dbAllocDmapLev(bmp
, dp
, (int) nblocks
, l2nb
, results
);
1761 mark_metapage_dirty(mp
);
1763 release_metapage(mp
);
1768 /* allocation request involving multiple dmaps. it must start on
1771 assert((blkno
& (BPERDMAP
- 1)) == 0);
1773 /* allocate the blocks dmap by dmap.
1775 for (n
= nblocks
, b
= blkno
; n
> 0; n
-= nb
, b
+= nb
) {
1776 /* get the buffer for the dmap.
1778 lblkno
= BLKTODMAP(b
, bmp
->db_l2nbperpage
);
1779 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1784 dp
= (struct dmap
*) mp
->data
;
1786 /* the dmap better be all free.
1788 if (dp
->tree
.stree
[ROOT
] != L2BPERDMAP
) {
1789 release_metapage(mp
);
1790 jfs_error(bmp
->db_ipbmap
->i_sb
,
1791 "dbAllocCtl: the dmap is not all free");
1796 /* determine how many blocks to allocate from this dmap.
1798 nb
= min(n
, (s64
)BPERDMAP
);
1800 /* allocate the blocks from the dmap.
1802 if ((rc
= dbAllocDmap(bmp
, dp
, b
, nb
))) {
1803 release_metapage(mp
);
1807 /* write the buffer.
1812 /* set the results (starting block number) and return.
1817 /* something failed in handling an allocation request involving
1818 * multiple dmaps. we'll try to clean up by backing out any
1819 * allocation that has already happened for this request. if
1820 * we fail in backing out the allocation, we'll mark the file
1821 * system to indicate that blocks have been leaked.
1825 /* try to backout the allocations dmap by dmap.
1827 for (n
= nblocks
- n
, b
= blkno
; n
> 0;
1828 n
-= BPERDMAP
, b
+= BPERDMAP
) {
1829 /* get the buffer for this dmap.
1831 lblkno
= BLKTODMAP(b
, bmp
->db_l2nbperpage
);
1832 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
1834 /* could not back out. mark the file system
1835 * to indicate that we have leaked blocks.
1837 jfs_error(bmp
->db_ipbmap
->i_sb
,
1838 "dbAllocCtl: I/O Error: Block Leakage.");
1841 dp
= (struct dmap
*) mp
->data
;
1843 /* free the blocks is this dmap.
1845 if (dbFreeDmap(bmp
, dp
, b
, BPERDMAP
)) {
1846 /* could not back out. mark the file system
1847 * to indicate that we have leaked blocks.
1849 release_metapage(mp
);
1850 jfs_error(bmp
->db_ipbmap
->i_sb
,
1851 "dbAllocCtl: Block Leakage.");
1855 /* write the buffer.
1865 * NAME: dbAllocDmapLev()
1867 * FUNCTION: attempt to allocate a specified number of contiguous blocks
1868 * from a specified dmap.
1870 * this routine checks if the contiguous blocks are available.
1871 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1875 * mp - pointer to bmap descriptor
1876 * dp - pointer to dmap to attempt to allocate blocks from.
1877 * l2nb - log2 number of contiguous block desired.
1878 * nblocks - actual number of contiguous block desired.
1879 * results - on successful return, set to the starting block number
1880 * of the newly allocated range.
1884 * -ENOSPC - insufficient disk resources
1887 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1888 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1891 dbAllocDmapLev(struct bmap
* bmp
,
1892 struct dmap
* dp
, int nblocks
, int l2nb
, s64
* results
)
1897 /* can't be more than a dmaps worth of blocks */
1898 assert(l2nb
<= L2BPERDMAP
);
1900 /* search the tree within the dmap page for sufficient
1901 * free space. if sufficient free space is found, dbFindLeaf()
1902 * returns the index of the leaf at which free space was found.
1904 if (dbFindLeaf((dmtree_t
*) & dp
->tree
, l2nb
, &leafidx
))
1907 /* determine the block number within the file system corresponding
1908 * to the leaf at which free space was found.
1910 blkno
= le64_to_cpu(dp
->start
) + (leafidx
<< L2DBWORD
);
1912 /* if not all bits of the dmap word are free, get the starting
1913 * bit number within the dmap word of the required string of free
1914 * bits and adjust the block number with this value.
1916 if (dp
->tree
.stree
[leafidx
+ LEAFIND
] < BUDMIN
)
1917 blkno
+= dbFindBits(le32_to_cpu(dp
->wmap
[leafidx
]), l2nb
);
1919 /* allocate the blocks */
1920 if ((rc
= dbAllocDmap(bmp
, dp
, blkno
, nblocks
)) == 0)
1928 * NAME: dbAllocDmap()
1930 * FUNCTION: adjust the disk allocation map to reflect the allocation
1931 * of a specified block range within a dmap.
1933 * this routine allocates the specified blocks from the dmap
1934 * through a call to dbAllocBits(). if the allocation of the
1935 * block range causes the maximum string of free blocks within
1936 * the dmap to change (i.e. the value of the root of the dmap's
1937 * dmtree), this routine will cause this change to be reflected
1938 * up through the appropriate levels of the dmap control pages
1939 * by a call to dbAdjCtl() for the L0 dmap control page that
1943 * bmp - pointer to bmap descriptor
1944 * dp - pointer to dmap to allocate the block range from.
1945 * blkno - starting block number of the block to be allocated.
1946 * nblocks - number of blocks to be allocated.
1952 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
1954 static int dbAllocDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
1960 /* save the current value of the root (i.e. maximum free string)
1963 oldroot
= dp
->tree
.stree
[ROOT
];
1965 /* allocate the specified (blocks) bits */
1966 dbAllocBits(bmp
, dp
, blkno
, nblocks
);
1968 /* if the root has not changed, done. */
1969 if (dp
->tree
.stree
[ROOT
] == oldroot
)
1972 /* root changed. bubble the change up to the dmap control pages.
1973 * if the adjustment of the upper level control pages fails,
1974 * backout the bit allocation (thus making everything consistent).
1976 if ((rc
= dbAdjCtl(bmp
, blkno
, dp
->tree
.stree
[ROOT
], 1, 0)))
1977 dbFreeBits(bmp
, dp
, blkno
, nblocks
);
1984 * NAME: dbFreeDmap()
1986 * FUNCTION: adjust the disk allocation map to reflect the allocation
1987 * of a specified block range within a dmap.
1989 * this routine frees the specified blocks from the dmap through
1990 * a call to dbFreeBits(). if the deallocation of the block range
1991 * causes the maximum string of free blocks within the dmap to
1992 * change (i.e. the value of the root of the dmap's dmtree), this
1993 * routine will cause this change to be reflected up through the
1994 * appropriate levels of the dmap control pages by a call to
1995 * dbAdjCtl() for the L0 dmap control page that covers this dmap.
1998 * bmp - pointer to bmap descriptor
1999 * dp - pointer to dmap to free the block range from.
2000 * blkno - starting block number of the block to be freed.
2001 * nblocks - number of blocks to be freed.
2007 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2009 static int dbFreeDmap(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
2015 /* save the current value of the root (i.e. maximum free string)
2018 oldroot
= dp
->tree
.stree
[ROOT
];
2020 /* free the specified (blocks) bits */
2021 rc
= dbFreeBits(bmp
, dp
, blkno
, nblocks
);
2023 /* if error or the root has not changed, done. */
2024 if (rc
|| (dp
->tree
.stree
[ROOT
] == oldroot
))
2027 /* root changed. bubble the change up to the dmap control pages.
2028 * if the adjustment of the upper level control pages fails,
2029 * backout the deallocation.
2031 if ((rc
= dbAdjCtl(bmp
, blkno
, dp
->tree
.stree
[ROOT
], 0, 0))) {
2032 word
= (blkno
& (BPERDMAP
- 1)) >> L2DBWORD
;
2034 /* as part of backing out the deallocation, we will have
2035 * to back split the dmap tree if the deallocation caused
2036 * the freed blocks to become part of a larger binary buddy
2039 if (dp
->tree
.stree
[word
] == NOFREE
)
2040 dbBackSplit((dmtree_t
*) & dp
->tree
, word
);
2042 dbAllocBits(bmp
, dp
, blkno
, nblocks
);
2050 * NAME: dbAllocBits()
2052 * FUNCTION: allocate a specified block range from a dmap.
2054 * this routine updates the dmap to reflect the working
2055 * state allocation of the specified block range. it directly
2056 * updates the bits of the working map and causes the adjustment
2057 * of the binary buddy system described by the dmap's dmtree
2058 * leaves to reflect the bits allocated. it also causes the
2059 * dmap's dmtree, as a whole, to reflect the allocated range.
2062 * bmp - pointer to bmap descriptor
2063 * dp - pointer to dmap to allocate bits from.
2064 * blkno - starting block number of the bits to be allocated.
2065 * nblocks - number of bits to be allocated.
2067 * RETURN VALUES: none
2069 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2071 static void dbAllocBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
2074 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, nw
, agno
;
2075 dmtree_t
*tp
= (dmtree_t
*) & dp
->tree
;
2079 /* pick up a pointer to the leaves of the dmap tree */
2080 leaf
= dp
->tree
.stree
+ LEAFIND
;
2082 /* determine the bit number and word within the dmap of the
2085 dbitno
= blkno
& (BPERDMAP
- 1);
2086 word
= dbitno
>> L2DBWORD
;
2088 /* block range better be within the dmap */
2089 assert(dbitno
+ nblocks
<= BPERDMAP
);
2091 /* allocate the bits of the dmap's words corresponding to the block
2092 * range. not all bits of the first and last words may be contained
2093 * within the block range. if this is the case, we'll work against
2094 * those words (i.e. partial first and/or last) on an individual basis
2095 * (a single pass), allocating the bits of interest by hand and
2096 * updating the leaf corresponding to the dmap word. a single pass
2097 * will be used for all dmap words fully contained within the
2098 * specified range. within this pass, the bits of all fully contained
2099 * dmap words will be marked as free in a single shot and the leaves
2100 * will be updated. a single leaf may describe the free space of
2101 * multiple dmap words, so we may update only a subset of the actual
2102 * leaves corresponding to the dmap words of the block range.
2104 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
2105 /* determine the bit number within the word and
2106 * the number of bits within the word.
2108 wbitno
= dbitno
& (DBWORD
- 1);
2109 nb
= min(rembits
, DBWORD
- wbitno
);
2111 /* check if only part of a word is to be allocated.
2114 /* allocate (set to 1) the appropriate bits within
2117 dp
->wmap
[word
] |= cpu_to_le32(ONES
<< (DBWORD
- nb
)
2120 /* update the leaf for this dmap word. in addition
2121 * to setting the leaf value to the binary buddy max
2122 * of the updated dmap word, dbSplit() will split
2123 * the binary system of the leaves if need be.
2125 dbSplit(tp
, word
, BUDMIN
,
2126 dbMaxBud((u8
*) & dp
->wmap
[word
]));
2130 /* one or more dmap words are fully contained
2131 * within the block range. determine how many
2132 * words and allocate (set to 1) the bits of these
2135 nwords
= rembits
>> L2DBWORD
;
2136 memset(&dp
->wmap
[word
], (int) ONES
, nwords
* 4);
2138 /* determine how many bits.
2140 nb
= nwords
<< L2DBWORD
;
2142 /* now update the appropriate leaves to reflect
2143 * the allocated words.
2145 for (; nwords
> 0; nwords
-= nw
) {
2146 if (leaf
[word
] < BUDMIN
) {
2147 jfs_error(bmp
->db_ipbmap
->i_sb
,
2148 "dbAllocBits: leaf page "
2153 /* determine what the leaf value should be
2154 * updated to as the minimum of the l2 number
2155 * of bits being allocated and the l2 number
2156 * of bits currently described by this leaf.
2158 size
= min((int)leaf
[word
], NLSTOL2BSZ(nwords
));
2160 /* update the leaf to reflect the allocation.
2161 * in addition to setting the leaf value to
2162 * NOFREE, dbSplit() will split the binary
2163 * system of the leaves to reflect the current
2164 * allocation (size).
2166 dbSplit(tp
, word
, size
, NOFREE
);
2168 /* get the number of dmap words handled */
2169 nw
= BUDSIZE(size
, BUDMIN
);
2175 /* update the free count for this dmap */
2176 le32_add_cpu(&dp
->nfree
, -nblocks
);
2180 /* if this allocation group is completely free,
2181 * update the maximum allocation group number if this allocation
2182 * group is the new max.
2184 agno
= blkno
>> bmp
->db_agl2size
;
2185 if (agno
> bmp
->db_maxag
)
2186 bmp
->db_maxag
= agno
;
2188 /* update the free count for the allocation group and map */
2189 bmp
->db_agfree
[agno
] -= nblocks
;
2190 bmp
->db_nfree
-= nblocks
;
2197 * NAME: dbFreeBits()
2199 * FUNCTION: free a specified block range from a dmap.
2201 * this routine updates the dmap to reflect the working
2202 * state allocation of the specified block range. it directly
2203 * updates the bits of the working map and causes the adjustment
2204 * of the binary buddy system described by the dmap's dmtree
2205 * leaves to reflect the bits freed. it also causes the dmap's
2206 * dmtree, as a whole, to reflect the deallocated range.
2209 * bmp - pointer to bmap descriptor
2210 * dp - pointer to dmap to free bits from.
2211 * blkno - starting block number of the bits to be freed.
2212 * nblocks - number of bits to be freed.
2214 * RETURN VALUES: 0 for success
2216 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2218 static int dbFreeBits(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
2221 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, nw
, agno
;
2222 dmtree_t
*tp
= (dmtree_t
*) & dp
->tree
;
2226 /* determine the bit number and word within the dmap of the
2229 dbitno
= blkno
& (BPERDMAP
- 1);
2230 word
= dbitno
>> L2DBWORD
;
2232 /* block range better be within the dmap.
2234 assert(dbitno
+ nblocks
<= BPERDMAP
);
2236 /* free the bits of the dmaps words corresponding to the block range.
2237 * not all bits of the first and last words may be contained within
2238 * the block range. if this is the case, we'll work against those
2239 * words (i.e. partial first and/or last) on an individual basis
2240 * (a single pass), freeing the bits of interest by hand and updating
2241 * the leaf corresponding to the dmap word. a single pass will be used
2242 * for all dmap words fully contained within the specified range.
2243 * within this pass, the bits of all fully contained dmap words will
2244 * be marked as free in a single shot and the leaves will be updated. a
2245 * single leaf may describe the free space of multiple dmap words,
2246 * so we may update only a subset of the actual leaves corresponding
2247 * to the dmap words of the block range.
2249 * dbJoin() is used to update leaf values and will join the binary
2250 * buddy system of the leaves if the new leaf values indicate this
2253 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
2254 /* determine the bit number within the word and
2255 * the number of bits within the word.
2257 wbitno
= dbitno
& (DBWORD
- 1);
2258 nb
= min(rembits
, DBWORD
- wbitno
);
2260 /* check if only part of a word is to be freed.
2263 /* free (zero) the appropriate bits within this
2267 cpu_to_le32(~(ONES
<< (DBWORD
- nb
)
2270 /* update the leaf for this dmap word.
2272 rc
= dbJoin(tp
, word
,
2273 dbMaxBud((u8
*) & dp
->wmap
[word
]));
2279 /* one or more dmap words are fully contained
2280 * within the block range. determine how many
2281 * words and free (zero) the bits of these words.
2283 nwords
= rembits
>> L2DBWORD
;
2284 memset(&dp
->wmap
[word
], 0, nwords
* 4);
2286 /* determine how many bits.
2288 nb
= nwords
<< L2DBWORD
;
2290 /* now update the appropriate leaves to reflect
2293 for (; nwords
> 0; nwords
-= nw
) {
2294 /* determine what the leaf value should be
2295 * updated to as the minimum of the l2 number
2296 * of bits being freed and the l2 (max) number
2297 * of bits that can be described by this leaf.
2301 (word
, L2LPERDMAP
, BUDMIN
),
2302 NLSTOL2BSZ(nwords
));
2306 rc
= dbJoin(tp
, word
, size
);
2310 /* get the number of dmap words handled.
2312 nw
= BUDSIZE(size
, BUDMIN
);
2318 /* update the free count for this dmap.
2320 le32_add_cpu(&dp
->nfree
, nblocks
);
2324 /* update the free count for the allocation group and
2327 agno
= blkno
>> bmp
->db_agl2size
;
2328 bmp
->db_nfree
+= nblocks
;
2329 bmp
->db_agfree
[agno
] += nblocks
;
2331 /* check if this allocation group is not completely free and
2332 * if it is currently the maximum (rightmost) allocation group.
2333 * if so, establish the new maximum allocation group number by
2334 * searching left for the first allocation group with allocation.
2336 if ((bmp
->db_agfree
[agno
] == bmp
->db_agsize
&& agno
== bmp
->db_maxag
) ||
2337 (agno
== bmp
->db_numag
- 1 &&
2338 bmp
->db_agfree
[agno
] == (bmp
-> db_mapsize
& (BPERDMAP
- 1)))) {
2339 while (bmp
->db_maxag
> 0) {
2341 if (bmp
->db_agfree
[bmp
->db_maxag
] !=
2346 /* re-establish the allocation group preference if the
2347 * current preference is right of the maximum allocation
2350 if (bmp
->db_agpref
> bmp
->db_maxag
)
2351 bmp
->db_agpref
= bmp
->db_maxag
;
2363 * FUNCTION: adjust a dmap control page at a specified level to reflect
2364 * the change in a lower level dmap or dmap control page's
2365 * maximum string of free blocks (i.e. a change in the root
2366 * of the lower level object's dmtree) due to the allocation
2367 * or deallocation of a range of blocks with a single dmap.
2369 * on entry, this routine is provided with the new value of
2370 * the lower level dmap or dmap control page root and the
2371 * starting block number of the block range whose allocation
2372 * or deallocation resulted in the root change. this range
2373 * is respresented by a single leaf of the current dmapctl
2374 * and the leaf will be updated with this value, possibly
2375 * causing a binary buddy system within the leaves to be
2376 * split or joined. the update may also cause the dmapctl's
2377 * dmtree to be updated.
2379 * if the adjustment of the dmap control page, itself, causes its
2380 * root to change, this change will be bubbled up to the next dmap
2381 * control level by a recursive call to this routine, specifying
2382 * the new root value and the next dmap control page level to
2385 * bmp - pointer to bmap descriptor
2386 * blkno - the first block of a block range within a dmap. it is
2387 * the allocation or deallocation of this block range that
2388 * requires the dmap control page to be adjusted.
2389 * newval - the new value of the lower level dmap or dmap control
2391 * alloc - 'true' if adjustment is due to an allocation.
2392 * level - current level of dmap control page (i.e. L0, L1, L2) to
2399 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2402 dbAdjCtl(struct bmap
* bmp
, s64 blkno
, int newval
, int alloc
, int level
)
2404 struct metapage
*mp
;
2408 struct dmapctl
*dcp
;
2411 /* get the buffer for the dmap control page for the specified
2412 * block number and control page level.
2414 lblkno
= BLKTOCTL(blkno
, bmp
->db_l2nbperpage
, level
);
2415 mp
= read_metapage(bmp
->db_ipbmap
, lblkno
, PSIZE
, 0);
2418 dcp
= (struct dmapctl
*) mp
->data
;
2420 if (dcp
->leafidx
!= cpu_to_le32(CTLLEAFIND
)) {
2421 jfs_error(bmp
->db_ipbmap
->i_sb
,
2422 "dbAdjCtl: Corrupt dmapctl page");
2423 release_metapage(mp
);
2427 /* determine the leaf number corresponding to the block and
2428 * the index within the dmap control tree.
2430 leafno
= BLKTOCTLLEAF(blkno
, dcp
->budmin
);
2431 ti
= leafno
+ le32_to_cpu(dcp
->leafidx
);
2433 /* save the current leaf value and the current root level (i.e.
2434 * maximum l2 free string described by this dmapctl).
2436 oldval
= dcp
->stree
[ti
];
2437 oldroot
= dcp
->stree
[ROOT
];
2439 /* check if this is a control page update for an allocation.
2440 * if so, update the leaf to reflect the new leaf value using
2441 * dbSplit(); otherwise (deallocation), use dbJoin() to update
2442 * the leaf with the new value. in addition to updating the
2443 * leaf, dbSplit() will also split the binary buddy system of
2444 * the leaves, if required, and bubble new values within the
2445 * dmapctl tree, if required. similarly, dbJoin() will join
2446 * the binary buddy system of leaves and bubble new values up
2447 * the dmapctl tree as required by the new leaf value.
2450 /* check if we are in the middle of a binary buddy
2451 * system. this happens when we are performing the
2452 * first allocation out of an allocation group that
2453 * is part (not the first part) of a larger binary
2454 * buddy system. if we are in the middle, back split
2455 * the system prior to calling dbSplit() which assumes
2456 * that it is at the front of a binary buddy system.
2458 if (oldval
== NOFREE
) {
2459 rc
= dbBackSplit((dmtree_t
*) dcp
, leafno
);
2462 oldval
= dcp
->stree
[ti
];
2464 dbSplit((dmtree_t
*) dcp
, leafno
, dcp
->budmin
, newval
);
2466 rc
= dbJoin((dmtree_t
*) dcp
, leafno
, newval
);
2471 /* check if the root of the current dmap control page changed due
2472 * to the update and if the current dmap control page is not at
2473 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e.
2474 * root changed and this is not the top level), call this routine
2475 * again (recursion) for the next higher level of the mapping to
2476 * reflect the change in root for the current dmap control page.
2478 if (dcp
->stree
[ROOT
] != oldroot
) {
2479 /* are we below the top level of the map. if so,
2480 * bubble the root up to the next higher level.
2482 if (level
< bmp
->db_maxlevel
) {
2483 /* bubble up the new root of this dmap control page to
2487 dbAdjCtl(bmp
, blkno
, dcp
->stree
[ROOT
], alloc
,
2489 /* something went wrong in bubbling up the new
2490 * root value, so backout the changes to the
2491 * current dmap control page.
2494 dbJoin((dmtree_t
*) dcp
, leafno
,
2497 /* the dbJoin() above might have
2498 * caused a larger binary buddy system
2499 * to form and we may now be in the
2500 * middle of it. if this is the case,
2501 * back split the buddies.
2503 if (dcp
->stree
[ti
] == NOFREE
)
2504 dbBackSplit((dmtree_t
*)
2506 dbSplit((dmtree_t
*) dcp
, leafno
,
2507 dcp
->budmin
, oldval
);
2510 /* release the buffer and return the error.
2512 release_metapage(mp
);
2516 /* we're at the top level of the map. update
2517 * the bmap control page to reflect the size
2518 * of the maximum free buddy system.
2520 assert(level
== bmp
->db_maxlevel
);
2521 if (bmp
->db_maxfreebud
!= oldroot
) {
2522 jfs_error(bmp
->db_ipbmap
->i_sb
,
2523 "dbAdjCtl: the maximum free buddy is "
2524 "not the old root");
2526 bmp
->db_maxfreebud
= dcp
->stree
[ROOT
];
2530 /* write the buffer.
2541 * FUNCTION: update the leaf of a dmtree with a new value, splitting
2542 * the leaf from the binary buddy system of the dmtree's
2543 * leaves, as required.
2546 * tp - pointer to the tree containing the leaf.
2547 * leafno - the number of the leaf to be updated.
2548 * splitsz - the size the binary buddy system starting at the leaf
2549 * must be split to, specified as the log2 number of blocks.
2550 * newval - the new value for the leaf.
2552 * RETURN VALUES: none
2554 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2556 static void dbSplit(dmtree_t
* tp
, int leafno
, int splitsz
, int newval
)
2560 s8
*leaf
= tp
->dmt_stree
+ le32_to_cpu(tp
->dmt_leafidx
);
2562 /* check if the leaf needs to be split.
2564 if (leaf
[leafno
] > tp
->dmt_budmin
) {
2565 /* the split occurs by cutting the buddy system in half
2566 * at the specified leaf until we reach the specified
2567 * size. pick up the starting split size (current size
2568 * - 1 in l2) and the corresponding buddy size.
2570 cursz
= leaf
[leafno
] - 1;
2571 budsz
= BUDSIZE(cursz
, tp
->dmt_budmin
);
2573 /* split until we reach the specified size.
2575 while (cursz
>= splitsz
) {
2576 /* update the buddy's leaf with its new value.
2578 dbAdjTree(tp
, leafno
^ budsz
, cursz
);
2580 /* on to the next size and buddy.
2587 /* adjust the dmap tree to reflect the specified leaf's new
2590 dbAdjTree(tp
, leafno
, newval
);
2595 * NAME: dbBackSplit()
2597 * FUNCTION: back split the binary buddy system of dmtree leaves
2598 * that hold a specified leaf until the specified leaf
2599 * starts its own binary buddy system.
2601 * the allocators typically perform allocations at the start
2602 * of binary buddy systems and dbSplit() is used to accomplish
2603 * any required splits. in some cases, however, allocation
2604 * may occur in the middle of a binary system and requires a
2605 * back split, with the split proceeding out from the middle of
2606 * the system (less efficient) rather than the start of the
2607 * system (more efficient). the cases in which a back split
2608 * is required are rare and are limited to the first allocation
2609 * within an allocation group which is a part (not first part)
2610 * of a larger binary buddy system and a few exception cases
2611 * in which a previous join operation must be backed out.
2614 * tp - pointer to the tree containing the leaf.
2615 * leafno - the number of the leaf to be updated.
2617 * RETURN VALUES: none
2619 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2621 static int dbBackSplit(dmtree_t
* tp
, int leafno
)
2623 int budsz
, bud
, w
, bsz
, size
;
2625 s8
*leaf
= tp
->dmt_stree
+ le32_to_cpu(tp
->dmt_leafidx
);
2627 /* leaf should be part (not first part) of a binary
2630 assert(leaf
[leafno
] == NOFREE
);
2632 /* the back split is accomplished by iteratively finding the leaf
2633 * that starts the buddy system that contains the specified leaf and
2634 * splitting that system in two. this iteration continues until
2635 * the specified leaf becomes the start of a buddy system.
2637 * determine maximum possible l2 size for the specified leaf.
2640 LITOL2BSZ(leafno
, le32_to_cpu(tp
->dmt_l2nleafs
),
2643 /* determine the number of leaves covered by this size. this
2644 * is the buddy size that we will start with as we search for
2645 * the buddy system that contains the specified leaf.
2647 budsz
= BUDSIZE(size
, tp
->dmt_budmin
);
2651 while (leaf
[leafno
] == NOFREE
) {
2652 /* find the leftmost buddy leaf.
2654 for (w
= leafno
, bsz
= budsz
;; bsz
<<= 1,
2655 w
= (w
< bud
) ? w
: bud
) {
2656 if (bsz
>= le32_to_cpu(tp
->dmt_nleafs
)) {
2657 jfs_err("JFS: block map error in dbBackSplit");
2661 /* determine the buddy.
2665 /* check if this buddy is the start of the system.
2667 if (leaf
[bud
] != NOFREE
) {
2668 /* split the leaf at the start of the
2671 cursz
= leaf
[bud
] - 1;
2672 dbSplit(tp
, bud
, cursz
, cursz
);
2678 if (leaf
[leafno
] != size
) {
2679 jfs_err("JFS: wrong leaf value in dbBackSplit");
2689 * FUNCTION: update the leaf of a dmtree with a new value, joining
2690 * the leaf with other leaves of the dmtree into a multi-leaf
2691 * binary buddy system, as required.
2694 * tp - pointer to the tree containing the leaf.
2695 * leafno - the number of the leaf to be updated.
2696 * newval - the new value for the leaf.
2698 * RETURN VALUES: none
2700 static int dbJoin(dmtree_t
* tp
, int leafno
, int newval
)
2705 /* can the new leaf value require a join with other leaves ?
2707 if (newval
>= tp
->dmt_budmin
) {
2708 /* pickup a pointer to the leaves of the tree.
2710 leaf
= tp
->dmt_stree
+ le32_to_cpu(tp
->dmt_leafidx
);
2712 /* try to join the specified leaf into a large binary
2713 * buddy system. the join proceeds by attempting to join
2714 * the specified leafno with its buddy (leaf) at new value.
2715 * if the join occurs, we attempt to join the left leaf
2716 * of the joined buddies with its buddy at new value + 1.
2717 * we continue to join until we find a buddy that cannot be
2718 * joined (does not have a value equal to the size of the
2719 * last join) or until all leaves have been joined into a
2722 * get the buddy size (number of words covered) of
2725 budsz
= BUDSIZE(newval
, tp
->dmt_budmin
);
2729 while (budsz
< le32_to_cpu(tp
->dmt_nleafs
)) {
2730 /* get the buddy leaf.
2732 buddy
= leafno
^ budsz
;
2734 /* if the leaf's new value is greater than its
2735 * buddy's value, we join no more.
2737 if (newval
> leaf
[buddy
])
2740 /* It shouldn't be less */
2741 if (newval
< leaf
[buddy
])
2744 /* check which (leafno or buddy) is the left buddy.
2745 * the left buddy gets to claim the blocks resulting
2746 * from the join while the right gets to claim none.
2747 * the left buddy is also eligable to participate in
2748 * a join at the next higher level while the right
2752 if (leafno
< buddy
) {
2753 /* leafno is the left buddy.
2755 dbAdjTree(tp
, buddy
, NOFREE
);
2757 /* buddy is the left buddy and becomes
2760 dbAdjTree(tp
, leafno
, NOFREE
);
2764 /* on to try the next join.
2771 /* update the leaf value.
2773 dbAdjTree(tp
, leafno
, newval
);
2782 * FUNCTION: update a leaf of a dmtree with a new value, adjusting
2783 * the dmtree, as required, to reflect the new leaf value.
2784 * the combination of any buddies must already be done before
2788 * tp - pointer to the tree to be adjusted.
2789 * leafno - the number of the leaf to be updated.
2790 * newval - the new value for the leaf.
2792 * RETURN VALUES: none
2794 static void dbAdjTree(dmtree_t
* tp
, int leafno
, int newval
)
2799 /* pick up the index of the leaf for this leafno.
2801 lp
= leafno
+ le32_to_cpu(tp
->dmt_leafidx
);
2803 /* is the current value the same as the old value ? if so,
2804 * there is nothing to do.
2806 if (tp
->dmt_stree
[lp
] == newval
)
2809 /* set the new value.
2811 tp
->dmt_stree
[lp
] = newval
;
2813 /* bubble the new value up the tree as required.
2815 for (k
= 0; k
< le32_to_cpu(tp
->dmt_height
); k
++) {
2816 /* get the index of the first leaf of the 4 leaf
2817 * group containing the specified leaf (leafno).
2819 lp
= ((lp
- 1) & ~0x03) + 1;
2821 /* get the index of the parent of this 4 leaf group.
2825 /* determine the maximum of the 4 leaves.
2827 max
= TREEMAX(&tp
->dmt_stree
[lp
]);
2829 /* if the maximum of the 4 is the same as the
2830 * parent's value, we're done.
2832 if (tp
->dmt_stree
[pp
] == max
)
2835 /* parent gets new value.
2837 tp
->dmt_stree
[pp
] = max
;
2839 /* parent becomes leaf for next go-round.
2847 * NAME: dbFindLeaf()
2849 * FUNCTION: search a dmtree_t for sufficient free blocks, returning
2850 * the index of a leaf describing the free blocks if
2851 * sufficient free blocks are found.
2853 * the search starts at the top of the dmtree_t tree and
2854 * proceeds down the tree to the leftmost leaf with sufficient
2858 * tp - pointer to the tree to be searched.
2859 * l2nb - log2 number of free blocks to search for.
2860 * leafidx - return pointer to be set to the index of the leaf
2861 * describing at least l2nb free blocks if sufficient
2862 * free blocks are found.
2866 * -ENOSPC - insufficient free blocks.
2868 static int dbFindLeaf(dmtree_t
* tp
, int l2nb
, int *leafidx
)
2870 int ti
, n
= 0, k
, x
= 0;
2872 /* first check the root of the tree to see if there is
2873 * sufficient free space.
2875 if (l2nb
> tp
->dmt_stree
[ROOT
])
2878 /* sufficient free space available. now search down the tree
2879 * starting at the next level for the leftmost leaf that
2880 * describes sufficient free space.
2882 for (k
= le32_to_cpu(tp
->dmt_height
), ti
= 1;
2883 k
> 0; k
--, ti
= ((ti
+ n
) << 2) + 1) {
2884 /* search the four nodes at this level, starting from
2887 for (x
= ti
, n
= 0; n
< 4; n
++) {
2888 /* sufficient free space found. move to the next
2889 * level (or quit if this is the last level).
2891 if (l2nb
<= tp
->dmt_stree
[x
+ n
])
2895 /* better have found something since the higher
2896 * levels of the tree said it was here.
2901 /* set the return to the leftmost leaf describing sufficient
2904 *leafidx
= x
+ n
- le32_to_cpu(tp
->dmt_leafidx
);
2911 * NAME: dbFindBits()
2913 * FUNCTION: find a specified number of binary buddy free bits within a
2914 * dmap bitmap word value.
2916 * this routine searches the bitmap value for (1 << l2nb) free
2917 * bits at (1 << l2nb) alignments within the value.
2920 * word - dmap bitmap word value.
2921 * l2nb - number of free bits specified as a log2 number.
2924 * starting bit number of free bits.
2926 static int dbFindBits(u32 word
, int l2nb
)
2931 /* get the number of bits.
2934 assert(nb
<= DBWORD
);
2936 /* complement the word so we can use a mask (i.e. 0s represent
2937 * free bits) and compute the mask.
2940 mask
= ONES
<< (DBWORD
- nb
);
2942 /* scan the word for nb free bits at nb alignments.
2944 for (bitno
= 0; mask
!= 0; bitno
+= nb
, mask
>>= nb
) {
2945 if ((mask
& word
) == mask
)
2951 /* return the bit number.
2958 * NAME: dbMaxBud(u8 *cp)
2960 * FUNCTION: determine the largest binary buddy string of free
2961 * bits within 32-bits of the map.
2964 * cp - pointer to the 32-bit value.
2967 * largest binary buddy of free bits within a dmap word.
2969 static int dbMaxBud(u8
* cp
)
2971 signed char tmp1
, tmp2
;
2973 /* check if the wmap word is all free. if so, the
2974 * free buddy size is BUDMIN.
2976 if (*((uint
*) cp
) == 0)
2979 /* check if the wmap word is half free. if so, the
2980 * free buddy size is BUDMIN-1.
2982 if (*((u16
*) cp
) == 0 || *((u16
*) cp
+ 1) == 0)
2983 return (BUDMIN
- 1);
2985 /* not all free or half free. determine the free buddy
2986 * size thru table lookup using quarters of the wmap word.
2988 tmp1
= max(budtab
[cp
[2]], budtab
[cp
[3]]);
2989 tmp2
= max(budtab
[cp
[0]], budtab
[cp
[1]]);
2990 return (max(tmp1
, tmp2
));
2995 * NAME: cnttz(uint word)
2997 * FUNCTION: determine the number of trailing zeros within a 32-bit
3001 * value - 32-bit value to be examined.
3004 * count of trailing zeros
3006 static int cnttz(u32 word
)
3010 for (n
= 0; n
< 32; n
++, word
>>= 1) {
3020 * NAME: cntlz(u32 value)
3022 * FUNCTION: determine the number of leading zeros within a 32-bit
3026 * value - 32-bit value to be examined.
3029 * count of leading zeros
3031 static int cntlz(u32 value
)
3035 for (n
= 0; n
< 32; n
++, value
<<= 1) {
3036 if (value
& HIGHORDER
)
3044 * NAME: blkstol2(s64 nb)
3046 * FUNCTION: convert a block count to its log2 value. if the block
3047 * count is not a l2 multiple, it is rounded up to the next
3048 * larger l2 multiple.
3051 * nb - number of blocks
3054 * log2 number of blocks
3056 static int blkstol2(s64 nb
)
3059 s64 mask
; /* meant to be signed */
3061 mask
= (s64
) 1 << (64 - 1);
3063 /* count the leading bits.
3065 for (l2nb
= 0; l2nb
< 64; l2nb
++, mask
>>= 1) {
3066 /* leading bit found.
3069 /* determine the l2 value.
3071 l2nb
= (64 - 1) - l2nb
;
3073 /* check if we need to round up.
3082 return 0; /* fix compiler warning */
3087 * NAME: dbAllocBottomUp()
3089 * FUNCTION: alloc the specified block range from the working block
3092 * the blocks will be alloc from the working map one dmap
3096 * ip - pointer to in-core inode;
3097 * blkno - starting block number to be freed.
3098 * nblocks - number of blocks to be freed.
3104 int dbAllocBottomUp(struct inode
*ip
, s64 blkno
, s64 nblocks
)
3106 struct metapage
*mp
;
3110 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
3111 struct bmap
*bmp
= JFS_SBI(ip
->i_sb
)->bmap
;
3113 IREAD_LOCK(ipbmap
, RDWRLOCK_DMAP
);
3115 /* block to be allocated better be within the mapsize. */
3116 ASSERT(nblocks
<= bmp
->db_mapsize
- blkno
);
3119 * allocate the blocks a dmap at a time.
3122 for (rem
= nblocks
; rem
> 0; rem
-= nb
, blkno
+= nb
) {
3123 /* release previous dmap if any */
3128 /* get the buffer for the current dmap. */
3129 lblkno
= BLKTODMAP(blkno
, bmp
->db_l2nbperpage
);
3130 mp
= read_metapage(ipbmap
, lblkno
, PSIZE
, 0);
3132 IREAD_UNLOCK(ipbmap
);
3135 dp
= (struct dmap
*) mp
->data
;
3137 /* determine the number of blocks to be allocated from
3140 nb
= min(rem
, BPERDMAP
- (blkno
& (BPERDMAP
- 1)));
3142 /* allocate the blocks. */
3143 if ((rc
= dbAllocDmapBU(bmp
, dp
, blkno
, nb
))) {
3144 release_metapage(mp
);
3145 IREAD_UNLOCK(ipbmap
);
3150 /* write the last buffer. */
3153 IREAD_UNLOCK(ipbmap
);
3159 static int dbAllocDmapBU(struct bmap
* bmp
, struct dmap
* dp
, s64 blkno
,
3163 int dbitno
, word
, rembits
, nb
, nwords
, wbitno
, agno
;
3165 struct dmaptree
*tp
= (struct dmaptree
*) & dp
->tree
;
3167 /* save the current value of the root (i.e. maximum free string)
3170 oldroot
= tp
->stree
[ROOT
];
3172 /* pick up a pointer to the leaves of the dmap tree */
3173 leaf
= tp
->stree
+ LEAFIND
;
3175 /* determine the bit number and word within the dmap of the
3178 dbitno
= blkno
& (BPERDMAP
- 1);
3179 word
= dbitno
>> L2DBWORD
;
3181 /* block range better be within the dmap */
3182 assert(dbitno
+ nblocks
<= BPERDMAP
);
3184 /* allocate the bits of the dmap's words corresponding to the block
3185 * range. not all bits of the first and last words may be contained
3186 * within the block range. if this is the case, we'll work against
3187 * those words (i.e. partial first and/or last) on an individual basis
3188 * (a single pass), allocating the bits of interest by hand and
3189 * updating the leaf corresponding to the dmap word. a single pass
3190 * will be used for all dmap words fully contained within the
3191 * specified range. within this pass, the bits of all fully contained
3192 * dmap words will be marked as free in a single shot and the leaves
3193 * will be updated. a single leaf may describe the free space of
3194 * multiple dmap words, so we may update only a subset of the actual
3195 * leaves corresponding to the dmap words of the block range.
3197 for (rembits
= nblocks
; rembits
> 0; rembits
-= nb
, dbitno
+= nb
) {
3198 /* determine the bit number within the word and
3199 * the number of bits within the word.
3201 wbitno
= dbitno
& (DBWORD
- 1);
3202 nb
= min(rembits
, DBWORD
- wbitno
);
3204 /* check if only part of a word is to be allocated.
3207 /* allocate (set to 1) the appropriate bits within
3210 dp
->wmap
[word
] |= cpu_to_le32(ONES
<< (DBWORD
- nb
)
3215 /* one or more dmap words are fully contained
3216 * within the block range. determine how many
3217 * words and allocate (set to 1) the bits of these
3220 nwords
= rembits
>> L2DBWORD
;
3221 memset(&dp
->wmap
[word
], (int) ONES
, nwords
* 4);
3223 /* determine how many bits */
3224 nb
= nwords
<< L2DBWORD
;
3229 /* update the free count for this dmap */
3230 le32_add_cpu(&dp
->nfree
, -nblocks
);
3232 /* reconstruct summary tree */
3237 /* if this allocation group is completely free,
3238 * update the highest active allocation group number
3239 * if this allocation group is the new max.
3241 agno
= blkno
>> bmp
->db_agl2size
;
3242 if (agno
> bmp
->db_maxag
)
3243 bmp
->db_maxag
= agno
;
3245 /* update the free count for the allocation group and map */
3246 bmp
->db_agfree
[agno
] -= nblocks
;
3247 bmp
->db_nfree
-= nblocks
;
3251 /* if the root has not changed, done. */
3252 if (tp
->stree
[ROOT
] == oldroot
)
3255 /* root changed. bubble the change up to the dmap control pages.
3256 * if the adjustment of the upper level control pages fails,
3257 * backout the bit allocation (thus making everything consistent).
3259 if ((rc
= dbAdjCtl(bmp
, blkno
, tp
->stree
[ROOT
], 1, 0)))
3260 dbFreeBits(bmp
, dp
, blkno
, nblocks
);
3267 * NAME: dbExtendFS()
3269 * FUNCTION: extend bmap from blkno for nblocks;
3270 * dbExtendFS() updates bmap ready for dbAllocBottomUp();
3274 * L1---------------------------------L1
3276 * L0---------L0---------L0 L0---------L0---------L0
3278 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm;
3279 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3281 * <---old---><----------------------------extend----------------------->
3283 int dbExtendFS(struct inode
*ipbmap
, s64 blkno
, s64 nblocks
)
3285 struct jfs_sb_info
*sbi
= JFS_SBI(ipbmap
->i_sb
);
3286 int nbperpage
= sbi
->nbperpage
;
3287 int i
, i0
= true, j
, j0
= true, k
, n
;
3290 struct metapage
*mp
, *l2mp
, *l1mp
= NULL
, *l0mp
= NULL
;
3291 struct dmapctl
*l2dcp
, *l1dcp
, *l0dcp
;
3293 s8
*l0leaf
, *l1leaf
, *l2leaf
;
3294 struct bmap
*bmp
= sbi
->bmap
;
3295 int agno
, l2agsize
, oldl2agsize
;
3298 newsize
= blkno
+ nblocks
;
3300 jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld",
3301 (long long) blkno
, (long long) nblocks
, (long long) newsize
);
3304 * initialize bmap control page.
3306 * all the data in bmap control page should exclude
3307 * the mkfs hidden dmap page.
3310 /* update mapsize */
3311 bmp
->db_mapsize
= newsize
;
3312 bmp
->db_maxlevel
= BMAPSZTOLEV(bmp
->db_mapsize
);
3314 /* compute new AG size */
3315 l2agsize
= dbGetL2AGSize(newsize
);
3316 oldl2agsize
= bmp
->db_agl2size
;
3318 bmp
->db_agl2size
= l2agsize
;
3319 bmp
->db_agsize
= 1 << l2agsize
;
3321 /* compute new number of AG */
3322 agno
= bmp
->db_numag
;
3323 bmp
->db_numag
= newsize
>> l2agsize
;
3324 bmp
->db_numag
+= ((u32
) newsize
% (u32
) bmp
->db_agsize
) ? 1 : 0;
3327 * reconfigure db_agfree[]
3328 * from old AG configuration to new AG configuration;
3330 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3331 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3332 * note: new AG size = old AG size * (2**x).
3334 if (l2agsize
== oldl2agsize
)
3336 k
= 1 << (l2agsize
- oldl2agsize
);
3337 ag_rem
= bmp
->db_agfree
[0]; /* save agfree[0] */
3338 for (i
= 0, n
= 0; i
< agno
; n
++) {
3339 bmp
->db_agfree
[n
] = 0; /* init collection point */
3341 /* coalesce contiguous k AGs; */
3342 for (j
= 0; j
< k
&& i
< agno
; j
++, i
++) {
3343 /* merge AGi to AGn */
3344 bmp
->db_agfree
[n
] += bmp
->db_agfree
[i
];
3347 bmp
->db_agfree
[0] += ag_rem
; /* restore agfree[0] */
3349 for (; n
< MAXAG
; n
++)
3350 bmp
->db_agfree
[n
] = 0;
3353 * update highest active ag number
3356 bmp
->db_maxag
= bmp
->db_maxag
/ k
;
3361 * update bit maps and corresponding level control pages;
3362 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3366 p
= BMAPBLKNO
+ nbperpage
; /* L2 page */
3367 l2mp
= read_metapage(ipbmap
, p
, PSIZE
, 0);
3369 jfs_error(ipbmap
->i_sb
, "dbExtendFS: L2 page could not be read");
3372 l2dcp
= (struct dmapctl
*) l2mp
->data
;
3374 /* compute start L1 */
3375 k
= blkno
>> L2MAXL1SIZE
;
3376 l2leaf
= l2dcp
->stree
+ CTLLEAFIND
+ k
;
3377 p
= BLKTOL1(blkno
, sbi
->l2nbperpage
); /* L1 page */
3380 * extend each L1 in L2
3382 for (; k
< LPERCTL
; k
++, p
+= nbperpage
) {
3385 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3386 l1mp
= read_metapage(ipbmap
, p
, PSIZE
, 0);
3389 l1dcp
= (struct dmapctl
*) l1mp
->data
;
3391 /* compute start L0 */
3392 j
= (blkno
& (MAXL1SIZE
- 1)) >> L2MAXL0SIZE
;
3393 l1leaf
= l1dcp
->stree
+ CTLLEAFIND
+ j
;
3394 p
= BLKTOL0(blkno
, sbi
->l2nbperpage
);
3397 /* assign/init L1 page */
3398 l1mp
= get_metapage(ipbmap
, p
, PSIZE
, 0);
3402 l1dcp
= (struct dmapctl
*) l1mp
->data
;
3404 /* compute start L0 */
3406 l1leaf
= l1dcp
->stree
+ CTLLEAFIND
;
3407 p
+= nbperpage
; /* 1st L0 of L1.k */
3411 * extend each L0 in L1
3413 for (; j
< LPERCTL
; j
++) {
3416 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3418 l0mp
= read_metapage(ipbmap
, p
, PSIZE
, 0);
3421 l0dcp
= (struct dmapctl
*) l0mp
->data
;
3423 /* compute start dmap */
3424 i
= (blkno
& (MAXL0SIZE
- 1)) >>
3426 l0leaf
= l0dcp
->stree
+ CTLLEAFIND
+ i
;
3427 p
= BLKTODMAP(blkno
,
3431 /* assign/init L0 page */
3432 l0mp
= get_metapage(ipbmap
, p
, PSIZE
, 0);
3436 l0dcp
= (struct dmapctl
*) l0mp
->data
;
3438 /* compute start dmap */
3440 l0leaf
= l0dcp
->stree
+ CTLLEAFIND
;
3441 p
+= nbperpage
; /* 1st dmap of L0.j */
3445 * extend each dmap in L0
3447 for (; i
< LPERCTL
; i
++) {
3449 * reconstruct the dmap page, and
3450 * initialize corresponding parent L0 leaf
3452 if ((n
= blkno
& (BPERDMAP
- 1))) {
3453 /* read in dmap page: */
3454 mp
= read_metapage(ipbmap
, p
,
3458 n
= min(nblocks
, (s64
)BPERDMAP
- n
);
3460 /* assign/init dmap page */
3461 mp
= read_metapage(ipbmap
, p
,
3466 n
= min(nblocks
, (s64
)BPERDMAP
);
3469 dp
= (struct dmap
*) mp
->data
;
3470 *l0leaf
= dbInitDmap(dp
, blkno
, n
);
3473 agno
= le64_to_cpu(dp
->start
) >> l2agsize
;
3474 bmp
->db_agfree
[agno
] += n
;
3485 } /* for each dmap in a L0 */
3488 * build current L0 page from its leaves, and
3489 * initialize corresponding parent L1 leaf
3491 *l1leaf
= dbInitDmapCtl(l0dcp
, 0, ++i
);
3492 write_metapage(l0mp
);
3496 l1leaf
++; /* continue for next L0 */
3498 /* more than 1 L0 ? */
3500 break; /* build L1 page */
3502 /* summarize in global bmap page */
3503 bmp
->db_maxfreebud
= *l1leaf
;
3504 release_metapage(l1mp
);
3505 release_metapage(l2mp
);
3509 } /* for each L0 in a L1 */
3512 * build current L1 page from its leaves, and
3513 * initialize corresponding parent L2 leaf
3515 *l2leaf
= dbInitDmapCtl(l1dcp
, 1, ++j
);
3516 write_metapage(l1mp
);
3520 l2leaf
++; /* continue for next L1 */
3522 /* more than 1 L1 ? */
3524 break; /* build L2 page */
3526 /* summarize in global bmap page */
3527 bmp
->db_maxfreebud
= *l2leaf
;
3528 release_metapage(l2mp
);
3532 } /* for each L1 in a L2 */
3534 jfs_error(ipbmap
->i_sb
,
3535 "dbExtendFS: function has not returned as expected");
3538 release_metapage(l0mp
);
3540 release_metapage(l1mp
);
3541 release_metapage(l2mp
);
3545 * finalize bmap control page
3556 void dbFinalizeBmap(struct inode
*ipbmap
)
3558 struct bmap
*bmp
= JFS_SBI(ipbmap
->i_sb
)->bmap
;
3559 int actags
, inactags
, l2nl
;
3560 s64 ag_rem
, actfree
, inactfree
, avgfree
;
3564 * finalize bmap control page
3568 * compute db_agpref: preferred ag to allocate from
3569 * (the leftmost ag with average free space in it);
3572 /* get the number of active ags and inacitve ags */
3573 actags
= bmp
->db_maxag
+ 1;
3574 inactags
= bmp
->db_numag
- actags
;
3575 ag_rem
= bmp
->db_mapsize
& (bmp
->db_agsize
- 1); /* ??? */
3577 /* determine how many blocks are in the inactive allocation
3578 * groups. in doing this, we must account for the fact that
3579 * the rightmost group might be a partial group (i.e. file
3580 * system size is not a multiple of the group size).
3582 inactfree
= (inactags
&& ag_rem
) ?
3583 ((inactags
- 1) << bmp
->db_agl2size
) + ag_rem
3584 : inactags
<< bmp
->db_agl2size
;
3586 /* determine how many free blocks are in the active
3587 * allocation groups plus the average number of free blocks
3588 * within the active ags.
3590 actfree
= bmp
->db_nfree
- inactfree
;
3591 avgfree
= (u32
) actfree
/ (u32
) actags
;
3593 /* if the preferred allocation group has not average free space.
3594 * re-establish the preferred group as the leftmost
3595 * group with average free space.
3597 if (bmp
->db_agfree
[bmp
->db_agpref
] < avgfree
) {
3598 for (bmp
->db_agpref
= 0; bmp
->db_agpref
< actags
;
3600 if (bmp
->db_agfree
[bmp
->db_agpref
] >= avgfree
)
3603 if (bmp
->db_agpref
>= bmp
->db_numag
) {
3604 jfs_error(ipbmap
->i_sb
,
3605 "cannot find ag with average freespace");
3610 * compute db_aglevel, db_agheight, db_width, db_agstart:
3611 * an ag is covered in aglevel dmapctl summary tree,
3612 * at agheight level height (from leaf) with agwidth number of nodes
3613 * each, which starts at agstart index node of the smmary tree node
3616 bmp
->db_aglevel
= BMAPSZTOLEV(bmp
->db_agsize
);
3618 bmp
->db_agl2size
- (L2BPERDMAP
+ bmp
->db_aglevel
* L2LPERCTL
);
3619 bmp
->db_agheight
= l2nl
>> 1;
3620 bmp
->db_agwidth
= 1 << (l2nl
- (bmp
->db_agheight
<< 1));
3621 for (i
= 5 - bmp
->db_agheight
, bmp
->db_agstart
= 0, n
= 1; i
> 0;
3623 bmp
->db_agstart
+= n
;
3631 * NAME: dbInitDmap()/ujfs_idmap_page()
3633 * FUNCTION: initialize working/persistent bitmap of the dmap page
3634 * for the specified number of blocks:
3636 * at entry, the bitmaps had been initialized as free (ZEROS);
3637 * The number of blocks will only account for the actually
3638 * existing blocks. Blocks which don't actually exist in
3639 * the aggregate will be marked as allocated (ONES);
3642 * dp - pointer to page of map
3643 * nblocks - number of blocks this page
3647 static int dbInitDmap(struct dmap
* dp
, s64 Blkno
, int nblocks
)
3649 int blkno
, w
, b
, r
, nw
, nb
, i
;
3651 /* starting block number within the dmap */
3652 blkno
= Blkno
& (BPERDMAP
- 1);
3655 dp
->nblocks
= dp
->nfree
= cpu_to_le32(nblocks
);
3656 dp
->start
= cpu_to_le64(Blkno
);
3658 if (nblocks
== BPERDMAP
) {
3659 memset(&dp
->wmap
[0], 0, LPERDMAP
* 4);
3660 memset(&dp
->pmap
[0], 0, LPERDMAP
* 4);
3664 le32_add_cpu(&dp
->nblocks
, nblocks
);
3665 le32_add_cpu(&dp
->nfree
, nblocks
);
3668 /* word number containing start block number */
3669 w
= blkno
>> L2DBWORD
;
3672 * free the bits corresponding to the block range (ZEROS):
3673 * note: not all bits of the first and last words may be contained
3674 * within the block range.
3676 for (r
= nblocks
; r
> 0; r
-= nb
, blkno
+= nb
) {
3677 /* number of bits preceding range to be freed in the word */
3678 b
= blkno
& (DBWORD
- 1);
3679 /* number of bits to free in the word */
3680 nb
= min(r
, DBWORD
- b
);
3682 /* is partial word to be freed ? */
3684 /* free (set to 0) from the bitmap word */
3685 dp
->wmap
[w
] &= cpu_to_le32(~(ONES
<< (DBWORD
- nb
)
3687 dp
->pmap
[w
] &= cpu_to_le32(~(ONES
<< (DBWORD
- nb
)
3690 /* skip the word freed */
3693 /* free (set to 0) contiguous bitmap words */
3695 memset(&dp
->wmap
[w
], 0, nw
* 4);
3696 memset(&dp
->pmap
[w
], 0, nw
* 4);
3698 /* skip the words freed */
3699 nb
= nw
<< L2DBWORD
;
3705 * mark bits following the range to be freed (non-existing
3706 * blocks) as allocated (ONES)
3709 if (blkno
== BPERDMAP
)
3712 /* the first word beyond the end of existing blocks */
3713 w
= blkno
>> L2DBWORD
;
3715 /* does nblocks fall on a 32-bit boundary ? */
3716 b
= blkno
& (DBWORD
- 1);
3718 /* mark a partial word allocated */
3719 dp
->wmap
[w
] = dp
->pmap
[w
] = cpu_to_le32(ONES
>> b
);
3723 /* set the rest of the words in the page to allocated (ONES) */
3724 for (i
= w
; i
< LPERDMAP
; i
++)
3725 dp
->pmap
[i
] = dp
->wmap
[i
] = cpu_to_le32(ONES
);
3731 return (dbInitDmapTree(dp
));
3736 * NAME: dbInitDmapTree()/ujfs_complete_dmap()
3738 * FUNCTION: initialize summary tree of the specified dmap:
3740 * at entry, bitmap of the dmap has been initialized;
3743 * dp - dmap to complete
3744 * blkno - starting block number for this dmap
3745 * treemax - will be filled in with max free for this dmap
3747 * RETURNS: max free string at the root of the tree
3749 static int dbInitDmapTree(struct dmap
* dp
)
3751 struct dmaptree
*tp
;
3755 /* init fixed info of tree */
3757 tp
->nleafs
= cpu_to_le32(LPERDMAP
);
3758 tp
->l2nleafs
= cpu_to_le32(L2LPERDMAP
);
3759 tp
->leafidx
= cpu_to_le32(LEAFIND
);
3760 tp
->height
= cpu_to_le32(4);
3761 tp
->budmin
= BUDMIN
;
3763 /* init each leaf from corresponding wmap word:
3764 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3765 * bitmap word are allocated.
3767 cp
= tp
->stree
+ le32_to_cpu(tp
->leafidx
);
3768 for (i
= 0; i
< LPERDMAP
; i
++)
3769 *cp
++ = dbMaxBud((u8
*) & dp
->wmap
[i
]);
3771 /* build the dmap's binary buddy summary tree */
3772 return (dbInitTree(tp
));
3777 * NAME: dbInitTree()/ujfs_adjtree()
3779 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl.
3781 * at entry, the leaves of the tree has been initialized
3782 * from corresponding bitmap word or root of summary tree
3783 * of the child control page;
3784 * configure binary buddy system at the leaf level, then
3785 * bubble up the values of the leaf nodes up the tree.
3788 * cp - Pointer to the root of the tree
3789 * l2leaves- Number of leaf nodes as a power of 2
3790 * l2min - Number of blocks that can be covered by a leaf
3793 * RETURNS: max free string at the root of the tree
3795 static int dbInitTree(struct dmaptree
* dtp
)
3797 int l2max
, l2free
, bsize
, nextb
, i
;
3798 int child
, parent
, nparent
;
3803 /* Determine the maximum free string possible for the leaves */
3804 l2max
= le32_to_cpu(dtp
->l2nleafs
) + dtp
->budmin
;
3807 * configure the leaf levevl into binary buddy system
3809 * Try to combine buddies starting with a buddy size of 1
3810 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3811 * can be combined if both buddies have a maximum free of l2min;
3812 * the combination will result in the left-most buddy leaf having
3813 * a maximum free of l2min+1.
3814 * After processing all buddies for a given size, process buddies
3815 * at the next higher buddy size (i.e. current size * 2) and
3816 * the next maximum free (current free + 1).
3817 * This continues until the maximum possible buddy combination
3818 * yields maximum free.
3820 for (l2free
= dtp
->budmin
, bsize
= 1; l2free
< l2max
;
3821 l2free
++, bsize
= nextb
) {
3822 /* get next buddy size == current buddy pair size */
3825 /* scan each adjacent buddy pair at current buddy size */
3826 for (i
= 0, cp
= tp
+ le32_to_cpu(dtp
->leafidx
);
3827 i
< le32_to_cpu(dtp
->nleafs
);
3828 i
+= nextb
, cp
+= nextb
) {
3829 /* coalesce if both adjacent buddies are max free */
3830 if (*cp
== l2free
&& *(cp
+ bsize
) == l2free
) {
3831 *cp
= l2free
+ 1; /* left take right */
3832 *(cp
+ bsize
) = -1; /* right give left */
3838 * bubble summary information of leaves up the tree.
3840 * Starting at the leaf node level, the four nodes described by
3841 * the higher level parent node are compared for a maximum free and
3842 * this maximum becomes the value of the parent node.
3843 * when all lower level nodes are processed in this fashion then
3844 * move up to the next level (parent becomes a lower level node) and
3845 * continue the process for that level.
3847 for (child
= le32_to_cpu(dtp
->leafidx
),
3848 nparent
= le32_to_cpu(dtp
->nleafs
) >> 2;
3849 nparent
> 0; nparent
>>= 2, child
= parent
) {
3850 /* get index of 1st node of parent level */
3851 parent
= (child
- 1) >> 2;
3853 /* set the value of the parent node as the maximum
3854 * of the four nodes of the current level.
3856 for (i
= 0, cp
= tp
+ child
, cp1
= tp
+ parent
;
3857 i
< nparent
; i
++, cp
+= 4, cp1
++)
3868 * function: initialize dmapctl page
3870 static int dbInitDmapCtl(struct dmapctl
* dcp
, int level
, int i
)
3871 { /* start leaf index not covered by range */
3874 dcp
->nleafs
= cpu_to_le32(LPERCTL
);
3875 dcp
->l2nleafs
= cpu_to_le32(L2LPERCTL
);
3876 dcp
->leafidx
= cpu_to_le32(CTLLEAFIND
);
3877 dcp
->height
= cpu_to_le32(5);
3878 dcp
->budmin
= L2BPERDMAP
+ L2LPERCTL
* level
;
3881 * initialize the leaves of current level that were not covered
3882 * by the specified input block range (i.e. the leaves have no
3883 * low level dmapctl or dmap).
3885 cp
= &dcp
->stree
[CTLLEAFIND
+ i
];
3886 for (; i
< LPERCTL
; i
++)
3889 /* build the dmap's binary buddy summary tree */
3890 return (dbInitTree((struct dmaptree
*) dcp
));
3895 * NAME: dbGetL2AGSize()/ujfs_getagl2size()
3897 * FUNCTION: Determine log2(allocation group size) from aggregate size
3900 * nblocks - Number of blocks in aggregate
3902 * RETURNS: log2(allocation group size) in aggregate blocks
3904 static int dbGetL2AGSize(s64 nblocks
)
3910 if (nblocks
< BPERDMAP
* MAXAG
)
3911 return (L2BPERDMAP
);
3913 /* round up aggregate size to power of 2 */
3914 m
= ((u64
) 1 << (64 - 1));
3915 for (l2sz
= 64; l2sz
>= 0; l2sz
--, m
>>= 1) {
3920 sz
= (s64
) 1 << l2sz
;
3924 /* agsize = roundupSize/max_number_of_ag */
3925 return (l2sz
- L2MAXAG
);
3930 * NAME: dbMapFileSizeToMapSize()
3932 * FUNCTION: compute number of blocks the block allocation map file
3933 * can cover from the map file size;
3935 * RETURNS: Number of blocks which can be covered by this block map file;
3939 * maximum number of map pages at each level including control pages
3941 #define MAXL0PAGES (1 + LPERCTL)
3942 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES)
3943 #define MAXL2PAGES (1 + LPERCTL * MAXL1PAGES)
3946 * convert number of map pages to the zero origin top dmapctl level
3948 #define BMAPPGTOLEV(npages) \
3949 (((npages) <= 3 + MAXL0PAGES) ? 0 : \
3950 ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
3952 s64
dbMapFileSizeToMapSize(struct inode
* ipbmap
)
3954 struct super_block
*sb
= ipbmap
->i_sb
;
3958 int complete
, factor
;
3960 nblocks
= ipbmap
->i_size
>> JFS_SBI(sb
)->l2bsize
;
3961 npages
= nblocks
>> JFS_SBI(sb
)->l2nbperpage
;
3962 level
= BMAPPGTOLEV(npages
);
3964 /* At each level, accumulate the number of dmap pages covered by
3965 * the number of full child levels below it;
3966 * repeat for the last incomplete child level.
3969 npages
--; /* skip the first global control page */
3970 /* skip higher level control pages above top level covered by map */
3971 npages
-= (2 - level
);
3972 npages
--; /* skip top level's control page */
3973 for (i
= level
; i
>= 0; i
--) {
3975 (i
== 2) ? MAXL1PAGES
: ((i
== 1) ? MAXL0PAGES
: 1);
3976 complete
= (u32
) npages
/ factor
;
3977 ndmaps
+= complete
* ((i
== 2) ? LPERCTL
* LPERCTL
:
3978 ((i
== 1) ? LPERCTL
: 1));
3980 /* pages in last/incomplete child */
3981 npages
= (u32
) npages
% factor
;
3982 /* skip incomplete child's level control page */
3986 /* convert the number of dmaps into the number of blocks
3987 * which can be covered by the dmaps;
3989 nblocks
= ndmaps
<< L2BPERDMAP
;