2 * Copyright (C) International Business Machines Corp., 2000-2004
3 * Portions Copyright (C) Tino Reichardt, 2012
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 #include <linux/slab.h>
22 #include "jfs_incore.h"
23 #include "jfs_superblock.h"
27 #include "jfs_metapage.h"
28 #include "jfs_debug.h"
29 #include "jfs_discard.h"
32 * SERIALIZATION of the Block Allocation Map.
34 * the working state of the block allocation map is accessed in
37 * 1) allocation and free requests that start at the dmap
38 * level and move up through the dmap control pages (i.e.
39 * the vast majority of requests).
41 * 2) allocation requests that start at dmap control page
42 * level and work down towards the dmaps.
44 * the serialization scheme used here is as follows.
46 * requests which start at the bottom are serialized against each
47 * other through buffers and each requests holds onto its buffers
48 * as it works it way up from a single dmap to the required level
49 * of dmap control page.
50 * requests that start at the top are serialized against each other
51 * and request that start from the bottom by the multiple read/single
52 * write inode lock of the bmap inode. requests starting at the top
53 * take this lock in write mode while request starting at the bottom
54 * take the lock in read mode. a single top-down request may proceed
55 * exclusively while multiple bottoms-up requests may proceed
56 * simultaneously (under the protection of busy buffers).
58 * in addition to information found in dmaps and dmap control pages,
59 * the working state of the block allocation map also includes read/
60 * write information maintained in the bmap descriptor (i.e. total
61 * free block count, allocation group level free block counts).
62 * a single exclusive lock (BMAP_LOCK) is used to guard this information
63 * in the face of multiple-bottoms up requests.
64 * (lock ordering: IREAD_LOCK, BMAP_LOCK);
66 * accesses to the persistent state of the block allocation map (limited
67 * to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
70 #define BMAP_LOCK_INIT(bmp) mutex_init(&bmp->db_bmaplock)
71 #define BMAP_LOCK(bmp) mutex_lock(&bmp->db_bmaplock)
72 #define BMAP_UNLOCK(bmp) mutex_unlock(&bmp->db_bmaplock)
77 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
79 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval);
80 static int dbBackSplit(dmtree_t * tp, int leafno);
81 static int dbJoin(dmtree_t * tp, int leafno, int newval);
82 static void dbAdjTree(dmtree_t * tp, int leafno, int newval);
83 static int dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc,
85 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results);
86 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
88 static int dbAllocNear(struct bmap * bmp, struct dmap * dp, s64 blkno,
90 int l2nb, s64 * results);
91 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
93 static int dbAllocDmapLev(struct bmap * bmp, struct dmap * dp, int nblocks,
96 static int dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb,
98 static int dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno,
100 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks);
101 static int dbFindBits(u32 word, int l2nb);
102 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno);
103 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx);
104 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
106 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
108 static int dbMaxBud(u8 * cp);
109 static int blkstol2(s64 nb);
111 static int cntlz(u32 value);
112 static int cnttz(u32 word);
114 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
116 static int dbInitDmap(struct dmap * dp, s64 blkno, int nblocks);
117 static int dbInitDmapTree(struct dmap * dp);
118 static int dbInitTree(struct dmaptree * dtp);
119 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i);
120 static int dbGetL2AGSize(s64 nblocks);
125 * table used for determining buddy sizes within characters of
126 * dmap bitmap words. the characters themselves serve as indexes
127 * into the table, with the table elements yielding the maximum
128 * binary buddy of free bits within the character.
130 static const s8 budtab[256] = {
131 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
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, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
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, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
139 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
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, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
143 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
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, 0,
146 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
164 * -EINVAL - wrong bmap data
166 int dbMount(struct inode *ipbmap)
169 struct dbmap_disk *dbmp_le;
174 * allocate/initialize the in-memory bmap descriptor
176 /* allocate memory for the in-memory bmap descriptor */
177 bmp = kmalloc(sizeof(struct bmap), GFP_KERNEL);
181 /* read the on-disk bmap descriptor. */
182 mp = read_metapage(ipbmap,
183 BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
190 /* copy the on-disk bmap descriptor to its in-memory version. */
191 dbmp_le = (struct dbmap_disk *) mp->data;
192 bmp->db_mapsize = le64_to_cpu(dbmp_le->dn_mapsize);
193 bmp->db_nfree = le64_to_cpu(dbmp_le->dn_nfree);
194 bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage);
195 bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag);
196 if (!bmp->db_numag) {
197 release_metapage(mp);
202 bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel);
203 bmp->db_maxag = le32_to_cpu(dbmp_le->dn_maxag);
204 bmp->db_agpref = le32_to_cpu(dbmp_le->dn_agpref);
205 bmp->db_aglevel = le32_to_cpu(dbmp_le->dn_aglevel);
206 bmp->db_agheight = le32_to_cpu(dbmp_le->dn_agheight);
207 bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth);
208 bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart);
209 bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size);
210 for (i = 0; i < MAXAG; i++)
211 bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
212 bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
213 bmp->db_maxfreebud = dbmp_le->dn_maxfreebud;
215 /* release the buffer. */
216 release_metapage(mp);
218 /* bind the bmap inode and the bmap descriptor to each other. */
219 bmp->db_ipbmap = ipbmap;
220 JFS_SBI(ipbmap->i_sb)->bmap = bmp;
222 memset(bmp->db_active, 0, sizeof(bmp->db_active));
225 * allocate/initialize the bmap lock
236 * FUNCTION: terminate the block allocation map in preparation for
237 * file system unmount.
239 * the in-core bmap descriptor is written to disk and
240 * the memory for this descriptor is freed.
243 * ipbmap - pointer to in-core inode for the block map.
249 int dbUnmount(struct inode *ipbmap, int mounterror)
251 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
253 if (!(mounterror || isReadOnly(ipbmap)))
257 * Invalidate the page cache buffers
259 truncate_inode_pages(ipbmap->i_mapping, 0);
261 /* free the memory for the in-memory bmap. */
270 int dbSync(struct inode *ipbmap)
272 struct dbmap_disk *dbmp_le;
273 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
278 * write bmap global control page
280 /* get the buffer for the on-disk bmap descriptor. */
281 mp = read_metapage(ipbmap,
282 BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
285 jfs_err("dbSync: read_metapage failed!");
288 /* copy the in-memory version of the bmap to the on-disk version */
289 dbmp_le = (struct dbmap_disk *) mp->data;
290 dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize);
291 dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree);
292 dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage);
293 dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag);
294 dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel);
295 dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag);
296 dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref);
297 dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel);
298 dbmp_le->dn_agheight = cpu_to_le32(bmp->db_agheight);
299 dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth);
300 dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart);
301 dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size);
302 for (i = 0; i < MAXAG; i++)
303 dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]);
304 dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize);
305 dbmp_le->dn_maxfreebud = bmp->db_maxfreebud;
307 /* write the buffer */
311 * write out dirty pages of bmap
313 filemap_write_and_wait(ipbmap->i_mapping);
315 diWriteSpecial(ipbmap, 0);
323 * FUNCTION: free the specified block range from the working block
326 * the blocks will be free from the working map one dmap
330 * ip - pointer to in-core inode;
331 * blkno - starting block number to be freed.
332 * nblocks - number of blocks to be freed.
338 int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
344 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
345 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
346 struct super_block *sb = ipbmap->i_sb;
348 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
350 /* block to be freed better be within the mapsize. */
351 if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
352 IREAD_UNLOCK(ipbmap);
353 printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
354 (unsigned long long) blkno,
355 (unsigned long long) nblocks);
356 jfs_error(ip->i_sb, "block to be freed is outside the map\n");
361 * TRIM the blocks, when mounted with discard option
363 if (JFS_SBI(sb)->flag & JFS_DISCARD)
364 if (JFS_SBI(sb)->minblks_trim <= nblocks)
365 jfs_issue_discard(ipbmap, blkno, nblocks);
368 * free the blocks a dmap at a time.
371 for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
372 /* release previous dmap if any */
377 /* get the buffer for the current dmap. */
378 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
379 mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
381 IREAD_UNLOCK(ipbmap);
384 dp = (struct dmap *) mp->data;
386 /* determine the number of blocks to be freed from
389 nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
391 /* free the blocks. */
392 if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
393 jfs_error(ip->i_sb, "error in block map\n");
394 release_metapage(mp);
395 IREAD_UNLOCK(ipbmap);
400 /* write the last buffer. */
404 IREAD_UNLOCK(ipbmap);
411 * NAME: dbUpdatePMap()
413 * FUNCTION: update the allocation state (free or allocate) of the
414 * specified block range in the persistent block allocation map.
416 * the blocks will be updated in the persistent map one
420 * ipbmap - pointer to in-core inode for the block map.
421 * free - 'true' if block range is to be freed from the persistent
422 * map; 'false' if it is to be allocated.
423 * blkno - starting block number of the range.
424 * nblocks - number of contiguous blocks in the range.
425 * tblk - transaction block;
432 dbUpdatePMap(struct inode *ipbmap,
433 int free, s64 blkno, s64 nblocks, struct tblock * tblk)
435 int nblks, dbitno, wbitno, rbits;
436 int word, nbits, nwords;
437 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
438 s64 lblkno, rem, lastlblkno;
443 int lsn, difft, diffp;
446 /* the blocks better be within the mapsize. */
447 if (blkno + nblocks > bmp->db_mapsize) {
448 printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
449 (unsigned long long) blkno,
450 (unsigned long long) nblocks);
451 jfs_error(ipbmap->i_sb, "blocks are outside the map\n");
455 /* compute delta of transaction lsn from log syncpt */
457 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
458 logdiff(difft, lsn, log);
461 * update the block state a dmap at a time.
465 for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) {
466 /* get the buffer for the current dmap. */
467 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
468 if (lblkno != lastlblkno) {
473 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE,
477 metapage_wait_for_io(mp);
479 dp = (struct dmap *) mp->data;
481 /* determine the bit number and word within the dmap of
482 * the starting block. also determine how many blocks
483 * are to be updated within this dmap.
485 dbitno = blkno & (BPERDMAP - 1);
486 word = dbitno >> L2DBWORD;
487 nblks = min(rem, (s64)BPERDMAP - dbitno);
489 /* update the bits of the dmap words. the first and last
490 * words may only have a subset of their bits updated. if
491 * this is the case, we'll work against that word (i.e.
492 * partial first and/or last) only in a single pass. a
493 * single pass will also be used to update all words that
494 * are to have all their bits updated.
496 for (rbits = nblks; rbits > 0;
497 rbits -= nbits, dbitno += nbits) {
498 /* determine the bit number within the word and
499 * the number of bits within the word.
501 wbitno = dbitno & (DBWORD - 1);
502 nbits = min(rbits, DBWORD - wbitno);
504 /* check if only part of the word is to be updated. */
505 if (nbits < DBWORD) {
506 /* update (free or allocate) the bits
510 (ONES << (DBWORD - nbits) >> wbitno);
520 /* one or more words are to have all
521 * their bits updated. determine how
522 * many words and how many bits.
524 nwords = rbits >> L2DBWORD;
525 nbits = nwords << L2DBWORD;
527 /* update (free or allocate) the bits
531 memset(&dp->pmap[word], 0,
534 memset(&dp->pmap[word], (int) ONES,
544 if (lblkno == lastlblkno)
549 LOGSYNC_LOCK(log, flags);
551 /* inherit older/smaller lsn */
552 logdiff(diffp, mp->lsn, log);
556 /* move bp after tblock in logsync list */
557 list_move(&mp->synclist, &tblk->synclist);
560 /* inherit younger/larger clsn */
561 logdiff(difft, tblk->clsn, log);
562 logdiff(diffp, mp->clsn, log);
564 mp->clsn = tblk->clsn;
569 /* insert bp after tblock in logsync list */
571 list_add(&mp->synclist, &tblk->synclist);
573 mp->clsn = tblk->clsn;
575 LOGSYNC_UNLOCK(log, flags);
578 /* write the last buffer. */
590 * FUNCTION: find the preferred allocation group for new allocations.
592 * Within the allocation groups, we maintain a preferred
593 * allocation group which consists of a group with at least
594 * average free space. It is the preferred group that we target
595 * new inode allocation towards. The tie-in between inode
596 * allocation and block allocation occurs as we allocate the
597 * first (data) block of an inode and specify the inode (block)
598 * as the allocation hint for this block.
600 * We try to avoid having more than one open file growing in
601 * an allocation group, as this will lead to fragmentation.
602 * This differs from the old OS/2 method of trying to keep
603 * empty ags around for large allocations.
606 * ipbmap - pointer to in-core inode for the block map.
609 * the preferred allocation group number.
611 int dbNextAG(struct inode *ipbmap)
618 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
622 /* determine the average number of free blocks within the ags. */
623 avgfree = (u32)bmp->db_nfree / bmp->db_numag;
626 * if the current preferred ag does not have an active allocator
627 * and has at least average freespace, return it
629 agpref = bmp->db_agpref;
630 if ((atomic_read(&bmp->db_active[agpref]) == 0) &&
631 (bmp->db_agfree[agpref] >= avgfree))
634 /* From the last preferred ag, find the next one with at least
635 * average free space.
637 for (i = 0 ; i < bmp->db_numag; i++, agpref++) {
638 if (agpref == bmp->db_numag)
641 if (atomic_read(&bmp->db_active[agpref]))
642 /* open file is currently growing in this ag */
644 if (bmp->db_agfree[agpref] >= avgfree) {
645 /* Return this one */
646 bmp->db_agpref = agpref;
648 } else if (bmp->db_agfree[agpref] > hwm) {
649 /* Less than avg. freespace, but best so far */
650 hwm = bmp->db_agfree[agpref];
656 * If no inactive ag was found with average freespace, use the
660 bmp->db_agpref = next_best;
661 /* else leave db_agpref unchanged */
665 /* return the preferred group.
667 return (bmp->db_agpref);
673 * FUNCTION: attempt to allocate a specified number of contiguous free
674 * blocks from the working allocation block map.
676 * the block allocation policy uses hints and a multi-step
679 * for allocation requests smaller than the number of blocks
680 * per dmap, we first try to allocate the new blocks
681 * immediately following the hint. if these blocks are not
682 * available, we try to allocate blocks near the hint. if
683 * no blocks near the hint are available, we next try to
684 * allocate within the same dmap as contains the hint.
686 * if no blocks are available in the dmap or the allocation
687 * request is larger than the dmap size, we try to allocate
688 * within the same allocation group as contains the hint. if
689 * this does not succeed, we finally try to allocate anywhere
690 * within the aggregate.
692 * we also try to allocate anywhere within the aggregate for
693 * for allocation requests larger than the allocation group
694 * size or requests that specify no hint value.
697 * ip - pointer to in-core inode;
698 * hint - allocation hint.
699 * nblocks - number of contiguous blocks in the range.
700 * results - on successful return, set to the starting block number
701 * of the newly allocated contiguous range.
705 * -ENOSPC - insufficient disk resources
708 int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results)
711 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
720 /* assert that nblocks is valid */
723 /* get the log2 number of blocks to be allocated.
724 * if the number of blocks is not a log2 multiple,
725 * it will be rounded up to the next log2 multiple.
727 l2nb = BLKSTOL2(nblocks);
729 bmp = JFS_SBI(ip->i_sb)->bmap;
731 mapSize = bmp->db_mapsize;
733 /* the hint should be within the map */
734 if (hint >= mapSize) {
735 jfs_error(ip->i_sb, "the hint is outside the map\n");
739 /* if the number of blocks to be allocated is greater than the
740 * allocation group size, try to allocate anywhere.
742 if (l2nb > bmp->db_agl2size) {
743 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
745 rc = dbAllocAny(bmp, nblocks, l2nb, results);
751 * If no hint, let dbNextAG recommend an allocation group
756 /* we would like to allocate close to the hint. adjust the
757 * hint to the block following the hint since the allocators
758 * will start looking for free space starting at this point.
762 if (blkno >= bmp->db_mapsize)
765 agno = blkno >> bmp->db_agl2size;
767 /* check if blkno crosses over into a new allocation group.
768 * if so, check if we should allow allocations within this
771 if ((blkno & (bmp->db_agsize - 1)) == 0)
772 /* check if the AG is currently being written to.
773 * if so, call dbNextAG() to find a non-busy
774 * AG with sufficient free space.
776 if (atomic_read(&bmp->db_active[agno]))
779 /* check if the allocation request size can be satisfied from a
780 * single dmap. if so, try to allocate from the dmap containing
781 * the hint using a tiered strategy.
783 if (nblocks <= BPERDMAP) {
784 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
786 /* get the buffer for the dmap containing the hint.
789 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
790 mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
794 dp = (struct dmap *) mp->data;
796 /* first, try to satisfy the allocation request with the
797 * blocks beginning at the hint.
799 if ((rc = dbAllocNext(bmp, dp, blkno, (int) nblocks))
803 mark_metapage_dirty(mp);
806 release_metapage(mp);
810 writers = atomic_read(&bmp->db_active[agno]);
812 ((writers == 1) && (JFS_IP(ip)->active_ag != agno))) {
814 * Someone else is writing in this allocation
815 * group. To avoid fragmenting, try another ag
817 release_metapage(mp);
818 IREAD_UNLOCK(ipbmap);
822 /* next, try to satisfy the allocation request with blocks
826 dbAllocNear(bmp, dp, blkno, (int) nblocks, l2nb, results))
829 mark_metapage_dirty(mp);
831 release_metapage(mp);
835 /* try to satisfy the allocation request with blocks within
836 * the same dmap as the hint.
838 if ((rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results))
841 mark_metapage_dirty(mp);
843 release_metapage(mp);
847 release_metapage(mp);
848 IREAD_UNLOCK(ipbmap);
851 /* try to satisfy the allocation request with blocks within
852 * the same allocation group as the hint.
854 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
855 if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) != -ENOSPC)
858 IWRITE_UNLOCK(ipbmap);
863 * Let dbNextAG recommend a preferred allocation group
865 agno = dbNextAG(ipbmap);
866 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
868 /* Try to allocate within this allocation group. if that fails, try to
869 * allocate anywhere in the map.
871 if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) == -ENOSPC)
872 rc = dbAllocAny(bmp, nblocks, l2nb, results);
875 IWRITE_UNLOCK(ipbmap);
880 IREAD_UNLOCK(ipbmap);
887 * NAME: dbAllocExact()
889 * FUNCTION: try to allocate the requested extent;
892 * ip - pointer to in-core inode;
893 * blkno - extent address;
894 * nblocks - extent length;
898 * -ENOSPC - insufficient disk resources
901 int dbAllocExact(struct inode *ip, s64 blkno, int nblocks)
904 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
905 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
910 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
913 * validate extent request:
915 * note: defragfs policy:
916 * max 64 blocks will be moved.
917 * allocation request size must be satisfied from a single dmap.
919 if (nblocks <= 0 || nblocks > BPERDMAP || blkno >= bmp->db_mapsize) {
920 IREAD_UNLOCK(ipbmap);
924 if (nblocks > ((s64) 1 << bmp->db_maxfreebud)) {
925 /* the free space is no longer available */
926 IREAD_UNLOCK(ipbmap);
930 /* read in the dmap covering the extent */
931 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
932 mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
934 IREAD_UNLOCK(ipbmap);
937 dp = (struct dmap *) mp->data;
939 /* try to allocate the requested extent */
940 rc = dbAllocNext(bmp, dp, blkno, nblocks);
942 IREAD_UNLOCK(ipbmap);
945 mark_metapage_dirty(mp);
947 release_metapage(mp);
956 * FUNCTION: attempt to extend a current allocation by a specified
959 * this routine attempts to satisfy the allocation request
960 * by first trying to extend the existing allocation in
961 * place by allocating the additional blocks as the blocks
962 * immediately following the current allocation. if these
963 * blocks are not available, this routine will attempt to
964 * allocate a new set of contiguous blocks large enough
965 * to cover the existing allocation plus the additional
966 * number of blocks required.
969 * ip - pointer to in-core inode requiring allocation.
970 * blkno - starting block of the current allocation.
971 * nblocks - number of contiguous blocks within the current
973 * addnblocks - number of blocks to add to the allocation.
974 * results - on successful return, set to the starting block number
975 * of the existing allocation if the existing allocation
976 * was extended in place or to a newly allocated contiguous
977 * range if the existing allocation could not be extended
982 * -ENOSPC - insufficient disk resources
986 dbReAlloc(struct inode *ip,
987 s64 blkno, s64 nblocks, s64 addnblocks, s64 * results)
991 /* try to extend the allocation in place.
993 if ((rc = dbExtend(ip, blkno, nblocks, addnblocks)) == 0) {
1001 /* could not extend the allocation in place, so allocate a
1002 * new set of blocks for the entire request (i.e. try to get
1003 * a range of contiguous blocks large enough to cover the
1004 * existing allocation plus the additional blocks.)
1007 (ip, blkno + nblocks - 1, addnblocks + nblocks, results));
1014 * FUNCTION: attempt to extend a current allocation by a specified
1017 * this routine attempts to satisfy the allocation request
1018 * by first trying to extend the existing allocation in
1019 * place by allocating the additional blocks as the blocks
1020 * immediately following the current allocation.
1023 * ip - pointer to in-core inode requiring allocation.
1024 * blkno - starting block of the current allocation.
1025 * nblocks - number of contiguous blocks within the current
1027 * addnblocks - number of blocks to add to the allocation.
1031 * -ENOSPC - insufficient disk resources
1034 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks)
1036 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
1037 s64 lblkno, lastblkno, extblkno;
1039 struct metapage *mp;
1042 struct inode *ipbmap = sbi->ipbmap;
1046 * We don't want a non-aligned extent to cross a page boundary
1048 if (((rel_block = blkno & (sbi->nbperpage - 1))) &&
1049 (rel_block + nblocks + addnblocks > sbi->nbperpage))
1052 /* get the last block of the current allocation */
1053 lastblkno = blkno + nblocks - 1;
1055 /* determine the block number of the block following
1056 * the existing allocation.
1058 extblkno = lastblkno + 1;
1060 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
1062 /* better be within the file system */
1064 if (lastblkno < 0 || lastblkno >= bmp->db_mapsize) {
1065 IREAD_UNLOCK(ipbmap);
1066 jfs_error(ip->i_sb, "the block is outside the filesystem\n");
1070 /* we'll attempt to extend the current allocation in place by
1071 * allocating the additional blocks as the blocks immediately
1072 * following the current allocation. we only try to extend the
1073 * current allocation in place if the number of additional blocks
1074 * can fit into a dmap, the last block of the current allocation
1075 * is not the last block of the file system, and the start of the
1076 * inplace extension is not on an allocation group boundary.
1078 if (addnblocks > BPERDMAP || extblkno >= bmp->db_mapsize ||
1079 (extblkno & (bmp->db_agsize - 1)) == 0) {
1080 IREAD_UNLOCK(ipbmap);
1084 /* get the buffer for the dmap containing the first block
1087 lblkno = BLKTODMAP(extblkno, bmp->db_l2nbperpage);
1088 mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
1090 IREAD_UNLOCK(ipbmap);
1094 dp = (struct dmap *) mp->data;
1096 /* try to allocate the blocks immediately following the
1097 * current allocation.
1099 rc = dbAllocNext(bmp, dp, extblkno, (int) addnblocks);
1101 IREAD_UNLOCK(ipbmap);
1103 /* were we successful ? */
1107 /* we were not successful */
1108 release_metapage(mp);
1115 * NAME: dbAllocNext()
1117 * FUNCTION: attempt to allocate the blocks of the specified block
1118 * range within a dmap.
1121 * bmp - pointer to bmap descriptor
1122 * dp - pointer to dmap.
1123 * blkno - starting block number of the range.
1124 * nblocks - number of contiguous free blocks of the range.
1128 * -ENOSPC - insufficient disk resources
1131 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1133 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
1136 int dbitno, word, rembits, nb, nwords, wbitno, nw;
1141 if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1142 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n");
1146 /* pick up a pointer to the leaves of the dmap tree.
1148 leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1150 /* determine the bit number and word within the dmap of the
1153 dbitno = blkno & (BPERDMAP - 1);
1154 word = dbitno >> L2DBWORD;
1156 /* check if the specified block range is contained within
1159 if (dbitno + nblocks > BPERDMAP)
1162 /* check if the starting leaf indicates that anything
1165 if (leaf[word] == NOFREE)
1168 /* check the dmaps words corresponding to block range to see
1169 * if the block range is free. not all bits of the first and
1170 * last words may be contained within the block range. if this
1171 * is the case, we'll work against those words (i.e. partial first
1172 * and/or last) on an individual basis (a single pass) and examine
1173 * the actual bits to determine if they are free. a single pass
1174 * will be used for all dmap words fully contained within the
1175 * specified range. within this pass, the leaves of the dmap
1176 * tree will be examined to determine if the blocks are free. a
1177 * single leaf may describe the free space of multiple dmap
1178 * words, so we may visit only a subset of the actual leaves
1179 * corresponding to the dmap words of the block range.
1181 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
1182 /* determine the bit number within the word and
1183 * the number of bits within the word.
1185 wbitno = dbitno & (DBWORD - 1);
1186 nb = min(rembits, DBWORD - wbitno);
1188 /* check if only part of the word is to be examined.
1191 /* check if the bits are free.
1193 mask = (ONES << (DBWORD - nb) >> wbitno);
1194 if ((mask & ~le32_to_cpu(dp->wmap[word])) != mask)
1199 /* one or more dmap words are fully contained
1200 * within the block range. determine how many
1201 * words and how many bits.
1203 nwords = rembits >> L2DBWORD;
1204 nb = nwords << L2DBWORD;
1206 /* now examine the appropriate leaves to determine
1207 * if the blocks are free.
1209 while (nwords > 0) {
1210 /* does the leaf describe any free space ?
1212 if (leaf[word] < BUDMIN)
1215 /* determine the l2 number of bits provided
1219 min_t(int, leaf[word], NLSTOL2BSZ(nwords));
1221 /* determine how many words were handled.
1223 nw = BUDSIZE(l2size, BUDMIN);
1231 /* allocate the blocks.
1233 return (dbAllocDmap(bmp, dp, blkno, nblocks));
1238 * NAME: dbAllocNear()
1240 * FUNCTION: attempt to allocate a number of contiguous free blocks near
1241 * a specified block (hint) within a dmap.
1243 * starting with the dmap leaf that covers the hint, we'll
1244 * check the next four contiguous leaves for sufficient free
1245 * space. if sufficient free space is found, we'll allocate
1246 * the desired free space.
1249 * bmp - pointer to bmap descriptor
1250 * dp - pointer to dmap.
1251 * blkno - block number to allocate near.
1252 * nblocks - actual number of contiguous free blocks desired.
1253 * l2nb - log2 number of contiguous free blocks desired.
1254 * results - on successful return, set to the starting block number
1255 * of the newly allocated range.
1259 * -ENOSPC - insufficient disk resources
1262 * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1265 dbAllocNear(struct bmap * bmp,
1266 struct dmap * dp, s64 blkno, int nblocks, int l2nb, s64 * results)
1268 int word, lword, rc;
1271 if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1272 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n");
1276 leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1278 /* determine the word within the dmap that holds the hint
1279 * (i.e. blkno). also, determine the last word in the dmap
1280 * that we'll include in our examination.
1282 word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
1283 lword = min(word + 4, LPERDMAP);
1285 /* examine the leaves for sufficient free space.
1287 for (; word < lword; word++) {
1288 /* does the leaf describe sufficient free space ?
1290 if (leaf[word] < l2nb)
1293 /* determine the block number within the file system
1294 * of the first block described by this dmap word.
1296 blkno = le64_to_cpu(dp->start) + (word << L2DBWORD);
1298 /* if not all bits of the dmap word are free, get the
1299 * starting bit number within the dmap word of the required
1300 * string of free bits and adjust the block number with the
1303 if (leaf[word] < BUDMIN)
1305 dbFindBits(le32_to_cpu(dp->wmap[word]), l2nb);
1307 /* allocate the blocks.
1309 if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1322 * FUNCTION: attempt to allocate the specified number of contiguous
1323 * free blocks within the specified allocation group.
1325 * unless the allocation group size is equal to the number
1326 * of blocks per dmap, the dmap control pages will be used to
1327 * find the required free space, if available. we start the
1328 * search at the highest dmap control page level which
1329 * distinctly describes the allocation group's free space
1330 * (i.e. the highest level at which the allocation group's
1331 * free space is not mixed in with that of any other group).
1332 * in addition, we start the search within this level at a
1333 * height of the dmapctl dmtree at which the nodes distinctly
1334 * describe the allocation group's free space. at this height,
1335 * the allocation group's free space may be represented by 1
1336 * or two sub-trees, depending on the allocation group size.
1337 * we search the top nodes of these subtrees left to right for
1338 * sufficient free space. if sufficient free space is found,
1339 * the subtree is searched to find the leftmost leaf that
1340 * has free space. once we have made it to the leaf, we
1341 * move the search to the next lower level dmap control page
1342 * corresponding to this leaf. we continue down the dmap control
1343 * pages until we find the dmap that contains or starts the
1344 * sufficient free space and we allocate at this dmap.
1346 * if the allocation group size is equal to the dmap size,
1347 * we'll start at the dmap corresponding to the allocation
1348 * group and attempt the allocation at this level.
1350 * the dmap control page search is also not performed if the
1351 * allocation group is completely free and we go to the first
1352 * dmap of the allocation group to do the allocation. this is
1353 * done because the allocation group may be part (not the first
1354 * part) of a larger binary buddy system, causing the dmap
1355 * control pages to indicate no free space (NOFREE) within
1356 * the allocation group.
1359 * bmp - pointer to bmap descriptor
1360 * agno - allocation group number.
1361 * nblocks - actual number of contiguous free blocks desired.
1362 * l2nb - log2 number of contiguous free blocks desired.
1363 * results - on successful return, set to the starting block number
1364 * of the newly allocated range.
1368 * -ENOSPC - insufficient disk resources
1371 * note: IWRITE_LOCK(ipmap) held on entry/exit;
1374 dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results)
1376 struct metapage *mp;
1377 struct dmapctl *dcp;
1378 int rc, ti, i, k, m, n, agperlev;
1382 /* allocation request should not be for more than the
1383 * allocation group size.
1385 if (l2nb > bmp->db_agl2size) {
1386 jfs_error(bmp->db_ipbmap->i_sb,
1387 "allocation request is larger than the allocation group size\n");
1391 /* determine the starting block number of the allocation
1394 blkno = (s64) agno << bmp->db_agl2size;
1396 /* check if the allocation group size is the minimum allocation
1397 * group size or if the allocation group is completely free. if
1398 * the allocation group size is the minimum size of BPERDMAP (i.e.
1399 * 1 dmap), there is no need to search the dmap control page (below)
1400 * that fully describes the allocation group since the allocation
1401 * group is already fully described by a dmap. in this case, we
1402 * just call dbAllocCtl() to search the dmap tree and allocate the
1403 * required space if available.
1405 * if the allocation group is completely free, dbAllocCtl() is
1406 * also called to allocate the required space. this is done for
1407 * two reasons. first, it makes no sense searching the dmap control
1408 * pages for free space when we know that free space exists. second,
1409 * the dmap control pages may indicate that the allocation group
1410 * has no free space if the allocation group is part (not the first
1411 * part) of a larger binary buddy system.
1413 if (bmp->db_agsize == BPERDMAP
1414 || bmp->db_agfree[agno] == bmp->db_agsize) {
1415 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1416 if ((rc == -ENOSPC) &&
1417 (bmp->db_agfree[agno] == bmp->db_agsize)) {
1418 printk(KERN_ERR "blkno = %Lx, blocks = %Lx\n",
1419 (unsigned long long) blkno,
1420 (unsigned long long) nblocks);
1421 jfs_error(bmp->db_ipbmap->i_sb,
1422 "dbAllocCtl failed in free AG\n");
1427 /* the buffer for the dmap control page that fully describes the
1430 lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, bmp->db_aglevel);
1431 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1434 dcp = (struct dmapctl *) mp->data;
1435 budmin = dcp->budmin;
1437 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1438 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n");
1439 release_metapage(mp);
1443 /* search the subtree(s) of the dmap control page that describes
1444 * the allocation group, looking for sufficient free space. to begin,
1445 * determine how many allocation groups are represented in a dmap
1446 * control page at the control page level (i.e. L0, L1, L2) that
1447 * fully describes an allocation group. next, determine the starting
1448 * tree index of this allocation group within the control page.
1451 (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp->db_agwidth;
1452 ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
1454 /* dmap control page trees fan-out by 4 and a single allocation
1455 * group may be described by 1 or 2 subtrees within the ag level
1456 * dmap control page, depending upon the ag size. examine the ag's
1457 * subtrees for sufficient free space, starting with the leftmost
1460 for (i = 0; i < bmp->db_agwidth; i++, ti++) {
1461 /* is there sufficient free space ?
1463 if (l2nb > dcp->stree[ti])
1466 /* sufficient free space found in a subtree. now search down
1467 * the subtree to find the leftmost leaf that describes this
1470 for (k = bmp->db_agheight; k > 0; k--) {
1471 for (n = 0, m = (ti << 2) + 1; n < 4; n++) {
1472 if (l2nb <= dcp->stree[m + n]) {
1478 jfs_error(bmp->db_ipbmap->i_sb,
1479 "failed descending stree\n");
1480 release_metapage(mp);
1485 /* determine the block number within the file system
1486 * that corresponds to this leaf.
1488 if (bmp->db_aglevel == 2)
1490 else if (bmp->db_aglevel == 1)
1491 blkno &= ~(MAXL1SIZE - 1);
1492 else /* bmp->db_aglevel == 0 */
1493 blkno &= ~(MAXL0SIZE - 1);
1496 ((s64) (ti - le32_to_cpu(dcp->leafidx))) << budmin;
1498 /* release the buffer in preparation for going down
1499 * the next level of dmap control pages.
1501 release_metapage(mp);
1503 /* check if we need to continue to search down the lower
1504 * level dmap control pages. we need to if the number of
1505 * blocks required is less than maximum number of blocks
1506 * described at the next lower level.
1508 if (l2nb < budmin) {
1510 /* search the lower level dmap control pages to get
1511 * the starting block number of the dmap that
1512 * contains or starts off the free space.
1515 dbFindCtl(bmp, l2nb, bmp->db_aglevel - 1,
1517 if (rc == -ENOSPC) {
1518 jfs_error(bmp->db_ipbmap->i_sb,
1519 "control page inconsistent\n");
1526 /* allocate the blocks.
1528 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1529 if (rc == -ENOSPC) {
1530 jfs_error(bmp->db_ipbmap->i_sb,
1531 "unable to allocate blocks\n");
1537 /* no space in the allocation group. release the buffer and
1540 release_metapage(mp);
1547 * NAME: dbAllocAny()
1549 * FUNCTION: attempt to allocate the specified number of contiguous
1550 * free blocks anywhere in the file system.
1552 * dbAllocAny() attempts to find the sufficient free space by
1553 * searching down the dmap control pages, starting with the
1554 * highest level (i.e. L0, L1, L2) control page. if free space
1555 * large enough to satisfy the desired free space is found, the
1556 * desired free space is allocated.
1559 * bmp - pointer to bmap descriptor
1560 * nblocks - actual number of contiguous free blocks desired.
1561 * l2nb - log2 number of contiguous free blocks desired.
1562 * results - on successful return, set to the starting block number
1563 * of the newly allocated range.
1567 * -ENOSPC - insufficient disk resources
1570 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1572 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results)
1577 /* starting with the top level dmap control page, search
1578 * down the dmap control levels for sufficient free space.
1579 * if free space is found, dbFindCtl() returns the starting
1580 * block number of the dmap that contains or starts off the
1581 * range of free space.
1583 if ((rc = dbFindCtl(bmp, l2nb, bmp->db_maxlevel, &blkno)))
1586 /* allocate the blocks.
1588 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1589 if (rc == -ENOSPC) {
1590 jfs_error(bmp->db_ipbmap->i_sb, "unable to allocate blocks\n");
1598 * NAME: dbDiscardAG()
1600 * FUNCTION: attempt to discard (TRIM) all free blocks of specific AG
1603 * 1) allocate blocks, as large as possible and save them
1604 * while holding IWRITE_LOCK on ipbmap
1605 * 2) trim all these saved block/length values
1606 * 3) mark the blocks free again
1609 * - we work only on one ag at some time, minimizing how long we
1610 * need to lock ipbmap
1611 * - reading / writing the fs is possible most time, even on
1615 * - we write two times to the dmapctl and dmap pages
1616 * - but for me, this seems the best way, better ideas?
1620 * ip - pointer to in-core inode
1622 * minlen - minimum value of contiguous blocks
1625 * s64 - actual number of blocks trimmed
1627 s64 dbDiscardAG(struct inode *ip, int agno, s64 minlen)
1629 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
1630 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
1634 struct super_block *sb = ipbmap->i_sb;
1641 /* max blkno / nblocks pairs to trim */
1642 int count = 0, range_cnt;
1645 /* prevent others from writing new stuff here, while trimming */
1646 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
1648 nblocks = bmp->db_agfree[agno];
1649 max_ranges = nblocks;
1650 do_div(max_ranges, minlen);
1651 range_cnt = min_t(u64, max_ranges + 1, 32 * 1024);
1652 totrim = kmalloc(sizeof(struct range2trim) * range_cnt, GFP_NOFS);
1653 if (totrim == NULL) {
1654 jfs_error(bmp->db_ipbmap->i_sb, "no memory for trim array\n");
1655 IWRITE_UNLOCK(ipbmap);
1660 while (nblocks >= minlen) {
1661 l2nb = BLKSTOL2(nblocks);
1663 /* 0 = okay, -EIO = fatal, -ENOSPC -> try smaller block */
1664 rc = dbAllocAG(bmp, agno, nblocks, l2nb, &blkno);
1667 tt->nblocks = nblocks;
1670 /* the whole ag is free, trim now */
1671 if (bmp->db_agfree[agno] == 0)
1674 /* give a hint for the next while */
1675 nblocks = bmp->db_agfree[agno];
1677 } else if (rc == -ENOSPC) {
1678 /* search for next smaller log2 block */
1679 l2nb = BLKSTOL2(nblocks) - 1;
1680 nblocks = 1LL << l2nb;
1682 /* Trim any already allocated blocks */
1683 jfs_error(bmp->db_ipbmap->i_sb, "-EIO\n");
1687 /* check, if our trim array is full */
1688 if (unlikely(count >= range_cnt - 1))
1691 IWRITE_UNLOCK(ipbmap);
1693 tt->nblocks = 0; /* mark the current end */
1694 for (tt = totrim; tt->nblocks != 0; tt++) {
1695 /* when mounted with online discard, dbFree() will
1696 * call jfs_issue_discard() itself */
1697 if (!(JFS_SBI(sb)->flag & JFS_DISCARD))
1698 jfs_issue_discard(ip, tt->blkno, tt->nblocks);
1699 dbFree(ip, tt->blkno, tt->nblocks);
1700 trimmed += tt->nblocks;
1710 * FUNCTION: starting at a specified dmap control page level and block
1711 * number, search down the dmap control levels for a range of
1712 * contiguous free blocks large enough to satisfy an allocation
1713 * request for the specified number of free blocks.
1715 * if sufficient contiguous free blocks are found, this routine
1716 * returns the starting block number within a dmap page that
1717 * contains or starts a range of contiqious free blocks that
1718 * is sufficient in size.
1721 * bmp - pointer to bmap descriptor
1722 * level - starting dmap control page level.
1723 * l2nb - log2 number of contiguous free blocks desired.
1724 * *blkno - on entry, starting block number for conducting the search.
1725 * on successful return, the first block within a dmap page
1726 * that contains or starts a range of contiguous free blocks.
1730 * -ENOSPC - insufficient disk resources
1733 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1735 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno)
1737 int rc, leafidx, lev;
1739 struct dmapctl *dcp;
1741 struct metapage *mp;
1743 /* starting at the specified dmap control page level and block
1744 * number, search down the dmap control levels for the starting
1745 * block number of a dmap page that contains or starts off
1746 * sufficient free blocks.
1748 for (lev = level, b = *blkno; lev >= 0; lev--) {
1749 /* get the buffer of the dmap control page for the block
1750 * number and level (i.e. L0, L1, L2).
1752 lblkno = BLKTOCTL(b, bmp->db_l2nbperpage, lev);
1753 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1756 dcp = (struct dmapctl *) mp->data;
1757 budmin = dcp->budmin;
1759 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1760 jfs_error(bmp->db_ipbmap->i_sb,
1761 "Corrupt dmapctl page\n");
1762 release_metapage(mp);
1766 /* search the tree within the dmap control page for
1767 * sufficient free space. if sufficient free space is found,
1768 * dbFindLeaf() returns the index of the leaf at which
1769 * free space was found.
1771 rc = dbFindLeaf((dmtree_t *) dcp, l2nb, &leafidx);
1773 /* release the buffer.
1775 release_metapage(mp);
1781 jfs_error(bmp->db_ipbmap->i_sb,
1782 "dmap inconsistent\n");
1788 /* adjust the block number to reflect the location within
1789 * the dmap control page (i.e. the leaf) at which free
1792 b += (((s64) leafidx) << budmin);
1794 /* we stop the search at this dmap control page level if
1795 * the number of blocks required is greater than or equal
1796 * to the maximum number of blocks described at the next
1809 * NAME: dbAllocCtl()
1811 * FUNCTION: attempt to allocate a specified number of contiguous
1812 * blocks starting within a specific dmap.
1814 * this routine is called by higher level routines that search
1815 * the dmap control pages above the actual dmaps for contiguous
1816 * free space. the result of successful searches by these
1817 * routines are the starting block numbers within dmaps, with
1818 * the dmaps themselves containing the desired contiguous free
1819 * space or starting a contiguous free space of desired size
1820 * that is made up of the blocks of one or more dmaps. these
1821 * calls should not fail due to insufficent resources.
1823 * this routine is called in some cases where it is not known
1824 * whether it will fail due to insufficient resources. more
1825 * specifically, this occurs when allocating from an allocation
1826 * group whose size is equal to the number of blocks per dmap.
1827 * in this case, the dmap control pages are not examined prior
1828 * to calling this routine (to save pathlength) and the call
1831 * for a request size that fits within a dmap, this routine relies
1832 * upon the dmap's dmtree to find the requested contiguous free
1833 * space. for request sizes that are larger than a dmap, the
1834 * requested free space will start at the first block of the
1835 * first dmap (i.e. blkno).
1838 * bmp - pointer to bmap descriptor
1839 * nblocks - actual number of contiguous free blocks to allocate.
1840 * l2nb - log2 number of contiguous free blocks to allocate.
1841 * blkno - starting block number of the dmap to start the allocation
1843 * results - on successful return, set to the starting block number
1844 * of the newly allocated range.
1848 * -ENOSPC - insufficient disk resources
1851 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1854 dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, s64 * results)
1858 struct metapage *mp;
1861 /* check if the allocation request is confined to a single dmap.
1863 if (l2nb <= L2BPERDMAP) {
1864 /* get the buffer for the dmap.
1866 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
1867 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1870 dp = (struct dmap *) mp->data;
1872 /* try to allocate the blocks.
1874 rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results);
1876 mark_metapage_dirty(mp);
1878 release_metapage(mp);
1883 /* allocation request involving multiple dmaps. it must start on
1886 assert((blkno & (BPERDMAP - 1)) == 0);
1888 /* allocate the blocks dmap by dmap.
1890 for (n = nblocks, b = blkno; n > 0; n -= nb, b += nb) {
1891 /* get the buffer for the dmap.
1893 lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1894 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1899 dp = (struct dmap *) mp->data;
1901 /* the dmap better be all free.
1903 if (dp->tree.stree[ROOT] != L2BPERDMAP) {
1904 release_metapage(mp);
1905 jfs_error(bmp->db_ipbmap->i_sb,
1906 "the dmap is not all free\n");
1911 /* determine how many blocks to allocate from this dmap.
1913 nb = min_t(s64, n, BPERDMAP);
1915 /* allocate the blocks from the dmap.
1917 if ((rc = dbAllocDmap(bmp, dp, b, nb))) {
1918 release_metapage(mp);
1922 /* write the buffer.
1927 /* set the results (starting block number) and return.
1932 /* something failed in handling an allocation request involving
1933 * multiple dmaps. we'll try to clean up by backing out any
1934 * allocation that has already happened for this request. if
1935 * we fail in backing out the allocation, we'll mark the file
1936 * system to indicate that blocks have been leaked.
1940 /* try to backout the allocations dmap by dmap.
1942 for (n = nblocks - n, b = blkno; n > 0;
1943 n -= BPERDMAP, b += BPERDMAP) {
1944 /* get the buffer for this dmap.
1946 lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1947 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1949 /* could not back out. mark the file system
1950 * to indicate that we have leaked blocks.
1952 jfs_error(bmp->db_ipbmap->i_sb,
1953 "I/O Error: Block Leakage\n");
1956 dp = (struct dmap *) mp->data;
1958 /* free the blocks is this dmap.
1960 if (dbFreeDmap(bmp, dp, b, BPERDMAP)) {
1961 /* could not back out. mark the file system
1962 * to indicate that we have leaked blocks.
1964 release_metapage(mp);
1965 jfs_error(bmp->db_ipbmap->i_sb, "Block Leakage\n");
1969 /* write the buffer.
1979 * NAME: dbAllocDmapLev()
1981 * FUNCTION: attempt to allocate a specified number of contiguous blocks
1982 * from a specified dmap.
1984 * this routine checks if the contiguous blocks are available.
1985 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1989 * mp - pointer to bmap descriptor
1990 * dp - pointer to dmap to attempt to allocate blocks from.
1991 * l2nb - log2 number of contiguous block desired.
1992 * nblocks - actual number of contiguous block desired.
1993 * results - on successful return, set to the starting block number
1994 * of the newly allocated range.
1998 * -ENOSPC - insufficient disk resources
2001 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
2002 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
2005 dbAllocDmapLev(struct bmap * bmp,
2006 struct dmap * dp, int nblocks, int l2nb, s64 * results)
2011 /* can't be more than a dmaps worth of blocks */
2012 assert(l2nb <= L2BPERDMAP);
2014 /* search the tree within the dmap page for sufficient
2015 * free space. if sufficient free space is found, dbFindLeaf()
2016 * returns the index of the leaf at which free space was found.
2018 if (dbFindLeaf((dmtree_t *) & dp->tree, l2nb, &leafidx))
2021 /* determine the block number within the file system corresponding
2022 * to the leaf at which free space was found.
2024 blkno = le64_to_cpu(dp->start) + (leafidx << L2DBWORD);
2026 /* if not all bits of the dmap word are free, get the starting
2027 * bit number within the dmap word of the required string of free
2028 * bits and adjust the block number with this value.
2030 if (dp->tree.stree[leafidx + LEAFIND] < BUDMIN)
2031 blkno += dbFindBits(le32_to_cpu(dp->wmap[leafidx]), l2nb);
2033 /* allocate the blocks */
2034 if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
2042 * NAME: dbAllocDmap()
2044 * FUNCTION: adjust the disk allocation map to reflect the allocation
2045 * of a specified block range within a dmap.
2047 * this routine allocates the specified blocks from the dmap
2048 * through a call to dbAllocBits(). if the allocation of the
2049 * block range causes the maximum string of free blocks within
2050 * the dmap to change (i.e. the value of the root of the dmap's
2051 * dmtree), this routine will cause this change to be reflected
2052 * up through the appropriate levels of the dmap control pages
2053 * by a call to dbAdjCtl() for the L0 dmap control page that
2057 * bmp - pointer to bmap descriptor
2058 * dp - pointer to dmap to allocate the block range from.
2059 * blkno - starting block number of the block to be allocated.
2060 * nblocks - number of blocks to be allocated.
2066 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2068 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2074 /* save the current value of the root (i.e. maximum free string)
2077 oldroot = dp->tree.stree[ROOT];
2079 /* allocate the specified (blocks) bits */
2080 dbAllocBits(bmp, dp, blkno, nblocks);
2082 /* if the root has not changed, done. */
2083 if (dp->tree.stree[ROOT] == oldroot)
2086 /* root changed. bubble the change up to the dmap control pages.
2087 * if the adjustment of the upper level control pages fails,
2088 * backout the bit allocation (thus making everything consistent).
2090 if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 1, 0)))
2091 dbFreeBits(bmp, dp, blkno, nblocks);
2098 * NAME: dbFreeDmap()
2100 * FUNCTION: adjust the disk allocation map to reflect the allocation
2101 * of a specified block range within a dmap.
2103 * this routine frees the specified blocks from the dmap through
2104 * a call to dbFreeBits(). if the deallocation of the block range
2105 * causes the maximum string of free blocks within the dmap to
2106 * change (i.e. the value of the root of the dmap's dmtree), this
2107 * routine will cause this change to be reflected up through the
2108 * appropriate levels of the dmap control pages by a call to
2109 * dbAdjCtl() for the L0 dmap control page that covers this dmap.
2112 * bmp - pointer to bmap descriptor
2113 * dp - pointer to dmap to free the block range from.
2114 * blkno - starting block number of the block to be freed.
2115 * nblocks - number of blocks to be freed.
2121 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2123 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2129 /* save the current value of the root (i.e. maximum free string)
2132 oldroot = dp->tree.stree[ROOT];
2134 /* free the specified (blocks) bits */
2135 rc = dbFreeBits(bmp, dp, blkno, nblocks);
2137 /* if error or the root has not changed, done. */
2138 if (rc || (dp->tree.stree[ROOT] == oldroot))
2141 /* root changed. bubble the change up to the dmap control pages.
2142 * if the adjustment of the upper level control pages fails,
2143 * backout the deallocation.
2145 if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 0, 0))) {
2146 word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
2148 /* as part of backing out the deallocation, we will have
2149 * to back split the dmap tree if the deallocation caused
2150 * the freed blocks to become part of a larger binary buddy
2153 if (dp->tree.stree[word] == NOFREE)
2154 dbBackSplit((dmtree_t *) & dp->tree, word);
2156 dbAllocBits(bmp, dp, blkno, nblocks);
2164 * NAME: dbAllocBits()
2166 * FUNCTION: allocate a specified block range from a dmap.
2168 * this routine updates the dmap to reflect the working
2169 * state allocation of the specified block range. it directly
2170 * updates the bits of the working map and causes the adjustment
2171 * of the binary buddy system described by the dmap's dmtree
2172 * leaves to reflect the bits allocated. it also causes the
2173 * dmap's dmtree, as a whole, to reflect the allocated range.
2176 * bmp - pointer to bmap descriptor
2177 * dp - pointer to dmap to allocate bits from.
2178 * blkno - starting block number of the bits to be allocated.
2179 * nblocks - number of bits to be allocated.
2181 * RETURN VALUES: none
2183 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2185 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2188 int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2189 dmtree_t *tp = (dmtree_t *) & dp->tree;
2193 /* pick up a pointer to the leaves of the dmap tree */
2194 leaf = dp->tree.stree + LEAFIND;
2196 /* determine the bit number and word within the dmap of the
2199 dbitno = blkno & (BPERDMAP - 1);
2200 word = dbitno >> L2DBWORD;
2202 /* block range better be within the dmap */
2203 assert(dbitno + nblocks <= BPERDMAP);
2205 /* allocate the bits of the dmap's words corresponding to the block
2206 * range. not all bits of the first and last words may be contained
2207 * within the block range. if this is the case, we'll work against
2208 * those words (i.e. partial first and/or last) on an individual basis
2209 * (a single pass), allocating the bits of interest by hand and
2210 * updating the leaf corresponding to the dmap word. a single pass
2211 * will be used for all dmap words fully contained within the
2212 * specified range. within this pass, the bits of all fully contained
2213 * dmap words will be marked as free in a single shot and the leaves
2214 * will be updated. a single leaf may describe the free space of
2215 * multiple dmap words, so we may update only a subset of the actual
2216 * leaves corresponding to the dmap words of the block range.
2218 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2219 /* determine the bit number within the word and
2220 * the number of bits within the word.
2222 wbitno = dbitno & (DBWORD - 1);
2223 nb = min(rembits, DBWORD - wbitno);
2225 /* check if only part of a word is to be allocated.
2228 /* allocate (set to 1) the appropriate bits within
2231 dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
2234 /* update the leaf for this dmap word. in addition
2235 * to setting the leaf value to the binary buddy max
2236 * of the updated dmap word, dbSplit() will split
2237 * the binary system of the leaves if need be.
2239 dbSplit(tp, word, BUDMIN,
2240 dbMaxBud((u8 *) & dp->wmap[word]));
2244 /* one or more dmap words are fully contained
2245 * within the block range. determine how many
2246 * words and allocate (set to 1) the bits of these
2249 nwords = rembits >> L2DBWORD;
2250 memset(&dp->wmap[word], (int) ONES, nwords * 4);
2252 /* determine how many bits.
2254 nb = nwords << L2DBWORD;
2256 /* now update the appropriate leaves to reflect
2257 * the allocated words.
2259 for (; nwords > 0; nwords -= nw) {
2260 if (leaf[word] < BUDMIN) {
2261 jfs_error(bmp->db_ipbmap->i_sb,
2262 "leaf page corrupt\n");
2266 /* determine what the leaf value should be
2267 * updated to as the minimum of the l2 number
2268 * of bits being allocated and the l2 number
2269 * of bits currently described by this leaf.
2271 size = min_t(int, leaf[word],
2272 NLSTOL2BSZ(nwords));
2274 /* update the leaf to reflect the allocation.
2275 * in addition to setting the leaf value to
2276 * NOFREE, dbSplit() will split the binary
2277 * system of the leaves to reflect the current
2278 * allocation (size).
2280 dbSplit(tp, word, size, NOFREE);
2282 /* get the number of dmap words handled */
2283 nw = BUDSIZE(size, BUDMIN);
2289 /* update the free count for this dmap */
2290 le32_add_cpu(&dp->nfree, -nblocks);
2294 /* if this allocation group is completely free,
2295 * update the maximum allocation group number if this allocation
2296 * group is the new max.
2298 agno = blkno >> bmp->db_agl2size;
2299 if (agno > bmp->db_maxag)
2300 bmp->db_maxag = agno;
2302 /* update the free count for the allocation group and map */
2303 bmp->db_agfree[agno] -= nblocks;
2304 bmp->db_nfree -= nblocks;
2311 * NAME: dbFreeBits()
2313 * FUNCTION: free a specified block range from a dmap.
2315 * this routine updates the dmap to reflect the working
2316 * state allocation of the specified block range. it directly
2317 * updates the bits of the working map and causes the adjustment
2318 * of the binary buddy system described by the dmap's dmtree
2319 * leaves to reflect the bits freed. it also causes the dmap's
2320 * dmtree, as a whole, to reflect the deallocated range.
2323 * bmp - pointer to bmap descriptor
2324 * dp - pointer to dmap to free bits from.
2325 * blkno - starting block number of the bits to be freed.
2326 * nblocks - number of bits to be freed.
2328 * RETURN VALUES: 0 for success
2330 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2332 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2335 int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2336 dmtree_t *tp = (dmtree_t *) & dp->tree;
2340 /* determine the bit number and word within the dmap of the
2343 dbitno = blkno & (BPERDMAP - 1);
2344 word = dbitno >> L2DBWORD;
2346 /* block range better be within the dmap.
2348 assert(dbitno + nblocks <= BPERDMAP);
2350 /* free the bits of the dmaps words corresponding to the block range.
2351 * not all bits of the first and last words may be contained within
2352 * the block range. if this is the case, we'll work against those
2353 * words (i.e. partial first and/or last) on an individual basis
2354 * (a single pass), freeing the bits of interest by hand and updating
2355 * the leaf corresponding to the dmap word. a single pass will be used
2356 * for all dmap words fully contained within the specified range.
2357 * within this pass, the bits of all fully contained dmap words will
2358 * be marked as free in a single shot and the leaves will be updated. a
2359 * single leaf may describe the free space of multiple dmap words,
2360 * so we may update only a subset of the actual leaves corresponding
2361 * to the dmap words of the block range.
2363 * dbJoin() is used to update leaf values and will join the binary
2364 * buddy system of the leaves if the new leaf values indicate this
2367 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2368 /* determine the bit number within the word and
2369 * the number of bits within the word.
2371 wbitno = dbitno & (DBWORD - 1);
2372 nb = min(rembits, DBWORD - wbitno);
2374 /* check if only part of a word is to be freed.
2377 /* free (zero) the appropriate bits within this
2381 cpu_to_le32(~(ONES << (DBWORD - nb)
2384 /* update the leaf for this dmap word.
2386 rc = dbJoin(tp, word,
2387 dbMaxBud((u8 *) & dp->wmap[word]));
2393 /* one or more dmap words are fully contained
2394 * within the block range. determine how many
2395 * words and free (zero) the bits of these words.
2397 nwords = rembits >> L2DBWORD;
2398 memset(&dp->wmap[word], 0, nwords * 4);
2400 /* determine how many bits.
2402 nb = nwords << L2DBWORD;
2404 /* now update the appropriate leaves to reflect
2407 for (; nwords > 0; nwords -= nw) {
2408 /* determine what the leaf value should be
2409 * updated to as the minimum of the l2 number
2410 * of bits being freed and the l2 (max) number
2411 * of bits that can be described by this leaf.
2415 (word, L2LPERDMAP, BUDMIN),
2416 NLSTOL2BSZ(nwords));
2420 rc = dbJoin(tp, word, size);
2424 /* get the number of dmap words handled.
2426 nw = BUDSIZE(size, BUDMIN);
2432 /* update the free count for this dmap.
2434 le32_add_cpu(&dp->nfree, nblocks);
2438 /* update the free count for the allocation group and
2441 agno = blkno >> bmp->db_agl2size;
2442 bmp->db_nfree += nblocks;
2443 bmp->db_agfree[agno] += nblocks;
2445 /* check if this allocation group is not completely free and
2446 * if it is currently the maximum (rightmost) allocation group.
2447 * if so, establish the new maximum allocation group number by
2448 * searching left for the first allocation group with allocation.
2450 if ((bmp->db_agfree[agno] == bmp->db_agsize && agno == bmp->db_maxag) ||
2451 (agno == bmp->db_numag - 1 &&
2452 bmp->db_agfree[agno] == (bmp-> db_mapsize & (BPERDMAP - 1)))) {
2453 while (bmp->db_maxag > 0) {
2455 if (bmp->db_agfree[bmp->db_maxag] !=
2460 /* re-establish the allocation group preference if the
2461 * current preference is right of the maximum allocation
2464 if (bmp->db_agpref > bmp->db_maxag)
2465 bmp->db_agpref = bmp->db_maxag;
2477 * FUNCTION: adjust a dmap control page at a specified level to reflect
2478 * the change in a lower level dmap or dmap control page's
2479 * maximum string of free blocks (i.e. a change in the root
2480 * of the lower level object's dmtree) due to the allocation
2481 * or deallocation of a range of blocks with a single dmap.
2483 * on entry, this routine is provided with the new value of
2484 * the lower level dmap or dmap control page root and the
2485 * starting block number of the block range whose allocation
2486 * or deallocation resulted in the root change. this range
2487 * is respresented by a single leaf of the current dmapctl
2488 * and the leaf will be updated with this value, possibly
2489 * causing a binary buddy system within the leaves to be
2490 * split or joined. the update may also cause the dmapctl's
2491 * dmtree to be updated.
2493 * if the adjustment of the dmap control page, itself, causes its
2494 * root to change, this change will be bubbled up to the next dmap
2495 * control level by a recursive call to this routine, specifying
2496 * the new root value and the next dmap control page level to
2499 * bmp - pointer to bmap descriptor
2500 * blkno - the first block of a block range within a dmap. it is
2501 * the allocation or deallocation of this block range that
2502 * requires the dmap control page to be adjusted.
2503 * newval - the new value of the lower level dmap or dmap control
2505 * alloc - 'true' if adjustment is due to an allocation.
2506 * level - current level of dmap control page (i.e. L0, L1, L2) to
2513 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2516 dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, int level)
2518 struct metapage *mp;
2522 struct dmapctl *dcp;
2525 /* get the buffer for the dmap control page for the specified
2526 * block number and control page level.
2528 lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, level);
2529 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
2532 dcp = (struct dmapctl *) mp->data;
2534 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
2535 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n");
2536 release_metapage(mp);
2540 /* determine the leaf number corresponding to the block and
2541 * the index within the dmap control tree.
2543 leafno = BLKTOCTLLEAF(blkno, dcp->budmin);
2544 ti = leafno + le32_to_cpu(dcp->leafidx);
2546 /* save the current leaf value and the current root level (i.e.
2547 * maximum l2 free string described by this dmapctl).
2549 oldval = dcp->stree[ti];
2550 oldroot = dcp->stree[ROOT];
2552 /* check if this is a control page update for an allocation.
2553 * if so, update the leaf to reflect the new leaf value using
2554 * dbSplit(); otherwise (deallocation), use dbJoin() to update
2555 * the leaf with the new value. in addition to updating the
2556 * leaf, dbSplit() will also split the binary buddy system of
2557 * the leaves, if required, and bubble new values within the
2558 * dmapctl tree, if required. similarly, dbJoin() will join
2559 * the binary buddy system of leaves and bubble new values up
2560 * the dmapctl tree as required by the new leaf value.
2563 /* check if we are in the middle of a binary buddy
2564 * system. this happens when we are performing the
2565 * first allocation out of an allocation group that
2566 * is part (not the first part) of a larger binary
2567 * buddy system. if we are in the middle, back split
2568 * the system prior to calling dbSplit() which assumes
2569 * that it is at the front of a binary buddy system.
2571 if (oldval == NOFREE) {
2572 rc = dbBackSplit((dmtree_t *) dcp, leafno);
2575 oldval = dcp->stree[ti];
2577 dbSplit((dmtree_t *) dcp, leafno, dcp->budmin, newval);
2579 rc = dbJoin((dmtree_t *) dcp, leafno, newval);
2584 /* check if the root of the current dmap control page changed due
2585 * to the update and if the current dmap control page is not at
2586 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e.
2587 * root changed and this is not the top level), call this routine
2588 * again (recursion) for the next higher level of the mapping to
2589 * reflect the change in root for the current dmap control page.
2591 if (dcp->stree[ROOT] != oldroot) {
2592 /* are we below the top level of the map. if so,
2593 * bubble the root up to the next higher level.
2595 if (level < bmp->db_maxlevel) {
2596 /* bubble up the new root of this dmap control page to
2600 dbAdjCtl(bmp, blkno, dcp->stree[ROOT], alloc,
2602 /* something went wrong in bubbling up the new
2603 * root value, so backout the changes to the
2604 * current dmap control page.
2607 dbJoin((dmtree_t *) dcp, leafno,
2610 /* the dbJoin() above might have
2611 * caused a larger binary buddy system
2612 * to form and we may now be in the
2613 * middle of it. if this is the case,
2614 * back split the buddies.
2616 if (dcp->stree[ti] == NOFREE)
2617 dbBackSplit((dmtree_t *)
2619 dbSplit((dmtree_t *) dcp, leafno,
2620 dcp->budmin, oldval);
2623 /* release the buffer and return the error.
2625 release_metapage(mp);
2629 /* we're at the top level of the map. update
2630 * the bmap control page to reflect the size
2631 * of the maximum free buddy system.
2633 assert(level == bmp->db_maxlevel);
2634 if (bmp->db_maxfreebud != oldroot) {
2635 jfs_error(bmp->db_ipbmap->i_sb,
2636 "the maximum free buddy is not the old root\n");
2638 bmp->db_maxfreebud = dcp->stree[ROOT];
2642 /* write the buffer.
2653 * FUNCTION: update the leaf of a dmtree with a new value, splitting
2654 * the leaf from the binary buddy system of the dmtree's
2655 * leaves, as required.
2658 * tp - pointer to the tree containing the leaf.
2659 * leafno - the number of the leaf to be updated.
2660 * splitsz - the size the binary buddy system starting at the leaf
2661 * must be split to, specified as the log2 number of blocks.
2662 * newval - the new value for the leaf.
2664 * RETURN VALUES: none
2666 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2668 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval)
2672 s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2674 /* check if the leaf needs to be split.
2676 if (leaf[leafno] > tp->dmt_budmin) {
2677 /* the split occurs by cutting the buddy system in half
2678 * at the specified leaf until we reach the specified
2679 * size. pick up the starting split size (current size
2680 * - 1 in l2) and the corresponding buddy size.
2682 cursz = leaf[leafno] - 1;
2683 budsz = BUDSIZE(cursz, tp->dmt_budmin);
2685 /* split until we reach the specified size.
2687 while (cursz >= splitsz) {
2688 /* update the buddy's leaf with its new value.
2690 dbAdjTree(tp, leafno ^ budsz, cursz);
2692 /* on to the next size and buddy.
2699 /* adjust the dmap tree to reflect the specified leaf's new
2702 dbAdjTree(tp, leafno, newval);
2707 * NAME: dbBackSplit()
2709 * FUNCTION: back split the binary buddy system of dmtree leaves
2710 * that hold a specified leaf until the specified leaf
2711 * starts its own binary buddy system.
2713 * the allocators typically perform allocations at the start
2714 * of binary buddy systems and dbSplit() is used to accomplish
2715 * any required splits. in some cases, however, allocation
2716 * may occur in the middle of a binary system and requires a
2717 * back split, with the split proceeding out from the middle of
2718 * the system (less efficient) rather than the start of the
2719 * system (more efficient). the cases in which a back split
2720 * is required are rare and are limited to the first allocation
2721 * within an allocation group which is a part (not first part)
2722 * of a larger binary buddy system and a few exception cases
2723 * in which a previous join operation must be backed out.
2726 * tp - pointer to the tree containing the leaf.
2727 * leafno - the number of the leaf to be updated.
2729 * RETURN VALUES: none
2731 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2733 static int dbBackSplit(dmtree_t * tp, int leafno)
2735 int budsz, bud, w, bsz, size;
2737 s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2739 /* leaf should be part (not first part) of a binary
2742 assert(leaf[leafno] == NOFREE);
2744 /* the back split is accomplished by iteratively finding the leaf
2745 * that starts the buddy system that contains the specified leaf and
2746 * splitting that system in two. this iteration continues until
2747 * the specified leaf becomes the start of a buddy system.
2749 * determine maximum possible l2 size for the specified leaf.
2752 LITOL2BSZ(leafno, le32_to_cpu(tp->dmt_l2nleafs),
2755 /* determine the number of leaves covered by this size. this
2756 * is the buddy size that we will start with as we search for
2757 * the buddy system that contains the specified leaf.
2759 budsz = BUDSIZE(size, tp->dmt_budmin);
2763 while (leaf[leafno] == NOFREE) {
2764 /* find the leftmost buddy leaf.
2766 for (w = leafno, bsz = budsz;; bsz <<= 1,
2767 w = (w < bud) ? w : bud) {
2768 if (bsz >= le32_to_cpu(tp->dmt_nleafs)) {
2769 jfs_err("JFS: block map error in dbBackSplit");
2773 /* determine the buddy.
2777 /* check if this buddy is the start of the system.
2779 if (leaf[bud] != NOFREE) {
2780 /* split the leaf at the start of the
2783 cursz = leaf[bud] - 1;
2784 dbSplit(tp, bud, cursz, cursz);
2790 if (leaf[leafno] != size) {
2791 jfs_err("JFS: wrong leaf value in dbBackSplit");
2801 * FUNCTION: update the leaf of a dmtree with a new value, joining
2802 * the leaf with other leaves of the dmtree into a multi-leaf
2803 * binary buddy system, as required.
2806 * tp - pointer to the tree containing the leaf.
2807 * leafno - the number of the leaf to be updated.
2808 * newval - the new value for the leaf.
2810 * RETURN VALUES: none
2812 static int dbJoin(dmtree_t * tp, int leafno, int newval)
2817 /* can the new leaf value require a join with other leaves ?
2819 if (newval >= tp->dmt_budmin) {
2820 /* pickup a pointer to the leaves of the tree.
2822 leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2824 /* try to join the specified leaf into a large binary
2825 * buddy system. the join proceeds by attempting to join
2826 * the specified leafno with its buddy (leaf) at new value.
2827 * if the join occurs, we attempt to join the left leaf
2828 * of the joined buddies with its buddy at new value + 1.
2829 * we continue to join until we find a buddy that cannot be
2830 * joined (does not have a value equal to the size of the
2831 * last join) or until all leaves have been joined into a
2834 * get the buddy size (number of words covered) of
2837 budsz = BUDSIZE(newval, tp->dmt_budmin);
2841 while (budsz < le32_to_cpu(tp->dmt_nleafs)) {
2842 /* get the buddy leaf.
2844 buddy = leafno ^ budsz;
2846 /* if the leaf's new value is greater than its
2847 * buddy's value, we join no more.
2849 if (newval > leaf[buddy])
2852 /* It shouldn't be less */
2853 if (newval < leaf[buddy])
2856 /* check which (leafno or buddy) is the left buddy.
2857 * the left buddy gets to claim the blocks resulting
2858 * from the join while the right gets to claim none.
2859 * the left buddy is also eligible to participate in
2860 * a join at the next higher level while the right
2864 if (leafno < buddy) {
2865 /* leafno is the left buddy.
2867 dbAdjTree(tp, buddy, NOFREE);
2869 /* buddy is the left buddy and becomes
2872 dbAdjTree(tp, leafno, NOFREE);
2876 /* on to try the next join.
2883 /* update the leaf value.
2885 dbAdjTree(tp, leafno, newval);
2894 * FUNCTION: update a leaf of a dmtree with a new value, adjusting
2895 * the dmtree, as required, to reflect the new leaf value.
2896 * the combination of any buddies must already be done before
2900 * tp - pointer to the tree to be adjusted.
2901 * leafno - the number of the leaf to be updated.
2902 * newval - the new value for the leaf.
2904 * RETURN VALUES: none
2906 static void dbAdjTree(dmtree_t * tp, int leafno, int newval)
2911 /* pick up the index of the leaf for this leafno.
2913 lp = leafno + le32_to_cpu(tp->dmt_leafidx);
2915 /* is the current value the same as the old value ? if so,
2916 * there is nothing to do.
2918 if (tp->dmt_stree[lp] == newval)
2921 /* set the new value.
2923 tp->dmt_stree[lp] = newval;
2925 /* bubble the new value up the tree as required.
2927 for (k = 0; k < le32_to_cpu(tp->dmt_height); k++) {
2928 /* get the index of the first leaf of the 4 leaf
2929 * group containing the specified leaf (leafno).
2931 lp = ((lp - 1) & ~0x03) + 1;
2933 /* get the index of the parent of this 4 leaf group.
2937 /* determine the maximum of the 4 leaves.
2939 max = TREEMAX(&tp->dmt_stree[lp]);
2941 /* if the maximum of the 4 is the same as the
2942 * parent's value, we're done.
2944 if (tp->dmt_stree[pp] == max)
2947 /* parent gets new value.
2949 tp->dmt_stree[pp] = max;
2951 /* parent becomes leaf for next go-round.
2959 * NAME: dbFindLeaf()
2961 * FUNCTION: search a dmtree_t for sufficient free blocks, returning
2962 * the index of a leaf describing the free blocks if
2963 * sufficient free blocks are found.
2965 * the search starts at the top of the dmtree_t tree and
2966 * proceeds down the tree to the leftmost leaf with sufficient
2970 * tp - pointer to the tree to be searched.
2971 * l2nb - log2 number of free blocks to search for.
2972 * leafidx - return pointer to be set to the index of the leaf
2973 * describing at least l2nb free blocks if sufficient
2974 * free blocks are found.
2978 * -ENOSPC - insufficient free blocks.
2980 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx)
2982 int ti, n = 0, k, x = 0;
2984 /* first check the root of the tree to see if there is
2985 * sufficient free space.
2987 if (l2nb > tp->dmt_stree[ROOT])
2990 /* sufficient free space available. now search down the tree
2991 * starting at the next level for the leftmost leaf that
2992 * describes sufficient free space.
2994 for (k = le32_to_cpu(tp->dmt_height), ti = 1;
2995 k > 0; k--, ti = ((ti + n) << 2) + 1) {
2996 /* search the four nodes at this level, starting from
2999 for (x = ti, n = 0; n < 4; n++) {
3000 /* sufficient free space found. move to the next
3001 * level (or quit if this is the last level).
3003 if (l2nb <= tp->dmt_stree[x + n])
3007 /* better have found something since the higher
3008 * levels of the tree said it was here.
3013 /* set the return to the leftmost leaf describing sufficient
3016 *leafidx = x + n - le32_to_cpu(tp->dmt_leafidx);
3023 * NAME: dbFindBits()
3025 * FUNCTION: find a specified number of binary buddy free bits within a
3026 * dmap bitmap word value.
3028 * this routine searches the bitmap value for (1 << l2nb) free
3029 * bits at (1 << l2nb) alignments within the value.
3032 * word - dmap bitmap word value.
3033 * l2nb - number of free bits specified as a log2 number.
3036 * starting bit number of free bits.
3038 static int dbFindBits(u32 word, int l2nb)
3043 /* get the number of bits.
3046 assert(nb <= DBWORD);
3048 /* complement the word so we can use a mask (i.e. 0s represent
3049 * free bits) and compute the mask.
3052 mask = ONES << (DBWORD - nb);
3054 /* scan the word for nb free bits at nb alignments.
3056 for (bitno = 0; mask != 0; bitno += nb, mask >>= nb) {
3057 if ((mask & word) == mask)
3063 /* return the bit number.
3070 * NAME: dbMaxBud(u8 *cp)
3072 * FUNCTION: determine the largest binary buddy string of free
3073 * bits within 32-bits of the map.
3076 * cp - pointer to the 32-bit value.
3079 * largest binary buddy of free bits within a dmap word.
3081 static int dbMaxBud(u8 * cp)
3083 signed char tmp1, tmp2;
3085 /* check if the wmap word is all free. if so, the
3086 * free buddy size is BUDMIN.
3088 if (*((uint *) cp) == 0)
3091 /* check if the wmap word is half free. if so, the
3092 * free buddy size is BUDMIN-1.
3094 if (*((u16 *) cp) == 0 || *((u16 *) cp + 1) == 0)
3095 return (BUDMIN - 1);
3097 /* not all free or half free. determine the free buddy
3098 * size thru table lookup using quarters of the wmap word.
3100 tmp1 = max(budtab[cp[2]], budtab[cp[3]]);
3101 tmp2 = max(budtab[cp[0]], budtab[cp[1]]);
3102 return (max(tmp1, tmp2));
3107 * NAME: cnttz(uint word)
3109 * FUNCTION: determine the number of trailing zeros within a 32-bit
3113 * value - 32-bit value to be examined.
3116 * count of trailing zeros
3118 static int cnttz(u32 word)
3122 for (n = 0; n < 32; n++, word >>= 1) {
3132 * NAME: cntlz(u32 value)
3134 * FUNCTION: determine the number of leading zeros within a 32-bit
3138 * value - 32-bit value to be examined.
3141 * count of leading zeros
3143 static int cntlz(u32 value)
3147 for (n = 0; n < 32; n++, value <<= 1) {
3148 if (value & HIGHORDER)
3156 * NAME: blkstol2(s64 nb)
3158 * FUNCTION: convert a block count to its log2 value. if the block
3159 * count is not a l2 multiple, it is rounded up to the next
3160 * larger l2 multiple.
3163 * nb - number of blocks
3166 * log2 number of blocks
3168 static int blkstol2(s64 nb)
3171 s64 mask; /* meant to be signed */
3173 mask = (s64) 1 << (64 - 1);
3175 /* count the leading bits.
3177 for (l2nb = 0; l2nb < 64; l2nb++, mask >>= 1) {
3178 /* leading bit found.
3181 /* determine the l2 value.
3183 l2nb = (64 - 1) - l2nb;
3185 /* check if we need to round up.
3194 return 0; /* fix compiler warning */
3199 * NAME: dbAllocBottomUp()
3201 * FUNCTION: alloc the specified block range from the working block
3204 * the blocks will be alloc from the working map one dmap
3208 * ip - pointer to in-core inode;
3209 * blkno - starting block number to be freed.
3210 * nblocks - number of blocks to be freed.
3216 int dbAllocBottomUp(struct inode *ip, s64 blkno, s64 nblocks)
3218 struct metapage *mp;
3222 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
3223 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
3225 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
3227 /* block to be allocated better be within the mapsize. */
3228 ASSERT(nblocks <= bmp->db_mapsize - blkno);
3231 * allocate the blocks a dmap at a time.
3234 for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
3235 /* release previous dmap if any */
3240 /* get the buffer for the current dmap. */
3241 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
3242 mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
3244 IREAD_UNLOCK(ipbmap);
3247 dp = (struct dmap *) mp->data;
3249 /* determine the number of blocks to be allocated from
3252 nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
3254 /* allocate the blocks. */
3255 if ((rc = dbAllocDmapBU(bmp, dp, blkno, nb))) {
3256 release_metapage(mp);
3257 IREAD_UNLOCK(ipbmap);
3262 /* write the last buffer. */
3265 IREAD_UNLOCK(ipbmap);
3271 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
3275 int dbitno, word, rembits, nb, nwords, wbitno, agno;
3277 struct dmaptree *tp = (struct dmaptree *) & dp->tree;
3279 /* save the current value of the root (i.e. maximum free string)
3282 oldroot = tp->stree[ROOT];
3284 /* determine the bit number and word within the dmap of the
3287 dbitno = blkno & (BPERDMAP - 1);
3288 word = dbitno >> L2DBWORD;
3290 /* block range better be within the dmap */
3291 assert(dbitno + nblocks <= BPERDMAP);
3293 /* allocate the bits of the dmap's words corresponding to the block
3294 * range. not all bits of the first and last words may be contained
3295 * within the block range. if this is the case, we'll work against
3296 * those words (i.e. partial first and/or last) on an individual basis
3297 * (a single pass), allocating the bits of interest by hand and
3298 * updating the leaf corresponding to the dmap word. a single pass
3299 * will be used for all dmap words fully contained within the
3300 * specified range. within this pass, the bits of all fully contained
3301 * dmap words will be marked as free in a single shot and the leaves
3302 * will be updated. a single leaf may describe the free space of
3303 * multiple dmap words, so we may update only a subset of the actual
3304 * leaves corresponding to the dmap words of the block range.
3306 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
3307 /* determine the bit number within the word and
3308 * the number of bits within the word.
3310 wbitno = dbitno & (DBWORD - 1);
3311 nb = min(rembits, DBWORD - wbitno);
3313 /* check if only part of a word is to be allocated.
3316 /* allocate (set to 1) the appropriate bits within
3319 dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
3324 /* one or more dmap words are fully contained
3325 * within the block range. determine how many
3326 * words and allocate (set to 1) the bits of these
3329 nwords = rembits >> L2DBWORD;
3330 memset(&dp->wmap[word], (int) ONES, nwords * 4);
3332 /* determine how many bits */
3333 nb = nwords << L2DBWORD;
3338 /* update the free count for this dmap */
3339 le32_add_cpu(&dp->nfree, -nblocks);
3341 /* reconstruct summary tree */
3346 /* if this allocation group is completely free,
3347 * update the highest active allocation group number
3348 * if this allocation group is the new max.
3350 agno = blkno >> bmp->db_agl2size;
3351 if (agno > bmp->db_maxag)
3352 bmp->db_maxag = agno;
3354 /* update the free count for the allocation group and map */
3355 bmp->db_agfree[agno] -= nblocks;
3356 bmp->db_nfree -= nblocks;
3360 /* if the root has not changed, done. */
3361 if (tp->stree[ROOT] == oldroot)
3364 /* root changed. bubble the change up to the dmap control pages.
3365 * if the adjustment of the upper level control pages fails,
3366 * backout the bit allocation (thus making everything consistent).
3368 if ((rc = dbAdjCtl(bmp, blkno, tp->stree[ROOT], 1, 0)))
3369 dbFreeBits(bmp, dp, blkno, nblocks);
3376 * NAME: dbExtendFS()
3378 * FUNCTION: extend bmap from blkno for nblocks;
3379 * dbExtendFS() updates bmap ready for dbAllocBottomUp();
3383 * L1---------------------------------L1
3385 * L0---------L0---------L0 L0---------L0---------L0
3387 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm;
3388 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3390 * <---old---><----------------------------extend----------------------->
3392 int dbExtendFS(struct inode *ipbmap, s64 blkno, s64 nblocks)
3394 struct jfs_sb_info *sbi = JFS_SBI(ipbmap->i_sb);
3395 int nbperpage = sbi->nbperpage;
3396 int i, i0 = true, j, j0 = true, k, n;
3399 struct metapage *mp, *l2mp, *l1mp = NULL, *l0mp = NULL;
3400 struct dmapctl *l2dcp, *l1dcp, *l0dcp;
3402 s8 *l0leaf, *l1leaf, *l2leaf;
3403 struct bmap *bmp = sbi->bmap;
3404 int agno, l2agsize, oldl2agsize;
3407 newsize = blkno + nblocks;
3409 jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld",
3410 (long long) blkno, (long long) nblocks, (long long) newsize);
3413 * initialize bmap control page.
3415 * all the data in bmap control page should exclude
3416 * the mkfs hidden dmap page.
3419 /* update mapsize */
3420 bmp->db_mapsize = newsize;
3421 bmp->db_maxlevel = BMAPSZTOLEV(bmp->db_mapsize);
3423 /* compute new AG size */
3424 l2agsize = dbGetL2AGSize(newsize);
3425 oldl2agsize = bmp->db_agl2size;
3427 bmp->db_agl2size = l2agsize;
3428 bmp->db_agsize = 1 << l2agsize;
3430 /* compute new number of AG */
3431 agno = bmp->db_numag;
3432 bmp->db_numag = newsize >> l2agsize;
3433 bmp->db_numag += ((u32) newsize % (u32) bmp->db_agsize) ? 1 : 0;
3436 * reconfigure db_agfree[]
3437 * from old AG configuration to new AG configuration;
3439 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3440 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3441 * note: new AG size = old AG size * (2**x).
3443 if (l2agsize == oldl2agsize)
3445 k = 1 << (l2agsize - oldl2agsize);
3446 ag_rem = bmp->db_agfree[0]; /* save agfree[0] */
3447 for (i = 0, n = 0; i < agno; n++) {
3448 bmp->db_agfree[n] = 0; /* init collection point */
3450 /* coalesce contiguous k AGs; */
3451 for (j = 0; j < k && i < agno; j++, i++) {
3452 /* merge AGi to AGn */
3453 bmp->db_agfree[n] += bmp->db_agfree[i];
3456 bmp->db_agfree[0] += ag_rem; /* restore agfree[0] */
3458 for (; n < MAXAG; n++)
3459 bmp->db_agfree[n] = 0;
3462 * update highest active ag number
3465 bmp->db_maxag = bmp->db_maxag / k;
3470 * update bit maps and corresponding level control pages;
3471 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3475 p = BMAPBLKNO + nbperpage; /* L2 page */
3476 l2mp = read_metapage(ipbmap, p, PSIZE, 0);
3478 jfs_error(ipbmap->i_sb, "L2 page could not be read\n");
3481 l2dcp = (struct dmapctl *) l2mp->data;
3483 /* compute start L1 */
3484 k = blkno >> L2MAXL1SIZE;
3485 l2leaf = l2dcp->stree + CTLLEAFIND + k;
3486 p = BLKTOL1(blkno, sbi->l2nbperpage); /* L1 page */
3489 * extend each L1 in L2
3491 for (; k < LPERCTL; k++, p += nbperpage) {
3494 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3495 l1mp = read_metapage(ipbmap, p, PSIZE, 0);
3498 l1dcp = (struct dmapctl *) l1mp->data;
3500 /* compute start L0 */
3501 j = (blkno & (MAXL1SIZE - 1)) >> L2MAXL0SIZE;
3502 l1leaf = l1dcp->stree + CTLLEAFIND + j;
3503 p = BLKTOL0(blkno, sbi->l2nbperpage);
3506 /* assign/init L1 page */
3507 l1mp = get_metapage(ipbmap, p, PSIZE, 0);
3511 l1dcp = (struct dmapctl *) l1mp->data;
3513 /* compute start L0 */
3515 l1leaf = l1dcp->stree + CTLLEAFIND;
3516 p += nbperpage; /* 1st L0 of L1.k */
3520 * extend each L0 in L1
3522 for (; j < LPERCTL; j++) {
3525 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3527 l0mp = read_metapage(ipbmap, p, PSIZE, 0);
3530 l0dcp = (struct dmapctl *) l0mp->data;
3532 /* compute start dmap */
3533 i = (blkno & (MAXL0SIZE - 1)) >>
3535 l0leaf = l0dcp->stree + CTLLEAFIND + i;
3536 p = BLKTODMAP(blkno,
3540 /* assign/init L0 page */
3541 l0mp = get_metapage(ipbmap, p, PSIZE, 0);
3545 l0dcp = (struct dmapctl *) l0mp->data;
3547 /* compute start dmap */
3549 l0leaf = l0dcp->stree + CTLLEAFIND;
3550 p += nbperpage; /* 1st dmap of L0.j */
3554 * extend each dmap in L0
3556 for (; i < LPERCTL; i++) {
3558 * reconstruct the dmap page, and
3559 * initialize corresponding parent L0 leaf
3561 if ((n = blkno & (BPERDMAP - 1))) {
3562 /* read in dmap page: */
3563 mp = read_metapage(ipbmap, p,
3567 n = min(nblocks, (s64)BPERDMAP - n);
3569 /* assign/init dmap page */
3570 mp = read_metapage(ipbmap, p,
3575 n = min_t(s64, nblocks, BPERDMAP);
3578 dp = (struct dmap *) mp->data;
3579 *l0leaf = dbInitDmap(dp, blkno, n);
3582 agno = le64_to_cpu(dp->start) >> l2agsize;
3583 bmp->db_agfree[agno] += n;
3594 } /* for each dmap in a L0 */
3597 * build current L0 page from its leaves, and
3598 * initialize corresponding parent L1 leaf
3600 *l1leaf = dbInitDmapCtl(l0dcp, 0, ++i);
3601 write_metapage(l0mp);
3605 l1leaf++; /* continue for next L0 */
3607 /* more than 1 L0 ? */
3609 break; /* build L1 page */
3611 /* summarize in global bmap page */
3612 bmp->db_maxfreebud = *l1leaf;
3613 release_metapage(l1mp);
3614 release_metapage(l2mp);
3618 } /* for each L0 in a L1 */
3621 * build current L1 page from its leaves, and
3622 * initialize corresponding parent L2 leaf
3624 *l2leaf = dbInitDmapCtl(l1dcp, 1, ++j);
3625 write_metapage(l1mp);
3629 l2leaf++; /* continue for next L1 */
3631 /* more than 1 L1 ? */
3633 break; /* build L2 page */
3635 /* summarize in global bmap page */
3636 bmp->db_maxfreebud = *l2leaf;
3637 release_metapage(l2mp);
3641 } /* for each L1 in a L2 */
3643 jfs_error(ipbmap->i_sb, "function has not returned as expected\n");
3646 release_metapage(l0mp);
3648 release_metapage(l1mp);
3649 release_metapage(l2mp);
3653 * finalize bmap control page
3664 void dbFinalizeBmap(struct inode *ipbmap)
3666 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
3667 int actags, inactags, l2nl;
3668 s64 ag_rem, actfree, inactfree, avgfree;
3672 * finalize bmap control page
3676 * compute db_agpref: preferred ag to allocate from
3677 * (the leftmost ag with average free space in it);
3680 /* get the number of active ags and inacitve ags */
3681 actags = bmp->db_maxag + 1;
3682 inactags = bmp->db_numag - actags;
3683 ag_rem = bmp->db_mapsize & (bmp->db_agsize - 1); /* ??? */
3685 /* determine how many blocks are in the inactive allocation
3686 * groups. in doing this, we must account for the fact that
3687 * the rightmost group might be a partial group (i.e. file
3688 * system size is not a multiple of the group size).
3690 inactfree = (inactags && ag_rem) ?
3691 ((inactags - 1) << bmp->db_agl2size) + ag_rem
3692 : inactags << bmp->db_agl2size;
3694 /* determine how many free blocks are in the active
3695 * allocation groups plus the average number of free blocks
3696 * within the active ags.
3698 actfree = bmp->db_nfree - inactfree;
3699 avgfree = (u32) actfree / (u32) actags;
3701 /* if the preferred allocation group has not average free space.
3702 * re-establish the preferred group as the leftmost
3703 * group with average free space.
3705 if (bmp->db_agfree[bmp->db_agpref] < avgfree) {
3706 for (bmp->db_agpref = 0; bmp->db_agpref < actags;
3708 if (bmp->db_agfree[bmp->db_agpref] >= avgfree)
3711 if (bmp->db_agpref >= bmp->db_numag) {
3712 jfs_error(ipbmap->i_sb,
3713 "cannot find ag with average freespace\n");
3718 * compute db_aglevel, db_agheight, db_width, db_agstart:
3719 * an ag is covered in aglevel dmapctl summary tree,
3720 * at agheight level height (from leaf) with agwidth number of nodes
3721 * each, which starts at agstart index node of the smmary tree node
3724 bmp->db_aglevel = BMAPSZTOLEV(bmp->db_agsize);
3726 bmp->db_agl2size - (L2BPERDMAP + bmp->db_aglevel * L2LPERCTL);
3727 bmp->db_agheight = l2nl >> 1;
3728 bmp->db_agwidth = 1 << (l2nl - (bmp->db_agheight << 1));
3729 for (i = 5 - bmp->db_agheight, bmp->db_agstart = 0, n = 1; i > 0;
3731 bmp->db_agstart += n;
3739 * NAME: dbInitDmap()/ujfs_idmap_page()
3741 * FUNCTION: initialize working/persistent bitmap of the dmap page
3742 * for the specified number of blocks:
3744 * at entry, the bitmaps had been initialized as free (ZEROS);
3745 * The number of blocks will only account for the actually
3746 * existing blocks. Blocks which don't actually exist in
3747 * the aggregate will be marked as allocated (ONES);
3750 * dp - pointer to page of map
3751 * nblocks - number of blocks this page
3755 static int dbInitDmap(struct dmap * dp, s64 Blkno, int nblocks)
3757 int blkno, w, b, r, nw, nb, i;
3759 /* starting block number within the dmap */
3760 blkno = Blkno & (BPERDMAP - 1);
3763 dp->nblocks = dp->nfree = cpu_to_le32(nblocks);
3764 dp->start = cpu_to_le64(Blkno);
3766 if (nblocks == BPERDMAP) {
3767 memset(&dp->wmap[0], 0, LPERDMAP * 4);
3768 memset(&dp->pmap[0], 0, LPERDMAP * 4);
3772 le32_add_cpu(&dp->nblocks, nblocks);
3773 le32_add_cpu(&dp->nfree, nblocks);
3776 /* word number containing start block number */
3777 w = blkno >> L2DBWORD;
3780 * free the bits corresponding to the block range (ZEROS):
3781 * note: not all bits of the first and last words may be contained
3782 * within the block range.
3784 for (r = nblocks; r > 0; r -= nb, blkno += nb) {
3785 /* number of bits preceding range to be freed in the word */
3786 b = blkno & (DBWORD - 1);
3787 /* number of bits to free in the word */
3788 nb = min(r, DBWORD - b);
3790 /* is partial word to be freed ? */
3792 /* free (set to 0) from the bitmap word */
3793 dp->wmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3795 dp->pmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3798 /* skip the word freed */
3801 /* free (set to 0) contiguous bitmap words */
3803 memset(&dp->wmap[w], 0, nw * 4);
3804 memset(&dp->pmap[w], 0, nw * 4);
3806 /* skip the words freed */
3807 nb = nw << L2DBWORD;
3813 * mark bits following the range to be freed (non-existing
3814 * blocks) as allocated (ONES)
3817 if (blkno == BPERDMAP)
3820 /* the first word beyond the end of existing blocks */
3821 w = blkno >> L2DBWORD;
3823 /* does nblocks fall on a 32-bit boundary ? */
3824 b = blkno & (DBWORD - 1);
3826 /* mark a partial word allocated */
3827 dp->wmap[w] = dp->pmap[w] = cpu_to_le32(ONES >> b);
3831 /* set the rest of the words in the page to allocated (ONES) */
3832 for (i = w; i < LPERDMAP; i++)
3833 dp->pmap[i] = dp->wmap[i] = cpu_to_le32(ONES);
3839 return (dbInitDmapTree(dp));
3844 * NAME: dbInitDmapTree()/ujfs_complete_dmap()
3846 * FUNCTION: initialize summary tree of the specified dmap:
3848 * at entry, bitmap of the dmap has been initialized;
3851 * dp - dmap to complete
3852 * blkno - starting block number for this dmap
3853 * treemax - will be filled in with max free for this dmap
3855 * RETURNS: max free string at the root of the tree
3857 static int dbInitDmapTree(struct dmap * dp)
3859 struct dmaptree *tp;
3863 /* init fixed info of tree */
3865 tp->nleafs = cpu_to_le32(LPERDMAP);
3866 tp->l2nleafs = cpu_to_le32(L2LPERDMAP);
3867 tp->leafidx = cpu_to_le32(LEAFIND);
3868 tp->height = cpu_to_le32(4);
3869 tp->budmin = BUDMIN;
3871 /* init each leaf from corresponding wmap word:
3872 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3873 * bitmap word are allocated.
3875 cp = tp->stree + le32_to_cpu(tp->leafidx);
3876 for (i = 0; i < LPERDMAP; i++)
3877 *cp++ = dbMaxBud((u8 *) & dp->wmap[i]);
3879 /* build the dmap's binary buddy summary tree */
3880 return (dbInitTree(tp));
3885 * NAME: dbInitTree()/ujfs_adjtree()
3887 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl.
3889 * at entry, the leaves of the tree has been initialized
3890 * from corresponding bitmap word or root of summary tree
3891 * of the child control page;
3892 * configure binary buddy system at the leaf level, then
3893 * bubble up the values of the leaf nodes up the tree.
3896 * cp - Pointer to the root of the tree
3897 * l2leaves- Number of leaf nodes as a power of 2
3898 * l2min - Number of blocks that can be covered by a leaf
3901 * RETURNS: max free string at the root of the tree
3903 static int dbInitTree(struct dmaptree * dtp)
3905 int l2max, l2free, bsize, nextb, i;
3906 int child, parent, nparent;
3911 /* Determine the maximum free string possible for the leaves */
3912 l2max = le32_to_cpu(dtp->l2nleafs) + dtp->budmin;
3915 * configure the leaf levevl into binary buddy system
3917 * Try to combine buddies starting with a buddy size of 1
3918 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3919 * can be combined if both buddies have a maximum free of l2min;
3920 * the combination will result in the left-most buddy leaf having
3921 * a maximum free of l2min+1.
3922 * After processing all buddies for a given size, process buddies
3923 * at the next higher buddy size (i.e. current size * 2) and
3924 * the next maximum free (current free + 1).
3925 * This continues until the maximum possible buddy combination
3926 * yields maximum free.
3928 for (l2free = dtp->budmin, bsize = 1; l2free < l2max;
3929 l2free++, bsize = nextb) {
3930 /* get next buddy size == current buddy pair size */
3933 /* scan each adjacent buddy pair at current buddy size */
3934 for (i = 0, cp = tp + le32_to_cpu(dtp->leafidx);
3935 i < le32_to_cpu(dtp->nleafs);
3936 i += nextb, cp += nextb) {
3937 /* coalesce if both adjacent buddies are max free */
3938 if (*cp == l2free && *(cp + bsize) == l2free) {
3939 *cp = l2free + 1; /* left take right */
3940 *(cp + bsize) = -1; /* right give left */
3946 * bubble summary information of leaves up the tree.
3948 * Starting at the leaf node level, the four nodes described by
3949 * the higher level parent node are compared for a maximum free and
3950 * this maximum becomes the value of the parent node.
3951 * when all lower level nodes are processed in this fashion then
3952 * move up to the next level (parent becomes a lower level node) and
3953 * continue the process for that level.
3955 for (child = le32_to_cpu(dtp->leafidx),
3956 nparent = le32_to_cpu(dtp->nleafs) >> 2;
3957 nparent > 0; nparent >>= 2, child = parent) {
3958 /* get index of 1st node of parent level */
3959 parent = (child - 1) >> 2;
3961 /* set the value of the parent node as the maximum
3962 * of the four nodes of the current level.
3964 for (i = 0, cp = tp + child, cp1 = tp + parent;
3965 i < nparent; i++, cp += 4, cp1++)
3976 * function: initialize dmapctl page
3978 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i)
3979 { /* start leaf index not covered by range */
3982 dcp->nleafs = cpu_to_le32(LPERCTL);
3983 dcp->l2nleafs = cpu_to_le32(L2LPERCTL);
3984 dcp->leafidx = cpu_to_le32(CTLLEAFIND);
3985 dcp->height = cpu_to_le32(5);
3986 dcp->budmin = L2BPERDMAP + L2LPERCTL * level;
3989 * initialize the leaves of current level that were not covered
3990 * by the specified input block range (i.e. the leaves have no
3991 * low level dmapctl or dmap).
3993 cp = &dcp->stree[CTLLEAFIND + i];
3994 for (; i < LPERCTL; i++)
3997 /* build the dmap's binary buddy summary tree */
3998 return (dbInitTree((struct dmaptree *) dcp));
4003 * NAME: dbGetL2AGSize()/ujfs_getagl2size()
4005 * FUNCTION: Determine log2(allocation group size) from aggregate size
4008 * nblocks - Number of blocks in aggregate
4010 * RETURNS: log2(allocation group size) in aggregate blocks
4012 static int dbGetL2AGSize(s64 nblocks)
4018 if (nblocks < BPERDMAP * MAXAG)
4019 return (L2BPERDMAP);
4021 /* round up aggregate size to power of 2 */
4022 m = ((u64) 1 << (64 - 1));
4023 for (l2sz = 64; l2sz >= 0; l2sz--, m >>= 1) {
4028 sz = (s64) 1 << l2sz;
4032 /* agsize = roundupSize/max_number_of_ag */
4033 return (l2sz - L2MAXAG);
4038 * NAME: dbMapFileSizeToMapSize()
4040 * FUNCTION: compute number of blocks the block allocation map file
4041 * can cover from the map file size;
4043 * RETURNS: Number of blocks which can be covered by this block map file;
4047 * maximum number of map pages at each level including control pages
4049 #define MAXL0PAGES (1 + LPERCTL)
4050 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES)
4051 #define MAXL2PAGES (1 + LPERCTL * MAXL1PAGES)
4054 * convert number of map pages to the zero origin top dmapctl level
4056 #define BMAPPGTOLEV(npages) \
4057 (((npages) <= 3 + MAXL0PAGES) ? 0 : \
4058 ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
4060 s64 dbMapFileSizeToMapSize(struct inode * ipbmap)
4062 struct super_block *sb = ipbmap->i_sb;
4066 int complete, factor;
4068 nblocks = ipbmap->i_size >> JFS_SBI(sb)->l2bsize;
4069 npages = nblocks >> JFS_SBI(sb)->l2nbperpage;
4070 level = BMAPPGTOLEV(npages);
4072 /* At each level, accumulate the number of dmap pages covered by
4073 * the number of full child levels below it;
4074 * repeat for the last incomplete child level.
4077 npages--; /* skip the first global control page */
4078 /* skip higher level control pages above top level covered by map */
4079 npages -= (2 - level);
4080 npages--; /* skip top level's control page */
4081 for (i = level; i >= 0; i--) {
4083 (i == 2) ? MAXL1PAGES : ((i == 1) ? MAXL0PAGES : 1);
4084 complete = (u32) npages / factor;
4085 ndmaps += complete * ((i == 2) ? LPERCTL * LPERCTL :
4086 ((i == 1) ? LPERCTL : 1));
4088 /* pages in last/incomplete child */
4089 npages = (u32) npages % factor;
4090 /* skip incomplete child's level control page */
4094 /* convert the number of dmaps into the number of blocks
4095 * which can be covered by the dmaps;
4097 nblocks = ndmaps << L2BPERDMAP;