xref: /openbmc/linux/fs/jfs/jfs_dmap.c (revision 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2)
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2004
3  *
4  *   This program is free software;  you can redistribute it and/or modify
5  *   it under the terms of the GNU General Public License as published by
6  *   the Free Software Foundation; either version 2 of the License, or
7  *   (at your option) any later version.
8  *
9  *   This program is distributed in the hope that it will be useful,
10  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
11  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
12  *   the GNU General Public License for more details.
13  *
14  *   You should have received a copy of the GNU General Public License
15  *   along with this program;  if not, write to the Free Software
16  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17  */
18 
19 #include <linux/fs.h>
20 #include "jfs_incore.h"
21 #include "jfs_superblock.h"
22 #include "jfs_dmap.h"
23 #include "jfs_imap.h"
24 #include "jfs_lock.h"
25 #include "jfs_metapage.h"
26 #include "jfs_debug.h"
27 
28 /*
29  *	Debug code for double-checking block map
30  */
31 /* #define	_JFS_DEBUG_DMAP	1 */
32 
33 #ifdef	_JFS_DEBUG_DMAP
34 #define DBINITMAP(size,ipbmap,results) \
35 	DBinitmap(size,ipbmap,results)
36 #define DBALLOC(dbmap,mapsize,blkno,nblocks) \
37 	DBAlloc(dbmap,mapsize,blkno,nblocks)
38 #define DBFREE(dbmap,mapsize,blkno,nblocks) \
39 	DBFree(dbmap,mapsize,blkno,nblocks)
40 #define DBALLOCCK(dbmap,mapsize,blkno,nblocks) \
41 	DBAllocCK(dbmap,mapsize,blkno,nblocks)
42 #define DBFREECK(dbmap,mapsize,blkno,nblocks) \
43 	DBFreeCK(dbmap,mapsize,blkno,nblocks)
44 
45 static void DBinitmap(s64, struct inode *, u32 **);
46 static void DBAlloc(uint *, s64, s64, s64);
47 static void DBFree(uint *, s64, s64, s64);
48 static void DBAllocCK(uint *, s64, s64, s64);
49 static void DBFreeCK(uint *, s64, s64, s64);
50 #else
51 #define DBINITMAP(size,ipbmap,results)
52 #define DBALLOC(dbmap, mapsize, blkno, nblocks)
53 #define DBFREE(dbmap, mapsize, blkno, nblocks)
54 #define DBALLOCCK(dbmap, mapsize, blkno, nblocks)
55 #define DBFREECK(dbmap, mapsize, blkno, nblocks)
56 #endif				/* _JFS_DEBUG_DMAP */
57 
58 /*
59  *	SERIALIZATION of the Block Allocation Map.
60  *
61  *	the working state of the block allocation map is accessed in
62  *	two directions:
63  *
64  *	1) allocation and free requests that start at the dmap
65  *	   level and move up through the dmap control pages (i.e.
66  *	   the vast majority of requests).
67  *
68  * 	2) allocation requests that start at dmap control page
69  *	   level and work down towards the dmaps.
70  *
71  *	the serialization scheme used here is as follows.
72  *
73  *	requests which start at the bottom are serialized against each
74  *	other through buffers and each requests holds onto its buffers
75  *	as it works it way up from a single dmap to the required level
76  *	of dmap control page.
77  *	requests that start at the top are serialized against each other
78  *	and request that start from the bottom by the multiple read/single
79  *	write inode lock of the bmap inode. requests starting at the top
80  *	take this lock in write mode while request starting at the bottom
81  *	take the lock in read mode.  a single top-down request may proceed
82  *	exclusively while multiple bottoms-up requests may proceed
83  * 	simultaneously (under the protection of busy buffers).
84  *
85  *	in addition to information found in dmaps and dmap control pages,
86  *	the working state of the block allocation map also includes read/
87  *	write information maintained in the bmap descriptor (i.e. total
88  *	free block count, allocation group level free block counts).
89  *	a single exclusive lock (BMAP_LOCK) is used to guard this information
90  *	in the face of multiple-bottoms up requests.
91  *	(lock ordering: IREAD_LOCK, BMAP_LOCK);
92  *
93  *	accesses to the persistent state of the block allocation map (limited
94  *	to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
95  */
96 
97 #define BMAP_LOCK_INIT(bmp)	init_MUTEX(&bmp->db_bmaplock)
98 #define BMAP_LOCK(bmp)		down(&bmp->db_bmaplock)
99 #define BMAP_UNLOCK(bmp)	up(&bmp->db_bmaplock)
100 
101 /*
102  * forward references
103  */
104 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
105 			int nblocks);
106 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval);
107 static void dbBackSplit(dmtree_t * tp, int leafno);
108 static void dbJoin(dmtree_t * tp, int leafno, int newval);
109 static void dbAdjTree(dmtree_t * tp, int leafno, int newval);
110 static int dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc,
111 		    int level);
112 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results);
113 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
114 		       int nblocks);
115 static int dbAllocNear(struct bmap * bmp, struct dmap * dp, s64 blkno,
116 		       int nblocks,
117 		       int l2nb, s64 * results);
118 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
119 		       int nblocks);
120 static int dbAllocDmapLev(struct bmap * bmp, struct dmap * dp, int nblocks,
121 			  int l2nb,
122 			  s64 * results);
123 static int dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb,
124 		     s64 * results);
125 static int dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno,
126 		      s64 * results);
127 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks);
128 static int dbFindBits(u32 word, int l2nb);
129 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno);
130 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx);
131 static void dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
132 		       int nblocks);
133 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
134 		      int nblocks);
135 static int dbMaxBud(u8 * cp);
136 s64 dbMapFileSizeToMapSize(struct inode *ipbmap);
137 static int blkstol2(s64 nb);
138 
139 static int cntlz(u32 value);
140 static int cnttz(u32 word);
141 
142 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
143 			 int nblocks);
144 static int dbInitDmap(struct dmap * dp, s64 blkno, int nblocks);
145 static int dbInitDmapTree(struct dmap * dp);
146 static int dbInitTree(struct dmaptree * dtp);
147 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i);
148 static int dbGetL2AGSize(s64 nblocks);
149 
150 /*
151  *	buddy table
152  *
153  * table used for determining buddy sizes within characters of
154  * dmap bitmap words.  the characters themselves serve as indexes
155  * into the table, with the table elements yielding the maximum
156  * binary buddy of free bits within the character.
157  */
158 static s8 budtab[256] = {
159 	3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
160 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
161 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
162 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
163 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
164 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
165 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
166 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
167 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
168 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
169 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
170 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
171 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
172 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
173 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
174 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1
175 };
176 
177 
178 /*
179  * NAME:    	dbMount()
180  *
181  * FUNCTION:	initializate the block allocation map.
182  *
183  *		memory is allocated for the in-core bmap descriptor and
184  *		the in-core descriptor is initialized from disk.
185  *
186  * PARAMETERS:
187  *      ipbmap	-  pointer to in-core inode for the block map.
188  *
189  * RETURN VALUES:
190  *      0	- success
191  *      -ENOMEM	- insufficient memory
192  *      -EIO	- i/o error
193  */
194 int dbMount(struct inode *ipbmap)
195 {
196 	struct bmap *bmp;
197 	struct dbmap_disk *dbmp_le;
198 	struct metapage *mp;
199 	int i;
200 
201 	/*
202 	 * allocate/initialize the in-memory bmap descriptor
203 	 */
204 	/* allocate memory for the in-memory bmap descriptor */
205 	bmp = kmalloc(sizeof(struct bmap), GFP_KERNEL);
206 	if (bmp == NULL)
207 		return -ENOMEM;
208 
209 	/* read the on-disk bmap descriptor. */
210 	mp = read_metapage(ipbmap,
211 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
212 			   PSIZE, 0);
213 	if (mp == NULL) {
214 		kfree(bmp);
215 		return -EIO;
216 	}
217 
218 	/* copy the on-disk bmap descriptor to its in-memory version. */
219 	dbmp_le = (struct dbmap_disk *) mp->data;
220 	bmp->db_mapsize = le64_to_cpu(dbmp_le->dn_mapsize);
221 	bmp->db_nfree = le64_to_cpu(dbmp_le->dn_nfree);
222 	bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage);
223 	bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag);
224 	bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel);
225 	bmp->db_maxag = le32_to_cpu(dbmp_le->dn_maxag);
226 	bmp->db_agpref = le32_to_cpu(dbmp_le->dn_agpref);
227 	bmp->db_aglevel = le32_to_cpu(dbmp_le->dn_aglevel);
228 	bmp->db_agheigth = le32_to_cpu(dbmp_le->dn_agheigth);
229 	bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth);
230 	bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart);
231 	bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size);
232 	for (i = 0; i < MAXAG; i++)
233 		bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
234 	bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
235 	bmp->db_maxfreebud = dbmp_le->dn_maxfreebud;
236 
237 	/* release the buffer. */
238 	release_metapage(mp);
239 
240 	/* bind the bmap inode and the bmap descriptor to each other. */
241 	bmp->db_ipbmap = ipbmap;
242 	JFS_SBI(ipbmap->i_sb)->bmap = bmp;
243 
244 	memset(bmp->db_active, 0, sizeof(bmp->db_active));
245 	DBINITMAP(bmp->db_mapsize, ipbmap, &bmp->db_DBmap);
246 
247 	/*
248 	 * allocate/initialize the bmap lock
249 	 */
250 	BMAP_LOCK_INIT(bmp);
251 
252 	return (0);
253 }
254 
255 
256 /*
257  * NAME:    	dbUnmount()
258  *
259  * FUNCTION:	terminate the block allocation map in preparation for
260  *		file system unmount.
261  *
262  * 		the in-core bmap descriptor is written to disk and
263  *		the memory for this descriptor is freed.
264  *
265  * PARAMETERS:
266  *      ipbmap	-  pointer to in-core inode for the block map.
267  *
268  * RETURN VALUES:
269  *      0	- success
270  *      -EIO	- i/o error
271  */
272 int dbUnmount(struct inode *ipbmap, int mounterror)
273 {
274 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
275 	int i;
276 
277 	if (!(mounterror || isReadOnly(ipbmap)))
278 		dbSync(ipbmap);
279 
280 	/*
281 	 * Invalidate the page cache buffers
282 	 */
283 	truncate_inode_pages(ipbmap->i_mapping, 0);
284 
285 	/*
286 	 * Sanity Check
287 	 */
288 	for (i = 0; i < bmp->db_numag; i++)
289 		if (atomic_read(&bmp->db_active[i]))
290 			printk(KERN_ERR "dbUnmount: db_active[%d] = %d\n",
291 			       i, atomic_read(&bmp->db_active[i]));
292 
293 	/* free the memory for the in-memory bmap. */
294 	kfree(bmp);
295 
296 	return (0);
297 }
298 
299 /*
300  *	dbSync()
301  */
302 int dbSync(struct inode *ipbmap)
303 {
304 	struct dbmap_disk *dbmp_le;
305 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
306 	struct metapage *mp;
307 	int i;
308 
309 	/*
310 	 * write bmap global control page
311 	 */
312 	/* get the buffer for the on-disk bmap descriptor. */
313 	mp = read_metapage(ipbmap,
314 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
315 			   PSIZE, 0);
316 	if (mp == NULL) {
317 		jfs_err("dbSync: read_metapage failed!");
318 		return -EIO;
319 	}
320 	/* copy the in-memory version of the bmap to the on-disk version */
321 	dbmp_le = (struct dbmap_disk *) mp->data;
322 	dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize);
323 	dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree);
324 	dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage);
325 	dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag);
326 	dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel);
327 	dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag);
328 	dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref);
329 	dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel);
330 	dbmp_le->dn_agheigth = cpu_to_le32(bmp->db_agheigth);
331 	dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth);
332 	dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart);
333 	dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size);
334 	for (i = 0; i < MAXAG; i++)
335 		dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]);
336 	dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize);
337 	dbmp_le->dn_maxfreebud = bmp->db_maxfreebud;
338 
339 	/* write the buffer */
340 	write_metapage(mp);
341 
342 	/*
343 	 * write out dirty pages of bmap
344 	 */
345 	filemap_fdatawrite(ipbmap->i_mapping);
346 	filemap_fdatawait(ipbmap->i_mapping);
347 
348 	ipbmap->i_state |= I_DIRTY;
349 	diWriteSpecial(ipbmap, 0);
350 
351 	return (0);
352 }
353 
354 
355 /*
356  * NAME:    	dbFree()
357  *
358  * FUNCTION:	free the specified block range from the working block
359  *		allocation map.
360  *
361  *		the blocks will be free from the working map one dmap
362  *		at a time.
363  *
364  * PARAMETERS:
365  *      ip	-  pointer to in-core inode;
366  *      blkno	-  starting block number to be freed.
367  *      nblocks	-  number of blocks to be freed.
368  *
369  * RETURN VALUES:
370  *      0	- success
371  *      -EIO	- i/o error
372  */
373 int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
374 {
375 	struct metapage *mp;
376 	struct dmap *dp;
377 	int nb, rc;
378 	s64 lblkno, rem;
379 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
380 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
381 
382 	IREAD_LOCK(ipbmap);
383 
384 	/* block to be freed better be within the mapsize. */
385 	if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
386 		IREAD_UNLOCK(ipbmap);
387 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
388 		       (unsigned long long) blkno,
389 		       (unsigned long long) nblocks);
390 		jfs_error(ip->i_sb,
391 			  "dbFree: block to be freed is outside the map");
392 		return -EIO;
393 	}
394 
395 	/*
396 	 * free the blocks a dmap at a time.
397 	 */
398 	mp = NULL;
399 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
400 		/* release previous dmap if any */
401 		if (mp) {
402 			write_metapage(mp);
403 		}
404 
405 		/* get the buffer for the current dmap. */
406 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
407 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
408 		if (mp == NULL) {
409 			IREAD_UNLOCK(ipbmap);
410 			return -EIO;
411 		}
412 		dp = (struct dmap *) mp->data;
413 
414 		/* determine the number of blocks to be freed from
415 		 * this dmap.
416 		 */
417 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
418 
419 		DBALLOCCK(bmp->db_DBmap, bmp->db_mapsize, blkno, nb);
420 
421 		/* free the blocks. */
422 		if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
423 			release_metapage(mp);
424 			IREAD_UNLOCK(ipbmap);
425 			return (rc);
426 		}
427 
428 		DBFREE(bmp->db_DBmap, bmp->db_mapsize, blkno, nb);
429 	}
430 
431 	/* write the last buffer. */
432 	write_metapage(mp);
433 
434 	IREAD_UNLOCK(ipbmap);
435 
436 	return (0);
437 }
438 
439 
440 /*
441  * NAME:	dbUpdatePMap()
442  *
443  * FUNCTION:    update the allocation state (free or allocate) of the
444  *		specified block range in the persistent block allocation map.
445  *
446  *		the blocks will be updated in the persistent map one
447  *		dmap at a time.
448  *
449  * PARAMETERS:
450  *      ipbmap	-  pointer to in-core inode for the block map.
451  *      free	- TRUE if block range is to be freed from the persistent
452  *		  map; FALSE if it is to   be allocated.
453  *      blkno	-  starting block number of the range.
454  *      nblocks	-  number of contiguous blocks in the range.
455  *      tblk	-  transaction block;
456  *
457  * RETURN VALUES:
458  *      0	- success
459  *      -EIO	- i/o error
460  */
461 int
462 dbUpdatePMap(struct inode *ipbmap,
463 	     int free, s64 blkno, s64 nblocks, struct tblock * tblk)
464 {
465 	int nblks, dbitno, wbitno, rbits;
466 	int word, nbits, nwords;
467 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
468 	s64 lblkno, rem, lastlblkno;
469 	u32 mask;
470 	struct dmap *dp;
471 	struct metapage *mp;
472 	struct jfs_log *log;
473 	int lsn, difft, diffp;
474 
475 	/* the blocks better be within the mapsize. */
476 	if (blkno + nblocks > bmp->db_mapsize) {
477 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
478 		       (unsigned long long) blkno,
479 		       (unsigned long long) nblocks);
480 		jfs_error(ipbmap->i_sb,
481 			  "dbUpdatePMap: blocks are outside the map");
482 		return -EIO;
483 	}
484 
485 	/* compute delta of transaction lsn from log syncpt */
486 	lsn = tblk->lsn;
487 	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
488 	logdiff(difft, lsn, log);
489 
490 	/*
491 	 * update the block state a dmap at a time.
492 	 */
493 	mp = NULL;
494 	lastlblkno = 0;
495 	for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) {
496 		/* get the buffer for the current dmap. */
497 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
498 		if (lblkno != lastlblkno) {
499 			if (mp) {
500 				write_metapage(mp);
501 			}
502 
503 			mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE,
504 					   0);
505 			if (mp == NULL)
506 				return -EIO;
507 		}
508 		dp = (struct dmap *) mp->data;
509 
510 		/* determine the bit number and word within the dmap of
511 		 * the starting block.  also determine how many blocks
512 		 * are to be updated within this dmap.
513 		 */
514 		dbitno = blkno & (BPERDMAP - 1);
515 		word = dbitno >> L2DBWORD;
516 		nblks = min(rem, (s64)BPERDMAP - dbitno);
517 
518 		/* update the bits of the dmap words. the first and last
519 		 * words may only have a subset of their bits updated. if
520 		 * this is the case, we'll work against that word (i.e.
521 		 * partial first and/or last) only in a single pass.  a
522 		 * single pass will also be used to update all words that
523 		 * are to have all their bits updated.
524 		 */
525 		for (rbits = nblks; rbits > 0;
526 		     rbits -= nbits, dbitno += nbits) {
527 			/* determine the bit number within the word and
528 			 * the number of bits within the word.
529 			 */
530 			wbitno = dbitno & (DBWORD - 1);
531 			nbits = min(rbits, DBWORD - wbitno);
532 
533 			/* check if only part of the word is to be updated. */
534 			if (nbits < DBWORD) {
535 				/* update (free or allocate) the bits
536 				 * in this word.
537 				 */
538 				mask =
539 				    (ONES << (DBWORD - nbits) >> wbitno);
540 				if (free)
541 					dp->pmap[word] &=
542 					    cpu_to_le32(~mask);
543 				else
544 					dp->pmap[word] |=
545 					    cpu_to_le32(mask);
546 
547 				word += 1;
548 			} else {
549 				/* one or more words are to have all
550 				 * their bits updated.  determine how
551 				 * many words and how many bits.
552 				 */
553 				nwords = rbits >> L2DBWORD;
554 				nbits = nwords << L2DBWORD;
555 
556 				/* update (free or allocate) the bits
557 				 * in these words.
558 				 */
559 				if (free)
560 					memset(&dp->pmap[word], 0,
561 					       nwords * 4);
562 				else
563 					memset(&dp->pmap[word], (int) ONES,
564 					       nwords * 4);
565 
566 				word += nwords;
567 			}
568 		}
569 
570 		/*
571 		 * update dmap lsn
572 		 */
573 		if (lblkno == lastlblkno)
574 			continue;
575 
576 		lastlblkno = lblkno;
577 
578 		if (mp->lsn != 0) {
579 			/* inherit older/smaller lsn */
580 			logdiff(diffp, mp->lsn, log);
581 			if (difft < diffp) {
582 				mp->lsn = lsn;
583 
584 				/* move bp after tblock in logsync list */
585 				LOGSYNC_LOCK(log);
586 				list_move(&mp->synclist, &tblk->synclist);
587 				LOGSYNC_UNLOCK(log);
588 			}
589 
590 			/* inherit younger/larger clsn */
591 			LOGSYNC_LOCK(log);
592 			logdiff(difft, tblk->clsn, log);
593 			logdiff(diffp, mp->clsn, log);
594 			if (difft > diffp)
595 				mp->clsn = tblk->clsn;
596 			LOGSYNC_UNLOCK(log);
597 		} else {
598 			mp->log = log;
599 			mp->lsn = lsn;
600 
601 			/* insert bp after tblock in logsync list */
602 			LOGSYNC_LOCK(log);
603 
604 			log->count++;
605 			list_add(&mp->synclist, &tblk->synclist);
606 
607 			mp->clsn = tblk->clsn;
608 			LOGSYNC_UNLOCK(log);
609 		}
610 	}
611 
612 	/* write the last buffer. */
613 	if (mp) {
614 		write_metapage(mp);
615 	}
616 
617 	return (0);
618 }
619 
620 
621 /*
622  * NAME:	dbNextAG()
623  *
624  * FUNCTION:    find the preferred allocation group for new allocations.
625  *
626  *		Within the allocation groups, we maintain a preferred
627  *		allocation group which consists of a group with at least
628  *		average free space.  It is the preferred group that we target
629  *		new inode allocation towards.  The tie-in between inode
630  *		allocation and block allocation occurs as we allocate the
631  *		first (data) block of an inode and specify the inode (block)
632  *		as the allocation hint for this block.
633  *
634  *		We try to avoid having more than one open file growing in
635  *		an allocation group, as this will lead to fragmentation.
636  *		This differs from the old OS/2 method of trying to keep
637  *		empty ags around for large allocations.
638  *
639  * PARAMETERS:
640  *      ipbmap	-  pointer to in-core inode for the block map.
641  *
642  * RETURN VALUES:
643  *      the preferred allocation group number.
644  */
645 int dbNextAG(struct inode *ipbmap)
646 {
647 	s64 avgfree;
648 	int agpref;
649 	s64 hwm = 0;
650 	int i;
651 	int next_best = -1;
652 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
653 
654 	BMAP_LOCK(bmp);
655 
656 	/* determine the average number of free blocks within the ags. */
657 	avgfree = (u32)bmp->db_nfree / bmp->db_numag;
658 
659 	/*
660 	 * if the current preferred ag does not have an active allocator
661 	 * and has at least average freespace, return it
662 	 */
663 	agpref = bmp->db_agpref;
664 	if ((atomic_read(&bmp->db_active[agpref]) == 0) &&
665 	    (bmp->db_agfree[agpref] >= avgfree))
666 		goto unlock;
667 
668 	/* From the last preferred ag, find the next one with at least
669 	 * average free space.
670 	 */
671 	for (i = 0 ; i < bmp->db_numag; i++, agpref++) {
672 		if (agpref == bmp->db_numag)
673 			agpref = 0;
674 
675 		if (atomic_read(&bmp->db_active[agpref]))
676 			/* open file is currently growing in this ag */
677 			continue;
678 		if (bmp->db_agfree[agpref] >= avgfree) {
679 			/* Return this one */
680 			bmp->db_agpref = agpref;
681 			goto unlock;
682 		} else if (bmp->db_agfree[agpref] > hwm) {
683 			/* Less than avg. freespace, but best so far */
684 			hwm = bmp->db_agfree[agpref];
685 			next_best = agpref;
686 		}
687 	}
688 
689 	/*
690 	 * If no inactive ag was found with average freespace, use the
691 	 * next best
692 	 */
693 	if (next_best != -1)
694 		bmp->db_agpref = next_best;
695 	/* else leave db_agpref unchanged */
696 unlock:
697 	BMAP_UNLOCK(bmp);
698 
699 	/* return the preferred group.
700 	 */
701 	return (bmp->db_agpref);
702 }
703 
704 /*
705  * NAME:	dbAlloc()
706  *
707  * FUNCTION:    attempt to allocate a specified number of contiguous free
708  *		blocks from the working allocation block map.
709  *
710  *		the block allocation policy uses hints and a multi-step
711  *		approach.
712  *
713  *	  	for allocation requests smaller than the number of blocks
714  *		per dmap, we first try to allocate the new blocks
715  *		immediately following the hint.  if these blocks are not
716  *		available, we try to allocate blocks near the hint.  if
717  *		no blocks near the hint are available, we next try to
718  *		allocate within the same dmap as contains the hint.
719  *
720  *		if no blocks are available in the dmap or the allocation
721  *		request is larger than the dmap size, we try to allocate
722  *		within the same allocation group as contains the hint. if
723  *		this does not succeed, we finally try to allocate anywhere
724  *		within the aggregate.
725  *
726  *		we also try to allocate anywhere within the aggregate for
727  *		for allocation requests larger than the allocation group
728  *		size or requests that specify no hint value.
729  *
730  * PARAMETERS:
731  *      ip	-  pointer to in-core inode;
732  *      hint	- allocation hint.
733  *      nblocks	- number of contiguous blocks in the range.
734  *      results	- on successful return, set to the starting block number
735  *		  of the newly allocated contiguous range.
736  *
737  * RETURN VALUES:
738  *      0	- success
739  *      -ENOSPC	- insufficient disk resources
740  *      -EIO	- i/o error
741  */
742 int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results)
743 {
744 	int rc, agno;
745 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
746 	struct bmap *bmp;
747 	struct metapage *mp;
748 	s64 lblkno, blkno;
749 	struct dmap *dp;
750 	int l2nb;
751 	s64 mapSize;
752 	int writers;
753 
754 	/* assert that nblocks is valid */
755 	assert(nblocks > 0);
756 
757 #ifdef _STILL_TO_PORT
758 	/* DASD limit check                                     F226941 */
759 	if (OVER_LIMIT(ip, nblocks))
760 		return -ENOSPC;
761 #endif				/* _STILL_TO_PORT */
762 
763 	/* get the log2 number of blocks to be allocated.
764 	 * if the number of blocks is not a log2 multiple,
765 	 * it will be rounded up to the next log2 multiple.
766 	 */
767 	l2nb = BLKSTOL2(nblocks);
768 
769 	bmp = JFS_SBI(ip->i_sb)->bmap;
770 
771 //retry:        /* serialize w.r.t.extendfs() */
772 	mapSize = bmp->db_mapsize;
773 
774 	/* the hint should be within the map */
775 	if (hint >= mapSize) {
776 		jfs_error(ip->i_sb, "dbAlloc: the hint is outside the map");
777 		return -EIO;
778 	}
779 
780 	/* if the number of blocks to be allocated is greater than the
781 	 * allocation group size, try to allocate anywhere.
782 	 */
783 	if (l2nb > bmp->db_agl2size) {
784 		IWRITE_LOCK(ipbmap);
785 
786 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
787 		if (rc == 0) {
788 			DBALLOC(bmp->db_DBmap, bmp->db_mapsize, *results,
789 				nblocks);
790 		}
791 
792 		goto write_unlock;
793 	}
794 
795 	/*
796 	 * If no hint, let dbNextAG recommend an allocation group
797 	 */
798 	if (hint == 0)
799 		goto pref_ag;
800 
801 	/* we would like to allocate close to the hint.  adjust the
802 	 * hint to the block following the hint since the allocators
803 	 * will start looking for free space starting at this point.
804 	 */
805 	blkno = hint + 1;
806 
807 	if (blkno >= bmp->db_mapsize)
808 		goto pref_ag;
809 
810 	agno = blkno >> bmp->db_agl2size;
811 
812 	/* check if blkno crosses over into a new allocation group.
813 	 * if so, check if we should allow allocations within this
814 	 * allocation group.
815 	 */
816 	if ((blkno & (bmp->db_agsize - 1)) == 0)
817 		/* check if the AG is currenly being written to.
818 		 * if so, call dbNextAG() to find a non-busy
819 		 * AG with sufficient free space.
820 		 */
821 		if (atomic_read(&bmp->db_active[agno]))
822 			goto pref_ag;
823 
824 	/* check if the allocation request size can be satisfied from a
825 	 * single dmap.  if so, try to allocate from the dmap containing
826 	 * the hint using a tiered strategy.
827 	 */
828 	if (nblocks <= BPERDMAP) {
829 		IREAD_LOCK(ipbmap);
830 
831 		/* get the buffer for the dmap containing the hint.
832 		 */
833 		rc = -EIO;
834 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
835 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
836 		if (mp == NULL)
837 			goto read_unlock;
838 
839 		dp = (struct dmap *) mp->data;
840 
841 		/* first, try to satisfy the allocation request with the
842 		 * blocks beginning at the hint.
843 		 */
844 		if ((rc = dbAllocNext(bmp, dp, blkno, (int) nblocks))
845 		    != -ENOSPC) {
846 			if (rc == 0) {
847 				*results = blkno;
848 				DBALLOC(bmp->db_DBmap, bmp->db_mapsize,
849 					*results, nblocks);
850 				mark_metapage_dirty(mp);
851 			}
852 
853 			release_metapage(mp);
854 			goto read_unlock;
855 		}
856 
857 		writers = atomic_read(&bmp->db_active[agno]);
858 		if ((writers > 1) ||
859 		    ((writers == 1) && (JFS_IP(ip)->active_ag != agno))) {
860 			/*
861 			 * Someone else is writing in this allocation
862 			 * group.  To avoid fragmenting, try another ag
863 			 */
864 			release_metapage(mp);
865 			IREAD_UNLOCK(ipbmap);
866 			goto pref_ag;
867 		}
868 
869 		/* next, try to satisfy the allocation request with blocks
870 		 * near the hint.
871 		 */
872 		if ((rc =
873 		     dbAllocNear(bmp, dp, blkno, (int) nblocks, l2nb, results))
874 		    != -ENOSPC) {
875 			if (rc == 0) {
876 				DBALLOC(bmp->db_DBmap, bmp->db_mapsize,
877 					*results, nblocks);
878 				mark_metapage_dirty(mp);
879 			}
880 
881 			release_metapage(mp);
882 			goto read_unlock;
883 		}
884 
885 		/* try to satisfy the allocation request with blocks within
886 		 * the same dmap as the hint.
887 		 */
888 		if ((rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results))
889 		    != -ENOSPC) {
890 			if (rc == 0) {
891 				DBALLOC(bmp->db_DBmap, bmp->db_mapsize,
892 					*results, nblocks);
893 				mark_metapage_dirty(mp);
894 			}
895 
896 			release_metapage(mp);
897 			goto read_unlock;
898 		}
899 
900 		release_metapage(mp);
901 		IREAD_UNLOCK(ipbmap);
902 	}
903 
904 	/* try to satisfy the allocation request with blocks within
905 	 * the same allocation group as the hint.
906 	 */
907 	IWRITE_LOCK(ipbmap);
908 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results))
909 	    != -ENOSPC) {
910 		if (rc == 0)
911 			DBALLOC(bmp->db_DBmap, bmp->db_mapsize,
912 				*results, nblocks);
913 		goto write_unlock;
914 	}
915 	IWRITE_UNLOCK(ipbmap);
916 
917 
918       pref_ag:
919 	/*
920 	 * Let dbNextAG recommend a preferred allocation group
921 	 */
922 	agno = dbNextAG(ipbmap);
923 	IWRITE_LOCK(ipbmap);
924 
925 	/* Try to allocate within this allocation group.  if that fails, try to
926 	 * allocate anywhere in the map.
927 	 */
928 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) == -ENOSPC)
929 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
930 	if (rc == 0) {
931 		DBALLOC(bmp->db_DBmap, bmp->db_mapsize, *results, nblocks);
932 	}
933 
934       write_unlock:
935 	IWRITE_UNLOCK(ipbmap);
936 
937 	return (rc);
938 
939       read_unlock:
940 	IREAD_UNLOCK(ipbmap);
941 
942 	return (rc);
943 }
944 
945 #ifdef _NOTYET
946 /*
947  * NAME:	dbAllocExact()
948  *
949  * FUNCTION:    try to allocate the requested extent;
950  *
951  * PARAMETERS:
952  *      ip	- pointer to in-core inode;
953  *      blkno	- extent address;
954  *      nblocks	- extent length;
955  *
956  * RETURN VALUES:
957  *      0	- success
958  *      -ENOSPC	- insufficient disk resources
959  *      -EIO	- i/o error
960  */
961 int dbAllocExact(struct inode *ip, s64 blkno, int nblocks)
962 {
963 	int rc;
964 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
965 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
966 	struct dmap *dp;
967 	s64 lblkno;
968 	struct metapage *mp;
969 
970 	IREAD_LOCK(ipbmap);
971 
972 	/*
973 	 * validate extent request:
974 	 *
975 	 * note: defragfs policy:
976 	 *  max 64 blocks will be moved.
977 	 *  allocation request size must be satisfied from a single dmap.
978 	 */
979 	if (nblocks <= 0 || nblocks > BPERDMAP || blkno >= bmp->db_mapsize) {
980 		IREAD_UNLOCK(ipbmap);
981 		return -EINVAL;
982 	}
983 
984 	if (nblocks > ((s64) 1 << bmp->db_maxfreebud)) {
985 		/* the free space is no longer available */
986 		IREAD_UNLOCK(ipbmap);
987 		return -ENOSPC;
988 	}
989 
990 	/* read in the dmap covering the extent */
991 	lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
992 	mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
993 	if (mp == NULL) {
994 		IREAD_UNLOCK(ipbmap);
995 		return -EIO;
996 	}
997 	dp = (struct dmap *) mp->data;
998 
999 	/* try to allocate the requested extent */
1000 	rc = dbAllocNext(bmp, dp, blkno, nblocks);
1001 
1002 	IREAD_UNLOCK(ipbmap);
1003 
1004 	if (rc == 0) {
1005 		DBALLOC(bmp->db_DBmap, bmp->db_mapsize, blkno, nblocks);
1006 		mark_metapage_dirty(mp);
1007 	}
1008 	release_metapage(mp);
1009 
1010 	return (rc);
1011 }
1012 #endif /* _NOTYET */
1013 
1014 /*
1015  * NAME:	dbReAlloc()
1016  *
1017  * FUNCTION:    attempt to extend a current allocation by a specified
1018  *		number of blocks.
1019  *
1020  *		this routine attempts to satisfy the allocation request
1021  *		by first trying to extend the existing allocation in
1022  *		place by allocating the additional blocks as the blocks
1023  *		immediately following the current allocation.  if these
1024  *		blocks are not available, this routine will attempt to
1025  *		allocate a new set of contiguous blocks large enough
1026  *		to cover the existing allocation plus the additional
1027  *		number of blocks required.
1028  *
1029  * PARAMETERS:
1030  *      ip	    -  pointer to in-core inode requiring allocation.
1031  *      blkno	    -  starting block of the current allocation.
1032  *      nblocks	    -  number of contiguous blocks within the current
1033  *		       allocation.
1034  *      addnblocks  -  number of blocks to add to the allocation.
1035  *      results	-      on successful return, set to the starting block number
1036  *		       of the existing allocation if the existing allocation
1037  *		       was extended in place or to a newly allocated contiguous
1038  *		       range if the existing allocation could not be extended
1039  *		       in place.
1040  *
1041  * RETURN VALUES:
1042  *      0	- success
1043  *      -ENOSPC	- insufficient disk resources
1044  *      -EIO	- i/o error
1045  */
1046 int
1047 dbReAlloc(struct inode *ip,
1048 	  s64 blkno, s64 nblocks, s64 addnblocks, s64 * results)
1049 {
1050 	int rc;
1051 
1052 	/* try to extend the allocation in place.
1053 	 */
1054 	if ((rc = dbExtend(ip, blkno, nblocks, addnblocks)) == 0) {
1055 		*results = blkno;
1056 		return (0);
1057 	} else {
1058 		if (rc != -ENOSPC)
1059 			return (rc);
1060 	}
1061 
1062 	/* could not extend the allocation in place, so allocate a
1063 	 * new set of blocks for the entire request (i.e. try to get
1064 	 * a range of contiguous blocks large enough to cover the
1065 	 * existing allocation plus the additional blocks.)
1066 	 */
1067 	return (dbAlloc
1068 		(ip, blkno + nblocks - 1, addnblocks + nblocks, results));
1069 }
1070 
1071 
1072 /*
1073  * NAME:	dbExtend()
1074  *
1075  * FUNCTION:    attempt to extend a current allocation by a specified
1076  *		number of blocks.
1077  *
1078  *		this routine attempts to satisfy the allocation request
1079  *		by first trying to extend the existing allocation in
1080  *		place by allocating the additional blocks as the blocks
1081  *		immediately following the current allocation.
1082  *
1083  * PARAMETERS:
1084  *      ip	    -  pointer to in-core inode requiring allocation.
1085  *      blkno	    -  starting block of the current allocation.
1086  *      nblocks	    -  number of contiguous blocks within the current
1087  *		       allocation.
1088  *      addnblocks  -  number of blocks to add to the allocation.
1089  *
1090  * RETURN VALUES:
1091  *      0	- success
1092  *      -ENOSPC	- insufficient disk resources
1093  *      -EIO	- i/o error
1094  */
1095 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks)
1096 {
1097 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
1098 	s64 lblkno, lastblkno, extblkno;
1099 	uint rel_block;
1100 	struct metapage *mp;
1101 	struct dmap *dp;
1102 	int rc;
1103 	struct inode *ipbmap = sbi->ipbmap;
1104 	struct bmap *bmp;
1105 
1106 	/*
1107 	 * We don't want a non-aligned extent to cross a page boundary
1108 	 */
1109 	if (((rel_block = blkno & (sbi->nbperpage - 1))) &&
1110 	    (rel_block + nblocks + addnblocks > sbi->nbperpage))
1111 		return -ENOSPC;
1112 
1113 	/* get the last block of the current allocation */
1114 	lastblkno = blkno + nblocks - 1;
1115 
1116 	/* determine the block number of the block following
1117 	 * the existing allocation.
1118 	 */
1119 	extblkno = lastblkno + 1;
1120 
1121 	IREAD_LOCK(ipbmap);
1122 
1123 	/* better be within the file system */
1124 	bmp = sbi->bmap;
1125 	if (lastblkno < 0 || lastblkno >= bmp->db_mapsize) {
1126 		IREAD_UNLOCK(ipbmap);
1127 		jfs_error(ip->i_sb,
1128 			  "dbExtend: the block is outside the filesystem");
1129 		return -EIO;
1130 	}
1131 
1132 	/* we'll attempt to extend the current allocation in place by
1133 	 * allocating the additional blocks as the blocks immediately
1134 	 * following the current allocation.  we only try to extend the
1135 	 * current allocation in place if the number of additional blocks
1136 	 * can fit into a dmap, the last block of the current allocation
1137 	 * is not the last block of the file system, and the start of the
1138 	 * inplace extension is not on an allocation group boundary.
1139 	 */
1140 	if (addnblocks > BPERDMAP || extblkno >= bmp->db_mapsize ||
1141 	    (extblkno & (bmp->db_agsize - 1)) == 0) {
1142 		IREAD_UNLOCK(ipbmap);
1143 		return -ENOSPC;
1144 	}
1145 
1146 	/* get the buffer for the dmap containing the first block
1147 	 * of the extension.
1148 	 */
1149 	lblkno = BLKTODMAP(extblkno, bmp->db_l2nbperpage);
1150 	mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
1151 	if (mp == NULL) {
1152 		IREAD_UNLOCK(ipbmap);
1153 		return -EIO;
1154 	}
1155 
1156 	DBALLOCCK(bmp->db_DBmap, bmp->db_mapsize, blkno, nblocks);
1157 	dp = (struct dmap *) mp->data;
1158 
1159 	/* try to allocate the blocks immediately following the
1160 	 * current allocation.
1161 	 */
1162 	rc = dbAllocNext(bmp, dp, extblkno, (int) addnblocks);
1163 
1164 	IREAD_UNLOCK(ipbmap);
1165 
1166 	/* were we successful ? */
1167 	if (rc == 0) {
1168 		DBALLOC(bmp->db_DBmap, bmp->db_mapsize, extblkno,
1169 			addnblocks);
1170 		write_metapage(mp);
1171 	} else
1172 		/* we were not successful */
1173 		release_metapage(mp);
1174 
1175 
1176 	return (rc);
1177 }
1178 
1179 
1180 /*
1181  * NAME:	dbAllocNext()
1182  *
1183  * FUNCTION:    attempt to allocate the blocks of the specified block
1184  *		range within a dmap.
1185  *
1186  * PARAMETERS:
1187  *      bmp	-  pointer to bmap descriptor
1188  *      dp	-  pointer to dmap.
1189  *      blkno	-  starting block number of the range.
1190  *      nblocks	-  number of contiguous free blocks of the range.
1191  *
1192  * RETURN VALUES:
1193  *      0	- success
1194  *      -ENOSPC	- insufficient disk resources
1195  *      -EIO	- i/o error
1196  *
1197  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1198  */
1199 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
1200 		       int nblocks)
1201 {
1202 	int dbitno, word, rembits, nb, nwords, wbitno, nw;
1203 	int l2size;
1204 	s8 *leaf;
1205 	u32 mask;
1206 
1207 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1208 		jfs_error(bmp->db_ipbmap->i_sb,
1209 			  "dbAllocNext: Corrupt dmap page");
1210 		return -EIO;
1211 	}
1212 
1213 	/* pick up a pointer to the leaves of the dmap tree.
1214 	 */
1215 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1216 
1217 	/* determine the bit number and word within the dmap of the
1218 	 * starting block.
1219 	 */
1220 	dbitno = blkno & (BPERDMAP - 1);
1221 	word = dbitno >> L2DBWORD;
1222 
1223 	/* check if the specified block range is contained within
1224 	 * this dmap.
1225 	 */
1226 	if (dbitno + nblocks > BPERDMAP)
1227 		return -ENOSPC;
1228 
1229 	/* check if the starting leaf indicates that anything
1230 	 * is free.
1231 	 */
1232 	if (leaf[word] == NOFREE)
1233 		return -ENOSPC;
1234 
1235 	/* check the dmaps words corresponding to block range to see
1236 	 * if the block range is free.  not all bits of the first and
1237 	 * last words may be contained within the block range.  if this
1238 	 * is the case, we'll work against those words (i.e. partial first
1239 	 * and/or last) on an individual basis (a single pass) and examine
1240 	 * the actual bits to determine if they are free.  a single pass
1241 	 * will be used for all dmap words fully contained within the
1242 	 * specified range.  within this pass, the leaves of the dmap
1243 	 * tree will be examined to determine if the blocks are free. a
1244 	 * single leaf may describe the free space of multiple dmap
1245 	 * words, so we may visit only a subset of the actual leaves
1246 	 * corresponding to the dmap words of the block range.
1247 	 */
1248 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
1249 		/* determine the bit number within the word and
1250 		 * the number of bits within the word.
1251 		 */
1252 		wbitno = dbitno & (DBWORD - 1);
1253 		nb = min(rembits, DBWORD - wbitno);
1254 
1255 		/* check if only part of the word is to be examined.
1256 		 */
1257 		if (nb < DBWORD) {
1258 			/* check if the bits are free.
1259 			 */
1260 			mask = (ONES << (DBWORD - nb) >> wbitno);
1261 			if ((mask & ~le32_to_cpu(dp->wmap[word])) != mask)
1262 				return -ENOSPC;
1263 
1264 			word += 1;
1265 		} else {
1266 			/* one or more dmap words are fully contained
1267 			 * within the block range.  determine how many
1268 			 * words and how many bits.
1269 			 */
1270 			nwords = rembits >> L2DBWORD;
1271 			nb = nwords << L2DBWORD;
1272 
1273 			/* now examine the appropriate leaves to determine
1274 			 * if the blocks are free.
1275 			 */
1276 			while (nwords > 0) {
1277 				/* does the leaf describe any free space ?
1278 				 */
1279 				if (leaf[word] < BUDMIN)
1280 					return -ENOSPC;
1281 
1282 				/* determine the l2 number of bits provided
1283 				 * by this leaf.
1284 				 */
1285 				l2size =
1286 				    min((int)leaf[word], NLSTOL2BSZ(nwords));
1287 
1288 				/* determine how many words were handled.
1289 				 */
1290 				nw = BUDSIZE(l2size, BUDMIN);
1291 
1292 				nwords -= nw;
1293 				word += nw;
1294 			}
1295 		}
1296 	}
1297 
1298 	/* allocate the blocks.
1299 	 */
1300 	return (dbAllocDmap(bmp, dp, blkno, nblocks));
1301 }
1302 
1303 
1304 /*
1305  * NAME:	dbAllocNear()
1306  *
1307  * FUNCTION:    attempt to allocate a number of contiguous free blocks near
1308  *		a specified block (hint) within a dmap.
1309  *
1310  *		starting with the dmap leaf that covers the hint, we'll
1311  *		check the next four contiguous leaves for sufficient free
1312  *		space.  if sufficient free space is found, we'll allocate
1313  *		the desired free space.
1314  *
1315  * PARAMETERS:
1316  *      bmp	-  pointer to bmap descriptor
1317  *      dp	-  pointer to dmap.
1318  *      blkno	-  block number to allocate near.
1319  *      nblocks	-  actual number of contiguous free blocks desired.
1320  *      l2nb	-  log2 number of contiguous free blocks desired.
1321  *      results	-  on successful return, set to the starting block number
1322  *		   of the newly allocated range.
1323  *
1324  * RETURN VALUES:
1325  *      0	- success
1326  *      -ENOSPC	- insufficient disk resources
1327  *      -EIO	- i/o error
1328  *
1329  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1330  */
1331 static int
1332 dbAllocNear(struct bmap * bmp,
1333 	    struct dmap * dp, s64 blkno, int nblocks, int l2nb, s64 * results)
1334 {
1335 	int word, lword, rc;
1336 	s8 *leaf;
1337 
1338 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1339 		jfs_error(bmp->db_ipbmap->i_sb,
1340 			  "dbAllocNear: Corrupt dmap page");
1341 		return -EIO;
1342 	}
1343 
1344 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1345 
1346 	/* determine the word within the dmap that holds the hint
1347 	 * (i.e. blkno).  also, determine the last word in the dmap
1348 	 * that we'll include in our examination.
1349 	 */
1350 	word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
1351 	lword = min(word + 4, LPERDMAP);
1352 
1353 	/* examine the leaves for sufficient free space.
1354 	 */
1355 	for (; word < lword; word++) {
1356 		/* does the leaf describe sufficient free space ?
1357 		 */
1358 		if (leaf[word] < l2nb)
1359 			continue;
1360 
1361 		/* determine the block number within the file system
1362 		 * of the first block described by this dmap word.
1363 		 */
1364 		blkno = le64_to_cpu(dp->start) + (word << L2DBWORD);
1365 
1366 		/* if not all bits of the dmap word are free, get the
1367 		 * starting bit number within the dmap word of the required
1368 		 * string of free bits and adjust the block number with the
1369 		 * value.
1370 		 */
1371 		if (leaf[word] < BUDMIN)
1372 			blkno +=
1373 			    dbFindBits(le32_to_cpu(dp->wmap[word]), l2nb);
1374 
1375 		/* allocate the blocks.
1376 		 */
1377 		if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1378 			*results = blkno;
1379 
1380 		return (rc);
1381 	}
1382 
1383 	return -ENOSPC;
1384 }
1385 
1386 
1387 /*
1388  * NAME:	dbAllocAG()
1389  *
1390  * FUNCTION:    attempt to allocate the specified number of contiguous
1391  *		free blocks within the specified allocation group.
1392  *
1393  *		unless the allocation group size is equal to the number
1394  *		of blocks per dmap, the dmap control pages will be used to
1395  *		find the required free space, if available.  we start the
1396  *		search at the highest dmap control page level which
1397  *		distinctly describes the allocation group's free space
1398  *		(i.e. the highest level at which the allocation group's
1399  *		free space is not mixed in with that of any other group).
1400  *		in addition, we start the search within this level at a
1401  *		height of the dmapctl dmtree at which the nodes distinctly
1402  *		describe the allocation group's free space.  at this height,
1403  *		the allocation group's free space may be represented by 1
1404  *		or two sub-trees, depending on the allocation group size.
1405  *		we search the top nodes of these subtrees left to right for
1406  *		sufficient free space.  if sufficient free space is found,
1407  *		the subtree is searched to find the leftmost leaf that
1408  *		has free space.  once we have made it to the leaf, we
1409  *		move the search to the next lower level dmap control page
1410  *		corresponding to this leaf.  we continue down the dmap control
1411  *		pages until we find the dmap that contains or starts the
1412  *		sufficient free space and we allocate at this dmap.
1413  *
1414  *		if the allocation group size is equal to the dmap size,
1415  *		we'll start at the dmap corresponding to the allocation
1416  *		group and attempt the allocation at this level.
1417  *
1418  *		the dmap control page search is also not performed if the
1419  *		allocation group is completely free and we go to the first
1420  *		dmap of the allocation group to do the allocation.  this is
1421  *		done because the allocation group may be part (not the first
1422  *		part) of a larger binary buddy system, causing the dmap
1423  *		control pages to indicate no free space (NOFREE) within
1424  *		the allocation group.
1425  *
1426  * PARAMETERS:
1427  *      bmp	-  pointer to bmap descriptor
1428  *	agno	- allocation group number.
1429  *      nblocks	-  actual number of contiguous free blocks desired.
1430  *      l2nb	-  log2 number of contiguous free blocks desired.
1431  *      results	-  on successful return, set to the starting block number
1432  *		   of the newly allocated range.
1433  *
1434  * RETURN VALUES:
1435  *      0	- success
1436  *      -ENOSPC	- insufficient disk resources
1437  *      -EIO	- i/o error
1438  *
1439  * note: IWRITE_LOCK(ipmap) held on entry/exit;
1440  */
1441 static int
1442 dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results)
1443 {
1444 	struct metapage *mp;
1445 	struct dmapctl *dcp;
1446 	int rc, ti, i, k, m, n, agperlev;
1447 	s64 blkno, lblkno;
1448 	int budmin;
1449 
1450 	/* allocation request should not be for more than the
1451 	 * allocation group size.
1452 	 */
1453 	if (l2nb > bmp->db_agl2size) {
1454 		jfs_error(bmp->db_ipbmap->i_sb,
1455 			  "dbAllocAG: allocation request is larger than the "
1456 			  "allocation group size");
1457 		return -EIO;
1458 	}
1459 
1460 	/* determine the starting block number of the allocation
1461 	 * group.
1462 	 */
1463 	blkno = (s64) agno << bmp->db_agl2size;
1464 
1465 	/* check if the allocation group size is the minimum allocation
1466 	 * group size or if the allocation group is completely free. if
1467 	 * the allocation group size is the minimum size of BPERDMAP (i.e.
1468 	 * 1 dmap), there is no need to search the dmap control page (below)
1469 	 * that fully describes the allocation group since the allocation
1470 	 * group is already fully described by a dmap.  in this case, we
1471 	 * just call dbAllocCtl() to search the dmap tree and allocate the
1472 	 * required space if available.
1473 	 *
1474 	 * if the allocation group is completely free, dbAllocCtl() is
1475 	 * also called to allocate the required space.  this is done for
1476 	 * two reasons.  first, it makes no sense searching the dmap control
1477 	 * pages for free space when we know that free space exists.  second,
1478 	 * the dmap control pages may indicate that the allocation group
1479 	 * has no free space if the allocation group is part (not the first
1480 	 * part) of a larger binary buddy system.
1481 	 */
1482 	if (bmp->db_agsize == BPERDMAP
1483 	    || bmp->db_agfree[agno] == bmp->db_agsize) {
1484 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1485 		if ((rc == -ENOSPC) &&
1486 		    (bmp->db_agfree[agno] == bmp->db_agsize)) {
1487 			printk(KERN_ERR "blkno = %Lx, blocks = %Lx\n",
1488 			       (unsigned long long) blkno,
1489 			       (unsigned long long) nblocks);
1490 			jfs_error(bmp->db_ipbmap->i_sb,
1491 				  "dbAllocAG: dbAllocCtl failed in free AG");
1492 		}
1493 		return (rc);
1494 	}
1495 
1496 	/* the buffer for the dmap control page that fully describes the
1497 	 * allocation group.
1498 	 */
1499 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, bmp->db_aglevel);
1500 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1501 	if (mp == NULL)
1502 		return -EIO;
1503 	dcp = (struct dmapctl *) mp->data;
1504 	budmin = dcp->budmin;
1505 
1506 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1507 		jfs_error(bmp->db_ipbmap->i_sb,
1508 			  "dbAllocAG: Corrupt dmapctl page");
1509 		release_metapage(mp);
1510 		return -EIO;
1511 	}
1512 
1513 	/* search the subtree(s) of the dmap control page that describes
1514 	 * the allocation group, looking for sufficient free space.  to begin,
1515 	 * determine how many allocation groups are represented in a dmap
1516 	 * control page at the control page level (i.e. L0, L1, L2) that
1517 	 * fully describes an allocation group. next, determine the starting
1518 	 * tree index of this allocation group within the control page.
1519 	 */
1520 	agperlev =
1521 	    (1 << (L2LPERCTL - (bmp->db_agheigth << 1))) / bmp->db_agwidth;
1522 	ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
1523 
1524 	/* dmap control page trees fan-out by 4 and a single allocation
1525 	 * group may be described by 1 or 2 subtrees within the ag level
1526 	 * dmap control page, depending upon the ag size. examine the ag's
1527 	 * subtrees for sufficient free space, starting with the leftmost
1528 	 * subtree.
1529 	 */
1530 	for (i = 0; i < bmp->db_agwidth; i++, ti++) {
1531 		/* is there sufficient free space ?
1532 		 */
1533 		if (l2nb > dcp->stree[ti])
1534 			continue;
1535 
1536 		/* sufficient free space found in a subtree. now search down
1537 		 * the subtree to find the leftmost leaf that describes this
1538 		 * free space.
1539 		 */
1540 		for (k = bmp->db_agheigth; k > 0; k--) {
1541 			for (n = 0, m = (ti << 2) + 1; n < 4; n++) {
1542 				if (l2nb <= dcp->stree[m + n]) {
1543 					ti = m + n;
1544 					break;
1545 				}
1546 			}
1547 			if (n == 4) {
1548 				jfs_error(bmp->db_ipbmap->i_sb,
1549 					  "dbAllocAG: failed descending stree");
1550 				release_metapage(mp);
1551 				return -EIO;
1552 			}
1553 		}
1554 
1555 		/* determine the block number within the file system
1556 		 * that corresponds to this leaf.
1557 		 */
1558 		if (bmp->db_aglevel == 2)
1559 			blkno = 0;
1560 		else if (bmp->db_aglevel == 1)
1561 			blkno &= ~(MAXL1SIZE - 1);
1562 		else		/* bmp->db_aglevel == 0 */
1563 			blkno &= ~(MAXL0SIZE - 1);
1564 
1565 		blkno +=
1566 		    ((s64) (ti - le32_to_cpu(dcp->leafidx))) << budmin;
1567 
1568 		/* release the buffer in preparation for going down
1569 		 * the next level of dmap control pages.
1570 		 */
1571 		release_metapage(mp);
1572 
1573 		/* check if we need to continue to search down the lower
1574 		 * level dmap control pages.  we need to if the number of
1575 		 * blocks required is less than maximum number of blocks
1576 		 * described at the next lower level.
1577 		 */
1578 		if (l2nb < budmin) {
1579 
1580 			/* search the lower level dmap control pages to get
1581 			 * the starting block number of the the dmap that
1582 			 * contains or starts off the free space.
1583 			 */
1584 			if ((rc =
1585 			     dbFindCtl(bmp, l2nb, bmp->db_aglevel - 1,
1586 				       &blkno))) {
1587 				if (rc == -ENOSPC) {
1588 					jfs_error(bmp->db_ipbmap->i_sb,
1589 						  "dbAllocAG: control page "
1590 						  "inconsistent");
1591 					return -EIO;
1592 				}
1593 				return (rc);
1594 			}
1595 		}
1596 
1597 		/* allocate the blocks.
1598 		 */
1599 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1600 		if (rc == -ENOSPC) {
1601 			jfs_error(bmp->db_ipbmap->i_sb,
1602 				  "dbAllocAG: unable to allocate blocks");
1603 			rc = -EIO;
1604 		}
1605 		return (rc);
1606 	}
1607 
1608 	/* no space in the allocation group.  release the buffer and
1609 	 * return -ENOSPC.
1610 	 */
1611 	release_metapage(mp);
1612 
1613 	return -ENOSPC;
1614 }
1615 
1616 
1617 /*
1618  * NAME:	dbAllocAny()
1619  *
1620  * FUNCTION:    attempt to allocate the specified number of contiguous
1621  *		free blocks anywhere in the file system.
1622  *
1623  *		dbAllocAny() attempts to find the sufficient free space by
1624  *		searching down the dmap control pages, starting with the
1625  *		highest level (i.e. L0, L1, L2) control page.  if free space
1626  *		large enough to satisfy the desired free space is found, the
1627  *		desired free space is allocated.
1628  *
1629  * PARAMETERS:
1630  *      bmp	-  pointer to bmap descriptor
1631  *      nblocks	 -  actual number of contiguous free blocks desired.
1632  *      l2nb	 -  log2 number of contiguous free blocks desired.
1633  *      results	-  on successful return, set to the starting block number
1634  *		   of the newly allocated range.
1635  *
1636  * RETURN VALUES:
1637  *      0	- success
1638  *      -ENOSPC	- insufficient disk resources
1639  *      -EIO	- i/o error
1640  *
1641  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1642  */
1643 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results)
1644 {
1645 	int rc;
1646 	s64 blkno = 0;
1647 
1648 	/* starting with the top level dmap control page, search
1649 	 * down the dmap control levels for sufficient free space.
1650 	 * if free space is found, dbFindCtl() returns the starting
1651 	 * block number of the dmap that contains or starts off the
1652 	 * range of free space.
1653 	 */
1654 	if ((rc = dbFindCtl(bmp, l2nb, bmp->db_maxlevel, &blkno)))
1655 		return (rc);
1656 
1657 	/* allocate the blocks.
1658 	 */
1659 	rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1660 	if (rc == -ENOSPC) {
1661 		jfs_error(bmp->db_ipbmap->i_sb,
1662 			  "dbAllocAny: unable to allocate blocks");
1663 		return -EIO;
1664 	}
1665 	return (rc);
1666 }
1667 
1668 
1669 /*
1670  * NAME:	dbFindCtl()
1671  *
1672  * FUNCTION:    starting at a specified dmap control page level and block
1673  *		number, search down the dmap control levels for a range of
1674  *	        contiguous free blocks large enough to satisfy an allocation
1675  *		request for the specified number of free blocks.
1676  *
1677  *		if sufficient contiguous free blocks are found, this routine
1678  *		returns the starting block number within a dmap page that
1679  *		contains or starts a range of contiqious free blocks that
1680  *		is sufficient in size.
1681  *
1682  * PARAMETERS:
1683  *      bmp	-  pointer to bmap descriptor
1684  *      level	-  starting dmap control page level.
1685  *      l2nb	-  log2 number of contiguous free blocks desired.
1686  *      *blkno	-  on entry, starting block number for conducting the search.
1687  *		   on successful return, the first block within a dmap page
1688  *		   that contains or starts a range of contiguous free blocks.
1689  *
1690  * RETURN VALUES:
1691  *      0	- success
1692  *      -ENOSPC	- insufficient disk resources
1693  *      -EIO	- i/o error
1694  *
1695  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1696  */
1697 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno)
1698 {
1699 	int rc, leafidx, lev;
1700 	s64 b, lblkno;
1701 	struct dmapctl *dcp;
1702 	int budmin;
1703 	struct metapage *mp;
1704 
1705 	/* starting at the specified dmap control page level and block
1706 	 * number, search down the dmap control levels for the starting
1707 	 * block number of a dmap page that contains or starts off
1708 	 * sufficient free blocks.
1709 	 */
1710 	for (lev = level, b = *blkno; lev >= 0; lev--) {
1711 		/* get the buffer of the dmap control page for the block
1712 		 * number and level (i.e. L0, L1, L2).
1713 		 */
1714 		lblkno = BLKTOCTL(b, bmp->db_l2nbperpage, lev);
1715 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1716 		if (mp == NULL)
1717 			return -EIO;
1718 		dcp = (struct dmapctl *) mp->data;
1719 		budmin = dcp->budmin;
1720 
1721 		if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1722 			jfs_error(bmp->db_ipbmap->i_sb,
1723 				  "dbFindCtl: Corrupt dmapctl page");
1724 			release_metapage(mp);
1725 			return -EIO;
1726 		}
1727 
1728 		/* search the tree within the dmap control page for
1729 		 * sufficent free space.  if sufficient free space is found,
1730 		 * dbFindLeaf() returns the index of the leaf at which
1731 		 * free space was found.
1732 		 */
1733 		rc = dbFindLeaf((dmtree_t *) dcp, l2nb, &leafidx);
1734 
1735 		/* release the buffer.
1736 		 */
1737 		release_metapage(mp);
1738 
1739 		/* space found ?
1740 		 */
1741 		if (rc) {
1742 			if (lev != level) {
1743 				jfs_error(bmp->db_ipbmap->i_sb,
1744 					  "dbFindCtl: dmap inconsistent");
1745 				return -EIO;
1746 			}
1747 			return -ENOSPC;
1748 		}
1749 
1750 		/* adjust the block number to reflect the location within
1751 		 * the dmap control page (i.e. the leaf) at which free
1752 		 * space was found.
1753 		 */
1754 		b += (((s64) leafidx) << budmin);
1755 
1756 		/* we stop the search at this dmap control page level if
1757 		 * the number of blocks required is greater than or equal
1758 		 * to the maximum number of blocks described at the next
1759 		 * (lower) level.
1760 		 */
1761 		if (l2nb >= budmin)
1762 			break;
1763 	}
1764 
1765 	*blkno = b;
1766 	return (0);
1767 }
1768 
1769 
1770 /*
1771  * NAME:	dbAllocCtl()
1772  *
1773  * FUNCTION:    attempt to allocate a specified number of contiguous
1774  *		blocks starting within a specific dmap.
1775  *
1776  *		this routine is called by higher level routines that search
1777  *		the dmap control pages above the actual dmaps for contiguous
1778  *		free space.  the result of successful searches by these
1779  * 		routines are the starting block numbers within dmaps, with
1780  *		the dmaps themselves containing the desired contiguous free
1781  *		space or starting a contiguous free space of desired size
1782  *		that is made up of the blocks of one or more dmaps. these
1783  *		calls should not fail due to insufficent resources.
1784  *
1785  *		this routine is called in some cases where it is not known
1786  *		whether it will fail due to insufficient resources.  more
1787  *		specifically, this occurs when allocating from an allocation
1788  *		group whose size is equal to the number of blocks per dmap.
1789  *		in this case, the dmap control pages are not examined prior
1790  *		to calling this routine (to save pathlength) and the call
1791  *		might fail.
1792  *
1793  *		for a request size that fits within a dmap, this routine relies
1794  *		upon the dmap's dmtree to find the requested contiguous free
1795  *		space.  for request sizes that are larger than a dmap, the
1796  *		requested free space will start at the first block of the
1797  *		first dmap (i.e. blkno).
1798  *
1799  * PARAMETERS:
1800  *      bmp	-  pointer to bmap descriptor
1801  *      nblocks	 -  actual number of contiguous free blocks to allocate.
1802  *      l2nb	 -  log2 number of contiguous free blocks to allocate.
1803  *      blkno	 -  starting block number of the dmap to start the allocation
1804  *		    from.
1805  *      results	-  on successful return, set to the starting block number
1806  *		   of the newly allocated range.
1807  *
1808  * RETURN VALUES:
1809  *      0	- success
1810  *      -ENOSPC	- insufficient disk resources
1811  *      -EIO	- i/o error
1812  *
1813  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1814  */
1815 static int
1816 dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, s64 * results)
1817 {
1818 	int rc, nb;
1819 	s64 b, lblkno, n;
1820 	struct metapage *mp;
1821 	struct dmap *dp;
1822 
1823 	/* check if the allocation request is confined to a single dmap.
1824 	 */
1825 	if (l2nb <= L2BPERDMAP) {
1826 		/* get the buffer for the dmap.
1827 		 */
1828 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
1829 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1830 		if (mp == NULL)
1831 			return -EIO;
1832 		dp = (struct dmap *) mp->data;
1833 
1834 		/* try to allocate the blocks.
1835 		 */
1836 		rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results);
1837 		if (rc == 0)
1838 			mark_metapage_dirty(mp);
1839 
1840 		release_metapage(mp);
1841 
1842 		return (rc);
1843 	}
1844 
1845 	/* allocation request involving multiple dmaps. it must start on
1846 	 * a dmap boundary.
1847 	 */
1848 	assert((blkno & (BPERDMAP - 1)) == 0);
1849 
1850 	/* allocate the blocks dmap by dmap.
1851 	 */
1852 	for (n = nblocks, b = blkno; n > 0; n -= nb, b += nb) {
1853 		/* get the buffer for the dmap.
1854 		 */
1855 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1856 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1857 		if (mp == NULL) {
1858 			rc = -EIO;
1859 			goto backout;
1860 		}
1861 		dp = (struct dmap *) mp->data;
1862 
1863 		/* the dmap better be all free.
1864 		 */
1865 		if (dp->tree.stree[ROOT] != L2BPERDMAP) {
1866 			release_metapage(mp);
1867 			jfs_error(bmp->db_ipbmap->i_sb,
1868 				  "dbAllocCtl: the dmap is not all free");
1869 			rc = -EIO;
1870 			goto backout;
1871 		}
1872 
1873 		/* determine how many blocks to allocate from this dmap.
1874 		 */
1875 		nb = min(n, (s64)BPERDMAP);
1876 
1877 		/* allocate the blocks from the dmap.
1878 		 */
1879 		if ((rc = dbAllocDmap(bmp, dp, b, nb))) {
1880 			release_metapage(mp);
1881 			goto backout;
1882 		}
1883 
1884 		/* write the buffer.
1885 		 */
1886 		write_metapage(mp);
1887 	}
1888 
1889 	/* set the results (starting block number) and return.
1890 	 */
1891 	*results = blkno;
1892 	return (0);
1893 
1894 	/* something failed in handling an allocation request involving
1895 	 * multiple dmaps.  we'll try to clean up by backing out any
1896 	 * allocation that has already happened for this request.  if
1897 	 * we fail in backing out the allocation, we'll mark the file
1898 	 * system to indicate that blocks have been leaked.
1899 	 */
1900       backout:
1901 
1902 	/* try to backout the allocations dmap by dmap.
1903 	 */
1904 	for (n = nblocks - n, b = blkno; n > 0;
1905 	     n -= BPERDMAP, b += BPERDMAP) {
1906 		/* get the buffer for this dmap.
1907 		 */
1908 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1909 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1910 		if (mp == NULL) {
1911 			/* could not back out.  mark the file system
1912 			 * to indicate that we have leaked blocks.
1913 			 */
1914 			jfs_error(bmp->db_ipbmap->i_sb,
1915 				  "dbAllocCtl: I/O Error: Block Leakage.");
1916 			continue;
1917 		}
1918 		dp = (struct dmap *) mp->data;
1919 
1920 		/* free the blocks is this dmap.
1921 		 */
1922 		if (dbFreeDmap(bmp, dp, b, BPERDMAP)) {
1923 			/* could not back out.  mark the file system
1924 			 * to indicate that we have leaked blocks.
1925 			 */
1926 			release_metapage(mp);
1927 			jfs_error(bmp->db_ipbmap->i_sb,
1928 				  "dbAllocCtl: Block Leakage.");
1929 			continue;
1930 		}
1931 
1932 		/* write the buffer.
1933 		 */
1934 		write_metapage(mp);
1935 	}
1936 
1937 	return (rc);
1938 }
1939 
1940 
1941 /*
1942  * NAME:	dbAllocDmapLev()
1943  *
1944  * FUNCTION:    attempt to allocate a specified number of contiguous blocks
1945  *		from a specified dmap.
1946  *
1947  *		this routine checks if the contiguous blocks are available.
1948  *		if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1949  *		returned.
1950  *
1951  * PARAMETERS:
1952  *      mp	-  pointer to bmap descriptor
1953  *      dp	-  pointer to dmap to attempt to allocate blocks from.
1954  *      l2nb	-  log2 number of contiguous block desired.
1955  *      nblocks	-  actual number of contiguous block desired.
1956  *      results	-  on successful return, set to the starting block number
1957  *		   of the newly allocated range.
1958  *
1959  * RETURN VALUES:
1960  *      0	- success
1961  *      -ENOSPC	- insufficient disk resources
1962  *      -EIO	- i/o error
1963  *
1964  * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1965  *	IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1966  */
1967 static int
1968 dbAllocDmapLev(struct bmap * bmp,
1969 	       struct dmap * dp, int nblocks, int l2nb, s64 * results)
1970 {
1971 	s64 blkno;
1972 	int leafidx, rc;
1973 
1974 	/* can't be more than a dmaps worth of blocks */
1975 	assert(l2nb <= L2BPERDMAP);
1976 
1977 	/* search the tree within the dmap page for sufficient
1978 	 * free space.  if sufficient free space is found, dbFindLeaf()
1979 	 * returns the index of the leaf at which free space was found.
1980 	 */
1981 	if (dbFindLeaf((dmtree_t *) & dp->tree, l2nb, &leafidx))
1982 		return -ENOSPC;
1983 
1984 	/* determine the block number within the file system corresponding
1985 	 * to the leaf at which free space was found.
1986 	 */
1987 	blkno = le64_to_cpu(dp->start) + (leafidx << L2DBWORD);
1988 
1989 	/* if not all bits of the dmap word are free, get the starting
1990 	 * bit number within the dmap word of the required string of free
1991 	 * bits and adjust the block number with this value.
1992 	 */
1993 	if (dp->tree.stree[leafidx + LEAFIND] < BUDMIN)
1994 		blkno += dbFindBits(le32_to_cpu(dp->wmap[leafidx]), l2nb);
1995 
1996 	/* allocate the blocks */
1997 	if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1998 		*results = blkno;
1999 
2000 	return (rc);
2001 }
2002 
2003 
2004 /*
2005  * NAME:	dbAllocDmap()
2006  *
2007  * FUNCTION:    adjust the disk allocation map to reflect the allocation
2008  *		of a specified block range within a dmap.
2009  *
2010  *		this routine allocates the specified blocks from the dmap
2011  *		through a call to dbAllocBits(). if the allocation of the
2012  *		block range causes the maximum string of free blocks within
2013  *		the dmap to change (i.e. the value of the root of the dmap's
2014  *		dmtree), this routine will cause this change to be reflected
2015  *		up through the appropriate levels of the dmap control pages
2016  *		by a call to dbAdjCtl() for the L0 dmap control page that
2017  *		covers this dmap.
2018  *
2019  * PARAMETERS:
2020  *      bmp	-  pointer to bmap descriptor
2021  *      dp	-  pointer to dmap to allocate the block range from.
2022  *      blkno	-  starting block number of the block to be allocated.
2023  *      nblocks	-  number of blocks to be allocated.
2024  *
2025  * RETURN VALUES:
2026  *      0	- success
2027  *      -EIO	- i/o error
2028  *
2029  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2030  */
2031 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2032 		       int nblocks)
2033 {
2034 	s8 oldroot;
2035 	int rc;
2036 
2037 	/* save the current value of the root (i.e. maximum free string)
2038 	 * of the dmap tree.
2039 	 */
2040 	oldroot = dp->tree.stree[ROOT];
2041 
2042 	/* allocate the specified (blocks) bits */
2043 	dbAllocBits(bmp, dp, blkno, nblocks);
2044 
2045 	/* if the root has not changed, done. */
2046 	if (dp->tree.stree[ROOT] == oldroot)
2047 		return (0);
2048 
2049 	/* root changed. bubble the change up to the dmap control pages.
2050 	 * if the adjustment of the upper level control pages fails,
2051 	 * backout the bit allocation (thus making everything consistent).
2052 	 */
2053 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 1, 0)))
2054 		dbFreeBits(bmp, dp, blkno, nblocks);
2055 
2056 	return (rc);
2057 }
2058 
2059 
2060 /*
2061  * NAME:	dbFreeDmap()
2062  *
2063  * FUNCTION:    adjust the disk allocation map to reflect the allocation
2064  *		of a specified block range within a dmap.
2065  *
2066  *		this routine frees the specified blocks from the dmap through
2067  *		a call to dbFreeBits(). if the deallocation of the block range
2068  *		causes the maximum string of free blocks within the dmap to
2069  *		change (i.e. the value of the root of the dmap's dmtree), this
2070  *		routine will cause this change to be reflected up through the
2071  *	        appropriate levels of the dmap control pages by a call to
2072  *		dbAdjCtl() for the L0 dmap control page that covers this dmap.
2073  *
2074  * PARAMETERS:
2075  *      bmp	-  pointer to bmap descriptor
2076  *      dp	-  pointer to dmap to free the block range from.
2077  *      blkno	-  starting block number of the block to be freed.
2078  *      nblocks	-  number of blocks to be freed.
2079  *
2080  * RETURN VALUES:
2081  *      0	- success
2082  *      -EIO	- i/o error
2083  *
2084  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2085  */
2086 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2087 		      int nblocks)
2088 {
2089 	s8 oldroot;
2090 	int rc, word;
2091 
2092 	/* save the current value of the root (i.e. maximum free string)
2093 	 * of the dmap tree.
2094 	 */
2095 	oldroot = dp->tree.stree[ROOT];
2096 
2097 	/* free the specified (blocks) bits */
2098 	dbFreeBits(bmp, dp, blkno, nblocks);
2099 
2100 	/* if the root has not changed, done. */
2101 	if (dp->tree.stree[ROOT] == oldroot)
2102 		return (0);
2103 
2104 	/* root changed. bubble the change up to the dmap control pages.
2105 	 * if the adjustment of the upper level control pages fails,
2106 	 * backout the deallocation.
2107 	 */
2108 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 0, 0))) {
2109 		word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
2110 
2111 		/* as part of backing out the deallocation, we will have
2112 		 * to back split the dmap tree if the deallocation caused
2113 		 * the freed blocks to become part of a larger binary buddy
2114 		 * system.
2115 		 */
2116 		if (dp->tree.stree[word] == NOFREE)
2117 			dbBackSplit((dmtree_t *) & dp->tree, word);
2118 
2119 		dbAllocBits(bmp, dp, blkno, nblocks);
2120 	}
2121 
2122 	return (rc);
2123 }
2124 
2125 
2126 /*
2127  * NAME:	dbAllocBits()
2128  *
2129  * FUNCTION:    allocate a specified block range from a dmap.
2130  *
2131  *		this routine updates the dmap to reflect the working
2132  *		state allocation of the specified block range. it directly
2133  *		updates the bits of the working map and causes the adjustment
2134  *		of the binary buddy system described by the dmap's dmtree
2135  *		leaves to reflect the bits allocated.  it also causes the
2136  *		dmap's dmtree, as a whole, to reflect the allocated range.
2137  *
2138  * PARAMETERS:
2139  *      bmp	-  pointer to bmap descriptor
2140  *      dp	-  pointer to dmap to allocate bits from.
2141  *      blkno	-  starting block number of the bits to be allocated.
2142  *      nblocks	-  number of bits to be allocated.
2143  *
2144  * RETURN VALUES: none
2145  *
2146  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2147  */
2148 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2149 			int nblocks)
2150 {
2151 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2152 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2153 	int size;
2154 	s8 *leaf;
2155 
2156 	/* pick up a pointer to the leaves of the dmap tree */
2157 	leaf = dp->tree.stree + LEAFIND;
2158 
2159 	/* determine the bit number and word within the dmap of the
2160 	 * starting block.
2161 	 */
2162 	dbitno = blkno & (BPERDMAP - 1);
2163 	word = dbitno >> L2DBWORD;
2164 
2165 	/* block range better be within the dmap */
2166 	assert(dbitno + nblocks <= BPERDMAP);
2167 
2168 	/* allocate the bits of the dmap's words corresponding to the block
2169 	 * range. not all bits of the first and last words may be contained
2170 	 * within the block range.  if this is the case, we'll work against
2171 	 * those words (i.e. partial first and/or last) on an individual basis
2172 	 * (a single pass), allocating the bits of interest by hand and
2173 	 * updating the leaf corresponding to the dmap word. a single pass
2174 	 * will be used for all dmap words fully contained within the
2175 	 * specified range.  within this pass, the bits of all fully contained
2176 	 * dmap words will be marked as free in a single shot and the leaves
2177 	 * will be updated. a single leaf may describe the free space of
2178 	 * multiple dmap words, so we may update only a subset of the actual
2179 	 * leaves corresponding to the dmap words of the block range.
2180 	 */
2181 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2182 		/* determine the bit number within the word and
2183 		 * the number of bits within the word.
2184 		 */
2185 		wbitno = dbitno & (DBWORD - 1);
2186 		nb = min(rembits, DBWORD - wbitno);
2187 
2188 		/* check if only part of a word is to be allocated.
2189 		 */
2190 		if (nb < DBWORD) {
2191 			/* allocate (set to 1) the appropriate bits within
2192 			 * this dmap word.
2193 			 */
2194 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
2195 						      >> wbitno);
2196 
2197 			/* update the leaf for this dmap word. in addition
2198 			 * to setting the leaf value to the binary buddy max
2199 			 * of the updated dmap word, dbSplit() will split
2200 			 * the binary system of the leaves if need be.
2201 			 */
2202 			dbSplit(tp, word, BUDMIN,
2203 				dbMaxBud((u8 *) & dp->wmap[word]));
2204 
2205 			word += 1;
2206 		} else {
2207 			/* one or more dmap words are fully contained
2208 			 * within the block range.  determine how many
2209 			 * words and allocate (set to 1) the bits of these
2210 			 * words.
2211 			 */
2212 			nwords = rembits >> L2DBWORD;
2213 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
2214 
2215 			/* determine how many bits.
2216 			 */
2217 			nb = nwords << L2DBWORD;
2218 
2219 			/* now update the appropriate leaves to reflect
2220 			 * the allocated words.
2221 			 */
2222 			for (; nwords > 0; nwords -= nw) {
2223 			        if (leaf[word] < BUDMIN) {
2224 					jfs_error(bmp->db_ipbmap->i_sb,
2225 						  "dbAllocBits: leaf page "
2226 						  "corrupt");
2227 					break;
2228 				}
2229 
2230 				/* determine what the leaf value should be
2231 				 * updated to as the minimum of the l2 number
2232 				 * of bits being allocated and the l2 number
2233 				 * of bits currently described by this leaf.
2234 				 */
2235 				size = min((int)leaf[word], NLSTOL2BSZ(nwords));
2236 
2237 				/* update the leaf to reflect the allocation.
2238 				 * in addition to setting the leaf value to
2239 				 * NOFREE, dbSplit() will split the binary
2240 				 * system of the leaves to reflect the current
2241 				 * allocation (size).
2242 				 */
2243 				dbSplit(tp, word, size, NOFREE);
2244 
2245 				/* get the number of dmap words handled */
2246 				nw = BUDSIZE(size, BUDMIN);
2247 				word += nw;
2248 			}
2249 		}
2250 	}
2251 
2252 	/* update the free count for this dmap */
2253 	dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) - nblocks);
2254 
2255 	BMAP_LOCK(bmp);
2256 
2257 	/* if this allocation group is completely free,
2258 	 * update the maximum allocation group number if this allocation
2259 	 * group is the new max.
2260 	 */
2261 	agno = blkno >> bmp->db_agl2size;
2262 	if (agno > bmp->db_maxag)
2263 		bmp->db_maxag = agno;
2264 
2265 	/* update the free count for the allocation group and map */
2266 	bmp->db_agfree[agno] -= nblocks;
2267 	bmp->db_nfree -= nblocks;
2268 
2269 	BMAP_UNLOCK(bmp);
2270 }
2271 
2272 
2273 /*
2274  * NAME:	dbFreeBits()
2275  *
2276  * FUNCTION:    free a specified block range from a dmap.
2277  *
2278  *		this routine updates the dmap to reflect the working
2279  *		state allocation of the specified block range. it directly
2280  *		updates the bits of the working map and causes the adjustment
2281  *		of the binary buddy system described by the dmap's dmtree
2282  *		leaves to reflect the bits freed.  it also causes the dmap's
2283  *		dmtree, as a whole, to reflect the deallocated range.
2284  *
2285  * PARAMETERS:
2286  *      bmp	-  pointer to bmap descriptor
2287  *      dp	-  pointer to dmap to free bits from.
2288  *      blkno	-  starting block number of the bits to be freed.
2289  *      nblocks	-  number of bits to be freed.
2290  *
2291  * RETURN VALUES: none
2292  *
2293  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2294  */
2295 static void dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2296 		       int nblocks)
2297 {
2298 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2299 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2300 	int size;
2301 
2302 	/* determine the bit number and word within the dmap of the
2303 	 * starting block.
2304 	 */
2305 	dbitno = blkno & (BPERDMAP - 1);
2306 	word = dbitno >> L2DBWORD;
2307 
2308 	/* block range better be within the dmap.
2309 	 */
2310 	assert(dbitno + nblocks <= BPERDMAP);
2311 
2312 	/* free the bits of the dmaps words corresponding to the block range.
2313 	 * not all bits of the first and last words may be contained within
2314 	 * the block range.  if this is the case, we'll work against those
2315 	 * words (i.e. partial first and/or last) on an individual basis
2316 	 * (a single pass), freeing the bits of interest by hand and updating
2317 	 * the leaf corresponding to the dmap word. a single pass will be used
2318 	 * for all dmap words fully contained within the specified range.
2319 	 * within this pass, the bits of all fully contained dmap words will
2320 	 * be marked as free in a single shot and the leaves will be updated. a
2321 	 * single leaf may describe the free space of multiple dmap words,
2322 	 * so we may update only a subset of the actual leaves corresponding
2323 	 * to the dmap words of the block range.
2324 	 *
2325 	 * dbJoin() is used to update leaf values and will join the binary
2326 	 * buddy system of the leaves if the new leaf values indicate this
2327 	 * should be done.
2328 	 */
2329 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2330 		/* determine the bit number within the word and
2331 		 * the number of bits within the word.
2332 		 */
2333 		wbitno = dbitno & (DBWORD - 1);
2334 		nb = min(rembits, DBWORD - wbitno);
2335 
2336 		/* check if only part of a word is to be freed.
2337 		 */
2338 		if (nb < DBWORD) {
2339 			/* free (zero) the appropriate bits within this
2340 			 * dmap word.
2341 			 */
2342 			dp->wmap[word] &=
2343 			    cpu_to_le32(~(ONES << (DBWORD - nb)
2344 					  >> wbitno));
2345 
2346 			/* update the leaf for this dmap word.
2347 			 */
2348 			dbJoin(tp, word,
2349 			       dbMaxBud((u8 *) & dp->wmap[word]));
2350 
2351 			word += 1;
2352 		} else {
2353 			/* one or more dmap words are fully contained
2354 			 * within the block range.  determine how many
2355 			 * words and free (zero) the bits of these words.
2356 			 */
2357 			nwords = rembits >> L2DBWORD;
2358 			memset(&dp->wmap[word], 0, nwords * 4);
2359 
2360 			/* determine how many bits.
2361 			 */
2362 			nb = nwords << L2DBWORD;
2363 
2364 			/* now update the appropriate leaves to reflect
2365 			 * the freed words.
2366 			 */
2367 			for (; nwords > 0; nwords -= nw) {
2368 				/* determine what the leaf value should be
2369 				 * updated to as the minimum of the l2 number
2370 				 * of bits being freed and the l2 (max) number
2371 				 * of bits that can be described by this leaf.
2372 				 */
2373 				size =
2374 				    min(LITOL2BSZ
2375 					(word, L2LPERDMAP, BUDMIN),
2376 					NLSTOL2BSZ(nwords));
2377 
2378 				/* update the leaf.
2379 				 */
2380 				dbJoin(tp, word, size);
2381 
2382 				/* get the number of dmap words handled.
2383 				 */
2384 				nw = BUDSIZE(size, BUDMIN);
2385 				word += nw;
2386 			}
2387 		}
2388 	}
2389 
2390 	/* update the free count for this dmap.
2391 	 */
2392 	dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) + nblocks);
2393 
2394 	BMAP_LOCK(bmp);
2395 
2396 	/* update the free count for the allocation group and
2397 	 * map.
2398 	 */
2399 	agno = blkno >> bmp->db_agl2size;
2400 	bmp->db_nfree += nblocks;
2401 	bmp->db_agfree[agno] += nblocks;
2402 
2403 	/* check if this allocation group is not completely free and
2404 	 * if it is currently the maximum (rightmost) allocation group.
2405 	 * if so, establish the new maximum allocation group number by
2406 	 * searching left for the first allocation group with allocation.
2407 	 */
2408 	if ((bmp->db_agfree[agno] == bmp->db_agsize && agno == bmp->db_maxag) ||
2409 	    (agno == bmp->db_numag - 1 &&
2410 	     bmp->db_agfree[agno] == (bmp-> db_mapsize & (BPERDMAP - 1)))) {
2411 		while (bmp->db_maxag > 0) {
2412 			bmp->db_maxag -= 1;
2413 			if (bmp->db_agfree[bmp->db_maxag] !=
2414 			    bmp->db_agsize)
2415 				break;
2416 		}
2417 
2418 		/* re-establish the allocation group preference if the
2419 		 * current preference is right of the maximum allocation
2420 		 * group.
2421 		 */
2422 		if (bmp->db_agpref > bmp->db_maxag)
2423 			bmp->db_agpref = bmp->db_maxag;
2424 	}
2425 
2426 	BMAP_UNLOCK(bmp);
2427 }
2428 
2429 
2430 /*
2431  * NAME:	dbAdjCtl()
2432  *
2433  * FUNCTION:	adjust a dmap control page at a specified level to reflect
2434  *		the change in a lower level dmap or dmap control page's
2435  *		maximum string of free blocks (i.e. a change in the root
2436  *		of the lower level object's dmtree) due to the allocation
2437  *		or deallocation of a range of blocks with a single dmap.
2438  *
2439  *		on entry, this routine is provided with the new value of
2440  *		the lower level dmap or dmap control page root and the
2441  *		starting block number of the block range whose allocation
2442  *		or deallocation resulted in the root change.  this range
2443  *		is respresented by a single leaf of the current dmapctl
2444  *		and the leaf will be updated with this value, possibly
2445  *		causing a binary buddy system within the leaves to be
2446  *		split or joined.  the update may also cause the dmapctl's
2447  *		dmtree to be updated.
2448  *
2449  *		if the adjustment of the dmap control page, itself, causes its
2450  *		root to change, this change will be bubbled up to the next dmap
2451  *		control level by a recursive call to this routine, specifying
2452  *		the new root value and the next dmap control page level to
2453  *		be adjusted.
2454  * PARAMETERS:
2455  *      bmp	-  pointer to bmap descriptor
2456  *      blkno	-  the first block of a block range within a dmap.  it is
2457  *		   the allocation or deallocation of this block range that
2458  *		   requires the dmap control page to be adjusted.
2459  *      newval	-  the new value of the lower level dmap or dmap control
2460  *		   page root.
2461  *      alloc	-  TRUE if adjustment is due to an allocation.
2462  *      level	-  current level of dmap control page (i.e. L0, L1, L2) to
2463  *		   be adjusted.
2464  *
2465  * RETURN VALUES:
2466  *      0	- success
2467  *      -EIO	- i/o error
2468  *
2469  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2470  */
2471 static int
2472 dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, int level)
2473 {
2474 	struct metapage *mp;
2475 	s8 oldroot;
2476 	int oldval;
2477 	s64 lblkno;
2478 	struct dmapctl *dcp;
2479 	int rc, leafno, ti;
2480 
2481 	/* get the buffer for the dmap control page for the specified
2482 	 * block number and control page level.
2483 	 */
2484 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, level);
2485 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
2486 	if (mp == NULL)
2487 		return -EIO;
2488 	dcp = (struct dmapctl *) mp->data;
2489 
2490 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
2491 		jfs_error(bmp->db_ipbmap->i_sb,
2492 			  "dbAdjCtl: Corrupt dmapctl page");
2493 		release_metapage(mp);
2494 		return -EIO;
2495 	}
2496 
2497 	/* determine the leaf number corresponding to the block and
2498 	 * the index within the dmap control tree.
2499 	 */
2500 	leafno = BLKTOCTLLEAF(blkno, dcp->budmin);
2501 	ti = leafno + le32_to_cpu(dcp->leafidx);
2502 
2503 	/* save the current leaf value and the current root level (i.e.
2504 	 * maximum l2 free string described by this dmapctl).
2505 	 */
2506 	oldval = dcp->stree[ti];
2507 	oldroot = dcp->stree[ROOT];
2508 
2509 	/* check if this is a control page update for an allocation.
2510 	 * if so, update the leaf to reflect the new leaf value using
2511 	 * dbSplit(); otherwise (deallocation), use dbJoin() to udpate
2512 	 * the leaf with the new value.  in addition to updating the
2513 	 * leaf, dbSplit() will also split the binary buddy system of
2514 	 * the leaves, if required, and bubble new values within the
2515 	 * dmapctl tree, if required.  similarly, dbJoin() will join
2516 	 * the binary buddy system of leaves and bubble new values up
2517 	 * the dmapctl tree as required by the new leaf value.
2518 	 */
2519 	if (alloc) {
2520 		/* check if we are in the middle of a binary buddy
2521 		 * system.  this happens when we are performing the
2522 		 * first allocation out of an allocation group that
2523 		 * is part (not the first part) of a larger binary
2524 		 * buddy system.  if we are in the middle, back split
2525 		 * the system prior to calling dbSplit() which assumes
2526 		 * that it is at the front of a binary buddy system.
2527 		 */
2528 		if (oldval == NOFREE) {
2529 			dbBackSplit((dmtree_t *) dcp, leafno);
2530 			oldval = dcp->stree[ti];
2531 		}
2532 		dbSplit((dmtree_t *) dcp, leafno, dcp->budmin, newval);
2533 	} else {
2534 		dbJoin((dmtree_t *) dcp, leafno, newval);
2535 	}
2536 
2537 	/* check if the root of the current dmap control page changed due
2538 	 * to the update and if the current dmap control page is not at
2539 	 * the current top level (i.e. L0, L1, L2) of the map.  if so (i.e.
2540 	 * root changed and this is not the top level), call this routine
2541 	 * again (recursion) for the next higher level of the mapping to
2542 	 * reflect the change in root for the current dmap control page.
2543 	 */
2544 	if (dcp->stree[ROOT] != oldroot) {
2545 		/* are we below the top level of the map.  if so,
2546 		 * bubble the root up to the next higher level.
2547 		 */
2548 		if (level < bmp->db_maxlevel) {
2549 			/* bubble up the new root of this dmap control page to
2550 			 * the next level.
2551 			 */
2552 			if ((rc =
2553 			     dbAdjCtl(bmp, blkno, dcp->stree[ROOT], alloc,
2554 				      level + 1))) {
2555 				/* something went wrong in bubbling up the new
2556 				 * root value, so backout the changes to the
2557 				 * current dmap control page.
2558 				 */
2559 				if (alloc) {
2560 					dbJoin((dmtree_t *) dcp, leafno,
2561 					       oldval);
2562 				} else {
2563 					/* the dbJoin() above might have
2564 					 * caused a larger binary buddy system
2565 					 * to form and we may now be in the
2566 					 * middle of it.  if this is the case,
2567 					 * back split the buddies.
2568 					 */
2569 					if (dcp->stree[ti] == NOFREE)
2570 						dbBackSplit((dmtree_t *)
2571 							    dcp, leafno);
2572 					dbSplit((dmtree_t *) dcp, leafno,
2573 						dcp->budmin, oldval);
2574 				}
2575 
2576 				/* release the buffer and return the error.
2577 				 */
2578 				release_metapage(mp);
2579 				return (rc);
2580 			}
2581 		} else {
2582 			/* we're at the top level of the map. update
2583 			 * the bmap control page to reflect the size
2584 			 * of the maximum free buddy system.
2585 			 */
2586 			assert(level == bmp->db_maxlevel);
2587 			if (bmp->db_maxfreebud != oldroot) {
2588 				jfs_error(bmp->db_ipbmap->i_sb,
2589 					  "dbAdjCtl: the maximum free buddy is "
2590 					  "not the old root");
2591 			}
2592 			bmp->db_maxfreebud = dcp->stree[ROOT];
2593 		}
2594 	}
2595 
2596 	/* write the buffer.
2597 	 */
2598 	write_metapage(mp);
2599 
2600 	return (0);
2601 }
2602 
2603 
2604 /*
2605  * NAME:	dbSplit()
2606  *
2607  * FUNCTION:    update the leaf of a dmtree with a new value, splitting
2608  *		the leaf from the binary buddy system of the dmtree's
2609  *		leaves, as required.
2610  *
2611  * PARAMETERS:
2612  *      tp	- pointer to the tree containing the leaf.
2613  *      leafno	- the number of the leaf to be updated.
2614  *      splitsz	- the size the binary buddy system starting at the leaf
2615  *		  must be split to, specified as the log2 number of blocks.
2616  *      newval	- the new value for the leaf.
2617  *
2618  * RETURN VALUES: none
2619  *
2620  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2621  */
2622 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval)
2623 {
2624 	int budsz;
2625 	int cursz;
2626 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2627 
2628 	/* check if the leaf needs to be split.
2629 	 */
2630 	if (leaf[leafno] > tp->dmt_budmin) {
2631 		/* the split occurs by cutting the buddy system in half
2632 		 * at the specified leaf until we reach the specified
2633 		 * size.  pick up the starting split size (current size
2634 		 * - 1 in l2) and the corresponding buddy size.
2635 		 */
2636 		cursz = leaf[leafno] - 1;
2637 		budsz = BUDSIZE(cursz, tp->dmt_budmin);
2638 
2639 		/* split until we reach the specified size.
2640 		 */
2641 		while (cursz >= splitsz) {
2642 			/* update the buddy's leaf with its new value.
2643 			 */
2644 			dbAdjTree(tp, leafno ^ budsz, cursz);
2645 
2646 			/* on to the next size and buddy.
2647 			 */
2648 			cursz -= 1;
2649 			budsz >>= 1;
2650 		}
2651 	}
2652 
2653 	/* adjust the dmap tree to reflect the specified leaf's new
2654 	 * value.
2655 	 */
2656 	dbAdjTree(tp, leafno, newval);
2657 }
2658 
2659 
2660 /*
2661  * NAME:	dbBackSplit()
2662  *
2663  * FUNCTION:    back split the binary buddy system of dmtree leaves
2664  *		that hold a specified leaf until the specified leaf
2665  *		starts its own binary buddy system.
2666  *
2667  *		the allocators typically perform allocations at the start
2668  *		of binary buddy systems and dbSplit() is used to accomplish
2669  *		any required splits.  in some cases, however, allocation
2670  *		may occur in the middle of a binary system and requires a
2671  *		back split, with the split proceeding out from the middle of
2672  *		the system (less efficient) rather than the start of the
2673  *		system (more efficient).  the cases in which a back split
2674  *		is required are rare and are limited to the first allocation
2675  *		within an allocation group which is a part (not first part)
2676  *		of a larger binary buddy system and a few exception cases
2677  *		in which a previous join operation must be backed out.
2678  *
2679  * PARAMETERS:
2680  *      tp	- pointer to the tree containing the leaf.
2681  *      leafno	- the number of the leaf to be updated.
2682  *
2683  * RETURN VALUES: none
2684  *
2685  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2686  */
2687 static void dbBackSplit(dmtree_t * tp, int leafno)
2688 {
2689 	int budsz, bud, w, bsz, size;
2690 	int cursz;
2691 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2692 
2693 	/* leaf should be part (not first part) of a binary
2694 	 * buddy system.
2695 	 */
2696 	assert(leaf[leafno] == NOFREE);
2697 
2698 	/* the back split is accomplished by iteratively finding the leaf
2699 	 * that starts the buddy system that contains the specified leaf and
2700 	 * splitting that system in two.  this iteration continues until
2701 	 * the specified leaf becomes the start of a buddy system.
2702 	 *
2703 	 * determine maximum possible l2 size for the specified leaf.
2704 	 */
2705 	size =
2706 	    LITOL2BSZ(leafno, le32_to_cpu(tp->dmt_l2nleafs),
2707 		      tp->dmt_budmin);
2708 
2709 	/* determine the number of leaves covered by this size.  this
2710 	 * is the buddy size that we will start with as we search for
2711 	 * the buddy system that contains the specified leaf.
2712 	 */
2713 	budsz = BUDSIZE(size, tp->dmt_budmin);
2714 
2715 	/* back split.
2716 	 */
2717 	while (leaf[leafno] == NOFREE) {
2718 		/* find the leftmost buddy leaf.
2719 		 */
2720 		for (w = leafno, bsz = budsz;; bsz <<= 1,
2721 		     w = (w < bud) ? w : bud) {
2722 			assert(bsz < le32_to_cpu(tp->dmt_nleafs));
2723 
2724 			/* determine the buddy.
2725 			 */
2726 			bud = w ^ bsz;
2727 
2728 			/* check if this buddy is the start of the system.
2729 			 */
2730 			if (leaf[bud] != NOFREE) {
2731 				/* split the leaf at the start of the
2732 				 * system in two.
2733 				 */
2734 				cursz = leaf[bud] - 1;
2735 				dbSplit(tp, bud, cursz, cursz);
2736 				break;
2737 			}
2738 		}
2739 	}
2740 
2741 	assert(leaf[leafno] == size);
2742 }
2743 
2744 
2745 /*
2746  * NAME:	dbJoin()
2747  *
2748  * FUNCTION:    update the leaf of a dmtree with a new value, joining
2749  *		the leaf with other leaves of the dmtree into a multi-leaf
2750  *		binary buddy system, as required.
2751  *
2752  * PARAMETERS:
2753  *      tp	- pointer to the tree containing the leaf.
2754  *      leafno	- the number of the leaf to be updated.
2755  *      newval	- the new value for the leaf.
2756  *
2757  * RETURN VALUES: none
2758  */
2759 static void dbJoin(dmtree_t * tp, int leafno, int newval)
2760 {
2761 	int budsz, buddy;
2762 	s8 *leaf;
2763 
2764 	/* can the new leaf value require a join with other leaves ?
2765 	 */
2766 	if (newval >= tp->dmt_budmin) {
2767 		/* pickup a pointer to the leaves of the tree.
2768 		 */
2769 		leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2770 
2771 		/* try to join the specified leaf into a large binary
2772 		 * buddy system.  the join proceeds by attempting to join
2773 		 * the specified leafno with its buddy (leaf) at new value.
2774 		 * if the join occurs, we attempt to join the left leaf
2775 		 * of the joined buddies with its buddy at new value + 1.
2776 		 * we continue to join until we find a buddy that cannot be
2777 		 * joined (does not have a value equal to the size of the
2778 		 * last join) or until all leaves have been joined into a
2779 		 * single system.
2780 		 *
2781 		 * get the buddy size (number of words covered) of
2782 		 * the new value.
2783 		 */
2784 		budsz = BUDSIZE(newval, tp->dmt_budmin);
2785 
2786 		/* try to join.
2787 		 */
2788 		while (budsz < le32_to_cpu(tp->dmt_nleafs)) {
2789 			/* get the buddy leaf.
2790 			 */
2791 			buddy = leafno ^ budsz;
2792 
2793 			/* if the leaf's new value is greater than its
2794 			 * buddy's value, we join no more.
2795 			 */
2796 			if (newval > leaf[buddy])
2797 				break;
2798 
2799 			assert(newval == leaf[buddy]);
2800 
2801 			/* check which (leafno or buddy) is the left buddy.
2802 			 * the left buddy gets to claim the blocks resulting
2803 			 * from the join while the right gets to claim none.
2804 			 * the left buddy is also eligable to participate in
2805 			 * a join at the next higher level while the right
2806 			 * is not.
2807 			 *
2808 			 */
2809 			if (leafno < buddy) {
2810 				/* leafno is the left buddy.
2811 				 */
2812 				dbAdjTree(tp, buddy, NOFREE);
2813 			} else {
2814 				/* buddy is the left buddy and becomes
2815 				 * leafno.
2816 				 */
2817 				dbAdjTree(tp, leafno, NOFREE);
2818 				leafno = buddy;
2819 			}
2820 
2821 			/* on to try the next join.
2822 			 */
2823 			newval += 1;
2824 			budsz <<= 1;
2825 		}
2826 	}
2827 
2828 	/* update the leaf value.
2829 	 */
2830 	dbAdjTree(tp, leafno, newval);
2831 }
2832 
2833 
2834 /*
2835  * NAME:	dbAdjTree()
2836  *
2837  * FUNCTION:    update a leaf of a dmtree with a new value, adjusting
2838  *		the dmtree, as required, to reflect the new leaf value.
2839  *		the combination of any buddies must already be done before
2840  *		this is called.
2841  *
2842  * PARAMETERS:
2843  *      tp	- pointer to the tree to be adjusted.
2844  *      leafno	- the number of the leaf to be updated.
2845  *      newval	- the new value for the leaf.
2846  *
2847  * RETURN VALUES: none
2848  */
2849 static void dbAdjTree(dmtree_t * tp, int leafno, int newval)
2850 {
2851 	int lp, pp, k;
2852 	int max;
2853 
2854 	/* pick up the index of the leaf for this leafno.
2855 	 */
2856 	lp = leafno + le32_to_cpu(tp->dmt_leafidx);
2857 
2858 	/* is the current value the same as the old value ?  if so,
2859 	 * there is nothing to do.
2860 	 */
2861 	if (tp->dmt_stree[lp] == newval)
2862 		return;
2863 
2864 	/* set the new value.
2865 	 */
2866 	tp->dmt_stree[lp] = newval;
2867 
2868 	/* bubble the new value up the tree as required.
2869 	 */
2870 	for (k = 0; k < le32_to_cpu(tp->dmt_height); k++) {
2871 		/* get the index of the first leaf of the 4 leaf
2872 		 * group containing the specified leaf (leafno).
2873 		 */
2874 		lp = ((lp - 1) & ~0x03) + 1;
2875 
2876 		/* get the index of the parent of this 4 leaf group.
2877 		 */
2878 		pp = (lp - 1) >> 2;
2879 
2880 		/* determine the maximum of the 4 leaves.
2881 		 */
2882 		max = TREEMAX(&tp->dmt_stree[lp]);
2883 
2884 		/* if the maximum of the 4 is the same as the
2885 		 * parent's value, we're done.
2886 		 */
2887 		if (tp->dmt_stree[pp] == max)
2888 			break;
2889 
2890 		/* parent gets new value.
2891 		 */
2892 		tp->dmt_stree[pp] = max;
2893 
2894 		/* parent becomes leaf for next go-round.
2895 		 */
2896 		lp = pp;
2897 	}
2898 }
2899 
2900 
2901 /*
2902  * NAME:	dbFindLeaf()
2903  *
2904  * FUNCTION:    search a dmtree_t for sufficient free blocks, returning
2905  *		the index of a leaf describing the free blocks if
2906  *		sufficient free blocks are found.
2907  *
2908  *		the search starts at the top of the dmtree_t tree and
2909  *		proceeds down the tree to the leftmost leaf with sufficient
2910  *		free space.
2911  *
2912  * PARAMETERS:
2913  *      tp	- pointer to the tree to be searched.
2914  *      l2nb	- log2 number of free blocks to search for.
2915  *	leafidx	- return pointer to be set to the index of the leaf
2916  *		  describing at least l2nb free blocks if sufficient
2917  *		  free blocks are found.
2918  *
2919  * RETURN VALUES:
2920  *      0	- success
2921  *      -ENOSPC	- insufficient free blocks.
2922  */
2923 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx)
2924 {
2925 	int ti, n = 0, k, x = 0;
2926 
2927 	/* first check the root of the tree to see if there is
2928 	 * sufficient free space.
2929 	 */
2930 	if (l2nb > tp->dmt_stree[ROOT])
2931 		return -ENOSPC;
2932 
2933 	/* sufficient free space available. now search down the tree
2934 	 * starting at the next level for the leftmost leaf that
2935 	 * describes sufficient free space.
2936 	 */
2937 	for (k = le32_to_cpu(tp->dmt_height), ti = 1;
2938 	     k > 0; k--, ti = ((ti + n) << 2) + 1) {
2939 		/* search the four nodes at this level, starting from
2940 		 * the left.
2941 		 */
2942 		for (x = ti, n = 0; n < 4; n++) {
2943 			/* sufficient free space found.  move to the next
2944 			 * level (or quit if this is the last level).
2945 			 */
2946 			if (l2nb <= tp->dmt_stree[x + n])
2947 				break;
2948 		}
2949 
2950 		/* better have found something since the higher
2951 		 * levels of the tree said it was here.
2952 		 */
2953 		assert(n < 4);
2954 	}
2955 
2956 	/* set the return to the leftmost leaf describing sufficient
2957 	 * free space.
2958 	 */
2959 	*leafidx = x + n - le32_to_cpu(tp->dmt_leafidx);
2960 
2961 	return (0);
2962 }
2963 
2964 
2965 /*
2966  * NAME:	dbFindBits()
2967  *
2968  * FUNCTION:    find a specified number of binary buddy free bits within a
2969  *		dmap bitmap word value.
2970  *
2971  *		this routine searches the bitmap value for (1 << l2nb) free
2972  *		bits at (1 << l2nb) alignments within the value.
2973  *
2974  * PARAMETERS:
2975  *      word	-  dmap bitmap word value.
2976  *      l2nb	-  number of free bits specified as a log2 number.
2977  *
2978  * RETURN VALUES:
2979  *      starting bit number of free bits.
2980  */
2981 static int dbFindBits(u32 word, int l2nb)
2982 {
2983 	int bitno, nb;
2984 	u32 mask;
2985 
2986 	/* get the number of bits.
2987 	 */
2988 	nb = 1 << l2nb;
2989 	assert(nb <= DBWORD);
2990 
2991 	/* complement the word so we can use a mask (i.e. 0s represent
2992 	 * free bits) and compute the mask.
2993 	 */
2994 	word = ~word;
2995 	mask = ONES << (DBWORD - nb);
2996 
2997 	/* scan the word for nb free bits at nb alignments.
2998 	 */
2999 	for (bitno = 0; mask != 0; bitno += nb, mask >>= nb) {
3000 		if ((mask & word) == mask)
3001 			break;
3002 	}
3003 
3004 	ASSERT(bitno < 32);
3005 
3006 	/* return the bit number.
3007 	 */
3008 	return (bitno);
3009 }
3010 
3011 
3012 /*
3013  * NAME:	dbMaxBud(u8 *cp)
3014  *
3015  * FUNCTION:    determine the largest binary buddy string of free
3016  *		bits within 32-bits of the map.
3017  *
3018  * PARAMETERS:
3019  *      cp	-  pointer to the 32-bit value.
3020  *
3021  * RETURN VALUES:
3022  *      largest binary buddy of free bits within a dmap word.
3023  */
3024 static int dbMaxBud(u8 * cp)
3025 {
3026 	signed char tmp1, tmp2;
3027 
3028 	/* check if the wmap word is all free. if so, the
3029 	 * free buddy size is BUDMIN.
3030 	 */
3031 	if (*((uint *) cp) == 0)
3032 		return (BUDMIN);
3033 
3034 	/* check if the wmap word is half free. if so, the
3035 	 * free buddy size is BUDMIN-1.
3036 	 */
3037 	if (*((u16 *) cp) == 0 || *((u16 *) cp + 1) == 0)
3038 		return (BUDMIN - 1);
3039 
3040 	/* not all free or half free. determine the free buddy
3041 	 * size thru table lookup using quarters of the wmap word.
3042 	 */
3043 	tmp1 = max(budtab[cp[2]], budtab[cp[3]]);
3044 	tmp2 = max(budtab[cp[0]], budtab[cp[1]]);
3045 	return (max(tmp1, tmp2));
3046 }
3047 
3048 
3049 /*
3050  * NAME:	cnttz(uint word)
3051  *
3052  * FUNCTION:    determine the number of trailing zeros within a 32-bit
3053  *		value.
3054  *
3055  * PARAMETERS:
3056  *      value	-  32-bit value to be examined.
3057  *
3058  * RETURN VALUES:
3059  *      count of trailing zeros
3060  */
3061 static int cnttz(u32 word)
3062 {
3063 	int n;
3064 
3065 	for (n = 0; n < 32; n++, word >>= 1) {
3066 		if (word & 0x01)
3067 			break;
3068 	}
3069 
3070 	return (n);
3071 }
3072 
3073 
3074 /*
3075  * NAME:	cntlz(u32 value)
3076  *
3077  * FUNCTION:    determine the number of leading zeros within a 32-bit
3078  *		value.
3079  *
3080  * PARAMETERS:
3081  *      value	-  32-bit value to be examined.
3082  *
3083  * RETURN VALUES:
3084  *      count of leading zeros
3085  */
3086 static int cntlz(u32 value)
3087 {
3088 	int n;
3089 
3090 	for (n = 0; n < 32; n++, value <<= 1) {
3091 		if (value & HIGHORDER)
3092 			break;
3093 	}
3094 	return (n);
3095 }
3096 
3097 
3098 /*
3099  * NAME:	blkstol2(s64 nb)
3100  *
3101  * FUNCTION:	convert a block count to its log2 value. if the block
3102  *	        count is not a l2 multiple, it is rounded up to the next
3103  *		larger l2 multiple.
3104  *
3105  * PARAMETERS:
3106  *      nb	-  number of blocks
3107  *
3108  * RETURN VALUES:
3109  *      log2 number of blocks
3110  */
3111 int blkstol2(s64 nb)
3112 {
3113 	int l2nb;
3114 	s64 mask;		/* meant to be signed */
3115 
3116 	mask = (s64) 1 << (64 - 1);
3117 
3118 	/* count the leading bits.
3119 	 */
3120 	for (l2nb = 0; l2nb < 64; l2nb++, mask >>= 1) {
3121 		/* leading bit found.
3122 		 */
3123 		if (nb & mask) {
3124 			/* determine the l2 value.
3125 			 */
3126 			l2nb = (64 - 1) - l2nb;
3127 
3128 			/* check if we need to round up.
3129 			 */
3130 			if (~mask & nb)
3131 				l2nb++;
3132 
3133 			return (l2nb);
3134 		}
3135 	}
3136 	assert(0);
3137 	return 0;		/* fix compiler warning */
3138 }
3139 
3140 
3141 /*
3142  * NAME:    	dbAllocBottomUp()
3143  *
3144  * FUNCTION:	alloc the specified block range from the working block
3145  *		allocation map.
3146  *
3147  *		the blocks will be alloc from the working map one dmap
3148  *		at a time.
3149  *
3150  * PARAMETERS:
3151  *      ip	-  pointer to in-core inode;
3152  *      blkno	-  starting block number to be freed.
3153  *      nblocks	-  number of blocks to be freed.
3154  *
3155  * RETURN VALUES:
3156  *      0	- success
3157  *      -EIO	- i/o error
3158  */
3159 int dbAllocBottomUp(struct inode *ip, s64 blkno, s64 nblocks)
3160 {
3161 	struct metapage *mp;
3162 	struct dmap *dp;
3163 	int nb, rc;
3164 	s64 lblkno, rem;
3165 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
3166 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
3167 
3168 	IREAD_LOCK(ipbmap);
3169 
3170 	/* block to be allocated better be within the mapsize. */
3171 	ASSERT(nblocks <= bmp->db_mapsize - blkno);
3172 
3173 	/*
3174 	 * allocate the blocks a dmap at a time.
3175 	 */
3176 	mp = NULL;
3177 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
3178 		/* release previous dmap if any */
3179 		if (mp) {
3180 			write_metapage(mp);
3181 		}
3182 
3183 		/* get the buffer for the current dmap. */
3184 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
3185 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
3186 		if (mp == NULL) {
3187 			IREAD_UNLOCK(ipbmap);
3188 			return -EIO;
3189 		}
3190 		dp = (struct dmap *) mp->data;
3191 
3192 		/* determine the number of blocks to be allocated from
3193 		 * this dmap.
3194 		 */
3195 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
3196 
3197 		DBFREECK(bmp->db_DBmap, bmp->db_mapsize, blkno, nb);
3198 
3199 		/* allocate the blocks. */
3200 		if ((rc = dbAllocDmapBU(bmp, dp, blkno, nb))) {
3201 			release_metapage(mp);
3202 			IREAD_UNLOCK(ipbmap);
3203 			return (rc);
3204 		}
3205 
3206 		DBALLOC(bmp->db_DBmap, bmp->db_mapsize, blkno, nb);
3207 	}
3208 
3209 	/* write the last buffer. */
3210 	write_metapage(mp);
3211 
3212 	IREAD_UNLOCK(ipbmap);
3213 
3214 	return (0);
3215 }
3216 
3217 
3218 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
3219 			 int nblocks)
3220 {
3221 	int rc;
3222 	int dbitno, word, rembits, nb, nwords, wbitno, agno;
3223 	s8 oldroot, *leaf;
3224 	struct dmaptree *tp = (struct dmaptree *) & dp->tree;
3225 
3226 	/* save the current value of the root (i.e. maximum free string)
3227 	 * of the dmap tree.
3228 	 */
3229 	oldroot = tp->stree[ROOT];
3230 
3231 	/* pick up a pointer to the leaves of the dmap tree */
3232 	leaf = tp->stree + LEAFIND;
3233 
3234 	/* determine the bit number and word within the dmap of the
3235 	 * starting block.
3236 	 */
3237 	dbitno = blkno & (BPERDMAP - 1);
3238 	word = dbitno >> L2DBWORD;
3239 
3240 	/* block range better be within the dmap */
3241 	assert(dbitno + nblocks <= BPERDMAP);
3242 
3243 	/* allocate the bits of the dmap's words corresponding to the block
3244 	 * range. not all bits of the first and last words may be contained
3245 	 * within the block range.  if this is the case, we'll work against
3246 	 * those words (i.e. partial first and/or last) on an individual basis
3247 	 * (a single pass), allocating the bits of interest by hand and
3248 	 * updating the leaf corresponding to the dmap word. a single pass
3249 	 * will be used for all dmap words fully contained within the
3250 	 * specified range.  within this pass, the bits of all fully contained
3251 	 * dmap words will be marked as free in a single shot and the leaves
3252 	 * will be updated. a single leaf may describe the free space of
3253 	 * multiple dmap words, so we may update only a subset of the actual
3254 	 * leaves corresponding to the dmap words of the block range.
3255 	 */
3256 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
3257 		/* determine the bit number within the word and
3258 		 * the number of bits within the word.
3259 		 */
3260 		wbitno = dbitno & (DBWORD - 1);
3261 		nb = min(rembits, DBWORD - wbitno);
3262 
3263 		/* check if only part of a word is to be allocated.
3264 		 */
3265 		if (nb < DBWORD) {
3266 			/* allocate (set to 1) the appropriate bits within
3267 			 * this dmap word.
3268 			 */
3269 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
3270 						      >> wbitno);
3271 
3272 			word++;
3273 		} else {
3274 			/* one or more dmap words are fully contained
3275 			 * within the block range.  determine how many
3276 			 * words and allocate (set to 1) the bits of these
3277 			 * words.
3278 			 */
3279 			nwords = rembits >> L2DBWORD;
3280 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
3281 
3282 			/* determine how many bits */
3283 			nb = nwords << L2DBWORD;
3284 			word += nwords;
3285 		}
3286 	}
3287 
3288 	/* update the free count for this dmap */
3289 	dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) - nblocks);
3290 
3291 	/* reconstruct summary tree */
3292 	dbInitDmapTree(dp);
3293 
3294 	BMAP_LOCK(bmp);
3295 
3296 	/* if this allocation group is completely free,
3297 	 * update the highest active allocation group number
3298 	 * if this allocation group is the new max.
3299 	 */
3300 	agno = blkno >> bmp->db_agl2size;
3301 	if (agno > bmp->db_maxag)
3302 		bmp->db_maxag = agno;
3303 
3304 	/* update the free count for the allocation group and map */
3305 	bmp->db_agfree[agno] -= nblocks;
3306 	bmp->db_nfree -= nblocks;
3307 
3308 	BMAP_UNLOCK(bmp);
3309 
3310 	/* if the root has not changed, done. */
3311 	if (tp->stree[ROOT] == oldroot)
3312 		return (0);
3313 
3314 	/* root changed. bubble the change up to the dmap control pages.
3315 	 * if the adjustment of the upper level control pages fails,
3316 	 * backout the bit allocation (thus making everything consistent).
3317 	 */
3318 	if ((rc = dbAdjCtl(bmp, blkno, tp->stree[ROOT], 1, 0)))
3319 		dbFreeBits(bmp, dp, blkno, nblocks);
3320 
3321 	return (rc);
3322 }
3323 
3324 
3325 /*
3326  * NAME:	dbExtendFS()
3327  *
3328  * FUNCTION:	extend bmap from blkno for nblocks;
3329  * 		dbExtendFS() updates bmap ready for dbAllocBottomUp();
3330  *
3331  * L2
3332  *  |
3333  *   L1---------------------------------L1
3334  *    |                                  |
3335  *     L0---------L0---------L0           L0---------L0---------L0
3336  *      |          |          |            |          |          |
3337  *       d0,...,dn  d0,...,dn  d0,...,dn    d0,...,dn  d0,...,dn  d0,.,dm;
3338  * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3339  *
3340  * <---old---><----------------------------extend----------------------->
3341  */
3342 int dbExtendFS(struct inode *ipbmap, s64 blkno,	s64 nblocks)
3343 {
3344 	struct jfs_sb_info *sbi = JFS_SBI(ipbmap->i_sb);
3345 	int nbperpage = sbi->nbperpage;
3346 	int i, i0 = TRUE, j, j0 = TRUE, k, n;
3347 	s64 newsize;
3348 	s64 p;
3349 	struct metapage *mp, *l2mp, *l1mp = NULL, *l0mp = NULL;
3350 	struct dmapctl *l2dcp, *l1dcp, *l0dcp;
3351 	struct dmap *dp;
3352 	s8 *l0leaf, *l1leaf, *l2leaf;
3353 	struct bmap *bmp = sbi->bmap;
3354 	int agno, l2agsize, oldl2agsize;
3355 	s64 ag_rem;
3356 
3357 	newsize = blkno + nblocks;
3358 
3359 	jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld",
3360 		 (long long) blkno, (long long) nblocks, (long long) newsize);
3361 
3362 	/*
3363 	 *      initialize bmap control page.
3364 	 *
3365 	 * all the data in bmap control page should exclude
3366 	 * the mkfs hidden dmap page.
3367 	 */
3368 
3369 	/* update mapsize */
3370 	bmp->db_mapsize = newsize;
3371 	bmp->db_maxlevel = BMAPSZTOLEV(bmp->db_mapsize);
3372 
3373 	/* compute new AG size */
3374 	l2agsize = dbGetL2AGSize(newsize);
3375 	oldl2agsize = bmp->db_agl2size;
3376 
3377 	bmp->db_agl2size = l2agsize;
3378 	bmp->db_agsize = 1 << l2agsize;
3379 
3380 	/* compute new number of AG */
3381 	agno = bmp->db_numag;
3382 	bmp->db_numag = newsize >> l2agsize;
3383 	bmp->db_numag += ((u32) newsize % (u32) bmp->db_agsize) ? 1 : 0;
3384 
3385 	/*
3386 	 *      reconfigure db_agfree[]
3387 	 * from old AG configuration to new AG configuration;
3388 	 *
3389 	 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3390 	 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3391 	 * note: new AG size = old AG size * (2**x).
3392 	 */
3393 	if (l2agsize == oldl2agsize)
3394 		goto extend;
3395 	k = 1 << (l2agsize - oldl2agsize);
3396 	ag_rem = bmp->db_agfree[0];	/* save agfree[0] */
3397 	for (i = 0, n = 0; i < agno; n++) {
3398 		bmp->db_agfree[n] = 0;	/* init collection point */
3399 
3400 		/* coalesce cotiguous k AGs; */
3401 		for (j = 0; j < k && i < agno; j++, i++) {
3402 			/* merge AGi to AGn */
3403 			bmp->db_agfree[n] += bmp->db_agfree[i];
3404 		}
3405 	}
3406 	bmp->db_agfree[0] += ag_rem;	/* restore agfree[0] */
3407 
3408 	for (; n < MAXAG; n++)
3409 		bmp->db_agfree[n] = 0;
3410 
3411 	/*
3412 	 * update highest active ag number
3413 	 */
3414 
3415 	bmp->db_maxag = bmp->db_maxag / k;
3416 
3417 	/*
3418 	 *      extend bmap
3419 	 *
3420 	 * update bit maps and corresponding level control pages;
3421 	 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3422 	 */
3423       extend:
3424 	/* get L2 page */
3425 	p = BMAPBLKNO + nbperpage;	/* L2 page */
3426 	l2mp = read_metapage(ipbmap, p, PSIZE, 0);
3427 	if (!l2mp) {
3428 		jfs_error(ipbmap->i_sb, "dbExtendFS: L2 page could not be read");
3429 		return -EIO;
3430 	}
3431 	l2dcp = (struct dmapctl *) l2mp->data;
3432 
3433 	/* compute start L1 */
3434 	k = blkno >> L2MAXL1SIZE;
3435 	l2leaf = l2dcp->stree + CTLLEAFIND + k;
3436 	p = BLKTOL1(blkno, sbi->l2nbperpage);	/* L1 page */
3437 
3438 	/*
3439 	 * extend each L1 in L2
3440 	 */
3441 	for (; k < LPERCTL; k++, p += nbperpage) {
3442 		/* get L1 page */
3443 		if (j0) {
3444 			/* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3445 			l1mp = read_metapage(ipbmap, p, PSIZE, 0);
3446 			if (l1mp == NULL)
3447 				goto errout;
3448 			l1dcp = (struct dmapctl *) l1mp->data;
3449 
3450 			/* compute start L0 */
3451 			j = (blkno & (MAXL1SIZE - 1)) >> L2MAXL0SIZE;
3452 			l1leaf = l1dcp->stree + CTLLEAFIND + j;
3453 			p = BLKTOL0(blkno, sbi->l2nbperpage);
3454 			j0 = FALSE;
3455 		} else {
3456 			/* assign/init L1 page */
3457 			l1mp = get_metapage(ipbmap, p, PSIZE, 0);
3458 			if (l1mp == NULL)
3459 				goto errout;
3460 
3461 			l1dcp = (struct dmapctl *) l1mp->data;
3462 
3463 			/* compute start L0 */
3464 			j = 0;
3465 			l1leaf = l1dcp->stree + CTLLEAFIND;
3466 			p += nbperpage;	/* 1st L0 of L1.k  */
3467 		}
3468 
3469 		/*
3470 		 * extend each L0 in L1
3471 		 */
3472 		for (; j < LPERCTL; j++) {
3473 			/* get L0 page */
3474 			if (i0) {
3475 				/* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3476 
3477 				l0mp = read_metapage(ipbmap, p, PSIZE, 0);
3478 				if (l0mp == NULL)
3479 					goto errout;
3480 				l0dcp = (struct dmapctl *) l0mp->data;
3481 
3482 				/* compute start dmap */
3483 				i = (blkno & (MAXL0SIZE - 1)) >>
3484 				    L2BPERDMAP;
3485 				l0leaf = l0dcp->stree + CTLLEAFIND + i;
3486 				p = BLKTODMAP(blkno,
3487 					      sbi->l2nbperpage);
3488 				i0 = FALSE;
3489 			} else {
3490 				/* assign/init L0 page */
3491 				l0mp = get_metapage(ipbmap, p, PSIZE, 0);
3492 				if (l0mp == NULL)
3493 					goto errout;
3494 
3495 				l0dcp = (struct dmapctl *) l0mp->data;
3496 
3497 				/* compute start dmap */
3498 				i = 0;
3499 				l0leaf = l0dcp->stree + CTLLEAFIND;
3500 				p += nbperpage;	/* 1st dmap of L0.j */
3501 			}
3502 
3503 			/*
3504 			 * extend each dmap in L0
3505 			 */
3506 			for (; i < LPERCTL; i++) {
3507 				/*
3508 				 * reconstruct the dmap page, and
3509 				 * initialize corresponding parent L0 leaf
3510 				 */
3511 				if ((n = blkno & (BPERDMAP - 1))) {
3512 					/* read in dmap page: */
3513 					mp = read_metapage(ipbmap, p,
3514 							   PSIZE, 0);
3515 					if (mp == NULL)
3516 						goto errout;
3517 					n = min(nblocks, (s64)BPERDMAP - n);
3518 				} else {
3519 					/* assign/init dmap page */
3520 					mp = read_metapage(ipbmap, p,
3521 							   PSIZE, 0);
3522 					if (mp == NULL)
3523 						goto errout;
3524 
3525 					n = min(nblocks, (s64)BPERDMAP);
3526 				}
3527 
3528 				dp = (struct dmap *) mp->data;
3529 				*l0leaf = dbInitDmap(dp, blkno, n);
3530 
3531 				bmp->db_nfree += n;
3532 				agno = le64_to_cpu(dp->start) >> l2agsize;
3533 				bmp->db_agfree[agno] += n;
3534 
3535 				write_metapage(mp);
3536 
3537 				l0leaf++;
3538 				p += nbperpage;
3539 
3540 				blkno += n;
3541 				nblocks -= n;
3542 				if (nblocks == 0)
3543 					break;
3544 			}	/* for each dmap in a L0 */
3545 
3546 			/*
3547 			 * build current L0 page from its leaves, and
3548 			 * initialize corresponding parent L1 leaf
3549 			 */
3550 			*l1leaf = dbInitDmapCtl(l0dcp, 0, ++i);
3551 			write_metapage(l0mp);
3552 			l0mp = NULL;
3553 
3554 			if (nblocks)
3555 				l1leaf++;	/* continue for next L0 */
3556 			else {
3557 				/* more than 1 L0 ? */
3558 				if (j > 0)
3559 					break;	/* build L1 page */
3560 				else {
3561 					/* summarize in global bmap page */
3562 					bmp->db_maxfreebud = *l1leaf;
3563 					release_metapage(l1mp);
3564 					release_metapage(l2mp);
3565 					goto finalize;
3566 				}
3567 			}
3568 		}		/* for each L0 in a L1 */
3569 
3570 		/*
3571 		 * build current L1 page from its leaves, and
3572 		 * initialize corresponding parent L2 leaf
3573 		 */
3574 		*l2leaf = dbInitDmapCtl(l1dcp, 1, ++j);
3575 		write_metapage(l1mp);
3576 		l1mp = NULL;
3577 
3578 		if (nblocks)
3579 			l2leaf++;	/* continue for next L1 */
3580 		else {
3581 			/* more than 1 L1 ? */
3582 			if (k > 0)
3583 				break;	/* build L2 page */
3584 			else {
3585 				/* summarize in global bmap page */
3586 				bmp->db_maxfreebud = *l2leaf;
3587 				release_metapage(l2mp);
3588 				goto finalize;
3589 			}
3590 		}
3591 	}			/* for each L1 in a L2 */
3592 
3593 	jfs_error(ipbmap->i_sb,
3594 		  "dbExtendFS: function has not returned as expected");
3595 errout:
3596 	if (l0mp)
3597 		release_metapage(l0mp);
3598 	if (l1mp)
3599 		release_metapage(l1mp);
3600 	release_metapage(l2mp);
3601 	return -EIO;
3602 
3603 	/*
3604 	 *      finalize bmap control page
3605 	 */
3606 finalize:
3607 
3608 	return 0;
3609 }
3610 
3611 
3612 /*
3613  *	dbFinalizeBmap()
3614  */
3615 void dbFinalizeBmap(struct inode *ipbmap)
3616 {
3617 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
3618 	int actags, inactags, l2nl;
3619 	s64 ag_rem, actfree, inactfree, avgfree;
3620 	int i, n;
3621 
3622 	/*
3623 	 *      finalize bmap control page
3624 	 */
3625 //finalize:
3626 	/*
3627 	 * compute db_agpref: preferred ag to allocate from
3628 	 * (the leftmost ag with average free space in it);
3629 	 */
3630 //agpref:
3631 	/* get the number of active ags and inacitve ags */
3632 	actags = bmp->db_maxag + 1;
3633 	inactags = bmp->db_numag - actags;
3634 	ag_rem = bmp->db_mapsize & (bmp->db_agsize - 1);	/* ??? */
3635 
3636 	/* determine how many blocks are in the inactive allocation
3637 	 * groups. in doing this, we must account for the fact that
3638 	 * the rightmost group might be a partial group (i.e. file
3639 	 * system size is not a multiple of the group size).
3640 	 */
3641 	inactfree = (inactags && ag_rem) ?
3642 	    ((inactags - 1) << bmp->db_agl2size) + ag_rem
3643 	    : inactags << bmp->db_agl2size;
3644 
3645 	/* determine how many free blocks are in the active
3646 	 * allocation groups plus the average number of free blocks
3647 	 * within the active ags.
3648 	 */
3649 	actfree = bmp->db_nfree - inactfree;
3650 	avgfree = (u32) actfree / (u32) actags;
3651 
3652 	/* if the preferred allocation group has not average free space.
3653 	 * re-establish the preferred group as the leftmost
3654 	 * group with average free space.
3655 	 */
3656 	if (bmp->db_agfree[bmp->db_agpref] < avgfree) {
3657 		for (bmp->db_agpref = 0; bmp->db_agpref < actags;
3658 		     bmp->db_agpref++) {
3659 			if (bmp->db_agfree[bmp->db_agpref] >= avgfree)
3660 				break;
3661 		}
3662 		if (bmp->db_agpref >= bmp->db_numag) {
3663 			jfs_error(ipbmap->i_sb,
3664 				  "cannot find ag with average freespace");
3665 		}
3666 	}
3667 
3668 	/*
3669 	 * compute db_aglevel, db_agheigth, db_width, db_agstart:
3670 	 * an ag is covered in aglevel dmapctl summary tree,
3671 	 * at agheight level height (from leaf) with agwidth number of nodes
3672 	 * each, which starts at agstart index node of the smmary tree node
3673 	 * array;
3674 	 */
3675 	bmp->db_aglevel = BMAPSZTOLEV(bmp->db_agsize);
3676 	l2nl =
3677 	    bmp->db_agl2size - (L2BPERDMAP + bmp->db_aglevel * L2LPERCTL);
3678 	bmp->db_agheigth = l2nl >> 1;
3679 	bmp->db_agwidth = 1 << (l2nl - (bmp->db_agheigth << 1));
3680 	for (i = 5 - bmp->db_agheigth, bmp->db_agstart = 0, n = 1; i > 0;
3681 	     i--) {
3682 		bmp->db_agstart += n;
3683 		n <<= 2;
3684 	}
3685 
3686 }
3687 
3688 
3689 /*
3690  * NAME:	dbInitDmap()/ujfs_idmap_page()
3691  *
3692  * FUNCTION:	initialize working/persistent bitmap of the dmap page
3693  *		for the specified number of blocks:
3694  *
3695  *		at entry, the bitmaps had been initialized as free (ZEROS);
3696  *		The number of blocks will only account for the actually
3697  *		existing blocks. Blocks which don't actually exist in
3698  *		the aggregate will be marked as allocated (ONES);
3699  *
3700  * PARAMETERS:
3701  *	dp	- pointer to page of map
3702  *	nblocks	- number of blocks this page
3703  *
3704  * RETURNS: NONE
3705  */
3706 static int dbInitDmap(struct dmap * dp, s64 Blkno, int nblocks)
3707 {
3708 	int blkno, w, b, r, nw, nb, i;
3709 
3710 	/* starting block number within the dmap */
3711 	blkno = Blkno & (BPERDMAP - 1);
3712 
3713 	if (blkno == 0) {
3714 		dp->nblocks = dp->nfree = cpu_to_le32(nblocks);
3715 		dp->start = cpu_to_le64(Blkno);
3716 
3717 		if (nblocks == BPERDMAP) {
3718 			memset(&dp->wmap[0], 0, LPERDMAP * 4);
3719 			memset(&dp->pmap[0], 0, LPERDMAP * 4);
3720 			goto initTree;
3721 		}
3722 	} else {
3723 		dp->nblocks =
3724 		    cpu_to_le32(le32_to_cpu(dp->nblocks) + nblocks);
3725 		dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) + nblocks);
3726 	}
3727 
3728 	/* word number containing start block number */
3729 	w = blkno >> L2DBWORD;
3730 
3731 	/*
3732 	 * free the bits corresponding to the block range (ZEROS):
3733 	 * note: not all bits of the first and last words may be contained
3734 	 * within the block range.
3735 	 */
3736 	for (r = nblocks; r > 0; r -= nb, blkno += nb) {
3737 		/* number of bits preceding range to be freed in the word */
3738 		b = blkno & (DBWORD - 1);
3739 		/* number of bits to free in the word */
3740 		nb = min(r, DBWORD - b);
3741 
3742 		/* is partial word to be freed ? */
3743 		if (nb < DBWORD) {
3744 			/* free (set to 0) from the bitmap word */
3745 			dp->wmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3746 						     >> b));
3747 			dp->pmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3748 						     >> b));
3749 
3750 			/* skip the word freed */
3751 			w++;
3752 		} else {
3753 			/* free (set to 0) contiguous bitmap words */
3754 			nw = r >> L2DBWORD;
3755 			memset(&dp->wmap[w], 0, nw * 4);
3756 			memset(&dp->pmap[w], 0, nw * 4);
3757 
3758 			/* skip the words freed */
3759 			nb = nw << L2DBWORD;
3760 			w += nw;
3761 		}
3762 	}
3763 
3764 	/*
3765 	 * mark bits following the range to be freed (non-existing
3766 	 * blocks) as allocated (ONES)
3767 	 */
3768 
3769 	if (blkno == BPERDMAP)
3770 		goto initTree;
3771 
3772 	/* the first word beyond the end of existing blocks */
3773 	w = blkno >> L2DBWORD;
3774 
3775 	/* does nblocks fall on a 32-bit boundary ? */
3776 	b = blkno & (DBWORD - 1);
3777 	if (b) {
3778 		/* mark a partial word allocated */
3779 		dp->wmap[w] = dp->pmap[w] = cpu_to_le32(ONES >> b);
3780 		w++;
3781 	}
3782 
3783 	/* set the rest of the words in the page to allocated (ONES) */
3784 	for (i = w; i < LPERDMAP; i++)
3785 		dp->pmap[i] = dp->wmap[i] = cpu_to_le32(ONES);
3786 
3787 	/*
3788 	 * init tree
3789 	 */
3790       initTree:
3791 	return (dbInitDmapTree(dp));
3792 }
3793 
3794 
3795 /*
3796  * NAME:	dbInitDmapTree()/ujfs_complete_dmap()
3797  *
3798  * FUNCTION:	initialize summary tree of the specified dmap:
3799  *
3800  *		at entry, bitmap of the dmap has been initialized;
3801  *
3802  * PARAMETERS:
3803  *	dp	- dmap to complete
3804  *	blkno	- starting block number for this dmap
3805  *	treemax	- will be filled in with max free for this dmap
3806  *
3807  * RETURNS:	max free string at the root of the tree
3808  */
3809 static int dbInitDmapTree(struct dmap * dp)
3810 {
3811 	struct dmaptree *tp;
3812 	s8 *cp;
3813 	int i;
3814 
3815 	/* init fixed info of tree */
3816 	tp = &dp->tree;
3817 	tp->nleafs = cpu_to_le32(LPERDMAP);
3818 	tp->l2nleafs = cpu_to_le32(L2LPERDMAP);
3819 	tp->leafidx = cpu_to_le32(LEAFIND);
3820 	tp->height = cpu_to_le32(4);
3821 	tp->budmin = BUDMIN;
3822 
3823 	/* init each leaf from corresponding wmap word:
3824 	 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3825 	 * bitmap word are allocated.
3826 	 */
3827 	cp = tp->stree + le32_to_cpu(tp->leafidx);
3828 	for (i = 0; i < LPERDMAP; i++)
3829 		*cp++ = dbMaxBud((u8 *) & dp->wmap[i]);
3830 
3831 	/* build the dmap's binary buddy summary tree */
3832 	return (dbInitTree(tp));
3833 }
3834 
3835 
3836 /*
3837  * NAME:	dbInitTree()/ujfs_adjtree()
3838  *
3839  * FUNCTION:	initialize binary buddy summary tree of a dmap or dmapctl.
3840  *
3841  *		at entry, the leaves of the tree has been initialized
3842  *		from corresponding bitmap word or root of summary tree
3843  *		of the child control page;
3844  *		configure binary buddy system at the leaf level, then
3845  *		bubble up the values of the leaf nodes up the tree.
3846  *
3847  * PARAMETERS:
3848  *	cp	- Pointer to the root of the tree
3849  *	l2leaves- Number of leaf nodes as a power of 2
3850  *	l2min	- Number of blocks that can be covered by a leaf
3851  *		  as a power of 2
3852  *
3853  * RETURNS: max free string at the root of the tree
3854  */
3855 static int dbInitTree(struct dmaptree * dtp)
3856 {
3857 	int l2max, l2free, bsize, nextb, i;
3858 	int child, parent, nparent;
3859 	s8 *tp, *cp, *cp1;
3860 
3861 	tp = dtp->stree;
3862 
3863 	/* Determine the maximum free string possible for the leaves */
3864 	l2max = le32_to_cpu(dtp->l2nleafs) + dtp->budmin;
3865 
3866 	/*
3867 	 * configure the leaf levevl into binary buddy system
3868 	 *
3869 	 * Try to combine buddies starting with a buddy size of 1
3870 	 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3871 	 * can be combined if both buddies have a maximum free of l2min;
3872 	 * the combination will result in the left-most buddy leaf having
3873 	 * a maximum free of l2min+1.
3874 	 * After processing all buddies for a given size, process buddies
3875 	 * at the next higher buddy size (i.e. current size * 2) and
3876 	 * the next maximum free (current free + 1).
3877 	 * This continues until the maximum possible buddy combination
3878 	 * yields maximum free.
3879 	 */
3880 	for (l2free = dtp->budmin, bsize = 1; l2free < l2max;
3881 	     l2free++, bsize = nextb) {
3882 		/* get next buddy size == current buddy pair size */
3883 		nextb = bsize << 1;
3884 
3885 		/* scan each adjacent buddy pair at current buddy size */
3886 		for (i = 0, cp = tp + le32_to_cpu(dtp->leafidx);
3887 		     i < le32_to_cpu(dtp->nleafs);
3888 		     i += nextb, cp += nextb) {
3889 			/* coalesce if both adjacent buddies are max free */
3890 			if (*cp == l2free && *(cp + bsize) == l2free) {
3891 				*cp = l2free + 1;	/* left take right */
3892 				*(cp + bsize) = -1;	/* right give left */
3893 			}
3894 		}
3895 	}
3896 
3897 	/*
3898 	 * bubble summary information of leaves up the tree.
3899 	 *
3900 	 * Starting at the leaf node level, the four nodes described by
3901 	 * the higher level parent node are compared for a maximum free and
3902 	 * this maximum becomes the value of the parent node.
3903 	 * when all lower level nodes are processed in this fashion then
3904 	 * move up to the next level (parent becomes a lower level node) and
3905 	 * continue the process for that level.
3906 	 */
3907 	for (child = le32_to_cpu(dtp->leafidx),
3908 	     nparent = le32_to_cpu(dtp->nleafs) >> 2;
3909 	     nparent > 0; nparent >>= 2, child = parent) {
3910 		/* get index of 1st node of parent level */
3911 		parent = (child - 1) >> 2;
3912 
3913 		/* set the value of the parent node as the maximum
3914 		 * of the four nodes of the current level.
3915 		 */
3916 		for (i = 0, cp = tp + child, cp1 = tp + parent;
3917 		     i < nparent; i++, cp += 4, cp1++)
3918 			*cp1 = TREEMAX(cp);
3919 	}
3920 
3921 	return (*tp);
3922 }
3923 
3924 
3925 /*
3926  *	dbInitDmapCtl()
3927  *
3928  * function: initialize dmapctl page
3929  */
3930 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i)
3931 {				/* start leaf index not covered by range */
3932 	s8 *cp;
3933 
3934 	dcp->nleafs = cpu_to_le32(LPERCTL);
3935 	dcp->l2nleafs = cpu_to_le32(L2LPERCTL);
3936 	dcp->leafidx = cpu_to_le32(CTLLEAFIND);
3937 	dcp->height = cpu_to_le32(5);
3938 	dcp->budmin = L2BPERDMAP + L2LPERCTL * level;
3939 
3940 	/*
3941 	 * initialize the leaves of current level that were not covered
3942 	 * by the specified input block range (i.e. the leaves have no
3943 	 * low level dmapctl or dmap).
3944 	 */
3945 	cp = &dcp->stree[CTLLEAFIND + i];
3946 	for (; i < LPERCTL; i++)
3947 		*cp++ = NOFREE;
3948 
3949 	/* build the dmap's binary buddy summary tree */
3950 	return (dbInitTree((struct dmaptree *) dcp));
3951 }
3952 
3953 
3954 /*
3955  * NAME:	dbGetL2AGSize()/ujfs_getagl2size()
3956  *
3957  * FUNCTION:	Determine log2(allocation group size) from aggregate size
3958  *
3959  * PARAMETERS:
3960  *	nblocks	- Number of blocks in aggregate
3961  *
3962  * RETURNS: log2(allocation group size) in aggregate blocks
3963  */
3964 static int dbGetL2AGSize(s64 nblocks)
3965 {
3966 	s64 sz;
3967 	s64 m;
3968 	int l2sz;
3969 
3970 	if (nblocks < BPERDMAP * MAXAG)
3971 		return (L2BPERDMAP);
3972 
3973 	/* round up aggregate size to power of 2 */
3974 	m = ((u64) 1 << (64 - 1));
3975 	for (l2sz = 64; l2sz >= 0; l2sz--, m >>= 1) {
3976 		if (m & nblocks)
3977 			break;
3978 	}
3979 
3980 	sz = (s64) 1 << l2sz;
3981 	if (sz < nblocks)
3982 		l2sz += 1;
3983 
3984 	/* agsize = roundupSize/max_number_of_ag */
3985 	return (l2sz - L2MAXAG);
3986 }
3987 
3988 
3989 /*
3990  * NAME:	dbMapFileSizeToMapSize()
3991  *
3992  * FUNCTION:	compute number of blocks the block allocation map file
3993  *		can cover from the map file size;
3994  *
3995  * RETURNS:	Number of blocks which can be covered by this block map file;
3996  */
3997 
3998 /*
3999  * maximum number of map pages at each level including control pages
4000  */
4001 #define MAXL0PAGES	(1 + LPERCTL)
4002 #define MAXL1PAGES	(1 + LPERCTL * MAXL0PAGES)
4003 #define MAXL2PAGES	(1 + LPERCTL * MAXL1PAGES)
4004 
4005 /*
4006  * convert number of map pages to the zero origin top dmapctl level
4007  */
4008 #define BMAPPGTOLEV(npages)	\
4009 	(((npages) <= 3 + MAXL0PAGES) ? 0 \
4010        : ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
4011 
4012 s64 dbMapFileSizeToMapSize(struct inode * ipbmap)
4013 {
4014 	struct super_block *sb = ipbmap->i_sb;
4015 	s64 nblocks;
4016 	s64 npages, ndmaps;
4017 	int level, i;
4018 	int complete, factor;
4019 
4020 	nblocks = ipbmap->i_size >> JFS_SBI(sb)->l2bsize;
4021 	npages = nblocks >> JFS_SBI(sb)->l2nbperpage;
4022 	level = BMAPPGTOLEV(npages);
4023 
4024 	/* At each level, accumulate the number of dmap pages covered by
4025 	 * the number of full child levels below it;
4026 	 * repeat for the last incomplete child level.
4027 	 */
4028 	ndmaps = 0;
4029 	npages--;		/* skip the first global control page */
4030 	/* skip higher level control pages above top level covered by map */
4031 	npages -= (2 - level);
4032 	npages--;		/* skip top level's control page */
4033 	for (i = level; i >= 0; i--) {
4034 		factor =
4035 		    (i == 2) ? MAXL1PAGES : ((i == 1) ? MAXL0PAGES : 1);
4036 		complete = (u32) npages / factor;
4037 		ndmaps += complete * ((i == 2) ? LPERCTL * LPERCTL
4038 				      : ((i == 1) ? LPERCTL : 1));
4039 
4040 		/* pages in last/incomplete child */
4041 		npages = (u32) npages % factor;
4042 		/* skip incomplete child's level control page */
4043 		npages--;
4044 	}
4045 
4046 	/* convert the number of dmaps into the number of blocks
4047 	 * which can be covered by the dmaps;
4048 	 */
4049 	nblocks = ndmaps << L2BPERDMAP;
4050 
4051 	return (nblocks);
4052 }
4053 
4054 
4055 #ifdef	_JFS_DEBUG_DMAP
4056 /*
4057  *	DBinitmap()
4058  */
4059 static void DBinitmap(s64 size, struct inode *ipbmap, u32 ** results)
4060 {
4061 	int npages;
4062 	u32 *dbmap, *d;
4063 	int n;
4064 	s64 lblkno, cur_block;
4065 	struct dmap *dp;
4066 	struct metapage *mp;
4067 
4068 	npages = size / 32768;
4069 	npages += (size % 32768) ? 1 : 0;
4070 
4071 	dbmap = (u32 *) xmalloc(npages * 4096, L2PSIZE, kernel_heap);
4072 	if (dbmap == NULL)
4073 		BUG();	/* Not robust since this is only unused debug code */
4074 
4075 	for (n = 0, d = dbmap; n < npages; n++, d += 1024)
4076 		bzero(d, 4096);
4077 
4078 	/* Need to initialize from disk map pages
4079 	 */
4080 	for (d = dbmap, cur_block = 0; cur_block < size;
4081 	     cur_block += BPERDMAP, d += LPERDMAP) {
4082 		lblkno = BLKTODMAP(cur_block,
4083 				   JFS_SBI(ipbmap->i_sb)->bmap->
4084 				   db_l2nbperpage);
4085 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
4086 		if (mp == NULL) {
4087 			jfs_error(ipbmap->i_sb,
4088 				  "DBinitmap: could not read disk map page");
4089 			continue;
4090 		}
4091 		dp = (struct dmap *) mp->data;
4092 
4093 		for (n = 0; n < LPERDMAP; n++)
4094 			d[n] = le32_to_cpu(dp->wmap[n]);
4095 
4096 		release_metapage(mp);
4097 	}
4098 
4099 	*results = dbmap;
4100 }
4101 
4102 
4103 /*
4104  *	DBAlloc()
4105  */
4106 void DBAlloc(uint * dbmap, s64 mapsize, s64 blkno, s64 nblocks)
4107 {
4108 	int word, nb, bitno;
4109 	u32 mask;
4110 
4111 	assert(blkno > 0 && blkno < mapsize);
4112 	assert(nblocks > 0 && nblocks <= mapsize);
4113 
4114 	assert(blkno + nblocks <= mapsize);
4115 
4116 	dbmap += (blkno / 32);
4117 	while (nblocks > 0) {
4118 		bitno = blkno & (32 - 1);
4119 		nb = min(nblocks, 32 - bitno);
4120 
4121 		mask = (0xffffffff << (32 - nb) >> bitno);
4122 		assert((mask & *dbmap) == 0);
4123 		*dbmap |= mask;
4124 
4125 		dbmap++;
4126 		blkno += nb;
4127 		nblocks -= nb;
4128 	}
4129 }
4130 
4131 
4132 /*
4133  *	DBFree()
4134  */
4135 static void DBFree(uint * dbmap, s64 mapsize, s64 blkno, s64 nblocks)
4136 {
4137 	int word, nb, bitno;
4138 	u32 mask;
4139 
4140 	assert(blkno > 0 && blkno < mapsize);
4141 	assert(nblocks > 0 && nblocks <= mapsize);
4142 
4143 	assert(blkno + nblocks <= mapsize);
4144 
4145 	dbmap += (blkno / 32);
4146 	while (nblocks > 0) {
4147 		bitno = blkno & (32 - 1);
4148 		nb = min(nblocks, 32 - bitno);
4149 
4150 		mask = (0xffffffff << (32 - nb) >> bitno);
4151 		assert((mask & *dbmap) == mask);
4152 		*dbmap &= ~mask;
4153 
4154 		dbmap++;
4155 		blkno += nb;
4156 		nblocks -= nb;
4157 	}
4158 }
4159 
4160 
4161 /*
4162  *	DBAllocCK()
4163  */
4164 static void DBAllocCK(uint * dbmap, s64 mapsize, s64 blkno, s64 nblocks)
4165 {
4166 	int word, nb, bitno;
4167 	u32 mask;
4168 
4169 	assert(blkno > 0 && blkno < mapsize);
4170 	assert(nblocks > 0 && nblocks <= mapsize);
4171 
4172 	assert(blkno + nblocks <= mapsize);
4173 
4174 	dbmap += (blkno / 32);
4175 	while (nblocks > 0) {
4176 		bitno = blkno & (32 - 1);
4177 		nb = min(nblocks, 32 - bitno);
4178 
4179 		mask = (0xffffffff << (32 - nb) >> bitno);
4180 		assert((mask & *dbmap) == mask);
4181 
4182 		dbmap++;
4183 		blkno += nb;
4184 		nblocks -= nb;
4185 	}
4186 }
4187 
4188 
4189 /*
4190  *	DBFreeCK()
4191  */
4192 static void DBFreeCK(uint * dbmap, s64 mapsize, s64 blkno, s64 nblocks)
4193 {
4194 	int word, nb, bitno;
4195 	u32 mask;
4196 
4197 	assert(blkno > 0 && blkno < mapsize);
4198 	assert(nblocks > 0 && nblocks <= mapsize);
4199 
4200 	assert(blkno + nblocks <= mapsize);
4201 
4202 	dbmap += (blkno / 32);
4203 	while (nblocks > 0) {
4204 		bitno = blkno & (32 - 1);
4205 		nb = min(nblocks, 32 - bitno);
4206 
4207 		mask = (0xffffffff << (32 - nb) >> bitno);
4208 		assert((mask & *dbmap) == 0);
4209 
4210 		dbmap++;
4211 		blkno += nb;
4212 		nblocks -= nb;
4213 	}
4214 }
4215 
4216 
4217 /*
4218  *	dbPrtMap()
4219  */
4220 static void dbPrtMap(struct bmap * bmp)
4221 {
4222 	printk("   mapsize:   %d%d\n", bmp->db_mapsize);
4223 	printk("   nfree:     %d%d\n", bmp->db_nfree);
4224 	printk("   numag:     %d\n", bmp->db_numag);
4225 	printk("   agsize:    %d%d\n", bmp->db_agsize);
4226 	printk("   agl2size:  %d\n", bmp->db_agl2size);
4227 	printk("   agwidth:   %d\n", bmp->db_agwidth);
4228 	printk("   agstart:   %d\n", bmp->db_agstart);
4229 	printk("   agheigth:  %d\n", bmp->db_agheigth);
4230 	printk("   aglevel:   %d\n", bmp->db_aglevel);
4231 	printk("   maxlevel:  %d\n", bmp->db_maxlevel);
4232 	printk("   maxag:     %d\n", bmp->db_maxag);
4233 	printk("   agpref:    %d\n", bmp->db_agpref);
4234 	printk("   l2nbppg:   %d\n", bmp->db_l2nbperpage);
4235 }
4236 
4237 
4238 /*
4239  *	dbPrtCtl()
4240  */
4241 static void dbPrtCtl(struct dmapctl * dcp)
4242 {
4243 	int i, j, n;
4244 
4245 	printk("   height:    %08x\n", le32_to_cpu(dcp->height));
4246 	printk("   leafidx:   %08x\n", le32_to_cpu(dcp->leafidx));
4247 	printk("   budmin:    %08x\n", dcp->budmin);
4248 	printk("   nleafs:    %08x\n", le32_to_cpu(dcp->nleafs));
4249 	printk("   l2nleafs:  %08x\n", le32_to_cpu(dcp->l2nleafs));
4250 
4251 	printk("\n Tree:\n");
4252 	for (i = 0; i < CTLLEAFIND; i += 8) {
4253 		n = min(8, CTLLEAFIND - i);
4254 
4255 		for (j = 0; j < n; j++)
4256 			printf("  [%03x]: %02x", i + j,
4257 			       (char) dcp->stree[i + j]);
4258 		printf("\n");
4259 	}
4260 
4261 	printk("\n Tree Leaves:\n");
4262 	for (i = 0; i < LPERCTL; i += 8) {
4263 		n = min(8, LPERCTL - i);
4264 
4265 		for (j = 0; j < n; j++)
4266 			printf("  [%03x]: %02x",
4267 			       i + j,
4268 			       (char) dcp->stree[i + j + CTLLEAFIND]);
4269 		printf("\n");
4270 	}
4271 }
4272 #endif				/* _JFS_DEBUG_DMAP */
4273