xref: /openbmc/linux/fs/jfs/jfs_dmap.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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  *	SERIALIZATION of the Block Allocation Map.
30  *
31  *	the working state of the block allocation map is accessed in
32  *	two directions:
33  *
34  *	1) allocation and free requests that start at the dmap
35  *	   level and move up through the dmap control pages (i.e.
36  *	   the vast majority of requests).
37  *
38  *	2) allocation requests that start at dmap control page
39  *	   level and work down towards the dmaps.
40  *
41  *	the serialization scheme used here is as follows.
42  *
43  *	requests which start at the bottom are serialized against each
44  *	other through buffers and each requests holds onto its buffers
45  *	as it works it way up from a single dmap to the required level
46  *	of dmap control page.
47  *	requests that start at the top are serialized against each other
48  *	and request that start from the bottom by the multiple read/single
49  *	write inode lock of the bmap inode. requests starting at the top
50  *	take this lock in write mode while request starting at the bottom
51  *	take the lock in read mode.  a single top-down request may proceed
52  *	exclusively while multiple bottoms-up requests may proceed
53  *	simultaneously (under the protection of busy buffers).
54  *
55  *	in addition to information found in dmaps and dmap control pages,
56  *	the working state of the block allocation map also includes read/
57  *	write information maintained in the bmap descriptor (i.e. total
58  *	free block count, allocation group level free block counts).
59  *	a single exclusive lock (BMAP_LOCK) is used to guard this information
60  *	in the face of multiple-bottoms up requests.
61  *	(lock ordering: IREAD_LOCK, BMAP_LOCK);
62  *
63  *	accesses to the persistent state of the block allocation map (limited
64  *	to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
65  */
66 
67 #define BMAP_LOCK_INIT(bmp)	mutex_init(&bmp->db_bmaplock)
68 #define BMAP_LOCK(bmp)		mutex_lock(&bmp->db_bmaplock)
69 #define BMAP_UNLOCK(bmp)	mutex_unlock(&bmp->db_bmaplock)
70 
71 /*
72  * forward references
73  */
74 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
75 			int nblocks);
76 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval);
77 static int dbBackSplit(dmtree_t * tp, int leafno);
78 static int dbJoin(dmtree_t * tp, int leafno, int newval);
79 static void dbAdjTree(dmtree_t * tp, int leafno, int newval);
80 static int dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc,
81 		    int level);
82 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results);
83 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
84 		       int nblocks);
85 static int dbAllocNear(struct bmap * bmp, struct dmap * dp, s64 blkno,
86 		       int nblocks,
87 		       int l2nb, s64 * results);
88 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
89 		       int nblocks);
90 static int dbAllocDmapLev(struct bmap * bmp, struct dmap * dp, int nblocks,
91 			  int l2nb,
92 			  s64 * results);
93 static int dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb,
94 		     s64 * results);
95 static int dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno,
96 		      s64 * results);
97 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks);
98 static int dbFindBits(u32 word, int l2nb);
99 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno);
100 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx);
101 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
102 		      int nblocks);
103 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
104 		      int nblocks);
105 static int dbMaxBud(u8 * cp);
106 s64 dbMapFileSizeToMapSize(struct inode *ipbmap);
107 static int blkstol2(s64 nb);
108 
109 static int cntlz(u32 value);
110 static int cnttz(u32 word);
111 
112 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
113 			 int nblocks);
114 static int dbInitDmap(struct dmap * dp, s64 blkno, int nblocks);
115 static int dbInitDmapTree(struct dmap * dp);
116 static int dbInitTree(struct dmaptree * dtp);
117 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i);
118 static int dbGetL2AGSize(s64 nblocks);
119 
120 /*
121  *	buddy table
122  *
123  * table used for determining buddy sizes within characters of
124  * dmap bitmap words.  the characters themselves serve as indexes
125  * into the table, with the table elements yielding the maximum
126  * binary buddy of free bits within the character.
127  */
128 static const s8 budtab[256] = {
129 	3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
130 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
131 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
132 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
133 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
134 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
135 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
136 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
137 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
138 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
139 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
140 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
141 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
142 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
143 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
144 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1
145 };
146 
147 
148 /*
149  * NAME:	dbMount()
150  *
151  * FUNCTION:	initializate the block allocation map.
152  *
153  *		memory is allocated for the in-core bmap descriptor and
154  *		the in-core descriptor is initialized from disk.
155  *
156  * PARAMETERS:
157  *      ipbmap	-  pointer to in-core inode for the block map.
158  *
159  * RETURN VALUES:
160  *      0	- success
161  *      -ENOMEM	- insufficient memory
162  *      -EIO	- i/o error
163  */
164 int dbMount(struct inode *ipbmap)
165 {
166 	struct bmap *bmp;
167 	struct dbmap_disk *dbmp_le;
168 	struct metapage *mp;
169 	int i;
170 
171 	/*
172 	 * allocate/initialize the in-memory bmap descriptor
173 	 */
174 	/* allocate memory for the in-memory bmap descriptor */
175 	bmp = kmalloc(sizeof(struct bmap), GFP_KERNEL);
176 	if (bmp == NULL)
177 		return -ENOMEM;
178 
179 	/* read the on-disk bmap descriptor. */
180 	mp = read_metapage(ipbmap,
181 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
182 			   PSIZE, 0);
183 	if (mp == NULL) {
184 		kfree(bmp);
185 		return -EIO;
186 	}
187 
188 	/* copy the on-disk bmap descriptor to its in-memory version. */
189 	dbmp_le = (struct dbmap_disk *) mp->data;
190 	bmp->db_mapsize = le64_to_cpu(dbmp_le->dn_mapsize);
191 	bmp->db_nfree = le64_to_cpu(dbmp_le->dn_nfree);
192 	bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage);
193 	bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag);
194 	bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel);
195 	bmp->db_maxag = le32_to_cpu(dbmp_le->dn_maxag);
196 	bmp->db_agpref = le32_to_cpu(dbmp_le->dn_agpref);
197 	bmp->db_aglevel = le32_to_cpu(dbmp_le->dn_aglevel);
198 	bmp->db_agheigth = le32_to_cpu(dbmp_le->dn_agheigth);
199 	bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth);
200 	bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart);
201 	bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size);
202 	for (i = 0; i < MAXAG; i++)
203 		bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
204 	bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
205 	bmp->db_maxfreebud = dbmp_le->dn_maxfreebud;
206 
207 	/* release the buffer. */
208 	release_metapage(mp);
209 
210 	/* bind the bmap inode and the bmap descriptor to each other. */
211 	bmp->db_ipbmap = ipbmap;
212 	JFS_SBI(ipbmap->i_sb)->bmap = bmp;
213 
214 	memset(bmp->db_active, 0, sizeof(bmp->db_active));
215 
216 	/*
217 	 * allocate/initialize the bmap lock
218 	 */
219 	BMAP_LOCK_INIT(bmp);
220 
221 	return (0);
222 }
223 
224 
225 /*
226  * NAME:	dbUnmount()
227  *
228  * FUNCTION:	terminate the block allocation map in preparation for
229  *		file system unmount.
230  *
231  *		the in-core bmap descriptor is written to disk and
232  *		the memory for this descriptor is freed.
233  *
234  * PARAMETERS:
235  *      ipbmap	-  pointer to in-core inode for the block map.
236  *
237  * RETURN VALUES:
238  *      0	- success
239  *      -EIO	- i/o error
240  */
241 int dbUnmount(struct inode *ipbmap, int mounterror)
242 {
243 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
244 
245 	if (!(mounterror || isReadOnly(ipbmap)))
246 		dbSync(ipbmap);
247 
248 	/*
249 	 * Invalidate the page cache buffers
250 	 */
251 	truncate_inode_pages(ipbmap->i_mapping, 0);
252 
253 	/* free the memory for the in-memory bmap. */
254 	kfree(bmp);
255 
256 	return (0);
257 }
258 
259 /*
260  *	dbSync()
261  */
262 int dbSync(struct inode *ipbmap)
263 {
264 	struct dbmap_disk *dbmp_le;
265 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
266 	struct metapage *mp;
267 	int i;
268 
269 	/*
270 	 * write bmap global control page
271 	 */
272 	/* get the buffer for the on-disk bmap descriptor. */
273 	mp = read_metapage(ipbmap,
274 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
275 			   PSIZE, 0);
276 	if (mp == NULL) {
277 		jfs_err("dbSync: read_metapage failed!");
278 		return -EIO;
279 	}
280 	/* copy the in-memory version of the bmap to the on-disk version */
281 	dbmp_le = (struct dbmap_disk *) mp->data;
282 	dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize);
283 	dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree);
284 	dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage);
285 	dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag);
286 	dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel);
287 	dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag);
288 	dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref);
289 	dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel);
290 	dbmp_le->dn_agheigth = cpu_to_le32(bmp->db_agheigth);
291 	dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth);
292 	dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart);
293 	dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size);
294 	for (i = 0; i < MAXAG; i++)
295 		dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]);
296 	dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize);
297 	dbmp_le->dn_maxfreebud = bmp->db_maxfreebud;
298 
299 	/* write the buffer */
300 	write_metapage(mp);
301 
302 	/*
303 	 * write out dirty pages of bmap
304 	 */
305 	filemap_write_and_wait(ipbmap->i_mapping);
306 
307 	diWriteSpecial(ipbmap, 0);
308 
309 	return (0);
310 }
311 
312 
313 /*
314  * NAME:	dbFree()
315  *
316  * FUNCTION:	free the specified block range from the working block
317  *		allocation map.
318  *
319  *		the blocks will be free from the working map one dmap
320  *		at a time.
321  *
322  * PARAMETERS:
323  *      ip	-  pointer to in-core inode;
324  *      blkno	-  starting block number to be freed.
325  *      nblocks	-  number of blocks to be freed.
326  *
327  * RETURN VALUES:
328  *      0	- success
329  *      -EIO	- i/o error
330  */
331 int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
332 {
333 	struct metapage *mp;
334 	struct dmap *dp;
335 	int nb, rc;
336 	s64 lblkno, rem;
337 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
338 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
339 
340 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
341 
342 	/* block to be freed better be within the mapsize. */
343 	if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
344 		IREAD_UNLOCK(ipbmap);
345 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
346 		       (unsigned long long) blkno,
347 		       (unsigned long long) nblocks);
348 		jfs_error(ip->i_sb,
349 			  "dbFree: block to be freed is outside the map");
350 		return -EIO;
351 	}
352 
353 	/*
354 	 * free the blocks a dmap at a time.
355 	 */
356 	mp = NULL;
357 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
358 		/* release previous dmap if any */
359 		if (mp) {
360 			write_metapage(mp);
361 		}
362 
363 		/* get the buffer for the current dmap. */
364 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
365 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
366 		if (mp == NULL) {
367 			IREAD_UNLOCK(ipbmap);
368 			return -EIO;
369 		}
370 		dp = (struct dmap *) mp->data;
371 
372 		/* determine the number of blocks to be freed from
373 		 * this dmap.
374 		 */
375 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
376 
377 		/* free the blocks. */
378 		if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
379 			jfs_error(ip->i_sb, "dbFree: error in block map\n");
380 			release_metapage(mp);
381 			IREAD_UNLOCK(ipbmap);
382 			return (rc);
383 		}
384 	}
385 
386 	/* write the last buffer. */
387 	write_metapage(mp);
388 
389 	IREAD_UNLOCK(ipbmap);
390 
391 	return (0);
392 }
393 
394 
395 /*
396  * NAME:	dbUpdatePMap()
397  *
398  * FUNCTION:    update the allocation state (free or allocate) of the
399  *		specified block range in the persistent block allocation map.
400  *
401  *		the blocks will be updated in the persistent map one
402  *		dmap at a time.
403  *
404  * PARAMETERS:
405  *      ipbmap	-  pointer to in-core inode for the block map.
406  *      free	-  'true' if block range is to be freed from the persistent
407  *		   map; 'false' if it is to   be allocated.
408  *      blkno	-  starting block number of the range.
409  *      nblocks	-  number of contiguous blocks in the range.
410  *      tblk	-  transaction block;
411  *
412  * RETURN VALUES:
413  *      0	- success
414  *      -EIO	- i/o error
415  */
416 int
417 dbUpdatePMap(struct inode *ipbmap,
418 	     int free, s64 blkno, s64 nblocks, struct tblock * tblk)
419 {
420 	int nblks, dbitno, wbitno, rbits;
421 	int word, nbits, nwords;
422 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
423 	s64 lblkno, rem, lastlblkno;
424 	u32 mask;
425 	struct dmap *dp;
426 	struct metapage *mp;
427 	struct jfs_log *log;
428 	int lsn, difft, diffp;
429 	unsigned long flags;
430 
431 	/* the blocks better be within the mapsize. */
432 	if (blkno + nblocks > bmp->db_mapsize) {
433 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
434 		       (unsigned long long) blkno,
435 		       (unsigned long long) nblocks);
436 		jfs_error(ipbmap->i_sb,
437 			  "dbUpdatePMap: blocks are outside the map");
438 		return -EIO;
439 	}
440 
441 	/* compute delta of transaction lsn from log syncpt */
442 	lsn = tblk->lsn;
443 	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
444 	logdiff(difft, lsn, log);
445 
446 	/*
447 	 * update the block state a dmap at a time.
448 	 */
449 	mp = NULL;
450 	lastlblkno = 0;
451 	for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) {
452 		/* get the buffer for the current dmap. */
453 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
454 		if (lblkno != lastlblkno) {
455 			if (mp) {
456 				write_metapage(mp);
457 			}
458 
459 			mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE,
460 					   0);
461 			if (mp == NULL)
462 				return -EIO;
463 			metapage_wait_for_io(mp);
464 		}
465 		dp = (struct dmap *) mp->data;
466 
467 		/* determine the bit number and word within the dmap of
468 		 * the starting block.  also determine how many blocks
469 		 * are to be updated within this dmap.
470 		 */
471 		dbitno = blkno & (BPERDMAP - 1);
472 		word = dbitno >> L2DBWORD;
473 		nblks = min(rem, (s64)BPERDMAP - dbitno);
474 
475 		/* update the bits of the dmap words. the first and last
476 		 * words may only have a subset of their bits updated. if
477 		 * this is the case, we'll work against that word (i.e.
478 		 * partial first and/or last) only in a single pass.  a
479 		 * single pass will also be used to update all words that
480 		 * are to have all their bits updated.
481 		 */
482 		for (rbits = nblks; rbits > 0;
483 		     rbits -= nbits, dbitno += nbits) {
484 			/* determine the bit number within the word and
485 			 * the number of bits within the word.
486 			 */
487 			wbitno = dbitno & (DBWORD - 1);
488 			nbits = min(rbits, DBWORD - wbitno);
489 
490 			/* check if only part of the word is to be updated. */
491 			if (nbits < DBWORD) {
492 				/* update (free or allocate) the bits
493 				 * in this word.
494 				 */
495 				mask =
496 				    (ONES << (DBWORD - nbits) >> wbitno);
497 				if (free)
498 					dp->pmap[word] &=
499 					    cpu_to_le32(~mask);
500 				else
501 					dp->pmap[word] |=
502 					    cpu_to_le32(mask);
503 
504 				word += 1;
505 			} else {
506 				/* one or more words are to have all
507 				 * their bits updated.  determine how
508 				 * many words and how many bits.
509 				 */
510 				nwords = rbits >> L2DBWORD;
511 				nbits = nwords << L2DBWORD;
512 
513 				/* update (free or allocate) the bits
514 				 * in these words.
515 				 */
516 				if (free)
517 					memset(&dp->pmap[word], 0,
518 					       nwords * 4);
519 				else
520 					memset(&dp->pmap[word], (int) ONES,
521 					       nwords * 4);
522 
523 				word += nwords;
524 			}
525 		}
526 
527 		/*
528 		 * update dmap lsn
529 		 */
530 		if (lblkno == lastlblkno)
531 			continue;
532 
533 		lastlblkno = lblkno;
534 
535 		LOGSYNC_LOCK(log, flags);
536 		if (mp->lsn != 0) {
537 			/* inherit older/smaller lsn */
538 			logdiff(diffp, mp->lsn, log);
539 			if (difft < diffp) {
540 				mp->lsn = lsn;
541 
542 				/* move bp after tblock in logsync list */
543 				list_move(&mp->synclist, &tblk->synclist);
544 			}
545 
546 			/* inherit younger/larger clsn */
547 			logdiff(difft, tblk->clsn, log);
548 			logdiff(diffp, mp->clsn, log);
549 			if (difft > diffp)
550 				mp->clsn = tblk->clsn;
551 		} else {
552 			mp->log = log;
553 			mp->lsn = lsn;
554 
555 			/* insert bp after tblock in logsync list */
556 			log->count++;
557 			list_add(&mp->synclist, &tblk->synclist);
558 
559 			mp->clsn = tblk->clsn;
560 		}
561 		LOGSYNC_UNLOCK(log, flags);
562 	}
563 
564 	/* write the last buffer. */
565 	if (mp) {
566 		write_metapage(mp);
567 	}
568 
569 	return (0);
570 }
571 
572 
573 /*
574  * NAME:	dbNextAG()
575  *
576  * FUNCTION:    find the preferred allocation group for new allocations.
577  *
578  *		Within the allocation groups, we maintain a preferred
579  *		allocation group which consists of a group with at least
580  *		average free space.  It is the preferred group that we target
581  *		new inode allocation towards.  The tie-in between inode
582  *		allocation and block allocation occurs as we allocate the
583  *		first (data) block of an inode and specify the inode (block)
584  *		as the allocation hint for this block.
585  *
586  *		We try to avoid having more than one open file growing in
587  *		an allocation group, as this will lead to fragmentation.
588  *		This differs from the old OS/2 method of trying to keep
589  *		empty ags around for large allocations.
590  *
591  * PARAMETERS:
592  *      ipbmap	-  pointer to in-core inode for the block map.
593  *
594  * RETURN VALUES:
595  *      the preferred allocation group number.
596  */
597 int dbNextAG(struct inode *ipbmap)
598 {
599 	s64 avgfree;
600 	int agpref;
601 	s64 hwm = 0;
602 	int i;
603 	int next_best = -1;
604 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
605 
606 	BMAP_LOCK(bmp);
607 
608 	/* determine the average number of free blocks within the ags. */
609 	avgfree = (u32)bmp->db_nfree / bmp->db_numag;
610 
611 	/*
612 	 * if the current preferred ag does not have an active allocator
613 	 * and has at least average freespace, return it
614 	 */
615 	agpref = bmp->db_agpref;
616 	if ((atomic_read(&bmp->db_active[agpref]) == 0) &&
617 	    (bmp->db_agfree[agpref] >= avgfree))
618 		goto unlock;
619 
620 	/* From the last preferred ag, find the next one with at least
621 	 * average free space.
622 	 */
623 	for (i = 0 ; i < bmp->db_numag; i++, agpref++) {
624 		if (agpref == bmp->db_numag)
625 			agpref = 0;
626 
627 		if (atomic_read(&bmp->db_active[agpref]))
628 			/* open file is currently growing in this ag */
629 			continue;
630 		if (bmp->db_agfree[agpref] >= avgfree) {
631 			/* Return this one */
632 			bmp->db_agpref = agpref;
633 			goto unlock;
634 		} else if (bmp->db_agfree[agpref] > hwm) {
635 			/* Less than avg. freespace, but best so far */
636 			hwm = bmp->db_agfree[agpref];
637 			next_best = agpref;
638 		}
639 	}
640 
641 	/*
642 	 * If no inactive ag was found with average freespace, use the
643 	 * next best
644 	 */
645 	if (next_best != -1)
646 		bmp->db_agpref = next_best;
647 	/* else leave db_agpref unchanged */
648 unlock:
649 	BMAP_UNLOCK(bmp);
650 
651 	/* return the preferred group.
652 	 */
653 	return (bmp->db_agpref);
654 }
655 
656 /*
657  * NAME:	dbAlloc()
658  *
659  * FUNCTION:    attempt to allocate a specified number of contiguous free
660  *		blocks from the working allocation block map.
661  *
662  *		the block allocation policy uses hints and a multi-step
663  *		approach.
664  *
665  *		for allocation requests smaller than the number of blocks
666  *		per dmap, we first try to allocate the new blocks
667  *		immediately following the hint.  if these blocks are not
668  *		available, we try to allocate blocks near the hint.  if
669  *		no blocks near the hint are available, we next try to
670  *		allocate within the same dmap as contains the hint.
671  *
672  *		if no blocks are available in the dmap or the allocation
673  *		request is larger than the dmap size, we try to allocate
674  *		within the same allocation group as contains the hint. if
675  *		this does not succeed, we finally try to allocate anywhere
676  *		within the aggregate.
677  *
678  *		we also try to allocate anywhere within the aggregate for
679  *		for allocation requests larger than the allocation group
680  *		size or requests that specify no hint value.
681  *
682  * PARAMETERS:
683  *      ip	-  pointer to in-core inode;
684  *      hint	- allocation hint.
685  *      nblocks	- number of contiguous blocks in the range.
686  *      results	- on successful return, set to the starting block number
687  *		  of the newly allocated contiguous range.
688  *
689  * RETURN VALUES:
690  *      0	- success
691  *      -ENOSPC	- insufficient disk resources
692  *      -EIO	- i/o error
693  */
694 int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results)
695 {
696 	int rc, agno;
697 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
698 	struct bmap *bmp;
699 	struct metapage *mp;
700 	s64 lblkno, blkno;
701 	struct dmap *dp;
702 	int l2nb;
703 	s64 mapSize;
704 	int writers;
705 
706 	/* assert that nblocks is valid */
707 	assert(nblocks > 0);
708 
709 #ifdef _STILL_TO_PORT
710 	/* DASD limit check                                     F226941 */
711 	if (OVER_LIMIT(ip, nblocks))
712 		return -ENOSPC;
713 #endif				/* _STILL_TO_PORT */
714 
715 	/* get the log2 number of blocks to be allocated.
716 	 * if the number of blocks is not a log2 multiple,
717 	 * it will be rounded up to the next log2 multiple.
718 	 */
719 	l2nb = BLKSTOL2(nblocks);
720 
721 	bmp = JFS_SBI(ip->i_sb)->bmap;
722 
723 //retry:        /* serialize w.r.t.extendfs() */
724 	mapSize = bmp->db_mapsize;
725 
726 	/* the hint should be within the map */
727 	if (hint >= mapSize) {
728 		jfs_error(ip->i_sb, "dbAlloc: the hint is outside the map");
729 		return -EIO;
730 	}
731 
732 	/* if the number of blocks to be allocated is greater than the
733 	 * allocation group size, try to allocate anywhere.
734 	 */
735 	if (l2nb > bmp->db_agl2size) {
736 		IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
737 
738 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
739 
740 		goto write_unlock;
741 	}
742 
743 	/*
744 	 * If no hint, let dbNextAG recommend an allocation group
745 	 */
746 	if (hint == 0)
747 		goto pref_ag;
748 
749 	/* we would like to allocate close to the hint.  adjust the
750 	 * hint to the block following the hint since the allocators
751 	 * will start looking for free space starting at this point.
752 	 */
753 	blkno = hint + 1;
754 
755 	if (blkno >= bmp->db_mapsize)
756 		goto pref_ag;
757 
758 	agno = blkno >> bmp->db_agl2size;
759 
760 	/* check if blkno crosses over into a new allocation group.
761 	 * if so, check if we should allow allocations within this
762 	 * allocation group.
763 	 */
764 	if ((blkno & (bmp->db_agsize - 1)) == 0)
765 		/* check if the AG is currenly being written to.
766 		 * if so, call dbNextAG() to find a non-busy
767 		 * AG with sufficient free space.
768 		 */
769 		if (atomic_read(&bmp->db_active[agno]))
770 			goto pref_ag;
771 
772 	/* check if the allocation request size can be satisfied from a
773 	 * single dmap.  if so, try to allocate from the dmap containing
774 	 * the hint using a tiered strategy.
775 	 */
776 	if (nblocks <= BPERDMAP) {
777 		IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
778 
779 		/* get the buffer for the dmap containing the hint.
780 		 */
781 		rc = -EIO;
782 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
783 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
784 		if (mp == NULL)
785 			goto read_unlock;
786 
787 		dp = (struct dmap *) mp->data;
788 
789 		/* first, try to satisfy the allocation request with the
790 		 * blocks beginning at the hint.
791 		 */
792 		if ((rc = dbAllocNext(bmp, dp, blkno, (int) nblocks))
793 		    != -ENOSPC) {
794 			if (rc == 0) {
795 				*results = blkno;
796 				mark_metapage_dirty(mp);
797 			}
798 
799 			release_metapage(mp);
800 			goto read_unlock;
801 		}
802 
803 		writers = atomic_read(&bmp->db_active[agno]);
804 		if ((writers > 1) ||
805 		    ((writers == 1) && (JFS_IP(ip)->active_ag != agno))) {
806 			/*
807 			 * Someone else is writing in this allocation
808 			 * group.  To avoid fragmenting, try another ag
809 			 */
810 			release_metapage(mp);
811 			IREAD_UNLOCK(ipbmap);
812 			goto pref_ag;
813 		}
814 
815 		/* next, try to satisfy the allocation request with blocks
816 		 * near the hint.
817 		 */
818 		if ((rc =
819 		     dbAllocNear(bmp, dp, blkno, (int) nblocks, l2nb, results))
820 		    != -ENOSPC) {
821 			if (rc == 0)
822 				mark_metapage_dirty(mp);
823 
824 			release_metapage(mp);
825 			goto read_unlock;
826 		}
827 
828 		/* try to satisfy the allocation request with blocks within
829 		 * the same dmap as the hint.
830 		 */
831 		if ((rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results))
832 		    != -ENOSPC) {
833 			if (rc == 0)
834 				mark_metapage_dirty(mp);
835 
836 			release_metapage(mp);
837 			goto read_unlock;
838 		}
839 
840 		release_metapage(mp);
841 		IREAD_UNLOCK(ipbmap);
842 	}
843 
844 	/* try to satisfy the allocation request with blocks within
845 	 * the same allocation group as the hint.
846 	 */
847 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
848 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) != -ENOSPC)
849 		goto write_unlock;
850 
851 	IWRITE_UNLOCK(ipbmap);
852 
853 
854       pref_ag:
855 	/*
856 	 * Let dbNextAG recommend a preferred allocation group
857 	 */
858 	agno = dbNextAG(ipbmap);
859 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
860 
861 	/* Try to allocate within this allocation group.  if that fails, try to
862 	 * allocate anywhere in the map.
863 	 */
864 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) == -ENOSPC)
865 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
866 
867       write_unlock:
868 	IWRITE_UNLOCK(ipbmap);
869 
870 	return (rc);
871 
872       read_unlock:
873 	IREAD_UNLOCK(ipbmap);
874 
875 	return (rc);
876 }
877 
878 #ifdef _NOTYET
879 /*
880  * NAME:	dbAllocExact()
881  *
882  * FUNCTION:    try to allocate the requested extent;
883  *
884  * PARAMETERS:
885  *      ip	- pointer to in-core inode;
886  *      blkno	- extent address;
887  *      nblocks	- extent length;
888  *
889  * RETURN VALUES:
890  *      0	- success
891  *      -ENOSPC	- insufficient disk resources
892  *      -EIO	- i/o error
893  */
894 int dbAllocExact(struct inode *ip, s64 blkno, int nblocks)
895 {
896 	int rc;
897 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
898 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
899 	struct dmap *dp;
900 	s64 lblkno;
901 	struct metapage *mp;
902 
903 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
904 
905 	/*
906 	 * validate extent request:
907 	 *
908 	 * note: defragfs policy:
909 	 *  max 64 blocks will be moved.
910 	 *  allocation request size must be satisfied from a single dmap.
911 	 */
912 	if (nblocks <= 0 || nblocks > BPERDMAP || blkno >= bmp->db_mapsize) {
913 		IREAD_UNLOCK(ipbmap);
914 		return -EINVAL;
915 	}
916 
917 	if (nblocks > ((s64) 1 << bmp->db_maxfreebud)) {
918 		/* the free space is no longer available */
919 		IREAD_UNLOCK(ipbmap);
920 		return -ENOSPC;
921 	}
922 
923 	/* read in the dmap covering the extent */
924 	lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
925 	mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
926 	if (mp == NULL) {
927 		IREAD_UNLOCK(ipbmap);
928 		return -EIO;
929 	}
930 	dp = (struct dmap *) mp->data;
931 
932 	/* try to allocate the requested extent */
933 	rc = dbAllocNext(bmp, dp, blkno, nblocks);
934 
935 	IREAD_UNLOCK(ipbmap);
936 
937 	if (rc == 0)
938 		mark_metapage_dirty(mp);
939 
940 	release_metapage(mp);
941 
942 	return (rc);
943 }
944 #endif /* _NOTYET */
945 
946 /*
947  * NAME:	dbReAlloc()
948  *
949  * FUNCTION:    attempt to extend a current allocation by a specified
950  *		number of blocks.
951  *
952  *		this routine attempts to satisfy the allocation request
953  *		by first trying to extend the existing allocation in
954  *		place by allocating the additional blocks as the blocks
955  *		immediately following the current allocation.  if these
956  *		blocks are not available, this routine will attempt to
957  *		allocate a new set of contiguous blocks large enough
958  *		to cover the existing allocation plus the additional
959  *		number of blocks required.
960  *
961  * PARAMETERS:
962  *      ip	    -  pointer to in-core inode requiring allocation.
963  *      blkno	    -  starting block of the current allocation.
964  *      nblocks	    -  number of contiguous blocks within the current
965  *		       allocation.
966  *      addnblocks  -  number of blocks to add to the allocation.
967  *      results	-      on successful return, set to the starting block number
968  *		       of the existing allocation if the existing allocation
969  *		       was extended in place or to a newly allocated contiguous
970  *		       range if the existing allocation could not be extended
971  *		       in place.
972  *
973  * RETURN VALUES:
974  *      0	- success
975  *      -ENOSPC	- insufficient disk resources
976  *      -EIO	- i/o error
977  */
978 int
979 dbReAlloc(struct inode *ip,
980 	  s64 blkno, s64 nblocks, s64 addnblocks, s64 * results)
981 {
982 	int rc;
983 
984 	/* try to extend the allocation in place.
985 	 */
986 	if ((rc = dbExtend(ip, blkno, nblocks, addnblocks)) == 0) {
987 		*results = blkno;
988 		return (0);
989 	} else {
990 		if (rc != -ENOSPC)
991 			return (rc);
992 	}
993 
994 	/* could not extend the allocation in place, so allocate a
995 	 * new set of blocks for the entire request (i.e. try to get
996 	 * a range of contiguous blocks large enough to cover the
997 	 * existing allocation plus the additional blocks.)
998 	 */
999 	return (dbAlloc
1000 		(ip, blkno + nblocks - 1, addnblocks + nblocks, results));
1001 }
1002 
1003 
1004 /*
1005  * NAME:	dbExtend()
1006  *
1007  * FUNCTION:    attempt to extend a current allocation by a specified
1008  *		number of blocks.
1009  *
1010  *		this routine attempts to satisfy the allocation request
1011  *		by first trying to extend the existing allocation in
1012  *		place by allocating the additional blocks as the blocks
1013  *		immediately following the current allocation.
1014  *
1015  * PARAMETERS:
1016  *      ip	    -  pointer to in-core inode requiring allocation.
1017  *      blkno	    -  starting block of the current allocation.
1018  *      nblocks	    -  number of contiguous blocks within the current
1019  *		       allocation.
1020  *      addnblocks  -  number of blocks to add to the allocation.
1021  *
1022  * RETURN VALUES:
1023  *      0	- success
1024  *      -ENOSPC	- insufficient disk resources
1025  *      -EIO	- i/o error
1026  */
1027 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks)
1028 {
1029 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
1030 	s64 lblkno, lastblkno, extblkno;
1031 	uint rel_block;
1032 	struct metapage *mp;
1033 	struct dmap *dp;
1034 	int rc;
1035 	struct inode *ipbmap = sbi->ipbmap;
1036 	struct bmap *bmp;
1037 
1038 	/*
1039 	 * We don't want a non-aligned extent to cross a page boundary
1040 	 */
1041 	if (((rel_block = blkno & (sbi->nbperpage - 1))) &&
1042 	    (rel_block + nblocks + addnblocks > sbi->nbperpage))
1043 		return -ENOSPC;
1044 
1045 	/* get the last block of the current allocation */
1046 	lastblkno = blkno + nblocks - 1;
1047 
1048 	/* determine the block number of the block following
1049 	 * the existing allocation.
1050 	 */
1051 	extblkno = lastblkno + 1;
1052 
1053 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
1054 
1055 	/* better be within the file system */
1056 	bmp = sbi->bmap;
1057 	if (lastblkno < 0 || lastblkno >= bmp->db_mapsize) {
1058 		IREAD_UNLOCK(ipbmap);
1059 		jfs_error(ip->i_sb,
1060 			  "dbExtend: the block is outside the filesystem");
1061 		return -EIO;
1062 	}
1063 
1064 	/* we'll attempt to extend the current allocation in place by
1065 	 * allocating the additional blocks as the blocks immediately
1066 	 * following the current allocation.  we only try to extend the
1067 	 * current allocation in place if the number of additional blocks
1068 	 * can fit into a dmap, the last block of the current allocation
1069 	 * is not the last block of the file system, and the start of the
1070 	 * inplace extension is not on an allocation group boundary.
1071 	 */
1072 	if (addnblocks > BPERDMAP || extblkno >= bmp->db_mapsize ||
1073 	    (extblkno & (bmp->db_agsize - 1)) == 0) {
1074 		IREAD_UNLOCK(ipbmap);
1075 		return -ENOSPC;
1076 	}
1077 
1078 	/* get the buffer for the dmap containing the first block
1079 	 * of the extension.
1080 	 */
1081 	lblkno = BLKTODMAP(extblkno, bmp->db_l2nbperpage);
1082 	mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
1083 	if (mp == NULL) {
1084 		IREAD_UNLOCK(ipbmap);
1085 		return -EIO;
1086 	}
1087 
1088 	dp = (struct dmap *) mp->data;
1089 
1090 	/* try to allocate the blocks immediately following the
1091 	 * current allocation.
1092 	 */
1093 	rc = dbAllocNext(bmp, dp, extblkno, (int) addnblocks);
1094 
1095 	IREAD_UNLOCK(ipbmap);
1096 
1097 	/* were we successful ? */
1098 	if (rc == 0)
1099 		write_metapage(mp);
1100 	else
1101 		/* we were not successful */
1102 		release_metapage(mp);
1103 
1104 
1105 	return (rc);
1106 }
1107 
1108 
1109 /*
1110  * NAME:	dbAllocNext()
1111  *
1112  * FUNCTION:    attempt to allocate the blocks of the specified block
1113  *		range within a dmap.
1114  *
1115  * PARAMETERS:
1116  *      bmp	-  pointer to bmap descriptor
1117  *      dp	-  pointer to dmap.
1118  *      blkno	-  starting block number of the range.
1119  *      nblocks	-  number of contiguous free blocks of the range.
1120  *
1121  * RETURN VALUES:
1122  *      0	- success
1123  *      -ENOSPC	- insufficient disk resources
1124  *      -EIO	- i/o error
1125  *
1126  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1127  */
1128 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
1129 		       int nblocks)
1130 {
1131 	int dbitno, word, rembits, nb, nwords, wbitno, nw;
1132 	int l2size;
1133 	s8 *leaf;
1134 	u32 mask;
1135 
1136 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1137 		jfs_error(bmp->db_ipbmap->i_sb,
1138 			  "dbAllocNext: Corrupt dmap page");
1139 		return -EIO;
1140 	}
1141 
1142 	/* pick up a pointer to the leaves of the dmap tree.
1143 	 */
1144 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1145 
1146 	/* determine the bit number and word within the dmap of the
1147 	 * starting block.
1148 	 */
1149 	dbitno = blkno & (BPERDMAP - 1);
1150 	word = dbitno >> L2DBWORD;
1151 
1152 	/* check if the specified block range is contained within
1153 	 * this dmap.
1154 	 */
1155 	if (dbitno + nblocks > BPERDMAP)
1156 		return -ENOSPC;
1157 
1158 	/* check if the starting leaf indicates that anything
1159 	 * is free.
1160 	 */
1161 	if (leaf[word] == NOFREE)
1162 		return -ENOSPC;
1163 
1164 	/* check the dmaps words corresponding to block range to see
1165 	 * if the block range is free.  not all bits of the first and
1166 	 * last words may be contained within the block range.  if this
1167 	 * is the case, we'll work against those words (i.e. partial first
1168 	 * and/or last) on an individual basis (a single pass) and examine
1169 	 * the actual bits to determine if they are free.  a single pass
1170 	 * will be used for all dmap words fully contained within the
1171 	 * specified range.  within this pass, the leaves of the dmap
1172 	 * tree will be examined to determine if the blocks are free. a
1173 	 * single leaf may describe the free space of multiple dmap
1174 	 * words, so we may visit only a subset of the actual leaves
1175 	 * corresponding to the dmap words of the block range.
1176 	 */
1177 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
1178 		/* determine the bit number within the word and
1179 		 * the number of bits within the word.
1180 		 */
1181 		wbitno = dbitno & (DBWORD - 1);
1182 		nb = min(rembits, DBWORD - wbitno);
1183 
1184 		/* check if only part of the word is to be examined.
1185 		 */
1186 		if (nb < DBWORD) {
1187 			/* check if the bits are free.
1188 			 */
1189 			mask = (ONES << (DBWORD - nb) >> wbitno);
1190 			if ((mask & ~le32_to_cpu(dp->wmap[word])) != mask)
1191 				return -ENOSPC;
1192 
1193 			word += 1;
1194 		} else {
1195 			/* one or more dmap words are fully contained
1196 			 * within the block range.  determine how many
1197 			 * words and how many bits.
1198 			 */
1199 			nwords = rembits >> L2DBWORD;
1200 			nb = nwords << L2DBWORD;
1201 
1202 			/* now examine the appropriate leaves to determine
1203 			 * if the blocks are free.
1204 			 */
1205 			while (nwords > 0) {
1206 				/* does the leaf describe any free space ?
1207 				 */
1208 				if (leaf[word] < BUDMIN)
1209 					return -ENOSPC;
1210 
1211 				/* determine the l2 number of bits provided
1212 				 * by this leaf.
1213 				 */
1214 				l2size =
1215 				    min((int)leaf[word], NLSTOL2BSZ(nwords));
1216 
1217 				/* determine how many words were handled.
1218 				 */
1219 				nw = BUDSIZE(l2size, BUDMIN);
1220 
1221 				nwords -= nw;
1222 				word += nw;
1223 			}
1224 		}
1225 	}
1226 
1227 	/* allocate the blocks.
1228 	 */
1229 	return (dbAllocDmap(bmp, dp, blkno, nblocks));
1230 }
1231 
1232 
1233 /*
1234  * NAME:	dbAllocNear()
1235  *
1236  * FUNCTION:    attempt to allocate a number of contiguous free blocks near
1237  *		a specified block (hint) within a dmap.
1238  *
1239  *		starting with the dmap leaf that covers the hint, we'll
1240  *		check the next four contiguous leaves for sufficient free
1241  *		space.  if sufficient free space is found, we'll allocate
1242  *		the desired free space.
1243  *
1244  * PARAMETERS:
1245  *      bmp	-  pointer to bmap descriptor
1246  *      dp	-  pointer to dmap.
1247  *      blkno	-  block number to allocate near.
1248  *      nblocks	-  actual number of contiguous free blocks desired.
1249  *      l2nb	-  log2 number of contiguous free blocks desired.
1250  *      results	-  on successful return, set to the starting block number
1251  *		   of the newly allocated range.
1252  *
1253  * RETURN VALUES:
1254  *      0	- success
1255  *      -ENOSPC	- insufficient disk resources
1256  *      -EIO	- i/o error
1257  *
1258  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1259  */
1260 static int
1261 dbAllocNear(struct bmap * bmp,
1262 	    struct dmap * dp, s64 blkno, int nblocks, int l2nb, s64 * results)
1263 {
1264 	int word, lword, rc;
1265 	s8 *leaf;
1266 
1267 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1268 		jfs_error(bmp->db_ipbmap->i_sb,
1269 			  "dbAllocNear: Corrupt dmap page");
1270 		return -EIO;
1271 	}
1272 
1273 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1274 
1275 	/* determine the word within the dmap that holds the hint
1276 	 * (i.e. blkno).  also, determine the last word in the dmap
1277 	 * that we'll include in our examination.
1278 	 */
1279 	word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
1280 	lword = min(word + 4, LPERDMAP);
1281 
1282 	/* examine the leaves for sufficient free space.
1283 	 */
1284 	for (; word < lword; word++) {
1285 		/* does the leaf describe sufficient free space ?
1286 		 */
1287 		if (leaf[word] < l2nb)
1288 			continue;
1289 
1290 		/* determine the block number within the file system
1291 		 * of the first block described by this dmap word.
1292 		 */
1293 		blkno = le64_to_cpu(dp->start) + (word << L2DBWORD);
1294 
1295 		/* if not all bits of the dmap word are free, get the
1296 		 * starting bit number within the dmap word of the required
1297 		 * string of free bits and adjust the block number with the
1298 		 * value.
1299 		 */
1300 		if (leaf[word] < BUDMIN)
1301 			blkno +=
1302 			    dbFindBits(le32_to_cpu(dp->wmap[word]), l2nb);
1303 
1304 		/* allocate the blocks.
1305 		 */
1306 		if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1307 			*results = blkno;
1308 
1309 		return (rc);
1310 	}
1311 
1312 	return -ENOSPC;
1313 }
1314 
1315 
1316 /*
1317  * NAME:	dbAllocAG()
1318  *
1319  * FUNCTION:    attempt to allocate the specified number of contiguous
1320  *		free blocks within the specified allocation group.
1321  *
1322  *		unless the allocation group size is equal to the number
1323  *		of blocks per dmap, the dmap control pages will be used to
1324  *		find the required free space, if available.  we start the
1325  *		search at the highest dmap control page level which
1326  *		distinctly describes the allocation group's free space
1327  *		(i.e. the highest level at which the allocation group's
1328  *		free space is not mixed in with that of any other group).
1329  *		in addition, we start the search within this level at a
1330  *		height of the dmapctl dmtree at which the nodes distinctly
1331  *		describe the allocation group's free space.  at this height,
1332  *		the allocation group's free space may be represented by 1
1333  *		or two sub-trees, depending on the allocation group size.
1334  *		we search the top nodes of these subtrees left to right for
1335  *		sufficient free space.  if sufficient free space is found,
1336  *		the subtree is searched to find the leftmost leaf that
1337  *		has free space.  once we have made it to the leaf, we
1338  *		move the search to the next lower level dmap control page
1339  *		corresponding to this leaf.  we continue down the dmap control
1340  *		pages until we find the dmap that contains or starts the
1341  *		sufficient free space and we allocate at this dmap.
1342  *
1343  *		if the allocation group size is equal to the dmap size,
1344  *		we'll start at the dmap corresponding to the allocation
1345  *		group and attempt the allocation at this level.
1346  *
1347  *		the dmap control page search is also not performed if the
1348  *		allocation group is completely free and we go to the first
1349  *		dmap of the allocation group to do the allocation.  this is
1350  *		done because the allocation group may be part (not the first
1351  *		part) of a larger binary buddy system, causing the dmap
1352  *		control pages to indicate no free space (NOFREE) within
1353  *		the allocation group.
1354  *
1355  * PARAMETERS:
1356  *      bmp	-  pointer to bmap descriptor
1357  *	agno	- allocation group number.
1358  *      nblocks	-  actual number of contiguous free blocks desired.
1359  *      l2nb	-  log2 number of contiguous free blocks desired.
1360  *      results	-  on successful return, set to the starting block number
1361  *		   of the newly allocated range.
1362  *
1363  * RETURN VALUES:
1364  *      0	- success
1365  *      -ENOSPC	- insufficient disk resources
1366  *      -EIO	- i/o error
1367  *
1368  * note: IWRITE_LOCK(ipmap) held on entry/exit;
1369  */
1370 static int
1371 dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results)
1372 {
1373 	struct metapage *mp;
1374 	struct dmapctl *dcp;
1375 	int rc, ti, i, k, m, n, agperlev;
1376 	s64 blkno, lblkno;
1377 	int budmin;
1378 
1379 	/* allocation request should not be for more than the
1380 	 * allocation group size.
1381 	 */
1382 	if (l2nb > bmp->db_agl2size) {
1383 		jfs_error(bmp->db_ipbmap->i_sb,
1384 			  "dbAllocAG: allocation request is larger than the "
1385 			  "allocation group size");
1386 		return -EIO;
1387 	}
1388 
1389 	/* determine the starting block number of the allocation
1390 	 * group.
1391 	 */
1392 	blkno = (s64) agno << bmp->db_agl2size;
1393 
1394 	/* check if the allocation group size is the minimum allocation
1395 	 * group size or if the allocation group is completely free. if
1396 	 * the allocation group size is the minimum size of BPERDMAP (i.e.
1397 	 * 1 dmap), there is no need to search the dmap control page (below)
1398 	 * that fully describes the allocation group since the allocation
1399 	 * group is already fully described by a dmap.  in this case, we
1400 	 * just call dbAllocCtl() to search the dmap tree and allocate the
1401 	 * required space if available.
1402 	 *
1403 	 * if the allocation group is completely free, dbAllocCtl() is
1404 	 * also called to allocate the required space.  this is done for
1405 	 * two reasons.  first, it makes no sense searching the dmap control
1406 	 * pages for free space when we know that free space exists.  second,
1407 	 * the dmap control pages may indicate that the allocation group
1408 	 * has no free space if the allocation group is part (not the first
1409 	 * part) of a larger binary buddy system.
1410 	 */
1411 	if (bmp->db_agsize == BPERDMAP
1412 	    || bmp->db_agfree[agno] == bmp->db_agsize) {
1413 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1414 		if ((rc == -ENOSPC) &&
1415 		    (bmp->db_agfree[agno] == bmp->db_agsize)) {
1416 			printk(KERN_ERR "blkno = %Lx, blocks = %Lx\n",
1417 			       (unsigned long long) blkno,
1418 			       (unsigned long long) nblocks);
1419 			jfs_error(bmp->db_ipbmap->i_sb,
1420 				  "dbAllocAG: dbAllocCtl failed in free AG");
1421 		}
1422 		return (rc);
1423 	}
1424 
1425 	/* the buffer for the dmap control page that fully describes the
1426 	 * allocation group.
1427 	 */
1428 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, bmp->db_aglevel);
1429 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1430 	if (mp == NULL)
1431 		return -EIO;
1432 	dcp = (struct dmapctl *) mp->data;
1433 	budmin = dcp->budmin;
1434 
1435 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1436 		jfs_error(bmp->db_ipbmap->i_sb,
1437 			  "dbAllocAG: Corrupt dmapctl page");
1438 		release_metapage(mp);
1439 		return -EIO;
1440 	}
1441 
1442 	/* search the subtree(s) of the dmap control page that describes
1443 	 * the allocation group, looking for sufficient free space.  to begin,
1444 	 * determine how many allocation groups are represented in a dmap
1445 	 * control page at the control page level (i.e. L0, L1, L2) that
1446 	 * fully describes an allocation group. next, determine the starting
1447 	 * tree index of this allocation group within the control page.
1448 	 */
1449 	agperlev =
1450 	    (1 << (L2LPERCTL - (bmp->db_agheigth << 1))) / bmp->db_agwidth;
1451 	ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
1452 
1453 	/* dmap control page trees fan-out by 4 and a single allocation
1454 	 * group may be described by 1 or 2 subtrees within the ag level
1455 	 * dmap control page, depending upon the ag size. examine the ag's
1456 	 * subtrees for sufficient free space, starting with the leftmost
1457 	 * subtree.
1458 	 */
1459 	for (i = 0; i < bmp->db_agwidth; i++, ti++) {
1460 		/* is there sufficient free space ?
1461 		 */
1462 		if (l2nb > dcp->stree[ti])
1463 			continue;
1464 
1465 		/* sufficient free space found in a subtree. now search down
1466 		 * the subtree to find the leftmost leaf that describes this
1467 		 * free space.
1468 		 */
1469 		for (k = bmp->db_agheigth; k > 0; k--) {
1470 			for (n = 0, m = (ti << 2) + 1; n < 4; n++) {
1471 				if (l2nb <= dcp->stree[m + n]) {
1472 					ti = m + n;
1473 					break;
1474 				}
1475 			}
1476 			if (n == 4) {
1477 				jfs_error(bmp->db_ipbmap->i_sb,
1478 					  "dbAllocAG: failed descending stree");
1479 				release_metapage(mp);
1480 				return -EIO;
1481 			}
1482 		}
1483 
1484 		/* determine the block number within the file system
1485 		 * that corresponds to this leaf.
1486 		 */
1487 		if (bmp->db_aglevel == 2)
1488 			blkno = 0;
1489 		else if (bmp->db_aglevel == 1)
1490 			blkno &= ~(MAXL1SIZE - 1);
1491 		else		/* bmp->db_aglevel == 0 */
1492 			blkno &= ~(MAXL0SIZE - 1);
1493 
1494 		blkno +=
1495 		    ((s64) (ti - le32_to_cpu(dcp->leafidx))) << budmin;
1496 
1497 		/* release the buffer in preparation for going down
1498 		 * the next level of dmap control pages.
1499 		 */
1500 		release_metapage(mp);
1501 
1502 		/* check if we need to continue to search down the lower
1503 		 * level dmap control pages.  we need to if the number of
1504 		 * blocks required is less than maximum number of blocks
1505 		 * described at the next lower level.
1506 		 */
1507 		if (l2nb < budmin) {
1508 
1509 			/* search the lower level dmap control pages to get
1510 			 * the starting block number of the dmap that
1511 			 * contains or starts off the free space.
1512 			 */
1513 			if ((rc =
1514 			     dbFindCtl(bmp, l2nb, bmp->db_aglevel - 1,
1515 				       &blkno))) {
1516 				if (rc == -ENOSPC) {
1517 					jfs_error(bmp->db_ipbmap->i_sb,
1518 						  "dbAllocAG: control page "
1519 						  "inconsistent");
1520 					return -EIO;
1521 				}
1522 				return (rc);
1523 			}
1524 		}
1525 
1526 		/* allocate the blocks.
1527 		 */
1528 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1529 		if (rc == -ENOSPC) {
1530 			jfs_error(bmp->db_ipbmap->i_sb,
1531 				  "dbAllocAG: unable to allocate blocks");
1532 			rc = -EIO;
1533 		}
1534 		return (rc);
1535 	}
1536 
1537 	/* no space in the allocation group.  release the buffer and
1538 	 * return -ENOSPC.
1539 	 */
1540 	release_metapage(mp);
1541 
1542 	return -ENOSPC;
1543 }
1544 
1545 
1546 /*
1547  * NAME:	dbAllocAny()
1548  *
1549  * FUNCTION:    attempt to allocate the specified number of contiguous
1550  *		free blocks anywhere in the file system.
1551  *
1552  *		dbAllocAny() attempts to find the sufficient free space by
1553  *		searching down the dmap control pages, starting with the
1554  *		highest level (i.e. L0, L1, L2) control page.  if free space
1555  *		large enough to satisfy the desired free space is found, the
1556  *		desired free space is allocated.
1557  *
1558  * PARAMETERS:
1559  *      bmp	-  pointer to bmap descriptor
1560  *      nblocks	 -  actual number of contiguous free blocks desired.
1561  *      l2nb	 -  log2 number of contiguous free blocks desired.
1562  *      results	-  on successful return, set to the starting block number
1563  *		   of the newly allocated range.
1564  *
1565  * RETURN VALUES:
1566  *      0	- success
1567  *      -ENOSPC	- insufficient disk resources
1568  *      -EIO	- i/o error
1569  *
1570  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1571  */
1572 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results)
1573 {
1574 	int rc;
1575 	s64 blkno = 0;
1576 
1577 	/* starting with the top level dmap control page, search
1578 	 * down the dmap control levels for sufficient free space.
1579 	 * if free space is found, dbFindCtl() returns the starting
1580 	 * block number of the dmap that contains or starts off the
1581 	 * range of free space.
1582 	 */
1583 	if ((rc = dbFindCtl(bmp, l2nb, bmp->db_maxlevel, &blkno)))
1584 		return (rc);
1585 
1586 	/* allocate the blocks.
1587 	 */
1588 	rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1589 	if (rc == -ENOSPC) {
1590 		jfs_error(bmp->db_ipbmap->i_sb,
1591 			  "dbAllocAny: unable to allocate blocks");
1592 		return -EIO;
1593 	}
1594 	return (rc);
1595 }
1596 
1597 
1598 /*
1599  * NAME:	dbFindCtl()
1600  *
1601  * FUNCTION:    starting at a specified dmap control page level and block
1602  *		number, search down the dmap control levels for a range of
1603  *	        contiguous free blocks large enough to satisfy an allocation
1604  *		request for the specified number of free blocks.
1605  *
1606  *		if sufficient contiguous free blocks are found, this routine
1607  *		returns the starting block number within a dmap page that
1608  *		contains or starts a range of contiqious free blocks that
1609  *		is sufficient in size.
1610  *
1611  * PARAMETERS:
1612  *      bmp	-  pointer to bmap descriptor
1613  *      level	-  starting dmap control page level.
1614  *      l2nb	-  log2 number of contiguous free blocks desired.
1615  *      *blkno	-  on entry, starting block number for conducting the search.
1616  *		   on successful return, the first block within a dmap page
1617  *		   that contains or starts a range of contiguous free blocks.
1618  *
1619  * RETURN VALUES:
1620  *      0	- success
1621  *      -ENOSPC	- insufficient disk resources
1622  *      -EIO	- i/o error
1623  *
1624  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1625  */
1626 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno)
1627 {
1628 	int rc, leafidx, lev;
1629 	s64 b, lblkno;
1630 	struct dmapctl *dcp;
1631 	int budmin;
1632 	struct metapage *mp;
1633 
1634 	/* starting at the specified dmap control page level and block
1635 	 * number, search down the dmap control levels for the starting
1636 	 * block number of a dmap page that contains or starts off
1637 	 * sufficient free blocks.
1638 	 */
1639 	for (lev = level, b = *blkno; lev >= 0; lev--) {
1640 		/* get the buffer of the dmap control page for the block
1641 		 * number and level (i.e. L0, L1, L2).
1642 		 */
1643 		lblkno = BLKTOCTL(b, bmp->db_l2nbperpage, lev);
1644 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1645 		if (mp == NULL)
1646 			return -EIO;
1647 		dcp = (struct dmapctl *) mp->data;
1648 		budmin = dcp->budmin;
1649 
1650 		if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1651 			jfs_error(bmp->db_ipbmap->i_sb,
1652 				  "dbFindCtl: Corrupt dmapctl page");
1653 			release_metapage(mp);
1654 			return -EIO;
1655 		}
1656 
1657 		/* search the tree within the dmap control page for
1658 		 * sufficent free space.  if sufficient free space is found,
1659 		 * dbFindLeaf() returns the index of the leaf at which
1660 		 * free space was found.
1661 		 */
1662 		rc = dbFindLeaf((dmtree_t *) dcp, l2nb, &leafidx);
1663 
1664 		/* release the buffer.
1665 		 */
1666 		release_metapage(mp);
1667 
1668 		/* space found ?
1669 		 */
1670 		if (rc) {
1671 			if (lev != level) {
1672 				jfs_error(bmp->db_ipbmap->i_sb,
1673 					  "dbFindCtl: dmap inconsistent");
1674 				return -EIO;
1675 			}
1676 			return -ENOSPC;
1677 		}
1678 
1679 		/* adjust the block number to reflect the location within
1680 		 * the dmap control page (i.e. the leaf) at which free
1681 		 * space was found.
1682 		 */
1683 		b += (((s64) leafidx) << budmin);
1684 
1685 		/* we stop the search at this dmap control page level if
1686 		 * the number of blocks required is greater than or equal
1687 		 * to the maximum number of blocks described at the next
1688 		 * (lower) level.
1689 		 */
1690 		if (l2nb >= budmin)
1691 			break;
1692 	}
1693 
1694 	*blkno = b;
1695 	return (0);
1696 }
1697 
1698 
1699 /*
1700  * NAME:	dbAllocCtl()
1701  *
1702  * FUNCTION:    attempt to allocate a specified number of contiguous
1703  *		blocks starting within a specific dmap.
1704  *
1705  *		this routine is called by higher level routines that search
1706  *		the dmap control pages above the actual dmaps for contiguous
1707  *		free space.  the result of successful searches by these
1708  *		routines are the starting block numbers within dmaps, with
1709  *		the dmaps themselves containing the desired contiguous free
1710  *		space or starting a contiguous free space of desired size
1711  *		that is made up of the blocks of one or more dmaps. these
1712  *		calls should not fail due to insufficent resources.
1713  *
1714  *		this routine is called in some cases where it is not known
1715  *		whether it will fail due to insufficient resources.  more
1716  *		specifically, this occurs when allocating from an allocation
1717  *		group whose size is equal to the number of blocks per dmap.
1718  *		in this case, the dmap control pages are not examined prior
1719  *		to calling this routine (to save pathlength) and the call
1720  *		might fail.
1721  *
1722  *		for a request size that fits within a dmap, this routine relies
1723  *		upon the dmap's dmtree to find the requested contiguous free
1724  *		space.  for request sizes that are larger than a dmap, the
1725  *		requested free space will start at the first block of the
1726  *		first dmap (i.e. blkno).
1727  *
1728  * PARAMETERS:
1729  *      bmp	-  pointer to bmap descriptor
1730  *      nblocks	 -  actual number of contiguous free blocks to allocate.
1731  *      l2nb	 -  log2 number of contiguous free blocks to allocate.
1732  *      blkno	 -  starting block number of the dmap to start the allocation
1733  *		    from.
1734  *      results	-  on successful return, set to the starting block number
1735  *		   of the newly allocated range.
1736  *
1737  * RETURN VALUES:
1738  *      0	- success
1739  *      -ENOSPC	- insufficient disk resources
1740  *      -EIO	- i/o error
1741  *
1742  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1743  */
1744 static int
1745 dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, s64 * results)
1746 {
1747 	int rc, nb;
1748 	s64 b, lblkno, n;
1749 	struct metapage *mp;
1750 	struct dmap *dp;
1751 
1752 	/* check if the allocation request is confined to a single dmap.
1753 	 */
1754 	if (l2nb <= L2BPERDMAP) {
1755 		/* get the buffer for the dmap.
1756 		 */
1757 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
1758 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1759 		if (mp == NULL)
1760 			return -EIO;
1761 		dp = (struct dmap *) mp->data;
1762 
1763 		/* try to allocate the blocks.
1764 		 */
1765 		rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results);
1766 		if (rc == 0)
1767 			mark_metapage_dirty(mp);
1768 
1769 		release_metapage(mp);
1770 
1771 		return (rc);
1772 	}
1773 
1774 	/* allocation request involving multiple dmaps. it must start on
1775 	 * a dmap boundary.
1776 	 */
1777 	assert((blkno & (BPERDMAP - 1)) == 0);
1778 
1779 	/* allocate the blocks dmap by dmap.
1780 	 */
1781 	for (n = nblocks, b = blkno; n > 0; n -= nb, b += nb) {
1782 		/* get the buffer for the dmap.
1783 		 */
1784 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1785 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1786 		if (mp == NULL) {
1787 			rc = -EIO;
1788 			goto backout;
1789 		}
1790 		dp = (struct dmap *) mp->data;
1791 
1792 		/* the dmap better be all free.
1793 		 */
1794 		if (dp->tree.stree[ROOT] != L2BPERDMAP) {
1795 			release_metapage(mp);
1796 			jfs_error(bmp->db_ipbmap->i_sb,
1797 				  "dbAllocCtl: the dmap is not all free");
1798 			rc = -EIO;
1799 			goto backout;
1800 		}
1801 
1802 		/* determine how many blocks to allocate from this dmap.
1803 		 */
1804 		nb = min(n, (s64)BPERDMAP);
1805 
1806 		/* allocate the blocks from the dmap.
1807 		 */
1808 		if ((rc = dbAllocDmap(bmp, dp, b, nb))) {
1809 			release_metapage(mp);
1810 			goto backout;
1811 		}
1812 
1813 		/* write the buffer.
1814 		 */
1815 		write_metapage(mp);
1816 	}
1817 
1818 	/* set the results (starting block number) and return.
1819 	 */
1820 	*results = blkno;
1821 	return (0);
1822 
1823 	/* something failed in handling an allocation request involving
1824 	 * multiple dmaps.  we'll try to clean up by backing out any
1825 	 * allocation that has already happened for this request.  if
1826 	 * we fail in backing out the allocation, we'll mark the file
1827 	 * system to indicate that blocks have been leaked.
1828 	 */
1829       backout:
1830 
1831 	/* try to backout the allocations dmap by dmap.
1832 	 */
1833 	for (n = nblocks - n, b = blkno; n > 0;
1834 	     n -= BPERDMAP, b += BPERDMAP) {
1835 		/* get the buffer for this dmap.
1836 		 */
1837 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1838 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1839 		if (mp == NULL) {
1840 			/* could not back out.  mark the file system
1841 			 * to indicate that we have leaked blocks.
1842 			 */
1843 			jfs_error(bmp->db_ipbmap->i_sb,
1844 				  "dbAllocCtl: I/O Error: Block Leakage.");
1845 			continue;
1846 		}
1847 		dp = (struct dmap *) mp->data;
1848 
1849 		/* free the blocks is this dmap.
1850 		 */
1851 		if (dbFreeDmap(bmp, dp, b, BPERDMAP)) {
1852 			/* could not back out.  mark the file system
1853 			 * to indicate that we have leaked blocks.
1854 			 */
1855 			release_metapage(mp);
1856 			jfs_error(bmp->db_ipbmap->i_sb,
1857 				  "dbAllocCtl: Block Leakage.");
1858 			continue;
1859 		}
1860 
1861 		/* write the buffer.
1862 		 */
1863 		write_metapage(mp);
1864 	}
1865 
1866 	return (rc);
1867 }
1868 
1869 
1870 /*
1871  * NAME:	dbAllocDmapLev()
1872  *
1873  * FUNCTION:    attempt to allocate a specified number of contiguous blocks
1874  *		from a specified dmap.
1875  *
1876  *		this routine checks if the contiguous blocks are available.
1877  *		if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1878  *		returned.
1879  *
1880  * PARAMETERS:
1881  *      mp	-  pointer to bmap descriptor
1882  *      dp	-  pointer to dmap to attempt to allocate blocks from.
1883  *      l2nb	-  log2 number of contiguous block desired.
1884  *      nblocks	-  actual number of contiguous block desired.
1885  *      results	-  on successful return, set to the starting block number
1886  *		   of the newly allocated range.
1887  *
1888  * RETURN VALUES:
1889  *      0	- success
1890  *      -ENOSPC	- insufficient disk resources
1891  *      -EIO	- i/o error
1892  *
1893  * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1894  *	IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1895  */
1896 static int
1897 dbAllocDmapLev(struct bmap * bmp,
1898 	       struct dmap * dp, int nblocks, int l2nb, s64 * results)
1899 {
1900 	s64 blkno;
1901 	int leafidx, rc;
1902 
1903 	/* can't be more than a dmaps worth of blocks */
1904 	assert(l2nb <= L2BPERDMAP);
1905 
1906 	/* search the tree within the dmap page for sufficient
1907 	 * free space.  if sufficient free space is found, dbFindLeaf()
1908 	 * returns the index of the leaf at which free space was found.
1909 	 */
1910 	if (dbFindLeaf((dmtree_t *) & dp->tree, l2nb, &leafidx))
1911 		return -ENOSPC;
1912 
1913 	/* determine the block number within the file system corresponding
1914 	 * to the leaf at which free space was found.
1915 	 */
1916 	blkno = le64_to_cpu(dp->start) + (leafidx << L2DBWORD);
1917 
1918 	/* if not all bits of the dmap word are free, get the starting
1919 	 * bit number within the dmap word of the required string of free
1920 	 * bits and adjust the block number with this value.
1921 	 */
1922 	if (dp->tree.stree[leafidx + LEAFIND] < BUDMIN)
1923 		blkno += dbFindBits(le32_to_cpu(dp->wmap[leafidx]), l2nb);
1924 
1925 	/* allocate the blocks */
1926 	if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1927 		*results = blkno;
1928 
1929 	return (rc);
1930 }
1931 
1932 
1933 /*
1934  * NAME:	dbAllocDmap()
1935  *
1936  * FUNCTION:    adjust the disk allocation map to reflect the allocation
1937  *		of a specified block range within a dmap.
1938  *
1939  *		this routine allocates the specified blocks from the dmap
1940  *		through a call to dbAllocBits(). if the allocation of the
1941  *		block range causes the maximum string of free blocks within
1942  *		the dmap to change (i.e. the value of the root of the dmap's
1943  *		dmtree), this routine will cause this change to be reflected
1944  *		up through the appropriate levels of the dmap control pages
1945  *		by a call to dbAdjCtl() for the L0 dmap control page that
1946  *		covers this dmap.
1947  *
1948  * PARAMETERS:
1949  *      bmp	-  pointer to bmap descriptor
1950  *      dp	-  pointer to dmap to allocate the block range from.
1951  *      blkno	-  starting block number of the block to be allocated.
1952  *      nblocks	-  number of blocks to be allocated.
1953  *
1954  * RETURN VALUES:
1955  *      0	- success
1956  *      -EIO	- i/o error
1957  *
1958  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
1959  */
1960 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
1961 		       int nblocks)
1962 {
1963 	s8 oldroot;
1964 	int rc;
1965 
1966 	/* save the current value of the root (i.e. maximum free string)
1967 	 * of the dmap tree.
1968 	 */
1969 	oldroot = dp->tree.stree[ROOT];
1970 
1971 	/* allocate the specified (blocks) bits */
1972 	dbAllocBits(bmp, dp, blkno, nblocks);
1973 
1974 	/* if the root has not changed, done. */
1975 	if (dp->tree.stree[ROOT] == oldroot)
1976 		return (0);
1977 
1978 	/* root changed. bubble the change up to the dmap control pages.
1979 	 * if the adjustment of the upper level control pages fails,
1980 	 * backout the bit allocation (thus making everything consistent).
1981 	 */
1982 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 1, 0)))
1983 		dbFreeBits(bmp, dp, blkno, nblocks);
1984 
1985 	return (rc);
1986 }
1987 
1988 
1989 /*
1990  * NAME:	dbFreeDmap()
1991  *
1992  * FUNCTION:    adjust the disk allocation map to reflect the allocation
1993  *		of a specified block range within a dmap.
1994  *
1995  *		this routine frees the specified blocks from the dmap through
1996  *		a call to dbFreeBits(). if the deallocation of the block range
1997  *		causes the maximum string of free blocks within the dmap to
1998  *		change (i.e. the value of the root of the dmap's dmtree), this
1999  *		routine will cause this change to be reflected up through the
2000  *	        appropriate levels of the dmap control pages by a call to
2001  *		dbAdjCtl() for the L0 dmap control page that covers this dmap.
2002  *
2003  * PARAMETERS:
2004  *      bmp	-  pointer to bmap descriptor
2005  *      dp	-  pointer to dmap to free the block range from.
2006  *      blkno	-  starting block number of the block to be freed.
2007  *      nblocks	-  number of blocks to be freed.
2008  *
2009  * RETURN VALUES:
2010  *      0	- success
2011  *      -EIO	- i/o error
2012  *
2013  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2014  */
2015 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2016 		      int nblocks)
2017 {
2018 	s8 oldroot;
2019 	int rc = 0, word;
2020 
2021 	/* save the current value of the root (i.e. maximum free string)
2022 	 * of the dmap tree.
2023 	 */
2024 	oldroot = dp->tree.stree[ROOT];
2025 
2026 	/* free the specified (blocks) bits */
2027 	rc = dbFreeBits(bmp, dp, blkno, nblocks);
2028 
2029 	/* if error or the root has not changed, done. */
2030 	if (rc || (dp->tree.stree[ROOT] == oldroot))
2031 		return (rc);
2032 
2033 	/* root changed. bubble the change up to the dmap control pages.
2034 	 * if the adjustment of the upper level control pages fails,
2035 	 * backout the deallocation.
2036 	 */
2037 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 0, 0))) {
2038 		word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
2039 
2040 		/* as part of backing out the deallocation, we will have
2041 		 * to back split the dmap tree if the deallocation caused
2042 		 * the freed blocks to become part of a larger binary buddy
2043 		 * system.
2044 		 */
2045 		if (dp->tree.stree[word] == NOFREE)
2046 			dbBackSplit((dmtree_t *) & dp->tree, word);
2047 
2048 		dbAllocBits(bmp, dp, blkno, nblocks);
2049 	}
2050 
2051 	return (rc);
2052 }
2053 
2054 
2055 /*
2056  * NAME:	dbAllocBits()
2057  *
2058  * FUNCTION:    allocate a specified block range from a dmap.
2059  *
2060  *		this routine updates the dmap to reflect the working
2061  *		state allocation of the specified block range. it directly
2062  *		updates the bits of the working map and causes the adjustment
2063  *		of the binary buddy system described by the dmap's dmtree
2064  *		leaves to reflect the bits allocated.  it also causes the
2065  *		dmap's dmtree, as a whole, to reflect the allocated range.
2066  *
2067  * PARAMETERS:
2068  *      bmp	-  pointer to bmap descriptor
2069  *      dp	-  pointer to dmap to allocate bits from.
2070  *      blkno	-  starting block number of the bits to be allocated.
2071  *      nblocks	-  number of bits to be allocated.
2072  *
2073  * RETURN VALUES: none
2074  *
2075  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2076  */
2077 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2078 			int nblocks)
2079 {
2080 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2081 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2082 	int size;
2083 	s8 *leaf;
2084 
2085 	/* pick up a pointer to the leaves of the dmap tree */
2086 	leaf = dp->tree.stree + LEAFIND;
2087 
2088 	/* determine the bit number and word within the dmap of the
2089 	 * starting block.
2090 	 */
2091 	dbitno = blkno & (BPERDMAP - 1);
2092 	word = dbitno >> L2DBWORD;
2093 
2094 	/* block range better be within the dmap */
2095 	assert(dbitno + nblocks <= BPERDMAP);
2096 
2097 	/* allocate the bits of the dmap's words corresponding to the block
2098 	 * range. not all bits of the first and last words may be contained
2099 	 * within the block range.  if this is the case, we'll work against
2100 	 * those words (i.e. partial first and/or last) on an individual basis
2101 	 * (a single pass), allocating the bits of interest by hand and
2102 	 * updating the leaf corresponding to the dmap word. a single pass
2103 	 * will be used for all dmap words fully contained within the
2104 	 * specified range.  within this pass, the bits of all fully contained
2105 	 * dmap words will be marked as free in a single shot and the leaves
2106 	 * will be updated. a single leaf may describe the free space of
2107 	 * multiple dmap words, so we may update only a subset of the actual
2108 	 * leaves corresponding to the dmap words of the block range.
2109 	 */
2110 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2111 		/* determine the bit number within the word and
2112 		 * the number of bits within the word.
2113 		 */
2114 		wbitno = dbitno & (DBWORD - 1);
2115 		nb = min(rembits, DBWORD - wbitno);
2116 
2117 		/* check if only part of a word is to be allocated.
2118 		 */
2119 		if (nb < DBWORD) {
2120 			/* allocate (set to 1) the appropriate bits within
2121 			 * this dmap word.
2122 			 */
2123 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
2124 						      >> wbitno);
2125 
2126 			/* update the leaf for this dmap word. in addition
2127 			 * to setting the leaf value to the binary buddy max
2128 			 * of the updated dmap word, dbSplit() will split
2129 			 * the binary system of the leaves if need be.
2130 			 */
2131 			dbSplit(tp, word, BUDMIN,
2132 				dbMaxBud((u8 *) & dp->wmap[word]));
2133 
2134 			word += 1;
2135 		} else {
2136 			/* one or more dmap words are fully contained
2137 			 * within the block range.  determine how many
2138 			 * words and allocate (set to 1) the bits of these
2139 			 * words.
2140 			 */
2141 			nwords = rembits >> L2DBWORD;
2142 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
2143 
2144 			/* determine how many bits.
2145 			 */
2146 			nb = nwords << L2DBWORD;
2147 
2148 			/* now update the appropriate leaves to reflect
2149 			 * the allocated words.
2150 			 */
2151 			for (; nwords > 0; nwords -= nw) {
2152 			        if (leaf[word] < BUDMIN) {
2153 					jfs_error(bmp->db_ipbmap->i_sb,
2154 						  "dbAllocBits: leaf page "
2155 						  "corrupt");
2156 					break;
2157 				}
2158 
2159 				/* determine what the leaf value should be
2160 				 * updated to as the minimum of the l2 number
2161 				 * of bits being allocated and the l2 number
2162 				 * of bits currently described by this leaf.
2163 				 */
2164 				size = min((int)leaf[word], NLSTOL2BSZ(nwords));
2165 
2166 				/* update the leaf to reflect the allocation.
2167 				 * in addition to setting the leaf value to
2168 				 * NOFREE, dbSplit() will split the binary
2169 				 * system of the leaves to reflect the current
2170 				 * allocation (size).
2171 				 */
2172 				dbSplit(tp, word, size, NOFREE);
2173 
2174 				/* get the number of dmap words handled */
2175 				nw = BUDSIZE(size, BUDMIN);
2176 				word += nw;
2177 			}
2178 		}
2179 	}
2180 
2181 	/* update the free count for this dmap */
2182 	dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) - nblocks);
2183 
2184 	BMAP_LOCK(bmp);
2185 
2186 	/* if this allocation group is completely free,
2187 	 * update the maximum allocation group number if this allocation
2188 	 * group is the new max.
2189 	 */
2190 	agno = blkno >> bmp->db_agl2size;
2191 	if (agno > bmp->db_maxag)
2192 		bmp->db_maxag = agno;
2193 
2194 	/* update the free count for the allocation group and map */
2195 	bmp->db_agfree[agno] -= nblocks;
2196 	bmp->db_nfree -= nblocks;
2197 
2198 	BMAP_UNLOCK(bmp);
2199 }
2200 
2201 
2202 /*
2203  * NAME:	dbFreeBits()
2204  *
2205  * FUNCTION:    free a specified block range from a dmap.
2206  *
2207  *		this routine updates the dmap to reflect the working
2208  *		state allocation of the specified block range. it directly
2209  *		updates the bits of the working map and causes the adjustment
2210  *		of the binary buddy system described by the dmap's dmtree
2211  *		leaves to reflect the bits freed.  it also causes the dmap's
2212  *		dmtree, as a whole, to reflect the deallocated range.
2213  *
2214  * PARAMETERS:
2215  *      bmp	-  pointer to bmap descriptor
2216  *      dp	-  pointer to dmap to free bits from.
2217  *      blkno	-  starting block number of the bits to be freed.
2218  *      nblocks	-  number of bits to be freed.
2219  *
2220  * RETURN VALUES: 0 for success
2221  *
2222  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2223  */
2224 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2225 		       int nblocks)
2226 {
2227 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2228 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2229 	int rc = 0;
2230 	int size;
2231 
2232 	/* determine the bit number and word within the dmap of the
2233 	 * starting block.
2234 	 */
2235 	dbitno = blkno & (BPERDMAP - 1);
2236 	word = dbitno >> L2DBWORD;
2237 
2238 	/* block range better be within the dmap.
2239 	 */
2240 	assert(dbitno + nblocks <= BPERDMAP);
2241 
2242 	/* free the bits of the dmaps words corresponding to the block range.
2243 	 * not all bits of the first and last words may be contained within
2244 	 * the block range.  if this is the case, we'll work against those
2245 	 * words (i.e. partial first and/or last) on an individual basis
2246 	 * (a single pass), freeing the bits of interest by hand and updating
2247 	 * the leaf corresponding to the dmap word. a single pass will be used
2248 	 * for all dmap words fully contained within the specified range.
2249 	 * within this pass, the bits of all fully contained dmap words will
2250 	 * be marked as free in a single shot and the leaves will be updated. a
2251 	 * single leaf may describe the free space of multiple dmap words,
2252 	 * so we may update only a subset of the actual leaves corresponding
2253 	 * to the dmap words of the block range.
2254 	 *
2255 	 * dbJoin() is used to update leaf values and will join the binary
2256 	 * buddy system of the leaves if the new leaf values indicate this
2257 	 * should be done.
2258 	 */
2259 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2260 		/* determine the bit number within the word and
2261 		 * the number of bits within the word.
2262 		 */
2263 		wbitno = dbitno & (DBWORD - 1);
2264 		nb = min(rembits, DBWORD - wbitno);
2265 
2266 		/* check if only part of a word is to be freed.
2267 		 */
2268 		if (nb < DBWORD) {
2269 			/* free (zero) the appropriate bits within this
2270 			 * dmap word.
2271 			 */
2272 			dp->wmap[word] &=
2273 			    cpu_to_le32(~(ONES << (DBWORD - nb)
2274 					  >> wbitno));
2275 
2276 			/* update the leaf for this dmap word.
2277 			 */
2278 			rc = dbJoin(tp, word,
2279 				    dbMaxBud((u8 *) & dp->wmap[word]));
2280 			if (rc)
2281 				return rc;
2282 
2283 			word += 1;
2284 		} else {
2285 			/* one or more dmap words are fully contained
2286 			 * within the block range.  determine how many
2287 			 * words and free (zero) the bits of these words.
2288 			 */
2289 			nwords = rembits >> L2DBWORD;
2290 			memset(&dp->wmap[word], 0, nwords * 4);
2291 
2292 			/* determine how many bits.
2293 			 */
2294 			nb = nwords << L2DBWORD;
2295 
2296 			/* now update the appropriate leaves to reflect
2297 			 * the freed words.
2298 			 */
2299 			for (; nwords > 0; nwords -= nw) {
2300 				/* determine what the leaf value should be
2301 				 * updated to as the minimum of the l2 number
2302 				 * of bits being freed and the l2 (max) number
2303 				 * of bits that can be described by this leaf.
2304 				 */
2305 				size =
2306 				    min(LITOL2BSZ
2307 					(word, L2LPERDMAP, BUDMIN),
2308 					NLSTOL2BSZ(nwords));
2309 
2310 				/* update the leaf.
2311 				 */
2312 				rc = dbJoin(tp, word, size);
2313 				if (rc)
2314 					return rc;
2315 
2316 				/* get the number of dmap words handled.
2317 				 */
2318 				nw = BUDSIZE(size, BUDMIN);
2319 				word += nw;
2320 			}
2321 		}
2322 	}
2323 
2324 	/* update the free count for this dmap.
2325 	 */
2326 	dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) + nblocks);
2327 
2328 	BMAP_LOCK(bmp);
2329 
2330 	/* update the free count for the allocation group and
2331 	 * map.
2332 	 */
2333 	agno = blkno >> bmp->db_agl2size;
2334 	bmp->db_nfree += nblocks;
2335 	bmp->db_agfree[agno] += nblocks;
2336 
2337 	/* check if this allocation group is not completely free and
2338 	 * if it is currently the maximum (rightmost) allocation group.
2339 	 * if so, establish the new maximum allocation group number by
2340 	 * searching left for the first allocation group with allocation.
2341 	 */
2342 	if ((bmp->db_agfree[agno] == bmp->db_agsize && agno == bmp->db_maxag) ||
2343 	    (agno == bmp->db_numag - 1 &&
2344 	     bmp->db_agfree[agno] == (bmp-> db_mapsize & (BPERDMAP - 1)))) {
2345 		while (bmp->db_maxag > 0) {
2346 			bmp->db_maxag -= 1;
2347 			if (bmp->db_agfree[bmp->db_maxag] !=
2348 			    bmp->db_agsize)
2349 				break;
2350 		}
2351 
2352 		/* re-establish the allocation group preference if the
2353 		 * current preference is right of the maximum allocation
2354 		 * group.
2355 		 */
2356 		if (bmp->db_agpref > bmp->db_maxag)
2357 			bmp->db_agpref = bmp->db_maxag;
2358 	}
2359 
2360 	BMAP_UNLOCK(bmp);
2361 
2362 	return 0;
2363 }
2364 
2365 
2366 /*
2367  * NAME:	dbAdjCtl()
2368  *
2369  * FUNCTION:	adjust a dmap control page at a specified level to reflect
2370  *		the change in a lower level dmap or dmap control page's
2371  *		maximum string of free blocks (i.e. a change in the root
2372  *		of the lower level object's dmtree) due to the allocation
2373  *		or deallocation of a range of blocks with a single dmap.
2374  *
2375  *		on entry, this routine is provided with the new value of
2376  *		the lower level dmap or dmap control page root and the
2377  *		starting block number of the block range whose allocation
2378  *		or deallocation resulted in the root change.  this range
2379  *		is respresented by a single leaf of the current dmapctl
2380  *		and the leaf will be updated with this value, possibly
2381  *		causing a binary buddy system within the leaves to be
2382  *		split or joined.  the update may also cause the dmapctl's
2383  *		dmtree to be updated.
2384  *
2385  *		if the adjustment of the dmap control page, itself, causes its
2386  *		root to change, this change will be bubbled up to the next dmap
2387  *		control level by a recursive call to this routine, specifying
2388  *		the new root value and the next dmap control page level to
2389  *		be adjusted.
2390  * PARAMETERS:
2391  *      bmp	-  pointer to bmap descriptor
2392  *      blkno	-  the first block of a block range within a dmap.  it is
2393  *		   the allocation or deallocation of this block range that
2394  *		   requires the dmap control page to be adjusted.
2395  *      newval	-  the new value of the lower level dmap or dmap control
2396  *		   page root.
2397  *      alloc	-  'true' if adjustment is due to an allocation.
2398  *      level	-  current level of dmap control page (i.e. L0, L1, L2) to
2399  *		   be adjusted.
2400  *
2401  * RETURN VALUES:
2402  *      0	- success
2403  *      -EIO	- i/o error
2404  *
2405  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2406  */
2407 static int
2408 dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, int level)
2409 {
2410 	struct metapage *mp;
2411 	s8 oldroot;
2412 	int oldval;
2413 	s64 lblkno;
2414 	struct dmapctl *dcp;
2415 	int rc, leafno, ti;
2416 
2417 	/* get the buffer for the dmap control page for the specified
2418 	 * block number and control page level.
2419 	 */
2420 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, level);
2421 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
2422 	if (mp == NULL)
2423 		return -EIO;
2424 	dcp = (struct dmapctl *) mp->data;
2425 
2426 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
2427 		jfs_error(bmp->db_ipbmap->i_sb,
2428 			  "dbAdjCtl: Corrupt dmapctl page");
2429 		release_metapage(mp);
2430 		return -EIO;
2431 	}
2432 
2433 	/* determine the leaf number corresponding to the block and
2434 	 * the index within the dmap control tree.
2435 	 */
2436 	leafno = BLKTOCTLLEAF(blkno, dcp->budmin);
2437 	ti = leafno + le32_to_cpu(dcp->leafidx);
2438 
2439 	/* save the current leaf value and the current root level (i.e.
2440 	 * maximum l2 free string described by this dmapctl).
2441 	 */
2442 	oldval = dcp->stree[ti];
2443 	oldroot = dcp->stree[ROOT];
2444 
2445 	/* check if this is a control page update for an allocation.
2446 	 * if so, update the leaf to reflect the new leaf value using
2447 	 * dbSplit(); otherwise (deallocation), use dbJoin() to udpate
2448 	 * the leaf with the new value.  in addition to updating the
2449 	 * leaf, dbSplit() will also split the binary buddy system of
2450 	 * the leaves, if required, and bubble new values within the
2451 	 * dmapctl tree, if required.  similarly, dbJoin() will join
2452 	 * the binary buddy system of leaves and bubble new values up
2453 	 * the dmapctl tree as required by the new leaf value.
2454 	 */
2455 	if (alloc) {
2456 		/* check if we are in the middle of a binary buddy
2457 		 * system.  this happens when we are performing the
2458 		 * first allocation out of an allocation group that
2459 		 * is part (not the first part) of a larger binary
2460 		 * buddy system.  if we are in the middle, back split
2461 		 * the system prior to calling dbSplit() which assumes
2462 		 * that it is at the front of a binary buddy system.
2463 		 */
2464 		if (oldval == NOFREE) {
2465 			rc = dbBackSplit((dmtree_t *) dcp, leafno);
2466 			if (rc)
2467 				return rc;
2468 			oldval = dcp->stree[ti];
2469 		}
2470 		dbSplit((dmtree_t *) dcp, leafno, dcp->budmin, newval);
2471 	} else {
2472 		rc = dbJoin((dmtree_t *) dcp, leafno, newval);
2473 		if (rc)
2474 			return rc;
2475 	}
2476 
2477 	/* check if the root of the current dmap control page changed due
2478 	 * to the update and if the current dmap control page is not at
2479 	 * the current top level (i.e. L0, L1, L2) of the map.  if so (i.e.
2480 	 * root changed and this is not the top level), call this routine
2481 	 * again (recursion) for the next higher level of the mapping to
2482 	 * reflect the change in root for the current dmap control page.
2483 	 */
2484 	if (dcp->stree[ROOT] != oldroot) {
2485 		/* are we below the top level of the map.  if so,
2486 		 * bubble the root up to the next higher level.
2487 		 */
2488 		if (level < bmp->db_maxlevel) {
2489 			/* bubble up the new root of this dmap control page to
2490 			 * the next level.
2491 			 */
2492 			if ((rc =
2493 			     dbAdjCtl(bmp, blkno, dcp->stree[ROOT], alloc,
2494 				      level + 1))) {
2495 				/* something went wrong in bubbling up the new
2496 				 * root value, so backout the changes to the
2497 				 * current dmap control page.
2498 				 */
2499 				if (alloc) {
2500 					dbJoin((dmtree_t *) dcp, leafno,
2501 					       oldval);
2502 				} else {
2503 					/* the dbJoin() above might have
2504 					 * caused a larger binary buddy system
2505 					 * to form and we may now be in the
2506 					 * middle of it.  if this is the case,
2507 					 * back split the buddies.
2508 					 */
2509 					if (dcp->stree[ti] == NOFREE)
2510 						dbBackSplit((dmtree_t *)
2511 							    dcp, leafno);
2512 					dbSplit((dmtree_t *) dcp, leafno,
2513 						dcp->budmin, oldval);
2514 				}
2515 
2516 				/* release the buffer and return the error.
2517 				 */
2518 				release_metapage(mp);
2519 				return (rc);
2520 			}
2521 		} else {
2522 			/* we're at the top level of the map. update
2523 			 * the bmap control page to reflect the size
2524 			 * of the maximum free buddy system.
2525 			 */
2526 			assert(level == bmp->db_maxlevel);
2527 			if (bmp->db_maxfreebud != oldroot) {
2528 				jfs_error(bmp->db_ipbmap->i_sb,
2529 					  "dbAdjCtl: the maximum free buddy is "
2530 					  "not the old root");
2531 			}
2532 			bmp->db_maxfreebud = dcp->stree[ROOT];
2533 		}
2534 	}
2535 
2536 	/* write the buffer.
2537 	 */
2538 	write_metapage(mp);
2539 
2540 	return (0);
2541 }
2542 
2543 
2544 /*
2545  * NAME:	dbSplit()
2546  *
2547  * FUNCTION:    update the leaf of a dmtree with a new value, splitting
2548  *		the leaf from the binary buddy system of the dmtree's
2549  *		leaves, as required.
2550  *
2551  * PARAMETERS:
2552  *      tp	- pointer to the tree containing the leaf.
2553  *      leafno	- the number of the leaf to be updated.
2554  *      splitsz	- the size the binary buddy system starting at the leaf
2555  *		  must be split to, specified as the log2 number of blocks.
2556  *      newval	- the new value for the leaf.
2557  *
2558  * RETURN VALUES: none
2559  *
2560  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2561  */
2562 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval)
2563 {
2564 	int budsz;
2565 	int cursz;
2566 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2567 
2568 	/* check if the leaf needs to be split.
2569 	 */
2570 	if (leaf[leafno] > tp->dmt_budmin) {
2571 		/* the split occurs by cutting the buddy system in half
2572 		 * at the specified leaf until we reach the specified
2573 		 * size.  pick up the starting split size (current size
2574 		 * - 1 in l2) and the corresponding buddy size.
2575 		 */
2576 		cursz = leaf[leafno] - 1;
2577 		budsz = BUDSIZE(cursz, tp->dmt_budmin);
2578 
2579 		/* split until we reach the specified size.
2580 		 */
2581 		while (cursz >= splitsz) {
2582 			/* update the buddy's leaf with its new value.
2583 			 */
2584 			dbAdjTree(tp, leafno ^ budsz, cursz);
2585 
2586 			/* on to the next size and buddy.
2587 			 */
2588 			cursz -= 1;
2589 			budsz >>= 1;
2590 		}
2591 	}
2592 
2593 	/* adjust the dmap tree to reflect the specified leaf's new
2594 	 * value.
2595 	 */
2596 	dbAdjTree(tp, leafno, newval);
2597 }
2598 
2599 
2600 /*
2601  * NAME:	dbBackSplit()
2602  *
2603  * FUNCTION:    back split the binary buddy system of dmtree leaves
2604  *		that hold a specified leaf until the specified leaf
2605  *		starts its own binary buddy system.
2606  *
2607  *		the allocators typically perform allocations at the start
2608  *		of binary buddy systems and dbSplit() is used to accomplish
2609  *		any required splits.  in some cases, however, allocation
2610  *		may occur in the middle of a binary system and requires a
2611  *		back split, with the split proceeding out from the middle of
2612  *		the system (less efficient) rather than the start of the
2613  *		system (more efficient).  the cases in which a back split
2614  *		is required are rare and are limited to the first allocation
2615  *		within an allocation group which is a part (not first part)
2616  *		of a larger binary buddy system and a few exception cases
2617  *		in which a previous join operation must be backed out.
2618  *
2619  * PARAMETERS:
2620  *      tp	- pointer to the tree containing the leaf.
2621  *      leafno	- the number of the leaf to be updated.
2622  *
2623  * RETURN VALUES: none
2624  *
2625  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2626  */
2627 static int dbBackSplit(dmtree_t * tp, int leafno)
2628 {
2629 	int budsz, bud, w, bsz, size;
2630 	int cursz;
2631 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2632 
2633 	/* leaf should be part (not first part) of a binary
2634 	 * buddy system.
2635 	 */
2636 	assert(leaf[leafno] == NOFREE);
2637 
2638 	/* the back split is accomplished by iteratively finding the leaf
2639 	 * that starts the buddy system that contains the specified leaf and
2640 	 * splitting that system in two.  this iteration continues until
2641 	 * the specified leaf becomes the start of a buddy system.
2642 	 *
2643 	 * determine maximum possible l2 size for the specified leaf.
2644 	 */
2645 	size =
2646 	    LITOL2BSZ(leafno, le32_to_cpu(tp->dmt_l2nleafs),
2647 		      tp->dmt_budmin);
2648 
2649 	/* determine the number of leaves covered by this size.  this
2650 	 * is the buddy size that we will start with as we search for
2651 	 * the buddy system that contains the specified leaf.
2652 	 */
2653 	budsz = BUDSIZE(size, tp->dmt_budmin);
2654 
2655 	/* back split.
2656 	 */
2657 	while (leaf[leafno] == NOFREE) {
2658 		/* find the leftmost buddy leaf.
2659 		 */
2660 		for (w = leafno, bsz = budsz;; bsz <<= 1,
2661 		     w = (w < bud) ? w : bud) {
2662 			if (bsz >= le32_to_cpu(tp->dmt_nleafs)) {
2663 				jfs_err("JFS: block map error in dbBackSplit");
2664 				return -EIO;
2665 			}
2666 
2667 			/* determine the buddy.
2668 			 */
2669 			bud = w ^ bsz;
2670 
2671 			/* check if this buddy is the start of the system.
2672 			 */
2673 			if (leaf[bud] != NOFREE) {
2674 				/* split the leaf at the start of the
2675 				 * system in two.
2676 				 */
2677 				cursz = leaf[bud] - 1;
2678 				dbSplit(tp, bud, cursz, cursz);
2679 				break;
2680 			}
2681 		}
2682 	}
2683 
2684 	if (leaf[leafno] != size) {
2685 		jfs_err("JFS: wrong leaf value in dbBackSplit");
2686 		return -EIO;
2687 	}
2688 	return 0;
2689 }
2690 
2691 
2692 /*
2693  * NAME:	dbJoin()
2694  *
2695  * FUNCTION:    update the leaf of a dmtree with a new value, joining
2696  *		the leaf with other leaves of the dmtree into a multi-leaf
2697  *		binary buddy system, as required.
2698  *
2699  * PARAMETERS:
2700  *      tp	- pointer to the tree containing the leaf.
2701  *      leafno	- the number of the leaf to be updated.
2702  *      newval	- the new value for the leaf.
2703  *
2704  * RETURN VALUES: none
2705  */
2706 static int dbJoin(dmtree_t * tp, int leafno, int newval)
2707 {
2708 	int budsz, buddy;
2709 	s8 *leaf;
2710 
2711 	/* can the new leaf value require a join with other leaves ?
2712 	 */
2713 	if (newval >= tp->dmt_budmin) {
2714 		/* pickup a pointer to the leaves of the tree.
2715 		 */
2716 		leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2717 
2718 		/* try to join the specified leaf into a large binary
2719 		 * buddy system.  the join proceeds by attempting to join
2720 		 * the specified leafno with its buddy (leaf) at new value.
2721 		 * if the join occurs, we attempt to join the left leaf
2722 		 * of the joined buddies with its buddy at new value + 1.
2723 		 * we continue to join until we find a buddy that cannot be
2724 		 * joined (does not have a value equal to the size of the
2725 		 * last join) or until all leaves have been joined into a
2726 		 * single system.
2727 		 *
2728 		 * get the buddy size (number of words covered) of
2729 		 * the new value.
2730 		 */
2731 		budsz = BUDSIZE(newval, tp->dmt_budmin);
2732 
2733 		/* try to join.
2734 		 */
2735 		while (budsz < le32_to_cpu(tp->dmt_nleafs)) {
2736 			/* get the buddy leaf.
2737 			 */
2738 			buddy = leafno ^ budsz;
2739 
2740 			/* if the leaf's new value is greater than its
2741 			 * buddy's value, we join no more.
2742 			 */
2743 			if (newval > leaf[buddy])
2744 				break;
2745 
2746 			/* It shouldn't be less */
2747 			if (newval < leaf[buddy])
2748 				return -EIO;
2749 
2750 			/* check which (leafno or buddy) is the left buddy.
2751 			 * the left buddy gets to claim the blocks resulting
2752 			 * from the join while the right gets to claim none.
2753 			 * the left buddy is also eligable to participate in
2754 			 * a join at the next higher level while the right
2755 			 * is not.
2756 			 *
2757 			 */
2758 			if (leafno < buddy) {
2759 				/* leafno is the left buddy.
2760 				 */
2761 				dbAdjTree(tp, buddy, NOFREE);
2762 			} else {
2763 				/* buddy is the left buddy and becomes
2764 				 * leafno.
2765 				 */
2766 				dbAdjTree(tp, leafno, NOFREE);
2767 				leafno = buddy;
2768 			}
2769 
2770 			/* on to try the next join.
2771 			 */
2772 			newval += 1;
2773 			budsz <<= 1;
2774 		}
2775 	}
2776 
2777 	/* update the leaf value.
2778 	 */
2779 	dbAdjTree(tp, leafno, newval);
2780 
2781 	return 0;
2782 }
2783 
2784 
2785 /*
2786  * NAME:	dbAdjTree()
2787  *
2788  * FUNCTION:    update a leaf of a dmtree with a new value, adjusting
2789  *		the dmtree, as required, to reflect the new leaf value.
2790  *		the combination of any buddies must already be done before
2791  *		this is called.
2792  *
2793  * PARAMETERS:
2794  *      tp	- pointer to the tree to be adjusted.
2795  *      leafno	- the number of the leaf to be updated.
2796  *      newval	- the new value for the leaf.
2797  *
2798  * RETURN VALUES: none
2799  */
2800 static void dbAdjTree(dmtree_t * tp, int leafno, int newval)
2801 {
2802 	int lp, pp, k;
2803 	int max;
2804 
2805 	/* pick up the index of the leaf for this leafno.
2806 	 */
2807 	lp = leafno + le32_to_cpu(tp->dmt_leafidx);
2808 
2809 	/* is the current value the same as the old value ?  if so,
2810 	 * there is nothing to do.
2811 	 */
2812 	if (tp->dmt_stree[lp] == newval)
2813 		return;
2814 
2815 	/* set the new value.
2816 	 */
2817 	tp->dmt_stree[lp] = newval;
2818 
2819 	/* bubble the new value up the tree as required.
2820 	 */
2821 	for (k = 0; k < le32_to_cpu(tp->dmt_height); k++) {
2822 		/* get the index of the first leaf of the 4 leaf
2823 		 * group containing the specified leaf (leafno).
2824 		 */
2825 		lp = ((lp - 1) & ~0x03) + 1;
2826 
2827 		/* get the index of the parent of this 4 leaf group.
2828 		 */
2829 		pp = (lp - 1) >> 2;
2830 
2831 		/* determine the maximum of the 4 leaves.
2832 		 */
2833 		max = TREEMAX(&tp->dmt_stree[lp]);
2834 
2835 		/* if the maximum of the 4 is the same as the
2836 		 * parent's value, we're done.
2837 		 */
2838 		if (tp->dmt_stree[pp] == max)
2839 			break;
2840 
2841 		/* parent gets new value.
2842 		 */
2843 		tp->dmt_stree[pp] = max;
2844 
2845 		/* parent becomes leaf for next go-round.
2846 		 */
2847 		lp = pp;
2848 	}
2849 }
2850 
2851 
2852 /*
2853  * NAME:	dbFindLeaf()
2854  *
2855  * FUNCTION:    search a dmtree_t for sufficient free blocks, returning
2856  *		the index of a leaf describing the free blocks if
2857  *		sufficient free blocks are found.
2858  *
2859  *		the search starts at the top of the dmtree_t tree and
2860  *		proceeds down the tree to the leftmost leaf with sufficient
2861  *		free space.
2862  *
2863  * PARAMETERS:
2864  *      tp	- pointer to the tree to be searched.
2865  *      l2nb	- log2 number of free blocks to search for.
2866  *	leafidx	- return pointer to be set to the index of the leaf
2867  *		  describing at least l2nb free blocks if sufficient
2868  *		  free blocks are found.
2869  *
2870  * RETURN VALUES:
2871  *      0	- success
2872  *      -ENOSPC	- insufficient free blocks.
2873  */
2874 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx)
2875 {
2876 	int ti, n = 0, k, x = 0;
2877 
2878 	/* first check the root of the tree to see if there is
2879 	 * sufficient free space.
2880 	 */
2881 	if (l2nb > tp->dmt_stree[ROOT])
2882 		return -ENOSPC;
2883 
2884 	/* sufficient free space available. now search down the tree
2885 	 * starting at the next level for the leftmost leaf that
2886 	 * describes sufficient free space.
2887 	 */
2888 	for (k = le32_to_cpu(tp->dmt_height), ti = 1;
2889 	     k > 0; k--, ti = ((ti + n) << 2) + 1) {
2890 		/* search the four nodes at this level, starting from
2891 		 * the left.
2892 		 */
2893 		for (x = ti, n = 0; n < 4; n++) {
2894 			/* sufficient free space found.  move to the next
2895 			 * level (or quit if this is the last level).
2896 			 */
2897 			if (l2nb <= tp->dmt_stree[x + n])
2898 				break;
2899 		}
2900 
2901 		/* better have found something since the higher
2902 		 * levels of the tree said it was here.
2903 		 */
2904 		assert(n < 4);
2905 	}
2906 
2907 	/* set the return to the leftmost leaf describing sufficient
2908 	 * free space.
2909 	 */
2910 	*leafidx = x + n - le32_to_cpu(tp->dmt_leafidx);
2911 
2912 	return (0);
2913 }
2914 
2915 
2916 /*
2917  * NAME:	dbFindBits()
2918  *
2919  * FUNCTION:    find a specified number of binary buddy free bits within a
2920  *		dmap bitmap word value.
2921  *
2922  *		this routine searches the bitmap value for (1 << l2nb) free
2923  *		bits at (1 << l2nb) alignments within the value.
2924  *
2925  * PARAMETERS:
2926  *      word	-  dmap bitmap word value.
2927  *      l2nb	-  number of free bits specified as a log2 number.
2928  *
2929  * RETURN VALUES:
2930  *      starting bit number of free bits.
2931  */
2932 static int dbFindBits(u32 word, int l2nb)
2933 {
2934 	int bitno, nb;
2935 	u32 mask;
2936 
2937 	/* get the number of bits.
2938 	 */
2939 	nb = 1 << l2nb;
2940 	assert(nb <= DBWORD);
2941 
2942 	/* complement the word so we can use a mask (i.e. 0s represent
2943 	 * free bits) and compute the mask.
2944 	 */
2945 	word = ~word;
2946 	mask = ONES << (DBWORD - nb);
2947 
2948 	/* scan the word for nb free bits at nb alignments.
2949 	 */
2950 	for (bitno = 0; mask != 0; bitno += nb, mask >>= nb) {
2951 		if ((mask & word) == mask)
2952 			break;
2953 	}
2954 
2955 	ASSERT(bitno < 32);
2956 
2957 	/* return the bit number.
2958 	 */
2959 	return (bitno);
2960 }
2961 
2962 
2963 /*
2964  * NAME:	dbMaxBud(u8 *cp)
2965  *
2966  * FUNCTION:    determine the largest binary buddy string of free
2967  *		bits within 32-bits of the map.
2968  *
2969  * PARAMETERS:
2970  *      cp	-  pointer to the 32-bit value.
2971  *
2972  * RETURN VALUES:
2973  *      largest binary buddy of free bits within a dmap word.
2974  */
2975 static int dbMaxBud(u8 * cp)
2976 {
2977 	signed char tmp1, tmp2;
2978 
2979 	/* check if the wmap word is all free. if so, the
2980 	 * free buddy size is BUDMIN.
2981 	 */
2982 	if (*((uint *) cp) == 0)
2983 		return (BUDMIN);
2984 
2985 	/* check if the wmap word is half free. if so, the
2986 	 * free buddy size is BUDMIN-1.
2987 	 */
2988 	if (*((u16 *) cp) == 0 || *((u16 *) cp + 1) == 0)
2989 		return (BUDMIN - 1);
2990 
2991 	/* not all free or half free. determine the free buddy
2992 	 * size thru table lookup using quarters of the wmap word.
2993 	 */
2994 	tmp1 = max(budtab[cp[2]], budtab[cp[3]]);
2995 	tmp2 = max(budtab[cp[0]], budtab[cp[1]]);
2996 	return (max(tmp1, tmp2));
2997 }
2998 
2999 
3000 /*
3001  * NAME:	cnttz(uint word)
3002  *
3003  * FUNCTION:    determine the number of trailing zeros within a 32-bit
3004  *		value.
3005  *
3006  * PARAMETERS:
3007  *      value	-  32-bit value to be examined.
3008  *
3009  * RETURN VALUES:
3010  *      count of trailing zeros
3011  */
3012 static int cnttz(u32 word)
3013 {
3014 	int n;
3015 
3016 	for (n = 0; n < 32; n++, word >>= 1) {
3017 		if (word & 0x01)
3018 			break;
3019 	}
3020 
3021 	return (n);
3022 }
3023 
3024 
3025 /*
3026  * NAME:	cntlz(u32 value)
3027  *
3028  * FUNCTION:    determine the number of leading zeros within a 32-bit
3029  *		value.
3030  *
3031  * PARAMETERS:
3032  *      value	-  32-bit value to be examined.
3033  *
3034  * RETURN VALUES:
3035  *      count of leading zeros
3036  */
3037 static int cntlz(u32 value)
3038 {
3039 	int n;
3040 
3041 	for (n = 0; n < 32; n++, value <<= 1) {
3042 		if (value & HIGHORDER)
3043 			break;
3044 	}
3045 	return (n);
3046 }
3047 
3048 
3049 /*
3050  * NAME:	blkstol2(s64 nb)
3051  *
3052  * FUNCTION:	convert a block count to its log2 value. if the block
3053  *	        count is not a l2 multiple, it is rounded up to the next
3054  *		larger l2 multiple.
3055  *
3056  * PARAMETERS:
3057  *      nb	-  number of blocks
3058  *
3059  * RETURN VALUES:
3060  *      log2 number of blocks
3061  */
3062 static int blkstol2(s64 nb)
3063 {
3064 	int l2nb;
3065 	s64 mask;		/* meant to be signed */
3066 
3067 	mask = (s64) 1 << (64 - 1);
3068 
3069 	/* count the leading bits.
3070 	 */
3071 	for (l2nb = 0; l2nb < 64; l2nb++, mask >>= 1) {
3072 		/* leading bit found.
3073 		 */
3074 		if (nb & mask) {
3075 			/* determine the l2 value.
3076 			 */
3077 			l2nb = (64 - 1) - l2nb;
3078 
3079 			/* check if we need to round up.
3080 			 */
3081 			if (~mask & nb)
3082 				l2nb++;
3083 
3084 			return (l2nb);
3085 		}
3086 	}
3087 	assert(0);
3088 	return 0;		/* fix compiler warning */
3089 }
3090 
3091 
3092 /*
3093  * NAME:	dbAllocBottomUp()
3094  *
3095  * FUNCTION:	alloc the specified block range from the working block
3096  *		allocation map.
3097  *
3098  *		the blocks will be alloc from the working map one dmap
3099  *		at a time.
3100  *
3101  * PARAMETERS:
3102  *      ip	-  pointer to in-core inode;
3103  *      blkno	-  starting block number to be freed.
3104  *      nblocks	-  number of blocks to be freed.
3105  *
3106  * RETURN VALUES:
3107  *      0	- success
3108  *      -EIO	- i/o error
3109  */
3110 int dbAllocBottomUp(struct inode *ip, s64 blkno, s64 nblocks)
3111 {
3112 	struct metapage *mp;
3113 	struct dmap *dp;
3114 	int nb, rc;
3115 	s64 lblkno, rem;
3116 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
3117 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
3118 
3119 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
3120 
3121 	/* block to be allocated better be within the mapsize. */
3122 	ASSERT(nblocks <= bmp->db_mapsize - blkno);
3123 
3124 	/*
3125 	 * allocate the blocks a dmap at a time.
3126 	 */
3127 	mp = NULL;
3128 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
3129 		/* release previous dmap if any */
3130 		if (mp) {
3131 			write_metapage(mp);
3132 		}
3133 
3134 		/* get the buffer for the current dmap. */
3135 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
3136 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
3137 		if (mp == NULL) {
3138 			IREAD_UNLOCK(ipbmap);
3139 			return -EIO;
3140 		}
3141 		dp = (struct dmap *) mp->data;
3142 
3143 		/* determine the number of blocks to be allocated from
3144 		 * this dmap.
3145 		 */
3146 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
3147 
3148 		/* allocate the blocks. */
3149 		if ((rc = dbAllocDmapBU(bmp, dp, blkno, nb))) {
3150 			release_metapage(mp);
3151 			IREAD_UNLOCK(ipbmap);
3152 			return (rc);
3153 		}
3154 	}
3155 
3156 	/* write the last buffer. */
3157 	write_metapage(mp);
3158 
3159 	IREAD_UNLOCK(ipbmap);
3160 
3161 	return (0);
3162 }
3163 
3164 
3165 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
3166 			 int nblocks)
3167 {
3168 	int rc;
3169 	int dbitno, word, rembits, nb, nwords, wbitno, agno;
3170 	s8 oldroot, *leaf;
3171 	struct dmaptree *tp = (struct dmaptree *) & dp->tree;
3172 
3173 	/* save the current value of the root (i.e. maximum free string)
3174 	 * of the dmap tree.
3175 	 */
3176 	oldroot = tp->stree[ROOT];
3177 
3178 	/* pick up a pointer to the leaves of the dmap tree */
3179 	leaf = tp->stree + LEAFIND;
3180 
3181 	/* determine the bit number and word within the dmap of the
3182 	 * starting block.
3183 	 */
3184 	dbitno = blkno & (BPERDMAP - 1);
3185 	word = dbitno >> L2DBWORD;
3186 
3187 	/* block range better be within the dmap */
3188 	assert(dbitno + nblocks <= BPERDMAP);
3189 
3190 	/* allocate the bits of the dmap's words corresponding to the block
3191 	 * range. not all bits of the first and last words may be contained
3192 	 * within the block range.  if this is the case, we'll work against
3193 	 * those words (i.e. partial first and/or last) on an individual basis
3194 	 * (a single pass), allocating the bits of interest by hand and
3195 	 * updating the leaf corresponding to the dmap word. a single pass
3196 	 * will be used for all dmap words fully contained within the
3197 	 * specified range.  within this pass, the bits of all fully contained
3198 	 * dmap words will be marked as free in a single shot and the leaves
3199 	 * will be updated. a single leaf may describe the free space of
3200 	 * multiple dmap words, so we may update only a subset of the actual
3201 	 * leaves corresponding to the dmap words of the block range.
3202 	 */
3203 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
3204 		/* determine the bit number within the word and
3205 		 * the number of bits within the word.
3206 		 */
3207 		wbitno = dbitno & (DBWORD - 1);
3208 		nb = min(rembits, DBWORD - wbitno);
3209 
3210 		/* check if only part of a word is to be allocated.
3211 		 */
3212 		if (nb < DBWORD) {
3213 			/* allocate (set to 1) the appropriate bits within
3214 			 * this dmap word.
3215 			 */
3216 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
3217 						      >> wbitno);
3218 
3219 			word++;
3220 		} else {
3221 			/* one or more dmap words are fully contained
3222 			 * within the block range.  determine how many
3223 			 * words and allocate (set to 1) the bits of these
3224 			 * words.
3225 			 */
3226 			nwords = rembits >> L2DBWORD;
3227 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
3228 
3229 			/* determine how many bits */
3230 			nb = nwords << L2DBWORD;
3231 			word += nwords;
3232 		}
3233 	}
3234 
3235 	/* update the free count for this dmap */
3236 	dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) - nblocks);
3237 
3238 	/* reconstruct summary tree */
3239 	dbInitDmapTree(dp);
3240 
3241 	BMAP_LOCK(bmp);
3242 
3243 	/* if this allocation group is completely free,
3244 	 * update the highest active allocation group number
3245 	 * if this allocation group is the new max.
3246 	 */
3247 	agno = blkno >> bmp->db_agl2size;
3248 	if (agno > bmp->db_maxag)
3249 		bmp->db_maxag = agno;
3250 
3251 	/* update the free count for the allocation group and map */
3252 	bmp->db_agfree[agno] -= nblocks;
3253 	bmp->db_nfree -= nblocks;
3254 
3255 	BMAP_UNLOCK(bmp);
3256 
3257 	/* if the root has not changed, done. */
3258 	if (tp->stree[ROOT] == oldroot)
3259 		return (0);
3260 
3261 	/* root changed. bubble the change up to the dmap control pages.
3262 	 * if the adjustment of the upper level control pages fails,
3263 	 * backout the bit allocation (thus making everything consistent).
3264 	 */
3265 	if ((rc = dbAdjCtl(bmp, blkno, tp->stree[ROOT], 1, 0)))
3266 		dbFreeBits(bmp, dp, blkno, nblocks);
3267 
3268 	return (rc);
3269 }
3270 
3271 
3272 /*
3273  * NAME:	dbExtendFS()
3274  *
3275  * FUNCTION:	extend bmap from blkno for nblocks;
3276  *		dbExtendFS() updates bmap ready for dbAllocBottomUp();
3277  *
3278  * L2
3279  *  |
3280  *   L1---------------------------------L1
3281  *    |                                  |
3282  *     L0---------L0---------L0           L0---------L0---------L0
3283  *      |          |          |            |          |          |
3284  *       d0,...,dn  d0,...,dn  d0,...,dn    d0,...,dn  d0,...,dn  d0,.,dm;
3285  * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3286  *
3287  * <---old---><----------------------------extend----------------------->
3288  */
3289 int dbExtendFS(struct inode *ipbmap, s64 blkno,	s64 nblocks)
3290 {
3291 	struct jfs_sb_info *sbi = JFS_SBI(ipbmap->i_sb);
3292 	int nbperpage = sbi->nbperpage;
3293 	int i, i0 = true, j, j0 = true, k, n;
3294 	s64 newsize;
3295 	s64 p;
3296 	struct metapage *mp, *l2mp, *l1mp = NULL, *l0mp = NULL;
3297 	struct dmapctl *l2dcp, *l1dcp, *l0dcp;
3298 	struct dmap *dp;
3299 	s8 *l0leaf, *l1leaf, *l2leaf;
3300 	struct bmap *bmp = sbi->bmap;
3301 	int agno, l2agsize, oldl2agsize;
3302 	s64 ag_rem;
3303 
3304 	newsize = blkno + nblocks;
3305 
3306 	jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld",
3307 		 (long long) blkno, (long long) nblocks, (long long) newsize);
3308 
3309 	/*
3310 	 *      initialize bmap control page.
3311 	 *
3312 	 * all the data in bmap control page should exclude
3313 	 * the mkfs hidden dmap page.
3314 	 */
3315 
3316 	/* update mapsize */
3317 	bmp->db_mapsize = newsize;
3318 	bmp->db_maxlevel = BMAPSZTOLEV(bmp->db_mapsize);
3319 
3320 	/* compute new AG size */
3321 	l2agsize = dbGetL2AGSize(newsize);
3322 	oldl2agsize = bmp->db_agl2size;
3323 
3324 	bmp->db_agl2size = l2agsize;
3325 	bmp->db_agsize = 1 << l2agsize;
3326 
3327 	/* compute new number of AG */
3328 	agno = bmp->db_numag;
3329 	bmp->db_numag = newsize >> l2agsize;
3330 	bmp->db_numag += ((u32) newsize % (u32) bmp->db_agsize) ? 1 : 0;
3331 
3332 	/*
3333 	 *      reconfigure db_agfree[]
3334 	 * from old AG configuration to new AG configuration;
3335 	 *
3336 	 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3337 	 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3338 	 * note: new AG size = old AG size * (2**x).
3339 	 */
3340 	if (l2agsize == oldl2agsize)
3341 		goto extend;
3342 	k = 1 << (l2agsize - oldl2agsize);
3343 	ag_rem = bmp->db_agfree[0];	/* save agfree[0] */
3344 	for (i = 0, n = 0; i < agno; n++) {
3345 		bmp->db_agfree[n] = 0;	/* init collection point */
3346 
3347 		/* coalesce cotiguous k AGs; */
3348 		for (j = 0; j < k && i < agno; j++, i++) {
3349 			/* merge AGi to AGn */
3350 			bmp->db_agfree[n] += bmp->db_agfree[i];
3351 		}
3352 	}
3353 	bmp->db_agfree[0] += ag_rem;	/* restore agfree[0] */
3354 
3355 	for (; n < MAXAG; n++)
3356 		bmp->db_agfree[n] = 0;
3357 
3358 	/*
3359 	 * update highest active ag number
3360 	 */
3361 
3362 	bmp->db_maxag = bmp->db_maxag / k;
3363 
3364 	/*
3365 	 *      extend bmap
3366 	 *
3367 	 * update bit maps and corresponding level control pages;
3368 	 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3369 	 */
3370       extend:
3371 	/* get L2 page */
3372 	p = BMAPBLKNO + nbperpage;	/* L2 page */
3373 	l2mp = read_metapage(ipbmap, p, PSIZE, 0);
3374 	if (!l2mp) {
3375 		jfs_error(ipbmap->i_sb, "dbExtendFS: L2 page could not be read");
3376 		return -EIO;
3377 	}
3378 	l2dcp = (struct dmapctl *) l2mp->data;
3379 
3380 	/* compute start L1 */
3381 	k = blkno >> L2MAXL1SIZE;
3382 	l2leaf = l2dcp->stree + CTLLEAFIND + k;
3383 	p = BLKTOL1(blkno, sbi->l2nbperpage);	/* L1 page */
3384 
3385 	/*
3386 	 * extend each L1 in L2
3387 	 */
3388 	for (; k < LPERCTL; k++, p += nbperpage) {
3389 		/* get L1 page */
3390 		if (j0) {
3391 			/* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3392 			l1mp = read_metapage(ipbmap, p, PSIZE, 0);
3393 			if (l1mp == NULL)
3394 				goto errout;
3395 			l1dcp = (struct dmapctl *) l1mp->data;
3396 
3397 			/* compute start L0 */
3398 			j = (blkno & (MAXL1SIZE - 1)) >> L2MAXL0SIZE;
3399 			l1leaf = l1dcp->stree + CTLLEAFIND + j;
3400 			p = BLKTOL0(blkno, sbi->l2nbperpage);
3401 			j0 = false;
3402 		} else {
3403 			/* assign/init L1 page */
3404 			l1mp = get_metapage(ipbmap, p, PSIZE, 0);
3405 			if (l1mp == NULL)
3406 				goto errout;
3407 
3408 			l1dcp = (struct dmapctl *) l1mp->data;
3409 
3410 			/* compute start L0 */
3411 			j = 0;
3412 			l1leaf = l1dcp->stree + CTLLEAFIND;
3413 			p += nbperpage;	/* 1st L0 of L1.k  */
3414 		}
3415 
3416 		/*
3417 		 * extend each L0 in L1
3418 		 */
3419 		for (; j < LPERCTL; j++) {
3420 			/* get L0 page */
3421 			if (i0) {
3422 				/* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3423 
3424 				l0mp = read_metapage(ipbmap, p, PSIZE, 0);
3425 				if (l0mp == NULL)
3426 					goto errout;
3427 				l0dcp = (struct dmapctl *) l0mp->data;
3428 
3429 				/* compute start dmap */
3430 				i = (blkno & (MAXL0SIZE - 1)) >>
3431 				    L2BPERDMAP;
3432 				l0leaf = l0dcp->stree + CTLLEAFIND + i;
3433 				p = BLKTODMAP(blkno,
3434 					      sbi->l2nbperpage);
3435 				i0 = false;
3436 			} else {
3437 				/* assign/init L0 page */
3438 				l0mp = get_metapage(ipbmap, p, PSIZE, 0);
3439 				if (l0mp == NULL)
3440 					goto errout;
3441 
3442 				l0dcp = (struct dmapctl *) l0mp->data;
3443 
3444 				/* compute start dmap */
3445 				i = 0;
3446 				l0leaf = l0dcp->stree + CTLLEAFIND;
3447 				p += nbperpage;	/* 1st dmap of L0.j */
3448 			}
3449 
3450 			/*
3451 			 * extend each dmap in L0
3452 			 */
3453 			for (; i < LPERCTL; i++) {
3454 				/*
3455 				 * reconstruct the dmap page, and
3456 				 * initialize corresponding parent L0 leaf
3457 				 */
3458 				if ((n = blkno & (BPERDMAP - 1))) {
3459 					/* read in dmap page: */
3460 					mp = read_metapage(ipbmap, p,
3461 							   PSIZE, 0);
3462 					if (mp == NULL)
3463 						goto errout;
3464 					n = min(nblocks, (s64)BPERDMAP - n);
3465 				} else {
3466 					/* assign/init dmap page */
3467 					mp = read_metapage(ipbmap, p,
3468 							   PSIZE, 0);
3469 					if (mp == NULL)
3470 						goto errout;
3471 
3472 					n = min(nblocks, (s64)BPERDMAP);
3473 				}
3474 
3475 				dp = (struct dmap *) mp->data;
3476 				*l0leaf = dbInitDmap(dp, blkno, n);
3477 
3478 				bmp->db_nfree += n;
3479 				agno = le64_to_cpu(dp->start) >> l2agsize;
3480 				bmp->db_agfree[agno] += n;
3481 
3482 				write_metapage(mp);
3483 
3484 				l0leaf++;
3485 				p += nbperpage;
3486 
3487 				blkno += n;
3488 				nblocks -= n;
3489 				if (nblocks == 0)
3490 					break;
3491 			}	/* for each dmap in a L0 */
3492 
3493 			/*
3494 			 * build current L0 page from its leaves, and
3495 			 * initialize corresponding parent L1 leaf
3496 			 */
3497 			*l1leaf = dbInitDmapCtl(l0dcp, 0, ++i);
3498 			write_metapage(l0mp);
3499 			l0mp = NULL;
3500 
3501 			if (nblocks)
3502 				l1leaf++;	/* continue for next L0 */
3503 			else {
3504 				/* more than 1 L0 ? */
3505 				if (j > 0)
3506 					break;	/* build L1 page */
3507 				else {
3508 					/* summarize in global bmap page */
3509 					bmp->db_maxfreebud = *l1leaf;
3510 					release_metapage(l1mp);
3511 					release_metapage(l2mp);
3512 					goto finalize;
3513 				}
3514 			}
3515 		}		/* for each L0 in a L1 */
3516 
3517 		/*
3518 		 * build current L1 page from its leaves, and
3519 		 * initialize corresponding parent L2 leaf
3520 		 */
3521 		*l2leaf = dbInitDmapCtl(l1dcp, 1, ++j);
3522 		write_metapage(l1mp);
3523 		l1mp = NULL;
3524 
3525 		if (nblocks)
3526 			l2leaf++;	/* continue for next L1 */
3527 		else {
3528 			/* more than 1 L1 ? */
3529 			if (k > 0)
3530 				break;	/* build L2 page */
3531 			else {
3532 				/* summarize in global bmap page */
3533 				bmp->db_maxfreebud = *l2leaf;
3534 				release_metapage(l2mp);
3535 				goto finalize;
3536 			}
3537 		}
3538 	}			/* for each L1 in a L2 */
3539 
3540 	jfs_error(ipbmap->i_sb,
3541 		  "dbExtendFS: function has not returned as expected");
3542 errout:
3543 	if (l0mp)
3544 		release_metapage(l0mp);
3545 	if (l1mp)
3546 		release_metapage(l1mp);
3547 	release_metapage(l2mp);
3548 	return -EIO;
3549 
3550 	/*
3551 	 *      finalize bmap control page
3552 	 */
3553 finalize:
3554 
3555 	return 0;
3556 }
3557 
3558 
3559 /*
3560  *	dbFinalizeBmap()
3561  */
3562 void dbFinalizeBmap(struct inode *ipbmap)
3563 {
3564 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
3565 	int actags, inactags, l2nl;
3566 	s64 ag_rem, actfree, inactfree, avgfree;
3567 	int i, n;
3568 
3569 	/*
3570 	 *      finalize bmap control page
3571 	 */
3572 //finalize:
3573 	/*
3574 	 * compute db_agpref: preferred ag to allocate from
3575 	 * (the leftmost ag with average free space in it);
3576 	 */
3577 //agpref:
3578 	/* get the number of active ags and inacitve ags */
3579 	actags = bmp->db_maxag + 1;
3580 	inactags = bmp->db_numag - actags;
3581 	ag_rem = bmp->db_mapsize & (bmp->db_agsize - 1);	/* ??? */
3582 
3583 	/* determine how many blocks are in the inactive allocation
3584 	 * groups. in doing this, we must account for the fact that
3585 	 * the rightmost group might be a partial group (i.e. file
3586 	 * system size is not a multiple of the group size).
3587 	 */
3588 	inactfree = (inactags && ag_rem) ?
3589 	    ((inactags - 1) << bmp->db_agl2size) + ag_rem
3590 	    : inactags << bmp->db_agl2size;
3591 
3592 	/* determine how many free blocks are in the active
3593 	 * allocation groups plus the average number of free blocks
3594 	 * within the active ags.
3595 	 */
3596 	actfree = bmp->db_nfree - inactfree;
3597 	avgfree = (u32) actfree / (u32) actags;
3598 
3599 	/* if the preferred allocation group has not average free space.
3600 	 * re-establish the preferred group as the leftmost
3601 	 * group with average free space.
3602 	 */
3603 	if (bmp->db_agfree[bmp->db_agpref] < avgfree) {
3604 		for (bmp->db_agpref = 0; bmp->db_agpref < actags;
3605 		     bmp->db_agpref++) {
3606 			if (bmp->db_agfree[bmp->db_agpref] >= avgfree)
3607 				break;
3608 		}
3609 		if (bmp->db_agpref >= bmp->db_numag) {
3610 			jfs_error(ipbmap->i_sb,
3611 				  "cannot find ag with average freespace");
3612 		}
3613 	}
3614 
3615 	/*
3616 	 * compute db_aglevel, db_agheigth, db_width, db_agstart:
3617 	 * an ag is covered in aglevel dmapctl summary tree,
3618 	 * at agheight level height (from leaf) with agwidth number of nodes
3619 	 * each, which starts at agstart index node of the smmary tree node
3620 	 * array;
3621 	 */
3622 	bmp->db_aglevel = BMAPSZTOLEV(bmp->db_agsize);
3623 	l2nl =
3624 	    bmp->db_agl2size - (L2BPERDMAP + bmp->db_aglevel * L2LPERCTL);
3625 	bmp->db_agheigth = l2nl >> 1;
3626 	bmp->db_agwidth = 1 << (l2nl - (bmp->db_agheigth << 1));
3627 	for (i = 5 - bmp->db_agheigth, bmp->db_agstart = 0, n = 1; i > 0;
3628 	     i--) {
3629 		bmp->db_agstart += n;
3630 		n <<= 2;
3631 	}
3632 
3633 }
3634 
3635 
3636 /*
3637  * NAME:	dbInitDmap()/ujfs_idmap_page()
3638  *
3639  * FUNCTION:	initialize working/persistent bitmap of the dmap page
3640  *		for the specified number of blocks:
3641  *
3642  *		at entry, the bitmaps had been initialized as free (ZEROS);
3643  *		The number of blocks will only account for the actually
3644  *		existing blocks. Blocks which don't actually exist in
3645  *		the aggregate will be marked as allocated (ONES);
3646  *
3647  * PARAMETERS:
3648  *	dp	- pointer to page of map
3649  *	nblocks	- number of blocks this page
3650  *
3651  * RETURNS: NONE
3652  */
3653 static int dbInitDmap(struct dmap * dp, s64 Blkno, int nblocks)
3654 {
3655 	int blkno, w, b, r, nw, nb, i;
3656 
3657 	/* starting block number within the dmap */
3658 	blkno = Blkno & (BPERDMAP - 1);
3659 
3660 	if (blkno == 0) {
3661 		dp->nblocks = dp->nfree = cpu_to_le32(nblocks);
3662 		dp->start = cpu_to_le64(Blkno);
3663 
3664 		if (nblocks == BPERDMAP) {
3665 			memset(&dp->wmap[0], 0, LPERDMAP * 4);
3666 			memset(&dp->pmap[0], 0, LPERDMAP * 4);
3667 			goto initTree;
3668 		}
3669 	} else {
3670 		dp->nblocks =
3671 		    cpu_to_le32(le32_to_cpu(dp->nblocks) + nblocks);
3672 		dp->nfree = cpu_to_le32(le32_to_cpu(dp->nfree) + nblocks);
3673 	}
3674 
3675 	/* word number containing start block number */
3676 	w = blkno >> L2DBWORD;
3677 
3678 	/*
3679 	 * free the bits corresponding to the block range (ZEROS):
3680 	 * note: not all bits of the first and last words may be contained
3681 	 * within the block range.
3682 	 */
3683 	for (r = nblocks; r > 0; r -= nb, blkno += nb) {
3684 		/* number of bits preceding range to be freed in the word */
3685 		b = blkno & (DBWORD - 1);
3686 		/* number of bits to free in the word */
3687 		nb = min(r, DBWORD - b);
3688 
3689 		/* is partial word to be freed ? */
3690 		if (nb < DBWORD) {
3691 			/* free (set to 0) from the bitmap word */
3692 			dp->wmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3693 						     >> b));
3694 			dp->pmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3695 						     >> b));
3696 
3697 			/* skip the word freed */
3698 			w++;
3699 		} else {
3700 			/* free (set to 0) contiguous bitmap words */
3701 			nw = r >> L2DBWORD;
3702 			memset(&dp->wmap[w], 0, nw * 4);
3703 			memset(&dp->pmap[w], 0, nw * 4);
3704 
3705 			/* skip the words freed */
3706 			nb = nw << L2DBWORD;
3707 			w += nw;
3708 		}
3709 	}
3710 
3711 	/*
3712 	 * mark bits following the range to be freed (non-existing
3713 	 * blocks) as allocated (ONES)
3714 	 */
3715 
3716 	if (blkno == BPERDMAP)
3717 		goto initTree;
3718 
3719 	/* the first word beyond the end of existing blocks */
3720 	w = blkno >> L2DBWORD;
3721 
3722 	/* does nblocks fall on a 32-bit boundary ? */
3723 	b = blkno & (DBWORD - 1);
3724 	if (b) {
3725 		/* mark a partial word allocated */
3726 		dp->wmap[w] = dp->pmap[w] = cpu_to_le32(ONES >> b);
3727 		w++;
3728 	}
3729 
3730 	/* set the rest of the words in the page to allocated (ONES) */
3731 	for (i = w; i < LPERDMAP; i++)
3732 		dp->pmap[i] = dp->wmap[i] = cpu_to_le32(ONES);
3733 
3734 	/*
3735 	 * init tree
3736 	 */
3737       initTree:
3738 	return (dbInitDmapTree(dp));
3739 }
3740 
3741 
3742 /*
3743  * NAME:	dbInitDmapTree()/ujfs_complete_dmap()
3744  *
3745  * FUNCTION:	initialize summary tree of the specified dmap:
3746  *
3747  *		at entry, bitmap of the dmap has been initialized;
3748  *
3749  * PARAMETERS:
3750  *	dp	- dmap to complete
3751  *	blkno	- starting block number for this dmap
3752  *	treemax	- will be filled in with max free for this dmap
3753  *
3754  * RETURNS:	max free string at the root of the tree
3755  */
3756 static int dbInitDmapTree(struct dmap * dp)
3757 {
3758 	struct dmaptree *tp;
3759 	s8 *cp;
3760 	int i;
3761 
3762 	/* init fixed info of tree */
3763 	tp = &dp->tree;
3764 	tp->nleafs = cpu_to_le32(LPERDMAP);
3765 	tp->l2nleafs = cpu_to_le32(L2LPERDMAP);
3766 	tp->leafidx = cpu_to_le32(LEAFIND);
3767 	tp->height = cpu_to_le32(4);
3768 	tp->budmin = BUDMIN;
3769 
3770 	/* init each leaf from corresponding wmap word:
3771 	 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3772 	 * bitmap word are allocated.
3773 	 */
3774 	cp = tp->stree + le32_to_cpu(tp->leafidx);
3775 	for (i = 0; i < LPERDMAP; i++)
3776 		*cp++ = dbMaxBud((u8 *) & dp->wmap[i]);
3777 
3778 	/* build the dmap's binary buddy summary tree */
3779 	return (dbInitTree(tp));
3780 }
3781 
3782 
3783 /*
3784  * NAME:	dbInitTree()/ujfs_adjtree()
3785  *
3786  * FUNCTION:	initialize binary buddy summary tree of a dmap or dmapctl.
3787  *
3788  *		at entry, the leaves of the tree has been initialized
3789  *		from corresponding bitmap word or root of summary tree
3790  *		of the child control page;
3791  *		configure binary buddy system at the leaf level, then
3792  *		bubble up the values of the leaf nodes up the tree.
3793  *
3794  * PARAMETERS:
3795  *	cp	- Pointer to the root of the tree
3796  *	l2leaves- Number of leaf nodes as a power of 2
3797  *	l2min	- Number of blocks that can be covered by a leaf
3798  *		  as a power of 2
3799  *
3800  * RETURNS: max free string at the root of the tree
3801  */
3802 static int dbInitTree(struct dmaptree * dtp)
3803 {
3804 	int l2max, l2free, bsize, nextb, i;
3805 	int child, parent, nparent;
3806 	s8 *tp, *cp, *cp1;
3807 
3808 	tp = dtp->stree;
3809 
3810 	/* Determine the maximum free string possible for the leaves */
3811 	l2max = le32_to_cpu(dtp->l2nleafs) + dtp->budmin;
3812 
3813 	/*
3814 	 * configure the leaf levevl into binary buddy system
3815 	 *
3816 	 * Try to combine buddies starting with a buddy size of 1
3817 	 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3818 	 * can be combined if both buddies have a maximum free of l2min;
3819 	 * the combination will result in the left-most buddy leaf having
3820 	 * a maximum free of l2min+1.
3821 	 * After processing all buddies for a given size, process buddies
3822 	 * at the next higher buddy size (i.e. current size * 2) and
3823 	 * the next maximum free (current free + 1).
3824 	 * This continues until the maximum possible buddy combination
3825 	 * yields maximum free.
3826 	 */
3827 	for (l2free = dtp->budmin, bsize = 1; l2free < l2max;
3828 	     l2free++, bsize = nextb) {
3829 		/* get next buddy size == current buddy pair size */
3830 		nextb = bsize << 1;
3831 
3832 		/* scan each adjacent buddy pair at current buddy size */
3833 		for (i = 0, cp = tp + le32_to_cpu(dtp->leafidx);
3834 		     i < le32_to_cpu(dtp->nleafs);
3835 		     i += nextb, cp += nextb) {
3836 			/* coalesce if both adjacent buddies are max free */
3837 			if (*cp == l2free && *(cp + bsize) == l2free) {
3838 				*cp = l2free + 1;	/* left take right */
3839 				*(cp + bsize) = -1;	/* right give left */
3840 			}
3841 		}
3842 	}
3843 
3844 	/*
3845 	 * bubble summary information of leaves up the tree.
3846 	 *
3847 	 * Starting at the leaf node level, the four nodes described by
3848 	 * the higher level parent node are compared for a maximum free and
3849 	 * this maximum becomes the value of the parent node.
3850 	 * when all lower level nodes are processed in this fashion then
3851 	 * move up to the next level (parent becomes a lower level node) and
3852 	 * continue the process for that level.
3853 	 */
3854 	for (child = le32_to_cpu(dtp->leafidx),
3855 	     nparent = le32_to_cpu(dtp->nleafs) >> 2;
3856 	     nparent > 0; nparent >>= 2, child = parent) {
3857 		/* get index of 1st node of parent level */
3858 		parent = (child - 1) >> 2;
3859 
3860 		/* set the value of the parent node as the maximum
3861 		 * of the four nodes of the current level.
3862 		 */
3863 		for (i = 0, cp = tp + child, cp1 = tp + parent;
3864 		     i < nparent; i++, cp += 4, cp1++)
3865 			*cp1 = TREEMAX(cp);
3866 	}
3867 
3868 	return (*tp);
3869 }
3870 
3871 
3872 /*
3873  *	dbInitDmapCtl()
3874  *
3875  * function: initialize dmapctl page
3876  */
3877 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i)
3878 {				/* start leaf index not covered by range */
3879 	s8 *cp;
3880 
3881 	dcp->nleafs = cpu_to_le32(LPERCTL);
3882 	dcp->l2nleafs = cpu_to_le32(L2LPERCTL);
3883 	dcp->leafidx = cpu_to_le32(CTLLEAFIND);
3884 	dcp->height = cpu_to_le32(5);
3885 	dcp->budmin = L2BPERDMAP + L2LPERCTL * level;
3886 
3887 	/*
3888 	 * initialize the leaves of current level that were not covered
3889 	 * by the specified input block range (i.e. the leaves have no
3890 	 * low level dmapctl or dmap).
3891 	 */
3892 	cp = &dcp->stree[CTLLEAFIND + i];
3893 	for (; i < LPERCTL; i++)
3894 		*cp++ = NOFREE;
3895 
3896 	/* build the dmap's binary buddy summary tree */
3897 	return (dbInitTree((struct dmaptree *) dcp));
3898 }
3899 
3900 
3901 /*
3902  * NAME:	dbGetL2AGSize()/ujfs_getagl2size()
3903  *
3904  * FUNCTION:	Determine log2(allocation group size) from aggregate size
3905  *
3906  * PARAMETERS:
3907  *	nblocks	- Number of blocks in aggregate
3908  *
3909  * RETURNS: log2(allocation group size) in aggregate blocks
3910  */
3911 static int dbGetL2AGSize(s64 nblocks)
3912 {
3913 	s64 sz;
3914 	s64 m;
3915 	int l2sz;
3916 
3917 	if (nblocks < BPERDMAP * MAXAG)
3918 		return (L2BPERDMAP);
3919 
3920 	/* round up aggregate size to power of 2 */
3921 	m = ((u64) 1 << (64 - 1));
3922 	for (l2sz = 64; l2sz >= 0; l2sz--, m >>= 1) {
3923 		if (m & nblocks)
3924 			break;
3925 	}
3926 
3927 	sz = (s64) 1 << l2sz;
3928 	if (sz < nblocks)
3929 		l2sz += 1;
3930 
3931 	/* agsize = roundupSize/max_number_of_ag */
3932 	return (l2sz - L2MAXAG);
3933 }
3934 
3935 
3936 /*
3937  * NAME:	dbMapFileSizeToMapSize()
3938  *
3939  * FUNCTION:	compute number of blocks the block allocation map file
3940  *		can cover from the map file size;
3941  *
3942  * RETURNS:	Number of blocks which can be covered by this block map file;
3943  */
3944 
3945 /*
3946  * maximum number of map pages at each level including control pages
3947  */
3948 #define MAXL0PAGES	(1 + LPERCTL)
3949 #define MAXL1PAGES	(1 + LPERCTL * MAXL0PAGES)
3950 #define MAXL2PAGES	(1 + LPERCTL * MAXL1PAGES)
3951 
3952 /*
3953  * convert number of map pages to the zero origin top dmapctl level
3954  */
3955 #define BMAPPGTOLEV(npages)	\
3956 	(((npages) <= 3 + MAXL0PAGES) ? 0 \
3957        : ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
3958 
3959 s64 dbMapFileSizeToMapSize(struct inode * ipbmap)
3960 {
3961 	struct super_block *sb = ipbmap->i_sb;
3962 	s64 nblocks;
3963 	s64 npages, ndmaps;
3964 	int level, i;
3965 	int complete, factor;
3966 
3967 	nblocks = ipbmap->i_size >> JFS_SBI(sb)->l2bsize;
3968 	npages = nblocks >> JFS_SBI(sb)->l2nbperpage;
3969 	level = BMAPPGTOLEV(npages);
3970 
3971 	/* At each level, accumulate the number of dmap pages covered by
3972 	 * the number of full child levels below it;
3973 	 * repeat for the last incomplete child level.
3974 	 */
3975 	ndmaps = 0;
3976 	npages--;		/* skip the first global control page */
3977 	/* skip higher level control pages above top level covered by map */
3978 	npages -= (2 - level);
3979 	npages--;		/* skip top level's control page */
3980 	for (i = level; i >= 0; i--) {
3981 		factor =
3982 		    (i == 2) ? MAXL1PAGES : ((i == 1) ? MAXL0PAGES : 1);
3983 		complete = (u32) npages / factor;
3984 		ndmaps += complete * ((i == 2) ? LPERCTL * LPERCTL
3985 				      : ((i == 1) ? LPERCTL : 1));
3986 
3987 		/* pages in last/incomplete child */
3988 		npages = (u32) npages % factor;
3989 		/* skip incomplete child's level control page */
3990 		npages--;
3991 	}
3992 
3993 	/* convert the number of dmaps into the number of blocks
3994 	 * which can be covered by the dmaps;
3995 	 */
3996 	nblocks = ndmaps << L2BPERDMAP;
3997 
3998 	return (nblocks);
3999 }
4000