xref: /openbmc/linux/fs/jfs/jfs_dmap.c (revision d6e2d652)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *   Copyright (C) International Business Machines Corp., 2000-2004
4  *   Portions Copyright (C) Tino Reichardt, 2012
5  */
6 
7 #include <linux/fs.h>
8 #include <linux/slab.h>
9 #include "jfs_incore.h"
10 #include "jfs_superblock.h"
11 #include "jfs_dmap.h"
12 #include "jfs_imap.h"
13 #include "jfs_lock.h"
14 #include "jfs_metapage.h"
15 #include "jfs_debug.h"
16 #include "jfs_discard.h"
17 
18 /*
19  *	SERIALIZATION of the Block Allocation Map.
20  *
21  *	the working state of the block allocation map is accessed in
22  *	two directions:
23  *
24  *	1) allocation and free requests that start at the dmap
25  *	   level and move up through the dmap control pages (i.e.
26  *	   the vast majority of requests).
27  *
28  *	2) allocation requests that start at dmap control page
29  *	   level and work down towards the dmaps.
30  *
31  *	the serialization scheme used here is as follows.
32  *
33  *	requests which start at the bottom are serialized against each
34  *	other through buffers and each requests holds onto its buffers
35  *	as it works it way up from a single dmap to the required level
36  *	of dmap control page.
37  *	requests that start at the top are serialized against each other
38  *	and request that start from the bottom by the multiple read/single
39  *	write inode lock of the bmap inode. requests starting at the top
40  *	take this lock in write mode while request starting at the bottom
41  *	take the lock in read mode.  a single top-down request may proceed
42  *	exclusively while multiple bottoms-up requests may proceed
43  *	simultaneously (under the protection of busy buffers).
44  *
45  *	in addition to information found in dmaps and dmap control pages,
46  *	the working state of the block allocation map also includes read/
47  *	write information maintained in the bmap descriptor (i.e. total
48  *	free block count, allocation group level free block counts).
49  *	a single exclusive lock (BMAP_LOCK) is used to guard this information
50  *	in the face of multiple-bottoms up requests.
51  *	(lock ordering: IREAD_LOCK, BMAP_LOCK);
52  *
53  *	accesses to the persistent state of the block allocation map (limited
54  *	to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
55  */
56 
57 #define BMAP_LOCK_INIT(bmp)	mutex_init(&bmp->db_bmaplock)
58 #define BMAP_LOCK(bmp)		mutex_lock(&bmp->db_bmaplock)
59 #define BMAP_UNLOCK(bmp)	mutex_unlock(&bmp->db_bmaplock)
60 
61 /*
62  * forward references
63  */
64 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
65 			int nblocks);
66 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval);
67 static int dbBackSplit(dmtree_t * tp, int leafno);
68 static int dbJoin(dmtree_t * tp, int leafno, int newval);
69 static void dbAdjTree(dmtree_t * tp, int leafno, int newval);
70 static int dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc,
71 		    int level);
72 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results);
73 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
74 		       int nblocks);
75 static int dbAllocNear(struct bmap * bmp, struct dmap * dp, s64 blkno,
76 		       int nblocks,
77 		       int l2nb, s64 * results);
78 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
79 		       int nblocks);
80 static int dbAllocDmapLev(struct bmap * bmp, struct dmap * dp, int nblocks,
81 			  int l2nb,
82 			  s64 * results);
83 static int dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb,
84 		     s64 * results);
85 static int dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno,
86 		      s64 * results);
87 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks);
88 static int dbFindBits(u32 word, int l2nb);
89 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno);
90 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx);
91 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
92 		      int nblocks);
93 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
94 		      int nblocks);
95 static int dbMaxBud(u8 * cp);
96 static int blkstol2(s64 nb);
97 
98 static int cntlz(u32 value);
99 static int cnttz(u32 word);
100 
101 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
102 			 int nblocks);
103 static int dbInitDmap(struct dmap * dp, s64 blkno, int nblocks);
104 static int dbInitDmapTree(struct dmap * dp);
105 static int dbInitTree(struct dmaptree * dtp);
106 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i);
107 static int dbGetL2AGSize(s64 nblocks);
108 
109 /*
110  *	buddy table
111  *
112  * table used for determining buddy sizes within characters of
113  * dmap bitmap words.  the characters themselves serve as indexes
114  * into the table, with the table elements yielding the maximum
115  * binary buddy of free bits within the character.
116  */
117 static const s8 budtab[256] = {
118 	3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
119 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
120 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
121 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
122 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
123 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
124 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
125 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
126 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
127 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
128 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
129 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
130 	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
131 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
132 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
133 	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1
134 };
135 
136 /*
137  * NAME:	dbMount()
138  *
139  * FUNCTION:	initializate the block allocation map.
140  *
141  *		memory is allocated for the in-core bmap descriptor and
142  *		the in-core descriptor is initialized from disk.
143  *
144  * PARAMETERS:
145  *	ipbmap	- pointer to in-core inode for the block map.
146  *
147  * RETURN VALUES:
148  *	0	- success
149  *	-ENOMEM	- insufficient memory
150  *	-EIO	- i/o error
151  *	-EINVAL - wrong bmap data
152  */
153 int dbMount(struct inode *ipbmap)
154 {
155 	struct bmap *bmp;
156 	struct dbmap_disk *dbmp_le;
157 	struct metapage *mp;
158 	int i, err;
159 
160 	/*
161 	 * allocate/initialize the in-memory bmap descriptor
162 	 */
163 	/* allocate memory for the in-memory bmap descriptor */
164 	bmp = kmalloc(sizeof(struct bmap), GFP_KERNEL);
165 	if (bmp == NULL)
166 		return -ENOMEM;
167 
168 	/* read the on-disk bmap descriptor. */
169 	mp = read_metapage(ipbmap,
170 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
171 			   PSIZE, 0);
172 	if (mp == NULL) {
173 		err = -EIO;
174 		goto err_kfree_bmp;
175 	}
176 
177 	/* copy the on-disk bmap descriptor to its in-memory version. */
178 	dbmp_le = (struct dbmap_disk *) mp->data;
179 	bmp->db_mapsize = le64_to_cpu(dbmp_le->dn_mapsize);
180 	bmp->db_nfree = le64_to_cpu(dbmp_le->dn_nfree);
181 
182 	bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage);
183 	if (bmp->db_l2nbperpage > L2PSIZE - L2MINBLOCKSIZE) {
184 		err = -EINVAL;
185 		goto err_release_metapage;
186 	}
187 
188 	bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag);
189 	if (!bmp->db_numag) {
190 		err = -EINVAL;
191 		goto err_release_metapage;
192 	}
193 
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_agheight = le32_to_cpu(dbmp_le->dn_agheight);
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 	if (bmp->db_agl2size > L2MAXL2SIZE - L2MAXAG ||
203 	    bmp->db_agl2size < 0) {
204 		err = -EINVAL;
205 		goto err_release_metapage;
206 	}
207 
208 	if (((bmp->db_mapsize - 1) >> bmp->db_agl2size) > MAXAG) {
209 		err = -EINVAL;
210 		goto err_release_metapage;
211 	}
212 
213 	for (i = 0; i < MAXAG; i++)
214 		bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
215 	bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
216 	bmp->db_maxfreebud = dbmp_le->dn_maxfreebud;
217 
218 	/* release the buffer. */
219 	release_metapage(mp);
220 
221 	/* bind the bmap inode and the bmap descriptor to each other. */
222 	bmp->db_ipbmap = ipbmap;
223 	JFS_SBI(ipbmap->i_sb)->bmap = bmp;
224 
225 	memset(bmp->db_active, 0, sizeof(bmp->db_active));
226 
227 	/*
228 	 * allocate/initialize the bmap lock
229 	 */
230 	BMAP_LOCK_INIT(bmp);
231 
232 	return (0);
233 
234 err_release_metapage:
235 	release_metapage(mp);
236 err_kfree_bmp:
237 	kfree(bmp);
238 	return err;
239 }
240 
241 
242 /*
243  * NAME:	dbUnmount()
244  *
245  * FUNCTION:	terminate the block allocation map in preparation for
246  *		file system unmount.
247  *
248  *		the in-core bmap descriptor is written to disk and
249  *		the memory for this descriptor is freed.
250  *
251  * PARAMETERS:
252  *	ipbmap	- pointer to in-core inode for the block map.
253  *
254  * RETURN VALUES:
255  *	0	- success
256  *	-EIO	- i/o error
257  */
258 int dbUnmount(struct inode *ipbmap, int mounterror)
259 {
260 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
261 
262 	if (!(mounterror || isReadOnly(ipbmap)))
263 		dbSync(ipbmap);
264 
265 	/*
266 	 * Invalidate the page cache buffers
267 	 */
268 	truncate_inode_pages(ipbmap->i_mapping, 0);
269 
270 	/* free the memory for the in-memory bmap. */
271 	kfree(bmp);
272 
273 	return (0);
274 }
275 
276 /*
277  *	dbSync()
278  */
279 int dbSync(struct inode *ipbmap)
280 {
281 	struct dbmap_disk *dbmp_le;
282 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
283 	struct metapage *mp;
284 	int i;
285 
286 	/*
287 	 * write bmap global control page
288 	 */
289 	/* get the buffer for the on-disk bmap descriptor. */
290 	mp = read_metapage(ipbmap,
291 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
292 			   PSIZE, 0);
293 	if (mp == NULL) {
294 		jfs_err("dbSync: read_metapage failed!");
295 		return -EIO;
296 	}
297 	/* copy the in-memory version of the bmap to the on-disk version */
298 	dbmp_le = (struct dbmap_disk *) mp->data;
299 	dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize);
300 	dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree);
301 	dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage);
302 	dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag);
303 	dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel);
304 	dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag);
305 	dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref);
306 	dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel);
307 	dbmp_le->dn_agheight = cpu_to_le32(bmp->db_agheight);
308 	dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth);
309 	dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart);
310 	dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size);
311 	for (i = 0; i < MAXAG; i++)
312 		dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]);
313 	dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize);
314 	dbmp_le->dn_maxfreebud = bmp->db_maxfreebud;
315 
316 	/* write the buffer */
317 	write_metapage(mp);
318 
319 	/*
320 	 * write out dirty pages of bmap
321 	 */
322 	filemap_write_and_wait(ipbmap->i_mapping);
323 
324 	diWriteSpecial(ipbmap, 0);
325 
326 	return (0);
327 }
328 
329 /*
330  * NAME:	dbFree()
331  *
332  * FUNCTION:	free the specified block range from the working block
333  *		allocation map.
334  *
335  *		the blocks will be free from the working map one dmap
336  *		at a time.
337  *
338  * PARAMETERS:
339  *	ip	- pointer to in-core inode;
340  *	blkno	- starting block number to be freed.
341  *	nblocks	- number of blocks to be freed.
342  *
343  * RETURN VALUES:
344  *	0	- success
345  *	-EIO	- i/o error
346  */
347 int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
348 {
349 	struct metapage *mp;
350 	struct dmap *dp;
351 	int nb, rc;
352 	s64 lblkno, rem;
353 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
354 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
355 	struct super_block *sb = ipbmap->i_sb;
356 
357 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
358 
359 	/* block to be freed better be within the mapsize. */
360 	if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
361 		IREAD_UNLOCK(ipbmap);
362 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
363 		       (unsigned long long) blkno,
364 		       (unsigned long long) nblocks);
365 		jfs_error(ip->i_sb, "block to be freed is outside the map\n");
366 		return -EIO;
367 	}
368 
369 	/**
370 	 * TRIM the blocks, when mounted with discard option
371 	 */
372 	if (JFS_SBI(sb)->flag & JFS_DISCARD)
373 		if (JFS_SBI(sb)->minblks_trim <= nblocks)
374 			jfs_issue_discard(ipbmap, blkno, nblocks);
375 
376 	/*
377 	 * free the blocks a dmap at a time.
378 	 */
379 	mp = NULL;
380 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
381 		/* release previous dmap if any */
382 		if (mp) {
383 			write_metapage(mp);
384 		}
385 
386 		/* get the buffer for the current dmap. */
387 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
388 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
389 		if (mp == NULL) {
390 			IREAD_UNLOCK(ipbmap);
391 			return -EIO;
392 		}
393 		dp = (struct dmap *) mp->data;
394 
395 		/* determine the number of blocks to be freed from
396 		 * this dmap.
397 		 */
398 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
399 
400 		/* free the blocks. */
401 		if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
402 			jfs_error(ip->i_sb, "error in block map\n");
403 			release_metapage(mp);
404 			IREAD_UNLOCK(ipbmap);
405 			return (rc);
406 		}
407 	}
408 
409 	/* write the last buffer. */
410 	if (mp)
411 		write_metapage(mp);
412 
413 	IREAD_UNLOCK(ipbmap);
414 
415 	return (0);
416 }
417 
418 
419 /*
420  * NAME:	dbUpdatePMap()
421  *
422  * FUNCTION:	update the allocation state (free or allocate) of the
423  *		specified block range in the persistent block allocation map.
424  *
425  *		the blocks will be updated in the persistent map one
426  *		dmap at a time.
427  *
428  * PARAMETERS:
429  *	ipbmap	- pointer to in-core inode for the block map.
430  *	free	- 'true' if block range is to be freed from the persistent
431  *		  map; 'false' if it is to be allocated.
432  *	blkno	- starting block number of the range.
433  *	nblocks	- number of contiguous blocks in the range.
434  *	tblk	- transaction block;
435  *
436  * RETURN VALUES:
437  *	0	- success
438  *	-EIO	- i/o error
439  */
440 int
441 dbUpdatePMap(struct inode *ipbmap,
442 	     int free, s64 blkno, s64 nblocks, struct tblock * tblk)
443 {
444 	int nblks, dbitno, wbitno, rbits;
445 	int word, nbits, nwords;
446 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
447 	s64 lblkno, rem, lastlblkno;
448 	u32 mask;
449 	struct dmap *dp;
450 	struct metapage *mp;
451 	struct jfs_log *log;
452 	int lsn, difft, diffp;
453 	unsigned long flags;
454 
455 	/* the blocks better be within the mapsize. */
456 	if (blkno + nblocks > bmp->db_mapsize) {
457 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
458 		       (unsigned long long) blkno,
459 		       (unsigned long long) nblocks);
460 		jfs_error(ipbmap->i_sb, "blocks are outside the map\n");
461 		return -EIO;
462 	}
463 
464 	/* compute delta of transaction lsn from log syncpt */
465 	lsn = tblk->lsn;
466 	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
467 	logdiff(difft, lsn, log);
468 
469 	/*
470 	 * update the block state a dmap at a time.
471 	 */
472 	mp = NULL;
473 	lastlblkno = 0;
474 	for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) {
475 		/* get the buffer for the current dmap. */
476 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
477 		if (lblkno != lastlblkno) {
478 			if (mp) {
479 				write_metapage(mp);
480 			}
481 
482 			mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE,
483 					   0);
484 			if (mp == NULL)
485 				return -EIO;
486 			metapage_wait_for_io(mp);
487 		}
488 		dp = (struct dmap *) mp->data;
489 
490 		/* determine the bit number and word within the dmap of
491 		 * the starting block.  also determine how many blocks
492 		 * are to be updated within this dmap.
493 		 */
494 		dbitno = blkno & (BPERDMAP - 1);
495 		word = dbitno >> L2DBWORD;
496 		nblks = min(rem, (s64)BPERDMAP - dbitno);
497 
498 		/* update the bits of the dmap words. the first and last
499 		 * words may only have a subset of their bits updated. if
500 		 * this is the case, we'll work against that word (i.e.
501 		 * partial first and/or last) only in a single pass.  a
502 		 * single pass will also be used to update all words that
503 		 * are to have all their bits updated.
504 		 */
505 		for (rbits = nblks; rbits > 0;
506 		     rbits -= nbits, dbitno += nbits) {
507 			/* determine the bit number within the word and
508 			 * the number of bits within the word.
509 			 */
510 			wbitno = dbitno & (DBWORD - 1);
511 			nbits = min(rbits, DBWORD - wbitno);
512 
513 			/* check if only part of the word is to be updated. */
514 			if (nbits < DBWORD) {
515 				/* update (free or allocate) the bits
516 				 * in this word.
517 				 */
518 				mask =
519 				    (ONES << (DBWORD - nbits) >> wbitno);
520 				if (free)
521 					dp->pmap[word] &=
522 					    cpu_to_le32(~mask);
523 				else
524 					dp->pmap[word] |=
525 					    cpu_to_le32(mask);
526 
527 				word += 1;
528 			} else {
529 				/* one or more words are to have all
530 				 * their bits updated.  determine how
531 				 * many words and how many bits.
532 				 */
533 				nwords = rbits >> L2DBWORD;
534 				nbits = nwords << L2DBWORD;
535 
536 				/* update (free or allocate) the bits
537 				 * in these words.
538 				 */
539 				if (free)
540 					memset(&dp->pmap[word], 0,
541 					       nwords * 4);
542 				else
543 					memset(&dp->pmap[word], (int) ONES,
544 					       nwords * 4);
545 
546 				word += nwords;
547 			}
548 		}
549 
550 		/*
551 		 * update dmap lsn
552 		 */
553 		if (lblkno == lastlblkno)
554 			continue;
555 
556 		lastlblkno = lblkno;
557 
558 		LOGSYNC_LOCK(log, flags);
559 		if (mp->lsn != 0) {
560 			/* inherit older/smaller lsn */
561 			logdiff(diffp, mp->lsn, log);
562 			if (difft < diffp) {
563 				mp->lsn = lsn;
564 
565 				/* move bp after tblock in logsync list */
566 				list_move(&mp->synclist, &tblk->synclist);
567 			}
568 
569 			/* inherit younger/larger clsn */
570 			logdiff(difft, tblk->clsn, log);
571 			logdiff(diffp, mp->clsn, log);
572 			if (difft > diffp)
573 				mp->clsn = tblk->clsn;
574 		} else {
575 			mp->log = log;
576 			mp->lsn = lsn;
577 
578 			/* insert bp after tblock in logsync list */
579 			log->count++;
580 			list_add(&mp->synclist, &tblk->synclist);
581 
582 			mp->clsn = tblk->clsn;
583 		}
584 		LOGSYNC_UNLOCK(log, flags);
585 	}
586 
587 	/* write the last buffer. */
588 	if (mp) {
589 		write_metapage(mp);
590 	}
591 
592 	return (0);
593 }
594 
595 
596 /*
597  * NAME:	dbNextAG()
598  *
599  * FUNCTION:	find the preferred allocation group for new allocations.
600  *
601  *		Within the allocation groups, we maintain a preferred
602  *		allocation group which consists of a group with at least
603  *		average free space.  It is the preferred group that we target
604  *		new inode allocation towards.  The tie-in between inode
605  *		allocation and block allocation occurs as we allocate the
606  *		first (data) block of an inode and specify the inode (block)
607  *		as the allocation hint for this block.
608  *
609  *		We try to avoid having more than one open file growing in
610  *		an allocation group, as this will lead to fragmentation.
611  *		This differs from the old OS/2 method of trying to keep
612  *		empty ags around for large allocations.
613  *
614  * PARAMETERS:
615  *	ipbmap	- pointer to in-core inode for the block map.
616  *
617  * RETURN VALUES:
618  *	the preferred allocation group number.
619  */
620 int dbNextAG(struct inode *ipbmap)
621 {
622 	s64 avgfree;
623 	int agpref;
624 	s64 hwm = 0;
625 	int i;
626 	int next_best = -1;
627 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
628 
629 	BMAP_LOCK(bmp);
630 
631 	/* determine the average number of free blocks within the ags. */
632 	avgfree = (u32)bmp->db_nfree / bmp->db_numag;
633 
634 	/*
635 	 * if the current preferred ag does not have an active allocator
636 	 * and has at least average freespace, return it
637 	 */
638 	agpref = bmp->db_agpref;
639 	if ((atomic_read(&bmp->db_active[agpref]) == 0) &&
640 	    (bmp->db_agfree[agpref] >= avgfree))
641 		goto unlock;
642 
643 	/* From the last preferred ag, find the next one with at least
644 	 * average free space.
645 	 */
646 	for (i = 0 ; i < bmp->db_numag; i++, agpref++) {
647 		if (agpref == bmp->db_numag)
648 			agpref = 0;
649 
650 		if (atomic_read(&bmp->db_active[agpref]))
651 			/* open file is currently growing in this ag */
652 			continue;
653 		if (bmp->db_agfree[agpref] >= avgfree) {
654 			/* Return this one */
655 			bmp->db_agpref = agpref;
656 			goto unlock;
657 		} else if (bmp->db_agfree[agpref] > hwm) {
658 			/* Less than avg. freespace, but best so far */
659 			hwm = bmp->db_agfree[agpref];
660 			next_best = agpref;
661 		}
662 	}
663 
664 	/*
665 	 * If no inactive ag was found with average freespace, use the
666 	 * next best
667 	 */
668 	if (next_best != -1)
669 		bmp->db_agpref = next_best;
670 	/* else leave db_agpref unchanged */
671 unlock:
672 	BMAP_UNLOCK(bmp);
673 
674 	/* return the preferred group.
675 	 */
676 	return (bmp->db_agpref);
677 }
678 
679 /*
680  * NAME:	dbAlloc()
681  *
682  * FUNCTION:	attempt to allocate a specified number of contiguous free
683  *		blocks from the working allocation block map.
684  *
685  *		the block allocation policy uses hints and a multi-step
686  *		approach.
687  *
688  *		for allocation requests smaller than the number of blocks
689  *		per dmap, we first try to allocate the new blocks
690  *		immediately following the hint.  if these blocks are not
691  *		available, we try to allocate blocks near the hint.  if
692  *		no blocks near the hint are available, we next try to
693  *		allocate within the same dmap as contains the hint.
694  *
695  *		if no blocks are available in the dmap or the allocation
696  *		request is larger than the dmap size, we try to allocate
697  *		within the same allocation group as contains the hint. if
698  *		this does not succeed, we finally try to allocate anywhere
699  *		within the aggregate.
700  *
701  *		we also try to allocate anywhere within the aggregate
702  *		for allocation requests larger than the allocation group
703  *		size or requests that specify no hint value.
704  *
705  * PARAMETERS:
706  *	ip	- pointer to in-core inode;
707  *	hint	- allocation hint.
708  *	nblocks	- number of contiguous blocks in the range.
709  *	results	- on successful return, set to the starting block number
710  *		  of the newly allocated contiguous range.
711  *
712  * RETURN VALUES:
713  *	0	- success
714  *	-ENOSPC	- insufficient disk resources
715  *	-EIO	- i/o error
716  */
717 int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results)
718 {
719 	int rc, agno;
720 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
721 	struct bmap *bmp;
722 	struct metapage *mp;
723 	s64 lblkno, blkno;
724 	struct dmap *dp;
725 	int l2nb;
726 	s64 mapSize;
727 	int writers;
728 
729 	/* assert that nblocks is valid */
730 	assert(nblocks > 0);
731 
732 	/* get the log2 number of blocks to be allocated.
733 	 * if the number of blocks is not a log2 multiple,
734 	 * it will be rounded up to the next log2 multiple.
735 	 */
736 	l2nb = BLKSTOL2(nblocks);
737 
738 	bmp = JFS_SBI(ip->i_sb)->bmap;
739 
740 	mapSize = bmp->db_mapsize;
741 
742 	/* the hint should be within the map */
743 	if (hint >= mapSize) {
744 		jfs_error(ip->i_sb, "the hint is outside the map\n");
745 		return -EIO;
746 	}
747 
748 	/* if the number of blocks to be allocated is greater than the
749 	 * allocation group size, try to allocate anywhere.
750 	 */
751 	if (l2nb > bmp->db_agl2size) {
752 		IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
753 
754 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
755 
756 		goto write_unlock;
757 	}
758 
759 	/*
760 	 * If no hint, let dbNextAG recommend an allocation group
761 	 */
762 	if (hint == 0)
763 		goto pref_ag;
764 
765 	/* we would like to allocate close to the hint.  adjust the
766 	 * hint to the block following the hint since the allocators
767 	 * will start looking for free space starting at this point.
768 	 */
769 	blkno = hint + 1;
770 
771 	if (blkno >= bmp->db_mapsize)
772 		goto pref_ag;
773 
774 	agno = blkno >> bmp->db_agl2size;
775 
776 	/* check if blkno crosses over into a new allocation group.
777 	 * if so, check if we should allow allocations within this
778 	 * allocation group.
779 	 */
780 	if ((blkno & (bmp->db_agsize - 1)) == 0)
781 		/* check if the AG is currently being written to.
782 		 * if so, call dbNextAG() to find a non-busy
783 		 * AG with sufficient free space.
784 		 */
785 		if (atomic_read(&bmp->db_active[agno]))
786 			goto pref_ag;
787 
788 	/* check if the allocation request size can be satisfied from a
789 	 * single dmap.  if so, try to allocate from the dmap containing
790 	 * the hint using a tiered strategy.
791 	 */
792 	if (nblocks <= BPERDMAP) {
793 		IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
794 
795 		/* get the buffer for the dmap containing the hint.
796 		 */
797 		rc = -EIO;
798 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
799 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
800 		if (mp == NULL)
801 			goto read_unlock;
802 
803 		dp = (struct dmap *) mp->data;
804 
805 		/* first, try to satisfy the allocation request with the
806 		 * blocks beginning at the hint.
807 		 */
808 		if ((rc = dbAllocNext(bmp, dp, blkno, (int) nblocks))
809 		    != -ENOSPC) {
810 			if (rc == 0) {
811 				*results = blkno;
812 				mark_metapage_dirty(mp);
813 			}
814 
815 			release_metapage(mp);
816 			goto read_unlock;
817 		}
818 
819 		writers = atomic_read(&bmp->db_active[agno]);
820 		if ((writers > 1) ||
821 		    ((writers == 1) && (JFS_IP(ip)->active_ag != agno))) {
822 			/*
823 			 * Someone else is writing in this allocation
824 			 * group.  To avoid fragmenting, try another ag
825 			 */
826 			release_metapage(mp);
827 			IREAD_UNLOCK(ipbmap);
828 			goto pref_ag;
829 		}
830 
831 		/* next, try to satisfy the allocation request with blocks
832 		 * near the hint.
833 		 */
834 		if ((rc =
835 		     dbAllocNear(bmp, dp, blkno, (int) nblocks, l2nb, results))
836 		    != -ENOSPC) {
837 			if (rc == 0)
838 				mark_metapage_dirty(mp);
839 
840 			release_metapage(mp);
841 			goto read_unlock;
842 		}
843 
844 		/* try to satisfy the allocation request with blocks within
845 		 * the same dmap as the hint.
846 		 */
847 		if ((rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results))
848 		    != -ENOSPC) {
849 			if (rc == 0)
850 				mark_metapage_dirty(mp);
851 
852 			release_metapage(mp);
853 			goto read_unlock;
854 		}
855 
856 		release_metapage(mp);
857 		IREAD_UNLOCK(ipbmap);
858 	}
859 
860 	/* try to satisfy the allocation request with blocks within
861 	 * the same allocation group as the hint.
862 	 */
863 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
864 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) != -ENOSPC)
865 		goto write_unlock;
866 
867 	IWRITE_UNLOCK(ipbmap);
868 
869 
870       pref_ag:
871 	/*
872 	 * Let dbNextAG recommend a preferred allocation group
873 	 */
874 	agno = dbNextAG(ipbmap);
875 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
876 
877 	/* Try to allocate within this allocation group.  if that fails, try to
878 	 * allocate anywhere in the map.
879 	 */
880 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) == -ENOSPC)
881 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
882 
883       write_unlock:
884 	IWRITE_UNLOCK(ipbmap);
885 
886 	return (rc);
887 
888       read_unlock:
889 	IREAD_UNLOCK(ipbmap);
890 
891 	return (rc);
892 }
893 
894 /*
895  * NAME:	dbReAlloc()
896  *
897  * FUNCTION:	attempt to extend a current allocation by a specified
898  *		number of blocks.
899  *
900  *		this routine attempts to satisfy the allocation request
901  *		by first trying to extend the existing allocation in
902  *		place by allocating the additional blocks as the blocks
903  *		immediately following the current allocation.  if these
904  *		blocks are not available, this routine will attempt to
905  *		allocate a new set of contiguous blocks large enough
906  *		to cover the existing allocation plus the additional
907  *		number of blocks required.
908  *
909  * PARAMETERS:
910  *	ip	    -  pointer to in-core inode requiring allocation.
911  *	blkno	    -  starting block of the current allocation.
912  *	nblocks	    -  number of contiguous blocks within the current
913  *		       allocation.
914  *	addnblocks  -  number of blocks to add to the allocation.
915  *	results	-      on successful return, set to the starting block number
916  *		       of the existing allocation if the existing allocation
917  *		       was extended in place or to a newly allocated contiguous
918  *		       range if the existing allocation could not be extended
919  *		       in place.
920  *
921  * RETURN VALUES:
922  *	0	- success
923  *	-ENOSPC	- insufficient disk resources
924  *	-EIO	- i/o error
925  */
926 int
927 dbReAlloc(struct inode *ip,
928 	  s64 blkno, s64 nblocks, s64 addnblocks, s64 * results)
929 {
930 	int rc;
931 
932 	/* try to extend the allocation in place.
933 	 */
934 	if ((rc = dbExtend(ip, blkno, nblocks, addnblocks)) == 0) {
935 		*results = blkno;
936 		return (0);
937 	} else {
938 		if (rc != -ENOSPC)
939 			return (rc);
940 	}
941 
942 	/* could not extend the allocation in place, so allocate a
943 	 * new set of blocks for the entire request (i.e. try to get
944 	 * a range of contiguous blocks large enough to cover the
945 	 * existing allocation plus the additional blocks.)
946 	 */
947 	return (dbAlloc
948 		(ip, blkno + nblocks - 1, addnblocks + nblocks, results));
949 }
950 
951 
952 /*
953  * NAME:	dbExtend()
954  *
955  * FUNCTION:	attempt to extend a current allocation by a specified
956  *		number of blocks.
957  *
958  *		this routine attempts to satisfy the allocation request
959  *		by first trying to extend the existing allocation in
960  *		place by allocating the additional blocks as the blocks
961  *		immediately following the current allocation.
962  *
963  * PARAMETERS:
964  *	ip	    -  pointer to in-core inode requiring allocation.
965  *	blkno	    -  starting block of the current allocation.
966  *	nblocks	    -  number of contiguous blocks within the current
967  *		       allocation.
968  *	addnblocks  -  number of blocks to add to the allocation.
969  *
970  * RETURN VALUES:
971  *	0	- success
972  *	-ENOSPC	- insufficient disk resources
973  *	-EIO	- i/o error
974  */
975 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks)
976 {
977 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
978 	s64 lblkno, lastblkno, extblkno;
979 	uint rel_block;
980 	struct metapage *mp;
981 	struct dmap *dp;
982 	int rc;
983 	struct inode *ipbmap = sbi->ipbmap;
984 	struct bmap *bmp;
985 
986 	/*
987 	 * We don't want a non-aligned extent to cross a page boundary
988 	 */
989 	if (((rel_block = blkno & (sbi->nbperpage - 1))) &&
990 	    (rel_block + nblocks + addnblocks > sbi->nbperpage))
991 		return -ENOSPC;
992 
993 	/* get the last block of the current allocation */
994 	lastblkno = blkno + nblocks - 1;
995 
996 	/* determine the block number of the block following
997 	 * the existing allocation.
998 	 */
999 	extblkno = lastblkno + 1;
1000 
1001 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
1002 
1003 	/* better be within the file system */
1004 	bmp = sbi->bmap;
1005 	if (lastblkno < 0 || lastblkno >= bmp->db_mapsize) {
1006 		IREAD_UNLOCK(ipbmap);
1007 		jfs_error(ip->i_sb, "the block is outside the filesystem\n");
1008 		return -EIO;
1009 	}
1010 
1011 	/* we'll attempt to extend the current allocation in place by
1012 	 * allocating the additional blocks as the blocks immediately
1013 	 * following the current allocation.  we only try to extend the
1014 	 * current allocation in place if the number of additional blocks
1015 	 * can fit into a dmap, the last block of the current allocation
1016 	 * is not the last block of the file system, and the start of the
1017 	 * inplace extension is not on an allocation group boundary.
1018 	 */
1019 	if (addnblocks > BPERDMAP || extblkno >= bmp->db_mapsize ||
1020 	    (extblkno & (bmp->db_agsize - 1)) == 0) {
1021 		IREAD_UNLOCK(ipbmap);
1022 		return -ENOSPC;
1023 	}
1024 
1025 	/* get the buffer for the dmap containing the first block
1026 	 * of the extension.
1027 	 */
1028 	lblkno = BLKTODMAP(extblkno, bmp->db_l2nbperpage);
1029 	mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
1030 	if (mp == NULL) {
1031 		IREAD_UNLOCK(ipbmap);
1032 		return -EIO;
1033 	}
1034 
1035 	dp = (struct dmap *) mp->data;
1036 
1037 	/* try to allocate the blocks immediately following the
1038 	 * current allocation.
1039 	 */
1040 	rc = dbAllocNext(bmp, dp, extblkno, (int) addnblocks);
1041 
1042 	IREAD_UNLOCK(ipbmap);
1043 
1044 	/* were we successful ? */
1045 	if (rc == 0)
1046 		write_metapage(mp);
1047 	else
1048 		/* we were not successful */
1049 		release_metapage(mp);
1050 
1051 	return (rc);
1052 }
1053 
1054 
1055 /*
1056  * NAME:	dbAllocNext()
1057  *
1058  * FUNCTION:	attempt to allocate the blocks of the specified block
1059  *		range within a dmap.
1060  *
1061  * PARAMETERS:
1062  *	bmp	-  pointer to bmap descriptor
1063  *	dp	-  pointer to dmap.
1064  *	blkno	-  starting block number of the range.
1065  *	nblocks	-  number of contiguous free blocks of the range.
1066  *
1067  * RETURN VALUES:
1068  *	0	- success
1069  *	-ENOSPC	- insufficient disk resources
1070  *	-EIO	- i/o error
1071  *
1072  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1073  */
1074 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
1075 		       int nblocks)
1076 {
1077 	int dbitno, word, rembits, nb, nwords, wbitno, nw;
1078 	int l2size;
1079 	s8 *leaf;
1080 	u32 mask;
1081 
1082 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1083 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n");
1084 		return -EIO;
1085 	}
1086 
1087 	/* pick up a pointer to the leaves of the dmap tree.
1088 	 */
1089 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1090 
1091 	/* determine the bit number and word within the dmap of the
1092 	 * starting block.
1093 	 */
1094 	dbitno = blkno & (BPERDMAP - 1);
1095 	word = dbitno >> L2DBWORD;
1096 
1097 	/* check if the specified block range is contained within
1098 	 * this dmap.
1099 	 */
1100 	if (dbitno + nblocks > BPERDMAP)
1101 		return -ENOSPC;
1102 
1103 	/* check if the starting leaf indicates that anything
1104 	 * is free.
1105 	 */
1106 	if (leaf[word] == NOFREE)
1107 		return -ENOSPC;
1108 
1109 	/* check the dmaps words corresponding to block range to see
1110 	 * if the block range is free.  not all bits of the first and
1111 	 * last words may be contained within the block range.  if this
1112 	 * is the case, we'll work against those words (i.e. partial first
1113 	 * and/or last) on an individual basis (a single pass) and examine
1114 	 * the actual bits to determine if they are free.  a single pass
1115 	 * will be used for all dmap words fully contained within the
1116 	 * specified range.  within this pass, the leaves of the dmap
1117 	 * tree will be examined to determine if the blocks are free. a
1118 	 * single leaf may describe the free space of multiple dmap
1119 	 * words, so we may visit only a subset of the actual leaves
1120 	 * corresponding to the dmap words of the block range.
1121 	 */
1122 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
1123 		/* determine the bit number within the word and
1124 		 * the number of bits within the word.
1125 		 */
1126 		wbitno = dbitno & (DBWORD - 1);
1127 		nb = min(rembits, DBWORD - wbitno);
1128 
1129 		/* check if only part of the word is to be examined.
1130 		 */
1131 		if (nb < DBWORD) {
1132 			/* check if the bits are free.
1133 			 */
1134 			mask = (ONES << (DBWORD - nb) >> wbitno);
1135 			if ((mask & ~le32_to_cpu(dp->wmap[word])) != mask)
1136 				return -ENOSPC;
1137 
1138 			word += 1;
1139 		} else {
1140 			/* one or more dmap words are fully contained
1141 			 * within the block range.  determine how many
1142 			 * words and how many bits.
1143 			 */
1144 			nwords = rembits >> L2DBWORD;
1145 			nb = nwords << L2DBWORD;
1146 
1147 			/* now examine the appropriate leaves to determine
1148 			 * if the blocks are free.
1149 			 */
1150 			while (nwords > 0) {
1151 				/* does the leaf describe any free space ?
1152 				 */
1153 				if (leaf[word] < BUDMIN)
1154 					return -ENOSPC;
1155 
1156 				/* determine the l2 number of bits provided
1157 				 * by this leaf.
1158 				 */
1159 				l2size =
1160 				    min_t(int, leaf[word], NLSTOL2BSZ(nwords));
1161 
1162 				/* determine how many words were handled.
1163 				 */
1164 				nw = BUDSIZE(l2size, BUDMIN);
1165 
1166 				nwords -= nw;
1167 				word += nw;
1168 			}
1169 		}
1170 	}
1171 
1172 	/* allocate the blocks.
1173 	 */
1174 	return (dbAllocDmap(bmp, dp, blkno, nblocks));
1175 }
1176 
1177 
1178 /*
1179  * NAME:	dbAllocNear()
1180  *
1181  * FUNCTION:	attempt to allocate a number of contiguous free blocks near
1182  *		a specified block (hint) within a dmap.
1183  *
1184  *		starting with the dmap leaf that covers the hint, we'll
1185  *		check the next four contiguous leaves for sufficient free
1186  *		space.  if sufficient free space is found, we'll allocate
1187  *		the desired free space.
1188  *
1189  * PARAMETERS:
1190  *	bmp	-  pointer to bmap descriptor
1191  *	dp	-  pointer to dmap.
1192  *	blkno	-  block number to allocate near.
1193  *	nblocks	-  actual number of contiguous free blocks desired.
1194  *	l2nb	-  log2 number of contiguous free blocks desired.
1195  *	results	-  on successful return, set to the starting block number
1196  *		   of the newly allocated range.
1197  *
1198  * RETURN VALUES:
1199  *	0	- success
1200  *	-ENOSPC	- insufficient disk resources
1201  *	-EIO	- i/o error
1202  *
1203  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1204  */
1205 static int
1206 dbAllocNear(struct bmap * bmp,
1207 	    struct dmap * dp, s64 blkno, int nblocks, int l2nb, s64 * results)
1208 {
1209 	int word, lword, rc;
1210 	s8 *leaf;
1211 
1212 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1213 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n");
1214 		return -EIO;
1215 	}
1216 
1217 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1218 
1219 	/* determine the word within the dmap that holds the hint
1220 	 * (i.e. blkno).  also, determine the last word in the dmap
1221 	 * that we'll include in our examination.
1222 	 */
1223 	word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
1224 	lword = min(word + 4, LPERDMAP);
1225 
1226 	/* examine the leaves for sufficient free space.
1227 	 */
1228 	for (; word < lword; word++) {
1229 		/* does the leaf describe sufficient free space ?
1230 		 */
1231 		if (leaf[word] < l2nb)
1232 			continue;
1233 
1234 		/* determine the block number within the file system
1235 		 * of the first block described by this dmap word.
1236 		 */
1237 		blkno = le64_to_cpu(dp->start) + (word << L2DBWORD);
1238 
1239 		/* if not all bits of the dmap word are free, get the
1240 		 * starting bit number within the dmap word of the required
1241 		 * string of free bits and adjust the block number with the
1242 		 * value.
1243 		 */
1244 		if (leaf[word] < BUDMIN)
1245 			blkno +=
1246 			    dbFindBits(le32_to_cpu(dp->wmap[word]), l2nb);
1247 
1248 		/* allocate the blocks.
1249 		 */
1250 		if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1251 			*results = blkno;
1252 
1253 		return (rc);
1254 	}
1255 
1256 	return -ENOSPC;
1257 }
1258 
1259 
1260 /*
1261  * NAME:	dbAllocAG()
1262  *
1263  * FUNCTION:	attempt to allocate the specified number of contiguous
1264  *		free blocks within the specified allocation group.
1265  *
1266  *		unless the allocation group size is equal to the number
1267  *		of blocks per dmap, the dmap control pages will be used to
1268  *		find the required free space, if available.  we start the
1269  *		search at the highest dmap control page level which
1270  *		distinctly describes the allocation group's free space
1271  *		(i.e. the highest level at which the allocation group's
1272  *		free space is not mixed in with that of any other group).
1273  *		in addition, we start the search within this level at a
1274  *		height of the dmapctl dmtree at which the nodes distinctly
1275  *		describe the allocation group's free space.  at this height,
1276  *		the allocation group's free space may be represented by 1
1277  *		or two sub-trees, depending on the allocation group size.
1278  *		we search the top nodes of these subtrees left to right for
1279  *		sufficient free space.  if sufficient free space is found,
1280  *		the subtree is searched to find the leftmost leaf that
1281  *		has free space.  once we have made it to the leaf, we
1282  *		move the search to the next lower level dmap control page
1283  *		corresponding to this leaf.  we continue down the dmap control
1284  *		pages until we find the dmap that contains or starts the
1285  *		sufficient free space and we allocate at this dmap.
1286  *
1287  *		if the allocation group size is equal to the dmap size,
1288  *		we'll start at the dmap corresponding to the allocation
1289  *		group and attempt the allocation at this level.
1290  *
1291  *		the dmap control page search is also not performed if the
1292  *		allocation group is completely free and we go to the first
1293  *		dmap of the allocation group to do the allocation.  this is
1294  *		done because the allocation group may be part (not the first
1295  *		part) of a larger binary buddy system, causing the dmap
1296  *		control pages to indicate no free space (NOFREE) within
1297  *		the allocation group.
1298  *
1299  * PARAMETERS:
1300  *	bmp	-  pointer to bmap descriptor
1301  *	agno	- allocation group number.
1302  *	nblocks	-  actual number of contiguous free blocks desired.
1303  *	l2nb	-  log2 number of contiguous free blocks desired.
1304  *	results	-  on successful return, set to the starting block number
1305  *		   of the newly allocated range.
1306  *
1307  * RETURN VALUES:
1308  *	0	- success
1309  *	-ENOSPC	- insufficient disk resources
1310  *	-EIO	- i/o error
1311  *
1312  * note: IWRITE_LOCK(ipmap) held on entry/exit;
1313  */
1314 static int
1315 dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results)
1316 {
1317 	struct metapage *mp;
1318 	struct dmapctl *dcp;
1319 	int rc, ti, i, k, m, n, agperlev;
1320 	s64 blkno, lblkno;
1321 	int budmin;
1322 
1323 	/* allocation request should not be for more than the
1324 	 * allocation group size.
1325 	 */
1326 	if (l2nb > bmp->db_agl2size) {
1327 		jfs_error(bmp->db_ipbmap->i_sb,
1328 			  "allocation request is larger than the allocation group size\n");
1329 		return -EIO;
1330 	}
1331 
1332 	/* determine the starting block number of the allocation
1333 	 * group.
1334 	 */
1335 	blkno = (s64) agno << bmp->db_agl2size;
1336 
1337 	/* check if the allocation group size is the minimum allocation
1338 	 * group size or if the allocation group is completely free. if
1339 	 * the allocation group size is the minimum size of BPERDMAP (i.e.
1340 	 * 1 dmap), there is no need to search the dmap control page (below)
1341 	 * that fully describes the allocation group since the allocation
1342 	 * group is already fully described by a dmap.  in this case, we
1343 	 * just call dbAllocCtl() to search the dmap tree and allocate the
1344 	 * required space if available.
1345 	 *
1346 	 * if the allocation group is completely free, dbAllocCtl() is
1347 	 * also called to allocate the required space.  this is done for
1348 	 * two reasons.  first, it makes no sense searching the dmap control
1349 	 * pages for free space when we know that free space exists.  second,
1350 	 * the dmap control pages may indicate that the allocation group
1351 	 * has no free space if the allocation group is part (not the first
1352 	 * part) of a larger binary buddy system.
1353 	 */
1354 	if (bmp->db_agsize == BPERDMAP
1355 	    || bmp->db_agfree[agno] == bmp->db_agsize) {
1356 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1357 		if ((rc == -ENOSPC) &&
1358 		    (bmp->db_agfree[agno] == bmp->db_agsize)) {
1359 			printk(KERN_ERR "blkno = %Lx, blocks = %Lx\n",
1360 			       (unsigned long long) blkno,
1361 			       (unsigned long long) nblocks);
1362 			jfs_error(bmp->db_ipbmap->i_sb,
1363 				  "dbAllocCtl failed in free AG\n");
1364 		}
1365 		return (rc);
1366 	}
1367 
1368 	/* the buffer for the dmap control page that fully describes the
1369 	 * allocation group.
1370 	 */
1371 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, bmp->db_aglevel);
1372 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1373 	if (mp == NULL)
1374 		return -EIO;
1375 	dcp = (struct dmapctl *) mp->data;
1376 	budmin = dcp->budmin;
1377 
1378 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1379 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n");
1380 		release_metapage(mp);
1381 		return -EIO;
1382 	}
1383 
1384 	/* search the subtree(s) of the dmap control page that describes
1385 	 * the allocation group, looking for sufficient free space.  to begin,
1386 	 * determine how many allocation groups are represented in a dmap
1387 	 * control page at the control page level (i.e. L0, L1, L2) that
1388 	 * fully describes an allocation group. next, determine the starting
1389 	 * tree index of this allocation group within the control page.
1390 	 */
1391 	agperlev =
1392 	    (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp->db_agwidth;
1393 	ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
1394 
1395 	/* dmap control page trees fan-out by 4 and a single allocation
1396 	 * group may be described by 1 or 2 subtrees within the ag level
1397 	 * dmap control page, depending upon the ag size. examine the ag's
1398 	 * subtrees for sufficient free space, starting with the leftmost
1399 	 * subtree.
1400 	 */
1401 	for (i = 0; i < bmp->db_agwidth; i++, ti++) {
1402 		/* is there sufficient free space ?
1403 		 */
1404 		if (l2nb > dcp->stree[ti])
1405 			continue;
1406 
1407 		/* sufficient free space found in a subtree. now search down
1408 		 * the subtree to find the leftmost leaf that describes this
1409 		 * free space.
1410 		 */
1411 		for (k = bmp->db_agheight; k > 0; k--) {
1412 			for (n = 0, m = (ti << 2) + 1; n < 4; n++) {
1413 				if (l2nb <= dcp->stree[m + n]) {
1414 					ti = m + n;
1415 					break;
1416 				}
1417 			}
1418 			if (n == 4) {
1419 				jfs_error(bmp->db_ipbmap->i_sb,
1420 					  "failed descending stree\n");
1421 				release_metapage(mp);
1422 				return -EIO;
1423 			}
1424 		}
1425 
1426 		/* determine the block number within the file system
1427 		 * that corresponds to this leaf.
1428 		 */
1429 		if (bmp->db_aglevel == 2)
1430 			blkno = 0;
1431 		else if (bmp->db_aglevel == 1)
1432 			blkno &= ~(MAXL1SIZE - 1);
1433 		else		/* bmp->db_aglevel == 0 */
1434 			blkno &= ~(MAXL0SIZE - 1);
1435 
1436 		blkno +=
1437 		    ((s64) (ti - le32_to_cpu(dcp->leafidx))) << budmin;
1438 
1439 		/* release the buffer in preparation for going down
1440 		 * the next level of dmap control pages.
1441 		 */
1442 		release_metapage(mp);
1443 
1444 		/* check if we need to continue to search down the lower
1445 		 * level dmap control pages.  we need to if the number of
1446 		 * blocks required is less than maximum number of blocks
1447 		 * described at the next lower level.
1448 		 */
1449 		if (l2nb < budmin) {
1450 
1451 			/* search the lower level dmap control pages to get
1452 			 * the starting block number of the dmap that
1453 			 * contains or starts off the free space.
1454 			 */
1455 			if ((rc =
1456 			     dbFindCtl(bmp, l2nb, bmp->db_aglevel - 1,
1457 				       &blkno))) {
1458 				if (rc == -ENOSPC) {
1459 					jfs_error(bmp->db_ipbmap->i_sb,
1460 						  "control page inconsistent\n");
1461 					return -EIO;
1462 				}
1463 				return (rc);
1464 			}
1465 		}
1466 
1467 		/* allocate the blocks.
1468 		 */
1469 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1470 		if (rc == -ENOSPC) {
1471 			jfs_error(bmp->db_ipbmap->i_sb,
1472 				  "unable to allocate blocks\n");
1473 			rc = -EIO;
1474 		}
1475 		return (rc);
1476 	}
1477 
1478 	/* no space in the allocation group.  release the buffer and
1479 	 * return -ENOSPC.
1480 	 */
1481 	release_metapage(mp);
1482 
1483 	return -ENOSPC;
1484 }
1485 
1486 
1487 /*
1488  * NAME:	dbAllocAny()
1489  *
1490  * FUNCTION:	attempt to allocate the specified number of contiguous
1491  *		free blocks anywhere in the file system.
1492  *
1493  *		dbAllocAny() attempts to find the sufficient free space by
1494  *		searching down the dmap control pages, starting with the
1495  *		highest level (i.e. L0, L1, L2) control page.  if free space
1496  *		large enough to satisfy the desired free space is found, the
1497  *		desired free space is allocated.
1498  *
1499  * PARAMETERS:
1500  *	bmp	-  pointer to bmap descriptor
1501  *	nblocks	 -  actual number of contiguous free blocks desired.
1502  *	l2nb	 -  log2 number of contiguous free blocks desired.
1503  *	results	-  on successful return, set to the starting block number
1504  *		   of the newly allocated range.
1505  *
1506  * RETURN VALUES:
1507  *	0	- success
1508  *	-ENOSPC	- insufficient disk resources
1509  *	-EIO	- i/o error
1510  *
1511  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1512  */
1513 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results)
1514 {
1515 	int rc;
1516 	s64 blkno = 0;
1517 
1518 	/* starting with the top level dmap control page, search
1519 	 * down the dmap control levels for sufficient free space.
1520 	 * if free space is found, dbFindCtl() returns the starting
1521 	 * block number of the dmap that contains or starts off the
1522 	 * range of free space.
1523 	 */
1524 	if ((rc = dbFindCtl(bmp, l2nb, bmp->db_maxlevel, &blkno)))
1525 		return (rc);
1526 
1527 	/* allocate the blocks.
1528 	 */
1529 	rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1530 	if (rc == -ENOSPC) {
1531 		jfs_error(bmp->db_ipbmap->i_sb, "unable to allocate blocks\n");
1532 		return -EIO;
1533 	}
1534 	return (rc);
1535 }
1536 
1537 
1538 /*
1539  * NAME:	dbDiscardAG()
1540  *
1541  * FUNCTION:	attempt to discard (TRIM) all free blocks of specific AG
1542  *
1543  *		algorithm:
1544  *		1) allocate blocks, as large as possible and save them
1545  *		   while holding IWRITE_LOCK on ipbmap
1546  *		2) trim all these saved block/length values
1547  *		3) mark the blocks free again
1548  *
1549  *		benefit:
1550  *		- we work only on one ag at some time, minimizing how long we
1551  *		  need to lock ipbmap
1552  *		- reading / writing the fs is possible most time, even on
1553  *		  trimming
1554  *
1555  *		downside:
1556  *		- we write two times to the dmapctl and dmap pages
1557  *		- but for me, this seems the best way, better ideas?
1558  *		/TR 2012
1559  *
1560  * PARAMETERS:
1561  *	ip	- pointer to in-core inode
1562  *	agno	- ag to trim
1563  *	minlen	- minimum value of contiguous blocks
1564  *
1565  * RETURN VALUES:
1566  *	s64	- actual number of blocks trimmed
1567  */
1568 s64 dbDiscardAG(struct inode *ip, int agno, s64 minlen)
1569 {
1570 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
1571 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
1572 	s64 nblocks, blkno;
1573 	u64 trimmed = 0;
1574 	int rc, l2nb;
1575 	struct super_block *sb = ipbmap->i_sb;
1576 
1577 	struct range2trim {
1578 		u64 blkno;
1579 		u64 nblocks;
1580 	} *totrim, *tt;
1581 
1582 	/* max blkno / nblocks pairs to trim */
1583 	int count = 0, range_cnt;
1584 	u64 max_ranges;
1585 
1586 	/* prevent others from writing new stuff here, while trimming */
1587 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
1588 
1589 	nblocks = bmp->db_agfree[agno];
1590 	max_ranges = nblocks;
1591 	do_div(max_ranges, minlen);
1592 	range_cnt = min_t(u64, max_ranges + 1, 32 * 1024);
1593 	totrim = kmalloc_array(range_cnt, sizeof(struct range2trim), GFP_NOFS);
1594 	if (totrim == NULL) {
1595 		jfs_error(bmp->db_ipbmap->i_sb, "no memory for trim array\n");
1596 		IWRITE_UNLOCK(ipbmap);
1597 		return 0;
1598 	}
1599 
1600 	tt = totrim;
1601 	while (nblocks >= minlen) {
1602 		l2nb = BLKSTOL2(nblocks);
1603 
1604 		/* 0 = okay, -EIO = fatal, -ENOSPC -> try smaller block */
1605 		rc = dbAllocAG(bmp, agno, nblocks, l2nb, &blkno);
1606 		if (rc == 0) {
1607 			tt->blkno = blkno;
1608 			tt->nblocks = nblocks;
1609 			tt++; count++;
1610 
1611 			/* the whole ag is free, trim now */
1612 			if (bmp->db_agfree[agno] == 0)
1613 				break;
1614 
1615 			/* give a hint for the next while */
1616 			nblocks = bmp->db_agfree[agno];
1617 			continue;
1618 		} else if (rc == -ENOSPC) {
1619 			/* search for next smaller log2 block */
1620 			l2nb = BLKSTOL2(nblocks) - 1;
1621 			nblocks = 1LL << l2nb;
1622 		} else {
1623 			/* Trim any already allocated blocks */
1624 			jfs_error(bmp->db_ipbmap->i_sb, "-EIO\n");
1625 			break;
1626 		}
1627 
1628 		/* check, if our trim array is full */
1629 		if (unlikely(count >= range_cnt - 1))
1630 			break;
1631 	}
1632 	IWRITE_UNLOCK(ipbmap);
1633 
1634 	tt->nblocks = 0; /* mark the current end */
1635 	for (tt = totrim; tt->nblocks != 0; tt++) {
1636 		/* when mounted with online discard, dbFree() will
1637 		 * call jfs_issue_discard() itself */
1638 		if (!(JFS_SBI(sb)->flag & JFS_DISCARD))
1639 			jfs_issue_discard(ip, tt->blkno, tt->nblocks);
1640 		dbFree(ip, tt->blkno, tt->nblocks);
1641 		trimmed += tt->nblocks;
1642 	}
1643 	kfree(totrim);
1644 
1645 	return trimmed;
1646 }
1647 
1648 /*
1649  * NAME:	dbFindCtl()
1650  *
1651  * FUNCTION:	starting at a specified dmap control page level and block
1652  *		number, search down the dmap control levels for a range of
1653  *		contiguous free blocks large enough to satisfy an allocation
1654  *		request for the specified number of free blocks.
1655  *
1656  *		if sufficient contiguous free blocks are found, this routine
1657  *		returns the starting block number within a dmap page that
1658  *		contains or starts a range of contiqious free blocks that
1659  *		is sufficient in size.
1660  *
1661  * PARAMETERS:
1662  *	bmp	-  pointer to bmap descriptor
1663  *	level	-  starting dmap control page level.
1664  *	l2nb	-  log2 number of contiguous free blocks desired.
1665  *	*blkno	-  on entry, starting block number for conducting the search.
1666  *		   on successful return, the first block within a dmap page
1667  *		   that contains or starts a range of contiguous free blocks.
1668  *
1669  * RETURN VALUES:
1670  *	0	- success
1671  *	-ENOSPC	- insufficient disk resources
1672  *	-EIO	- i/o error
1673  *
1674  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1675  */
1676 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno)
1677 {
1678 	int rc, leafidx, lev;
1679 	s64 b, lblkno;
1680 	struct dmapctl *dcp;
1681 	int budmin;
1682 	struct metapage *mp;
1683 
1684 	/* starting at the specified dmap control page level and block
1685 	 * number, search down the dmap control levels for the starting
1686 	 * block number of a dmap page that contains or starts off
1687 	 * sufficient free blocks.
1688 	 */
1689 	for (lev = level, b = *blkno; lev >= 0; lev--) {
1690 		/* get the buffer of the dmap control page for the block
1691 		 * number and level (i.e. L0, L1, L2).
1692 		 */
1693 		lblkno = BLKTOCTL(b, bmp->db_l2nbperpage, lev);
1694 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1695 		if (mp == NULL)
1696 			return -EIO;
1697 		dcp = (struct dmapctl *) mp->data;
1698 		budmin = dcp->budmin;
1699 
1700 		if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1701 			jfs_error(bmp->db_ipbmap->i_sb,
1702 				  "Corrupt dmapctl page\n");
1703 			release_metapage(mp);
1704 			return -EIO;
1705 		}
1706 
1707 		/* search the tree within the dmap control page for
1708 		 * sufficient free space.  if sufficient free space is found,
1709 		 * dbFindLeaf() returns the index of the leaf at which
1710 		 * free space was found.
1711 		 */
1712 		rc = dbFindLeaf((dmtree_t *) dcp, l2nb, &leafidx);
1713 
1714 		/* release the buffer.
1715 		 */
1716 		release_metapage(mp);
1717 
1718 		/* space found ?
1719 		 */
1720 		if (rc) {
1721 			if (lev != level) {
1722 				jfs_error(bmp->db_ipbmap->i_sb,
1723 					  "dmap inconsistent\n");
1724 				return -EIO;
1725 			}
1726 			return -ENOSPC;
1727 		}
1728 
1729 		/* adjust the block number to reflect the location within
1730 		 * the dmap control page (i.e. the leaf) at which free
1731 		 * space was found.
1732 		 */
1733 		b += (((s64) leafidx) << budmin);
1734 
1735 		/* we stop the search at this dmap control page level if
1736 		 * the number of blocks required is greater than or equal
1737 		 * to the maximum number of blocks described at the next
1738 		 * (lower) level.
1739 		 */
1740 		if (l2nb >= budmin)
1741 			break;
1742 	}
1743 
1744 	*blkno = b;
1745 	return (0);
1746 }
1747 
1748 
1749 /*
1750  * NAME:	dbAllocCtl()
1751  *
1752  * FUNCTION:	attempt to allocate a specified number of contiguous
1753  *		blocks starting within a specific dmap.
1754  *
1755  *		this routine is called by higher level routines that search
1756  *		the dmap control pages above the actual dmaps for contiguous
1757  *		free space.  the result of successful searches by these
1758  *		routines are the starting block numbers within dmaps, with
1759  *		the dmaps themselves containing the desired contiguous free
1760  *		space or starting a contiguous free space of desired size
1761  *		that is made up of the blocks of one or more dmaps. these
1762  *		calls should not fail due to insufficent resources.
1763  *
1764  *		this routine is called in some cases where it is not known
1765  *		whether it will fail due to insufficient resources.  more
1766  *		specifically, this occurs when allocating from an allocation
1767  *		group whose size is equal to the number of blocks per dmap.
1768  *		in this case, the dmap control pages are not examined prior
1769  *		to calling this routine (to save pathlength) and the call
1770  *		might fail.
1771  *
1772  *		for a request size that fits within a dmap, this routine relies
1773  *		upon the dmap's dmtree to find the requested contiguous free
1774  *		space.  for request sizes that are larger than a dmap, the
1775  *		requested free space will start at the first block of the
1776  *		first dmap (i.e. blkno).
1777  *
1778  * PARAMETERS:
1779  *	bmp	-  pointer to bmap descriptor
1780  *	nblocks	 -  actual number of contiguous free blocks to allocate.
1781  *	l2nb	 -  log2 number of contiguous free blocks to allocate.
1782  *	blkno	 -  starting block number of the dmap to start the allocation
1783  *		    from.
1784  *	results	-  on successful return, set to the starting block number
1785  *		   of the newly allocated range.
1786  *
1787  * RETURN VALUES:
1788  *	0	- success
1789  *	-ENOSPC	- insufficient disk resources
1790  *	-EIO	- i/o error
1791  *
1792  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1793  */
1794 static int
1795 dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, s64 * results)
1796 {
1797 	int rc, nb;
1798 	s64 b, lblkno, n;
1799 	struct metapage *mp;
1800 	struct dmap *dp;
1801 
1802 	/* check if the allocation request is confined to a single dmap.
1803 	 */
1804 	if (l2nb <= L2BPERDMAP) {
1805 		/* get the buffer for the dmap.
1806 		 */
1807 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
1808 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1809 		if (mp == NULL)
1810 			return -EIO;
1811 		dp = (struct dmap *) mp->data;
1812 
1813 		/* try to allocate the blocks.
1814 		 */
1815 		rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results);
1816 		if (rc == 0)
1817 			mark_metapage_dirty(mp);
1818 
1819 		release_metapage(mp);
1820 
1821 		return (rc);
1822 	}
1823 
1824 	/* allocation request involving multiple dmaps. it must start on
1825 	 * a dmap boundary.
1826 	 */
1827 	assert((blkno & (BPERDMAP - 1)) == 0);
1828 
1829 	/* allocate the blocks dmap by dmap.
1830 	 */
1831 	for (n = nblocks, b = blkno; n > 0; n -= nb, b += nb) {
1832 		/* get the buffer for the dmap.
1833 		 */
1834 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1835 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1836 		if (mp == NULL) {
1837 			rc = -EIO;
1838 			goto backout;
1839 		}
1840 		dp = (struct dmap *) mp->data;
1841 
1842 		/* the dmap better be all free.
1843 		 */
1844 		if (dp->tree.stree[ROOT] != L2BPERDMAP) {
1845 			release_metapage(mp);
1846 			jfs_error(bmp->db_ipbmap->i_sb,
1847 				  "the dmap is not all free\n");
1848 			rc = -EIO;
1849 			goto backout;
1850 		}
1851 
1852 		/* determine how many blocks to allocate from this dmap.
1853 		 */
1854 		nb = min_t(s64, n, BPERDMAP);
1855 
1856 		/* allocate the blocks from the dmap.
1857 		 */
1858 		if ((rc = dbAllocDmap(bmp, dp, b, nb))) {
1859 			release_metapage(mp);
1860 			goto backout;
1861 		}
1862 
1863 		/* write the buffer.
1864 		 */
1865 		write_metapage(mp);
1866 	}
1867 
1868 	/* set the results (starting block number) and return.
1869 	 */
1870 	*results = blkno;
1871 	return (0);
1872 
1873 	/* something failed in handling an allocation request involving
1874 	 * multiple dmaps.  we'll try to clean up by backing out any
1875 	 * allocation that has already happened for this request.  if
1876 	 * we fail in backing out the allocation, we'll mark the file
1877 	 * system to indicate that blocks have been leaked.
1878 	 */
1879       backout:
1880 
1881 	/* try to backout the allocations dmap by dmap.
1882 	 */
1883 	for (n = nblocks - n, b = blkno; n > 0;
1884 	     n -= BPERDMAP, b += BPERDMAP) {
1885 		/* get the buffer for this dmap.
1886 		 */
1887 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1888 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1889 		if (mp == NULL) {
1890 			/* could not back out.  mark the file system
1891 			 * to indicate that we have leaked blocks.
1892 			 */
1893 			jfs_error(bmp->db_ipbmap->i_sb,
1894 				  "I/O Error: Block Leakage\n");
1895 			continue;
1896 		}
1897 		dp = (struct dmap *) mp->data;
1898 
1899 		/* free the blocks is this dmap.
1900 		 */
1901 		if (dbFreeDmap(bmp, dp, b, BPERDMAP)) {
1902 			/* could not back out.  mark the file system
1903 			 * to indicate that we have leaked blocks.
1904 			 */
1905 			release_metapage(mp);
1906 			jfs_error(bmp->db_ipbmap->i_sb, "Block Leakage\n");
1907 			continue;
1908 		}
1909 
1910 		/* write the buffer.
1911 		 */
1912 		write_metapage(mp);
1913 	}
1914 
1915 	return (rc);
1916 }
1917 
1918 
1919 /*
1920  * NAME:	dbAllocDmapLev()
1921  *
1922  * FUNCTION:	attempt to allocate a specified number of contiguous blocks
1923  *		from a specified dmap.
1924  *
1925  *		this routine checks if the contiguous blocks are available.
1926  *		if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1927  *		returned.
1928  *
1929  * PARAMETERS:
1930  *	mp	-  pointer to bmap descriptor
1931  *	dp	-  pointer to dmap to attempt to allocate blocks from.
1932  *	l2nb	-  log2 number of contiguous block desired.
1933  *	nblocks	-  actual number of contiguous block desired.
1934  *	results	-  on successful return, set to the starting block number
1935  *		   of the newly allocated range.
1936  *
1937  * RETURN VALUES:
1938  *	0	- success
1939  *	-ENOSPC	- insufficient disk resources
1940  *	-EIO	- i/o error
1941  *
1942  * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1943  *	IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1944  */
1945 static int
1946 dbAllocDmapLev(struct bmap * bmp,
1947 	       struct dmap * dp, int nblocks, int l2nb, s64 * results)
1948 {
1949 	s64 blkno;
1950 	int leafidx, rc;
1951 
1952 	/* can't be more than a dmaps worth of blocks */
1953 	assert(l2nb <= L2BPERDMAP);
1954 
1955 	/* search the tree within the dmap page for sufficient
1956 	 * free space.  if sufficient free space is found, dbFindLeaf()
1957 	 * returns the index of the leaf at which free space was found.
1958 	 */
1959 	if (dbFindLeaf((dmtree_t *) & dp->tree, l2nb, &leafidx))
1960 		return -ENOSPC;
1961 
1962 	if (leafidx < 0)
1963 		return -EIO;
1964 
1965 	/* determine the block number within the file system corresponding
1966 	 * to the leaf at which free space was found.
1967 	 */
1968 	blkno = le64_to_cpu(dp->start) + (leafidx << L2DBWORD);
1969 
1970 	/* if not all bits of the dmap word are free, get the starting
1971 	 * bit number within the dmap word of the required string of free
1972 	 * bits and adjust the block number with this value.
1973 	 */
1974 	if (dp->tree.stree[leafidx + LEAFIND] < BUDMIN)
1975 		blkno += dbFindBits(le32_to_cpu(dp->wmap[leafidx]), l2nb);
1976 
1977 	/* allocate the blocks */
1978 	if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1979 		*results = blkno;
1980 
1981 	return (rc);
1982 }
1983 
1984 
1985 /*
1986  * NAME:	dbAllocDmap()
1987  *
1988  * FUNCTION:	adjust the disk allocation map to reflect the allocation
1989  *		of a specified block range within a dmap.
1990  *
1991  *		this routine allocates the specified blocks from the dmap
1992  *		through a call to dbAllocBits(). if the allocation of the
1993  *		block range causes the maximum string of free blocks within
1994  *		the dmap to change (i.e. the value of the root of the dmap's
1995  *		dmtree), this routine will cause this change to be reflected
1996  *		up through the appropriate levels of the dmap control pages
1997  *		by a call to dbAdjCtl() for the L0 dmap control page that
1998  *		covers this dmap.
1999  *
2000  * PARAMETERS:
2001  *	bmp	-  pointer to bmap descriptor
2002  *	dp	-  pointer to dmap to allocate the block range from.
2003  *	blkno	-  starting block number of the block to be allocated.
2004  *	nblocks	-  number of blocks to be allocated.
2005  *
2006  * RETURN VALUES:
2007  *	0	- success
2008  *	-EIO	- i/o error
2009  *
2010  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2011  */
2012 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2013 		       int nblocks)
2014 {
2015 	s8 oldroot;
2016 	int rc;
2017 
2018 	/* save the current value of the root (i.e. maximum free string)
2019 	 * of the dmap tree.
2020 	 */
2021 	oldroot = dp->tree.stree[ROOT];
2022 
2023 	/* allocate the specified (blocks) bits */
2024 	dbAllocBits(bmp, dp, blkno, nblocks);
2025 
2026 	/* if the root has not changed, done. */
2027 	if (dp->tree.stree[ROOT] == oldroot)
2028 		return (0);
2029 
2030 	/* root changed. bubble the change up to the dmap control pages.
2031 	 * if the adjustment of the upper level control pages fails,
2032 	 * backout the bit allocation (thus making everything consistent).
2033 	 */
2034 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 1, 0)))
2035 		dbFreeBits(bmp, dp, blkno, nblocks);
2036 
2037 	return (rc);
2038 }
2039 
2040 
2041 /*
2042  * NAME:	dbFreeDmap()
2043  *
2044  * FUNCTION:	adjust the disk allocation map to reflect the allocation
2045  *		of a specified block range within a dmap.
2046  *
2047  *		this routine frees the specified blocks from the dmap through
2048  *		a call to dbFreeBits(). if the deallocation of the block range
2049  *		causes the maximum string of free blocks within the dmap to
2050  *		change (i.e. the value of the root of the dmap's dmtree), this
2051  *		routine will cause this change to be reflected up through the
2052  *		appropriate levels of the dmap control pages by a call to
2053  *		dbAdjCtl() for the L0 dmap control page that covers this dmap.
2054  *
2055  * PARAMETERS:
2056  *	bmp	-  pointer to bmap descriptor
2057  *	dp	-  pointer to dmap to free the block range from.
2058  *	blkno	-  starting block number of the block to be freed.
2059  *	nblocks	-  number of blocks to be freed.
2060  *
2061  * RETURN VALUES:
2062  *	0	- success
2063  *	-EIO	- i/o error
2064  *
2065  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2066  */
2067 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2068 		      int nblocks)
2069 {
2070 	s8 oldroot;
2071 	int rc = 0, word;
2072 
2073 	/* save the current value of the root (i.e. maximum free string)
2074 	 * of the dmap tree.
2075 	 */
2076 	oldroot = dp->tree.stree[ROOT];
2077 
2078 	/* free the specified (blocks) bits */
2079 	rc = dbFreeBits(bmp, dp, blkno, nblocks);
2080 
2081 	/* if error or the root has not changed, done. */
2082 	if (rc || (dp->tree.stree[ROOT] == oldroot))
2083 		return (rc);
2084 
2085 	/* root changed. bubble the change up to the dmap control pages.
2086 	 * if the adjustment of the upper level control pages fails,
2087 	 * backout the deallocation.
2088 	 */
2089 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 0, 0))) {
2090 		word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
2091 
2092 		/* as part of backing out the deallocation, we will have
2093 		 * to back split the dmap tree if the deallocation caused
2094 		 * the freed blocks to become part of a larger binary buddy
2095 		 * system.
2096 		 */
2097 		if (dp->tree.stree[word] == NOFREE)
2098 			dbBackSplit((dmtree_t *) & dp->tree, word);
2099 
2100 		dbAllocBits(bmp, dp, blkno, nblocks);
2101 	}
2102 
2103 	return (rc);
2104 }
2105 
2106 
2107 /*
2108  * NAME:	dbAllocBits()
2109  *
2110  * FUNCTION:	allocate a specified block range from a dmap.
2111  *
2112  *		this routine updates the dmap to reflect the working
2113  *		state allocation of the specified block range. it directly
2114  *		updates the bits of the working map and causes the adjustment
2115  *		of the binary buddy system described by the dmap's dmtree
2116  *		leaves to reflect the bits allocated.  it also causes the
2117  *		dmap's dmtree, as a whole, to reflect the allocated range.
2118  *
2119  * PARAMETERS:
2120  *	bmp	-  pointer to bmap descriptor
2121  *	dp	-  pointer to dmap to allocate bits from.
2122  *	blkno	-  starting block number of the bits to be allocated.
2123  *	nblocks	-  number of bits to be allocated.
2124  *
2125  * RETURN VALUES: none
2126  *
2127  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2128  */
2129 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2130 			int nblocks)
2131 {
2132 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2133 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2134 	int size;
2135 	s8 *leaf;
2136 
2137 	/* pick up a pointer to the leaves of the dmap tree */
2138 	leaf = dp->tree.stree + LEAFIND;
2139 
2140 	/* determine the bit number and word within the dmap of the
2141 	 * starting block.
2142 	 */
2143 	dbitno = blkno & (BPERDMAP - 1);
2144 	word = dbitno >> L2DBWORD;
2145 
2146 	/* block range better be within the dmap */
2147 	assert(dbitno + nblocks <= BPERDMAP);
2148 
2149 	/* allocate the bits of the dmap's words corresponding to the block
2150 	 * range. not all bits of the first and last words may be contained
2151 	 * within the block range.  if this is the case, we'll work against
2152 	 * those words (i.e. partial first and/or last) on an individual basis
2153 	 * (a single pass), allocating the bits of interest by hand and
2154 	 * updating the leaf corresponding to the dmap word. a single pass
2155 	 * will be used for all dmap words fully contained within the
2156 	 * specified range.  within this pass, the bits of all fully contained
2157 	 * dmap words will be marked as free in a single shot and the leaves
2158 	 * will be updated. a single leaf may describe the free space of
2159 	 * multiple dmap words, so we may update only a subset of the actual
2160 	 * leaves corresponding to the dmap words of the block range.
2161 	 */
2162 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2163 		/* determine the bit number within the word and
2164 		 * the number of bits within the word.
2165 		 */
2166 		wbitno = dbitno & (DBWORD - 1);
2167 		nb = min(rembits, DBWORD - wbitno);
2168 
2169 		/* check if only part of a word is to be allocated.
2170 		 */
2171 		if (nb < DBWORD) {
2172 			/* allocate (set to 1) the appropriate bits within
2173 			 * this dmap word.
2174 			 */
2175 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
2176 						      >> wbitno);
2177 
2178 			/* update the leaf for this dmap word. in addition
2179 			 * to setting the leaf value to the binary buddy max
2180 			 * of the updated dmap word, dbSplit() will split
2181 			 * the binary system of the leaves if need be.
2182 			 */
2183 			dbSplit(tp, word, BUDMIN,
2184 				dbMaxBud((u8 *) & dp->wmap[word]));
2185 
2186 			word += 1;
2187 		} else {
2188 			/* one or more dmap words are fully contained
2189 			 * within the block range.  determine how many
2190 			 * words and allocate (set to 1) the bits of these
2191 			 * words.
2192 			 */
2193 			nwords = rembits >> L2DBWORD;
2194 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
2195 
2196 			/* determine how many bits.
2197 			 */
2198 			nb = nwords << L2DBWORD;
2199 
2200 			/* now update the appropriate leaves to reflect
2201 			 * the allocated words.
2202 			 */
2203 			for (; nwords > 0; nwords -= nw) {
2204 				if (leaf[word] < BUDMIN) {
2205 					jfs_error(bmp->db_ipbmap->i_sb,
2206 						  "leaf page corrupt\n");
2207 					break;
2208 				}
2209 
2210 				/* determine what the leaf value should be
2211 				 * updated to as the minimum of the l2 number
2212 				 * of bits being allocated and the l2 number
2213 				 * of bits currently described by this leaf.
2214 				 */
2215 				size = min_t(int, leaf[word],
2216 					     NLSTOL2BSZ(nwords));
2217 
2218 				/* update the leaf to reflect the allocation.
2219 				 * in addition to setting the leaf value to
2220 				 * NOFREE, dbSplit() will split the binary
2221 				 * system of the leaves to reflect the current
2222 				 * allocation (size).
2223 				 */
2224 				dbSplit(tp, word, size, NOFREE);
2225 
2226 				/* get the number of dmap words handled */
2227 				nw = BUDSIZE(size, BUDMIN);
2228 				word += nw;
2229 			}
2230 		}
2231 	}
2232 
2233 	/* update the free count for this dmap */
2234 	le32_add_cpu(&dp->nfree, -nblocks);
2235 
2236 	BMAP_LOCK(bmp);
2237 
2238 	/* if this allocation group is completely free,
2239 	 * update the maximum allocation group number if this allocation
2240 	 * group is the new max.
2241 	 */
2242 	agno = blkno >> bmp->db_agl2size;
2243 	if (agno > bmp->db_maxag)
2244 		bmp->db_maxag = agno;
2245 
2246 	/* update the free count for the allocation group and map */
2247 	bmp->db_agfree[agno] -= nblocks;
2248 	bmp->db_nfree -= nblocks;
2249 
2250 	BMAP_UNLOCK(bmp);
2251 }
2252 
2253 
2254 /*
2255  * NAME:	dbFreeBits()
2256  *
2257  * FUNCTION:	free a specified block range from a dmap.
2258  *
2259  *		this routine updates the dmap to reflect the working
2260  *		state allocation of the specified block range. it directly
2261  *		updates the bits of the working map and causes the adjustment
2262  *		of the binary buddy system described by the dmap's dmtree
2263  *		leaves to reflect the bits freed.  it also causes the dmap's
2264  *		dmtree, as a whole, to reflect the deallocated range.
2265  *
2266  * PARAMETERS:
2267  *	bmp	-  pointer to bmap descriptor
2268  *	dp	-  pointer to dmap to free bits from.
2269  *	blkno	-  starting block number of the bits to be freed.
2270  *	nblocks	-  number of bits to be freed.
2271  *
2272  * RETURN VALUES: 0 for success
2273  *
2274  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2275  */
2276 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2277 		       int nblocks)
2278 {
2279 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2280 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2281 	int rc = 0;
2282 	int size;
2283 
2284 	/* determine the bit number and word within the dmap of the
2285 	 * starting block.
2286 	 */
2287 	dbitno = blkno & (BPERDMAP - 1);
2288 	word = dbitno >> L2DBWORD;
2289 
2290 	/* block range better be within the dmap.
2291 	 */
2292 	assert(dbitno + nblocks <= BPERDMAP);
2293 
2294 	/* free the bits of the dmaps words corresponding to the block range.
2295 	 * not all bits of the first and last words may be contained within
2296 	 * the block range.  if this is the case, we'll work against those
2297 	 * words (i.e. partial first and/or last) on an individual basis
2298 	 * (a single pass), freeing the bits of interest by hand and updating
2299 	 * the leaf corresponding to the dmap word. a single pass will be used
2300 	 * for all dmap words fully contained within the specified range.
2301 	 * within this pass, the bits of all fully contained dmap words will
2302 	 * be marked as free in a single shot and the leaves will be updated. a
2303 	 * single leaf may describe the free space of multiple dmap words,
2304 	 * so we may update only a subset of the actual leaves corresponding
2305 	 * to the dmap words of the block range.
2306 	 *
2307 	 * dbJoin() is used to update leaf values and will join the binary
2308 	 * buddy system of the leaves if the new leaf values indicate this
2309 	 * should be done.
2310 	 */
2311 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2312 		/* determine the bit number within the word and
2313 		 * the number of bits within the word.
2314 		 */
2315 		wbitno = dbitno & (DBWORD - 1);
2316 		nb = min(rembits, DBWORD - wbitno);
2317 
2318 		/* check if only part of a word is to be freed.
2319 		 */
2320 		if (nb < DBWORD) {
2321 			/* free (zero) the appropriate bits within this
2322 			 * dmap word.
2323 			 */
2324 			dp->wmap[word] &=
2325 			    cpu_to_le32(~(ONES << (DBWORD - nb)
2326 					  >> wbitno));
2327 
2328 			/* update the leaf for this dmap word.
2329 			 */
2330 			rc = dbJoin(tp, word,
2331 				    dbMaxBud((u8 *) & dp->wmap[word]));
2332 			if (rc)
2333 				return rc;
2334 
2335 			word += 1;
2336 		} else {
2337 			/* one or more dmap words are fully contained
2338 			 * within the block range.  determine how many
2339 			 * words and free (zero) the bits of these words.
2340 			 */
2341 			nwords = rembits >> L2DBWORD;
2342 			memset(&dp->wmap[word], 0, nwords * 4);
2343 
2344 			/* determine how many bits.
2345 			 */
2346 			nb = nwords << L2DBWORD;
2347 
2348 			/* now update the appropriate leaves to reflect
2349 			 * the freed words.
2350 			 */
2351 			for (; nwords > 0; nwords -= nw) {
2352 				/* determine what the leaf value should be
2353 				 * updated to as the minimum of the l2 number
2354 				 * of bits being freed and the l2 (max) number
2355 				 * of bits that can be described by this leaf.
2356 				 */
2357 				size =
2358 				    min(LITOL2BSZ
2359 					(word, L2LPERDMAP, BUDMIN),
2360 					NLSTOL2BSZ(nwords));
2361 
2362 				/* update the leaf.
2363 				 */
2364 				rc = dbJoin(tp, word, size);
2365 				if (rc)
2366 					return rc;
2367 
2368 				/* get the number of dmap words handled.
2369 				 */
2370 				nw = BUDSIZE(size, BUDMIN);
2371 				word += nw;
2372 			}
2373 		}
2374 	}
2375 
2376 	/* update the free count for this dmap.
2377 	 */
2378 	le32_add_cpu(&dp->nfree, nblocks);
2379 
2380 	BMAP_LOCK(bmp);
2381 
2382 	/* update the free count for the allocation group and
2383 	 * map.
2384 	 */
2385 	agno = blkno >> bmp->db_agl2size;
2386 	bmp->db_nfree += nblocks;
2387 	bmp->db_agfree[agno] += nblocks;
2388 
2389 	/* check if this allocation group is not completely free and
2390 	 * if it is currently the maximum (rightmost) allocation group.
2391 	 * if so, establish the new maximum allocation group number by
2392 	 * searching left for the first allocation group with allocation.
2393 	 */
2394 	if ((bmp->db_agfree[agno] == bmp->db_agsize && agno == bmp->db_maxag) ||
2395 	    (agno == bmp->db_numag - 1 &&
2396 	     bmp->db_agfree[agno] == (bmp-> db_mapsize & (BPERDMAP - 1)))) {
2397 		while (bmp->db_maxag > 0) {
2398 			bmp->db_maxag -= 1;
2399 			if (bmp->db_agfree[bmp->db_maxag] !=
2400 			    bmp->db_agsize)
2401 				break;
2402 		}
2403 
2404 		/* re-establish the allocation group preference if the
2405 		 * current preference is right of the maximum allocation
2406 		 * group.
2407 		 */
2408 		if (bmp->db_agpref > bmp->db_maxag)
2409 			bmp->db_agpref = bmp->db_maxag;
2410 	}
2411 
2412 	BMAP_UNLOCK(bmp);
2413 
2414 	return 0;
2415 }
2416 
2417 
2418 /*
2419  * NAME:	dbAdjCtl()
2420  *
2421  * FUNCTION:	adjust a dmap control page at a specified level to reflect
2422  *		the change in a lower level dmap or dmap control page's
2423  *		maximum string of free blocks (i.e. a change in the root
2424  *		of the lower level object's dmtree) due to the allocation
2425  *		or deallocation of a range of blocks with a single dmap.
2426  *
2427  *		on entry, this routine is provided with the new value of
2428  *		the lower level dmap or dmap control page root and the
2429  *		starting block number of the block range whose allocation
2430  *		or deallocation resulted in the root change.  this range
2431  *		is respresented by a single leaf of the current dmapctl
2432  *		and the leaf will be updated with this value, possibly
2433  *		causing a binary buddy system within the leaves to be
2434  *		split or joined.  the update may also cause the dmapctl's
2435  *		dmtree to be updated.
2436  *
2437  *		if the adjustment of the dmap control page, itself, causes its
2438  *		root to change, this change will be bubbled up to the next dmap
2439  *		control level by a recursive call to this routine, specifying
2440  *		the new root value and the next dmap control page level to
2441  *		be adjusted.
2442  * PARAMETERS:
2443  *	bmp	-  pointer to bmap descriptor
2444  *	blkno	-  the first block of a block range within a dmap.  it is
2445  *		   the allocation or deallocation of this block range that
2446  *		   requires the dmap control page to be adjusted.
2447  *	newval	-  the new value of the lower level dmap or dmap control
2448  *		   page root.
2449  *	alloc	-  'true' if adjustment is due to an allocation.
2450  *	level	-  current level of dmap control page (i.e. L0, L1, L2) to
2451  *		   be adjusted.
2452  *
2453  * RETURN VALUES:
2454  *	0	- success
2455  *	-EIO	- i/o error
2456  *
2457  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2458  */
2459 static int
2460 dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, int level)
2461 {
2462 	struct metapage *mp;
2463 	s8 oldroot;
2464 	int oldval;
2465 	s64 lblkno;
2466 	struct dmapctl *dcp;
2467 	int rc, leafno, ti;
2468 
2469 	/* get the buffer for the dmap control page for the specified
2470 	 * block number and control page level.
2471 	 */
2472 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, level);
2473 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
2474 	if (mp == NULL)
2475 		return -EIO;
2476 	dcp = (struct dmapctl *) mp->data;
2477 
2478 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
2479 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n");
2480 		release_metapage(mp);
2481 		return -EIO;
2482 	}
2483 
2484 	/* determine the leaf number corresponding to the block and
2485 	 * the index within the dmap control tree.
2486 	 */
2487 	leafno = BLKTOCTLLEAF(blkno, dcp->budmin);
2488 	ti = leafno + le32_to_cpu(dcp->leafidx);
2489 
2490 	/* save the current leaf value and the current root level (i.e.
2491 	 * maximum l2 free string described by this dmapctl).
2492 	 */
2493 	oldval = dcp->stree[ti];
2494 	oldroot = dcp->stree[ROOT];
2495 
2496 	/* check if this is a control page update for an allocation.
2497 	 * if so, update the leaf to reflect the new leaf value using
2498 	 * dbSplit(); otherwise (deallocation), use dbJoin() to update
2499 	 * the leaf with the new value.  in addition to updating the
2500 	 * leaf, dbSplit() will also split the binary buddy system of
2501 	 * the leaves, if required, and bubble new values within the
2502 	 * dmapctl tree, if required.  similarly, dbJoin() will join
2503 	 * the binary buddy system of leaves and bubble new values up
2504 	 * the dmapctl tree as required by the new leaf value.
2505 	 */
2506 	if (alloc) {
2507 		/* check if we are in the middle of a binary buddy
2508 		 * system.  this happens when we are performing the
2509 		 * first allocation out of an allocation group that
2510 		 * is part (not the first part) of a larger binary
2511 		 * buddy system.  if we are in the middle, back split
2512 		 * the system prior to calling dbSplit() which assumes
2513 		 * that it is at the front of a binary buddy system.
2514 		 */
2515 		if (oldval == NOFREE) {
2516 			rc = dbBackSplit((dmtree_t *) dcp, leafno);
2517 			if (rc) {
2518 				release_metapage(mp);
2519 				return rc;
2520 			}
2521 			oldval = dcp->stree[ti];
2522 		}
2523 		dbSplit((dmtree_t *) dcp, leafno, dcp->budmin, newval);
2524 	} else {
2525 		rc = dbJoin((dmtree_t *) dcp, leafno, newval);
2526 		if (rc) {
2527 			release_metapage(mp);
2528 			return rc;
2529 		}
2530 	}
2531 
2532 	/* check if the root of the current dmap control page changed due
2533 	 * to the update and if the current dmap control page is not at
2534 	 * the current top level (i.e. L0, L1, L2) of the map.  if so (i.e.
2535 	 * root changed and this is not the top level), call this routine
2536 	 * again (recursion) for the next higher level of the mapping to
2537 	 * reflect the change in root for the current dmap control page.
2538 	 */
2539 	if (dcp->stree[ROOT] != oldroot) {
2540 		/* are we below the top level of the map.  if so,
2541 		 * bubble the root up to the next higher level.
2542 		 */
2543 		if (level < bmp->db_maxlevel) {
2544 			/* bubble up the new root of this dmap control page to
2545 			 * the next level.
2546 			 */
2547 			if ((rc =
2548 			     dbAdjCtl(bmp, blkno, dcp->stree[ROOT], alloc,
2549 				      level + 1))) {
2550 				/* something went wrong in bubbling up the new
2551 				 * root value, so backout the changes to the
2552 				 * current dmap control page.
2553 				 */
2554 				if (alloc) {
2555 					dbJoin((dmtree_t *) dcp, leafno,
2556 					       oldval);
2557 				} else {
2558 					/* the dbJoin() above might have
2559 					 * caused a larger binary buddy system
2560 					 * to form and we may now be in the
2561 					 * middle of it.  if this is the case,
2562 					 * back split the buddies.
2563 					 */
2564 					if (dcp->stree[ti] == NOFREE)
2565 						dbBackSplit((dmtree_t *)
2566 							    dcp, leafno);
2567 					dbSplit((dmtree_t *) dcp, leafno,
2568 						dcp->budmin, oldval);
2569 				}
2570 
2571 				/* release the buffer and return the error.
2572 				 */
2573 				release_metapage(mp);
2574 				return (rc);
2575 			}
2576 		} else {
2577 			/* we're at the top level of the map. update
2578 			 * the bmap control page to reflect the size
2579 			 * of the maximum free buddy system.
2580 			 */
2581 			assert(level == bmp->db_maxlevel);
2582 			if (bmp->db_maxfreebud != oldroot) {
2583 				jfs_error(bmp->db_ipbmap->i_sb,
2584 					  "the maximum free buddy is not the old root\n");
2585 			}
2586 			bmp->db_maxfreebud = dcp->stree[ROOT];
2587 		}
2588 	}
2589 
2590 	/* write the buffer.
2591 	 */
2592 	write_metapage(mp);
2593 
2594 	return (0);
2595 }
2596 
2597 
2598 /*
2599  * NAME:	dbSplit()
2600  *
2601  * FUNCTION:	update the leaf of a dmtree with a new value, splitting
2602  *		the leaf from the binary buddy system of the dmtree's
2603  *		leaves, as required.
2604  *
2605  * PARAMETERS:
2606  *	tp	- pointer to the tree containing the leaf.
2607  *	leafno	- the number of the leaf to be updated.
2608  *	splitsz	- the size the binary buddy system starting at the leaf
2609  *		  must be split to, specified as the log2 number of blocks.
2610  *	newval	- the new value for the leaf.
2611  *
2612  * RETURN VALUES: none
2613  *
2614  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2615  */
2616 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval)
2617 {
2618 	int budsz;
2619 	int cursz;
2620 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2621 
2622 	/* check if the leaf needs to be split.
2623 	 */
2624 	if (leaf[leafno] > tp->dmt_budmin) {
2625 		/* the split occurs by cutting the buddy system in half
2626 		 * at the specified leaf until we reach the specified
2627 		 * size.  pick up the starting split size (current size
2628 		 * - 1 in l2) and the corresponding buddy size.
2629 		 */
2630 		cursz = leaf[leafno] - 1;
2631 		budsz = BUDSIZE(cursz, tp->dmt_budmin);
2632 
2633 		/* split until we reach the specified size.
2634 		 */
2635 		while (cursz >= splitsz) {
2636 			/* update the buddy's leaf with its new value.
2637 			 */
2638 			dbAdjTree(tp, leafno ^ budsz, cursz);
2639 
2640 			/* on to the next size and buddy.
2641 			 */
2642 			cursz -= 1;
2643 			budsz >>= 1;
2644 		}
2645 	}
2646 
2647 	/* adjust the dmap tree to reflect the specified leaf's new
2648 	 * value.
2649 	 */
2650 	dbAdjTree(tp, leafno, newval);
2651 }
2652 
2653 
2654 /*
2655  * NAME:	dbBackSplit()
2656  *
2657  * FUNCTION:	back split the binary buddy system of dmtree leaves
2658  *		that hold a specified leaf until the specified leaf
2659  *		starts its own binary buddy system.
2660  *
2661  *		the allocators typically perform allocations at the start
2662  *		of binary buddy systems and dbSplit() is used to accomplish
2663  *		any required splits.  in some cases, however, allocation
2664  *		may occur in the middle of a binary system and requires a
2665  *		back split, with the split proceeding out from the middle of
2666  *		the system (less efficient) rather than the start of the
2667  *		system (more efficient).  the cases in which a back split
2668  *		is required are rare and are limited to the first allocation
2669  *		within an allocation group which is a part (not first part)
2670  *		of a larger binary buddy system and a few exception cases
2671  *		in which a previous join operation must be backed out.
2672  *
2673  * PARAMETERS:
2674  *	tp	- pointer to the tree containing the leaf.
2675  *	leafno	- the number of the leaf to be updated.
2676  *
2677  * RETURN VALUES: none
2678  *
2679  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2680  */
2681 static int dbBackSplit(dmtree_t * tp, int leafno)
2682 {
2683 	int budsz, bud, w, bsz, size;
2684 	int cursz;
2685 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2686 
2687 	/* leaf should be part (not first part) of a binary
2688 	 * buddy system.
2689 	 */
2690 	assert(leaf[leafno] == NOFREE);
2691 
2692 	/* the back split is accomplished by iteratively finding the leaf
2693 	 * that starts the buddy system that contains the specified leaf and
2694 	 * splitting that system in two.  this iteration continues until
2695 	 * the specified leaf becomes the start of a buddy system.
2696 	 *
2697 	 * determine maximum possible l2 size for the specified leaf.
2698 	 */
2699 	size =
2700 	    LITOL2BSZ(leafno, le32_to_cpu(tp->dmt_l2nleafs),
2701 		      tp->dmt_budmin);
2702 
2703 	/* determine the number of leaves covered by this size.  this
2704 	 * is the buddy size that we will start with as we search for
2705 	 * the buddy system that contains the specified leaf.
2706 	 */
2707 	budsz = BUDSIZE(size, tp->dmt_budmin);
2708 
2709 	/* back split.
2710 	 */
2711 	while (leaf[leafno] == NOFREE) {
2712 		/* find the leftmost buddy leaf.
2713 		 */
2714 		for (w = leafno, bsz = budsz;; bsz <<= 1,
2715 		     w = (w < bud) ? w : bud) {
2716 			if (bsz >= le32_to_cpu(tp->dmt_nleafs)) {
2717 				jfs_err("JFS: block map error in dbBackSplit");
2718 				return -EIO;
2719 			}
2720 
2721 			/* determine the buddy.
2722 			 */
2723 			bud = w ^ bsz;
2724 
2725 			/* check if this buddy is the start of the system.
2726 			 */
2727 			if (leaf[bud] != NOFREE) {
2728 				/* split the leaf at the start of the
2729 				 * system in two.
2730 				 */
2731 				cursz = leaf[bud] - 1;
2732 				dbSplit(tp, bud, cursz, cursz);
2733 				break;
2734 			}
2735 		}
2736 	}
2737 
2738 	if (leaf[leafno] != size) {
2739 		jfs_err("JFS: wrong leaf value in dbBackSplit");
2740 		return -EIO;
2741 	}
2742 	return 0;
2743 }
2744 
2745 
2746 /*
2747  * NAME:	dbJoin()
2748  *
2749  * FUNCTION:	update the leaf of a dmtree with a new value, joining
2750  *		the leaf with other leaves of the dmtree into a multi-leaf
2751  *		binary buddy system, as required.
2752  *
2753  * PARAMETERS:
2754  *	tp	- pointer to the tree containing the leaf.
2755  *	leafno	- the number of the leaf to be updated.
2756  *	newval	- the new value for the leaf.
2757  *
2758  * RETURN VALUES: none
2759  */
2760 static int dbJoin(dmtree_t * tp, int leafno, int newval)
2761 {
2762 	int budsz, buddy;
2763 	s8 *leaf;
2764 
2765 	/* can the new leaf value require a join with other leaves ?
2766 	 */
2767 	if (newval >= tp->dmt_budmin) {
2768 		/* pickup a pointer to the leaves of the tree.
2769 		 */
2770 		leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2771 
2772 		/* try to join the specified leaf into a large binary
2773 		 * buddy system.  the join proceeds by attempting to join
2774 		 * the specified leafno with its buddy (leaf) at new value.
2775 		 * if the join occurs, we attempt to join the left leaf
2776 		 * of the joined buddies with its buddy at new value + 1.
2777 		 * we continue to join until we find a buddy that cannot be
2778 		 * joined (does not have a value equal to the size of the
2779 		 * last join) or until all leaves have been joined into a
2780 		 * single system.
2781 		 *
2782 		 * get the buddy size (number of words covered) of
2783 		 * the new value.
2784 		 */
2785 		budsz = BUDSIZE(newval, tp->dmt_budmin);
2786 
2787 		/* try to join.
2788 		 */
2789 		while (budsz < le32_to_cpu(tp->dmt_nleafs)) {
2790 			/* get the buddy leaf.
2791 			 */
2792 			buddy = leafno ^ budsz;
2793 
2794 			/* if the leaf's new value is greater than its
2795 			 * buddy's value, we join no more.
2796 			 */
2797 			if (newval > leaf[buddy])
2798 				break;
2799 
2800 			/* It shouldn't be less */
2801 			if (newval < leaf[buddy])
2802 				return -EIO;
2803 
2804 			/* check which (leafno or buddy) is the left buddy.
2805 			 * the left buddy gets to claim the blocks resulting
2806 			 * from the join while the right gets to claim none.
2807 			 * the left buddy is also eligible to participate in
2808 			 * a join at the next higher level while the right
2809 			 * is not.
2810 			 *
2811 			 */
2812 			if (leafno < buddy) {
2813 				/* leafno is the left buddy.
2814 				 */
2815 				dbAdjTree(tp, buddy, NOFREE);
2816 			} else {
2817 				/* buddy is the left buddy and becomes
2818 				 * leafno.
2819 				 */
2820 				dbAdjTree(tp, leafno, NOFREE);
2821 				leafno = buddy;
2822 			}
2823 
2824 			/* on to try the next join.
2825 			 */
2826 			newval += 1;
2827 			budsz <<= 1;
2828 		}
2829 	}
2830 
2831 	/* update the leaf value.
2832 	 */
2833 	dbAdjTree(tp, leafno, newval);
2834 
2835 	return 0;
2836 }
2837 
2838 
2839 /*
2840  * NAME:	dbAdjTree()
2841  *
2842  * FUNCTION:	update a leaf of a dmtree with a new value, adjusting
2843  *		the dmtree, as required, to reflect the new leaf value.
2844  *		the combination of any buddies must already be done before
2845  *		this is called.
2846  *
2847  * PARAMETERS:
2848  *	tp	- pointer to the tree to be adjusted.
2849  *	leafno	- the number of the leaf to be updated.
2850  *	newval	- the new value for the leaf.
2851  *
2852  * RETURN VALUES: none
2853  */
2854 static void dbAdjTree(dmtree_t * tp, int leafno, int newval)
2855 {
2856 	int lp, pp, k;
2857 	int max;
2858 
2859 	/* pick up the index of the leaf for this leafno.
2860 	 */
2861 	lp = leafno + le32_to_cpu(tp->dmt_leafidx);
2862 
2863 	/* is the current value the same as the old value ?  if so,
2864 	 * there is nothing to do.
2865 	 */
2866 	if (tp->dmt_stree[lp] == newval)
2867 		return;
2868 
2869 	/* set the new value.
2870 	 */
2871 	tp->dmt_stree[lp] = newval;
2872 
2873 	/* bubble the new value up the tree as required.
2874 	 */
2875 	for (k = 0; k < le32_to_cpu(tp->dmt_height); k++) {
2876 		/* get the index of the first leaf of the 4 leaf
2877 		 * group containing the specified leaf (leafno).
2878 		 */
2879 		lp = ((lp - 1) & ~0x03) + 1;
2880 
2881 		/* get the index of the parent of this 4 leaf group.
2882 		 */
2883 		pp = (lp - 1) >> 2;
2884 
2885 		/* determine the maximum of the 4 leaves.
2886 		 */
2887 		max = TREEMAX(&tp->dmt_stree[lp]);
2888 
2889 		/* if the maximum of the 4 is the same as the
2890 		 * parent's value, we're done.
2891 		 */
2892 		if (tp->dmt_stree[pp] == max)
2893 			break;
2894 
2895 		/* parent gets new value.
2896 		 */
2897 		tp->dmt_stree[pp] = max;
2898 
2899 		/* parent becomes leaf for next go-round.
2900 		 */
2901 		lp = pp;
2902 	}
2903 }
2904 
2905 
2906 /*
2907  * NAME:	dbFindLeaf()
2908  *
2909  * FUNCTION:	search a dmtree_t for sufficient free blocks, returning
2910  *		the index of a leaf describing the free blocks if
2911  *		sufficient free blocks are found.
2912  *
2913  *		the search starts at the top of the dmtree_t tree and
2914  *		proceeds down the tree to the leftmost leaf with sufficient
2915  *		free space.
2916  *
2917  * PARAMETERS:
2918  *	tp	- pointer to the tree to be searched.
2919  *	l2nb	- log2 number of free blocks to search for.
2920  *	leafidx	- return pointer to be set to the index of the leaf
2921  *		  describing at least l2nb free blocks if sufficient
2922  *		  free blocks are found.
2923  *
2924  * RETURN VALUES:
2925  *	0	- success
2926  *	-ENOSPC	- insufficient free blocks.
2927  */
2928 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx)
2929 {
2930 	int ti, n = 0, k, x = 0;
2931 
2932 	/* first check the root of the tree to see if there is
2933 	 * sufficient free space.
2934 	 */
2935 	if (l2nb > tp->dmt_stree[ROOT])
2936 		return -ENOSPC;
2937 
2938 	/* sufficient free space available. now search down the tree
2939 	 * starting at the next level for the leftmost leaf that
2940 	 * describes sufficient free space.
2941 	 */
2942 	for (k = le32_to_cpu(tp->dmt_height), ti = 1;
2943 	     k > 0; k--, ti = ((ti + n) << 2) + 1) {
2944 		/* search the four nodes at this level, starting from
2945 		 * the left.
2946 		 */
2947 		for (x = ti, n = 0; n < 4; n++) {
2948 			/* sufficient free space found.  move to the next
2949 			 * level (or quit if this is the last level).
2950 			 */
2951 			if (l2nb <= tp->dmt_stree[x + n])
2952 				break;
2953 		}
2954 
2955 		/* better have found something since the higher
2956 		 * levels of the tree said it was here.
2957 		 */
2958 		assert(n < 4);
2959 	}
2960 
2961 	/* set the return to the leftmost leaf describing sufficient
2962 	 * free space.
2963 	 */
2964 	*leafidx = x + n - le32_to_cpu(tp->dmt_leafidx);
2965 
2966 	return (0);
2967 }
2968 
2969 
2970 /*
2971  * NAME:	dbFindBits()
2972  *
2973  * FUNCTION:	find a specified number of binary buddy free bits within a
2974  *		dmap bitmap word value.
2975  *
2976  *		this routine searches the bitmap value for (1 << l2nb) free
2977  *		bits at (1 << l2nb) alignments within the value.
2978  *
2979  * PARAMETERS:
2980  *	word	-  dmap bitmap word value.
2981  *	l2nb	-  number of free bits specified as a log2 number.
2982  *
2983  * RETURN VALUES:
2984  *	starting bit number of free bits.
2985  */
2986 static int dbFindBits(u32 word, int l2nb)
2987 {
2988 	int bitno, nb;
2989 	u32 mask;
2990 
2991 	/* get the number of bits.
2992 	 */
2993 	nb = 1 << l2nb;
2994 	assert(nb <= DBWORD);
2995 
2996 	/* complement the word so we can use a mask (i.e. 0s represent
2997 	 * free bits) and compute the mask.
2998 	 */
2999 	word = ~word;
3000 	mask = ONES << (DBWORD - nb);
3001 
3002 	/* scan the word for nb free bits at nb alignments.
3003 	 */
3004 	for (bitno = 0; mask != 0; bitno += nb, mask >>= nb) {
3005 		if ((mask & word) == mask)
3006 			break;
3007 	}
3008 
3009 	ASSERT(bitno < 32);
3010 
3011 	/* return the bit number.
3012 	 */
3013 	return (bitno);
3014 }
3015 
3016 
3017 /*
3018  * NAME:	dbMaxBud(u8 *cp)
3019  *
3020  * FUNCTION:	determine the largest binary buddy string of free
3021  *		bits within 32-bits of the map.
3022  *
3023  * PARAMETERS:
3024  *	cp	-  pointer to the 32-bit value.
3025  *
3026  * RETURN VALUES:
3027  *	largest binary buddy of free bits within a dmap word.
3028  */
3029 static int dbMaxBud(u8 * cp)
3030 {
3031 	signed char tmp1, tmp2;
3032 
3033 	/* check if the wmap word is all free. if so, the
3034 	 * free buddy size is BUDMIN.
3035 	 */
3036 	if (*((uint *) cp) == 0)
3037 		return (BUDMIN);
3038 
3039 	/* check if the wmap word is half free. if so, the
3040 	 * free buddy size is BUDMIN-1.
3041 	 */
3042 	if (*((u16 *) cp) == 0 || *((u16 *) cp + 1) == 0)
3043 		return (BUDMIN - 1);
3044 
3045 	/* not all free or half free. determine the free buddy
3046 	 * size thru table lookup using quarters of the wmap word.
3047 	 */
3048 	tmp1 = max(budtab[cp[2]], budtab[cp[3]]);
3049 	tmp2 = max(budtab[cp[0]], budtab[cp[1]]);
3050 	return (max(tmp1, tmp2));
3051 }
3052 
3053 
3054 /*
3055  * NAME:	cnttz(uint word)
3056  *
3057  * FUNCTION:	determine the number of trailing zeros within a 32-bit
3058  *		value.
3059  *
3060  * PARAMETERS:
3061  *	value	-  32-bit value to be examined.
3062  *
3063  * RETURN VALUES:
3064  *	count of trailing zeros
3065  */
3066 static int cnttz(u32 word)
3067 {
3068 	int n;
3069 
3070 	for (n = 0; n < 32; n++, word >>= 1) {
3071 		if (word & 0x01)
3072 			break;
3073 	}
3074 
3075 	return (n);
3076 }
3077 
3078 
3079 /*
3080  * NAME:	cntlz(u32 value)
3081  *
3082  * FUNCTION:	determine the number of leading zeros within a 32-bit
3083  *		value.
3084  *
3085  * PARAMETERS:
3086  *	value	-  32-bit value to be examined.
3087  *
3088  * RETURN VALUES:
3089  *	count of leading zeros
3090  */
3091 static int cntlz(u32 value)
3092 {
3093 	int n;
3094 
3095 	for (n = 0; n < 32; n++, value <<= 1) {
3096 		if (value & HIGHORDER)
3097 			break;
3098 	}
3099 	return (n);
3100 }
3101 
3102 
3103 /*
3104  * NAME:	blkstol2(s64 nb)
3105  *
3106  * FUNCTION:	convert a block count to its log2 value. if the block
3107  *		count is not a l2 multiple, it is rounded up to the next
3108  *		larger l2 multiple.
3109  *
3110  * PARAMETERS:
3111  *	nb	-  number of blocks
3112  *
3113  * RETURN VALUES:
3114  *	log2 number of blocks
3115  */
3116 static int blkstol2(s64 nb)
3117 {
3118 	int l2nb;
3119 	s64 mask;		/* meant to be signed */
3120 
3121 	mask = (s64) 1 << (64 - 1);
3122 
3123 	/* count the leading bits.
3124 	 */
3125 	for (l2nb = 0; l2nb < 64; l2nb++, mask >>= 1) {
3126 		/* leading bit found.
3127 		 */
3128 		if (nb & mask) {
3129 			/* determine the l2 value.
3130 			 */
3131 			l2nb = (64 - 1) - l2nb;
3132 
3133 			/* check if we need to round up.
3134 			 */
3135 			if (~mask & nb)
3136 				l2nb++;
3137 
3138 			return (l2nb);
3139 		}
3140 	}
3141 	assert(0);
3142 	return 0;		/* fix compiler warning */
3143 }
3144 
3145 
3146 /*
3147  * NAME:	dbAllocBottomUp()
3148  *
3149  * FUNCTION:	alloc the specified block range from the working block
3150  *		allocation map.
3151  *
3152  *		the blocks will be alloc from the working map one dmap
3153  *		at a time.
3154  *
3155  * PARAMETERS:
3156  *	ip	-  pointer to in-core inode;
3157  *	blkno	-  starting block number to be freed.
3158  *	nblocks	-  number of blocks to be freed.
3159  *
3160  * RETURN VALUES:
3161  *	0	- success
3162  *	-EIO	- i/o error
3163  */
3164 int dbAllocBottomUp(struct inode *ip, s64 blkno, s64 nblocks)
3165 {
3166 	struct metapage *mp;
3167 	struct dmap *dp;
3168 	int nb, rc;
3169 	s64 lblkno, rem;
3170 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
3171 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
3172 
3173 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
3174 
3175 	/* block to be allocated better be within the mapsize. */
3176 	ASSERT(nblocks <= bmp->db_mapsize - blkno);
3177 
3178 	/*
3179 	 * allocate the blocks a dmap at a time.
3180 	 */
3181 	mp = NULL;
3182 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
3183 		/* release previous dmap if any */
3184 		if (mp) {
3185 			write_metapage(mp);
3186 		}
3187 
3188 		/* get the buffer for the current dmap. */
3189 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
3190 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
3191 		if (mp == NULL) {
3192 			IREAD_UNLOCK(ipbmap);
3193 			return -EIO;
3194 		}
3195 		dp = (struct dmap *) mp->data;
3196 
3197 		/* determine the number of blocks to be allocated from
3198 		 * this dmap.
3199 		 */
3200 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
3201 
3202 		/* allocate the blocks. */
3203 		if ((rc = dbAllocDmapBU(bmp, dp, blkno, nb))) {
3204 			release_metapage(mp);
3205 			IREAD_UNLOCK(ipbmap);
3206 			return (rc);
3207 		}
3208 	}
3209 
3210 	/* write the last buffer. */
3211 	write_metapage(mp);
3212 
3213 	IREAD_UNLOCK(ipbmap);
3214 
3215 	return (0);
3216 }
3217 
3218 
3219 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
3220 			 int nblocks)
3221 {
3222 	int rc;
3223 	int dbitno, word, rembits, nb, nwords, wbitno, agno;
3224 	s8 oldroot;
3225 	struct dmaptree *tp = (struct dmaptree *) & dp->tree;
3226 
3227 	/* save the current value of the root (i.e. maximum free string)
3228 	 * of the dmap tree.
3229 	 */
3230 	oldroot = tp->stree[ROOT];
3231 
3232 	/* determine the bit number and word within the dmap of the
3233 	 * starting block.
3234 	 */
3235 	dbitno = blkno & (BPERDMAP - 1);
3236 	word = dbitno >> L2DBWORD;
3237 
3238 	/* block range better be within the dmap */
3239 	assert(dbitno + nblocks <= BPERDMAP);
3240 
3241 	/* allocate the bits of the dmap's words corresponding to the block
3242 	 * range. not all bits of the first and last words may be contained
3243 	 * within the block range.  if this is the case, we'll work against
3244 	 * those words (i.e. partial first and/or last) on an individual basis
3245 	 * (a single pass), allocating the bits of interest by hand and
3246 	 * updating the leaf corresponding to the dmap word. a single pass
3247 	 * will be used for all dmap words fully contained within the
3248 	 * specified range.  within this pass, the bits of all fully contained
3249 	 * dmap words will be marked as free in a single shot and the leaves
3250 	 * will be updated. a single leaf may describe the free space of
3251 	 * multiple dmap words, so we may update only a subset of the actual
3252 	 * leaves corresponding to the dmap words of the block range.
3253 	 */
3254 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
3255 		/* determine the bit number within the word and
3256 		 * the number of bits within the word.
3257 		 */
3258 		wbitno = dbitno & (DBWORD - 1);
3259 		nb = min(rembits, DBWORD - wbitno);
3260 
3261 		/* check if only part of a word is to be allocated.
3262 		 */
3263 		if (nb < DBWORD) {
3264 			/* allocate (set to 1) the appropriate bits within
3265 			 * this dmap word.
3266 			 */
3267 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
3268 						      >> wbitno);
3269 
3270 			word++;
3271 		} else {
3272 			/* one or more dmap words are fully contained
3273 			 * within the block range.  determine how many
3274 			 * words and allocate (set to 1) the bits of these
3275 			 * words.
3276 			 */
3277 			nwords = rembits >> L2DBWORD;
3278 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
3279 
3280 			/* determine how many bits */
3281 			nb = nwords << L2DBWORD;
3282 			word += nwords;
3283 		}
3284 	}
3285 
3286 	/* update the free count for this dmap */
3287 	le32_add_cpu(&dp->nfree, -nblocks);
3288 
3289 	/* reconstruct summary tree */
3290 	dbInitDmapTree(dp);
3291 
3292 	BMAP_LOCK(bmp);
3293 
3294 	/* if this allocation group is completely free,
3295 	 * update the highest active allocation group number
3296 	 * if this allocation group is the new max.
3297 	 */
3298 	agno = blkno >> bmp->db_agl2size;
3299 	if (agno > bmp->db_maxag)
3300 		bmp->db_maxag = agno;
3301 
3302 	/* update the free count for the allocation group and map */
3303 	bmp->db_agfree[agno] -= nblocks;
3304 	bmp->db_nfree -= nblocks;
3305 
3306 	BMAP_UNLOCK(bmp);
3307 
3308 	/* if the root has not changed, done. */
3309 	if (tp->stree[ROOT] == oldroot)
3310 		return (0);
3311 
3312 	/* root changed. bubble the change up to the dmap control pages.
3313 	 * if the adjustment of the upper level control pages fails,
3314 	 * backout the bit allocation (thus making everything consistent).
3315 	 */
3316 	if ((rc = dbAdjCtl(bmp, blkno, tp->stree[ROOT], 1, 0)))
3317 		dbFreeBits(bmp, dp, blkno, nblocks);
3318 
3319 	return (rc);
3320 }
3321 
3322 
3323 /*
3324  * NAME:	dbExtendFS()
3325  *
3326  * FUNCTION:	extend bmap from blkno for nblocks;
3327  *		dbExtendFS() updates bmap ready for dbAllocBottomUp();
3328  *
3329  * L2
3330  *  |
3331  *   L1---------------------------------L1
3332  *    |					 |
3333  *     L0---------L0---------L0		  L0---------L0---------L0
3334  *      |	   |	      |		   |	      |		 |
3335  *	 d0,...,dn  d0,...,dn  d0,...,dn    d0,...,dn  d0,...,dn  d0,.,dm;
3336  * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3337  *
3338  * <---old---><----------------------------extend----------------------->
3339  */
3340 int dbExtendFS(struct inode *ipbmap, s64 blkno,	s64 nblocks)
3341 {
3342 	struct jfs_sb_info *sbi = JFS_SBI(ipbmap->i_sb);
3343 	int nbperpage = sbi->nbperpage;
3344 	int i, i0 = true, j, j0 = true, k, n;
3345 	s64 newsize;
3346 	s64 p;
3347 	struct metapage *mp, *l2mp, *l1mp = NULL, *l0mp = NULL;
3348 	struct dmapctl *l2dcp, *l1dcp, *l0dcp;
3349 	struct dmap *dp;
3350 	s8 *l0leaf, *l1leaf, *l2leaf;
3351 	struct bmap *bmp = sbi->bmap;
3352 	int agno, l2agsize, oldl2agsize;
3353 	s64 ag_rem;
3354 
3355 	newsize = blkno + nblocks;
3356 
3357 	jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld",
3358 		 (long long) blkno, (long long) nblocks, (long long) newsize);
3359 
3360 	/*
3361 	 *	initialize bmap control page.
3362 	 *
3363 	 * all the data in bmap control page should exclude
3364 	 * the mkfs hidden dmap page.
3365 	 */
3366 
3367 	/* update mapsize */
3368 	bmp->db_mapsize = newsize;
3369 	bmp->db_maxlevel = BMAPSZTOLEV(bmp->db_mapsize);
3370 
3371 	/* compute new AG size */
3372 	l2agsize = dbGetL2AGSize(newsize);
3373 	oldl2agsize = bmp->db_agl2size;
3374 
3375 	bmp->db_agl2size = l2agsize;
3376 	bmp->db_agsize = 1 << l2agsize;
3377 
3378 	/* compute new number of AG */
3379 	agno = bmp->db_numag;
3380 	bmp->db_numag = newsize >> l2agsize;
3381 	bmp->db_numag += ((u32) newsize % (u32) bmp->db_agsize) ? 1 : 0;
3382 
3383 	/*
3384 	 *	reconfigure db_agfree[]
3385 	 * from old AG configuration to new AG configuration;
3386 	 *
3387 	 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3388 	 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3389 	 * note: new AG size = old AG size * (2**x).
3390 	 */
3391 	if (l2agsize == oldl2agsize)
3392 		goto extend;
3393 	k = 1 << (l2agsize - oldl2agsize);
3394 	ag_rem = bmp->db_agfree[0];	/* save agfree[0] */
3395 	for (i = 0, n = 0; i < agno; n++) {
3396 		bmp->db_agfree[n] = 0;	/* init collection point */
3397 
3398 		/* coalesce contiguous k AGs; */
3399 		for (j = 0; j < k && i < agno; j++, i++) {
3400 			/* merge AGi to AGn */
3401 			bmp->db_agfree[n] += bmp->db_agfree[i];
3402 		}
3403 	}
3404 	bmp->db_agfree[0] += ag_rem;	/* restore agfree[0] */
3405 
3406 	for (; n < MAXAG; n++)
3407 		bmp->db_agfree[n] = 0;
3408 
3409 	/*
3410 	 * update highest active ag number
3411 	 */
3412 
3413 	bmp->db_maxag = bmp->db_maxag / k;
3414 
3415 	/*
3416 	 *	extend bmap
3417 	 *
3418 	 * update bit maps and corresponding level control pages;
3419 	 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3420 	 */
3421       extend:
3422 	/* get L2 page */
3423 	p = BMAPBLKNO + nbperpage;	/* L2 page */
3424 	l2mp = read_metapage(ipbmap, p, PSIZE, 0);
3425 	if (!l2mp) {
3426 		jfs_error(ipbmap->i_sb, "L2 page could not be read\n");
3427 		return -EIO;
3428 	}
3429 	l2dcp = (struct dmapctl *) l2mp->data;
3430 
3431 	/* compute start L1 */
3432 	k = blkno >> L2MAXL1SIZE;
3433 	l2leaf = l2dcp->stree + CTLLEAFIND + k;
3434 	p = BLKTOL1(blkno, sbi->l2nbperpage);	/* L1 page */
3435 
3436 	/*
3437 	 * extend each L1 in L2
3438 	 */
3439 	for (; k < LPERCTL; k++, p += nbperpage) {
3440 		/* get L1 page */
3441 		if (j0) {
3442 			/* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3443 			l1mp = read_metapage(ipbmap, p, PSIZE, 0);
3444 			if (l1mp == NULL)
3445 				goto errout;
3446 			l1dcp = (struct dmapctl *) l1mp->data;
3447 
3448 			/* compute start L0 */
3449 			j = (blkno & (MAXL1SIZE - 1)) >> L2MAXL0SIZE;
3450 			l1leaf = l1dcp->stree + CTLLEAFIND + j;
3451 			p = BLKTOL0(blkno, sbi->l2nbperpage);
3452 			j0 = false;
3453 		} else {
3454 			/* assign/init L1 page */
3455 			l1mp = get_metapage(ipbmap, p, PSIZE, 0);
3456 			if (l1mp == NULL)
3457 				goto errout;
3458 
3459 			l1dcp = (struct dmapctl *) l1mp->data;
3460 
3461 			/* compute start L0 */
3462 			j = 0;
3463 			l1leaf = l1dcp->stree + CTLLEAFIND;
3464 			p += nbperpage;	/* 1st L0 of L1.k */
3465 		}
3466 
3467 		/*
3468 		 * extend each L0 in L1
3469 		 */
3470 		for (; j < LPERCTL; j++) {
3471 			/* get L0 page */
3472 			if (i0) {
3473 				/* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3474 
3475 				l0mp = read_metapage(ipbmap, p, PSIZE, 0);
3476 				if (l0mp == NULL)
3477 					goto errout;
3478 				l0dcp = (struct dmapctl *) l0mp->data;
3479 
3480 				/* compute start dmap */
3481 				i = (blkno & (MAXL0SIZE - 1)) >>
3482 				    L2BPERDMAP;
3483 				l0leaf = l0dcp->stree + CTLLEAFIND + i;
3484 				p = BLKTODMAP(blkno,
3485 					      sbi->l2nbperpage);
3486 				i0 = false;
3487 			} else {
3488 				/* assign/init L0 page */
3489 				l0mp = get_metapage(ipbmap, p, PSIZE, 0);
3490 				if (l0mp == NULL)
3491 					goto errout;
3492 
3493 				l0dcp = (struct dmapctl *) l0mp->data;
3494 
3495 				/* compute start dmap */
3496 				i = 0;
3497 				l0leaf = l0dcp->stree + CTLLEAFIND;
3498 				p += nbperpage;	/* 1st dmap of L0.j */
3499 			}
3500 
3501 			/*
3502 			 * extend each dmap in L0
3503 			 */
3504 			for (; i < LPERCTL; i++) {
3505 				/*
3506 				 * reconstruct the dmap page, and
3507 				 * initialize corresponding parent L0 leaf
3508 				 */
3509 				if ((n = blkno & (BPERDMAP - 1))) {
3510 					/* read in dmap page: */
3511 					mp = read_metapage(ipbmap, p,
3512 							   PSIZE, 0);
3513 					if (mp == NULL)
3514 						goto errout;
3515 					n = min(nblocks, (s64)BPERDMAP - n);
3516 				} else {
3517 					/* assign/init dmap page */
3518 					mp = read_metapage(ipbmap, p,
3519 							   PSIZE, 0);
3520 					if (mp == NULL)
3521 						goto errout;
3522 
3523 					n = min_t(s64, nblocks, BPERDMAP);
3524 				}
3525 
3526 				dp = (struct dmap *) mp->data;
3527 				*l0leaf = dbInitDmap(dp, blkno, n);
3528 
3529 				bmp->db_nfree += n;
3530 				agno = le64_to_cpu(dp->start) >> l2agsize;
3531 				bmp->db_agfree[agno] += n;
3532 
3533 				write_metapage(mp);
3534 
3535 				l0leaf++;
3536 				p += nbperpage;
3537 
3538 				blkno += n;
3539 				nblocks -= n;
3540 				if (nblocks == 0)
3541 					break;
3542 			}	/* for each dmap in a L0 */
3543 
3544 			/*
3545 			 * build current L0 page from its leaves, and
3546 			 * initialize corresponding parent L1 leaf
3547 			 */
3548 			*l1leaf = dbInitDmapCtl(l0dcp, 0, ++i);
3549 			write_metapage(l0mp);
3550 			l0mp = NULL;
3551 
3552 			if (nblocks)
3553 				l1leaf++;	/* continue for next L0 */
3554 			else {
3555 				/* more than 1 L0 ? */
3556 				if (j > 0)
3557 					break;	/* build L1 page */
3558 				else {
3559 					/* summarize in global bmap page */
3560 					bmp->db_maxfreebud = *l1leaf;
3561 					release_metapage(l1mp);
3562 					release_metapage(l2mp);
3563 					goto finalize;
3564 				}
3565 			}
3566 		}		/* for each L0 in a L1 */
3567 
3568 		/*
3569 		 * build current L1 page from its leaves, and
3570 		 * initialize corresponding parent L2 leaf
3571 		 */
3572 		*l2leaf = dbInitDmapCtl(l1dcp, 1, ++j);
3573 		write_metapage(l1mp);
3574 		l1mp = NULL;
3575 
3576 		if (nblocks)
3577 			l2leaf++;	/* continue for next L1 */
3578 		else {
3579 			/* more than 1 L1 ? */
3580 			if (k > 0)
3581 				break;	/* build L2 page */
3582 			else {
3583 				/* summarize in global bmap page */
3584 				bmp->db_maxfreebud = *l2leaf;
3585 				release_metapage(l2mp);
3586 				goto finalize;
3587 			}
3588 		}
3589 	}			/* for each L1 in a L2 */
3590 
3591 	jfs_error(ipbmap->i_sb, "function has not returned as expected\n");
3592 errout:
3593 	if (l0mp)
3594 		release_metapage(l0mp);
3595 	if (l1mp)
3596 		release_metapage(l1mp);
3597 	release_metapage(l2mp);
3598 	return -EIO;
3599 
3600 	/*
3601 	 *	finalize bmap control page
3602 	 */
3603 finalize:
3604 
3605 	return 0;
3606 }
3607 
3608 
3609 /*
3610  *	dbFinalizeBmap()
3611  */
3612 void dbFinalizeBmap(struct inode *ipbmap)
3613 {
3614 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
3615 	int actags, inactags, l2nl;
3616 	s64 ag_rem, actfree, inactfree, avgfree;
3617 	int i, n;
3618 
3619 	/*
3620 	 *	finalize bmap control page
3621 	 */
3622 //finalize:
3623 	/*
3624 	 * compute db_agpref: preferred ag to allocate from
3625 	 * (the leftmost ag with average free space in it);
3626 	 */
3627 //agpref:
3628 	/* get the number of active ags and inactive ags */
3629 	actags = bmp->db_maxag + 1;
3630 	inactags = bmp->db_numag - actags;
3631 	ag_rem = bmp->db_mapsize & (bmp->db_agsize - 1);	/* ??? */
3632 
3633 	/* determine how many blocks are in the inactive allocation
3634 	 * groups. in doing this, we must account for the fact that
3635 	 * the rightmost group might be a partial group (i.e. file
3636 	 * system size is not a multiple of the group size).
3637 	 */
3638 	inactfree = (inactags && ag_rem) ?
3639 	    ((inactags - 1) << bmp->db_agl2size) + ag_rem
3640 	    : inactags << bmp->db_agl2size;
3641 
3642 	/* determine how many free blocks are in the active
3643 	 * allocation groups plus the average number of free blocks
3644 	 * within the active ags.
3645 	 */
3646 	actfree = bmp->db_nfree - inactfree;
3647 	avgfree = (u32) actfree / (u32) actags;
3648 
3649 	/* if the preferred allocation group has not average free space.
3650 	 * re-establish the preferred group as the leftmost
3651 	 * group with average free space.
3652 	 */
3653 	if (bmp->db_agfree[bmp->db_agpref] < avgfree) {
3654 		for (bmp->db_agpref = 0; bmp->db_agpref < actags;
3655 		     bmp->db_agpref++) {
3656 			if (bmp->db_agfree[bmp->db_agpref] >= avgfree)
3657 				break;
3658 		}
3659 		if (bmp->db_agpref >= bmp->db_numag) {
3660 			jfs_error(ipbmap->i_sb,
3661 				  "cannot find ag with average freespace\n");
3662 		}
3663 	}
3664 
3665 	/*
3666 	 * compute db_aglevel, db_agheight, db_width, db_agstart:
3667 	 * an ag is covered in aglevel dmapctl summary tree,
3668 	 * at agheight level height (from leaf) with agwidth number of nodes
3669 	 * each, which starts at agstart index node of the smmary tree node
3670 	 * array;
3671 	 */
3672 	bmp->db_aglevel = BMAPSZTOLEV(bmp->db_agsize);
3673 	l2nl =
3674 	    bmp->db_agl2size - (L2BPERDMAP + bmp->db_aglevel * L2LPERCTL);
3675 	bmp->db_agheight = l2nl >> 1;
3676 	bmp->db_agwidth = 1 << (l2nl - (bmp->db_agheight << 1));
3677 	for (i = 5 - bmp->db_agheight, bmp->db_agstart = 0, n = 1; i > 0;
3678 	     i--) {
3679 		bmp->db_agstart += n;
3680 		n <<= 2;
3681 	}
3682 
3683 }
3684 
3685 
3686 /*
3687  * NAME:	dbInitDmap()/ujfs_idmap_page()
3688  *
3689  * FUNCTION:	initialize working/persistent bitmap of the dmap page
3690  *		for the specified number of blocks:
3691  *
3692  *		at entry, the bitmaps had been initialized as free (ZEROS);
3693  *		The number of blocks will only account for the actually
3694  *		existing blocks. Blocks which don't actually exist in
3695  *		the aggregate will be marked as allocated (ONES);
3696  *
3697  * PARAMETERS:
3698  *	dp	- pointer to page of map
3699  *	nblocks	- number of blocks this page
3700  *
3701  * RETURNS: NONE
3702  */
3703 static int dbInitDmap(struct dmap * dp, s64 Blkno, int nblocks)
3704 {
3705 	int blkno, w, b, r, nw, nb, i;
3706 
3707 	/* starting block number within the dmap */
3708 	blkno = Blkno & (BPERDMAP - 1);
3709 
3710 	if (blkno == 0) {
3711 		dp->nblocks = dp->nfree = cpu_to_le32(nblocks);
3712 		dp->start = cpu_to_le64(Blkno);
3713 
3714 		if (nblocks == BPERDMAP) {
3715 			memset(&dp->wmap[0], 0, LPERDMAP * 4);
3716 			memset(&dp->pmap[0], 0, LPERDMAP * 4);
3717 			goto initTree;
3718 		}
3719 	} else {
3720 		le32_add_cpu(&dp->nblocks, nblocks);
3721 		le32_add_cpu(&dp->nfree, nblocks);
3722 	}
3723 
3724 	/* word number containing start block number */
3725 	w = blkno >> L2DBWORD;
3726 
3727 	/*
3728 	 * free the bits corresponding to the block range (ZEROS):
3729 	 * note: not all bits of the first and last words may be contained
3730 	 * within the block range.
3731 	 */
3732 	for (r = nblocks; r > 0; r -= nb, blkno += nb) {
3733 		/* number of bits preceding range to be freed in the word */
3734 		b = blkno & (DBWORD - 1);
3735 		/* number of bits to free in the word */
3736 		nb = min(r, DBWORD - b);
3737 
3738 		/* is partial word to be freed ? */
3739 		if (nb < DBWORD) {
3740 			/* free (set to 0) from the bitmap word */
3741 			dp->wmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3742 						     >> b));
3743 			dp->pmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3744 						     >> b));
3745 
3746 			/* skip the word freed */
3747 			w++;
3748 		} else {
3749 			/* free (set to 0) contiguous bitmap words */
3750 			nw = r >> L2DBWORD;
3751 			memset(&dp->wmap[w], 0, nw * 4);
3752 			memset(&dp->pmap[w], 0, nw * 4);
3753 
3754 			/* skip the words freed */
3755 			nb = nw << L2DBWORD;
3756 			w += nw;
3757 		}
3758 	}
3759 
3760 	/*
3761 	 * mark bits following the range to be freed (non-existing
3762 	 * blocks) as allocated (ONES)
3763 	 */
3764 
3765 	if (blkno == BPERDMAP)
3766 		goto initTree;
3767 
3768 	/* the first word beyond the end of existing blocks */
3769 	w = blkno >> L2DBWORD;
3770 
3771 	/* does nblocks fall on a 32-bit boundary ? */
3772 	b = blkno & (DBWORD - 1);
3773 	if (b) {
3774 		/* mark a partial word allocated */
3775 		dp->wmap[w] = dp->pmap[w] = cpu_to_le32(ONES >> b);
3776 		w++;
3777 	}
3778 
3779 	/* set the rest of the words in the page to allocated (ONES) */
3780 	for (i = w; i < LPERDMAP; i++)
3781 		dp->pmap[i] = dp->wmap[i] = cpu_to_le32(ONES);
3782 
3783 	/*
3784 	 * init tree
3785 	 */
3786       initTree:
3787 	return (dbInitDmapTree(dp));
3788 }
3789 
3790 
3791 /*
3792  * NAME:	dbInitDmapTree()/ujfs_complete_dmap()
3793  *
3794  * FUNCTION:	initialize summary tree of the specified dmap:
3795  *
3796  *		at entry, bitmap of the dmap has been initialized;
3797  *
3798  * PARAMETERS:
3799  *	dp	- dmap to complete
3800  *	blkno	- starting block number for this dmap
3801  *	treemax	- will be filled in with max free for this dmap
3802  *
3803  * RETURNS:	max free string at the root of the tree
3804  */
3805 static int dbInitDmapTree(struct dmap * dp)
3806 {
3807 	struct dmaptree *tp;
3808 	s8 *cp;
3809 	int i;
3810 
3811 	/* init fixed info of tree */
3812 	tp = &dp->tree;
3813 	tp->nleafs = cpu_to_le32(LPERDMAP);
3814 	tp->l2nleafs = cpu_to_le32(L2LPERDMAP);
3815 	tp->leafidx = cpu_to_le32(LEAFIND);
3816 	tp->height = cpu_to_le32(4);
3817 	tp->budmin = BUDMIN;
3818 
3819 	/* init each leaf from corresponding wmap word:
3820 	 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3821 	 * bitmap word are allocated.
3822 	 */
3823 	cp = tp->stree + le32_to_cpu(tp->leafidx);
3824 	for (i = 0; i < LPERDMAP; i++)
3825 		*cp++ = dbMaxBud((u8 *) & dp->wmap[i]);
3826 
3827 	/* build the dmap's binary buddy summary tree */
3828 	return (dbInitTree(tp));
3829 }
3830 
3831 
3832 /*
3833  * NAME:	dbInitTree()/ujfs_adjtree()
3834  *
3835  * FUNCTION:	initialize binary buddy summary tree of a dmap or dmapctl.
3836  *
3837  *		at entry, the leaves of the tree has been initialized
3838  *		from corresponding bitmap word or root of summary tree
3839  *		of the child control page;
3840  *		configure binary buddy system at the leaf level, then
3841  *		bubble up the values of the leaf nodes up the tree.
3842  *
3843  * PARAMETERS:
3844  *	cp	- Pointer to the root of the tree
3845  *	l2leaves- Number of leaf nodes as a power of 2
3846  *	l2min	- Number of blocks that can be covered by a leaf
3847  *		  as a power of 2
3848  *
3849  * RETURNS: max free string at the root of the tree
3850  */
3851 static int dbInitTree(struct dmaptree * dtp)
3852 {
3853 	int l2max, l2free, bsize, nextb, i;
3854 	int child, parent, nparent;
3855 	s8 *tp, *cp, *cp1;
3856 
3857 	tp = dtp->stree;
3858 
3859 	/* Determine the maximum free string possible for the leaves */
3860 	l2max = le32_to_cpu(dtp->l2nleafs) + dtp->budmin;
3861 
3862 	/*
3863 	 * configure the leaf level into binary buddy system
3864 	 *
3865 	 * Try to combine buddies starting with a buddy size of 1
3866 	 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3867 	 * can be combined if both buddies have a maximum free of l2min;
3868 	 * the combination will result in the left-most buddy leaf having
3869 	 * a maximum free of l2min+1.
3870 	 * After processing all buddies for a given size, process buddies
3871 	 * at the next higher buddy size (i.e. current size * 2) and
3872 	 * the next maximum free (current free + 1).
3873 	 * This continues until the maximum possible buddy combination
3874 	 * yields maximum free.
3875 	 */
3876 	for (l2free = dtp->budmin, bsize = 1; l2free < l2max;
3877 	     l2free++, bsize = nextb) {
3878 		/* get next buddy size == current buddy pair size */
3879 		nextb = bsize << 1;
3880 
3881 		/* scan each adjacent buddy pair at current buddy size */
3882 		for (i = 0, cp = tp + le32_to_cpu(dtp->leafidx);
3883 		     i < le32_to_cpu(dtp->nleafs);
3884 		     i += nextb, cp += nextb) {
3885 			/* coalesce if both adjacent buddies are max free */
3886 			if (*cp == l2free && *(cp + bsize) == l2free) {
3887 				*cp = l2free + 1;	/* left take right */
3888 				*(cp + bsize) = -1;	/* right give left */
3889 			}
3890 		}
3891 	}
3892 
3893 	/*
3894 	 * bubble summary information of leaves up the tree.
3895 	 *
3896 	 * Starting at the leaf node level, the four nodes described by
3897 	 * the higher level parent node are compared for a maximum free and
3898 	 * this maximum becomes the value of the parent node.
3899 	 * when all lower level nodes are processed in this fashion then
3900 	 * move up to the next level (parent becomes a lower level node) and
3901 	 * continue the process for that level.
3902 	 */
3903 	for (child = le32_to_cpu(dtp->leafidx),
3904 	     nparent = le32_to_cpu(dtp->nleafs) >> 2;
3905 	     nparent > 0; nparent >>= 2, child = parent) {
3906 		/* get index of 1st node of parent level */
3907 		parent = (child - 1) >> 2;
3908 
3909 		/* set the value of the parent node as the maximum
3910 		 * of the four nodes of the current level.
3911 		 */
3912 		for (i = 0, cp = tp + child, cp1 = tp + parent;
3913 		     i < nparent; i++, cp += 4, cp1++)
3914 			*cp1 = TREEMAX(cp);
3915 	}
3916 
3917 	return (*tp);
3918 }
3919 
3920 
3921 /*
3922  *	dbInitDmapCtl()
3923  *
3924  * function: initialize dmapctl page
3925  */
3926 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i)
3927 {				/* start leaf index not covered by range */
3928 	s8 *cp;
3929 
3930 	dcp->nleafs = cpu_to_le32(LPERCTL);
3931 	dcp->l2nleafs = cpu_to_le32(L2LPERCTL);
3932 	dcp->leafidx = cpu_to_le32(CTLLEAFIND);
3933 	dcp->height = cpu_to_le32(5);
3934 	dcp->budmin = L2BPERDMAP + L2LPERCTL * level;
3935 
3936 	/*
3937 	 * initialize the leaves of current level that were not covered
3938 	 * by the specified input block range (i.e. the leaves have no
3939 	 * low level dmapctl or dmap).
3940 	 */
3941 	cp = &dcp->stree[CTLLEAFIND + i];
3942 	for (; i < LPERCTL; i++)
3943 		*cp++ = NOFREE;
3944 
3945 	/* build the dmap's binary buddy summary tree */
3946 	return (dbInitTree((struct dmaptree *) dcp));
3947 }
3948 
3949 
3950 /*
3951  * NAME:	dbGetL2AGSize()/ujfs_getagl2size()
3952  *
3953  * FUNCTION:	Determine log2(allocation group size) from aggregate size
3954  *
3955  * PARAMETERS:
3956  *	nblocks	- Number of blocks in aggregate
3957  *
3958  * RETURNS: log2(allocation group size) in aggregate blocks
3959  */
3960 static int dbGetL2AGSize(s64 nblocks)
3961 {
3962 	s64 sz;
3963 	s64 m;
3964 	int l2sz;
3965 
3966 	if (nblocks < BPERDMAP * MAXAG)
3967 		return (L2BPERDMAP);
3968 
3969 	/* round up aggregate size to power of 2 */
3970 	m = ((u64) 1 << (64 - 1));
3971 	for (l2sz = 64; l2sz >= 0; l2sz--, m >>= 1) {
3972 		if (m & nblocks)
3973 			break;
3974 	}
3975 
3976 	sz = (s64) 1 << l2sz;
3977 	if (sz < nblocks)
3978 		l2sz += 1;
3979 
3980 	/* agsize = roundupSize/max_number_of_ag */
3981 	return (l2sz - L2MAXAG);
3982 }
3983 
3984 
3985 /*
3986  * NAME:	dbMapFileSizeToMapSize()
3987  *
3988  * FUNCTION:	compute number of blocks the block allocation map file
3989  *		can cover from the map file size;
3990  *
3991  * RETURNS:	Number of blocks which can be covered by this block map file;
3992  */
3993 
3994 /*
3995  * maximum number of map pages at each level including control pages
3996  */
3997 #define MAXL0PAGES	(1 + LPERCTL)
3998 #define MAXL1PAGES	(1 + LPERCTL * MAXL0PAGES)
3999 
4000 /*
4001  * convert number of map pages to the zero origin top dmapctl level
4002  */
4003 #define BMAPPGTOLEV(npages)	\
4004 	(((npages) <= 3 + MAXL0PAGES) ? 0 : \
4005 	 ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
4006 
4007 s64 dbMapFileSizeToMapSize(struct inode * ipbmap)
4008 {
4009 	struct super_block *sb = ipbmap->i_sb;
4010 	s64 nblocks;
4011 	s64 npages, ndmaps;
4012 	int level, i;
4013 	int complete, factor;
4014 
4015 	nblocks = ipbmap->i_size >> JFS_SBI(sb)->l2bsize;
4016 	npages = nblocks >> JFS_SBI(sb)->l2nbperpage;
4017 	level = BMAPPGTOLEV(npages);
4018 
4019 	/* At each level, accumulate the number of dmap pages covered by
4020 	 * the number of full child levels below it;
4021 	 * repeat for the last incomplete child level.
4022 	 */
4023 	ndmaps = 0;
4024 	npages--;		/* skip the first global control page */
4025 	/* skip higher level control pages above top level covered by map */
4026 	npages -= (2 - level);
4027 	npages--;		/* skip top level's control page */
4028 	for (i = level; i >= 0; i--) {
4029 		factor =
4030 		    (i == 2) ? MAXL1PAGES : ((i == 1) ? MAXL0PAGES : 1);
4031 		complete = (u32) npages / factor;
4032 		ndmaps += complete * ((i == 2) ? LPERCTL * LPERCTL :
4033 				      ((i == 1) ? LPERCTL : 1));
4034 
4035 		/* pages in last/incomplete child */
4036 		npages = (u32) npages % factor;
4037 		/* skip incomplete child's level control page */
4038 		npages--;
4039 	}
4040 
4041 	/* convert the number of dmaps into the number of blocks
4042 	 * which can be covered by the dmaps;
4043 	 */
4044 	nblocks = ndmaps << L2BPERDMAP;
4045 
4046 	return (nblocks);
4047 }
4048