xref: /linux/fs/jfs/jfs_dmap.c (revision dbcedec3a31119d7594baacc743300d127c99c56)
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, bool is_ctl);
67 static int dbBackSplit(dmtree_t *tp, int leafno, bool is_ctl);
68 static int dbJoin(dmtree_t *tp, int leafno, int newval, bool is_ctl);
69 static void dbAdjTree(dmtree_t *tp, int leafno, int newval, bool is_ctl);
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, bool is_ctl);
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 		bmp->db_l2nbperpage < 0) {
185 		err = -EINVAL;
186 		goto err_release_metapage;
187 	}
188 
189 	bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag);
190 	if (!bmp->db_numag) {
191 		err = -EINVAL;
192 		goto err_release_metapage;
193 	}
194 
195 	bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel);
196 	bmp->db_maxag = le32_to_cpu(dbmp_le->dn_maxag);
197 	bmp->db_agpref = le32_to_cpu(dbmp_le->dn_agpref);
198 	if (bmp->db_maxag >= MAXAG || bmp->db_maxag < 0 ||
199 		bmp->db_agpref >= MAXAG || bmp->db_agpref < 0) {
200 		err = -EINVAL;
201 		goto err_release_metapage;
202 	}
203 
204 	bmp->db_aglevel = le32_to_cpu(dbmp_le->dn_aglevel);
205 	bmp->db_agheight = le32_to_cpu(dbmp_le->dn_agheight);
206 	bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth);
207 	bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart);
208 	bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size);
209 	if (bmp->db_agl2size > L2MAXL2SIZE - L2MAXAG ||
210 	    bmp->db_agl2size < 0) {
211 		err = -EINVAL;
212 		goto err_release_metapage;
213 	}
214 
215 	if (((bmp->db_mapsize - 1) >> bmp->db_agl2size) > MAXAG) {
216 		err = -EINVAL;
217 		goto err_release_metapage;
218 	}
219 
220 	for (i = 0; i < MAXAG; i++)
221 		bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
222 	bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
223 	bmp->db_maxfreebud = dbmp_le->dn_maxfreebud;
224 
225 	/* release the buffer. */
226 	release_metapage(mp);
227 
228 	/* bind the bmap inode and the bmap descriptor to each other. */
229 	bmp->db_ipbmap = ipbmap;
230 	JFS_SBI(ipbmap->i_sb)->bmap = bmp;
231 
232 	memset(bmp->db_active, 0, sizeof(bmp->db_active));
233 
234 	/*
235 	 * allocate/initialize the bmap lock
236 	 */
237 	BMAP_LOCK_INIT(bmp);
238 
239 	return (0);
240 
241 err_release_metapage:
242 	release_metapage(mp);
243 err_kfree_bmp:
244 	kfree(bmp);
245 	return err;
246 }
247 
248 
249 /*
250  * NAME:	dbUnmount()
251  *
252  * FUNCTION:	terminate the block allocation map in preparation for
253  *		file system unmount.
254  *
255  *		the in-core bmap descriptor is written to disk and
256  *		the memory for this descriptor is freed.
257  *
258  * PARAMETERS:
259  *	ipbmap	- pointer to in-core inode for the block map.
260  *
261  * RETURN VALUES:
262  *	0	- success
263  *	-EIO	- i/o error
264  */
265 int dbUnmount(struct inode *ipbmap, int mounterror)
266 {
267 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
268 
269 	if (!(mounterror || isReadOnly(ipbmap)))
270 		dbSync(ipbmap);
271 
272 	/*
273 	 * Invalidate the page cache buffers
274 	 */
275 	truncate_inode_pages(ipbmap->i_mapping, 0);
276 
277 	/* free the memory for the in-memory bmap. */
278 	kfree(bmp);
279 	JFS_SBI(ipbmap->i_sb)->bmap = NULL;
280 
281 	return (0);
282 }
283 
284 /*
285  *	dbSync()
286  */
287 int dbSync(struct inode *ipbmap)
288 {
289 	struct dbmap_disk *dbmp_le;
290 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
291 	struct metapage *mp;
292 	int i;
293 
294 	/*
295 	 * write bmap global control page
296 	 */
297 	/* get the buffer for the on-disk bmap descriptor. */
298 	mp = read_metapage(ipbmap,
299 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
300 			   PSIZE, 0);
301 	if (mp == NULL) {
302 		jfs_err("dbSync: read_metapage failed!");
303 		return -EIO;
304 	}
305 	/* copy the in-memory version of the bmap to the on-disk version */
306 	dbmp_le = (struct dbmap_disk *) mp->data;
307 	dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize);
308 	dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree);
309 	dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage);
310 	dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag);
311 	dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel);
312 	dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag);
313 	dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref);
314 	dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel);
315 	dbmp_le->dn_agheight = cpu_to_le32(bmp->db_agheight);
316 	dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth);
317 	dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart);
318 	dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size);
319 	for (i = 0; i < MAXAG; i++)
320 		dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]);
321 	dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize);
322 	dbmp_le->dn_maxfreebud = bmp->db_maxfreebud;
323 
324 	/* write the buffer */
325 	write_metapage(mp);
326 
327 	/*
328 	 * write out dirty pages of bmap
329 	 */
330 	filemap_write_and_wait(ipbmap->i_mapping);
331 
332 	diWriteSpecial(ipbmap, 0);
333 
334 	return (0);
335 }
336 
337 /*
338  * NAME:	dbFree()
339  *
340  * FUNCTION:	free the specified block range from the working block
341  *		allocation map.
342  *
343  *		the blocks will be free from the working map one dmap
344  *		at a time.
345  *
346  * PARAMETERS:
347  *	ip	- pointer to in-core inode;
348  *	blkno	- starting block number to be freed.
349  *	nblocks	- number of blocks to be freed.
350  *
351  * RETURN VALUES:
352  *	0	- success
353  *	-EIO	- i/o error
354  */
355 int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
356 {
357 	struct metapage *mp;
358 	struct dmap *dp;
359 	int nb, rc;
360 	s64 lblkno, rem;
361 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
362 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
363 	struct super_block *sb = ipbmap->i_sb;
364 
365 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
366 
367 	/* block to be freed better be within the mapsize. */
368 	if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
369 		IREAD_UNLOCK(ipbmap);
370 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
371 		       (unsigned long long) blkno,
372 		       (unsigned long long) nblocks);
373 		jfs_error(ip->i_sb, "block to be freed is outside the map\n");
374 		return -EIO;
375 	}
376 
377 	/**
378 	 * TRIM the blocks, when mounted with discard option
379 	 */
380 	if (JFS_SBI(sb)->flag & JFS_DISCARD)
381 		if (JFS_SBI(sb)->minblks_trim <= nblocks)
382 			jfs_issue_discard(ipbmap, blkno, nblocks);
383 
384 	/*
385 	 * free the blocks a dmap at a time.
386 	 */
387 	mp = NULL;
388 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
389 		/* release previous dmap if any */
390 		if (mp) {
391 			write_metapage(mp);
392 		}
393 
394 		/* get the buffer for the current dmap. */
395 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
396 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
397 		if (mp == NULL) {
398 			IREAD_UNLOCK(ipbmap);
399 			return -EIO;
400 		}
401 		dp = (struct dmap *) mp->data;
402 
403 		/* determine the number of blocks to be freed from
404 		 * this dmap.
405 		 */
406 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
407 
408 		/* free the blocks. */
409 		if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
410 			jfs_error(ip->i_sb, "error in block map\n");
411 			release_metapage(mp);
412 			IREAD_UNLOCK(ipbmap);
413 			return (rc);
414 		}
415 	}
416 
417 	/* write the last buffer. */
418 	if (mp)
419 		write_metapage(mp);
420 
421 	IREAD_UNLOCK(ipbmap);
422 
423 	return (0);
424 }
425 
426 
427 /*
428  * NAME:	dbUpdatePMap()
429  *
430  * FUNCTION:	update the allocation state (free or allocate) of the
431  *		specified block range in the persistent block allocation map.
432  *
433  *		the blocks will be updated in the persistent map one
434  *		dmap at a time.
435  *
436  * PARAMETERS:
437  *	ipbmap	- pointer to in-core inode for the block map.
438  *	free	- 'true' if block range is to be freed from the persistent
439  *		  map; 'false' if it is to be allocated.
440  *	blkno	- starting block number of the range.
441  *	nblocks	- number of contiguous blocks in the range.
442  *	tblk	- transaction block;
443  *
444  * RETURN VALUES:
445  *	0	- success
446  *	-EIO	- i/o error
447  */
448 int
449 dbUpdatePMap(struct inode *ipbmap,
450 	     int free, s64 blkno, s64 nblocks, struct tblock * tblk)
451 {
452 	int nblks, dbitno, wbitno, rbits;
453 	int word, nbits, nwords;
454 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
455 	s64 lblkno, rem, lastlblkno;
456 	u32 mask;
457 	struct dmap *dp;
458 	struct metapage *mp;
459 	struct jfs_log *log;
460 	int lsn, difft, diffp;
461 	unsigned long flags;
462 
463 	/* the blocks better be within the mapsize. */
464 	if (blkno + nblocks > bmp->db_mapsize) {
465 		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
466 		       (unsigned long long) blkno,
467 		       (unsigned long long) nblocks);
468 		jfs_error(ipbmap->i_sb, "blocks are outside the map\n");
469 		return -EIO;
470 	}
471 
472 	/* compute delta of transaction lsn from log syncpt */
473 	lsn = tblk->lsn;
474 	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
475 	logdiff(difft, lsn, log);
476 
477 	/*
478 	 * update the block state a dmap at a time.
479 	 */
480 	mp = NULL;
481 	lastlblkno = 0;
482 	for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) {
483 		/* get the buffer for the current dmap. */
484 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
485 		if (lblkno != lastlblkno) {
486 			if (mp) {
487 				write_metapage(mp);
488 			}
489 
490 			mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE,
491 					   0);
492 			if (mp == NULL)
493 				return -EIO;
494 			metapage_wait_for_io(mp);
495 		}
496 		dp = (struct dmap *) mp->data;
497 
498 		/* determine the bit number and word within the dmap of
499 		 * the starting block.  also determine how many blocks
500 		 * are to be updated within this dmap.
501 		 */
502 		dbitno = blkno & (BPERDMAP - 1);
503 		word = dbitno >> L2DBWORD;
504 		nblks = min(rem, (s64)BPERDMAP - dbitno);
505 
506 		/* update the bits of the dmap words. the first and last
507 		 * words may only have a subset of their bits updated. if
508 		 * this is the case, we'll work against that word (i.e.
509 		 * partial first and/or last) only in a single pass.  a
510 		 * single pass will also be used to update all words that
511 		 * are to have all their bits updated.
512 		 */
513 		for (rbits = nblks; rbits > 0;
514 		     rbits -= nbits, dbitno += nbits) {
515 			/* determine the bit number within the word and
516 			 * the number of bits within the word.
517 			 */
518 			wbitno = dbitno & (DBWORD - 1);
519 			nbits = min(rbits, DBWORD - wbitno);
520 
521 			/* check if only part of the word is to be updated. */
522 			if (nbits < DBWORD) {
523 				/* update (free or allocate) the bits
524 				 * in this word.
525 				 */
526 				mask =
527 				    (ONES << (DBWORD - nbits) >> wbitno);
528 				if (free)
529 					dp->pmap[word] &=
530 					    cpu_to_le32(~mask);
531 				else
532 					dp->pmap[word] |=
533 					    cpu_to_le32(mask);
534 
535 				word += 1;
536 			} else {
537 				/* one or more words are to have all
538 				 * their bits updated.  determine how
539 				 * many words and how many bits.
540 				 */
541 				nwords = rbits >> L2DBWORD;
542 				nbits = nwords << L2DBWORD;
543 
544 				/* update (free or allocate) the bits
545 				 * in these words.
546 				 */
547 				if (free)
548 					memset(&dp->pmap[word], 0,
549 					       nwords * 4);
550 				else
551 					memset(&dp->pmap[word], (int) ONES,
552 					       nwords * 4);
553 
554 				word += nwords;
555 			}
556 		}
557 
558 		/*
559 		 * update dmap lsn
560 		 */
561 		if (lblkno == lastlblkno)
562 			continue;
563 
564 		lastlblkno = lblkno;
565 
566 		LOGSYNC_LOCK(log, flags);
567 		if (mp->lsn != 0) {
568 			/* inherit older/smaller lsn */
569 			logdiff(diffp, mp->lsn, log);
570 			if (difft < diffp) {
571 				mp->lsn = lsn;
572 
573 				/* move bp after tblock in logsync list */
574 				list_move(&mp->synclist, &tblk->synclist);
575 			}
576 
577 			/* inherit younger/larger clsn */
578 			logdiff(difft, tblk->clsn, log);
579 			logdiff(diffp, mp->clsn, log);
580 			if (difft > diffp)
581 				mp->clsn = tblk->clsn;
582 		} else {
583 			mp->log = log;
584 			mp->lsn = lsn;
585 
586 			/* insert bp after tblock in logsync list */
587 			log->count++;
588 			list_add(&mp->synclist, &tblk->synclist);
589 
590 			mp->clsn = tblk->clsn;
591 		}
592 		LOGSYNC_UNLOCK(log, flags);
593 	}
594 
595 	/* write the last buffer. */
596 	if (mp) {
597 		write_metapage(mp);
598 	}
599 
600 	return (0);
601 }
602 
603 
604 /*
605  * NAME:	dbNextAG()
606  *
607  * FUNCTION:	find the preferred allocation group for new allocations.
608  *
609  *		Within the allocation groups, we maintain a preferred
610  *		allocation group which consists of a group with at least
611  *		average free space.  It is the preferred group that we target
612  *		new inode allocation towards.  The tie-in between inode
613  *		allocation and block allocation occurs as we allocate the
614  *		first (data) block of an inode and specify the inode (block)
615  *		as the allocation hint for this block.
616  *
617  *		We try to avoid having more than one open file growing in
618  *		an allocation group, as this will lead to fragmentation.
619  *		This differs from the old OS/2 method of trying to keep
620  *		empty ags around for large allocations.
621  *
622  * PARAMETERS:
623  *	ipbmap	- pointer to in-core inode for the block map.
624  *
625  * RETURN VALUES:
626  *	the preferred allocation group number.
627  */
628 int dbNextAG(struct inode *ipbmap)
629 {
630 	s64 avgfree;
631 	int agpref;
632 	s64 hwm = 0;
633 	int i;
634 	int next_best = -1;
635 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
636 
637 	BMAP_LOCK(bmp);
638 
639 	/* determine the average number of free blocks within the ags. */
640 	avgfree = (u32)bmp->db_nfree / bmp->db_numag;
641 
642 	/*
643 	 * if the current preferred ag does not have an active allocator
644 	 * and has at least average freespace, return it
645 	 */
646 	agpref = bmp->db_agpref;
647 	if ((atomic_read(&bmp->db_active[agpref]) == 0) &&
648 	    (bmp->db_agfree[agpref] >= avgfree))
649 		goto unlock;
650 
651 	/* From the last preferred ag, find the next one with at least
652 	 * average free space.
653 	 */
654 	for (i = 0 ; i < bmp->db_numag; i++, agpref++) {
655 		if (agpref == bmp->db_numag)
656 			agpref = 0;
657 
658 		if (atomic_read(&bmp->db_active[agpref]))
659 			/* open file is currently growing in this ag */
660 			continue;
661 		if (bmp->db_agfree[agpref] >= avgfree) {
662 			/* Return this one */
663 			bmp->db_agpref = agpref;
664 			goto unlock;
665 		} else if (bmp->db_agfree[agpref] > hwm) {
666 			/* Less than avg. freespace, but best so far */
667 			hwm = bmp->db_agfree[agpref];
668 			next_best = agpref;
669 		}
670 	}
671 
672 	/*
673 	 * If no inactive ag was found with average freespace, use the
674 	 * next best
675 	 */
676 	if (next_best != -1)
677 		bmp->db_agpref = next_best;
678 	/* else leave db_agpref unchanged */
679 unlock:
680 	BMAP_UNLOCK(bmp);
681 
682 	/* return the preferred group.
683 	 */
684 	return (bmp->db_agpref);
685 }
686 
687 /*
688  * NAME:	dbAlloc()
689  *
690  * FUNCTION:	attempt to allocate a specified number of contiguous free
691  *		blocks from the working allocation block map.
692  *
693  *		the block allocation policy uses hints and a multi-step
694  *		approach.
695  *
696  *		for allocation requests smaller than the number of blocks
697  *		per dmap, we first try to allocate the new blocks
698  *		immediately following the hint.  if these blocks are not
699  *		available, we try to allocate blocks near the hint.  if
700  *		no blocks near the hint are available, we next try to
701  *		allocate within the same dmap as contains the hint.
702  *
703  *		if no blocks are available in the dmap or the allocation
704  *		request is larger than the dmap size, we try to allocate
705  *		within the same allocation group as contains the hint. if
706  *		this does not succeed, we finally try to allocate anywhere
707  *		within the aggregate.
708  *
709  *		we also try to allocate anywhere within the aggregate
710  *		for allocation requests larger than the allocation group
711  *		size or requests that specify no hint value.
712  *
713  * PARAMETERS:
714  *	ip	- pointer to in-core inode;
715  *	hint	- allocation hint.
716  *	nblocks	- number of contiguous blocks in the range.
717  *	results	- on successful return, set to the starting block number
718  *		  of the newly allocated contiguous range.
719  *
720  * RETURN VALUES:
721  *	0	- success
722  *	-ENOSPC	- insufficient disk resources
723  *	-EIO	- i/o error
724  */
725 int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results)
726 {
727 	int rc, agno;
728 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
729 	struct bmap *bmp;
730 	struct metapage *mp;
731 	s64 lblkno, blkno;
732 	struct dmap *dp;
733 	int l2nb;
734 	s64 mapSize;
735 	int writers;
736 
737 	/* assert that nblocks is valid */
738 	assert(nblocks > 0);
739 
740 	/* get the log2 number of blocks to be allocated.
741 	 * if the number of blocks is not a log2 multiple,
742 	 * it will be rounded up to the next log2 multiple.
743 	 */
744 	l2nb = BLKSTOL2(nblocks);
745 
746 	bmp = JFS_SBI(ip->i_sb)->bmap;
747 
748 	mapSize = bmp->db_mapsize;
749 
750 	/* the hint should be within the map */
751 	if (hint >= mapSize) {
752 		jfs_error(ip->i_sb, "the hint is outside the map\n");
753 		return -EIO;
754 	}
755 
756 	/* if the number of blocks to be allocated is greater than the
757 	 * allocation group size, try to allocate anywhere.
758 	 */
759 	if (l2nb > bmp->db_agl2size) {
760 		IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
761 
762 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
763 
764 		goto write_unlock;
765 	}
766 
767 	/*
768 	 * If no hint, let dbNextAG recommend an allocation group
769 	 */
770 	if (hint == 0)
771 		goto pref_ag;
772 
773 	/* we would like to allocate close to the hint.  adjust the
774 	 * hint to the block following the hint since the allocators
775 	 * will start looking for free space starting at this point.
776 	 */
777 	blkno = hint + 1;
778 
779 	if (blkno >= bmp->db_mapsize)
780 		goto pref_ag;
781 
782 	agno = blkno >> bmp->db_agl2size;
783 
784 	/* check if blkno crosses over into a new allocation group.
785 	 * if so, check if we should allow allocations within this
786 	 * allocation group.
787 	 */
788 	if ((blkno & (bmp->db_agsize - 1)) == 0)
789 		/* check if the AG is currently being written to.
790 		 * if so, call dbNextAG() to find a non-busy
791 		 * AG with sufficient free space.
792 		 */
793 		if (atomic_read(&bmp->db_active[agno]))
794 			goto pref_ag;
795 
796 	/* check if the allocation request size can be satisfied from a
797 	 * single dmap.  if so, try to allocate from the dmap containing
798 	 * the hint using a tiered strategy.
799 	 */
800 	if (nblocks <= BPERDMAP) {
801 		IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
802 
803 		/* get the buffer for the dmap containing the hint.
804 		 */
805 		rc = -EIO;
806 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
807 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
808 		if (mp == NULL)
809 			goto read_unlock;
810 
811 		dp = (struct dmap *) mp->data;
812 
813 		/* first, try to satisfy the allocation request with the
814 		 * blocks beginning at the hint.
815 		 */
816 		if ((rc = dbAllocNext(bmp, dp, blkno, (int) nblocks))
817 		    != -ENOSPC) {
818 			if (rc == 0) {
819 				*results = blkno;
820 				mark_metapage_dirty(mp);
821 			}
822 
823 			release_metapage(mp);
824 			goto read_unlock;
825 		}
826 
827 		writers = atomic_read(&bmp->db_active[agno]);
828 		if ((writers > 1) ||
829 		    ((writers == 1) && (JFS_IP(ip)->active_ag != agno))) {
830 			/*
831 			 * Someone else is writing in this allocation
832 			 * group.  To avoid fragmenting, try another ag
833 			 */
834 			release_metapage(mp);
835 			IREAD_UNLOCK(ipbmap);
836 			goto pref_ag;
837 		}
838 
839 		/* next, try to satisfy the allocation request with blocks
840 		 * near the hint.
841 		 */
842 		if ((rc =
843 		     dbAllocNear(bmp, dp, blkno, (int) nblocks, l2nb, results))
844 		    != -ENOSPC) {
845 			if (rc == 0)
846 				mark_metapage_dirty(mp);
847 
848 			release_metapage(mp);
849 			goto read_unlock;
850 		}
851 
852 		/* try to satisfy the allocation request with blocks within
853 		 * the same dmap as the hint.
854 		 */
855 		if ((rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results))
856 		    != -ENOSPC) {
857 			if (rc == 0)
858 				mark_metapage_dirty(mp);
859 
860 			release_metapage(mp);
861 			goto read_unlock;
862 		}
863 
864 		release_metapage(mp);
865 		IREAD_UNLOCK(ipbmap);
866 	}
867 
868 	/* try to satisfy the allocation request with blocks within
869 	 * the same allocation group as the hint.
870 	 */
871 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
872 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) != -ENOSPC)
873 		goto write_unlock;
874 
875 	IWRITE_UNLOCK(ipbmap);
876 
877 
878       pref_ag:
879 	/*
880 	 * Let dbNextAG recommend a preferred allocation group
881 	 */
882 	agno = dbNextAG(ipbmap);
883 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
884 
885 	/* Try to allocate within this allocation group.  if that fails, try to
886 	 * allocate anywhere in the map.
887 	 */
888 	if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) == -ENOSPC)
889 		rc = dbAllocAny(bmp, nblocks, l2nb, results);
890 
891       write_unlock:
892 	IWRITE_UNLOCK(ipbmap);
893 
894 	return (rc);
895 
896       read_unlock:
897 	IREAD_UNLOCK(ipbmap);
898 
899 	return (rc);
900 }
901 
902 /*
903  * NAME:	dbReAlloc()
904  *
905  * FUNCTION:	attempt to extend a current allocation by a specified
906  *		number of blocks.
907  *
908  *		this routine attempts to satisfy the allocation request
909  *		by first trying to extend the existing allocation in
910  *		place by allocating the additional blocks as the blocks
911  *		immediately following the current allocation.  if these
912  *		blocks are not available, this routine will attempt to
913  *		allocate a new set of contiguous blocks large enough
914  *		to cover the existing allocation plus the additional
915  *		number of blocks required.
916  *
917  * PARAMETERS:
918  *	ip	    -  pointer to in-core inode requiring allocation.
919  *	blkno	    -  starting block of the current allocation.
920  *	nblocks	    -  number of contiguous blocks within the current
921  *		       allocation.
922  *	addnblocks  -  number of blocks to add to the allocation.
923  *	results	-      on successful return, set to the starting block number
924  *		       of the existing allocation if the existing allocation
925  *		       was extended in place or to a newly allocated contiguous
926  *		       range if the existing allocation could not be extended
927  *		       in place.
928  *
929  * RETURN VALUES:
930  *	0	- success
931  *	-ENOSPC	- insufficient disk resources
932  *	-EIO	- i/o error
933  */
934 int
935 dbReAlloc(struct inode *ip,
936 	  s64 blkno, s64 nblocks, s64 addnblocks, s64 * results)
937 {
938 	int rc;
939 
940 	/* try to extend the allocation in place.
941 	 */
942 	if ((rc = dbExtend(ip, blkno, nblocks, addnblocks)) == 0) {
943 		*results = blkno;
944 		return (0);
945 	} else {
946 		if (rc != -ENOSPC)
947 			return (rc);
948 	}
949 
950 	/* could not extend the allocation in place, so allocate a
951 	 * new set of blocks for the entire request (i.e. try to get
952 	 * a range of contiguous blocks large enough to cover the
953 	 * existing allocation plus the additional blocks.)
954 	 */
955 	return (dbAlloc
956 		(ip, blkno + nblocks - 1, addnblocks + nblocks, results));
957 }
958 
959 
960 /*
961  * NAME:	dbExtend()
962  *
963  * FUNCTION:	attempt to extend a current allocation by a specified
964  *		number of blocks.
965  *
966  *		this routine attempts to satisfy the allocation request
967  *		by first trying to extend the existing allocation in
968  *		place by allocating the additional blocks as the blocks
969  *		immediately following the current allocation.
970  *
971  * PARAMETERS:
972  *	ip	    -  pointer to in-core inode requiring allocation.
973  *	blkno	    -  starting block of the current allocation.
974  *	nblocks	    -  number of contiguous blocks within the current
975  *		       allocation.
976  *	addnblocks  -  number of blocks to add to the allocation.
977  *
978  * RETURN VALUES:
979  *	0	- success
980  *	-ENOSPC	- insufficient disk resources
981  *	-EIO	- i/o error
982  */
983 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks)
984 {
985 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
986 	s64 lblkno, lastblkno, extblkno;
987 	uint rel_block;
988 	struct metapage *mp;
989 	struct dmap *dp;
990 	int rc;
991 	struct inode *ipbmap = sbi->ipbmap;
992 	struct bmap *bmp;
993 
994 	/*
995 	 * We don't want a non-aligned extent to cross a page boundary
996 	 */
997 	if (((rel_block = blkno & (sbi->nbperpage - 1))) &&
998 	    (rel_block + nblocks + addnblocks > sbi->nbperpage))
999 		return -ENOSPC;
1000 
1001 	/* get the last block of the current allocation */
1002 	lastblkno = blkno + nblocks - 1;
1003 
1004 	/* determine the block number of the block following
1005 	 * the existing allocation.
1006 	 */
1007 	extblkno = lastblkno + 1;
1008 
1009 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
1010 
1011 	/* better be within the file system */
1012 	bmp = sbi->bmap;
1013 	if (lastblkno < 0 || lastblkno >= bmp->db_mapsize) {
1014 		IREAD_UNLOCK(ipbmap);
1015 		jfs_error(ip->i_sb, "the block is outside the filesystem\n");
1016 		return -EIO;
1017 	}
1018 
1019 	/* we'll attempt to extend the current allocation in place by
1020 	 * allocating the additional blocks as the blocks immediately
1021 	 * following the current allocation.  we only try to extend the
1022 	 * current allocation in place if the number of additional blocks
1023 	 * can fit into a dmap, the last block of the current allocation
1024 	 * is not the last block of the file system, and the start of the
1025 	 * inplace extension is not on an allocation group boundary.
1026 	 */
1027 	if (addnblocks > BPERDMAP || extblkno >= bmp->db_mapsize ||
1028 	    (extblkno & (bmp->db_agsize - 1)) == 0) {
1029 		IREAD_UNLOCK(ipbmap);
1030 		return -ENOSPC;
1031 	}
1032 
1033 	/* get the buffer for the dmap containing the first block
1034 	 * of the extension.
1035 	 */
1036 	lblkno = BLKTODMAP(extblkno, bmp->db_l2nbperpage);
1037 	mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
1038 	if (mp == NULL) {
1039 		IREAD_UNLOCK(ipbmap);
1040 		return -EIO;
1041 	}
1042 
1043 	dp = (struct dmap *) mp->data;
1044 
1045 	/* try to allocate the blocks immediately following the
1046 	 * current allocation.
1047 	 */
1048 	rc = dbAllocNext(bmp, dp, extblkno, (int) addnblocks);
1049 
1050 	IREAD_UNLOCK(ipbmap);
1051 
1052 	/* were we successful ? */
1053 	if (rc == 0)
1054 		write_metapage(mp);
1055 	else
1056 		/* we were not successful */
1057 		release_metapage(mp);
1058 
1059 	return (rc);
1060 }
1061 
1062 
1063 /*
1064  * NAME:	dbAllocNext()
1065  *
1066  * FUNCTION:	attempt to allocate the blocks of the specified block
1067  *		range within a dmap.
1068  *
1069  * PARAMETERS:
1070  *	bmp	-  pointer to bmap descriptor
1071  *	dp	-  pointer to dmap.
1072  *	blkno	-  starting block number of the range.
1073  *	nblocks	-  number of contiguous free blocks of the range.
1074  *
1075  * RETURN VALUES:
1076  *	0	- success
1077  *	-ENOSPC	- insufficient disk resources
1078  *	-EIO	- i/o error
1079  *
1080  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1081  */
1082 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
1083 		       int nblocks)
1084 {
1085 	int dbitno, word, rembits, nb, nwords, wbitno, nw;
1086 	int l2size;
1087 	s8 *leaf;
1088 	u32 mask;
1089 
1090 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1091 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n");
1092 		return -EIO;
1093 	}
1094 
1095 	/* pick up a pointer to the leaves of the dmap tree.
1096 	 */
1097 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1098 
1099 	/* determine the bit number and word within the dmap of the
1100 	 * starting block.
1101 	 */
1102 	dbitno = blkno & (BPERDMAP - 1);
1103 	word = dbitno >> L2DBWORD;
1104 
1105 	/* check if the specified block range is contained within
1106 	 * this dmap.
1107 	 */
1108 	if (dbitno + nblocks > BPERDMAP)
1109 		return -ENOSPC;
1110 
1111 	/* check if the starting leaf indicates that anything
1112 	 * is free.
1113 	 */
1114 	if (leaf[word] == NOFREE)
1115 		return -ENOSPC;
1116 
1117 	/* check the dmaps words corresponding to block range to see
1118 	 * if the block range is free.  not all bits of the first and
1119 	 * last words may be contained within the block range.  if this
1120 	 * is the case, we'll work against those words (i.e. partial first
1121 	 * and/or last) on an individual basis (a single pass) and examine
1122 	 * the actual bits to determine if they are free.  a single pass
1123 	 * will be used for all dmap words fully contained within the
1124 	 * specified range.  within this pass, the leaves of the dmap
1125 	 * tree will be examined to determine if the blocks are free. a
1126 	 * single leaf may describe the free space of multiple dmap
1127 	 * words, so we may visit only a subset of the actual leaves
1128 	 * corresponding to the dmap words of the block range.
1129 	 */
1130 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
1131 		/* determine the bit number within the word and
1132 		 * the number of bits within the word.
1133 		 */
1134 		wbitno = dbitno & (DBWORD - 1);
1135 		nb = min(rembits, DBWORD - wbitno);
1136 
1137 		/* check if only part of the word is to be examined.
1138 		 */
1139 		if (nb < DBWORD) {
1140 			/* check if the bits are free.
1141 			 */
1142 			mask = (ONES << (DBWORD - nb) >> wbitno);
1143 			if ((mask & ~le32_to_cpu(dp->wmap[word])) != mask)
1144 				return -ENOSPC;
1145 
1146 			word += 1;
1147 		} else {
1148 			/* one or more dmap words are fully contained
1149 			 * within the block range.  determine how many
1150 			 * words and how many bits.
1151 			 */
1152 			nwords = rembits >> L2DBWORD;
1153 			nb = nwords << L2DBWORD;
1154 
1155 			/* now examine the appropriate leaves to determine
1156 			 * if the blocks are free.
1157 			 */
1158 			while (nwords > 0) {
1159 				/* does the leaf describe any free space ?
1160 				 */
1161 				if (leaf[word] < BUDMIN)
1162 					return -ENOSPC;
1163 
1164 				/* determine the l2 number of bits provided
1165 				 * by this leaf.
1166 				 */
1167 				l2size =
1168 				    min_t(int, leaf[word], NLSTOL2BSZ(nwords));
1169 
1170 				/* determine how many words were handled.
1171 				 */
1172 				nw = BUDSIZE(l2size, BUDMIN);
1173 
1174 				nwords -= nw;
1175 				word += nw;
1176 			}
1177 		}
1178 	}
1179 
1180 	/* allocate the blocks.
1181 	 */
1182 	return (dbAllocDmap(bmp, dp, blkno, nblocks));
1183 }
1184 
1185 
1186 /*
1187  * NAME:	dbAllocNear()
1188  *
1189  * FUNCTION:	attempt to allocate a number of contiguous free blocks near
1190  *		a specified block (hint) within a dmap.
1191  *
1192  *		starting with the dmap leaf that covers the hint, we'll
1193  *		check the next four contiguous leaves for sufficient free
1194  *		space.  if sufficient free space is found, we'll allocate
1195  *		the desired free space.
1196  *
1197  * PARAMETERS:
1198  *	bmp	-  pointer to bmap descriptor
1199  *	dp	-  pointer to dmap.
1200  *	blkno	-  block number to allocate near.
1201  *	nblocks	-  actual number of contiguous free blocks desired.
1202  *	l2nb	-  log2 number of contiguous free blocks desired.
1203  *	results	-  on successful return, set to the starting block number
1204  *		   of the newly allocated range.
1205  *
1206  * RETURN VALUES:
1207  *	0	- success
1208  *	-ENOSPC	- insufficient disk resources
1209  *	-EIO	- i/o error
1210  *
1211  * serialization: IREAD_LOCK(ipbmap) held on entry/exit;
1212  */
1213 static int
1214 dbAllocNear(struct bmap * bmp,
1215 	    struct dmap * dp, s64 blkno, int nblocks, int l2nb, s64 * results)
1216 {
1217 	int word, lword, rc;
1218 	s8 *leaf;
1219 
1220 	if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) {
1221 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n");
1222 		return -EIO;
1223 	}
1224 
1225 	leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx);
1226 
1227 	/* determine the word within the dmap that holds the hint
1228 	 * (i.e. blkno).  also, determine the last word in the dmap
1229 	 * that we'll include in our examination.
1230 	 */
1231 	word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
1232 	lword = min(word + 4, LPERDMAP);
1233 
1234 	/* examine the leaves for sufficient free space.
1235 	 */
1236 	for (; word < lword; word++) {
1237 		/* does the leaf describe sufficient free space ?
1238 		 */
1239 		if (leaf[word] < l2nb)
1240 			continue;
1241 
1242 		/* determine the block number within the file system
1243 		 * of the first block described by this dmap word.
1244 		 */
1245 		blkno = le64_to_cpu(dp->start) + (word << L2DBWORD);
1246 
1247 		/* if not all bits of the dmap word are free, get the
1248 		 * starting bit number within the dmap word of the required
1249 		 * string of free bits and adjust the block number with the
1250 		 * value.
1251 		 */
1252 		if (leaf[word] < BUDMIN)
1253 			blkno +=
1254 			    dbFindBits(le32_to_cpu(dp->wmap[word]), l2nb);
1255 
1256 		/* allocate the blocks.
1257 		 */
1258 		if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1259 			*results = blkno;
1260 
1261 		return (rc);
1262 	}
1263 
1264 	return -ENOSPC;
1265 }
1266 
1267 
1268 /*
1269  * NAME:	dbAllocAG()
1270  *
1271  * FUNCTION:	attempt to allocate the specified number of contiguous
1272  *		free blocks within the specified allocation group.
1273  *
1274  *		unless the allocation group size is equal to the number
1275  *		of blocks per dmap, the dmap control pages will be used to
1276  *		find the required free space, if available.  we start the
1277  *		search at the highest dmap control page level which
1278  *		distinctly describes the allocation group's free space
1279  *		(i.e. the highest level at which the allocation group's
1280  *		free space is not mixed in with that of any other group).
1281  *		in addition, we start the search within this level at a
1282  *		height of the dmapctl dmtree at which the nodes distinctly
1283  *		describe the allocation group's free space.  at this height,
1284  *		the allocation group's free space may be represented by 1
1285  *		or two sub-trees, depending on the allocation group size.
1286  *		we search the top nodes of these subtrees left to right for
1287  *		sufficient free space.  if sufficient free space is found,
1288  *		the subtree is searched to find the leftmost leaf that
1289  *		has free space.  once we have made it to the leaf, we
1290  *		move the search to the next lower level dmap control page
1291  *		corresponding to this leaf.  we continue down the dmap control
1292  *		pages until we find the dmap that contains or starts the
1293  *		sufficient free space and we allocate at this dmap.
1294  *
1295  *		if the allocation group size is equal to the dmap size,
1296  *		we'll start at the dmap corresponding to the allocation
1297  *		group and attempt the allocation at this level.
1298  *
1299  *		the dmap control page search is also not performed if the
1300  *		allocation group is completely free and we go to the first
1301  *		dmap of the allocation group to do the allocation.  this is
1302  *		done because the allocation group may be part (not the first
1303  *		part) of a larger binary buddy system, causing the dmap
1304  *		control pages to indicate no free space (NOFREE) within
1305  *		the allocation group.
1306  *
1307  * PARAMETERS:
1308  *	bmp	-  pointer to bmap descriptor
1309  *	agno	- allocation group number.
1310  *	nblocks	-  actual number of contiguous free blocks desired.
1311  *	l2nb	-  log2 number of contiguous free blocks desired.
1312  *	results	-  on successful return, set to the starting block number
1313  *		   of the newly allocated range.
1314  *
1315  * RETURN VALUES:
1316  *	0	- success
1317  *	-ENOSPC	- insufficient disk resources
1318  *	-EIO	- i/o error
1319  *
1320  * note: IWRITE_LOCK(ipmap) held on entry/exit;
1321  */
1322 static int
1323 dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results)
1324 {
1325 	struct metapage *mp;
1326 	struct dmapctl *dcp;
1327 	int rc, ti, i, k, m, n, agperlev;
1328 	s64 blkno, lblkno;
1329 	int budmin;
1330 
1331 	/* allocation request should not be for more than the
1332 	 * allocation group size.
1333 	 */
1334 	if (l2nb > bmp->db_agl2size) {
1335 		jfs_error(bmp->db_ipbmap->i_sb,
1336 			  "allocation request is larger than the allocation group size\n");
1337 		return -EIO;
1338 	}
1339 
1340 	/* determine the starting block number of the allocation
1341 	 * group.
1342 	 */
1343 	blkno = (s64) agno << bmp->db_agl2size;
1344 
1345 	/* check if the allocation group size is the minimum allocation
1346 	 * group size or if the allocation group is completely free. if
1347 	 * the allocation group size is the minimum size of BPERDMAP (i.e.
1348 	 * 1 dmap), there is no need to search the dmap control page (below)
1349 	 * that fully describes the allocation group since the allocation
1350 	 * group is already fully described by a dmap.  in this case, we
1351 	 * just call dbAllocCtl() to search the dmap tree and allocate the
1352 	 * required space if available.
1353 	 *
1354 	 * if the allocation group is completely free, dbAllocCtl() is
1355 	 * also called to allocate the required space.  this is done for
1356 	 * two reasons.  first, it makes no sense searching the dmap control
1357 	 * pages for free space when we know that free space exists.  second,
1358 	 * the dmap control pages may indicate that the allocation group
1359 	 * has no free space if the allocation group is part (not the first
1360 	 * part) of a larger binary buddy system.
1361 	 */
1362 	if (bmp->db_agsize == BPERDMAP
1363 	    || bmp->db_agfree[agno] == bmp->db_agsize) {
1364 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1365 		if ((rc == -ENOSPC) &&
1366 		    (bmp->db_agfree[agno] == bmp->db_agsize)) {
1367 			printk(KERN_ERR "blkno = %Lx, blocks = %Lx\n",
1368 			       (unsigned long long) blkno,
1369 			       (unsigned long long) nblocks);
1370 			jfs_error(bmp->db_ipbmap->i_sb,
1371 				  "dbAllocCtl failed in free AG\n");
1372 		}
1373 		return (rc);
1374 	}
1375 
1376 	/* the buffer for the dmap control page that fully describes the
1377 	 * allocation group.
1378 	 */
1379 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, bmp->db_aglevel);
1380 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1381 	if (mp == NULL)
1382 		return -EIO;
1383 	dcp = (struct dmapctl *) mp->data;
1384 	budmin = dcp->budmin;
1385 
1386 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1387 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n");
1388 		release_metapage(mp);
1389 		return -EIO;
1390 	}
1391 
1392 	/* search the subtree(s) of the dmap control page that describes
1393 	 * the allocation group, looking for sufficient free space.  to begin,
1394 	 * determine how many allocation groups are represented in a dmap
1395 	 * control page at the control page level (i.e. L0, L1, L2) that
1396 	 * fully describes an allocation group. next, determine the starting
1397 	 * tree index of this allocation group within the control page.
1398 	 */
1399 	agperlev =
1400 	    (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp->db_agwidth;
1401 	ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
1402 
1403 	/* dmap control page trees fan-out by 4 and a single allocation
1404 	 * group may be described by 1 or 2 subtrees within the ag level
1405 	 * dmap control page, depending upon the ag size. examine the ag's
1406 	 * subtrees for sufficient free space, starting with the leftmost
1407 	 * subtree.
1408 	 */
1409 	for (i = 0; i < bmp->db_agwidth; i++, ti++) {
1410 		/* is there sufficient free space ?
1411 		 */
1412 		if (l2nb > dcp->stree[ti])
1413 			continue;
1414 
1415 		/* sufficient free space found in a subtree. now search down
1416 		 * the subtree to find the leftmost leaf that describes this
1417 		 * free space.
1418 		 */
1419 		for (k = bmp->db_agheight; k > 0; k--) {
1420 			for (n = 0, m = (ti << 2) + 1; n < 4; n++) {
1421 				if (l2nb <= dcp->stree[m + n]) {
1422 					ti = m + n;
1423 					break;
1424 				}
1425 			}
1426 			if (n == 4) {
1427 				jfs_error(bmp->db_ipbmap->i_sb,
1428 					  "failed descending stree\n");
1429 				release_metapage(mp);
1430 				return -EIO;
1431 			}
1432 		}
1433 
1434 		/* determine the block number within the file system
1435 		 * that corresponds to this leaf.
1436 		 */
1437 		if (bmp->db_aglevel == 2)
1438 			blkno = 0;
1439 		else if (bmp->db_aglevel == 1)
1440 			blkno &= ~(MAXL1SIZE - 1);
1441 		else		/* bmp->db_aglevel == 0 */
1442 			blkno &= ~(MAXL0SIZE - 1);
1443 
1444 		blkno +=
1445 		    ((s64) (ti - le32_to_cpu(dcp->leafidx))) << budmin;
1446 
1447 		/* release the buffer in preparation for going down
1448 		 * the next level of dmap control pages.
1449 		 */
1450 		release_metapage(mp);
1451 
1452 		/* check if we need to continue to search down the lower
1453 		 * level dmap control pages.  we need to if the number of
1454 		 * blocks required is less than maximum number of blocks
1455 		 * described at the next lower level.
1456 		 */
1457 		if (l2nb < budmin) {
1458 
1459 			/* search the lower level dmap control pages to get
1460 			 * the starting block number of the dmap that
1461 			 * contains or starts off the free space.
1462 			 */
1463 			if ((rc =
1464 			     dbFindCtl(bmp, l2nb, bmp->db_aglevel - 1,
1465 				       &blkno))) {
1466 				if (rc == -ENOSPC) {
1467 					jfs_error(bmp->db_ipbmap->i_sb,
1468 						  "control page inconsistent\n");
1469 					return -EIO;
1470 				}
1471 				return (rc);
1472 			}
1473 		}
1474 
1475 		/* allocate the blocks.
1476 		 */
1477 		rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1478 		if (rc == -ENOSPC) {
1479 			jfs_error(bmp->db_ipbmap->i_sb,
1480 				  "unable to allocate blocks\n");
1481 			rc = -EIO;
1482 		}
1483 		return (rc);
1484 	}
1485 
1486 	/* no space in the allocation group.  release the buffer and
1487 	 * return -ENOSPC.
1488 	 */
1489 	release_metapage(mp);
1490 
1491 	return -ENOSPC;
1492 }
1493 
1494 
1495 /*
1496  * NAME:	dbAllocAny()
1497  *
1498  * FUNCTION:	attempt to allocate the specified number of contiguous
1499  *		free blocks anywhere in the file system.
1500  *
1501  *		dbAllocAny() attempts to find the sufficient free space by
1502  *		searching down the dmap control pages, starting with the
1503  *		highest level (i.e. L0, L1, L2) control page.  if free space
1504  *		large enough to satisfy the desired free space is found, the
1505  *		desired free space is allocated.
1506  *
1507  * PARAMETERS:
1508  *	bmp	-  pointer to bmap descriptor
1509  *	nblocks	 -  actual number of contiguous free blocks desired.
1510  *	l2nb	 -  log2 number of contiguous free blocks desired.
1511  *	results	-  on successful return, set to the starting block number
1512  *		   of the newly allocated range.
1513  *
1514  * RETURN VALUES:
1515  *	0	- success
1516  *	-ENOSPC	- insufficient disk resources
1517  *	-EIO	- i/o error
1518  *
1519  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1520  */
1521 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results)
1522 {
1523 	int rc;
1524 	s64 blkno = 0;
1525 
1526 	/* starting with the top level dmap control page, search
1527 	 * down the dmap control levels for sufficient free space.
1528 	 * if free space is found, dbFindCtl() returns the starting
1529 	 * block number of the dmap that contains or starts off the
1530 	 * range of free space.
1531 	 */
1532 	if ((rc = dbFindCtl(bmp, l2nb, bmp->db_maxlevel, &blkno)))
1533 		return (rc);
1534 
1535 	/* allocate the blocks.
1536 	 */
1537 	rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results);
1538 	if (rc == -ENOSPC) {
1539 		jfs_error(bmp->db_ipbmap->i_sb, "unable to allocate blocks\n");
1540 		return -EIO;
1541 	}
1542 	return (rc);
1543 }
1544 
1545 
1546 /*
1547  * NAME:	dbDiscardAG()
1548  *
1549  * FUNCTION:	attempt to discard (TRIM) all free blocks of specific AG
1550  *
1551  *		algorithm:
1552  *		1) allocate blocks, as large as possible and save them
1553  *		   while holding IWRITE_LOCK on ipbmap
1554  *		2) trim all these saved block/length values
1555  *		3) mark the blocks free again
1556  *
1557  *		benefit:
1558  *		- we work only on one ag at some time, minimizing how long we
1559  *		  need to lock ipbmap
1560  *		- reading / writing the fs is possible most time, even on
1561  *		  trimming
1562  *
1563  *		downside:
1564  *		- we write two times to the dmapctl and dmap pages
1565  *		- but for me, this seems the best way, better ideas?
1566  *		/TR 2012
1567  *
1568  * PARAMETERS:
1569  *	ip	- pointer to in-core inode
1570  *	agno	- ag to trim
1571  *	minlen	- minimum value of contiguous blocks
1572  *
1573  * RETURN VALUES:
1574  *	s64	- actual number of blocks trimmed
1575  */
1576 s64 dbDiscardAG(struct inode *ip, int agno, s64 minlen)
1577 {
1578 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
1579 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
1580 	s64 nblocks, blkno;
1581 	u64 trimmed = 0;
1582 	int rc, l2nb;
1583 	struct super_block *sb = ipbmap->i_sb;
1584 
1585 	struct range2trim {
1586 		u64 blkno;
1587 		u64 nblocks;
1588 	} *totrim, *tt;
1589 
1590 	/* max blkno / nblocks pairs to trim */
1591 	int count = 0, range_cnt;
1592 	u64 max_ranges;
1593 
1594 	/* prevent others from writing new stuff here, while trimming */
1595 	IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
1596 
1597 	nblocks = bmp->db_agfree[agno];
1598 	max_ranges = nblocks;
1599 	do_div(max_ranges, minlen);
1600 	range_cnt = min_t(u64, max_ranges + 1, 32 * 1024);
1601 	totrim = kmalloc_array(range_cnt, sizeof(struct range2trim), GFP_NOFS);
1602 	if (totrim == NULL) {
1603 		jfs_error(bmp->db_ipbmap->i_sb, "no memory for trim array\n");
1604 		IWRITE_UNLOCK(ipbmap);
1605 		return 0;
1606 	}
1607 
1608 	tt = totrim;
1609 	while (nblocks >= minlen) {
1610 		l2nb = BLKSTOL2(nblocks);
1611 
1612 		/* 0 = okay, -EIO = fatal, -ENOSPC -> try smaller block */
1613 		rc = dbAllocAG(bmp, agno, nblocks, l2nb, &blkno);
1614 		if (rc == 0) {
1615 			tt->blkno = blkno;
1616 			tt->nblocks = nblocks;
1617 			tt++; count++;
1618 
1619 			/* the whole ag is free, trim now */
1620 			if (bmp->db_agfree[agno] == 0)
1621 				break;
1622 
1623 			/* give a hint for the next while */
1624 			nblocks = bmp->db_agfree[agno];
1625 			continue;
1626 		} else if (rc == -ENOSPC) {
1627 			/* search for next smaller log2 block */
1628 			l2nb = BLKSTOL2(nblocks) - 1;
1629 			nblocks = 1LL << l2nb;
1630 		} else {
1631 			/* Trim any already allocated blocks */
1632 			jfs_error(bmp->db_ipbmap->i_sb, "-EIO\n");
1633 			break;
1634 		}
1635 
1636 		/* check, if our trim array is full */
1637 		if (unlikely(count >= range_cnt - 1))
1638 			break;
1639 	}
1640 	IWRITE_UNLOCK(ipbmap);
1641 
1642 	tt->nblocks = 0; /* mark the current end */
1643 	for (tt = totrim; tt->nblocks != 0; tt++) {
1644 		/* when mounted with online discard, dbFree() will
1645 		 * call jfs_issue_discard() itself */
1646 		if (!(JFS_SBI(sb)->flag & JFS_DISCARD))
1647 			jfs_issue_discard(ip, tt->blkno, tt->nblocks);
1648 		dbFree(ip, tt->blkno, tt->nblocks);
1649 		trimmed += tt->nblocks;
1650 	}
1651 	kfree(totrim);
1652 
1653 	return trimmed;
1654 }
1655 
1656 /*
1657  * NAME:	dbFindCtl()
1658  *
1659  * FUNCTION:	starting at a specified dmap control page level and block
1660  *		number, search down the dmap control levels for a range of
1661  *		contiguous free blocks large enough to satisfy an allocation
1662  *		request for the specified number of free blocks.
1663  *
1664  *		if sufficient contiguous free blocks are found, this routine
1665  *		returns the starting block number within a dmap page that
1666  *		contains or starts a range of contiqious free blocks that
1667  *		is sufficient in size.
1668  *
1669  * PARAMETERS:
1670  *	bmp	-  pointer to bmap descriptor
1671  *	level	-  starting dmap control page level.
1672  *	l2nb	-  log2 number of contiguous free blocks desired.
1673  *	*blkno	-  on entry, starting block number for conducting the search.
1674  *		   on successful return, the first block within a dmap page
1675  *		   that contains or starts a range of contiguous free blocks.
1676  *
1677  * RETURN VALUES:
1678  *	0	- success
1679  *	-ENOSPC	- insufficient disk resources
1680  *	-EIO	- i/o error
1681  *
1682  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1683  */
1684 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno)
1685 {
1686 	int rc, leafidx, lev;
1687 	s64 b, lblkno;
1688 	struct dmapctl *dcp;
1689 	int budmin;
1690 	struct metapage *mp;
1691 
1692 	/* starting at the specified dmap control page level and block
1693 	 * number, search down the dmap control levels for the starting
1694 	 * block number of a dmap page that contains or starts off
1695 	 * sufficient free blocks.
1696 	 */
1697 	for (lev = level, b = *blkno; lev >= 0; lev--) {
1698 		/* get the buffer of the dmap control page for the block
1699 		 * number and level (i.e. L0, L1, L2).
1700 		 */
1701 		lblkno = BLKTOCTL(b, bmp->db_l2nbperpage, lev);
1702 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1703 		if (mp == NULL)
1704 			return -EIO;
1705 		dcp = (struct dmapctl *) mp->data;
1706 		budmin = dcp->budmin;
1707 
1708 		if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
1709 			jfs_error(bmp->db_ipbmap->i_sb,
1710 				  "Corrupt dmapctl page\n");
1711 			release_metapage(mp);
1712 			return -EIO;
1713 		}
1714 
1715 		/* search the tree within the dmap control page for
1716 		 * sufficient free space.  if sufficient free space is found,
1717 		 * dbFindLeaf() returns the index of the leaf at which
1718 		 * free space was found.
1719 		 */
1720 		rc = dbFindLeaf((dmtree_t *) dcp, l2nb, &leafidx, true);
1721 
1722 		/* release the buffer.
1723 		 */
1724 		release_metapage(mp);
1725 
1726 		/* space found ?
1727 		 */
1728 		if (rc) {
1729 			if (lev != level) {
1730 				jfs_error(bmp->db_ipbmap->i_sb,
1731 					  "dmap inconsistent\n");
1732 				return -EIO;
1733 			}
1734 			return -ENOSPC;
1735 		}
1736 
1737 		/* adjust the block number to reflect the location within
1738 		 * the dmap control page (i.e. the leaf) at which free
1739 		 * space was found.
1740 		 */
1741 		b += (((s64) leafidx) << budmin);
1742 
1743 		/* we stop the search at this dmap control page level if
1744 		 * the number of blocks required is greater than or equal
1745 		 * to the maximum number of blocks described at the next
1746 		 * (lower) level.
1747 		 */
1748 		if (l2nb >= budmin)
1749 			break;
1750 	}
1751 
1752 	*blkno = b;
1753 	return (0);
1754 }
1755 
1756 
1757 /*
1758  * NAME:	dbAllocCtl()
1759  *
1760  * FUNCTION:	attempt to allocate a specified number of contiguous
1761  *		blocks starting within a specific dmap.
1762  *
1763  *		this routine is called by higher level routines that search
1764  *		the dmap control pages above the actual dmaps for contiguous
1765  *		free space.  the result of successful searches by these
1766  *		routines are the starting block numbers within dmaps, with
1767  *		the dmaps themselves containing the desired contiguous free
1768  *		space or starting a contiguous free space of desired size
1769  *		that is made up of the blocks of one or more dmaps. these
1770  *		calls should not fail due to insufficent resources.
1771  *
1772  *		this routine is called in some cases where it is not known
1773  *		whether it will fail due to insufficient resources.  more
1774  *		specifically, this occurs when allocating from an allocation
1775  *		group whose size is equal to the number of blocks per dmap.
1776  *		in this case, the dmap control pages are not examined prior
1777  *		to calling this routine (to save pathlength) and the call
1778  *		might fail.
1779  *
1780  *		for a request size that fits within a dmap, this routine relies
1781  *		upon the dmap's dmtree to find the requested contiguous free
1782  *		space.  for request sizes that are larger than a dmap, the
1783  *		requested free space will start at the first block of the
1784  *		first dmap (i.e. blkno).
1785  *
1786  * PARAMETERS:
1787  *	bmp	-  pointer to bmap descriptor
1788  *	nblocks	 -  actual number of contiguous free blocks to allocate.
1789  *	l2nb	 -  log2 number of contiguous free blocks to allocate.
1790  *	blkno	 -  starting block number of the dmap to start the allocation
1791  *		    from.
1792  *	results	-  on successful return, set to the starting block number
1793  *		   of the newly allocated range.
1794  *
1795  * RETURN VALUES:
1796  *	0	- success
1797  *	-ENOSPC	- insufficient disk resources
1798  *	-EIO	- i/o error
1799  *
1800  * serialization: IWRITE_LOCK(ipbmap) held on entry/exit;
1801  */
1802 static int
1803 dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, s64 * results)
1804 {
1805 	int rc, nb;
1806 	s64 b, lblkno, n;
1807 	struct metapage *mp;
1808 	struct dmap *dp;
1809 
1810 	/* check if the allocation request is confined to a single dmap.
1811 	 */
1812 	if (l2nb <= L2BPERDMAP) {
1813 		/* get the buffer for the dmap.
1814 		 */
1815 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
1816 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1817 		if (mp == NULL)
1818 			return -EIO;
1819 		dp = (struct dmap *) mp->data;
1820 
1821 		/* try to allocate the blocks.
1822 		 */
1823 		rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results);
1824 		if (rc == 0)
1825 			mark_metapage_dirty(mp);
1826 
1827 		release_metapage(mp);
1828 
1829 		return (rc);
1830 	}
1831 
1832 	/* allocation request involving multiple dmaps. it must start on
1833 	 * a dmap boundary.
1834 	 */
1835 	assert((blkno & (BPERDMAP - 1)) == 0);
1836 
1837 	/* allocate the blocks dmap by dmap.
1838 	 */
1839 	for (n = nblocks, b = blkno; n > 0; n -= nb, b += nb) {
1840 		/* get the buffer for the dmap.
1841 		 */
1842 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1843 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1844 		if (mp == NULL) {
1845 			rc = -EIO;
1846 			goto backout;
1847 		}
1848 		dp = (struct dmap *) mp->data;
1849 
1850 		/* the dmap better be all free.
1851 		 */
1852 		if (dp->tree.stree[ROOT] != L2BPERDMAP) {
1853 			release_metapage(mp);
1854 			jfs_error(bmp->db_ipbmap->i_sb,
1855 				  "the dmap is not all free\n");
1856 			rc = -EIO;
1857 			goto backout;
1858 		}
1859 
1860 		/* determine how many blocks to allocate from this dmap.
1861 		 */
1862 		nb = min_t(s64, n, BPERDMAP);
1863 
1864 		/* allocate the blocks from the dmap.
1865 		 */
1866 		if ((rc = dbAllocDmap(bmp, dp, b, nb))) {
1867 			release_metapage(mp);
1868 			goto backout;
1869 		}
1870 
1871 		/* write the buffer.
1872 		 */
1873 		write_metapage(mp);
1874 	}
1875 
1876 	/* set the results (starting block number) and return.
1877 	 */
1878 	*results = blkno;
1879 	return (0);
1880 
1881 	/* something failed in handling an allocation request involving
1882 	 * multiple dmaps.  we'll try to clean up by backing out any
1883 	 * allocation that has already happened for this request.  if
1884 	 * we fail in backing out the allocation, we'll mark the file
1885 	 * system to indicate that blocks have been leaked.
1886 	 */
1887       backout:
1888 
1889 	/* try to backout the allocations dmap by dmap.
1890 	 */
1891 	for (n = nblocks - n, b = blkno; n > 0;
1892 	     n -= BPERDMAP, b += BPERDMAP) {
1893 		/* get the buffer for this dmap.
1894 		 */
1895 		lblkno = BLKTODMAP(b, bmp->db_l2nbperpage);
1896 		mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
1897 		if (mp == NULL) {
1898 			/* could not back out.  mark the file system
1899 			 * to indicate that we have leaked blocks.
1900 			 */
1901 			jfs_error(bmp->db_ipbmap->i_sb,
1902 				  "I/O Error: Block Leakage\n");
1903 			continue;
1904 		}
1905 		dp = (struct dmap *) mp->data;
1906 
1907 		/* free the blocks is this dmap.
1908 		 */
1909 		if (dbFreeDmap(bmp, dp, b, BPERDMAP)) {
1910 			/* could not back out.  mark the file system
1911 			 * to indicate that we have leaked blocks.
1912 			 */
1913 			release_metapage(mp);
1914 			jfs_error(bmp->db_ipbmap->i_sb, "Block Leakage\n");
1915 			continue;
1916 		}
1917 
1918 		/* write the buffer.
1919 		 */
1920 		write_metapage(mp);
1921 	}
1922 
1923 	return (rc);
1924 }
1925 
1926 
1927 /*
1928  * NAME:	dbAllocDmapLev()
1929  *
1930  * FUNCTION:	attempt to allocate a specified number of contiguous blocks
1931  *		from a specified dmap.
1932  *
1933  *		this routine checks if the contiguous blocks are available.
1934  *		if so, nblocks of blocks are allocated; otherwise, ENOSPC is
1935  *		returned.
1936  *
1937  * PARAMETERS:
1938  *	mp	-  pointer to bmap descriptor
1939  *	dp	-  pointer to dmap to attempt to allocate blocks from.
1940  *	l2nb	-  log2 number of contiguous block desired.
1941  *	nblocks	-  actual number of contiguous block desired.
1942  *	results	-  on successful return, set to the starting block number
1943  *		   of the newly allocated range.
1944  *
1945  * RETURN VALUES:
1946  *	0	- success
1947  *	-ENOSPC	- insufficient disk resources
1948  *	-EIO	- i/o error
1949  *
1950  * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or
1951  *	IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit;
1952  */
1953 static int
1954 dbAllocDmapLev(struct bmap * bmp,
1955 	       struct dmap * dp, int nblocks, int l2nb, s64 * results)
1956 {
1957 	s64 blkno;
1958 	int leafidx, rc;
1959 
1960 	/* can't be more than a dmaps worth of blocks */
1961 	assert(l2nb <= L2BPERDMAP);
1962 
1963 	/* search the tree within the dmap page for sufficient
1964 	 * free space.  if sufficient free space is found, dbFindLeaf()
1965 	 * returns the index of the leaf at which free space was found.
1966 	 */
1967 	if (dbFindLeaf((dmtree_t *) &dp->tree, l2nb, &leafidx, false))
1968 		return -ENOSPC;
1969 
1970 	if (leafidx < 0)
1971 		return -EIO;
1972 
1973 	/* determine the block number within the file system corresponding
1974 	 * to the leaf at which free space was found.
1975 	 */
1976 	blkno = le64_to_cpu(dp->start) + (leafidx << L2DBWORD);
1977 
1978 	/* if not all bits of the dmap word are free, get the starting
1979 	 * bit number within the dmap word of the required string of free
1980 	 * bits and adjust the block number with this value.
1981 	 */
1982 	if (dp->tree.stree[leafidx + LEAFIND] < BUDMIN)
1983 		blkno += dbFindBits(le32_to_cpu(dp->wmap[leafidx]), l2nb);
1984 
1985 	/* allocate the blocks */
1986 	if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0)
1987 		*results = blkno;
1988 
1989 	return (rc);
1990 }
1991 
1992 
1993 /*
1994  * NAME:	dbAllocDmap()
1995  *
1996  * FUNCTION:	adjust the disk allocation map to reflect the allocation
1997  *		of a specified block range within a dmap.
1998  *
1999  *		this routine allocates the specified blocks from the dmap
2000  *		through a call to dbAllocBits(). if the allocation of the
2001  *		block range causes the maximum string of free blocks within
2002  *		the dmap to change (i.e. the value of the root of the dmap's
2003  *		dmtree), this routine will cause this change to be reflected
2004  *		up through the appropriate levels of the dmap control pages
2005  *		by a call to dbAdjCtl() for the L0 dmap control page that
2006  *		covers this dmap.
2007  *
2008  * PARAMETERS:
2009  *	bmp	-  pointer to bmap descriptor
2010  *	dp	-  pointer to dmap to allocate the block range from.
2011  *	blkno	-  starting block number of the block to be allocated.
2012  *	nblocks	-  number of blocks to be allocated.
2013  *
2014  * RETURN VALUES:
2015  *	0	- success
2016  *	-EIO	- i/o error
2017  *
2018  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2019  */
2020 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2021 		       int nblocks)
2022 {
2023 	s8 oldroot;
2024 	int rc;
2025 
2026 	/* save the current value of the root (i.e. maximum free string)
2027 	 * of the dmap tree.
2028 	 */
2029 	oldroot = dp->tree.stree[ROOT];
2030 
2031 	/* allocate the specified (blocks) bits */
2032 	dbAllocBits(bmp, dp, blkno, nblocks);
2033 
2034 	/* if the root has not changed, done. */
2035 	if (dp->tree.stree[ROOT] == oldroot)
2036 		return (0);
2037 
2038 	/* root changed. bubble the change up to the dmap control pages.
2039 	 * if the adjustment of the upper level control pages fails,
2040 	 * backout the bit allocation (thus making everything consistent).
2041 	 */
2042 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 1, 0)))
2043 		dbFreeBits(bmp, dp, blkno, nblocks);
2044 
2045 	return (rc);
2046 }
2047 
2048 
2049 /*
2050  * NAME:	dbFreeDmap()
2051  *
2052  * FUNCTION:	adjust the disk allocation map to reflect the allocation
2053  *		of a specified block range within a dmap.
2054  *
2055  *		this routine frees the specified blocks from the dmap through
2056  *		a call to dbFreeBits(). if the deallocation of the block range
2057  *		causes the maximum string of free blocks within the dmap to
2058  *		change (i.e. the value of the root of the dmap's dmtree), this
2059  *		routine will cause this change to be reflected up through the
2060  *		appropriate levels of the dmap control pages by a call to
2061  *		dbAdjCtl() for the L0 dmap control page that covers this dmap.
2062  *
2063  * PARAMETERS:
2064  *	bmp	-  pointer to bmap descriptor
2065  *	dp	-  pointer to dmap to free the block range from.
2066  *	blkno	-  starting block number of the block to be freed.
2067  *	nblocks	-  number of blocks to be freed.
2068  *
2069  * RETURN VALUES:
2070  *	0	- success
2071  *	-EIO	- i/o error
2072  *
2073  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2074  */
2075 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
2076 		      int nblocks)
2077 {
2078 	s8 oldroot;
2079 	int rc = 0, word;
2080 
2081 	/* save the current value of the root (i.e. maximum free string)
2082 	 * of the dmap tree.
2083 	 */
2084 	oldroot = dp->tree.stree[ROOT];
2085 
2086 	/* free the specified (blocks) bits */
2087 	rc = dbFreeBits(bmp, dp, blkno, nblocks);
2088 
2089 	/* if error or the root has not changed, done. */
2090 	if (rc || (dp->tree.stree[ROOT] == oldroot))
2091 		return (rc);
2092 
2093 	/* root changed. bubble the change up to the dmap control pages.
2094 	 * if the adjustment of the upper level control pages fails,
2095 	 * backout the deallocation.
2096 	 */
2097 	if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 0, 0))) {
2098 		word = (blkno & (BPERDMAP - 1)) >> L2DBWORD;
2099 
2100 		/* as part of backing out the deallocation, we will have
2101 		 * to back split the dmap tree if the deallocation caused
2102 		 * the freed blocks to become part of a larger binary buddy
2103 		 * system.
2104 		 */
2105 		if (dp->tree.stree[word] == NOFREE)
2106 			dbBackSplit((dmtree_t *)&dp->tree, word, false);
2107 
2108 		dbAllocBits(bmp, dp, blkno, nblocks);
2109 	}
2110 
2111 	return (rc);
2112 }
2113 
2114 
2115 /*
2116  * NAME:	dbAllocBits()
2117  *
2118  * FUNCTION:	allocate a specified block range from a dmap.
2119  *
2120  *		this routine updates the dmap to reflect the working
2121  *		state allocation of the specified block range. it directly
2122  *		updates the bits of the working map and causes the adjustment
2123  *		of the binary buddy system described by the dmap's dmtree
2124  *		leaves to reflect the bits allocated.  it also causes the
2125  *		dmap's dmtree, as a whole, to reflect the allocated range.
2126  *
2127  * PARAMETERS:
2128  *	bmp	-  pointer to bmap descriptor
2129  *	dp	-  pointer to dmap to allocate bits from.
2130  *	blkno	-  starting block number of the bits to be allocated.
2131  *	nblocks	-  number of bits to be allocated.
2132  *
2133  * RETURN VALUES: none
2134  *
2135  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2136  */
2137 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2138 			int nblocks)
2139 {
2140 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2141 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2142 	int size;
2143 	s8 *leaf;
2144 
2145 	/* pick up a pointer to the leaves of the dmap tree */
2146 	leaf = dp->tree.stree + LEAFIND;
2147 
2148 	/* determine the bit number and word within the dmap of the
2149 	 * starting block.
2150 	 */
2151 	dbitno = blkno & (BPERDMAP - 1);
2152 	word = dbitno >> L2DBWORD;
2153 
2154 	/* block range better be within the dmap */
2155 	assert(dbitno + nblocks <= BPERDMAP);
2156 
2157 	/* allocate the bits of the dmap's words corresponding to the block
2158 	 * range. not all bits of the first and last words may be contained
2159 	 * within the block range.  if this is the case, we'll work against
2160 	 * those words (i.e. partial first and/or last) on an individual basis
2161 	 * (a single pass), allocating the bits of interest by hand and
2162 	 * updating the leaf corresponding to the dmap word. a single pass
2163 	 * will be used for all dmap words fully contained within the
2164 	 * specified range.  within this pass, the bits of all fully contained
2165 	 * dmap words will be marked as free in a single shot and the leaves
2166 	 * will be updated. a single leaf may describe the free space of
2167 	 * multiple dmap words, so we may update only a subset of the actual
2168 	 * leaves corresponding to the dmap words of the block range.
2169 	 */
2170 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2171 		/* determine the bit number within the word and
2172 		 * the number of bits within the word.
2173 		 */
2174 		wbitno = dbitno & (DBWORD - 1);
2175 		nb = min(rembits, DBWORD - wbitno);
2176 
2177 		/* check if only part of a word is to be allocated.
2178 		 */
2179 		if (nb < DBWORD) {
2180 			/* allocate (set to 1) the appropriate bits within
2181 			 * this dmap word.
2182 			 */
2183 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
2184 						      >> wbitno);
2185 
2186 			/* update the leaf for this dmap word. in addition
2187 			 * to setting the leaf value to the binary buddy max
2188 			 * of the updated dmap word, dbSplit() will split
2189 			 * the binary system of the leaves if need be.
2190 			 */
2191 			dbSplit(tp, word, BUDMIN,
2192 				dbMaxBud((u8 *)&dp->wmap[word]), false);
2193 
2194 			word += 1;
2195 		} else {
2196 			/* one or more dmap words are fully contained
2197 			 * within the block range.  determine how many
2198 			 * words and allocate (set to 1) the bits of these
2199 			 * words.
2200 			 */
2201 			nwords = rembits >> L2DBWORD;
2202 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
2203 
2204 			/* determine how many bits.
2205 			 */
2206 			nb = nwords << L2DBWORD;
2207 
2208 			/* now update the appropriate leaves to reflect
2209 			 * the allocated words.
2210 			 */
2211 			for (; nwords > 0; nwords -= nw) {
2212 				if (leaf[word] < BUDMIN) {
2213 					jfs_error(bmp->db_ipbmap->i_sb,
2214 						  "leaf page corrupt\n");
2215 					break;
2216 				}
2217 
2218 				/* determine what the leaf value should be
2219 				 * updated to as the minimum of the l2 number
2220 				 * of bits being allocated and the l2 number
2221 				 * of bits currently described by this leaf.
2222 				 */
2223 				size = min_t(int, leaf[word],
2224 					     NLSTOL2BSZ(nwords));
2225 
2226 				/* update the leaf to reflect the allocation.
2227 				 * in addition to setting the leaf value to
2228 				 * NOFREE, dbSplit() will split the binary
2229 				 * system of the leaves to reflect the current
2230 				 * allocation (size).
2231 				 */
2232 				dbSplit(tp, word, size, NOFREE, false);
2233 
2234 				/* get the number of dmap words handled */
2235 				nw = BUDSIZE(size, BUDMIN);
2236 				word += nw;
2237 			}
2238 		}
2239 	}
2240 
2241 	/* update the free count for this dmap */
2242 	le32_add_cpu(&dp->nfree, -nblocks);
2243 
2244 	BMAP_LOCK(bmp);
2245 
2246 	/* if this allocation group is completely free,
2247 	 * update the maximum allocation group number if this allocation
2248 	 * group is the new max.
2249 	 */
2250 	agno = blkno >> bmp->db_agl2size;
2251 	if (agno > bmp->db_maxag)
2252 		bmp->db_maxag = agno;
2253 
2254 	/* update the free count for the allocation group and map */
2255 	bmp->db_agfree[agno] -= nblocks;
2256 	bmp->db_nfree -= nblocks;
2257 
2258 	BMAP_UNLOCK(bmp);
2259 }
2260 
2261 
2262 /*
2263  * NAME:	dbFreeBits()
2264  *
2265  * FUNCTION:	free a specified block range from a dmap.
2266  *
2267  *		this routine updates the dmap to reflect the working
2268  *		state allocation of the specified block range. it directly
2269  *		updates the bits of the working map and causes the adjustment
2270  *		of the binary buddy system described by the dmap's dmtree
2271  *		leaves to reflect the bits freed.  it also causes the dmap's
2272  *		dmtree, as a whole, to reflect the deallocated range.
2273  *
2274  * PARAMETERS:
2275  *	bmp	-  pointer to bmap descriptor
2276  *	dp	-  pointer to dmap to free bits from.
2277  *	blkno	-  starting block number of the bits to be freed.
2278  *	nblocks	-  number of bits to be freed.
2279  *
2280  * RETURN VALUES: 0 for success
2281  *
2282  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2283  */
2284 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
2285 		       int nblocks)
2286 {
2287 	int dbitno, word, rembits, nb, nwords, wbitno, nw, agno;
2288 	dmtree_t *tp = (dmtree_t *) & dp->tree;
2289 	int rc = 0;
2290 	int size;
2291 
2292 	/* determine the bit number and word within the dmap of the
2293 	 * starting block.
2294 	 */
2295 	dbitno = blkno & (BPERDMAP - 1);
2296 	word = dbitno >> L2DBWORD;
2297 
2298 	/* block range better be within the dmap.
2299 	 */
2300 	assert(dbitno + nblocks <= BPERDMAP);
2301 
2302 	/* free the bits of the dmaps words corresponding to the block range.
2303 	 * not all bits of the first and last words may be contained within
2304 	 * the block range.  if this is the case, we'll work against those
2305 	 * words (i.e. partial first and/or last) on an individual basis
2306 	 * (a single pass), freeing the bits of interest by hand and updating
2307 	 * the leaf corresponding to the dmap word. a single pass will be used
2308 	 * for all dmap words fully contained within the specified range.
2309 	 * within this pass, the bits of all fully contained dmap words will
2310 	 * be marked as free in a single shot and the leaves will be updated. a
2311 	 * single leaf may describe the free space of multiple dmap words,
2312 	 * so we may update only a subset of the actual leaves corresponding
2313 	 * to the dmap words of the block range.
2314 	 *
2315 	 * dbJoin() is used to update leaf values and will join the binary
2316 	 * buddy system of the leaves if the new leaf values indicate this
2317 	 * should be done.
2318 	 */
2319 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
2320 		/* determine the bit number within the word and
2321 		 * the number of bits within the word.
2322 		 */
2323 		wbitno = dbitno & (DBWORD - 1);
2324 		nb = min(rembits, DBWORD - wbitno);
2325 
2326 		/* check if only part of a word is to be freed.
2327 		 */
2328 		if (nb < DBWORD) {
2329 			/* free (zero) the appropriate bits within this
2330 			 * dmap word.
2331 			 */
2332 			dp->wmap[word] &=
2333 			    cpu_to_le32(~(ONES << (DBWORD - nb)
2334 					  >> wbitno));
2335 
2336 			/* update the leaf for this dmap word.
2337 			 */
2338 			rc = dbJoin(tp, word,
2339 				    dbMaxBud((u8 *)&dp->wmap[word]), false);
2340 			if (rc)
2341 				return rc;
2342 
2343 			word += 1;
2344 		} else {
2345 			/* one or more dmap words are fully contained
2346 			 * within the block range.  determine how many
2347 			 * words and free (zero) the bits of these words.
2348 			 */
2349 			nwords = rembits >> L2DBWORD;
2350 			memset(&dp->wmap[word], 0, nwords * 4);
2351 
2352 			/* determine how many bits.
2353 			 */
2354 			nb = nwords << L2DBWORD;
2355 
2356 			/* now update the appropriate leaves to reflect
2357 			 * the freed words.
2358 			 */
2359 			for (; nwords > 0; nwords -= nw) {
2360 				/* determine what the leaf value should be
2361 				 * updated to as the minimum of the l2 number
2362 				 * of bits being freed and the l2 (max) number
2363 				 * of bits that can be described by this leaf.
2364 				 */
2365 				size =
2366 				    min(LITOL2BSZ
2367 					(word, L2LPERDMAP, BUDMIN),
2368 					NLSTOL2BSZ(nwords));
2369 
2370 				/* update the leaf.
2371 				 */
2372 				rc = dbJoin(tp, word, size, false);
2373 				if (rc)
2374 					return rc;
2375 
2376 				/* get the number of dmap words handled.
2377 				 */
2378 				nw = BUDSIZE(size, BUDMIN);
2379 				word += nw;
2380 			}
2381 		}
2382 	}
2383 
2384 	/* update the free count for this dmap.
2385 	 */
2386 	le32_add_cpu(&dp->nfree, nblocks);
2387 
2388 	BMAP_LOCK(bmp);
2389 
2390 	/* update the free count for the allocation group and
2391 	 * map.
2392 	 */
2393 	agno = blkno >> bmp->db_agl2size;
2394 	bmp->db_nfree += nblocks;
2395 	bmp->db_agfree[agno] += nblocks;
2396 
2397 	/* check if this allocation group is not completely free and
2398 	 * if it is currently the maximum (rightmost) allocation group.
2399 	 * if so, establish the new maximum allocation group number by
2400 	 * searching left for the first allocation group with allocation.
2401 	 */
2402 	if ((bmp->db_agfree[agno] == bmp->db_agsize && agno == bmp->db_maxag) ||
2403 	    (agno == bmp->db_numag - 1 &&
2404 	     bmp->db_agfree[agno] == (bmp-> db_mapsize & (BPERDMAP - 1)))) {
2405 		while (bmp->db_maxag > 0) {
2406 			bmp->db_maxag -= 1;
2407 			if (bmp->db_agfree[bmp->db_maxag] !=
2408 			    bmp->db_agsize)
2409 				break;
2410 		}
2411 
2412 		/* re-establish the allocation group preference if the
2413 		 * current preference is right of the maximum allocation
2414 		 * group.
2415 		 */
2416 		if (bmp->db_agpref > bmp->db_maxag)
2417 			bmp->db_agpref = bmp->db_maxag;
2418 	}
2419 
2420 	BMAP_UNLOCK(bmp);
2421 
2422 	return 0;
2423 }
2424 
2425 
2426 /*
2427  * NAME:	dbAdjCtl()
2428  *
2429  * FUNCTION:	adjust a dmap control page at a specified level to reflect
2430  *		the change in a lower level dmap or dmap control page's
2431  *		maximum string of free blocks (i.e. a change in the root
2432  *		of the lower level object's dmtree) due to the allocation
2433  *		or deallocation of a range of blocks with a single dmap.
2434  *
2435  *		on entry, this routine is provided with the new value of
2436  *		the lower level dmap or dmap control page root and the
2437  *		starting block number of the block range whose allocation
2438  *		or deallocation resulted in the root change.  this range
2439  *		is respresented by a single leaf of the current dmapctl
2440  *		and the leaf will be updated with this value, possibly
2441  *		causing a binary buddy system within the leaves to be
2442  *		split or joined.  the update may also cause the dmapctl's
2443  *		dmtree to be updated.
2444  *
2445  *		if the adjustment of the dmap control page, itself, causes its
2446  *		root to change, this change will be bubbled up to the next dmap
2447  *		control level by a recursive call to this routine, specifying
2448  *		the new root value and the next dmap control page level to
2449  *		be adjusted.
2450  * PARAMETERS:
2451  *	bmp	-  pointer to bmap descriptor
2452  *	blkno	-  the first block of a block range within a dmap.  it is
2453  *		   the allocation or deallocation of this block range that
2454  *		   requires the dmap control page to be adjusted.
2455  *	newval	-  the new value of the lower level dmap or dmap control
2456  *		   page root.
2457  *	alloc	-  'true' if adjustment is due to an allocation.
2458  *	level	-  current level of dmap control page (i.e. L0, L1, L2) to
2459  *		   be adjusted.
2460  *
2461  * RETURN VALUES:
2462  *	0	- success
2463  *	-EIO	- i/o error
2464  *
2465  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2466  */
2467 static int
2468 dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, int level)
2469 {
2470 	struct metapage *mp;
2471 	s8 oldroot;
2472 	int oldval;
2473 	s64 lblkno;
2474 	struct dmapctl *dcp;
2475 	int rc, leafno, ti;
2476 
2477 	/* get the buffer for the dmap control page for the specified
2478 	 * block number and control page level.
2479 	 */
2480 	lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, level);
2481 	mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0);
2482 	if (mp == NULL)
2483 		return -EIO;
2484 	dcp = (struct dmapctl *) mp->data;
2485 
2486 	if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) {
2487 		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n");
2488 		release_metapage(mp);
2489 		return -EIO;
2490 	}
2491 
2492 	/* determine the leaf number corresponding to the block and
2493 	 * the index within the dmap control tree.
2494 	 */
2495 	leafno = BLKTOCTLLEAF(blkno, dcp->budmin);
2496 	ti = leafno + le32_to_cpu(dcp->leafidx);
2497 
2498 	/* save the current leaf value and the current root level (i.e.
2499 	 * maximum l2 free string described by this dmapctl).
2500 	 */
2501 	oldval = dcp->stree[ti];
2502 	oldroot = dcp->stree[ROOT];
2503 
2504 	/* check if this is a control page update for an allocation.
2505 	 * if so, update the leaf to reflect the new leaf value using
2506 	 * dbSplit(); otherwise (deallocation), use dbJoin() to update
2507 	 * the leaf with the new value.  in addition to updating the
2508 	 * leaf, dbSplit() will also split the binary buddy system of
2509 	 * the leaves, if required, and bubble new values within the
2510 	 * dmapctl tree, if required.  similarly, dbJoin() will join
2511 	 * the binary buddy system of leaves and bubble new values up
2512 	 * the dmapctl tree as required by the new leaf value.
2513 	 */
2514 	if (alloc) {
2515 		/* check if we are in the middle of a binary buddy
2516 		 * system.  this happens when we are performing the
2517 		 * first allocation out of an allocation group that
2518 		 * is part (not the first part) of a larger binary
2519 		 * buddy system.  if we are in the middle, back split
2520 		 * the system prior to calling dbSplit() which assumes
2521 		 * that it is at the front of a binary buddy system.
2522 		 */
2523 		if (oldval == NOFREE) {
2524 			rc = dbBackSplit((dmtree_t *)dcp, leafno, true);
2525 			if (rc) {
2526 				release_metapage(mp);
2527 				return rc;
2528 			}
2529 			oldval = dcp->stree[ti];
2530 		}
2531 		dbSplit((dmtree_t *) dcp, leafno, dcp->budmin, newval, true);
2532 	} else {
2533 		rc = dbJoin((dmtree_t *) dcp, leafno, newval, true);
2534 		if (rc) {
2535 			release_metapage(mp);
2536 			return rc;
2537 		}
2538 	}
2539 
2540 	/* check if the root of the current dmap control page changed due
2541 	 * to the update and if the current dmap control page is not at
2542 	 * the current top level (i.e. L0, L1, L2) of the map.  if so (i.e.
2543 	 * root changed and this is not the top level), call this routine
2544 	 * again (recursion) for the next higher level of the mapping to
2545 	 * reflect the change in root for the current dmap control page.
2546 	 */
2547 	if (dcp->stree[ROOT] != oldroot) {
2548 		/* are we below the top level of the map.  if so,
2549 		 * bubble the root up to the next higher level.
2550 		 */
2551 		if (level < bmp->db_maxlevel) {
2552 			/* bubble up the new root of this dmap control page to
2553 			 * the next level.
2554 			 */
2555 			if ((rc =
2556 			     dbAdjCtl(bmp, blkno, dcp->stree[ROOT], alloc,
2557 				      level + 1))) {
2558 				/* something went wrong in bubbling up the new
2559 				 * root value, so backout the changes to the
2560 				 * current dmap control page.
2561 				 */
2562 				if (alloc) {
2563 					dbJoin((dmtree_t *) dcp, leafno,
2564 					       oldval, true);
2565 				} else {
2566 					/* the dbJoin() above might have
2567 					 * caused a larger binary buddy system
2568 					 * to form and we may now be in the
2569 					 * middle of it.  if this is the case,
2570 					 * back split the buddies.
2571 					 */
2572 					if (dcp->stree[ti] == NOFREE)
2573 						dbBackSplit((dmtree_t *)
2574 							    dcp, leafno, true);
2575 					dbSplit((dmtree_t *) dcp, leafno,
2576 						dcp->budmin, oldval, true);
2577 				}
2578 
2579 				/* release the buffer and return the error.
2580 				 */
2581 				release_metapage(mp);
2582 				return (rc);
2583 			}
2584 		} else {
2585 			/* we're at the top level of the map. update
2586 			 * the bmap control page to reflect the size
2587 			 * of the maximum free buddy system.
2588 			 */
2589 			assert(level == bmp->db_maxlevel);
2590 			if (bmp->db_maxfreebud != oldroot) {
2591 				jfs_error(bmp->db_ipbmap->i_sb,
2592 					  "the maximum free buddy is not the old root\n");
2593 			}
2594 			bmp->db_maxfreebud = dcp->stree[ROOT];
2595 		}
2596 	}
2597 
2598 	/* write the buffer.
2599 	 */
2600 	write_metapage(mp);
2601 
2602 	return (0);
2603 }
2604 
2605 
2606 /*
2607  * NAME:	dbSplit()
2608  *
2609  * FUNCTION:	update the leaf of a dmtree with a new value, splitting
2610  *		the leaf from the binary buddy system of the dmtree's
2611  *		leaves, as required.
2612  *
2613  * PARAMETERS:
2614  *	tp	- pointer to the tree containing the leaf.
2615  *	leafno	- the number of the leaf to be updated.
2616  *	splitsz	- the size the binary buddy system starting at the leaf
2617  *		  must be split to, specified as the log2 number of blocks.
2618  *	newval	- the new value for the leaf.
2619  *
2620  * RETURN VALUES: none
2621  *
2622  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2623  */
2624 static void dbSplit(dmtree_t *tp, int leafno, int splitsz, int newval, bool is_ctl)
2625 {
2626 	int budsz;
2627 	int cursz;
2628 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2629 
2630 	/* check if the leaf needs to be split.
2631 	 */
2632 	if (leaf[leafno] > tp->dmt_budmin) {
2633 		/* the split occurs by cutting the buddy system in half
2634 		 * at the specified leaf until we reach the specified
2635 		 * size.  pick up the starting split size (current size
2636 		 * - 1 in l2) and the corresponding buddy size.
2637 		 */
2638 		cursz = leaf[leafno] - 1;
2639 		budsz = BUDSIZE(cursz, tp->dmt_budmin);
2640 
2641 		/* split until we reach the specified size.
2642 		 */
2643 		while (cursz >= splitsz) {
2644 			/* update the buddy's leaf with its new value.
2645 			 */
2646 			dbAdjTree(tp, leafno ^ budsz, cursz, is_ctl);
2647 
2648 			/* on to the next size and buddy.
2649 			 */
2650 			cursz -= 1;
2651 			budsz >>= 1;
2652 		}
2653 	}
2654 
2655 	/* adjust the dmap tree to reflect the specified leaf's new
2656 	 * value.
2657 	 */
2658 	dbAdjTree(tp, leafno, newval, is_ctl);
2659 }
2660 
2661 
2662 /*
2663  * NAME:	dbBackSplit()
2664  *
2665  * FUNCTION:	back split the binary buddy system of dmtree leaves
2666  *		that hold a specified leaf until the specified leaf
2667  *		starts its own binary buddy system.
2668  *
2669  *		the allocators typically perform allocations at the start
2670  *		of binary buddy systems and dbSplit() is used to accomplish
2671  *		any required splits.  in some cases, however, allocation
2672  *		may occur in the middle of a binary system and requires a
2673  *		back split, with the split proceeding out from the middle of
2674  *		the system (less efficient) rather than the start of the
2675  *		system (more efficient).  the cases in which a back split
2676  *		is required are rare and are limited to the first allocation
2677  *		within an allocation group which is a part (not first part)
2678  *		of a larger binary buddy system and a few exception cases
2679  *		in which a previous join operation must be backed out.
2680  *
2681  * PARAMETERS:
2682  *	tp	- pointer to the tree containing the leaf.
2683  *	leafno	- the number of the leaf to be updated.
2684  *
2685  * RETURN VALUES: none
2686  *
2687  * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit;
2688  */
2689 static int dbBackSplit(dmtree_t *tp, int leafno, bool is_ctl)
2690 {
2691 	int budsz, bud, w, bsz, size;
2692 	int cursz;
2693 	s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2694 
2695 	/* leaf should be part (not first part) of a binary
2696 	 * buddy system.
2697 	 */
2698 	assert(leaf[leafno] == NOFREE);
2699 
2700 	/* the back split is accomplished by iteratively finding the leaf
2701 	 * that starts the buddy system that contains the specified leaf and
2702 	 * splitting that system in two.  this iteration continues until
2703 	 * the specified leaf becomes the start of a buddy system.
2704 	 *
2705 	 * determine maximum possible l2 size for the specified leaf.
2706 	 */
2707 	size =
2708 	    LITOL2BSZ(leafno, le32_to_cpu(tp->dmt_l2nleafs),
2709 		      tp->dmt_budmin);
2710 
2711 	/* determine the number of leaves covered by this size.  this
2712 	 * is the buddy size that we will start with as we search for
2713 	 * the buddy system that contains the specified leaf.
2714 	 */
2715 	budsz = BUDSIZE(size, tp->dmt_budmin);
2716 
2717 	/* back split.
2718 	 */
2719 	while (leaf[leafno] == NOFREE) {
2720 		/* find the leftmost buddy leaf.
2721 		 */
2722 		for (w = leafno, bsz = budsz;; bsz <<= 1,
2723 		     w = (w < bud) ? w : bud) {
2724 			if (bsz >= le32_to_cpu(tp->dmt_nleafs)) {
2725 				jfs_err("JFS: block map error in dbBackSplit");
2726 				return -EIO;
2727 			}
2728 
2729 			/* determine the buddy.
2730 			 */
2731 			bud = w ^ bsz;
2732 
2733 			/* check if this buddy is the start of the system.
2734 			 */
2735 			if (leaf[bud] != NOFREE) {
2736 				/* split the leaf at the start of the
2737 				 * system in two.
2738 				 */
2739 				cursz = leaf[bud] - 1;
2740 				dbSplit(tp, bud, cursz, cursz, is_ctl);
2741 				break;
2742 			}
2743 		}
2744 	}
2745 
2746 	if (leaf[leafno] != size) {
2747 		jfs_err("JFS: wrong leaf value in dbBackSplit");
2748 		return -EIO;
2749 	}
2750 	return 0;
2751 }
2752 
2753 
2754 /*
2755  * NAME:	dbJoin()
2756  *
2757  * FUNCTION:	update the leaf of a dmtree with a new value, joining
2758  *		the leaf with other leaves of the dmtree into a multi-leaf
2759  *		binary buddy system, as required.
2760  *
2761  * PARAMETERS:
2762  *	tp	- pointer to the tree containing the leaf.
2763  *	leafno	- the number of the leaf to be updated.
2764  *	newval	- the new value for the leaf.
2765  *
2766  * RETURN VALUES: none
2767  */
2768 static int dbJoin(dmtree_t *tp, int leafno, int newval, bool is_ctl)
2769 {
2770 	int budsz, buddy;
2771 	s8 *leaf;
2772 
2773 	/* can the new leaf value require a join with other leaves ?
2774 	 */
2775 	if (newval >= tp->dmt_budmin) {
2776 		/* pickup a pointer to the leaves of the tree.
2777 		 */
2778 		leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx);
2779 
2780 		/* try to join the specified leaf into a large binary
2781 		 * buddy system.  the join proceeds by attempting to join
2782 		 * the specified leafno with its buddy (leaf) at new value.
2783 		 * if the join occurs, we attempt to join the left leaf
2784 		 * of the joined buddies with its buddy at new value + 1.
2785 		 * we continue to join until we find a buddy that cannot be
2786 		 * joined (does not have a value equal to the size of the
2787 		 * last join) or until all leaves have been joined into a
2788 		 * single system.
2789 		 *
2790 		 * get the buddy size (number of words covered) of
2791 		 * the new value.
2792 		 */
2793 		budsz = BUDSIZE(newval, tp->dmt_budmin);
2794 
2795 		/* try to join.
2796 		 */
2797 		while (budsz < le32_to_cpu(tp->dmt_nleafs)) {
2798 			/* get the buddy leaf.
2799 			 */
2800 			buddy = leafno ^ budsz;
2801 
2802 			/* if the leaf's new value is greater than its
2803 			 * buddy's value, we join no more.
2804 			 */
2805 			if (newval > leaf[buddy])
2806 				break;
2807 
2808 			/* It shouldn't be less */
2809 			if (newval < leaf[buddy])
2810 				return -EIO;
2811 
2812 			/* check which (leafno or buddy) is the left buddy.
2813 			 * the left buddy gets to claim the blocks resulting
2814 			 * from the join while the right gets to claim none.
2815 			 * the left buddy is also eligible to participate in
2816 			 * a join at the next higher level while the right
2817 			 * is not.
2818 			 *
2819 			 */
2820 			if (leafno < buddy) {
2821 				/* leafno is the left buddy.
2822 				 */
2823 				dbAdjTree(tp, buddy, NOFREE, is_ctl);
2824 			} else {
2825 				/* buddy is the left buddy and becomes
2826 				 * leafno.
2827 				 */
2828 				dbAdjTree(tp, leafno, NOFREE, is_ctl);
2829 				leafno = buddy;
2830 			}
2831 
2832 			/* on to try the next join.
2833 			 */
2834 			newval += 1;
2835 			budsz <<= 1;
2836 		}
2837 	}
2838 
2839 	/* update the leaf value.
2840 	 */
2841 	dbAdjTree(tp, leafno, newval, is_ctl);
2842 
2843 	return 0;
2844 }
2845 
2846 
2847 /*
2848  * NAME:	dbAdjTree()
2849  *
2850  * FUNCTION:	update a leaf of a dmtree with a new value, adjusting
2851  *		the dmtree, as required, to reflect the new leaf value.
2852  *		the combination of any buddies must already be done before
2853  *		this is called.
2854  *
2855  * PARAMETERS:
2856  *	tp	- pointer to the tree to be adjusted.
2857  *	leafno	- the number of the leaf to be updated.
2858  *	newval	- the new value for the leaf.
2859  *
2860  * RETURN VALUES: none
2861  */
2862 static void dbAdjTree(dmtree_t *tp, int leafno, int newval, bool is_ctl)
2863 {
2864 	int lp, pp, k;
2865 	int max, size;
2866 
2867 	size = is_ctl ? CTLTREESIZE : TREESIZE;
2868 
2869 	/* pick up the index of the leaf for this leafno.
2870 	 */
2871 	lp = leafno + le32_to_cpu(tp->dmt_leafidx);
2872 
2873 	if (WARN_ON_ONCE(lp >= size || lp < 0))
2874 		return;
2875 
2876 	/* is the current value the same as the old value ?  if so,
2877 	 * there is nothing to do.
2878 	 */
2879 	if (tp->dmt_stree[lp] == newval)
2880 		return;
2881 
2882 	/* set the new value.
2883 	 */
2884 	tp->dmt_stree[lp] = newval;
2885 
2886 	/* bubble the new value up the tree as required.
2887 	 */
2888 	for (k = 0; k < le32_to_cpu(tp->dmt_height); k++) {
2889 		/* get the index of the first leaf of the 4 leaf
2890 		 * group containing the specified leaf (leafno).
2891 		 */
2892 		lp = ((lp - 1) & ~0x03) + 1;
2893 
2894 		/* get the index of the parent of this 4 leaf group.
2895 		 */
2896 		pp = (lp - 1) >> 2;
2897 
2898 		/* determine the maximum of the 4 leaves.
2899 		 */
2900 		max = TREEMAX(&tp->dmt_stree[lp]);
2901 
2902 		/* if the maximum of the 4 is the same as the
2903 		 * parent's value, we're done.
2904 		 */
2905 		if (tp->dmt_stree[pp] == max)
2906 			break;
2907 
2908 		/* parent gets new value.
2909 		 */
2910 		tp->dmt_stree[pp] = max;
2911 
2912 		/* parent becomes leaf for next go-round.
2913 		 */
2914 		lp = pp;
2915 	}
2916 }
2917 
2918 
2919 /*
2920  * NAME:	dbFindLeaf()
2921  *
2922  * FUNCTION:	search a dmtree_t for sufficient free blocks, returning
2923  *		the index of a leaf describing the free blocks if
2924  *		sufficient free blocks are found.
2925  *
2926  *		the search starts at the top of the dmtree_t tree and
2927  *		proceeds down the tree to the leftmost leaf with sufficient
2928  *		free space.
2929  *
2930  * PARAMETERS:
2931  *	tp	- pointer to the tree to be searched.
2932  *	l2nb	- log2 number of free blocks to search for.
2933  *	leafidx	- return pointer to be set to the index of the leaf
2934  *		  describing at least l2nb free blocks if sufficient
2935  *		  free blocks are found.
2936  *	is_ctl	- determines if the tree is of type ctl
2937  *
2938  * RETURN VALUES:
2939  *	0	- success
2940  *	-ENOSPC	- insufficient free blocks.
2941  */
2942 static int dbFindLeaf(dmtree_t *tp, int l2nb, int *leafidx, bool is_ctl)
2943 {
2944 	int ti, n = 0, k, x = 0;
2945 	int max_size;
2946 
2947 	max_size = is_ctl ? CTLTREESIZE : TREESIZE;
2948 
2949 	/* first check the root of the tree to see if there is
2950 	 * sufficient free space.
2951 	 */
2952 	if (l2nb > tp->dmt_stree[ROOT])
2953 		return -ENOSPC;
2954 
2955 	/* sufficient free space available. now search down the tree
2956 	 * starting at the next level for the leftmost leaf that
2957 	 * describes sufficient free space.
2958 	 */
2959 	for (k = le32_to_cpu(tp->dmt_height), ti = 1;
2960 	     k > 0; k--, ti = ((ti + n) << 2) + 1) {
2961 		/* search the four nodes at this level, starting from
2962 		 * the left.
2963 		 */
2964 		for (x = ti, n = 0; n < 4; n++) {
2965 			/* sufficient free space found.  move to the next
2966 			 * level (or quit if this is the last level).
2967 			 */
2968 			if (x + n > max_size)
2969 				return -ENOSPC;
2970 			if (l2nb <= tp->dmt_stree[x + n])
2971 				break;
2972 		}
2973 
2974 		/* better have found something since the higher
2975 		 * levels of the tree said it was here.
2976 		 */
2977 		assert(n < 4);
2978 	}
2979 
2980 	/* set the return to the leftmost leaf describing sufficient
2981 	 * free space.
2982 	 */
2983 	*leafidx = x + n - le32_to_cpu(tp->dmt_leafidx);
2984 
2985 	return (0);
2986 }
2987 
2988 
2989 /*
2990  * NAME:	dbFindBits()
2991  *
2992  * FUNCTION:	find a specified number of binary buddy free bits within a
2993  *		dmap bitmap word value.
2994  *
2995  *		this routine searches the bitmap value for (1 << l2nb) free
2996  *		bits at (1 << l2nb) alignments within the value.
2997  *
2998  * PARAMETERS:
2999  *	word	-  dmap bitmap word value.
3000  *	l2nb	-  number of free bits specified as a log2 number.
3001  *
3002  * RETURN VALUES:
3003  *	starting bit number of free bits.
3004  */
3005 static int dbFindBits(u32 word, int l2nb)
3006 {
3007 	int bitno, nb;
3008 	u32 mask;
3009 
3010 	/* get the number of bits.
3011 	 */
3012 	nb = 1 << l2nb;
3013 	assert(nb <= DBWORD);
3014 
3015 	/* complement the word so we can use a mask (i.e. 0s represent
3016 	 * free bits) and compute the mask.
3017 	 */
3018 	word = ~word;
3019 	mask = ONES << (DBWORD - nb);
3020 
3021 	/* scan the word for nb free bits at nb alignments.
3022 	 */
3023 	for (bitno = 0; mask != 0; bitno += nb, mask >>= nb) {
3024 		if ((mask & word) == mask)
3025 			break;
3026 	}
3027 
3028 	ASSERT(bitno < 32);
3029 
3030 	/* return the bit number.
3031 	 */
3032 	return (bitno);
3033 }
3034 
3035 
3036 /*
3037  * NAME:	dbMaxBud(u8 *cp)
3038  *
3039  * FUNCTION:	determine the largest binary buddy string of free
3040  *		bits within 32-bits of the map.
3041  *
3042  * PARAMETERS:
3043  *	cp	-  pointer to the 32-bit value.
3044  *
3045  * RETURN VALUES:
3046  *	largest binary buddy of free bits within a dmap word.
3047  */
3048 static int dbMaxBud(u8 * cp)
3049 {
3050 	signed char tmp1, tmp2;
3051 
3052 	/* check if the wmap word is all free. if so, the
3053 	 * free buddy size is BUDMIN.
3054 	 */
3055 	if (*((uint *) cp) == 0)
3056 		return (BUDMIN);
3057 
3058 	/* check if the wmap word is half free. if so, the
3059 	 * free buddy size is BUDMIN-1.
3060 	 */
3061 	if (*((u16 *) cp) == 0 || *((u16 *) cp + 1) == 0)
3062 		return (BUDMIN - 1);
3063 
3064 	/* not all free or half free. determine the free buddy
3065 	 * size thru table lookup using quarters of the wmap word.
3066 	 */
3067 	tmp1 = max(budtab[cp[2]], budtab[cp[3]]);
3068 	tmp2 = max(budtab[cp[0]], budtab[cp[1]]);
3069 	return (max(tmp1, tmp2));
3070 }
3071 
3072 
3073 /*
3074  * NAME:	cnttz(uint word)
3075  *
3076  * FUNCTION:	determine the number of trailing zeros within a 32-bit
3077  *		value.
3078  *
3079  * PARAMETERS:
3080  *	value	-  32-bit value to be examined.
3081  *
3082  * RETURN VALUES:
3083  *	count of trailing zeros
3084  */
3085 static int cnttz(u32 word)
3086 {
3087 	int n;
3088 
3089 	for (n = 0; n < 32; n++, word >>= 1) {
3090 		if (word & 0x01)
3091 			break;
3092 	}
3093 
3094 	return (n);
3095 }
3096 
3097 
3098 /*
3099  * NAME:	cntlz(u32 value)
3100  *
3101  * FUNCTION:	determine the number of leading zeros within a 32-bit
3102  *		value.
3103  *
3104  * PARAMETERS:
3105  *	value	-  32-bit value to be examined.
3106  *
3107  * RETURN VALUES:
3108  *	count of leading zeros
3109  */
3110 static int cntlz(u32 value)
3111 {
3112 	int n;
3113 
3114 	for (n = 0; n < 32; n++, value <<= 1) {
3115 		if (value & HIGHORDER)
3116 			break;
3117 	}
3118 	return (n);
3119 }
3120 
3121 
3122 /*
3123  * NAME:	blkstol2(s64 nb)
3124  *
3125  * FUNCTION:	convert a block count to its log2 value. if the block
3126  *		count is not a l2 multiple, it is rounded up to the next
3127  *		larger l2 multiple.
3128  *
3129  * PARAMETERS:
3130  *	nb	-  number of blocks
3131  *
3132  * RETURN VALUES:
3133  *	log2 number of blocks
3134  */
3135 static int blkstol2(s64 nb)
3136 {
3137 	int l2nb;
3138 	s64 mask;		/* meant to be signed */
3139 
3140 	mask = (s64) 1 << (64 - 1);
3141 
3142 	/* count the leading bits.
3143 	 */
3144 	for (l2nb = 0; l2nb < 64; l2nb++, mask >>= 1) {
3145 		/* leading bit found.
3146 		 */
3147 		if (nb & mask) {
3148 			/* determine the l2 value.
3149 			 */
3150 			l2nb = (64 - 1) - l2nb;
3151 
3152 			/* check if we need to round up.
3153 			 */
3154 			if (~mask & nb)
3155 				l2nb++;
3156 
3157 			return (l2nb);
3158 		}
3159 	}
3160 	assert(0);
3161 	return 0;		/* fix compiler warning */
3162 }
3163 
3164 
3165 /*
3166  * NAME:	dbAllocBottomUp()
3167  *
3168  * FUNCTION:	alloc the specified block range from the working block
3169  *		allocation map.
3170  *
3171  *		the blocks will be alloc from the working map one dmap
3172  *		at a time.
3173  *
3174  * PARAMETERS:
3175  *	ip	-  pointer to in-core inode;
3176  *	blkno	-  starting block number to be freed.
3177  *	nblocks	-  number of blocks to be freed.
3178  *
3179  * RETURN VALUES:
3180  *	0	- success
3181  *	-EIO	- i/o error
3182  */
3183 int dbAllocBottomUp(struct inode *ip, s64 blkno, s64 nblocks)
3184 {
3185 	struct metapage *mp;
3186 	struct dmap *dp;
3187 	int nb, rc;
3188 	s64 lblkno, rem;
3189 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
3190 	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
3191 
3192 	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
3193 
3194 	/* block to be allocated better be within the mapsize. */
3195 	ASSERT(nblocks <= bmp->db_mapsize - blkno);
3196 
3197 	/*
3198 	 * allocate the blocks a dmap at a time.
3199 	 */
3200 	mp = NULL;
3201 	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
3202 		/* release previous dmap if any */
3203 		if (mp) {
3204 			write_metapage(mp);
3205 		}
3206 
3207 		/* get the buffer for the current dmap. */
3208 		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
3209 		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
3210 		if (mp == NULL) {
3211 			IREAD_UNLOCK(ipbmap);
3212 			return -EIO;
3213 		}
3214 		dp = (struct dmap *) mp->data;
3215 
3216 		/* determine the number of blocks to be allocated from
3217 		 * this dmap.
3218 		 */
3219 		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
3220 
3221 		/* allocate the blocks. */
3222 		if ((rc = dbAllocDmapBU(bmp, dp, blkno, nb))) {
3223 			release_metapage(mp);
3224 			IREAD_UNLOCK(ipbmap);
3225 			return (rc);
3226 		}
3227 	}
3228 
3229 	/* write the last buffer. */
3230 	write_metapage(mp);
3231 
3232 	IREAD_UNLOCK(ipbmap);
3233 
3234 	return (0);
3235 }
3236 
3237 
3238 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
3239 			 int nblocks)
3240 {
3241 	int rc;
3242 	int dbitno, word, rembits, nb, nwords, wbitno, agno;
3243 	s8 oldroot;
3244 	struct dmaptree *tp = (struct dmaptree *) & dp->tree;
3245 
3246 	/* save the current value of the root (i.e. maximum free string)
3247 	 * of the dmap tree.
3248 	 */
3249 	oldroot = tp->stree[ROOT];
3250 
3251 	/* determine the bit number and word within the dmap of the
3252 	 * starting block.
3253 	 */
3254 	dbitno = blkno & (BPERDMAP - 1);
3255 	word = dbitno >> L2DBWORD;
3256 
3257 	/* block range better be within the dmap */
3258 	assert(dbitno + nblocks <= BPERDMAP);
3259 
3260 	/* allocate the bits of the dmap's words corresponding to the block
3261 	 * range. not all bits of the first and last words may be contained
3262 	 * within the block range.  if this is the case, we'll work against
3263 	 * those words (i.e. partial first and/or last) on an individual basis
3264 	 * (a single pass), allocating the bits of interest by hand and
3265 	 * updating the leaf corresponding to the dmap word. a single pass
3266 	 * will be used for all dmap words fully contained within the
3267 	 * specified range.  within this pass, the bits of all fully contained
3268 	 * dmap words will be marked as free in a single shot and the leaves
3269 	 * will be updated. a single leaf may describe the free space of
3270 	 * multiple dmap words, so we may update only a subset of the actual
3271 	 * leaves corresponding to the dmap words of the block range.
3272 	 */
3273 	for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) {
3274 		/* determine the bit number within the word and
3275 		 * the number of bits within the word.
3276 		 */
3277 		wbitno = dbitno & (DBWORD - 1);
3278 		nb = min(rembits, DBWORD - wbitno);
3279 
3280 		/* check if only part of a word is to be allocated.
3281 		 */
3282 		if (nb < DBWORD) {
3283 			/* allocate (set to 1) the appropriate bits within
3284 			 * this dmap word.
3285 			 */
3286 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
3287 						      >> wbitno);
3288 
3289 			word++;
3290 		} else {
3291 			/* one or more dmap words are fully contained
3292 			 * within the block range.  determine how many
3293 			 * words and allocate (set to 1) the bits of these
3294 			 * words.
3295 			 */
3296 			nwords = rembits >> L2DBWORD;
3297 			memset(&dp->wmap[word], (int) ONES, nwords * 4);
3298 
3299 			/* determine how many bits */
3300 			nb = nwords << L2DBWORD;
3301 			word += nwords;
3302 		}
3303 	}
3304 
3305 	/* update the free count for this dmap */
3306 	le32_add_cpu(&dp->nfree, -nblocks);
3307 
3308 	/* reconstruct summary tree */
3309 	dbInitDmapTree(dp);
3310 
3311 	BMAP_LOCK(bmp);
3312 
3313 	/* if this allocation group is completely free,
3314 	 * update the highest active allocation group number
3315 	 * if this allocation group is the new max.
3316 	 */
3317 	agno = blkno >> bmp->db_agl2size;
3318 	if (agno > bmp->db_maxag)
3319 		bmp->db_maxag = agno;
3320 
3321 	/* update the free count for the allocation group and map */
3322 	bmp->db_agfree[agno] -= nblocks;
3323 	bmp->db_nfree -= nblocks;
3324 
3325 	BMAP_UNLOCK(bmp);
3326 
3327 	/* if the root has not changed, done. */
3328 	if (tp->stree[ROOT] == oldroot)
3329 		return (0);
3330 
3331 	/* root changed. bubble the change up to the dmap control pages.
3332 	 * if the adjustment of the upper level control pages fails,
3333 	 * backout the bit allocation (thus making everything consistent).
3334 	 */
3335 	if ((rc = dbAdjCtl(bmp, blkno, tp->stree[ROOT], 1, 0)))
3336 		dbFreeBits(bmp, dp, blkno, nblocks);
3337 
3338 	return (rc);
3339 }
3340 
3341 
3342 /*
3343  * NAME:	dbExtendFS()
3344  *
3345  * FUNCTION:	extend bmap from blkno for nblocks;
3346  *		dbExtendFS() updates bmap ready for dbAllocBottomUp();
3347  *
3348  * L2
3349  *  |
3350  *   L1---------------------------------L1
3351  *    |					 |
3352  *     L0---------L0---------L0		  L0---------L0---------L0
3353  *      |	   |	      |		   |	      |		 |
3354  *	 d0,...,dn  d0,...,dn  d0,...,dn    d0,...,dn  d0,...,dn  d0,.,dm;
3355  * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm
3356  *
3357  * <---old---><----------------------------extend----------------------->
3358  */
3359 int dbExtendFS(struct inode *ipbmap, s64 blkno,	s64 nblocks)
3360 {
3361 	struct jfs_sb_info *sbi = JFS_SBI(ipbmap->i_sb);
3362 	int nbperpage = sbi->nbperpage;
3363 	int i, i0 = true, j, j0 = true, k, n;
3364 	s64 newsize;
3365 	s64 p;
3366 	struct metapage *mp, *l2mp, *l1mp = NULL, *l0mp = NULL;
3367 	struct dmapctl *l2dcp, *l1dcp, *l0dcp;
3368 	struct dmap *dp;
3369 	s8 *l0leaf, *l1leaf, *l2leaf;
3370 	struct bmap *bmp = sbi->bmap;
3371 	int agno, l2agsize, oldl2agsize;
3372 	s64 ag_rem;
3373 
3374 	newsize = blkno + nblocks;
3375 
3376 	jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld",
3377 		 (long long) blkno, (long long) nblocks, (long long) newsize);
3378 
3379 	/*
3380 	 *	initialize bmap control page.
3381 	 *
3382 	 * all the data in bmap control page should exclude
3383 	 * the mkfs hidden dmap page.
3384 	 */
3385 
3386 	/* update mapsize */
3387 	bmp->db_mapsize = newsize;
3388 	bmp->db_maxlevel = BMAPSZTOLEV(bmp->db_mapsize);
3389 
3390 	/* compute new AG size */
3391 	l2agsize = dbGetL2AGSize(newsize);
3392 	oldl2agsize = bmp->db_agl2size;
3393 
3394 	bmp->db_agl2size = l2agsize;
3395 	bmp->db_agsize = 1 << l2agsize;
3396 
3397 	/* compute new number of AG */
3398 	agno = bmp->db_numag;
3399 	bmp->db_numag = newsize >> l2agsize;
3400 	bmp->db_numag += ((u32) newsize % (u32) bmp->db_agsize) ? 1 : 0;
3401 
3402 	/*
3403 	 *	reconfigure db_agfree[]
3404 	 * from old AG configuration to new AG configuration;
3405 	 *
3406 	 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
3407 	 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
3408 	 * note: new AG size = old AG size * (2**x).
3409 	 */
3410 	if (l2agsize == oldl2agsize)
3411 		goto extend;
3412 	k = 1 << (l2agsize - oldl2agsize);
3413 	ag_rem = bmp->db_agfree[0];	/* save agfree[0] */
3414 	for (i = 0, n = 0; i < agno; n++) {
3415 		bmp->db_agfree[n] = 0;	/* init collection point */
3416 
3417 		/* coalesce contiguous k AGs; */
3418 		for (j = 0; j < k && i < agno; j++, i++) {
3419 			/* merge AGi to AGn */
3420 			bmp->db_agfree[n] += bmp->db_agfree[i];
3421 		}
3422 	}
3423 	bmp->db_agfree[0] += ag_rem;	/* restore agfree[0] */
3424 
3425 	for (; n < MAXAG; n++)
3426 		bmp->db_agfree[n] = 0;
3427 
3428 	/*
3429 	 * update highest active ag number
3430 	 */
3431 
3432 	bmp->db_maxag = bmp->db_maxag / k;
3433 
3434 	/*
3435 	 *	extend bmap
3436 	 *
3437 	 * update bit maps and corresponding level control pages;
3438 	 * global control page db_nfree, db_agfree[agno], db_maxfreebud;
3439 	 */
3440       extend:
3441 	/* get L2 page */
3442 	p = BMAPBLKNO + nbperpage;	/* L2 page */
3443 	l2mp = read_metapage(ipbmap, p, PSIZE, 0);
3444 	if (!l2mp) {
3445 		jfs_error(ipbmap->i_sb, "L2 page could not be read\n");
3446 		return -EIO;
3447 	}
3448 	l2dcp = (struct dmapctl *) l2mp->data;
3449 
3450 	/* compute start L1 */
3451 	k = blkno >> L2MAXL1SIZE;
3452 	l2leaf = l2dcp->stree + CTLLEAFIND + k;
3453 	p = BLKTOL1(blkno, sbi->l2nbperpage);	/* L1 page */
3454 
3455 	/*
3456 	 * extend each L1 in L2
3457 	 */
3458 	for (; k < LPERCTL; k++, p += nbperpage) {
3459 		/* get L1 page */
3460 		if (j0) {
3461 			/* read in L1 page: (blkno & (MAXL1SIZE - 1)) */
3462 			l1mp = read_metapage(ipbmap, p, PSIZE, 0);
3463 			if (l1mp == NULL)
3464 				goto errout;
3465 			l1dcp = (struct dmapctl *) l1mp->data;
3466 
3467 			/* compute start L0 */
3468 			j = (blkno & (MAXL1SIZE - 1)) >> L2MAXL0SIZE;
3469 			l1leaf = l1dcp->stree + CTLLEAFIND + j;
3470 			p = BLKTOL0(blkno, sbi->l2nbperpage);
3471 			j0 = false;
3472 		} else {
3473 			/* assign/init L1 page */
3474 			l1mp = get_metapage(ipbmap, p, PSIZE, 0);
3475 			if (l1mp == NULL)
3476 				goto errout;
3477 
3478 			l1dcp = (struct dmapctl *) l1mp->data;
3479 
3480 			/* compute start L0 */
3481 			j = 0;
3482 			l1leaf = l1dcp->stree + CTLLEAFIND;
3483 			p += nbperpage;	/* 1st L0 of L1.k */
3484 		}
3485 
3486 		/*
3487 		 * extend each L0 in L1
3488 		 */
3489 		for (; j < LPERCTL; j++) {
3490 			/* get L0 page */
3491 			if (i0) {
3492 				/* read in L0 page: (blkno & (MAXL0SIZE - 1)) */
3493 
3494 				l0mp = read_metapage(ipbmap, p, PSIZE, 0);
3495 				if (l0mp == NULL)
3496 					goto errout;
3497 				l0dcp = (struct dmapctl *) l0mp->data;
3498 
3499 				/* compute start dmap */
3500 				i = (blkno & (MAXL0SIZE - 1)) >>
3501 				    L2BPERDMAP;
3502 				l0leaf = l0dcp->stree + CTLLEAFIND + i;
3503 				p = BLKTODMAP(blkno,
3504 					      sbi->l2nbperpage);
3505 				i0 = false;
3506 			} else {
3507 				/* assign/init L0 page */
3508 				l0mp = get_metapage(ipbmap, p, PSIZE, 0);
3509 				if (l0mp == NULL)
3510 					goto errout;
3511 
3512 				l0dcp = (struct dmapctl *) l0mp->data;
3513 
3514 				/* compute start dmap */
3515 				i = 0;
3516 				l0leaf = l0dcp->stree + CTLLEAFIND;
3517 				p += nbperpage;	/* 1st dmap of L0.j */
3518 			}
3519 
3520 			/*
3521 			 * extend each dmap in L0
3522 			 */
3523 			for (; i < LPERCTL; i++) {
3524 				/*
3525 				 * reconstruct the dmap page, and
3526 				 * initialize corresponding parent L0 leaf
3527 				 */
3528 				if ((n = blkno & (BPERDMAP - 1))) {
3529 					/* read in dmap page: */
3530 					mp = read_metapage(ipbmap, p,
3531 							   PSIZE, 0);
3532 					if (mp == NULL)
3533 						goto errout;
3534 					n = min(nblocks, (s64)BPERDMAP - n);
3535 				} else {
3536 					/* assign/init dmap page */
3537 					mp = read_metapage(ipbmap, p,
3538 							   PSIZE, 0);
3539 					if (mp == NULL)
3540 						goto errout;
3541 
3542 					n = min_t(s64, nblocks, BPERDMAP);
3543 				}
3544 
3545 				dp = (struct dmap *) mp->data;
3546 				*l0leaf = dbInitDmap(dp, blkno, n);
3547 
3548 				bmp->db_nfree += n;
3549 				agno = le64_to_cpu(dp->start) >> l2agsize;
3550 				bmp->db_agfree[agno] += n;
3551 
3552 				write_metapage(mp);
3553 
3554 				l0leaf++;
3555 				p += nbperpage;
3556 
3557 				blkno += n;
3558 				nblocks -= n;
3559 				if (nblocks == 0)
3560 					break;
3561 			}	/* for each dmap in a L0 */
3562 
3563 			/*
3564 			 * build current L0 page from its leaves, and
3565 			 * initialize corresponding parent L1 leaf
3566 			 */
3567 			*l1leaf = dbInitDmapCtl(l0dcp, 0, ++i);
3568 			write_metapage(l0mp);
3569 			l0mp = NULL;
3570 
3571 			if (nblocks)
3572 				l1leaf++;	/* continue for next L0 */
3573 			else {
3574 				/* more than 1 L0 ? */
3575 				if (j > 0)
3576 					break;	/* build L1 page */
3577 				else {
3578 					/* summarize in global bmap page */
3579 					bmp->db_maxfreebud = *l1leaf;
3580 					release_metapage(l1mp);
3581 					release_metapage(l2mp);
3582 					goto finalize;
3583 				}
3584 			}
3585 		}		/* for each L0 in a L1 */
3586 
3587 		/*
3588 		 * build current L1 page from its leaves, and
3589 		 * initialize corresponding parent L2 leaf
3590 		 */
3591 		*l2leaf = dbInitDmapCtl(l1dcp, 1, ++j);
3592 		write_metapage(l1mp);
3593 		l1mp = NULL;
3594 
3595 		if (nblocks)
3596 			l2leaf++;	/* continue for next L1 */
3597 		else {
3598 			/* more than 1 L1 ? */
3599 			if (k > 0)
3600 				break;	/* build L2 page */
3601 			else {
3602 				/* summarize in global bmap page */
3603 				bmp->db_maxfreebud = *l2leaf;
3604 				release_metapage(l2mp);
3605 				goto finalize;
3606 			}
3607 		}
3608 	}			/* for each L1 in a L2 */
3609 
3610 	jfs_error(ipbmap->i_sb, "function has not returned as expected\n");
3611 errout:
3612 	if (l0mp)
3613 		release_metapage(l0mp);
3614 	if (l1mp)
3615 		release_metapage(l1mp);
3616 	release_metapage(l2mp);
3617 	return -EIO;
3618 
3619 	/*
3620 	 *	finalize bmap control page
3621 	 */
3622 finalize:
3623 
3624 	return 0;
3625 }
3626 
3627 
3628 /*
3629  *	dbFinalizeBmap()
3630  */
3631 void dbFinalizeBmap(struct inode *ipbmap)
3632 {
3633 	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
3634 	int actags, inactags, l2nl;
3635 	s64 ag_rem, actfree, inactfree, avgfree;
3636 	int i, n;
3637 
3638 	/*
3639 	 *	finalize bmap control page
3640 	 */
3641 //finalize:
3642 	/*
3643 	 * compute db_agpref: preferred ag to allocate from
3644 	 * (the leftmost ag with average free space in it);
3645 	 */
3646 //agpref:
3647 	/* get the number of active ags and inactive ags */
3648 	actags = bmp->db_maxag + 1;
3649 	inactags = bmp->db_numag - actags;
3650 	ag_rem = bmp->db_mapsize & (bmp->db_agsize - 1);	/* ??? */
3651 
3652 	/* determine how many blocks are in the inactive allocation
3653 	 * groups. in doing this, we must account for the fact that
3654 	 * the rightmost group might be a partial group (i.e. file
3655 	 * system size is not a multiple of the group size).
3656 	 */
3657 	inactfree = (inactags && ag_rem) ?
3658 	    ((inactags - 1) << bmp->db_agl2size) + ag_rem
3659 	    : inactags << bmp->db_agl2size;
3660 
3661 	/* determine how many free blocks are in the active
3662 	 * allocation groups plus the average number of free blocks
3663 	 * within the active ags.
3664 	 */
3665 	actfree = bmp->db_nfree - inactfree;
3666 	avgfree = (u32) actfree / (u32) actags;
3667 
3668 	/* if the preferred allocation group has not average free space.
3669 	 * re-establish the preferred group as the leftmost
3670 	 * group with average free space.
3671 	 */
3672 	if (bmp->db_agfree[bmp->db_agpref] < avgfree) {
3673 		for (bmp->db_agpref = 0; bmp->db_agpref < actags;
3674 		     bmp->db_agpref++) {
3675 			if (bmp->db_agfree[bmp->db_agpref] >= avgfree)
3676 				break;
3677 		}
3678 		if (bmp->db_agpref >= bmp->db_numag) {
3679 			jfs_error(ipbmap->i_sb,
3680 				  "cannot find ag with average freespace\n");
3681 		}
3682 	}
3683 
3684 	/*
3685 	 * compute db_aglevel, db_agheight, db_width, db_agstart:
3686 	 * an ag is covered in aglevel dmapctl summary tree,
3687 	 * at agheight level height (from leaf) with agwidth number of nodes
3688 	 * each, which starts at agstart index node of the smmary tree node
3689 	 * array;
3690 	 */
3691 	bmp->db_aglevel = BMAPSZTOLEV(bmp->db_agsize);
3692 	l2nl =
3693 	    bmp->db_agl2size - (L2BPERDMAP + bmp->db_aglevel * L2LPERCTL);
3694 	bmp->db_agheight = l2nl >> 1;
3695 	bmp->db_agwidth = 1 << (l2nl - (bmp->db_agheight << 1));
3696 	for (i = 5 - bmp->db_agheight, bmp->db_agstart = 0, n = 1; i > 0;
3697 	     i--) {
3698 		bmp->db_agstart += n;
3699 		n <<= 2;
3700 	}
3701 
3702 }
3703 
3704 
3705 /*
3706  * NAME:	dbInitDmap()/ujfs_idmap_page()
3707  *
3708  * FUNCTION:	initialize working/persistent bitmap of the dmap page
3709  *		for the specified number of blocks:
3710  *
3711  *		at entry, the bitmaps had been initialized as free (ZEROS);
3712  *		The number of blocks will only account for the actually
3713  *		existing blocks. Blocks which don't actually exist in
3714  *		the aggregate will be marked as allocated (ONES);
3715  *
3716  * PARAMETERS:
3717  *	dp	- pointer to page of map
3718  *	nblocks	- number of blocks this page
3719  *
3720  * RETURNS: NONE
3721  */
3722 static int dbInitDmap(struct dmap * dp, s64 Blkno, int nblocks)
3723 {
3724 	int blkno, w, b, r, nw, nb, i;
3725 
3726 	/* starting block number within the dmap */
3727 	blkno = Blkno & (BPERDMAP - 1);
3728 
3729 	if (blkno == 0) {
3730 		dp->nblocks = dp->nfree = cpu_to_le32(nblocks);
3731 		dp->start = cpu_to_le64(Blkno);
3732 
3733 		if (nblocks == BPERDMAP) {
3734 			memset(&dp->wmap[0], 0, LPERDMAP * 4);
3735 			memset(&dp->pmap[0], 0, LPERDMAP * 4);
3736 			goto initTree;
3737 		}
3738 	} else {
3739 		le32_add_cpu(&dp->nblocks, nblocks);
3740 		le32_add_cpu(&dp->nfree, nblocks);
3741 	}
3742 
3743 	/* word number containing start block number */
3744 	w = blkno >> L2DBWORD;
3745 
3746 	/*
3747 	 * free the bits corresponding to the block range (ZEROS):
3748 	 * note: not all bits of the first and last words may be contained
3749 	 * within the block range.
3750 	 */
3751 	for (r = nblocks; r > 0; r -= nb, blkno += nb) {
3752 		/* number of bits preceding range to be freed in the word */
3753 		b = blkno & (DBWORD - 1);
3754 		/* number of bits to free in the word */
3755 		nb = min(r, DBWORD - b);
3756 
3757 		/* is partial word to be freed ? */
3758 		if (nb < DBWORD) {
3759 			/* free (set to 0) from the bitmap word */
3760 			dp->wmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3761 						     >> b));
3762 			dp->pmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb)
3763 						     >> b));
3764 
3765 			/* skip the word freed */
3766 			w++;
3767 		} else {
3768 			/* free (set to 0) contiguous bitmap words */
3769 			nw = r >> L2DBWORD;
3770 			memset(&dp->wmap[w], 0, nw * 4);
3771 			memset(&dp->pmap[w], 0, nw * 4);
3772 
3773 			/* skip the words freed */
3774 			nb = nw << L2DBWORD;
3775 			w += nw;
3776 		}
3777 	}
3778 
3779 	/*
3780 	 * mark bits following the range to be freed (non-existing
3781 	 * blocks) as allocated (ONES)
3782 	 */
3783 
3784 	if (blkno == BPERDMAP)
3785 		goto initTree;
3786 
3787 	/* the first word beyond the end of existing blocks */
3788 	w = blkno >> L2DBWORD;
3789 
3790 	/* does nblocks fall on a 32-bit boundary ? */
3791 	b = blkno & (DBWORD - 1);
3792 	if (b) {
3793 		/* mark a partial word allocated */
3794 		dp->wmap[w] = dp->pmap[w] = cpu_to_le32(ONES >> b);
3795 		w++;
3796 	}
3797 
3798 	/* set the rest of the words in the page to allocated (ONES) */
3799 	for (i = w; i < LPERDMAP; i++)
3800 		dp->pmap[i] = dp->wmap[i] = cpu_to_le32(ONES);
3801 
3802 	/*
3803 	 * init tree
3804 	 */
3805       initTree:
3806 	return (dbInitDmapTree(dp));
3807 }
3808 
3809 
3810 /*
3811  * NAME:	dbInitDmapTree()/ujfs_complete_dmap()
3812  *
3813  * FUNCTION:	initialize summary tree of the specified dmap:
3814  *
3815  *		at entry, bitmap of the dmap has been initialized;
3816  *
3817  * PARAMETERS:
3818  *	dp	- dmap to complete
3819  *	blkno	- starting block number for this dmap
3820  *	treemax	- will be filled in with max free for this dmap
3821  *
3822  * RETURNS:	max free string at the root of the tree
3823  */
3824 static int dbInitDmapTree(struct dmap * dp)
3825 {
3826 	struct dmaptree *tp;
3827 	s8 *cp;
3828 	int i;
3829 
3830 	/* init fixed info of tree */
3831 	tp = &dp->tree;
3832 	tp->nleafs = cpu_to_le32(LPERDMAP);
3833 	tp->l2nleafs = cpu_to_le32(L2LPERDMAP);
3834 	tp->leafidx = cpu_to_le32(LEAFIND);
3835 	tp->height = cpu_to_le32(4);
3836 	tp->budmin = BUDMIN;
3837 
3838 	/* init each leaf from corresponding wmap word:
3839 	 * note: leaf is set to NOFREE(-1) if all blocks of corresponding
3840 	 * bitmap word are allocated.
3841 	 */
3842 	cp = tp->stree + le32_to_cpu(tp->leafidx);
3843 	for (i = 0; i < LPERDMAP; i++)
3844 		*cp++ = dbMaxBud((u8 *) & dp->wmap[i]);
3845 
3846 	/* build the dmap's binary buddy summary tree */
3847 	return (dbInitTree(tp));
3848 }
3849 
3850 
3851 /*
3852  * NAME:	dbInitTree()/ujfs_adjtree()
3853  *
3854  * FUNCTION:	initialize binary buddy summary tree of a dmap or dmapctl.
3855  *
3856  *		at entry, the leaves of the tree has been initialized
3857  *		from corresponding bitmap word or root of summary tree
3858  *		of the child control page;
3859  *		configure binary buddy system at the leaf level, then
3860  *		bubble up the values of the leaf nodes up the tree.
3861  *
3862  * PARAMETERS:
3863  *	cp	- Pointer to the root of the tree
3864  *	l2leaves- Number of leaf nodes as a power of 2
3865  *	l2min	- Number of blocks that can be covered by a leaf
3866  *		  as a power of 2
3867  *
3868  * RETURNS: max free string at the root of the tree
3869  */
3870 static int dbInitTree(struct dmaptree * dtp)
3871 {
3872 	int l2max, l2free, bsize, nextb, i;
3873 	int child, parent, nparent;
3874 	s8 *tp, *cp, *cp1;
3875 
3876 	tp = dtp->stree;
3877 
3878 	/* Determine the maximum free string possible for the leaves */
3879 	l2max = le32_to_cpu(dtp->l2nleafs) + dtp->budmin;
3880 
3881 	/*
3882 	 * configure the leaf level into binary buddy system
3883 	 *
3884 	 * Try to combine buddies starting with a buddy size of 1
3885 	 * (i.e. two leaves). At a buddy size of 1 two buddy leaves
3886 	 * can be combined if both buddies have a maximum free of l2min;
3887 	 * the combination will result in the left-most buddy leaf having
3888 	 * a maximum free of l2min+1.
3889 	 * After processing all buddies for a given size, process buddies
3890 	 * at the next higher buddy size (i.e. current size * 2) and
3891 	 * the next maximum free (current free + 1).
3892 	 * This continues until the maximum possible buddy combination
3893 	 * yields maximum free.
3894 	 */
3895 	for (l2free = dtp->budmin, bsize = 1; l2free < l2max;
3896 	     l2free++, bsize = nextb) {
3897 		/* get next buddy size == current buddy pair size */
3898 		nextb = bsize << 1;
3899 
3900 		/* scan each adjacent buddy pair at current buddy size */
3901 		for (i = 0, cp = tp + le32_to_cpu(dtp->leafidx);
3902 		     i < le32_to_cpu(dtp->nleafs);
3903 		     i += nextb, cp += nextb) {
3904 			/* coalesce if both adjacent buddies are max free */
3905 			if (*cp == l2free && *(cp + bsize) == l2free) {
3906 				*cp = l2free + 1;	/* left take right */
3907 				*(cp + bsize) = -1;	/* right give left */
3908 			}
3909 		}
3910 	}
3911 
3912 	/*
3913 	 * bubble summary information of leaves up the tree.
3914 	 *
3915 	 * Starting at the leaf node level, the four nodes described by
3916 	 * the higher level parent node are compared for a maximum free and
3917 	 * this maximum becomes the value of the parent node.
3918 	 * when all lower level nodes are processed in this fashion then
3919 	 * move up to the next level (parent becomes a lower level node) and
3920 	 * continue the process for that level.
3921 	 */
3922 	for (child = le32_to_cpu(dtp->leafidx),
3923 	     nparent = le32_to_cpu(dtp->nleafs) >> 2;
3924 	     nparent > 0; nparent >>= 2, child = parent) {
3925 		/* get index of 1st node of parent level */
3926 		parent = (child - 1) >> 2;
3927 
3928 		/* set the value of the parent node as the maximum
3929 		 * of the four nodes of the current level.
3930 		 */
3931 		for (i = 0, cp = tp + child, cp1 = tp + parent;
3932 		     i < nparent; i++, cp += 4, cp1++)
3933 			*cp1 = TREEMAX(cp);
3934 	}
3935 
3936 	return (*tp);
3937 }
3938 
3939 
3940 /*
3941  *	dbInitDmapCtl()
3942  *
3943  * function: initialize dmapctl page
3944  */
3945 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i)
3946 {				/* start leaf index not covered by range */
3947 	s8 *cp;
3948 
3949 	dcp->nleafs = cpu_to_le32(LPERCTL);
3950 	dcp->l2nleafs = cpu_to_le32(L2LPERCTL);
3951 	dcp->leafidx = cpu_to_le32(CTLLEAFIND);
3952 	dcp->height = cpu_to_le32(5);
3953 	dcp->budmin = L2BPERDMAP + L2LPERCTL * level;
3954 
3955 	/*
3956 	 * initialize the leaves of current level that were not covered
3957 	 * by the specified input block range (i.e. the leaves have no
3958 	 * low level dmapctl or dmap).
3959 	 */
3960 	cp = &dcp->stree[CTLLEAFIND + i];
3961 	for (; i < LPERCTL; i++)
3962 		*cp++ = NOFREE;
3963 
3964 	/* build the dmap's binary buddy summary tree */
3965 	return (dbInitTree((struct dmaptree *) dcp));
3966 }
3967 
3968 
3969 /*
3970  * NAME:	dbGetL2AGSize()/ujfs_getagl2size()
3971  *
3972  * FUNCTION:	Determine log2(allocation group size) from aggregate size
3973  *
3974  * PARAMETERS:
3975  *	nblocks	- Number of blocks in aggregate
3976  *
3977  * RETURNS: log2(allocation group size) in aggregate blocks
3978  */
3979 static int dbGetL2AGSize(s64 nblocks)
3980 {
3981 	s64 sz;
3982 	s64 m;
3983 	int l2sz;
3984 
3985 	if (nblocks < BPERDMAP * MAXAG)
3986 		return (L2BPERDMAP);
3987 
3988 	/* round up aggregate size to power of 2 */
3989 	m = ((u64) 1 << (64 - 1));
3990 	for (l2sz = 64; l2sz >= 0; l2sz--, m >>= 1) {
3991 		if (m & nblocks)
3992 			break;
3993 	}
3994 
3995 	sz = (s64) 1 << l2sz;
3996 	if (sz < nblocks)
3997 		l2sz += 1;
3998 
3999 	/* agsize = roundupSize/max_number_of_ag */
4000 	return (l2sz - L2MAXAG);
4001 }
4002 
4003 
4004 /*
4005  * NAME:	dbMapFileSizeToMapSize()
4006  *
4007  * FUNCTION:	compute number of blocks the block allocation map file
4008  *		can cover from the map file size;
4009  *
4010  * RETURNS:	Number of blocks which can be covered by this block map file;
4011  */
4012 
4013 /*
4014  * maximum number of map pages at each level including control pages
4015  */
4016 #define MAXL0PAGES	(1 + LPERCTL)
4017 #define MAXL1PAGES	(1 + LPERCTL * MAXL0PAGES)
4018 
4019 /*
4020  * convert number of map pages to the zero origin top dmapctl level
4021  */
4022 #define BMAPPGTOLEV(npages)	\
4023 	(((npages) <= 3 + MAXL0PAGES) ? 0 : \
4024 	 ((npages) <= 2 + MAXL1PAGES) ? 1 : 2)
4025 
4026 s64 dbMapFileSizeToMapSize(struct inode * ipbmap)
4027 {
4028 	struct super_block *sb = ipbmap->i_sb;
4029 	s64 nblocks;
4030 	s64 npages, ndmaps;
4031 	int level, i;
4032 	int complete, factor;
4033 
4034 	nblocks = ipbmap->i_size >> JFS_SBI(sb)->l2bsize;
4035 	npages = nblocks >> JFS_SBI(sb)->l2nbperpage;
4036 	level = BMAPPGTOLEV(npages);
4037 
4038 	/* At each level, accumulate the number of dmap pages covered by
4039 	 * the number of full child levels below it;
4040 	 * repeat for the last incomplete child level.
4041 	 */
4042 	ndmaps = 0;
4043 	npages--;		/* skip the first global control page */
4044 	/* skip higher level control pages above top level covered by map */
4045 	npages -= (2 - level);
4046 	npages--;		/* skip top level's control page */
4047 	for (i = level; i >= 0; i--) {
4048 		factor =
4049 		    (i == 2) ? MAXL1PAGES : ((i == 1) ? MAXL0PAGES : 1);
4050 		complete = (u32) npages / factor;
4051 		ndmaps += complete * ((i == 2) ? LPERCTL * LPERCTL :
4052 				      ((i == 1) ? LPERCTL : 1));
4053 
4054 		/* pages in last/incomplete child */
4055 		npages = (u32) npages % factor;
4056 		/* skip incomplete child's level control page */
4057 		npages--;
4058 	}
4059 
4060 	/* convert the number of dmaps into the number of blocks
4061 	 * which can be covered by the dmaps;
4062 	 */
4063 	nblocks = ndmaps << L2BPERDMAP;
4064 
4065 	return (nblocks);
4066 }
4067