xref: /csrg-svn/sys/ufs/lfs/lfs_alloc.c (revision 24839)
1 /*
2  * Copyright (c) 1982 Regents of the University of California.
3  * All rights reserved.  The Berkeley software License Agreement
4  * specifies the terms and conditions for redistribution.
5  *
6  *	@(#)lfs_alloc.c	6.16 (Berkeley) 09/17/85
7  */
8 
9 #include "param.h"
10 #include "systm.h"
11 #include "mount.h"
12 #include "fs.h"
13 #include "buf.h"
14 #include "inode.h"
15 #include "dir.h"
16 #include "user.h"
17 #include "quota.h"
18 #include "kernel.h"
19 #include "syslog.h"
20 
21 extern u_long		hashalloc();
22 extern ino_t		ialloccg();
23 extern daddr_t		alloccg();
24 extern daddr_t		alloccgblk();
25 extern daddr_t		fragextend();
26 extern daddr_t		blkpref();
27 extern daddr_t		mapsearch();
28 extern int		inside[], around[];
29 extern unsigned char	*fragtbl[];
30 
31 /*
32  * Allocate a block in the file system.
33  *
34  * The size of the requested block is given, which must be some
35  * multiple of fs_fsize and <= fs_bsize.
36  * A preference may be optionally specified. If a preference is given
37  * the following hierarchy is used to allocate a block:
38  *   1) allocate the requested block.
39  *   2) allocate a rotationally optimal block in the same cylinder.
40  *   3) allocate a block in the same cylinder group.
41  *   4) quadradically rehash into other cylinder groups, until an
42  *      available block is located.
43  * If no block preference is given the following heirarchy is used
44  * to allocate a block:
45  *   1) allocate a block in the cylinder group that contains the
46  *      inode for the file.
47  *   2) quadradically rehash into other cylinder groups, until an
48  *      available block is located.
49  */
50 struct buf *
51 alloc(ip, bpref, size)
52 	register struct inode *ip;
53 	daddr_t bpref;
54 	int size;
55 {
56 	daddr_t bno;
57 	register struct fs *fs;
58 	register struct buf *bp;
59 	int cg;
60 
61 	fs = ip->i_fs;
62 	if ((unsigned)size > fs->fs_bsize || fragoff(fs, size) != 0) {
63 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
64 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
65 		panic("alloc: bad size");
66 	}
67 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
68 		goto nospace;
69 	if (u.u_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
70 		goto nospace;
71 #ifdef QUOTA
72 	u.u_error = chkdq(ip, (long)btodb(size), 0);
73 	if (u.u_error)
74 		return (NULL);
75 #endif
76 	if (bpref >= fs->fs_size)
77 		bpref = 0;
78 	if (bpref == 0)
79 		cg = itog(fs, ip->i_number);
80 	else
81 		cg = dtog(fs, bpref);
82 	bno = (daddr_t)hashalloc(ip, cg, (long)bpref, size,
83 		(u_long (*)())alloccg);
84 	if (bno <= 0)
85 		goto nospace;
86 	ip->i_blocks += btodb(size);
87 	ip->i_flag |= IUPD|ICHG;
88 	bp = getblk(ip->i_dev, fsbtodb(fs, bno), size);
89 	clrbuf(bp);
90 	return (bp);
91 nospace:
92 	fserr(fs, "file system full");
93 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
94 	u.u_error = ENOSPC;
95 	return (NULL);
96 }
97 
98 /*
99  * Reallocate a fragment to a bigger size
100  *
101  * The number and size of the old block is given, and a preference
102  * and new size is also specified. The allocator attempts to extend
103  * the original block. Failing that, the regular block allocator is
104  * invoked to get an appropriate block.
105  */
106 struct buf *
107 realloccg(ip, bprev, bpref, osize, nsize)
108 	register struct inode *ip;
109 	daddr_t bprev, bpref;
110 	int osize, nsize;
111 {
112 	daddr_t bno;
113 	register struct fs *fs;
114 	register struct buf *bp, *obp;
115 	int cg, request;
116 
117 	fs = ip->i_fs;
118 	if ((unsigned)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
119 	    (unsigned)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
120 		printf("dev = 0x%x, bsize = %d, osize = %d, nsize = %d, fs = %s\n",
121 		    ip->i_dev, fs->fs_bsize, osize, nsize, fs->fs_fsmnt);
122 		panic("realloccg: bad size");
123 	}
124 	if (u.u_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
125 		goto nospace;
126 	if (bprev == 0) {
127 		printf("dev = 0x%x, bsize = %d, bprev = %d, fs = %s\n",
128 		    ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
129 		panic("realloccg: bad bprev");
130 	}
131 #ifdef QUOTA
132 	u.u_error = chkdq(ip, (long)btodb(nsize - osize), 0);
133 	if (u.u_error)
134 		return (NULL);
135 #endif
136 	cg = dtog(fs, bprev);
137 	bno = fragextend(ip, cg, (long)bprev, osize, nsize);
138 	if (bno != 0) {
139 		do {
140 			bp = bread(ip->i_dev, fsbtodb(fs, bno), osize);
141 			if (bp->b_flags & B_ERROR) {
142 				brelse(bp);
143 				return (NULL);
144 			}
145 		} while (brealloc(bp, nsize) == 0);
146 		bp->b_flags |= B_DONE;
147 		bzero(bp->b_un.b_addr + osize, (unsigned)nsize - osize);
148 		ip->i_blocks += btodb(nsize - osize);
149 		ip->i_flag |= IUPD|ICHG;
150 		return (bp);
151 	}
152 	if (bpref >= fs->fs_size)
153 		bpref = 0;
154 	if (fs->fs_optim == FS_OPTSPACE)
155 		request = nsize;
156 	else /* if (fs->fs_optim == FS_OPTTIME) */
157 		request = fs->fs_bsize;
158 	bno = (daddr_t)hashalloc(ip, cg, (long)bpref, request,
159 		(u_long (*)())alloccg);
160 	if (bno > 0) {
161 		obp = bread(ip->i_dev, fsbtodb(fs, bprev), osize);
162 		if (obp->b_flags & B_ERROR) {
163 			brelse(obp);
164 			return (NULL);
165 		}
166 		bp = getblk(ip->i_dev, fsbtodb(fs, bno), nsize);
167 		bcopy(obp->b_un.b_addr, bp->b_un.b_addr, (u_int)osize);
168 		bzero(bp->b_un.b_addr + osize, (unsigned)nsize - osize);
169 		if (obp->b_flags & B_DELWRI) {
170 			obp->b_flags &= ~B_DELWRI;
171 			u.u_ru.ru_oublock--;		/* delete charge */
172 		}
173 		brelse(obp);
174 		free(ip, bprev, (off_t)osize);
175 		if (nsize < request)
176 			free(ip, bno + numfrags(fs, nsize),
177 				(off_t)(request - nsize));
178 		ip->i_blocks += btodb(nsize - osize);
179 		ip->i_flag |= IUPD|ICHG;
180 		return (bp);
181 	}
182 nospace:
183 	/*
184 	 * no space available
185 	 */
186 	fserr(fs, "file system full");
187 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
188 	u.u_error = ENOSPC;
189 	return (NULL);
190 }
191 
192 /*
193  * Allocate an inode in the file system.
194  *
195  * A preference may be optionally specified. If a preference is given
196  * the following hierarchy is used to allocate an inode:
197  *   1) allocate the requested inode.
198  *   2) allocate an inode in the same cylinder group.
199  *   3) quadradically rehash into other cylinder groups, until an
200  *      available inode is located.
201  * If no inode preference is given the following heirarchy is used
202  * to allocate an inode:
203  *   1) allocate an inode in cylinder group 0.
204  *   2) quadradically rehash into other cylinder groups, until an
205  *      available inode is located.
206  */
207 struct inode *
208 ialloc(pip, ipref, mode)
209 	register struct inode *pip;
210 	ino_t ipref;
211 	int mode;
212 {
213 	ino_t ino;
214 	register struct fs *fs;
215 	register struct inode *ip;
216 	int cg;
217 
218 	fs = pip->i_fs;
219 	if (fs->fs_cstotal.cs_nifree == 0)
220 		goto noinodes;
221 #ifdef QUOTA
222 	u.u_error = chkiq(pip->i_dev, (struct inode *)NULL, u.u_uid, 0);
223 	if (u.u_error)
224 		return (NULL);
225 #endif
226 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
227 		ipref = 0;
228 	cg = itog(fs, ipref);
229 	ino = (ino_t)hashalloc(pip, cg, (long)ipref, mode, ialloccg);
230 	if (ino == 0)
231 		goto noinodes;
232 	ip = iget(pip->i_dev, pip->i_fs, ino);
233 	if (ip == NULL) {
234 		ifree(pip, ino, 0);
235 		return (NULL);
236 	}
237 	if (ip->i_mode) {
238 		printf("mode = 0%o, inum = %d, fs = %s\n",
239 		    ip->i_mode, ip->i_number, fs->fs_fsmnt);
240 		panic("ialloc: dup alloc");
241 	}
242 	if (ip->i_blocks) {				/* XXX */
243 		printf("free inode %s/%d had %d blocks\n",
244 		    fs->fs_fsmnt, ino, ip->i_blocks);
245 		ip->i_blocks = 0;
246 	}
247 	return (ip);
248 noinodes:
249 	fserr(fs, "out of inodes");
250 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
251 	u.u_error = ENOSPC;
252 	return (NULL);
253 }
254 
255 /*
256  * Find a cylinder to place a directory.
257  *
258  * The policy implemented by this algorithm is to select from
259  * among those cylinder groups with above the average number of
260  * free inodes, the one with the smallest number of directories.
261  */
262 ino_t
263 dirpref(fs)
264 	register struct fs *fs;
265 {
266 	int cg, minndir, mincg, avgifree;
267 
268 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
269 	minndir = fs->fs_ipg;
270 	mincg = 0;
271 	for (cg = 0; cg < fs->fs_ncg; cg++)
272 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
273 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
274 			mincg = cg;
275 			minndir = fs->fs_cs(fs, cg).cs_ndir;
276 		}
277 	return ((ino_t)(fs->fs_ipg * mincg));
278 }
279 
280 /*
281  * Select the desired position for the next block in a file.  The file is
282  * logically divided into sections. The first section is composed of the
283  * direct blocks. Each additional section contains fs_maxbpg blocks.
284  *
285  * If no blocks have been allocated in the first section, the policy is to
286  * request a block in the same cylinder group as the inode that describes
287  * the file. If no blocks have been allocated in any other section, the
288  * policy is to place the section in a cylinder group with a greater than
289  * average number of free blocks.  An appropriate cylinder group is found
290  * by using a rotor that sweeps the cylinder groups. When a new group of
291  * blocks is needed, the sweep begins in the cylinder group following the
292  * cylinder group from which the previous allocation was made. The sweep
293  * continues until a cylinder group with greater than the average number
294  * of free blocks is found. If the allocation is for the first block in an
295  * indirect block, the information on the previous allocation is unavailable;
296  * here a best guess is made based upon the logical block number being
297  * allocated.
298  *
299  * If a section is already partially allocated, the policy is to
300  * contiguously allocate fs_maxcontig blocks.  The end of one of these
301  * contiguous blocks and the beginning of the next is physically separated
302  * so that the disk head will be in transit between them for at least
303  * fs_rotdelay milliseconds.  This is to allow time for the processor to
304  * schedule another I/O transfer.
305  */
306 daddr_t
307 blkpref(ip, lbn, indx, bap)
308 	struct inode *ip;
309 	daddr_t lbn;
310 	int indx;
311 	daddr_t *bap;
312 {
313 	register struct fs *fs;
314 	register int cg;
315 	int avgbfree, startcg;
316 	daddr_t nextblk;
317 
318 	fs = ip->i_fs;
319 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
320 		if (lbn < NDADDR) {
321 			cg = itog(fs, ip->i_number);
322 			return (fs->fs_fpg * cg + fs->fs_frag);
323 		}
324 		/*
325 		 * Find a cylinder with greater than average number of
326 		 * unused data blocks.
327 		 */
328 		if (indx == 0 || bap[indx - 1] == 0)
329 			startcg = itog(fs, ip->i_number) + lbn / fs->fs_maxbpg;
330 		else
331 			startcg = dtog(fs, bap[indx - 1]) + 1;
332 		startcg %= fs->fs_ncg;
333 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
334 		for (cg = startcg; cg < fs->fs_ncg; cg++)
335 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
336 				fs->fs_cgrotor = cg;
337 				return (fs->fs_fpg * cg + fs->fs_frag);
338 			}
339 		for (cg = 0; cg <= startcg; cg++)
340 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
341 				fs->fs_cgrotor = cg;
342 				return (fs->fs_fpg * cg + fs->fs_frag);
343 			}
344 		return (NULL);
345 	}
346 	/*
347 	 * One or more previous blocks have been laid out. If less
348 	 * than fs_maxcontig previous blocks are contiguous, the
349 	 * next block is requested contiguously, otherwise it is
350 	 * requested rotationally delayed by fs_rotdelay milliseconds.
351 	 */
352 	nextblk = bap[indx - 1] + fs->fs_frag;
353 	if (indx > fs->fs_maxcontig &&
354 	    bap[indx - fs->fs_maxcontig] + blkstofrags(fs, fs->fs_maxcontig)
355 	    != nextblk)
356 		return (nextblk);
357 	if (fs->fs_rotdelay != 0)
358 		/*
359 		 * Here we convert ms of delay to frags as:
360 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
361 		 *	((sect/frag) * (ms/sec))
362 		 * then round up to the next block.
363 		 */
364 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
365 		    (NSPF(fs) * 1000), fs->fs_frag);
366 	return (nextblk);
367 }
368 
369 /*
370  * Implement the cylinder overflow algorithm.
371  *
372  * The policy implemented by this algorithm is:
373  *   1) allocate the block in its requested cylinder group.
374  *   2) quadradically rehash on the cylinder group number.
375  *   3) brute force search for a free block.
376  */
377 /*VARARGS5*/
378 u_long
379 hashalloc(ip, cg, pref, size, allocator)
380 	struct inode *ip;
381 	int cg;
382 	long pref;
383 	int size;	/* size for data blocks, mode for inodes */
384 	u_long (*allocator)();
385 {
386 	register struct fs *fs;
387 	long result;
388 	int i, icg = cg;
389 
390 	fs = ip->i_fs;
391 	/*
392 	 * 1: preferred cylinder group
393 	 */
394 	result = (*allocator)(ip, cg, pref, size);
395 	if (result)
396 		return (result);
397 	/*
398 	 * 2: quadratic rehash
399 	 */
400 	for (i = 1; i < fs->fs_ncg; i *= 2) {
401 		cg += i;
402 		if (cg >= fs->fs_ncg)
403 			cg -= fs->fs_ncg;
404 		result = (*allocator)(ip, cg, 0, size);
405 		if (result)
406 			return (result);
407 	}
408 	/*
409 	 * 3: brute force search
410 	 * Note that we start at i == 2, since 0 was checked initially,
411 	 * and 1 is always checked in the quadratic rehash.
412 	 */
413 	cg = (icg + 2) % fs->fs_ncg;
414 	for (i = 2; i < fs->fs_ncg; i++) {
415 		result = (*allocator)(ip, cg, 0, size);
416 		if (result)
417 			return (result);
418 		cg++;
419 		if (cg == fs->fs_ncg)
420 			cg = 0;
421 	}
422 	return (NULL);
423 }
424 
425 /*
426  * Determine whether a fragment can be extended.
427  *
428  * Check to see if the necessary fragments are available, and
429  * if they are, allocate them.
430  */
431 daddr_t
432 fragextend(ip, cg, bprev, osize, nsize)
433 	struct inode *ip;
434 	int cg;
435 	long bprev;
436 	int osize, nsize;
437 {
438 	register struct fs *fs;
439 	register struct buf *bp;
440 	register struct cg *cgp;
441 	long bno;
442 	int frags, bbase;
443 	int i;
444 
445 	fs = ip->i_fs;
446 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
447 		return (NULL);
448 	frags = numfrags(fs, nsize);
449 	bbase = fragnum(fs, bprev);
450 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
451 		/* cannot extend across a block boundry */
452 		return (NULL);
453 	}
454 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
455 	cgp = bp->b_un.b_cg;
456 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
457 		brelse(bp);
458 		return (NULL);
459 	}
460 	cgp->cg_time = time.tv_sec;
461 	bno = dtogd(fs, bprev);
462 	for (i = numfrags(fs, osize); i < frags; i++)
463 		if (isclr(cgp->cg_free, bno + i)) {
464 			brelse(bp);
465 			return (NULL);
466 		}
467 	/*
468 	 * the current fragment can be extended
469 	 * deduct the count on fragment being extended into
470 	 * increase the count on the remaining fragment (if any)
471 	 * allocate the extended piece
472 	 */
473 	for (i = frags; i < fs->fs_frag - bbase; i++)
474 		if (isclr(cgp->cg_free, bno + i))
475 			break;
476 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
477 	if (i != frags)
478 		cgp->cg_frsum[i - frags]++;
479 	for (i = numfrags(fs, osize); i < frags; i++) {
480 		clrbit(cgp->cg_free, bno + i);
481 		cgp->cg_cs.cs_nffree--;
482 		fs->fs_cstotal.cs_nffree--;
483 		fs->fs_cs(fs, cg).cs_nffree--;
484 	}
485 	fs->fs_fmod++;
486 	bdwrite(bp);
487 	return (bprev);
488 }
489 
490 /*
491  * Determine whether a block can be allocated.
492  *
493  * Check to see if a block of the apprpriate size is available,
494  * and if it is, allocate it.
495  */
496 daddr_t
497 alloccg(ip, cg, bpref, size)
498 	struct inode *ip;
499 	int cg;
500 	daddr_t bpref;
501 	int size;
502 {
503 	register struct fs *fs;
504 	register struct buf *bp;
505 	register struct cg *cgp;
506 	int bno, frags;
507 	int allocsiz;
508 	register int i;
509 
510 	fs = ip->i_fs;
511 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
512 		return (NULL);
513 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
514 	cgp = bp->b_un.b_cg;
515 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC ||
516 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
517 		brelse(bp);
518 		return (NULL);
519 	}
520 	cgp->cg_time = time.tv_sec;
521 	if (size == fs->fs_bsize) {
522 		bno = alloccgblk(fs, cgp, bpref);
523 		bdwrite(bp);
524 		return (bno);
525 	}
526 	/*
527 	 * check to see if any fragments are already available
528 	 * allocsiz is the size which will be allocated, hacking
529 	 * it down to a smaller size if necessary
530 	 */
531 	frags = numfrags(fs, size);
532 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
533 		if (cgp->cg_frsum[allocsiz] != 0)
534 			break;
535 	if (allocsiz == fs->fs_frag) {
536 		/*
537 		 * no fragments were available, so a block will be
538 		 * allocated, and hacked up
539 		 */
540 		if (cgp->cg_cs.cs_nbfree == 0) {
541 			brelse(bp);
542 			return (NULL);
543 		}
544 		bno = alloccgblk(fs, cgp, bpref);
545 		bpref = dtogd(fs, bno);
546 		for (i = frags; i < fs->fs_frag; i++)
547 			setbit(cgp->cg_free, bpref + i);
548 		i = fs->fs_frag - frags;
549 		cgp->cg_cs.cs_nffree += i;
550 		fs->fs_cstotal.cs_nffree += i;
551 		fs->fs_cs(fs, cg).cs_nffree += i;
552 		fs->fs_fmod++;
553 		cgp->cg_frsum[i]++;
554 		bdwrite(bp);
555 		return (bno);
556 	}
557 	bno = mapsearch(fs, cgp, bpref, allocsiz);
558 	if (bno < 0) {
559 		brelse(bp);
560 		return (NULL);
561 	}
562 	for (i = 0; i < frags; i++)
563 		clrbit(cgp->cg_free, bno + i);
564 	cgp->cg_cs.cs_nffree -= frags;
565 	fs->fs_cstotal.cs_nffree -= frags;
566 	fs->fs_cs(fs, cg).cs_nffree -= frags;
567 	fs->fs_fmod++;
568 	cgp->cg_frsum[allocsiz]--;
569 	if (frags != allocsiz)
570 		cgp->cg_frsum[allocsiz - frags]++;
571 	bdwrite(bp);
572 	return (cg * fs->fs_fpg + bno);
573 }
574 
575 /*
576  * Allocate a block in a cylinder group.
577  *
578  * This algorithm implements the following policy:
579  *   1) allocate the requested block.
580  *   2) allocate a rotationally optimal block in the same cylinder.
581  *   3) allocate the next available block on the block rotor for the
582  *      specified cylinder group.
583  * Note that this routine only allocates fs_bsize blocks; these
584  * blocks may be fragmented by the routine that allocates them.
585  */
586 daddr_t
587 alloccgblk(fs, cgp, bpref)
588 	register struct fs *fs;
589 	register struct cg *cgp;
590 	daddr_t bpref;
591 {
592 	daddr_t bno;
593 	int cylno, pos, delta;
594 	short *cylbp;
595 	register int i;
596 
597 	if (bpref == 0) {
598 		bpref = cgp->cg_rotor;
599 		goto norot;
600 	}
601 	bpref = blknum(fs, bpref);
602 	bpref = dtogd(fs, bpref);
603 	/*
604 	 * if the requested block is available, use it
605 	 */
606 	if (isblock(fs, cgp->cg_free, fragstoblks(fs, bpref))) {
607 		bno = bpref;
608 		goto gotit;
609 	}
610 	/*
611 	 * check for a block available on the same cylinder
612 	 */
613 	cylno = cbtocylno(fs, bpref);
614 	if (cgp->cg_btot[cylno] == 0)
615 		goto norot;
616 	if (fs->fs_cpc == 0) {
617 		/*
618 		 * block layout info is not available, so just have
619 		 * to take any block in this cylinder.
620 		 */
621 		bpref = howmany(fs->fs_spc * cylno, NSPF(fs));
622 		goto norot;
623 	}
624 	/*
625 	 * check the summary information to see if a block is
626 	 * available in the requested cylinder starting at the
627 	 * requested rotational position and proceeding around.
628 	 */
629 	cylbp = cgp->cg_b[cylno];
630 	pos = cbtorpos(fs, bpref);
631 	for (i = pos; i < NRPOS; i++)
632 		if (cylbp[i] > 0)
633 			break;
634 	if (i == NRPOS)
635 		for (i = 0; i < pos; i++)
636 			if (cylbp[i] > 0)
637 				break;
638 	if (cylbp[i] > 0) {
639 		/*
640 		 * found a rotational position, now find the actual
641 		 * block. A panic if none is actually there.
642 		 */
643 		pos = cylno % fs->fs_cpc;
644 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
645 		if (fs->fs_postbl[pos][i] == -1) {
646 			printf("pos = %d, i = %d, fs = %s\n",
647 			    pos, i, fs->fs_fsmnt);
648 			panic("alloccgblk: cyl groups corrupted");
649 		}
650 		for (i = fs->fs_postbl[pos][i];; ) {
651 			if (isblock(fs, cgp->cg_free, bno + i)) {
652 				bno = blkstofrags(fs, (bno + i));
653 				goto gotit;
654 			}
655 			delta = fs->fs_rotbl[i];
656 			if (delta <= 0 || delta > MAXBPC - i)
657 				break;
658 			i += delta;
659 		}
660 		printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
661 		panic("alloccgblk: can't find blk in cyl");
662 	}
663 norot:
664 	/*
665 	 * no blocks in the requested cylinder, so take next
666 	 * available one in this cylinder group.
667 	 */
668 	bno = mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
669 	if (bno < 0)
670 		return (NULL);
671 	cgp->cg_rotor = bno;
672 gotit:
673 	clrblock(fs, cgp->cg_free, (long)fragstoblks(fs, bno));
674 	cgp->cg_cs.cs_nbfree--;
675 	fs->fs_cstotal.cs_nbfree--;
676 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
677 	cylno = cbtocylno(fs, bno);
678 	cgp->cg_b[cylno][cbtorpos(fs, bno)]--;
679 	cgp->cg_btot[cylno]--;
680 	fs->fs_fmod++;
681 	return (cgp->cg_cgx * fs->fs_fpg + bno);
682 }
683 
684 /*
685  * Determine whether an inode can be allocated.
686  *
687  * Check to see if an inode is available, and if it is,
688  * allocate it using the following policy:
689  *   1) allocate the requested inode.
690  *   2) allocate the next available inode after the requested
691  *      inode in the specified cylinder group.
692  */
693 ino_t
694 ialloccg(ip, cg, ipref, mode)
695 	struct inode *ip;
696 	int cg;
697 	daddr_t ipref;
698 	int mode;
699 {
700 	register struct fs *fs;
701 	register struct cg *cgp;
702 	struct buf *bp;
703 	int start, len, loc, map, i;
704 
705 	fs = ip->i_fs;
706 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
707 		return (NULL);
708 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
709 	cgp = bp->b_un.b_cg;
710 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC ||
711 	    cgp->cg_cs.cs_nifree == 0) {
712 		brelse(bp);
713 		return (NULL);
714 	}
715 	cgp->cg_time = time.tv_sec;
716 	if (ipref) {
717 		ipref %= fs->fs_ipg;
718 		if (isclr(cgp->cg_iused, ipref))
719 			goto gotit;
720 	}
721 	start = cgp->cg_irotor / NBBY;
722 	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
723 	loc = skpc(0xff, len, &cgp->cg_iused[start]);
724 	if (loc == 0) {
725 		len = start + 1;
726 		start = 0;
727 		loc = skpc(0xff, len, &cgp->cg_iused[0]);
728 		if (loc == 0) {
729 			printf("cg = %s, irotor = %d, fs = %s\n",
730 			    cg, cgp->cg_irotor, fs->fs_fsmnt);
731 			panic("ialloccg: map corrupted");
732 			/* NOTREACHED */
733 		}
734 	}
735 	i = start + len - loc;
736 	map = cgp->cg_iused[i];
737 	ipref = i * NBBY;
738 	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
739 		if ((map & i) == 0) {
740 			cgp->cg_irotor = ipref;
741 			goto gotit;
742 		}
743 	}
744 	printf("fs = %s\n", fs->fs_fsmnt);
745 	panic("ialloccg: block not in map");
746 	/* NOTREACHED */
747 gotit:
748 	setbit(cgp->cg_iused, ipref);
749 	cgp->cg_cs.cs_nifree--;
750 	fs->fs_cstotal.cs_nifree--;
751 	fs->fs_cs(fs, cg).cs_nifree--;
752 	fs->fs_fmod++;
753 	if ((mode & IFMT) == IFDIR) {
754 		cgp->cg_cs.cs_ndir++;
755 		fs->fs_cstotal.cs_ndir++;
756 		fs->fs_cs(fs, cg).cs_ndir++;
757 	}
758 	bdwrite(bp);
759 	return (cg * fs->fs_ipg + ipref);
760 }
761 
762 /*
763  * Free a block or fragment.
764  *
765  * The specified block or fragment is placed back in the
766  * free map. If a fragment is deallocated, a possible
767  * block reassembly is checked.
768  */
769 free(ip, bno, size)
770 	register struct inode *ip;
771 	daddr_t bno;
772 	off_t size;
773 {
774 	register struct fs *fs;
775 	register struct cg *cgp;
776 	register struct buf *bp;
777 	int cg, blk, frags, bbase;
778 	register int i;
779 
780 	fs = ip->i_fs;
781 	if ((unsigned)size > fs->fs_bsize || fragoff(fs, size) != 0) {
782 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
783 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
784 		panic("free: bad size");
785 	}
786 	cg = dtog(fs, bno);
787 	if (badblock(fs, bno)) {
788 		printf("bad block %d, ino %d\n", bno, ip->i_number);
789 		return;
790 	}
791 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
792 	cgp = bp->b_un.b_cg;
793 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
794 		brelse(bp);
795 		return;
796 	}
797 	cgp->cg_time = time.tv_sec;
798 	bno = dtogd(fs, bno);
799 	if (size == fs->fs_bsize) {
800 		if (isblock(fs, cgp->cg_free, fragstoblks(fs, bno))) {
801 			printf("dev = 0x%x, block = %d, fs = %s\n",
802 			    ip->i_dev, bno, fs->fs_fsmnt);
803 			panic("free: freeing free block");
804 		}
805 		setblock(fs, cgp->cg_free, fragstoblks(fs, bno));
806 		cgp->cg_cs.cs_nbfree++;
807 		fs->fs_cstotal.cs_nbfree++;
808 		fs->fs_cs(fs, cg).cs_nbfree++;
809 		i = cbtocylno(fs, bno);
810 		cgp->cg_b[i][cbtorpos(fs, bno)]++;
811 		cgp->cg_btot[i]++;
812 	} else {
813 		bbase = bno - fragnum(fs, bno);
814 		/*
815 		 * decrement the counts associated with the old frags
816 		 */
817 		blk = blkmap(fs, cgp->cg_free, bbase);
818 		fragacct(fs, blk, cgp->cg_frsum, -1);
819 		/*
820 		 * deallocate the fragment
821 		 */
822 		frags = numfrags(fs, size);
823 		for (i = 0; i < frags; i++) {
824 			if (isset(cgp->cg_free, bno + i)) {
825 				printf("dev = 0x%x, block = %d, fs = %s\n",
826 				    ip->i_dev, bno + i, fs->fs_fsmnt);
827 				panic("free: freeing free frag");
828 			}
829 			setbit(cgp->cg_free, bno + i);
830 		}
831 		cgp->cg_cs.cs_nffree += i;
832 		fs->fs_cstotal.cs_nffree += i;
833 		fs->fs_cs(fs, cg).cs_nffree += i;
834 		/*
835 		 * add back in counts associated with the new frags
836 		 */
837 		blk = blkmap(fs, cgp->cg_free, bbase);
838 		fragacct(fs, blk, cgp->cg_frsum, 1);
839 		/*
840 		 * if a complete block has been reassembled, account for it
841 		 */
842 		if (isblock(fs, cgp->cg_free, fragstoblks(fs, bbase))) {
843 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
844 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
845 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
846 			cgp->cg_cs.cs_nbfree++;
847 			fs->fs_cstotal.cs_nbfree++;
848 			fs->fs_cs(fs, cg).cs_nbfree++;
849 			i = cbtocylno(fs, bbase);
850 			cgp->cg_b[i][cbtorpos(fs, bbase)]++;
851 			cgp->cg_btot[i]++;
852 		}
853 	}
854 	fs->fs_fmod++;
855 	bdwrite(bp);
856 }
857 
858 /*
859  * Free an inode.
860  *
861  * The specified inode is placed back in the free map.
862  */
863 ifree(ip, ino, mode)
864 	struct inode *ip;
865 	ino_t ino;
866 	int mode;
867 {
868 	register struct fs *fs;
869 	register struct cg *cgp;
870 	register struct buf *bp;
871 	int cg;
872 
873 	fs = ip->i_fs;
874 	if ((unsigned)ino >= fs->fs_ipg*fs->fs_ncg) {
875 		printf("dev = 0x%x, ino = %d, fs = %s\n",
876 		    ip->i_dev, ino, fs->fs_fsmnt);
877 		panic("ifree: range");
878 	}
879 	cg = itog(fs, ino);
880 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
881 	cgp = bp->b_un.b_cg;
882 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
883 		brelse(bp);
884 		return;
885 	}
886 	cgp->cg_time = time.tv_sec;
887 	ino %= fs->fs_ipg;
888 	if (isclr(cgp->cg_iused, ino)) {
889 		printf("dev = 0x%x, ino = %d, fs = %s\n",
890 		    ip->i_dev, ino, fs->fs_fsmnt);
891 		panic("ifree: freeing free inode");
892 	}
893 	clrbit(cgp->cg_iused, ino);
894 	if (ino < cgp->cg_irotor)
895 		cgp->cg_irotor = ino;
896 	cgp->cg_cs.cs_nifree++;
897 	fs->fs_cstotal.cs_nifree++;
898 	fs->fs_cs(fs, cg).cs_nifree++;
899 	if ((mode & IFMT) == IFDIR) {
900 		cgp->cg_cs.cs_ndir--;
901 		fs->fs_cstotal.cs_ndir--;
902 		fs->fs_cs(fs, cg).cs_ndir--;
903 	}
904 	fs->fs_fmod++;
905 	bdwrite(bp);
906 }
907 
908 /*
909  * Find a block of the specified size in the specified cylinder group.
910  *
911  * It is a panic if a request is made to find a block if none are
912  * available.
913  */
914 daddr_t
915 mapsearch(fs, cgp, bpref, allocsiz)
916 	register struct fs *fs;
917 	register struct cg *cgp;
918 	daddr_t bpref;
919 	int allocsiz;
920 {
921 	daddr_t bno;
922 	int start, len, loc, i;
923 	int blk, field, subfield, pos;
924 
925 	/*
926 	 * find the fragment by searching through the free block
927 	 * map for an appropriate bit pattern
928 	 */
929 	if (bpref)
930 		start = dtogd(fs, bpref) / NBBY;
931 	else
932 		start = cgp->cg_frotor / NBBY;
933 	len = howmany(fs->fs_fpg, NBBY) - start;
934 	loc = scanc((unsigned)len, (caddr_t)&cgp->cg_free[start],
935 		(caddr_t)fragtbl[fs->fs_frag],
936 		(int)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
937 	if (loc == 0) {
938 		len = start + 1;
939 		start = 0;
940 		loc = scanc((unsigned)len, (caddr_t)&cgp->cg_free[0],
941 			(caddr_t)fragtbl[fs->fs_frag],
942 			(int)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
943 		if (loc == 0) {
944 			printf("start = %d, len = %d, fs = %s\n",
945 			    start, len, fs->fs_fsmnt);
946 			panic("alloccg: map corrupted");
947 			/* NOTREACHED */
948 		}
949 	}
950 	bno = (start + len - loc) * NBBY;
951 	cgp->cg_frotor = bno;
952 	/*
953 	 * found the byte in the map
954 	 * sift through the bits to find the selected frag
955 	 */
956 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
957 		blk = blkmap(fs, cgp->cg_free, bno);
958 		blk <<= 1;
959 		field = around[allocsiz];
960 		subfield = inside[allocsiz];
961 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
962 			if ((blk & field) == subfield)
963 				return (bno + pos);
964 			field <<= 1;
965 			subfield <<= 1;
966 		}
967 	}
968 	printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt);
969 	panic("alloccg: block not in map");
970 	return (-1);
971 }
972 
973 /*
974  * Fserr prints the name of a file system with an error diagnostic.
975  *
976  * The form of the error message is:
977  *	fs: error message
978  */
979 fserr(fs, cp)
980 	struct fs *fs;
981 	char *cp;
982 {
983 
984 	log(LOG_ERR, "%s: %s\n", fs->fs_fsmnt, cp);
985 }
986