xref: /csrg-svn/sys/ufs/lfs/lfs_alloc.c (revision 17658)
1 /*	lfs_alloc.c	6.8	85/01/07	*/
2 
3 #include "param.h"
4 #include "systm.h"
5 #include "mount.h"
6 #include "fs.h"
7 #include "conf.h"
8 #include "buf.h"
9 #include "inode.h"
10 #include "dir.h"
11 #include "user.h"
12 #include "quota.h"
13 #include "kernel.h"
14 
15 extern u_long		hashalloc();
16 extern ino_t		ialloccg();
17 extern daddr_t		alloccg();
18 extern daddr_t		alloccgblk();
19 extern daddr_t		fragextend();
20 extern daddr_t		blkpref();
21 extern daddr_t		mapsearch();
22 extern int		inside[], around[];
23 extern unsigned char	*fragtbl[];
24 
25 /*
26  * Allocate a block in the file system.
27  *
28  * The size of the requested block is given, which must be some
29  * multiple of fs_fsize and <= fs_bsize.
30  * A preference may be optionally specified. If a preference is given
31  * the following hierarchy is used to allocate a block:
32  *   1) allocate the requested block.
33  *   2) allocate a rotationally optimal block in the same cylinder.
34  *   3) allocate a block in the same cylinder group.
35  *   4) quadradically rehash into other cylinder groups, until an
36  *      available block is located.
37  * If no block preference is given the following heirarchy is used
38  * to allocate a block:
39  *   1) allocate a block in the cylinder group that contains the
40  *      inode for the file.
41  *   2) quadradically rehash into other cylinder groups, until an
42  *      available block is located.
43  */
44 struct buf *
45 alloc(ip, bpref, size)
46 	register struct inode *ip;
47 	daddr_t bpref;
48 	int size;
49 {
50 	daddr_t bno;
51 	register struct fs *fs;
52 	register struct buf *bp;
53 	int cg;
54 
55 	fs = ip->i_fs;
56 	if ((unsigned)size > fs->fs_bsize || fragoff(fs, size) != 0) {
57 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
58 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
59 		panic("alloc: bad size");
60 	}
61 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
62 		goto nospace;
63 	if (u.u_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
64 		goto nospace;
65 #ifdef QUOTA
66 	u.u_error = chkdq(ip, (long)btodb(size), 0);
67 	if (u.u_error)
68 		return (NULL);
69 #endif
70 	if (bpref >= fs->fs_size)
71 		bpref = 0;
72 	if (bpref == 0)
73 		cg = itog(fs, ip->i_number);
74 	else
75 		cg = dtog(fs, bpref);
76 	bno = (daddr_t)hashalloc(ip, cg, (long)bpref, size,
77 		(u_long (*)())alloccg);
78 	if (bno <= 0)
79 		goto nospace;
80 	ip->i_blocks += btodb(size);
81 	ip->i_flag |= IUPD|ICHG;
82 	bp = getblk(ip->i_dev, fsbtodb(fs, bno), size);
83 	clrbuf(bp);
84 	return (bp);
85 nospace:
86 	fserr(fs, "file system full");
87 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
88 	u.u_error = ENOSPC;
89 	return (NULL);
90 }
91 
92 /*
93  * Reallocate a fragment to a bigger size
94  *
95  * The number and size of the old block is given, and a preference
96  * and new size is also specified. The allocator attempts to extend
97  * the original block. Failing that, the regular block allocator is
98  * invoked to get an appropriate block.
99  */
100 struct buf *
101 realloccg(ip, bprev, bpref, osize, nsize)
102 	register struct inode *ip;
103 	daddr_t bprev, bpref;
104 	int osize, nsize;
105 {
106 	daddr_t bno;
107 	register struct fs *fs;
108 	register struct buf *bp, *obp;
109 	int cg;
110 
111 	fs = ip->i_fs;
112 	if ((unsigned)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
113 	    (unsigned)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
114 		printf("dev = 0x%x, bsize = %d, osize = %d, nsize = %d, fs = %s\n",
115 		    ip->i_dev, fs->fs_bsize, osize, nsize, fs->fs_fsmnt);
116 		panic("realloccg: bad size");
117 	}
118 	if (u.u_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
119 		goto nospace;
120 	if (bprev == 0) {
121 		printf("dev = 0x%x, bsize = %d, bprev = %d, fs = %s\n",
122 		    ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
123 		panic("realloccg: bad bprev");
124 	}
125 #ifdef QUOTA
126 	u.u_error = chkdq(ip, (long)btodb(nsize - osize), 0);
127 	if (u.u_error)
128 		return (NULL);
129 #endif
130 	cg = dtog(fs, bprev);
131 	bno = fragextend(ip, cg, (long)bprev, osize, nsize);
132 	if (bno != 0) {
133 		do {
134 			bp = bread(ip->i_dev, fsbtodb(fs, bno), osize);
135 			if (bp->b_flags & B_ERROR) {
136 				brelse(bp);
137 				return (NULL);
138 			}
139 		} while (brealloc(bp, nsize) == 0);
140 		bp->b_flags |= B_DONE;
141 		bzero(bp->b_un.b_addr + osize, (unsigned)nsize - osize);
142 		ip->i_blocks += btodb(nsize - osize);
143 		ip->i_flag |= IUPD|ICHG;
144 		return (bp);
145 	}
146 	if (bpref >= fs->fs_size)
147 		bpref = 0;
148 	bno = (daddr_t)hashalloc(ip, cg, (long)bpref, nsize,
149 		(u_long (*)())alloccg);
150 	if (bno > 0) {
151 		obp = bread(ip->i_dev, fsbtodb(fs, bprev), osize);
152 		if (obp->b_flags & B_ERROR) {
153 			brelse(obp);
154 			return (NULL);
155 		}
156 		bp = getblk(ip->i_dev, fsbtodb(fs, bno), nsize);
157 		bcopy(obp->b_un.b_addr, bp->b_un.b_addr, (u_int)osize);
158 		bzero(bp->b_un.b_addr + osize, (unsigned)nsize - osize);
159 		if (obp->b_flags & B_DELWRI) {
160 			obp->b_flags &= ~B_DELWRI;
161 			u.u_ru.ru_oublock--;		/* delete charge */
162 		}
163 		brelse(obp);
164 		free(ip, bprev, (off_t)osize);
165 		ip->i_blocks += btodb(nsize - osize);
166 		ip->i_flag |= IUPD|ICHG;
167 		return (bp);
168 	}
169 nospace:
170 	/*
171 	 * no space available
172 	 */
173 	fserr(fs, "file system full");
174 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
175 	u.u_error = ENOSPC;
176 	return (NULL);
177 }
178 
179 /*
180  * Allocate an inode in the file system.
181  *
182  * A preference may be optionally specified. If a preference is given
183  * the following hierarchy is used to allocate an inode:
184  *   1) allocate the requested inode.
185  *   2) allocate an inode in the same cylinder group.
186  *   3) quadradically rehash into other cylinder groups, until an
187  *      available inode is located.
188  * If no inode preference is given the following heirarchy is used
189  * to allocate an inode:
190  *   1) allocate an inode in cylinder group 0.
191  *   2) quadradically rehash into other cylinder groups, until an
192  *      available inode is located.
193  */
194 struct inode *
195 ialloc(pip, ipref, mode)
196 	register struct inode *pip;
197 	ino_t ipref;
198 	int mode;
199 {
200 	ino_t ino;
201 	register struct fs *fs;
202 	register struct inode *ip;
203 	int cg;
204 
205 	fs = pip->i_fs;
206 	if (fs->fs_cstotal.cs_nifree == 0)
207 		goto noinodes;
208 #ifdef QUOTA
209 	u.u_error = chkiq(pip->i_dev, (struct inode *)NULL, u.u_uid, 0);
210 	if (u.u_error)
211 		return (NULL);
212 #endif
213 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
214 		ipref = 0;
215 	cg = itog(fs, ipref);
216 	ino = (ino_t)hashalloc(pip, cg, (long)ipref, mode, ialloccg);
217 	if (ino == 0)
218 		goto noinodes;
219 	ip = iget(pip->i_dev, pip->i_fs, ino);
220 	if (ip == NULL) {
221 		ifree(pip, ino, 0);
222 		return (NULL);
223 	}
224 	if (ip->i_mode) {
225 		printf("mode = 0%o, inum = %d, fs = %s\n",
226 		    ip->i_mode, ip->i_number, fs->fs_fsmnt);
227 		panic("ialloc: dup alloc");
228 	}
229 	if (ip->i_blocks) {				/* XXX */
230 		printf("free inode %s/%d had %d blocks\n",
231 		    fs->fs_fsmnt, ino, ip->i_blocks);
232 		ip->i_blocks = 0;
233 	}
234 	return (ip);
235 noinodes:
236 	fserr(fs, "out of inodes");
237 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
238 	u.u_error = ENOSPC;
239 	return (NULL);
240 }
241 
242 /*
243  * Find a cylinder to place a directory.
244  *
245  * The policy implemented by this algorithm is to select from
246  * among those cylinder groups with above the average number of
247  * free inodes, the one with the smallest number of directories.
248  */
249 ino_t
250 dirpref(fs)
251 	register struct fs *fs;
252 {
253 	int cg, minndir, mincg, avgifree;
254 
255 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
256 	minndir = fs->fs_ipg;
257 	mincg = 0;
258 	for (cg = 0; cg < fs->fs_ncg; cg++)
259 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
260 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
261 			mincg = cg;
262 			minndir = fs->fs_cs(fs, cg).cs_ndir;
263 		}
264 	return ((ino_t)(fs->fs_ipg * mincg));
265 }
266 
267 /*
268  * Select the desired position for the next block in a file.  The file is
269  * logically divided into sections. The first section is composed of the
270  * direct blocks. Each additional section contains fs_maxbpg blocks.
271  *
272  * If no blocks have been allocated in the first section, the policy is to
273  * request a block in the same cylinder group as the inode that describes
274  * the file. If no blocks have been allocated in any other section, the
275  * policy is to place the section in a cylinder group with a greater than
276  * average number of free blocks.  An appropriate cylinder group is found
277  * by maintaining a rotor that sweeps the cylinder groups. When a new
278  * group of blocks is needed, the rotor is advanced until a cylinder group
279  * with greater than the average number of free blocks is found.
280  *
281  * If a section is already partially allocated, the policy is to
282  * contiguously allocate fs_maxcontig blocks.  The end of one of these
283  * contiguous blocks and the beginning of the next is physically separated
284  * so that the disk head will be in transit between them for at least
285  * fs_rotdelay milliseconds.  This is to allow time for the processor to
286  * schedule another I/O transfer.
287  */
288 daddr_t
289 blkpref(ip, lbn, indx, bap)
290 	struct inode *ip;
291 	daddr_t lbn;
292 	int indx;
293 	daddr_t *bap;
294 {
295 	register struct fs *fs;
296 	int cg, avgbfree;
297 	daddr_t nextblk;
298 
299 	fs = ip->i_fs;
300 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
301 		if (lbn < NDADDR) {
302 			cg = itog(fs, ip->i_number);
303 			return (fs->fs_fpg * cg + fs->fs_frag);
304 		}
305 		/*
306 		 * Find a cylinder with greater than average number of
307 		 * unused data blocks.
308 		 */
309 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
310 		for (cg = fs->fs_cgrotor + 1; cg < fs->fs_ncg; cg++)
311 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
312 				fs->fs_cgrotor = cg;
313 				return (fs->fs_fpg * cg + fs->fs_frag);
314 			}
315 		for (cg = 0; cg <= fs->fs_cgrotor; cg++)
316 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
317 				fs->fs_cgrotor = cg;
318 				return (fs->fs_fpg * cg + fs->fs_frag);
319 			}
320 		return (NULL);
321 	}
322 	/*
323 	 * One or more previous blocks have been laid out. If less
324 	 * than fs_maxcontig previous blocks are contiguous, the
325 	 * next block is requested contiguously, otherwise it is
326 	 * requested rotationally delayed by fs_rotdelay milliseconds.
327 	 */
328 	nextblk = bap[indx - 1] + fs->fs_frag;
329 	if (indx > fs->fs_maxcontig &&
330 	    bap[indx - fs->fs_maxcontig] + blkstofrags(fs, fs->fs_maxcontig)
331 	    != nextblk)
332 		return (nextblk);
333 	if (fs->fs_rotdelay != 0)
334 		/*
335 		 * Here we convert ms of delay to frags as:
336 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
337 		 *	((sect/frag) * (ms/sec))
338 		 * then round up to the next block.
339 		 */
340 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
341 		    (NSPF(fs) * 1000), fs->fs_frag);
342 	return (nextblk);
343 }
344 
345 /*
346  * Implement the cylinder overflow algorithm.
347  *
348  * The policy implemented by this algorithm is:
349  *   1) allocate the block in its requested cylinder group.
350  *   2) quadradically rehash on the cylinder group number.
351  *   3) brute force search for a free block.
352  */
353 /*VARARGS5*/
354 u_long
355 hashalloc(ip, cg, pref, size, allocator)
356 	struct inode *ip;
357 	int cg;
358 	long pref;
359 	int size;	/* size for data blocks, mode for inodes */
360 	u_long (*allocator)();
361 {
362 	register struct fs *fs;
363 	long result;
364 	int i, icg = cg;
365 
366 	fs = ip->i_fs;
367 	/*
368 	 * 1: preferred cylinder group
369 	 */
370 	result = (*allocator)(ip, cg, pref, size);
371 	if (result)
372 		return (result);
373 	/*
374 	 * 2: quadratic rehash
375 	 */
376 	for (i = 1; i < fs->fs_ncg; i *= 2) {
377 		cg += i;
378 		if (cg >= fs->fs_ncg)
379 			cg -= fs->fs_ncg;
380 		result = (*allocator)(ip, cg, 0, size);
381 		if (result)
382 			return (result);
383 	}
384 	/*
385 	 * 3: brute force search
386 	 * Note that we start at i == 2, since 0 was checked initially,
387 	 * and 1 is always checked in the quadratic rehash.
388 	 */
389 	cg = (icg + 2) % fs->fs_ncg;
390 	for (i = 2; i < fs->fs_ncg; i++) {
391 		result = (*allocator)(ip, cg, 0, size);
392 		if (result)
393 			return (result);
394 		cg++;
395 		if (cg == fs->fs_ncg)
396 			cg = 0;
397 	}
398 	return (NULL);
399 }
400 
401 /*
402  * Determine whether a fragment can be extended.
403  *
404  * Check to see if the necessary fragments are available, and
405  * if they are, allocate them.
406  */
407 daddr_t
408 fragextend(ip, cg, bprev, osize, nsize)
409 	struct inode *ip;
410 	int cg;
411 	long bprev;
412 	int osize, nsize;
413 {
414 	register struct fs *fs;
415 	register struct buf *bp;
416 	register struct cg *cgp;
417 	long bno;
418 	int frags, bbase;
419 	int i;
420 
421 	fs = ip->i_fs;
422 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
423 		return (NULL);
424 	frags = numfrags(fs, nsize);
425 	bbase = fragnum(fs, bprev);
426 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
427 		/* cannot extend across a block boundry */
428 		return (NULL);
429 	}
430 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
431 	cgp = bp->b_un.b_cg;
432 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
433 		brelse(bp);
434 		return (NULL);
435 	}
436 	cgp->cg_time = time.tv_sec;
437 	bno = dtogd(fs, bprev);
438 	for (i = numfrags(fs, osize); i < frags; i++)
439 		if (isclr(cgp->cg_free, bno + i)) {
440 			brelse(bp);
441 			return (NULL);
442 		}
443 	/*
444 	 * the current fragment can be extended
445 	 * deduct the count on fragment being extended into
446 	 * increase the count on the remaining fragment (if any)
447 	 * allocate the extended piece
448 	 */
449 	for (i = frags; i < fs->fs_frag - bbase; i++)
450 		if (isclr(cgp->cg_free, bno + i))
451 			break;
452 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
453 	if (i != frags)
454 		cgp->cg_frsum[i - frags]++;
455 	for (i = numfrags(fs, osize); i < frags; i++) {
456 		clrbit(cgp->cg_free, bno + i);
457 		cgp->cg_cs.cs_nffree--;
458 		fs->fs_cstotal.cs_nffree--;
459 		fs->fs_cs(fs, cg).cs_nffree--;
460 	}
461 	fs->fs_fmod++;
462 	bdwrite(bp);
463 	return (bprev);
464 }
465 
466 /*
467  * Determine whether a block can be allocated.
468  *
469  * Check to see if a block of the apprpriate size is available,
470  * and if it is, allocate it.
471  */
472 daddr_t
473 alloccg(ip, cg, bpref, size)
474 	struct inode *ip;
475 	int cg;
476 	daddr_t bpref;
477 	int size;
478 {
479 	register struct fs *fs;
480 	register struct buf *bp;
481 	register struct cg *cgp;
482 	int bno, frags;
483 	int allocsiz;
484 	register int i;
485 
486 	fs = ip->i_fs;
487 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
488 		return (NULL);
489 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
490 	cgp = bp->b_un.b_cg;
491 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC ||
492 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
493 		brelse(bp);
494 		return (NULL);
495 	}
496 	cgp->cg_time = time.tv_sec;
497 	if (size == fs->fs_bsize) {
498 		bno = alloccgblk(fs, cgp, bpref);
499 		bdwrite(bp);
500 		return (bno);
501 	}
502 	/*
503 	 * check to see if any fragments are already available
504 	 * allocsiz is the size which will be allocated, hacking
505 	 * it down to a smaller size if necessary
506 	 */
507 	frags = numfrags(fs, size);
508 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
509 		if (cgp->cg_frsum[allocsiz] != 0)
510 			break;
511 	if (allocsiz == fs->fs_frag) {
512 		/*
513 		 * no fragments were available, so a block will be
514 		 * allocated, and hacked up
515 		 */
516 		if (cgp->cg_cs.cs_nbfree == 0) {
517 			brelse(bp);
518 			return (NULL);
519 		}
520 		bno = alloccgblk(fs, cgp, bpref);
521 		bpref = dtogd(fs, bno);
522 		for (i = frags; i < fs->fs_frag; i++)
523 			setbit(cgp->cg_free, bpref + i);
524 		i = fs->fs_frag - frags;
525 		cgp->cg_cs.cs_nffree += i;
526 		fs->fs_cstotal.cs_nffree += i;
527 		fs->fs_cs(fs, cg).cs_nffree += i;
528 		fs->fs_fmod++;
529 		cgp->cg_frsum[i]++;
530 		bdwrite(bp);
531 		return (bno);
532 	}
533 	bno = mapsearch(fs, cgp, bpref, allocsiz);
534 	if (bno < 0) {
535 		brelse(bp);
536 		return (NULL);
537 	}
538 	for (i = 0; i < frags; i++)
539 		clrbit(cgp->cg_free, bno + i);
540 	cgp->cg_cs.cs_nffree -= frags;
541 	fs->fs_cstotal.cs_nffree -= frags;
542 	fs->fs_cs(fs, cg).cs_nffree -= frags;
543 	fs->fs_fmod++;
544 	cgp->cg_frsum[allocsiz]--;
545 	if (frags != allocsiz)
546 		cgp->cg_frsum[allocsiz - frags]++;
547 	bdwrite(bp);
548 	return (cg * fs->fs_fpg + bno);
549 }
550 
551 /*
552  * Allocate a block in a cylinder group.
553  *
554  * This algorithm implements the following policy:
555  *   1) allocate the requested block.
556  *   2) allocate a rotationally optimal block in the same cylinder.
557  *   3) allocate the next available block on the block rotor for the
558  *      specified cylinder group.
559  * Note that this routine only allocates fs_bsize blocks; these
560  * blocks may be fragmented by the routine that allocates them.
561  */
562 daddr_t
563 alloccgblk(fs, cgp, bpref)
564 	register struct fs *fs;
565 	register struct cg *cgp;
566 	daddr_t bpref;
567 {
568 	daddr_t bno;
569 	int cylno, pos, delta;
570 	short *cylbp;
571 	register int i;
572 
573 	if (bpref == 0) {
574 		bpref = cgp->cg_rotor;
575 		goto norot;
576 	}
577 	bpref = blknum(fs, bpref);
578 	bpref = dtogd(fs, bpref);
579 	/*
580 	 * if the requested block is available, use it
581 	 */
582 	if (isblock(fs, cgp->cg_free, fragstoblks(fs, bpref))) {
583 		bno = bpref;
584 		goto gotit;
585 	}
586 	/*
587 	 * check for a block available on the same cylinder
588 	 */
589 	cylno = cbtocylno(fs, bpref);
590 	if (cgp->cg_btot[cylno] == 0)
591 		goto norot;
592 	if (fs->fs_cpc == 0) {
593 		/*
594 		 * block layout info is not available, so just have
595 		 * to take any block in this cylinder.
596 		 */
597 		bpref = howmany(fs->fs_spc * cylno, NSPF(fs));
598 		goto norot;
599 	}
600 	/*
601 	 * check the summary information to see if a block is
602 	 * available in the requested cylinder starting at the
603 	 * requested rotational position and proceeding around.
604 	 */
605 	cylbp = cgp->cg_b[cylno];
606 	pos = cbtorpos(fs, bpref);
607 	for (i = pos; i < NRPOS; i++)
608 		if (cylbp[i] > 0)
609 			break;
610 	if (i == NRPOS)
611 		for (i = 0; i < pos; i++)
612 			if (cylbp[i] > 0)
613 				break;
614 	if (cylbp[i] > 0) {
615 		/*
616 		 * found a rotational position, now find the actual
617 		 * block. A panic if none is actually there.
618 		 */
619 		pos = cylno % fs->fs_cpc;
620 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
621 		if (fs->fs_postbl[pos][i] == -1) {
622 			printf("pos = %d, i = %d, fs = %s\n",
623 			    pos, i, fs->fs_fsmnt);
624 			panic("alloccgblk: cyl groups corrupted");
625 		}
626 		for (i = fs->fs_postbl[pos][i];; ) {
627 			if (isblock(fs, cgp->cg_free, bno + i)) {
628 				bno = blkstofrags(fs, (bno + i));
629 				goto gotit;
630 			}
631 			delta = fs->fs_rotbl[i];
632 			if (delta <= 0 || delta > MAXBPC - i)
633 				break;
634 			i += delta;
635 		}
636 		printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
637 		panic("alloccgblk: can't find blk in cyl");
638 	}
639 norot:
640 	/*
641 	 * no blocks in the requested cylinder, so take next
642 	 * available one in this cylinder group.
643 	 */
644 	bno = mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
645 	if (bno < 0)
646 		return (NULL);
647 	cgp->cg_rotor = bno;
648 gotit:
649 	clrblock(fs, cgp->cg_free, (long)fragstoblks(fs, bno));
650 	cgp->cg_cs.cs_nbfree--;
651 	fs->fs_cstotal.cs_nbfree--;
652 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
653 	cylno = cbtocylno(fs, bno);
654 	cgp->cg_b[cylno][cbtorpos(fs, bno)]--;
655 	cgp->cg_btot[cylno]--;
656 	fs->fs_fmod++;
657 	return (cgp->cg_cgx * fs->fs_fpg + bno);
658 }
659 
660 /*
661  * Determine whether an inode can be allocated.
662  *
663  * Check to see if an inode is available, and if it is,
664  * allocate it using the following policy:
665  *   1) allocate the requested inode.
666  *   2) allocate the next available inode after the requested
667  *      inode in the specified cylinder group.
668  */
669 ino_t
670 ialloccg(ip, cg, ipref, mode)
671 	struct inode *ip;
672 	int cg;
673 	daddr_t ipref;
674 	int mode;
675 {
676 	register struct fs *fs;
677 	register struct cg *cgp;
678 	struct buf *bp;
679 	int start, len, loc, map, i;
680 
681 	fs = ip->i_fs;
682 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
683 		return (NULL);
684 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
685 	cgp = bp->b_un.b_cg;
686 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC ||
687 	    cgp->cg_cs.cs_nifree == 0) {
688 		brelse(bp);
689 		return (NULL);
690 	}
691 	cgp->cg_time = time.tv_sec;
692 	if (ipref) {
693 		ipref %= fs->fs_ipg;
694 		if (isclr(cgp->cg_iused, ipref))
695 			goto gotit;
696 	}
697 	start = cgp->cg_irotor / NBBY;
698 	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
699 	loc = skpc(0xff, len, &cgp->cg_iused[start]);
700 	if (loc == 0) {
701 		printf("cg = %s, irotor = %d, fs = %s\n",
702 		    cg, cgp->cg_irotor, fs->fs_fsmnt);
703 		panic("ialloccg: map corrupted");
704 		/* NOTREACHED */
705 	}
706 	i = start + len - loc;
707 	map = cgp->cg_iused[i];
708 	ipref = i * NBBY;
709 	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
710 		if ((map & i) == 0) {
711 			cgp->cg_irotor = ipref;
712 			goto gotit;
713 		}
714 	}
715 	printf("fs = %s\n", fs->fs_fsmnt);
716 	panic("ialloccg: block not in map");
717 	/* NOTREACHED */
718 gotit:
719 	setbit(cgp->cg_iused, ipref);
720 	cgp->cg_cs.cs_nifree--;
721 	fs->fs_cstotal.cs_nifree--;
722 	fs->fs_cs(fs, cg).cs_nifree--;
723 	fs->fs_fmod++;
724 	if ((mode & IFMT) == IFDIR) {
725 		cgp->cg_cs.cs_ndir++;
726 		fs->fs_cstotal.cs_ndir++;
727 		fs->fs_cs(fs, cg).cs_ndir++;
728 	}
729 	bdwrite(bp);
730 	return (cg * fs->fs_ipg + ipref);
731 }
732 
733 /*
734  * Free a block or fragment.
735  *
736  * The specified block or fragment is placed back in the
737  * free map. If a fragment is deallocated, a possible
738  * block reassembly is checked.
739  */
740 free(ip, bno, size)
741 	register struct inode *ip;
742 	daddr_t bno;
743 	off_t size;
744 {
745 	register struct fs *fs;
746 	register struct cg *cgp;
747 	register struct buf *bp;
748 	int cg, blk, frags, bbase;
749 	register int i;
750 
751 	fs = ip->i_fs;
752 	if ((unsigned)size > fs->fs_bsize || fragoff(fs, size) != 0) {
753 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
754 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
755 		panic("free: bad size");
756 	}
757 	cg = dtog(fs, bno);
758 	if (badblock(fs, bno)) {
759 		printf("bad block %d, ino %d\n", bno, ip->i_number);
760 		return;
761 	}
762 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
763 	cgp = bp->b_un.b_cg;
764 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
765 		brelse(bp);
766 		return;
767 	}
768 	cgp->cg_time = time.tv_sec;
769 	bno = dtogd(fs, bno);
770 	if (size == fs->fs_bsize) {
771 		if (isblock(fs, cgp->cg_free, fragstoblks(fs, bno))) {
772 			printf("dev = 0x%x, block = %d, fs = %s\n",
773 			    ip->i_dev, bno, fs->fs_fsmnt);
774 			panic("free: freeing free block");
775 		}
776 		setblock(fs, cgp->cg_free, fragstoblks(fs, bno));
777 		cgp->cg_cs.cs_nbfree++;
778 		fs->fs_cstotal.cs_nbfree++;
779 		fs->fs_cs(fs, cg).cs_nbfree++;
780 		i = cbtocylno(fs, bno);
781 		cgp->cg_b[i][cbtorpos(fs, bno)]++;
782 		cgp->cg_btot[i]++;
783 	} else {
784 		bbase = bno - fragnum(fs, bno);
785 		/*
786 		 * decrement the counts associated with the old frags
787 		 */
788 		blk = blkmap(fs, cgp->cg_free, bbase);
789 		fragacct(fs, blk, cgp->cg_frsum, -1);
790 		/*
791 		 * deallocate the fragment
792 		 */
793 		frags = numfrags(fs, size);
794 		for (i = 0; i < frags; i++) {
795 			if (isset(cgp->cg_free, bno + i)) {
796 				printf("dev = 0x%x, block = %d, fs = %s\n",
797 				    ip->i_dev, bno + i, fs->fs_fsmnt);
798 				panic("free: freeing free frag");
799 			}
800 			setbit(cgp->cg_free, bno + i);
801 		}
802 		cgp->cg_cs.cs_nffree += i;
803 		fs->fs_cstotal.cs_nffree += i;
804 		fs->fs_cs(fs, cg).cs_nffree += i;
805 		/*
806 		 * add back in counts associated with the new frags
807 		 */
808 		blk = blkmap(fs, cgp->cg_free, bbase);
809 		fragacct(fs, blk, cgp->cg_frsum, 1);
810 		/*
811 		 * if a complete block has been reassembled, account for it
812 		 */
813 		if (isblock(fs, cgp->cg_free, fragstoblks(fs, bbase))) {
814 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
815 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
816 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
817 			cgp->cg_cs.cs_nbfree++;
818 			fs->fs_cstotal.cs_nbfree++;
819 			fs->fs_cs(fs, cg).cs_nbfree++;
820 			i = cbtocylno(fs, bbase);
821 			cgp->cg_b[i][cbtorpos(fs, bbase)]++;
822 			cgp->cg_btot[i]++;
823 		}
824 	}
825 	fs->fs_fmod++;
826 	bdwrite(bp);
827 }
828 
829 /*
830  * Free an inode.
831  *
832  * The specified inode is placed back in the free map.
833  */
834 ifree(ip, ino, mode)
835 	struct inode *ip;
836 	ino_t ino;
837 	int mode;
838 {
839 	register struct fs *fs;
840 	register struct cg *cgp;
841 	register struct buf *bp;
842 	int cg;
843 
844 	fs = ip->i_fs;
845 	if ((unsigned)ino >= fs->fs_ipg*fs->fs_ncg) {
846 		printf("dev = 0x%x, ino = %d, fs = %s\n",
847 		    ip->i_dev, ino, fs->fs_fsmnt);
848 		panic("ifree: range");
849 	}
850 	cg = itog(fs, ino);
851 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
852 	cgp = bp->b_un.b_cg;
853 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
854 		brelse(bp);
855 		return;
856 	}
857 	cgp->cg_time = time.tv_sec;
858 	ino %= fs->fs_ipg;
859 	if (isclr(cgp->cg_iused, ino)) {
860 		printf("dev = 0x%x, ino = %d, fs = %s\n",
861 		    ip->i_dev, ino, fs->fs_fsmnt);
862 		panic("ifree: freeing free inode");
863 	}
864 	clrbit(cgp->cg_iused, ino);
865 	if (ino < cgp->cg_irotor)
866 		cgp->cg_irotor = ino;
867 	cgp->cg_cs.cs_nifree++;
868 	fs->fs_cstotal.cs_nifree++;
869 	fs->fs_cs(fs, cg).cs_nifree++;
870 	if ((mode & IFMT) == IFDIR) {
871 		cgp->cg_cs.cs_ndir--;
872 		fs->fs_cstotal.cs_ndir--;
873 		fs->fs_cs(fs, cg).cs_ndir--;
874 	}
875 	fs->fs_fmod++;
876 	bdwrite(bp);
877 }
878 
879 /*
880  * Find a block of the specified size in the specified cylinder group.
881  *
882  * It is a panic if a request is made to find a block if none are
883  * available.
884  */
885 daddr_t
886 mapsearch(fs, cgp, bpref, allocsiz)
887 	register struct fs *fs;
888 	register struct cg *cgp;
889 	daddr_t bpref;
890 	int allocsiz;
891 {
892 	daddr_t bno;
893 	int start, len, loc, i;
894 	int blk, field, subfield, pos;
895 
896 	/*
897 	 * find the fragment by searching through the free block
898 	 * map for an appropriate bit pattern
899 	 */
900 	if (bpref)
901 		start = dtogd(fs, bpref) / NBBY;
902 	else
903 		start = cgp->cg_frotor / NBBY;
904 	len = howmany(fs->fs_fpg, NBBY) - start;
905 	loc = scanc((unsigned)len, (caddr_t)&cgp->cg_free[start],
906 		(caddr_t)fragtbl[fs->fs_frag],
907 		(int)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
908 	if (loc == 0) {
909 		len = start + 1;
910 		start = 0;
911 		loc = scanc((unsigned)len, (caddr_t)&cgp->cg_free[start],
912 			(caddr_t)fragtbl[fs->fs_frag],
913 			(int)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
914 		if (loc == 0) {
915 			printf("start = %d, len = %d, fs = %s\n",
916 			    start, len, fs->fs_fsmnt);
917 			panic("alloccg: map corrupted");
918 			return (-1);
919 		}
920 	}
921 	bno = (start + len - loc) * NBBY;
922 	cgp->cg_frotor = bno;
923 	/*
924 	 * found the byte in the map
925 	 * sift through the bits to find the selected frag
926 	 */
927 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
928 		blk = blkmap(fs, cgp->cg_free, bno);
929 		blk <<= 1;
930 		field = around[allocsiz];
931 		subfield = inside[allocsiz];
932 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
933 			if ((blk & field) == subfield)
934 				return (bno + pos);
935 			field <<= 1;
936 			subfield <<= 1;
937 		}
938 	}
939 	printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt);
940 	panic("alloccg: block not in map");
941 	return (-1);
942 }
943 
944 /*
945  * Fserr prints the name of a file system with an error diagnostic.
946  *
947  * The form of the error message is:
948  *	fs: error message
949  */
950 fserr(fs, cp)
951 	struct fs *fs;
952 	char *cp;
953 {
954 
955 	printf("%s: %s\n", fs->fs_fsmnt, cp);
956 }
957