xref: /csrg-svn/sys/ufs/ffs/ffs_alloc.c (revision 17494)
1 /*	ffs_alloc.c	6.7	84/12/10	*/
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), osize);
157 		swapbuf(obp, bp);
158 		obp->b_flags &= ~B_DELWRI;
159 		obp->b_flags |= B_INVAL;
160 		brelse(obp);
161 		free(ip, bprev, (off_t)osize);
162 		while (brealloc(bp, nsize) == 0) {
163 			bp = bread(ip->i_dev, fsbtodb(fs, bno), osize);
164 			if (bp->b_flags & B_ERROR) {
165 				brelse(bp);
166 				return (NULL);
167 			}
168 		}
169 		bzero(bp->b_un.b_addr + osize, (unsigned)nsize - osize);
170 		ip->i_blocks += btodb(nsize - osize);
171 		ip->i_flag |= IUPD|ICHG;
172 		return (bp);
173 	}
174 nospace:
175 	/*
176 	 * no space available
177 	 */
178 	fserr(fs, "file system full");
179 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
180 	u.u_error = ENOSPC;
181 	return (NULL);
182 }
183 
184 /*
185  * Allocate an inode in the file system.
186  *
187  * A preference may be optionally specified. If a preference is given
188  * the following hierarchy is used to allocate an inode:
189  *   1) allocate the requested inode.
190  *   2) allocate an inode in the same cylinder group.
191  *   3) quadradically rehash into other cylinder groups, until an
192  *      available inode is located.
193  * If no inode preference is given the following heirarchy is used
194  * to allocate an inode:
195  *   1) allocate an inode in cylinder group 0.
196  *   2) quadradically rehash into other cylinder groups, until an
197  *      available inode is located.
198  */
199 struct inode *
200 ialloc(pip, ipref, mode)
201 	register struct inode *pip;
202 	ino_t ipref;
203 	int mode;
204 {
205 	ino_t ino;
206 	register struct fs *fs;
207 	register struct inode *ip;
208 	int cg;
209 
210 	fs = pip->i_fs;
211 	if (fs->fs_cstotal.cs_nifree == 0)
212 		goto noinodes;
213 #ifdef QUOTA
214 	u.u_error = chkiq(pip->i_dev, (struct inode *)NULL, u.u_uid, 0);
215 	if (u.u_error)
216 		return (NULL);
217 #endif
218 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
219 		ipref = 0;
220 	cg = itog(fs, ipref);
221 	ino = (ino_t)hashalloc(pip, cg, (long)ipref, mode, ialloccg);
222 	if (ino == 0)
223 		goto noinodes;
224 	ip = iget(pip->i_dev, pip->i_fs, ino);
225 	if (ip == NULL) {
226 		ifree(pip, ino, 0);
227 		return (NULL);
228 	}
229 	if (ip->i_mode) {
230 		printf("mode = 0%o, inum = %d, fs = %s\n",
231 		    ip->i_mode, ip->i_number, fs->fs_fsmnt);
232 		panic("ialloc: dup alloc");
233 	}
234 	if (ip->i_blocks) {				/* XXX */
235 		printf("free inode %s/%d had %d blocks\n",
236 		    fs->fs_fsmnt, ino, ip->i_blocks);
237 		ip->i_blocks = 0;
238 	}
239 	return (ip);
240 noinodes:
241 	fserr(fs, "out of inodes");
242 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
243 	u.u_error = ENOSPC;
244 	return (NULL);
245 }
246 
247 /*
248  * Find a cylinder to place a directory.
249  *
250  * The policy implemented by this algorithm is to select from
251  * among those cylinder groups with above the average number of
252  * free inodes, the one with the smallest number of directories.
253  */
254 ino_t
255 dirpref(fs)
256 	register struct fs *fs;
257 {
258 	int cg, minndir, mincg, avgifree;
259 
260 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
261 	minndir = fs->fs_ipg;
262 	mincg = 0;
263 	for (cg = 0; cg < fs->fs_ncg; cg++)
264 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
265 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
266 			mincg = cg;
267 			minndir = fs->fs_cs(fs, cg).cs_ndir;
268 		}
269 	return ((ino_t)(fs->fs_ipg * mincg));
270 }
271 
272 /*
273  * Select the desired position for the next block in a file.  The file is
274  * logically divided into sections. The first section is composed of the
275  * direct blocks. Each additional section contains fs_maxbpg blocks.
276  *
277  * If no blocks have been allocated in the first section, the policy is to
278  * request a block in the same cylinder group as the inode that describes
279  * the file. If no blocks have been allocated in any other section, the
280  * policy is to place the section in a cylinder group with a greater than
281  * average number of free blocks.  An appropriate cylinder group is found
282  * by maintaining a rotor that sweeps the cylinder groups. When a new
283  * group of blocks is needed, the rotor is advanced until a cylinder group
284  * with greater than the average number of free blocks is found.
285  *
286  * If a section is already partially allocated, the policy is to
287  * contiguously allocate fs_maxcontig blocks.  The end of one of these
288  * contiguous blocks and the beginning of the next is physically separated
289  * so that the disk head will be in transit between them for at least
290  * fs_rotdelay milliseconds.  This is to allow time for the processor to
291  * schedule another I/O transfer.
292  */
293 daddr_t
294 blkpref(ip, lbn, indx, bap)
295 	struct inode *ip;
296 	daddr_t lbn;
297 	int indx;
298 	daddr_t *bap;
299 {
300 	register struct fs *fs;
301 	int cg, avgbfree;
302 	daddr_t nextblk;
303 
304 	fs = ip->i_fs;
305 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
306 		if (lbn < NDADDR) {
307 			cg = itog(fs, ip->i_number);
308 			return (fs->fs_fpg * cg + fs->fs_frag);
309 		}
310 		/*
311 		 * Find a cylinder with greater than average number of
312 		 * unused data blocks.
313 		 */
314 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
315 		for (cg = fs->fs_cgrotor + 1; cg < fs->fs_ncg; 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 		for (cg = 0; cg <= fs->fs_cgrotor; cg++)
321 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
322 				fs->fs_cgrotor = cg;
323 				return (fs->fs_fpg * cg + fs->fs_frag);
324 			}
325 		return (NULL);
326 	}
327 	/*
328 	 * One or more previous blocks have been laid out. If less
329 	 * than fs_maxcontig previous blocks are contiguous, the
330 	 * next block is requested contiguously, otherwise it is
331 	 * requested rotationally delayed by fs_rotdelay milliseconds.
332 	 */
333 	nextblk = bap[indx - 1] + fs->fs_frag;
334 	if (indx > fs->fs_maxcontig &&
335 	    bap[indx - fs->fs_maxcontig] + blkstofrags(fs, fs->fs_maxcontig)
336 	    != nextblk)
337 		return (nextblk);
338 	if (fs->fs_rotdelay != 0)
339 		/*
340 		 * Here we convert ms of delay to frags as:
341 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
342 		 *	((sect/frag) * (ms/sec))
343 		 * then round up to the next block.
344 		 */
345 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
346 		    (NSPF(fs) * 1000), fs->fs_frag);
347 	return (nextblk);
348 }
349 
350 /*
351  * Implement the cylinder overflow algorithm.
352  *
353  * The policy implemented by this algorithm is:
354  *   1) allocate the block in its requested cylinder group.
355  *   2) quadradically rehash on the cylinder group number.
356  *   3) brute force search for a free block.
357  */
358 /*VARARGS5*/
359 u_long
360 hashalloc(ip, cg, pref, size, allocator)
361 	struct inode *ip;
362 	int cg;
363 	long pref;
364 	int size;	/* size for data blocks, mode for inodes */
365 	u_long (*allocator)();
366 {
367 	register struct fs *fs;
368 	long result;
369 	int i, icg = cg;
370 
371 	fs = ip->i_fs;
372 	/*
373 	 * 1: preferred cylinder group
374 	 */
375 	result = (*allocator)(ip, cg, pref, size);
376 	if (result)
377 		return (result);
378 	/*
379 	 * 2: quadratic rehash
380 	 */
381 	for (i = 1; i < fs->fs_ncg; i *= 2) {
382 		cg += i;
383 		if (cg >= fs->fs_ncg)
384 			cg -= fs->fs_ncg;
385 		result = (*allocator)(ip, cg, 0, size);
386 		if (result)
387 			return (result);
388 	}
389 	/*
390 	 * 3: brute force search
391 	 * Note that we start at i == 2, since 0 was checked initially,
392 	 * and 1 is always checked in the quadratic rehash.
393 	 */
394 	cg = (icg + 2) % fs->fs_ncg;
395 	for (i = 2; i < fs->fs_ncg; i++) {
396 		result = (*allocator)(ip, cg, 0, size);
397 		if (result)
398 			return (result);
399 		cg++;
400 		if (cg == fs->fs_ncg)
401 			cg = 0;
402 	}
403 	return (NULL);
404 }
405 
406 /*
407  * Determine whether a fragment can be extended.
408  *
409  * Check to see if the necessary fragments are available, and
410  * if they are, allocate them.
411  */
412 daddr_t
413 fragextend(ip, cg, bprev, osize, nsize)
414 	struct inode *ip;
415 	int cg;
416 	long bprev;
417 	int osize, nsize;
418 {
419 	register struct fs *fs;
420 	register struct buf *bp;
421 	register struct cg *cgp;
422 	long bno;
423 	int frags, bbase;
424 	int i;
425 
426 	fs = ip->i_fs;
427 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
428 		return (NULL);
429 	frags = numfrags(fs, nsize);
430 	bbase = fragnum(fs, bprev);
431 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
432 		/* cannot extend across a block boundry */
433 		return (NULL);
434 	}
435 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
436 	cgp = bp->b_un.b_cg;
437 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
438 		brelse(bp);
439 		return (NULL);
440 	}
441 	cgp->cg_time = time.tv_sec;
442 	bno = dtogd(fs, bprev);
443 	for (i = numfrags(fs, osize); i < frags; i++)
444 		if (isclr(cgp->cg_free, bno + i)) {
445 			brelse(bp);
446 			return (NULL);
447 		}
448 	/*
449 	 * the current fragment can be extended
450 	 * deduct the count on fragment being extended into
451 	 * increase the count on the remaining fragment (if any)
452 	 * allocate the extended piece
453 	 */
454 	for (i = frags; i < fs->fs_frag - bbase; i++)
455 		if (isclr(cgp->cg_free, bno + i))
456 			break;
457 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
458 	if (i != frags)
459 		cgp->cg_frsum[i - frags]++;
460 	for (i = numfrags(fs, osize); i < frags; i++) {
461 		clrbit(cgp->cg_free, bno + i);
462 		cgp->cg_cs.cs_nffree--;
463 		fs->fs_cstotal.cs_nffree--;
464 		fs->fs_cs(fs, cg).cs_nffree--;
465 	}
466 	fs->fs_fmod++;
467 	bdwrite(bp);
468 	return (bprev);
469 }
470 
471 /*
472  * Determine whether a block can be allocated.
473  *
474  * Check to see if a block of the apprpriate size is available,
475  * and if it is, allocate it.
476  */
477 daddr_t
478 alloccg(ip, cg, bpref, size)
479 	struct inode *ip;
480 	int cg;
481 	daddr_t bpref;
482 	int size;
483 {
484 	register struct fs *fs;
485 	register struct buf *bp;
486 	register struct cg *cgp;
487 	int bno, frags;
488 	int allocsiz;
489 	register int i;
490 
491 	fs = ip->i_fs;
492 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
493 		return (NULL);
494 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
495 	cgp = bp->b_un.b_cg;
496 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC ||
497 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
498 		brelse(bp);
499 		return (NULL);
500 	}
501 	cgp->cg_time = time.tv_sec;
502 	if (size == fs->fs_bsize) {
503 		bno = alloccgblk(fs, cgp, bpref);
504 		bdwrite(bp);
505 		return (bno);
506 	}
507 	/*
508 	 * check to see if any fragments are already available
509 	 * allocsiz is the size which will be allocated, hacking
510 	 * it down to a smaller size if necessary
511 	 */
512 	frags = numfrags(fs, size);
513 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
514 		if (cgp->cg_frsum[allocsiz] != 0)
515 			break;
516 	if (allocsiz == fs->fs_frag) {
517 		/*
518 		 * no fragments were available, so a block will be
519 		 * allocated, and hacked up
520 		 */
521 		if (cgp->cg_cs.cs_nbfree == 0) {
522 			brelse(bp);
523 			return (NULL);
524 		}
525 		bno = alloccgblk(fs, cgp, bpref);
526 		bpref = dtogd(fs, bno);
527 		for (i = frags; i < fs->fs_frag; i++)
528 			setbit(cgp->cg_free, bpref + i);
529 		i = fs->fs_frag - frags;
530 		cgp->cg_cs.cs_nffree += i;
531 		fs->fs_cstotal.cs_nffree += i;
532 		fs->fs_cs(fs, cg).cs_nffree += i;
533 		fs->fs_fmod++;
534 		cgp->cg_frsum[i]++;
535 		bdwrite(bp);
536 		return (bno);
537 	}
538 	bno = mapsearch(fs, cgp, bpref, allocsiz);
539 	if (bno < 0) {
540 		brelse(bp);
541 		return (NULL);
542 	}
543 	for (i = 0; i < frags; i++)
544 		clrbit(cgp->cg_free, bno + i);
545 	cgp->cg_cs.cs_nffree -= frags;
546 	fs->fs_cstotal.cs_nffree -= frags;
547 	fs->fs_cs(fs, cg).cs_nffree -= frags;
548 	fs->fs_fmod++;
549 	cgp->cg_frsum[allocsiz]--;
550 	if (frags != allocsiz)
551 		cgp->cg_frsum[allocsiz - frags]++;
552 	bdwrite(bp);
553 	return (cg * fs->fs_fpg + bno);
554 }
555 
556 /*
557  * Allocate a block in a cylinder group.
558  *
559  * This algorithm implements the following policy:
560  *   1) allocate the requested block.
561  *   2) allocate a rotationally optimal block in the same cylinder.
562  *   3) allocate the next available block on the block rotor for the
563  *      specified cylinder group.
564  * Note that this routine only allocates fs_bsize blocks; these
565  * blocks may be fragmented by the routine that allocates them.
566  */
567 daddr_t
568 alloccgblk(fs, cgp, bpref)
569 	register struct fs *fs;
570 	register struct cg *cgp;
571 	daddr_t bpref;
572 {
573 	daddr_t bno;
574 	int cylno, pos, delta;
575 	short *cylbp;
576 	register int i;
577 
578 	if (bpref == 0) {
579 		bpref = cgp->cg_rotor;
580 		goto norot;
581 	}
582 	bpref = blknum(fs, bpref);
583 	bpref = dtogd(fs, bpref);
584 	/*
585 	 * if the requested block is available, use it
586 	 */
587 	if (isblock(fs, cgp->cg_free, fragstoblks(fs, bpref))) {
588 		bno = bpref;
589 		goto gotit;
590 	}
591 	/*
592 	 * check for a block available on the same cylinder
593 	 */
594 	cylno = cbtocylno(fs, bpref);
595 	if (cgp->cg_btot[cylno] == 0)
596 		goto norot;
597 	if (fs->fs_cpc == 0) {
598 		/*
599 		 * block layout info is not available, so just have
600 		 * to take any block in this cylinder.
601 		 */
602 		bpref = howmany(fs->fs_spc * cylno, NSPF(fs));
603 		goto norot;
604 	}
605 	/*
606 	 * check the summary information to see if a block is
607 	 * available in the requested cylinder starting at the
608 	 * requested rotational position and proceeding around.
609 	 */
610 	cylbp = cgp->cg_b[cylno];
611 	pos = cbtorpos(fs, bpref);
612 	for (i = pos; i < NRPOS; i++)
613 		if (cylbp[i] > 0)
614 			break;
615 	if (i == NRPOS)
616 		for (i = 0; i < pos; i++)
617 			if (cylbp[i] > 0)
618 				break;
619 	if (cylbp[i] > 0) {
620 		/*
621 		 * found a rotational position, now find the actual
622 		 * block. A panic if none is actually there.
623 		 */
624 		pos = cylno % fs->fs_cpc;
625 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
626 		if (fs->fs_postbl[pos][i] == -1) {
627 			printf("pos = %d, i = %d, fs = %s\n",
628 			    pos, i, fs->fs_fsmnt);
629 			panic("alloccgblk: cyl groups corrupted");
630 		}
631 		for (i = fs->fs_postbl[pos][i];; ) {
632 			if (isblock(fs, cgp->cg_free, bno + i)) {
633 				bno = blkstofrags(fs, (bno + i));
634 				goto gotit;
635 			}
636 			delta = fs->fs_rotbl[i];
637 			if (delta <= 0 || delta > MAXBPC - i)
638 				break;
639 			i += delta;
640 		}
641 		printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
642 		panic("alloccgblk: can't find blk in cyl");
643 	}
644 norot:
645 	/*
646 	 * no blocks in the requested cylinder, so take next
647 	 * available one in this cylinder group.
648 	 */
649 	bno = mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
650 	if (bno < 0)
651 		return (NULL);
652 	cgp->cg_rotor = bno;
653 gotit:
654 	clrblock(fs, cgp->cg_free, (long)fragstoblks(fs, bno));
655 	cgp->cg_cs.cs_nbfree--;
656 	fs->fs_cstotal.cs_nbfree--;
657 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
658 	cylno = cbtocylno(fs, bno);
659 	cgp->cg_b[cylno][cbtorpos(fs, bno)]--;
660 	cgp->cg_btot[cylno]--;
661 	fs->fs_fmod++;
662 	return (cgp->cg_cgx * fs->fs_fpg + bno);
663 }
664 
665 /*
666  * Determine whether an inode can be allocated.
667  *
668  * Check to see if an inode is available, and if it is,
669  * allocate it using the following policy:
670  *   1) allocate the requested inode.
671  *   2) allocate the next available inode after the requested
672  *      inode in the specified cylinder group.
673  */
674 ino_t
675 ialloccg(ip, cg, ipref, mode)
676 	struct inode *ip;
677 	int cg;
678 	daddr_t ipref;
679 	int mode;
680 {
681 	register struct fs *fs;
682 	register struct cg *cgp;
683 	struct buf *bp;
684 	int start, len, loc, map, i;
685 
686 	fs = ip->i_fs;
687 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
688 		return (NULL);
689 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
690 	cgp = bp->b_un.b_cg;
691 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC ||
692 	    cgp->cg_cs.cs_nifree == 0) {
693 		brelse(bp);
694 		return (NULL);
695 	}
696 	cgp->cg_time = time.tv_sec;
697 	if (ipref) {
698 		ipref %= fs->fs_ipg;
699 		if (isclr(cgp->cg_iused, ipref))
700 			goto gotit;
701 	}
702 	start = cgp->cg_irotor / NBBY;
703 	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
704 	loc = skpc(0xff, len, &cgp->cg_iused[start]);
705 	if (loc == 0) {
706 		printf("cg = %s, irotor = %d, fs = %s\n",
707 		    cg, cgp->cg_irotor, fs->fs_fsmnt);
708 		panic("ialloccg: map corrupted");
709 		/* NOTREACHED */
710 	}
711 	i = start + len - loc;
712 	map = cgp->cg_iused[i];
713 	ipref = i * NBBY;
714 	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
715 		if ((map & i) == 0) {
716 			cgp->cg_irotor = ipref;
717 			goto gotit;
718 		}
719 	}
720 	printf("fs = %s\n", fs->fs_fsmnt);
721 	panic("ialloccg: block not in map");
722 	/* NOTREACHED */
723 gotit:
724 	setbit(cgp->cg_iused, ipref);
725 	cgp->cg_cs.cs_nifree--;
726 	fs->fs_cstotal.cs_nifree--;
727 	fs->fs_cs(fs, cg).cs_nifree--;
728 	fs->fs_fmod++;
729 	if ((mode & IFMT) == IFDIR) {
730 		cgp->cg_cs.cs_ndir++;
731 		fs->fs_cstotal.cs_ndir++;
732 		fs->fs_cs(fs, cg).cs_ndir++;
733 	}
734 	bdwrite(bp);
735 	return (cg * fs->fs_ipg + ipref);
736 }
737 
738 /*
739  * Free a block or fragment.
740  *
741  * The specified block or fragment is placed back in the
742  * free map. If a fragment is deallocated, a possible
743  * block reassembly is checked.
744  */
745 free(ip, bno, size)
746 	register struct inode *ip;
747 	daddr_t bno;
748 	off_t size;
749 {
750 	register struct fs *fs;
751 	register struct cg *cgp;
752 	register struct buf *bp;
753 	int cg, blk, frags, bbase;
754 	register int i;
755 
756 	fs = ip->i_fs;
757 	if ((unsigned)size > fs->fs_bsize || fragoff(fs, size) != 0) {
758 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
759 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
760 		panic("free: bad size");
761 	}
762 	cg = dtog(fs, bno);
763 	if (badblock(fs, bno)) {
764 		printf("bad block %d, ino %d\n", bno, ip->i_number);
765 		return;
766 	}
767 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
768 	cgp = bp->b_un.b_cg;
769 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
770 		brelse(bp);
771 		return;
772 	}
773 	cgp->cg_time = time.tv_sec;
774 	bno = dtogd(fs, bno);
775 	if (size == fs->fs_bsize) {
776 		if (isblock(fs, cgp->cg_free, fragstoblks(fs, bno))) {
777 			printf("dev = 0x%x, block = %d, fs = %s\n",
778 			    ip->i_dev, bno, fs->fs_fsmnt);
779 			panic("free: freeing free block");
780 		}
781 		setblock(fs, cgp->cg_free, fragstoblks(fs, bno));
782 		cgp->cg_cs.cs_nbfree++;
783 		fs->fs_cstotal.cs_nbfree++;
784 		fs->fs_cs(fs, cg).cs_nbfree++;
785 		i = cbtocylno(fs, bno);
786 		cgp->cg_b[i][cbtorpos(fs, bno)]++;
787 		cgp->cg_btot[i]++;
788 	} else {
789 		bbase = bno - fragnum(fs, bno);
790 		/*
791 		 * decrement the counts associated with the old frags
792 		 */
793 		blk = blkmap(fs, cgp->cg_free, bbase);
794 		fragacct(fs, blk, cgp->cg_frsum, -1);
795 		/*
796 		 * deallocate the fragment
797 		 */
798 		frags = numfrags(fs, size);
799 		for (i = 0; i < frags; i++) {
800 			if (isset(cgp->cg_free, bno + i)) {
801 				printf("dev = 0x%x, block = %d, fs = %s\n",
802 				    ip->i_dev, bno + i, fs->fs_fsmnt);
803 				panic("free: freeing free frag");
804 			}
805 			setbit(cgp->cg_free, bno + i);
806 		}
807 		cgp->cg_cs.cs_nffree += i;
808 		fs->fs_cstotal.cs_nffree += i;
809 		fs->fs_cs(fs, cg).cs_nffree += i;
810 		/*
811 		 * add back in counts associated with the new frags
812 		 */
813 		blk = blkmap(fs, cgp->cg_free, bbase);
814 		fragacct(fs, blk, cgp->cg_frsum, 1);
815 		/*
816 		 * if a complete block has been reassembled, account for it
817 		 */
818 		if (isblock(fs, cgp->cg_free, fragstoblks(fs, bbase))) {
819 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
820 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
821 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
822 			cgp->cg_cs.cs_nbfree++;
823 			fs->fs_cstotal.cs_nbfree++;
824 			fs->fs_cs(fs, cg).cs_nbfree++;
825 			i = cbtocylno(fs, bbase);
826 			cgp->cg_b[i][cbtorpos(fs, bbase)]++;
827 			cgp->cg_btot[i]++;
828 		}
829 	}
830 	fs->fs_fmod++;
831 	bdwrite(bp);
832 }
833 
834 /*
835  * Free an inode.
836  *
837  * The specified inode is placed back in the free map.
838  */
839 ifree(ip, ino, mode)
840 	struct inode *ip;
841 	ino_t ino;
842 	int mode;
843 {
844 	register struct fs *fs;
845 	register struct cg *cgp;
846 	register struct buf *bp;
847 	int cg;
848 
849 	fs = ip->i_fs;
850 	if ((unsigned)ino >= fs->fs_ipg*fs->fs_ncg) {
851 		printf("dev = 0x%x, ino = %d, fs = %s\n",
852 		    ip->i_dev, ino, fs->fs_fsmnt);
853 		panic("ifree: range");
854 	}
855 	cg = itog(fs, ino);
856 	bp = bread(ip->i_dev, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
857 	cgp = bp->b_un.b_cg;
858 	if (bp->b_flags & B_ERROR || cgp->cg_magic != CG_MAGIC) {
859 		brelse(bp);
860 		return;
861 	}
862 	cgp->cg_time = time.tv_sec;
863 	ino %= fs->fs_ipg;
864 	if (isclr(cgp->cg_iused, ino)) {
865 		printf("dev = 0x%x, ino = %d, fs = %s\n",
866 		    ip->i_dev, ino, fs->fs_fsmnt);
867 		panic("ifree: freeing free inode");
868 	}
869 	clrbit(cgp->cg_iused, ino);
870 	if (ino < cgp->cg_irotor)
871 		cgp->cg_irotor = ino;
872 	cgp->cg_cs.cs_nifree++;
873 	fs->fs_cstotal.cs_nifree++;
874 	fs->fs_cs(fs, cg).cs_nifree++;
875 	if ((mode & IFMT) == IFDIR) {
876 		cgp->cg_cs.cs_ndir--;
877 		fs->fs_cstotal.cs_ndir--;
878 		fs->fs_cs(fs, cg).cs_ndir--;
879 	}
880 	fs->fs_fmod++;
881 	bdwrite(bp);
882 }
883 
884 /*
885  * Find a block of the specified size in the specified cylinder group.
886  *
887  * It is a panic if a request is made to find a block if none are
888  * available.
889  */
890 daddr_t
891 mapsearch(fs, cgp, bpref, allocsiz)
892 	register struct fs *fs;
893 	register struct cg *cgp;
894 	daddr_t bpref;
895 	int allocsiz;
896 {
897 	daddr_t bno;
898 	int start, len, loc, i;
899 	int blk, field, subfield, pos;
900 
901 	/*
902 	 * find the fragment by searching through the free block
903 	 * map for an appropriate bit pattern
904 	 */
905 	if (bpref)
906 		start = dtogd(fs, bpref) / NBBY;
907 	else
908 		start = cgp->cg_frotor / NBBY;
909 	len = howmany(fs->fs_fpg, NBBY) - start;
910 	loc = scanc((unsigned)len, (caddr_t)&cgp->cg_free[start],
911 		(caddr_t)fragtbl[fs->fs_frag],
912 		(int)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
913 	if (loc == 0) {
914 		len = start + 1;
915 		start = 0;
916 		loc = scanc((unsigned)len, (caddr_t)&cgp->cg_free[start],
917 			(caddr_t)fragtbl[fs->fs_frag],
918 			(int)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
919 		if (loc == 0) {
920 			printf("start = %d, len = %d, fs = %s\n",
921 			    start, len, fs->fs_fsmnt);
922 			panic("alloccg: map corrupted");
923 			return (-1);
924 		}
925 	}
926 	bno = (start + len - loc) * NBBY;
927 	cgp->cg_frotor = bno;
928 	/*
929 	 * found the byte in the map
930 	 * sift through the bits to find the selected frag
931 	 */
932 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
933 		blk = blkmap(fs, cgp->cg_free, bno);
934 		blk <<= 1;
935 		field = around[allocsiz];
936 		subfield = inside[allocsiz];
937 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
938 			if ((blk & field) == subfield)
939 				return (bno + pos);
940 			field <<= 1;
941 			subfield <<= 1;
942 		}
943 	}
944 	printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt);
945 	panic("alloccg: block not in map");
946 	return (-1);
947 }
948 
949 /*
950  * Fserr prints the name of a file system with an error diagnostic.
951  *
952  * The form of the error message is:
953  *	fs: error message
954  */
955 fserr(fs, cp)
956 	struct fs *fs;
957 	char *cp;
958 {
959 
960 	printf("%s: %s\n", fs->fs_fsmnt, cp);
961 }
962