xref: /csrg-svn/sys/ufs/ffs/ffs_alloc.c (revision 5960)
1 /* Copyright (c) 1981 Regents of the University of California */
2 
3 static char vers[] = "@(#)ffs_alloc.c 1.18 02/25/82";
4 
5 /*	alloc.c	4.8	81/03/08	*/
6 
7 #include "../h/param.h"
8 #include "../h/systm.h"
9 #include "../h/mount.h"
10 #include "../h/fs.h"
11 #include "../h/conf.h"
12 #include "../h/buf.h"
13 #include "../h/inode.h"
14 #include "../h/user.h"
15 
16 extern u_long		hashalloc();
17 extern ino_t		ialloccg();
18 extern daddr_t		alloccg();
19 extern daddr_t		alloccgblk();
20 extern daddr_t		fragextend();
21 extern daddr_t		blkpref();
22 extern daddr_t		mapsearch();
23 extern int		inside[], around[];
24 extern unsigned char	*fragtbl[];
25 
26 /*
27  * Allocate a block in the file system.
28  *
29  * The size of the requested block is given, which must be some
30  * multiple of fs_fsize and <= fs_bsize.
31  * A preference may be optionally specified. If a preference is given
32  * the following hierarchy is used to allocate a block:
33  *   1) allocate the requested block.
34  *   2) allocate a rotationally optimal block in the same cylinder.
35  *   3) allocate a block in the same cylinder group.
36  *   4) quadradically rehash into other cylinder groups, until an
37  *      available block is located.
38  * If no block preference is given the following heirarchy is used
39  * to allocate a block:
40  *   1) allocate a block in the cylinder group that contains the
41  *      inode for the file.
42  *   2) quadradically rehash into other cylinder groups, until an
43  *      available block is located.
44  */
45 struct buf *
46 alloc(dev, ip, bpref, size)
47 	dev_t dev;
48 	register struct inode *ip;
49 	daddr_t bpref;
50 	int size;
51 {
52 	daddr_t bno;
53 	register struct fs *fs;
54 	register struct buf *bp;
55 	int cg;
56 
57 	fs = getfs(dev);
58 	if ((unsigned)size > fs->fs_bsize || fragoff(fs, size) != 0)
59 		panic("alloc: bad size");
60 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
61 		goto nospace;
62 	if (u.u_uid != 0 &&
63 	    fs->fs_cstotal.cs_nbfree * fs->fs_frag + fs->fs_cstotal.cs_nffree <
64 	      fs->fs_dsize * fs->fs_minfree / 100)
65 		goto nospace;
66 	if (bpref >= fs->fs_size)
67 		bpref = 0;
68 	if (bpref == 0)
69 		cg = itog(fs, ip->i_number);
70 	else
71 		cg = dtog(fs, bpref);
72 	bno = (daddr_t)hashalloc(dev, fs, cg, (long)bpref, size, alloccg);
73 	if (bno == 0)
74 		goto nospace;
75 	bp = getblk(dev, fsbtodb(fs, bno), size);
76 	clrbuf(bp);
77 	return (bp);
78 nospace:
79 	fserr(fs, "file system full");
80 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
81 	u.u_error = ENOSPC;
82 	return (NULL);
83 }
84 
85 /*
86  * Reallocate a fragment to a bigger size
87  *
88  * The number and size of the old block is given, and a preference
89  * and new size is also specified. The allocator attempts to extend
90  * the original block. Failing that, the regular block allocator is
91  * invoked to get an appropriate block.
92  */
93 struct buf *
94 realloccg(dev, bprev, bpref, osize, nsize)
95 	dev_t dev;
96 	daddr_t bprev, bpref;
97 	int osize, nsize;
98 {
99 	daddr_t bno;
100 	register struct fs *fs;
101 	register struct buf *bp, *obp;
102 	caddr_t cp;
103 	int cg;
104 
105 	fs = getfs(dev);
106 	if ((unsigned)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
107 	    (unsigned)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0)
108 		panic("realloccg: bad size");
109 	if (u.u_uid != 0 &&
110 	    fs->fs_cstotal.cs_nbfree * fs->fs_frag + fs->fs_cstotal.cs_nffree <
111 	      fs->fs_dsize * fs->fs_minfree / 100)
112 		goto nospace;
113 	if (bprev != 0)
114 		cg = dtog(fs, bprev);
115 	else
116 		panic("realloccg: bad bprev");
117 	bno = fragextend(dev, fs, cg, (long)bprev, osize, nsize);
118 	if (bno != 0) {
119 		bp = bread(dev, fsbtodb(fs, bno), osize);
120 		if (bp->b_flags & B_ERROR) {
121 			brelse(bp);
122 			return 0;
123 		}
124 		bp->b_bcount = nsize;
125 		blkclr(bp->b_un.b_addr + osize, nsize - osize);
126 		return (bp);
127 	}
128 	if (bpref >= fs->fs_size)
129 		bpref = 0;
130 	bno = (daddr_t)hashalloc(dev, fs, cg, (long)bpref, nsize, alloccg);
131 	if (bno != 0) {
132 		/*
133 		 * make a new copy
134 		 */
135 		obp = bread(dev, fsbtodb(fs, bprev), osize);
136 		if (obp->b_flags & B_ERROR) {
137 			brelse(obp);
138 			return 0;
139 		}
140 		bp = getblk(dev, fsbtodb(fs, bno), nsize);
141 		cp = bp->b_un.b_addr;
142 		bp->b_un.b_addr = obp->b_un.b_addr;
143 		obp->b_un.b_addr = cp;
144 		obp->b_flags |= B_INVAL;
145 		brelse(obp);
146 		fre(dev, bprev, (off_t)osize);
147 		blkclr(bp->b_un.b_addr + osize, nsize - osize);
148 		return(bp);
149 	}
150 nospace:
151 	/*
152 	 * no space available
153 	 */
154 	fserr(fs, "file system full");
155 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
156 	u.u_error = ENOSPC;
157 	return (NULL);
158 }
159 
160 /*
161  * Allocate an inode in the file system.
162  *
163  * A preference may be optionally specified. If a preference is given
164  * the following hierarchy is used to allocate an inode:
165  *   1) allocate the requested inode.
166  *   2) allocate an inode in the same cylinder group.
167  *   3) quadradically rehash into other cylinder groups, until an
168  *      available inode is located.
169  * If no inode preference is given the following heirarchy is used
170  * to allocate an inode:
171  *   1) allocate an inode in cylinder group 0.
172  *   2) quadradically rehash into other cylinder groups, until an
173  *      available inode is located.
174  */
175 struct inode *
176 ialloc(dev, ipref, mode)
177 	dev_t dev;
178 	ino_t ipref;
179 	int mode;
180 {
181 	ino_t ino;
182 	register struct fs *fs;
183 	register struct inode *ip;
184 	int cg;
185 
186 	fs = getfs(dev);
187 	if (fs->fs_cstotal.cs_nifree == 0)
188 		goto noinodes;
189 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
190 		ipref = 0;
191 	cg = itog(fs, ipref);
192 	ino = (ino_t)hashalloc(dev, fs, cg, (long)ipref, mode, ialloccg);
193 	if (ino == 0)
194 		goto noinodes;
195 	ip = iget(dev, ino);
196 	if (ip == NULL) {
197 		ifree(dev, ino, 0);
198 		return (NULL);
199 	}
200 	if (ip->i_mode)
201 		panic("ialloc: dup alloc");
202 	return (ip);
203 noinodes:
204 	fserr(fs, "out of inodes");
205 	uprintf("\n%s: create failed, no inodes free\n", fs->fs_fsmnt);
206 	u.u_error = ENOSPC;
207 	return (NULL);
208 }
209 
210 /*
211  * Find a cylinder to place a directory.
212  *
213  * The policy implemented by this algorithm is to select from
214  * among those cylinder groups with above the average number of
215  * free inodes, the one with the smallest number of directories.
216  */
217 dirpref(dev)
218 	dev_t dev;
219 {
220 	register struct fs *fs;
221 	int cg, minndir, mincg, avgifree;
222 
223 	fs = getfs(dev);
224 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
225 	minndir = fs->fs_ipg;
226 	mincg = 0;
227 	for (cg = 0; cg < fs->fs_ncg; cg++)
228 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
229 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
230 			mincg = cg;
231 			minndir = fs->fs_cs(fs, cg).cs_ndir;
232 		}
233 	return (fs->fs_ipg * mincg);
234 }
235 
236 /*
237  * Select a cylinder to place a large block of data.
238  *
239  * The policy implemented by this algorithm is to maintain a
240  * rotor that sweeps the cylinder groups. When a block is
241  * needed, the rotor is advanced until a cylinder group with
242  * greater than the average number of free blocks is found.
243  */
244 daddr_t
245 blkpref(dev)
246 	dev_t dev;
247 {
248 	register struct fs *fs;
249 	int cg, avgbfree;
250 
251 	fs = getfs(dev);
252 	avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
253 	for (cg = fs->fs_cgrotor + 1; cg < fs->fs_ncg; cg++)
254 		if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
255 			fs->fs_cgrotor = cg;
256 			return (fs->fs_fpg * cg + fs->fs_frag);
257 		}
258 	for (cg = 0; cg <= fs->fs_cgrotor; cg++)
259 		if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
260 			fs->fs_cgrotor = cg;
261 			return (fs->fs_fpg * cg + fs->fs_frag);
262 		}
263 	return (0);
264 }
265 
266 /*
267  * Implement the cylinder overflow algorithm.
268  *
269  * The policy implemented by this algorithm is:
270  *   1) allocate the block in its requested cylinder group.
271  *   2) quadradically rehash on the cylinder group number.
272  *   3) brute force search for a free block.
273  */
274 /*VARARGS5*/
275 u_long
276 hashalloc(dev, fs, cg, pref, size, allocator)
277 	dev_t dev;
278 	register struct fs *fs;
279 	int cg;
280 	long pref;
281 	int size;	/* size for data blocks, mode for inodes */
282 	u_long (*allocator)();
283 {
284 	long result;
285 	int i, icg = cg;
286 
287 	/*
288 	 * 1: preferred cylinder group
289 	 */
290 	result = (*allocator)(dev, fs, cg, pref, size);
291 	if (result)
292 		return (result);
293 	/*
294 	 * 2: quadratic rehash
295 	 */
296 	for (i = 1; i < fs->fs_ncg; i *= 2) {
297 		cg += i;
298 		if (cg >= fs->fs_ncg)
299 			cg -= fs->fs_ncg;
300 		result = (*allocator)(dev, fs, cg, 0, size);
301 		if (result)
302 			return (result);
303 	}
304 	/*
305 	 * 3: brute force search
306 	 */
307 	cg = icg;
308 	for (i = 0; i < fs->fs_ncg; i++) {
309 		result = (*allocator)(dev, fs, cg, 0, size);
310 		if (result)
311 			return (result);
312 		cg++;
313 		if (cg == fs->fs_ncg)
314 			cg = 0;
315 	}
316 	return (0);
317 }
318 
319 /*
320  * Determine whether a fragment can be extended.
321  *
322  * Check to see if the necessary fragments are available, and
323  * if they are, allocate them.
324  */
325 daddr_t
326 fragextend(dev, fs, cg, bprev, osize, nsize)
327 	dev_t dev;
328 	register struct fs *fs;
329 	int cg;
330 	long bprev;
331 	int osize, nsize;
332 {
333 	register struct buf *bp;
334 	register struct cg *cgp;
335 	long bno;
336 	int frags, bbase;
337 	int i;
338 
339 	frags = numfrags(fs, nsize);
340 	bbase = fragoff(fs, bprev);
341 	if (bbase > (bprev + frags - 1) % fs->fs_frag) {
342 		/* cannot extend across a block boundry */
343 		return (0);
344 	}
345 	bp = bread(dev, fsbtodb(fs, cgtod(fs, cg)), fs->fs_bsize);
346 	if (bp->b_flags & B_ERROR) {
347 		brelse(bp);
348 		return 0;
349 	}
350 	cgp = bp->b_un.b_cg;
351 	bno = dtogd(fs, bprev);
352 	for (i = numfrags(fs, osize); i < frags; i++)
353 		if (isclr(cgp->cg_free, bno + i)) {
354 			brelse(bp);
355 			return (0);
356 		}
357 	/*
358 	 * the current fragment can be extended
359 	 * deduct the count on fragment being extended into
360 	 * increase the count on the remaining fragment (if any)
361 	 * allocate the extended piece
362 	 */
363 	for (i = frags; i < fs->fs_frag - bbase; i++)
364 		if (isclr(cgp->cg_free, bno + i))
365 			break;
366 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
367 	if (i != frags)
368 		cgp->cg_frsum[i - frags]++;
369 	for (i = numfrags(fs, osize); i < frags; i++) {
370 		clrbit(cgp->cg_free, bno + i);
371 		cgp->cg_cs.cs_nffree--;
372 		fs->fs_cstotal.cs_nffree--;
373 		fs->fs_cs(fs, cg).cs_nffree--;
374 	}
375 	fs->fs_fmod++;
376 	bdwrite(bp);
377 	return (bprev);
378 }
379 
380 /*
381  * Determine whether a block can be allocated.
382  *
383  * Check to see if a block of the apprpriate size is available,
384  * and if it is, allocate it.
385  */
386 daddr_t
387 alloccg(dev, fs, cg, bpref, size)
388 	dev_t dev;
389 	register struct fs *fs;
390 	int cg;
391 	daddr_t bpref;
392 	int size;
393 {
394 	register struct buf *bp;
395 	register struct cg *cgp;
396 	int bno, frags;
397 	int allocsiz;
398 	register int i;
399 
400 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
401 		return (0);
402 	bp = bread(dev, fsbtodb(fs, cgtod(fs, cg)), fs->fs_bsize);
403 	if (bp->b_flags & B_ERROR) {
404 		brelse(bp);
405 		return 0;
406 	}
407 	cgp = bp->b_un.b_cg;
408 	if (size == fs->fs_bsize) {
409 		bno = alloccgblk(fs, cgp, bpref);
410 		bdwrite(bp);
411 		return (bno);
412 	}
413 	/*
414 	 * check to see if any fragments are already available
415 	 * allocsiz is the size which will be allocated, hacking
416 	 * it down to a smaller size if necessary
417 	 */
418 	frags = numfrags(fs, size);
419 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
420 		if (cgp->cg_frsum[allocsiz] != 0)
421 			break;
422 	if (allocsiz == fs->fs_frag) {
423 		/*
424 		 * no fragments were available, so a block will be
425 		 * allocated, and hacked up
426 		 */
427 		if (cgp->cg_cs.cs_nbfree == 0) {
428 			brelse(bp);
429 			return (0);
430 		}
431 		bno = alloccgblk(fs, cgp, bpref);
432 		bpref = dtogd(fs, bno);
433 		for (i = frags; i < fs->fs_frag; i++)
434 			setbit(cgp->cg_free, bpref + i);
435 		i = fs->fs_frag - frags;
436 		cgp->cg_cs.cs_nffree += i;
437 		fs->fs_cstotal.cs_nffree += i;
438 		fs->fs_cs(fs, cg).cs_nffree += i;
439 		cgp->cg_frsum[i]++;
440 		bdwrite(bp);
441 		return (bno);
442 	}
443 	bno = mapsearch(fs, cgp, bpref, allocsiz);
444 	if (bno == 0)
445 		return (0);
446 	for (i = 0; i < frags; i++)
447 		clrbit(cgp->cg_free, bno + i);
448 	cgp->cg_cs.cs_nffree -= frags;
449 	fs->fs_cstotal.cs_nffree -= frags;
450 	fs->fs_cs(fs, cg).cs_nffree -= frags;
451 	cgp->cg_frsum[allocsiz]--;
452 	if (frags != allocsiz)
453 		cgp->cg_frsum[allocsiz - frags]++;
454 	bdwrite(bp);
455 	return (cg * fs->fs_fpg + bno);
456 }
457 
458 /*
459  * Allocate a block in a cylinder group.
460  *
461  * This algorithm implements the following policy:
462  *   1) allocate the requested block.
463  *   2) allocate a rotationally optimal block in the same cylinder.
464  *   3) allocate the next available block on the block rotor for the
465  *      specified cylinder group.
466  * Note that this routine only allocates fs_bsize blocks; these
467  * blocks may be fragmented by the routine that allocates them.
468  */
469 daddr_t
470 alloccgblk(fs, cgp, bpref)
471 	struct fs *fs;
472 	register struct cg *cgp;
473 	daddr_t bpref;
474 {
475 	daddr_t bno;
476 	int cylno, pos;
477 	short *cylbp;
478 	register int i;
479 
480 	if (bpref == 0) {
481 		bpref = cgp->cg_rotor;
482 		goto norot;
483 	}
484 	bpref &= ~(fs->fs_frag - 1);
485 	bpref = dtogd(fs, bpref);
486 	/*
487 	 * if the requested block is available, use it
488 	 */
489 	if (isblock(fs, cgp->cg_free, bpref/fs->fs_frag)) {
490 		bno = bpref;
491 		goto gotit;
492 	}
493 	/*
494 	 * check for a block available on the same cylinder
495 	 */
496 	cylno = cbtocylno(fs, bpref);
497 	if (cgp->cg_btot[cylno] == 0)
498 		goto norot;
499 	if (fs->fs_cpc == 0) {
500 		/*
501 		 * block layout info is not available, so just have
502 		 * to take any block in this cylinder.
503 		 */
504 		bpref = howmany(fs->fs_spc * cylno, NSPF(fs));
505 		goto norot;
506 	}
507 	/*
508 	 * find a block that is rotationally optimal
509 	 */
510 	cylbp = cgp->cg_b[cylno];
511 	if (fs->fs_rotdelay == 0) {
512 		pos = cbtorpos(fs, bpref);
513 	} else {
514 		/*
515 		 * here we convert ms of delay to frags as:
516 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
517 		 *	((sect/frag) * (ms/sec))
518 		 * then round up to the next rotational position
519 		 */
520 		bpref += fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
521 		    (NSPF(fs) * 1000);
522 		pos = cbtorpos(fs, bpref);
523 		pos = (pos + 1) % NRPOS;
524 	}
525 	/*
526 	 * check the summary information to see if a block is
527 	 * available in the requested cylinder starting at the
528 	 * optimal rotational position and proceeding around.
529 	 */
530 	for (i = pos; i < NRPOS; i++)
531 		if (cylbp[i] > 0)
532 			break;
533 	if (i == NRPOS)
534 		for (i = 0; i < pos; i++)
535 			if (cylbp[i] > 0)
536 				break;
537 	if (cylbp[i] > 0) {
538 		/*
539 		 * found a rotational position, now find the actual
540 		 * block. A panic if none is actually there.
541 		 */
542 		pos = cylno % fs->fs_cpc;
543 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
544 		if (fs->fs_postbl[pos][i] == -1)
545 			panic("alloccgblk: cyl groups corrupted");
546 		for (i = fs->fs_postbl[pos][i]; ; i += fs->fs_rotbl[i]) {
547 			if (isblock(fs, cgp->cg_free, bno + i)) {
548 				bno = (bno + i) * fs->fs_frag;
549 				goto gotit;
550 			}
551 			if (fs->fs_rotbl[i] == 0)
552 				break;
553 		}
554 		panic("alloccgblk: can't find blk in cyl");
555 	}
556 norot:
557 	/*
558 	 * no blocks in the requested cylinder, so take next
559 	 * available one in this cylinder group.
560 	 */
561 	bno = mapsearch(fs, cgp, bpref, fs->fs_frag);
562 	if (bno == 0)
563 		return (0);
564 	cgp->cg_rotor = bno;
565 gotit:
566 	clrblock(fs, cgp->cg_free, bno/fs->fs_frag);
567 	cgp->cg_cs.cs_nbfree--;
568 	fs->fs_cstotal.cs_nbfree--;
569 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
570 	cylno = cbtocylno(fs, bno);
571 	cgp->cg_b[cylno][cbtorpos(fs, bno)]--;
572 	cgp->cg_btot[cylno]--;
573 	fs->fs_fmod++;
574 	return (cgp->cg_cgx * fs->fs_fpg + bno);
575 }
576 
577 /*
578  * Determine whether an inode can be allocated.
579  *
580  * Check to see if an inode is available, and if it is,
581  * allocate it using the following policy:
582  *   1) allocate the requested inode.
583  *   2) allocate the next available inode after the requested
584  *      inode in the specified cylinder group.
585  */
586 ino_t
587 ialloccg(dev, fs, cg, ipref, mode)
588 	dev_t dev;
589 	register struct fs *fs;
590 	int cg;
591 	daddr_t ipref;
592 	int mode;
593 {
594 	register struct buf *bp;
595 	register struct cg *cgp;
596 	int i;
597 
598 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
599 		return (0);
600 	bp = bread(dev, fsbtodb(fs, cgtod(fs, cg)), fs->fs_bsize);
601 	if (bp->b_flags & B_ERROR) {
602 		brelse(bp);
603 		return 0;
604 	}
605 	cgp = bp->b_un.b_cg;
606 	if (ipref) {
607 		ipref %= fs->fs_ipg;
608 		if (isclr(cgp->cg_iused, ipref))
609 			goto gotit;
610 	} else
611 		ipref = cgp->cg_irotor;
612 	for (i = 0; i < fs->fs_ipg; i++) {
613 		ipref++;
614 		if (ipref >= fs->fs_ipg)
615 			ipref = 0;
616 		if (isclr(cgp->cg_iused, ipref)) {
617 			cgp->cg_irotor = ipref;
618 			goto gotit;
619 		}
620 	}
621 	brelse(bp);
622 	return (0);
623 gotit:
624 	setbit(cgp->cg_iused, ipref);
625 	cgp->cg_cs.cs_nifree--;
626 	fs->fs_cstotal.cs_nifree--;
627 	fs->fs_cs(fs, cg).cs_nifree--;
628 	fs->fs_fmod++;
629 	if ((mode & IFMT) == IFDIR) {
630 		cgp->cg_cs.cs_ndir++;
631 		fs->fs_cstotal.cs_ndir++;
632 		fs->fs_cs(fs, cg).cs_ndir++;
633 	}
634 	bdwrite(bp);
635 	return (cg * fs->fs_ipg + ipref);
636 }
637 
638 /*
639  * Free a block or fragment.
640  *
641  * The specified block or fragment is placed back in the
642  * free map. If a fragment is deallocated, a possible
643  * block reassembly is checked.
644  */
645 fre(dev, bno, size)
646 	dev_t dev;
647 	daddr_t bno;
648 	off_t size;
649 {
650 	register struct fs *fs;
651 	register struct cg *cgp;
652 	register struct buf *bp;
653 	int cg, blk, frags, bbase;
654 	register int i;
655 
656 	fs = getfs(dev);
657 	if ((unsigned)size > fs->fs_bsize || fragoff(fs, size) != 0)
658 		panic("free: bad size");
659 	cg = dtog(fs, bno);
660 	if (badblock(fs, bno))
661 		return;
662 	bp = bread(dev, fsbtodb(fs, cgtod(fs, cg)), fs->fs_bsize);
663 	if (bp->b_flags & B_ERROR) {
664 		brelse(bp);
665 		return;
666 	}
667 	cgp = bp->b_un.b_cg;
668 	bno = dtogd(fs, bno);
669 	if (size == fs->fs_bsize) {
670 		if (isblock(fs, cgp->cg_free, bno/fs->fs_frag))
671 			panic("free: freeing free block");
672 		setblock(fs, cgp->cg_free, bno/fs->fs_frag);
673 		cgp->cg_cs.cs_nbfree++;
674 		fs->fs_cstotal.cs_nbfree++;
675 		fs->fs_cs(fs, cg).cs_nbfree++;
676 		i = cbtocylno(fs, bno);
677 		cgp->cg_b[i][cbtorpos(fs, bno)]++;
678 		cgp->cg_btot[i]++;
679 	} else {
680 		bbase = bno - (bno % fs->fs_frag);
681 		/*
682 		 * decrement the counts associated with the old frags
683 		 */
684 		blk = ((cgp->cg_free[bbase / NBBY] >> (bbase % NBBY)) &
685 		       (0xff >> (NBBY - fs->fs_frag)));
686 		fragacct(fs, blk, cgp->cg_frsum, -1);
687 		/*
688 		 * deallocate the fragment
689 		 */
690 		frags = numfrags(fs, size);
691 		for (i = 0; i < frags; i++) {
692 			if (isset(cgp->cg_free, bno + i))
693 				panic("free: freeing free frag");
694 			setbit(cgp->cg_free, bno + i);
695 			cgp->cg_cs.cs_nffree++;
696 			fs->fs_cstotal.cs_nffree++;
697 			fs->fs_cs(fs, cg).cs_nffree++;
698 		}
699 		/*
700 		 * add back in counts associated with the new frags
701 		 */
702 		blk = ((cgp->cg_free[bbase / NBBY] >> (bbase % NBBY)) &
703 		       (0xff >> (NBBY - fs->fs_frag)));
704 		fragacct(fs, blk, cgp->cg_frsum, 1);
705 		/*
706 		 * if a complete block has been reassembled, account for it
707 		 */
708 		if (isblock(fs, cgp->cg_free, bbase / fs->fs_frag)) {
709 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
710 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
711 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
712 			cgp->cg_cs.cs_nbfree++;
713 			fs->fs_cstotal.cs_nbfree++;
714 			fs->fs_cs(fs, cg).cs_nbfree++;
715 			i = cbtocylno(fs, bbase);
716 			cgp->cg_b[i][cbtorpos(fs, bbase)]++;
717 			cgp->cg_btot[i]++;
718 		}
719 	}
720 	fs->fs_fmod++;
721 	bdwrite(bp);
722 }
723 
724 /*
725  * Free an inode.
726  *
727  * The specified inode is placed back in the free map.
728  */
729 ifree(dev, ino, mode)
730 	dev_t dev;
731 	ino_t ino;
732 	int mode;
733 {
734 	register struct fs *fs;
735 	register struct cg *cgp;
736 	register struct buf *bp;
737 	int cg;
738 
739 	fs = getfs(dev);
740 	if ((unsigned)ino >= fs->fs_ipg*fs->fs_ncg)
741 		panic("ifree: range");
742 	cg = itog(fs, ino);
743 	bp = bread(dev, fsbtodb(fs, cgtod(fs, cg)), fs->fs_bsize);
744 	if (bp->b_flags & B_ERROR) {
745 		brelse(bp);
746 		return;
747 	}
748 	cgp = bp->b_un.b_cg;
749 	ino %= fs->fs_ipg;
750 	if (isclr(cgp->cg_iused, ino))
751 		panic("ifree: freeing free inode");
752 	clrbit(cgp->cg_iused, ino);
753 	cgp->cg_cs.cs_nifree++;
754 	fs->fs_cstotal.cs_nifree++;
755 	fs->fs_cs(fs, cg).cs_nifree++;
756 	if ((mode & IFMT) == IFDIR) {
757 		cgp->cg_cs.cs_ndir--;
758 		fs->fs_cstotal.cs_ndir--;
759 		fs->fs_cs(fs, cg).cs_ndir--;
760 	}
761 	fs->fs_fmod++;
762 	bdwrite(bp);
763 }
764 
765 /*
766  * Find a block of the specified size in the specified cylinder group.
767  *
768  * It is a panic if a request is made to find a block if none are
769  * available.
770  */
771 daddr_t
772 mapsearch(fs, cgp, bpref, allocsiz)
773 	register struct fs *fs;
774 	register struct cg *cgp;
775 	daddr_t bpref;
776 	int allocsiz;
777 {
778 	daddr_t bno;
779 	int start, len, loc, i;
780 	int blk, field, subfield, pos;
781 
782 	/*
783 	 * find the fragment by searching through the free block
784 	 * map for an appropriate bit pattern
785 	 */
786 	if (bpref)
787 		start = dtogd(fs, bpref) / NBBY;
788 	else
789 		start = cgp->cg_frotor / NBBY;
790 	len = howmany(fs->fs_fpg, NBBY) - start;
791 	loc = scanc(len, &cgp->cg_free[start], fragtbl[fs->fs_frag],
792 		1 << (allocsiz - 1));
793 	if (loc == 0) {
794 		loc = fs->fs_dblkno / NBBY;
795 		len = start - loc + 1;
796 		start = loc;
797 		loc = scanc(len, &cgp->cg_free[start], fragtbl[fs->fs_frag],
798 			1 << (allocsiz - 1));
799 		if (loc == 0) {
800 			panic("alloccg: map corrupted");
801 			return (0);
802 		}
803 	}
804 	bno = (start + len - loc) * NBBY;
805 	cgp->cg_frotor = bno;
806 	/*
807 	 * found the byte in the map
808 	 * sift through the bits to find the selected frag
809 	 */
810 	for (i = 0; i < NBBY; i += fs->fs_frag) {
811 		blk = (cgp->cg_free[bno / NBBY] >> i) &
812 		      (0xff >> NBBY - fs->fs_frag);
813 		blk <<= 1;
814 		field = around[allocsiz];
815 		subfield = inside[allocsiz];
816 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
817 			if ((blk & field) == subfield) {
818 				return (bno + i + pos);
819 			}
820 			field <<= 1;
821 			subfield <<= 1;
822 		}
823 	}
824 	panic("alloccg: block not in map");
825 	return (0);
826 }
827 
828 /*
829  * Update the frsum fields to reflect addition or deletion
830  * of some frags.
831  */
832 fragacct(fs, fragmap, fraglist, cnt)
833 	struct fs *fs;
834 	int fragmap;
835 	long fraglist[];
836 	int cnt;
837 {
838 	int inblk;
839 	register int field, subfield;
840 	register int siz, pos;
841 
842 	inblk = (int)(fragtbl[fs->fs_frag][fragmap]) << 1;
843 	fragmap <<= 1;
844 	for (siz = 1; siz < fs->fs_frag; siz++) {
845 		if (((1 << siz) & inblk) == 0)
846 			continue;
847 		field = around[siz];
848 		subfield = inside[siz];
849 		for (pos = siz; pos <= fs->fs_frag; pos++) {
850 			if ((fragmap & field) == subfield) {
851 				fraglist[siz] += cnt;
852 				pos += siz;
853 				field <<= siz;
854 				subfield <<= siz;
855 			}
856 			field <<= 1;
857 			subfield <<= 1;
858 		}
859 	}
860 }
861 
862 /*
863  * Check that a specified block number is in range.
864  */
865 badblock(fs, bn)
866 	register struct fs *fs;
867 	daddr_t bn;
868 {
869 
870 	if ((unsigned)bn >= fs->fs_size || bn < cgdmin(fs, dtog(fs, bn))) {
871 		fserr(fs, "bad block");
872 		return (1);
873 	}
874 	return (0);
875 }
876 
877 /*
878  * Getfs maps a device number into a pointer to the incore super block.
879  *
880  * The algorithm is a linear search through the mount table. A
881  * consistency check of the super block magic number is performed.
882  *
883  * panic: no fs -- the device is not mounted.
884  *	this "cannot happen"
885  */
886 struct fs *
887 getfs(dev)
888 	dev_t dev;
889 {
890 	register struct mount *mp;
891 	register struct fs *fs;
892 
893 	for (mp = &mount[0]; mp < &mount[NMOUNT]; mp++)
894 		if (mp->m_bufp != NULL && mp->m_dev == dev) {
895 			fs = mp->m_bufp->b_un.b_fs;
896 			if (fs->fs_magic != FS_MAGIC)
897 				panic("getfs: bad magic");
898 			return (fs);
899 		}
900 	panic("getfs: no fs");
901 	return (NULL);
902 }
903 
904 /*
905  * Fserr prints the name of a file system with an error diagnostic.
906  *
907  * The form of the error message is:
908  *	fs: error message
909  */
910 fserr(fs, cp)
911 	struct fs *fs;
912 	char *cp;
913 {
914 
915 	printf("%s: %s\n", fs->fs_fsmnt, cp);
916 }
917 
918 /*
919  * Getfsx returns the index in the file system
920  * table of the specified device.  The swap device
921  * is also assigned a pseudo-index.  The index may
922  * be used as a compressed indication of the location
923  * of a block, recording
924  *	<getfsx(dev),blkno>
925  * rather than
926  *	<dev, blkno>
927  * provided the information need remain valid only
928  * as long as the file system is mounted.
929  */
930 getfsx(dev)
931 	dev_t dev;
932 {
933 	register struct mount *mp;
934 
935 	if (dev == swapdev)
936 		return (MSWAPX);
937 	for(mp = &mount[0]; mp < &mount[NMOUNT]; mp++)
938 		if (mp->m_dev == dev)
939 			return (mp - &mount[0]);
940 	return (-1);
941 }
942 
943 /*
944  * Update is the internal name of 'sync'.  It goes through the disk
945  * queues to initiate sandbagged IO; goes through the inodes to write
946  * modified nodes; and it goes through the mount table to initiate
947  * the writing of the modified super blocks.
948  */
949 update()
950 {
951 	register struct inode *ip;
952 	register struct mount *mp;
953 	register struct buf *bp;
954 	struct fs *fs;
955 	time_t tim;
956 	int i, blks;
957 
958 	if (updlock)
959 		return;
960 	updlock++;
961 	/*
962 	 * Write back modified superblocks.
963 	 * Consistency check that the superblock
964 	 * of each file system is still in the buffer cache.
965 	 */
966 	for (mp = &mount[0]; mp < &mount[NMOUNT]; mp++)
967 		if (mp->m_bufp != NULL) {
968 			fs = mp->m_bufp->b_un.b_fs;
969 			if (fs->fs_fmod == 0)
970 				continue;
971 			if (fs->fs_ronly != 0)
972 				panic("update: rofs mod");
973 			bp = getblk(mp->m_dev, SBLOCK, SBSIZE);
974 			fs->fs_fmod = 0;
975 			fs->fs_time = TIME;
976 			if (bp->b_un.b_fs != fs)
977 				panic("update: bad b_fs");
978 			bwrite(bp);
979 			blks = howmany(fs->fs_cssize, fs->fs_bsize);
980 			for (i = 0; i < blks; i++) {
981 				bp = getblk(mp->m_dev,
982 				    fsbtodb(fs, fs->fs_csaddr + (i * fs->fs_frag)),
983 				    fs->fs_bsize);
984 				bwrite(bp);
985 			}
986 		}
987 	/*
988 	 * Write back each (modified) inode.
989 	 */
990 	for (ip = inode; ip < inodeNINODE; ip++)
991 		if((ip->i_flag&ILOCK)==0 && ip->i_count) {
992 			ip->i_flag |= ILOCK;
993 			ip->i_count++;
994 			tim = TIME;
995 			iupdat(ip, &tim, &tim, 0);
996 			iput(ip);
997 		}
998 	updlock = 0;
999 	/*
1000 	 * Force stale buffer cache information to be flushed,
1001 	 * for all devices.
1002 	 */
1003 	bflush(NODEV);
1004 }
1005 
1006 /*
1007  * block operations
1008  *
1009  * check if a block is available
1010  */
1011 isblock(fs, cp, h)
1012 	struct fs *fs;
1013 	unsigned char *cp;
1014 	int h;
1015 {
1016 	unsigned char mask;
1017 
1018 	switch (fs->fs_frag) {
1019 	case 8:
1020 		return (cp[h] == 0xff);
1021 	case 4:
1022 		mask = 0x0f << ((h & 0x1) << 2);
1023 		return ((cp[h >> 1] & mask) == mask);
1024 	case 2:
1025 		mask = 0x03 << ((h & 0x3) << 1);
1026 		return ((cp[h >> 2] & mask) == mask);
1027 	case 1:
1028 		mask = 0x01 << (h & 0x7);
1029 		return ((cp[h >> 3] & mask) == mask);
1030 	default:
1031 		panic("isblock bad fs_frag");
1032 		return;
1033 	}
1034 }
1035 
1036 /*
1037  * take a block out of the map
1038  */
1039 clrblock(fs, cp, h)
1040 	struct fs *fs;
1041 	unsigned char *cp;
1042 	int h;
1043 {
1044 	switch ((fs)->fs_frag) {
1045 	case 8:
1046 		cp[h] = 0;
1047 		return;
1048 	case 4:
1049 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1050 		return;
1051 	case 2:
1052 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1053 		return;
1054 	case 1:
1055 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1056 		return;
1057 	default:
1058 		panic("clrblock bad fs_frag");
1059 		return;
1060 	}
1061 }
1062 
1063 /*
1064  * put a block into the map
1065  */
1066 setblock(fs, cp, h)
1067 	struct fs *fs;
1068 	unsigned char *cp;
1069 	int h;
1070 {
1071 	switch (fs->fs_frag) {
1072 	case 8:
1073 		cp[h] = 0xff;
1074 		return;
1075 	case 4:
1076 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1077 		return;
1078 	case 2:
1079 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1080 		return;
1081 	case 1:
1082 		cp[h >> 3] |= (0x01 << (h & 0x7));
1083 		return;
1084 	default:
1085 		panic("setblock bad fs_frag");
1086 		return;
1087 	}
1088 }
1089