xref: /original-bsd/sys/ufs/ffs/ffs_alloc.c (revision f737e041)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)ffs_alloc.c	8.8 (Berkeley) 02/21/94
8  */
9 
10 #include <sys/param.h>
11 #include <sys/systm.h>
12 #include <sys/buf.h>
13 #include <sys/proc.h>
14 #include <sys/vnode.h>
15 #include <sys/mount.h>
16 #include <sys/kernel.h>
17 #include <sys/syslog.h>
18 
19 #include <vm/vm.h>
20 
21 #include <ufs/ufs/quota.h>
22 #include <ufs/ufs/inode.h>
23 
24 #include <ufs/ffs/fs.h>
25 #include <ufs/ffs/ffs_extern.h>
26 
27 extern u_long nextgennumber;
28 
29 static daddr_t	ffs_alloccg __P((struct inode *, int, daddr_t, int));
30 static daddr_t	ffs_alloccgblk __P((struct fs *, struct cg *, daddr_t));
31 static daddr_t	ffs_clusteralloc __P((struct inode *, int, daddr_t, int));
32 static ino_t	ffs_dirpref __P((struct fs *));
33 static daddr_t	ffs_fragextend __P((struct inode *, int, long, int, int));
34 static void	ffs_fserr __P((struct fs *, u_int, char *));
35 static u_long	ffs_hashalloc
36 		    __P((struct inode *, int, long, int, u_long (*)()));
37 static ino_t	ffs_nodealloccg __P((struct inode *, int, daddr_t, int));
38 static daddr_t	ffs_mapsearch __P((struct fs *, struct cg *, daddr_t, int));
39 
40 /*
41  * Allocate a block in the file system.
42  *
43  * The size of the requested block is given, which must be some
44  * multiple of fs_fsize and <= fs_bsize.
45  * A preference may be optionally specified. If a preference is given
46  * the following hierarchy is used to allocate a block:
47  *   1) allocate the requested block.
48  *   2) allocate a rotationally optimal block in the same cylinder.
49  *   3) allocate a block in the same cylinder group.
50  *   4) quadradically rehash into other cylinder groups, until an
51  *      available block is located.
52  * If no block preference is given the following heirarchy is used
53  * to allocate a block:
54  *   1) allocate a block in the cylinder group that contains the
55  *      inode for the file.
56  *   2) quadradically rehash into other cylinder groups, until an
57  *      available block is located.
58  */
59 ffs_alloc(ip, lbn, bpref, size, cred, bnp)
60 	register struct inode *ip;
61 	daddr_t lbn, bpref;
62 	int size;
63 	struct ucred *cred;
64 	daddr_t *bnp;
65 {
66 	register struct fs *fs;
67 	daddr_t bno;
68 	int cg, error;
69 
70 	*bnp = 0;
71 	fs = ip->i_fs;
72 #ifdef DIAGNOSTIC
73 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
74 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
75 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
76 		panic("ffs_alloc: bad size");
77 	}
78 	if (cred == NOCRED)
79 		panic("ffs_alloc: missing credential\n");
80 #endif /* DIAGNOSTIC */
81 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
82 		goto nospace;
83 	if (cred->cr_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
84 		goto nospace;
85 #ifdef QUOTA
86 	if (error = chkdq(ip, (long)btodb(size), cred, 0))
87 		return (error);
88 #endif
89 	if (bpref >= fs->fs_size)
90 		bpref = 0;
91 	if (bpref == 0)
92 		cg = ino_to_cg(fs, ip->i_number);
93 	else
94 		cg = dtog(fs, bpref);
95 	bno = (daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
96 	    (u_long (*)())ffs_alloccg);
97 	if (bno > 0) {
98 		ip->i_blocks += btodb(size);
99 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
100 		*bnp = bno;
101 		return (0);
102 	}
103 #ifdef QUOTA
104 	/*
105 	 * Restore user's disk quota because allocation failed.
106 	 */
107 	(void) chkdq(ip, (long)-btodb(size), cred, FORCE);
108 #endif
109 nospace:
110 	ffs_fserr(fs, cred->cr_uid, "file system full");
111 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
112 	return (ENOSPC);
113 }
114 
115 /*
116  * Reallocate a fragment to a bigger size
117  *
118  * The number and size of the old block is given, and a preference
119  * and new size is also specified. The allocator attempts to extend
120  * the original block. Failing that, the regular block allocator is
121  * invoked to get an appropriate block.
122  */
123 ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
124 	register struct inode *ip;
125 	daddr_t lbprev;
126 	daddr_t bpref;
127 	int osize, nsize;
128 	struct ucred *cred;
129 	struct buf **bpp;
130 {
131 	register struct fs *fs;
132 	struct buf *bp;
133 	int cg, request, error;
134 	daddr_t bprev, bno;
135 
136 	*bpp = 0;
137 	fs = ip->i_fs;
138 #ifdef DIAGNOSTIC
139 	if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
140 	    (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
141 		printf(
142 		    "dev = 0x%x, bsize = %d, osize = %d, nsize = %d, fs = %s\n",
143 		    ip->i_dev, fs->fs_bsize, osize, nsize, fs->fs_fsmnt);
144 		panic("ffs_realloccg: bad size");
145 	}
146 	if (cred == NOCRED)
147 		panic("ffs_realloccg: missing credential\n");
148 #endif /* DIAGNOSTIC */
149 	if (cred->cr_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
150 		goto nospace;
151 	if ((bprev = ip->i_db[lbprev]) == 0) {
152 		printf("dev = 0x%x, bsize = %d, bprev = %d, fs = %s\n",
153 		    ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
154 		panic("ffs_realloccg: bad bprev");
155 	}
156 	/*
157 	 * Allocate the extra space in the buffer.
158 	 */
159 	if (error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp)) {
160 		brelse(bp);
161 		return (error);
162 	}
163 #ifdef QUOTA
164 	if (error = chkdq(ip, (long)btodb(nsize - osize), cred, 0)) {
165 		brelse(bp);
166 		return (error);
167 	}
168 #endif
169 	/*
170 	 * Check for extension in the existing location.
171 	 */
172 	cg = dtog(fs, bprev);
173 	if (bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize)) {
174 		if (bp->b_blkno != fsbtodb(fs, bno))
175 			panic("bad blockno");
176 		ip->i_blocks += btodb(nsize - osize);
177 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
178 		allocbuf(bp, nsize);
179 		bp->b_flags |= B_DONE;
180 		bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
181 		*bpp = bp;
182 		return (0);
183 	}
184 	/*
185 	 * Allocate a new disk location.
186 	 */
187 	if (bpref >= fs->fs_size)
188 		bpref = 0;
189 	switch ((int)fs->fs_optim) {
190 	case FS_OPTSPACE:
191 		/*
192 		 * Allocate an exact sized fragment. Although this makes
193 		 * best use of space, we will waste time relocating it if
194 		 * the file continues to grow. If the fragmentation is
195 		 * less than half of the minimum free reserve, we choose
196 		 * to begin optimizing for time.
197 		 */
198 		request = nsize;
199 		if (fs->fs_minfree < 5 ||
200 		    fs->fs_cstotal.cs_nffree >
201 		    fs->fs_dsize * fs->fs_minfree / (2 * 100))
202 			break;
203 		log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
204 			fs->fs_fsmnt);
205 		fs->fs_optim = FS_OPTTIME;
206 		break;
207 	case FS_OPTTIME:
208 		/*
209 		 * At this point we have discovered a file that is trying to
210 		 * grow a small fragment to a larger fragment. To save time,
211 		 * we allocate a full sized block, then free the unused portion.
212 		 * If the file continues to grow, the `ffs_fragextend' call
213 		 * above will be able to grow it in place without further
214 		 * copying. If aberrant programs cause disk fragmentation to
215 		 * grow within 2% of the free reserve, we choose to begin
216 		 * optimizing for space.
217 		 */
218 		request = fs->fs_bsize;
219 		if (fs->fs_cstotal.cs_nffree <
220 		    fs->fs_dsize * (fs->fs_minfree - 2) / 100)
221 			break;
222 		log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
223 			fs->fs_fsmnt);
224 		fs->fs_optim = FS_OPTSPACE;
225 		break;
226 	default:
227 		printf("dev = 0x%x, optim = %d, fs = %s\n",
228 		    ip->i_dev, fs->fs_optim, fs->fs_fsmnt);
229 		panic("ffs_realloccg: bad optim");
230 		/* NOTREACHED */
231 	}
232 	bno = (daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
233 	    (u_long (*)())ffs_alloccg);
234 	if (bno > 0) {
235 		bp->b_blkno = fsbtodb(fs, bno);
236 		(void) vnode_pager_uncache(ITOV(ip));
237 		ffs_blkfree(ip, bprev, (long)osize);
238 		if (nsize < request)
239 			ffs_blkfree(ip, bno + numfrags(fs, nsize),
240 			    (long)(request - nsize));
241 		ip->i_blocks += btodb(nsize - osize);
242 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
243 		allocbuf(bp, nsize);
244 		bp->b_flags |= B_DONE;
245 		bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
246 		*bpp = bp;
247 		return (0);
248 	}
249 #ifdef QUOTA
250 	/*
251 	 * Restore user's disk quota because allocation failed.
252 	 */
253 	(void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
254 #endif
255 	brelse(bp);
256 nospace:
257 	/*
258 	 * no space available
259 	 */
260 	ffs_fserr(fs, cred->cr_uid, "file system full");
261 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
262 	return (ENOSPC);
263 }
264 
265 /*
266  * Reallocate a sequence of blocks into a contiguous sequence of blocks.
267  *
268  * The vnode and an array of buffer pointers for a range of sequential
269  * logical blocks to be made contiguous is given. The allocator attempts
270  * to find a range of sequential blocks starting as close as possible to
271  * an fs_rotdelay offset from the end of the allocation for the logical
272  * block immediately preceeding the current range. If successful, the
273  * physical block numbers in the buffer pointers and in the inode are
274  * changed to reflect the new allocation. If unsuccessful, the allocation
275  * is left unchanged. The success in doing the reallocation is returned.
276  * Note that the error return is not reflected back to the user. Rather
277  * the previous block allocation will be used.
278  */
279 #include <sys/sysctl.h>
280 int doasyncfree = 1;
281 struct ctldebug debug14 = { "doasyncfree", &doasyncfree };
282 int
283 ffs_reallocblks(ap)
284 	struct vop_reallocblks_args /* {
285 		struct vnode *a_vp;
286 		struct cluster_save *a_buflist;
287 	} */ *ap;
288 {
289 	struct fs *fs;
290 	struct inode *ip;
291 	struct vnode *vp;
292 	struct buf *sbp, *ebp;
293 	daddr_t *bap, *sbap, *ebap;
294 	struct cluster_save *buflist;
295 	daddr_t start_lbn, end_lbn, soff, eoff, newblk, blkno;
296 	struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
297 	int i, len, start_lvl, end_lvl, pref, ssize;
298 
299 	vp = ap->a_vp;
300 	ip = VTOI(vp);
301 	fs = ip->i_fs;
302 	if (fs->fs_contigsumsize <= 0)
303 		return (ENOSPC);
304 	buflist = ap->a_buflist;
305 	len = buflist->bs_nchildren;
306 	start_lbn = buflist->bs_children[0]->b_lblkno;
307 	end_lbn = start_lbn + len - 1;
308 #ifdef DIAGNOSTIC
309 	for (i = 1; i < len; i++)
310 		if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
311 			panic("ffs_reallocblks: non-cluster");
312 #endif
313 	/*
314 	 * If the latest allocation is in a new cylinder group, assume that
315 	 * the filesystem has decided to move and do not force it back to
316 	 * the previous cylinder group.
317 	 */
318 	if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
319 	    dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
320 		return (ENOSPC);
321 	if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
322 	    ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
323 		return (ENOSPC);
324 	/*
325 	 * Get the starting offset and block map for the first block.
326 	 */
327 	if (start_lvl == 0) {
328 		sbap = &ip->i_db[0];
329 		soff = start_lbn;
330 	} else {
331 		idp = &start_ap[start_lvl - 1];
332 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
333 			brelse(sbp);
334 			return (ENOSPC);
335 		}
336 		sbap = (daddr_t *)sbp->b_data;
337 		soff = idp->in_off;
338 	}
339 	/*
340 	 * Find the preferred location for the cluster.
341 	 */
342 	pref = ffs_blkpref(ip, start_lbn, soff, sbap);
343 	/*
344 	 * If the block range spans two block maps, get the second map.
345 	 */
346 	if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
347 		ssize = len;
348 	} else {
349 #ifdef DIAGNOSTIC
350 		if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
351 			panic("ffs_reallocblk: start == end");
352 #endif
353 		ssize = len - (idp->in_off + 1);
354 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
355 			goto fail;
356 		ebap = (daddr_t *)ebp->b_data;
357 	}
358 	/*
359 	 * Search the block map looking for an allocation of the desired size.
360 	 */
361 	if ((newblk = (daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
362 	    len, (u_long (*)())ffs_clusteralloc)) == 0)
363 		goto fail;
364 	/*
365 	 * We have found a new contiguous block.
366 	 *
367 	 * First we have to replace the old block pointers with the new
368 	 * block pointers in the inode and indirect blocks associated
369 	 * with the file.
370 	 */
371 	blkno = newblk;
372 	for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
373 		if (i == ssize)
374 			bap = ebap;
375 #ifdef DIAGNOSTIC
376 		if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap))
377 			panic("ffs_reallocblks: alloc mismatch");
378 #endif
379 		*bap++ = blkno;
380 	}
381 	/*
382 	 * Next we must write out the modified inode and indirect blocks.
383 	 * For strict correctness, the writes should be synchronous since
384 	 * the old block values may have been written to disk. In practise
385 	 * they are almost never written, but if we are concerned about
386 	 * strict correctness, the `doasyncfree' flag should be set to zero.
387 	 *
388 	 * The test on `doasyncfree' should be changed to test a flag
389 	 * that shows whether the associated buffers and inodes have
390 	 * been written. The flag should be set when the cluster is
391 	 * started and cleared whenever the buffer or inode is flushed.
392 	 * We can then check below to see if it is set, and do the
393 	 * synchronous write only when it has been cleared.
394 	 */
395 	if (sbap != &ip->i_db[0]) {
396 		if (doasyncfree)
397 			bdwrite(sbp);
398 		else
399 			bwrite(sbp);
400 	} else {
401 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
402 		if (!doasyncfree)
403 			VOP_UPDATE(vp, &time, &time, MNT_WAIT);
404 	}
405 	if (ssize < len)
406 		if (doasyncfree)
407 			bdwrite(ebp);
408 		else
409 			bwrite(ebp);
410 	/*
411 	 * Last, free the old blocks and assign the new blocks to the buffers.
412 	 */
413 	for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
414 		ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
415 		    fs->fs_bsize);
416 		buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
417 	}
418 	return (0);
419 
420 fail:
421 	if (ssize < len)
422 		brelse(ebp);
423 	if (sbap != &ip->i_db[0])
424 		brelse(sbp);
425 	return (ENOSPC);
426 }
427 
428 /*
429  * Allocate an inode in the file system.
430  *
431  * If allocating a directory, use ffs_dirpref to select the inode.
432  * If allocating in a directory, the following hierarchy is followed:
433  *   1) allocate the preferred inode.
434  *   2) allocate an inode in the same cylinder group.
435  *   3) quadradically rehash into other cylinder groups, until an
436  *      available inode is located.
437  * If no inode preference is given the following heirarchy is used
438  * to allocate an inode:
439  *   1) allocate an inode in cylinder group 0.
440  *   2) quadradically rehash into other cylinder groups, until an
441  *      available inode is located.
442  */
443 ffs_valloc(ap)
444 	struct vop_valloc_args /* {
445 		struct vnode *a_pvp;
446 		int a_mode;
447 		struct ucred *a_cred;
448 		struct vnode **a_vpp;
449 	} */ *ap;
450 {
451 	register struct vnode *pvp = ap->a_pvp;
452 	register struct inode *pip;
453 	register struct fs *fs;
454 	register struct inode *ip;
455 	mode_t mode = ap->a_mode;
456 	ino_t ino, ipref;
457 	int cg, error;
458 
459 	*ap->a_vpp = NULL;
460 	pip = VTOI(pvp);
461 	fs = pip->i_fs;
462 	if (fs->fs_cstotal.cs_nifree == 0)
463 		goto noinodes;
464 
465 	if ((mode & IFMT) == IFDIR)
466 		ipref = ffs_dirpref(fs);
467 	else
468 		ipref = pip->i_number;
469 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
470 		ipref = 0;
471 	cg = ino_to_cg(fs, ipref);
472 	ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode, ffs_nodealloccg);
473 	if (ino == 0)
474 		goto noinodes;
475 	error = VFS_VGET(pvp->v_mount, ino, ap->a_vpp);
476 	if (error) {
477 		VOP_VFREE(pvp, ino, mode);
478 		return (error);
479 	}
480 	ip = VTOI(*ap->a_vpp);
481 	if (ip->i_mode) {
482 		printf("mode = 0%o, inum = %d, fs = %s\n",
483 		    ip->i_mode, ip->i_number, fs->fs_fsmnt);
484 		panic("ffs_valloc: dup alloc");
485 	}
486 	if (ip->i_blocks) {				/* XXX */
487 		printf("free inode %s/%d had %d blocks\n",
488 		    fs->fs_fsmnt, ino, ip->i_blocks);
489 		ip->i_blocks = 0;
490 	}
491 	ip->i_flags = 0;
492 	/*
493 	 * Set up a new generation number for this inode.
494 	 */
495 	if (++nextgennumber < (u_long)time.tv_sec)
496 		nextgennumber = time.tv_sec;
497 	ip->i_gen = nextgennumber;
498 	return (0);
499 noinodes:
500 	ffs_fserr(fs, ap->a_cred->cr_uid, "out of inodes");
501 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
502 	return (ENOSPC);
503 }
504 
505 /*
506  * Find a cylinder to place a directory.
507  *
508  * The policy implemented by this algorithm is to select from
509  * among those cylinder groups with above the average number of
510  * free inodes, the one with the smallest number of directories.
511  */
512 static ino_t
513 ffs_dirpref(fs)
514 	register struct fs *fs;
515 {
516 	int cg, minndir, mincg, avgifree;
517 
518 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
519 	minndir = fs->fs_ipg;
520 	mincg = 0;
521 	for (cg = 0; cg < fs->fs_ncg; cg++)
522 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
523 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
524 			mincg = cg;
525 			minndir = fs->fs_cs(fs, cg).cs_ndir;
526 		}
527 	return ((ino_t)(fs->fs_ipg * mincg));
528 }
529 
530 /*
531  * Select the desired position for the next block in a file.  The file is
532  * logically divided into sections. The first section is composed of the
533  * direct blocks. Each additional section contains fs_maxbpg blocks.
534  *
535  * If no blocks have been allocated in the first section, the policy is to
536  * request a block in the same cylinder group as the inode that describes
537  * the file. If no blocks have been allocated in any other section, the
538  * policy is to place the section in a cylinder group with a greater than
539  * average number of free blocks.  An appropriate cylinder group is found
540  * by using a rotor that sweeps the cylinder groups. When a new group of
541  * blocks is needed, the sweep begins in the cylinder group following the
542  * cylinder group from which the previous allocation was made. The sweep
543  * continues until a cylinder group with greater than the average number
544  * of free blocks is found. If the allocation is for the first block in an
545  * indirect block, the information on the previous allocation is unavailable;
546  * here a best guess is made based upon the logical block number being
547  * allocated.
548  *
549  * If a section is already partially allocated, the policy is to
550  * contiguously allocate fs_maxcontig blocks.  The end of one of these
551  * contiguous blocks and the beginning of the next is physically separated
552  * so that the disk head will be in transit between them for at least
553  * fs_rotdelay milliseconds.  This is to allow time for the processor to
554  * schedule another I/O transfer.
555  */
556 daddr_t
557 ffs_blkpref(ip, lbn, indx, bap)
558 	struct inode *ip;
559 	daddr_t lbn;
560 	int indx;
561 	daddr_t *bap;
562 {
563 	register struct fs *fs;
564 	register int cg;
565 	int avgbfree, startcg;
566 	daddr_t nextblk;
567 
568 	fs = ip->i_fs;
569 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
570 		if (lbn < NDADDR) {
571 			cg = ino_to_cg(fs, ip->i_number);
572 			return (fs->fs_fpg * cg + fs->fs_frag);
573 		}
574 		/*
575 		 * Find a cylinder with greater than average number of
576 		 * unused data blocks.
577 		 */
578 		if (indx == 0 || bap[indx - 1] == 0)
579 			startcg =
580 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
581 		else
582 			startcg = dtog(fs, bap[indx - 1]) + 1;
583 		startcg %= fs->fs_ncg;
584 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
585 		for (cg = startcg; cg < fs->fs_ncg; cg++)
586 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
587 				fs->fs_cgrotor = cg;
588 				return (fs->fs_fpg * cg + fs->fs_frag);
589 			}
590 		for (cg = 0; cg <= startcg; cg++)
591 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
592 				fs->fs_cgrotor = cg;
593 				return (fs->fs_fpg * cg + fs->fs_frag);
594 			}
595 		return (NULL);
596 	}
597 	/*
598 	 * One or more previous blocks have been laid out. If less
599 	 * than fs_maxcontig previous blocks are contiguous, the
600 	 * next block is requested contiguously, otherwise it is
601 	 * requested rotationally delayed by fs_rotdelay milliseconds.
602 	 */
603 	nextblk = bap[indx - 1] + fs->fs_frag;
604 	if (indx < fs->fs_maxcontig || bap[indx - fs->fs_maxcontig] +
605 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
606 		return (nextblk);
607 	if (fs->fs_rotdelay != 0)
608 		/*
609 		 * Here we convert ms of delay to frags as:
610 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
611 		 *	((sect/frag) * (ms/sec))
612 		 * then round up to the next block.
613 		 */
614 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
615 		    (NSPF(fs) * 1000), fs->fs_frag);
616 	return (nextblk);
617 }
618 
619 /*
620  * Implement the cylinder overflow algorithm.
621  *
622  * The policy implemented by this algorithm is:
623  *   1) allocate the block in its requested cylinder group.
624  *   2) quadradically rehash on the cylinder group number.
625  *   3) brute force search for a free block.
626  */
627 /*VARARGS5*/
628 static u_long
629 ffs_hashalloc(ip, cg, pref, size, allocator)
630 	struct inode *ip;
631 	int cg;
632 	long pref;
633 	int size;	/* size for data blocks, mode for inodes */
634 	u_long (*allocator)();
635 {
636 	register struct fs *fs;
637 	long result;
638 	int i, icg = cg;
639 
640 	fs = ip->i_fs;
641 	/*
642 	 * 1: preferred cylinder group
643 	 */
644 	result = (*allocator)(ip, cg, pref, size);
645 	if (result)
646 		return (result);
647 	/*
648 	 * 2: quadratic rehash
649 	 */
650 	for (i = 1; i < fs->fs_ncg; i *= 2) {
651 		cg += i;
652 		if (cg >= fs->fs_ncg)
653 			cg -= fs->fs_ncg;
654 		result = (*allocator)(ip, cg, 0, size);
655 		if (result)
656 			return (result);
657 	}
658 	/*
659 	 * 3: brute force search
660 	 * Note that we start at i == 2, since 0 was checked initially,
661 	 * and 1 is always checked in the quadratic rehash.
662 	 */
663 	cg = (icg + 2) % fs->fs_ncg;
664 	for (i = 2; i < fs->fs_ncg; i++) {
665 		result = (*allocator)(ip, cg, 0, size);
666 		if (result)
667 			return (result);
668 		cg++;
669 		if (cg == fs->fs_ncg)
670 			cg = 0;
671 	}
672 	return (NULL);
673 }
674 
675 /*
676  * Determine whether a fragment can be extended.
677  *
678  * Check to see if the necessary fragments are available, and
679  * if they are, allocate them.
680  */
681 static daddr_t
682 ffs_fragextend(ip, cg, bprev, osize, nsize)
683 	struct inode *ip;
684 	int cg;
685 	long bprev;
686 	int osize, nsize;
687 {
688 	register struct fs *fs;
689 	register struct cg *cgp;
690 	struct buf *bp;
691 	long bno;
692 	int frags, bbase;
693 	int i, error;
694 
695 	fs = ip->i_fs;
696 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
697 		return (NULL);
698 	frags = numfrags(fs, nsize);
699 	bbase = fragnum(fs, bprev);
700 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
701 		/* cannot extend across a block boundary */
702 		return (NULL);
703 	}
704 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
705 		(int)fs->fs_cgsize, NOCRED, &bp);
706 	if (error) {
707 		brelse(bp);
708 		return (NULL);
709 	}
710 	cgp = (struct cg *)bp->b_data;
711 	if (!cg_chkmagic(cgp)) {
712 		brelse(bp);
713 		return (NULL);
714 	}
715 	cgp->cg_time = time.tv_sec;
716 	bno = dtogd(fs, bprev);
717 	for (i = numfrags(fs, osize); i < frags; i++)
718 		if (isclr(cg_blksfree(cgp), bno + i)) {
719 			brelse(bp);
720 			return (NULL);
721 		}
722 	/*
723 	 * the current fragment can be extended
724 	 * deduct the count on fragment being extended into
725 	 * increase the count on the remaining fragment (if any)
726 	 * allocate the extended piece
727 	 */
728 	for (i = frags; i < fs->fs_frag - bbase; i++)
729 		if (isclr(cg_blksfree(cgp), bno + i))
730 			break;
731 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
732 	if (i != frags)
733 		cgp->cg_frsum[i - frags]++;
734 	for (i = numfrags(fs, osize); i < frags; i++) {
735 		clrbit(cg_blksfree(cgp), bno + i);
736 		cgp->cg_cs.cs_nffree--;
737 		fs->fs_cstotal.cs_nffree--;
738 		fs->fs_cs(fs, cg).cs_nffree--;
739 	}
740 	fs->fs_fmod = 1;
741 	bdwrite(bp);
742 	return (bprev);
743 }
744 
745 /*
746  * Determine whether a block can be allocated.
747  *
748  * Check to see if a block of the appropriate size is available,
749  * and if it is, allocate it.
750  */
751 static daddr_t
752 ffs_alloccg(ip, cg, bpref, size)
753 	struct inode *ip;
754 	int cg;
755 	daddr_t bpref;
756 	int size;
757 {
758 	register struct fs *fs;
759 	register struct cg *cgp;
760 	struct buf *bp;
761 	register int i;
762 	int error, bno, frags, allocsiz;
763 
764 	fs = ip->i_fs;
765 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
766 		return (NULL);
767 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
768 		(int)fs->fs_cgsize, NOCRED, &bp);
769 	if (error) {
770 		brelse(bp);
771 		return (NULL);
772 	}
773 	cgp = (struct cg *)bp->b_data;
774 	if (!cg_chkmagic(cgp) ||
775 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
776 		brelse(bp);
777 		return (NULL);
778 	}
779 	cgp->cg_time = time.tv_sec;
780 	if (size == fs->fs_bsize) {
781 		bno = ffs_alloccgblk(fs, cgp, bpref);
782 		bdwrite(bp);
783 		return (bno);
784 	}
785 	/*
786 	 * check to see if any fragments are already available
787 	 * allocsiz is the size which will be allocated, hacking
788 	 * it down to a smaller size if necessary
789 	 */
790 	frags = numfrags(fs, size);
791 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
792 		if (cgp->cg_frsum[allocsiz] != 0)
793 			break;
794 	if (allocsiz == fs->fs_frag) {
795 		/*
796 		 * no fragments were available, so a block will be
797 		 * allocated, and hacked up
798 		 */
799 		if (cgp->cg_cs.cs_nbfree == 0) {
800 			brelse(bp);
801 			return (NULL);
802 		}
803 		bno = ffs_alloccgblk(fs, cgp, bpref);
804 		bpref = dtogd(fs, bno);
805 		for (i = frags; i < fs->fs_frag; i++)
806 			setbit(cg_blksfree(cgp), bpref + i);
807 		i = fs->fs_frag - frags;
808 		cgp->cg_cs.cs_nffree += i;
809 		fs->fs_cstotal.cs_nffree += i;
810 		fs->fs_cs(fs, cg).cs_nffree += i;
811 		fs->fs_fmod = 1;
812 		cgp->cg_frsum[i]++;
813 		bdwrite(bp);
814 		return (bno);
815 	}
816 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
817 	if (bno < 0) {
818 		brelse(bp);
819 		return (NULL);
820 	}
821 	for (i = 0; i < frags; i++)
822 		clrbit(cg_blksfree(cgp), bno + i);
823 	cgp->cg_cs.cs_nffree -= frags;
824 	fs->fs_cstotal.cs_nffree -= frags;
825 	fs->fs_cs(fs, cg).cs_nffree -= frags;
826 	fs->fs_fmod = 1;
827 	cgp->cg_frsum[allocsiz]--;
828 	if (frags != allocsiz)
829 		cgp->cg_frsum[allocsiz - frags]++;
830 	bdwrite(bp);
831 	return (cg * fs->fs_fpg + bno);
832 }
833 
834 /*
835  * Allocate a block in a cylinder group.
836  *
837  * This algorithm implements the following policy:
838  *   1) allocate the requested block.
839  *   2) allocate a rotationally optimal block in the same cylinder.
840  *   3) allocate the next available block on the block rotor for the
841  *      specified cylinder group.
842  * Note that this routine only allocates fs_bsize blocks; these
843  * blocks may be fragmented by the routine that allocates them.
844  */
845 static daddr_t
846 ffs_alloccgblk(fs, cgp, bpref)
847 	register struct fs *fs;
848 	register struct cg *cgp;
849 	daddr_t bpref;
850 {
851 	daddr_t bno, blkno;
852 	int cylno, pos, delta;
853 	short *cylbp;
854 	register int i;
855 
856 	if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
857 		bpref = cgp->cg_rotor;
858 		goto norot;
859 	}
860 	bpref = blknum(fs, bpref);
861 	bpref = dtogd(fs, bpref);
862 	/*
863 	 * if the requested block is available, use it
864 	 */
865 	if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
866 		bno = bpref;
867 		goto gotit;
868 	}
869 	/*
870 	 * check for a block available on the same cylinder
871 	 */
872 	cylno = cbtocylno(fs, bpref);
873 	if (cg_blktot(cgp)[cylno] == 0)
874 		goto norot;
875 	if (fs->fs_cpc == 0) {
876 		/*
877 		 * Block layout information is not available.
878 		 * Leaving bpref unchanged means we take the
879 		 * next available free block following the one
880 		 * we just allocated. Hopefully this will at
881 		 * least hit a track cache on drives of unknown
882 		 * geometry (e.g. SCSI).
883 		 */
884 		goto norot;
885 	}
886 	/*
887 	 * check the summary information to see if a block is
888 	 * available in the requested cylinder starting at the
889 	 * requested rotational position and proceeding around.
890 	 */
891 	cylbp = cg_blks(fs, cgp, cylno);
892 	pos = cbtorpos(fs, bpref);
893 	for (i = pos; i < fs->fs_nrpos; i++)
894 		if (cylbp[i] > 0)
895 			break;
896 	if (i == fs->fs_nrpos)
897 		for (i = 0; i < pos; i++)
898 			if (cylbp[i] > 0)
899 				break;
900 	if (cylbp[i] > 0) {
901 		/*
902 		 * found a rotational position, now find the actual
903 		 * block. A panic if none is actually there.
904 		 */
905 		pos = cylno % fs->fs_cpc;
906 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
907 		if (fs_postbl(fs, pos)[i] == -1) {
908 			printf("pos = %d, i = %d, fs = %s\n",
909 			    pos, i, fs->fs_fsmnt);
910 			panic("ffs_alloccgblk: cyl groups corrupted");
911 		}
912 		for (i = fs_postbl(fs, pos)[i];; ) {
913 			if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
914 				bno = blkstofrags(fs, (bno + i));
915 				goto gotit;
916 			}
917 			delta = fs_rotbl(fs)[i];
918 			if (delta <= 0 ||
919 			    delta + i > fragstoblks(fs, fs->fs_fpg))
920 				break;
921 			i += delta;
922 		}
923 		printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
924 		panic("ffs_alloccgblk: can't find blk in cyl");
925 	}
926 norot:
927 	/*
928 	 * no blocks in the requested cylinder, so take next
929 	 * available one in this cylinder group.
930 	 */
931 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
932 	if (bno < 0)
933 		return (NULL);
934 	cgp->cg_rotor = bno;
935 gotit:
936 	blkno = fragstoblks(fs, bno);
937 	ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
938 	ffs_clusteracct(fs, cgp, blkno, -1);
939 	cgp->cg_cs.cs_nbfree--;
940 	fs->fs_cstotal.cs_nbfree--;
941 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
942 	cylno = cbtocylno(fs, bno);
943 	cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
944 	cg_blktot(cgp)[cylno]--;
945 	fs->fs_fmod = 1;
946 	return (cgp->cg_cgx * fs->fs_fpg + bno);
947 }
948 
949 /*
950  * Determine whether a cluster can be allocated.
951  *
952  * We do not currently check for optimal rotational layout if there
953  * are multiple choices in the same cylinder group. Instead we just
954  * take the first one that we find following bpref.
955  */
956 static daddr_t
957 ffs_clusteralloc(ip, cg, bpref, len)
958 	struct inode *ip;
959 	int cg;
960 	daddr_t bpref;
961 	int len;
962 {
963 	register struct fs *fs;
964 	register struct cg *cgp;
965 	struct buf *bp;
966 	int i, run, bno, bit, map;
967 	u_char *mapp;
968 
969 	fs = ip->i_fs;
970 	if (fs->fs_cs(fs, cg).cs_nbfree < len)
971 		return (NULL);
972 	if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
973 	    NOCRED, &bp))
974 		goto fail;
975 	cgp = (struct cg *)bp->b_data;
976 	if (!cg_chkmagic(cgp))
977 		goto fail;
978 	/*
979 	 * Check to see if a cluster of the needed size (or bigger) is
980 	 * available in this cylinder group.
981 	 */
982 	for (i = len; i <= fs->fs_contigsumsize; i++)
983 		if (cg_clustersum(cgp)[i] > 0)
984 			break;
985 	if (i > fs->fs_contigsumsize)
986 		goto fail;
987 	/*
988 	 * Search the cluster map to find a big enough cluster.
989 	 * We take the first one that we find, even if it is larger
990 	 * than we need as we prefer to get one close to the previous
991 	 * block allocation. We do not search before the current
992 	 * preference point as we do not want to allocate a block
993 	 * that is allocated before the previous one (as we will
994 	 * then have to wait for another pass of the elevator
995 	 * algorithm before it will be read). We prefer to fail and
996 	 * be recalled to try an allocation in the next cylinder group.
997 	 */
998 	if (dtog(fs, bpref) != cg)
999 		bpref = 0;
1000 	else
1001 		bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1002 	mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1003 	map = *mapp++;
1004 	bit = 1 << (bpref % NBBY);
1005 	for (run = 0, i = bpref; i < cgp->cg_nclusterblks; i++) {
1006 		if ((map & bit) == 0) {
1007 			run = 0;
1008 		} else {
1009 			run++;
1010 			if (run == len)
1011 				break;
1012 		}
1013 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
1014 			bit <<= 1;
1015 		} else {
1016 			map = *mapp++;
1017 			bit = 1;
1018 		}
1019 	}
1020 	if (i == cgp->cg_nclusterblks)
1021 		goto fail;
1022 	/*
1023 	 * Allocate the cluster that we have found.
1024 	 */
1025 	bno = cg * fs->fs_fpg + blkstofrags(fs, i - run + 1);
1026 	len = blkstofrags(fs, len);
1027 	for (i = 0; i < len; i += fs->fs_frag)
1028 		if (ffs_alloccgblk(fs, cgp, bno + i) != bno + i)
1029 			panic("ffs_clusteralloc: lost block");
1030 	brelse(bp);
1031 	return (bno);
1032 
1033 fail:
1034 	brelse(bp);
1035 	return (0);
1036 }
1037 
1038 /*
1039  * Determine whether an inode can be allocated.
1040  *
1041  * Check to see if an inode is available, and if it is,
1042  * allocate it using the following policy:
1043  *   1) allocate the requested inode.
1044  *   2) allocate the next available inode after the requested
1045  *      inode in the specified cylinder group.
1046  */
1047 static ino_t
1048 ffs_nodealloccg(ip, cg, ipref, mode)
1049 	struct inode *ip;
1050 	int cg;
1051 	daddr_t ipref;
1052 	int mode;
1053 {
1054 	register struct fs *fs;
1055 	register struct cg *cgp;
1056 	struct buf *bp;
1057 	int error, start, len, loc, map, i;
1058 
1059 	fs = ip->i_fs;
1060 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
1061 		return (NULL);
1062 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1063 		(int)fs->fs_cgsize, NOCRED, &bp);
1064 	if (error) {
1065 		brelse(bp);
1066 		return (NULL);
1067 	}
1068 	cgp = (struct cg *)bp->b_data;
1069 	if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1070 		brelse(bp);
1071 		return (NULL);
1072 	}
1073 	cgp->cg_time = time.tv_sec;
1074 	if (ipref) {
1075 		ipref %= fs->fs_ipg;
1076 		if (isclr(cg_inosused(cgp), ipref))
1077 			goto gotit;
1078 	}
1079 	start = cgp->cg_irotor / NBBY;
1080 	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1081 	loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1082 	if (loc == 0) {
1083 		len = start + 1;
1084 		start = 0;
1085 		loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1086 		if (loc == 0) {
1087 			printf("cg = %d, irotor = %d, fs = %s\n",
1088 			    cg, cgp->cg_irotor, fs->fs_fsmnt);
1089 			panic("ffs_nodealloccg: map corrupted");
1090 			/* NOTREACHED */
1091 		}
1092 	}
1093 	i = start + len - loc;
1094 	map = cg_inosused(cgp)[i];
1095 	ipref = i * NBBY;
1096 	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1097 		if ((map & i) == 0) {
1098 			cgp->cg_irotor = ipref;
1099 			goto gotit;
1100 		}
1101 	}
1102 	printf("fs = %s\n", fs->fs_fsmnt);
1103 	panic("ffs_nodealloccg: block not in map");
1104 	/* NOTREACHED */
1105 gotit:
1106 	setbit(cg_inosused(cgp), ipref);
1107 	cgp->cg_cs.cs_nifree--;
1108 	fs->fs_cstotal.cs_nifree--;
1109 	fs->fs_cs(fs, cg).cs_nifree--;
1110 	fs->fs_fmod = 1;
1111 	if ((mode & IFMT) == IFDIR) {
1112 		cgp->cg_cs.cs_ndir++;
1113 		fs->fs_cstotal.cs_ndir++;
1114 		fs->fs_cs(fs, cg).cs_ndir++;
1115 	}
1116 	bdwrite(bp);
1117 	return (cg * fs->fs_ipg + ipref);
1118 }
1119 
1120 /*
1121  * Free a block or fragment.
1122  *
1123  * The specified block or fragment is placed back in the
1124  * free map. If a fragment is deallocated, a possible
1125  * block reassembly is checked.
1126  */
1127 ffs_blkfree(ip, bno, size)
1128 	register struct inode *ip;
1129 	daddr_t bno;
1130 	long size;
1131 {
1132 	register struct fs *fs;
1133 	register struct cg *cgp;
1134 	struct buf *bp;
1135 	daddr_t blkno;
1136 	int i, error, cg, blk, frags, bbase;
1137 
1138 	fs = ip->i_fs;
1139 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1140 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
1141 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
1142 		panic("blkfree: bad size");
1143 	}
1144 	cg = dtog(fs, bno);
1145 	if ((u_int)bno >= fs->fs_size) {
1146 		printf("bad block %d, ino %d\n", bno, ip->i_number);
1147 		ffs_fserr(fs, ip->i_uid, "bad block");
1148 		return;
1149 	}
1150 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1151 		(int)fs->fs_cgsize, NOCRED, &bp);
1152 	if (error) {
1153 		brelse(bp);
1154 		return;
1155 	}
1156 	cgp = (struct cg *)bp->b_data;
1157 	if (!cg_chkmagic(cgp)) {
1158 		brelse(bp);
1159 		return;
1160 	}
1161 	cgp->cg_time = time.tv_sec;
1162 	bno = dtogd(fs, bno);
1163 	if (size == fs->fs_bsize) {
1164 		blkno = fragstoblks(fs, bno);
1165 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1166 			printf("dev = 0x%x, block = %d, fs = %s\n",
1167 			    ip->i_dev, bno, fs->fs_fsmnt);
1168 			panic("blkfree: freeing free block");
1169 		}
1170 		ffs_setblock(fs, cg_blksfree(cgp), blkno);
1171 		ffs_clusteracct(fs, cgp, blkno, 1);
1172 		cgp->cg_cs.cs_nbfree++;
1173 		fs->fs_cstotal.cs_nbfree++;
1174 		fs->fs_cs(fs, cg).cs_nbfree++;
1175 		i = cbtocylno(fs, bno);
1176 		cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1177 		cg_blktot(cgp)[i]++;
1178 	} else {
1179 		bbase = bno - fragnum(fs, bno);
1180 		/*
1181 		 * decrement the counts associated with the old frags
1182 		 */
1183 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
1184 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1185 		/*
1186 		 * deallocate the fragment
1187 		 */
1188 		frags = numfrags(fs, size);
1189 		for (i = 0; i < frags; i++) {
1190 			if (isset(cg_blksfree(cgp), bno + i)) {
1191 				printf("dev = 0x%x, block = %d, fs = %s\n",
1192 				    ip->i_dev, bno + i, fs->fs_fsmnt);
1193 				panic("blkfree: freeing free frag");
1194 			}
1195 			setbit(cg_blksfree(cgp), bno + i);
1196 		}
1197 		cgp->cg_cs.cs_nffree += i;
1198 		fs->fs_cstotal.cs_nffree += i;
1199 		fs->fs_cs(fs, cg).cs_nffree += i;
1200 		/*
1201 		 * add back in counts associated with the new frags
1202 		 */
1203 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
1204 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1205 		/*
1206 		 * if a complete block has been reassembled, account for it
1207 		 */
1208 		blkno = fragstoblks(fs, bbase);
1209 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1210 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
1211 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1212 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1213 			ffs_clusteracct(fs, cgp, blkno, 1);
1214 			cgp->cg_cs.cs_nbfree++;
1215 			fs->fs_cstotal.cs_nbfree++;
1216 			fs->fs_cs(fs, cg).cs_nbfree++;
1217 			i = cbtocylno(fs, bbase);
1218 			cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1219 			cg_blktot(cgp)[i]++;
1220 		}
1221 	}
1222 	fs->fs_fmod = 1;
1223 	bdwrite(bp);
1224 }
1225 
1226 /*
1227  * Free an inode.
1228  *
1229  * The specified inode is placed back in the free map.
1230  */
1231 int
1232 ffs_vfree(ap)
1233 	struct vop_vfree_args /* {
1234 		struct vnode *a_pvp;
1235 		ino_t a_ino;
1236 		int a_mode;
1237 	} */ *ap;
1238 {
1239 	register struct fs *fs;
1240 	register struct cg *cgp;
1241 	register struct inode *pip;
1242 	ino_t ino = ap->a_ino;
1243 	struct buf *bp;
1244 	int error, cg;
1245 
1246 	pip = VTOI(ap->a_pvp);
1247 	fs = pip->i_fs;
1248 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1249 		panic("ifree: range: dev = 0x%x, ino = %d, fs = %s\n",
1250 		    pip->i_dev, ino, fs->fs_fsmnt);
1251 	cg = ino_to_cg(fs, ino);
1252 	error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1253 		(int)fs->fs_cgsize, NOCRED, &bp);
1254 	if (error) {
1255 		brelse(bp);
1256 		return (0);
1257 	}
1258 	cgp = (struct cg *)bp->b_data;
1259 	if (!cg_chkmagic(cgp)) {
1260 		brelse(bp);
1261 		return (0);
1262 	}
1263 	cgp->cg_time = time.tv_sec;
1264 	ino %= fs->fs_ipg;
1265 	if (isclr(cg_inosused(cgp), ino)) {
1266 		printf("dev = 0x%x, ino = %d, fs = %s\n",
1267 		    pip->i_dev, ino, fs->fs_fsmnt);
1268 		if (fs->fs_ronly == 0)
1269 			panic("ifree: freeing free inode");
1270 	}
1271 	clrbit(cg_inosused(cgp), ino);
1272 	if (ino < cgp->cg_irotor)
1273 		cgp->cg_irotor = ino;
1274 	cgp->cg_cs.cs_nifree++;
1275 	fs->fs_cstotal.cs_nifree++;
1276 	fs->fs_cs(fs, cg).cs_nifree++;
1277 	if ((ap->a_mode & IFMT) == IFDIR) {
1278 		cgp->cg_cs.cs_ndir--;
1279 		fs->fs_cstotal.cs_ndir--;
1280 		fs->fs_cs(fs, cg).cs_ndir--;
1281 	}
1282 	fs->fs_fmod = 1;
1283 	bdwrite(bp);
1284 	return (0);
1285 }
1286 
1287 /*
1288  * Find a block of the specified size in the specified cylinder group.
1289  *
1290  * It is a panic if a request is made to find a block if none are
1291  * available.
1292  */
1293 static daddr_t
1294 ffs_mapsearch(fs, cgp, bpref, allocsiz)
1295 	register struct fs *fs;
1296 	register struct cg *cgp;
1297 	daddr_t bpref;
1298 	int allocsiz;
1299 {
1300 	daddr_t bno;
1301 	int start, len, loc, i;
1302 	int blk, field, subfield, pos;
1303 
1304 	/*
1305 	 * find the fragment by searching through the free block
1306 	 * map for an appropriate bit pattern
1307 	 */
1308 	if (bpref)
1309 		start = dtogd(fs, bpref) / NBBY;
1310 	else
1311 		start = cgp->cg_frotor / NBBY;
1312 	len = howmany(fs->fs_fpg, NBBY) - start;
1313 	loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1314 		(u_char *)fragtbl[fs->fs_frag],
1315 		(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1316 	if (loc == 0) {
1317 		len = start + 1;
1318 		start = 0;
1319 		loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1320 			(u_char *)fragtbl[fs->fs_frag],
1321 			(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1322 		if (loc == 0) {
1323 			printf("start = %d, len = %d, fs = %s\n",
1324 			    start, len, fs->fs_fsmnt);
1325 			panic("ffs_alloccg: map corrupted");
1326 			/* NOTREACHED */
1327 		}
1328 	}
1329 	bno = (start + len - loc) * NBBY;
1330 	cgp->cg_frotor = bno;
1331 	/*
1332 	 * found the byte in the map
1333 	 * sift through the bits to find the selected frag
1334 	 */
1335 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1336 		blk = blkmap(fs, cg_blksfree(cgp), bno);
1337 		blk <<= 1;
1338 		field = around[allocsiz];
1339 		subfield = inside[allocsiz];
1340 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1341 			if ((blk & field) == subfield)
1342 				return (bno + pos);
1343 			field <<= 1;
1344 			subfield <<= 1;
1345 		}
1346 	}
1347 	printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt);
1348 	panic("ffs_alloccg: block not in map");
1349 	return (-1);
1350 }
1351 
1352 /*
1353  * Update the cluster map because of an allocation or free.
1354  *
1355  * Cnt == 1 means free; cnt == -1 means allocating.
1356  */
1357 ffs_clusteracct(fs, cgp, blkno, cnt)
1358 	struct fs *fs;
1359 	struct cg *cgp;
1360 	daddr_t blkno;
1361 	int cnt;
1362 {
1363 	long *sump;
1364 	u_char *freemapp, *mapp;
1365 	int i, start, end, forw, back, map, bit;
1366 
1367 	if (fs->fs_contigsumsize <= 0)
1368 		return;
1369 	freemapp = cg_clustersfree(cgp);
1370 	sump = cg_clustersum(cgp);
1371 	/*
1372 	 * Allocate or clear the actual block.
1373 	 */
1374 	if (cnt > 0)
1375 		setbit(freemapp, blkno);
1376 	else
1377 		clrbit(freemapp, blkno);
1378 	/*
1379 	 * Find the size of the cluster going forward.
1380 	 */
1381 	start = blkno + 1;
1382 	end = start + fs->fs_contigsumsize;
1383 	if (end >= cgp->cg_nclusterblks)
1384 		end = cgp->cg_nclusterblks;
1385 	mapp = &freemapp[start / NBBY];
1386 	map = *mapp++;
1387 	bit = 1 << (start % NBBY);
1388 	for (i = start; i < end; i++) {
1389 		if ((map & bit) == 0)
1390 			break;
1391 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
1392 			bit <<= 1;
1393 		} else {
1394 			map = *mapp++;
1395 			bit = 1;
1396 		}
1397 	}
1398 	forw = i - start;
1399 	/*
1400 	 * Find the size of the cluster going backward.
1401 	 */
1402 	start = blkno - 1;
1403 	end = start - fs->fs_contigsumsize;
1404 	if (end < 0)
1405 		end = -1;
1406 	mapp = &freemapp[start / NBBY];
1407 	map = *mapp--;
1408 	bit = 1 << (start % NBBY);
1409 	for (i = start; i > end; i--) {
1410 		if ((map & bit) == 0)
1411 			break;
1412 		if ((i & (NBBY - 1)) != 0) {
1413 			bit >>= 1;
1414 		} else {
1415 			map = *mapp--;
1416 			bit = 1 << (NBBY - 1);
1417 		}
1418 	}
1419 	back = start - i;
1420 	/*
1421 	 * Account for old cluster and the possibly new forward and
1422 	 * back clusters.
1423 	 */
1424 	i = back + forw + 1;
1425 	if (i > fs->fs_contigsumsize)
1426 		i = fs->fs_contigsumsize;
1427 	sump[i] += cnt;
1428 	if (back > 0)
1429 		sump[back] -= cnt;
1430 	if (forw > 0)
1431 		sump[forw] -= cnt;
1432 }
1433 
1434 /*
1435  * Fserr prints the name of a file system with an error diagnostic.
1436  *
1437  * The form of the error message is:
1438  *	fs: error message
1439  */
1440 static void
1441 ffs_fserr(fs, uid, cp)
1442 	struct fs *fs;
1443 	u_int uid;
1444 	char *cp;
1445 {
1446 
1447 	log(LOG_ERR, "uid %d on %s: %s\n", uid, fs->fs_fsmnt, cp);
1448 }
1449