xref: /original-bsd/sys/ufs/ffs/ffs_alloc.c (revision 9507f4d3)
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	8.6 (Berkeley) 02/05/94
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 int
280 ffs_reallocblks(ap)
281 	struct vop_reallocblks_args /* {
282 		struct vnode *a_vp;
283 		struct cluster_save *a_buflist;
284 	} */ *ap;
285 {
286 	struct fs *fs;
287 	struct inode *ip;
288 	struct vnode *vp;
289 	struct buf *sbp, *ebp;
290 	daddr_t *bap, *sbap, *ebap;
291 	struct cluster_save *buflist;
292 	daddr_t start_lbn, end_lbn, soff, eoff, newblk, blkno;
293 	struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
294 	int i, len, start_lvl, end_lvl, pref, ssize;
295 
296 	vp = ap->a_vp;
297 	ip = VTOI(vp);
298 	fs = ip->i_fs;
299 	if (fs->fs_contigsumsize <= 0)
300 		return (ENOSPC);
301 	buflist = ap->a_buflist;
302 	len = buflist->bs_nchildren;
303 	start_lbn = buflist->bs_children[0]->b_lblkno;
304 	end_lbn = start_lbn + len - 1;
305 #ifdef DIAGNOSTIC
306 	for (i = 1; i < len; i++)
307 		if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
308 			panic("ffs_reallocblks: non-cluster");
309 #endif
310 	/*
311 	 * If the latest allocation is in a new cylinder group, assume that
312 	 * the filesystem has decided to move and do not force it back to
313 	 * the previous cylinder group.
314 	 */
315 	if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
316 	    dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
317 		return (ENOSPC);
318 	if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
319 	    ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
320 		return (ENOSPC);
321 	/*
322 	 * Get the starting offset and block map for the first block.
323 	 */
324 	if (start_lvl == 0) {
325 		sbap = &ip->i_db[0];
326 		soff = start_lbn;
327 	} else {
328 		idp = &start_ap[start_lvl - 1];
329 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
330 			brelse(sbp);
331 			return (ENOSPC);
332 		}
333 		sbap = (daddr_t *)sbp->b_data;
334 		soff = idp->in_off;
335 	}
336 	/*
337 	 * Find the preferred location for the cluster.
338 	 */
339 	pref = ffs_blkpref(ip, start_lbn, soff, sbap);
340 	/*
341 	 * If the block range spans two block maps, get the second map.
342 	 */
343 	if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off >= len) {
344 		ssize = len;
345 	} else {
346 		ssize = len - (idp->in_off + 1);
347 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
348 			goto fail;
349 		ebap = (daddr_t *)ebp->b_data;
350 	}
351 	/*
352 	 * Search the block map looking for an allocation of the desired size.
353 	 */
354 	if ((newblk = (daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
355 	    len, (u_long (*)())ffs_clusteralloc)) == 0)
356 		goto fail;
357 	/*
358 	 * We have found a new contiguous block.
359 	 *
360 	 * First we have to replace the old block pointers with the new
361 	 * block pointers in the inode and indirect blocks associated
362 	 * with the file.
363 	 */
364 	blkno = newblk;
365 	for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
366 		if (i == ssize)
367 			bap = ebap;
368 #ifdef DIAGNOSTIC
369 		if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap))
370 			panic("ffs_reallocblks: alloc mismatch");
371 #endif
372 		*bap++ = blkno;
373 	}
374 	/*
375 	 * Next we must write out the modified inode and indirect blocks.
376 	 * The writes are synchronous since the old block values may have
377 	 * been written to disk.
378 	 *
379 	 * These writes should be changed to be bdwrites (and the VOP_UPDATE
380 	 * dropped) when a flag has been added to the buffers and inodes
381 	 * to detect when they have been written. It should be set when the
382 	 * cluster is started and cleared whenever the buffer or inode is
383 	 * flushed. We can then check below to see if it is set, and do
384 	 * the synchronous write only when it has been cleared.
385 	 */
386 	if (sbap != &ip->i_db[0]) {
387 		bwrite(sbp);
388 	} else {
389 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
390 		VOP_UPDATE(vp, &time, &time, MNT_WAIT);
391 	}
392 	if (ssize < len)
393 		bwrite(ebp);
394 	/*
395 	 * Last, free the old blocks and assign the new blocks to the buffers.
396 	 */
397 	for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
398 		ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
399 		    fs->fs_bsize);
400 		buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
401 	}
402 	return (0);
403 
404 fail:
405 	if (ssize < len)
406 		brelse(ebp);
407 	if (sbap != &ip->i_db[0])
408 		brelse(sbp);
409 	return (ENOSPC);
410 }
411 
412 /*
413  * Allocate an inode in the file system.
414  *
415  * If allocating a directory, use ffs_dirpref to select the inode.
416  * If allocating in a directory, the following hierarchy is followed:
417  *   1) allocate the preferred inode.
418  *   2) allocate an inode in the same cylinder group.
419  *   3) quadradically rehash into other cylinder groups, until an
420  *      available inode is located.
421  * If no inode preference is given the following heirarchy is used
422  * to allocate an inode:
423  *   1) allocate an inode in cylinder group 0.
424  *   2) quadradically rehash into other cylinder groups, until an
425  *      available inode is located.
426  */
427 ffs_valloc(ap)
428 	struct vop_valloc_args /* {
429 		struct vnode *a_pvp;
430 		int a_mode;
431 		struct ucred *a_cred;
432 		struct vnode **a_vpp;
433 	} */ *ap;
434 {
435 	register struct vnode *pvp = ap->a_pvp;
436 	register struct inode *pip;
437 	register struct fs *fs;
438 	register struct inode *ip;
439 	mode_t mode = ap->a_mode;
440 	ino_t ino, ipref;
441 	int cg, error;
442 
443 	*ap->a_vpp = NULL;
444 	pip = VTOI(pvp);
445 	fs = pip->i_fs;
446 	if (fs->fs_cstotal.cs_nifree == 0)
447 		goto noinodes;
448 
449 	if ((mode & IFMT) == IFDIR)
450 		ipref = ffs_dirpref(fs);
451 	else
452 		ipref = pip->i_number;
453 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
454 		ipref = 0;
455 	cg = ino_to_cg(fs, ipref);
456 	ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode, ffs_nodealloccg);
457 	if (ino == 0)
458 		goto noinodes;
459 	error = VFS_VGET(pvp->v_mount, ino, ap->a_vpp);
460 	if (error) {
461 		VOP_VFREE(pvp, ino, mode);
462 		return (error);
463 	}
464 	ip = VTOI(*ap->a_vpp);
465 	if (ip->i_mode) {
466 		printf("mode = 0%o, inum = %d, fs = %s\n",
467 		    ip->i_mode, ip->i_number, fs->fs_fsmnt);
468 		panic("ffs_valloc: dup alloc");
469 	}
470 	if (ip->i_blocks) {				/* XXX */
471 		printf("free inode %s/%d had %d blocks\n",
472 		    fs->fs_fsmnt, ino, ip->i_blocks);
473 		ip->i_blocks = 0;
474 	}
475 	ip->i_flags = 0;
476 	/*
477 	 * Set up a new generation number for this inode.
478 	 */
479 	if (++nextgennumber < (u_long)time.tv_sec)
480 		nextgennumber = time.tv_sec;
481 	ip->i_gen = nextgennumber;
482 	return (0);
483 noinodes:
484 	ffs_fserr(fs, ap->a_cred->cr_uid, "out of inodes");
485 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
486 	return (ENOSPC);
487 }
488 
489 /*
490  * Find a cylinder to place a directory.
491  *
492  * The policy implemented by this algorithm is to select from
493  * among those cylinder groups with above the average number of
494  * free inodes, the one with the smallest number of directories.
495  */
496 static ino_t
497 ffs_dirpref(fs)
498 	register struct fs *fs;
499 {
500 	int cg, minndir, mincg, avgifree;
501 
502 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
503 	minndir = fs->fs_ipg;
504 	mincg = 0;
505 	for (cg = 0; cg < fs->fs_ncg; cg++)
506 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
507 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
508 			mincg = cg;
509 			minndir = fs->fs_cs(fs, cg).cs_ndir;
510 		}
511 	return ((ino_t)(fs->fs_ipg * mincg));
512 }
513 
514 /*
515  * Select the desired position for the next block in a file.  The file is
516  * logically divided into sections. The first section is composed of the
517  * direct blocks. Each additional section contains fs_maxbpg blocks.
518  *
519  * If no blocks have been allocated in the first section, the policy is to
520  * request a block in the same cylinder group as the inode that describes
521  * the file. If no blocks have been allocated in any other section, the
522  * policy is to place the section in a cylinder group with a greater than
523  * average number of free blocks.  An appropriate cylinder group is found
524  * by using a rotor that sweeps the cylinder groups. When a new group of
525  * blocks is needed, the sweep begins in the cylinder group following the
526  * cylinder group from which the previous allocation was made. The sweep
527  * continues until a cylinder group with greater than the average number
528  * of free blocks is found. If the allocation is for the first block in an
529  * indirect block, the information on the previous allocation is unavailable;
530  * here a best guess is made based upon the logical block number being
531  * allocated.
532  *
533  * If a section is already partially allocated, the policy is to
534  * contiguously allocate fs_maxcontig blocks.  The end of one of these
535  * contiguous blocks and the beginning of the next is physically separated
536  * so that the disk head will be in transit between them for at least
537  * fs_rotdelay milliseconds.  This is to allow time for the processor to
538  * schedule another I/O transfer.
539  */
540 daddr_t
541 ffs_blkpref(ip, lbn, indx, bap)
542 	struct inode *ip;
543 	daddr_t lbn;
544 	int indx;
545 	daddr_t *bap;
546 {
547 	register struct fs *fs;
548 	register int cg;
549 	int avgbfree, startcg;
550 	daddr_t nextblk;
551 
552 	fs = ip->i_fs;
553 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
554 		if (lbn < NDADDR) {
555 			cg = ino_to_cg(fs, ip->i_number);
556 			return (fs->fs_fpg * cg + fs->fs_frag);
557 		}
558 		/*
559 		 * Find a cylinder with greater than average number of
560 		 * unused data blocks.
561 		 */
562 		if (indx == 0 || bap[indx - 1] == 0)
563 			startcg =
564 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
565 		else
566 			startcg = dtog(fs, bap[indx - 1]) + 1;
567 		startcg %= fs->fs_ncg;
568 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
569 		for (cg = startcg; cg < fs->fs_ncg; cg++)
570 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
571 				fs->fs_cgrotor = cg;
572 				return (fs->fs_fpg * cg + fs->fs_frag);
573 			}
574 		for (cg = 0; cg <= startcg; cg++)
575 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
576 				fs->fs_cgrotor = cg;
577 				return (fs->fs_fpg * cg + fs->fs_frag);
578 			}
579 		return (NULL);
580 	}
581 	/*
582 	 * One or more previous blocks have been laid out. If less
583 	 * than fs_maxcontig previous blocks are contiguous, the
584 	 * next block is requested contiguously, otherwise it is
585 	 * requested rotationally delayed by fs_rotdelay milliseconds.
586 	 */
587 	nextblk = bap[indx - 1] + fs->fs_frag;
588 	if (indx < fs->fs_maxcontig || bap[indx - fs->fs_maxcontig] +
589 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
590 		return (nextblk);
591 	if (fs->fs_rotdelay != 0)
592 		/*
593 		 * Here we convert ms of delay to frags as:
594 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
595 		 *	((sect/frag) * (ms/sec))
596 		 * then round up to the next block.
597 		 */
598 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
599 		    (NSPF(fs) * 1000), fs->fs_frag);
600 	return (nextblk);
601 }
602 
603 /*
604  * Implement the cylinder overflow algorithm.
605  *
606  * The policy implemented by this algorithm is:
607  *   1) allocate the block in its requested cylinder group.
608  *   2) quadradically rehash on the cylinder group number.
609  *   3) brute force search for a free block.
610  */
611 /*VARARGS5*/
612 static u_long
613 ffs_hashalloc(ip, cg, pref, size, allocator)
614 	struct inode *ip;
615 	int cg;
616 	long pref;
617 	int size;	/* size for data blocks, mode for inodes */
618 	u_long (*allocator)();
619 {
620 	register struct fs *fs;
621 	long result;
622 	int i, icg = cg;
623 
624 	fs = ip->i_fs;
625 	/*
626 	 * 1: preferred cylinder group
627 	 */
628 	result = (*allocator)(ip, cg, pref, size);
629 	if (result)
630 		return (result);
631 	/*
632 	 * 2: quadratic rehash
633 	 */
634 	for (i = 1; i < fs->fs_ncg; i *= 2) {
635 		cg += i;
636 		if (cg >= fs->fs_ncg)
637 			cg -= fs->fs_ncg;
638 		result = (*allocator)(ip, cg, 0, size);
639 		if (result)
640 			return (result);
641 	}
642 	/*
643 	 * 3: brute force search
644 	 * Note that we start at i == 2, since 0 was checked initially,
645 	 * and 1 is always checked in the quadratic rehash.
646 	 */
647 	cg = (icg + 2) % fs->fs_ncg;
648 	for (i = 2; i < fs->fs_ncg; i++) {
649 		result = (*allocator)(ip, cg, 0, size);
650 		if (result)
651 			return (result);
652 		cg++;
653 		if (cg == fs->fs_ncg)
654 			cg = 0;
655 	}
656 	return (NULL);
657 }
658 
659 /*
660  * Determine whether a fragment can be extended.
661  *
662  * Check to see if the necessary fragments are available, and
663  * if they are, allocate them.
664  */
665 static daddr_t
666 ffs_fragextend(ip, cg, bprev, osize, nsize)
667 	struct inode *ip;
668 	int cg;
669 	long bprev;
670 	int osize, nsize;
671 {
672 	register struct fs *fs;
673 	register struct cg *cgp;
674 	struct buf *bp;
675 	long bno;
676 	int frags, bbase;
677 	int i, error;
678 
679 	fs = ip->i_fs;
680 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
681 		return (NULL);
682 	frags = numfrags(fs, nsize);
683 	bbase = fragnum(fs, bprev);
684 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
685 		/* cannot extend across a block boundary */
686 		return (NULL);
687 	}
688 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
689 		(int)fs->fs_cgsize, NOCRED, &bp);
690 	if (error) {
691 		brelse(bp);
692 		return (NULL);
693 	}
694 	cgp = (struct cg *)bp->b_data;
695 	if (!cg_chkmagic(cgp)) {
696 		brelse(bp);
697 		return (NULL);
698 	}
699 	cgp->cg_time = time.tv_sec;
700 	bno = dtogd(fs, bprev);
701 	for (i = numfrags(fs, osize); i < frags; i++)
702 		if (isclr(cg_blksfree(cgp), bno + i)) {
703 			brelse(bp);
704 			return (NULL);
705 		}
706 	/*
707 	 * the current fragment can be extended
708 	 * deduct the count on fragment being extended into
709 	 * increase the count on the remaining fragment (if any)
710 	 * allocate the extended piece
711 	 */
712 	for (i = frags; i < fs->fs_frag - bbase; i++)
713 		if (isclr(cg_blksfree(cgp), bno + i))
714 			break;
715 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
716 	if (i != frags)
717 		cgp->cg_frsum[i - frags]++;
718 	for (i = numfrags(fs, osize); i < frags; i++) {
719 		clrbit(cg_blksfree(cgp), bno + i);
720 		cgp->cg_cs.cs_nffree--;
721 		fs->fs_cstotal.cs_nffree--;
722 		fs->fs_cs(fs, cg).cs_nffree--;
723 	}
724 	fs->fs_fmod = 1;
725 	bdwrite(bp);
726 	return (bprev);
727 }
728 
729 /*
730  * Determine whether a block can be allocated.
731  *
732  * Check to see if a block of the appropriate size is available,
733  * and if it is, allocate it.
734  */
735 static daddr_t
736 ffs_alloccg(ip, cg, bpref, size)
737 	struct inode *ip;
738 	int cg;
739 	daddr_t bpref;
740 	int size;
741 {
742 	register struct fs *fs;
743 	register struct cg *cgp;
744 	struct buf *bp;
745 	register int i;
746 	int error, bno, frags, allocsiz;
747 
748 	fs = ip->i_fs;
749 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
750 		return (NULL);
751 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
752 		(int)fs->fs_cgsize, NOCRED, &bp);
753 	if (error) {
754 		brelse(bp);
755 		return (NULL);
756 	}
757 	cgp = (struct cg *)bp->b_data;
758 	if (!cg_chkmagic(cgp) ||
759 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
760 		brelse(bp);
761 		return (NULL);
762 	}
763 	cgp->cg_time = time.tv_sec;
764 	if (size == fs->fs_bsize) {
765 		bno = ffs_alloccgblk(fs, cgp, bpref);
766 		bdwrite(bp);
767 		return (bno);
768 	}
769 	/*
770 	 * check to see if any fragments are already available
771 	 * allocsiz is the size which will be allocated, hacking
772 	 * it down to a smaller size if necessary
773 	 */
774 	frags = numfrags(fs, size);
775 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
776 		if (cgp->cg_frsum[allocsiz] != 0)
777 			break;
778 	if (allocsiz == fs->fs_frag) {
779 		/*
780 		 * no fragments were available, so a block will be
781 		 * allocated, and hacked up
782 		 */
783 		if (cgp->cg_cs.cs_nbfree == 0) {
784 			brelse(bp);
785 			return (NULL);
786 		}
787 		bno = ffs_alloccgblk(fs, cgp, bpref);
788 		bpref = dtogd(fs, bno);
789 		for (i = frags; i < fs->fs_frag; i++)
790 			setbit(cg_blksfree(cgp), bpref + i);
791 		i = fs->fs_frag - frags;
792 		cgp->cg_cs.cs_nffree += i;
793 		fs->fs_cstotal.cs_nffree += i;
794 		fs->fs_cs(fs, cg).cs_nffree += i;
795 		fs->fs_fmod = 1;
796 		cgp->cg_frsum[i]++;
797 		bdwrite(bp);
798 		return (bno);
799 	}
800 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
801 	if (bno < 0) {
802 		brelse(bp);
803 		return (NULL);
804 	}
805 	for (i = 0; i < frags; i++)
806 		clrbit(cg_blksfree(cgp), bno + i);
807 	cgp->cg_cs.cs_nffree -= frags;
808 	fs->fs_cstotal.cs_nffree -= frags;
809 	fs->fs_cs(fs, cg).cs_nffree -= frags;
810 	fs->fs_fmod = 1;
811 	cgp->cg_frsum[allocsiz]--;
812 	if (frags != allocsiz)
813 		cgp->cg_frsum[allocsiz - frags]++;
814 	bdwrite(bp);
815 	return (cg * fs->fs_fpg + bno);
816 }
817 
818 /*
819  * Allocate a block in a cylinder group.
820  *
821  * This algorithm implements the following policy:
822  *   1) allocate the requested block.
823  *   2) allocate a rotationally optimal block in the same cylinder.
824  *   3) allocate the next available block on the block rotor for the
825  *      specified cylinder group.
826  * Note that this routine only allocates fs_bsize blocks; these
827  * blocks may be fragmented by the routine that allocates them.
828  */
829 static daddr_t
830 ffs_alloccgblk(fs, cgp, bpref)
831 	register struct fs *fs;
832 	register struct cg *cgp;
833 	daddr_t bpref;
834 {
835 	daddr_t bno, blkno;
836 	int cylno, pos, delta;
837 	short *cylbp;
838 	register int i;
839 
840 	if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
841 		bpref = cgp->cg_rotor;
842 		goto norot;
843 	}
844 	bpref = blknum(fs, bpref);
845 	bpref = dtogd(fs, bpref);
846 	/*
847 	 * if the requested block is available, use it
848 	 */
849 	if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
850 		bno = bpref;
851 		goto gotit;
852 	}
853 	/*
854 	 * check for a block available on the same cylinder
855 	 */
856 	cylno = cbtocylno(fs, bpref);
857 	if (cg_blktot(cgp)[cylno] == 0)
858 		goto norot;
859 	if (fs->fs_cpc == 0) {
860 		/*
861 		 * Block layout information is not available.
862 		 * Leaving bpref unchanged means we take the
863 		 * next available free block following the one
864 		 * we just allocated. Hopefully this will at
865 		 * least hit a track cache on drives of unknown
866 		 * geometry (e.g. SCSI).
867 		 */
868 		goto norot;
869 	}
870 	/*
871 	 * check the summary information to see if a block is
872 	 * available in the requested cylinder starting at the
873 	 * requested rotational position and proceeding around.
874 	 */
875 	cylbp = cg_blks(fs, cgp, cylno);
876 	pos = cbtorpos(fs, bpref);
877 	for (i = pos; i < fs->fs_nrpos; i++)
878 		if (cylbp[i] > 0)
879 			break;
880 	if (i == fs->fs_nrpos)
881 		for (i = 0; i < pos; i++)
882 			if (cylbp[i] > 0)
883 				break;
884 	if (cylbp[i] > 0) {
885 		/*
886 		 * found a rotational position, now find the actual
887 		 * block. A panic if none is actually there.
888 		 */
889 		pos = cylno % fs->fs_cpc;
890 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
891 		if (fs_postbl(fs, pos)[i] == -1) {
892 			printf("pos = %d, i = %d, fs = %s\n",
893 			    pos, i, fs->fs_fsmnt);
894 			panic("ffs_alloccgblk: cyl groups corrupted");
895 		}
896 		for (i = fs_postbl(fs, pos)[i];; ) {
897 			if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
898 				bno = blkstofrags(fs, (bno + i));
899 				goto gotit;
900 			}
901 			delta = fs_rotbl(fs)[i];
902 			if (delta <= 0 ||
903 			    delta + i > fragstoblks(fs, fs->fs_fpg))
904 				break;
905 			i += delta;
906 		}
907 		printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
908 		panic("ffs_alloccgblk: can't find blk in cyl");
909 	}
910 norot:
911 	/*
912 	 * no blocks in the requested cylinder, so take next
913 	 * available one in this cylinder group.
914 	 */
915 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
916 	if (bno < 0)
917 		return (NULL);
918 	cgp->cg_rotor = bno;
919 gotit:
920 	blkno = fragstoblks(fs, bno);
921 	ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
922 	ffs_clusteracct(fs, cgp, blkno, -1);
923 	cgp->cg_cs.cs_nbfree--;
924 	fs->fs_cstotal.cs_nbfree--;
925 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
926 	cylno = cbtocylno(fs, bno);
927 	cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
928 	cg_blktot(cgp)[cylno]--;
929 	fs->fs_fmod = 1;
930 	return (cgp->cg_cgx * fs->fs_fpg + bno);
931 }
932 
933 /*
934  * Determine whether a cluster can be allocated.
935  *
936  * We do not currently check for optimal rotational layout if there
937  * are multiple choices in the same cylinder group. Instead we just
938  * take the first one that we find following bpref.
939  */
940 static daddr_t
941 ffs_clusteralloc(ip, cg, bpref, len)
942 	struct inode *ip;
943 	int cg;
944 	daddr_t bpref;
945 	int len;
946 {
947 	register struct fs *fs;
948 	register struct cg *cgp;
949 	struct buf *bp;
950 	int i, run, bno, bit, map;
951 	u_char *mapp;
952 
953 	fs = ip->i_fs;
954 	if (fs->fs_cs(fs, cg).cs_nbfree < len)
955 		return (NULL);
956 	if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
957 	    NOCRED, &bp))
958 		goto fail;
959 	cgp = (struct cg *)bp->b_data;
960 	if (!cg_chkmagic(cgp))
961 		goto fail;
962 	/*
963 	 * Check to see if a cluster of the needed size (or bigger) is
964 	 * available in this cylinder group.
965 	 */
966 	for (i = len; i <= fs->fs_contigsumsize; i++)
967 		if (cg_clustersum(cgp)[i] > 0)
968 			break;
969 	if (i > fs->fs_contigsumsize)
970 		goto fail;
971 	/*
972 	 * Search the cluster map to find a big enough cluster.
973 	 * We take the first one that we find, even if it is larger
974 	 * than we need as we prefer to get one close to the previous
975 	 * block allocation. We do not search before the current
976 	 * preference point as we do not want to allocate a block
977 	 * that is allocated before the previous one (as we will
978 	 * then have to wait for another pass of the elevator
979 	 * algorithm before it will be read). We prefer to fail and
980 	 * be recalled to try an allocation in the next cylinder group.
981 	 */
982 	if (dtog(fs, bpref) != cg)
983 		bpref = 0;
984 	else
985 		bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
986 	mapp = &cg_clustersfree(cgp)[bpref / NBBY];
987 	map = *mapp++;
988 	bit = 1 << (bpref % NBBY);
989 	for (run = 0, i = bpref; i < cgp->cg_nclusterblks; i++) {
990 		if ((map & bit) == 0) {
991 			run = 0;
992 		} else {
993 			run++;
994 			if (run == len)
995 				break;
996 		}
997 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
998 			bit <<= 1;
999 		} else {
1000 			map = *mapp++;
1001 			bit = 1;
1002 		}
1003 	}
1004 	if (i == cgp->cg_nclusterblks)
1005 		goto fail;
1006 	/*
1007 	 * Allocate the cluster that we have found.
1008 	 */
1009 	bno = cg * fs->fs_fpg + blkstofrags(fs, i - run + 1);
1010 	len = blkstofrags(fs, len);
1011 	for (i = 0; i < len; i += fs->fs_frag)
1012 		if (ffs_alloccgblk(fs, cgp, bno + i) != bno + i)
1013 			panic("ffs_clusteralloc: lost block");
1014 	brelse(bp);
1015 	return (bno);
1016 
1017 fail:
1018 	brelse(bp);
1019 	return (0);
1020 }
1021 
1022 /*
1023  * Determine whether an inode can be allocated.
1024  *
1025  * Check to see if an inode is available, and if it is,
1026  * allocate it using the following policy:
1027  *   1) allocate the requested inode.
1028  *   2) allocate the next available inode after the requested
1029  *      inode in the specified cylinder group.
1030  */
1031 static ino_t
1032 ffs_nodealloccg(ip, cg, ipref, mode)
1033 	struct inode *ip;
1034 	int cg;
1035 	daddr_t ipref;
1036 	int mode;
1037 {
1038 	register struct fs *fs;
1039 	register struct cg *cgp;
1040 	struct buf *bp;
1041 	int error, start, len, loc, map, i;
1042 
1043 	fs = ip->i_fs;
1044 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
1045 		return (NULL);
1046 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1047 		(int)fs->fs_cgsize, NOCRED, &bp);
1048 	if (error) {
1049 		brelse(bp);
1050 		return (NULL);
1051 	}
1052 	cgp = (struct cg *)bp->b_data;
1053 	if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1054 		brelse(bp);
1055 		return (NULL);
1056 	}
1057 	cgp->cg_time = time.tv_sec;
1058 	if (ipref) {
1059 		ipref %= fs->fs_ipg;
1060 		if (isclr(cg_inosused(cgp), ipref))
1061 			goto gotit;
1062 	}
1063 	start = cgp->cg_irotor / NBBY;
1064 	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1065 	loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1066 	if (loc == 0) {
1067 		len = start + 1;
1068 		start = 0;
1069 		loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1070 		if (loc == 0) {
1071 			printf("cg = %d, irotor = %d, fs = %s\n",
1072 			    cg, cgp->cg_irotor, fs->fs_fsmnt);
1073 			panic("ffs_nodealloccg: map corrupted");
1074 			/* NOTREACHED */
1075 		}
1076 	}
1077 	i = start + len - loc;
1078 	map = cg_inosused(cgp)[i];
1079 	ipref = i * NBBY;
1080 	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1081 		if ((map & i) == 0) {
1082 			cgp->cg_irotor = ipref;
1083 			goto gotit;
1084 		}
1085 	}
1086 	printf("fs = %s\n", fs->fs_fsmnt);
1087 	panic("ffs_nodealloccg: block not in map");
1088 	/* NOTREACHED */
1089 gotit:
1090 	setbit(cg_inosused(cgp), ipref);
1091 	cgp->cg_cs.cs_nifree--;
1092 	fs->fs_cstotal.cs_nifree--;
1093 	fs->fs_cs(fs, cg).cs_nifree--;
1094 	fs->fs_fmod = 1;
1095 	if ((mode & IFMT) == IFDIR) {
1096 		cgp->cg_cs.cs_ndir++;
1097 		fs->fs_cstotal.cs_ndir++;
1098 		fs->fs_cs(fs, cg).cs_ndir++;
1099 	}
1100 	bdwrite(bp);
1101 	return (cg * fs->fs_ipg + ipref);
1102 }
1103 
1104 /*
1105  * Free a block or fragment.
1106  *
1107  * The specified block or fragment is placed back in the
1108  * free map. If a fragment is deallocated, a possible
1109  * block reassembly is checked.
1110  */
1111 ffs_blkfree(ip, bno, size)
1112 	register struct inode *ip;
1113 	daddr_t bno;
1114 	long size;
1115 {
1116 	register struct fs *fs;
1117 	register struct cg *cgp;
1118 	struct buf *bp;
1119 	daddr_t blkno;
1120 	int i, error, cg, blk, frags, bbase;
1121 
1122 	fs = ip->i_fs;
1123 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1124 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
1125 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
1126 		panic("blkfree: bad size");
1127 	}
1128 	cg = dtog(fs, bno);
1129 	if ((u_int)bno >= fs->fs_size) {
1130 		printf("bad block %d, ino %d\n", bno, ip->i_number);
1131 		ffs_fserr(fs, ip->i_uid, "bad block");
1132 		return;
1133 	}
1134 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1135 		(int)fs->fs_cgsize, NOCRED, &bp);
1136 	if (error) {
1137 		brelse(bp);
1138 		return;
1139 	}
1140 	cgp = (struct cg *)bp->b_data;
1141 	if (!cg_chkmagic(cgp)) {
1142 		brelse(bp);
1143 		return;
1144 	}
1145 	cgp->cg_time = time.tv_sec;
1146 	bno = dtogd(fs, bno);
1147 	if (size == fs->fs_bsize) {
1148 		blkno = fragstoblks(fs, bno);
1149 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1150 			printf("dev = 0x%x, block = %d, fs = %s\n",
1151 			    ip->i_dev, bno, fs->fs_fsmnt);
1152 			panic("blkfree: freeing free block");
1153 		}
1154 		ffs_setblock(fs, cg_blksfree(cgp), blkno);
1155 		ffs_clusteracct(fs, cgp, blkno, 1);
1156 		cgp->cg_cs.cs_nbfree++;
1157 		fs->fs_cstotal.cs_nbfree++;
1158 		fs->fs_cs(fs, cg).cs_nbfree++;
1159 		i = cbtocylno(fs, bno);
1160 		cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1161 		cg_blktot(cgp)[i]++;
1162 	} else {
1163 		bbase = bno - fragnum(fs, bno);
1164 		/*
1165 		 * decrement the counts associated with the old frags
1166 		 */
1167 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
1168 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1169 		/*
1170 		 * deallocate the fragment
1171 		 */
1172 		frags = numfrags(fs, size);
1173 		for (i = 0; i < frags; i++) {
1174 			if (isset(cg_blksfree(cgp), bno + i)) {
1175 				printf("dev = 0x%x, block = %d, fs = %s\n",
1176 				    ip->i_dev, bno + i, fs->fs_fsmnt);
1177 				panic("blkfree: freeing free frag");
1178 			}
1179 			setbit(cg_blksfree(cgp), bno + i);
1180 		}
1181 		cgp->cg_cs.cs_nffree += i;
1182 		fs->fs_cstotal.cs_nffree += i;
1183 		fs->fs_cs(fs, cg).cs_nffree += i;
1184 		/*
1185 		 * add back in counts associated with the new frags
1186 		 */
1187 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
1188 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1189 		/*
1190 		 * if a complete block has been reassembled, account for it
1191 		 */
1192 		blkno = fragstoblks(fs, bbase);
1193 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1194 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
1195 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1196 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1197 			ffs_clusteracct(fs, cgp, blkno, 1);
1198 			cgp->cg_cs.cs_nbfree++;
1199 			fs->fs_cstotal.cs_nbfree++;
1200 			fs->fs_cs(fs, cg).cs_nbfree++;
1201 			i = cbtocylno(fs, bbase);
1202 			cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1203 			cg_blktot(cgp)[i]++;
1204 		}
1205 	}
1206 	fs->fs_fmod = 1;
1207 	bdwrite(bp);
1208 }
1209 
1210 /*
1211  * Free an inode.
1212  *
1213  * The specified inode is placed back in the free map.
1214  */
1215 int
1216 ffs_vfree(ap)
1217 	struct vop_vfree_args /* {
1218 		struct vnode *a_pvp;
1219 		ino_t a_ino;
1220 		int a_mode;
1221 	} */ *ap;
1222 {
1223 	register struct fs *fs;
1224 	register struct cg *cgp;
1225 	register struct inode *pip;
1226 	ino_t ino = ap->a_ino;
1227 	struct buf *bp;
1228 	int error, cg;
1229 
1230 	pip = VTOI(ap->a_pvp);
1231 	fs = pip->i_fs;
1232 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1233 		panic("ifree: range: dev = 0x%x, ino = %d, fs = %s\n",
1234 		    pip->i_dev, ino, fs->fs_fsmnt);
1235 	cg = ino_to_cg(fs, ino);
1236 	error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1237 		(int)fs->fs_cgsize, NOCRED, &bp);
1238 	if (error) {
1239 		brelse(bp);
1240 		return (0);
1241 	}
1242 	cgp = (struct cg *)bp->b_data;
1243 	if (!cg_chkmagic(cgp)) {
1244 		brelse(bp);
1245 		return (0);
1246 	}
1247 	cgp->cg_time = time.tv_sec;
1248 	ino %= fs->fs_ipg;
1249 	if (isclr(cg_inosused(cgp), ino)) {
1250 		printf("dev = 0x%x, ino = %d, fs = %s\n",
1251 		    pip->i_dev, ino, fs->fs_fsmnt);
1252 		if (fs->fs_ronly == 0)
1253 			panic("ifree: freeing free inode");
1254 	}
1255 	clrbit(cg_inosused(cgp), ino);
1256 	if (ino < cgp->cg_irotor)
1257 		cgp->cg_irotor = ino;
1258 	cgp->cg_cs.cs_nifree++;
1259 	fs->fs_cstotal.cs_nifree++;
1260 	fs->fs_cs(fs, cg).cs_nifree++;
1261 	if ((ap->a_mode & IFMT) == IFDIR) {
1262 		cgp->cg_cs.cs_ndir--;
1263 		fs->fs_cstotal.cs_ndir--;
1264 		fs->fs_cs(fs, cg).cs_ndir--;
1265 	}
1266 	fs->fs_fmod = 1;
1267 	bdwrite(bp);
1268 	return (0);
1269 }
1270 
1271 /*
1272  * Find a block of the specified size in the specified cylinder group.
1273  *
1274  * It is a panic if a request is made to find a block if none are
1275  * available.
1276  */
1277 static daddr_t
1278 ffs_mapsearch(fs, cgp, bpref, allocsiz)
1279 	register struct fs *fs;
1280 	register struct cg *cgp;
1281 	daddr_t bpref;
1282 	int allocsiz;
1283 {
1284 	daddr_t bno;
1285 	int start, len, loc, i;
1286 	int blk, field, subfield, pos;
1287 
1288 	/*
1289 	 * find the fragment by searching through the free block
1290 	 * map for an appropriate bit pattern
1291 	 */
1292 	if (bpref)
1293 		start = dtogd(fs, bpref) / NBBY;
1294 	else
1295 		start = cgp->cg_frotor / NBBY;
1296 	len = howmany(fs->fs_fpg, NBBY) - start;
1297 	loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1298 		(u_char *)fragtbl[fs->fs_frag],
1299 		(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1300 	if (loc == 0) {
1301 		len = start + 1;
1302 		start = 0;
1303 		loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1304 			(u_char *)fragtbl[fs->fs_frag],
1305 			(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1306 		if (loc == 0) {
1307 			printf("start = %d, len = %d, fs = %s\n",
1308 			    start, len, fs->fs_fsmnt);
1309 			panic("ffs_alloccg: map corrupted");
1310 			/* NOTREACHED */
1311 		}
1312 	}
1313 	bno = (start + len - loc) * NBBY;
1314 	cgp->cg_frotor = bno;
1315 	/*
1316 	 * found the byte in the map
1317 	 * sift through the bits to find the selected frag
1318 	 */
1319 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1320 		blk = blkmap(fs, cg_blksfree(cgp), bno);
1321 		blk <<= 1;
1322 		field = around[allocsiz];
1323 		subfield = inside[allocsiz];
1324 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1325 			if ((blk & field) == subfield)
1326 				return (bno + pos);
1327 			field <<= 1;
1328 			subfield <<= 1;
1329 		}
1330 	}
1331 	printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt);
1332 	panic("ffs_alloccg: block not in map");
1333 	return (-1);
1334 }
1335 
1336 /*
1337  * Update the cluster map because of an allocation or free.
1338  *
1339  * Cnt == 1 means free; cnt == -1 means allocating.
1340  */
1341 ffs_clusteracct(fs, cgp, blkno, cnt)
1342 	struct fs *fs;
1343 	struct cg *cgp;
1344 	daddr_t blkno;
1345 	int cnt;
1346 {
1347 	long *sump;
1348 	u_char *freemapp, *mapp;
1349 	int i, start, end, forw, back, map, bit;
1350 
1351 	if (fs->fs_contigsumsize <= 0)
1352 		return;
1353 	freemapp = cg_clustersfree(cgp);
1354 	sump = cg_clustersum(cgp);
1355 	/*
1356 	 * Allocate or clear the actual block.
1357 	 */
1358 	if (cnt > 0)
1359 		setbit(freemapp, blkno);
1360 	else
1361 		clrbit(freemapp, blkno);
1362 	/*
1363 	 * Find the size of the cluster going forward.
1364 	 */
1365 	start = blkno + 1;
1366 	end = start + fs->fs_contigsumsize;
1367 	if (end >= cgp->cg_nclusterblks)
1368 		end = cgp->cg_nclusterblks;
1369 	mapp = &freemapp[start / NBBY];
1370 	map = *mapp++;
1371 	bit = 1 << (start % NBBY);
1372 	for (i = start; i < end; i++) {
1373 		if ((map & bit) == 0)
1374 			break;
1375 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
1376 			bit <<= 1;
1377 		} else {
1378 			map = *mapp++;
1379 			bit = 1;
1380 		}
1381 	}
1382 	forw = i - start;
1383 	/*
1384 	 * Find the size of the cluster going backward.
1385 	 */
1386 	start = blkno - 1;
1387 	end = start - fs->fs_contigsumsize;
1388 	if (end < 0)
1389 		end = -1;
1390 	mapp = &freemapp[start / NBBY];
1391 	map = *mapp--;
1392 	bit = 1 << (start % NBBY);
1393 	for (i = start; i > end; i--) {
1394 		if ((map & bit) == 0)
1395 			break;
1396 		if ((i & (NBBY - 1)) != 0) {
1397 			bit >>= 1;
1398 		} else {
1399 			map = *mapp--;
1400 			bit = 1 << (NBBY - 1);
1401 		}
1402 	}
1403 	back = start - i;
1404 	/*
1405 	 * Account for old cluster and the possibly new forward and
1406 	 * back clusters.
1407 	 */
1408 	i = back + forw + 1;
1409 	if (i > fs->fs_contigsumsize)
1410 		i = fs->fs_contigsumsize;
1411 	sump[i] += cnt;
1412 	if (back > 0)
1413 		sump[back] -= cnt;
1414 	if (forw > 0)
1415 		sump[forw] -= cnt;
1416 }
1417 
1418 /*
1419  * Fserr prints the name of a file system with an error diagnostic.
1420  *
1421  * The form of the error message is:
1422  *	fs: error message
1423  */
1424 static void
1425 ffs_fserr(fs, uid, cp)
1426 	struct fs *fs;
1427 	u_int uid;
1428 	char *cp;
1429 {
1430 
1431 	log(LOG_ERR, "uid %d on %s: %s\n", uid, fs->fs_fsmnt, cp);
1432 }
1433