xref: /freebsd/sys/ufs/ffs/ffs_alloc.c (revision df7f5d4de4592a8948a25ce01e5bddfbb7ce39dc)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)ffs_alloc.c	8.18 (Berkeley) 5/26/95
34  * $Id: ffs_alloc.c,v 1.30 1997/02/22 09:47:00 peter Exp $
35  */
36 
37 #include "opt_quota.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/buf.h>
42 #include <sys/proc.h>
43 #include <sys/vnode.h>
44 #include <sys/mount.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/syslog.h>
48 
49 #include <vm/vm.h>
50 
51 #include <ufs/ufs/quota.h>
52 #include <ufs/ufs/inode.h>
53 #include <ufs/ufs/ufs_extern.h>
54 
55 #include <ufs/ffs/fs.h>
56 #include <ufs/ffs/ffs_extern.h>
57 
58 extern u_long nextgennumber;
59 
60 typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref,
61 				  int size));
62 
63 static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int));
64 static ufs_daddr_t ffs_alloccgblk __P((struct fs *, struct cg *, ufs_daddr_t));
65 static void	ffs_clusteracct	__P((struct fs *, struct cg *, ufs_daddr_t,
66 				     int));
67 static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
68 	    int));
69 static ino_t	ffs_dirpref __P((struct fs *));
70 static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
71 static void	ffs_fserr __P((struct fs *, u_int, char *));
72 static u_long	ffs_hashalloc
73 		    __P((struct inode *, int, long, int, allocfcn_t *));
74 static ino_t	ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int));
75 static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t,
76 	    int));
77 
78 /*
79  * Allocate a block in the file system.
80  *
81  * The size of the requested block is given, which must be some
82  * multiple of fs_fsize and <= fs_bsize.
83  * A preference may be optionally specified. If a preference is given
84  * the following hierarchy is used to allocate a block:
85  *   1) allocate the requested block.
86  *   2) allocate a rotationally optimal block in the same cylinder.
87  *   3) allocate a block in the same cylinder group.
88  *   4) quadradically rehash into other cylinder groups, until an
89  *      available block is located.
90  * If no block preference is given the following heirarchy is used
91  * to allocate a block:
92  *   1) allocate a block in the cylinder group that contains the
93  *      inode for the file.
94  *   2) quadradically rehash into other cylinder groups, until an
95  *      available block is located.
96  */
97 int
98 ffs_alloc(ip, lbn, bpref, size, cred, bnp)
99 	register struct inode *ip;
100 	ufs_daddr_t lbn, bpref;
101 	int size;
102 	struct ucred *cred;
103 	ufs_daddr_t *bnp;
104 {
105 	register struct fs *fs;
106 	ufs_daddr_t bno;
107 	int cg;
108 #ifdef QUOTA
109 	int error;
110 #endif
111 
112 	*bnp = 0;
113 	fs = ip->i_fs;
114 #ifdef DIAGNOSTIC
115 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
116 		printf("dev = 0x%lx, bsize = %ld, size = %d, fs = %s\n",
117 		    (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
118 		panic("ffs_alloc: bad size");
119 	}
120 	if (cred == NOCRED)
121 		panic("ffs_alloc: missing credential");
122 #endif /* DIAGNOSTIC */
123 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
124 		goto nospace;
125 	if (cred->cr_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
126 		goto nospace;
127 #ifdef QUOTA
128 	error = chkdq(ip, (long)btodb(size), cred, 0);
129 	if (error)
130 		return (error);
131 #endif
132 	if (bpref >= fs->fs_size)
133 		bpref = 0;
134 	if (bpref == 0)
135 		cg = ino_to_cg(fs, ip->i_number);
136 	else
137 		cg = dtog(fs, bpref);
138 	bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
139 					 ffs_alloccg);
140 	if (bno > 0) {
141 		ip->i_blocks += btodb(size);
142 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
143 		*bnp = bno;
144 		return (0);
145 	}
146 #ifdef QUOTA
147 	/*
148 	 * Restore user's disk quota because allocation failed.
149 	 */
150 	(void) chkdq(ip, (long)-btodb(size), cred, FORCE);
151 #endif
152 nospace:
153 	ffs_fserr(fs, cred->cr_uid, "file system full");
154 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
155 	return (ENOSPC);
156 }
157 
158 /*
159  * Reallocate a fragment to a bigger size
160  *
161  * The number and size of the old block is given, and a preference
162  * and new size is also specified. The allocator attempts to extend
163  * the original block. Failing that, the regular block allocator is
164  * invoked to get an appropriate block.
165  */
166 int
167 ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
168 	register struct inode *ip;
169 	ufs_daddr_t lbprev;
170 	ufs_daddr_t bpref;
171 	int osize, nsize;
172 	struct ucred *cred;
173 	struct buf **bpp;
174 {
175 	register struct fs *fs;
176 	struct buf *bp;
177 	int cg, request, error;
178 	ufs_daddr_t bprev, bno;
179 
180 	*bpp = 0;
181 	fs = ip->i_fs;
182 #ifdef DIAGNOSTIC
183 	if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
184 	    (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
185 		printf(
186 		    "dev = 0x%lx, bsize = %ld, osize = %d, "
187 		    "nsize = %d, fs = %s\n",
188 		    (u_long)ip->i_dev, fs->fs_bsize, osize,
189 		    nsize, fs->fs_fsmnt);
190 		panic("ffs_realloccg: bad size");
191 	}
192 	if (cred == NOCRED)
193 		panic("ffs_realloccg: missing credential");
194 #endif /* DIAGNOSTIC */
195 	if (cred->cr_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
196 		goto nospace;
197 	if ((bprev = ip->i_db[lbprev]) == 0) {
198 		printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n",
199 		    (u_long) ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
200 		panic("ffs_realloccg: bad bprev");
201 	}
202 	/*
203 	 * Allocate the extra space in the buffer.
204 	 */
205 	error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp);
206 	if (error) {
207 		brelse(bp);
208 		return (error);
209 	}
210 
211 	if( bp->b_blkno == bp->b_lblkno) {
212 		if( lbprev >= NDADDR)
213 			panic("ffs_realloccg: lbprev out of range");
214 		bp->b_blkno = fsbtodb(fs, bprev);
215 	}
216 
217 #ifdef QUOTA
218 	error = chkdq(ip, (long)btodb(nsize - osize), cred, 0);
219 	if (error) {
220 		brelse(bp);
221 		return (error);
222 	}
223 #endif
224 	/*
225 	 * Check for extension in the existing location.
226 	 */
227 	cg = dtog(fs, bprev);
228 	bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
229 	if (bno) {
230 		if (bp->b_blkno != fsbtodb(fs, bno))
231 			panic("ffs_realloccg: bad blockno");
232 		ip->i_blocks += btodb(nsize - osize);
233 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
234 		allocbuf(bp, nsize);
235 		bp->b_flags |= B_DONE;
236 		bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
237 		*bpp = bp;
238 		return (0);
239 	}
240 	/*
241 	 * Allocate a new disk location.
242 	 */
243 	if (bpref >= fs->fs_size)
244 		bpref = 0;
245 	switch ((int)fs->fs_optim) {
246 	case FS_OPTSPACE:
247 		/*
248 		 * Allocate an exact sized fragment. Although this makes
249 		 * best use of space, we will waste time relocating it if
250 		 * the file continues to grow. If the fragmentation is
251 		 * less than half of the minimum free reserve, we choose
252 		 * to begin optimizing for time.
253 		 */
254 		request = nsize;
255 		if (fs->fs_minfree <= 5 ||
256 		    fs->fs_cstotal.cs_nffree >
257 		    fs->fs_dsize * fs->fs_minfree / (2 * 100))
258 			break;
259 		log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
260 			fs->fs_fsmnt);
261 		fs->fs_optim = FS_OPTTIME;
262 		break;
263 	case FS_OPTTIME:
264 		/*
265 		 * At this point we have discovered a file that is trying to
266 		 * grow a small fragment to a larger fragment. To save time,
267 		 * we allocate a full sized block, then free the unused portion.
268 		 * If the file continues to grow, the `ffs_fragextend' call
269 		 * above will be able to grow it in place without further
270 		 * copying. If aberrant programs cause disk fragmentation to
271 		 * grow within 2% of the free reserve, we choose to begin
272 		 * optimizing for space.
273 		 */
274 		request = fs->fs_bsize;
275 		if (fs->fs_cstotal.cs_nffree <
276 		    fs->fs_dsize * (fs->fs_minfree - 2) / 100)
277 			break;
278 		log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
279 			fs->fs_fsmnt);
280 		fs->fs_optim = FS_OPTSPACE;
281 		break;
282 	default:
283 		printf("dev = 0x%lx, optim = %ld, fs = %s\n",
284 		    (u_long)ip->i_dev, fs->fs_optim, fs->fs_fsmnt);
285 		panic("ffs_realloccg: bad optim");
286 		/* NOTREACHED */
287 	}
288 	bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
289 					 ffs_alloccg);
290 	if (bno > 0) {
291 		bp->b_blkno = fsbtodb(fs, bno);
292 		ffs_blkfree(ip, bprev, (long)osize);
293 		if (nsize < request)
294 			ffs_blkfree(ip, bno + numfrags(fs, nsize),
295 			    (long)(request - nsize));
296 		ip->i_blocks += btodb(nsize - osize);
297 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
298 		allocbuf(bp, nsize);
299 		bp->b_flags |= B_DONE;
300 		bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
301 		*bpp = bp;
302 		return (0);
303 	}
304 #ifdef QUOTA
305 	/*
306 	 * Restore user's disk quota because allocation failed.
307 	 */
308 	(void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
309 #endif
310 	brelse(bp);
311 nospace:
312 	/*
313 	 * no space available
314 	 */
315 	ffs_fserr(fs, cred->cr_uid, "file system full");
316 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
317 	return (ENOSPC);
318 }
319 
320 /*
321  * Reallocate a sequence of blocks into a contiguous sequence of blocks.
322  *
323  * The vnode and an array of buffer pointers for a range of sequential
324  * logical blocks to be made contiguous is given. The allocator attempts
325  * to find a range of sequential blocks starting as close as possible to
326  * an fs_rotdelay offset from the end of the allocation for the logical
327  * block immediately preceeding the current range. If successful, the
328  * physical block numbers in the buffer pointers and in the inode are
329  * changed to reflect the new allocation. If unsuccessful, the allocation
330  * is left unchanged. The success in doing the reallocation is returned.
331  * Note that the error return is not reflected back to the user. Rather
332  * the previous block allocation will be used.
333  */
334 static int doasyncfree = 1;
335 SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
336 
337 int doreallocblks = 1;
338 SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
339 
340 static int prtrealloc = 0;
341 
342 int
343 ffs_reallocblks(ap)
344 	struct vop_reallocblks_args /* {
345 		struct vnode *a_vp;
346 		struct cluster_save *a_buflist;
347 	} */ *ap;
348 {
349 #if !defined (not_yes)
350 	return (ENOSPC);
351 #else
352 	struct fs *fs;
353 	struct inode *ip;
354 	struct vnode *vp;
355 	struct buf *sbp, *ebp;
356 	ufs_daddr_t *bap, *sbap, *ebap = 0;
357 	struct cluster_save *buflist;
358 	ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
359 	struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
360 	int i, len, start_lvl, end_lvl, pref, ssize;
361 	struct timeval tv;
362 
363 	if (doreallocblks == 0)
364 		return (ENOSPC);
365 	vp = ap->a_vp;
366 	ip = VTOI(vp);
367 	fs = ip->i_fs;
368 	if (fs->fs_contigsumsize <= 0)
369 		return (ENOSPC);
370 	buflist = ap->a_buflist;
371 	len = buflist->bs_nchildren;
372 	start_lbn = buflist->bs_children[0]->b_lblkno;
373 	end_lbn = start_lbn + len - 1;
374 #ifdef DIAGNOSTIC
375 	for (i = 0; i < len; i++)
376 		if (!ffs_checkblk(ip,
377 		   dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
378 			panic("ffs_reallocblks: unallocated block 1");
379 	for (i = 1; i < len; i++)
380 		if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
381 			panic("ffs_reallocblks: non-logical cluster");
382 	blkno = buflist->bs_children[0]->b_blkno;
383 	ssize = fsbtodb(fs, fs->fs_frag);
384 	for (i = 1; i < len - 1; i++)
385 		if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
386 			panic("ffs_reallocblks: non-physical cluster %d", i);
387 #endif
388 	/*
389 	 * If the latest allocation is in a new cylinder group, assume that
390 	 * the filesystem has decided to move and do not force it back to
391 	 * the previous cylinder group.
392 	 */
393 	if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
394 	    dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
395 		return (ENOSPC);
396 	if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
397 	    ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
398 		return (ENOSPC);
399 	/*
400 	 * Get the starting offset and block map for the first block.
401 	 */
402 	if (start_lvl == 0) {
403 		sbap = &ip->i_db[0];
404 		soff = start_lbn;
405 	} else {
406 		idp = &start_ap[start_lvl - 1];
407 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
408 			brelse(sbp);
409 			return (ENOSPC);
410 		}
411 		sbap = (ufs_daddr_t *)sbp->b_data;
412 		soff = idp->in_off;
413 	}
414 	/*
415 	 * Find the preferred location for the cluster.
416 	 */
417 	pref = ffs_blkpref(ip, start_lbn, soff, sbap);
418 	/*
419 	 * If the block range spans two block maps, get the second map.
420 	 */
421 	if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
422 		ssize = len;
423 	} else {
424 #ifdef DIAGNOSTIC
425 		if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
426 			panic("ffs_reallocblk: start == end");
427 #endif
428 		ssize = len - (idp->in_off + 1);
429 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
430 			goto fail;
431 		ebap = (ufs_daddr_t *)ebp->b_data;
432 	}
433 	/*
434 	 * Search the block map looking for an allocation of the desired size.
435 	 */
436 	if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
437 	    len, ffs_clusteralloc)) == 0)
438 		goto fail;
439 	/*
440 	 * We have found a new contiguous block.
441 	 *
442 	 * First we have to replace the old block pointers with the new
443 	 * block pointers in the inode and indirect blocks associated
444 	 * with the file.
445 	 */
446 #ifdef DEBUG
447 	if (prtrealloc)
448 		printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
449 		    start_lbn, end_lbn);
450 #endif
451 	blkno = newblk;
452 	for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
453 		if (i == ssize)
454 			bap = ebap;
455 #ifdef DIAGNOSTIC
456 		if (!ffs_checkblk(ip,
457 		   dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
458 			panic("ffs_reallocblks: unallocated block 2");
459 		if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
460 			panic("ffs_reallocblks: alloc mismatch");
461 #endif
462 #ifdef DEBUG
463 		if (prtrealloc)
464 			printf(" %d,", *bap);
465 #endif
466 		*bap++ = blkno;
467 	}
468 	/*
469 	 * Next we must write out the modified inode and indirect blocks.
470 	 * For strict correctness, the writes should be synchronous since
471 	 * the old block values may have been written to disk. In practise
472 	 * they are almost never written, but if we are concerned about
473 	 * strict correctness, the `doasyncfree' flag should be set to zero.
474 	 *
475 	 * The test on `doasyncfree' should be changed to test a flag
476 	 * that shows whether the associated buffers and inodes have
477 	 * been written. The flag should be set when the cluster is
478 	 * started and cleared whenever the buffer or inode is flushed.
479 	 * We can then check below to see if it is set, and do the
480 	 * synchronous write only when it has been cleared.
481 	 */
482 	if (sbap != &ip->i_db[0]) {
483 		if (doasyncfree)
484 			bdwrite(sbp);
485 		else
486 			bwrite(sbp);
487 	} else {
488 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
489 		if (!doasyncfree) {
490 			tv = time;
491 			VOP_UPDATE(vp, &tv, &tv, 1);
492 		}
493 	}
494 	if (ssize < len)
495 		if (doasyncfree)
496 			bdwrite(ebp);
497 		else
498 			bwrite(ebp);
499 	/*
500 	 * Last, free the old blocks and assign the new blocks to the buffers.
501 	 */
502 #ifdef DEBUG
503 	if (prtrealloc)
504 		printf("\n\tnew:");
505 #endif
506 	for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
507 		ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
508 		    fs->fs_bsize);
509 		buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
510 #ifdef DEBUG
511 		if (!ffs_checkblk(ip,
512 		   dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
513 			panic("ffs_reallocblks: unallocated block 3");
514 		if (prtrealloc)
515 			printf(" %d,", blkno);
516 #endif
517 	}
518 #ifdef DEBUG
519 	if (prtrealloc) {
520 		prtrealloc--;
521 		printf("\n");
522 	}
523 #endif
524 	return (0);
525 
526 fail:
527 	if (ssize < len)
528 		brelse(ebp);
529 	if (sbap != &ip->i_db[0])
530 		brelse(sbp);
531 	return (ENOSPC);
532 #endif
533 }
534 
535 /*
536  * Allocate an inode in the file system.
537  *
538  * If allocating a directory, use ffs_dirpref to select the inode.
539  * If allocating in a directory, the following hierarchy is followed:
540  *   1) allocate the preferred inode.
541  *   2) allocate an inode in the same cylinder group.
542  *   3) quadradically rehash into other cylinder groups, until an
543  *      available inode is located.
544  * If no inode preference is given the following heirarchy is used
545  * to allocate an inode:
546  *   1) allocate an inode in cylinder group 0.
547  *   2) quadradically rehash into other cylinder groups, until an
548  *      available inode is located.
549  */
550 int
551 ffs_valloc(ap)
552 	struct vop_valloc_args /* {
553 		struct vnode *a_pvp;
554 		int a_mode;
555 		struct ucred *a_cred;
556 		struct vnode **a_vpp;
557 	} */ *ap;
558 {
559 	register struct vnode *pvp = ap->a_pvp;
560 	register struct inode *pip;
561 	register struct fs *fs;
562 	register struct inode *ip;
563 	mode_t mode = ap->a_mode;
564 	ino_t ino, ipref;
565 	int cg, error;
566 
567 	*ap->a_vpp = NULL;
568 	pip = VTOI(pvp);
569 	fs = pip->i_fs;
570 	if (fs->fs_cstotal.cs_nifree == 0)
571 		goto noinodes;
572 
573 	if ((mode & IFMT) == IFDIR)
574 		ipref = ffs_dirpref(fs);
575 	else
576 		ipref = pip->i_number;
577 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
578 		ipref = 0;
579 	cg = ino_to_cg(fs, ipref);
580 	ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
581 					(allocfcn_t *)ffs_nodealloccg);
582 	if (ino == 0)
583 		goto noinodes;
584 	error = VFS_VGET(pvp->v_mount, ino, ap->a_vpp);
585 	if (error) {
586 		VOP_VFREE(pvp, ino, mode);
587 		return (error);
588 	}
589 	ip = VTOI(*ap->a_vpp);
590 	if (ip->i_mode) {
591 		printf("mode = 0%o, inum = %ld, fs = %s\n",
592 		    ip->i_mode, ip->i_number, fs->fs_fsmnt);
593 		panic("ffs_valloc: dup alloc");
594 	}
595 	if (ip->i_blocks) {				/* XXX */
596 		printf("free inode %s/%ld had %ld blocks\n",
597 		    fs->fs_fsmnt, ino, ip->i_blocks);
598 		ip->i_blocks = 0;
599 	}
600 	ip->i_flags = 0;
601 	/*
602 	 * Set up a new generation number for this inode.
603 	 */
604 	if (++nextgennumber < (u_long)time.tv_sec)
605 		nextgennumber = time.tv_sec;
606 	ip->i_gen = nextgennumber;
607 	return (0);
608 noinodes:
609 	ffs_fserr(fs, ap->a_cred->cr_uid, "out of inodes");
610 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
611 	return (ENOSPC);
612 }
613 
614 /*
615  * Find a cylinder to place a directory.
616  *
617  * The policy implemented by this algorithm is to select from
618  * among those cylinder groups with above the average number of
619  * free inodes, the one with the smallest number of directories.
620  */
621 static ino_t
622 ffs_dirpref(fs)
623 	register struct fs *fs;
624 {
625 	int cg, minndir, mincg, avgifree;
626 
627 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
628 	minndir = fs->fs_ipg;
629 	mincg = 0;
630 	for (cg = 0; cg < fs->fs_ncg; cg++)
631 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
632 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
633 			mincg = cg;
634 			minndir = fs->fs_cs(fs, cg).cs_ndir;
635 		}
636 	return ((ino_t)(fs->fs_ipg * mincg));
637 }
638 
639 /*
640  * Select the desired position for the next block in a file.  The file is
641  * logically divided into sections. The first section is composed of the
642  * direct blocks. Each additional section contains fs_maxbpg blocks.
643  *
644  * If no blocks have been allocated in the first section, the policy is to
645  * request a block in the same cylinder group as the inode that describes
646  * the file. If no blocks have been allocated in any other section, the
647  * policy is to place the section in a cylinder group with a greater than
648  * average number of free blocks.  An appropriate cylinder group is found
649  * by using a rotor that sweeps the cylinder groups. When a new group of
650  * blocks is needed, the sweep begins in the cylinder group following the
651  * cylinder group from which the previous allocation was made. The sweep
652  * continues until a cylinder group with greater than the average number
653  * of free blocks is found. If the allocation is for the first block in an
654  * indirect block, the information on the previous allocation is unavailable;
655  * here a best guess is made based upon the logical block number being
656  * allocated.
657  *
658  * If a section is already partially allocated, the policy is to
659  * contiguously allocate fs_maxcontig blocks.  The end of one of these
660  * contiguous blocks and the beginning of the next is physically separated
661  * so that the disk head will be in transit between them for at least
662  * fs_rotdelay milliseconds.  This is to allow time for the processor to
663  * schedule another I/O transfer.
664  */
665 ufs_daddr_t
666 ffs_blkpref(ip, lbn, indx, bap)
667 	struct inode *ip;
668 	ufs_daddr_t lbn;
669 	int indx;
670 	ufs_daddr_t *bap;
671 {
672 	register struct fs *fs;
673 	register int cg;
674 	int avgbfree, startcg;
675 	ufs_daddr_t nextblk;
676 
677 	fs = ip->i_fs;
678 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
679 		if (lbn < NDADDR) {
680 			cg = ino_to_cg(fs, ip->i_number);
681 			return (fs->fs_fpg * cg + fs->fs_frag);
682 		}
683 		/*
684 		 * Find a cylinder with greater than average number of
685 		 * unused data blocks.
686 		 */
687 		if (indx == 0 || bap[indx - 1] == 0)
688 			startcg =
689 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
690 		else
691 			startcg = dtog(fs, bap[indx - 1]) + 1;
692 		startcg %= fs->fs_ncg;
693 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
694 		for (cg = startcg; cg < fs->fs_ncg; cg++)
695 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
696 				fs->fs_cgrotor = cg;
697 				return (fs->fs_fpg * cg + fs->fs_frag);
698 			}
699 		for (cg = 0; cg <= startcg; cg++)
700 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
701 				fs->fs_cgrotor = cg;
702 				return (fs->fs_fpg * cg + fs->fs_frag);
703 			}
704 		return (0);
705 	}
706 	/*
707 	 * One or more previous blocks have been laid out. If less
708 	 * than fs_maxcontig previous blocks are contiguous, the
709 	 * next block is requested contiguously, otherwise it is
710 	 * requested rotationally delayed by fs_rotdelay milliseconds.
711 	 */
712 	nextblk = bap[indx - 1] + fs->fs_frag;
713 	if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
714 	    bap[indx - fs->fs_maxcontig] +
715 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
716 		return (nextblk);
717 	/*
718 	 * Here we convert ms of delay to frags as:
719 	 * (frags) = (ms) * (rev/sec) * (sect/rev) /
720 	 *	((sect/frag) * (ms/sec))
721 	 * then round up to the next block.
722 	 */
723 	nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
724 	    (NSPF(fs) * 1000), fs->fs_frag);
725 	return (nextblk);
726 }
727 
728 /*
729  * Implement the cylinder overflow algorithm.
730  *
731  * The policy implemented by this algorithm is:
732  *   1) allocate the block in its requested cylinder group.
733  *   2) quadradically rehash on the cylinder group number.
734  *   3) brute force search for a free block.
735  */
736 /*VARARGS5*/
737 static u_long
738 ffs_hashalloc(ip, cg, pref, size, allocator)
739 	struct inode *ip;
740 	int cg;
741 	long pref;
742 	int size;	/* size for data blocks, mode for inodes */
743 	allocfcn_t *allocator;
744 {
745 	register struct fs *fs;
746 	long result;	/* XXX why not same type as we return? */
747 	int i, icg = cg;
748 
749 	fs = ip->i_fs;
750 	/*
751 	 * 1: preferred cylinder group
752 	 */
753 	result = (*allocator)(ip, cg, pref, size);
754 	if (result)
755 		return (result);
756 	/*
757 	 * 2: quadratic rehash
758 	 */
759 	for (i = 1; i < fs->fs_ncg; i *= 2) {
760 		cg += i;
761 		if (cg >= fs->fs_ncg)
762 			cg -= fs->fs_ncg;
763 		result = (*allocator)(ip, cg, 0, size);
764 		if (result)
765 			return (result);
766 	}
767 	/*
768 	 * 3: brute force search
769 	 * Note that we start at i == 2, since 0 was checked initially,
770 	 * and 1 is always checked in the quadratic rehash.
771 	 */
772 	cg = (icg + 2) % fs->fs_ncg;
773 	for (i = 2; i < fs->fs_ncg; i++) {
774 		result = (*allocator)(ip, cg, 0, size);
775 		if (result)
776 			return (result);
777 		cg++;
778 		if (cg == fs->fs_ncg)
779 			cg = 0;
780 	}
781 	return (0);
782 }
783 
784 /*
785  * Determine whether a fragment can be extended.
786  *
787  * Check to see if the necessary fragments are available, and
788  * if they are, allocate them.
789  */
790 static ufs_daddr_t
791 ffs_fragextend(ip, cg, bprev, osize, nsize)
792 	struct inode *ip;
793 	int cg;
794 	long bprev;
795 	int osize, nsize;
796 {
797 	register struct fs *fs;
798 	register struct cg *cgp;
799 	struct buf *bp;
800 	long bno;
801 	int frags, bbase;
802 	int i, error;
803 
804 	fs = ip->i_fs;
805 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
806 		return (0);
807 	frags = numfrags(fs, nsize);
808 	bbase = fragnum(fs, bprev);
809 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
810 		/* cannot extend across a block boundary */
811 		return (0);
812 	}
813 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
814 		(int)fs->fs_cgsize, NOCRED, &bp);
815 	if (error) {
816 		brelse(bp);
817 		return (0);
818 	}
819 	cgp = (struct cg *)bp->b_data;
820 	if (!cg_chkmagic(cgp)) {
821 		brelse(bp);
822 		return (0);
823 	}
824 	cgp->cg_time = time.tv_sec;
825 	bno = dtogd(fs, bprev);
826 	for (i = numfrags(fs, osize); i < frags; i++)
827 		if (isclr(cg_blksfree(cgp), bno + i)) {
828 			brelse(bp);
829 			return (0);
830 		}
831 	/*
832 	 * the current fragment can be extended
833 	 * deduct the count on fragment being extended into
834 	 * increase the count on the remaining fragment (if any)
835 	 * allocate the extended piece
836 	 */
837 	for (i = frags; i < fs->fs_frag - bbase; i++)
838 		if (isclr(cg_blksfree(cgp), bno + i))
839 			break;
840 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
841 	if (i != frags)
842 		cgp->cg_frsum[i - frags]++;
843 	for (i = numfrags(fs, osize); i < frags; i++) {
844 		clrbit(cg_blksfree(cgp), bno + i);
845 		cgp->cg_cs.cs_nffree--;
846 		fs->fs_cstotal.cs_nffree--;
847 		fs->fs_cs(fs, cg).cs_nffree--;
848 	}
849 	fs->fs_fmod = 1;
850 	bdwrite(bp);
851 	return (bprev);
852 }
853 
854 /*
855  * Determine whether a block can be allocated.
856  *
857  * Check to see if a block of the appropriate size is available,
858  * and if it is, allocate it.
859  */
860 static ufs_daddr_t
861 ffs_alloccg(ip, cg, bpref, size)
862 	struct inode *ip;
863 	int cg;
864 	ufs_daddr_t bpref;
865 	int size;
866 {
867 	register struct fs *fs;
868 	register struct cg *cgp;
869 	struct buf *bp;
870 	register int i;
871 	int error, bno, frags, allocsiz;
872 
873 	fs = ip->i_fs;
874 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
875 		return (0);
876 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
877 		(int)fs->fs_cgsize, NOCRED, &bp);
878 	if (error) {
879 		brelse(bp);
880 		return (0);
881 	}
882 	cgp = (struct cg *)bp->b_data;
883 	if (!cg_chkmagic(cgp) ||
884 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
885 		brelse(bp);
886 		return (0);
887 	}
888 	cgp->cg_time = time.tv_sec;
889 	if (size == fs->fs_bsize) {
890 		bno = ffs_alloccgblk(fs, cgp, bpref);
891 		bdwrite(bp);
892 		return (bno);
893 	}
894 	/*
895 	 * check to see if any fragments are already available
896 	 * allocsiz is the size which will be allocated, hacking
897 	 * it down to a smaller size if necessary
898 	 */
899 	frags = numfrags(fs, size);
900 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
901 		if (cgp->cg_frsum[allocsiz] != 0)
902 			break;
903 	if (allocsiz == fs->fs_frag) {
904 		/*
905 		 * no fragments were available, so a block will be
906 		 * allocated, and hacked up
907 		 */
908 		if (cgp->cg_cs.cs_nbfree == 0) {
909 			brelse(bp);
910 			return (0);
911 		}
912 		bno = ffs_alloccgblk(fs, cgp, bpref);
913 		bpref = dtogd(fs, bno);
914 		for (i = frags; i < fs->fs_frag; i++)
915 			setbit(cg_blksfree(cgp), bpref + i);
916 		i = fs->fs_frag - frags;
917 		cgp->cg_cs.cs_nffree += i;
918 		fs->fs_cstotal.cs_nffree += i;
919 		fs->fs_cs(fs, cg).cs_nffree += i;
920 		fs->fs_fmod = 1;
921 		cgp->cg_frsum[i]++;
922 		bdwrite(bp);
923 		return (bno);
924 	}
925 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
926 	if (bno < 0) {
927 		brelse(bp);
928 		return (0);
929 	}
930 	for (i = 0; i < frags; i++)
931 		clrbit(cg_blksfree(cgp), bno + i);
932 	cgp->cg_cs.cs_nffree -= frags;
933 	fs->fs_cstotal.cs_nffree -= frags;
934 	fs->fs_cs(fs, cg).cs_nffree -= frags;
935 	fs->fs_fmod = 1;
936 	cgp->cg_frsum[allocsiz]--;
937 	if (frags != allocsiz)
938 		cgp->cg_frsum[allocsiz - frags]++;
939 	bdwrite(bp);
940 	return (cg * fs->fs_fpg + bno);
941 }
942 
943 /*
944  * Allocate a block in a cylinder group.
945  *
946  * This algorithm implements the following policy:
947  *   1) allocate the requested block.
948  *   2) allocate a rotationally optimal block in the same cylinder.
949  *   3) allocate the next available block on the block rotor for the
950  *      specified cylinder group.
951  * Note that this routine only allocates fs_bsize blocks; these
952  * blocks may be fragmented by the routine that allocates them.
953  */
954 static ufs_daddr_t
955 ffs_alloccgblk(fs, cgp, bpref)
956 	register struct fs *fs;
957 	register struct cg *cgp;
958 	ufs_daddr_t bpref;
959 {
960 	ufs_daddr_t bno, blkno;
961 	int cylno, pos, delta;
962 	short *cylbp;
963 	register int i;
964 
965 	if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
966 		bpref = cgp->cg_rotor;
967 		goto norot;
968 	}
969 	bpref = blknum(fs, bpref);
970 	bpref = dtogd(fs, bpref);
971 	/*
972 	 * if the requested block is available, use it
973 	 */
974 	if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
975 		bno = bpref;
976 		goto gotit;
977 	}
978 	if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
979 		/*
980 		 * Block layout information is not available.
981 		 * Leaving bpref unchanged means we take the
982 		 * next available free block following the one
983 		 * we just allocated. Hopefully this will at
984 		 * least hit a track cache on drives of unknown
985 		 * geometry (e.g. SCSI).
986 		 */
987 		goto norot;
988 	}
989 	/*
990 	 * check for a block available on the same cylinder
991 	 */
992 	cylno = cbtocylno(fs, bpref);
993 	if (cg_blktot(cgp)[cylno] == 0)
994 		goto norot;
995 	/*
996 	 * check the summary information to see if a block is
997 	 * available in the requested cylinder starting at the
998 	 * requested rotational position and proceeding around.
999 	 */
1000 	cylbp = cg_blks(fs, cgp, cylno);
1001 	pos = cbtorpos(fs, bpref);
1002 	for (i = pos; i < fs->fs_nrpos; i++)
1003 		if (cylbp[i] > 0)
1004 			break;
1005 	if (i == fs->fs_nrpos)
1006 		for (i = 0; i < pos; i++)
1007 			if (cylbp[i] > 0)
1008 				break;
1009 	if (cylbp[i] > 0) {
1010 		/*
1011 		 * found a rotational position, now find the actual
1012 		 * block. A panic if none is actually there.
1013 		 */
1014 		pos = cylno % fs->fs_cpc;
1015 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1016 		if (fs_postbl(fs, pos)[i] == -1) {
1017 			printf("pos = %d, i = %d, fs = %s\n",
1018 			    pos, i, fs->fs_fsmnt);
1019 			panic("ffs_alloccgblk: cyl groups corrupted");
1020 		}
1021 		for (i = fs_postbl(fs, pos)[i];; ) {
1022 			if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1023 				bno = blkstofrags(fs, (bno + i));
1024 				goto gotit;
1025 			}
1026 			delta = fs_rotbl(fs)[i];
1027 			if (delta <= 0 ||
1028 			    delta + i > fragstoblks(fs, fs->fs_fpg))
1029 				break;
1030 			i += delta;
1031 		}
1032 		printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1033 		panic("ffs_alloccgblk: can't find blk in cyl");
1034 	}
1035 norot:
1036 	/*
1037 	 * no blocks in the requested cylinder, so take next
1038 	 * available one in this cylinder group.
1039 	 */
1040 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1041 	if (bno < 0)
1042 		return (0);
1043 	cgp->cg_rotor = bno;
1044 gotit:
1045 	blkno = fragstoblks(fs, bno);
1046 	ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1047 	ffs_clusteracct(fs, cgp, blkno, -1);
1048 	cgp->cg_cs.cs_nbfree--;
1049 	fs->fs_cstotal.cs_nbfree--;
1050 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1051 	cylno = cbtocylno(fs, bno);
1052 	cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1053 	cg_blktot(cgp)[cylno]--;
1054 	fs->fs_fmod = 1;
1055 	return (cgp->cg_cgx * fs->fs_fpg + bno);
1056 }
1057 
1058 #ifdef notyet
1059 /*
1060  * Determine whether a cluster can be allocated.
1061  *
1062  * We do not currently check for optimal rotational layout if there
1063  * are multiple choices in the same cylinder group. Instead we just
1064  * take the first one that we find following bpref.
1065  */
1066 static ufs_daddr_t
1067 ffs_clusteralloc(ip, cg, bpref, len)
1068 	struct inode *ip;
1069 	int cg;
1070 	ufs_daddr_t bpref;
1071 	int len;
1072 {
1073 	register struct fs *fs;
1074 	register struct cg *cgp;
1075 	struct buf *bp;
1076 	int i, got, run, bno, bit, map;
1077 	u_char *mapp;
1078 	int32_t *lp;
1079 
1080 	fs = ip->i_fs;
1081 	if (fs->fs_maxcluster[cg] < len)
1082 		return (NULL);
1083 	if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1084 	    NOCRED, &bp))
1085 		goto fail;
1086 	cgp = (struct cg *)bp->b_data;
1087 	if (!cg_chkmagic(cgp))
1088 		goto fail;
1089 	/*
1090 	 * Check to see if a cluster of the needed size (or bigger) is
1091 	 * available in this cylinder group.
1092 	 */
1093 	lp = &cg_clustersum(cgp)[len];
1094 	for (i = len; i <= fs->fs_contigsumsize; i++)
1095 		if (*lp++ > 0)
1096 			break;
1097 	if (i > fs->fs_contigsumsize) {
1098 		/*
1099 		 * This is the first time looking for a cluster in this
1100 		 * cylinder group. Update the cluster summary information
1101 		 * to reflect the true maximum sized cluster so that
1102 		 * future cluster allocation requests can avoid reading
1103 		 * the cylinder group map only to find no clusters.
1104 		 */
1105 		lp = &cg_clustersum(cgp)[len - 1];
1106 		for (i = len - 1; i > 0; i--)
1107 			if (*lp-- > 0)
1108 				break;
1109 		fs->fs_maxcluster[cg] = i;
1110 		goto fail;
1111 	}
1112 	/*
1113 	 * Search the cluster map to find a big enough cluster.
1114 	 * We take the first one that we find, even if it is larger
1115 	 * than we need as we prefer to get one close to the previous
1116 	 * block allocation. We do not search before the current
1117 	 * preference point as we do not want to allocate a block
1118 	 * that is allocated before the previous one (as we will
1119 	 * then have to wait for another pass of the elevator
1120 	 * algorithm before it will be read). We prefer to fail and
1121 	 * be recalled to try an allocation in the next cylinder group.
1122 	 */
1123 	if (dtog(fs, bpref) != cg)
1124 		bpref = 0;
1125 	else
1126 		bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1127 	mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1128 	map = *mapp++;
1129 	bit = 1 << (bpref % NBBY);
1130 	for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1131 		if ((map & bit) == 0) {
1132 			run = 0;
1133 		} else {
1134 			run++;
1135 			if (run == len)
1136 				break;
1137 		}
1138 		if ((got & (NBBY - 1)) != (NBBY - 1)) {
1139 			bit <<= 1;
1140 		} else {
1141 			map = *mapp++;
1142 			bit = 1;
1143 		}
1144 	}
1145 	if (got == cgp->cg_nclusterblks)
1146 		goto fail;
1147 	/*
1148 	 * Allocate the cluster that we have found.
1149 	 */
1150 	for (i = 1; i <= len; i++)
1151 		if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1152 			panic("ffs_clusteralloc: map mismatch");
1153 	bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1154 	if (dtog(fs, bno) != cg)
1155 		panic("ffs_clusteralloc: allocated out of group");
1156 	len = blkstofrags(fs, len);
1157 	for (i = 0; i < len; i += fs->fs_frag)
1158 		if ((got = ffs_alloccgblk(fs, cgp, bno + i)) != bno + i)
1159 			panic("ffs_clusteralloc: lost block");
1160 	bdwrite(bp);
1161 	return (bno);
1162 
1163 fail:
1164 	brelse(bp);
1165 	return (0);
1166 }
1167 #endif
1168 
1169 /*
1170  * Determine whether an inode can be allocated.
1171  *
1172  * Check to see if an inode is available, and if it is,
1173  * allocate it using the following policy:
1174  *   1) allocate the requested inode.
1175  *   2) allocate the next available inode after the requested
1176  *      inode in the specified cylinder group.
1177  */
1178 static ino_t
1179 ffs_nodealloccg(ip, cg, ipref, mode)
1180 	struct inode *ip;
1181 	int cg;
1182 	ufs_daddr_t ipref;
1183 	int mode;
1184 {
1185 	register struct fs *fs;
1186 	register struct cg *cgp;
1187 	struct buf *bp;
1188 	int error, start, len, loc, map, i;
1189 
1190 	fs = ip->i_fs;
1191 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
1192 		return (0);
1193 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1194 		(int)fs->fs_cgsize, NOCRED, &bp);
1195 	if (error) {
1196 		brelse(bp);
1197 		return (0);
1198 	}
1199 	cgp = (struct cg *)bp->b_data;
1200 	if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1201 		brelse(bp);
1202 		return (0);
1203 	}
1204 	cgp->cg_time = time.tv_sec;
1205 	if (ipref) {
1206 		ipref %= fs->fs_ipg;
1207 		if (isclr(cg_inosused(cgp), ipref))
1208 			goto gotit;
1209 	}
1210 	start = cgp->cg_irotor / NBBY;
1211 	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1212 	loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1213 	if (loc == 0) {
1214 		len = start + 1;
1215 		start = 0;
1216 		loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1217 		if (loc == 0) {
1218 			printf("cg = %d, irotor = %ld, fs = %s\n",
1219 			    cg, cgp->cg_irotor, fs->fs_fsmnt);
1220 			panic("ffs_nodealloccg: map corrupted");
1221 			/* NOTREACHED */
1222 		}
1223 	}
1224 	i = start + len - loc;
1225 	map = cg_inosused(cgp)[i];
1226 	ipref = i * NBBY;
1227 	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1228 		if ((map & i) == 0) {
1229 			cgp->cg_irotor = ipref;
1230 			goto gotit;
1231 		}
1232 	}
1233 	printf("fs = %s\n", fs->fs_fsmnt);
1234 	panic("ffs_nodealloccg: block not in map");
1235 	/* NOTREACHED */
1236 gotit:
1237 	setbit(cg_inosused(cgp), ipref);
1238 	cgp->cg_cs.cs_nifree--;
1239 	fs->fs_cstotal.cs_nifree--;
1240 	fs->fs_cs(fs, cg).cs_nifree--;
1241 	fs->fs_fmod = 1;
1242 	if ((mode & IFMT) == IFDIR) {
1243 		cgp->cg_cs.cs_ndir++;
1244 		fs->fs_cstotal.cs_ndir++;
1245 		fs->fs_cs(fs, cg).cs_ndir++;
1246 	}
1247 	bdwrite(bp);
1248 	return (cg * fs->fs_ipg + ipref);
1249 }
1250 
1251 /*
1252  * Free a block or fragment.
1253  *
1254  * The specified block or fragment is placed back in the
1255  * free map. If a fragment is deallocated, a possible
1256  * block reassembly is checked.
1257  */
1258 void
1259 ffs_blkfree(ip, bno, size)
1260 	register struct inode *ip;
1261 	ufs_daddr_t bno;
1262 	long size;
1263 {
1264 	register struct fs *fs;
1265 	register struct cg *cgp;
1266 	struct buf *bp;
1267 	ufs_daddr_t blkno;
1268 	int i, error, cg, blk, frags, bbase;
1269 
1270 	fs = ip->i_fs;
1271 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1272 		printf("dev = 0x%lx, bsize = %ld, size = %ld, fs = %s\n",
1273 		    (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
1274 		panic("ffs_blkfree: bad size");
1275 	}
1276 	cg = dtog(fs, bno);
1277 	if ((u_int)bno >= fs->fs_size) {
1278 		printf("bad block %ld, ino %ld\n", bno, ip->i_number);
1279 		ffs_fserr(fs, ip->i_uid, "bad block");
1280 		return;
1281 	}
1282 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1283 		(int)fs->fs_cgsize, NOCRED, &bp);
1284 	if (error) {
1285 		brelse(bp);
1286 		return;
1287 	}
1288 	cgp = (struct cg *)bp->b_data;
1289 	if (!cg_chkmagic(cgp)) {
1290 		brelse(bp);
1291 		return;
1292 	}
1293 	cgp->cg_time = time.tv_sec;
1294 	bno = dtogd(fs, bno);
1295 	if (size == fs->fs_bsize) {
1296 		blkno = fragstoblks(fs, bno);
1297 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1298 			printf("dev = 0x%lx, block = %ld, fs = %s\n",
1299 			    (u_long) ip->i_dev, bno, fs->fs_fsmnt);
1300 			panic("ffs_blkfree: freeing free block");
1301 		}
1302 		ffs_setblock(fs, cg_blksfree(cgp), blkno);
1303 		ffs_clusteracct(fs, cgp, blkno, 1);
1304 		cgp->cg_cs.cs_nbfree++;
1305 		fs->fs_cstotal.cs_nbfree++;
1306 		fs->fs_cs(fs, cg).cs_nbfree++;
1307 		i = cbtocylno(fs, bno);
1308 		cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1309 		cg_blktot(cgp)[i]++;
1310 	} else {
1311 		bbase = bno - fragnum(fs, bno);
1312 		/*
1313 		 * decrement the counts associated with the old frags
1314 		 */
1315 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
1316 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1317 		/*
1318 		 * deallocate the fragment
1319 		 */
1320 		frags = numfrags(fs, size);
1321 		for (i = 0; i < frags; i++) {
1322 			if (isset(cg_blksfree(cgp), bno + i)) {
1323 				printf("dev = 0x%lx, block = %ld, fs = %s\n",
1324 				    (u_long) ip->i_dev, bno + i, fs->fs_fsmnt);
1325 				panic("ffs_blkfree: freeing free frag");
1326 			}
1327 			setbit(cg_blksfree(cgp), bno + i);
1328 		}
1329 		cgp->cg_cs.cs_nffree += i;
1330 		fs->fs_cstotal.cs_nffree += i;
1331 		fs->fs_cs(fs, cg).cs_nffree += i;
1332 		/*
1333 		 * add back in counts associated with the new frags
1334 		 */
1335 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
1336 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1337 		/*
1338 		 * if a complete block has been reassembled, account for it
1339 		 */
1340 		blkno = fragstoblks(fs, bbase);
1341 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1342 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
1343 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1344 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1345 			ffs_clusteracct(fs, cgp, blkno, 1);
1346 			cgp->cg_cs.cs_nbfree++;
1347 			fs->fs_cstotal.cs_nbfree++;
1348 			fs->fs_cs(fs, cg).cs_nbfree++;
1349 			i = cbtocylno(fs, bbase);
1350 			cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1351 			cg_blktot(cgp)[i]++;
1352 		}
1353 	}
1354 	fs->fs_fmod = 1;
1355 	bdwrite(bp);
1356 }
1357 
1358 #ifdef DIAGNOSTIC
1359 /*
1360  * Verify allocation of a block or fragment. Returns true if block or
1361  * fragment is allocated, false if it is free.
1362  */
1363 int
1364 ffs_checkblk(ip, bno, size)
1365 	struct inode *ip;
1366 	ufs_daddr_t bno;
1367 	long size;
1368 {
1369 	struct fs *fs;
1370 	struct cg *cgp;
1371 	struct buf *bp;
1372 	int i, error, frags, free;
1373 
1374 	fs = ip->i_fs;
1375 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1376 		printf("bsize = %d, size = %d, fs = %s\n",
1377 		    fs->fs_bsize, size, fs->fs_fsmnt);
1378 		panic("ffs_checkblk: bad size");
1379 	}
1380 	if ((u_int)bno >= fs->fs_size)
1381 		panic("ffs_checkblk: bad block %d", bno);
1382 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1383 		(int)fs->fs_cgsize, NOCRED, &bp);
1384 	if (error)
1385 		panic("ffs_checkblk: cg bread failed");
1386 	cgp = (struct cg *)bp->b_data;
1387 	if (!cg_chkmagic(cgp))
1388 		panic("ffs_checkblk: cg magic mismatch");
1389 	bno = dtogd(fs, bno);
1390 	if (size == fs->fs_bsize) {
1391 		free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1392 	} else {
1393 		frags = numfrags(fs, size);
1394 		for (free = 0, i = 0; i < frags; i++)
1395 			if (isset(cg_blksfree(cgp), bno + i))
1396 				free++;
1397 		if (free != 0 && free != frags)
1398 			panic("ffs_checkblk: partially free fragment");
1399 	}
1400 	brelse(bp);
1401 	return (!free);
1402 }
1403 #endif /* DIAGNOSTIC */
1404 
1405 /*
1406  * Free an inode.
1407  *
1408  * The specified inode is placed back in the free map.
1409  */
1410 int
1411 ffs_vfree(ap)
1412 	struct vop_vfree_args /* {
1413 		struct vnode *a_pvp;
1414 		ino_t a_ino;
1415 		int a_mode;
1416 	} */ *ap;
1417 {
1418 	register struct fs *fs;
1419 	register struct cg *cgp;
1420 	register struct inode *pip;
1421 	ino_t ino = ap->a_ino;
1422 	struct buf *bp;
1423 	int error, cg;
1424 
1425 	pip = VTOI(ap->a_pvp);
1426 	fs = pip->i_fs;
1427 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1428 		panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1429 		    pip->i_dev, ino, fs->fs_fsmnt);
1430 	cg = ino_to_cg(fs, ino);
1431 	error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1432 		(int)fs->fs_cgsize, NOCRED, &bp);
1433 	if (error) {
1434 		brelse(bp);
1435 		return (0);
1436 	}
1437 	cgp = (struct cg *)bp->b_data;
1438 	if (!cg_chkmagic(cgp)) {
1439 		brelse(bp);
1440 		return (0);
1441 	}
1442 	cgp->cg_time = time.tv_sec;
1443 	ino %= fs->fs_ipg;
1444 	if (isclr(cg_inosused(cgp), ino)) {
1445 		printf("dev = 0x%lx, ino = %ld, fs = %s\n",
1446 		    (u_long)pip->i_dev, ino, fs->fs_fsmnt);
1447 		if (fs->fs_ronly == 0)
1448 			panic("ffs_vfree: freeing free inode");
1449 	}
1450 	clrbit(cg_inosused(cgp), ino);
1451 	if (ino < cgp->cg_irotor)
1452 		cgp->cg_irotor = ino;
1453 	cgp->cg_cs.cs_nifree++;
1454 	fs->fs_cstotal.cs_nifree++;
1455 	fs->fs_cs(fs, cg).cs_nifree++;
1456 	if ((ap->a_mode & IFMT) == IFDIR) {
1457 		cgp->cg_cs.cs_ndir--;
1458 		fs->fs_cstotal.cs_ndir--;
1459 		fs->fs_cs(fs, cg).cs_ndir--;
1460 	}
1461 	fs->fs_fmod = 1;
1462 	bdwrite(bp);
1463 	return (0);
1464 }
1465 
1466 /*
1467  * Find a block of the specified size in the specified cylinder group.
1468  *
1469  * It is a panic if a request is made to find a block if none are
1470  * available.
1471  */
1472 static ufs_daddr_t
1473 ffs_mapsearch(fs, cgp, bpref, allocsiz)
1474 	register struct fs *fs;
1475 	register struct cg *cgp;
1476 	ufs_daddr_t bpref;
1477 	int allocsiz;
1478 {
1479 	ufs_daddr_t bno;
1480 	int start, len, loc, i;
1481 	int blk, field, subfield, pos;
1482 
1483 	/*
1484 	 * find the fragment by searching through the free block
1485 	 * map for an appropriate bit pattern
1486 	 */
1487 	if (bpref)
1488 		start = dtogd(fs, bpref) / NBBY;
1489 	else
1490 		start = cgp->cg_frotor / NBBY;
1491 	len = howmany(fs->fs_fpg, NBBY) - start;
1492 	loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1493 		(u_char *)fragtbl[fs->fs_frag],
1494 		(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1495 	if (loc == 0) {
1496 		len = start + 1;
1497 		start = 0;
1498 		loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1499 			(u_char *)fragtbl[fs->fs_frag],
1500 			(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1501 		if (loc == 0) {
1502 			printf("start = %d, len = %d, fs = %s\n",
1503 			    start, len, fs->fs_fsmnt);
1504 			panic("ffs_alloccg: map corrupted");
1505 			/* NOTREACHED */
1506 		}
1507 	}
1508 	bno = (start + len - loc) * NBBY;
1509 	cgp->cg_frotor = bno;
1510 	/*
1511 	 * found the byte in the map
1512 	 * sift through the bits to find the selected frag
1513 	 */
1514 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1515 		blk = blkmap(fs, cg_blksfree(cgp), bno);
1516 		blk <<= 1;
1517 		field = around[allocsiz];
1518 		subfield = inside[allocsiz];
1519 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1520 			if ((blk & field) == subfield)
1521 				return (bno + pos);
1522 			field <<= 1;
1523 			subfield <<= 1;
1524 		}
1525 	}
1526 	printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1527 	panic("ffs_alloccg: block not in map");
1528 	return (-1);
1529 }
1530 
1531 /*
1532  * Update the cluster map because of an allocation or free.
1533  *
1534  * Cnt == 1 means free; cnt == -1 means allocating.
1535  */
1536 static void
1537 ffs_clusteracct(fs, cgp, blkno, cnt)
1538 	struct fs *fs;
1539 	struct cg *cgp;
1540 	ufs_daddr_t blkno;
1541 	int cnt;
1542 {
1543 	int32_t *sump;
1544 	int32_t *lp;
1545 	u_char *freemapp, *mapp;
1546 	int i, start, end, forw, back, map, bit;
1547 
1548 	if (fs->fs_contigsumsize <= 0)
1549 		return;
1550 	freemapp = cg_clustersfree(cgp);
1551 	sump = cg_clustersum(cgp);
1552 	/*
1553 	 * Allocate or clear the actual block.
1554 	 */
1555 	if (cnt > 0)
1556 		setbit(freemapp, blkno);
1557 	else
1558 		clrbit(freemapp, blkno);
1559 	/*
1560 	 * Find the size of the cluster going forward.
1561 	 */
1562 	start = blkno + 1;
1563 	end = start + fs->fs_contigsumsize;
1564 	if (end >= cgp->cg_nclusterblks)
1565 		end = cgp->cg_nclusterblks;
1566 	mapp = &freemapp[start / NBBY];
1567 	map = *mapp++;
1568 	bit = 1 << (start % NBBY);
1569 	for (i = start; i < end; i++) {
1570 		if ((map & bit) == 0)
1571 			break;
1572 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
1573 			bit <<= 1;
1574 		} else {
1575 			map = *mapp++;
1576 			bit = 1;
1577 		}
1578 	}
1579 	forw = i - start;
1580 	/*
1581 	 * Find the size of the cluster going backward.
1582 	 */
1583 	start = blkno - 1;
1584 	end = start - fs->fs_contigsumsize;
1585 	if (end < 0)
1586 		end = -1;
1587 	mapp = &freemapp[start / NBBY];
1588 	map = *mapp--;
1589 	bit = 1 << (start % NBBY);
1590 	for (i = start; i > end; i--) {
1591 		if ((map & bit) == 0)
1592 			break;
1593 		if ((i & (NBBY - 1)) != 0) {
1594 			bit >>= 1;
1595 		} else {
1596 			map = *mapp--;
1597 			bit = 1 << (NBBY - 1);
1598 		}
1599 	}
1600 	back = start - i;
1601 	/*
1602 	 * Account for old cluster and the possibly new forward and
1603 	 * back clusters.
1604 	 */
1605 	i = back + forw + 1;
1606 	if (i > fs->fs_contigsumsize)
1607 		i = fs->fs_contigsumsize;
1608 	sump[i] += cnt;
1609 	if (back > 0)
1610 		sump[back] -= cnt;
1611 	if (forw > 0)
1612 		sump[forw] -= cnt;
1613 	/*
1614 	 * Update cluster summary information.
1615 	 */
1616 	lp = &sump[fs->fs_contigsumsize];
1617 	for (i = fs->fs_contigsumsize; i > 0; i--)
1618 		if (*lp-- > 0)
1619 			break;
1620 	fs->fs_maxcluster[cgp->cg_cgx] = i;
1621 }
1622 
1623 /*
1624  * Fserr prints the name of a file system with an error diagnostic.
1625  *
1626  * The form of the error message is:
1627  *	fs: error message
1628  */
1629 static void
1630 ffs_fserr(fs, uid, cp)
1631 	struct fs *fs;
1632 	u_int uid;
1633 	char *cp;
1634 {
1635 	struct proc *p = curproc;	/* XXX */
1636 
1637 	log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1638 			p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1639 }
1640