xref: /dragonfly/sys/vfs/ufs/ffs_alloc.c (revision dda92f98)
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. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ffs_alloc.c	8.18 (Berkeley) 5/26/95
30  * $FreeBSD: src/sys/ufs/ffs/ffs_alloc.c,v 1.64.2.2 2001/09/21 19:15:21 dillon Exp $
31  */
32 
33 #include "opt_quota.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/buf.h>
38 #include <sys/conf.h>
39 #include <sys/malloc.h>
40 #include <sys/proc.h>
41 #include <sys/vnode.h>
42 #include <sys/mount.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/syslog.h>
46 
47 #include <sys/taskqueue.h>
48 #include <machine/inttypes.h>
49 
50 #include <sys/buf2.h>
51 
52 #include "quota.h"
53 #include "inode.h"
54 #include "ufs_extern.h"
55 #include "ufsmount.h"
56 
57 #include "fs.h"
58 #include "ffs_extern.h"
59 
60 typedef ufs_daddr_t allocfcn_t (struct inode *ip, int cg, ufs_daddr_t bpref,
61 				  int size);
62 
63 static ufs_daddr_t ffs_alloccg (struct inode *, int, ufs_daddr_t, int);
64 static ufs_daddr_t
65 	      ffs_alloccgblk (struct inode *, struct buf *, ufs_daddr_t);
66 static void ffs_blkfree_cg(struct fs *, struct vnode *, cdev_t , ino_t,
67 			   uint32_t , ufs_daddr_t, long );
68 #ifdef DIAGNOSTIC
69 static int	ffs_checkblk (struct inode *, ufs_daddr_t, long);
70 #endif
71 static void	ffs_clusteracct	(struct fs *, struct cg *, ufs_daddr_t,
72 				     int);
73 static ufs_daddr_t ffs_clusteralloc (struct inode *, int, ufs_daddr_t,
74 	    int);
75 static ino_t	ffs_dirpref (struct inode *);
76 static ufs_daddr_t ffs_fragextend (struct inode *, int, long, int, int);
77 static void	ffs_fserr (struct fs *, uint, char *);
78 static u_long	ffs_hashalloc
79 		    (struct inode *, int, long, int, allocfcn_t *);
80 static ino_t	ffs_nodealloccg (struct inode *, int, ufs_daddr_t, int);
81 static ufs_daddr_t ffs_mapsearch (struct fs *, struct cg *, ufs_daddr_t,
82 	    int);
83 
84 /*
85  * Allocate a block in the filesystem.
86  *
87  * The size of the requested block is given, which must be some
88  * multiple of fs_fsize and <= fs_bsize.
89  * A preference may be optionally specified. If a preference is given
90  * the following hierarchy is used to allocate a block:
91  *   1) allocate the requested block.
92  *   2) allocate a rotationally optimal block in the same cylinder.
93  *   3) allocate a block in the same cylinder group.
94  *   4) quadradically rehash into other cylinder groups, until an
95  *      available block is located.
96  * If no block preference is given the following heirarchy is used
97  * to allocate a block:
98  *   1) allocate a block in the cylinder group that contains the
99  *      inode for the file.
100  *   2) quadradically rehash into other cylinder groups, until an
101  *      available block is located.
102  */
103 int
ffs_alloc(struct inode * ip,ufs_daddr_t lbn,ufs_daddr_t bpref,int size,struct ucred * cred,ufs_daddr_t * bnp)104 ffs_alloc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t bpref, int size,
105 	  struct ucred *cred, ufs_daddr_t *bnp)
106 {
107 	struct fs *fs;
108 	ufs_daddr_t bno;
109 	int cg;
110 #ifdef QUOTA
111 	int error;
112 #endif
113 
114 	*bnp = 0;
115 	fs = ip->i_fs;
116 #ifdef DIAGNOSTIC
117 	if ((uint)size > fs->fs_bsize || fragoff(fs, size) != 0) {
118 		kprintf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
119 		    devtoname(ip->i_dev), (long)fs->fs_bsize, size,
120 		    fs->fs_fsmnt);
121 		panic("ffs_alloc: bad size");
122 	}
123 	if (cred == NOCRED)
124 		panic("ffs_alloc: missing credential");
125 #endif /* DIAGNOSTIC */
126 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
127 		goto nospace;
128 	if (cred->cr_uid != 0 &&
129 	    freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
130 		goto nospace;
131 #ifdef QUOTA
132 	error = ufs_chkdq(ip, (long)btodb(size), cred, 0);
133 	if (error)
134 		return (error);
135 #endif
136 	if (bpref >= fs->fs_size)
137 		bpref = 0;
138 	if (bpref == 0)
139 		cg = ino_to_cg(fs, ip->i_number);
140 	else
141 		cg = dtog(fs, bpref);
142 	bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
143 					 ffs_alloccg);
144 	if (bno > 0) {
145 		ip->i_blocks += btodb(size);
146 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
147 		*bnp = bno;
148 		return (0);
149 	}
150 #ifdef QUOTA
151 	/*
152 	 * Restore user's disk quota because allocation failed.
153 	 */
154 	(void) ufs_chkdq(ip, (long)-btodb(size), cred, FORCE);
155 #endif
156 nospace:
157 	ffs_fserr(fs, cred->cr_uid, "filesystem full");
158 	uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt);
159 	return (ENOSPC);
160 }
161 
162 /*
163  * Reallocate a fragment to a bigger size
164  *
165  * The number and size of the old block is given, and a preference
166  * and new size is also specified. The allocator attempts to extend
167  * the original block. Failing that, the regular block allocator is
168  * invoked to get an appropriate block.
169  */
170 int
ffs_realloccg(struct inode * ip,ufs_daddr_t lbprev,ufs_daddr_t bpref,int osize,int nsize,struct ucred * cred,struct buf ** bpp)171 ffs_realloccg(struct inode *ip, ufs_daddr_t lbprev, ufs_daddr_t bpref,
172 	      int osize, int nsize, struct ucred *cred, struct buf **bpp)
173 {
174 	struct fs *fs;
175 	struct buf *bp;
176 	int cg, request, error;
177 	ufs_daddr_t bprev, bno;
178 
179 	*bpp = NULL;
180 	fs = ip->i_fs;
181 #ifdef DIAGNOSTIC
182 	if ((uint)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
183 	    (uint)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
184 		kprintf(
185 		"dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
186 		    devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
187 		    nsize, fs->fs_fsmnt);
188 		panic("ffs_realloccg: bad size");
189 	}
190 	if (cred == NOCRED)
191 		panic("ffs_realloccg: missing credential");
192 #endif /* DIAGNOSTIC */
193 	if (cred->cr_uid != 0 &&
194 	    freespace(fs, fs->fs_minfree) -  numfrags(fs, nsize - osize) < 0)
195 		goto nospace;
196 	if ((bprev = ip->i_db[lbprev]) == 0) {
197 		kprintf("dev = %s, bsize = %ld, bprev = %ld, fs = %s\n",
198 		    devtoname(ip->i_dev), (long)fs->fs_bsize, (long)bprev,
199 		    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), lblktodoff(fs, lbprev), osize, &bp);
206 	if (error) {
207 		brelse(bp);
208 		return (error);
209 	}
210 
211 	if(bp->b_bio2.bio_offset == NOOFFSET) {
212 		if (lbprev >= UFS_NDADDR)
213 			panic("ffs_realloccg: lbprev out of range");
214 		bp->b_bio2.bio_offset = fsbtodoff(fs, bprev);
215 	}
216 
217 #ifdef QUOTA
218 	error = ufs_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_bio2.bio_offset != fsbtodoff(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 		bzero((char *)bp->b_data + osize, (uint)nsize - osize);
236 		*bpp = bp;
237 		return (0);
238 	}
239 	/*
240 	 * Allocate a new disk location.
241 	 */
242 	if (bpref >= fs->fs_size)
243 		bpref = 0;
244 	switch ((int)fs->fs_optim) {
245 	case FS_OPTSPACE:
246 		/*
247 		 * Allocate an exact sized fragment. Although this makes
248 		 * best use of space, we will waste time relocating it if
249 		 * the file continues to grow. If the fragmentation is
250 		 * less than half of the minimum free reserve, we choose
251 		 * to begin optimizing for time.
252 		 */
253 		request = nsize;
254 		if (fs->fs_minfree <= 5 ||
255 		    fs->fs_cstotal.cs_nffree >
256 		    (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
257 			break;
258 		log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
259 			fs->fs_fsmnt);
260 		fs->fs_optim = FS_OPTTIME;
261 		break;
262 	case FS_OPTTIME:
263 		/*
264 		 * At this point we have discovered a file that is trying to
265 		 * grow a small fragment to a larger fragment. To save time,
266 		 * we allocate a full sized block, then free the unused portion.
267 		 * If the file continues to grow, the `ffs_fragextend' call
268 		 * above will be able to grow it in place without further
269 		 * copying. If aberrant programs cause disk fragmentation to
270 		 * grow within 2% of the free reserve, we choose to begin
271 		 * optimizing for space.
272 		 */
273 		request = fs->fs_bsize;
274 		if (fs->fs_cstotal.cs_nffree <
275 		    (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
276 			break;
277 		log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
278 			fs->fs_fsmnt);
279 		fs->fs_optim = FS_OPTSPACE;
280 		break;
281 	default:
282 		kprintf("dev = %s, optim = %ld, fs = %s\n",
283 		    devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
284 		panic("ffs_realloccg: bad optim");
285 		/* NOTREACHED */
286 	}
287 	bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
288 					 ffs_alloccg);
289 	if (bno > 0) {
290 		bp->b_bio2.bio_offset = fsbtodoff(fs, bno);
291 		if (!DOINGSOFTDEP(ITOV(ip)))
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 		bzero((char *)bp->b_data + osize, (uint)nsize - osize);
300 		*bpp = bp;
301 		return (0);
302 	}
303 #ifdef QUOTA
304 	/*
305 	 * Restore user's disk quota because allocation failed.
306 	 */
307 	(void) ufs_chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
308 #endif
309 	brelse(bp);
310 nospace:
311 	/*
312 	 * no space available
313 	 */
314 	ffs_fserr(fs, cred->cr_uid, "filesystem full");
315 	uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt);
316 	return (ENOSPC);
317 }
318 
319 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
320 
321 /*
322  * Reallocate a sequence of blocks into a contiguous sequence of blocks.
323  *
324  * The vnode and an array of buffer pointers for a range of sequential
325  * logical blocks to be made contiguous is given. The allocator attempts
326  * to find a range of sequential blocks starting as close as possible to
327  * an fs_rotdelay offset from the end of the allocation for the logical
328  * block immediately preceeding the current range. If successful, the
329  * physical block numbers in the buffer pointers and in the inode are
330  * changed to reflect the new allocation. If unsuccessful, the allocation
331  * is left unchanged. The success in doing the reallocation is returned.
332  * Note that the error return is not reflected back to the user. Rather
333  * the previous block allocation will be used.
334  */
335 static int doasyncfree = 1;
336 SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
337 
338 static int doreallocblks = 1;
339 SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
340 
341 #ifdef DEBUG
342 static volatile int prtrealloc = 0;
343 #endif
344 
345 /*
346  * ffs_reallocblks(struct vnode *a_vp, struct cluster_save *a_buflist)
347  */
348 int
ffs_reallocblks(struct vop_reallocblks_args * ap)349 ffs_reallocblks(struct vop_reallocblks_args *ap)
350 {
351 	struct fs *fs;
352 	struct inode *ip;
353 	struct vnode *vp;
354 	struct buf *sbp, *ebp;
355 	ufs_daddr_t *bap, *sbap, *ebap = NULL;
356 	struct cluster_save *buflist;
357 	ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
358 #ifdef DIAGNOSTIC
359 	off_t boffset;
360 #endif
361 	struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp;
362 	int i, len, slen, start_lvl, end_lvl, pref, ssize;
363 
364 	if (doreallocblks == 0)
365 		return (ENOSPC);
366 	vp = ap->a_vp;
367 	ip = VTOI(vp);
368 	fs = ip->i_fs;
369 	if (fs->fs_contigsumsize <= 0)
370 		return (ENOSPC);
371 	buflist = ap->a_buflist;
372 	len = buflist->bs_nchildren;
373 	start_lbn = lblkno(fs, buflist->bs_children[0]->b_loffset);
374 	end_lbn = start_lbn + len - 1;
375 #ifdef DIAGNOSTIC
376 	for (i = 0; i < len; i++)
377 		if (!ffs_checkblk(ip,
378 		   dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset), fs->fs_bsize))
379 			panic("ffs_reallocblks: unallocated block 1");
380 	for (i = 1; i < len; i++) {
381 		if (buflist->bs_children[i]->b_loffset != lblktodoff(fs, start_lbn) + lblktodoff(fs, i))
382 			panic("ffs_reallocblks: non-logical cluster");
383 	}
384 	boffset = buflist->bs_children[0]->b_bio2.bio_offset;
385 	ssize = (int)fsbtodoff(fs, fs->fs_frag);
386 	for (i = 1; i < len - 1; i++)
387 		if (buflist->bs_children[i]->b_bio2.bio_offset != boffset + (i * ssize))
388 			panic("ffs_reallocblks: non-physical cluster %d", i);
389 #endif
390 	/*
391 	 * If the latest allocation is in a new cylinder group, assume that
392 	 * the filesystem has decided to move and do not force it back to
393 	 * the previous cylinder group.
394 	 */
395 	if (dtog(fs, dofftofsb(fs, buflist->bs_children[0]->b_bio2.bio_offset)) !=
396 	    dtog(fs, dofftofsb(fs, buflist->bs_children[len - 1]->b_bio2.bio_offset)))
397 		return (ENOSPC);
398 	if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
399 	    ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
400 		return (ENOSPC);
401 	/*
402 	 * Get the starting offset and block map for the first block and
403 	 * the number of blocks that will fit into sbap starting at soff.
404 	 */
405 	if (start_lvl == 0) {
406 		sbap = &ip->i_db[0];
407 		soff = start_lbn;
408 		slen = UFS_NDADDR - soff;
409 	} else {
410 		idp = &start_ap[start_lvl - 1];
411 		if (bread(vp, lblktodoff(fs, idp->in_lbn), (int)fs->fs_bsize, &sbp)) {
412 			brelse(sbp);
413 			return (ENOSPC);
414 		}
415 		sbap = (ufs_daddr_t *)sbp->b_data;
416 		soff = idp->in_off;
417 		slen = fs->fs_nindir - soff;
418 	}
419 	/*
420 	 * Find the preferred location for the cluster.
421 	 */
422 	pref = ffs_blkpref(ip, start_lbn, soff, sbap);
423 
424 	/*
425 	 * If the block range spans two block maps, get the second map.
426 	 */
427 	if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
428 		ssize = len;
429 	} else {
430 #ifdef DIAGNOSTIC
431 		if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
432 			panic("ffs_reallocblk: start == end");
433 #endif
434 		ssize = len - (idp->in_off + 1);
435 		if (bread(vp, lblktodoff(fs, idp->in_lbn), (int)fs->fs_bsize, &ebp))
436 			goto fail;
437 		ebap = (ufs_daddr_t *)ebp->b_data;
438 	}
439 
440 	/*
441 	 * Make sure we aren't spanning more then two blockmaps.  ssize is
442 	 * our calculation of the span we have to scan in the first blockmap,
443 	 * while slen is our calculation of the number of entries available
444 	 * in the first blockmap (from soff).
445 	 */
446 	if (ssize > slen) {
447 		panic("ffs_reallocblks: range spans more than two blockmaps!"
448 			" start_lbn %ld len %d (%d/%d)",
449 			(long)start_lbn, len, slen, ssize);
450 	}
451 	/*
452 	 * Search the block map looking for an allocation of the desired size.
453 	 */
454 	if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
455 	    len, ffs_clusteralloc)) == 0)
456 		goto fail;
457 	/*
458 	 * We have found a new contiguous block.
459 	 *
460 	 * First we have to replace the old block pointers with the new
461 	 * block pointers in the inode and indirect blocks associated
462 	 * with the file.
463 	 */
464 #ifdef DEBUG
465 	if (prtrealloc)
466 		kprintf("realloc: ino %ju, lbns %d-%d\n\told:",
467 		    (uintmax_t)ip->i_number, start_lbn, end_lbn);
468 #endif
469 	blkno = newblk;
470 	for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
471 		if (i == ssize) {
472 			bap = ebap;
473 			soff = -i;
474 		}
475 #ifdef DIAGNOSTIC
476 		if (!ffs_checkblk(ip,
477 		   dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset), fs->fs_bsize))
478 			panic("ffs_reallocblks: unallocated block 2");
479 		if (dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset) != *bap)
480 			panic("ffs_reallocblks: alloc mismatch");
481 #endif
482 #ifdef DEBUG
483 		if (prtrealloc)
484 			kprintf(" %d,", *bap);
485 #endif
486 		if (DOINGSOFTDEP(vp)) {
487 			if (sbap == &ip->i_db[0] && i < ssize)
488 				softdep_setup_allocdirect(ip, start_lbn + i,
489 				    blkno, *bap, fs->fs_bsize, fs->fs_bsize,
490 				    buflist->bs_children[i]);
491 			else
492 				softdep_setup_allocindir_page(ip, start_lbn + i,
493 				    i < ssize ? sbp : ebp, soff + i, blkno,
494 				    *bap, buflist->bs_children[i]);
495 		}
496 		*bap++ = blkno;
497 	}
498 	/*
499 	 * Next we must write out the modified inode and indirect blocks.
500 	 * For strict correctness, the writes should be synchronous since
501 	 * the old block values may have been written to disk. In practise
502 	 * they are almost never written, but if we are concerned about
503 	 * strict correctness, the `doasyncfree' flag should be set to zero.
504 	 *
505 	 * The test on `doasyncfree' should be changed to test a flag
506 	 * that shows whether the associated buffers and inodes have
507 	 * been written. The flag should be set when the cluster is
508 	 * started and cleared whenever the buffer or inode is flushed.
509 	 * We can then check below to see if it is set, and do the
510 	 * synchronous write only when it has been cleared.
511 	 */
512 	if (sbap != &ip->i_db[0]) {
513 		if (doasyncfree)
514 			bdwrite(sbp);
515 		else
516 			bwrite(sbp);
517 	} else {
518 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
519 		if (!doasyncfree)
520 			ffs_update(vp, 1);
521 	}
522 	if (ssize < len) {
523 		if (doasyncfree)
524 			bdwrite(ebp);
525 		else
526 			bwrite(ebp);
527 	}
528 	/*
529 	 * Last, free the old blocks and assign the new blocks to the buffers.
530 	 */
531 #ifdef DEBUG
532 	if (prtrealloc)
533 		kprintf("\n\tnew:");
534 #endif
535 	for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
536 		if (!DOINGSOFTDEP(vp) &&
537 		    buflist->bs_children[i]->b_bio2.bio_offset != NOOFFSET) {
538 			ffs_blkfree(ip,
539 			    dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset),
540 			    fs->fs_bsize);
541 		}
542 		buflist->bs_children[i]->b_bio2.bio_offset = fsbtodoff(fs, blkno);
543 #ifdef DIAGNOSTIC
544 		if (!ffs_checkblk(ip,
545 		   dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset), fs->fs_bsize))
546 			panic("ffs_reallocblks: unallocated block 3");
547 #endif
548 #ifdef DEBUG
549 		if (prtrealloc)
550 			kprintf(" %d,", blkno);
551 #endif
552 	}
553 #ifdef DEBUG
554 	if (prtrealloc) {
555 		prtrealloc--;
556 		kprintf("\n");
557 	}
558 #endif
559 	return (0);
560 
561 fail:
562 	if (ssize < len)
563 		brelse(ebp);
564 	if (sbap != &ip->i_db[0])
565 		brelse(sbp);
566 	return (ENOSPC);
567 }
568 
569 /*
570  * Allocate an inode in the filesystem.
571  *
572  * If allocating a directory, use ffs_dirpref to select the inode.
573  * If allocating in a directory, the following hierarchy is followed:
574  *   1) allocate the preferred inode.
575  *   2) allocate an inode in the same cylinder group.
576  *   3) quadradically rehash into other cylinder groups, until an
577  *      available inode is located.
578  * If no inode preference is given the following heirarchy is used
579  * to allocate an inode:
580  *   1) allocate an inode in cylinder group 0.
581  *   2) quadradically rehash into other cylinder groups, until an
582  *      available inode is located.
583  */
584 int
ffs_valloc(struct vnode * pvp,int mode,struct ucred * cred,struct vnode ** vpp)585 ffs_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp)
586 {
587 	struct inode *pip;
588 	struct fs *fs;
589 	struct inode *ip;
590 	ino_t ino, ipref;
591 	int cg, error;
592 
593 	*vpp = NULL;
594 	pip = VTOI(pvp);
595 	fs = pip->i_fs;
596 	if (fs->fs_cstotal.cs_nifree == 0)
597 		goto noinodes;
598 
599 	if ((mode & IFMT) == IFDIR)
600 		ipref = ffs_dirpref(pip);
601 	else
602 		ipref = pip->i_number;
603 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
604 		ipref = 0;
605 	cg = ino_to_cg(fs, ipref);
606 	/*
607 	 * Track number of dirs created one after another
608 	 * in a same cg without intervening by files.
609 	 */
610 	if ((mode & IFMT) == IFDIR) {
611 		if (fs->fs_contigdirs[cg] < 255)
612 			fs->fs_contigdirs[cg]++;
613 	} else {
614 		if (fs->fs_contigdirs[cg] > 0)
615 			fs->fs_contigdirs[cg]--;
616 	}
617 	ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
618 					(allocfcn_t *)ffs_nodealloccg);
619 	if (ino == 0)
620 		goto noinodes;
621 	error = VFS_VGET(pvp->v_mount, NULL, ino, vpp);
622 	if (error) {
623 		ffs_vfree(pvp, ino, mode);
624 		return (error);
625 	}
626 	ip = VTOI(*vpp);
627 	if (ip->i_mode) {
628 		kprintf("mode = 0%o, inum = %lu, fs = %s\n",
629 		    ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
630 		panic("ffs_valloc: dup alloc");
631 	}
632 	if (ip->i_blocks) {				/* XXX */
633 		kprintf("free inode %s/%lu had %ld blocks\n",
634 		    fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks);
635 		ip->i_blocks = 0;
636 	}
637 	ip->i_flags = 0;
638 	/*
639 	 * Set up a new generation number for this inode.
640 	 */
641 	if (ip->i_gen == 0 || ++ip->i_gen == 0)
642 		ip->i_gen = krandom() / 2 + 1;
643 	return (0);
644 noinodes:
645 	ffs_fserr(fs, cred->cr_uid, "out of inodes");
646 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
647 	return (ENOSPC);
648 }
649 
650 /*
651  * Find a cylinder group to place a directory.
652  *
653  * The policy implemented by this algorithm is to allocate a
654  * directory inode in the same cylinder group as its parent
655  * directory, but also to reserve space for its files inodes
656  * and data. Restrict the number of directories which may be
657  * allocated one after another in the same cylinder group
658  * without intervening allocation of files.
659  *
660  * If we allocate a first level directory then force allocation
661  * in another cylinder group.
662  */
663 static ino_t
ffs_dirpref(struct inode * pip)664 ffs_dirpref(struct inode *pip)
665 {
666 	struct fs *fs;
667 	int cg, prefcg, dirsize, cgsize;
668 	int64_t dirsize64;
669 	int avgifree, avgbfree, avgndir, curdirsize;
670 	int minifree, minbfree, maxndir;
671 	int mincg, minndir;
672 	int maxcontigdirs;
673 
674 	fs = pip->i_fs;
675 
676 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
677 	avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
678 	avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
679 
680 	/*
681 	 * Force allocation in another cg if creating a first level dir.
682 	 */
683 	if (ITOV(pip)->v_flag & VROOT) {
684 		prefcg = karc4random() % fs->fs_ncg;
685 		mincg = prefcg;
686 		minndir = fs->fs_ipg;
687 		for (cg = prefcg; cg < fs->fs_ncg; cg++)
688 			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
689 			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
690 			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
691 				mincg = cg;
692 				minndir = fs->fs_cs(fs, cg).cs_ndir;
693 			}
694 		for (cg = 0; cg < prefcg; cg++)
695 			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
696 			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
697 			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
698 				mincg = cg;
699 				minndir = fs->fs_cs(fs, cg).cs_ndir;
700 			}
701 		return ((ino_t)(fs->fs_ipg * mincg));
702 	}
703 
704 	/*
705 	 * Count various limits which used for
706 	 * optimal allocation of a directory inode.
707 	 */
708 	maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
709 	minifree = avgifree - avgifree / 4;
710 	if (minifree < 1)
711 		minifree = 1;
712 	minbfree = avgbfree - avgbfree / 4;
713 	if (minbfree < 1)
714 		minbfree = 1;
715 	cgsize = fs->fs_fsize * fs->fs_fpg;
716 
717 	/*
718 	 * fs_avgfilesize and fs_avgfpdir are user-settable entities and
719 	 * multiplying them may overflow a 32 bit integer.
720 	 */
721 	dirsize64 = fs->fs_avgfilesize * (int64_t)fs->fs_avgfpdir;
722 	if (dirsize64 > 0x7fffffff) {
723 		maxcontigdirs = 1;
724 	} else {
725 		dirsize = (int)dirsize64;
726 		curdirsize = avgndir ?
727 			(cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
728 		if (dirsize < curdirsize)
729 			dirsize = curdirsize;
730 		maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
731 		if (fs->fs_avgfpdir > 0)
732 			maxcontigdirs = min(maxcontigdirs,
733 				    fs->fs_ipg / fs->fs_avgfpdir);
734 		if (maxcontigdirs == 0)
735 			maxcontigdirs = 1;
736 	}
737 
738 	/*
739 	 * Limit number of dirs in one cg and reserve space for
740 	 * regular files, but only if we have no deficit in
741 	 * inodes or space.
742 	 */
743 	prefcg = ino_to_cg(fs, pip->i_number);
744 	for (cg = prefcg; cg < fs->fs_ncg; cg++)
745 		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
746 		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
747 	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
748 			if (fs->fs_contigdirs[cg] < maxcontigdirs)
749 				return ((ino_t)(fs->fs_ipg * cg));
750 		}
751 	for (cg = 0; cg < prefcg; cg++)
752 		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
753 		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
754 	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
755 			if (fs->fs_contigdirs[cg] < maxcontigdirs)
756 				return ((ino_t)(fs->fs_ipg * cg));
757 		}
758 	/*
759 	 * This is a backstop when we have deficit in space.
760 	 */
761 	for (cg = prefcg; cg < fs->fs_ncg; cg++)
762 		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
763 			return ((ino_t)(fs->fs_ipg * cg));
764 	for (cg = 0; cg < prefcg; cg++)
765 		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
766 			break;
767 	return ((ino_t)(fs->fs_ipg * cg));
768 }
769 
770 /*
771  * Select the desired position for the next block in a file.  The file is
772  * logically divided into sections. The first section is composed of the
773  * direct blocks. Each additional section contains fs_maxbpg blocks.
774  *
775  * If no blocks have been allocated in the first section, the policy is to
776  * request a block in the same cylinder group as the inode that describes
777  * the file. If no blocks have been allocated in any other section, the
778  * policy is to place the section in a cylinder group with a greater than
779  * average number of free blocks.  An appropriate cylinder group is found
780  * by using a rotor that sweeps the cylinder groups. When a new group of
781  * blocks is needed, the sweep begins in the cylinder group following the
782  * cylinder group from which the previous allocation was made. The sweep
783  * continues until a cylinder group with greater than the average number
784  * of free blocks is found. If the allocation is for the first block in an
785  * indirect block, the information on the previous allocation is unavailable;
786  * here a best guess is made based upon the logical block number being
787  * allocated.
788  *
789  * If a section is already partially allocated, the policy is to
790  * contiguously allocate fs_maxcontig blocks.  The end of one of these
791  * contiguous blocks and the beginning of the next is physically separated
792  * so that the disk head will be in transit between them for at least
793  * fs_rotdelay milliseconds.  This is to allow time for the processor to
794  * schedule another I/O transfer.
795  */
796 ufs_daddr_t
ffs_blkpref(struct inode * ip,ufs_daddr_t lbn,int indx,ufs_daddr_t * bap)797 ffs_blkpref(struct inode *ip, ufs_daddr_t lbn, int indx, ufs_daddr_t *bap)
798 {
799 	struct fs *fs;
800 	int cg;
801 	int avgbfree, startcg;
802 	ufs_daddr_t nextblk;
803 
804 	fs = ip->i_fs;
805 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
806 		if (lbn < UFS_NDADDR + NINDIR(fs)) {
807 			cg = ino_to_cg(fs, ip->i_number);
808 			return (fs->fs_fpg * cg + fs->fs_frag);
809 		}
810 		/*
811 		 * Find a cylinder with greater than average number of
812 		 * unused data blocks.
813 		 */
814 		if (indx == 0 || bap[indx - 1] == 0)
815 			startcg =
816 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
817 		else
818 			startcg = dtog(fs, bap[indx - 1]) + 1;
819 		startcg %= fs->fs_ncg;
820 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
821 		for (cg = startcg; cg < fs->fs_ncg; cg++)
822 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
823 				fs->fs_cgrotor = cg;
824 				return (fs->fs_fpg * cg + fs->fs_frag);
825 			}
826 		for (cg = 0; cg <= startcg; cg++)
827 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
828 				fs->fs_cgrotor = cg;
829 				return (fs->fs_fpg * cg + fs->fs_frag);
830 			}
831 		return (0);
832 	}
833 	/*
834 	 * One or more previous blocks have been laid out. If less
835 	 * than fs_maxcontig previous blocks are contiguous, the
836 	 * next block is requested contiguously, otherwise it is
837 	 * requested rotationally delayed by fs_rotdelay milliseconds.
838 	 */
839 	nextblk = bap[indx - 1] + fs->fs_frag;
840 	if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
841 	    bap[indx - fs->fs_maxcontig] +
842 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
843 		return (nextblk);
844 	/*
845 	 * Here we convert ms of delay to frags as:
846 	 * (frags) = (ms) * (rev/sec) * (sect/rev) /
847 	 *	((sect/frag) * (ms/sec))
848 	 * then round up to the next block.
849 	 */
850 	nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
851 	    (NSPF(fs) * 1000), fs->fs_frag);
852 	return (nextblk);
853 }
854 
855 /*
856  * Implement the cylinder overflow algorithm.
857  *
858  * The policy implemented by this algorithm is:
859  *   1) allocate the block in its requested cylinder group.
860  *   2) quadradically rehash on the cylinder group number.
861  *   3) brute force search for a free block.
862  */
863 /*VARARGS5*/
864 static u_long
ffs_hashalloc(struct inode * ip,int cg,long pref,int size,allocfcn_t * allocator)865 ffs_hashalloc(struct inode *ip, int cg, long pref,
866 	      int size,	/* size for data blocks, mode for inodes */
867 	      allocfcn_t *allocator)
868 {
869 	struct fs *fs;
870 	long result;	/* XXX why not same type as we return? */
871 	int i, icg = cg;
872 
873 	fs = ip->i_fs;
874 	/*
875 	 * 1: preferred cylinder group
876 	 */
877 	result = (*allocator)(ip, cg, pref, size);
878 	if (result)
879 		return (result);
880 	/*
881 	 * 2: quadratic rehash
882 	 */
883 	for (i = 1; i < fs->fs_ncg; i *= 2) {
884 		cg += i;
885 		if (cg >= fs->fs_ncg)
886 			cg -= fs->fs_ncg;
887 		result = (*allocator)(ip, cg, 0, size);
888 		if (result)
889 			return (result);
890 	}
891 	/*
892 	 * 3: brute force search
893 	 * Note that we start at i == 2, since 0 was checked initially,
894 	 * and 1 is always checked in the quadratic rehash.
895 	 */
896 	cg = (icg + 2) % fs->fs_ncg;
897 	for (i = 2; i < fs->fs_ncg; i++) {
898 		result = (*allocator)(ip, cg, 0, size);
899 		if (result)
900 			return (result);
901 		cg++;
902 		if (cg == fs->fs_ncg)
903 			cg = 0;
904 	}
905 	return (0);
906 }
907 
908 /*
909  * Determine whether a fragment can be extended.
910  *
911  * Check to see if the necessary fragments are available, and
912  * if they are, allocate them.
913  */
914 static ufs_daddr_t
ffs_fragextend(struct inode * ip,int cg,long bprev,int osize,int nsize)915 ffs_fragextend(struct inode *ip, int cg, long bprev, int osize, int nsize)
916 {
917 	struct fs *fs;
918 	struct cg *cgp;
919 	struct buf *bp;
920 	long bno;
921 	int frags, bbase;
922 	int i, error;
923 	uint8_t *blksfree;
924 
925 	fs = ip->i_fs;
926 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
927 		return (0);
928 	frags = numfrags(fs, nsize);
929 	bbase = fragnum(fs, bprev);
930 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
931 		/* cannot extend across a block boundary */
932 		return (0);
933 	}
934 	KKASSERT(blknum(fs, bprev) == blknum(fs, bprev + frags - 1));
935 	error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
936 		(int)fs->fs_cgsize, &bp);
937 	if (error) {
938 		brelse(bp);
939 		return (0);
940 	}
941 	cgp = (struct cg *)bp->b_data;
942 	if (!cg_chkmagic(cgp)) {
943 		brelse(bp);
944 		return (0);
945 	}
946 	cgp->cg_time = time_second;
947 	bno = dtogd(fs, bprev);
948 	blksfree = cg_blksfree(cgp);
949 	for (i = numfrags(fs, osize); i < frags; i++) {
950 		if (isclr(blksfree, bno + i)) {
951 			brelse(bp);
952 			return (0);
953 		}
954 	}
955 
956 	/*
957 	 * the current fragment can be extended
958 	 * deduct the count on fragment being extended into
959 	 * increase the count on the remaining fragment (if any)
960 	 * allocate the extended piece
961 	 *
962 	 * ---oooooooooonnnnnnn111----
963 	 *    [-----frags-----]
964 	 *    ^                       ^
965 	 *    bbase                   fs_frag
966 	 */
967 	for (i = frags; i < fs->fs_frag - bbase; i++) {
968 		if (isclr(blksfree, bno + i))
969 			break;
970 	}
971 
972 	/*
973 	 * Size of original free frag is [i - numfrags(fs, osize)]
974 	 * Size of remaining free frag is [i - frags]
975 	 */
976 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
977 	if (i != frags)
978 		cgp->cg_frsum[i - frags]++;
979 	for (i = numfrags(fs, osize); i < frags; i++) {
980 		clrbit(blksfree, bno + i);
981 		cgp->cg_cs.cs_nffree--;
982 		fs->fs_cstotal.cs_nffree--;
983 		fs->fs_cs(fs, cg).cs_nffree--;
984 	}
985 	fs->fs_fmod = 1;
986 	if (DOINGSOFTDEP(ITOV(ip)))
987 		softdep_setup_blkmapdep(bp, fs, bprev);
988 	bdwrite(bp);
989 	return (bprev);
990 }
991 
992 /*
993  * Determine whether a block can be allocated.
994  *
995  * Check to see if a block of the appropriate size is available,
996  * and if it is, allocate it.
997  */
998 static ufs_daddr_t
ffs_alloccg(struct inode * ip,int cg,ufs_daddr_t bpref,int size)999 ffs_alloccg(struct inode *ip, int cg, ufs_daddr_t bpref, int size)
1000 {
1001 	struct fs *fs;
1002 	struct cg *cgp;
1003 	struct buf *bp;
1004 	int i;
1005 	ufs_daddr_t bno, blkno;
1006 	int allocsiz, error, frags;
1007 	uint8_t *blksfree;
1008 
1009 	fs = ip->i_fs;
1010 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
1011 		return (0);
1012 	error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1013 		(int)fs->fs_cgsize, &bp);
1014 	if (error) {
1015 		brelse(bp);
1016 		return (0);
1017 	}
1018 	cgp = (struct cg *)bp->b_data;
1019 	if (!cg_chkmagic(cgp) ||
1020 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
1021 		brelse(bp);
1022 		return (0);
1023 	}
1024 	cgp->cg_time = time_second;
1025 	if (size == fs->fs_bsize) {
1026 		bno = ffs_alloccgblk(ip, bp, bpref);
1027 		bdwrite(bp);
1028 		return (bno);
1029 	}
1030 	/*
1031 	 * Check to see if any fragments of sufficient size are already
1032 	 * available.  Fit the data into a larger fragment if necessary,
1033 	 * before allocating a whole new block.
1034 	 */
1035 	blksfree = cg_blksfree(cgp);
1036 	frags = numfrags(fs, size);
1037 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) {
1038 		if (cgp->cg_frsum[allocsiz] != 0)
1039 			break;
1040 	}
1041 	if (allocsiz == fs->fs_frag) {
1042 		/*
1043 		 * No fragments were available, allocate a whole block and
1044 		 * cut the requested fragment (of size frags) out of it.
1045 		 */
1046 		if (cgp->cg_cs.cs_nbfree == 0) {
1047 			brelse(bp);
1048 			return (0);
1049 		}
1050 		bno = ffs_alloccgblk(ip, bp, bpref);
1051 		bpref = dtogd(fs, bno);
1052 		for (i = frags; i < fs->fs_frag; i++)
1053 			setbit(blksfree, bpref + i);
1054 
1055 		/*
1056 		 * Calculate the number of free frags still remaining after
1057 		 * we have cut out the requested allocation.  Indicate that
1058 		 * a fragment of that size is now available for future
1059 		 * allocation.
1060 		 */
1061 		i = fs->fs_frag - frags;
1062 		cgp->cg_cs.cs_nffree += i;
1063 		fs->fs_cstotal.cs_nffree += i;
1064 		fs->fs_cs(fs, cg).cs_nffree += i;
1065 		fs->fs_fmod = 1;
1066 		cgp->cg_frsum[i]++;
1067 		bdwrite(bp);
1068 		return (bno);
1069 	}
1070 
1071 	/*
1072 	 * cg_frsum[] has told us that a free fragment of allocsiz size is
1073 	 * available.  Find it, then clear the bitmap bits associated with
1074 	 * the size we want.
1075 	 */
1076 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
1077 	if (bno < 0) {
1078 		brelse(bp);
1079 		return (0);
1080 	}
1081 	for (i = 0; i < frags; i++)
1082 		clrbit(blksfree, bno + i);
1083 	cgp->cg_cs.cs_nffree -= frags;
1084 	fs->fs_cstotal.cs_nffree -= frags;
1085 	fs->fs_cs(fs, cg).cs_nffree -= frags;
1086 	fs->fs_fmod = 1;
1087 
1088 	/*
1089 	 * Account for the allocation.  The original searched size that we
1090 	 * found is no longer available.  If we cut out a smaller piece then
1091 	 * a smaller fragment is now available.
1092 	 */
1093 	cgp->cg_frsum[allocsiz]--;
1094 	if (frags != allocsiz)
1095 		cgp->cg_frsum[allocsiz - frags]++;
1096 	blkno = cg * fs->fs_fpg + bno;
1097 	if (DOINGSOFTDEP(ITOV(ip)))
1098 		softdep_setup_blkmapdep(bp, fs, blkno);
1099 	bdwrite(bp);
1100 	return ((u_long)blkno);
1101 }
1102 
1103 /*
1104  * Allocate a block in a cylinder group.
1105  *
1106  * This algorithm implements the following policy:
1107  *   1) allocate the requested block.
1108  *   2) allocate a rotationally optimal block in the same cylinder.
1109  *   3) allocate the next available block on the block rotor for the
1110  *      specified cylinder group.
1111  * Note that this routine only allocates fs_bsize blocks; these
1112  * blocks may be fragmented by the routine that allocates them.
1113  */
1114 static ufs_daddr_t
ffs_alloccgblk(struct inode * ip,struct buf * bp,ufs_daddr_t bpref)1115 ffs_alloccgblk(struct inode *ip, struct buf *bp, ufs_daddr_t bpref)
1116 {
1117 	struct fs *fs;
1118 	struct cg *cgp;
1119 	ufs_daddr_t bno, blkno;
1120 	int cylno, pos, delta;
1121 	short *cylbp;
1122 	int i;
1123 	uint8_t *blksfree;
1124 
1125 	fs = ip->i_fs;
1126 	cgp = (struct cg *)bp->b_data;
1127 	blksfree = cg_blksfree(cgp);
1128 	if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
1129 		bpref = cgp->cg_rotor;
1130 		goto norot;
1131 	}
1132 	bpref = blknum(fs, bpref);
1133 	bpref = dtogd(fs, bpref);
1134 	/*
1135 	 * if the requested block is available, use it
1136 	 */
1137 	if (ffs_isblock(fs, blksfree, fragstoblks(fs, bpref))) {
1138 		bno = bpref;
1139 		goto gotit;
1140 	}
1141 	if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
1142 		/*
1143 		 * Block layout information is not available.
1144 		 * Leaving bpref unchanged means we take the
1145 		 * next available free block following the one
1146 		 * we just allocated. Hopefully this will at
1147 		 * least hit a track cache on drives of unknown
1148 		 * geometry (e.g. SCSI).
1149 		 */
1150 		goto norot;
1151 	}
1152 	/*
1153 	 * check for a block available on the same cylinder
1154 	 */
1155 	cylno = cbtocylno(fs, bpref);
1156 	if (cg_blktot(cgp)[cylno] == 0)
1157 		goto norot;
1158 	/*
1159 	 * check the summary information to see if a block is
1160 	 * available in the requested cylinder starting at the
1161 	 * requested rotational position and proceeding around.
1162 	 */
1163 	cylbp = cg_blks(fs, cgp, cylno);
1164 	pos = cbtorpos(fs, bpref);
1165 	for (i = pos; i < fs->fs_nrpos; i++)
1166 		if (cylbp[i] > 0)
1167 			break;
1168 	if (i == fs->fs_nrpos)
1169 		for (i = 0; i < pos; i++)
1170 			if (cylbp[i] > 0)
1171 				break;
1172 	if (cylbp[i] > 0) {
1173 		/*
1174 		 * found a rotational position, now find the actual
1175 		 * block. A panic if none is actually there.
1176 		 */
1177 		pos = cylno % fs->fs_cpc;
1178 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1179 		if (fs_postbl(fs, pos)[i] == -1) {
1180 			kprintf("pos = %d, i = %d, fs = %s\n",
1181 			    pos, i, fs->fs_fsmnt);
1182 			panic("ffs_alloccgblk: cyl groups corrupted");
1183 		}
1184 		for (i = fs_postbl(fs, pos)[i];; ) {
1185 			if (ffs_isblock(fs, blksfree, bno + i)) {
1186 				bno = blkstofrags(fs, (bno + i));
1187 				goto gotit;
1188 			}
1189 			delta = fs_rotbl(fs)[i];
1190 			if (delta <= 0 ||
1191 			    delta + i > fragstoblks(fs, fs->fs_fpg))
1192 				break;
1193 			i += delta;
1194 		}
1195 		kprintf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1196 		panic("ffs_alloccgblk: can't find blk in cyl");
1197 	}
1198 norot:
1199 	/*
1200 	 * no blocks in the requested cylinder, so take next
1201 	 * available one in this cylinder group.
1202 	 */
1203 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1204 	if (bno < 0)
1205 		return (0);
1206 	cgp->cg_rotor = bno;
1207 gotit:
1208 	blkno = fragstoblks(fs, bno);
1209 	ffs_clrblock(fs, blksfree, (long)blkno);
1210 	ffs_clusteracct(fs, cgp, blkno, -1);
1211 	cgp->cg_cs.cs_nbfree--;
1212 	fs->fs_cstotal.cs_nbfree--;
1213 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1214 	cylno = cbtocylno(fs, bno);
1215 	cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1216 	cg_blktot(cgp)[cylno]--;
1217 	fs->fs_fmod = 1;
1218 	blkno = cgp->cg_cgx * fs->fs_fpg + bno;
1219 	if (DOINGSOFTDEP(ITOV(ip)))
1220 		softdep_setup_blkmapdep(bp, fs, blkno);
1221 	return (blkno);
1222 }
1223 
1224 /*
1225  * Determine whether a cluster can be allocated.
1226  *
1227  * We do not currently check for optimal rotational layout if there
1228  * are multiple choices in the same cylinder group. Instead we just
1229  * take the first one that we find following bpref.
1230  */
1231 static ufs_daddr_t
ffs_clusteralloc(struct inode * ip,int cg,ufs_daddr_t bpref,int len)1232 ffs_clusteralloc(struct inode *ip, int cg, ufs_daddr_t bpref, int len)
1233 {
1234 	struct fs *fs;
1235 	struct cg *cgp;
1236 	struct buf *bp;
1237 	int i, got, run, bno, bit, map;
1238 	u_char *mapp;
1239 	int32_t *lp;
1240 	uint8_t *blksfree;
1241 
1242 	fs = ip->i_fs;
1243 	if (fs->fs_maxcluster[cg] < len)
1244 		return (0);
1245 	if (bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1246 		  (int)fs->fs_cgsize, &bp)) {
1247 		goto fail;
1248 	}
1249 	cgp = (struct cg *)bp->b_data;
1250 	if (!cg_chkmagic(cgp))
1251 		goto fail;
1252 
1253 	/*
1254 	 * Check to see if a cluster of the needed size (or bigger) is
1255 	 * available in this cylinder group.
1256 	 */
1257 	lp = &cg_clustersum(cgp)[len];
1258 	for (i = len; i <= fs->fs_contigsumsize; i++)
1259 		if (*lp++ > 0)
1260 			break;
1261 	if (i > fs->fs_contigsumsize) {
1262 		/*
1263 		 * This is the first time looking for a cluster in this
1264 		 * cylinder group. Update the cluster summary information
1265 		 * to reflect the true maximum sized cluster so that
1266 		 * future cluster allocation requests can avoid reading
1267 		 * the cylinder group map only to find no clusters.
1268 		 */
1269 		lp = &cg_clustersum(cgp)[len - 1];
1270 		for (i = len - 1; i > 0; i--)
1271 			if (*lp-- > 0)
1272 				break;
1273 		fs->fs_maxcluster[cg] = i;
1274 		goto fail;
1275 	}
1276 	/*
1277 	 * Search the cluster map to find a big enough cluster.
1278 	 * We take the first one that we find, even if it is larger
1279 	 * than we need as we prefer to get one close to the previous
1280 	 * block allocation. We do not search before the current
1281 	 * preference point as we do not want to allocate a block
1282 	 * that is allocated before the previous one (as we will
1283 	 * then have to wait for another pass of the elevator
1284 	 * algorithm before it will be read). We prefer to fail and
1285 	 * be recalled to try an allocation in the next cylinder group.
1286 	 */
1287 	if (dtog(fs, bpref) != cg)
1288 		bpref = 0;
1289 	else
1290 		bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1291 	mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1292 	map = *mapp++;
1293 	bit = 1 << (bpref % NBBY);
1294 	for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1295 		if ((map & bit) == 0) {
1296 			run = 0;
1297 		} else {
1298 			run++;
1299 			if (run == len)
1300 				break;
1301 		}
1302 		if ((got & (NBBY - 1)) != (NBBY - 1)) {
1303 			bit <<= 1;
1304 		} else {
1305 			map = *mapp++;
1306 			bit = 1;
1307 		}
1308 	}
1309 	if (got >= cgp->cg_nclusterblks)
1310 		goto fail;
1311 	/*
1312 	 * Allocate the cluster that we have found.
1313 	 */
1314 	blksfree = cg_blksfree(cgp);
1315 	for (i = 1; i <= len; i++) {
1316 		if (!ffs_isblock(fs, blksfree, got - run + i))
1317 			panic("ffs_clusteralloc: map mismatch");
1318 	}
1319 	bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1320 	if (dtog(fs, bno) != cg)
1321 		panic("ffs_clusteralloc: allocated out of group");
1322 	len = blkstofrags(fs, len);
1323 	for (i = 0; i < len; i += fs->fs_frag) {
1324 		if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i)
1325 			panic("ffs_clusteralloc: lost block");
1326 	}
1327 	bdwrite(bp);
1328 	return (bno);
1329 
1330 fail:
1331 	brelse(bp);
1332 	return (0);
1333 }
1334 
1335 /*
1336  * Determine whether an inode can be allocated.
1337  *
1338  * Check to see if an inode is available, and if it is,
1339  * allocate it using the following policy:
1340  *   1) allocate the requested inode.
1341  *   2) allocate the next available inode after the requested
1342  *      inode in the specified cylinder group.
1343  *   3) the inode must not already be in the inode hash table.  We
1344  *	can encounter such a case because the vnode reclamation sequence
1345  *	frees the bit
1346  *   3) the inode must not already be in the inode hash, otherwise it
1347  *	may be in the process of being deallocated.  This can occur
1348  *	because the bitmap is updated before the inode is removed from
1349  *	hash.  If we were to reallocate the inode the caller could wind
1350  *	up returning a vnode/inode combination which is in an indeterminate
1351  *	state.
1352  */
1353 static ino_t
ffs_nodealloccg(struct inode * ip,int cg,ufs_daddr_t ipref,int mode)1354 ffs_nodealloccg(struct inode *ip, int cg, ufs_daddr_t ipref, int mode)
1355 {
1356 	struct ufsmount *ump;
1357 	struct fs *fs;
1358 	struct cg *cgp;
1359 	struct buf *bp;
1360 	uint8_t *inosused;
1361 	uint8_t map;
1362 	int error, len, arraysize, i;
1363 	int icheckmiss;
1364 	ufs_daddr_t ibase;
1365 	struct vnode *vp;
1366 
1367 	vp = ITOV(ip);
1368 	ump = VFSTOUFS(vp->v_mount);
1369 	fs = ip->i_fs;
1370 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
1371 		return (0);
1372 	error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1373 		      (int)fs->fs_cgsize, &bp);
1374 	if (error) {
1375 		brelse(bp);
1376 		return (0);
1377 	}
1378 	cgp = (struct cg *)bp->b_data;
1379 	if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1380 		brelse(bp);
1381 		return (0);
1382 	}
1383 	inosused = cg_inosused(cgp);
1384 	icheckmiss = 0;
1385 
1386 	/*
1387 	 * Quick check, reuse the most recently free inode or continue
1388 	 * a scan from where we left off the last time.
1389 	 */
1390 	ibase = cg * fs->fs_ipg;
1391 	if (ipref) {
1392 		ipref %= fs->fs_ipg;
1393 		if (isclr(inosused, ipref)) {
1394 			if (ufs_ihashcheck(ump, ip->i_dev, ibase + ipref) == 0)
1395 				goto gotit;
1396 		}
1397 	}
1398 
1399 	/*
1400 	 * Scan the inode bitmap starting at irotor, be sure to handle
1401 	 * the edge case by going back to the beginning of the array.
1402 	 *
1403 	 * If the number of inodes is not byte-aligned, the unused bits
1404 	 * should be set to 1.  This will be sanity checked in gotit.  Note
1405 	 * that we have to be sure not to overlap the beginning and end
1406 	 * when irotor is in the middle of a byte as this will cause the
1407 	 * same bitmap byte to be checked twice.  To solve this problem we
1408 	 * just convert everything to a byte index for the loop.
1409 	 */
1410 	ipref = (cgp->cg_irotor % fs->fs_ipg) >> 3;	/* byte index */
1411 	len = (fs->fs_ipg + 7) >> 3;			/* byte size */
1412 	arraysize = len;
1413 
1414 	while (len > 0) {
1415 		map = inosused[ipref];
1416 		if (map != 255) {
1417 			for (i = 0; i < NBBY; ++i) {
1418 				/*
1419 				 * If we find a free bit we have to make sure
1420 				 * that the inode is not in the middle of
1421 				 * being destroyed.  The inode should not exist
1422 				 * in the inode hash.
1423 				 *
1424 				 * Adjust the rotor to try to hit the
1425 				 * quick-check up above.
1426 				 */
1427 				if ((map & (1 << i)) == 0) {
1428 					if (ufs_ihashcheck(ump, ip->i_dev, ibase + (ipref << 3) + i) == 0) {
1429 						ipref = (ipref << 3) + i;
1430 						cgp->cg_irotor = (ipref + 1) % fs->fs_ipg;
1431 						goto gotit;
1432 					}
1433 					++icheckmiss;
1434 				}
1435 			}
1436 		}
1437 
1438 		/*
1439 		 * Setup for the next byte, start at the beginning again if
1440 		 * we hit the end of the array.
1441 		 */
1442 		if (++ipref == arraysize)
1443 			ipref = 0;
1444 		--len;
1445 	}
1446 	if (icheckmiss == cgp->cg_cs.cs_nifree) {
1447 		brelse(bp);
1448 		return(0);
1449 	}
1450 	kprintf("fs = %s\n", fs->fs_fsmnt);
1451 	panic("ffs_nodealloccg: block not in map, icheckmiss/nfree %d/%d",
1452 		icheckmiss, cgp->cg_cs.cs_nifree);
1453 	/* NOTREACHED */
1454 
1455 	/*
1456 	 * ipref is a bit index as of the gotit label.
1457 	 */
1458 gotit:
1459 	KKASSERT(ipref >= 0 && ipref < fs->fs_ipg);
1460 	cgp->cg_time = time_second;
1461 	if (DOINGSOFTDEP(ITOV(ip)))
1462 		softdep_setup_inomapdep(bp, ip, ibase + ipref);
1463 	setbit(inosused, ipref);
1464 	cgp->cg_cs.cs_nifree--;
1465 	fs->fs_cstotal.cs_nifree--;
1466 	fs->fs_cs(fs, cg).cs_nifree--;
1467 	fs->fs_fmod = 1;
1468 	if ((mode & IFMT) == IFDIR) {
1469 		cgp->cg_cs.cs_ndir++;
1470 		fs->fs_cstotal.cs_ndir++;
1471 		fs->fs_cs(fs, cg).cs_ndir++;
1472 	}
1473 	bdwrite(bp);
1474 	return (ibase + ipref);
1475 }
1476 
1477 /*
1478  * Free a block or fragment.
1479  *
1480  * The specified block or fragment is placed back in the
1481  * free map. If a fragment is deallocated, a possible
1482  * block reassembly is checked.
1483  */
1484 void
ffs_blkfree_cg(struct fs * fs,struct vnode * i_devvp,cdev_t i_dev,ino_t i_number,uint32_t i_din_uid,ufs_daddr_t bno,long size)1485 ffs_blkfree_cg(struct fs * fs, struct vnode * i_devvp, cdev_t i_dev, ino_t i_number,
1486 	        uint32_t i_din_uid, ufs_daddr_t bno, long size)
1487 {
1488 	struct cg *cgp;
1489 	struct buf *bp;
1490 	ufs_daddr_t blkno;
1491 	int i, error, cg, blk, frags, bbase;
1492 	uint8_t *blksfree;
1493 
1494 #if 0
1495 	/*
1496 	 * ffs_blkfree() handles TRIM if UFS is mounted with the 'trim'
1497 	 * option, do not issue an unconditional duplicate here!
1498 	 * VOP_FREEBLKS(i_devvp, fsbtodoff(fs, bno), size);
1499 	 */
1500 #endif
1501 	if ((uint)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
1502 	    fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
1503 		kprintf("dev=%s, bno = %ld, bsize = %ld, size = %ld, fs = %s\n",
1504 		    devtoname(i_dev), (long)bno, (long)fs->fs_bsize, size,
1505 		    fs->fs_fsmnt);
1506 		panic("ffs_blkfree: bad size");
1507 	}
1508 	cg = dtog(fs, bno);
1509 	if ((uint)bno >= fs->fs_size) {
1510 		kprintf("bad block %ld, ino %lu\n",
1511 		    (long)bno, (u_long)i_number);
1512 		ffs_fserr(fs, i_din_uid, "bad block");
1513 		return;
1514 	}
1515 
1516 	/*
1517 	 * Load the cylinder group
1518 	 */
1519 	error = bread(i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1520 		      (int)fs->fs_cgsize, &bp);
1521 	if (error) {
1522 		brelse(bp);
1523 		return;
1524 	}
1525 	cgp = (struct cg *)bp->b_data;
1526 	if (!cg_chkmagic(cgp)) {
1527 		brelse(bp);
1528 		return;
1529 	}
1530 	cgp->cg_time = time_second;
1531 	bno = dtogd(fs, bno);
1532 	blksfree = cg_blksfree(cgp);
1533 
1534 	if (size == fs->fs_bsize) {
1535 		/*
1536 		 * Free a whole block
1537 		 */
1538 		blkno = fragstoblks(fs, bno);
1539 		if (!ffs_isfreeblock(fs, blksfree, blkno)) {
1540 			kprintf("dev = %s, block = %ld, fs = %s\n",
1541 			    devtoname(i_dev), (long)bno, fs->fs_fsmnt);
1542 			panic("ffs_blkfree: freeing free block");
1543 		}
1544 		ffs_setblock(fs, blksfree, blkno);
1545 		ffs_clusteracct(fs, cgp, blkno, 1);
1546 		cgp->cg_cs.cs_nbfree++;
1547 		fs->fs_cstotal.cs_nbfree++;
1548 		fs->fs_cs(fs, cg).cs_nbfree++;
1549 		i = cbtocylno(fs, bno);
1550 		cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1551 		cg_blktot(cgp)[i]++;
1552 	} else {
1553 		/*
1554 		 * Free a fragment within a block.
1555 		 *
1556 		 * bno is the starting block number of the fragment being
1557 		 * freed.
1558 		 *
1559 		 * bbase is the starting block number for the filesystem
1560 		 * block containing the fragment.
1561 		 *
1562 		 * blk is the current bitmap for the fragments within the
1563 		 * filesystem block containing the fragment.
1564 		 *
1565 		 * frags is the number of fragments being freed
1566 		 *
1567 		 * Call ffs_fragacct() to account for the removal of all
1568 		 * current fragments, then adjust the bitmap to free the
1569 		 * requested fragment, and finally call ffs_fragacct() again
1570 		 * to regenerate the accounting.
1571 		 */
1572 		bbase = bno - fragnum(fs, bno);
1573 		blk = blkmap(fs, blksfree, bbase);
1574 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1575 		frags = numfrags(fs, size);
1576 		for (i = 0; i < frags; i++) {
1577 			if (isset(blksfree, bno + i)) {
1578 				kprintf("dev = %s, block = %ld, fs = %s\n",
1579 				    devtoname(i_dev), (long)(bno + i),
1580 				    fs->fs_fsmnt);
1581 				panic("ffs_blkfree: freeing free frag");
1582 			}
1583 			setbit(blksfree, bno + i);
1584 		}
1585 		cgp->cg_cs.cs_nffree += i;
1586 		fs->fs_cstotal.cs_nffree += i;
1587 		fs->fs_cs(fs, cg).cs_nffree += i;
1588 
1589 		/*
1590 		 * Add back in counts associated with the new frags
1591 		 */
1592 		blk = blkmap(fs, blksfree, bbase);
1593 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1594 
1595 		/*
1596 		 * If a complete block has been reassembled, account for it
1597 		 */
1598 		blkno = fragstoblks(fs, bbase);
1599 		if (ffs_isblock(fs, blksfree, blkno)) {
1600 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
1601 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1602 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1603 			ffs_clusteracct(fs, cgp, blkno, 1);
1604 			cgp->cg_cs.cs_nbfree++;
1605 			fs->fs_cstotal.cs_nbfree++;
1606 			fs->fs_cs(fs, cg).cs_nbfree++;
1607 			i = cbtocylno(fs, bbase);
1608 			cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1609 			cg_blktot(cgp)[i]++;
1610 		}
1611 	}
1612 	fs->fs_fmod = 1;
1613 	bdwrite(bp);
1614 }
1615 
1616 struct ffs_blkfree_trim_params {
1617 	struct task task;
1618 	ufs_daddr_t bno;
1619 	long size;
1620 
1621 	/*
1622 	 * With TRIM,  inode pointer is gone in the callback but we still need
1623 	 * the following fields for  ffs_blkfree_cg()
1624 	 */
1625 	struct vnode *i_devvp;
1626 	struct fs *i_fs;
1627 	cdev_t i_dev;
1628 	ino_t i_number;
1629 	uint32_t i_din_uid;
1630 };
1631 
1632 
1633 static void
ffs_blkfree_trim_task(void * ctx,int pending)1634 ffs_blkfree_trim_task(void *ctx, int pending)
1635 {
1636 	struct ffs_blkfree_trim_params *tp;
1637 
1638 	tp = ctx;
1639 	ffs_blkfree_cg(tp->i_fs, tp->i_devvp, tp->i_dev, tp->i_number,
1640 	    tp->i_din_uid, tp->bno, tp->size);
1641 	kfree(tp, M_TEMP);
1642 }
1643 
1644 
1645 
1646 static void
ffs_blkfree_trim_completed(struct bio * biop)1647 ffs_blkfree_trim_completed(struct bio *biop)
1648 {
1649 	struct buf *bp = biop->bio_buf;
1650 	struct ffs_blkfree_trim_params *tp;
1651 
1652 	tp = bp->b_bio1.bio_caller_info1.ptr;
1653 	TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp);
1654 	tp = biop->bio_caller_info1.ptr;
1655 	taskqueue_enqueue(taskqueue_swi, &tp->task);
1656 	biodone(biop);
1657 }
1658 
1659 
1660 /*
1661  * If TRIM is enabled, we TRIM the blocks first then free them. We do this
1662  * after TRIM is finished and the callback handler is called. The logic here
1663  * is that we free the blocks before updating the bitmap so that we don't
1664  * reuse a block before we actually trim it, which would result in trimming
1665  * a valid block.
1666  */
1667 void
ffs_blkfree(struct inode * ip,ufs_daddr_t bno,long size)1668 ffs_blkfree(struct inode *ip, ufs_daddr_t bno, long size)
1669 {
1670 	struct mount *mp = ip->i_devvp->v_mount;
1671 	struct ffs_blkfree_trim_params *tp;
1672 
1673 	if (!(mp->mnt_flag & MNT_TRIM)) {
1674 		ffs_blkfree_cg(ip->i_fs, ip->i_devvp,ip->i_dev,ip->i_number,
1675 		    ip->i_uid, bno, size);
1676 		return;
1677 	}
1678 
1679 	struct buf *bp;
1680 
1681 	tp = kmalloc(sizeof(struct ffs_blkfree_trim_params), M_TEMP, M_WAITOK);
1682 	tp->bno = bno;
1683 	tp->i_fs= ip->i_fs;
1684 	tp->i_devvp = ip->i_devvp;
1685 	tp->i_dev = ip->i_dev;
1686 	tp->i_din_uid = ip->i_uid;
1687 	tp->i_number = ip->i_number;
1688 	tp->size = size;
1689 
1690 	bp = getnewbuf(0, 0, 0, 1);
1691 	BUF_KERNPROC(bp);
1692 	bp->b_cmd = BUF_CMD_FREEBLKS;
1693 	bp->b_bio1.bio_offset =  fsbtodoff(ip->i_fs, bno);
1694 	bp->b_bcount = size;
1695 	bp->b_bio1.bio_caller_info1.ptr = tp;
1696 	bp->b_bio1.bio_done = ffs_blkfree_trim_completed;
1697 	vn_strategy(ip->i_devvp, &bp->b_bio1);
1698 }
1699 
1700 #ifdef DIAGNOSTIC
1701 /*
1702  * Verify allocation of a block or fragment. Returns true if block or
1703  * fragment is allocated, false if it is free.
1704  */
1705 static int
ffs_checkblk(struct inode * ip,ufs_daddr_t bno,long size)1706 ffs_checkblk(struct inode *ip, ufs_daddr_t bno, long size)
1707 {
1708 	struct fs *fs;
1709 	struct cg *cgp;
1710 	struct buf *bp;
1711 	int i, error, frags, free;
1712 	uint8_t *blksfree;
1713 
1714 	fs = ip->i_fs;
1715 	if ((uint)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1716 		kprintf("bsize = %ld, size = %ld, fs = %s\n",
1717 		    (long)fs->fs_bsize, size, fs->fs_fsmnt);
1718 		panic("ffs_checkblk: bad size");
1719 	}
1720 	if ((uint)bno >= fs->fs_size)
1721 		panic("ffs_checkblk: bad block %d", bno);
1722 	error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, dtog(fs, bno))),
1723 		      (int)fs->fs_cgsize, &bp);
1724 	if (error)
1725 		panic("ffs_checkblk: cg bread failed");
1726 	cgp = (struct cg *)bp->b_data;
1727 	if (!cg_chkmagic(cgp))
1728 		panic("ffs_checkblk: cg magic mismatch");
1729 	blksfree = cg_blksfree(cgp);
1730 	bno = dtogd(fs, bno);
1731 	if (size == fs->fs_bsize) {
1732 		free = ffs_isblock(fs, blksfree, fragstoblks(fs, bno));
1733 	} else {
1734 		frags = numfrags(fs, size);
1735 		for (free = 0, i = 0; i < frags; i++)
1736 			if (isset(blksfree, bno + i))
1737 				free++;
1738 		if (free != 0 && free != frags)
1739 			panic("ffs_checkblk: partially free fragment");
1740 	}
1741 	brelse(bp);
1742 	return (!free);
1743 }
1744 #endif /* DIAGNOSTIC */
1745 
1746 /*
1747  * Free an inode.
1748  */
1749 int
ffs_vfree(struct vnode * pvp,ino_t ino,int mode)1750 ffs_vfree(struct vnode *pvp, ino_t ino, int mode)
1751 {
1752 	if (DOINGSOFTDEP(pvp)) {
1753 		softdep_freefile(pvp, ino, mode);
1754 		return (0);
1755 	}
1756 	return (ffs_freefile(pvp, ino, mode));
1757 }
1758 
1759 /*
1760  * Do the actual free operation.
1761  * The specified inode is placed back in the free map.
1762  */
1763 int
ffs_freefile(struct vnode * pvp,ino_t ino,int mode)1764 ffs_freefile(struct vnode *pvp, ino_t ino, int mode)
1765 {
1766 	struct fs *fs;
1767 	struct cg *cgp;
1768 	struct inode *pip;
1769 	struct buf *bp;
1770 	int error, cg;
1771 	uint8_t *inosused;
1772 
1773 	pip = VTOI(pvp);
1774 	fs = pip->i_fs;
1775 	if ((uint)ino >= fs->fs_ipg * fs->fs_ncg)
1776 		panic("ffs_vfree: range: dev = (%d,%d), ino = %"PRId64", fs = %s",
1777 		    major(pip->i_dev), minor(pip->i_dev), ino, fs->fs_fsmnt);
1778 	cg = ino_to_cg(fs, ino);
1779 	error = bread(pip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1780 		      (int)fs->fs_cgsize, &bp);
1781 	if (error) {
1782 		brelse(bp);
1783 		return (error);
1784 	}
1785 	cgp = (struct cg *)bp->b_data;
1786 	if (!cg_chkmagic(cgp)) {
1787 		brelse(bp);
1788 		return (0);
1789 	}
1790 	cgp->cg_time = time_second;
1791 	inosused = cg_inosused(cgp);
1792 	ino %= fs->fs_ipg;
1793 	if (isclr(inosused, ino)) {
1794 		kprintf("dev = %s, ino = %lu, fs = %s\n",
1795 		    devtoname(pip->i_dev), (u_long)ino, fs->fs_fsmnt);
1796 		if (fs->fs_ronly == 0)
1797 			panic("ffs_vfree: freeing free inode");
1798 	}
1799 	clrbit(inosused, ino);
1800 	if (ino < cgp->cg_irotor)
1801 		cgp->cg_irotor = ino;
1802 	cgp->cg_cs.cs_nifree++;
1803 	fs->fs_cstotal.cs_nifree++;
1804 	fs->fs_cs(fs, cg).cs_nifree++;
1805 	if ((mode & IFMT) == IFDIR) {
1806 		cgp->cg_cs.cs_ndir--;
1807 		fs->fs_cstotal.cs_ndir--;
1808 		fs->fs_cs(fs, cg).cs_ndir--;
1809 	}
1810 	fs->fs_fmod = 1;
1811 	bdwrite(bp);
1812 	return (0);
1813 }
1814 
1815 /*
1816  * Find a block of the specified size in the specified cylinder group.
1817  *
1818  * It is a panic if a request is made to find a block if none are
1819  * available.
1820  */
1821 static ufs_daddr_t
ffs_mapsearch(struct fs * fs,struct cg * cgp,ufs_daddr_t bpref,int allocsiz)1822 ffs_mapsearch(struct fs *fs, struct cg *cgp, ufs_daddr_t bpref, int allocsiz)
1823 {
1824 	ufs_daddr_t bno;
1825 	int start, len, loc, i;
1826 	int blk, field, subfield, pos;
1827 	uint8_t *blksfree;
1828 
1829 	/*
1830 	 * find the fragment by searching through the free block
1831 	 * map for an appropriate bit pattern.
1832 	 */
1833 	if (bpref)
1834 		start = dtogd(fs, bpref) / NBBY;
1835 	else
1836 		start = cgp->cg_frotor / NBBY;
1837 	blksfree = cg_blksfree(cgp);
1838 	len = howmany(fs->fs_fpg, NBBY) - start;
1839 	loc = scanc((uint)len, (u_char *)&blksfree[start],
1840 		(u_char *)fragtbl[fs->fs_frag],
1841 		(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1842 	if (loc == 0) {
1843 		len = start + 1;	/* XXX why overlap here? */
1844 		start = 0;
1845 		loc = scanc((uint)len, (u_char *)&blksfree[0],
1846 			(u_char *)fragtbl[fs->fs_frag],
1847 			(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1848 		if (loc == 0) {
1849 			kprintf("start = %d, len = %d, fs = %s\n",
1850 			    start, len, fs->fs_fsmnt);
1851 			panic("ffs_alloccg: map corrupted");
1852 			/* NOTREACHED */
1853 		}
1854 	}
1855 	bno = (start + len - loc) * NBBY;
1856 	cgp->cg_frotor = bno;
1857 	/*
1858 	 * found the byte in the map
1859 	 * sift through the bits to find the selected frag
1860 	 */
1861 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1862 		blk = blkmap(fs, blksfree, bno);
1863 		blk <<= 1;
1864 		field = around[allocsiz];
1865 		subfield = inside[allocsiz];
1866 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1867 			if ((blk & field) == subfield)
1868 				return (bno + pos);
1869 			field <<= 1;
1870 			subfield <<= 1;
1871 		}
1872 	}
1873 	kprintf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1874 	panic("ffs_alloccg: block not in map");
1875 	return (-1);
1876 }
1877 
1878 /*
1879  * Update the cluster map because of an allocation or free.
1880  *
1881  * Cnt == 1 means free; cnt == -1 means allocating.
1882  */
1883 static void
ffs_clusteracct(struct fs * fs,struct cg * cgp,ufs_daddr_t blkno,int cnt)1884 ffs_clusteracct(struct fs *fs, struct cg *cgp, ufs_daddr_t blkno, int cnt)
1885 {
1886 	int32_t *sump;
1887 	int32_t *lp;
1888 	u_char *freemapp, *mapp;
1889 	int i, start, end, forw, back, map, bit;
1890 
1891 	if (fs->fs_contigsumsize <= 0)
1892 		return;
1893 	freemapp = cg_clustersfree(cgp);
1894 	sump = cg_clustersum(cgp);
1895 	/*
1896 	 * Allocate or clear the actual block.
1897 	 */
1898 	if (cnt > 0)
1899 		setbit(freemapp, blkno);
1900 	else
1901 		clrbit(freemapp, blkno);
1902 	/*
1903 	 * Find the size of the cluster going forward.
1904 	 */
1905 	start = blkno + 1;
1906 	end = start + fs->fs_contigsumsize;
1907 	if (end >= cgp->cg_nclusterblks)
1908 		end = cgp->cg_nclusterblks;
1909 	mapp = &freemapp[start / NBBY];
1910 	map = *mapp++;
1911 	bit = 1 << (start % NBBY);
1912 	for (i = start; i < end; i++) {
1913 		if ((map & bit) == 0)
1914 			break;
1915 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
1916 			bit <<= 1;
1917 		} else {
1918 			map = *mapp++;
1919 			bit = 1;
1920 		}
1921 	}
1922 	forw = i - start;
1923 	/*
1924 	 * Find the size of the cluster going backward.
1925 	 */
1926 	start = blkno - 1;
1927 	end = start - fs->fs_contigsumsize;
1928 	if (end < 0)
1929 		end = -1;
1930 	mapp = &freemapp[start / NBBY];
1931 	map = *mapp--;
1932 	bit = 1 << (start % NBBY);
1933 	for (i = start; i > end; i--) {
1934 		if ((map & bit) == 0)
1935 			break;
1936 		if ((i & (NBBY - 1)) != 0) {
1937 			bit >>= 1;
1938 		} else {
1939 			map = *mapp--;
1940 			bit = 1 << (NBBY - 1);
1941 		}
1942 	}
1943 	back = start - i;
1944 	/*
1945 	 * Account for old cluster and the possibly new forward and
1946 	 * back clusters.
1947 	 */
1948 	i = back + forw + 1;
1949 	if (i > fs->fs_contigsumsize)
1950 		i = fs->fs_contigsumsize;
1951 	sump[i] += cnt;
1952 	if (back > 0)
1953 		sump[back] -= cnt;
1954 	if (forw > 0)
1955 		sump[forw] -= cnt;
1956 	/*
1957 	 * Update cluster summary information.
1958 	 */
1959 	lp = &sump[fs->fs_contigsumsize];
1960 	for (i = fs->fs_contigsumsize; i > 0; i--)
1961 		if (*lp-- > 0)
1962 			break;
1963 	fs->fs_maxcluster[cgp->cg_cgx] = i;
1964 }
1965 
1966 /*
1967  * Fserr prints the name of a filesystem with an error diagnostic.
1968  *
1969  * The form of the error message is:
1970  *	fs: error message
1971  */
1972 static void
ffs_fserr(struct fs * fs,uint uid,char * cp)1973 ffs_fserr(struct fs *fs, uint uid, char *cp)
1974 {
1975 	struct thread *td = curthread;
1976 	struct proc *p;
1977 
1978 	if ((p = td->td_proc) != NULL) {
1979 	    log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1980 		    p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1981 	} else {
1982 	    log(LOG_ERR, "system thread %p, uid %d on %s: %s\n",
1983 		    td, uid, fs->fs_fsmnt, cp);
1984 	}
1985 }
1986