xref: /netbsd-src/sys/ufs/ffs/ffs_alloc.c (revision 122b5006ee1bd67145794b4cde92f4fe4781a5ec)
1 /*	$NetBSD: ffs_alloc.c,v 1.170 2021/09/03 21:55:01 andvar Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Wasabi Systems, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 2002 Networks Associates Technology, Inc.
34  * All rights reserved.
35  *
36  * This software was developed for the FreeBSD Project by Marshall
37  * Kirk McKusick and Network Associates Laboratories, the Security
38  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
39  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
40  * research program
41  *
42  * Copyright (c) 1982, 1986, 1989, 1993
43  *	The Regents of the University of California.  All rights reserved.
44  *
45  * Redistribution and use in source and binary forms, with or without
46  * modification, are permitted provided that the following conditions
47  * are met:
48  * 1. Redistributions of source code must retain the above copyright
49  *    notice, this list of conditions and the following disclaimer.
50  * 2. Redistributions in binary form must reproduce the above copyright
51  *    notice, this list of conditions and the following disclaimer in the
52  *    documentation and/or other materials provided with the distribution.
53  * 3. Neither the name of the University nor the names of its contributors
54  *    may be used to endorse or promote products derived from this software
55  *    without specific prior written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67  * SUCH DAMAGE.
68  *
69  *	@(#)ffs_alloc.c	8.19 (Berkeley) 7/13/95
70  */
71 
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: ffs_alloc.c,v 1.170 2021/09/03 21:55:01 andvar Exp $");
74 
75 #if defined(_KERNEL_OPT)
76 #include "opt_ffs.h"
77 #include "opt_quota.h"
78 #include "opt_uvm_page_trkown.h"
79 #endif
80 
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/buf.h>
84 #include <sys/cprng.h>
85 #include <sys/kauth.h>
86 #include <sys/kernel.h>
87 #include <sys/mount.h>
88 #include <sys/proc.h>
89 #include <sys/syslog.h>
90 #include <sys/vnode.h>
91 #include <sys/wapbl.h>
92 #include <sys/cprng.h>
93 
94 #include <miscfs/specfs/specdev.h>
95 #include <ufs/ufs/quota.h>
96 #include <ufs/ufs/ufsmount.h>
97 #include <ufs/ufs/inode.h>
98 #include <ufs/ufs/ufs_extern.h>
99 #include <ufs/ufs/ufs_bswap.h>
100 #include <ufs/ufs/ufs_wapbl.h>
101 
102 #include <ufs/ffs/fs.h>
103 #include <ufs/ffs/ffs_extern.h>
104 
105 #ifdef UVM_PAGE_TRKOWN
106 #include <uvm/uvm_object.h>
107 #include <uvm/uvm_page.h>
108 #endif
109 
110 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int, int, int);
111 static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t, int, int);
112 static ino_t ffs_dirpref(struct inode *);
113 static daddr_t ffs_fragextend(struct inode *, int, daddr_t, int, int);
114 static void ffs_fserr(struct fs *, kauth_cred_t, const char *);
115 static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int, int, int,
116     daddr_t (*)(struct inode *, int, daddr_t, int, int, int));
117 static daddr_t ffs_nodealloccg(struct inode *, int, daddr_t, int, int, int);
118 static int32_t ffs_mapsearch(struct fs *, struct cg *,
119 				      daddr_t, int);
120 static void ffs_blkfree_common(struct ufsmount *, struct fs *, dev_t, struct buf *,
121     daddr_t, long, bool);
122 static void ffs_freefile_common(struct ufsmount *, struct fs *, dev_t, struct buf *, ino_t,
123     int, bool);
124 
125 /* if 1, changes in optimalization strategy are logged */
126 int ffs_log_changeopt = 0;
127 
128 /* in ffs_tables.c */
129 extern const int inside[], around[];
130 extern const u_char * const fragtbl[];
131 
132 /* Basic consistency check for block allocations */
133 static int
134 ffs_check_bad_allocation(const char *func, struct fs *fs, daddr_t bno,
135     long size, dev_t dev, ino_t inum)
136 {
137 	if ((u_int)size > fs->fs_bsize || ffs_fragoff(fs, size) != 0 ||
138 	    ffs_fragnum(fs, bno) + ffs_numfrags(fs, size) > fs->fs_frag) {
139 		panic("%s: bad size: dev = 0x%llx, bno = %" PRId64
140 		    " bsize = %d, size = %ld, fs = %s", func,
141 		    (long long)dev, bno, fs->fs_bsize, size, fs->fs_fsmnt);
142 	}
143 
144 	if (bno >= fs->fs_size) {
145 		printf("%s: bad block %" PRId64 ", ino %llu\n", func, bno,
146 		    (unsigned long long)inum);
147 		ffs_fserr(fs, NOCRED, "bad block");
148 		return EINVAL;
149 	}
150 	return 0;
151 }
152 
153 /*
154  * Allocate a block in the file system.
155  *
156  * The size of the requested block is given, which must be some
157  * multiple of fs_fsize and <= fs_bsize.
158  * A preference may be optionally specified. If a preference is given
159  * the following hierarchy is used to allocate a block:
160  *   1) allocate the requested block.
161  *   2) allocate a rotationally optimal block in the same cylinder.
162  *   3) allocate a block in the same cylinder group.
163  *   4) quadradically rehash into other cylinder groups, until an
164  *      available block is located.
165  * If no block preference is given the following hierarchy is used
166  * to allocate a block:
167  *   1) allocate a block in the cylinder group that contains the
168  *      inode for the file.
169  *   2) quadradically rehash into other cylinder groups, until an
170  *      available block is located.
171  *
172  * => called with um_lock held
173  * => releases um_lock before returning
174  */
175 int
176 ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size,
177     int flags, kauth_cred_t cred, daddr_t *bnp)
178 {
179 	struct ufsmount *ump;
180 	struct fs *fs;
181 	daddr_t bno;
182 	int cg;
183 #if defined(QUOTA) || defined(QUOTA2)
184 	int error;
185 #endif
186 
187 	fs = ip->i_fs;
188 	ump = ip->i_ump;
189 
190 	KASSERT(mutex_owned(&ump->um_lock));
191 
192 #ifdef UVM_PAGE_TRKOWN
193 
194 	/*
195 	 * Sanity-check that allocations within the file size
196 	 * do not allow other threads to read the stale contents
197 	 * of newly allocated blocks.
198 	 * Usually pages will exist to cover the new allocation.
199 	 * There is an optimization in ffs_write() where we skip
200 	 * creating pages if several conditions are met:
201 	 *  - the file must not be mapped (in any user address space).
202 	 *  - the write must cover whole pages and whole blocks.
203 	 * If those conditions are not met then pages must exist and
204 	 * be locked by the current thread.
205 	 */
206 
207 	struct vnode *vp = ITOV(ip);
208 	if (vp->v_type == VREG && (flags & IO_EXT) == 0 &&
209 	    ffs_lblktosize(fs, (voff_t)lbn) < round_page(vp->v_size) &&
210 	    ((vp->v_vflag & VV_MAPPED) != 0 || (size & PAGE_MASK) != 0 ||
211 	     ffs_blkoff(fs, size) != 0)) {
212 		struct vm_page *pg __diagused;
213 		struct uvm_object *uobj = &vp->v_uobj;
214 		voff_t off = trunc_page(ffs_lblktosize(fs, lbn));
215 		voff_t endoff = round_page(ffs_lblktosize(fs, lbn) + size);
216 
217 		rw_enter(uobj->vmobjlock, RW_WRITER);
218 		while (off < endoff) {
219 			pg = uvm_pagelookup(uobj, off);
220 			KASSERT((pg != NULL && pg->owner_tag != NULL &&
221 				 pg->owner == curproc->p_pid &&
222 				 pg->lowner == curlwp->l_lid));
223 			off += PAGE_SIZE;
224 		}
225 		rw_exit(uobj->vmobjlock);
226 	}
227 #endif
228 
229 	*bnp = 0;
230 
231 	KASSERTMSG((cred != NOCRED), "missing credential");
232 	KASSERTMSG(((u_int)size <= fs->fs_bsize),
233 	    "bad size: dev = 0x%llx, bsize = %d, size = %d, fs = %s",
234 	    (unsigned long long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
235 	KASSERTMSG((ffs_fragoff(fs, size) == 0),
236 	    "bad size: dev = 0x%llx, bsize = %d, size = %d, fs = %s",
237 	    (unsigned long long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
238 
239 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
240 		goto nospace;
241 	if (freespace(fs, fs->fs_minfree) <= 0 &&
242 	    kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL,
243 	    NULL, NULL) != 0)
244 		goto nospace;
245 #if defined(QUOTA) || defined(QUOTA2)
246 	mutex_exit(&ump->um_lock);
247 	if ((error = chkdq(ip, btodb(size), cred, 0)) != 0)
248 		return (error);
249 	mutex_enter(&ump->um_lock);
250 #endif
251 
252 	if (bpref >= fs->fs_size)
253 		bpref = 0;
254 	if (bpref == 0)
255 		cg = ino_to_cg(fs, ip->i_number);
256 	else
257 		cg = dtog(fs, bpref);
258 	bno = ffs_hashalloc(ip, cg, bpref, size, 0, flags, ffs_alloccg);
259 	if (bno > 0) {
260 		DIP_ADD(ip, blocks, btodb(size));
261 		if (flags & IO_EXT)
262 			ip->i_flag |= IN_CHANGE;
263 		else
264 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
265 		*bnp = bno;
266 		return (0);
267 	}
268 #if defined(QUOTA) || defined(QUOTA2)
269 	/*
270 	 * Restore user's disk quota because allocation failed.
271 	 */
272 	(void) chkdq(ip, -btodb(size), cred, FORCE);
273 #endif
274 	if (flags & B_CONTIG) {
275 		/*
276 		 * XXX ump->um_lock handling is "suspect" at best.
277 		 * For the case where ffs_hashalloc() fails early
278 		 * in the B_CONTIG case we reach here with um_lock
279 		 * already unlocked, so we can't release it again
280 		 * like in the normal error path.  See kern/39206.
281 		 *
282 		 *
283 		 * Fail silently - it's up to our caller to report
284 		 * errors.
285 		 */
286 		return (ENOSPC);
287 	}
288 nospace:
289 	mutex_exit(&ump->um_lock);
290 	ffs_fserr(fs, cred, "file system full");
291 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
292 	return (ENOSPC);
293 }
294 
295 /*
296  * Reallocate a fragment to a bigger size
297  *
298  * The number and size of the old block is given, and a preference
299  * and new size is also specified. The allocator attempts to extend
300  * the original block. Failing that, the regular block allocator is
301  * invoked to get an appropriate block.
302  *
303  * => called with um_lock held
304  * => return with um_lock released
305  */
306 int
307 ffs_realloccg(struct inode *ip, daddr_t lbprev, daddr_t bprev, daddr_t bpref,
308     int osize, int nsize, int flags, kauth_cred_t cred, struct buf **bpp,
309     daddr_t *blknop)
310 {
311 	struct ufsmount *ump;
312 	struct fs *fs;
313 	struct buf *bp;
314 	int cg, request, error;
315 	daddr_t bno;
316 
317 	fs = ip->i_fs;
318 	ump = ip->i_ump;
319 
320 	KASSERT(mutex_owned(&ump->um_lock));
321 
322 #ifdef UVM_PAGE_TRKOWN
323 
324 	/*
325 	 * Sanity-check that allocations within the file size
326 	 * do not allow other threads to read the stale contents
327 	 * of newly allocated blocks.
328 	 * Unlike in ffs_alloc(), here pages must always exist
329 	 * for such allocations, because only the last block of a file
330 	 * can be a fragment and ffs_write() will reallocate the
331 	 * fragment to the new size using ufs_balloc_range(),
332 	 * which always creates pages to cover blocks it allocates.
333 	 */
334 
335 	if (ITOV(ip)->v_type == VREG) {
336 		struct vm_page *pg __diagused;
337 		struct uvm_object *uobj = &ITOV(ip)->v_uobj;
338 		voff_t off = trunc_page(ffs_lblktosize(fs, lbprev));
339 		voff_t endoff = round_page(ffs_lblktosize(fs, lbprev) + osize);
340 
341 		rw_enter(uobj->vmobjlock, RW_WRITER);
342 		while (off < endoff) {
343 			pg = uvm_pagelookup(uobj, off);
344 			KASSERT(pg->owner == curproc->p_pid &&
345 				pg->lowner == curlwp->l_lid);
346 			off += PAGE_SIZE;
347 		}
348 		rw_exit(uobj->vmobjlock);
349 	}
350 #endif
351 
352 	KASSERTMSG((cred != NOCRED), "missing credential");
353 	KASSERTMSG(((u_int)osize <= fs->fs_bsize),
354 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
355 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
356 	    fs->fs_fsmnt);
357 	KASSERTMSG((ffs_fragoff(fs, osize) == 0),
358 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
359 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
360 	    fs->fs_fsmnt);
361 	KASSERTMSG(((u_int)nsize <= fs->fs_bsize),
362 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
363 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
364 	    fs->fs_fsmnt);
365 	KASSERTMSG((ffs_fragoff(fs, nsize) == 0),
366 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
367 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
368 	    fs->fs_fsmnt);
369 
370 	if (freespace(fs, fs->fs_minfree) <= 0 &&
371 	    kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL,
372 	    NULL, NULL) != 0) {
373 		mutex_exit(&ump->um_lock);
374 		goto nospace;
375 	}
376 
377 	if (bprev == 0) {
378 		panic("%s: bad bprev: dev = 0x%llx, bsize = %d, bprev = %"
379 		    PRId64 ", fs = %s", __func__,
380 		    (unsigned long long)ip->i_dev, fs->fs_bsize, bprev,
381 		    fs->fs_fsmnt);
382 	}
383 	mutex_exit(&ump->um_lock);
384 
385 	/*
386 	 * Allocate the extra space in the buffer.
387 	 */
388 	if (bpp != NULL &&
389 	    (error = bread(ITOV(ip), lbprev, osize, 0, &bp)) != 0) {
390 		return (error);
391 	}
392 #if defined(QUOTA) || defined(QUOTA2)
393 	if ((error = chkdq(ip, btodb(nsize - osize), cred, 0)) != 0) {
394 		if (bpp != NULL) {
395 			brelse(bp, 0);
396 		}
397 		return (error);
398 	}
399 #endif
400 	/*
401 	 * Check for extension in the existing location.
402 	 */
403 	cg = dtog(fs, bprev);
404 	mutex_enter(&ump->um_lock);
405 	if ((bno = ffs_fragextend(ip, cg, bprev, osize, nsize)) != 0) {
406 		DIP_ADD(ip, blocks, btodb(nsize - osize));
407 		if (flags & IO_EXT)
408 			ip->i_flag |= IN_CHANGE;
409 		else
410 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
411 
412 		if (bpp != NULL) {
413 			if (bp->b_blkno != FFS_FSBTODB(fs, bno)) {
414 				panic("%s: bad blockno %#llx != %#llx",
415 				    __func__, (unsigned long long) bp->b_blkno,
416 				    (unsigned long long)FFS_FSBTODB(fs, bno));
417 			}
418 			allocbuf(bp, nsize, 1);
419 			memset((char *)bp->b_data + osize, 0, nsize - osize);
420 			mutex_enter(bp->b_objlock);
421 			KASSERT(!cv_has_waiters(&bp->b_done));
422 			bp->b_oflags |= BO_DONE;
423 			mutex_exit(bp->b_objlock);
424 			*bpp = bp;
425 		}
426 		if (blknop != NULL) {
427 			*blknop = bno;
428 		}
429 		return (0);
430 	}
431 	/*
432 	 * Allocate a new disk location.
433 	 */
434 	if (bpref >= fs->fs_size)
435 		bpref = 0;
436 	switch ((int)fs->fs_optim) {
437 	case FS_OPTSPACE:
438 		/*
439 		 * Allocate an exact sized fragment. Although this makes
440 		 * best use of space, we will waste time relocating it if
441 		 * the file continues to grow. If the fragmentation is
442 		 * less than half of the minimum free reserve, we choose
443 		 * to begin optimizing for time.
444 		 */
445 		request = nsize;
446 		if (fs->fs_minfree < 5 ||
447 		    fs->fs_cstotal.cs_nffree >
448 		    fs->fs_dsize * fs->fs_minfree / (2 * 100))
449 			break;
450 
451 		if (ffs_log_changeopt) {
452 			log(LOG_NOTICE,
453 				"%s: optimization changed from SPACE to TIME\n",
454 				fs->fs_fsmnt);
455 		}
456 
457 		fs->fs_optim = FS_OPTTIME;
458 		break;
459 	case FS_OPTTIME:
460 		/*
461 		 * At this point we have discovered a file that is trying to
462 		 * grow a small fragment to a larger fragment. To save time,
463 		 * we allocate a full sized block, then free the unused portion.
464 		 * If the file continues to grow, the `ffs_fragextend' call
465 		 * above will be able to grow it in place without further
466 		 * copying. If aberrant programs cause disk fragmentation to
467 		 * grow within 2% of the free reserve, we choose to begin
468 		 * optimizing for space.
469 		 */
470 		request = fs->fs_bsize;
471 		if (fs->fs_cstotal.cs_nffree <
472 		    fs->fs_dsize * (fs->fs_minfree - 2) / 100)
473 			break;
474 
475 		if (ffs_log_changeopt) {
476 			log(LOG_NOTICE,
477 				"%s: optimization changed from TIME to SPACE\n",
478 				fs->fs_fsmnt);
479 		}
480 
481 		fs->fs_optim = FS_OPTSPACE;
482 		break;
483 	default:
484 		panic("%s: bad optim: dev = 0x%llx, optim = %d, fs = %s",
485 		    __func__, (unsigned long long)ip->i_dev, fs->fs_optim,
486 		    fs->fs_fsmnt);
487 		/* NOTREACHED */
488 	}
489 	bno = ffs_hashalloc(ip, cg, bpref, request, nsize, 0, ffs_alloccg);
490 	if (bno > 0) {
491 		/*
492 		 * Use forced deallocation registration, we can't handle
493 		 * failure here. This is safe, as this place is ever hit
494 		 * maximum once per write operation, when fragment is extended
495 		 * to longer fragment, or a full block.
496 		 */
497 		if ((ip->i_ump->um_mountp->mnt_wapbl) &&
498 		    (ITOV(ip)->v_type != VREG)) {
499 			/* this should never fail */
500 			error = UFS_WAPBL_REGISTER_DEALLOCATION_FORCE(
501 			    ip->i_ump->um_mountp, FFS_FSBTODB(fs, bprev),
502 			    osize);
503 			if (error)
504 				panic("ffs_realloccg: dealloc registration failed");
505 		} else {
506 			ffs_blkfree(fs, ip->i_devvp, bprev, (long)osize,
507 			    ip->i_number);
508 		}
509 		DIP_ADD(ip, blocks, btodb(nsize - osize));
510 		if (flags & IO_EXT)
511 			ip->i_flag |= IN_CHANGE;
512 		else
513 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
514 		if (bpp != NULL) {
515 			bp->b_blkno = FFS_FSBTODB(fs, bno);
516 			allocbuf(bp, nsize, 1);
517 			memset((char *)bp->b_data + osize, 0, (u_int)nsize - osize);
518 			mutex_enter(bp->b_objlock);
519 			KASSERT(!cv_has_waiters(&bp->b_done));
520 			bp->b_oflags |= BO_DONE;
521 			mutex_exit(bp->b_objlock);
522 			*bpp = bp;
523 		}
524 		if (blknop != NULL) {
525 			*blknop = bno;
526 		}
527 		return (0);
528 	}
529 	mutex_exit(&ump->um_lock);
530 
531 #if defined(QUOTA) || defined(QUOTA2)
532 	/*
533 	 * Restore user's disk quota because allocation failed.
534 	 */
535 	(void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
536 #endif
537 	if (bpp != NULL) {
538 		brelse(bp, 0);
539 	}
540 
541 nospace:
542 	/*
543 	 * no space available
544 	 */
545 	ffs_fserr(fs, cred, "file system full");
546 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
547 	return (ENOSPC);
548 }
549 
550 /*
551  * Allocate an inode in the file system.
552  *
553  * If allocating a directory, use ffs_dirpref to select the inode.
554  * If allocating in a directory, the following hierarchy is followed:
555  *   1) allocate the preferred inode.
556  *   2) allocate an inode in the same cylinder group.
557  *   3) quadradically rehash into other cylinder groups, until an
558  *      available inode is located.
559  * If no inode preference is given the following hierarchy is used
560  * to allocate an inode:
561  *   1) allocate an inode in cylinder group 0.
562  *   2) quadradically rehash into other cylinder groups, until an
563  *      available inode is located.
564  *
565  * => um_lock not held upon entry or return
566  */
567 int
568 ffs_valloc(struct vnode *pvp, int mode, kauth_cred_t cred, ino_t *inop)
569 {
570 	struct ufsmount *ump;
571 	struct inode *pip;
572 	struct fs *fs;
573 	ino_t ino, ipref;
574 	int cg, error;
575 
576 	UFS_WAPBL_JUNLOCK_ASSERT(pvp->v_mount);
577 
578 	pip = VTOI(pvp);
579 	fs = pip->i_fs;
580 	ump = pip->i_ump;
581 
582 	error = UFS_WAPBL_BEGIN(pvp->v_mount);
583 	if (error) {
584 		return error;
585 	}
586 	mutex_enter(&ump->um_lock);
587 	if (fs->fs_cstotal.cs_nifree == 0)
588 		goto noinodes;
589 
590 	if ((mode & IFMT) == IFDIR)
591 		ipref = ffs_dirpref(pip);
592 	else
593 		ipref = pip->i_number;
594 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
595 		ipref = 0;
596 	cg = ino_to_cg(fs, ipref);
597 	/*
598 	 * Track number of dirs created one after another
599 	 * in a same cg without intervening by files.
600 	 */
601 	if ((mode & IFMT) == IFDIR) {
602 		if (fs->fs_contigdirs[cg] < 255)
603 			fs->fs_contigdirs[cg]++;
604 	} else {
605 		if (fs->fs_contigdirs[cg] > 0)
606 			fs->fs_contigdirs[cg]--;
607 	}
608 	ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, 0, ffs_nodealloccg);
609 	if (ino == 0)
610 		goto noinodes;
611 	UFS_WAPBL_END(pvp->v_mount);
612 	*inop = ino;
613 	return 0;
614 
615 noinodes:
616 	mutex_exit(&ump->um_lock);
617 	UFS_WAPBL_END(pvp->v_mount);
618 	ffs_fserr(fs, cred, "out of inodes");
619 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
620 	return ENOSPC;
621 }
622 
623 /*
624  * Find a cylinder group in which to place a directory.
625  *
626  * The policy implemented by this algorithm is to allocate a
627  * directory inode in the same cylinder group as its parent
628  * directory, but also to reserve space for its files inodes
629  * and data. Restrict the number of directories which may be
630  * allocated one after another in the same cylinder group
631  * without intervening allocation of files.
632  *
633  * If we allocate a first level directory then force allocation
634  * in another cylinder group.
635  */
636 static ino_t
637 ffs_dirpref(struct inode *pip)
638 {
639 	register struct fs *fs;
640 	int cg, prefcg;
641 	int64_t dirsize, cgsize, curdsz;
642 	int avgifree, avgbfree, avgndir;
643 	int minifree, minbfree, maxndir;
644 	int mincg, minndir;
645 	int maxcontigdirs;
646 
647 	KASSERT(mutex_owned(&pip->i_ump->um_lock));
648 
649 	fs = pip->i_fs;
650 
651 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
652 	avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
653 	avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
654 
655 	/*
656 	 * Force allocation in another cg if creating a first level dir.
657 	 */
658 	if (ITOV(pip)->v_vflag & VV_ROOT) {
659 		prefcg = cprng_fast32() % fs->fs_ncg;
660 		mincg = prefcg;
661 		minndir = fs->fs_ipg;
662 		for (cg = prefcg; cg < fs->fs_ncg; cg++)
663 			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
664 			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
665 			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
666 				mincg = cg;
667 				minndir = fs->fs_cs(fs, cg).cs_ndir;
668 			}
669 		for (cg = 0; cg < prefcg; cg++)
670 			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
671 			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
672 			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
673 				mincg = cg;
674 				minndir = fs->fs_cs(fs, cg).cs_ndir;
675 			}
676 		return ((ino_t)(fs->fs_ipg * mincg));
677 	}
678 
679 	/*
680 	 * Count various limits which used for
681 	 * optimal allocation of a directory inode.
682 	 * Try cylinder groups with >75% avgifree and avgbfree.
683 	 * Avoid cylinder groups with no free blocks or inodes as that
684 	 * triggers an I/O-expensive cylinder group scan.
685 	 */
686 	maxndir = uimin(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
687 	minifree = avgifree - avgifree / 4;
688 	if (minifree < 1)
689 		minifree = 1;
690 	minbfree = avgbfree - avgbfree / 4;
691 	if (minbfree < 1)
692 		minbfree = 1;
693 	cgsize = (int64_t)fs->fs_fsize * fs->fs_fpg;
694 	dirsize = (int64_t)fs->fs_avgfilesize * fs->fs_avgfpdir;
695 	if (avgndir != 0) {
696 		curdsz = (cgsize - (int64_t)avgbfree * fs->fs_bsize) / avgndir;
697 		if (dirsize < curdsz)
698 			dirsize = curdsz;
699 	}
700 	if (cgsize < dirsize * 255)
701 		maxcontigdirs = (avgbfree * fs->fs_bsize) / dirsize;
702 	else
703 		maxcontigdirs = 255;
704 	if (fs->fs_avgfpdir > 0)
705 		maxcontigdirs = uimin(maxcontigdirs,
706 				    fs->fs_ipg / fs->fs_avgfpdir);
707 	if (maxcontigdirs == 0)
708 		maxcontigdirs = 1;
709 
710 	/*
711 	 * Limit number of dirs in one cg and reserve space for
712 	 * regular files, but only if we have no deficit in
713 	 * inodes or space.
714 	 */
715 	prefcg = ino_to_cg(fs, pip->i_number);
716 	for (cg = prefcg; cg < fs->fs_ncg; cg++)
717 		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
718 		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
719 	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
720 			if (fs->fs_contigdirs[cg] < maxcontigdirs)
721 				return ((ino_t)(fs->fs_ipg * cg));
722 		}
723 	for (cg = 0; cg < prefcg; cg++)
724 		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
725 		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
726 	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
727 			if (fs->fs_contigdirs[cg] < maxcontigdirs)
728 				return ((ino_t)(fs->fs_ipg * cg));
729 		}
730 	/*
731 	 * This is a backstop when we are deficient in space.
732 	 */
733 	for (cg = prefcg; cg < fs->fs_ncg; cg++)
734 		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
735 			return ((ino_t)(fs->fs_ipg * cg));
736 	for (cg = 0; cg < prefcg; cg++)
737 		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
738 			break;
739 	return ((ino_t)(fs->fs_ipg * cg));
740 }
741 
742 /*
743  * Select the desired position for the next block in a file.  The file is
744  * logically divided into sections. The first section is composed of the
745  * direct blocks. Each additional section contains fs_maxbpg blocks.
746  *
747  * If no blocks have been allocated in the first section, the policy is to
748  * request a block in the same cylinder group as the inode that describes
749  * the file. If no blocks have been allocated in any other section, the
750  * policy is to place the section in a cylinder group with a greater than
751  * average number of free blocks.  An appropriate cylinder group is found
752  * by using a rotor that sweeps the cylinder groups. When a new group of
753  * blocks is needed, the sweep begins in the cylinder group following the
754  * cylinder group from which the previous allocation was made. The sweep
755  * continues until a cylinder group with greater than the average number
756  * of free blocks is found. If the allocation is for the first block in an
757  * indirect block, the information on the previous allocation is unavailable;
758  * here a best guess is made based upon the logical block number being
759  * allocated.
760  *
761  * If a section is already partially allocated, the policy is to
762  * contiguously allocate fs_maxcontig blocks.  The end of one of these
763  * contiguous blocks and the beginning of the next is laid out
764  * contigously if possible.
765  *
766  * => um_lock held on entry and exit
767  */
768 daddr_t
769 ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int flags,
770     int32_t *bap /* XXX ondisk32 */)
771 {
772 	struct fs *fs;
773 	int cg;
774 	int avgbfree, startcg;
775 
776 	KASSERT(mutex_owned(&ip->i_ump->um_lock));
777 
778 	fs = ip->i_fs;
779 
780 	/*
781 	 * If allocating a contiguous file with B_CONTIG, use the hints
782 	 * in the inode extensions to return the desired block.
783 	 *
784 	 * For metadata (indirect blocks) return the address of where
785 	 * the first indirect block resides - we'll scan for the next
786 	 * available slot if we need to allocate more than one indirect
787 	 * block.  For data, return the address of the actual block
788 	 * relative to the address of the first data block.
789 	 */
790 	if (flags & B_CONTIG) {
791 		KASSERT(ip->i_ffs_first_data_blk != 0);
792 		KASSERT(ip->i_ffs_first_indir_blk != 0);
793 		if (flags & B_METAONLY)
794 			return ip->i_ffs_first_indir_blk;
795 		else
796 			return ip->i_ffs_first_data_blk + ffs_blkstofrags(fs, lbn);
797 	}
798 
799 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
800 		if (lbn < UFS_NDADDR + FFS_NINDIR(fs)) {
801 			cg = ino_to_cg(fs, ip->i_number);
802 			return (cgbase(fs, cg) + fs->fs_frag);
803 		}
804 		/*
805 		 * Find a cylinder with greater than average number of
806 		 * unused data blocks.
807 		 */
808 		if (indx == 0 || bap[indx - 1] == 0)
809 			startcg =
810 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
811 		else
812 			startcg = dtog(fs,
813 				ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
814 		startcg %= fs->fs_ncg;
815 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
816 		for (cg = startcg; cg < fs->fs_ncg; cg++)
817 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
818 				return (cgbase(fs, cg) + fs->fs_frag);
819 			}
820 		for (cg = 0; cg < startcg; cg++)
821 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
822 				return (cgbase(fs, cg) + fs->fs_frag);
823 			}
824 		return (0);
825 	}
826 	/*
827 	 * We just always try to lay things out contiguously.
828 	 */
829 	return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
830 }
831 
832 daddr_t
833 ffs_blkpref_ufs2(struct inode *ip, daddr_t lbn, int indx, int flags,
834     int64_t *bap)
835 {
836 	struct fs *fs;
837 	int cg;
838 	int avgbfree, startcg;
839 
840 	KASSERT(mutex_owned(&ip->i_ump->um_lock));
841 
842 	fs = ip->i_fs;
843 
844 	/*
845 	 * If allocating a contiguous file with B_CONTIG, use the hints
846 	 * in the inode extensions to return the desired block.
847 	 *
848 	 * For metadata (indirect blocks) return the address of where
849 	 * the first indirect block resides - we'll scan for the next
850 	 * available slot if we need to allocate more than one indirect
851 	 * block.  For data, return the address of the actual block
852 	 * relative to the address of the first data block.
853 	 */
854 	if (flags & B_CONTIG) {
855 		KASSERT(ip->i_ffs_first_data_blk != 0);
856 		KASSERT(ip->i_ffs_first_indir_blk != 0);
857 		if (flags & B_METAONLY)
858 			return ip->i_ffs_first_indir_blk;
859 		else
860 			return ip->i_ffs_first_data_blk + ffs_blkstofrags(fs, lbn);
861 	}
862 
863 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
864 		if (lbn < UFS_NDADDR + FFS_NINDIR(fs)) {
865 			cg = ino_to_cg(fs, ip->i_number);
866 			return (cgbase(fs, cg) + fs->fs_frag);
867 		}
868 		/*
869 		 * Find a cylinder with greater than average number of
870 		 * unused data blocks.
871 		 */
872 		if (indx == 0 || bap[indx - 1] == 0)
873 			startcg =
874 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
875 		else
876 			startcg = dtog(fs,
877 				ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
878 		startcg %= fs->fs_ncg;
879 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
880 		for (cg = startcg; cg < fs->fs_ncg; cg++)
881 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
882 				return (cgbase(fs, cg) + fs->fs_frag);
883 			}
884 		for (cg = 0; cg < startcg; cg++)
885 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
886 				return (cgbase(fs, cg) + fs->fs_frag);
887 			}
888 		return (0);
889 	}
890 	/*
891 	 * We just always try to lay things out contiguously.
892 	 */
893 	return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
894 }
895 
896 
897 /*
898  * Implement the cylinder overflow algorithm.
899  *
900  * The policy implemented by this algorithm is:
901  *   1) allocate the block in its requested cylinder group.
902  *   2) quadradically rehash on the cylinder group number.
903  *   3) brute force search for a free block.
904  *
905  * => called with um_lock held
906  * => returns with um_lock released on success, held on failure
907  *    (*allocator releases lock on success, retains lock on failure)
908  */
909 /*VARARGS5*/
910 static daddr_t
911 ffs_hashalloc(struct inode *ip, int cg, daddr_t pref,
912     int size /* size for data blocks, mode for inodes */,
913     int realsize,
914     int flags,
915     daddr_t (*allocator)(struct inode *, int, daddr_t, int, int, int))
916 {
917 	struct fs *fs;
918 	daddr_t result;
919 	int i, icg = cg;
920 
921 	fs = ip->i_fs;
922 	/*
923 	 * 1: preferred cylinder group
924 	 */
925 	result = (*allocator)(ip, cg, pref, size, realsize, flags);
926 	if (result)
927 		return (result);
928 
929 	if (flags & B_CONTIG)
930 		return (result);
931 	/*
932 	 * 2: quadratic rehash
933 	 */
934 	for (i = 1; i < fs->fs_ncg; i *= 2) {
935 		cg += i;
936 		if (cg >= fs->fs_ncg)
937 			cg -= fs->fs_ncg;
938 		result = (*allocator)(ip, cg, 0, size, realsize, flags);
939 		if (result)
940 			return (result);
941 	}
942 	/*
943 	 * 3: brute force search
944 	 * Note that we start at i == 2, since 0 was checked initially,
945 	 * and 1 is always checked in the quadratic rehash.
946 	 */
947 	cg = (icg + 2) % fs->fs_ncg;
948 	for (i = 2; i < fs->fs_ncg; i++) {
949 		result = (*allocator)(ip, cg, 0, size, realsize, flags);
950 		if (result)
951 			return (result);
952 		cg++;
953 		if (cg == fs->fs_ncg)
954 			cg = 0;
955 	}
956 	return (0);
957 }
958 
959 /*
960  * Determine whether a fragment can be extended.
961  *
962  * Check to see if the necessary fragments are available, and
963  * if they are, allocate them.
964  *
965  * => called with um_lock held
966  * => returns with um_lock released on success, held on failure
967  */
968 static daddr_t
969 ffs_fragextend(struct inode *ip, int cg, daddr_t bprev, int osize, int nsize)
970 {
971 	struct ufsmount *ump;
972 	struct fs *fs;
973 	struct cg *cgp;
974 	struct buf *bp;
975 	daddr_t bno;
976 	int frags, bbase;
977 	int i, error;
978 	u_int8_t *blksfree;
979 
980 	fs = ip->i_fs;
981 	ump = ip->i_ump;
982 
983 	KASSERT(mutex_owned(&ump->um_lock));
984 
985 	if (fs->fs_cs(fs, cg).cs_nffree < ffs_numfrags(fs, nsize - osize))
986 		return (0);
987 	frags = ffs_numfrags(fs, nsize);
988 	bbase = ffs_fragnum(fs, bprev);
989 	if (bbase > ffs_fragnum(fs, (bprev + frags - 1))) {
990 		/* cannot extend across a block boundary */
991 		return (0);
992 	}
993 	mutex_exit(&ump->um_lock);
994 	error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
995 		(int)fs->fs_cgsize, B_MODIFY, &bp);
996 	if (error)
997 		goto fail;
998 	cgp = (struct cg *)bp->b_data;
999 	if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs)))
1000 		goto fail;
1001 	cgp->cg_old_time = ufs_rw32(time_second, UFS_FSNEEDSWAP(fs));
1002 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
1003 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
1004 		cgp->cg_time = ufs_rw64(time_second, UFS_FSNEEDSWAP(fs));
1005 	bno = dtogd(fs, bprev);
1006 	blksfree = cg_blksfree(cgp, UFS_FSNEEDSWAP(fs));
1007 	for (i = ffs_numfrags(fs, osize); i < frags; i++)
1008 		if (isclr(blksfree, bno + i))
1009 			goto fail;
1010 	/*
1011 	 * the current fragment can be extended
1012 	 * deduct the count on fragment being extended into
1013 	 * increase the count on the remaining fragment (if any)
1014 	 * allocate the extended piece
1015 	 */
1016 	for (i = frags; i < fs->fs_frag - bbase; i++)
1017 		if (isclr(blksfree, bno + i))
1018 			break;
1019 	ufs_add32(cgp->cg_frsum[i - ffs_numfrags(fs, osize)], -1, UFS_FSNEEDSWAP(fs));
1020 	if (i != frags)
1021 		ufs_add32(cgp->cg_frsum[i - frags], 1, UFS_FSNEEDSWAP(fs));
1022 	mutex_enter(&ump->um_lock);
1023 	for (i = ffs_numfrags(fs, osize); i < frags; i++) {
1024 		clrbit(blksfree, bno + i);
1025 		ufs_add32(cgp->cg_cs.cs_nffree, -1, UFS_FSNEEDSWAP(fs));
1026 		fs->fs_cstotal.cs_nffree--;
1027 		fs->fs_cs(fs, cg).cs_nffree--;
1028 	}
1029 	fs->fs_fmod = 1;
1030 	ACTIVECG_CLR(fs, cg);
1031 	mutex_exit(&ump->um_lock);
1032 	bdwrite(bp);
1033 	return (bprev);
1034 
1035  fail:
1036  	if (bp != NULL)
1037 		brelse(bp, 0);
1038  	mutex_enter(&ump->um_lock);
1039  	return (0);
1040 }
1041 
1042 /*
1043  * Determine whether a block can be allocated.
1044  *
1045  * Check to see if a block of the appropriate size is available,
1046  * and if it is, allocate it.
1047  */
1048 static daddr_t
1049 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size, int realsize,
1050     int flags)
1051 {
1052 	struct ufsmount *ump;
1053 	struct fs *fs = ip->i_fs;
1054 	struct cg *cgp;
1055 	struct buf *bp;
1056 	int32_t bno;
1057 	daddr_t blkno;
1058 	int error, frags, allocsiz, i;
1059 	u_int8_t *blksfree;
1060 	const int needswap = UFS_FSNEEDSWAP(fs);
1061 
1062 	ump = ip->i_ump;
1063 
1064 	KASSERT(mutex_owned(&ump->um_lock));
1065 
1066 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
1067 		return (0);
1068 	mutex_exit(&ump->um_lock);
1069 	error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
1070 		(int)fs->fs_cgsize, B_MODIFY, &bp);
1071 	if (error)
1072 		goto fail;
1073 	cgp = (struct cg *)bp->b_data;
1074 	if (!cg_chkmagic(cgp, needswap) ||
1075 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
1076 		goto fail;
1077 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
1078 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
1079 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
1080 		cgp->cg_time = ufs_rw64(time_second, needswap);
1081 	if (size == fs->fs_bsize) {
1082 		mutex_enter(&ump->um_lock);
1083 		blkno = ffs_alloccgblk(ip, bp, bpref, realsize, flags);
1084 		ACTIVECG_CLR(fs, cg);
1085 		mutex_exit(&ump->um_lock);
1086 
1087 		/*
1088 		 * If actually needed size is lower, free the extra blocks now.
1089 		 * This is safe to call here, there is no outside reference
1090 		 * to this block yet. It is not necessary to keep um_lock
1091 		 * locked.
1092 		 */
1093 		if (realsize != 0 && realsize < size) {
1094 			ffs_blkfree_common(ip->i_ump, ip->i_fs,
1095 			    ip->i_devvp->v_rdev,
1096 			    bp, blkno + ffs_numfrags(fs, realsize),
1097 			    (long)(size - realsize), false);
1098 		}
1099 
1100 		bdwrite(bp);
1101 		return (blkno);
1102 	}
1103 	/*
1104 	 * check to see if any fragments are already available
1105 	 * allocsiz is the size which will be allocated, hacking
1106 	 * it down to a smaller size if necessary
1107 	 */
1108 	blksfree = cg_blksfree(cgp, needswap);
1109 	frags = ffs_numfrags(fs, size);
1110 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
1111 		if (cgp->cg_frsum[allocsiz] != 0)
1112 			break;
1113 	if (allocsiz == fs->fs_frag) {
1114 		/*
1115 		 * no fragments were available, so a block will be
1116 		 * allocated, and hacked up
1117 		 */
1118 		if (cgp->cg_cs.cs_nbfree == 0)
1119 			goto fail;
1120 		mutex_enter(&ump->um_lock);
1121 		blkno = ffs_alloccgblk(ip, bp, bpref, realsize, flags);
1122 		bno = dtogd(fs, blkno);
1123 		for (i = frags; i < fs->fs_frag; i++)
1124 			setbit(blksfree, bno + i);
1125 		i = fs->fs_frag - frags;
1126 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
1127 		fs->fs_cstotal.cs_nffree += i;
1128 		fs->fs_cs(fs, cg).cs_nffree += i;
1129 		fs->fs_fmod = 1;
1130 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
1131 		ACTIVECG_CLR(fs, cg);
1132 		mutex_exit(&ump->um_lock);
1133 		bdwrite(bp);
1134 		return (blkno);
1135 	}
1136 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
1137 #if 0
1138 	/*
1139 	 * XXX fvdl mapsearch will panic, and never return -1
1140 	 *          also: returning NULL as daddr_t ?
1141 	 */
1142 	if (bno < 0)
1143 		goto fail;
1144 #endif
1145 	for (i = 0; i < frags; i++)
1146 		clrbit(blksfree, bno + i);
1147 	mutex_enter(&ump->um_lock);
1148 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
1149 	fs->fs_cstotal.cs_nffree -= frags;
1150 	fs->fs_cs(fs, cg).cs_nffree -= frags;
1151 	fs->fs_fmod = 1;
1152 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
1153 	if (frags != allocsiz)
1154 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
1155 	blkno = cgbase(fs, cg) + bno;
1156 	ACTIVECG_CLR(fs, cg);
1157 	mutex_exit(&ump->um_lock);
1158 	bdwrite(bp);
1159 	return blkno;
1160 
1161  fail:
1162  	if (bp != NULL)
1163 		brelse(bp, 0);
1164  	mutex_enter(&ump->um_lock);
1165  	return (0);
1166 }
1167 
1168 /*
1169  * Allocate a block in a cylinder group.
1170  *
1171  * This algorithm implements the following policy:
1172  *   1) allocate the requested block.
1173  *   2) allocate a rotationally optimal block in the same cylinder.
1174  *   3) allocate the next available block on the block rotor for the
1175  *      specified cylinder group.
1176  * Note that this routine only allocates fs_bsize blocks; these
1177  * blocks may be fragmented by the routine that allocates them.
1178  */
1179 static daddr_t
1180 ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref, int realsize,
1181     int flags)
1182 {
1183 	struct fs *fs = ip->i_fs;
1184 	struct cg *cgp;
1185 	int cg;
1186 	daddr_t blkno;
1187 	int32_t bno;
1188 	u_int8_t *blksfree;
1189 	const int needswap = UFS_FSNEEDSWAP(fs);
1190 
1191 	KASSERT(mutex_owned(&ip->i_ump->um_lock));
1192 
1193 	cgp = (struct cg *)bp->b_data;
1194 	blksfree = cg_blksfree(cgp, needswap);
1195 	if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
1196 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
1197 	} else {
1198 		bpref = ffs_blknum(fs, bpref);
1199 		bno = dtogd(fs, bpref);
1200 		/*
1201 		 * if the requested block is available, use it
1202 		 */
1203 		if (ffs_isblock(fs, blksfree, ffs_fragstoblks(fs, bno)))
1204 			goto gotit;
1205 		/*
1206 		 * if the requested data block isn't available and we are
1207 		 * trying to allocate a contiguous file, return an error.
1208 		 */
1209 		if ((flags & (B_CONTIG | B_METAONLY)) == B_CONTIG)
1210 			return (0);
1211 	}
1212 
1213 	/*
1214 	 * Take the next available block in this cylinder group.
1215 	 */
1216 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1217 #if 0
1218 	/*
1219 	 * XXX jdolecek ffs_mapsearch() succeeds or panics
1220 	 */
1221 	if (bno < 0)
1222 		return (0);
1223 #endif
1224 	cgp->cg_rotor = ufs_rw32(bno, needswap);
1225 gotit:
1226 	blkno = ffs_fragstoblks(fs, bno);
1227 	ffs_clrblock(fs, blksfree, blkno);
1228 	ffs_clusteracct(fs, cgp, blkno, -1);
1229 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
1230 	fs->fs_cstotal.cs_nbfree--;
1231 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
1232 	if ((fs->fs_magic == FS_UFS1_MAGIC) &&
1233 	    ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
1234 		int cylno;
1235 		cylno = old_cbtocylno(fs, bno);
1236 		KASSERT(cylno >= 0);
1237 		KASSERT(cylno < fs->fs_old_ncyl);
1238 		KASSERT(old_cbtorpos(fs, bno) >= 0);
1239 		KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bno) < fs->fs_old_nrpos);
1240 		ufs_add16(old_cg_blks(fs, cgp, cylno, needswap)[old_cbtorpos(fs, bno)], -1,
1241 		    needswap);
1242 		ufs_add32(old_cg_blktot(cgp, needswap)[cylno], -1, needswap);
1243 	}
1244 	fs->fs_fmod = 1;
1245 	cg = ufs_rw32(cgp->cg_cgx, needswap);
1246 	blkno = cgbase(fs, cg) + bno;
1247 	return (blkno);
1248 }
1249 
1250 /*
1251  * Determine whether an inode can be allocated.
1252  *
1253  * Check to see if an inode is available, and if it is,
1254  * allocate it using the following policy:
1255  *   1) allocate the requested inode.
1256  *   2) allocate the next available inode after the requested
1257  *      inode in the specified cylinder group.
1258  */
1259 static daddr_t
1260 ffs_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode, int realsize,
1261     int flags)
1262 {
1263 	struct ufsmount *ump = ip->i_ump;
1264 	struct fs *fs = ip->i_fs;
1265 	struct cg *cgp;
1266 	struct buf *bp, *ibp;
1267 	u_int8_t *inosused;
1268 	int error, start, len, loc, map, i;
1269 	int32_t initediblk, maxiblk, irotor;
1270 	daddr_t nalloc;
1271 	struct ufs2_dinode *dp2;
1272 	const int needswap = UFS_FSNEEDSWAP(fs);
1273 
1274 	KASSERT(mutex_owned(&ump->um_lock));
1275 	UFS_WAPBL_JLOCK_ASSERT(ip->i_ump->um_mountp);
1276 
1277 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
1278 		return (0);
1279 	mutex_exit(&ump->um_lock);
1280 	ibp = NULL;
1281 	if (fs->fs_magic == FS_UFS2_MAGIC) {
1282 		initediblk = -1;
1283 	} else {
1284 		initediblk = fs->fs_ipg;
1285 	}
1286 	maxiblk = initediblk;
1287 
1288 retry:
1289 	error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
1290 		(int)fs->fs_cgsize, B_MODIFY, &bp);
1291 	if (error)
1292 		goto fail;
1293 	cgp = (struct cg *)bp->b_data;
1294 	if (!cg_chkmagic(cgp, needswap) || cgp->cg_cs.cs_nifree == 0)
1295 		goto fail;
1296 
1297 	if (ibp != NULL &&
1298 	    initediblk != ufs_rw32(cgp->cg_initediblk, needswap)) {
1299 		/* Another thread allocated more inodes so we retry the test. */
1300 		brelse(ibp, 0);
1301 		ibp = NULL;
1302 	}
1303 	/*
1304 	 * Check to see if we need to initialize more inodes.
1305 	 */
1306 	if (fs->fs_magic == FS_UFS2_MAGIC && ibp == NULL) {
1307 	        initediblk = ufs_rw32(cgp->cg_initediblk, needswap);
1308 		maxiblk = initediblk;
1309 		nalloc = fs->fs_ipg - ufs_rw32(cgp->cg_cs.cs_nifree, needswap);
1310 		if (nalloc + FFS_INOPB(fs) > initediblk &&
1311 		    initediblk < ufs_rw32(cgp->cg_niblk, needswap)) {
1312 			/*
1313 			 * We have to release the cg buffer here to prevent
1314 			 * a deadlock when reading the inode block will
1315 			 * run a copy-on-write that might use this cg.
1316 			 */
1317 			brelse(bp, 0);
1318 			bp = NULL;
1319 			error = ffs_getblk(ip->i_devvp, FFS_FSBTODB(fs,
1320 			    ino_to_fsba(fs, cg * fs->fs_ipg + initediblk)),
1321 			    FFS_NOBLK, fs->fs_bsize, false, &ibp);
1322 			if (error)
1323 				goto fail;
1324 
1325 			maxiblk += FFS_INOPB(fs);
1326 
1327 			goto retry;
1328 		}
1329 	}
1330 
1331 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
1332 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
1333 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
1334 		cgp->cg_time = ufs_rw64(time_second, needswap);
1335 	inosused = cg_inosused(cgp, needswap);
1336 
1337 	if (ipref) {
1338 		ipref %= fs->fs_ipg;
1339 		/* safeguard to stay in (to be) allocated range */
1340 		if (ipref < maxiblk && isclr(inosused, ipref))
1341 			goto gotit;
1342 	}
1343 
1344 	irotor = ufs_rw32(cgp->cg_irotor, needswap);
1345 
1346 	KASSERTMSG(irotor < initediblk, "%s: allocation botch: cg=%d, irotor %d"
1347 		   " out of bounds, initediblk=%d",
1348 		   __func__, cg, irotor, initediblk);
1349 
1350 	start = irotor / NBBY;
1351 	len = howmany(maxiblk - irotor, NBBY);
1352 	loc = skpc(0xff, len, &inosused[start]);
1353 	if (loc == 0) {
1354 		len = start + 1;
1355 		start = 0;
1356 		loc = skpc(0xff, len, &inosused[0]);
1357 		if (loc == 0) {
1358 			panic("%s: map corrupted: cg=%d, irotor=%d, fs=%s",
1359 			    __func__, cg, ufs_rw32(cgp->cg_irotor, needswap),
1360 			    fs->fs_fsmnt);
1361 			/* NOTREACHED */
1362 		}
1363 	}
1364 	i = start + len - loc;
1365 	map = inosused[i] ^ 0xff;
1366 	if (map == 0) {
1367 		panic("%s: block not in map: fs=%s", __func__, fs->fs_fsmnt);
1368 	}
1369 
1370 	ipref = i * NBBY + ffs(map) - 1;
1371 
1372 	cgp->cg_irotor = ufs_rw32(ipref, needswap);
1373 
1374 gotit:
1375 	KASSERTMSG(ipref < maxiblk, "%s: allocation botch: cg=%d attempt to "
1376 		   "allocate inode index %d beyond max allocated index %d"
1377 		   " of %d inodes/cg",
1378 		   __func__, cg, (int)ipref, maxiblk, cgp->cg_niblk);
1379 
1380 	UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp, cg * fs->fs_ipg + ipref,
1381 	    mode);
1382 	/*
1383 	 * Check to see if we need to initialize more inodes.
1384 	 */
1385 	if (ibp != NULL) {
1386 		KASSERT(initediblk == ufs_rw32(cgp->cg_initediblk, needswap));
1387 		memset(ibp->b_data, 0, fs->fs_bsize);
1388 		dp2 = (struct ufs2_dinode *)(ibp->b_data);
1389 		for (i = 0; i < FFS_INOPB(fs); i++) {
1390 			/*
1391 			 * Don't bother to swap, it's supposed to be
1392 			 * random, after all.
1393 			 */
1394 			dp2->di_gen = (cprng_fast32() & INT32_MAX) / 2 + 1;
1395 			dp2++;
1396 		}
1397 		initediblk += FFS_INOPB(fs);
1398 		cgp->cg_initediblk = ufs_rw32(initediblk, needswap);
1399 	}
1400 
1401 	mutex_enter(&ump->um_lock);
1402 	ACTIVECG_CLR(fs, cg);
1403 	setbit(inosused, ipref);
1404 	ufs_add32(cgp->cg_cs.cs_nifree, -1, needswap);
1405 	fs->fs_cstotal.cs_nifree--;
1406 	fs->fs_cs(fs, cg).cs_nifree--;
1407 	fs->fs_fmod = 1;
1408 	if ((mode & IFMT) == IFDIR) {
1409 		ufs_add32(cgp->cg_cs.cs_ndir, 1, needswap);
1410 		fs->fs_cstotal.cs_ndir++;
1411 		fs->fs_cs(fs, cg).cs_ndir++;
1412 	}
1413 	mutex_exit(&ump->um_lock);
1414 	if (ibp != NULL) {
1415 		bwrite(ibp);
1416 		bwrite(bp);
1417 	} else
1418 		bdwrite(bp);
1419 	return (cg * fs->fs_ipg + ipref);
1420  fail:
1421 	if (bp != NULL)
1422 		brelse(bp, 0);
1423 	if (ibp != NULL)
1424 		brelse(ibp, 0);
1425 	mutex_enter(&ump->um_lock);
1426 	return (0);
1427 }
1428 
1429 /*
1430  * Allocate a block or fragment.
1431  *
1432  * The specified block or fragment is removed from the
1433  * free map, possibly fragmenting a block in the process.
1434  *
1435  * This implementation should mirror fs_blkfree
1436  *
1437  * => um_lock not held on entry or exit
1438  */
1439 int
1440 ffs_blkalloc(struct inode *ip, daddr_t bno, long size)
1441 {
1442 	int error;
1443 
1444 	error = ffs_check_bad_allocation(__func__, ip->i_fs, bno, size,
1445 	    ip->i_dev, ip->i_uid);
1446 	if (error)
1447 		return error;
1448 
1449 	return ffs_blkalloc_ump(ip->i_ump, bno, size);
1450 }
1451 
1452 int
1453 ffs_blkalloc_ump(struct ufsmount *ump, daddr_t bno, long size)
1454 {
1455 	struct fs *fs = ump->um_fs;
1456 	struct cg *cgp;
1457 	struct buf *bp;
1458 	int32_t fragno, cgbno;
1459 	int i, error, cg, blk, frags, bbase;
1460 	u_int8_t *blksfree;
1461 	const int needswap = UFS_FSNEEDSWAP(fs);
1462 
1463 	KASSERT((u_int)size <= fs->fs_bsize && ffs_fragoff(fs, size) == 0 &&
1464 	    ffs_fragnum(fs, bno) + ffs_numfrags(fs, size) <= fs->fs_frag);
1465 	KASSERT(bno < fs->fs_size);
1466 
1467 	cg = dtog(fs, bno);
1468 	error = bread(ump->um_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
1469 		(int)fs->fs_cgsize, B_MODIFY, &bp);
1470 	if (error) {
1471 		return error;
1472 	}
1473 	cgp = (struct cg *)bp->b_data;
1474 	if (!cg_chkmagic(cgp, needswap)) {
1475 		brelse(bp, 0);
1476 		return EIO;
1477 	}
1478 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
1479 	cgp->cg_time = ufs_rw64(time_second, needswap);
1480 	cgbno = dtogd(fs, bno);
1481 	blksfree = cg_blksfree(cgp, needswap);
1482 
1483 	mutex_enter(&ump->um_lock);
1484 	if (size == fs->fs_bsize) {
1485 		fragno = ffs_fragstoblks(fs, cgbno);
1486 		if (!ffs_isblock(fs, blksfree, fragno)) {
1487 			mutex_exit(&ump->um_lock);
1488 			brelse(bp, 0);
1489 			return EBUSY;
1490 		}
1491 		ffs_clrblock(fs, blksfree, fragno);
1492 		ffs_clusteracct(fs, cgp, fragno, -1);
1493 		ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
1494 		fs->fs_cstotal.cs_nbfree--;
1495 		fs->fs_cs(fs, cg).cs_nbfree--;
1496 	} else {
1497 		bbase = cgbno - ffs_fragnum(fs, cgbno);
1498 
1499 		frags = ffs_numfrags(fs, size);
1500 		for (i = 0; i < frags; i++) {
1501 			if (isclr(blksfree, cgbno + i)) {
1502 				mutex_exit(&ump->um_lock);
1503 				brelse(bp, 0);
1504 				return EBUSY;
1505 			}
1506 		}
1507 		/*
1508 		 * if a complete block is being split, account for it
1509 		 */
1510 		fragno = ffs_fragstoblks(fs, bbase);
1511 		if (ffs_isblock(fs, blksfree, fragno)) {
1512 			ufs_add32(cgp->cg_cs.cs_nffree, fs->fs_frag, needswap);
1513 			fs->fs_cstotal.cs_nffree += fs->fs_frag;
1514 			fs->fs_cs(fs, cg).cs_nffree += fs->fs_frag;
1515 			ffs_clusteracct(fs, cgp, fragno, -1);
1516 			ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
1517 			fs->fs_cstotal.cs_nbfree--;
1518 			fs->fs_cs(fs, cg).cs_nbfree--;
1519 		}
1520 		/*
1521 		 * decrement the counts associated with the old frags
1522 		 */
1523 		blk = blkmap(fs, blksfree, bbase);
1524 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
1525 		/*
1526 		 * allocate the fragment
1527 		 */
1528 		for (i = 0; i < frags; i++) {
1529 			clrbit(blksfree, cgbno + i);
1530 		}
1531 		ufs_add32(cgp->cg_cs.cs_nffree, -i, needswap);
1532 		fs->fs_cstotal.cs_nffree -= i;
1533 		fs->fs_cs(fs, cg).cs_nffree -= i;
1534 		/*
1535 		 * add back in counts associated with the new frags
1536 		 */
1537 		blk = blkmap(fs, blksfree, bbase);
1538 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
1539 	}
1540 	fs->fs_fmod = 1;
1541 	ACTIVECG_CLR(fs, cg);
1542 	mutex_exit(&ump->um_lock);
1543 	bdwrite(bp);
1544 	return 0;
1545 }
1546 
1547 /*
1548  * Free a block or fragment.
1549  *
1550  * The specified block or fragment is placed back in the
1551  * free map. If a fragment is deallocated, a possible
1552  * block reassembly is checked.
1553  *
1554  * => um_lock not held on entry or exit
1555  */
1556 static void
1557 ffs_blkfree_cg(struct fs *fs, struct vnode *devvp, daddr_t bno, long size)
1558 {
1559 	struct cg *cgp;
1560 	struct buf *bp;
1561 	struct ufsmount *ump;
1562 	daddr_t cgblkno;
1563 	int error, cg;
1564 	dev_t dev;
1565 	const bool devvp_is_snapshot = (devvp->v_type != VBLK);
1566 	const int needswap = UFS_FSNEEDSWAP(fs);
1567 
1568 	KASSERT(!devvp_is_snapshot);
1569 
1570 	cg = dtog(fs, bno);
1571 	dev = devvp->v_rdev;
1572 	ump = VFSTOUFS(spec_node_getmountedfs(devvp));
1573 	KASSERT(fs == ump->um_fs);
1574 	cgblkno = FFS_FSBTODB(fs, cgtod(fs, cg));
1575 
1576 	error = bread(devvp, cgblkno, (int)fs->fs_cgsize,
1577 	    B_MODIFY, &bp);
1578 	if (error) {
1579 		return;
1580 	}
1581 	cgp = (struct cg *)bp->b_data;
1582 	if (!cg_chkmagic(cgp, needswap)) {
1583 		brelse(bp, 0);
1584 		return;
1585 	}
1586 
1587 	ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot);
1588 
1589 	bdwrite(bp);
1590 }
1591 
1592 struct discardopdata {
1593 	struct work wk; /* must be first */
1594 	struct vnode *devvp;
1595 	daddr_t bno;
1596 	long size;
1597 };
1598 
1599 struct discarddata {
1600 	struct fs *fs;
1601 	struct discardopdata *entry;
1602 	long maxsize;
1603 	kmutex_t entrylk;
1604 	struct workqueue *wq;
1605 	int wqcnt, wqdraining;
1606 	kmutex_t wqlk;
1607 	kcondvar_t wqcv;
1608 	/* timer for flush? */
1609 };
1610 
1611 static void
1612 ffs_blkfree_td(struct fs *fs, struct discardopdata *td)
1613 {
1614 	struct mount *mp = spec_node_getmountedfs(td->devvp);
1615 	long todo;
1616 	int error;
1617 
1618 	while (td->size) {
1619 		todo = uimin(td->size,
1620 		  ffs_lfragtosize(fs, (fs->fs_frag - ffs_fragnum(fs, td->bno))));
1621 		error = UFS_WAPBL_BEGIN(mp);
1622 		if (error) {
1623 			printf("ffs: failed to begin wapbl transaction"
1624 			    " for discard: %d\n", error);
1625 			break;
1626 		}
1627 		ffs_blkfree_cg(fs, td->devvp, td->bno, todo);
1628 		UFS_WAPBL_END(mp);
1629 		td->bno += ffs_numfrags(fs, todo);
1630 		td->size -= todo;
1631 	}
1632 }
1633 
1634 static void
1635 ffs_discardcb(struct work *wk, void *arg)
1636 {
1637 	struct discardopdata *td = (void *)wk;
1638 	struct discarddata *ts = arg;
1639 	struct fs *fs = ts->fs;
1640 	off_t start, len;
1641 #ifdef TRIMDEBUG
1642 	int error;
1643 #endif
1644 
1645 /* like FSBTODB but emits bytes; XXX move to fs.h */
1646 #ifndef FFS_FSBTOBYTES
1647 #define FFS_FSBTOBYTES(fs, b) ((b) << (fs)->fs_fshift)
1648 #endif
1649 
1650 	start = FFS_FSBTOBYTES(fs, td->bno);
1651 	len = td->size;
1652 #ifdef TRIMDEBUG
1653 	error =
1654 #endif
1655 		VOP_FDISCARD(td->devvp, start, len);
1656 #ifdef TRIMDEBUG
1657 	printf("trim(%" PRId64 ",%ld):%d\n", td->bno, td->size, error);
1658 #endif
1659 
1660 	ffs_blkfree_td(fs, td);
1661 	kmem_free(td, sizeof(*td));
1662 	mutex_enter(&ts->wqlk);
1663 	ts->wqcnt--;
1664 	if (ts->wqdraining && !ts->wqcnt)
1665 		cv_signal(&ts->wqcv);
1666 	mutex_exit(&ts->wqlk);
1667 }
1668 
1669 void *
1670 ffs_discard_init(struct vnode *devvp, struct fs *fs)
1671 {
1672 	struct discarddata *ts;
1673 	int error;
1674 
1675 	ts = kmem_zalloc(sizeof (*ts), KM_SLEEP);
1676 	error = workqueue_create(&ts->wq, "trimwq", ffs_discardcb, ts,
1677 				 PRI_USER, IPL_NONE, 0);
1678 	if (error) {
1679 		kmem_free(ts, sizeof (*ts));
1680 		return NULL;
1681 	}
1682 	mutex_init(&ts->entrylk, MUTEX_DEFAULT, IPL_NONE);
1683 	mutex_init(&ts->wqlk, MUTEX_DEFAULT, IPL_NONE);
1684 	cv_init(&ts->wqcv, "trimwqcv");
1685 	ts->maxsize = 100*1024; /* XXX */
1686 	ts->fs = fs;
1687 	return ts;
1688 }
1689 
1690 void
1691 ffs_discard_finish(void *vts, int flags)
1692 {
1693 	struct discarddata *ts = vts;
1694 	struct discardopdata *td = NULL;
1695 
1696 	/* wait for workqueue to drain */
1697 	mutex_enter(&ts->wqlk);
1698 	if (ts->wqcnt) {
1699 		ts->wqdraining = 1;
1700 		cv_wait(&ts->wqcv, &ts->wqlk);
1701 	}
1702 	mutex_exit(&ts->wqlk);
1703 
1704 	mutex_enter(&ts->entrylk);
1705 	if (ts->entry) {
1706 		td = ts->entry;
1707 		ts->entry = NULL;
1708 	}
1709 	mutex_exit(&ts->entrylk);
1710 	if (td) {
1711 		/* XXX don't tell disk, its optional */
1712 		ffs_blkfree_td(ts->fs, td);
1713 #ifdef TRIMDEBUG
1714 		printf("finish(%" PRId64 ",%ld)\n", td->bno, td->size);
1715 #endif
1716 		kmem_free(td, sizeof(*td));
1717 	}
1718 
1719 	cv_destroy(&ts->wqcv);
1720 	mutex_destroy(&ts->entrylk);
1721 	mutex_destroy(&ts->wqlk);
1722 	workqueue_destroy(ts->wq);
1723 	kmem_free(ts, sizeof(*ts));
1724 }
1725 
1726 void
1727 ffs_blkfree(struct fs *fs, struct vnode *devvp, daddr_t bno, long size,
1728     ino_t inum)
1729 {
1730 	struct ufsmount *ump;
1731 	int error;
1732 	dev_t dev;
1733 	struct discarddata *ts;
1734 	struct discardopdata *td;
1735 
1736 	dev = devvp->v_rdev;
1737 	ump = VFSTOUFS(spec_node_getmountedfs(devvp));
1738 	if (ffs_snapblkfree(fs, devvp, bno, size, inum))
1739 		return;
1740 
1741 	error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum);
1742 	if (error)
1743 		return;
1744 
1745 	if (!ump->um_discarddata) {
1746 		ffs_blkfree_cg(fs, devvp, bno, size);
1747 		return;
1748 	}
1749 
1750 #ifdef TRIMDEBUG
1751 	printf("blkfree(%" PRId64 ",%ld)\n", bno, size);
1752 #endif
1753 	ts = ump->um_discarddata;
1754 	td = NULL;
1755 
1756 	mutex_enter(&ts->entrylk);
1757 	if (ts->entry) {
1758 		td = ts->entry;
1759 		/* ffs deallocs backwards, check for prepend only */
1760 		if (td->bno == bno + ffs_numfrags(fs, size)
1761 		    && td->size + size <= ts->maxsize) {
1762 			td->bno = bno;
1763 			td->size += size;
1764 			if (td->size < ts->maxsize) {
1765 #ifdef TRIMDEBUG
1766 				printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size);
1767 #endif
1768 				mutex_exit(&ts->entrylk);
1769 				return;
1770 			}
1771 			size = 0; /* mark done */
1772 		}
1773 		ts->entry = NULL;
1774 	}
1775 	mutex_exit(&ts->entrylk);
1776 
1777 	if (td) {
1778 #ifdef TRIMDEBUG
1779 		printf("enq old(%" PRId64 ",%ld)\n", td->bno, td->size);
1780 #endif
1781 		mutex_enter(&ts->wqlk);
1782 		ts->wqcnt++;
1783 		mutex_exit(&ts->wqlk);
1784 		workqueue_enqueue(ts->wq, &td->wk, NULL);
1785 	}
1786 	if (!size)
1787 		return;
1788 
1789 	td = kmem_alloc(sizeof(*td), KM_SLEEP);
1790 	td->devvp = devvp;
1791 	td->bno = bno;
1792 	td->size = size;
1793 
1794 	if (td->size < ts->maxsize) { /* XXX always the case */
1795 		mutex_enter(&ts->entrylk);
1796 		if (!ts->entry) { /* possible race? */
1797 #ifdef TRIMDEBUG
1798 			printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size);
1799 #endif
1800 			ts->entry = td;
1801 			td = NULL;
1802 		}
1803 		mutex_exit(&ts->entrylk);
1804 	}
1805 	if (td) {
1806 #ifdef TRIMDEBUG
1807 		printf("enq new(%" PRId64 ",%ld)\n", td->bno, td->size);
1808 #endif
1809 		mutex_enter(&ts->wqlk);
1810 		ts->wqcnt++;
1811 		mutex_exit(&ts->wqlk);
1812 		workqueue_enqueue(ts->wq, &td->wk, NULL);
1813 	}
1814 }
1815 
1816 /*
1817  * Free a block or fragment from a snapshot cg copy.
1818  *
1819  * The specified block or fragment is placed back in the
1820  * free map. If a fragment is deallocated, a possible
1821  * block reassembly is checked.
1822  *
1823  * => um_lock not held on entry or exit
1824  */
1825 void
1826 ffs_blkfree_snap(struct fs *fs, struct vnode *devvp, daddr_t bno, long size,
1827     ino_t inum)
1828 {
1829 	struct cg *cgp;
1830 	struct buf *bp;
1831 	struct ufsmount *ump;
1832 	daddr_t cgblkno;
1833 	int error, cg;
1834 	dev_t dev;
1835 	const bool devvp_is_snapshot = (devvp->v_type != VBLK);
1836 	const int needswap = UFS_FSNEEDSWAP(fs);
1837 
1838 	KASSERT(devvp_is_snapshot);
1839 
1840 	cg = dtog(fs, bno);
1841 	dev = VTOI(devvp)->i_devvp->v_rdev;
1842 	ump = VFSTOUFS(devvp->v_mount);
1843 	cgblkno = ffs_fragstoblks(fs, cgtod(fs, cg));
1844 
1845 	error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum);
1846 	if (error)
1847 		return;
1848 
1849 	error = bread(devvp, cgblkno, (int)fs->fs_cgsize,
1850 	    B_MODIFY, &bp);
1851 	if (error) {
1852 		return;
1853 	}
1854 	cgp = (struct cg *)bp->b_data;
1855 	if (!cg_chkmagic(cgp, needswap)) {
1856 		brelse(bp, 0);
1857 		return;
1858 	}
1859 
1860 	ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot);
1861 
1862 	bdwrite(bp);
1863 }
1864 
1865 static void
1866 ffs_blkfree_common(struct ufsmount *ump, struct fs *fs, dev_t dev,
1867     struct buf *bp, daddr_t bno, long size, bool devvp_is_snapshot)
1868 {
1869 	struct cg *cgp;
1870 	int32_t fragno, cgbno;
1871 	int i, cg, blk, frags, bbase;
1872 	u_int8_t *blksfree;
1873 	const int needswap = UFS_FSNEEDSWAP(fs);
1874 
1875 	cg = dtog(fs, bno);
1876 	cgp = (struct cg *)bp->b_data;
1877 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
1878 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
1879 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
1880 		cgp->cg_time = ufs_rw64(time_second, needswap);
1881 	cgbno = dtogd(fs, bno);
1882 	blksfree = cg_blksfree(cgp, needswap);
1883 	mutex_enter(&ump->um_lock);
1884 	if (size == fs->fs_bsize) {
1885 		fragno = ffs_fragstoblks(fs, cgbno);
1886 		if (!ffs_isfreeblock(fs, blksfree, fragno)) {
1887 			if (devvp_is_snapshot) {
1888 				mutex_exit(&ump->um_lock);
1889 				return;
1890 			}
1891 			panic("%s: freeing free block: dev = 0x%llx, block = %"
1892 			    PRId64 ", fs = %s", __func__,
1893 			    (unsigned long long)dev, bno, fs->fs_fsmnt);
1894 		}
1895 		ffs_setblock(fs, blksfree, fragno);
1896 		ffs_clusteracct(fs, cgp, fragno, 1);
1897 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
1898 		fs->fs_cstotal.cs_nbfree++;
1899 		fs->fs_cs(fs, cg).cs_nbfree++;
1900 		if ((fs->fs_magic == FS_UFS1_MAGIC) &&
1901 		    ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
1902 			i = old_cbtocylno(fs, cgbno);
1903 			KASSERT(i >= 0);
1904 			KASSERT(i < fs->fs_old_ncyl);
1905 			KASSERT(old_cbtorpos(fs, cgbno) >= 0);
1906 			KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, cgbno) < fs->fs_old_nrpos);
1907 			ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs, cgbno)], 1,
1908 			    needswap);
1909 			ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap);
1910 		}
1911 	} else {
1912 		bbase = cgbno - ffs_fragnum(fs, cgbno);
1913 		/*
1914 		 * decrement the counts associated with the old frags
1915 		 */
1916 		blk = blkmap(fs, blksfree, bbase);
1917 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
1918 		/*
1919 		 * deallocate the fragment
1920 		 */
1921 		frags = ffs_numfrags(fs, size);
1922 		for (i = 0; i < frags; i++) {
1923 			if (isset(blksfree, cgbno + i)) {
1924 				panic("%s: freeing free frag: "
1925 				    "dev = 0x%llx, block = %" PRId64
1926 				    ", fs = %s", __func__,
1927 				    (unsigned long long)dev, bno + i,
1928 				    fs->fs_fsmnt);
1929 			}
1930 			setbit(blksfree, cgbno + i);
1931 		}
1932 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
1933 		fs->fs_cstotal.cs_nffree += i;
1934 		fs->fs_cs(fs, cg).cs_nffree += i;
1935 		/*
1936 		 * add back in counts associated with the new frags
1937 		 */
1938 		blk = blkmap(fs, blksfree, bbase);
1939 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
1940 		/*
1941 		 * if a complete block has been reassembled, account for it
1942 		 */
1943 		fragno = ffs_fragstoblks(fs, bbase);
1944 		if (ffs_isblock(fs, blksfree, fragno)) {
1945 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
1946 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1947 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1948 			ffs_clusteracct(fs, cgp, fragno, 1);
1949 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
1950 			fs->fs_cstotal.cs_nbfree++;
1951 			fs->fs_cs(fs, cg).cs_nbfree++;
1952 			if ((fs->fs_magic == FS_UFS1_MAGIC) &&
1953 			    ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
1954 				i = old_cbtocylno(fs, bbase);
1955 				KASSERT(i >= 0);
1956 				KASSERT(i < fs->fs_old_ncyl);
1957 				KASSERT(old_cbtorpos(fs, bbase) >= 0);
1958 				KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bbase) < fs->fs_old_nrpos);
1959 				ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs,
1960 				    bbase)], 1, needswap);
1961 				ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap);
1962 			}
1963 		}
1964 	}
1965 	fs->fs_fmod = 1;
1966 	ACTIVECG_CLR(fs, cg);
1967 	mutex_exit(&ump->um_lock);
1968 }
1969 
1970 /*
1971  * Free an inode.
1972  */
1973 int
1974 ffs_vfree(struct vnode *vp, ino_t ino, int mode)
1975 {
1976 
1977 	return ffs_freefile(vp->v_mount, ino, mode);
1978 }
1979 
1980 /*
1981  * Do the actual free operation.
1982  * The specified inode is placed back in the free map.
1983  *
1984  * => um_lock not held on entry or exit
1985  */
1986 int
1987 ffs_freefile(struct mount *mp, ino_t ino, int mode)
1988 {
1989 	struct ufsmount *ump = VFSTOUFS(mp);
1990 	struct fs *fs = ump->um_fs;
1991 	struct vnode *devvp;
1992 	struct cg *cgp;
1993 	struct buf *bp;
1994 	int error, cg;
1995 	daddr_t cgbno;
1996 	dev_t dev;
1997 	const int needswap = UFS_FSNEEDSWAP(fs);
1998 
1999 	cg = ino_to_cg(fs, ino);
2000 	devvp = ump->um_devvp;
2001 	dev = devvp->v_rdev;
2002 	cgbno = FFS_FSBTODB(fs, cgtod(fs, cg));
2003 
2004 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
2005 		panic("%s: range: dev = 0x%llx, ino = %llu, fs = %s", __func__,
2006 		    (long long)dev, (unsigned long long)ino, fs->fs_fsmnt);
2007 	error = bread(devvp, cgbno, (int)fs->fs_cgsize,
2008 	    B_MODIFY, &bp);
2009 	if (error) {
2010 		return (error);
2011 	}
2012 	cgp = (struct cg *)bp->b_data;
2013 	if (!cg_chkmagic(cgp, needswap)) {
2014 		brelse(bp, 0);
2015 		return (0);
2016 	}
2017 
2018 	ffs_freefile_common(ump, fs, dev, bp, ino, mode, false);
2019 
2020 	bdwrite(bp);
2021 
2022 	return 0;
2023 }
2024 
2025 int
2026 ffs_freefile_snap(struct fs *fs, struct vnode *devvp, ino_t ino, int mode)
2027 {
2028 	struct ufsmount *ump;
2029 	struct cg *cgp;
2030 	struct buf *bp;
2031 	int error, cg;
2032 	daddr_t cgbno;
2033 	dev_t dev;
2034 	const int needswap = UFS_FSNEEDSWAP(fs);
2035 
2036 	KASSERT(devvp->v_type != VBLK);
2037 
2038 	cg = ino_to_cg(fs, ino);
2039 	dev = VTOI(devvp)->i_devvp->v_rdev;
2040 	ump = VFSTOUFS(devvp->v_mount);
2041 	cgbno = ffs_fragstoblks(fs, cgtod(fs, cg));
2042 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
2043 		panic("%s: range: dev = 0x%llx, ino = %llu, fs = %s", __func__,
2044 		    (unsigned long long)dev, (unsigned long long)ino,
2045 		    fs->fs_fsmnt);
2046 	error = bread(devvp, cgbno, (int)fs->fs_cgsize,
2047 	    B_MODIFY, &bp);
2048 	if (error) {
2049 		return (error);
2050 	}
2051 	cgp = (struct cg *)bp->b_data;
2052 	if (!cg_chkmagic(cgp, needswap)) {
2053 		brelse(bp, 0);
2054 		return (0);
2055 	}
2056 	ffs_freefile_common(ump, fs, dev, bp, ino, mode, true);
2057 
2058 	bdwrite(bp);
2059 
2060 	return 0;
2061 }
2062 
2063 static void
2064 ffs_freefile_common(struct ufsmount *ump, struct fs *fs, dev_t dev,
2065     struct buf *bp, ino_t ino, int mode, bool devvp_is_snapshot)
2066 {
2067 	int cg;
2068 	struct cg *cgp;
2069 	u_int8_t *inosused;
2070 	const int needswap = UFS_FSNEEDSWAP(fs);
2071 
2072 	cg = ino_to_cg(fs, ino);
2073 	cgp = (struct cg *)bp->b_data;
2074 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
2075 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
2076 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
2077 		cgp->cg_time = ufs_rw64(time_second, needswap);
2078 	inosused = cg_inosused(cgp, needswap);
2079 	ino %= fs->fs_ipg;
2080 	if (isclr(inosused, ino)) {
2081 		printf("ifree: dev = 0x%llx, ino = %llu, fs = %s\n",
2082 		    (unsigned long long)dev, (unsigned long long)ino +
2083 		    cg * fs->fs_ipg, fs->fs_fsmnt);
2084 		if (fs->fs_ronly == 0)
2085 			panic("%s: freeing free inode", __func__);
2086 	}
2087 	clrbit(inosused, ino);
2088 	if (!devvp_is_snapshot)
2089 		UFS_WAPBL_UNREGISTER_INODE(ump->um_mountp,
2090 		    ino + cg * fs->fs_ipg, mode);
2091 	if (ino < ufs_rw32(cgp->cg_irotor, needswap))
2092 		cgp->cg_irotor = ufs_rw32(ino, needswap);
2093 	ufs_add32(cgp->cg_cs.cs_nifree, 1, needswap);
2094 	mutex_enter(&ump->um_lock);
2095 	fs->fs_cstotal.cs_nifree++;
2096 	fs->fs_cs(fs, cg).cs_nifree++;
2097 	if ((mode & IFMT) == IFDIR) {
2098 		ufs_add32(cgp->cg_cs.cs_ndir, -1, needswap);
2099 		fs->fs_cstotal.cs_ndir--;
2100 		fs->fs_cs(fs, cg).cs_ndir--;
2101 	}
2102 	fs->fs_fmod = 1;
2103 	ACTIVECG_CLR(fs, cg);
2104 	mutex_exit(&ump->um_lock);
2105 }
2106 
2107 /*
2108  * Check to see if a file is free.
2109  */
2110 int
2111 ffs_checkfreefile(struct fs *fs, struct vnode *devvp, ino_t ino)
2112 {
2113 	struct cg *cgp;
2114 	struct buf *bp;
2115 	daddr_t cgbno;
2116 	int ret, cg;
2117 	u_int8_t *inosused;
2118 	const bool devvp_is_snapshot = (devvp->v_type != VBLK);
2119 
2120 	KASSERT(devvp_is_snapshot);
2121 
2122 	cg = ino_to_cg(fs, ino);
2123 	if (devvp_is_snapshot)
2124 		cgbno = ffs_fragstoblks(fs, cgtod(fs, cg));
2125 	else
2126 		cgbno = FFS_FSBTODB(fs, cgtod(fs, cg));
2127 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
2128 		return 1;
2129 	if (bread(devvp, cgbno, (int)fs->fs_cgsize, 0, &bp)) {
2130 		return 1;
2131 	}
2132 	cgp = (struct cg *)bp->b_data;
2133 	if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) {
2134 		brelse(bp, 0);
2135 		return 1;
2136 	}
2137 	inosused = cg_inosused(cgp, UFS_FSNEEDSWAP(fs));
2138 	ino %= fs->fs_ipg;
2139 	ret = isclr(inosused, ino);
2140 	brelse(bp, 0);
2141 	return ret;
2142 }
2143 
2144 /*
2145  * Find a block of the specified size in the specified cylinder group.
2146  *
2147  * It is a panic if a request is made to find a block if none are
2148  * available.
2149  */
2150 static int32_t
2151 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
2152 {
2153 	int32_t bno;
2154 	int start, len, loc, i;
2155 	int blk, field, subfield, pos;
2156 	int ostart, olen;
2157 	u_int8_t *blksfree;
2158 	const int needswap = UFS_FSNEEDSWAP(fs);
2159 
2160 	/* KASSERT(mutex_owned(&ump->um_lock)); */
2161 
2162 	/*
2163 	 * find the fragment by searching through the free block
2164 	 * map for an appropriate bit pattern
2165 	 */
2166 	if (bpref)
2167 		start = dtogd(fs, bpref) / NBBY;
2168 	else
2169 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
2170 	blksfree = cg_blksfree(cgp, needswap);
2171 	len = howmany(fs->fs_fpg, NBBY) - start;
2172 	ostart = start;
2173 	olen = len;
2174 	loc = scanc((u_int)len,
2175 		(const u_char *)&blksfree[start],
2176 		(const u_char *)fragtbl[fs->fs_frag],
2177 		(1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1)))));
2178 	if (loc == 0) {
2179 		len = start + 1;
2180 		start = 0;
2181 		loc = scanc((u_int)len,
2182 			(const u_char *)&blksfree[0],
2183 			(const u_char *)fragtbl[fs->fs_frag],
2184 			(1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1)))));
2185 		if (loc == 0) {
2186 			panic("%s: map corrupted: start=%d, len=%d, "
2187 			    "fs = %s, offset=%d/%ld, cg %d", __func__,
2188 			    ostart, olen, fs->fs_fsmnt,
2189 			    ufs_rw32(cgp->cg_freeoff, needswap),
2190 			    (long)blksfree - (long)cgp, cgp->cg_cgx);
2191 			/* NOTREACHED */
2192 		}
2193 	}
2194 	bno = (start + len - loc) * NBBY;
2195 	cgp->cg_frotor = ufs_rw32(bno, needswap);
2196 	/*
2197 	 * found the byte in the map
2198 	 * sift through the bits to find the selected frag
2199 	 */
2200 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
2201 		blk = blkmap(fs, blksfree, bno);
2202 		blk <<= 1;
2203 		field = around[allocsiz];
2204 		subfield = inside[allocsiz];
2205 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
2206 			if ((blk & field) == subfield)
2207 				return (bno + pos);
2208 			field <<= 1;
2209 			subfield <<= 1;
2210 		}
2211 	}
2212 	panic("%s: block not in map: bno=%d, fs=%s", __func__,
2213 	    bno, fs->fs_fsmnt);
2214 	/* return (-1); */
2215 }
2216 
2217 /*
2218  * Fserr prints the name of a file system with an error diagnostic.
2219  *
2220  * The form of the error message is:
2221  *	fs: error message
2222  */
2223 static void
2224 ffs_fserr(struct fs *fs, kauth_cred_t cred, const char *cp)
2225 {
2226 	KASSERT(cred != NULL);
2227 
2228 	if (cred == NOCRED || cred == FSCRED) {
2229 		log(LOG_ERR, "pid %d, command %s, on %s: %s\n",
2230 		    curproc->p_pid, curproc->p_comm,
2231 		    fs->fs_fsmnt, cp);
2232 	} else {
2233 		log(LOG_ERR, "uid %d, pid %d, command %s, on %s: %s\n",
2234 		    kauth_cred_getuid(cred), curproc->p_pid, curproc->p_comm,
2235 		    fs->fs_fsmnt, cp);
2236 	}
2237 }
2238