xref: /netbsd-src/sys/ufs/lfs/lfs_subr.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: lfs_subr.c,v 1.55 2005/12/11 12:25:26 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*
39  * Copyright (c) 1991, 1993
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)lfs_subr.c	8.4 (Berkeley) 5/8/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: lfs_subr.c,v 1.55 2005/12/11 12:25:26 christos Exp $");
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/namei.h>
75 #include <sys/vnode.h>
76 #include <sys/buf.h>
77 #include <sys/mount.h>
78 #include <sys/malloc.h>
79 #include <sys/proc.h>
80 
81 #include <ufs/ufs/inode.h>
82 #include <ufs/lfs/lfs.h>
83 #include <ufs/lfs/lfs_extern.h>
84 
85 #include <uvm/uvm.h>
86 
87 /*
88  * Return buffer with the contents of block "offset" from the beginning of
89  * directory "ip".  If "res" is non-zero, fill it in with a pointer to the
90  * remaining space in the directory.
91  */
92 int
93 lfs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp)
94 {
95 	struct lfs *fs;
96 	struct inode *ip;
97 	struct buf *bp;
98 	daddr_t lbn;
99 	int bsize, error;
100 
101 	ip = VTOI(vp);
102 	fs = ip->i_lfs;
103 	lbn = lblkno(fs, offset);
104 	bsize = blksize(fs, ip, lbn);
105 
106 	*bpp = NULL;
107 	if ((error = bread(vp, lbn, bsize, NOCRED, &bp)) != 0) {
108 		brelse(bp);
109 		return (error);
110 	}
111 	if (res)
112 		*res = (char *)bp->b_data + blkoff(fs, offset);
113 	*bpp = bp;
114 	return (0);
115 }
116 
117 #ifdef DEBUG
118 const char *lfs_res_names[LFS_NB_COUNT] = {
119 	"summary",
120 	"superblock",
121 	"file block",
122 	"cluster",
123 	"clean",
124 	"blkiov",
125 };
126 #endif
127 
128 int lfs_res_qty[LFS_NB_COUNT] = {
129 	LFS_N_SUMMARIES,
130 	LFS_N_SBLOCKS,
131 	LFS_N_IBLOCKS,
132 	LFS_N_CLUSTERS,
133 	LFS_N_CLEAN,
134 	LFS_N_BLKIOV,
135 };
136 
137 void
138 lfs_setup_resblks(struct lfs *fs)
139 {
140 	int i, j;
141 	int maxbpp;
142 
143 	ASSERT_NO_SEGLOCK(fs);
144 	fs->lfs_resblk = (res_t *)malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
145 					  M_WAITOK);
146 	for (i = 0; i < LFS_N_TOTAL; i++) {
147 		fs->lfs_resblk[i].inuse = 0;
148 		fs->lfs_resblk[i].p = NULL;
149 	}
150 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
151 		LIST_INIT(fs->lfs_reshash + i);
152 
153 	/*
154 	 * These types of allocations can be larger than a page,
155 	 * so we can't use the pool subsystem for them.
156 	 */
157 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
158 		fs->lfs_resblk[i].size = fs->lfs_sumsize;
159 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
160 		fs->lfs_resblk[i].size = LFS_SBPAD;
161 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
162 		fs->lfs_resblk[i].size = fs->lfs_bsize;
163 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
164 		fs->lfs_resblk[i].size = MAXPHYS;
165 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
166 		fs->lfs_resblk[i].size = MAXPHYS;
167 	for (j = 0; j < LFS_N_BLKIOV; j++, i++)
168 		fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO);
169 
170 	for (i = 0; i < LFS_N_TOTAL; i++) {
171 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
172 					     M_SEGMENT, M_WAITOK);
173 	}
174 
175 	/*
176 	 * Initialize pools for small types (XXX is BPP small?)
177 	 */
178 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
179 		"lfsclpl", &pool_allocator_nointr);
180 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
181 		"lfssegpool", &pool_allocator_nointr);
182 	maxbpp = ((fs->lfs_sumsize - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
183 	maxbpp = MIN(maxbpp, segsize(fs) / fs->lfs_fsize + 2);
184 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
185 		"lfsbpppl", &pool_allocator_nointr);
186 }
187 
188 void
189 lfs_free_resblks(struct lfs *fs)
190 {
191 	int i;
192 
193 	pool_destroy(&fs->lfs_bpppool);
194 	pool_destroy(&fs->lfs_segpool);
195 	pool_destroy(&fs->lfs_clpool);
196 
197 	simple_lock(&fs->lfs_interlock);
198 	for (i = 0; i < LFS_N_TOTAL; i++) {
199 		while (fs->lfs_resblk[i].inuse)
200 			ltsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
201 				&fs->lfs_interlock);
202 		if (fs->lfs_resblk[i].p != NULL)
203 			free(fs->lfs_resblk[i].p, M_SEGMENT);
204 	}
205 	free(fs->lfs_resblk, M_SEGMENT);
206 	simple_unlock(&fs->lfs_interlock);
207 }
208 
209 static unsigned int
210 lfs_mhash(void *vp)
211 {
212 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
213 }
214 
215 /*
216  * Return memory of the given size for the given purpose, or use one of a
217  * number of spare last-resort buffers, if malloc returns NULL.
218  */
219 void *
220 lfs_malloc(struct lfs *fs, size_t size, int type)
221 {
222 	struct lfs_res_blk *re;
223 	void *r;
224 	int i, s, start;
225 	unsigned int h;
226 
227 	ASSERT_MAYBE_SEGLOCK(fs);
228 	r = NULL;
229 
230 	/* If no mem allocated for this type, it just waits */
231 	if (lfs_res_qty[type] == 0) {
232 		r = malloc(size, M_SEGMENT, M_WAITOK);
233 		return r;
234 	}
235 
236 	/* Otherwise try a quick malloc, and if it works, great */
237 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
238 		return r;
239 	}
240 
241 	/*
242 	 * If malloc returned NULL, we are forced to use one of our
243 	 * reserve blocks.  We have on hand at least one summary block,
244 	 * at least one cluster block, at least one superblock,
245 	 * and several indirect blocks.
246 	 */
247 
248 	simple_lock(&fs->lfs_interlock);
249 	/* skip over blocks of other types */
250 	for (i = 0, start = 0; i < type; i++)
251 		start += lfs_res_qty[i];
252 	while (r == NULL) {
253 		for (i = 0; i < lfs_res_qty[type]; i++) {
254 			if (fs->lfs_resblk[start + i].inuse == 0) {
255 				re = fs->lfs_resblk + start + i;
256 				re->inuse = 1;
257 				r = re->p;
258 				KASSERT(re->size >= size);
259 				h = lfs_mhash(r);
260 				s = splbio();
261 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
262 				splx(s);
263 				simple_unlock(&fs->lfs_interlock);
264 				return r;
265 			}
266 		}
267 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
268 		      lfs_res_names[type], lfs_res_qty[type]));
269 		ltsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
270 			&fs->lfs_interlock);
271 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
272 		      lfs_res_names[type]));
273 	}
274 	/* NOTREACHED */
275 	simple_unlock(&fs->lfs_interlock);
276 	return r;
277 }
278 
279 void
280 lfs_free(struct lfs *fs, void *p, int type)
281 {
282 	int s;
283 	unsigned int h;
284 	res_t *re;
285 #ifdef DEBUG
286 	int i;
287 #endif
288 
289 	ASSERT_MAYBE_SEGLOCK(fs);
290 	h = lfs_mhash(p);
291 	simple_lock(&fs->lfs_interlock);
292 	s = splbio();
293 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
294 		if (re->p == p) {
295 			KASSERT(re->inuse == 1);
296 			LIST_REMOVE(re, res);
297 			re->inuse = 0;
298 			wakeup(&fs->lfs_resblk);
299 			splx(s);
300 			simple_unlock(&fs->lfs_interlock);
301 			return;
302 		}
303 	}
304 #ifdef DEBUG
305 	for (i = 0; i < LFS_N_TOTAL; i++) {
306 		if (fs->lfs_resblk[i].p == p)
307 			panic("lfs_free: inconsistent reserved block");
308 	}
309 #endif
310 	splx(s);
311 	simple_unlock(&fs->lfs_interlock);
312 
313 	/*
314 	 * If we didn't find it, free it.
315 	 */
316 	free(p, M_SEGMENT);
317 }
318 
319 /*
320  * lfs_seglock --
321  *	Single thread the segment writer.
322  */
323 int
324 lfs_seglock(struct lfs *fs, unsigned long flags)
325 {
326 	struct segment *sp;
327 
328 	simple_lock(&fs->lfs_interlock);
329 	if (fs->lfs_seglock) {
330 		if (fs->lfs_lockpid == curproc->p_pid) {
331 			simple_unlock(&fs->lfs_interlock);
332 			++fs->lfs_seglock;
333 			fs->lfs_sp->seg_flags |= flags;
334 			return 0;
335 		} else if (flags & SEGM_PAGEDAEMON) {
336 			simple_unlock(&fs->lfs_interlock);
337 			return EWOULDBLOCK;
338 		} else {
339 			while (fs->lfs_seglock) {
340 				(void)ltsleep(&fs->lfs_seglock, PRIBIO + 1,
341 					"lfs seglock", 0, &fs->lfs_interlock);
342 			}
343 		}
344 	}
345 
346 	fs->lfs_seglock = 1;
347 	fs->lfs_lockpid = curproc->p_pid;
348 	simple_unlock(&fs->lfs_interlock);
349 	fs->lfs_cleanind = 0;
350 
351 #ifdef DEBUG
352 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
353 #endif
354 	/* Drain fragment size changes out */
355 	lockmgr(&fs->lfs_fraglock, LK_EXCLUSIVE, 0);
356 
357 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
358 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
359 	sp->seg_flags = flags;
360 	sp->vp = NULL;
361 	sp->seg_iocount = 0;
362 	(void) lfs_initseg(fs);
363 
364 	/*
365 	 * Keep a cumulative count of the outstanding I/O operations.  If the
366 	 * disk drive catches up with us it could go to zero before we finish,
367 	 * so we artificially increment it by one until we've scheduled all of
368 	 * the writes we intend to do.
369 	 */
370 	simple_lock(&fs->lfs_interlock);
371 	++fs->lfs_iocount;
372 	simple_unlock(&fs->lfs_interlock);
373 	return 0;
374 }
375 
376 static void lfs_unmark_dirop(struct lfs *);
377 
378 static void
379 lfs_unmark_dirop(struct lfs *fs)
380 {
381 	struct inode *ip, *nip;
382 	struct vnode *vp;
383 	int doit;
384 
385 	ASSERT_NO_SEGLOCK(fs);
386 	simple_lock(&fs->lfs_interlock);
387 	doit = !(fs->lfs_flags & LFS_UNDIROP);
388 	if (doit)
389 		fs->lfs_flags |= LFS_UNDIROP;
390 	if (!doit) {
391 		simple_unlock(&fs->lfs_interlock);
392 		return;
393 	}
394 
395 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
396 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
397 		simple_unlock(&fs->lfs_interlock);
398 		vp = ITOV(ip);
399 
400 		simple_lock(&vp->v_interlock);
401 		if (VOP_ISLOCKED(vp) &&
402 			   vp->v_lock.lk_lockholder != curproc->p_pid) {
403 			simple_lock(&fs->lfs_interlock);
404 			simple_unlock(&vp->v_interlock);
405 			continue;
406 		}
407 		if ((VTOI(vp)->i_flag & IN_ADIROP) == 0) {
408 			simple_lock(&fs->lfs_interlock);
409 			simple_lock(&lfs_subsys_lock);
410 			--lfs_dirvcount;
411 			simple_unlock(&lfs_subsys_lock);
412 			vp->v_flag &= ~VDIROP;
413 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
414 			simple_unlock(&fs->lfs_interlock);
415 			wakeup(&lfs_dirvcount);
416 			simple_unlock(&vp->v_interlock);
417 			simple_lock(&fs->lfs_interlock);
418 			fs->lfs_unlockvp = vp;
419 			simple_unlock(&fs->lfs_interlock);
420 			vrele(vp);
421 			simple_lock(&fs->lfs_interlock);
422 			fs->lfs_unlockvp = NULL;
423 			simple_unlock(&fs->lfs_interlock);
424 		} else
425 			simple_unlock(&vp->v_interlock);
426 		simple_lock(&fs->lfs_interlock);
427 	}
428 
429 	fs->lfs_flags &= ~LFS_UNDIROP;
430 	simple_unlock(&fs->lfs_interlock);
431 	wakeup(&fs->lfs_flags);
432 }
433 
434 static void
435 lfs_auto_segclean(struct lfs *fs)
436 {
437 	int i, error, s, waited;
438 
439 	ASSERT_SEGLOCK(fs);
440 	/*
441 	 * Now that we've swapped lfs_activesb, but while we still
442 	 * hold the segment lock, run through the segment list marking
443 	 * the empty ones clean.
444 	 * XXX - do we really need to do them all at once?
445 	 */
446 	waited = 0;
447 	for (i = 0; i < fs->lfs_nseg; i++) {
448 		if ((fs->lfs_suflags[0][i] &
449 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
450 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
451 		    (fs->lfs_suflags[1][i] &
452 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
453 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
454 
455 			/* Make sure the sb is written before we clean */
456 			simple_lock(&fs->lfs_interlock);
457 			s = splbio();
458 			while (waited == 0 && fs->lfs_sbactive)
459 				ltsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
460 					0, &fs->lfs_interlock);
461 			splx(s);
462 			simple_unlock(&fs->lfs_interlock);
463 			waited = 1;
464 
465 			if ((error = lfs_do_segclean(fs, i)) != 0) {
466 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
467 			}
468 		}
469 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
470 			fs->lfs_suflags[fs->lfs_activesb][i];
471 	}
472 }
473 
474 /*
475  * lfs_segunlock --
476  *	Single thread the segment writer.
477  */
478 void
479 lfs_segunlock(struct lfs *fs)
480 {
481 	struct segment *sp;
482 	unsigned long sync, ckp;
483 	struct buf *bp;
484 	int do_unmark_dirop = 0;
485 
486 	sp = fs->lfs_sp;
487 
488 	simple_lock(&fs->lfs_interlock);
489 	LOCK_ASSERT(LFS_SEGLOCK_HELD(fs));
490 	if (fs->lfs_seglock == 1) {
491 		if ((sp->seg_flags & SEGM_PROT) == 0)
492 			do_unmark_dirop = 1;
493 		simple_unlock(&fs->lfs_interlock);
494 		sync = sp->seg_flags & SEGM_SYNC;
495 		ckp = sp->seg_flags & SEGM_CKP;
496 		if (sp->bpp != sp->cbpp) {
497 			/* Free allocated segment summary */
498 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
499 			bp = *sp->bpp;
500 			lfs_freebuf(fs, bp);
501 		} else
502 			DLOG((DLOG_SEG, "lfs_segunlock: unlock to 0 with no summary"));
503 
504 		pool_put(&fs->lfs_bpppool, sp->bpp);
505 		sp->bpp = NULL;
506 
507 		/*
508 		 * If we're not sync, we're done with sp, get rid of it.
509 		 * Otherwise, we keep a local copy around but free
510 		 * fs->lfs_sp so another process can use it (we have to
511 		 * wait but they don't have to wait for us).
512 		 */
513 		if (!sync)
514 			pool_put(&fs->lfs_segpool, sp);
515 		fs->lfs_sp = NULL;
516 
517 		/*
518 		 * If the I/O count is non-zero, sleep until it reaches zero.
519 		 * At the moment, the user's process hangs around so we can
520 		 * sleep.
521 		 */
522 		simple_lock(&fs->lfs_interlock);
523 		if (--fs->lfs_iocount == 0)
524 			LFS_DEBUG_COUNTLOCKED("lfs_segunlock");
525 		if (fs->lfs_iocount <= 1)
526 			wakeup(&fs->lfs_iocount);
527 		simple_unlock(&fs->lfs_interlock);
528 		/*
529 		 * If we're not checkpointing, we don't have to block
530 		 * other processes to wait for a synchronous write
531 		 * to complete.
532 		 */
533 		if (!ckp) {
534 #ifdef DEBUG
535 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
536 #endif
537 			simple_lock(&fs->lfs_interlock);
538 			--fs->lfs_seglock;
539 			fs->lfs_lockpid = 0;
540 			simple_unlock(&fs->lfs_interlock);
541 			wakeup(&fs->lfs_seglock);
542 		}
543 		/*
544 		 * We let checkpoints happen asynchronously.  That means
545 		 * that during recovery, we have to roll forward between
546 		 * the two segments described by the first and second
547 		 * superblocks to make sure that the checkpoint described
548 		 * by a superblock completed.
549 		 */
550 		simple_lock(&fs->lfs_interlock);
551 		while (ckp && sync && fs->lfs_iocount)
552 			(void)ltsleep(&fs->lfs_iocount, PRIBIO + 1,
553 				      "lfs_iocount", 0, &fs->lfs_interlock);
554 		while (sync && sp->seg_iocount) {
555 			(void)ltsleep(&sp->seg_iocount, PRIBIO + 1,
556 				     "seg_iocount", 0, &fs->lfs_interlock);
557 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
558 		}
559 		simple_unlock(&fs->lfs_interlock);
560 		if (sync)
561 			pool_put(&fs->lfs_segpool, sp);
562 
563 		if (ckp) {
564 			fs->lfs_nactive = 0;
565 			/* If we *know* everything's on disk, write both sbs */
566 			/* XXX should wait for this one	 */
567 			if (sync)
568 				lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
569 			lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
570 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
571 				lfs_auto_segclean(fs);
572 				/* If sync, we can clean the remainder too */
573 				if (sync)
574 					lfs_auto_segclean(fs);
575 			}
576 			fs->lfs_activesb = 1 - fs->lfs_activesb;
577 #ifdef DEBUG
578 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
579 #endif
580 			simple_lock(&fs->lfs_interlock);
581 			--fs->lfs_seglock;
582 			fs->lfs_lockpid = 0;
583 			simple_unlock(&fs->lfs_interlock);
584 			wakeup(&fs->lfs_seglock);
585 		}
586 		/* Reenable fragment size changes */
587 		lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0);
588 		if (do_unmark_dirop)
589 			lfs_unmark_dirop(fs);
590 	} else if (fs->lfs_seglock == 0) {
591 		simple_unlock(&fs->lfs_interlock);
592 		panic ("Seglock not held");
593 	} else {
594 		--fs->lfs_seglock;
595 		simple_unlock(&fs->lfs_interlock);
596 	}
597 }
598 
599 /*
600  * drain dirops and start writer.
601  */
602 int
603 lfs_writer_enter(struct lfs *fs, const char *wmesg)
604 {
605 	int error = 0;
606 
607 	ASSERT_MAYBE_SEGLOCK(fs);
608 	simple_lock(&fs->lfs_interlock);
609 
610 	/* disallow dirops during flush */
611 	fs->lfs_writer++;
612 
613 	while (fs->lfs_dirops > 0) {
614 		++fs->lfs_diropwait;
615 		error = ltsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
616 				&fs->lfs_interlock);
617 		--fs->lfs_diropwait;
618 	}
619 
620 	if (error)
621 		fs->lfs_writer--;
622 
623 	simple_unlock(&fs->lfs_interlock);
624 
625 	return error;
626 }
627 
628 void
629 lfs_writer_leave(struct lfs *fs)
630 {
631 	boolean_t dowakeup;
632 
633 	ASSERT_MAYBE_SEGLOCK(fs);
634 	simple_lock(&fs->lfs_interlock);
635 	dowakeup = !(--fs->lfs_writer);
636 	simple_unlock(&fs->lfs_interlock);
637 	if (dowakeup)
638 		wakeup(&fs->lfs_dirops);
639 }
640