xref: /netbsd-src/sys/ufs/lfs/lfs_subr.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: lfs_subr.c,v 1.56 2006/01/14 17:41:17 yamt 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 1.56 2006/01/14 17:41:17 yamt 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 #ifdef DEBUG
88 const char *lfs_res_names[LFS_NB_COUNT] = {
89 	"summary",
90 	"superblock",
91 	"file block",
92 	"cluster",
93 	"clean",
94 	"blkiov",
95 };
96 #endif
97 
98 int lfs_res_qty[LFS_NB_COUNT] = {
99 	LFS_N_SUMMARIES,
100 	LFS_N_SBLOCKS,
101 	LFS_N_IBLOCKS,
102 	LFS_N_CLUSTERS,
103 	LFS_N_CLEAN,
104 	LFS_N_BLKIOV,
105 };
106 
107 void
108 lfs_setup_resblks(struct lfs *fs)
109 {
110 	int i, j;
111 	int maxbpp;
112 
113 	ASSERT_NO_SEGLOCK(fs);
114 	fs->lfs_resblk = (res_t *)malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
115 					  M_WAITOK);
116 	for (i = 0; i < LFS_N_TOTAL; i++) {
117 		fs->lfs_resblk[i].inuse = 0;
118 		fs->lfs_resblk[i].p = NULL;
119 	}
120 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
121 		LIST_INIT(fs->lfs_reshash + i);
122 
123 	/*
124 	 * These types of allocations can be larger than a page,
125 	 * so we can't use the pool subsystem for them.
126 	 */
127 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
128 		fs->lfs_resblk[i].size = fs->lfs_sumsize;
129 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
130 		fs->lfs_resblk[i].size = LFS_SBPAD;
131 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
132 		fs->lfs_resblk[i].size = fs->lfs_bsize;
133 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
134 		fs->lfs_resblk[i].size = MAXPHYS;
135 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
136 		fs->lfs_resblk[i].size = MAXPHYS;
137 	for (j = 0; j < LFS_N_BLKIOV; j++, i++)
138 		fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO);
139 
140 	for (i = 0; i < LFS_N_TOTAL; i++) {
141 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
142 					     M_SEGMENT, M_WAITOK);
143 	}
144 
145 	/*
146 	 * Initialize pools for small types (XXX is BPP small?)
147 	 */
148 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
149 		"lfsclpl", &pool_allocator_nointr);
150 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
151 		"lfssegpool", &pool_allocator_nointr);
152 	maxbpp = ((fs->lfs_sumsize - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
153 	maxbpp = MIN(maxbpp, segsize(fs) / fs->lfs_fsize + 2);
154 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
155 		"lfsbpppl", &pool_allocator_nointr);
156 }
157 
158 void
159 lfs_free_resblks(struct lfs *fs)
160 {
161 	int i;
162 
163 	pool_destroy(&fs->lfs_bpppool);
164 	pool_destroy(&fs->lfs_segpool);
165 	pool_destroy(&fs->lfs_clpool);
166 
167 	simple_lock(&fs->lfs_interlock);
168 	for (i = 0; i < LFS_N_TOTAL; i++) {
169 		while (fs->lfs_resblk[i].inuse)
170 			ltsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
171 				&fs->lfs_interlock);
172 		if (fs->lfs_resblk[i].p != NULL)
173 			free(fs->lfs_resblk[i].p, M_SEGMENT);
174 	}
175 	free(fs->lfs_resblk, M_SEGMENT);
176 	simple_unlock(&fs->lfs_interlock);
177 }
178 
179 static unsigned int
180 lfs_mhash(void *vp)
181 {
182 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
183 }
184 
185 /*
186  * Return memory of the given size for the given purpose, or use one of a
187  * number of spare last-resort buffers, if malloc returns NULL.
188  */
189 void *
190 lfs_malloc(struct lfs *fs, size_t size, int type)
191 {
192 	struct lfs_res_blk *re;
193 	void *r;
194 	int i, s, start;
195 	unsigned int h;
196 
197 	ASSERT_MAYBE_SEGLOCK(fs);
198 	r = NULL;
199 
200 	/* If no mem allocated for this type, it just waits */
201 	if (lfs_res_qty[type] == 0) {
202 		r = malloc(size, M_SEGMENT, M_WAITOK);
203 		return r;
204 	}
205 
206 	/* Otherwise try a quick malloc, and if it works, great */
207 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
208 		return r;
209 	}
210 
211 	/*
212 	 * If malloc returned NULL, we are forced to use one of our
213 	 * reserve blocks.  We have on hand at least one summary block,
214 	 * at least one cluster block, at least one superblock,
215 	 * and several indirect blocks.
216 	 */
217 
218 	simple_lock(&fs->lfs_interlock);
219 	/* skip over blocks of other types */
220 	for (i = 0, start = 0; i < type; i++)
221 		start += lfs_res_qty[i];
222 	while (r == NULL) {
223 		for (i = 0; i < lfs_res_qty[type]; i++) {
224 			if (fs->lfs_resblk[start + i].inuse == 0) {
225 				re = fs->lfs_resblk + start + i;
226 				re->inuse = 1;
227 				r = re->p;
228 				KASSERT(re->size >= size);
229 				h = lfs_mhash(r);
230 				s = splbio();
231 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
232 				splx(s);
233 				simple_unlock(&fs->lfs_interlock);
234 				return r;
235 			}
236 		}
237 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
238 		      lfs_res_names[type], lfs_res_qty[type]));
239 		ltsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
240 			&fs->lfs_interlock);
241 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
242 		      lfs_res_names[type]));
243 	}
244 	/* NOTREACHED */
245 	simple_unlock(&fs->lfs_interlock);
246 	return r;
247 }
248 
249 void
250 lfs_free(struct lfs *fs, void *p, int type)
251 {
252 	int s;
253 	unsigned int h;
254 	res_t *re;
255 #ifdef DEBUG
256 	int i;
257 #endif
258 
259 	ASSERT_MAYBE_SEGLOCK(fs);
260 	h = lfs_mhash(p);
261 	simple_lock(&fs->lfs_interlock);
262 	s = splbio();
263 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
264 		if (re->p == p) {
265 			KASSERT(re->inuse == 1);
266 			LIST_REMOVE(re, res);
267 			re->inuse = 0;
268 			wakeup(&fs->lfs_resblk);
269 			splx(s);
270 			simple_unlock(&fs->lfs_interlock);
271 			return;
272 		}
273 	}
274 #ifdef DEBUG
275 	for (i = 0; i < LFS_N_TOTAL; i++) {
276 		if (fs->lfs_resblk[i].p == p)
277 			panic("lfs_free: inconsistent reserved block");
278 	}
279 #endif
280 	splx(s);
281 	simple_unlock(&fs->lfs_interlock);
282 
283 	/*
284 	 * If we didn't find it, free it.
285 	 */
286 	free(p, M_SEGMENT);
287 }
288 
289 /*
290  * lfs_seglock --
291  *	Single thread the segment writer.
292  */
293 int
294 lfs_seglock(struct lfs *fs, unsigned long flags)
295 {
296 	struct segment *sp;
297 
298 	simple_lock(&fs->lfs_interlock);
299 	if (fs->lfs_seglock) {
300 		if (fs->lfs_lockpid == curproc->p_pid) {
301 			simple_unlock(&fs->lfs_interlock);
302 			++fs->lfs_seglock;
303 			fs->lfs_sp->seg_flags |= flags;
304 			return 0;
305 		} else if (flags & SEGM_PAGEDAEMON) {
306 			simple_unlock(&fs->lfs_interlock);
307 			return EWOULDBLOCK;
308 		} else {
309 			while (fs->lfs_seglock) {
310 				(void)ltsleep(&fs->lfs_seglock, PRIBIO + 1,
311 					"lfs seglock", 0, &fs->lfs_interlock);
312 			}
313 		}
314 	}
315 
316 	fs->lfs_seglock = 1;
317 	fs->lfs_lockpid = curproc->p_pid;
318 	simple_unlock(&fs->lfs_interlock);
319 	fs->lfs_cleanind = 0;
320 
321 #ifdef DEBUG
322 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
323 #endif
324 	/* Drain fragment size changes out */
325 	lockmgr(&fs->lfs_fraglock, LK_EXCLUSIVE, 0);
326 
327 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
328 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
329 	sp->seg_flags = flags;
330 	sp->vp = NULL;
331 	sp->seg_iocount = 0;
332 	(void) lfs_initseg(fs);
333 
334 	/*
335 	 * Keep a cumulative count of the outstanding I/O operations.  If the
336 	 * disk drive catches up with us it could go to zero before we finish,
337 	 * so we artificially increment it by one until we've scheduled all of
338 	 * the writes we intend to do.
339 	 */
340 	simple_lock(&fs->lfs_interlock);
341 	++fs->lfs_iocount;
342 	simple_unlock(&fs->lfs_interlock);
343 	return 0;
344 }
345 
346 static void lfs_unmark_dirop(struct lfs *);
347 
348 static void
349 lfs_unmark_dirop(struct lfs *fs)
350 {
351 	struct inode *ip, *nip;
352 	struct vnode *vp;
353 	int doit;
354 
355 	ASSERT_NO_SEGLOCK(fs);
356 	simple_lock(&fs->lfs_interlock);
357 	doit = !(fs->lfs_flags & LFS_UNDIROP);
358 	if (doit)
359 		fs->lfs_flags |= LFS_UNDIROP;
360 	if (!doit) {
361 		simple_unlock(&fs->lfs_interlock);
362 		return;
363 	}
364 
365 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
366 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
367 		simple_unlock(&fs->lfs_interlock);
368 		vp = ITOV(ip);
369 
370 		simple_lock(&vp->v_interlock);
371 		if (VOP_ISLOCKED(vp) &&
372 			   vp->v_lock.lk_lockholder != curproc->p_pid) {
373 			simple_lock(&fs->lfs_interlock);
374 			simple_unlock(&vp->v_interlock);
375 			continue;
376 		}
377 		if ((VTOI(vp)->i_flag & IN_ADIROP) == 0) {
378 			simple_lock(&fs->lfs_interlock);
379 			simple_lock(&lfs_subsys_lock);
380 			--lfs_dirvcount;
381 			simple_unlock(&lfs_subsys_lock);
382 			vp->v_flag &= ~VDIROP;
383 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
384 			simple_unlock(&fs->lfs_interlock);
385 			wakeup(&lfs_dirvcount);
386 			simple_unlock(&vp->v_interlock);
387 			simple_lock(&fs->lfs_interlock);
388 			fs->lfs_unlockvp = vp;
389 			simple_unlock(&fs->lfs_interlock);
390 			vrele(vp);
391 			simple_lock(&fs->lfs_interlock);
392 			fs->lfs_unlockvp = NULL;
393 			simple_unlock(&fs->lfs_interlock);
394 		} else
395 			simple_unlock(&vp->v_interlock);
396 		simple_lock(&fs->lfs_interlock);
397 	}
398 
399 	fs->lfs_flags &= ~LFS_UNDIROP;
400 	simple_unlock(&fs->lfs_interlock);
401 	wakeup(&fs->lfs_flags);
402 }
403 
404 static void
405 lfs_auto_segclean(struct lfs *fs)
406 {
407 	int i, error, s, waited;
408 
409 	ASSERT_SEGLOCK(fs);
410 	/*
411 	 * Now that we've swapped lfs_activesb, but while we still
412 	 * hold the segment lock, run through the segment list marking
413 	 * the empty ones clean.
414 	 * XXX - do we really need to do them all at once?
415 	 */
416 	waited = 0;
417 	for (i = 0; i < fs->lfs_nseg; i++) {
418 		if ((fs->lfs_suflags[0][i] &
419 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
420 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
421 		    (fs->lfs_suflags[1][i] &
422 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
423 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
424 
425 			/* Make sure the sb is written before we clean */
426 			simple_lock(&fs->lfs_interlock);
427 			s = splbio();
428 			while (waited == 0 && fs->lfs_sbactive)
429 				ltsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
430 					0, &fs->lfs_interlock);
431 			splx(s);
432 			simple_unlock(&fs->lfs_interlock);
433 			waited = 1;
434 
435 			if ((error = lfs_do_segclean(fs, i)) != 0) {
436 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
437 			}
438 		}
439 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
440 			fs->lfs_suflags[fs->lfs_activesb][i];
441 	}
442 }
443 
444 /*
445  * lfs_segunlock --
446  *	Single thread the segment writer.
447  */
448 void
449 lfs_segunlock(struct lfs *fs)
450 {
451 	struct segment *sp;
452 	unsigned long sync, ckp;
453 	struct buf *bp;
454 	int do_unmark_dirop = 0;
455 
456 	sp = fs->lfs_sp;
457 
458 	simple_lock(&fs->lfs_interlock);
459 	LOCK_ASSERT(LFS_SEGLOCK_HELD(fs));
460 	if (fs->lfs_seglock == 1) {
461 		if ((sp->seg_flags & SEGM_PROT) == 0)
462 			do_unmark_dirop = 1;
463 		simple_unlock(&fs->lfs_interlock);
464 		sync = sp->seg_flags & SEGM_SYNC;
465 		ckp = sp->seg_flags & SEGM_CKP;
466 		if (sp->bpp != sp->cbpp) {
467 			/* Free allocated segment summary */
468 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
469 			bp = *sp->bpp;
470 			lfs_freebuf(fs, bp);
471 		} else
472 			DLOG((DLOG_SEG, "lfs_segunlock: unlock to 0 with no summary"));
473 
474 		pool_put(&fs->lfs_bpppool, sp->bpp);
475 		sp->bpp = NULL;
476 
477 		/*
478 		 * If we're not sync, we're done with sp, get rid of it.
479 		 * Otherwise, we keep a local copy around but free
480 		 * fs->lfs_sp so another process can use it (we have to
481 		 * wait but they don't have to wait for us).
482 		 */
483 		if (!sync)
484 			pool_put(&fs->lfs_segpool, sp);
485 		fs->lfs_sp = NULL;
486 
487 		/*
488 		 * If the I/O count is non-zero, sleep until it reaches zero.
489 		 * At the moment, the user's process hangs around so we can
490 		 * sleep.
491 		 */
492 		simple_lock(&fs->lfs_interlock);
493 		if (--fs->lfs_iocount == 0)
494 			LFS_DEBUG_COUNTLOCKED("lfs_segunlock");
495 		if (fs->lfs_iocount <= 1)
496 			wakeup(&fs->lfs_iocount);
497 		simple_unlock(&fs->lfs_interlock);
498 		/*
499 		 * If we're not checkpointing, we don't have to block
500 		 * other processes to wait for a synchronous write
501 		 * to complete.
502 		 */
503 		if (!ckp) {
504 #ifdef DEBUG
505 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
506 #endif
507 			simple_lock(&fs->lfs_interlock);
508 			--fs->lfs_seglock;
509 			fs->lfs_lockpid = 0;
510 			simple_unlock(&fs->lfs_interlock);
511 			wakeup(&fs->lfs_seglock);
512 		}
513 		/*
514 		 * We let checkpoints happen asynchronously.  That means
515 		 * that during recovery, we have to roll forward between
516 		 * the two segments described by the first and second
517 		 * superblocks to make sure that the checkpoint described
518 		 * by a superblock completed.
519 		 */
520 		simple_lock(&fs->lfs_interlock);
521 		while (ckp && sync && fs->lfs_iocount)
522 			(void)ltsleep(&fs->lfs_iocount, PRIBIO + 1,
523 				      "lfs_iocount", 0, &fs->lfs_interlock);
524 		while (sync && sp->seg_iocount) {
525 			(void)ltsleep(&sp->seg_iocount, PRIBIO + 1,
526 				     "seg_iocount", 0, &fs->lfs_interlock);
527 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
528 		}
529 		simple_unlock(&fs->lfs_interlock);
530 		if (sync)
531 			pool_put(&fs->lfs_segpool, sp);
532 
533 		if (ckp) {
534 			fs->lfs_nactive = 0;
535 			/* If we *know* everything's on disk, write both sbs */
536 			/* XXX should wait for this one	 */
537 			if (sync)
538 				lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
539 			lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
540 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
541 				lfs_auto_segclean(fs);
542 				/* If sync, we can clean the remainder too */
543 				if (sync)
544 					lfs_auto_segclean(fs);
545 			}
546 			fs->lfs_activesb = 1 - fs->lfs_activesb;
547 #ifdef DEBUG
548 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
549 #endif
550 			simple_lock(&fs->lfs_interlock);
551 			--fs->lfs_seglock;
552 			fs->lfs_lockpid = 0;
553 			simple_unlock(&fs->lfs_interlock);
554 			wakeup(&fs->lfs_seglock);
555 		}
556 		/* Reenable fragment size changes */
557 		lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0);
558 		if (do_unmark_dirop)
559 			lfs_unmark_dirop(fs);
560 	} else if (fs->lfs_seglock == 0) {
561 		simple_unlock(&fs->lfs_interlock);
562 		panic ("Seglock not held");
563 	} else {
564 		--fs->lfs_seglock;
565 		simple_unlock(&fs->lfs_interlock);
566 	}
567 }
568 
569 /*
570  * drain dirops and start writer.
571  */
572 int
573 lfs_writer_enter(struct lfs *fs, const char *wmesg)
574 {
575 	int error = 0;
576 
577 	ASSERT_MAYBE_SEGLOCK(fs);
578 	simple_lock(&fs->lfs_interlock);
579 
580 	/* disallow dirops during flush */
581 	fs->lfs_writer++;
582 
583 	while (fs->lfs_dirops > 0) {
584 		++fs->lfs_diropwait;
585 		error = ltsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
586 				&fs->lfs_interlock);
587 		--fs->lfs_diropwait;
588 	}
589 
590 	if (error)
591 		fs->lfs_writer--;
592 
593 	simple_unlock(&fs->lfs_interlock);
594 
595 	return error;
596 }
597 
598 void
599 lfs_writer_leave(struct lfs *fs)
600 {
601 	boolean_t dowakeup;
602 
603 	ASSERT_MAYBE_SEGLOCK(fs);
604 	simple_lock(&fs->lfs_interlock);
605 	dowakeup = !(--fs->lfs_writer);
606 	simple_unlock(&fs->lfs_interlock);
607 	if (dowakeup)
608 		wakeup(&fs->lfs_dirops);
609 }
610