xref: /netbsd-src/sys/ufs/lfs/lfs_subr.c (revision 9fc453562f6ebe8eabdfd51e21ae0a0058906d4f)
1 /*	$NetBSD: lfs_subr.c,v 1.103 2020/09/05 16:30:13 riastradh 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  *
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  * Copyright (c) 1991, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	@(#)lfs_subr.c	8.4 (Berkeley) 5/8/95
60  */
61 
62 #include <sys/cdefs.h>
63 __KERNEL_RCSID(0, "$NetBSD: lfs_subr.c,v 1.103 2020/09/05 16:30:13 riastradh Exp $");
64 
65 #include <sys/param.h>
66 #include <sys/systm.h>
67 #include <sys/namei.h>
68 #include <sys/vnode.h>
69 #include <sys/buf.h>
70 #include <sys/mount.h>
71 #include <sys/malloc.h>
72 #include <sys/proc.h>
73 #include <sys/kauth.h>
74 
75 #include <ufs/lfs/ulfs_inode.h>
76 #include <ufs/lfs/lfs.h>
77 #include <ufs/lfs/lfs_accessors.h>
78 #include <ufs/lfs/lfs_kernel.h>
79 #include <ufs/lfs/lfs_extern.h>
80 
81 #ifdef DEBUG
82 const char *lfs_res_names[LFS_NB_COUNT] = {
83 	"summary",
84 	"superblock",
85 	"file block",
86 	"cluster",
87 	"clean",
88 	"blkiov",
89 };
90 #endif
91 
92 int lfs_res_qty[LFS_NB_COUNT] = {
93 	LFS_N_SUMMARIES,
94 	LFS_N_SBLOCKS,
95 	LFS_N_IBLOCKS,
96 	LFS_N_CLUSTERS,
97 	LFS_N_CLEAN,
98 	LFS_N_BLKIOV,
99 };
100 
101 void
lfs_setup_resblks(struct lfs * fs)102 lfs_setup_resblks(struct lfs *fs)
103 {
104 	int i, j;
105 	int maxbpp;
106 
107 	ASSERT_NO_SEGLOCK(fs);
108 	fs->lfs_resblk = malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
109 				M_WAITOK);
110 	for (i = 0; i < LFS_N_TOTAL; i++) {
111 		fs->lfs_resblk[i].inuse = 0;
112 		fs->lfs_resblk[i].p = NULL;
113 	}
114 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
115 		LIST_INIT(fs->lfs_reshash + i);
116 
117 	/*
118 	 * These types of allocations can be larger than a page,
119 	 * so we can't use the pool subsystem for them.
120 	 */
121 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
122 		fs->lfs_resblk[i].size = lfs_sb_getsumsize(fs);
123 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
124 		fs->lfs_resblk[i].size = LFS_SBPAD;
125 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
126 		fs->lfs_resblk[i].size = lfs_sb_getbsize(fs);
127 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
128 		fs->lfs_resblk[i].size = MAXPHYS;
129 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
130 		fs->lfs_resblk[i].size = MAXPHYS;
131 	for (j = 0; j < LFS_N_BLKIOV; j++, i++)
132 		fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO);
133 
134 	for (i = 0; i < LFS_N_TOTAL; i++) {
135 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
136 					     M_SEGMENT, M_WAITOK);
137 	}
138 
139 	/*
140 	 * Initialize pools for small types (XXX is BPP small?)
141 	 */
142 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
143 		"lfsclpl", &pool_allocator_nointr, IPL_NONE);
144 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
145 		"lfssegpool", &pool_allocator_nointr, IPL_NONE);
146 	/* XXX: should this int32 be 32/64? */
147 	maxbpp = ((lfs_sb_getsumsize(fs) - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
148 	maxbpp = MIN(maxbpp, lfs_segsize(fs) / lfs_sb_getfsize(fs) + 2);
149 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
150 		"lfsbpppl", &pool_allocator_nointr, IPL_NONE);
151 }
152 
153 void
lfs_free_resblks(struct lfs * fs)154 lfs_free_resblks(struct lfs *fs)
155 {
156 	int i;
157 
158 	pool_destroy(&fs->lfs_bpppool);
159 	pool_destroy(&fs->lfs_segpool);
160 	pool_destroy(&fs->lfs_clpool);
161 
162 	mutex_enter(&lfs_lock);
163 	for (i = 0; i < LFS_N_TOTAL; i++) {
164 		while (fs->lfs_resblk[i].inuse)
165 			mtsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
166 				&lfs_lock);
167 		if (fs->lfs_resblk[i].p != NULL)
168 			free(fs->lfs_resblk[i].p, M_SEGMENT);
169 	}
170 	free(fs->lfs_resblk, M_SEGMENT);
171 	mutex_exit(&lfs_lock);
172 }
173 
174 static unsigned int
lfs_mhash(void * vp)175 lfs_mhash(void *vp)
176 {
177 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
178 }
179 
180 /*
181  * Return memory of the given size for the given purpose, or use one of a
182  * number of spare last-resort buffers, if malloc returns NULL.
183  */
184 void *
lfs_malloc(struct lfs * fs,size_t size,int type)185 lfs_malloc(struct lfs *fs, size_t size, int type)
186 {
187 	struct lfs_res_blk *re;
188 	void *r;
189 	int i, start;
190 	unsigned int h;
191 
192 	ASSERT_MAYBE_SEGLOCK(fs);
193 	r = NULL;
194 
195 	/* If no mem allocated for this type, it just waits */
196 	if (lfs_res_qty[type] == 0) {
197 		r = malloc(size, M_SEGMENT, M_WAITOK);
198 		return r;
199 	}
200 
201 	/* Otherwise try a quick malloc, and if it works, great */
202 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
203 		return r;
204 	}
205 
206 	/*
207 	 * If malloc returned NULL, we are forced to use one of our
208 	 * reserve blocks.  We have on hand at least one summary block,
209 	 * at least one cluster block, at least one superblock,
210 	 * and several indirect blocks.
211 	 */
212 
213 	mutex_enter(&lfs_lock);
214 	/* skip over blocks of other types */
215 	for (i = 0, start = 0; i < type; i++)
216 		start += lfs_res_qty[i];
217 	while (r == NULL) {
218 		for (i = 0; i < lfs_res_qty[type]; i++) {
219 			if (fs->lfs_resblk[start + i].inuse == 0) {
220 				re = fs->lfs_resblk + start + i;
221 				re->inuse = 1;
222 				r = re->p;
223 				KASSERT(re->size >= size);
224 				h = lfs_mhash(r);
225 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
226 				mutex_exit(&lfs_lock);
227 				return r;
228 			}
229 		}
230 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
231 		      lfs_res_names[type], lfs_res_qty[type]));
232 		mtsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
233 			&lfs_lock);
234 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
235 		      lfs_res_names[type]));
236 	}
237 	/* NOTREACHED */
238 	mutex_exit(&lfs_lock);
239 	return r;
240 }
241 
242 void
lfs_free(struct lfs * fs,void * p,int type)243 lfs_free(struct lfs *fs, void *p, int type)
244 {
245 	unsigned int h;
246 	res_t *re;
247 
248 	ASSERT_MAYBE_SEGLOCK(fs);
249 	h = lfs_mhash(p);
250 	mutex_enter(&lfs_lock);
251 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
252 		if (re->p == p) {
253 			KASSERT(re->inuse == 1);
254 			LIST_REMOVE(re, res);
255 			re->inuse = 0;
256 			wakeup(&fs->lfs_resblk);
257 			mutex_exit(&lfs_lock);
258 			return;
259 		}
260 	}
261 
262 #ifdef notyet /* XXX this assert fires */
263 	for (int i = 0; i < LFS_N_TOTAL; i++) {
264 		KDASSERTMSG(fs->lfs_resblk[i].p == p,
265 		    "lfs_free: inconsistent reserved block");
266 	}
267 #endif
268 
269 	mutex_exit(&lfs_lock);
270 
271 	/*
272 	 * If we didn't find it, free it.
273 	 */
274 	free(p, M_SEGMENT);
275 }
276 
277 /*
278  * lfs_seglock --
279  *	Single thread the segment writer.
280  */
281 int
lfs_seglock(struct lfs * fs,unsigned long flags)282 lfs_seglock(struct lfs *fs, unsigned long flags)
283 {
284 	struct segment *sp;
285 
286 	mutex_enter(&lfs_lock);
287 	if (fs->lfs_seglock) {
288 		if (fs->lfs_lockpid == curproc->p_pid &&
289 		    fs->lfs_locklwp == curlwp->l_lid) {
290 			++fs->lfs_seglock;
291 			fs->lfs_sp->seg_flags |= flags;
292 			mutex_exit(&lfs_lock);
293 			return 0;
294 		} else if (flags & SEGM_PAGEDAEMON) {
295 			mutex_exit(&lfs_lock);
296 			return EWOULDBLOCK;
297 		} else {
298 			while (fs->lfs_seglock) {
299 				(void)mtsleep(&fs->lfs_seglock, PRIBIO + 1,
300 					"lfs_seglock", 0, &lfs_lock);
301 			}
302 		}
303 	}
304 
305 	fs->lfs_seglock = 1;
306 	fs->lfs_lockpid = curproc->p_pid;
307 	fs->lfs_locklwp = curlwp->l_lid;
308 	mutex_exit(&lfs_lock);
309 	fs->lfs_cleanind = 0;
310 
311 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
312 
313 	/* Drain fragment size changes out */
314 	rw_enter(&fs->lfs_fraglock, RW_WRITER);
315 
316 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
317 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
318 	sp->seg_flags = flags;
319 	sp->vp = NULL;
320 	sp->seg_iocount = 0;
321 	(void) lfs_initseg(fs);
322 
323 	/*
324 	 * Keep a cumulative count of the outstanding I/O operations.  If the
325 	 * disk drive catches up with us it could go to zero before we finish,
326 	 * so we artificially increment it by one until we've scheduled all of
327 	 * the writes we intend to do.
328 	 */
329 	mutex_enter(&lfs_lock);
330 	++fs->lfs_iocount;
331 	fs->lfs_startseg = lfs_sb_getcurseg(fs);
332 	mutex_exit(&lfs_lock);
333 	return 0;
334 }
335 
336 static void lfs_unmark_dirop(struct lfs *);
337 
338 static void
lfs_unmark_dirop(struct lfs * fs)339 lfs_unmark_dirop(struct lfs *fs)
340 {
341 	struct inode *ip, *marker;
342 	struct vnode *vp;
343 	int doit;
344 
345 	ASSERT_NO_SEGLOCK(fs);
346 	mutex_enter(&lfs_lock);
347 	doit = !(fs->lfs_flags & LFS_UNDIROP);
348 	if (doit)
349 		fs->lfs_flags |= LFS_UNDIROP;
350 	mutex_exit(&lfs_lock);
351 
352 	if (!doit)
353 		return;
354 
355 	marker = pool_get(&lfs_inode_pool, PR_WAITOK);
356 	KASSERT(fs != NULL);
357 	memset(marker, 0, sizeof(*marker));
358 	marker->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK);
359 	memset(marker->inode_ext.lfs, 0, sizeof(*marker->inode_ext.lfs));
360 	marker->i_state |= IN_MARKER;
361 
362 	mutex_enter(&lfs_lock);
363 	TAILQ_INSERT_HEAD(&fs->lfs_dchainhd, marker, i_lfs_dchain);
364 	while ((ip = TAILQ_NEXT(marker, i_lfs_dchain)) != NULL) {
365 		TAILQ_REMOVE(&fs->lfs_dchainhd, marker, i_lfs_dchain);
366 		TAILQ_INSERT_AFTER(&fs->lfs_dchainhd, ip, marker,
367 		    i_lfs_dchain);
368 		if (ip->i_state & IN_MARKER)
369 			continue;
370 		vp = ITOV(ip);
371 		if ((ip->i_state & (IN_ADIROP | IN_CDIROP)) == IN_CDIROP) {
372 			--lfs_dirvcount;
373 			--fs->lfs_dirvcount;
374 			vp->v_uflag &= ~VU_DIROP;
375 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
376 			wakeup(&lfs_dirvcount);
377 			fs->lfs_unlockvp = vp;
378 			mutex_exit(&lfs_lock);
379 			vrele(vp);
380 			mutex_enter(&lfs_lock);
381 			fs->lfs_unlockvp = NULL;
382 			ip->i_state &= ~IN_CDIROP;
383 		}
384 	}
385 	TAILQ_REMOVE(&fs->lfs_dchainhd, marker, i_lfs_dchain);
386 	fs->lfs_flags &= ~LFS_UNDIROP;
387 	wakeup(&fs->lfs_flags);
388 	mutex_exit(&lfs_lock);
389 
390 	pool_put(&lfs_inoext_pool, marker->inode_ext.lfs);
391 	pool_put(&lfs_inode_pool, marker);
392 }
393 
394 static void
lfs_auto_segclean(struct lfs * fs)395 lfs_auto_segclean(struct lfs *fs)
396 {
397 	int i, error, waited;
398 
399 	ASSERT_SEGLOCK(fs);
400 	/*
401 	 * Now that we've swapped lfs_activesb, but while we still
402 	 * hold the segment lock, run through the segment list marking
403 	 * the empty ones clean.
404 	 * XXX - do we really need to do them all at once?
405 	 */
406 	waited = 0;
407 	for (i = 0; i < lfs_sb_getnseg(fs); i++) {
408 		if ((fs->lfs_suflags[0][i] &
409 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
410 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
411 		    (fs->lfs_suflags[1][i] &
412 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
413 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
414 
415 			/* Make sure the sb is written before we clean */
416 			mutex_enter(&lfs_lock);
417 			while (waited == 0 && fs->lfs_sbactive)
418 				mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
419 					0, &lfs_lock);
420 			mutex_exit(&lfs_lock);
421 			waited = 1;
422 
423 			if ((error = lfs_do_segclean(fs, i)) != 0) {
424 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
425 			}
426 		}
427 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
428 			fs->lfs_suflags[fs->lfs_activesb][i];
429 	}
430 }
431 
432 /*
433  * lfs_segunlock --
434  *	Single thread the segment writer.
435  */
436 void
lfs_segunlock(struct lfs * fs)437 lfs_segunlock(struct lfs *fs)
438 {
439 	struct segment *sp;
440 	unsigned long sync, ckp;
441 	struct buf *bp;
442 	int do_unmark_dirop = 0;
443 
444 	sp = fs->lfs_sp;
445 
446 	mutex_enter(&lfs_lock);
447 
448 	if (!LFS_SEGLOCK_HELD(fs))
449 		panic("lfs seglock not held");
450 
451 	if (fs->lfs_seglock == 1) {
452 		if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0)
453 			do_unmark_dirop = 1;
454 		mutex_exit(&lfs_lock);
455 		sync = sp->seg_flags & SEGM_SYNC;
456 		ckp = sp->seg_flags & SEGM_CKP;
457 
458 		/* We should have a segment summary, and nothing else */
459 		KASSERT(sp->cbpp == sp->bpp + 1);
460 
461 		/* Free allocated segment summary */
462 		lfs_sb_suboffset(fs, lfs_btofsb(fs, lfs_sb_getsumsize(fs)));
463 		bp = *sp->bpp;
464 		lfs_freebuf(fs, bp);
465 
466 		pool_put(&fs->lfs_bpppool, sp->bpp);
467 		sp->bpp = NULL;
468 
469 		/*
470 		 * If we're not sync, we're done with sp, get rid of it.
471 		 * Otherwise, we keep a local copy around but free
472 		 * fs->lfs_sp so another process can use it (we have to
473 		 * wait but they don't have to wait for us).
474 		 */
475 		if (!sync)
476 			pool_put(&fs->lfs_segpool, sp);
477 		fs->lfs_sp = NULL;
478 
479 		/*
480 		 * If the I/O count is non-zero, sleep until it reaches zero.
481 		 * At the moment, the user's process hangs around so we can
482 		 * sleep.
483 		 */
484 		mutex_enter(&lfs_lock);
485 		if (--fs->lfs_iocount <= 1)
486 			wakeup(&fs->lfs_iocount);
487 		mutex_exit(&lfs_lock);
488 
489 		/*
490 		 * If we're not checkpointing, we don't have to block
491 		 * other processes to wait for a synchronous write
492 		 * to complete.
493 		 */
494 		if (!ckp) {
495 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
496 
497 			mutex_enter(&lfs_lock);
498 			--fs->lfs_seglock;
499 			fs->lfs_lockpid = 0;
500 			fs->lfs_locklwp = 0;
501 			mutex_exit(&lfs_lock);
502 			wakeup(&fs->lfs_seglock);
503 		}
504 		/*
505 		 * We let checkpoints happen asynchronously.  That means
506 		 * that during recovery, we have to roll forward between
507 		 * the two segments described by the first and second
508 		 * superblocks to make sure that the checkpoint described
509 		 * by a superblock completed.
510 		 */
511 		mutex_enter(&lfs_lock);
512 		while (ckp && sync && fs->lfs_iocount) {
513 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
514 				      "lfs_iocount", 0, &lfs_lock);
515 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", fs, fs->lfs_iocount));
516 		}
517 		while (sync && sp->seg_iocount) {
518 			(void)mtsleep(&sp->seg_iocount, PRIBIO + 1,
519 				     "seg_iocount", 0, &lfs_lock);
520 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
521 		}
522 		mutex_exit(&lfs_lock);
523 		if (sync)
524 			pool_put(&fs->lfs_segpool, sp);
525 
526 		if (ckp) {
527 			fs->lfs_nactive = 0;
528 			/* If we *know* everything's on disk, write both sbs */
529 			/* XXX should wait for this one	 */
530 			if (sync)
531 				lfs_writesuper(fs, lfs_sb_getsboff(fs, fs->lfs_activesb));
532 			lfs_writesuper(fs, lfs_sb_getsboff(fs, 1 - fs->lfs_activesb));
533 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
534 				lfs_auto_segclean(fs);
535 				/* If sync, we can clean the remainder too */
536 				if (sync)
537 					lfs_auto_segclean(fs);
538 			}
539 			fs->lfs_activesb = 1 - fs->lfs_activesb;
540 
541 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
542 
543 			mutex_enter(&lfs_lock);
544 			--fs->lfs_seglock;
545 			fs->lfs_lockpid = 0;
546 			fs->lfs_locklwp = 0;
547 			mutex_exit(&lfs_lock);
548 			wakeup(&fs->lfs_seglock);
549 		}
550 		/* Reenable fragment size changes */
551 		rw_exit(&fs->lfs_fraglock);
552 		if (do_unmark_dirop)
553 			lfs_unmark_dirop(fs);
554 	} else {
555 		--fs->lfs_seglock;
556 		KASSERT(fs->lfs_seglock != 0);
557 		mutex_exit(&lfs_lock);
558 	}
559 }
560 
561 /*
562  * Drain dirops and start writer.
563  *
564  * No simple_locks are held when we enter and none are held when we return.
565  */
566 void
lfs_writer_enter(struct lfs * fs,const char * wmesg)567 lfs_writer_enter(struct lfs *fs, const char *wmesg)
568 {
569 	int error __diagused;
570 
571 	ASSERT_NO_SEGLOCK(fs);
572 	mutex_enter(&lfs_lock);
573 
574 	/* disallow dirops during flush */
575 	fs->lfs_writer++;
576 
577 	while (fs->lfs_dirops > 0) {
578 		++fs->lfs_diropwait;
579 		error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
580 				&lfs_lock);
581 		KASSERT(error == 0);
582 		--fs->lfs_diropwait;
583 	}
584 
585 	mutex_exit(&lfs_lock);
586 }
587 
588 int
lfs_writer_tryenter(struct lfs * fs)589 lfs_writer_tryenter(struct lfs *fs)
590 {
591 	int writer_set;
592 
593 	ASSERT_MAYBE_SEGLOCK(fs);
594 	mutex_enter(&lfs_lock);
595 	writer_set = (fs->lfs_dirops == 0);
596 	if (writer_set)
597 		fs->lfs_writer++;
598 	mutex_exit(&lfs_lock);
599 
600 	return writer_set;
601 }
602 
603 void
lfs_writer_leave(struct lfs * fs)604 lfs_writer_leave(struct lfs *fs)
605 {
606 	bool dowakeup;
607 
608 	ASSERT_MAYBE_SEGLOCK(fs);
609 	mutex_enter(&lfs_lock);
610 	dowakeup = !(--fs->lfs_writer);
611 	if (dowakeup)
612 		cv_broadcast(&fs->lfs_diropscv);
613 	mutex_exit(&lfs_lock);
614 }
615 
616 /*
617  * Unlock, wait for the cleaner, then relock to where we were before.
618  * To be used only at a fairly high level, to address a paucity of free
619  * segments propagated back from lfs_gop_write().
620  */
621 void
lfs_segunlock_relock(struct lfs * fs)622 lfs_segunlock_relock(struct lfs *fs)
623 {
624 	int n = fs->lfs_seglock;
625 	u_int16_t seg_flags;
626 	CLEANERINFO *cip;
627 	struct buf *bp;
628 
629 	if (n == 0)
630 		return;
631 
632 	/* Write anything we've already gathered to disk */
633 	lfs_writeseg(fs, fs->lfs_sp);
634 
635 	/* Tell cleaner */
636 	LFS_CLEANERINFO(cip, fs, bp);
637 	lfs_ci_setflags(fs, cip,
638 			lfs_ci_getflags(fs, cip) | LFS_CLEANER_MUST_CLEAN);
639 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
640 
641 	/* Save segment flags for later */
642 	seg_flags = fs->lfs_sp->seg_flags;
643 
644 	fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */
645 	while(fs->lfs_seglock)
646 		lfs_segunlock(fs);
647 
648 	/* Wait for the cleaner */
649 	lfs_wakeup_cleaner(fs);
650 	mutex_enter(&lfs_lock);
651 	while (LFS_STARVED_FOR_SEGS(fs))
652 		mtsleep(&fs->lfs_availsleep, PRIBIO, "relock", 0,
653 			&lfs_lock);
654 	mutex_exit(&lfs_lock);
655 
656 	/* Put the segment lock back the way it was. */
657 	while(n--)
658 		lfs_seglock(fs, seg_flags);
659 
660 	/* Cleaner can relax now */
661 	LFS_CLEANERINFO(cip, fs, bp);
662 	lfs_ci_setflags(fs, cip,
663 			lfs_ci_getflags(fs, cip) & ~LFS_CLEANER_MUST_CLEAN);
664 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
665 
666 	return;
667 }
668 
669 /*
670  * Wake up the cleaner, provided that nowrap is not set.
671  */
672 void
lfs_wakeup_cleaner(struct lfs * fs)673 lfs_wakeup_cleaner(struct lfs *fs)
674 {
675 	if (fs->lfs_nowrap > 0)
676 		return;
677 
678 	cv_broadcast(&fs->lfs_nextsegsleep);
679 	cv_broadcast(&lfs_allclean_wakeup);
680 }
681