Home | History | Annotate | Line # | Download | only in lfs
lfs_subr.c revision 1.93
      1 /*	$NetBSD: lfs_subr.c,v 1.93 2017/06/08 01:23:01 chs 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 (at) 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 1.93 2017/06/08 01:23:01 chs 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 #include <uvm/uvm.h>
     82 
     83 #ifdef DEBUG
     84 const char *lfs_res_names[LFS_NB_COUNT] = {
     85 	"summary",
     86 	"superblock",
     87 	"file block",
     88 	"cluster",
     89 	"clean",
     90 	"blkiov",
     91 };
     92 #endif
     93 
     94 int lfs_res_qty[LFS_NB_COUNT] = {
     95 	LFS_N_SUMMARIES,
     96 	LFS_N_SBLOCKS,
     97 	LFS_N_IBLOCKS,
     98 	LFS_N_CLUSTERS,
     99 	LFS_N_CLEAN,
    100 	LFS_N_BLKIOV,
    101 };
    102 
    103 void
    104 lfs_setup_resblks(struct lfs *fs)
    105 {
    106 	int i, j;
    107 	int maxbpp;
    108 
    109 	ASSERT_NO_SEGLOCK(fs);
    110 	fs->lfs_resblk = malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
    111 				M_WAITOK);
    112 	for (i = 0; i < LFS_N_TOTAL; i++) {
    113 		fs->lfs_resblk[i].inuse = 0;
    114 		fs->lfs_resblk[i].p = NULL;
    115 	}
    116 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
    117 		LIST_INIT(fs->lfs_reshash + i);
    118 
    119 	/*
    120 	 * These types of allocations can be larger than a page,
    121 	 * so we can't use the pool subsystem for them.
    122 	 */
    123 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
    124 		fs->lfs_resblk[i].size = lfs_sb_getsumsize(fs);
    125 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
    126 		fs->lfs_resblk[i].size = LFS_SBPAD;
    127 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
    128 		fs->lfs_resblk[i].size = lfs_sb_getbsize(fs);
    129 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
    130 		fs->lfs_resblk[i].size = MAXPHYS;
    131 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
    132 		fs->lfs_resblk[i].size = MAXPHYS;
    133 	for (j = 0; j < LFS_N_BLKIOV; j++, i++)
    134 		fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO);
    135 
    136 	for (i = 0; i < LFS_N_TOTAL; i++) {
    137 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
    138 					     M_SEGMENT, M_WAITOK);
    139 	}
    140 
    141 	/*
    142 	 * Initialize pools for small types (XXX is BPP small?)
    143 	 */
    144 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
    145 		"lfsclpl", &pool_allocator_nointr, IPL_NONE);
    146 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
    147 		"lfssegpool", &pool_allocator_nointr, IPL_NONE);
    148 	/* XXX: should this int32 be 32/64? */
    149 	maxbpp = ((lfs_sb_getsumsize(fs) - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
    150 	maxbpp = MIN(maxbpp, lfs_segsize(fs) / lfs_sb_getfsize(fs) + 2);
    151 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
    152 		"lfsbpppl", &pool_allocator_nointr, IPL_NONE);
    153 }
    154 
    155 void
    156 lfs_free_resblks(struct lfs *fs)
    157 {
    158 	int i;
    159 
    160 	pool_destroy(&fs->lfs_bpppool);
    161 	pool_destroy(&fs->lfs_segpool);
    162 	pool_destroy(&fs->lfs_clpool);
    163 
    164 	mutex_enter(&lfs_lock);
    165 	for (i = 0; i < LFS_N_TOTAL; i++) {
    166 		while (fs->lfs_resblk[i].inuse)
    167 			mtsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
    168 				&lfs_lock);
    169 		if (fs->lfs_resblk[i].p != NULL)
    170 			free(fs->lfs_resblk[i].p, M_SEGMENT);
    171 	}
    172 	free(fs->lfs_resblk, M_SEGMENT);
    173 	mutex_exit(&lfs_lock);
    174 }
    175 
    176 static unsigned int
    177 lfs_mhash(void *vp)
    178 {
    179 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
    180 }
    181 
    182 /*
    183  * Return memory of the given size for the given purpose, or use one of a
    184  * number of spare last-resort buffers, if malloc returns NULL.
    185  */
    186 void *
    187 lfs_malloc(struct lfs *fs, size_t size, int type)
    188 {
    189 	struct lfs_res_blk *re;
    190 	void *r;
    191 	int i, start;
    192 	unsigned int h;
    193 
    194 	ASSERT_MAYBE_SEGLOCK(fs);
    195 	r = NULL;
    196 
    197 	/* If no mem allocated for this type, it just waits */
    198 	if (lfs_res_qty[type] == 0) {
    199 		r = malloc(size, M_SEGMENT, M_WAITOK);
    200 		return r;
    201 	}
    202 
    203 	/* Otherwise try a quick malloc, and if it works, great */
    204 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
    205 		return r;
    206 	}
    207 
    208 	/*
    209 	 * If malloc returned NULL, we are forced to use one of our
    210 	 * reserve blocks.  We have on hand at least one summary block,
    211 	 * at least one cluster block, at least one superblock,
    212 	 * and several indirect blocks.
    213 	 */
    214 
    215 	mutex_enter(&lfs_lock);
    216 	/* skip over blocks of other types */
    217 	for (i = 0, start = 0; i < type; i++)
    218 		start += lfs_res_qty[i];
    219 	while (r == NULL) {
    220 		for (i = 0; i < lfs_res_qty[type]; i++) {
    221 			if (fs->lfs_resblk[start + i].inuse == 0) {
    222 				re = fs->lfs_resblk + start + i;
    223 				re->inuse = 1;
    224 				r = re->p;
    225 				KASSERT(re->size >= size);
    226 				h = lfs_mhash(r);
    227 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
    228 				mutex_exit(&lfs_lock);
    229 				return r;
    230 			}
    231 		}
    232 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
    233 		      lfs_res_names[type], lfs_res_qty[type]));
    234 		mtsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
    235 			&lfs_lock);
    236 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
    237 		      lfs_res_names[type]));
    238 	}
    239 	/* NOTREACHED */
    240 	mutex_exit(&lfs_lock);
    241 	return r;
    242 }
    243 
    244 void
    245 lfs_free(struct lfs *fs, void *p, int type)
    246 {
    247 	unsigned int h;
    248 	res_t *re;
    249 
    250 	ASSERT_MAYBE_SEGLOCK(fs);
    251 	h = lfs_mhash(p);
    252 	mutex_enter(&lfs_lock);
    253 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
    254 		if (re->p == p) {
    255 			KASSERT(re->inuse == 1);
    256 			LIST_REMOVE(re, res);
    257 			re->inuse = 0;
    258 			wakeup(&fs->lfs_resblk);
    259 			mutex_exit(&lfs_lock);
    260 			return;
    261 		}
    262 	}
    263 
    264 	for (int i = 0; i < LFS_N_TOTAL; i++) {
    265 		KDASSERTMSG(fs->lfs_resblk[i].p == p,
    266 		    "lfs_free: inconsistent reserved block");
    267 	}
    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
    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
    339 lfs_unmark_dirop(struct lfs *fs)
    340 {
    341 	struct inode *ip, *nip;
    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 	if (!doit) {
    351 		mutex_exit(&lfs_lock);
    352 		return;
    353 	}
    354 
    355 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
    356 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
    357 		vp = ITOV(ip);
    358 		if ((ip->i_flag & (IN_ADIROP | IN_CDIROP)) == IN_CDIROP) {
    359 			--lfs_dirvcount;
    360 			--fs->lfs_dirvcount;
    361 			vp->v_uflag &= ~VU_DIROP;
    362 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
    363 			wakeup(&lfs_dirvcount);
    364 			fs->lfs_unlockvp = vp;
    365 			mutex_exit(&lfs_lock);
    366 			vrele(vp);
    367 			mutex_enter(&lfs_lock);
    368 			fs->lfs_unlockvp = NULL;
    369 			ip->i_flag &= ~IN_CDIROP;
    370 		}
    371 	}
    372 
    373 	fs->lfs_flags &= ~LFS_UNDIROP;
    374 	wakeup(&fs->lfs_flags);
    375 	mutex_exit(&lfs_lock);
    376 }
    377 
    378 static void
    379 lfs_auto_segclean(struct lfs *fs)
    380 {
    381 	int i, error, waited;
    382 
    383 	ASSERT_SEGLOCK(fs);
    384 	/*
    385 	 * Now that we've swapped lfs_activesb, but while we still
    386 	 * hold the segment lock, run through the segment list marking
    387 	 * the empty ones clean.
    388 	 * XXX - do we really need to do them all at once?
    389 	 */
    390 	waited = 0;
    391 	for (i = 0; i < lfs_sb_getnseg(fs); i++) {
    392 		if ((fs->lfs_suflags[0][i] &
    393 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    394 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
    395 		    (fs->lfs_suflags[1][i] &
    396 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    397 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
    398 
    399 			/* Make sure the sb is written before we clean */
    400 			mutex_enter(&lfs_lock);
    401 			while (waited == 0 && fs->lfs_sbactive)
    402 				mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
    403 					0, &lfs_lock);
    404 			mutex_exit(&lfs_lock);
    405 			waited = 1;
    406 
    407 			if ((error = lfs_do_segclean(fs, i)) != 0) {
    408 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
    409 			}
    410 		}
    411 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
    412 			fs->lfs_suflags[fs->lfs_activesb][i];
    413 	}
    414 }
    415 
    416 /*
    417  * lfs_segunlock --
    418  *	Single thread the segment writer.
    419  */
    420 void
    421 lfs_segunlock(struct lfs *fs)
    422 {
    423 	struct segment *sp;
    424 	unsigned long sync, ckp;
    425 	struct buf *bp;
    426 	int do_unmark_dirop = 0;
    427 
    428 	sp = fs->lfs_sp;
    429 
    430 	mutex_enter(&lfs_lock);
    431 	KASSERT(LFS_SEGLOCK_HELD(fs));
    432 	if (fs->lfs_seglock == 1) {
    433 		if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0)
    434 			do_unmark_dirop = 1;
    435 		mutex_exit(&lfs_lock);
    436 		sync = sp->seg_flags & SEGM_SYNC;
    437 		ckp = sp->seg_flags & SEGM_CKP;
    438 
    439 		/* We should have a segment summary, and nothing else */
    440 		KASSERT(sp->cbpp == sp->bpp + 1);
    441 
    442 		/* Free allocated segment summary */
    443 		lfs_sb_suboffset(fs, lfs_btofsb(fs, lfs_sb_getsumsize(fs)));
    444 		bp = *sp->bpp;
    445 		lfs_freebuf(fs, bp);
    446 
    447 		pool_put(&fs->lfs_bpppool, sp->bpp);
    448 		sp->bpp = NULL;
    449 
    450 		/*
    451 		 * If we're not sync, we're done with sp, get rid of it.
    452 		 * Otherwise, we keep a local copy around but free
    453 		 * fs->lfs_sp so another process can use it (we have to
    454 		 * wait but they don't have to wait for us).
    455 		 */
    456 		if (!sync)
    457 			pool_put(&fs->lfs_segpool, sp);
    458 		fs->lfs_sp = NULL;
    459 
    460 		/*
    461 		 * If the I/O count is non-zero, sleep until it reaches zero.
    462 		 * At the moment, the user's process hangs around so we can
    463 		 * sleep.
    464 		 */
    465 		mutex_enter(&lfs_lock);
    466 		if (--fs->lfs_iocount <= 1)
    467 			wakeup(&fs->lfs_iocount);
    468 		mutex_exit(&lfs_lock);
    469 
    470 		/*
    471 		 * If we're not checkpointing, we don't have to block
    472 		 * other processes to wait for a synchronous write
    473 		 * to complete.
    474 		 */
    475 		if (!ckp) {
    476 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    477 
    478 			mutex_enter(&lfs_lock);
    479 			--fs->lfs_seglock;
    480 			fs->lfs_lockpid = 0;
    481 			fs->lfs_locklwp = 0;
    482 			mutex_exit(&lfs_lock);
    483 			wakeup(&fs->lfs_seglock);
    484 		}
    485 		/*
    486 		 * We let checkpoints happen asynchronously.  That means
    487 		 * that during recovery, we have to roll forward between
    488 		 * the two segments described by the first and second
    489 		 * superblocks to make sure that the checkpoint described
    490 		 * by a superblock completed.
    491 		 */
    492 		mutex_enter(&lfs_lock);
    493 		while (ckp && sync && fs->lfs_iocount) {
    494 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
    495 				      "lfs_iocount", 0, &lfs_lock);
    496 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", fs, fs->lfs_iocount));
    497 		}
    498 		while (sync && sp->seg_iocount) {
    499 			(void)mtsleep(&sp->seg_iocount, PRIBIO + 1,
    500 				     "seg_iocount", 0, &lfs_lock);
    501 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
    502 		}
    503 		mutex_exit(&lfs_lock);
    504 		if (sync)
    505 			pool_put(&fs->lfs_segpool, sp);
    506 
    507 		if (ckp) {
    508 			fs->lfs_nactive = 0;
    509 			/* If we *know* everything's on disk, write both sbs */
    510 			/* XXX should wait for this one	 */
    511 			if (sync)
    512 				lfs_writesuper(fs, lfs_sb_getsboff(fs, fs->lfs_activesb));
    513 			lfs_writesuper(fs, lfs_sb_getsboff(fs, 1 - fs->lfs_activesb));
    514 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
    515 				lfs_auto_segclean(fs);
    516 				/* If sync, we can clean the remainder too */
    517 				if (sync)
    518 					lfs_auto_segclean(fs);
    519 			}
    520 			fs->lfs_activesb = 1 - fs->lfs_activesb;
    521 
    522 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    523 
    524 			mutex_enter(&lfs_lock);
    525 			--fs->lfs_seglock;
    526 			fs->lfs_lockpid = 0;
    527 			fs->lfs_locklwp = 0;
    528 			mutex_exit(&lfs_lock);
    529 			wakeup(&fs->lfs_seglock);
    530 		}
    531 		/* Reenable fragment size changes */
    532 		rw_exit(&fs->lfs_fraglock);
    533 		if (do_unmark_dirop)
    534 			lfs_unmark_dirop(fs);
    535 	} else if (fs->lfs_seglock == 0) {
    536 		mutex_exit(&lfs_lock);
    537 		panic ("Seglock not held");
    538 	} else {
    539 		--fs->lfs_seglock;
    540 		mutex_exit(&lfs_lock);
    541 	}
    542 }
    543 
    544 /*
    545  * Drain dirops and start writer.
    546  *
    547  * No simple_locks are held when we enter and none are held when we return.
    548  */
    549 int
    550 lfs_writer_enter(struct lfs *fs, const char *wmesg)
    551 {
    552 	int error = 0;
    553 
    554 	ASSERT_MAYBE_SEGLOCK(fs);
    555 	mutex_enter(&lfs_lock);
    556 
    557 	/* disallow dirops during flush */
    558 	fs->lfs_writer++;
    559 
    560 	while (fs->lfs_dirops > 0) {
    561 		++fs->lfs_diropwait;
    562 		error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
    563 				&lfs_lock);
    564 		--fs->lfs_diropwait;
    565 	}
    566 
    567 	if (error)
    568 		fs->lfs_writer--;
    569 
    570 	mutex_exit(&lfs_lock);
    571 
    572 	return error;
    573 }
    574 
    575 void
    576 lfs_writer_leave(struct lfs *fs)
    577 {
    578 	bool dowakeup;
    579 
    580 	ASSERT_MAYBE_SEGLOCK(fs);
    581 	mutex_enter(&lfs_lock);
    582 	dowakeup = !(--fs->lfs_writer);
    583 	if (dowakeup)
    584 		cv_broadcast(&fs->lfs_diropscv);
    585 	mutex_exit(&lfs_lock);
    586 }
    587 
    588 /*
    589  * Unlock, wait for the cleaner, then relock to where we were before.
    590  * To be used only at a fairly high level, to address a paucity of free
    591  * segments propagated back from lfs_gop_write().
    592  */
    593 void
    594 lfs_segunlock_relock(struct lfs *fs)
    595 {
    596 	int n = fs->lfs_seglock;
    597 	u_int16_t seg_flags;
    598 	CLEANERINFO *cip;
    599 	struct buf *bp;
    600 
    601 	if (n == 0)
    602 		return;
    603 
    604 	/* Write anything we've already gathered to disk */
    605 	lfs_writeseg(fs, fs->lfs_sp);
    606 
    607 	/* Tell cleaner */
    608 	LFS_CLEANERINFO(cip, fs, bp);
    609 	lfs_ci_setflags(fs, cip,
    610 			lfs_ci_getflags(fs, cip) | LFS_CLEANER_MUST_CLEAN);
    611 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    612 
    613 	/* Save segment flags for later */
    614 	seg_flags = fs->lfs_sp->seg_flags;
    615 
    616 	fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */
    617 	while(fs->lfs_seglock)
    618 		lfs_segunlock(fs);
    619 
    620 	/* Wait for the cleaner */
    621 	lfs_wakeup_cleaner(fs);
    622 	mutex_enter(&lfs_lock);
    623 	while (LFS_STARVED_FOR_SEGS(fs))
    624 		mtsleep(&fs->lfs_availsleep, PRIBIO, "relock", 0,
    625 			&lfs_lock);
    626 	mutex_exit(&lfs_lock);
    627 
    628 	/* Put the segment lock back the way it was. */
    629 	while(n--)
    630 		lfs_seglock(fs, seg_flags);
    631 
    632 	/* Cleaner can relax now */
    633 	LFS_CLEANERINFO(cip, fs, bp);
    634 	lfs_ci_setflags(fs, cip,
    635 			lfs_ci_getflags(fs, cip) & ~LFS_CLEANER_MUST_CLEAN);
    636 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    637 
    638 	return;
    639 }
    640 
    641 /*
    642  * Wake up the cleaner, provided that nowrap is not set.
    643  */
    644 void
    645 lfs_wakeup_cleaner(struct lfs *fs)
    646 {
    647 	if (fs->lfs_nowrap > 0)
    648 		return;
    649 
    650 	wakeup(&fs->lfs_nextsegsleep);
    651 	wakeup(&lfs_allclean_wakeup);
    652 }
    653