Home | History | Annotate | Line # | Download | only in lfs
lfs_subr.c revision 1.80
      1 /*	$NetBSD: lfs_subr.c,v 1.80 2013/07/28 01:05:52 dholland 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 1.80 2013/07/28 01:05:52 dholland 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_kernel.h>
     78 #include <ufs/lfs/lfs_extern.h>
     79 
     80 #include <uvm/uvm.h>
     81 
     82 #ifdef DEBUG
     83 const char *lfs_res_names[LFS_NB_COUNT] = {
     84 	"summary",
     85 	"superblock",
     86 	"file block",
     87 	"cluster",
     88 	"clean",
     89 	"blkiov",
     90 };
     91 #endif
     92 
     93 int lfs_res_qty[LFS_NB_COUNT] = {
     94 	LFS_N_SUMMARIES,
     95 	LFS_N_SBLOCKS,
     96 	LFS_N_IBLOCKS,
     97 	LFS_N_CLUSTERS,
     98 	LFS_N_CLEAN,
     99 	LFS_N_BLKIOV,
    100 };
    101 
    102 void
    103 lfs_setup_resblks(struct lfs *fs)
    104 {
    105 	int i, j;
    106 	int maxbpp;
    107 
    108 	ASSERT_NO_SEGLOCK(fs);
    109 	fs->lfs_resblk = (res_t *)malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
    110 					  M_WAITOK);
    111 	for (i = 0; i < LFS_N_TOTAL; i++) {
    112 		fs->lfs_resblk[i].inuse = 0;
    113 		fs->lfs_resblk[i].p = NULL;
    114 	}
    115 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
    116 		LIST_INIT(fs->lfs_reshash + i);
    117 
    118 	/*
    119 	 * These types of allocations can be larger than a page,
    120 	 * so we can't use the pool subsystem for them.
    121 	 */
    122 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
    123 		fs->lfs_resblk[i].size = fs->lfs_sumsize;
    124 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
    125 		fs->lfs_resblk[i].size = LFS_SBPAD;
    126 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
    127 		fs->lfs_resblk[i].size = fs->lfs_bsize;
    128 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
    129 		fs->lfs_resblk[i].size = MAXPHYS;
    130 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
    131 		fs->lfs_resblk[i].size = MAXPHYS;
    132 	for (j = 0; j < LFS_N_BLKIOV; j++, i++)
    133 		fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO);
    134 
    135 	for (i = 0; i < LFS_N_TOTAL; i++) {
    136 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
    137 					     M_SEGMENT, M_WAITOK);
    138 	}
    139 
    140 	/*
    141 	 * Initialize pools for small types (XXX is BPP small?)
    142 	 */
    143 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
    144 		"lfsclpl", &pool_allocator_nointr, IPL_NONE);
    145 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
    146 		"lfssegpool", &pool_allocator_nointr, IPL_NONE);
    147 	maxbpp = ((fs->lfs_sumsize - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
    148 	maxbpp = MIN(maxbpp, lfs_segsize(fs) / fs->lfs_fsize + 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
    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
    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 *
    185 lfs_malloc(struct lfs *fs, size_t size, int type)
    186 {
    187 	struct lfs_res_blk *re;
    188 	void *r;
    189 	int i, s, 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 				s = splbio();
    226 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
    227 				splx(s);
    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 	int s;
    248 	unsigned int h;
    249 	res_t *re;
    250 #ifdef DEBUG
    251 	int i;
    252 #endif
    253 
    254 	ASSERT_MAYBE_SEGLOCK(fs);
    255 	h = lfs_mhash(p);
    256 	mutex_enter(&lfs_lock);
    257 	s = splbio();
    258 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
    259 		if (re->p == p) {
    260 			KASSERT(re->inuse == 1);
    261 			LIST_REMOVE(re, res);
    262 			re->inuse = 0;
    263 			wakeup(&fs->lfs_resblk);
    264 			splx(s);
    265 			mutex_exit(&lfs_lock);
    266 			return;
    267 		}
    268 	}
    269 #ifdef DEBUG
    270 	for (i = 0; i < LFS_N_TOTAL; i++) {
    271 		if (fs->lfs_resblk[i].p == p)
    272 			panic("lfs_free: inconsistent reserved block");
    273 	}
    274 #endif
    275 	splx(s);
    276 	mutex_exit(&lfs_lock);
    277 
    278 	/*
    279 	 * If we didn't find it, free it.
    280 	 */
    281 	free(p, M_SEGMENT);
    282 }
    283 
    284 /*
    285  * lfs_seglock --
    286  *	Single thread the segment writer.
    287  */
    288 int
    289 lfs_seglock(struct lfs *fs, unsigned long flags)
    290 {
    291 	struct segment *sp;
    292 
    293 	mutex_enter(&lfs_lock);
    294 	if (fs->lfs_seglock) {
    295 		if (fs->lfs_lockpid == curproc->p_pid &&
    296 		    fs->lfs_locklwp == curlwp->l_lid) {
    297 			++fs->lfs_seglock;
    298 			fs->lfs_sp->seg_flags |= flags;
    299 			mutex_exit(&lfs_lock);
    300 			return 0;
    301 		} else if (flags & SEGM_PAGEDAEMON) {
    302 			mutex_exit(&lfs_lock);
    303 			return EWOULDBLOCK;
    304 		} else {
    305 			while (fs->lfs_seglock) {
    306 				(void)mtsleep(&fs->lfs_seglock, PRIBIO + 1,
    307 					"lfs_seglock", 0, &lfs_lock);
    308 			}
    309 		}
    310 	}
    311 
    312 	fs->lfs_seglock = 1;
    313 	fs->lfs_lockpid = curproc->p_pid;
    314 	fs->lfs_locklwp = curlwp->l_lid;
    315 	mutex_exit(&lfs_lock);
    316 	fs->lfs_cleanind = 0;
    317 
    318 #ifdef DEBUG
    319 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
    320 #endif
    321 	/* Drain fragment size changes out */
    322 	rw_enter(&fs->lfs_fraglock, RW_WRITER);
    323 
    324 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
    325 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
    326 	sp->seg_flags = flags;
    327 	sp->vp = NULL;
    328 	sp->seg_iocount = 0;
    329 	(void) lfs_initseg(fs);
    330 
    331 	/*
    332 	 * Keep a cumulative count of the outstanding I/O operations.  If the
    333 	 * disk drive catches up with us it could go to zero before we finish,
    334 	 * so we artificially increment it by one until we've scheduled all of
    335 	 * the writes we intend to do.
    336 	 */
    337 	mutex_enter(&lfs_lock);
    338 	++fs->lfs_iocount;
    339 	fs->lfs_startseg = fs->lfs_curseg;
    340 	mutex_exit(&lfs_lock);
    341 	return 0;
    342 }
    343 
    344 static void lfs_unmark_dirop(struct lfs *);
    345 
    346 static void
    347 lfs_unmark_dirop(struct lfs *fs)
    348 {
    349 	struct inode *ip, *nip;
    350 	struct vnode *vp;
    351 	int doit;
    352 
    353 	ASSERT_NO_SEGLOCK(fs);
    354 	mutex_enter(&lfs_lock);
    355 	doit = !(fs->lfs_flags & LFS_UNDIROP);
    356 	if (doit)
    357 		fs->lfs_flags |= LFS_UNDIROP;
    358 	if (!doit) {
    359 		mutex_exit(&lfs_lock);
    360 		return;
    361 	}
    362 
    363 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
    364 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
    365 		vp = ITOV(ip);
    366 		if ((ip->i_flag & (IN_ADIROP | IN_CDIROP)) == IN_CDIROP) {
    367 			--lfs_dirvcount;
    368 			--fs->lfs_dirvcount;
    369 			vp->v_uflag &= ~VU_DIROP;
    370 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
    371 			wakeup(&lfs_dirvcount);
    372 			fs->lfs_unlockvp = vp;
    373 			mutex_exit(&lfs_lock);
    374 			vrele(vp);
    375 			mutex_enter(&lfs_lock);
    376 			fs->lfs_unlockvp = NULL;
    377 			ip->i_flag &= ~IN_CDIROP;
    378 		}
    379 	}
    380 
    381 	fs->lfs_flags &= ~LFS_UNDIROP;
    382 	wakeup(&fs->lfs_flags);
    383 	mutex_exit(&lfs_lock);
    384 }
    385 
    386 static void
    387 lfs_auto_segclean(struct lfs *fs)
    388 {
    389 	int i, error, s, waited;
    390 
    391 	ASSERT_SEGLOCK(fs);
    392 	/*
    393 	 * Now that we've swapped lfs_activesb, but while we still
    394 	 * hold the segment lock, run through the segment list marking
    395 	 * the empty ones clean.
    396 	 * XXX - do we really need to do them all at once?
    397 	 */
    398 	waited = 0;
    399 	for (i = 0; i < fs->lfs_nseg; i++) {
    400 		if ((fs->lfs_suflags[0][i] &
    401 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    402 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
    403 		    (fs->lfs_suflags[1][i] &
    404 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    405 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
    406 
    407 			/* Make sure the sb is written before we clean */
    408 			mutex_enter(&lfs_lock);
    409 			s = splbio();
    410 			while (waited == 0 && fs->lfs_sbactive)
    411 				mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
    412 					0, &lfs_lock);
    413 			splx(s);
    414 			mutex_exit(&lfs_lock);
    415 			waited = 1;
    416 
    417 			if ((error = lfs_do_segclean(fs, i)) != 0) {
    418 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
    419 			}
    420 		}
    421 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
    422 			fs->lfs_suflags[fs->lfs_activesb][i];
    423 	}
    424 }
    425 
    426 /*
    427  * lfs_segunlock --
    428  *	Single thread the segment writer.
    429  */
    430 void
    431 lfs_segunlock(struct lfs *fs)
    432 {
    433 	struct segment *sp;
    434 	unsigned long sync, ckp;
    435 	struct buf *bp;
    436 	int do_unmark_dirop = 0;
    437 
    438 	sp = fs->lfs_sp;
    439 
    440 	mutex_enter(&lfs_lock);
    441 	KASSERT(LFS_SEGLOCK_HELD(fs));
    442 	if (fs->lfs_seglock == 1) {
    443 		if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0)
    444 			do_unmark_dirop = 1;
    445 		mutex_exit(&lfs_lock);
    446 		sync = sp->seg_flags & SEGM_SYNC;
    447 		ckp = sp->seg_flags & SEGM_CKP;
    448 
    449 		/* We should have a segment summary, and nothing else */
    450 		KASSERT(sp->cbpp == sp->bpp + 1);
    451 
    452 		/* Free allocated segment summary */
    453 		fs->lfs_offset -= lfs_btofsb(fs, fs->lfs_sumsize);
    454 		bp = *sp->bpp;
    455 		lfs_freebuf(fs, bp);
    456 
    457 		pool_put(&fs->lfs_bpppool, sp->bpp);
    458 		sp->bpp = NULL;
    459 
    460 		/*
    461 		 * If we're not sync, we're done with sp, get rid of it.
    462 		 * Otherwise, we keep a local copy around but free
    463 		 * fs->lfs_sp so another process can use it (we have to
    464 		 * wait but they don't have to wait for us).
    465 		 */
    466 		if (!sync)
    467 			pool_put(&fs->lfs_segpool, sp);
    468 		fs->lfs_sp = NULL;
    469 
    470 		/*
    471 		 * If the I/O count is non-zero, sleep until it reaches zero.
    472 		 * At the moment, the user's process hangs around so we can
    473 		 * sleep.
    474 		 */
    475 		mutex_enter(&lfs_lock);
    476 		if (--fs->lfs_iocount == 0) {
    477 			LFS_DEBUG_COUNTLOCKED("lfs_segunlock");
    478 		}
    479 		if (fs->lfs_iocount <= 1)
    480 			wakeup(&fs->lfs_iocount);
    481 		mutex_exit(&lfs_lock);
    482 		/*
    483 		 * If we're not checkpointing, we don't have to block
    484 		 * other processes to wait for a synchronous write
    485 		 * to complete.
    486 		 */
    487 		if (!ckp) {
    488 #ifdef DEBUG
    489 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    490 #endif
    491 			mutex_enter(&lfs_lock);
    492 			--fs->lfs_seglock;
    493 			fs->lfs_lockpid = 0;
    494 			fs->lfs_locklwp = 0;
    495 			mutex_exit(&lfs_lock);
    496 			wakeup(&fs->lfs_seglock);
    497 		}
    498 		/*
    499 		 * We let checkpoints happen asynchronously.  That means
    500 		 * that during recovery, we have to roll forward between
    501 		 * the two segments described by the first and second
    502 		 * superblocks to make sure that the checkpoint described
    503 		 * by a superblock completed.
    504 		 */
    505 		mutex_enter(&lfs_lock);
    506 		while (ckp && sync && fs->lfs_iocount) {
    507 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
    508 				      "lfs_iocount", 0, &lfs_lock);
    509 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", fs, fs->lfs_iocount));
    510 		}
    511 		while (sync && sp->seg_iocount) {
    512 			(void)mtsleep(&sp->seg_iocount, PRIBIO + 1,
    513 				     "seg_iocount", 0, &lfs_lock);
    514 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
    515 		}
    516 		mutex_exit(&lfs_lock);
    517 		if (sync)
    518 			pool_put(&fs->lfs_segpool, sp);
    519 
    520 		if (ckp) {
    521 			fs->lfs_nactive = 0;
    522 			/* If we *know* everything's on disk, write both sbs */
    523 			/* XXX should wait for this one	 */
    524 			if (sync)
    525 				lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
    526 			lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
    527 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
    528 				lfs_auto_segclean(fs);
    529 				/* If sync, we can clean the remainder too */
    530 				if (sync)
    531 					lfs_auto_segclean(fs);
    532 			}
    533 			fs->lfs_activesb = 1 - fs->lfs_activesb;
    534 #ifdef DEBUG
    535 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    536 #endif
    537 			mutex_enter(&lfs_lock);
    538 			--fs->lfs_seglock;
    539 			fs->lfs_lockpid = 0;
    540 			fs->lfs_locklwp = 0;
    541 			mutex_exit(&lfs_lock);
    542 			wakeup(&fs->lfs_seglock);
    543 		}
    544 		/* Reenable fragment size changes */
    545 		rw_exit(&fs->lfs_fraglock);
    546 		if (do_unmark_dirop)
    547 			lfs_unmark_dirop(fs);
    548 	} else if (fs->lfs_seglock == 0) {
    549 		mutex_exit(&lfs_lock);
    550 		panic ("Seglock not held");
    551 	} else {
    552 		--fs->lfs_seglock;
    553 		mutex_exit(&lfs_lock);
    554 	}
    555 }
    556 
    557 /*
    558  * Drain dirops and start writer.
    559  *
    560  * No simple_locks are held when we enter and none are held when we return.
    561  */
    562 int
    563 lfs_writer_enter(struct lfs *fs, const char *wmesg)
    564 {
    565 	int error = 0;
    566 
    567 	ASSERT_MAYBE_SEGLOCK(fs);
    568 	mutex_enter(&lfs_lock);
    569 
    570 	/* disallow dirops during flush */
    571 	fs->lfs_writer++;
    572 
    573 	while (fs->lfs_dirops > 0) {
    574 		++fs->lfs_diropwait;
    575 		error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
    576 				&lfs_lock);
    577 		--fs->lfs_diropwait;
    578 	}
    579 
    580 	if (error)
    581 		fs->lfs_writer--;
    582 
    583 	mutex_exit(&lfs_lock);
    584 
    585 	return error;
    586 }
    587 
    588 void
    589 lfs_writer_leave(struct lfs *fs)
    590 {
    591 	bool dowakeup;
    592 
    593 	ASSERT_MAYBE_SEGLOCK(fs);
    594 	mutex_enter(&lfs_lock);
    595 	dowakeup = !(--fs->lfs_writer);
    596 	mutex_exit(&lfs_lock);
    597 	if (dowakeup)
    598 		wakeup(&fs->lfs_dirops);
    599 }
    600 
    601 /*
    602  * Unlock, wait for the cleaner, then relock to where we were before.
    603  * To be used only at a fairly high level, to address a paucity of free
    604  * segments propagated back from lfs_gop_write().
    605  */
    606 void
    607 lfs_segunlock_relock(struct lfs *fs)
    608 {
    609 	int n = fs->lfs_seglock;
    610 	u_int16_t seg_flags;
    611 	CLEANERINFO *cip;
    612 	struct buf *bp;
    613 
    614 	if (n == 0)
    615 		return;
    616 
    617 	/* Write anything we've already gathered to disk */
    618 	lfs_writeseg(fs, fs->lfs_sp);
    619 
    620 	/* Tell cleaner */
    621 	LFS_CLEANERINFO(cip, fs, bp);
    622 	cip->flags |= LFS_CLEANER_MUST_CLEAN;
    623 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    624 
    625 	/* Save segment flags for later */
    626 	seg_flags = fs->lfs_sp->seg_flags;
    627 
    628 	fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */
    629 	while(fs->lfs_seglock)
    630 		lfs_segunlock(fs);
    631 
    632 	/* Wait for the cleaner */
    633 	lfs_wakeup_cleaner(fs);
    634 	mutex_enter(&lfs_lock);
    635 	while (LFS_STARVED_FOR_SEGS(fs))
    636 		mtsleep(&fs->lfs_avail, PRIBIO, "relock", 0,
    637 			&lfs_lock);
    638 	mutex_exit(&lfs_lock);
    639 
    640 	/* Put the segment lock back the way it was. */
    641 	while(n--)
    642 		lfs_seglock(fs, seg_flags);
    643 
    644 	/* Cleaner can relax now */
    645 	LFS_CLEANERINFO(cip, fs, bp);
    646 	cip->flags &= ~LFS_CLEANER_MUST_CLEAN;
    647 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    648 
    649 	return;
    650 }
    651 
    652 /*
    653  * Wake up the cleaner, provided that nowrap is not set.
    654  */
    655 void
    656 lfs_wakeup_cleaner(struct lfs *fs)
    657 {
    658 	if (fs->lfs_nowrap > 0)
    659 		return;
    660 
    661 	wakeup(&fs->lfs_nextseg);
    662 	wakeup(&lfs_allclean_wakeup);
    663 }
    664