Home | History | Annotate | Line # | Download | only in lfs
lfs_subr.c revision 1.79
      1 /*	$NetBSD: lfs_subr.c,v 1.79 2013/06/18 18:18:58 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 (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 1.79 2013/06/18 18:18:58 christos 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_extern.h>
     78 
     79 #include <uvm/uvm.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
    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 = (res_t *)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 = fs->lfs_sumsize;
    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 = fs->lfs_bsize;
    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 	maxbpp = ((fs->lfs_sumsize - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
    147 	maxbpp = MIN(maxbpp, lfs_segsize(fs) / fs->lfs_fsize + 2);
    148 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
    149 		"lfsbpppl", &pool_allocator_nointr, IPL_NONE);
    150 }
    151 
    152 void
    153 lfs_free_resblks(struct lfs *fs)
    154 {
    155 	int i;
    156 
    157 	pool_destroy(&fs->lfs_bpppool);
    158 	pool_destroy(&fs->lfs_segpool);
    159 	pool_destroy(&fs->lfs_clpool);
    160 
    161 	mutex_enter(&lfs_lock);
    162 	for (i = 0; i < LFS_N_TOTAL; i++) {
    163 		while (fs->lfs_resblk[i].inuse)
    164 			mtsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
    165 				&lfs_lock);
    166 		if (fs->lfs_resblk[i].p != NULL)
    167 			free(fs->lfs_resblk[i].p, M_SEGMENT);
    168 	}
    169 	free(fs->lfs_resblk, M_SEGMENT);
    170 	mutex_exit(&lfs_lock);
    171 }
    172 
    173 static unsigned int
    174 lfs_mhash(void *vp)
    175 {
    176 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
    177 }
    178 
    179 /*
    180  * Return memory of the given size for the given purpose, or use one of a
    181  * number of spare last-resort buffers, if malloc returns NULL.
    182  */
    183 void *
    184 lfs_malloc(struct lfs *fs, size_t size, int type)
    185 {
    186 	struct lfs_res_blk *re;
    187 	void *r;
    188 	int i, s, start;
    189 	unsigned int h;
    190 
    191 	ASSERT_MAYBE_SEGLOCK(fs);
    192 	r = NULL;
    193 
    194 	/* If no mem allocated for this type, it just waits */
    195 	if (lfs_res_qty[type] == 0) {
    196 		r = malloc(size, M_SEGMENT, M_WAITOK);
    197 		return r;
    198 	}
    199 
    200 	/* Otherwise try a quick malloc, and if it works, great */
    201 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
    202 		return r;
    203 	}
    204 
    205 	/*
    206 	 * If malloc returned NULL, we are forced to use one of our
    207 	 * reserve blocks.  We have on hand at least one summary block,
    208 	 * at least one cluster block, at least one superblock,
    209 	 * and several indirect blocks.
    210 	 */
    211 
    212 	mutex_enter(&lfs_lock);
    213 	/* skip over blocks of other types */
    214 	for (i = 0, start = 0; i < type; i++)
    215 		start += lfs_res_qty[i];
    216 	while (r == NULL) {
    217 		for (i = 0; i < lfs_res_qty[type]; i++) {
    218 			if (fs->lfs_resblk[start + i].inuse == 0) {
    219 				re = fs->lfs_resblk + start + i;
    220 				re->inuse = 1;
    221 				r = re->p;
    222 				KASSERT(re->size >= size);
    223 				h = lfs_mhash(r);
    224 				s = splbio();
    225 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
    226 				splx(s);
    227 				mutex_exit(&lfs_lock);
    228 				return r;
    229 			}
    230 		}
    231 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
    232 		      lfs_res_names[type], lfs_res_qty[type]));
    233 		mtsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
    234 			&lfs_lock);
    235 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
    236 		      lfs_res_names[type]));
    237 	}
    238 	/* NOTREACHED */
    239 	mutex_exit(&lfs_lock);
    240 	return r;
    241 }
    242 
    243 void
    244 lfs_free(struct lfs *fs, void *p, int type)
    245 {
    246 	int s;
    247 	unsigned int h;
    248 	res_t *re;
    249 #ifdef DEBUG
    250 	int i;
    251 #endif
    252 
    253 	ASSERT_MAYBE_SEGLOCK(fs);
    254 	h = lfs_mhash(p);
    255 	mutex_enter(&lfs_lock);
    256 	s = splbio();
    257 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
    258 		if (re->p == p) {
    259 			KASSERT(re->inuse == 1);
    260 			LIST_REMOVE(re, res);
    261 			re->inuse = 0;
    262 			wakeup(&fs->lfs_resblk);
    263 			splx(s);
    264 			mutex_exit(&lfs_lock);
    265 			return;
    266 		}
    267 	}
    268 #ifdef DEBUG
    269 	for (i = 0; i < LFS_N_TOTAL; i++) {
    270 		if (fs->lfs_resblk[i].p == p)
    271 			panic("lfs_free: inconsistent reserved block");
    272 	}
    273 #endif
    274 	splx(s);
    275 	mutex_exit(&lfs_lock);
    276 
    277 	/*
    278 	 * If we didn't find it, free it.
    279 	 */
    280 	free(p, M_SEGMENT);
    281 }
    282 
    283 /*
    284  * lfs_seglock --
    285  *	Single thread the segment writer.
    286  */
    287 int
    288 lfs_seglock(struct lfs *fs, unsigned long flags)
    289 {
    290 	struct segment *sp;
    291 
    292 	mutex_enter(&lfs_lock);
    293 	if (fs->lfs_seglock) {
    294 		if (fs->lfs_lockpid == curproc->p_pid &&
    295 		    fs->lfs_locklwp == curlwp->l_lid) {
    296 			++fs->lfs_seglock;
    297 			fs->lfs_sp->seg_flags |= flags;
    298 			mutex_exit(&lfs_lock);
    299 			return 0;
    300 		} else if (flags & SEGM_PAGEDAEMON) {
    301 			mutex_exit(&lfs_lock);
    302 			return EWOULDBLOCK;
    303 		} else {
    304 			while (fs->lfs_seglock) {
    305 				(void)mtsleep(&fs->lfs_seglock, PRIBIO + 1,
    306 					"lfs_seglock", 0, &lfs_lock);
    307 			}
    308 		}
    309 	}
    310 
    311 	fs->lfs_seglock = 1;
    312 	fs->lfs_lockpid = curproc->p_pid;
    313 	fs->lfs_locklwp = curlwp->l_lid;
    314 	mutex_exit(&lfs_lock);
    315 	fs->lfs_cleanind = 0;
    316 
    317 #ifdef DEBUG
    318 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
    319 #endif
    320 	/* Drain fragment size changes out */
    321 	rw_enter(&fs->lfs_fraglock, RW_WRITER);
    322 
    323 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
    324 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
    325 	sp->seg_flags = flags;
    326 	sp->vp = NULL;
    327 	sp->seg_iocount = 0;
    328 	(void) lfs_initseg(fs);
    329 
    330 	/*
    331 	 * Keep a cumulative count of the outstanding I/O operations.  If the
    332 	 * disk drive catches up with us it could go to zero before we finish,
    333 	 * so we artificially increment it by one until we've scheduled all of
    334 	 * the writes we intend to do.
    335 	 */
    336 	mutex_enter(&lfs_lock);
    337 	++fs->lfs_iocount;
    338 	fs->lfs_startseg = fs->lfs_curseg;
    339 	mutex_exit(&lfs_lock);
    340 	return 0;
    341 }
    342 
    343 static void lfs_unmark_dirop(struct lfs *);
    344 
    345 static void
    346 lfs_unmark_dirop(struct lfs *fs)
    347 {
    348 	struct inode *ip, *nip;
    349 	struct vnode *vp;
    350 	int doit;
    351 
    352 	ASSERT_NO_SEGLOCK(fs);
    353 	mutex_enter(&lfs_lock);
    354 	doit = !(fs->lfs_flags & LFS_UNDIROP);
    355 	if (doit)
    356 		fs->lfs_flags |= LFS_UNDIROP;
    357 	if (!doit) {
    358 		mutex_exit(&lfs_lock);
    359 		return;
    360 	}
    361 
    362 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
    363 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
    364 		vp = ITOV(ip);
    365 		if ((ip->i_flag & (IN_ADIROP | IN_CDIROP)) == IN_CDIROP) {
    366 			--lfs_dirvcount;
    367 			--fs->lfs_dirvcount;
    368 			vp->v_uflag &= ~VU_DIROP;
    369 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
    370 			wakeup(&lfs_dirvcount);
    371 			fs->lfs_unlockvp = vp;
    372 			mutex_exit(&lfs_lock);
    373 			vrele(vp);
    374 			mutex_enter(&lfs_lock);
    375 			fs->lfs_unlockvp = NULL;
    376 			ip->i_flag &= ~IN_CDIROP;
    377 		}
    378 	}
    379 
    380 	fs->lfs_flags &= ~LFS_UNDIROP;
    381 	wakeup(&fs->lfs_flags);
    382 	mutex_exit(&lfs_lock);
    383 }
    384 
    385 static void
    386 lfs_auto_segclean(struct lfs *fs)
    387 {
    388 	int i, error, s, waited;
    389 
    390 	ASSERT_SEGLOCK(fs);
    391 	/*
    392 	 * Now that we've swapped lfs_activesb, but while we still
    393 	 * hold the segment lock, run through the segment list marking
    394 	 * the empty ones clean.
    395 	 * XXX - do we really need to do them all at once?
    396 	 */
    397 	waited = 0;
    398 	for (i = 0; i < fs->lfs_nseg; i++) {
    399 		if ((fs->lfs_suflags[0][i] &
    400 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    401 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
    402 		    (fs->lfs_suflags[1][i] &
    403 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    404 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
    405 
    406 			/* Make sure the sb is written before we clean */
    407 			mutex_enter(&lfs_lock);
    408 			s = splbio();
    409 			while (waited == 0 && fs->lfs_sbactive)
    410 				mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
    411 					0, &lfs_lock);
    412 			splx(s);
    413 			mutex_exit(&lfs_lock);
    414 			waited = 1;
    415 
    416 			if ((error = lfs_do_segclean(fs, i)) != 0) {
    417 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
    418 			}
    419 		}
    420 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
    421 			fs->lfs_suflags[fs->lfs_activesb][i];
    422 	}
    423 }
    424 
    425 /*
    426  * lfs_segunlock --
    427  *	Single thread the segment writer.
    428  */
    429 void
    430 lfs_segunlock(struct lfs *fs)
    431 {
    432 	struct segment *sp;
    433 	unsigned long sync, ckp;
    434 	struct buf *bp;
    435 	int do_unmark_dirop = 0;
    436 
    437 	sp = fs->lfs_sp;
    438 
    439 	mutex_enter(&lfs_lock);
    440 	KASSERT(LFS_SEGLOCK_HELD(fs));
    441 	if (fs->lfs_seglock == 1) {
    442 		if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0)
    443 			do_unmark_dirop = 1;
    444 		mutex_exit(&lfs_lock);
    445 		sync = sp->seg_flags & SEGM_SYNC;
    446 		ckp = sp->seg_flags & SEGM_CKP;
    447 
    448 		/* We should have a segment summary, and nothing else */
    449 		KASSERT(sp->cbpp == sp->bpp + 1);
    450 
    451 		/* Free allocated segment summary */
    452 		fs->lfs_offset -= lfs_btofsb(fs, fs->lfs_sumsize);
    453 		bp = *sp->bpp;
    454 		lfs_freebuf(fs, bp);
    455 
    456 		pool_put(&fs->lfs_bpppool, sp->bpp);
    457 		sp->bpp = NULL;
    458 
    459 		/*
    460 		 * If we're not sync, we're done with sp, get rid of it.
    461 		 * Otherwise, we keep a local copy around but free
    462 		 * fs->lfs_sp so another process can use it (we have to
    463 		 * wait but they don't have to wait for us).
    464 		 */
    465 		if (!sync)
    466 			pool_put(&fs->lfs_segpool, sp);
    467 		fs->lfs_sp = NULL;
    468 
    469 		/*
    470 		 * If the I/O count is non-zero, sleep until it reaches zero.
    471 		 * At the moment, the user's process hangs around so we can
    472 		 * sleep.
    473 		 */
    474 		mutex_enter(&lfs_lock);
    475 		if (--fs->lfs_iocount == 0) {
    476 			LFS_DEBUG_COUNTLOCKED("lfs_segunlock");
    477 		}
    478 		if (fs->lfs_iocount <= 1)
    479 			wakeup(&fs->lfs_iocount);
    480 		mutex_exit(&lfs_lock);
    481 		/*
    482 		 * If we're not checkpointing, we don't have to block
    483 		 * other processes to wait for a synchronous write
    484 		 * to complete.
    485 		 */
    486 		if (!ckp) {
    487 #ifdef DEBUG
    488 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    489 #endif
    490 			mutex_enter(&lfs_lock);
    491 			--fs->lfs_seglock;
    492 			fs->lfs_lockpid = 0;
    493 			fs->lfs_locklwp = 0;
    494 			mutex_exit(&lfs_lock);
    495 			wakeup(&fs->lfs_seglock);
    496 		}
    497 		/*
    498 		 * We let checkpoints happen asynchronously.  That means
    499 		 * that during recovery, we have to roll forward between
    500 		 * the two segments described by the first and second
    501 		 * superblocks to make sure that the checkpoint described
    502 		 * by a superblock completed.
    503 		 */
    504 		mutex_enter(&lfs_lock);
    505 		while (ckp && sync && fs->lfs_iocount) {
    506 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
    507 				      "lfs_iocount", 0, &lfs_lock);
    508 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", fs, fs->lfs_iocount));
    509 		}
    510 		while (sync && sp->seg_iocount) {
    511 			(void)mtsleep(&sp->seg_iocount, PRIBIO + 1,
    512 				     "seg_iocount", 0, &lfs_lock);
    513 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
    514 		}
    515 		mutex_exit(&lfs_lock);
    516 		if (sync)
    517 			pool_put(&fs->lfs_segpool, sp);
    518 
    519 		if (ckp) {
    520 			fs->lfs_nactive = 0;
    521 			/* If we *know* everything's on disk, write both sbs */
    522 			/* XXX should wait for this one	 */
    523 			if (sync)
    524 				lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
    525 			lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
    526 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
    527 				lfs_auto_segclean(fs);
    528 				/* If sync, we can clean the remainder too */
    529 				if (sync)
    530 					lfs_auto_segclean(fs);
    531 			}
    532 			fs->lfs_activesb = 1 - fs->lfs_activesb;
    533 #ifdef DEBUG
    534 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    535 #endif
    536 			mutex_enter(&lfs_lock);
    537 			--fs->lfs_seglock;
    538 			fs->lfs_lockpid = 0;
    539 			fs->lfs_locklwp = 0;
    540 			mutex_exit(&lfs_lock);
    541 			wakeup(&fs->lfs_seglock);
    542 		}
    543 		/* Reenable fragment size changes */
    544 		rw_exit(&fs->lfs_fraglock);
    545 		if (do_unmark_dirop)
    546 			lfs_unmark_dirop(fs);
    547 	} else if (fs->lfs_seglock == 0) {
    548 		mutex_exit(&lfs_lock);
    549 		panic ("Seglock not held");
    550 	} else {
    551 		--fs->lfs_seglock;
    552 		mutex_exit(&lfs_lock);
    553 	}
    554 }
    555 
    556 /*
    557  * Drain dirops and start writer.
    558  *
    559  * No simple_locks are held when we enter and none are held when we return.
    560  */
    561 int
    562 lfs_writer_enter(struct lfs *fs, const char *wmesg)
    563 {
    564 	int error = 0;
    565 
    566 	ASSERT_MAYBE_SEGLOCK(fs);
    567 	mutex_enter(&lfs_lock);
    568 
    569 	/* disallow dirops during flush */
    570 	fs->lfs_writer++;
    571 
    572 	while (fs->lfs_dirops > 0) {
    573 		++fs->lfs_diropwait;
    574 		error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
    575 				&lfs_lock);
    576 		--fs->lfs_diropwait;
    577 	}
    578 
    579 	if (error)
    580 		fs->lfs_writer--;
    581 
    582 	mutex_exit(&lfs_lock);
    583 
    584 	return error;
    585 }
    586 
    587 void
    588 lfs_writer_leave(struct lfs *fs)
    589 {
    590 	bool dowakeup;
    591 
    592 	ASSERT_MAYBE_SEGLOCK(fs);
    593 	mutex_enter(&lfs_lock);
    594 	dowakeup = !(--fs->lfs_writer);
    595 	mutex_exit(&lfs_lock);
    596 	if (dowakeup)
    597 		wakeup(&fs->lfs_dirops);
    598 }
    599 
    600 /*
    601  * Unlock, wait for the cleaner, then relock to where we were before.
    602  * To be used only at a fairly high level, to address a paucity of free
    603  * segments propagated back from lfs_gop_write().
    604  */
    605 void
    606 lfs_segunlock_relock(struct lfs *fs)
    607 {
    608 	int n = fs->lfs_seglock;
    609 	u_int16_t seg_flags;
    610 	CLEANERINFO *cip;
    611 	struct buf *bp;
    612 
    613 	if (n == 0)
    614 		return;
    615 
    616 	/* Write anything we've already gathered to disk */
    617 	lfs_writeseg(fs, fs->lfs_sp);
    618 
    619 	/* Tell cleaner */
    620 	LFS_CLEANERINFO(cip, fs, bp);
    621 	cip->flags |= LFS_CLEANER_MUST_CLEAN;
    622 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    623 
    624 	/* Save segment flags for later */
    625 	seg_flags = fs->lfs_sp->seg_flags;
    626 
    627 	fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */
    628 	while(fs->lfs_seglock)
    629 		lfs_segunlock(fs);
    630 
    631 	/* Wait for the cleaner */
    632 	lfs_wakeup_cleaner(fs);
    633 	mutex_enter(&lfs_lock);
    634 	while (LFS_STARVED_FOR_SEGS(fs))
    635 		mtsleep(&fs->lfs_avail, PRIBIO, "relock", 0,
    636 			&lfs_lock);
    637 	mutex_exit(&lfs_lock);
    638 
    639 	/* Put the segment lock back the way it was. */
    640 	while(n--)
    641 		lfs_seglock(fs, seg_flags);
    642 
    643 	/* Cleaner can relax now */
    644 	LFS_CLEANERINFO(cip, fs, bp);
    645 	cip->flags &= ~LFS_CLEANER_MUST_CLEAN;
    646 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    647 
    648 	return;
    649 }
    650 
    651 /*
    652  * Wake up the cleaner, provided that nowrap is not set.
    653  */
    654 void
    655 lfs_wakeup_cleaner(struct lfs *fs)
    656 {
    657 	if (fs->lfs_nowrap > 0)
    658 		return;
    659 
    660 	wakeup(&fs->lfs_nextseg);
    661 	wakeup(&lfs_allclean_wakeup);
    662 }
    663