Home | History | Annotate | Line # | Download | only in lfs
lfs_subr.c revision 1.73.20.1
      1  1.73.20.1  uebayasi /*	$NetBSD: lfs_subr.c,v 1.73.20.1 2010/04/30 14:44:36 uebayasi Exp $	*/
      2        1.2       cgd 
      3        1.8  perseant /*-
      4       1.31  perseant  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
      5        1.8  perseant  * All rights reserved.
      6        1.8  perseant  *
      7        1.8  perseant  * This code is derived from software contributed to The NetBSD Foundation
      8        1.8  perseant  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9        1.8  perseant  *
     10        1.8  perseant  * Redistribution and use in source and binary forms, with or without
     11        1.8  perseant  * modification, are permitted provided that the following conditions
     12        1.8  perseant  * are met:
     13        1.8  perseant  * 1. Redistributions of source code must retain the above copyright
     14        1.8  perseant  *    notice, this list of conditions and the following disclaimer.
     15        1.8  perseant  * 2. Redistributions in binary form must reproduce the above copyright
     16        1.8  perseant  *    notice, this list of conditions and the following disclaimer in the
     17        1.8  perseant  *    documentation and/or other materials provided with the distribution.
     18        1.8  perseant  *
     19        1.8  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20        1.8  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21        1.8  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22        1.8  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23        1.8  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24        1.8  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25        1.8  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26        1.8  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27        1.8  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28        1.8  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29        1.8  perseant  * POSSIBILITY OF SUCH DAMAGE.
     30        1.8  perseant  */
     31        1.1   mycroft /*
     32        1.1   mycroft  * Copyright (c) 1991, 1993
     33        1.1   mycroft  *	The Regents of the University of California.  All rights reserved.
     34        1.1   mycroft  *
     35        1.1   mycroft  * Redistribution and use in source and binary forms, with or without
     36        1.1   mycroft  * modification, are permitted provided that the following conditions
     37        1.1   mycroft  * are met:
     38        1.1   mycroft  * 1. Redistributions of source code must retain the above copyright
     39        1.1   mycroft  *    notice, this list of conditions and the following disclaimer.
     40        1.1   mycroft  * 2. Redistributions in binary form must reproduce the above copyright
     41        1.1   mycroft  *    notice, this list of conditions and the following disclaimer in the
     42        1.1   mycroft  *    documentation and/or other materials provided with the distribution.
     43       1.43       agc  * 3. Neither the name of the University nor the names of its contributors
     44        1.1   mycroft  *    may be used to endorse or promote products derived from this software
     45        1.1   mycroft  *    without specific prior written permission.
     46        1.1   mycroft  *
     47        1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     48        1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     49        1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     50        1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     51        1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     52        1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     53        1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     54        1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     55        1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     56        1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     57        1.1   mycroft  * SUCH DAMAGE.
     58        1.1   mycroft  *
     59        1.6      fvdl  *	@(#)lfs_subr.c	8.4 (Berkeley) 5/8/95
     60        1.1   mycroft  */
     61       1.20     lukem 
     62       1.20     lukem #include <sys/cdefs.h>
     63  1.73.20.1  uebayasi __KERNEL_RCSID(0, "$NetBSD: lfs_subr.c,v 1.73.20.1 2010/04/30 14:44:36 uebayasi Exp $");
     64        1.1   mycroft 
     65        1.1   mycroft #include <sys/param.h>
     66        1.3  christos #include <sys/systm.h>
     67        1.1   mycroft #include <sys/namei.h>
     68        1.1   mycroft #include <sys/vnode.h>
     69        1.1   mycroft #include <sys/buf.h>
     70        1.1   mycroft #include <sys/mount.h>
     71        1.1   mycroft #include <sys/malloc.h>
     72        1.1   mycroft #include <sys/proc.h>
     73       1.61  perseant #include <sys/kauth.h>
     74        1.1   mycroft 
     75        1.1   mycroft #include <ufs/ufs/inode.h>
     76        1.1   mycroft #include <ufs/lfs/lfs.h>
     77        1.1   mycroft #include <ufs/lfs/lfs_extern.h>
     78        1.1   mycroft 
     79       1.31  perseant #include <uvm/uvm.h>
     80       1.31  perseant 
     81       1.50  perseant #ifdef DEBUG
     82       1.53  christos const char *lfs_res_names[LFS_NB_COUNT] = {
     83       1.31  perseant 	"summary",
     84       1.31  perseant 	"superblock",
     85       1.52  perseant 	"file block",
     86       1.31  perseant 	"cluster",
     87       1.31  perseant 	"clean",
     88       1.52  perseant 	"blkiov",
     89       1.31  perseant };
     90       1.31  perseant #endif
     91       1.31  perseant 
     92       1.31  perseant int lfs_res_qty[LFS_NB_COUNT] = {
     93       1.31  perseant 	LFS_N_SUMMARIES,
     94       1.31  perseant 	LFS_N_SBLOCKS,
     95       1.31  perseant 	LFS_N_IBLOCKS,
     96       1.31  perseant 	LFS_N_CLUSTERS,
     97       1.31  perseant 	LFS_N_CLEAN,
     98       1.52  perseant 	LFS_N_BLKIOV,
     99       1.31  perseant };
    100       1.31  perseant 
    101       1.31  perseant void
    102       1.31  perseant lfs_setup_resblks(struct lfs *fs)
    103       1.31  perseant {
    104       1.31  perseant 	int i, j;
    105       1.31  perseant 	int maxbpp;
    106       1.31  perseant 
    107       1.51  perseant 	ASSERT_NO_SEGLOCK(fs);
    108       1.31  perseant 	fs->lfs_resblk = (res_t *)malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
    109       1.33  perseant 					  M_WAITOK);
    110       1.31  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    111       1.31  perseant 		fs->lfs_resblk[i].inuse = 0;
    112       1.31  perseant 		fs->lfs_resblk[i].p = NULL;
    113       1.31  perseant 	}
    114       1.31  perseant 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
    115       1.31  perseant 		LIST_INIT(fs->lfs_reshash + i);
    116       1.31  perseant 
    117       1.31  perseant 	/*
    118       1.31  perseant 	 * These types of allocations can be larger than a page,
    119       1.31  perseant 	 * so we can't use the pool subsystem for them.
    120       1.31  perseant 	 */
    121       1.31  perseant 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
    122       1.34  perseant 		fs->lfs_resblk[i].size = fs->lfs_sumsize;
    123       1.31  perseant 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
    124       1.34  perseant 		fs->lfs_resblk[i].size = LFS_SBPAD;
    125       1.31  perseant 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
    126       1.34  perseant 		fs->lfs_resblk[i].size = fs->lfs_bsize;
    127       1.31  perseant 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
    128       1.34  perseant 		fs->lfs_resblk[i].size = MAXPHYS;
    129       1.31  perseant 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
    130       1.34  perseant 		fs->lfs_resblk[i].size = MAXPHYS;
    131       1.52  perseant 	for (j = 0; j < LFS_N_BLKIOV; j++, i++)
    132       1.52  perseant 		fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO);
    133       1.34  perseant 
    134       1.34  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    135       1.34  perseant 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
    136       1.34  perseant 					     M_SEGMENT, M_WAITOK);
    137       1.34  perseant 	}
    138       1.31  perseant 
    139       1.31  perseant 	/*
    140       1.31  perseant 	 * Initialize pools for small types (XXX is BPP small?)
    141       1.31  perseant 	 */
    142       1.46    simonb 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
    143       1.68        ad 		"lfsclpl", &pool_allocator_nointr, IPL_NONE);
    144       1.46    simonb 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
    145       1.68        ad 		"lfssegpool", &pool_allocator_nointr, IPL_NONE);
    146       1.31  perseant 	maxbpp = ((fs->lfs_sumsize - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
    147       1.47      yamt 	maxbpp = MIN(maxbpp, segsize(fs) / fs->lfs_fsize + 2);
    148       1.46    simonb 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
    149       1.68        ad 		"lfsbpppl", &pool_allocator_nointr, IPL_NONE);
    150       1.31  perseant }
    151       1.31  perseant 
    152       1.31  perseant void
    153       1.31  perseant lfs_free_resblks(struct lfs *fs)
    154       1.31  perseant {
    155       1.31  perseant 	int i;
    156       1.31  perseant 
    157       1.31  perseant 	pool_destroy(&fs->lfs_bpppool);
    158       1.31  perseant 	pool_destroy(&fs->lfs_segpool);
    159       1.31  perseant 	pool_destroy(&fs->lfs_clpool);
    160       1.31  perseant 
    161       1.72        ad 	mutex_enter(&lfs_lock);
    162       1.31  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    163       1.39  perseant 		while (fs->lfs_resblk[i].inuse)
    164       1.72        ad 			mtsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
    165       1.72        ad 				&lfs_lock);
    166       1.31  perseant 		if (fs->lfs_resblk[i].p != NULL)
    167       1.31  perseant 			free(fs->lfs_resblk[i].p, M_SEGMENT);
    168       1.31  perseant 	}
    169       1.31  perseant 	free(fs->lfs_resblk, M_SEGMENT);
    170       1.72        ad 	mutex_exit(&lfs_lock);
    171       1.31  perseant }
    172       1.31  perseant 
    173       1.31  perseant static unsigned int
    174       1.31  perseant lfs_mhash(void *vp)
    175       1.31  perseant {
    176       1.31  perseant 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
    177       1.31  perseant }
    178       1.31  perseant 
    179       1.31  perseant /*
    180       1.31  perseant  * Return memory of the given size for the given purpose, or use one of a
    181       1.31  perseant  * number of spare last-resort buffers, if malloc returns NULL.
    182       1.49     perry  */
    183       1.31  perseant void *
    184       1.31  perseant lfs_malloc(struct lfs *fs, size_t size, int type)
    185       1.31  perseant {
    186       1.31  perseant 	struct lfs_res_blk *re;
    187       1.31  perseant 	void *r;
    188       1.31  perseant 	int i, s, start;
    189       1.31  perseant 	unsigned int h;
    190       1.31  perseant 
    191       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    192       1.34  perseant 	r = NULL;
    193       1.34  perseant 
    194       1.31  perseant 	/* If no mem allocated for this type, it just waits */
    195       1.34  perseant 	if (lfs_res_qty[type] == 0) {
    196       1.34  perseant 		r = malloc(size, M_SEGMENT, M_WAITOK);
    197       1.34  perseant 		return r;
    198       1.34  perseant 	}
    199       1.31  perseant 
    200       1.31  perseant 	/* Otherwise try a quick malloc, and if it works, great */
    201       1.34  perseant 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
    202       1.31  perseant 		return r;
    203       1.34  perseant 	}
    204       1.31  perseant 
    205       1.31  perseant 	/*
    206       1.31  perseant 	 * If malloc returned NULL, we are forced to use one of our
    207       1.31  perseant 	 * reserve blocks.  We have on hand at least one summary block,
    208       1.31  perseant 	 * at least one cluster block, at least one superblock,
    209       1.31  perseant 	 * and several indirect blocks.
    210       1.31  perseant 	 */
    211       1.51  perseant 
    212       1.72        ad 	mutex_enter(&lfs_lock);
    213       1.31  perseant 	/* skip over blocks of other types */
    214       1.31  perseant 	for (i = 0, start = 0; i < type; i++)
    215       1.31  perseant 		start += lfs_res_qty[i];
    216       1.31  perseant 	while (r == NULL) {
    217       1.31  perseant 		for (i = 0; i < lfs_res_qty[type]; i++) {
    218       1.31  perseant 			if (fs->lfs_resblk[start + i].inuse == 0) {
    219       1.31  perseant 				re = fs->lfs_resblk + start + i;
    220       1.31  perseant 				re->inuse = 1;
    221       1.31  perseant 				r = re->p;
    222       1.34  perseant 				KASSERT(re->size >= size);
    223       1.31  perseant 				h = lfs_mhash(r);
    224       1.31  perseant 				s = splbio();
    225       1.31  perseant 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
    226       1.31  perseant 				splx(s);
    227       1.72        ad 				mutex_exit(&lfs_lock);
    228       1.31  perseant 				return r;
    229       1.31  perseant 			}
    230       1.31  perseant 		}
    231       1.51  perseant 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
    232       1.51  perseant 		      lfs_res_names[type], lfs_res_qty[type]));
    233       1.72        ad 		mtsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
    234       1.72        ad 			&lfs_lock);
    235       1.51  perseant 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
    236       1.51  perseant 		      lfs_res_names[type]));
    237       1.31  perseant 	}
    238       1.31  perseant 	/* NOTREACHED */
    239       1.72        ad 	mutex_exit(&lfs_lock);
    240       1.31  perseant 	return r;
    241       1.31  perseant }
    242       1.31  perseant 
    243       1.31  perseant void
    244       1.65  christos lfs_free(struct lfs *fs, void *p, int type)
    245       1.31  perseant {
    246       1.31  perseant 	int s;
    247       1.31  perseant 	unsigned int h;
    248       1.31  perseant 	res_t *re;
    249       1.32      yamt #ifdef DEBUG
    250       1.32      yamt 	int i;
    251       1.32      yamt #endif
    252       1.31  perseant 
    253       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    254       1.31  perseant 	h = lfs_mhash(p);
    255       1.72        ad 	mutex_enter(&lfs_lock);
    256       1.31  perseant 	s = splbio();
    257       1.31  perseant 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
    258       1.31  perseant 		if (re->p == p) {
    259       1.32      yamt 			KASSERT(re->inuse == 1);
    260       1.31  perseant 			LIST_REMOVE(re, res);
    261       1.31  perseant 			re->inuse = 0;
    262       1.31  perseant 			wakeup(&fs->lfs_resblk);
    263       1.31  perseant 			splx(s);
    264       1.72        ad 			mutex_exit(&lfs_lock);
    265       1.31  perseant 			return;
    266       1.31  perseant 		}
    267       1.31  perseant 	}
    268       1.32      yamt #ifdef DEBUG
    269       1.32      yamt 	for (i = 0; i < LFS_N_TOTAL; i++) {
    270       1.32      yamt 		if (fs->lfs_resblk[i].p == p)
    271       1.34  perseant 			panic("lfs_free: inconsistent reserved block");
    272       1.32      yamt 	}
    273       1.32      yamt #endif
    274       1.31  perseant 	splx(s);
    275       1.72        ad 	mutex_exit(&lfs_lock);
    276       1.51  perseant 
    277       1.31  perseant 	/*
    278       1.31  perseant 	 * If we didn't find it, free it.
    279       1.31  perseant 	 */
    280       1.31  perseant 	free(p, M_SEGMENT);
    281       1.31  perseant }
    282        1.1   mycroft 
    283        1.1   mycroft /*
    284        1.1   mycroft  * lfs_seglock --
    285        1.1   mycroft  *	Single thread the segment writer.
    286        1.1   mycroft  */
    287       1.31  perseant int
    288       1.18  perseant lfs_seglock(struct lfs *fs, unsigned long flags)
    289        1.1   mycroft {
    290        1.1   mycroft 	struct segment *sp;
    291       1.49     perry 
    292       1.72        ad 	mutex_enter(&lfs_lock);
    293        1.7   thorpej 	if (fs->lfs_seglock) {
    294       1.58  perseant 		if (fs->lfs_lockpid == curproc->p_pid &&
    295       1.58  perseant 		    fs->lfs_locklwp == curlwp->l_lid) {
    296        1.1   mycroft 			++fs->lfs_seglock;
    297        1.1   mycroft 			fs->lfs_sp->seg_flags |= flags;
    298  1.73.20.1  uebayasi 			mutex_exit(&lfs_lock);
    299       1.31  perseant 			return 0;
    300       1.38  perseant 		} else if (flags & SEGM_PAGEDAEMON) {
    301       1.72        ad 			mutex_exit(&lfs_lock);
    302       1.31  perseant 			return EWOULDBLOCK;
    303       1.51  perseant 		} else {
    304       1.51  perseant 			while (fs->lfs_seglock) {
    305       1.72        ad 				(void)mtsleep(&fs->lfs_seglock, PRIBIO + 1,
    306       1.72        ad 					"lfs_seglock", 0, &lfs_lock);
    307       1.51  perseant 			}
    308       1.51  perseant 		}
    309        1.7   thorpej 	}
    310       1.49     perry 
    311        1.1   mycroft 	fs->lfs_seglock = 1;
    312        1.1   mycroft 	fs->lfs_lockpid = curproc->p_pid;
    313       1.58  perseant 	fs->lfs_locklwp = curlwp->l_lid;
    314       1.72        ad 	mutex_exit(&lfs_lock);
    315       1.36  perseant 	fs->lfs_cleanind = 0;
    316       1.36  perseant 
    317       1.51  perseant #ifdef DEBUG
    318       1.51  perseant 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
    319       1.51  perseant #endif
    320       1.27  perseant 	/* Drain fragment size changes out */
    321       1.66        ad 	rw_enter(&fs->lfs_fraglock, RW_WRITER);
    322       1.27  perseant 
    323       1.31  perseant 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
    324       1.31  perseant 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
    325        1.1   mycroft 	sp->seg_flags = flags;
    326        1.1   mycroft 	sp->vp = NULL;
    327       1.26  perseant 	sp->seg_iocount = 0;
    328        1.1   mycroft 	(void) lfs_initseg(fs);
    329       1.49     perry 
    330        1.1   mycroft 	/*
    331        1.1   mycroft 	 * Keep a cumulative count of the outstanding I/O operations.  If the
    332        1.1   mycroft 	 * disk drive catches up with us it could go to zero before we finish,
    333        1.1   mycroft 	 * so we artificially increment it by one until we've scheduled all of
    334        1.1   mycroft 	 * the writes we intend to do.
    335        1.1   mycroft 	 */
    336       1.72        ad 	mutex_enter(&lfs_lock);
    337        1.1   mycroft 	++fs->lfs_iocount;
    338       1.72        ad 	mutex_exit(&lfs_lock);
    339       1.31  perseant 	return 0;
    340        1.1   mycroft }
    341        1.8  perseant 
    342       1.31  perseant static void lfs_unmark_dirop(struct lfs *);
    343       1.31  perseant 
    344       1.31  perseant static void
    345       1.31  perseant lfs_unmark_dirop(struct lfs *fs)
    346       1.31  perseant {
    347       1.31  perseant 	struct inode *ip, *nip;
    348       1.31  perseant 	struct vnode *vp;
    349       1.40  perseant 	int doit;
    350       1.31  perseant 
    351       1.51  perseant 	ASSERT_NO_SEGLOCK(fs);
    352       1.72        ad 	mutex_enter(&lfs_lock);
    353       1.40  perseant 	doit = !(fs->lfs_flags & LFS_UNDIROP);
    354       1.40  perseant 	if (doit)
    355       1.40  perseant 		fs->lfs_flags |= LFS_UNDIROP;
    356       1.51  perseant 	if (!doit) {
    357       1.72        ad 		mutex_exit(&lfs_lock);
    358       1.40  perseant 		return;
    359       1.51  perseant 	}
    360       1.40  perseant 
    361       1.31  perseant 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
    362       1.31  perseant 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
    363       1.31  perseant 		vp = ITOV(ip);
    364       1.72        ad 		if (VOP_ISLOCKED(vp) == LK_EXCLOTHER)
    365       1.31  perseant 			continue;
    366       1.61  perseant 		if ((VTOI(vp)->i_flag & (IN_ADIROP | IN_ALLMOD)) == 0) {
    367       1.31  perseant 			--lfs_dirvcount;
    368       1.59  perseant 			--fs->lfs_dirvcount;
    369       1.71        ad 			vp->v_uflag &= ~VU_DIROP;
    370       1.31  perseant 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
    371       1.31  perseant 			wakeup(&lfs_dirvcount);
    372       1.31  perseant 			fs->lfs_unlockvp = vp;
    373       1.72        ad 			mutex_exit(&lfs_lock);
    374       1.31  perseant 			vrele(vp);
    375       1.72        ad 			mutex_enter(&lfs_lock);
    376       1.31  perseant 			fs->lfs_unlockvp = NULL;
    377       1.72        ad 		}
    378       1.31  perseant 	}
    379       1.40  perseant 
    380       1.40  perseant 	fs->lfs_flags &= ~LFS_UNDIROP;
    381       1.51  perseant 	wakeup(&fs->lfs_flags);
    382       1.72        ad 	mutex_exit(&lfs_lock);
    383       1.31  perseant }
    384       1.31  perseant 
    385       1.31  perseant static void
    386       1.31  perseant lfs_auto_segclean(struct lfs *fs)
    387       1.31  perseant {
    388       1.48  perseant 	int i, error, s, waited;
    389       1.31  perseant 
    390       1.51  perseant 	ASSERT_SEGLOCK(fs);
    391       1.31  perseant 	/*
    392       1.31  perseant 	 * Now that we've swapped lfs_activesb, but while we still
    393       1.31  perseant 	 * hold the segment lock, run through the segment list marking
    394       1.31  perseant 	 * the empty ones clean.
    395       1.31  perseant 	 * XXX - do we really need to do them all at once?
    396       1.31  perseant 	 */
    397       1.48  perseant 	waited = 0;
    398       1.31  perseant 	for (i = 0; i < fs->lfs_nseg; i++) {
    399       1.31  perseant 		if ((fs->lfs_suflags[0][i] &
    400       1.31  perseant 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    401       1.31  perseant 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
    402       1.31  perseant 		    (fs->lfs_suflags[1][i] &
    403       1.31  perseant 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    404       1.31  perseant 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
    405       1.31  perseant 
    406       1.48  perseant 			/* Make sure the sb is written before we clean */
    407       1.72        ad 			mutex_enter(&lfs_lock);
    408       1.48  perseant 			s = splbio();
    409       1.48  perseant 			while (waited == 0 && fs->lfs_sbactive)
    410       1.72        ad 				mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
    411       1.72        ad 					0, &lfs_lock);
    412       1.48  perseant 			splx(s);
    413       1.72        ad 			mutex_exit(&lfs_lock);
    414       1.48  perseant 			waited = 1;
    415       1.48  perseant 
    416       1.31  perseant 			if ((error = lfs_do_segclean(fs, i)) != 0) {
    417       1.50  perseant 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
    418       1.31  perseant 			}
    419       1.31  perseant 		}
    420       1.31  perseant 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
    421       1.31  perseant 			fs->lfs_suflags[fs->lfs_activesb][i];
    422       1.31  perseant 	}
    423       1.31  perseant }
    424       1.31  perseant 
    425        1.1   mycroft /*
    426        1.1   mycroft  * lfs_segunlock --
    427        1.1   mycroft  *	Single thread the segment writer.
    428        1.1   mycroft  */
    429        1.1   mycroft void
    430       1.18  perseant lfs_segunlock(struct lfs *fs)
    431        1.1   mycroft {
    432        1.1   mycroft 	struct segment *sp;
    433        1.1   mycroft 	unsigned long sync, ckp;
    434       1.22  perseant 	struct buf *bp;
    435       1.40  perseant 	int do_unmark_dirop = 0;
    436       1.49     perry 
    437       1.15  perseant 	sp = fs->lfs_sp;
    438       1.15  perseant 
    439       1.72        ad 	mutex_enter(&lfs_lock);
    440       1.72        ad 	KASSERT(LFS_SEGLOCK_HELD(fs));
    441       1.15  perseant 	if (fs->lfs_seglock == 1) {
    442       1.57  perseant 		if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0 &&
    443       1.57  perseant 		    LFS_STARVED_FOR_SEGS(fs) == 0)
    444       1.40  perseant 			do_unmark_dirop = 1;
    445       1.72        ad 		mutex_exit(&lfs_lock);
    446        1.1   mycroft 		sync = sp->seg_flags & SEGM_SYNC;
    447        1.1   mycroft 		ckp = sp->seg_flags & SEGM_CKP;
    448       1.57  perseant 
    449       1.57  perseant 		/* We should have a segment summary, and nothing else */
    450       1.57  perseant 		KASSERT(sp->cbpp == sp->bpp + 1);
    451       1.57  perseant 
    452       1.57  perseant 		/* Free allocated segment summary */
    453       1.57  perseant 		fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
    454       1.57  perseant 		bp = *sp->bpp;
    455       1.57  perseant 		lfs_freebuf(fs, bp);
    456        1.8  perseant 
    457       1.31  perseant 		pool_put(&fs->lfs_bpppool, sp->bpp);
    458       1.18  perseant 		sp->bpp = NULL;
    459       1.36  perseant 
    460       1.36  perseant 		/*
    461       1.36  perseant 		 * If we're not sync, we're done with sp, get rid of it.
    462       1.36  perseant 		 * Otherwise, we keep a local copy around but free
    463       1.36  perseant 		 * fs->lfs_sp so another process can use it (we have to
    464       1.36  perseant 		 * wait but they don't have to wait for us).
    465       1.36  perseant 		 */
    466       1.26  perseant 		if (!sync)
    467       1.31  perseant 			pool_put(&fs->lfs_segpool, sp);
    468       1.18  perseant 		fs->lfs_sp = NULL;
    469        1.1   mycroft 
    470        1.1   mycroft 		/*
    471        1.1   mycroft 		 * If the I/O count is non-zero, sleep until it reaches zero.
    472        1.1   mycroft 		 * At the moment, the user's process hangs around so we can
    473        1.1   mycroft 		 * sleep.
    474        1.1   mycroft 		 */
    475       1.72        ad 		mutex_enter(&lfs_lock);
    476       1.63  christos 		if (--fs->lfs_iocount == 0) {
    477       1.44      yamt 			LFS_DEBUG_COUNTLOCKED("lfs_segunlock");
    478       1.63  christos 		}
    479       1.36  perseant 		if (fs->lfs_iocount <= 1)
    480       1.22  perseant 			wakeup(&fs->lfs_iocount);
    481       1.72        ad 		mutex_exit(&lfs_lock);
    482        1.1   mycroft 		/*
    483       1.26  perseant 		 * If we're not checkpointing, we don't have to block
    484       1.26  perseant 		 * other processes to wait for a synchronous write
    485       1.26  perseant 		 * to complete.
    486       1.26  perseant 		 */
    487       1.26  perseant 		if (!ckp) {
    488       1.51  perseant #ifdef DEBUG
    489       1.51  perseant 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    490       1.51  perseant #endif
    491       1.72        ad 			mutex_enter(&lfs_lock);
    492       1.26  perseant 			--fs->lfs_seglock;
    493       1.26  perseant 			fs->lfs_lockpid = 0;
    494       1.58  perseant 			fs->lfs_locklwp = 0;
    495       1.72        ad 			mutex_exit(&lfs_lock);
    496       1.26  perseant 			wakeup(&fs->lfs_seglock);
    497       1.26  perseant 		}
    498       1.26  perseant 		/*
    499        1.1   mycroft 		 * We let checkpoints happen asynchronously.  That means
    500        1.1   mycroft 		 * that during recovery, we have to roll forward between
    501        1.1   mycroft 		 * the two segments described by the first and second
    502        1.1   mycroft 		 * superblocks to make sure that the checkpoint described
    503        1.1   mycroft 		 * by a superblock completed.
    504        1.1   mycroft 		 */
    505       1.72        ad 		mutex_enter(&lfs_lock);
    506  1.73.20.1  uebayasi 		while (ckp && sync && fs->lfs_iocount) {
    507       1.72        ad 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
    508       1.72        ad 				      "lfs_iocount", 0, &lfs_lock);
    509  1.73.20.1  uebayasi 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", fs, fs->lfs_iocount));
    510  1.73.20.1  uebayasi 		}
    511       1.26  perseant 		while (sync && sp->seg_iocount) {
    512       1.72        ad 			(void)mtsleep(&sp->seg_iocount, PRIBIO + 1,
    513       1.72        ad 				     "seg_iocount", 0, &lfs_lock);
    514       1.50  perseant 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
    515       1.26  perseant 		}
    516       1.72        ad 		mutex_exit(&lfs_lock);
    517       1.26  perseant 		if (sync)
    518       1.31  perseant 			pool_put(&fs->lfs_segpool, sp);
    519       1.36  perseant 
    520        1.1   mycroft 		if (ckp) {
    521        1.1   mycroft 			fs->lfs_nactive = 0;
    522        1.8  perseant 			/* If we *know* everything's on disk, write both sbs */
    523       1.33  perseant 			/* XXX should wait for this one	 */
    524       1.21       chs 			if (sync)
    525       1.31  perseant 				lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
    526       1.31  perseant 			lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
    527       1.48  perseant 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
    528       1.35  perseant 				lfs_auto_segclean(fs);
    529       1.48  perseant 				/* If sync, we can clean the remainder too */
    530       1.48  perseant 				if (sync)
    531       1.48  perseant 					lfs_auto_segclean(fs);
    532       1.48  perseant 			}
    533        1.8  perseant 			fs->lfs_activesb = 1 - fs->lfs_activesb;
    534       1.51  perseant #ifdef DEBUG
    535       1.51  perseant 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    536       1.51  perseant #endif
    537       1.72        ad 			mutex_enter(&lfs_lock);
    538       1.26  perseant 			--fs->lfs_seglock;
    539       1.26  perseant 			fs->lfs_lockpid = 0;
    540       1.58  perseant 			fs->lfs_locklwp = 0;
    541       1.72        ad 			mutex_exit(&lfs_lock);
    542       1.26  perseant 			wakeup(&fs->lfs_seglock);
    543        1.1   mycroft 		}
    544       1.27  perseant 		/* Reenable fragment size changes */
    545       1.66        ad 		rw_exit(&fs->lfs_fraglock);
    546       1.40  perseant 		if (do_unmark_dirop)
    547       1.40  perseant 			lfs_unmark_dirop(fs);
    548        1.1   mycroft 	} else if (fs->lfs_seglock == 0) {
    549       1.72        ad 		mutex_exit(&lfs_lock);
    550        1.1   mycroft 		panic ("Seglock not held");
    551        1.1   mycroft 	} else {
    552        1.1   mycroft 		--fs->lfs_seglock;
    553       1.72        ad 		mutex_exit(&lfs_lock);
    554        1.1   mycroft 	}
    555       1.41      yamt }
    556       1.41      yamt 
    557       1.41      yamt /*
    558       1.69  perseant  * Drain dirops and start writer.
    559       1.69  perseant  *
    560       1.69  perseant  * No simple_locks are held when we enter and none are held when we return.
    561       1.41      yamt  */
    562       1.41      yamt int
    563       1.41      yamt lfs_writer_enter(struct lfs *fs, const char *wmesg)
    564       1.41      yamt {
    565       1.41      yamt 	int error = 0;
    566       1.41      yamt 
    567       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    568       1.72        ad 	mutex_enter(&lfs_lock);
    569       1.41      yamt 
    570       1.41      yamt 	/* disallow dirops during flush */
    571       1.41      yamt 	fs->lfs_writer++;
    572       1.41      yamt 
    573       1.41      yamt 	while (fs->lfs_dirops > 0) {
    574       1.49     perry 		++fs->lfs_diropwait;
    575       1.72        ad 		error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
    576       1.72        ad 				&lfs_lock);
    577       1.49     perry 		--fs->lfs_diropwait;
    578       1.41      yamt 	}
    579       1.41      yamt 
    580       1.41      yamt 	if (error)
    581       1.41      yamt 		fs->lfs_writer--;
    582       1.41      yamt 
    583       1.72        ad 	mutex_exit(&lfs_lock);
    584       1.41      yamt 
    585       1.41      yamt 	return error;
    586       1.41      yamt }
    587       1.41      yamt 
    588       1.41      yamt void
    589       1.41      yamt lfs_writer_leave(struct lfs *fs)
    590       1.41      yamt {
    591       1.67   thorpej 	bool dowakeup;
    592       1.41      yamt 
    593       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    594       1.72        ad 	mutex_enter(&lfs_lock);
    595       1.41      yamt 	dowakeup = !(--fs->lfs_writer);
    596       1.72        ad 	mutex_exit(&lfs_lock);
    597       1.41      yamt 	if (dowakeup)
    598       1.41      yamt 		wakeup(&fs->lfs_dirops);
    599        1.1   mycroft }
    600       1.57  perseant 
    601       1.57  perseant /*
    602       1.57  perseant  * Unlock, wait for the cleaner, then relock to where we were before.
    603       1.57  perseant  * To be used only at a fairly high level, to address a paucity of free
    604       1.57  perseant  * segments propagated back from lfs_gop_write().
    605       1.57  perseant  */
    606       1.57  perseant void
    607       1.57  perseant lfs_segunlock_relock(struct lfs *fs)
    608       1.57  perseant {
    609       1.57  perseant 	int n = fs->lfs_seglock;
    610       1.57  perseant 	u_int16_t seg_flags;
    611       1.61  perseant 	CLEANERINFO *cip;
    612       1.61  perseant 	struct buf *bp;
    613       1.57  perseant 
    614       1.57  perseant 	if (n == 0)
    615       1.57  perseant 		return;
    616       1.57  perseant 
    617       1.57  perseant 	/* Write anything we've already gathered to disk */
    618       1.57  perseant 	lfs_writeseg(fs, fs->lfs_sp);
    619       1.57  perseant 
    620       1.61  perseant 	/* Tell cleaner */
    621       1.61  perseant 	LFS_CLEANERINFO(cip, fs, bp);
    622       1.61  perseant 	cip->flags |= LFS_CLEANER_MUST_CLEAN;
    623       1.61  perseant 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    624       1.61  perseant 
    625       1.57  perseant 	/* Save segment flags for later */
    626       1.57  perseant 	seg_flags = fs->lfs_sp->seg_flags;
    627       1.57  perseant 
    628       1.57  perseant 	fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */
    629       1.57  perseant 	while(fs->lfs_seglock)
    630       1.57  perseant 		lfs_segunlock(fs);
    631       1.57  perseant 
    632       1.57  perseant 	/* Wait for the cleaner */
    633       1.60  perseant 	lfs_wakeup_cleaner(fs);
    634       1.72        ad 	mutex_enter(&lfs_lock);
    635       1.57  perseant 	while (LFS_STARVED_FOR_SEGS(fs))
    636       1.72        ad 		mtsleep(&fs->lfs_avail, PRIBIO, "relock", 0,
    637       1.72        ad 			&lfs_lock);
    638       1.72        ad 	mutex_exit(&lfs_lock);
    639       1.57  perseant 
    640       1.57  perseant 	/* Put the segment lock back the way it was. */
    641       1.57  perseant 	while(n--)
    642       1.57  perseant 		lfs_seglock(fs, seg_flags);
    643       1.57  perseant 
    644       1.61  perseant 	/* Cleaner can relax now */
    645       1.61  perseant 	LFS_CLEANERINFO(cip, fs, bp);
    646       1.61  perseant 	cip->flags &= ~LFS_CLEANER_MUST_CLEAN;
    647       1.61  perseant 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    648       1.61  perseant 
    649       1.57  perseant 	return;
    650       1.57  perseant }
    651       1.60  perseant 
    652       1.60  perseant /*
    653       1.60  perseant  * Wake up the cleaner, provided that nowrap is not set.
    654       1.60  perseant  */
    655       1.60  perseant void
    656       1.60  perseant lfs_wakeup_cleaner(struct lfs *fs)
    657       1.60  perseant {
    658       1.60  perseant 	if (fs->lfs_nowrap > 0)
    659       1.60  perseant 		return;
    660       1.60  perseant 
    661       1.60  perseant 	wakeup(&fs->lfs_nextseg);
    662       1.60  perseant 	wakeup(&lfs_allclean_wakeup);
    663       1.60  perseant }
    664