Home | History | Annotate | Line # | Download | only in lfs
lfs_subr.c revision 1.71.10.1
      1  1.71.10.1    bouyer /*	$NetBSD: lfs_subr.c,v 1.71.10.1 2008/01/02 21:58:26 bouyer 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  * 3. All advertising materials mentioning features or use of this software
     19        1.8  perseant  *    must display the following acknowledgement:
     20       1.33  perseant  *	This product includes software developed by the NetBSD
     21       1.33  perseant  *	Foundation, Inc. and its contributors.
     22        1.8  perseant  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23        1.8  perseant  *    contributors may be used to endorse or promote products derived
     24        1.8  perseant  *    from this software without specific prior written permission.
     25        1.8  perseant  *
     26        1.8  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27        1.8  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28        1.8  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29        1.8  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30        1.8  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31        1.8  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32        1.8  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33        1.8  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34        1.8  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35        1.8  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36        1.8  perseant  * POSSIBILITY OF SUCH DAMAGE.
     37        1.8  perseant  */
     38        1.1   mycroft /*
     39        1.1   mycroft  * Copyright (c) 1991, 1993
     40        1.1   mycroft  *	The Regents of the University of California.  All rights reserved.
     41        1.1   mycroft  *
     42        1.1   mycroft  * Redistribution and use in source and binary forms, with or without
     43        1.1   mycroft  * modification, are permitted provided that the following conditions
     44        1.1   mycroft  * are met:
     45        1.1   mycroft  * 1. Redistributions of source code must retain the above copyright
     46        1.1   mycroft  *    notice, this list of conditions and the following disclaimer.
     47        1.1   mycroft  * 2. Redistributions in binary form must reproduce the above copyright
     48        1.1   mycroft  *    notice, this list of conditions and the following disclaimer in the
     49        1.1   mycroft  *    documentation and/or other materials provided with the distribution.
     50       1.43       agc  * 3. Neither the name of the University nor the names of its contributors
     51        1.1   mycroft  *    may be used to endorse or promote products derived from this software
     52        1.1   mycroft  *    without specific prior written permission.
     53        1.1   mycroft  *
     54        1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     55        1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     56        1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     57        1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     58        1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     59        1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     60        1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     61        1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     62        1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     63        1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     64        1.1   mycroft  * SUCH DAMAGE.
     65        1.1   mycroft  *
     66        1.6      fvdl  *	@(#)lfs_subr.c	8.4 (Berkeley) 5/8/95
     67        1.1   mycroft  */
     68       1.20     lukem 
     69       1.20     lukem #include <sys/cdefs.h>
     70  1.71.10.1    bouyer __KERNEL_RCSID(0, "$NetBSD: lfs_subr.c,v 1.71.10.1 2008/01/02 21:58:26 bouyer Exp $");
     71        1.1   mycroft 
     72        1.1   mycroft #include <sys/param.h>
     73        1.3  christos #include <sys/systm.h>
     74        1.1   mycroft #include <sys/namei.h>
     75        1.1   mycroft #include <sys/vnode.h>
     76        1.1   mycroft #include <sys/buf.h>
     77        1.1   mycroft #include <sys/mount.h>
     78        1.1   mycroft #include <sys/malloc.h>
     79        1.1   mycroft #include <sys/proc.h>
     80       1.61  perseant #include <sys/kauth.h>
     81        1.1   mycroft 
     82        1.1   mycroft #include <ufs/ufs/inode.h>
     83        1.1   mycroft #include <ufs/lfs/lfs.h>
     84        1.1   mycroft #include <ufs/lfs/lfs_extern.h>
     85        1.1   mycroft 
     86       1.31  perseant #include <uvm/uvm.h>
     87       1.31  perseant 
     88       1.50  perseant #ifdef DEBUG
     89       1.53  christos const char *lfs_res_names[LFS_NB_COUNT] = {
     90       1.31  perseant 	"summary",
     91       1.31  perseant 	"superblock",
     92       1.52  perseant 	"file block",
     93       1.31  perseant 	"cluster",
     94       1.31  perseant 	"clean",
     95       1.52  perseant 	"blkiov",
     96       1.31  perseant };
     97       1.31  perseant #endif
     98       1.31  perseant 
     99       1.31  perseant int lfs_res_qty[LFS_NB_COUNT] = {
    100       1.31  perseant 	LFS_N_SUMMARIES,
    101       1.31  perseant 	LFS_N_SBLOCKS,
    102       1.31  perseant 	LFS_N_IBLOCKS,
    103       1.31  perseant 	LFS_N_CLUSTERS,
    104       1.31  perseant 	LFS_N_CLEAN,
    105       1.52  perseant 	LFS_N_BLKIOV,
    106       1.31  perseant };
    107       1.31  perseant 
    108       1.31  perseant void
    109       1.31  perseant lfs_setup_resblks(struct lfs *fs)
    110       1.31  perseant {
    111       1.31  perseant 	int i, j;
    112       1.31  perseant 	int maxbpp;
    113       1.31  perseant 
    114       1.51  perseant 	ASSERT_NO_SEGLOCK(fs);
    115       1.31  perseant 	fs->lfs_resblk = (res_t *)malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
    116       1.33  perseant 					  M_WAITOK);
    117       1.31  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    118       1.31  perseant 		fs->lfs_resblk[i].inuse = 0;
    119       1.31  perseant 		fs->lfs_resblk[i].p = NULL;
    120       1.31  perseant 	}
    121       1.31  perseant 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
    122       1.31  perseant 		LIST_INIT(fs->lfs_reshash + i);
    123       1.31  perseant 
    124       1.31  perseant 	/*
    125       1.31  perseant 	 * These types of allocations can be larger than a page,
    126       1.31  perseant 	 * so we can't use the pool subsystem for them.
    127       1.31  perseant 	 */
    128       1.31  perseant 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
    129       1.34  perseant 		fs->lfs_resblk[i].size = fs->lfs_sumsize;
    130       1.31  perseant 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
    131       1.34  perseant 		fs->lfs_resblk[i].size = LFS_SBPAD;
    132       1.31  perseant 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
    133       1.34  perseant 		fs->lfs_resblk[i].size = fs->lfs_bsize;
    134       1.31  perseant 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
    135       1.34  perseant 		fs->lfs_resblk[i].size = MAXPHYS;
    136       1.31  perseant 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
    137       1.34  perseant 		fs->lfs_resblk[i].size = MAXPHYS;
    138       1.52  perseant 	for (j = 0; j < LFS_N_BLKIOV; j++, i++)
    139       1.52  perseant 		fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO);
    140       1.34  perseant 
    141       1.34  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    142       1.34  perseant 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
    143       1.34  perseant 					     M_SEGMENT, M_WAITOK);
    144       1.34  perseant 	}
    145       1.31  perseant 
    146       1.31  perseant 	/*
    147       1.31  perseant 	 * Initialize pools for small types (XXX is BPP small?)
    148       1.31  perseant 	 */
    149       1.46    simonb 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
    150       1.68        ad 		"lfsclpl", &pool_allocator_nointr, IPL_NONE);
    151       1.46    simonb 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
    152       1.68        ad 		"lfssegpool", &pool_allocator_nointr, IPL_NONE);
    153       1.31  perseant 	maxbpp = ((fs->lfs_sumsize - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
    154       1.47      yamt 	maxbpp = MIN(maxbpp, segsize(fs) / fs->lfs_fsize + 2);
    155       1.46    simonb 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
    156       1.68        ad 		"lfsbpppl", &pool_allocator_nointr, IPL_NONE);
    157       1.31  perseant }
    158       1.31  perseant 
    159       1.31  perseant void
    160       1.31  perseant lfs_free_resblks(struct lfs *fs)
    161       1.31  perseant {
    162       1.31  perseant 	int i;
    163       1.31  perseant 
    164       1.31  perseant 	pool_destroy(&fs->lfs_bpppool);
    165       1.31  perseant 	pool_destroy(&fs->lfs_segpool);
    166       1.31  perseant 	pool_destroy(&fs->lfs_clpool);
    167       1.31  perseant 
    168  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    169       1.31  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    170       1.39  perseant 		while (fs->lfs_resblk[i].inuse)
    171  1.71.10.1    bouyer 			mtsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
    172  1.71.10.1    bouyer 				&lfs_lock);
    173       1.31  perseant 		if (fs->lfs_resblk[i].p != NULL)
    174       1.31  perseant 			free(fs->lfs_resblk[i].p, M_SEGMENT);
    175       1.31  perseant 	}
    176       1.31  perseant 	free(fs->lfs_resblk, M_SEGMENT);
    177  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    178       1.31  perseant }
    179       1.31  perseant 
    180       1.31  perseant static unsigned int
    181       1.31  perseant lfs_mhash(void *vp)
    182       1.31  perseant {
    183       1.31  perseant 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
    184       1.31  perseant }
    185       1.31  perseant 
    186       1.31  perseant /*
    187       1.31  perseant  * Return memory of the given size for the given purpose, or use one of a
    188       1.31  perseant  * number of spare last-resort buffers, if malloc returns NULL.
    189       1.49     perry  */
    190       1.31  perseant void *
    191       1.31  perseant lfs_malloc(struct lfs *fs, size_t size, int type)
    192       1.31  perseant {
    193       1.31  perseant 	struct lfs_res_blk *re;
    194       1.31  perseant 	void *r;
    195       1.31  perseant 	int i, s, start;
    196       1.31  perseant 	unsigned int h;
    197       1.31  perseant 
    198       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    199       1.34  perseant 	r = NULL;
    200       1.34  perseant 
    201       1.31  perseant 	/* If no mem allocated for this type, it just waits */
    202       1.34  perseant 	if (lfs_res_qty[type] == 0) {
    203       1.34  perseant 		r = malloc(size, M_SEGMENT, M_WAITOK);
    204       1.34  perseant 		return r;
    205       1.34  perseant 	}
    206       1.31  perseant 
    207       1.31  perseant 	/* Otherwise try a quick malloc, and if it works, great */
    208       1.34  perseant 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
    209       1.31  perseant 		return r;
    210       1.34  perseant 	}
    211       1.31  perseant 
    212       1.31  perseant 	/*
    213       1.31  perseant 	 * If malloc returned NULL, we are forced to use one of our
    214       1.31  perseant 	 * reserve blocks.  We have on hand at least one summary block,
    215       1.31  perseant 	 * at least one cluster block, at least one superblock,
    216       1.31  perseant 	 * and several indirect blocks.
    217       1.31  perseant 	 */
    218       1.51  perseant 
    219  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    220       1.31  perseant 	/* skip over blocks of other types */
    221       1.31  perseant 	for (i = 0, start = 0; i < type; i++)
    222       1.31  perseant 		start += lfs_res_qty[i];
    223       1.31  perseant 	while (r == NULL) {
    224       1.31  perseant 		for (i = 0; i < lfs_res_qty[type]; i++) {
    225       1.31  perseant 			if (fs->lfs_resblk[start + i].inuse == 0) {
    226       1.31  perseant 				re = fs->lfs_resblk + start + i;
    227       1.31  perseant 				re->inuse = 1;
    228       1.31  perseant 				r = re->p;
    229       1.34  perseant 				KASSERT(re->size >= size);
    230       1.31  perseant 				h = lfs_mhash(r);
    231       1.31  perseant 				s = splbio();
    232       1.31  perseant 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
    233       1.31  perseant 				splx(s);
    234  1.71.10.1    bouyer 				mutex_exit(&lfs_lock);
    235       1.31  perseant 				return r;
    236       1.31  perseant 			}
    237       1.31  perseant 		}
    238       1.51  perseant 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
    239       1.51  perseant 		      lfs_res_names[type], lfs_res_qty[type]));
    240  1.71.10.1    bouyer 		mtsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
    241  1.71.10.1    bouyer 			&lfs_lock);
    242       1.51  perseant 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
    243       1.51  perseant 		      lfs_res_names[type]));
    244       1.31  perseant 	}
    245       1.31  perseant 	/* NOTREACHED */
    246  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    247       1.31  perseant 	return r;
    248       1.31  perseant }
    249       1.31  perseant 
    250       1.31  perseant void
    251       1.65  christos lfs_free(struct lfs *fs, void *p, int type)
    252       1.31  perseant {
    253       1.31  perseant 	int s;
    254       1.31  perseant 	unsigned int h;
    255       1.31  perseant 	res_t *re;
    256       1.32      yamt #ifdef DEBUG
    257       1.32      yamt 	int i;
    258       1.32      yamt #endif
    259       1.31  perseant 
    260       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    261       1.31  perseant 	h = lfs_mhash(p);
    262  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    263       1.31  perseant 	s = splbio();
    264       1.31  perseant 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
    265       1.31  perseant 		if (re->p == p) {
    266       1.32      yamt 			KASSERT(re->inuse == 1);
    267       1.31  perseant 			LIST_REMOVE(re, res);
    268       1.31  perseant 			re->inuse = 0;
    269       1.31  perseant 			wakeup(&fs->lfs_resblk);
    270       1.31  perseant 			splx(s);
    271  1.71.10.1    bouyer 			mutex_exit(&lfs_lock);
    272       1.31  perseant 			return;
    273       1.31  perseant 		}
    274       1.31  perseant 	}
    275       1.32      yamt #ifdef DEBUG
    276       1.32      yamt 	for (i = 0; i < LFS_N_TOTAL; i++) {
    277       1.32      yamt 		if (fs->lfs_resblk[i].p == p)
    278       1.34  perseant 			panic("lfs_free: inconsistent reserved block");
    279       1.32      yamt 	}
    280       1.32      yamt #endif
    281       1.31  perseant 	splx(s);
    282  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    283       1.51  perseant 
    284       1.31  perseant 	/*
    285       1.31  perseant 	 * If we didn't find it, free it.
    286       1.31  perseant 	 */
    287       1.31  perseant 	free(p, M_SEGMENT);
    288       1.31  perseant }
    289        1.1   mycroft 
    290        1.1   mycroft /*
    291        1.1   mycroft  * lfs_seglock --
    292        1.1   mycroft  *	Single thread the segment writer.
    293        1.1   mycroft  */
    294       1.31  perseant int
    295       1.18  perseant lfs_seglock(struct lfs *fs, unsigned long flags)
    296        1.1   mycroft {
    297        1.1   mycroft 	struct segment *sp;
    298       1.49     perry 
    299  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    300        1.7   thorpej 	if (fs->lfs_seglock) {
    301       1.58  perseant 		if (fs->lfs_lockpid == curproc->p_pid &&
    302       1.58  perseant 		    fs->lfs_locklwp == curlwp->l_lid) {
    303  1.71.10.1    bouyer 			mutex_exit(&lfs_lock);
    304        1.1   mycroft 			++fs->lfs_seglock;
    305        1.1   mycroft 			fs->lfs_sp->seg_flags |= flags;
    306       1.31  perseant 			return 0;
    307       1.38  perseant 		} else if (flags & SEGM_PAGEDAEMON) {
    308  1.71.10.1    bouyer 			mutex_exit(&lfs_lock);
    309       1.31  perseant 			return EWOULDBLOCK;
    310       1.51  perseant 		} else {
    311       1.51  perseant 			while (fs->lfs_seglock) {
    312  1.71.10.1    bouyer 				(void)mtsleep(&fs->lfs_seglock, PRIBIO + 1,
    313  1.71.10.1    bouyer 					"lfs_seglock", 0, &lfs_lock);
    314       1.51  perseant 			}
    315       1.51  perseant 		}
    316        1.7   thorpej 	}
    317       1.49     perry 
    318        1.1   mycroft 	fs->lfs_seglock = 1;
    319        1.1   mycroft 	fs->lfs_lockpid = curproc->p_pid;
    320       1.58  perseant 	fs->lfs_locklwp = curlwp->l_lid;
    321  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    322       1.36  perseant 	fs->lfs_cleanind = 0;
    323       1.36  perseant 
    324       1.51  perseant #ifdef DEBUG
    325       1.51  perseant 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
    326       1.51  perseant #endif
    327       1.27  perseant 	/* Drain fragment size changes out */
    328       1.66        ad 	rw_enter(&fs->lfs_fraglock, RW_WRITER);
    329       1.27  perseant 
    330       1.31  perseant 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
    331       1.31  perseant 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
    332        1.1   mycroft 	sp->seg_flags = flags;
    333        1.1   mycroft 	sp->vp = NULL;
    334       1.26  perseant 	sp->seg_iocount = 0;
    335        1.1   mycroft 	(void) lfs_initseg(fs);
    336       1.49     perry 
    337        1.1   mycroft 	/*
    338        1.1   mycroft 	 * Keep a cumulative count of the outstanding I/O operations.  If the
    339        1.1   mycroft 	 * disk drive catches up with us it could go to zero before we finish,
    340        1.1   mycroft 	 * so we artificially increment it by one until we've scheduled all of
    341        1.1   mycroft 	 * the writes we intend to do.
    342        1.1   mycroft 	 */
    343  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    344        1.1   mycroft 	++fs->lfs_iocount;
    345  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    346       1.31  perseant 	return 0;
    347        1.1   mycroft }
    348        1.8  perseant 
    349       1.31  perseant static void lfs_unmark_dirop(struct lfs *);
    350       1.31  perseant 
    351       1.31  perseant static void
    352       1.31  perseant lfs_unmark_dirop(struct lfs *fs)
    353       1.31  perseant {
    354       1.31  perseant 	struct inode *ip, *nip;
    355       1.31  perseant 	struct vnode *vp;
    356       1.40  perseant 	int doit;
    357       1.31  perseant 
    358       1.51  perseant 	ASSERT_NO_SEGLOCK(fs);
    359  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    360       1.40  perseant 	doit = !(fs->lfs_flags & LFS_UNDIROP);
    361       1.40  perseant 	if (doit)
    362       1.40  perseant 		fs->lfs_flags |= LFS_UNDIROP;
    363       1.51  perseant 	if (!doit) {
    364  1.71.10.1    bouyer 		mutex_exit(&lfs_lock);
    365       1.40  perseant 		return;
    366       1.51  perseant 	}
    367       1.40  perseant 
    368       1.31  perseant 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
    369       1.31  perseant 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
    370       1.31  perseant 		vp = ITOV(ip);
    371  1.71.10.1    bouyer 		if (VOP_ISLOCKED(vp) == LK_EXCLOTHER)
    372       1.31  perseant 			continue;
    373       1.61  perseant 		if ((VTOI(vp)->i_flag & (IN_ADIROP | IN_ALLMOD)) == 0) {
    374       1.31  perseant 			--lfs_dirvcount;
    375       1.59  perseant 			--fs->lfs_dirvcount;
    376       1.71        ad 			vp->v_uflag &= ~VU_DIROP;
    377       1.31  perseant 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
    378       1.31  perseant 			wakeup(&lfs_dirvcount);
    379       1.31  perseant 			fs->lfs_unlockvp = vp;
    380  1.71.10.1    bouyer 			mutex_exit(&lfs_lock);
    381       1.31  perseant 			vrele(vp);
    382  1.71.10.1    bouyer 			mutex_enter(&lfs_lock);
    383       1.31  perseant 			fs->lfs_unlockvp = NULL;
    384  1.71.10.1    bouyer 		}
    385       1.31  perseant 	}
    386       1.40  perseant 
    387       1.40  perseant 	fs->lfs_flags &= ~LFS_UNDIROP;
    388       1.51  perseant 	wakeup(&fs->lfs_flags);
    389  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    390       1.31  perseant }
    391       1.31  perseant 
    392       1.31  perseant static void
    393       1.31  perseant lfs_auto_segclean(struct lfs *fs)
    394       1.31  perseant {
    395       1.48  perseant 	int i, error, s, waited;
    396       1.31  perseant 
    397       1.51  perseant 	ASSERT_SEGLOCK(fs);
    398       1.31  perseant 	/*
    399       1.31  perseant 	 * Now that we've swapped lfs_activesb, but while we still
    400       1.31  perseant 	 * hold the segment lock, run through the segment list marking
    401       1.31  perseant 	 * the empty ones clean.
    402       1.31  perseant 	 * XXX - do we really need to do them all at once?
    403       1.31  perseant 	 */
    404       1.48  perseant 	waited = 0;
    405       1.31  perseant 	for (i = 0; i < fs->lfs_nseg; i++) {
    406       1.31  perseant 		if ((fs->lfs_suflags[0][i] &
    407       1.31  perseant 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    408       1.31  perseant 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
    409       1.31  perseant 		    (fs->lfs_suflags[1][i] &
    410       1.31  perseant 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    411       1.31  perseant 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
    412       1.31  perseant 
    413       1.48  perseant 			/* Make sure the sb is written before we clean */
    414  1.71.10.1    bouyer 			mutex_enter(&lfs_lock);
    415       1.48  perseant 			s = splbio();
    416       1.48  perseant 			while (waited == 0 && fs->lfs_sbactive)
    417  1.71.10.1    bouyer 				mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
    418  1.71.10.1    bouyer 					0, &lfs_lock);
    419       1.48  perseant 			splx(s);
    420  1.71.10.1    bouyer 			mutex_exit(&lfs_lock);
    421       1.48  perseant 			waited = 1;
    422       1.48  perseant 
    423       1.31  perseant 			if ((error = lfs_do_segclean(fs, i)) != 0) {
    424       1.50  perseant 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
    425       1.31  perseant 			}
    426       1.31  perseant 		}
    427       1.31  perseant 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
    428       1.31  perseant 			fs->lfs_suflags[fs->lfs_activesb][i];
    429       1.31  perseant 	}
    430       1.31  perseant }
    431       1.31  perseant 
    432        1.1   mycroft /*
    433        1.1   mycroft  * lfs_segunlock --
    434        1.1   mycroft  *	Single thread the segment writer.
    435        1.1   mycroft  */
    436        1.1   mycroft void
    437       1.18  perseant lfs_segunlock(struct lfs *fs)
    438        1.1   mycroft {
    439        1.1   mycroft 	struct segment *sp;
    440        1.1   mycroft 	unsigned long sync, ckp;
    441       1.22  perseant 	struct buf *bp;
    442       1.40  perseant 	int do_unmark_dirop = 0;
    443       1.49     perry 
    444       1.15  perseant 	sp = fs->lfs_sp;
    445       1.15  perseant 
    446  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    447  1.71.10.1    bouyer 	KASSERT(LFS_SEGLOCK_HELD(fs));
    448       1.15  perseant 	if (fs->lfs_seglock == 1) {
    449       1.57  perseant 		if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0 &&
    450       1.57  perseant 		    LFS_STARVED_FOR_SEGS(fs) == 0)
    451       1.40  perseant 			do_unmark_dirop = 1;
    452  1.71.10.1    bouyer 		mutex_exit(&lfs_lock);
    453        1.1   mycroft 		sync = sp->seg_flags & SEGM_SYNC;
    454        1.1   mycroft 		ckp = sp->seg_flags & SEGM_CKP;
    455       1.57  perseant 
    456       1.57  perseant 		/* We should have a segment summary, and nothing else */
    457       1.57  perseant 		KASSERT(sp->cbpp == sp->bpp + 1);
    458       1.57  perseant 
    459       1.57  perseant 		/* Free allocated segment summary */
    460       1.57  perseant 		fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
    461       1.57  perseant 		bp = *sp->bpp;
    462       1.57  perseant 		lfs_freebuf(fs, bp);
    463        1.8  perseant 
    464       1.31  perseant 		pool_put(&fs->lfs_bpppool, sp->bpp);
    465       1.18  perseant 		sp->bpp = NULL;
    466       1.36  perseant 
    467       1.36  perseant 		/*
    468       1.36  perseant 		 * If we're not sync, we're done with sp, get rid of it.
    469       1.36  perseant 		 * Otherwise, we keep a local copy around but free
    470       1.36  perseant 		 * fs->lfs_sp so another process can use it (we have to
    471       1.36  perseant 		 * wait but they don't have to wait for us).
    472       1.36  perseant 		 */
    473       1.26  perseant 		if (!sync)
    474       1.31  perseant 			pool_put(&fs->lfs_segpool, sp);
    475       1.18  perseant 		fs->lfs_sp = NULL;
    476        1.1   mycroft 
    477        1.1   mycroft 		/*
    478        1.1   mycroft 		 * If the I/O count is non-zero, sleep until it reaches zero.
    479        1.1   mycroft 		 * At the moment, the user's process hangs around so we can
    480        1.1   mycroft 		 * sleep.
    481        1.1   mycroft 		 */
    482  1.71.10.1    bouyer 		mutex_enter(&lfs_lock);
    483       1.63  christos 		if (--fs->lfs_iocount == 0) {
    484       1.44      yamt 			LFS_DEBUG_COUNTLOCKED("lfs_segunlock");
    485       1.63  christos 		}
    486       1.36  perseant 		if (fs->lfs_iocount <= 1)
    487       1.22  perseant 			wakeup(&fs->lfs_iocount);
    488  1.71.10.1    bouyer 		mutex_exit(&lfs_lock);
    489        1.1   mycroft 		/*
    490       1.26  perseant 		 * If we're not checkpointing, we don't have to block
    491       1.26  perseant 		 * other processes to wait for a synchronous write
    492       1.26  perseant 		 * to complete.
    493       1.26  perseant 		 */
    494       1.26  perseant 		if (!ckp) {
    495       1.51  perseant #ifdef DEBUG
    496       1.51  perseant 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    497       1.51  perseant #endif
    498  1.71.10.1    bouyer 			mutex_enter(&lfs_lock);
    499       1.26  perseant 			--fs->lfs_seglock;
    500       1.26  perseant 			fs->lfs_lockpid = 0;
    501       1.58  perseant 			fs->lfs_locklwp = 0;
    502  1.71.10.1    bouyer 			mutex_exit(&lfs_lock);
    503       1.26  perseant 			wakeup(&fs->lfs_seglock);
    504       1.26  perseant 		}
    505       1.26  perseant 		/*
    506        1.1   mycroft 		 * We let checkpoints happen asynchronously.  That means
    507        1.1   mycroft 		 * that during recovery, we have to roll forward between
    508        1.1   mycroft 		 * the two segments described by the first and second
    509        1.1   mycroft 		 * superblocks to make sure that the checkpoint described
    510        1.1   mycroft 		 * by a superblock completed.
    511        1.1   mycroft 		 */
    512  1.71.10.1    bouyer 		mutex_enter(&lfs_lock);
    513       1.26  perseant 		while (ckp && sync && fs->lfs_iocount)
    514  1.71.10.1    bouyer 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
    515  1.71.10.1    bouyer 				      "lfs_iocount", 0, &lfs_lock);
    516       1.26  perseant 		while (sync && sp->seg_iocount) {
    517  1.71.10.1    bouyer 			(void)mtsleep(&sp->seg_iocount, PRIBIO + 1,
    518  1.71.10.1    bouyer 				     "seg_iocount", 0, &lfs_lock);
    519       1.50  perseant 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
    520       1.26  perseant 		}
    521  1.71.10.1    bouyer 		mutex_exit(&lfs_lock);
    522       1.26  perseant 		if (sync)
    523       1.31  perseant 			pool_put(&fs->lfs_segpool, sp);
    524       1.36  perseant 
    525        1.1   mycroft 		if (ckp) {
    526        1.1   mycroft 			fs->lfs_nactive = 0;
    527        1.8  perseant 			/* If we *know* everything's on disk, write both sbs */
    528       1.33  perseant 			/* XXX should wait for this one	 */
    529       1.21       chs 			if (sync)
    530       1.31  perseant 				lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
    531       1.31  perseant 			lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
    532       1.48  perseant 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
    533       1.35  perseant 				lfs_auto_segclean(fs);
    534       1.48  perseant 				/* If sync, we can clean the remainder too */
    535       1.48  perseant 				if (sync)
    536       1.48  perseant 					lfs_auto_segclean(fs);
    537       1.48  perseant 			}
    538        1.8  perseant 			fs->lfs_activesb = 1 - fs->lfs_activesb;
    539       1.51  perseant #ifdef DEBUG
    540       1.51  perseant 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    541       1.51  perseant #endif
    542  1.71.10.1    bouyer 			mutex_enter(&lfs_lock);
    543       1.26  perseant 			--fs->lfs_seglock;
    544       1.26  perseant 			fs->lfs_lockpid = 0;
    545       1.58  perseant 			fs->lfs_locklwp = 0;
    546  1.71.10.1    bouyer 			mutex_exit(&lfs_lock);
    547       1.26  perseant 			wakeup(&fs->lfs_seglock);
    548        1.1   mycroft 		}
    549       1.27  perseant 		/* Reenable fragment size changes */
    550       1.66        ad 		rw_exit(&fs->lfs_fraglock);
    551       1.40  perseant 		if (do_unmark_dirop)
    552       1.40  perseant 			lfs_unmark_dirop(fs);
    553        1.1   mycroft 	} else if (fs->lfs_seglock == 0) {
    554  1.71.10.1    bouyer 		mutex_exit(&lfs_lock);
    555        1.1   mycroft 		panic ("Seglock not held");
    556        1.1   mycroft 	} else {
    557        1.1   mycroft 		--fs->lfs_seglock;
    558  1.71.10.1    bouyer 		mutex_exit(&lfs_lock);
    559        1.1   mycroft 	}
    560       1.41      yamt }
    561       1.41      yamt 
    562       1.41      yamt /*
    563       1.69  perseant  * Drain dirops and start writer.
    564       1.69  perseant  *
    565       1.69  perseant  * No simple_locks are held when we enter and none are held when we return.
    566       1.41      yamt  */
    567       1.41      yamt int
    568       1.41      yamt lfs_writer_enter(struct lfs *fs, const char *wmesg)
    569       1.41      yamt {
    570       1.41      yamt 	int error = 0;
    571       1.41      yamt 
    572       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    573  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    574       1.41      yamt 
    575       1.41      yamt 	/* disallow dirops during flush */
    576       1.41      yamt 	fs->lfs_writer++;
    577       1.41      yamt 
    578       1.41      yamt 	while (fs->lfs_dirops > 0) {
    579       1.49     perry 		++fs->lfs_diropwait;
    580  1.71.10.1    bouyer 		error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
    581  1.71.10.1    bouyer 				&lfs_lock);
    582       1.49     perry 		--fs->lfs_diropwait;
    583       1.41      yamt 	}
    584       1.41      yamt 
    585       1.41      yamt 	if (error)
    586       1.41      yamt 		fs->lfs_writer--;
    587       1.41      yamt 
    588  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    589       1.41      yamt 
    590       1.41      yamt 	return error;
    591       1.41      yamt }
    592       1.41      yamt 
    593       1.41      yamt void
    594       1.41      yamt lfs_writer_leave(struct lfs *fs)
    595       1.41      yamt {
    596       1.67   thorpej 	bool dowakeup;
    597       1.41      yamt 
    598       1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    599  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    600       1.41      yamt 	dowakeup = !(--fs->lfs_writer);
    601  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    602       1.41      yamt 	if (dowakeup)
    603       1.41      yamt 		wakeup(&fs->lfs_dirops);
    604        1.1   mycroft }
    605       1.57  perseant 
    606       1.57  perseant /*
    607       1.57  perseant  * Unlock, wait for the cleaner, then relock to where we were before.
    608       1.57  perseant  * To be used only at a fairly high level, to address a paucity of free
    609       1.57  perseant  * segments propagated back from lfs_gop_write().
    610       1.57  perseant  */
    611       1.57  perseant void
    612       1.57  perseant lfs_segunlock_relock(struct lfs *fs)
    613       1.57  perseant {
    614       1.57  perseant 	int n = fs->lfs_seglock;
    615       1.57  perseant 	u_int16_t seg_flags;
    616       1.61  perseant 	CLEANERINFO *cip;
    617       1.61  perseant 	struct buf *bp;
    618       1.57  perseant 
    619       1.57  perseant 	if (n == 0)
    620       1.57  perseant 		return;
    621       1.57  perseant 
    622       1.57  perseant 	/* Write anything we've already gathered to disk */
    623       1.57  perseant 	lfs_writeseg(fs, fs->lfs_sp);
    624       1.57  perseant 
    625       1.61  perseant 	/* Tell cleaner */
    626       1.61  perseant 	LFS_CLEANERINFO(cip, fs, bp);
    627       1.61  perseant 	cip->flags |= LFS_CLEANER_MUST_CLEAN;
    628       1.61  perseant 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    629       1.61  perseant 
    630       1.57  perseant 	/* Save segment flags for later */
    631       1.57  perseant 	seg_flags = fs->lfs_sp->seg_flags;
    632       1.57  perseant 
    633       1.57  perseant 	fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */
    634       1.57  perseant 	while(fs->lfs_seglock)
    635       1.57  perseant 		lfs_segunlock(fs);
    636       1.57  perseant 
    637       1.57  perseant 	/* Wait for the cleaner */
    638       1.60  perseant 	lfs_wakeup_cleaner(fs);
    639  1.71.10.1    bouyer 	mutex_enter(&lfs_lock);
    640       1.57  perseant 	while (LFS_STARVED_FOR_SEGS(fs))
    641  1.71.10.1    bouyer 		mtsleep(&fs->lfs_avail, PRIBIO, "relock", 0,
    642  1.71.10.1    bouyer 			&lfs_lock);
    643  1.71.10.1    bouyer 	mutex_exit(&lfs_lock);
    644       1.57  perseant 
    645       1.57  perseant 	/* Put the segment lock back the way it was. */
    646       1.57  perseant 	while(n--)
    647       1.57  perseant 		lfs_seglock(fs, seg_flags);
    648       1.57  perseant 
    649       1.61  perseant 	/* Cleaner can relax now */
    650       1.61  perseant 	LFS_CLEANERINFO(cip, fs, bp);
    651       1.61  perseant 	cip->flags &= ~LFS_CLEANER_MUST_CLEAN;
    652       1.61  perseant 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
    653       1.61  perseant 
    654       1.57  perseant 	return;
    655       1.57  perseant }
    656       1.60  perseant 
    657       1.60  perseant /*
    658       1.60  perseant  * Wake up the cleaner, provided that nowrap is not set.
    659       1.60  perseant  */
    660       1.60  perseant void
    661       1.60  perseant lfs_wakeup_cleaner(struct lfs *fs)
    662       1.60  perseant {
    663       1.60  perseant 	if (fs->lfs_nowrap > 0)
    664       1.60  perseant 		return;
    665       1.60  perseant 
    666       1.60  perseant 	wakeup(&fs->lfs_nextseg);
    667       1.60  perseant 	wakeup(&lfs_allclean_wakeup);
    668       1.60  perseant }
    669