Home | History | Annotate | Line # | Download | only in lfs
lfs_subr.c revision 1.51
      1  1.51  perseant /*	$NetBSD: lfs_subr.c,v 1.51 2005/04/01 21:59:46 perseant 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  1.51  perseant __KERNEL_RCSID(0, "$NetBSD: lfs_subr.c,v 1.51 2005/04/01 21:59:46 perseant 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   1.1   mycroft 
     81   1.1   mycroft #include <ufs/ufs/inode.h>
     82   1.1   mycroft #include <ufs/lfs/lfs.h>
     83   1.1   mycroft #include <ufs/lfs/lfs_extern.h>
     84   1.1   mycroft 
     85  1.31  perseant #include <uvm/uvm.h>
     86  1.31  perseant 
     87   1.1   mycroft /*
     88   1.1   mycroft  * Return buffer with the contents of block "offset" from the beginning of
     89   1.1   mycroft  * directory "ip".  If "res" is non-zero, fill it in with a pointer to the
     90   1.1   mycroft  * remaining space in the directory.
     91   1.1   mycroft  */
     92   1.1   mycroft int
     93  1.18  perseant lfs_blkatoff(void *v)
     94   1.3  christos {
     95   1.1   mycroft 	struct vop_blkatoff_args /* {
     96   1.1   mycroft 		struct vnode *a_vp;
     97   1.1   mycroft 		off_t a_offset;
     98   1.1   mycroft 		char **a_res;
     99   1.1   mycroft 		struct buf **a_bpp;
    100   1.8  perseant 		} */ *ap = v;
    101  1.13  augustss 	struct lfs *fs;
    102   1.1   mycroft 	struct inode *ip;
    103   1.1   mycroft 	struct buf *bp;
    104  1.29      fvdl 	daddr_t lbn;
    105   1.1   mycroft 	int bsize, error;
    106  1.49     perry 
    107   1.1   mycroft 	ip = VTOI(ap->a_vp);
    108   1.1   mycroft 	fs = ip->i_lfs;
    109   1.1   mycroft 	lbn = lblkno(fs, ap->a_offset);
    110   1.6      fvdl 	bsize = blksize(fs, ip, lbn);
    111  1.49     perry 
    112   1.1   mycroft 	*ap->a_bpp = NULL;
    113   1.3  christos 	if ((error = bread(ap->a_vp, lbn, bsize, NOCRED, &bp)) != 0) {
    114   1.1   mycroft 		brelse(bp);
    115   1.1   mycroft 		return (error);
    116   1.1   mycroft 	}
    117   1.1   mycroft 	if (ap->a_res)
    118   1.1   mycroft 		*ap->a_res = (char *)bp->b_data + blkoff(fs, ap->a_offset);
    119   1.1   mycroft 	*ap->a_bpp = bp;
    120   1.1   mycroft 	return (0);
    121   1.1   mycroft }
    122   1.1   mycroft 
    123  1.50  perseant #ifdef DEBUG
    124  1.31  perseant char *lfs_res_names[LFS_NB_COUNT] = {
    125  1.31  perseant 	"summary",
    126  1.31  perseant 	"superblock",
    127  1.31  perseant 	"ifile block",
    128  1.31  perseant 	"cluster",
    129  1.31  perseant 	"clean",
    130  1.31  perseant };
    131  1.31  perseant #endif
    132  1.31  perseant 
    133  1.31  perseant int lfs_res_qty[LFS_NB_COUNT] = {
    134  1.31  perseant 	LFS_N_SUMMARIES,
    135  1.31  perseant 	LFS_N_SBLOCKS,
    136  1.31  perseant 	LFS_N_IBLOCKS,
    137  1.31  perseant 	LFS_N_CLUSTERS,
    138  1.31  perseant 	LFS_N_CLEAN,
    139  1.31  perseant };
    140  1.31  perseant 
    141  1.31  perseant void
    142  1.31  perseant lfs_setup_resblks(struct lfs *fs)
    143  1.31  perseant {
    144  1.31  perseant 	int i, j;
    145  1.31  perseant 	int maxbpp;
    146  1.31  perseant 
    147  1.51  perseant 	ASSERT_NO_SEGLOCK(fs);
    148  1.31  perseant 	fs->lfs_resblk = (res_t *)malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT,
    149  1.33  perseant 					  M_WAITOK);
    150  1.31  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    151  1.31  perseant 		fs->lfs_resblk[i].inuse = 0;
    152  1.31  perseant 		fs->lfs_resblk[i].p = NULL;
    153  1.31  perseant 	}
    154  1.31  perseant 	for (i = 0; i < LFS_RESHASH_WIDTH; i++)
    155  1.31  perseant 		LIST_INIT(fs->lfs_reshash + i);
    156  1.31  perseant 
    157  1.31  perseant 	/*
    158  1.31  perseant 	 * These types of allocations can be larger than a page,
    159  1.31  perseant 	 * so we can't use the pool subsystem for them.
    160  1.31  perseant 	 */
    161  1.31  perseant 	for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++)
    162  1.34  perseant 		fs->lfs_resblk[i].size = fs->lfs_sumsize;
    163  1.31  perseant 	for (j = 0; j < LFS_N_SBLOCKS; j++, i++)
    164  1.34  perseant 		fs->lfs_resblk[i].size = LFS_SBPAD;
    165  1.31  perseant 	for (j = 0; j < LFS_N_IBLOCKS; j++, i++)
    166  1.34  perseant 		fs->lfs_resblk[i].size = fs->lfs_bsize;
    167  1.31  perseant 	for (j = 0; j < LFS_N_CLUSTERS; j++, i++)
    168  1.34  perseant 		fs->lfs_resblk[i].size = MAXPHYS;
    169  1.31  perseant 	for (j = 0; j < LFS_N_CLEAN; j++, i++)
    170  1.34  perseant 		fs->lfs_resblk[i].size = MAXPHYS;
    171  1.34  perseant 
    172  1.34  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    173  1.34  perseant 		fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size,
    174  1.34  perseant 					     M_SEGMENT, M_WAITOK);
    175  1.34  perseant 	}
    176  1.31  perseant 
    177  1.31  perseant 	/*
    178  1.31  perseant 	 * Initialize pools for small types (XXX is BPP small?)
    179  1.31  perseant 	 */
    180  1.46    simonb 	pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0,
    181  1.46    simonb 		"lfsclpl", &pool_allocator_nointr);
    182  1.46    simonb 	pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0,
    183  1.46    simonb 		"lfssegpool", &pool_allocator_nointr);
    184  1.31  perseant 	maxbpp = ((fs->lfs_sumsize - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2);
    185  1.47      yamt 	maxbpp = MIN(maxbpp, segsize(fs) / fs->lfs_fsize + 2);
    186  1.46    simonb 	pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0,
    187  1.46    simonb 		"lfsbpppl", &pool_allocator_nointr);
    188  1.31  perseant }
    189  1.31  perseant 
    190  1.31  perseant void
    191  1.31  perseant lfs_free_resblks(struct lfs *fs)
    192  1.31  perseant {
    193  1.31  perseant 	int i;
    194  1.31  perseant 
    195  1.31  perseant 	pool_destroy(&fs->lfs_bpppool);
    196  1.31  perseant 	pool_destroy(&fs->lfs_segpool);
    197  1.31  perseant 	pool_destroy(&fs->lfs_clpool);
    198  1.31  perseant 
    199  1.51  perseant 	simple_lock(&fs->lfs_interlock);
    200  1.31  perseant 	for (i = 0; i < LFS_N_TOTAL; i++) {
    201  1.39  perseant 		while (fs->lfs_resblk[i].inuse)
    202  1.51  perseant 			ltsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0,
    203  1.51  perseant 				&fs->lfs_interlock);
    204  1.31  perseant 		if (fs->lfs_resblk[i].p != NULL)
    205  1.31  perseant 			free(fs->lfs_resblk[i].p, M_SEGMENT);
    206  1.31  perseant 	}
    207  1.31  perseant 	free(fs->lfs_resblk, M_SEGMENT);
    208  1.51  perseant 	simple_unlock(&fs->lfs_interlock);
    209  1.31  perseant }
    210  1.31  perseant 
    211  1.31  perseant static unsigned int
    212  1.31  perseant lfs_mhash(void *vp)
    213  1.31  perseant {
    214  1.31  perseant 	return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH;
    215  1.31  perseant }
    216  1.31  perseant 
    217  1.31  perseant /*
    218  1.31  perseant  * Return memory of the given size for the given purpose, or use one of a
    219  1.31  perseant  * number of spare last-resort buffers, if malloc returns NULL.
    220  1.49     perry  */
    221  1.31  perseant void *
    222  1.31  perseant lfs_malloc(struct lfs *fs, size_t size, int type)
    223  1.31  perseant {
    224  1.31  perseant 	struct lfs_res_blk *re;
    225  1.31  perseant 	void *r;
    226  1.31  perseant 	int i, s, start;
    227  1.31  perseant 	unsigned int h;
    228  1.31  perseant 
    229  1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    230  1.34  perseant 	r = NULL;
    231  1.34  perseant 
    232  1.31  perseant 	/* If no mem allocated for this type, it just waits */
    233  1.34  perseant 	if (lfs_res_qty[type] == 0) {
    234  1.34  perseant 		r = malloc(size, M_SEGMENT, M_WAITOK);
    235  1.34  perseant 		return r;
    236  1.34  perseant 	}
    237  1.31  perseant 
    238  1.31  perseant 	/* Otherwise try a quick malloc, and if it works, great */
    239  1.34  perseant 	if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) {
    240  1.31  perseant 		return r;
    241  1.34  perseant 	}
    242  1.31  perseant 
    243  1.31  perseant 	/*
    244  1.31  perseant 	 * If malloc returned NULL, we are forced to use one of our
    245  1.31  perseant 	 * reserve blocks.  We have on hand at least one summary block,
    246  1.31  perseant 	 * at least one cluster block, at least one superblock,
    247  1.31  perseant 	 * and several indirect blocks.
    248  1.31  perseant 	 */
    249  1.51  perseant 
    250  1.51  perseant 	simple_lock(&fs->lfs_interlock);
    251  1.31  perseant 	/* skip over blocks of other types */
    252  1.31  perseant 	for (i = 0, start = 0; i < type; i++)
    253  1.31  perseant 		start += lfs_res_qty[i];
    254  1.31  perseant 	while (r == NULL) {
    255  1.31  perseant 		for (i = 0; i < lfs_res_qty[type]; i++) {
    256  1.31  perseant 			if (fs->lfs_resblk[start + i].inuse == 0) {
    257  1.31  perseant 				re = fs->lfs_resblk + start + i;
    258  1.31  perseant 				re->inuse = 1;
    259  1.31  perseant 				r = re->p;
    260  1.34  perseant 				KASSERT(re->size >= size);
    261  1.31  perseant 				h = lfs_mhash(r);
    262  1.31  perseant 				s = splbio();
    263  1.31  perseant 				LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res);
    264  1.31  perseant 				splx(s);
    265  1.51  perseant 				simple_unlock(&fs->lfs_interlock);
    266  1.31  perseant 				return r;
    267  1.31  perseant 			}
    268  1.31  perseant 		}
    269  1.51  perseant 		DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n",
    270  1.51  perseant 		      lfs_res_names[type], lfs_res_qty[type]));
    271  1.51  perseant 		ltsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0,
    272  1.51  perseant 			&fs->lfs_interlock);
    273  1.51  perseant 		DLOG((DLOG_MALLOC, "done sleeping on %s\n",
    274  1.51  perseant 		      lfs_res_names[type]));
    275  1.31  perseant 	}
    276  1.31  perseant 	/* NOTREACHED */
    277  1.51  perseant 	simple_unlock(&fs->lfs_interlock);
    278  1.31  perseant 	return r;
    279  1.31  perseant }
    280  1.31  perseant 
    281  1.31  perseant void
    282  1.31  perseant lfs_free(struct lfs *fs, void *p, int type)
    283  1.31  perseant {
    284  1.31  perseant 	int s;
    285  1.31  perseant 	unsigned int h;
    286  1.31  perseant 	res_t *re;
    287  1.32      yamt #ifdef DEBUG
    288  1.32      yamt 	int i;
    289  1.32      yamt #endif
    290  1.31  perseant 
    291  1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    292  1.31  perseant 	h = lfs_mhash(p);
    293  1.51  perseant 	simple_lock(&fs->lfs_interlock);
    294  1.31  perseant 	s = splbio();
    295  1.31  perseant 	LIST_FOREACH(re, &fs->lfs_reshash[h], res) {
    296  1.31  perseant 		if (re->p == p) {
    297  1.32      yamt 			KASSERT(re->inuse == 1);
    298  1.31  perseant 			LIST_REMOVE(re, res);
    299  1.31  perseant 			re->inuse = 0;
    300  1.31  perseant 			wakeup(&fs->lfs_resblk);
    301  1.31  perseant 			splx(s);
    302  1.51  perseant 			simple_unlock(&fs->lfs_interlock);
    303  1.31  perseant 			return;
    304  1.31  perseant 		}
    305  1.31  perseant 	}
    306  1.32      yamt #ifdef DEBUG
    307  1.32      yamt 	for (i = 0; i < LFS_N_TOTAL; i++) {
    308  1.32      yamt 		if (fs->lfs_resblk[i].p == p)
    309  1.34  perseant 			panic("lfs_free: inconsistent reserved block");
    310  1.32      yamt 	}
    311  1.32      yamt #endif
    312  1.31  perseant 	splx(s);
    313  1.51  perseant 	simple_unlock(&fs->lfs_interlock);
    314  1.51  perseant 
    315  1.31  perseant 	/*
    316  1.31  perseant 	 * If we didn't find it, free it.
    317  1.31  perseant 	 */
    318  1.31  perseant 	free(p, M_SEGMENT);
    319  1.31  perseant }
    320   1.1   mycroft 
    321   1.1   mycroft /*
    322   1.1   mycroft  * lfs_seglock --
    323   1.1   mycroft  *	Single thread the segment writer.
    324   1.1   mycroft  */
    325  1.31  perseant int
    326  1.18  perseant lfs_seglock(struct lfs *fs, unsigned long flags)
    327   1.1   mycroft {
    328   1.1   mycroft 	struct segment *sp;
    329  1.49     perry 
    330  1.38  perseant 	simple_lock(&fs->lfs_interlock);
    331   1.7   thorpej 	if (fs->lfs_seglock) {
    332   1.1   mycroft 		if (fs->lfs_lockpid == curproc->p_pid) {
    333  1.38  perseant 			simple_unlock(&fs->lfs_interlock);
    334   1.1   mycroft 			++fs->lfs_seglock;
    335   1.1   mycroft 			fs->lfs_sp->seg_flags |= flags;
    336  1.31  perseant 			return 0;
    337  1.38  perseant 		} else if (flags & SEGM_PAGEDAEMON) {
    338  1.38  perseant 			simple_unlock(&fs->lfs_interlock);
    339  1.31  perseant 			return EWOULDBLOCK;
    340  1.51  perseant 		} else {
    341  1.51  perseant 			while (fs->lfs_seglock) {
    342  1.51  perseant 				(void)ltsleep(&fs->lfs_seglock, PRIBIO + 1,
    343  1.51  perseant 					"lfs seglock", 0, &fs->lfs_interlock);
    344  1.51  perseant 			}
    345  1.51  perseant 		}
    346   1.7   thorpej 	}
    347  1.49     perry 
    348   1.1   mycroft 	fs->lfs_seglock = 1;
    349   1.1   mycroft 	fs->lfs_lockpid = curproc->p_pid;
    350  1.38  perseant 	simple_unlock(&fs->lfs_interlock);
    351  1.36  perseant 	fs->lfs_cleanind = 0;
    352  1.36  perseant 
    353  1.51  perseant #ifdef DEBUG
    354  1.51  perseant 	LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid);
    355  1.51  perseant #endif
    356  1.27  perseant 	/* Drain fragment size changes out */
    357  1.27  perseant 	lockmgr(&fs->lfs_fraglock, LK_EXCLUSIVE, 0);
    358  1.27  perseant 
    359  1.31  perseant 	sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK);
    360  1.31  perseant 	sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK);
    361   1.1   mycroft 	sp->seg_flags = flags;
    362   1.1   mycroft 	sp->vp = NULL;
    363  1.26  perseant 	sp->seg_iocount = 0;
    364   1.1   mycroft 	(void) lfs_initseg(fs);
    365  1.49     perry 
    366   1.1   mycroft 	/*
    367   1.1   mycroft 	 * Keep a cumulative count of the outstanding I/O operations.  If the
    368   1.1   mycroft 	 * disk drive catches up with us it could go to zero before we finish,
    369   1.1   mycroft 	 * so we artificially increment it by one until we've scheduled all of
    370   1.1   mycroft 	 * the writes we intend to do.
    371   1.1   mycroft 	 */
    372  1.51  perseant 	simple_lock(&fs->lfs_interlock);
    373   1.1   mycroft 	++fs->lfs_iocount;
    374  1.51  perseant 	simple_unlock(&fs->lfs_interlock);
    375  1.31  perseant 	return 0;
    376   1.1   mycroft }
    377   1.8  perseant 
    378  1.31  perseant static void lfs_unmark_dirop(struct lfs *);
    379  1.31  perseant 
    380  1.31  perseant static void
    381  1.31  perseant lfs_unmark_dirop(struct lfs *fs)
    382  1.31  perseant {
    383  1.31  perseant 	struct inode *ip, *nip;
    384  1.31  perseant 	struct vnode *vp;
    385  1.40  perseant 	int doit;
    386  1.31  perseant 
    387  1.51  perseant 	ASSERT_NO_SEGLOCK(fs);
    388  1.40  perseant 	simple_lock(&fs->lfs_interlock);
    389  1.40  perseant 	doit = !(fs->lfs_flags & LFS_UNDIROP);
    390  1.40  perseant 	if (doit)
    391  1.40  perseant 		fs->lfs_flags |= LFS_UNDIROP;
    392  1.51  perseant 	if (!doit) {
    393  1.51  perseant 		simple_unlock(&fs->lfs_interlock);
    394  1.40  perseant 		return;
    395  1.51  perseant 	}
    396  1.40  perseant 
    397  1.31  perseant 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
    398  1.31  perseant 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
    399  1.51  perseant 		simple_unlock(&fs->lfs_interlock);
    400  1.31  perseant 		vp = ITOV(ip);
    401  1.31  perseant 
    402  1.51  perseant 		simple_lock(&vp->v_interlock);
    403  1.31  perseant 		if (VOP_ISLOCKED(vp) &&
    404  1.33  perseant 			   vp->v_lock.lk_lockholder != curproc->p_pid) {
    405  1.51  perseant 			simple_lock(&fs->lfs_interlock);
    406  1.51  perseant 			simple_unlock(&vp->v_interlock);
    407  1.31  perseant 			continue;
    408  1.31  perseant 		}
    409  1.31  perseant 		if ((VTOI(vp)->i_flag & IN_ADIROP) == 0) {
    410  1.51  perseant 			simple_lock(&fs->lfs_interlock);
    411  1.51  perseant 			simple_lock(&lfs_subsys_lock);
    412  1.31  perseant 			--lfs_dirvcount;
    413  1.51  perseant 			simple_unlock(&lfs_subsys_lock);
    414  1.31  perseant 			vp->v_flag &= ~VDIROP;
    415  1.31  perseant 			TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
    416  1.51  perseant 			simple_unlock(&fs->lfs_interlock);
    417  1.31  perseant 			wakeup(&lfs_dirvcount);
    418  1.51  perseant 			simple_unlock(&vp->v_interlock);
    419  1.51  perseant 			simple_lock(&fs->lfs_interlock);
    420  1.31  perseant 			fs->lfs_unlockvp = vp;
    421  1.51  perseant 			simple_unlock(&fs->lfs_interlock);
    422  1.31  perseant 			vrele(vp);
    423  1.51  perseant 			simple_lock(&fs->lfs_interlock);
    424  1.31  perseant 			fs->lfs_unlockvp = NULL;
    425  1.51  perseant 			simple_unlock(&fs->lfs_interlock);
    426  1.51  perseant 		} else
    427  1.51  perseant 			simple_unlock(&vp->v_interlock);
    428  1.51  perseant 		simple_lock(&fs->lfs_interlock);
    429  1.31  perseant 	}
    430  1.40  perseant 
    431  1.40  perseant 	fs->lfs_flags &= ~LFS_UNDIROP;
    432  1.40  perseant 	simple_unlock(&fs->lfs_interlock);
    433  1.51  perseant 	wakeup(&fs->lfs_flags);
    434  1.31  perseant }
    435  1.31  perseant 
    436  1.31  perseant static void
    437  1.31  perseant lfs_auto_segclean(struct lfs *fs)
    438  1.31  perseant {
    439  1.48  perseant 	int i, error, s, waited;
    440  1.31  perseant 
    441  1.51  perseant 	ASSERT_SEGLOCK(fs);
    442  1.31  perseant 	/*
    443  1.31  perseant 	 * Now that we've swapped lfs_activesb, but while we still
    444  1.31  perseant 	 * hold the segment lock, run through the segment list marking
    445  1.31  perseant 	 * the empty ones clean.
    446  1.31  perseant 	 * XXX - do we really need to do them all at once?
    447  1.31  perseant 	 */
    448  1.48  perseant 	waited = 0;
    449  1.31  perseant 	for (i = 0; i < fs->lfs_nseg; i++) {
    450  1.31  perseant 		if ((fs->lfs_suflags[0][i] &
    451  1.31  perseant 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    452  1.31  perseant 		    (SEGUSE_DIRTY | SEGUSE_EMPTY) &&
    453  1.31  perseant 		    (fs->lfs_suflags[1][i] &
    454  1.31  perseant 		     (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) ==
    455  1.31  perseant 		    (SEGUSE_DIRTY | SEGUSE_EMPTY)) {
    456  1.31  perseant 
    457  1.48  perseant 			/* Make sure the sb is written before we clean */
    458  1.51  perseant 			simple_lock(&fs->lfs_interlock);
    459  1.48  perseant 			s = splbio();
    460  1.48  perseant 			while (waited == 0 && fs->lfs_sbactive)
    461  1.51  perseant 				ltsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb",
    462  1.51  perseant 					0, &fs->lfs_interlock);
    463  1.48  perseant 			splx(s);
    464  1.51  perseant 			simple_unlock(&fs->lfs_interlock);
    465  1.48  perseant 			waited = 1;
    466  1.48  perseant 
    467  1.31  perseant 			if ((error = lfs_do_segclean(fs, i)) != 0) {
    468  1.50  perseant 				DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i));
    469  1.31  perseant 			}
    470  1.31  perseant 		}
    471  1.31  perseant 		fs->lfs_suflags[1 - fs->lfs_activesb][i] =
    472  1.31  perseant 			fs->lfs_suflags[fs->lfs_activesb][i];
    473  1.31  perseant 	}
    474  1.31  perseant }
    475  1.31  perseant 
    476   1.1   mycroft /*
    477   1.1   mycroft  * lfs_segunlock --
    478   1.1   mycroft  *	Single thread the segment writer.
    479   1.1   mycroft  */
    480   1.1   mycroft void
    481  1.18  perseant lfs_segunlock(struct lfs *fs)
    482   1.1   mycroft {
    483   1.1   mycroft 	struct segment *sp;
    484   1.1   mycroft 	unsigned long sync, ckp;
    485  1.22  perseant 	struct buf *bp;
    486  1.40  perseant 	int do_unmark_dirop = 0;
    487  1.49     perry 
    488  1.15  perseant 	sp = fs->lfs_sp;
    489  1.15  perseant 
    490  1.38  perseant 	simple_lock(&fs->lfs_interlock);
    491  1.51  perseant 	LOCK_ASSERT(LFS_SEGLOCK_HELD(fs));
    492  1.15  perseant 	if (fs->lfs_seglock == 1) {
    493  1.40  perseant 		if ((sp->seg_flags & SEGM_PROT) == 0)
    494  1.40  perseant 			do_unmark_dirop = 1;
    495  1.38  perseant 		simple_unlock(&fs->lfs_interlock);
    496   1.1   mycroft 		sync = sp->seg_flags & SEGM_SYNC;
    497   1.1   mycroft 		ckp = sp->seg_flags & SEGM_CKP;
    498   1.1   mycroft 		if (sp->bpp != sp->cbpp) {
    499   1.1   mycroft 			/* Free allocated segment summary */
    500  1.18  perseant 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
    501  1.22  perseant 			bp = *sp->bpp;
    502  1.31  perseant 			lfs_freebuf(fs, bp);
    503   1.1   mycroft 		} else
    504  1.50  perseant 			DLOG((DLOG_SEG, "lfs_segunlock: unlock to 0 with no summary"));
    505   1.8  perseant 
    506  1.31  perseant 		pool_put(&fs->lfs_bpppool, sp->bpp);
    507  1.18  perseant 		sp->bpp = NULL;
    508  1.36  perseant 
    509  1.36  perseant 		/*
    510  1.36  perseant 		 * If we're not sync, we're done with sp, get rid of it.
    511  1.36  perseant 		 * Otherwise, we keep a local copy around but free
    512  1.36  perseant 		 * fs->lfs_sp so another process can use it (we have to
    513  1.36  perseant 		 * wait but they don't have to wait for us).
    514  1.36  perseant 		 */
    515  1.26  perseant 		if (!sync)
    516  1.31  perseant 			pool_put(&fs->lfs_segpool, sp);
    517  1.18  perseant 		fs->lfs_sp = NULL;
    518   1.1   mycroft 
    519   1.1   mycroft 		/*
    520   1.1   mycroft 		 * If the I/O count is non-zero, sleep until it reaches zero.
    521   1.1   mycroft 		 * At the moment, the user's process hangs around so we can
    522   1.1   mycroft 		 * sleep.
    523   1.1   mycroft 		 */
    524  1.51  perseant 		simple_lock(&fs->lfs_interlock);
    525  1.44      yamt 		if (--fs->lfs_iocount == 0)
    526  1.44      yamt 			LFS_DEBUG_COUNTLOCKED("lfs_segunlock");
    527  1.36  perseant 		if (fs->lfs_iocount <= 1)
    528  1.22  perseant 			wakeup(&fs->lfs_iocount);
    529  1.51  perseant 		simple_unlock(&fs->lfs_interlock);
    530   1.1   mycroft 		/*
    531  1.26  perseant 		 * If we're not checkpointing, we don't have to block
    532  1.26  perseant 		 * other processes to wait for a synchronous write
    533  1.26  perseant 		 * to complete.
    534  1.26  perseant 		 */
    535  1.26  perseant 		if (!ckp) {
    536  1.51  perseant #ifdef DEBUG
    537  1.51  perseant 			LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    538  1.51  perseant #endif
    539  1.38  perseant 			simple_lock(&fs->lfs_interlock);
    540  1.26  perseant 			--fs->lfs_seglock;
    541  1.26  perseant 			fs->lfs_lockpid = 0;
    542  1.38  perseant 			simple_unlock(&fs->lfs_interlock);
    543  1.26  perseant 			wakeup(&fs->lfs_seglock);
    544  1.26  perseant 		}
    545  1.26  perseant 		/*
    546   1.1   mycroft 		 * We let checkpoints happen asynchronously.  That means
    547   1.1   mycroft 		 * that during recovery, we have to roll forward between
    548   1.1   mycroft 		 * the two segments described by the first and second
    549   1.1   mycroft 		 * superblocks to make sure that the checkpoint described
    550   1.1   mycroft 		 * by a superblock completed.
    551   1.1   mycroft 		 */
    552  1.51  perseant 		simple_lock(&fs->lfs_interlock);
    553  1.26  perseant 		while (ckp && sync && fs->lfs_iocount)
    554  1.51  perseant 			(void)ltsleep(&fs->lfs_iocount, PRIBIO + 1,
    555  1.51  perseant 				      "lfs_iocount", 0, &fs->lfs_interlock);
    556  1.26  perseant 		while (sync && sp->seg_iocount) {
    557  1.51  perseant 			(void)ltsleep(&sp->seg_iocount, PRIBIO + 1,
    558  1.51  perseant 				     "seg_iocount", 0, &fs->lfs_interlock);
    559  1.50  perseant 			DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount));
    560  1.26  perseant 		}
    561  1.51  perseant 		simple_unlock(&fs->lfs_interlock);
    562  1.26  perseant 		if (sync)
    563  1.31  perseant 			pool_put(&fs->lfs_segpool, sp);
    564  1.36  perseant 
    565   1.1   mycroft 		if (ckp) {
    566   1.1   mycroft 			fs->lfs_nactive = 0;
    567   1.8  perseant 			/* If we *know* everything's on disk, write both sbs */
    568  1.33  perseant 			/* XXX should wait for this one	 */
    569  1.21       chs 			if (sync)
    570  1.31  perseant 				lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
    571  1.31  perseant 			lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
    572  1.48  perseant 			if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) {
    573  1.35  perseant 				lfs_auto_segclean(fs);
    574  1.48  perseant 				/* If sync, we can clean the remainder too */
    575  1.48  perseant 				if (sync)
    576  1.48  perseant 					lfs_auto_segclean(fs);
    577  1.48  perseant 			}
    578   1.8  perseant 			fs->lfs_activesb = 1 - fs->lfs_activesb;
    579  1.51  perseant #ifdef DEBUG
    580  1.51  perseant 			LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    581  1.51  perseant #endif
    582  1.38  perseant 			simple_lock(&fs->lfs_interlock);
    583  1.26  perseant 			--fs->lfs_seglock;
    584  1.26  perseant 			fs->lfs_lockpid = 0;
    585  1.38  perseant 			simple_unlock(&fs->lfs_interlock);
    586  1.26  perseant 			wakeup(&fs->lfs_seglock);
    587   1.1   mycroft 		}
    588  1.27  perseant 		/* Reenable fragment size changes */
    589  1.27  perseant 		lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0);
    590  1.40  perseant 		if (do_unmark_dirop)
    591  1.40  perseant 			lfs_unmark_dirop(fs);
    592   1.1   mycroft 	} else if (fs->lfs_seglock == 0) {
    593  1.38  perseant 		simple_unlock(&fs->lfs_interlock);
    594   1.1   mycroft 		panic ("Seglock not held");
    595   1.1   mycroft 	} else {
    596   1.1   mycroft 		--fs->lfs_seglock;
    597  1.38  perseant 		simple_unlock(&fs->lfs_interlock);
    598   1.1   mycroft 	}
    599  1.41      yamt }
    600  1.41      yamt 
    601  1.41      yamt /*
    602  1.41      yamt  * drain dirops and start writer.
    603  1.41      yamt  */
    604  1.41      yamt int
    605  1.41      yamt lfs_writer_enter(struct lfs *fs, const char *wmesg)
    606  1.41      yamt {
    607  1.41      yamt 	int error = 0;
    608  1.41      yamt 
    609  1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    610  1.41      yamt 	simple_lock(&fs->lfs_interlock);
    611  1.41      yamt 
    612  1.41      yamt 	/* disallow dirops during flush */
    613  1.41      yamt 	fs->lfs_writer++;
    614  1.41      yamt 
    615  1.41      yamt 	while (fs->lfs_dirops > 0) {
    616  1.49     perry 		++fs->lfs_diropwait;
    617  1.41      yamt 		error = ltsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0,
    618  1.51  perseant 				&fs->lfs_interlock);
    619  1.49     perry 		--fs->lfs_diropwait;
    620  1.41      yamt 	}
    621  1.41      yamt 
    622  1.41      yamt 	if (error)
    623  1.41      yamt 		fs->lfs_writer--;
    624  1.41      yamt 
    625  1.41      yamt 	simple_unlock(&fs->lfs_interlock);
    626  1.41      yamt 
    627  1.41      yamt 	return error;
    628  1.41      yamt }
    629  1.41      yamt 
    630  1.41      yamt void
    631  1.41      yamt lfs_writer_leave(struct lfs *fs)
    632  1.41      yamt {
    633  1.41      yamt 	boolean_t dowakeup;
    634  1.41      yamt 
    635  1.51  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    636  1.41      yamt 	simple_lock(&fs->lfs_interlock);
    637  1.41      yamt 	dowakeup = !(--fs->lfs_writer);
    638  1.41      yamt 	simple_unlock(&fs->lfs_interlock);
    639  1.41      yamt 	if (dowakeup)
    640  1.41      yamt 		wakeup(&fs->lfs_dirops);
    641   1.1   mycroft }
    642