Home | History | Annotate | Line # | Download | only in lfs
lfs_bio.c revision 1.23.2.1
      1  1.23.2.1  perseant /*	$NetBSD: lfs_bio.c,v 1.23.2.1 2000/09/14 18:50:17 perseant Exp $	*/
      2       1.2       cgd 
      3       1.7  perseant /*-
      4  1.23.2.1  perseant  * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
      5       1.7  perseant  * All rights reserved.
      6       1.7  perseant  *
      7       1.7  perseant  * This code is derived from software contributed to The NetBSD Foundation
      8       1.7  perseant  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9       1.7  perseant  *
     10       1.7  perseant  * Redistribution and use in source and binary forms, with or without
     11       1.7  perseant  * modification, are permitted provided that the following conditions
     12       1.7  perseant  * are met:
     13       1.7  perseant  * 1. Redistributions of source code must retain the above copyright
     14       1.7  perseant  *    notice, this list of conditions and the following disclaimer.
     15       1.7  perseant  * 2. Redistributions in binary form must reproduce the above copyright
     16       1.7  perseant  *    notice, this list of conditions and the following disclaimer in the
     17       1.7  perseant  *    documentation and/or other materials provided with the distribution.
     18       1.7  perseant  * 3. All advertising materials mentioning features or use of this software
     19       1.7  perseant  *    must display the following acknowledgement:
     20       1.7  perseant  *      This product includes software developed by the NetBSD
     21       1.7  perseant  *      Foundation, Inc. and its contributors.
     22       1.7  perseant  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23       1.7  perseant  *    contributors may be used to endorse or promote products derived
     24       1.7  perseant  *    from this software without specific prior written permission.
     25       1.7  perseant  *
     26       1.7  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27       1.7  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28       1.7  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29       1.7  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30       1.7  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31       1.7  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32       1.7  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33       1.7  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34       1.7  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35       1.7  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36       1.7  perseant  * POSSIBILITY OF SUCH DAMAGE.
     37       1.7  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       1.1   mycroft  * 3. All advertising materials mentioning features or use of this software
     51       1.1   mycroft  *    must display the following acknowledgement:
     52       1.1   mycroft  *	This product includes software developed by the University of
     53       1.1   mycroft  *	California, Berkeley and its contributors.
     54       1.1   mycroft  * 4. Neither the name of the University nor the names of its contributors
     55       1.1   mycroft  *    may be used to endorse or promote products derived from this software
     56       1.1   mycroft  *    without specific prior written permission.
     57       1.1   mycroft  *
     58       1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     59       1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     60       1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     61       1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     62       1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     63       1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     64       1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     65       1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     66       1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     67       1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     68       1.1   mycroft  * SUCH DAMAGE.
     69       1.1   mycroft  *
     70       1.6      fvdl  *	@(#)lfs_bio.c	8.10 (Berkeley) 6/10/95
     71       1.1   mycroft  */
     72       1.1   mycroft 
     73       1.1   mycroft #include <sys/param.h>
     74       1.5  christos #include <sys/systm.h>
     75       1.1   mycroft #include <sys/proc.h>
     76       1.1   mycroft #include <sys/buf.h>
     77       1.1   mycroft #include <sys/vnode.h>
     78       1.1   mycroft #include <sys/resourcevar.h>
     79       1.1   mycroft #include <sys/mount.h>
     80       1.1   mycroft #include <sys/kernel.h>
     81       1.1   mycroft 
     82       1.1   mycroft #include <ufs/ufs/quota.h>
     83       1.1   mycroft #include <ufs/ufs/inode.h>
     84       1.1   mycroft #include <ufs/ufs/ufsmount.h>
     85       1.5  christos #include <ufs/ufs/ufs_extern.h>
     86       1.1   mycroft 
     87       1.7  perseant #include <sys/malloc.h>
     88       1.1   mycroft #include <ufs/lfs/lfs.h>
     89       1.1   mycroft #include <ufs/lfs/lfs_extern.h>
     90       1.1   mycroft 
     91       1.7  perseant /* Macros to clear/set/test flags. */
     92       1.7  perseant # define	SET(t, f)	(t) |= (f)
     93       1.7  perseant # define	CLR(t, f)	(t) &= ~(f)
     94       1.7  perseant # define	ISSET(t, f)	((t) & (f))
     95       1.7  perseant 
     96       1.1   mycroft /*
     97       1.1   mycroft  * LFS block write function.
     98       1.1   mycroft  *
     99       1.1   mycroft  * XXX
    100       1.1   mycroft  * No write cost accounting is done.
    101       1.1   mycroft  * This is almost certainly wrong for synchronous operations and NFS.
    102       1.1   mycroft  */
    103       1.1   mycroft int	lfs_allclean_wakeup;		/* Cleaner wakeup address. */
    104       1.7  perseant int	locked_queue_count   = 0;	/* XXX Count of locked-down buffers. */
    105       1.7  perseant long	locked_queue_bytes   = 0L;	/* XXX Total size of locked buffers. */
    106       1.7  perseant int	lfs_writing          = 0;	/* Set if already kicked off a writer
    107       1.1   mycroft 					   because of buffer space */
    108       1.7  perseant extern int lfs_dostats;
    109       1.7  perseant 
    110       1.7  perseant 
    111       1.1   mycroft /*
    112       1.7  perseant  * #define WRITE_THRESHHOLD    ((nbuf >> 1) - 10)
    113       1.7  perseant  * #define WAIT_THRESHHOLD     (nbuf - (nbuf >> 2) - 10)
    114       1.7  perseant  */
    115       1.7  perseant #define LFS_MAX_BUFS        ((nbuf >> 2) - 10)
    116       1.7  perseant #define LFS_WAIT_BUFS       ((nbuf >> 1) - (nbuf >> 3) - 10)
    117       1.7  perseant /* These are new ... is LFS taking up too much memory in its buffers? */
    118      1.15     ragge #define LFS_MAX_BYTES       (((bufpages >> 2) - 10) * NBPG)
    119      1.15     ragge #define LFS_WAIT_BYTES      (((bufpages >> 1) - (bufpages >> 3) - 10) * NBPG)
    120       1.7  perseant #define LFS_BUFWAIT         2
    121  1.23.2.1  perseant 
    122  1.23.2.1  perseant inline static int lfs_fits(struct lfs *, int);
    123  1.23.2.1  perseant 
    124  1.23.2.1  perseant /*
    125  1.23.2.1  perseant  * Try to reserve some blocks, prior to performing a sensitive operation that
    126  1.23.2.1  perseant  * requires the vnode lock to be honored.  If there is not enough space, give
    127  1.23.2.1  perseant  * up the vnode lock temporarily and wait for the space to become available.
    128  1.23.2.1  perseant  *
    129  1.23.2.1  perseant  * Called with vp locked.  (Note nowever that if nb < 0, vp is ignored.)
    130  1.23.2.1  perseant  */
    131  1.23.2.1  perseant int
    132  1.23.2.1  perseant lfs_reserve(fs, vp, nb)
    133  1.23.2.1  perseant 	struct lfs *fs;
    134  1.23.2.1  perseant 	struct vnode *vp;
    135  1.23.2.1  perseant 	int nb;
    136  1.23.2.1  perseant {
    137  1.23.2.1  perseant 	CLEANERINFO *cip;
    138  1.23.2.1  perseant 	struct buf *bp;
    139  1.23.2.1  perseant 	int error, slept;
    140  1.23.2.1  perseant 
    141  1.23.2.1  perseant 	slept = 0;
    142  1.23.2.1  perseant 	while (nb > 0 && !lfs_fits(fs, nb + fs->lfs_ravail)) {
    143  1.23.2.1  perseant 		VOP_UNLOCK(vp, 0);
    144  1.23.2.1  perseant 
    145  1.23.2.1  perseant 		if (!slept) {
    146  1.23.2.1  perseant #ifdef DEBUG
    147  1.23.2.1  perseant 			printf("lfs_reserve: waiting for %ld (bfree = %d,"
    148  1.23.2.1  perseant 			       " est_bfree = %d)\n",
    149  1.23.2.1  perseant 			       nb + fs->lfs_ravail, fs->lfs_bfree,
    150  1.23.2.1  perseant 			       LFS_EST_BFREE(fs));
    151  1.23.2.1  perseant #endif
    152  1.23.2.1  perseant 		}
    153  1.23.2.1  perseant 		++slept;
    154  1.23.2.1  perseant 
    155  1.23.2.1  perseant 		/* Wake up the cleaner */
    156  1.23.2.1  perseant 		LFS_CLEANERINFO(cip, fs, bp);
    157  1.23.2.1  perseant 		cip->bfree = fs->lfs_bfree;
    158  1.23.2.1  perseant 		cip->avail = fs->lfs_avail - fs->lfs_ravail;
    159  1.23.2.1  perseant 		(void) VOP_BWRITE(bp); /* Ifile */
    160  1.23.2.1  perseant 		wakeup(&lfs_allclean_wakeup);
    161  1.23.2.1  perseant 		wakeup(&fs->lfs_nextseg);
    162  1.23.2.1  perseant 
    163  1.23.2.1  perseant 		error = tsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_reserve",
    164  1.23.2.1  perseant 			       0);
    165  1.23.2.1  perseant 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX use lockstatus */
    166  1.23.2.1  perseant 		if (error)
    167  1.23.2.1  perseant 			return error;
    168  1.23.2.1  perseant 	}
    169  1.23.2.1  perseant 	if (slept)
    170  1.23.2.1  perseant 		printf("lfs_reserve: woke up\n");
    171  1.23.2.1  perseant 	fs->lfs_ravail += nb;
    172  1.23.2.1  perseant 	return 0;
    173  1.23.2.1  perseant }
    174  1.23.2.1  perseant 
    175       1.7  perseant /*
    176       1.7  perseant  *
    177       1.7  perseant  * XXX we don't let meta-data writes run out of space because they can
    178       1.7  perseant  * come from the segment writer.  We need to make sure that there is
    179       1.7  perseant  * enough space reserved so that there's room to write meta-data
    180       1.7  perseant  * blocks.
    181       1.7  perseant  *
    182       1.7  perseant  * Also, we don't let blocks that have come to us from the cleaner
    183       1.7  perseant  * run out of space.
    184       1.7  perseant  */
    185       1.7  perseant #define CANT_WAIT(BP,F) (IS_IFILE((BP)) || (BP)->b_lblkno<0 || ((F) & BW_CLEAN))
    186       1.1   mycroft 
    187       1.1   mycroft int
    188       1.5  christos lfs_bwrite(v)
    189       1.5  christos 	void *v;
    190       1.5  christos {
    191       1.1   mycroft 	struct vop_bwrite_args /* {
    192       1.1   mycroft 		struct buf *a_bp;
    193       1.5  christos 	} */ *ap = v;
    194      1.17  augustss 	struct buf *bp = ap->a_bp;
    195      1.23  perseant 	struct inode *ip;
    196      1.23  perseant 
    197      1.23  perseant 	ip = VTOI(bp->b_vp);
    198       1.7  perseant 
    199       1.7  perseant #ifdef DIAGNOSTIC
    200      1.23  perseant         if (VTOI(bp->b_vp)->i_lfs->lfs_ronly == 0 && (bp->b_flags & B_ASYNC)) {
    201       1.7  perseant 		panic("bawrite LFS buffer");
    202      1.16  perseant 	}
    203       1.7  perseant #endif /* DIAGNOSTIC */
    204       1.7  perseant 	return lfs_bwrite_ext(bp,0);
    205       1.7  perseant }
    206       1.7  perseant 
    207       1.7  perseant /*
    208       1.7  perseant  * Determine if there is enough room currently available to write db
    209       1.7  perseant  * disk blocks.  We need enough blocks for the new blocks, the current
    210       1.7  perseant  * inode blocks, a summary block, plus potentially the ifile inode and
    211       1.7  perseant  * the segment usage table, plus an ifile page.
    212       1.7  perseant  */
    213      1.23  perseant inline static int
    214      1.23  perseant lfs_fits(struct lfs *fs, int db)
    215       1.7  perseant {
    216  1.23.2.1  perseant 	int needed;
    217       1.7  perseant 
    218  1.23.2.1  perseant 	needed = db + btodb(LFS_SUMMARY_SIZE) +
    219  1.23.2.1  perseant 		fsbtodb(fs, howmany(fs->lfs_uinodes + 1, INOPB(fs)) +
    220  1.23.2.1  perseant 			    fs->lfs_segtabsz + 1);
    221  1.23.2.1  perseant 
    222  1.23.2.1  perseant 	if (needed >= fs->lfs_avail) {
    223  1.23.2.1  perseant #ifdef DEBUG
    224  1.23.2.1  perseant 		printf("lfs_fits: no fit: db = %d, uinodes = %d, "
    225  1.23.2.1  perseant 		       "needed = %d, avail = %d\n",
    226  1.23.2.1  perseant 		       db, fs->lfs_uinodes, needed, fs->lfs_avail);
    227  1.23.2.1  perseant #endif
    228       1.7  perseant 		return 0;
    229  1.23.2.1  perseant 	}
    230       1.7  perseant 	return 1;
    231       1.7  perseant }
    232       1.7  perseant 
    233       1.7  perseant int
    234       1.7  perseant lfs_bwrite_ext(bp, flags)
    235       1.7  perseant 	struct buf *bp;
    236       1.7  perseant 	int flags;
    237       1.7  perseant {
    238       1.1   mycroft 	struct lfs *fs;
    239       1.1   mycroft 	struct inode *ip;
    240       1.6      fvdl 	int db, error, s;
    241  1.23.2.1  perseant 	struct buf *cbp;
    242  1.23.2.1  perseant 	CLEANERINFO *cip;
    243       1.7  perseant 
    244       1.1   mycroft 	/*
    245      1.16  perseant 	 * Don't write *any* blocks if we're mounted read-only.
    246      1.16  perseant 	 * In particular the cleaner can't write blocks either.
    247      1.16  perseant 	 */
    248      1.16  perseant         if(VTOI(bp->b_vp)->i_lfs->lfs_ronly) {
    249      1.16  perseant 		bp->b_flags &= ~(B_DELWRI|B_LOCKED|B_READ|B_ERROR);
    250      1.16  perseant 		if(bp->b_flags & B_CALL)
    251      1.16  perseant 			bp->b_flags &= ~B_BUSY;
    252      1.16  perseant 		else
    253      1.16  perseant 			brelse(bp);
    254      1.16  perseant 		return EROFS;
    255      1.16  perseant 	}
    256      1.16  perseant 
    257      1.16  perseant 	/*
    258       1.1   mycroft 	 * Set the delayed write flag and use reassignbuf to move the buffer
    259       1.1   mycroft 	 * from the clean list to the dirty one.
    260       1.1   mycroft 	 *
    261       1.1   mycroft 	 * Set the B_LOCKED flag and unlock the buffer, causing brelse to move
    262       1.1   mycroft 	 * the buffer onto the LOCKED free list.  This is necessary, otherwise
    263       1.1   mycroft 	 * getnewbuf() would try to reclaim the buffers using bawrite, which
    264       1.1   mycroft 	 * isn't going to work.
    265       1.1   mycroft 	 *
    266       1.1   mycroft 	 * XXX we don't let meta-data writes run out of space because they can
    267       1.1   mycroft 	 * come from the segment writer.  We need to make sure that there is
    268       1.1   mycroft 	 * enough space reserved so that there's room to write meta-data
    269       1.1   mycroft 	 * blocks.
    270       1.1   mycroft 	 */
    271       1.1   mycroft 	if (!(bp->b_flags & B_LOCKED)) {
    272       1.1   mycroft 		fs = VFSTOUFS(bp->b_vp->v_mount)->um_lfs;
    273       1.6      fvdl 		db = fragstodb(fs, numfrags(fs, bp->b_bcount));
    274       1.7  perseant 		while (!lfs_fits(fs, db) && !CANT_WAIT(bp,flags)) {
    275  1.23.2.1  perseant 			/*
    276  1.23.2.1  perseant 			 * Out of space, need cleaner to run.
    277  1.23.2.1  perseant 			 * Update the cleaner info, then wake it up.
    278  1.23.2.1  perseant 			 * Note the cleanerinfo block is on the ifile
    279  1.23.2.1  perseant 			 * so it CANT_WAIT.
    280  1.23.2.1  perseant 			 */
    281  1.23.2.1  perseant 			LFS_CLEANERINFO(cip, fs, cbp);
    282  1.23.2.1  perseant 			cip->bfree = fs->lfs_bfree;
    283  1.23.2.1  perseant 			cip->avail = fs->lfs_avail;
    284  1.23.2.1  perseant 			(void) VOP_BWRITE(cbp);
    285  1.23.2.1  perseant 
    286  1.23.2.1  perseant 			printf("lfs_bwrite: out of available space, "
    287  1.23.2.1  perseant 			       "waiting on cleaner\n");
    288  1.23.2.1  perseant 
    289       1.1   mycroft 			wakeup(&lfs_allclean_wakeup);
    290       1.6      fvdl 			wakeup(&fs->lfs_nextseg);
    291      1.19   thorpej 			error = tsleep(&fs->lfs_avail, PCATCH | PUSER,
    292      1.19   thorpej 			    "cleaner", 0);
    293       1.5  christos 			if (error) {
    294       1.7  perseant 				/* printf("lfs_bwrite: error in tsleep"); */
    295       1.1   mycroft 				brelse(bp);
    296       1.1   mycroft 				return (error);
    297       1.1   mycroft 			}
    298       1.1   mycroft 		}
    299       1.7  perseant 
    300       1.7  perseant 		ip = VTOI(bp->b_vp);
    301  1.23.2.1  perseant 		if (bp->b_flags & B_CALL) {
    302  1.23.2.1  perseant 			LFS_SET_UINO(ip, IN_CLEANING);
    303       1.7  perseant 		} else {
    304  1.23.2.1  perseant 			LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE);
    305       1.7  perseant 		}
    306       1.6      fvdl 		fs->lfs_avail -= db;
    307       1.1   mycroft 		++locked_queue_count;
    308       1.7  perseant 		locked_queue_bytes += bp->b_bufsize;
    309      1.14      fvdl 		s = splbio();
    310      1.16  perseant 		bp->b_flags |= B_DELWRI | B_LOCKED;
    311       1.1   mycroft 		bp->b_flags &= ~(B_READ | B_ERROR);
    312       1.1   mycroft 		reassignbuf(bp, bp->b_vp);
    313       1.1   mycroft 		splx(s);
    314       1.9  perseant 
    315       1.1   mycroft 	}
    316       1.7  perseant 
    317       1.9  perseant 	if(bp->b_flags & B_CALL)
    318       1.7  perseant 		bp->b_flags &= ~B_BUSY;
    319       1.7  perseant 	else
    320       1.7  perseant 		brelse(bp);
    321       1.7  perseant 
    322       1.1   mycroft 	return (0);
    323       1.1   mycroft }
    324       1.1   mycroft 
    325  1.23.2.1  perseant void
    326  1.23.2.1  perseant lfs_flush_fs(fs, flags)
    327  1.23.2.1  perseant 	struct lfs *fs;
    328      1.11  perseant 	int flags;
    329      1.11  perseant {
    330  1.23.2.1  perseant 	if(fs->lfs_ronly == 0 && fs->lfs_dirops == 0)
    331      1.11  perseant 	{
    332      1.11  perseant 		/* disallow dirops during flush */
    333  1.23.2.1  perseant 		fs->lfs_writer++;
    334      1.11  perseant 
    335      1.11  perseant 		/*
    336      1.11  perseant 		 * We set the queue to 0 here because we
    337      1.11  perseant 		 * are about to write all the dirty
    338      1.11  perseant 		 * buffers we have.  If more come in
    339      1.11  perseant 		 * while we're writing the segment, they
    340      1.11  perseant 		 * may not get written, so we want the
    341      1.11  perseant 		 * count to reflect these new writes
    342      1.11  perseant 		 * after the segwrite completes.
    343      1.11  perseant 		 */
    344      1.11  perseant 		if(lfs_dostats)
    345      1.11  perseant 			++lfs_stats.flush_invoked;
    346  1.23.2.1  perseant 		lfs_segwrite(fs->lfs_ivnode->v_mount, flags);
    347      1.11  perseant 
    348      1.11  perseant 		/* XXX KS - allow dirops again */
    349  1.23.2.1  perseant 		if(--fs->lfs_writer==0)
    350  1.23.2.1  perseant 			wakeup(&fs->lfs_dirops);
    351      1.11  perseant 	}
    352      1.11  perseant }
    353      1.11  perseant 
    354       1.1   mycroft /*
    355       1.1   mycroft  * XXX
    356       1.1   mycroft  * This routine flushes buffers out of the B_LOCKED queue when LFS has too
    357       1.1   mycroft  * many locked down.  Eventually the pageout daemon will simply call LFS
    358       1.1   mycroft  * when pages need to be reclaimed.  Note, we have one static count of locked
    359       1.1   mycroft  * buffers, so we can't have more than a single file system.  To make this
    360       1.1   mycroft  * work for multiple file systems, put the count into the mount structure.
    361       1.1   mycroft  */
    362       1.1   mycroft void
    363       1.7  perseant lfs_flush(fs, flags)
    364       1.7  perseant 	struct lfs *fs;
    365       1.7  perseant 	int flags;
    366       1.1   mycroft {
    367  1.23.2.1  perseant 	int s;
    368      1.17  augustss 	struct mount *mp, *nmp;
    369       1.7  perseant 
    370       1.7  perseant 	if(lfs_dostats)
    371       1.7  perseant 		++lfs_stats.write_exceeded;
    372      1.16  perseant 	if (lfs_writing && flags==0) {/* XXX flags */
    373      1.16  perseant #ifdef DEBUG_LFS
    374      1.16  perseant 		printf("lfs_flush: not flushing because another flush is active\n");
    375      1.16  perseant #endif
    376       1.1   mycroft 		return;
    377      1.16  perseant 	}
    378       1.1   mycroft 	lfs_writing = 1;
    379       1.7  perseant 
    380       1.6      fvdl 	simple_lock(&mountlist_slock);
    381       1.6      fvdl 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
    382       1.6      fvdl 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
    383       1.6      fvdl 			nmp = mp->mnt_list.cqe_next;
    384       1.6      fvdl 			continue;
    385       1.6      fvdl 		}
    386       1.7  perseant 		if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS, MFSNAMELEN)==0)
    387  1.23.2.1  perseant 			lfs_flush_fs(((struct ufsmount *)mp->mnt_data)->ufsmount_u.lfs, flags);
    388       1.6      fvdl 		simple_lock(&mountlist_slock);
    389       1.6      fvdl 		nmp = mp->mnt_list.cqe_next;
    390       1.6      fvdl 		vfs_unbusy(mp);
    391       1.1   mycroft 	}
    392       1.6      fvdl 	simple_unlock(&mountlist_slock);
    393       1.7  perseant 
    394  1.23.2.1  perseant 	s = splbio();
    395  1.23.2.1  perseant 	lfs_countlocked(&locked_queue_count, &locked_queue_bytes);
    396  1.23.2.1  perseant 	splx(s);
    397       1.7  perseant 	wakeup(&locked_queue_count);
    398       1.7  perseant 
    399       1.1   mycroft 	lfs_writing = 0;
    400       1.1   mycroft }
    401       1.1   mycroft 
    402       1.1   mycroft int
    403       1.7  perseant lfs_check(vp, blkno, flags)
    404       1.1   mycroft 	struct vnode *vp;
    405       1.6      fvdl 	ufs_daddr_t blkno;
    406       1.7  perseant 	int flags;
    407       1.1   mycroft {
    408       1.1   mycroft 	int error;
    409      1.10  perseant 	struct lfs *fs;
    410      1.20  perseant 	struct inode *ip;
    411      1.13  perseant 	extern int lfs_dirvcount;
    412      1.10  perseant 
    413       1.1   mycroft 	error = 0;
    414      1.20  perseant 	ip = VTOI(vp);
    415       1.7  perseant 
    416       1.7  perseant 	/* If out of buffers, wait on writer */
    417       1.7  perseant 	/* XXX KS - if it's the Ifile, we're probably the cleaner! */
    418      1.20  perseant 	if (ip->i_number == LFS_IFILE_INUM)
    419      1.20  perseant 		return 0;
    420      1.20  perseant 	/* If we're being called from inside a dirop, don't sleep */
    421      1.20  perseant 	if (ip->i_flag & IN_ADIROP)
    422       1.7  perseant 		return 0;
    423       1.7  perseant 
    424      1.20  perseant 	fs = ip->i_lfs;
    425      1.20  perseant 
    426      1.20  perseant 	/*
    427      1.20  perseant 	 * If we would flush below, but dirops are active, sleep.
    428      1.20  perseant 	 * Note that a dirop cannot ever reach this code!
    429      1.20  perseant 	 */
    430      1.20  perseant 	while (fs->lfs_dirops > 0 &&
    431      1.20  perseant 	       (locked_queue_count > LFS_MAX_BUFS ||
    432      1.20  perseant                 locked_queue_bytes > LFS_MAX_BYTES ||
    433      1.20  perseant                 lfs_dirvcount > LFS_MAXDIROP || fs->lfs_diropwait > 0))
    434      1.20  perseant 	{
    435      1.20  perseant 		++fs->lfs_diropwait;
    436      1.20  perseant 		tsleep(&fs->lfs_writer, PRIBIO+1, "bufdirop", 0);
    437      1.20  perseant 		--fs->lfs_diropwait;
    438      1.20  perseant 	}
    439      1.10  perseant 
    440      1.20  perseant 	if (locked_queue_count > LFS_MAX_BUFS ||
    441      1.20  perseant 	    locked_queue_bytes > LFS_MAX_BYTES ||
    442      1.20  perseant 	    lfs_dirvcount > LFS_MAXDIROP || fs->lfs_diropwait > 0)
    443       1.7  perseant 	{
    444      1.10  perseant 		++fs->lfs_writer;
    445      1.10  perseant 		lfs_flush(fs, flags);
    446      1.10  perseant 		if(--fs->lfs_writer==0)
    447      1.10  perseant 			wakeup(&fs->lfs_dirops);
    448       1.7  perseant 	}
    449      1.11  perseant 
    450       1.7  perseant 	while  (locked_queue_count > LFS_WAIT_BUFS
    451       1.7  perseant 		|| locked_queue_bytes > LFS_WAIT_BYTES)
    452       1.7  perseant 	{
    453       1.7  perseant 		if(lfs_dostats)
    454       1.7  perseant 			++lfs_stats.wait_exceeded;
    455      1.16  perseant #ifdef DEBUG_LFS
    456      1.16  perseant 		printf("lfs_check: waiting: count=%d, bytes=%ld\n",
    457      1.16  perseant 			locked_queue_count, locked_queue_bytes);
    458      1.16  perseant #endif
    459       1.7  perseant 		error = tsleep(&locked_queue_count, PCATCH | PUSER,
    460       1.7  perseant 			       "buffers", hz * LFS_BUFWAIT);
    461      1.18  perseant 		/*
    462      1.18  perseant 		 * lfs_flush might not flush all the buffers, if some of the
    463      1.18  perseant 		 * inodes were locked.  Try flushing again to keep us from
    464      1.18  perseant 		 * blocking indefinitely.
    465      1.18  perseant 		 */
    466      1.20  perseant 		if (locked_queue_count > LFS_MAX_BUFS ||
    467      1.20  perseant 		    locked_queue_bytes > LFS_MAX_BYTES)
    468      1.18  perseant 		{
    469      1.18  perseant 			++fs->lfs_writer;
    470      1.18  perseant 			lfs_flush(fs, flags);
    471      1.18  perseant 			if(--fs->lfs_writer==0)
    472      1.18  perseant 				wakeup(&fs->lfs_dirops);
    473      1.18  perseant 		}
    474       1.7  perseant 	}
    475       1.7  perseant 	return (error);
    476       1.7  perseant }
    477       1.1   mycroft 
    478       1.7  perseant /*
    479       1.7  perseant  * Allocate a new buffer header.
    480       1.7  perseant  */
    481       1.7  perseant struct buf *
    482       1.7  perseant lfs_newbuf(vp, daddr, size)
    483       1.7  perseant 	struct vnode *vp;
    484       1.7  perseant 	ufs_daddr_t daddr;
    485       1.7  perseant 	size_t size;
    486       1.7  perseant {
    487       1.7  perseant 	struct buf *bp;
    488       1.7  perseant 	size_t nbytes;
    489       1.7  perseant 	int s;
    490       1.7  perseant 
    491       1.7  perseant 	nbytes = roundup(size, DEV_BSIZE);
    492       1.7  perseant 
    493       1.7  perseant 	bp = malloc(sizeof(struct buf), M_SEGMENT, M_WAITOK);
    494       1.7  perseant 	bzero(bp, sizeof(struct buf));
    495       1.7  perseant 	if (nbytes)
    496       1.7  perseant 		bp->b_data = malloc(nbytes, M_SEGMENT, M_WAITOK);
    497       1.7  perseant 	if(nbytes) {
    498       1.7  perseant 		bzero(bp->b_data, nbytes);
    499       1.7  perseant 	}
    500       1.7  perseant #ifdef DIAGNOSTIC
    501       1.7  perseant 	if(vp==NULL)
    502       1.7  perseant 		panic("vp is NULL in lfs_newbuf");
    503       1.7  perseant 	if(bp==NULL)
    504       1.7  perseant 		panic("bp is NULL after malloc in lfs_newbuf");
    505       1.1   mycroft #endif
    506       1.7  perseant 	s = splbio();
    507       1.7  perseant 	bgetvp(vp, bp);
    508       1.7  perseant 	splx(s);
    509       1.7  perseant 
    510       1.7  perseant 	bp->b_bufsize = size;
    511       1.7  perseant 	bp->b_bcount = size;
    512       1.7  perseant 	bp->b_lblkno = daddr;
    513       1.7  perseant 	bp->b_blkno = daddr;
    514       1.7  perseant 	bp->b_error = 0;
    515       1.7  perseant 	bp->b_resid = 0;
    516       1.7  perseant 	bp->b_iodone = lfs_callback;
    517       1.7  perseant 	bp->b_flags |= B_BUSY | B_CALL | B_NOCACHE;
    518       1.7  perseant 
    519       1.7  perseant 	return (bp);
    520       1.7  perseant }
    521       1.7  perseant 
    522       1.7  perseant void
    523       1.7  perseant lfs_freebuf(bp)
    524       1.7  perseant 	struct buf *bp;
    525       1.7  perseant {
    526       1.7  perseant 	int s;
    527       1.7  perseant 
    528       1.7  perseant 	s = splbio();
    529       1.7  perseant 	if(bp->b_vp)
    530       1.7  perseant 		brelvp(bp);
    531       1.7  perseant 	splx(s);
    532       1.7  perseant 	if (!(bp->b_flags & B_INVAL)) { /* B_INVAL indicates a "fake" buffer */
    533       1.7  perseant 		free(bp->b_data, M_SEGMENT);
    534       1.7  perseant 		bp->b_data = NULL;
    535       1.1   mycroft 	}
    536       1.7  perseant 	free(bp, M_SEGMENT);
    537       1.7  perseant }
    538       1.1   mycroft 
    539       1.7  perseant /*
    540       1.7  perseant  * Definitions for the buffer free lists.
    541       1.7  perseant  */
    542       1.7  perseant #define BQUEUES		4		/* number of free buffer queues */
    543       1.7  perseant 
    544       1.7  perseant #define BQ_LOCKED	0		/* super-blocks &c */
    545       1.7  perseant #define BQ_LRU		1		/* lru, useful buffers */
    546       1.7  perseant #define BQ_AGE		2		/* rubbish */
    547       1.7  perseant #define BQ_EMPTY	3		/* buffer headers with no memory */
    548       1.7  perseant 
    549       1.7  perseant extern TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
    550       1.7  perseant 
    551       1.7  perseant /*
    552       1.7  perseant  * Return a count of buffers on the "locked" queue.
    553  1.23.2.1  perseant  * Don't count malloced buffers, since they don't detract from the total.
    554       1.7  perseant  */
    555       1.7  perseant void
    556       1.7  perseant lfs_countlocked(count, bytes)
    557       1.7  perseant 	int *count;
    558       1.7  perseant 	long *bytes;
    559       1.7  perseant {
    560      1.17  augustss 	struct buf *bp;
    561      1.17  augustss 	int n = 0;
    562      1.17  augustss 	long int size = 0L;
    563       1.7  perseant 
    564       1.7  perseant 	for (bp = bufqueues[BQ_LOCKED].tqh_first; bp;
    565       1.7  perseant 	    bp = bp->b_freelist.tqe_next) {
    566  1.23.2.1  perseant #ifdef DEBUG_LOCKED_LIST
    567  1.23.2.1  perseant 		if (bp->b_flags & B_CALL) /* Malloced buffer */
    568  1.23.2.1  perseant 			continue;
    569  1.23.2.1  perseant #endif
    570       1.7  perseant 		n++;
    571       1.7  perseant 		size += bp->b_bufsize;
    572  1.23.2.1  perseant #ifdef DEBUG_LOCKED_LIST
    573  1.23.2.1  perseant 		if (n > nbuf)
    574  1.23.2.1  perseant 			panic("lfs_countlocked: this can't happen: more"
    575  1.23.2.1  perseant 			      " buffers locked than exist");
    576  1.23.2.1  perseant #endif
    577       1.7  perseant 	}
    578  1.23.2.1  perseant #ifdef DEBUG_LFS
    579  1.23.2.1  perseant 	/* Theoretically this function never really does anything */
    580  1.23.2.1  perseant 	if (n != *count)
    581  1.23.2.1  perseant 		printf("lfs_countlocked: adjusted buf count from %d to %d\n",
    582  1.23.2.1  perseant 		       *count, n);
    583  1.23.2.1  perseant 	if (size != *bytes)
    584  1.23.2.1  perseant 		printf("lfs_countlocked: adjusted byte count from %ld to %ld\n",
    585  1.23.2.1  perseant 		       *bytes, size);
    586  1.23.2.1  perseant #endif
    587       1.7  perseant 	*count = n;
    588       1.7  perseant 	*bytes = size;
    589       1.7  perseant 	return;
    590       1.1   mycroft }
    591