lfs_segment.c

来自「早期freebsd实现」· C语言 代码 · 共 1,112 行 · 第 1/2 页

/*
 * Copyright (c) 1991, 1993
 *	The Regents of the University of California.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *	@(#)lfs_segment.c	8.5 (Berkeley) 1/4/94
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/namei.h>
#include <sys/kernel.h>
#include <sys/resourcevar.h>
#include <sys/file.h>
#include <sys/stat.h>
#include <sys/buf.h>
#include <sys/proc.h>
#include <sys/conf.h>
#include <sys/vnode.h>
#include <sys/malloc.h>
#include <sys/mount.h>

#include <miscfs/specfs/specdev.h>
#include <miscfs/fifofs/fifo.h>

#include <ufs/ufs/quota.h>
#include <ufs/ufs/inode.h>
#include <ufs/ufs/dir.h>
#include <ufs/ufs/ufsmount.h>
#include <ufs/ufs/ufs_extern.h>

#include <ufs/lfs/lfs.h>
#include <ufs/lfs/lfs_extern.h>

extern int count_lock_queue __P((void));

#define MAX_ACTIVE	10
/*
 * Determine if it's OK to start a partial in this segment, or if we need
 * to go on to a new segment.
 */
#define	LFS_PARTIAL_FITS(fs) \
	((fs)->lfs_dbpseg - ((fs)->lfs_offset - (fs)->lfs_curseg) > \
	1 << (fs)->lfs_fsbtodb)

void	 lfs_callback __P((struct buf *));
void	 lfs_gather __P((struct lfs *, struct segment *,
	     struct vnode *, int (*) __P((struct lfs *, struct buf *))));
int	 lfs_gatherblock __P((struct segment *, struct buf *, int *));
void	 lfs_iset __P((struct inode *, daddr_t, time_t));
int	 lfs_match_data __P((struct lfs *, struct buf *));
int	 lfs_match_dindir __P((struct lfs *, struct buf *));
int	 lfs_match_indir __P((struct lfs *, struct buf *));
int	 lfs_match_tindir __P((struct lfs *, struct buf *));
void	 lfs_newseg __P((struct lfs *));
void	 lfs_shellsort __P((struct buf **, daddr_t *, register int));
void	 lfs_supercallback __P((struct buf *));
void	 lfs_updatemeta __P((struct segment *));
int	 lfs_vref __P((struct vnode *));
void	 lfs_vunref __P((struct vnode *));
void	 lfs_writefile __P((struct lfs *, struct segment *, struct vnode *));
int	 lfs_writeinode __P((struct lfs *, struct segment *, struct inode *));
int	 lfs_writeseg __P((struct lfs *, struct segment *));
void	 lfs_writesuper __P((struct lfs *));
void	 lfs_writevnodes __P((struct lfs *fs, struct mount *mp,
	    struct segment *sp, int dirops));

int	lfs_allclean_wakeup;		/* Cleaner wakeup address. */

/* Statistics Counters */
#define DOSTATS
struct lfs_stats lfs_stats;

/* op values to lfs_writevnodes */
#define	VN_REG	0
#define	VN_DIROP	1
#define	VN_EMPTY	2

/*
 * Ifile and meta data blocks are not marked busy, so segment writes MUST be
 * single threaded.  Currently, there are two paths into lfs_segwrite, sync()
 * and getnewbuf().  They both mark the file system busy.  Lfs_vflush()
 * explicitly marks the file system busy.  So lfs_segwrite is safe.  I think.
 */

int
lfs_vflush(vp)
	struct vnode *vp;
{
	struct inode *ip;
	struct lfs *fs;
	struct segment *sp;

	fs = VFSTOUFS(vp->v_mount)->um_lfs;
	if (fs->lfs_nactive > MAX_ACTIVE)
		return(lfs_segwrite(vp->v_mount, SEGM_SYNC|SEGM_CKP));
	lfs_seglock(fs, SEGM_SYNC);
	sp = fs->lfs_sp;


	ip = VTOI(vp);
	if (vp->v_dirtyblkhd.lh_first == NULL)
		lfs_writevnodes(fs, vp->v_mount, sp, VN_EMPTY);

	do {
		do {
			if (vp->v_dirtyblkhd.lh_first != NULL)
				lfs_writefile(fs, sp, vp);
		} while (lfs_writeinode(fs, sp, ip));

	} while (lfs_writeseg(fs, sp) && ip->i_number == LFS_IFILE_INUM);

#ifdef DOSTATS
	++lfs_stats.nwrites;
	if (sp->seg_flags & SEGM_SYNC)
		++lfs_stats.nsync_writes;
	if (sp->seg_flags & SEGM_CKP)
		++lfs_stats.ncheckpoints;
#endif
	lfs_segunlock(fs);
	return (0);
}

void
lfs_writevnodes(fs, mp, sp, op)
	struct lfs *fs;
	struct mount *mp;
	struct segment *sp;
	int op;
{
	struct inode *ip;
	struct vnode *vp;

loop:
	for (vp = mp->mnt_vnodelist.lh_first;
	     vp != NULL;
	     vp = vp->v_mntvnodes.le_next) {
		/*
		 * If the vnode that we are about to sync is no longer
		 * associated with this mount point, start over.
		 */
		if (vp->v_mount != mp)
			goto loop;

		/* XXX ignore dirops for now
		if (op == VN_DIROP && !(vp->v_flag & VDIROP) ||
		    op != VN_DIROP && (vp->v_flag & VDIROP))
			continue;
		*/

		if (op == VN_EMPTY && vp->v_dirtyblkhd.lh_first)
			continue;

		if (vp->v_type == VNON)
			continue;

		if (lfs_vref(vp))
			continue;

		/*
		 * Write the inode/file if dirty and it's not the
		 * the IFILE.
		 */
		ip = VTOI(vp);
		if ((ip->i_flag &
		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE) ||
		    vp->v_dirtyblkhd.lh_first != NULL) &&
		    ip->i_number != LFS_IFILE_INUM) {
			if (vp->v_dirtyblkhd.lh_first != NULL)
				lfs_writefile(fs, sp, vp);
			(void) lfs_writeinode(fs, sp, ip);
		}
		vp->v_flag &= ~VDIROP;
		lfs_vunref(vp);
	}
}

int
lfs_segwrite(mp, flags)
	struct mount *mp;
	int flags;			/* Do a checkpoint. */
{
	struct buf *bp;
	struct inode *ip;
	struct lfs *fs;
	struct segment *sp;
	struct vnode *vp;
	SEGUSE *segusep;
	daddr_t ibno;
	CLEANERINFO *cip;
	int clean, do_ckp, error, i;

	fs = VFSTOUFS(mp)->um_lfs;

 	/*
 	 * If we have fewer than 2 clean segments, wait until cleaner
	 * writes.
 	 */
	do {
		LFS_CLEANERINFO(cip, fs, bp);
		clean = cip->clean;
		brelse(bp);
		if (clean <= 2) {
			printf ("segs clean: %d\n", clean);
			wakeup(&lfs_allclean_wakeup);
			if (error = tsleep(&fs->lfs_avail, PRIBIO + 1,
			    "lfs writer", 0))
				return (error);
		}
	} while (clean <= 2 );

	/*
	 * Allocate a segment structure and enough space to hold pointers to
	 * the maximum possible number of buffers which can be described in a
	 * single summary block.
	 */
	do_ckp = flags & SEGM_CKP || fs->lfs_nactive > MAX_ACTIVE;
	lfs_seglock(fs, flags | (do_ckp ? SEGM_CKP : 0));
	sp = fs->lfs_sp;

	lfs_writevnodes(fs, mp, sp, VN_REG);

	/* XXX ignore ordering of dirops for now */
	/* XXX
	fs->lfs_writer = 1;
	if (fs->lfs_dirops && (error =
	    tsleep(&fs->lfs_writer, PRIBIO + 1, "lfs writer", 0))) {
		free(sp->bpp, M_SEGMENT);
		free(sp, M_SEGMENT); 
		fs->lfs_writer = 0;
		return (error);
	}

	lfs_writevnodes(fs, mp, sp, VN_DIROP);
	*/

	/*
	 * If we are doing a checkpoint, mark everything since the
	 * last checkpoint as no longer ACTIVE.
	 */
	if (do_ckp)
		for (ibno = fs->lfs_cleansz + fs->lfs_segtabsz;
		     --ibno >= fs->lfs_cleansz; ) {
			if (bread(fs->lfs_ivnode, ibno, fs->lfs_bsize,
			    NOCRED, &bp))

				panic("lfs: ifile read");
			segusep = (SEGUSE *)bp->b_data;
			for (i = fs->lfs_sepb; i--; segusep++)
				segusep->su_flags &= ~SEGUSE_ACTIVE;
				
			error = VOP_BWRITE(bp);
		}

	if (do_ckp || fs->lfs_doifile) {
redo:
		vp = fs->lfs_ivnode;
		while (vget(vp, 1));
		ip = VTOI(vp);
		if (vp->v_dirtyblkhd.lh_first != NULL)
			lfs_writefile(fs, sp, vp);
		(void)lfs_writeinode(fs, sp, ip);
		vput(vp);
		if (lfs_writeseg(fs, sp) && do_ckp)
			goto redo;
	} else
		(void) lfs_writeseg(fs, sp);

	/*
	 * If the I/O count is non-zero, sleep until it reaches zero.  At the
	 * moment, the user's process hangs around so we can sleep.
	 */
	/* XXX ignore dirops for now
	fs->lfs_writer = 0;
	fs->lfs_doifile = 0;
	wakeup(&fs->lfs_dirops);
	*/

#ifdef DOSTATS
	++lfs_stats.nwrites;
	if (sp->seg_flags & SEGM_SYNC)
		++lfs_stats.nsync_writes;
	if (sp->seg_flags & SEGM_CKP)
		++lfs_stats.ncheckpoints;
#endif
	lfs_segunlock(fs);
	return (0);
}

/*
 * Write the dirty blocks associated with a vnode.
 */
void
lfs_writefile(fs, sp, vp)
	struct lfs *fs;
	struct segment *sp;
	struct vnode *vp;
{
	struct buf *bp;
	struct finfo *fip;
	IFILE *ifp;

	if (sp->seg_bytes_left < fs->lfs_bsize ||
	    sp->sum_bytes_left < sizeof(struct finfo))
		(void) lfs_writeseg(fs, sp);

	sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(daddr_t);
	++((SEGSUM *)(sp->segsum))->ss_nfinfo;

	fip = sp->fip;
	fip->fi_nblocks = 0;
	fip->fi_ino = VTOI(vp)->i_number;
	LFS_IENTRY(ifp, fs, fip->fi_ino, bp);
	fip->fi_version = ifp->if_version;
	brelse(bp);

	/*
	 * It may not be necessary to write the meta-data blocks at this point,
	 * as the roll-forward recovery code should be able to reconstruct the
	 * list.
	 */
	lfs_gather(fs, sp, vp, lfs_match_data);
	lfs_gather(fs, sp, vp, lfs_match_indir);
	lfs_gather(fs, sp, vp, lfs_match_dindir);
#ifdef TRIPLE
	lfs_gather(fs, sp, vp, lfs_match_tindir);
#endif

	fip = sp->fip;
	if (fip->fi_nblocks != 0) {
		sp->fip =
		    (struct finfo *)((caddr_t)fip + sizeof(struct finfo) +
		    sizeof(daddr_t) * (fip->fi_nblocks - 1));
		sp->start_lbp = &sp->fip->fi_blocks[0];
	} else {
		sp->sum_bytes_left += sizeof(struct finfo) - sizeof(daddr_t);
		--((SEGSUM *)(sp->segsum))->ss_nfinfo;
	}
}

int
lfs_writeinode(fs, sp, ip)
	struct lfs *fs;
	struct segment *sp;
	struct inode *ip;
{
	struct buf *bp, *ibp;
	IFILE *ifp;
	SEGUSE *sup;
	daddr_t daddr;
	ino_t ino;
	int error, i, ndx;
	int redo_ifile = 0;

	if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)))
		return(0);

	/* Allocate a new inode block if necessary. */
	if (sp->ibp == NULL) {
		/* Allocate a new segment if necessary. */
		if (sp->seg_bytes_left < fs->lfs_bsize ||
		    sp->sum_bytes_left < sizeof(daddr_t))
			(void) lfs_writeseg(fs, sp);

		/* Get next inode block. */
		daddr = fs->lfs_offset;
		fs->lfs_offset += fsbtodb(fs, 1);
		sp->ibp = *sp->cbpp++ =
		    lfs_newbuf(VTOI(fs->lfs_ivnode)->i_devvp, daddr,
		    fs->lfs_bsize);
		/* Zero out inode numbers */
		for (i = 0; i < INOPB(fs); ++i)
			((struct dinode *)sp->ibp->b_data)[i].di_inumber = 0;
		++sp->start_bpp;
		fs->lfs_avail -= fsbtodb(fs, 1);
		/* Set remaining space counters. */
		sp->seg_bytes_left -= fs->lfs_bsize;
		sp->sum_bytes_left -= sizeof(daddr_t);
		ndx = LFS_SUMMARY_SIZE / sizeof(daddr_t) -
		    sp->ninodes / INOPB(fs) - 1;
		((daddr_t *)(sp->segsum))[ndx] = daddr;
	}

	/* Update the inode times and copy the inode onto the inode page. */
	if (ip->i_flag & IN_MODIFIED)
		--fs->lfs_uinodes;
	ITIMES(ip, &time, &time);
	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
	bp = sp->ibp;
	((struct dinode *)bp->b_data)[sp->ninodes % INOPB(fs)] = ip->i_din;
	/* Increment inode count in segment summary block. */
	++((SEGSUM *)(sp->segsum))->ss_ninos;

	/* If this page is full, set flag to allocate a new page. */
	if (++sp->ninodes % INOPB(fs) == 0)
		sp->ibp = NULL;

	/*
	 * If updating the ifile, update the super-block.  Update the disk
	 * address and access times for this inode in the ifile.
	 */
	ino = ip->i_number;
	if (ino == LFS_IFILE_INUM) {
		daddr = fs->lfs_idaddr;
		fs->lfs_idaddr = bp->b_blkno;
	} else {
		LFS_IENTRY(ifp, fs, ino, ibp);
		daddr = ifp->if_daddr;
		ifp->if_daddr = bp->b_blkno;
		error = VOP_BWRITE(ibp);
	}

	/*
	 * No need to update segment usage if there was no former inode address
	 * or if the last inode address is in the current partial segment.
	 */
	if (daddr != LFS_UNUSED_DADDR && 
	    !(daddr >= fs->lfs_lastpseg && daddr <= bp->b_blkno)) {
		LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp);
#ifdef DIAGNOSTIC
		if (sup->su_nbytes < sizeof(struct dinode)) {
			/* XXX -- Change to a panic. */
			printf("lfs: negative bytes (segment %d)\n",
			    datosn(fs, daddr));
			panic("negative bytes");
		}
#endif
		sup->su_nbytes -= sizeof(struct dinode);
		redo_ifile =
		    (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));
		error = VOP_BWRITE(bp);
	}
	return (redo_ifile);
}

int
lfs_gatherblock(sp, bp, sptr)
	struct segment *sp;
	struct buf *bp;
	int *sptr;
{
	struct lfs *fs;
	int version;

	/*
	 * If full, finish this segment.  We may be doing I/O, so
	 * release and reacquire the splbio().
	 */
#ifdef DIAGNOSTIC
	if (sp->vp == NULL)
		panic ("lfs_gatherblock: Null vp in segment");
#endif
	fs = sp->fs;
	if (sp->sum_bytes_left < sizeof(daddr_t) ||
	    sp->seg_bytes_left < fs->lfs_bsize) {
		if (sptr)
			splx(*sptr);
		lfs_updatemeta(sp);

		version = sp->fip->fi_version;
		(void) lfs_writeseg(fs, sp);

		sp->fip->fi_version = version;
		sp->fip->fi_ino = VTOI(sp->vp)->i_number;
		/* Add the current file to the segment summary. */
		++((SEGSUM *)(sp->segsum))->ss_nfinfo;
		sp->sum_bytes_left -= 
		    sizeof(struct finfo) - sizeof(daddr_t);

		if (sptr)
			*sptr = splbio();
		return(1);
	}

	/* Insert into the buffer list, update the FINFO block. */
	bp->b_flags |= B_GATHERED;
	*sp->cbpp++ = bp;
	sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno;

	sp->sum_bytes_left -= sizeof(daddr_t);
	sp->seg_bytes_left -= fs->lfs_bsize;
	return(0);
}

void
lfs_gather(fs, sp, vp, match)
	struct lfs *fs;
	struct segment *sp;
	struct vnode *vp;
	int (*match) __P((struct lfs *, struct buf *));
{
	struct buf *bp;
	int s;

	sp->vp = vp;
	s = splbio();
loop:	for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = bp->b_vnbufs.le_next) {
		if (bp->b_flags & B_BUSY || !match(fs, bp) ||
		    bp->b_flags & B_GATHERED)
			continue;
#ifdef DIAGNOSTIC
		if (!(bp->b_flags & B_DELWRI))
			panic("lfs_gather: bp not B_DELWRI");
		if (!(bp->b_flags & B_LOCKED))
			panic("lfs_gather: bp not B_LOCKED");
#endif
		if (lfs_gatherblock(sp, bp, &s))
			goto loop;
	}
	splx(s);
	lfs_updatemeta(sp);
	sp->vp = NULL;
}


/*
 * Update the metadata that points to the blocks listed in the FINFO
 * array.
 */
void
lfs_updatemeta(sp)
	struct segment *sp;
{
	SEGUSE *sup;
	struct buf *bp;
	struct lfs *fs;
	struct vnode *vp;
	struct indir a[NIADDR + 2], *ap;
	struct inode *ip;