vfs_bio.c

来自「基于组件方式开发操作系统的OSKIT源代码」· C语言 代码 · 共 871 行 · 第 1/2 页

	 */

	/* If it's locked, don't report an error; try again later. */
	if (ISSET(bp->b_flags, (B_LOCKED|B_ERROR)) == (B_LOCKED|B_ERROR))
		CLR(bp->b_flags, B_ERROR);

	/* If it's not cacheable, or an error, mark it invalid. */
	if (ISSET(bp->b_flags, (B_NOCACHE|B_ERROR)))
		SET(bp->b_flags, B_INVAL);

	if ((bp->b_bufsize <= 0) || ISSET(bp->b_flags, B_INVAL)) {
		/*
		 * If it's invalid or empty, dissociate it from its vnode
		 * and put on the head of the appropriate queue.
		 */
		if (bp->b_vp)
			brelvp(bp);
		CLR(bp->b_flags, B_DELWRI);
		if (bp->b_bufsize <= 0)
			/* no data */
			bufq = &bufqueues[BQ_EMPTY];
		else
			/* invalid data */
			bufq = &bufqueues[BQ_AGE];
		binsheadfree(bp, bufq);
	} else {
		/*
		 * It has valid data.  Put it on the end of the appropriate
		 * queue, so that it'll stick around for as long as possible.
		 */
		if (ISSET(bp->b_flags, B_LOCKED))
			/* locked in core */
			bufq = &bufqueues[BQ_LOCKED];
		else if (ISSET(bp->b_flags, B_AGE))
			/* stale but valid data */
			bufq = &bufqueues[BQ_AGE];
		else
			/* valid data */
			bufq = &bufqueues[BQ_LRU];
		binstailfree(bp, bufq);
	}

	/* Unlock the buffer. */
	CLR(bp->b_flags, (B_AGE | B_ASYNC | B_BUSY | B_NOCACHE));

	/* Allow disk interrupts. */
	splx(s);
}

/*
 * Determine if a block is in the cache.
 * Just look on what would be its hash chain.  If it's there, return
 * a pointer to it, unless it's marked invalid.  If it's marked invalid,
 * we normally don't return the buffer, unless the caller explicitly
 * wants us to.
 */
struct buf *
incore(vp, blkno)
	struct vnode *vp;
	daddr_t blkno;
{
	struct buf *bp;

	bp = BUFHASH(vp, blkno)->lh_first;

	/* Search hash chain */
	for (; bp != NULL; bp = bp->b_hash.le_next) {
		if (bp->b_lblkno == blkno && bp->b_vp == vp &&
		    !ISSET(bp->b_flags, B_INVAL))
		return (bp);
	}

	return (0);
}

/*
 * Get a block of requested size that is associated with
 * a given vnode and block offset. If it is found in the
 * block cache, mark it as having been found, make it busy
 * and return it. Otherwise, return an empty block of the
 * correct size. It is up to the caller to insure that the
 * cached blocks be of the correct size.
 */
struct buf *
getblk(vp, blkno, size, slpflag, slptimeo)
	register struct vnode *vp;
	daddr_t blkno;
	int size, slpflag, slptimeo;
{
	struct bufhashhdr *bh;
	struct buf *bp;
	int s, err;

	/*
	 * XXX
	 * The following is an inlined version of 'incore()', but with
	 * the 'invalid' test moved to after the 'busy' test.  It's
	 * necessary because there are some cases in which the NFS
	 * code sets B_INVAL prior to writing data to the server, but 
	 * in which the buffers actually contain valid data.  In this
	 * case, we can't allow the system to allocate a new buffer for
	 * the block until the write is finished.
	 */
	bh = BUFHASH(vp, blkno);
start:
        bp = bh->lh_first;
        for (; bp != NULL; bp = bp->b_hash.le_next) {
                if (bp->b_lblkno != blkno || bp->b_vp != vp)
			continue;

		s = splbio();
		if (ISSET(bp->b_flags, B_BUSY)) {
			SET(bp->b_flags, B_WANTED);
			err = tsleep(bp, slpflag | (PRIBIO + 1), "getblk",
			    slptimeo);
			splx(s);
			if (err)
				return (NULL);
			goto start;
		}

		if (!ISSET(bp->b_flags, B_INVAL)) {
			SET(bp->b_flags, (B_BUSY | B_CACHE));
			bremfree(bp);
			splx(s);
			break;
		}
		splx(s);
        }

	if (bp == NULL) {
		if ((bp = getnewbuf(slpflag, slptimeo)) == NULL)
			goto start;
		binshash(bp, bh);
		bp->b_blkno = bp->b_lblkno = blkno;
		s = splbio();
		bgetvp(vp, bp);
		splx(s);
	}
	allocbuf(bp, size);
	return (bp);
}

/*
 * Get an empty, disassociated buffer of given size.
 */
struct buf *
geteblk(size)
	int size;
{
	struct buf *bp; 

	while ((bp = getnewbuf(0, 0)) == 0)
		;
	SET(bp->b_flags, B_INVAL);
	binshash(bp, &invalhash);
	allocbuf(bp, size);

	return (bp);
}

/*
 * Expand or contract the actual memory allocated to a buffer.
 *
 * If the buffer shrinks, data is lost, so it's up to the
 * caller to have written it out *first*; this routine will not
 * start a write.  If the buffer grows, it's the callers
 * responsibility to fill out the buffer's additional contents.
 */
void
allocbuf(bp, size)
	struct buf *bp;
	int size;
{
	struct buf      *nbp;
	vm_size_t       desired_size;
	int	     s;

	desired_size = roundup(size, CLBYTES);
	if (desired_size > MAXBSIZE)
		panic("allocbuf: buffer larger than MAXBSIZE requested");

	if (bp->b_bufsize == desired_size)
		goto out;

#ifdef OSKIT
	if (bp->b_bufsize > desired_size)
		goto out;
	
	panic("allocbuf:  not enough space");
#else
	/*
	 * If the buffer is smaller than the desired size, we need to snarf
	 * it from other buffers.  Get buffers (via getnewbuf()), and
	 * steal their pages.
	 */
	while (bp->b_bufsize < desired_size) {
		int amt;

		/* find a buffer */
		while ((nbp = getnewbuf(0, 0)) == NULL)
			;
		SET(nbp->b_flags, B_INVAL);
		binshash(nbp, &invalhash);

		/* and steal its pages, up to the amount we need */
		amt = min(nbp->b_bufsize, (desired_size - bp->b_bufsize));
		pagemove((nbp->b_data + nbp->b_bufsize - amt),
			 bp->b_data + bp->b_bufsize, amt);
		bp->b_bufsize += amt;
		nbp->b_bufsize -= amt;

		/* reduce transfer count if we stole some data */
		if (nbp->b_bcount > nbp->b_bufsize)
			nbp->b_bcount = nbp->b_bufsize;

#ifdef DIAGNOSTIC
		if (nbp->b_bufsize < 0)
			panic("allocbuf: negative bufsize");
#endif

		brelse(nbp);
	}

	/*
	 * If we want a buffer smaller than the current size,
	 * shrink this buffer.  Grab a buf head from the EMPTY queue,
	 * move a page onto it, and put it on front of the AGE queue.
	 * If there are no free buffer headers, leave the buffer alone.
	 */
	if (bp->b_bufsize > desired_size) {
		s = splbio();
		if ((nbp = bufqueues[BQ_EMPTY].tqh_first) == NULL) {
			/* No free buffer head */
			splx(s);
			goto out;
		}
		bremfree(nbp);
		SET(nbp->b_flags, B_BUSY);
		splx(s);

		/* move the page to it and note this change */
		pagemove(bp->b_data + desired_size,
		    nbp->b_data, bp->b_bufsize - desired_size);
		nbp->b_bufsize = bp->b_bufsize - desired_size;
		bp->b_bufsize = desired_size;
		nbp->b_bcount = 0;
		SET(nbp->b_flags, B_INVAL);

		/* release the newly-filled buffer and leave */
		brelse(nbp);
	}
#endif OSKIT

out:
	bp->b_bcount = size;
}

/*
 * Find a buffer which is available for use.
 * Select something from a free list.
 * Preference is to AGE list, then LRU list.    
 */
struct buf *
getnewbuf(slpflag, slptimeo)
	int slpflag, slptimeo;
{
	register struct buf *bp;
	int s;

start:
	s = splbio();
	if ((bp = bufqueues[BQ_AGE].tqh_first) != NULL ||
	    (bp = bufqueues[BQ_LRU].tqh_first) != NULL) {
		bremfree(bp);
	} else {
		/* wait for a free buffer of any kind */
		needbuffer = 1;
		tsleep(&needbuffer, slpflag|(PRIBIO+1), "getnewbuf", slptimeo);
		splx(s);
		return (0);
	}

	/* Buffer is no longer on free lists. */
	SET(bp->b_flags, B_BUSY);

	/* If buffer was a delayed write, start it, and go back to the top. */
	if (ISSET(bp->b_flags, B_DELWRI)) {
		splx(s);
		bawrite (bp);
		goto start;
	}

	/* disassociate us from our vnode, if we had one... */
	if (bp->b_vp)
		brelvp(bp);
	splx(s);

	/* clear out various other fields */
	bp->b_flags = B_BUSY;
	bp->b_dev = NODEV;
	bp->b_blkno = bp->b_lblkno = 0;
	bp->b_iodone = 0;
	bp->b_error = 0;
	bp->b_resid = 0;
	bp->b_bcount = 0;
	bp->b_dirtyoff = bp->b_dirtyend = 0;
	bp->b_validoff = bp->b_validend = 0;

	/* nuke any credentials we were holding */
	if (bp->b_rcred != NOCRED) {
		crfree(bp->b_rcred);
		bp->b_rcred = NOCRED; 
	}
	if (bp->b_wcred != NOCRED) {
		crfree(bp->b_wcred);
		bp->b_wcred = NOCRED;
	}
	
	bremhash(bp);
	return (bp); 
}

/*
 * Wait for operations on the buffer to complete.
 * When they do, extract and return the I/O's error value.
 */
int
biowait(bp)
	struct buf *bp;
{
	int s;

	s = splbio();
	while (!ISSET(bp->b_flags, B_DONE))
		tsleep(bp, PRIBIO + 1, "biowait", 0);
	splx(s);

	/* check for interruption of I/O (e.g. via NFS), then errors. */
	if (ISSET(bp->b_flags, B_EINTR)) {
		CLR(bp->b_flags, B_EINTR);
		return (EINTR);
	} else if (ISSET(bp->b_flags, B_ERROR))
		return (bp->b_error ? bp->b_error : EIO);
	else
		return (0);
}

/*
 * Mark I/O complete on a buffer.
 *
 * If a callback has been requested, e.g. the pageout
 * daemon, do so. Otherwise, awaken waiting processes.
 *
 * [ Leffler, et al., says on p.247:
 *	"This routine wakes up the blocked process, frees the buffer
 *	for an asynchronous write, or, for a request by the pagedaemon
 *	process, invokes a procedure specified in the buffer structure" ]
 *
 * In real life, the pagedaemon (or other system processes) wants
 * to do async stuff to, and doesn't want the buffer brelse()'d.
 * (for swap pager, that puts swap buffers on the free lists (!!!),
 * for the vn device, that puts malloc'd buffers on the free lists!)
 */
void
biodone(bp)
	struct buf *bp;
{
	if (ISSET(bp->b_flags, B_DONE))
		panic("biodone already");
	SET(bp->b_flags, B_DONE);		/* note that it's done */

	if (!ISSET(bp->b_flags, B_READ))	/* wake up reader */
		vwakeup(bp);

	if (ISSET(bp->b_flags, B_CALL)) {	/* if necessary, call out */
		CLR(bp->b_flags, B_CALL);	/* but note callout done */
		(*bp->b_iodone)(bp);
	} else if (ISSET(bp->b_flags, B_ASYNC))	/* if async, release it */
		brelse(bp);
	else {					/* or just wakeup the buffer */
		CLR(bp->b_flags, B_WANTED);
		wakeup(bp);
	}
}

/*
 * Return a count of buffers on the "locked" queue.
 */
int
count_lock_queue()
{
	register struct buf *bp;
	register int n = 0;

	for (bp = bufqueues[BQ_LOCKED].tqh_first; bp;
	    bp = bp->b_freelist.tqe_next)
		n++;
	return (n);
}

#ifdef DEBUG
/*
 * Print out statistics on the current allocation of the buffer pool.
 * Can be enabled to print out on every ``sync'' by setting "syncprt"
 * in vfs_syscalls.c using sysctl.
 */
void
vfs_bufstats()
{
	int s, i, j, count;
	register struct buf *bp;
	register struct bqueues *dp;
	int counts[MAXBSIZE/CLBYTES+1];
	static char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE", "EMPTY" };

	for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) {
		count = 0;
		for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
			counts[j] = 0;
		s = splbio();
		for (bp = dp->tqh_first; bp; bp = bp->b_freelist.tqe_next) {
			counts[bp->b_bufsize/CLBYTES]++;
			count++;
		}
		splx(s);
		printf("%s: total-%d", bname[i], count);
		for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
			if (counts[j] != 0)
				printf(", %d-%d", j * CLBYTES, counts[j]);
		printf("\n");
	}
}
#endif /* DEBUG */