devbridge.c

这是一个同样来自贝尔实验室的和UNIX有着渊源的操作系统, 其简洁的设计和实现易于我们学习和理解

		break;
	case Ttun:
		port->data[0] = namec(dev, Aopen, OREAD, 0);
		port->data[1] = namec(dev2, Aopen, OWRITE, 0);
		break;
	}

	poperror();

	// commited to binding port
	b->port[port->id] = port;
	port->bridge = b;
	if(b->nport <= port->id)
		b->nport = port->id+1;

	// assumes kproc always succeeds
	kproc("etherread", etherread, port);	// poperror must be next
	port->ref++;
}

// assumes b is locked
static void
portunbind(Bridge *b, int argc, char *argv[])
{
	Port *port=nil;
	int type=0, i;
	char *usage = "usage: unbind ether|tunnel addr [ownhash]";
	char name[KNAMELEN];
	ulong ownhash;

	memset(name, 0, KNAMELEN);
	if(argc < 2 || argc > 3)
		error(usage);
	if(strcmp(argv[0], "ether") == 0) {
		type = Tether;
		strncpy(name, argv[1], KNAMELEN);
		name[KNAMELEN-1] = 0;
//		parseaddr(addr, argv[1], Eaddrlen);
	} else if(strcmp(argv[0], "tunnel") == 0) {
		type = Ttun;
		strncpy(name, argv[1], KNAMELEN);
		name[KNAMELEN-1] = 0;
//		parseip(addr, argv[1]);
	} else
		error(usage);
	if(argc == 3)
		ownhash = atoi(argv[2]);
	else
		ownhash = 0;
	for(i=0; i<b->nport; i++) {
		port = b->port[i];
		if(port != nil)
		if(port->type == type)
		if(memcmp(port->name, name, KNAMELEN) == 0)
			break;
	}
	if(i == b->nport)
		error("port not found");
	if(ownhash != 0 && port->ownhash != 0 && ownhash != port->ownhash)
		error("bad owner hash");

	port->closed = 1;
	b->port[i] = nil;	// port is now unbound
	cacheflushport(b, i);

	// try and stop reader
	if(port->readp)
		postnote(port->readp, 1, "unbind", 0);
	portfree(port);
}

// assumes b is locked
static Centry *
cachelookup(Bridge *b, uchar d[Eaddrlen])
{
	int i;
	uint h;
	Centry *p;
	long sec;

	// dont cache multicast or broadcast
	if(d[0] & 1)
		return 0;

	h = 0;
	for(i=0; i<Eaddrlen; i++) {
		h *= 7;
		h += d[i];
	}
	h %= CacheHash;
	p = b->cache + h;
	sec = TK2SEC(m->ticks);
	for(i=0; i<CacheLook; i++,p++) {
		if(memcmp(d, p->d, Eaddrlen) == 0) {
			p->dst++;
			if(sec >= p->expire) {
				log(b, Logcache, "expired cache entry: %E %d\n",
					d, p->port);
				return nil;
			}
			p->expire = sec + CacheTimeout;
			return p;
		}
	}
	log(b, Logcache, "cache miss: %E\n", d);
	return nil;
}

// assumes b is locked
static void
cacheupdate(Bridge *b, uchar d[Eaddrlen], int port)
{
	int i;
	uint h;
	Centry *p, *pp;
	long sec;

	// dont cache multicast or broadcast
	if(d[0] & 1) {
		log(b, Logcache, "bad source address: %E\n", d);
		return;
	}
	
	h = 0;
	for(i=0; i<Eaddrlen; i++) {
		h *= 7;
		h += d[i];
	}
	h %= CacheHash;
	p = b->cache + h;
	pp = p;
	sec = p->expire;

	// look for oldest entry
	for(i=0; i<CacheLook; i++,p++) {
		if(memcmp(p->d, d, Eaddrlen) == 0) {
			p->expire = TK2SEC(m->ticks) + CacheTimeout;
			if(p->port != port) {
				log(b, Logcache, "NIC changed port %d->%d: %E\n",
					p->port, port, d);
				p->port = port;
			}
			p->src++;
			return;
		}
		if(p->expire < sec) {
			sec = p->expire;
			pp = p;
		}
	}
	if(pp->expire != 0)
		log(b, Logcache, "bumping from cache: %E %d\n", pp->d, pp->port);
	pp->expire = TK2SEC(m->ticks) + CacheTimeout;
	memmove(pp->d, d, Eaddrlen);
	pp->port = port;
	pp->src = 1;
	pp->dst = 0;
	log(b, Logcache, "adding to cache: %E %d\n", pp->d, pp->port);
}

// assumes b is locked
static void
cacheflushport(Bridge *b, int port)
{
	Centry *ce;
	int i;

	ce = b->cache;
	for(i=0; i<CacheSize; i++,ce++) {
		if(ce->port != port)
			continue;
		memset(ce, 0, sizeof(Centry));
	}
}

static char *
cachedump(Bridge *b)
{
	int i, n;
	long sec, off;
	char *buf, *p, *ep;
	Centry *ce;
	char c;

	qlock(b);
	if(waserror()) {
		qunlock(b);
		nexterror();
	}
	sec = TK2SEC(m->ticks);
	n = 0;
	for(i=0; i<CacheSize; i++)
		if(b->cache[i].expire != 0)
			n++;
	
	n *= 51;	// change if print format is changed
	n += 10;	// some slop at the end
	buf = malloc(n);
	p = buf;
	ep = buf + n;
	ce = b->cache;
	off = seconds() - sec;
	for(i=0; i<CacheSize; i++,ce++) {
		if(ce->expire == 0)
			continue;	
		c = (sec < ce->expire)?'v':'e';
		p += snprint(p, ep-p, "%E %2d %10ld %10ld %10ld %c\n", ce->d,
			ce->port, ce->src, ce->dst, ce->expire+off, c);
	}
	*p = 0;
	poperror();
	qunlock(b);

	return buf;
}



// assumes b is locked
static void
ethermultiwrite(Bridge *b, Block *bp, Port *port)
{
	Port *oport;
	Block *bp2;
	Etherpkt *ep;
	int i, mcast, bcast;
	static uchar bcastaddr[Eaddrlen] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};

	if(waserror()) {
		if(bp)
			freeb(bp);
		nexterror();
	}
	
	ep = (Etherpkt*)bp->rp;
	mcast = ep->d[0] & 1;
	if(mcast)
		bcast = memcmp(ep->d, bcastaddr, Eaddrlen) == 0;
	else
		bcast = 0;

	oport = nil;
	for(i=0; i<b->nport; i++) {
		if(i == port->id || b->port[i] == nil)
			continue;
		if(mcast && !bcast && !b->port[i]->mcast)
			continue;
		if(mcast)
			b->port[i]->outmulti++;
		else
			b->port[i]->outunknown++;

		// delay one so that the last write does not copy
		if(oport != nil) {
			b->copy++;
			bp2 = copyblock(bp, blocklen(bp));
			if(!waserror()) {
				etherwrite(oport, bp2);
				poperror();
			}
		}
		oport = b->port[i];
	}

	// last write free block
	if(oport) {
		bp2 = bp; bp = nil; USED(bp);
		if(!waserror()) {
			etherwrite(oport, bp2);
			poperror();
		}
	} else
		freeb(bp);

	poperror();
}

static void
tcpmsshack(Etherpkt *epkt, int n)
{
	int hl;
	Iphdr *iphdr;
	Tcphdr *tcphdr;
	ulong mss;
	ulong cksum;
	int optlen;
	uchar *optr;

	// check it is an ip packet
	if(nhgets(epkt->type) != 0x800)
		return;
	iphdr = (Iphdr*)(epkt->data);
	n -= ETHERHDRSIZE;
	if(n < IPHDR)
		return;

	// check it is ok IP packet
	if(iphdr->vihl != (IP_VER|IP_HLEN)) {
		hl = (iphdr->vihl&0xF)<<2;
		if((iphdr->vihl&0xF0) != IP_VER || hl < (IP_HLEN<<2))
			return;
	} else
		hl = IP_HLEN<<2;

	// check TCP
	if(iphdr->proto != IP_TCPPROTO)
		return;
	n -= hl;
	if(n < sizeof(Tcphdr))
		return;
	tcphdr = (Tcphdr*)((uchar*)(iphdr) + hl);
	// MSS can only appear in SYN packet
	if(!(tcphdr->flag[1] & SYN))
		return;
	hl = (tcphdr->flag[0] & 0xf0)>>2;
	if(n < hl)
		return;

	// check for MSS option
	optr = (uchar*)(tcphdr) + sizeof(Tcphdr);
	n = hl - sizeof(Tcphdr);
	for(;;) {
		if(n <= 0 || *optr == EOLOPT)
			return;
		if(*optr == NOOPOPT) {
			n--;
			optr++;
			continue;
		}
		optlen = optr[1];
		if(optlen < 2 || optlen > n)
			return;
		if(*optr == MSSOPT && optlen == MSS_LENGTH)
			break;
		n -= optlen;
		optr += optlen;
	}

	mss = nhgets(optr+2);
	if(mss <= TcpMssMax)
		return;
	// fit checksum
	cksum = nhgets(tcphdr->cksum);
	if(optr-(uchar*)tcphdr & 1) {
print("tcpmsshack: odd alignment!\n");
		// odd alignments are a pain
		cksum += nhgets(optr+1);
		cksum -= (optr[1]<<8)|(TcpMssMax>>8);
		cksum += (cksum>>16);
		cksum &= 0xffff;
		cksum += nhgets(optr+3);
		cksum -= ((TcpMssMax&0xff)<<8)|optr[4];
		cksum += (cksum>>16);
	} else {
		cksum += mss;
		cksum -= TcpMssMax;
		cksum += (cksum>>16);
	}
	hnputs(tcphdr->cksum, cksum);
	hnputs(optr+2, TcpMssMax);
}

/*
 *  process to read from the ethernet
 */
static void
etherread(void *a)
{
	Port *port = a;
	Bridge *b = port->bridge;
	Block *bp, *bp2;
	Etherpkt *ep;
	Centry *ce;
	long md;
	
	qlock(b);
	port->readp = up;	/* hide identity under a rock for unbind */

	while(!port->closed){
		// release lock to read - error means it is time to quit
		qunlock(b);
		if(waserror()) {
print("etherread read error: %s\n", up->errstr);
			qlock(b);
			break;
		}
if(0)print("devbridge: etherread: reading\n");
		bp = devtab[port->data[0]->type]->bread(port->data[0], ETHERMAXTU, 0);
if(0)print("devbridge: etherread: blocklen = %d\n", blocklen(bp));
		poperror();
		qlock(b);
		if(bp == nil || port->closed)
			break;
		if(waserror()) {
//print("etherread bridge error\n");
			if(bp)
				freeb(bp);
			continue;
		}
		if(blocklen(bp) < ETHERMINTU)
			error("short packet");
		port->in++;

		ep = (Etherpkt*)bp->rp;
		cacheupdate(b, ep->s, port->id);
		if(b->tcpmss)
			tcpmsshack(ep, BLEN(bp));

		/*
		 * delay packets to simulate a slow link
		 */
		if(b->delay0 || b->delayn){
			md = b->delay0 + b->delayn * BLEN(bp);
			if(md > 0)
				microdelay(md);
		}

		if(ep->d[0] & 1) {
			log(b, Logmcast, "mulitcast: port=%d src=%E dst=%E type=%#.4ux\n",
				port->id, ep->s, ep->d, (ep->type[0]<<8)|ep->type[1] );
			port->inmulti++;
			bp2 = bp; bp = nil;
			ethermultiwrite(b, bp2, port);
		} else {
			ce = cachelookup(b, ep->d);
			if(ce == nil) {
				b->miss++;
				port->inunknown++;
				bp2 = bp; bp = nil;
				ethermultiwrite(b, bp2, port);
			}else if(ce->port != port->id){
				b->hit++;
				bp2 = bp; bp = nil;
				etherwrite(b->port[ce->port], bp2);
			}
		}

		poperror();
		if(bp)
			freeb(bp);
	}
//print("etherread: trying to exit\n");
	port->readp = nil;
	portfree(port);
	qunlock(b);
	pexit("hangup", 1);
}

static int
fragment(Etherpkt *epkt, int n)
{
	Iphdr *iphdr;

	if(n <= TunnelMtu)
		return 0;

	// check it is an ip packet
	if(nhgets(epkt->type) != 0x800)
		return 0;
	iphdr = (Iphdr*)(epkt->data);
	n -= ETHERHDRSIZE;
	if(n < IPHDR)
		return 0;

	// check it is ok IP packet - I don't handle IP options for the momment
	if(iphdr->vihl != (IP_VER|IP_HLEN))
		return 0;

	// check for don't fragment
	if(iphdr->frag[0] & (IP_DF>>8))
		return 0;

	// check for short block
	if(nhgets(iphdr->length) > n)
		return 0;

	return 1;
}


static void
etherwrite(Port *port, Block *bp)
{
	Iphdr *eh, *feh;
	Etherpkt *epkt;
	int n, lid, len, seglen, chunk, dlen, blklen, offset, mf;
	Block *xp, *nb;
	ushort fragoff, frag;

	port->out++;
	epkt = (Etherpkt*)bp->rp;
	n = blocklen(bp);
	if(port->type != Ttun || !fragment(epkt, n)) {
		devtab[port->data[1]->type]->bwrite(port->data[1], bp, 0);
		return;
	}
	port->outfrag++;
	if(waserror()){
		freeblist(bp);	
		nexterror();
	}

	seglen = (TunnelMtu - ETHERHDRSIZE - IPHDR) & ~7;
	eh = (Iphdr*)(epkt->data);
	len = nhgets(eh->length);
	frag = nhgets(eh->frag);
	mf = frag & IP_MF;
	frag <<= 3;
	dlen = len - IPHDR;
	xp = bp;
	lid = nhgets(eh->id);
	offset = ETHERHDRSIZE+IPHDR;
	while(xp != nil && offset && offset >= BLEN(xp)) {
		offset -= BLEN(xp);
		xp = xp->next;
	}
	xp->rp += offset;
	
if(0) print("seglen=%d, dlen=%d, mf=%x, frag=%d\n", seglen, dlen, mf, frag);
	for(fragoff = 0; fragoff < dlen; fragoff += seglen) {
		nb = allocb(ETHERHDRSIZE+IPHDR+seglen);
		
		feh = (Iphdr*)(nb->wp+ETHERHDRSIZE);

		memmove(nb->wp, epkt, ETHERHDRSIZE+IPHDR);
		nb->wp += ETHERHDRSIZE+IPHDR;

		if((fragoff + seglen) >= dlen) {
			seglen = dlen - fragoff;
			hnputs(feh->frag, (frag+fragoff)>>3 | mf);
		}
		else	
			hnputs(feh->frag, (frag+fragoff>>3) | IP_MF);

		hnputs(feh->length, seglen + IPHDR);
		hnputs(feh->id, lid);

		/* Copy up the data area */
		chunk = seglen;
		while(chunk) {
			blklen = chunk;
			if(BLEN(xp) < chunk)
				blklen = BLEN(xp);
			memmove(nb->wp, xp->rp, blklen);
			nb->wp += blklen;
			xp->rp += blklen;
			chunk -= blklen;
			if(xp->rp == xp->wp)
				xp = xp->next;
		} 

		feh->cksum[0] = 0;
		feh->cksum[1] = 0;
		hnputs(feh->cksum, ipcsum(&feh->vihl));
	
		// don't generate small packets
		if(BLEN(nb) < ETHERMINTU)
			nb->wp = nb->rp + ETHERMINTU;
		devtab[port->data[1]->type]->bwrite(port->data[1], nb, 0);
	}
	poperror();
	freeblist(bp);	
}

// hold b lock
static void
portfree(Port *port)
{
	port->ref--;
	if(port->ref < 0)
		panic("portfree: bad ref");
	if(port->ref > 0)
		return;

	if(port->data[0])
		cclose(port->data[0]);
	if(port->data[1])
		cclose(port->data[1]);
	memset(port, 0, sizeof(Port));
	free(port);
}

Dev bridgedevtab = {
	'B',
	"bridge",

	devreset,
	bridgeinit,
	devshutdown,
	bridgeattach,
	bridgewalk,
	bridgestat,
	bridgeopen,
	devcreate,
	bridgeclose,
	bridgeread,
	devbread,
	bridgewrite,
	devbwrite,
	devremove,
	devwstat,
};