devip.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

ipbread(Chan* ch, long n, ulong offset)
{
	Conv *c;
	Proto *x;
	Fs *f;

	switch(TYPE(ch->qid)){
	case Qdata:
		f = ipfs[ch->dev];
		x = f->p[PROTO(ch->qid)];
		c = x->conv[CONV(ch->qid)];
		return qbread(c->rq, n);
	default:
		return devbread(ch, n, offset);
	}
}

/*
 *  set local address to be that of the ifc closest to remote address
 */
static void
setladdr(Conv* c)
{
	findlocalip(c->p->f, c->laddr, c->raddr);
}

/*
 *  set a local port making sure the quad of raddr,rport,laddr,lport is unique
 */
char*
setluniqueport(Conv* c, int lport)
{
	Proto *p;
	Conv *xp;
	int x;

	p = c->p;

	qlock(p);
	for(x = 0; x < p->nc; x++){
		xp = p->conv[x];
		if(xp == nil)
			break;
		if(xp == c)
			continue;
		if((xp->state == Connected || xp->state == Announced)
		&& xp->lport == lport
		&& xp->rport == c->rport
		&& ipcmp(xp->raddr, c->raddr) == 0
		&& ipcmp(xp->laddr, c->laddr) == 0){
			qunlock(p);
			return "address in use";
		}
	}
	c->lport = lport;
	qunlock(p);
	return nil;
}


/*
 *  pick a local port and set it
 */
void
setlport(Conv* c)
{
	Proto *p;
	ushort *pp;
	int x, found;

	p = c->p;
	if(c->restricted)
		pp = &p->nextrport;
	else
		pp = &p->nextport;
	qlock(p);
	for(;;(*pp)++){
		/*
		 * Fsproto initialises p->nextport to 0 and the restricted
		 * ports (p->nextrport) to 600.
		 * Restricted ports must lie between 600 and 1024.
		 * For the initial condition or if the unrestricted port number
		 * has wrapped round, select a random port between 5000 and 1<<15
		 * to start at.
		 */
		if(c->restricted){
			if(*pp >= 1024)
				*pp = 600;
		}
		else while(*pp < 5000)
			*pp = nrand(1<<15);

		found = 0;
		for(x = 0; x < p->nc; x++){
			if(p->conv[x] == nil)
				break;
			if(p->conv[x]->lport == *pp){
				found = 1;
				break;
			}
		}
		if(!found)
			break;
	}
	c->lport = (*pp)++;
	qunlock(p);
}

/*
 *  set a local address and port from a string of the form
 *	[address!]port[!r]
 */
char*
setladdrport(Conv* c, char* str, int announcing)
{
	char *p;
	char *rv;
	ushort lport;
	uchar addr[IPaddrlen];

	rv = nil;

	/*
	 *  ignore restricted part if it exists.  it's
	 *  meaningless on local ports.
	 */
	p = strchr(str, '!');
	if(p != nil){
		*p++ = 0;
		if(strcmp(p, "r") == 0)
			p = nil;
	}

	c->lport = 0;
	if(p == nil){
		if(announcing)
			ipmove(c->laddr, IPnoaddr);
		else
			setladdr(c);
		p = str;
	} else {
		if(strcmp(str, "*") == 0)
			ipmove(c->laddr, IPnoaddr);
		else {
			parseip(addr, str);
			if(ipforme(c->p->f, addr))
				ipmove(c->laddr, addr);
			else
				return "not a local IP address";
		}
	}

	/* one process can get all connections */
	if(announcing && strcmp(p, "*") == 0){
		if(!iseve())
			error(Eperm);
		return setluniqueport(c, 0);
	}

	lport = atoi(p);
	if(lport <= 0)
		setlport(c);
	else
		rv = setluniqueport(c, lport);
	return rv;
}

static char*
setraddrport(Conv* c, char* str)
{
	char *p;

	p = strchr(str, '!');
	if(p == nil)
		return "malformed address";
	*p++ = 0;
	parseip(c->raddr, str);
	c->rport = atoi(p);
	p = strchr(p, '!');
	if(p){
		if(strstr(p, "!r") != nil)
			c->restricted = 1;
	}
	return nil;
}

/*
 *  called by protocol connect routine to set addresses
 */
char*
Fsstdconnect(Conv *c, char *argv[], int argc)
{
	char *p;

	switch(argc) {
	default:
		return "bad args to connect";
	case 2:
		p = setraddrport(c, argv[1]);
		if(p != nil)
			return p;
		setladdr(c);
		setlport(c);
		break;
	case 3:
		p = setraddrport(c, argv[1]);
		if(p != nil)
			return p;
		p = setladdrport(c, argv[2], 0);
		if(p != nil)
			return p;
	}

	if( (memcmp(c->raddr, v4prefix, IPv4off) == 0 &&
		memcmp(c->laddr, v4prefix, IPv4off) == 0)
		|| ipcmp(c->raddr, IPnoaddr) == 0)
		c->ipversion = V4;
	else
		c->ipversion = V6;

	return nil;
}
/*
 *  initiate connection and sleep till its set up
 */
static int
connected(void* a)
{
	return ((Conv*)a)->state == Connected;
}
static void
connectctlmsg(Proto *x, Conv *c, Cmdbuf *cb)
{
	char *p;

	if(c->state != 0)
		error(Econinuse);
	c->state = Connecting;
	c->cerr[0] = '\0';
	if(x->connect == nil)
		error("connect not supported");
	p = x->connect(c, cb->f, cb->nf);
	if(p != nil)
		error(p);

	qunlock(c);
	if(waserror()){
		qlock(c);
		nexterror();
	}
	sleep(&c->cr, connected, c);
	qlock(c);
	poperror();

	if(c->cerr[0] != '\0')
		error(c->cerr);
}

/*
 *  called by protocol announce routine to set addresses
 */
char*
Fsstdannounce(Conv* c, char* argv[], int argc)
{
	memset(c->raddr, 0, sizeof(c->raddr));
	c->rport = 0;
	switch(argc){
	default:
		break;
	case 2:
		return setladdrport(c, argv[1], 1);
	}
	return "bad args to announce";
}

/*
 *  initiate announcement and sleep till its set up
 */
static int
announced(void* a)
{
	return ((Conv*)a)->state == Announced;
}
static void
announcectlmsg(Proto *x, Conv *c, Cmdbuf *cb)
{
	char *p;

	if(c->state != 0)
		error(Econinuse);
	c->state = Announcing;
	c->cerr[0] = '\0';
	if(x->announce == nil)
		error("announce not supported");
	p = x->announce(c, cb->f, cb->nf);
	if(p != nil)
		error(p);

	qunlock(c);
	if(waserror()){
		qlock(c);
		nexterror();
	}
	sleep(&c->cr, announced, c);
	qlock(c);
	poperror();

	if(c->cerr[0] != '\0')
		error(c->cerr);
}

/*
 *  called by protocol bind routine to set addresses
 */
char*
Fsstdbind(Conv* c, char* argv[], int argc)
{
	switch(argc){
	default:
		break;
	case 2:
		return setladdrport(c, argv[1], 0);
	}
	return "bad args to bind";
}

static void
bindctlmsg(Proto *x, Conv *c, Cmdbuf *cb)
{
	char *p;

	if(x->bind == nil)
		p = Fsstdbind(c, cb->f, cb->nf);
	else
		p = x->bind(c, cb->f, cb->nf);
	if(p != nil)
		error(p);
}

static void
tosctlmsg(Conv *c, Cmdbuf *cb)
{
	if(cb->nf < 2)
		c->tos = 0;
	else
		c->tos = atoi(cb->f[1]);
}

static void
ttlctlmsg(Conv *c, Cmdbuf *cb)
{
	if(cb->nf < 2)
		c->ttl = MAXTTL;
	else
		c->ttl = atoi(cb->f[1]);
}

static long
ipwrite(Chan* ch, void *v, long n, vlong off)
{
	Conv *c;
	Proto *x;
	char *p;
	Cmdbuf *cb;
	uchar ia[IPaddrlen], ma[IPaddrlen];
	Fs *f;
	char *a;
	ulong offset = off;

	a = v;
	f = ipfs[ch->dev];

	switch(TYPE(ch->qid)){
	default:
		error(Eperm);
	case Qdata:
		x = f->p[PROTO(ch->qid)];
		c = x->conv[CONV(ch->qid)];

		if(c->wq == nil)
			error(Eperm);

		qwrite(c->wq, a, n);
		break;
	case Qarp:
		return arpwrite(f, a, n);
	case Qiproute:
		return routewrite(f, ch, a, n);
	case Qlog:
		netlogctl(f, a, n);
		return n;
	case Qndb:
		return ndbwrite(f, a, offset, n);
		break;
	case Qctl:
		x = f->p[PROTO(ch->qid)];
		c = x->conv[CONV(ch->qid)];
		cb = parsecmd(a, n);

		qlock(c);
		if(waserror()) {
			qunlock(c);
			free(cb);
			nexterror();
		}
		if(cb->nf < 1)
			error("short control request");
		if(strcmp(cb->f[0], "connect") == 0)
			connectctlmsg(x, c, cb);
		else if(strcmp(cb->f[0], "announce") == 0)
			announcectlmsg(x, c, cb);
		else if(strcmp(cb->f[0], "bind") == 0)
			bindctlmsg(x, c, cb);
		else if(strcmp(cb->f[0], "ttl") == 0)
			ttlctlmsg(c, cb);
		else if(strcmp(cb->f[0], "tos") == 0)
			tosctlmsg(c, cb);
		else if(strcmp(cb->f[0], "ignoreadvice") == 0)
			c->ignoreadvice = 1;
		else if(strcmp(cb->f[0], "addmulti") == 0){
			if(cb->nf < 2)
				error("addmulti needs interface address");
			if(cb->nf == 2){
				if(!ipismulticast(c->raddr))
					error("addmulti for a non multicast address");
				parseip(ia, cb->f[1]);
				ipifcaddmulti(c, c->raddr, ia);
			} else {
				parseip(ma, cb->f[2]);
				if(!ipismulticast(ma))
					error("addmulti for a non multicast address");
				parseip(ia, cb->f[1]);
				ipifcaddmulti(c, ma, ia);
			}
		} else if(strcmp(cb->f[0], "remmulti") == 0){
			if(cb->nf < 2)
				error("remmulti needs interface address");
			if(!ipismulticast(c->raddr))
				error("remmulti for a non multicast address");
			parseip(ia, cb->f[1]);
			ipifcremmulti(c, c->raddr, ia);
		} else if(x->ctl != nil) {
			p = x->ctl(c, cb->f, cb->nf);
			if(p != nil)
				error(p);
		} else
			error("unknown control request");
		qunlock(c);
		free(cb);
		poperror();
	}
	return n;
}

static long
ipbwrite(Chan* ch, Block* bp, ulong offset)
{
	Conv *c;
	Proto *x;
	Fs *f;
	int n;

	switch(TYPE(ch->qid)){
	case Qdata:
		f = ipfs[ch->dev];
		x = f->p[PROTO(ch->qid)];
		c = x->conv[CONV(ch->qid)];

		if(c->wq == nil)
			error(Eperm);

		if(bp->next)
			bp = concatblock(bp);
		n = BLEN(bp);
		qbwrite(c->wq, bp);
		return n;
	default:
		return devbwrite(ch, bp, offset);
	}
}

Dev ipdevtab = {
	'I',
	"ip",

	ipreset,
	devinit,
	devshutdown,
	ipattach,
	ipwalk,
	ipstat,
	ipopen,
	ipcreate,
	ipclose,
	ipread,
	ipbread,
	ipwrite,
	ipbwrite,
	ipremove,
	ipwstat,
};

int
Fsproto(Fs *f, Proto *p)
{
	if(f->np >= Maxproto)
		return -1;

	p->f = f;

	if(p->ipproto > 0){
		if(f->t2p[p->ipproto] != nil)
			return -1;
		f->t2p[p->ipproto] = p;
	}

	p->qid.type = QTDIR;
	p->qid.path = QID(f->np, 0, Qprotodir);
	p->conv = malloc(sizeof(Conv*)*(p->nc+1));
	if(p->conv == nil)
		panic("Fsproto");

	p->x = f->np;
	p->nextport = 0;
	p->nextrport = 600;
	f->p[f->np++] = p;

	return 0;
}

/*
 *  return true if this protocol is
 *  built in
 */
int
Fsbuiltinproto(Fs* f, uchar proto)
{
	return f->t2p[proto] != nil;
}

/*
 *  called with protocol locked
 */
Conv*
Fsprotoclone(Proto *p, char *user)
{
	Conv *c, **pp, **ep;

retry:
	c = nil;
	ep = &p->conv[p->nc];
	for(pp = p->conv; pp < ep; pp++) {
		c = *pp;
		if(c == nil){
			c = malloc(sizeof(Conv));
			if(c == nil)
				error(Enomem);
			qlock(c);
			c->p = p;
			c->x = pp - p->conv;
			if(p->ptclsize != 0){
				c->ptcl = malloc(p->ptclsize);
				if(c->ptcl == nil) {
					free(c);
					error(Enomem);
				}
			}
			*pp = c;
			p->ac++;
			c->eq = qopen(1024, Qmsg, 0, 0);
			(*p->create)(c);
			break;
		}
		if(canqlock(c)){
			/*
			 *  make sure both processes and protocol
			 *  are done with this Conv
			 */
			if(c->inuse == 0 && (p->inuse == nil || (*p->inuse)(c) == 0))
				break;

			qunlock(c);
		}
	}
	if(pp >= ep) {
		if(p->gc != nil && (*p->gc)(p))
			goto retry;
		return nil;
	}

	c->inuse = 1;
	kstrdup(&c->owner, user);
	c->perm = 0660;
	c->state = Idle;
	ipmove(c->laddr, IPnoaddr);
	ipmove(c->raddr, IPnoaddr);
	c->r = nil;
	c->rgen = 0;
	c->lport = 0;
	c->rport = 0;
	c->restricted = 0;
	c->ttl = MAXTTL;
	qreopen(c->rq);
	qreopen(c->wq);
	qreopen(c->eq);

	qunlock(c);
	return c;
}

int
Fsconnected(Conv* c, char* msg)
{
	if(msg != nil && *msg != '\0')
		strncpy(c->cerr, msg, ERRMAX-1);

	switch(c->state){

	case Announcing:
		c->state = Announced;
		break;

	case Connecting:
		c->state = Connected;
		break;
	}

	wakeup(&c->cr);
	return 0;
}

Proto*
Fsrcvpcol(Fs* f, uchar proto)
{
	if(f->ipmux)
		return f->ipmux;
	else
		return f->t2p[proto];
}

Proto*
Fsrcvpcolx(Fs *f, uchar proto)
{
	return f->t2p[proto];
}

/*
 *  called with protocol locked
 */
Conv*
Fsnewcall(Conv *c, uchar *raddr, ushort rport, uchar *laddr, ushort lport, uchar version)
{
	Conv *nc;
	Conv **l;
	int i;

	qlock(c);
	i = 0;
	for(l = &c->incall; *l; l = &(*l)->next)
		i++;
	if(i >= Maxincall) {
		qunlock(c);
		return nil;
	}

	/* find a free conversation */
	nc = Fsprotoclone(c->p, network);
	if(nc == nil) {
		qunlock(c);
		return nil;
	}
	ipmove(nc->raddr, raddr);
	nc->rport = rport;
	ipmove(nc->laddr, laddr);
	nc->lport = lport;
	nc->next = nil;
	*l = nc;
	nc->state = Connected;
	nc->ipversion = version;

	qunlock(c);

	wakeup(&c->listenr);

	return nc;
}

long
ndbwrite(Fs *f, char *a, ulong off, int n)
{
	if(off > strlen(f->ndb))
		error(Eio);
	if(off+n >= sizeof(f->ndb))
		error(Eio);
	memmove(f->ndb+off, a, n);
	f->ndb[off+n] = 0;
	f->ndbvers++;
	f->ndbmtime = seconds();
	return n;
}

ulong
scalednconv(void)
{
	if(cpuserver && conf.npage*BY2PG >= 128*MB)
		return Nchans*4;
	return Nchans;
}