/*
 * fw, request-filtering half - a per-namespace connection firewall.
 *
 * Serves a mirror of /net and mounts it back over /net.  Almost
 * everything passes straight through; the interesting part is a write
 * of "connect" or "announce" to a protocol ctl file, which is matched
 * against a rule list before it reaches the kernel.  A denial fails
 * the write, and dial(2) hands the text to whoever called it.
 *
 * The real /net needs no second name, and must not have one: any path
 * that still reaches it is a way around this.  lib9p forks the server
 * proc with RFNAMEG (see postsrv in /sys/src/lib9p/post.c), so the
 * server keeps a private copy of the namespace as it was before the
 * mount.  "/net" in here is the real one; "/net" out there is us.
 *
 * None of this holds unless the sandboxed process is also denied #I,
 * or it can bind the IP stack back in and ignore us.  The wrapper does
 * that with a write to /dev/drivers; see fw(8).
 */
#include <u.h>
#include <libc.h>
#include <fcall.h>
#include <thread.h>
#include <9p.h>
#include <bio.h>
#include <ndb.h>
#include <ip.h>
#include "rules.h"

/*
 * One per fid.  fd is the host file behind it, and is closed only when
 * the fid is clunked: for a ctl file that close is what tears down the
 * connection, so the two lifetimes have to be the same one.
 */
typedef struct Fnode Fnode;
struct Fnode
{
	char	*path;			/* relative to the root, "" is the root */
	int	fd;
	char	*dbuf;			/* directory, encoded at open */
	long	dlen;
};

static char	*orig = "/net";

/*
 * The real /net is a union of devip and the cs and dns mounts, whose
 * qids are allocated by different servers and can collide.  Hash the
 * path instead: stable across walks, unique across servers.
 */
static uvlong
hashpath(char *s)
{
	uvlong h;

	h = 14695981039346656037ULL;
	while(*s != '\0'){
		h ^= (uchar)*s++;
		h *= 1099511628211ULL;
	}
	return h;
}

static void
mkqid(Qid *q, char *path, Qid *real)
{
	q->path = hashpath(path);
	q->vers = real->vers;
	q->type = real->type;
}

static char*
realpath(char *path)
{
	if(*path == '\0')
		return estrdup(orig);
	return smprint("%s/%s", orig, path);
}

static char*
childpath(char *dir, char *name)
{
	char *p, *q;

	if(strcmp(name, "..") == 0){
		p = estrdup(dir);
		if((q = strrchr(p, '/')) != nil)
			*q = '\0';
		else
			*p = '\0';
		return p;
	}
	if(*dir == '\0')
		return estrdup(name);
	return smprint("%s/%s", dir, name);
}

/*
 * What is served, named; not what is hidden, listed.
 *
 * Listing what to hide has now been wrong twice.  "trans" installs
 * kernel address translations and devip gates it with iseve(), which is
 * fw's identity here and not the caller's, so a filtered program could
 * write it and reach the whole machine.  "log" is a trace of every
 * connection on the machine, and only its write was refused, when
 * reading it was the leak.  Neither was a hard problem: both were
 * simply missing from a list of things to deny.
 *
 * So it is the other way round.  Nothing is served unless it is named
 * below, and the next file devip grows is invisible until someone
 * decides otherwise - which is the direction a firewall should fail in,
 * and the same argument the rule parser makes about attributes it does
 * not recognise.
 */

/*
 * devip's protocol directories, by name.  Not "any directory with a
 * clone file": devether has one of those too, and #l bound into /net
 * would have become a protocol.  A protocol missing from this list is
 * one nobody can reach, which is the safe way to be out of date.
 * ipmux is left out on purpose - it is raw packet access.
 */
static char *protos[] = {
	"tcp", "udp", "il", "icmp", "icmpv6", "rudp", "gre", "esp",
	"ipifc",		/* read-only: see writable() */
	nil,
};

/* the files a conversation directory has */
static char *convfiles[] = {
	"ctl", "data", "err", "listen", "local", "remote", "status",
	nil,
};

/* of those, the ones a filtered program may write */
static char *rwconvfiles[] = {
	"ctl", "data", "err", "listen",
	nil,
};

/* everything else served from the root, and whether writing it is allowed */
static struct {
	char	*name;
	int	rw;
} rootfiles[] = {
	{ "cs",		1 },	/* asking for a translation is a write */
	{ "dns",	1 },
	{ "arp",	0 },
	{ "bootp",	0 },
	{ "iproute",	0 },
	{ "ipselftab",	0 },
	{ "ndb",	0 },
};

static int
inlist(char **l, char *s)
{
	int i;

	for(i = 0; l[i] != nil; i++)
		if(strcmp(l[i], s) == 0)
			return 1;
	return 0;
}

static int
rootfile(char *name)
{
	int i;

	for(i = 0; i < nelem(rootfiles); i++)
		if(strcmp(rootfiles[i].name, name) == 0)
			return i;
	return -1;
}

static int
isnum(char *s)
{
	if(*s == '\0')
		return 0;
	for(; *s != '\0'; s++)
		if(*s < '0' || *s > '9')
			return 0;
	return 1;
}

/*
 * Split a path into its components, of which a served path has at most
 * three.  A deeper path, an over-long one, or one with an empty
 * component is not a path this server ever handed out, so it is not one
 * it will honour: that also disposes of names like "tcp/../.." arriving
 * as a single walk element from a client speaking 9P directly.
 */
static int
splitpath(char *path, char *buf, int nbuf, char **f)
{
	int i, n;

	if(strlen(path) >= nbuf)
		return -1;
	strcpy(buf, path);
	n = getfields(buf, f, 4, 0, "/");
	if(n < 1 || n > 3)
		return -1;
	for(i = 0; i < n; i++)
		if(*f[i] == '\0')
			return -1;
	return n;
}

/*
 * A protocol this kernel actually has.  The name list says which ones
 * may be served; the clone file says which ones are there.
 */
static int
isproto(char *name)
{
	char *p;
	int ok;

	if(!inlist(protos, name))
		return 0;
	p = smprint("%s/%s/clone", orig, name);
	ok = access(p, AEXIST) == 0;
	free(p);
	return ok;
}

static int
served(char *path)
{
	char buf[128], *f[4];
	int n;

	if(*path == '\0')
		return 1;			/* the root */
	if((n = splitpath(path, buf, sizeof buf, f)) < 0)
		return 0;

	if(!isproto(f[0]))
		return n == 1 && rootfile(f[0]) >= 0;
	if(n == 1)
		return 1;			/* the protocol directory */

	if(isnum(f[1])){
		if(n == 2)
			return 1;		/* a conversation directory */
		/*
		 * An interface's data and snoop files are a wire.  On a
		 * machine also running fw -e one of them is *the* wire, so
		 * either hands a filtered program the packet stream the
		 * filter exists to control.
		 */
		if(strcmp(f[0], "ipifc") == 0)
			if(strcmp(f[2], "data") == 0 || strcmp(f[2], "snoop") == 0)
				return 0;
		return inlist(convfiles, f[2]);
	}
	if(n != 2)
		return 0;
	if(strcmp(f[0], "ipifc") == 0)
		return strcmp(f[1], "stats") == 0;	/* no new interfaces */
	return strcmp(f[1], "clone") == 0 || strcmp(f[1], "stats") == 0;
}

static int
writable(char *path)
{
	char buf[128], *f[4];
	int n, i;

	if((n = splitpath(path, buf, sizeof buf, f)) < 0)
		return 0;
	if(!isproto(f[0])){
		if(n != 1 || (i = rootfile(f[0])) < 0)
			return 0;
		return rootfiles[i].rw;
	}
	if(strcmp(f[0], "ipifc") == 0)
		return 0;		/* the interfaces are not ours to change */
	if(n == 2)
		return strcmp(f[1], "clone") == 0;
	if(n == 3 && isnum(f[1]))
		return inlist(rwconvfiles, f[2]);
	return 0;
}

static char*
protect(char *path, int mode)
{
	if(!served(path))
		return "fw: does not exist";
	if((mode & 3) != OREAD && !writable(path))
		return "fw: read-only under fw";
	return nil;
}

/*
 * If path is a protocol ctl file - "tcp/clone" or "tcp/1/ctl" - return
 * the protocol name.  Opening clone yields an fd that is itself the new
 * connection's ctl file, so both spellings take a connect write.
 */
static char*
ctlproto(char *path)
{
	char buf[64], *p, *q;
	int n;

	if((p = strchr(path, '/')) == nil)
		return nil;
	n = p - path;
	if(n <= 0 || n >= sizeof buf)
		return nil;
	memmove(buf, path, n);
	buf[n] = '\0';
	p++;
	if(strcmp(p, "clone") != 0){
		if((q = strchr(p, '/')) == nil || strcmp(q+1, "ctl") != 0)
			return nil;
	}
	if(!isproto(buf))
		return nil;
	return estrdup(buf);
}

/*
 * Control messages that only change how this one conversation behaves,
 * and so cannot reach the network by themselves.  Everything outside
 * this list is refused: see checkctl.
 */
static char *okverbs[] = {
	"bind",			/* the local address; announce is what opens */
	"ttl",
	"tos",
	"ignoreadvice",
	"close",
	"hangup",
	"keepalive",
	nil,
};

/*
 * If path is a protocol listen file - "tcp/1/listen" - return the
 * protocol name.  Opening one blocks until somebody connects, and the
 * fd it yields is the ctl file of the connection that arrived.
 */
static char*
listenproto(char *path)
{
	char buf[64], *p, *q;
	int n;

	if((p = strchr(path, '/')) == nil)
		return nil;
	n = p - path;
	if(n <= 0 || n >= sizeof buf)
		return nil;
	memmove(buf, path, n);
	buf[n] = '\0';
	if((q = strchr(p+1, '/')) == nil || strcmp(q+1, "listen") != 0)
		return nil;
	if(!isproto(buf))
		return nil;
	return estrdup(buf);
}

/* the address side of "10.9.9.1!1234", read out of local or remote */
static int
convaddr(char *proto, int conv, char *which, uchar *ip, int *port)
{
	char path[256], buf[128], *p;
	int fd, n;

	snprint(path, sizeof path, "%s/%s/%d/%s", orig, proto, conv, which);
	if((fd = open(path, OREAD)) < 0)
		return -1;
	n = read(fd, buf, sizeof buf - 1);
	close(fd);
	if(n <= 0)
		return -1;
	buf[n] = '\0';
	if((p = strchr(buf, '\n')) != nil)
		*p = '\0';
	if((p = strrchr(buf, '!')) == nil)
		return -1;
	*p++ = '\0';
	*port = atoi(p);
	return parseip(ip, buf);
}

/*
 * A listen has returned a connection.  Ask about the peer, now that
 * there is one.
 *
 * The announce could not be matched against a peer - at that moment
 * nobody had called - so any rule naming an address was skipped, and
 * "deny=in ip=..." meant nothing here while meaning something at the
 * packet layer.  A rule that silently does nothing is the failure this
 * program refuses to accept from a mistyped attribute, and it should
 * not accept it from itself.
 *
 * The handshake has already happened: the kernel answered before listen
 * returned, and no filter at this altitude can prevent that.  What it
 * can do is refuse the connection to the program and hang it up, which
 * is the difference between a rule that is late and a rule that is
 * decorative.
 */
static char*
checklisten(Match *m, char *proto, int fd)
{
	uchar ip[IPaddrlen], lip[IPaddrlen];
	char buf[64];
	int n, conv, lport;

	/*
	 * fd is the new conversation's ctl file and its number is what
	 * reading it gives.  At offset 0, so the program's own read - the
	 * one listen(2) makes to learn the same number - still sees it.
	 */
	if((n = pread(fd, buf, sizeof buf - 1, 0)) <= 0)
		return nil;
	buf[n] = '\0';
	conv = atoi(buf);

	memset(m, 0, sizeof *m);
	m->count = 1;
	m->verb = Vin;
	m->proto = proto;
	m->ip = ip;
	m->port = -1;
	m->lport = -1;
	if(convaddr(proto, conv, "remote", ip, &m->port) < 0)
		return nil;			/* gone already; let it be */
	if(convaddr(proto, conv, "local", lip, &lport) >= 0)
		m->lport = lport;	/* the port announced, for an lport rule */

	if(matchrule(m)){
		if(m->log)
			syslog(0, "fw", "allow %s listen %I!%d", proto, ip, m->port);
		return nil;
	}
	syslog(0, "fw", "deny %s listen %I!%d: %s", proto, ip, m->port, m->err);
	fprint(fd, "hangup");
	snprint(buf, sizeof buf, "%s", m->err);
	snprint(m->err, sizeof m->err, "fw: %s", buf);
	return m->err;
}

/*
 * connect takes addr!port with optional trailing fields; announce takes
 * a bare port, or addr!port with addr often "*".
 */
static char*
checkctl(Match *m, char *proto, char *msg, long n)
{
	char buf[512], dest[128], *f[8], *a[4], *addr;
	uchar ip[IPaddrlen], mask[IPaddrlen];
	int nf, na, i;

	memset(m, 0, sizeof *m);
	m->count = 1;
	m->proto = proto;
	m->ip = ip;
	if(n <= 0)
		return nil;
	if(n >= sizeof buf)
		n = sizeof buf - 1;
	memmove(buf, msg, n);
	buf[n] = '\0';

	if((nf = tokenize(buf, f, nelem(f))) < 1)
		return nil;
	if(strcmp(f[0], "connect") == 0)
		m->verb = Vout;
	else if(strcmp(f[0], "announce") == 0)
		m->verb = Vin;
	else{
		/*
		 * Anything else is refused unless it is known to be
		 * harmless.  Letting unknown control messages through was
		 * a hole, not a convenience: "headers" on a udp
		 * conversation turns it into one that carries its own
		 * destination, and the write queue is live from the
		 * moment it is cloned, so three writes send a datagram
		 * anywhere with no connect for a rule to match.  gre has
		 * "raw" and "forward"; rudp and icmpv6 have "headers"
		 * too.
		 *
		 * This is the same argument the rule parser already makes
		 * about attributes it does not recognise, applied to the
		 * place it was not.
		 */
		for(i = 0; okverbs[i] != nil; i++)
			if(strcmp(f[0], okverbs[i]) == 0)
				return nil;
		syslog(0, "fw", "deny %s %s: control message not permitted",
			proto, f[0]);
		snprint(m->err, sizeof m->err,
			"fw: %s: not a permitted control message", f[0]);
		return m->err;
	}
	if(nf < 2)
		return nil;			/* malformed; let the kernel say so */

	/*
	 * Keep the address before splitting it: getfields writes over the
	 * separators, so f[1] afterwards is only what precedes the first
	 * one, and the log said "connect 127.0.0.2" for a connection to a
	 * port it never named.
	 */
	snprint(dest, sizeof dest, "%s", f[1]);
	na = getfields(f[1], a, nelem(a), 0, "!");
	if(na < 1)
		return nil;
	if(na == 1){
		addr = "*";			/* announce 17019 */
		m->port = atoi(a[0]);
	}else{
		addr = a[0];
		m->port = strcmp(a[1], "*") == 0 ? -1 : atoi(a[1]);
	}

	/*
	 * Which end the port and the address belong to, so that a rule
	 * means the same here as it does against a packet.
	 *
	 * connect names the far end: its port is the peer's, and the
	 * local port is whatever the kernel picks, so unknown.
	 *
	 * announce names this end: its port is ours, its address is a
	 * local address to listen on, and the peer is nobody yet - we
	 * find out who connected only at listen time.  A rule naming a
	 * peer therefore cannot apply to an announce, which is right:
	 * at this point there is no peer to name.
	 */
	if(m->verb == Vin){
		m->lport = m->port;
		m->port = -1;
		m->anyip = 1;
	}else{
		m->lport = -1;
		m->anyip = strcmp(addr, "*") == 0;
		if(!m->anyip && parseipandmask(ip, mask, addr, nil) == -1){
			syslog(0, "fw", "deny %s %s %s: unparseable address",
				proto, f[0], dest);
			snprint(m->err, sizeof m->err, "fw: unparseable address");
			return m->err;
		}
	}
	if(!matchrule(m)){
		if(m->log)
			syslog(0, "fw", "deny %s %s %s: %s",
				proto, f[0], dest, m->err);
		snprint(buf, sizeof buf, "fw: %s", m->err);
		snprint(m->err, sizeof m->err, "%s", buf);
		return m->err;
	}
	if(m->log)
		syslog(0, "fw", "allow %s %s %s", proto, f[0], dest);
	return nil;
}

/*
 * A directory is read once at open, filtered, and re-encoded; reads
 * then slice that buffer at entry boundaries.  This gets the offset
 * rules right without a gen function, and an open directory is a
 * snapshot on Plan 9 anyway.
 */
static char*
slurpdir(Fnode *f, char *rp)
{
	char *buf, *cp;
	Dir *d;
	Qid q;
	long sz;
	int fd, i, n, m;

	if((fd = open(rp, OREAD)) < 0)
		return "fw: cannot open directory";
	n = dirreadall(fd, &d);
	close(fd);
	if(n < 0)
		return "fw: cannot read directory";

	buf = nil;
	sz = 0;
	for(i = 0; i < n; i++){
		cp = childpath(f->path, d[i].name);
		if(!served(cp)){
			free(cp);
			continue;
		}
		q = d[i].qid;
		mkqid(&d[i].qid, cp, &q);
		free(cp);
		m = sizeD2M(&d[i]);
		if((buf = realloc(buf, sz + m)) == nil)
			sysfatal("out of memory");
		convD2M(&d[i], (uchar*)buf + sz, m);
		sz += m;
	}
	free(d);
	f->dbuf = buf;
	f->dlen = sz;
	return nil;
}

static void
dirslice(Req *r, Fnode *f)
{
	long o, e, m;

	for(o = 0; o < f->dlen && o != r->ifcall.offset; o += m)
		m = GBIT16((uchar*)f->dbuf + o) + BIT16SZ;
	if(o != r->ifcall.offset || o >= f->dlen){
		r->ofcall.count = 0;
		return;
	}
	for(e = o; e < f->dlen; e += m){
		m = GBIT16((uchar*)f->dbuf + e) + BIT16SZ;
		if(e + m - o > r->ifcall.count)
			break;
	}
	memmove(r->ofcall.data, f->dbuf + o, e - o);
	r->ofcall.count = e - o;
}

static Fnode*
newfnode(char *path)
{
	Fnode *f;

	f = emalloc(sizeof *f);
	f->path = estrdup(path);
	f->fd = -1;
	return f;
}

static void
fsattach(Req *r)
{
	Fnode *f;
	Dir *d;

	if((d = dirstat(orig)) == nil){
		responderror(r);
		return;
	}
	f = newfnode("");
	mkqid(&r->fid->qid, "", &d->qid);
	free(d);
	r->fid->aux = f;
	r->ofcall.qid = r->fid->qid;
	respond(r, nil);
}

static char*
fsclone(Fid *old, Fid *new)
{
	Fnode *f;

	f = old->aux;
	new->aux = newfnode(f->path);
	return nil;
}

static char*
fswalk1(Fid *fid, char *name, Qid *q)
{
	Fnode *f;
	Dir *d;
	char *np, *rp;

	f = fid->aux;
	np = childpath(f->path, name);
	if(!served(np)){
		free(np);
		return "fw: does not exist";
	}
	rp = realpath(np);
	d = dirstat(rp);
	free(rp);
	if(d == nil){
		free(np);
		return "fw: does not exist";
	}
	mkqid(q, np, &d->qid);
	free(d);
	free(f->path);
	f->path = np;
	fid->qid = *q;
	return nil;
}

static void
fsopen(Req *r)
{
	Fnode *f;
	char *rp, *e, *proto;
	int mode;

	f = r->fid->aux;
	mode = r->ifcall.mode;
	if((e = protect(f->path, mode)) != nil){
		respond(r, e);
		return;
	}
	rp = realpath(f->path);
	if(r->fid->qid.type & QTDIR){
		e = slurpdir(f, rp);
		free(rp);
		respond(r, e);
		return;
	}
	/* opening listen blocks until someone connects */
	srvrelease(r->srv);
	f->fd = open(rp, mode & ~ORCLOSE);
	srvacquire(r->srv);
	free(rp);
	if(f->fd < 0){
		responderror(r);
		return;
	}
	if((proto = listenproto(f->path)) != nil){
		Match m;

		e = checklisten(&m, proto, f->fd);
		free(proto);
		if(e != nil){
			close(f->fd);
			f->fd = -1;
			respond(r, e);	/* m outlives the respond, which packs it */
			return;
		}
	}
	respond(r, nil);
}

static void
fsread(Req *r)
{
	Fnode *f;
	long n;

	f = r->fid->aux;
	if(r->fid->qid.type & QTDIR){
		dirslice(r, f);
		respond(r, nil);
		return;
	}
	if(f->fd < 0){
		respond(r, "fw: not open");
		return;
	}
	srvrelease(r->srv);
	n = pread(f->fd, r->ofcall.data, r->ifcall.count, r->ifcall.offset);
	srvacquire(r->srv);
	if(n < 0){
		responderror(r);
		return;
	}
	r->ofcall.count = n;
	respond(r, nil);
}

static void
fswrite(Req *r)
{
	Fnode *f;
	char *proto, *e;
	long n;

	f = r->fid->aux;
	if(f->fd < 0){
		respond(r, "fw: not open");
		return;
	}
	if((proto = ctlproto(f->path)) != nil){
		Match m;

		e = checkctl(&m, proto, r->ifcall.data, r->ifcall.count);
		free(proto);
		if(e != nil){
			respond(r, e);	/* m outlives the respond, which packs it */
			return;
		}
	}
	srvrelease(r->srv);
	n = pwrite(f->fd, r->ifcall.data, r->ifcall.count, r->ifcall.offset);
	srvacquire(r->srv);
	if(n < 0){
		responderror(r);
		return;
	}
	r->ofcall.count = n;
	respond(r, nil);
}

static void
fsstat(Req *r)
{
	Fnode *f;
	Dir *d;
	char *rp, *name;
	Qid q;

	f = r->fid->aux;
	rp = realpath(f->path);
	d = dirstat(rp);
	free(rp);
	if(d == nil){
		responderror(r);
		return;
	}
	q = d->qid;
	mkqid(&d->qid, f->path, &q);
	if((name = strrchr(f->path, '/')) != nil)
		name++;
	else if(*f->path != '\0')
		name = f->path;
	else
		name = "/";
	r->d = *d;
	r->d.name = estrdup(name);
	r->d.uid = estrdup(d->uid);
	r->d.gid = estrdup(d->gid);
	r->d.muid = estrdup(d->muid);
	free(d);
	respond(r, nil);
}

static void
fsdestroyfid(Fid *fid)
{
	Fnode *f;

	if((f = fid->aux) == nil)
		return;
	fid->aux = nil;
	if(f->fd >= 0)
		close(f->fd);
	free(f->dbuf);
	free(f->path);
	free(f);
}

static Srv fs =
{
	.attach = fsattach,
	.clone = fsclone,
	.walk1 = fswalk1,
	.open = fsopen,
	.read = fsread,
	.write = fswrite,
	.stat = fsstat,
	.destroyfid = fsdestroyfid,
};

/*
 * Serve a filtered view of "orig" at "mtpt".  Rules have already been
 * read; this is the request-filtering half of fw, kept in its own file
 * only because it is a different mechanism, not a different program.
 */
void
servenet(char *mtpt, char *srvname, char *net)
{
	orig = net;
	if(access(orig, AEXIST) < 0)
		sysfatal("%s: %r", orig);
	fprint(2, "fw: filtering requests on %s\n", mtpt);
	syslog(0, "fw", "started, filtering requests on %s", mtpt);
	threadpostmountsrv(&fs, srvname, mtpt, MREPL);
}