git: 36b10ac2cd18 - main - rtsock: do not use raw socket code

From: Gleb Smirnoff <glebius_at_FreeBSD.org>
Date: Thu, 11 Aug 2022 16:20:12 UTC
The branch main has been updated by glebius:

URL: https://cgit.FreeBSD.org/src/commit/?id=36b10ac2cd18a535cac20ccf51e3fc6c408671e8

commit 36b10ac2cd18a535cac20ccf51e3fc6c408671e8
Author:     Gleb Smirnoff <glebius@FreeBSD.org>
AuthorDate: 2022-08-11 16:19:36 +0000
Commit:     Gleb Smirnoff <glebius@FreeBSD.org>
CommitDate: 2022-08-11 16:19:36 +0000

    rtsock: do not use raw socket code
    
    This makes routing socket implementation self contained and removes one
    of the last dependencies on the raw socket code and pr_output method.
    
    There are very subtle API visible changes:
    - now routing socket would return EOPNOTSUPP instead of EINVAL on
      syscalls that are not supposed to be called on a routing socket.
    - routing socket buffer sizes are now controlled by net.rtsock
      sysctls instead of net.raw.  The latter were not documented
      anywhere, and even Internet search doesn't find any references
      or discussions related to these sysctls.
    
    Reviewed by:            melifaro
    Differential revision:  https://reviews.freebsd.org/D36122
---
 sys/net/rtsock.c | 230 ++++++++++++++++++++++---------------------------------
 1 file changed, 92 insertions(+), 138 deletions(-)

diff --git a/sys/net/rtsock.c b/sys/net/rtsock.c
index 837491a3821b..c9b521eed9ca 100644
--- a/sys/net/rtsock.c
+++ b/sys/net/rtsock.c
@@ -60,7 +60,6 @@
 #include <net/if_llatbl.h>
 #include <net/if_types.h>
 #include <net/netisr.h>
-#include <net/raw_cb.h>
 #include <net/route.h>
 #include <net/route/route_ctl.h>
 #include <net/route/route_var.h>
@@ -150,7 +149,7 @@ static struct	sockaddr sa_zero   = { sizeof(sa_zero), AF_INET, };
 int	(*carp_get_vhid_p)(struct ifaddr *);
 
 /*
- * Used by rtsock/raw_input callback code to decide whether to filter the update
+ * Used by rtsock callback code to decide whether to filter the update
  * notification to a socket bound to a particular FIB.
  */
 #define	RTS_FILTER_FIB	M_PROTO8
@@ -159,7 +158,14 @@ int	(*carp_get_vhid_p)(struct ifaddr *);
  */
 #define	m_rtsock_family	m_pkthdr.PH_loc.eight[0]
 
+struct rcb {
+	LIST_ENTRY(rcb) list;
+	struct socket	*rcb_socket;
+	sa_family_t	rcb_family;
+};
+
 typedef struct {
+	LIST_HEAD(, rcb)	cblist;
 	int	ip_count;	/* attached w/ AF_INET */
 	int	ip6_count;	/* attached w/ AF_INET6 */
 	int	any_count;	/* total attached */
@@ -198,7 +204,6 @@ static int	sysctl_dumpnhop(struct rtentry *rt, struct nhop_object *nh,
 			uint32_t weight, struct walkarg *w);
 static int	sysctl_iflist(int af, struct walkarg *w);
 static int	sysctl_ifmalist(int af, struct walkarg *w);
-static int	route_output(struct mbuf *m, struct socket *so, ...);
 static void	rt_getmetrics(const struct rtentry *rt,
 			const struct nhop_object *nh, struct rt_metrics *out);
 static void	rt_dispatch(struct mbuf *, sa_family_t);
@@ -267,84 +272,85 @@ VNET_SYSUNINIT(vnet_rts_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
     vnet_rts_uninit, 0);
 #endif
 
-static int
-raw_input_rts_cb(struct mbuf *m, struct sockproto *proto, struct sockaddr *src,
-    struct rawcb *rp)
+static void
+rts_append_data(struct socket *so, struct mbuf *m)
 {
-	int fibnum;
-
-	KASSERT(m != NULL, ("%s: m is NULL", __func__));
-	KASSERT(proto != NULL, ("%s: proto is NULL", __func__));
-	KASSERT(rp != NULL, ("%s: rp is NULL", __func__));
 
-	/* No filtering requested. */
-	if ((m->m_flags & RTS_FILTER_FIB) == 0)
-		return (0);
-
-	/* Check if it is a rts and the fib matches the one of the socket. */
-	fibnum = M_GETFIB(m);
-	if (proto->sp_family != PF_ROUTE ||
-	    rp->rcb_socket == NULL ||
-	    rp->rcb_socket->so_fibnum == fibnum)
-		return (0);
-
-	/* Filtering requested and no match, the socket shall be skipped. */
-	return (1);
+	if (sbappendaddr(&so->so_rcv, &route_src, m, NULL) == 0) {
+		soroverflow(so);
+		m_freem(m);
+	} else
+		sorwakeup(so);
 }
 
 static void
 rts_input(struct mbuf *m)
 {
-	struct sockproto route_proto;
+	struct rcb *rcb;
+	struct socket *last;
 
-	route_proto.sp_family = PF_ROUTE;
-	route_proto.sp_protocol = m->m_rtsock_family;
-
-	raw_input_ext(m, &route_proto, &route_src, raw_input_rts_cb);
-}
-
-/*
- * It really doesn't make any sense at all for this code to share much
- * with raw_usrreq.c, since its functionality is so restricted.  XXX
- */
-static void
-rts_abort(struct socket *so)
-{
+	last = NULL;
+	RTSOCK_LOCK();
+	LIST_FOREACH(rcb, &V_route_cb.cblist, list) {
+		if (rcb->rcb_family != AF_UNSPEC &&
+		    rcb->rcb_family != m->m_rtsock_family)
+			continue;
+		if ((m->m_flags & RTS_FILTER_FIB) &&
+		    M_GETFIB(m) != rcb->rcb_socket->so_fibnum)
+			continue;
+		if (last != NULL) {
+			struct mbuf *n;
 
-	raw_usrreqs.pru_abort(so);
+			n = m_copym(m, 0, M_COPYALL, M_NOWAIT);
+			if (n != NULL)
+				rts_append_data(last, n);
+		}
+		last = rcb->rcb_socket;
+	}
+	if (last != NULL)
+		rts_append_data(last, m);
+	else
+		m_freem(m);
+	RTSOCK_UNLOCK();
 }
 
 static void
 rts_close(struct socket *so)
 {
 
-	raw_usrreqs.pru_close(so);
+	soisdisconnected(so);
 }
 
-/* pru_accept is EOPNOTSUPP */
+static SYSCTL_NODE(_net, OID_AUTO, rtsock, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
+    "Routing socket infrastructure");
+static u_long rts_sendspace = 8192;
+SYSCTL_ULONG(_net_rtsock, OID_AUTO, sendspace, CTLFLAG_RW, &rts_sendspace, 0,
+    "Default routing socket send space");
+static u_long rts_recvspace = 8192;
+SYSCTL_ULONG(_net_rtsock, OID_AUTO, recvspace, CTLFLAG_RW, &rts_recvspace, 0,
+    "Default routing socket receive space");
 
 static int
 rts_attach(struct socket *so, int proto, struct thread *td)
 {
-	struct rawcb *rp;
+	struct rcb *rcb;
 	int error;
 
-	KASSERT(so->so_pcb == NULL, ("rts_attach: so_pcb != NULL"));
+	error = soreserve(so, rts_sendspace, rts_recvspace);
+	if (error)
+		return (error);
 
-	/* XXX */
-	rp = malloc(sizeof *rp, M_PCB, M_WAITOK | M_ZERO);
+	rcb = malloc(sizeof(*rcb), M_PCB, M_WAITOK);
+	rcb->rcb_socket = so;
+	rcb->rcb_family = proto;
 
-	so->so_pcb = (caddr_t)rp;
+	so->so_pcb = rcb;
 	so->so_fibnum = td->td_proc->p_fibnum;
-	error = raw_attach(so, proto);
-	rp = sotorawcb(so);
-	if (error) {
-		so->so_pcb = NULL;
-		free(rp, M_PCB);
-		return error;
-	}
+	so->so_options |= SO_USELOOPBACK;
+
 	RTSOCK_LOCK();
-	switch(rp->rcb_proto.sp_protocol) {
+	LIST_INSERT_HEAD(&V_route_cb.cblist, rcb, list);
+	switch (proto) {
 	case AF_INET:
 		V_route_cb.ip_count++;
 		break;
@@ -355,36 +361,18 @@ rts_attach(struct socket *so, int proto, struct thread *td)
 	V_route_cb.any_count++;
 	RTSOCK_UNLOCK();
 	soisconnected(so);
-	so->so_options |= SO_USELOOPBACK;
-	return 0;
-}
 
-static int
-rts_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
-{
-
-	return (raw_usrreqs.pru_bind(so, nam, td)); /* xxx just EINVAL */
-}
-
-static int
-rts_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
-{
-
-	return (raw_usrreqs.pru_connect(so, nam, td)); /* XXX just EINVAL */
+	return (0);
 }
 
-/* pru_connect2 is EOPNOTSUPP */
-/* pru_control is EOPNOTSUPP */
-
 static void
 rts_detach(struct socket *so)
 {
-	struct rawcb *rp = sotorawcb(so);
-
-	KASSERT(rp != NULL, ("rts_detach: rp == NULL"));
+	struct rcb *rcb = so->so_pcb;
 
 	RTSOCK_LOCK();
-	switch(rp->rcb_proto.sp_protocol) {
+	LIST_REMOVE(rcb, list);
+	switch(rcb->rcb_family) {
 	case AF_INET:
 		V_route_cb.ip_count--;
 		break;
@@ -394,66 +382,18 @@ rts_detach(struct socket *so)
 	}
 	V_route_cb.any_count--;
 	RTSOCK_UNLOCK();
-	raw_usrreqs.pru_detach(so);
+	free(rcb, M_PCB);
+	so->so_pcb = NULL;
 }
 
-static int
-rts_disconnect(struct socket *so)
-{
-
-	return (raw_usrreqs.pru_disconnect(so));
-}
-
-/* pru_listen is EOPNOTSUPP */
-
-static int
-rts_peeraddr(struct socket *so, struct sockaddr **nam)
-{
-
-	return (raw_usrreqs.pru_peeraddr(so, nam));
-}
-
-/* pru_rcvd is EOPNOTSUPP */
-/* pru_rcvoob is EOPNOTSUPP */
-
-static int
-rts_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
-	 struct mbuf *control, struct thread *td)
-{
-
-	return (raw_usrreqs.pru_send(so, flags, m, nam, control, td));
-}
-
-/* pru_sense is null */
-
 static int
 rts_shutdown(struct socket *so)
 {
 
-	return (raw_usrreqs.pru_shutdown(so));
-}
-
-static int
-rts_sockaddr(struct socket *so, struct sockaddr **nam)
-{
-
-	return (raw_usrreqs.pru_sockaddr(so, nam));
+	socantsendmore(so);
+	return (0);
 }
 
-static struct pr_usrreqs route_usrreqs = {
-	.pru_abort =		rts_abort,
-	.pru_attach =		rts_attach,
-	.pru_bind =		rts_bind,
-	.pru_connect =		rts_connect,
-	.pru_detach =		rts_detach,
-	.pru_disconnect =	rts_disconnect,
-	.pru_peeraddr =		rts_peeraddr,
-	.pru_send =		rts_send,
-	.pru_shutdown =		rts_shutdown,
-	.pru_sockaddr =		rts_sockaddr,
-	.pru_close =		rts_close,
-};
-
 #ifndef _SOCKADDR_UNION_DEFINED
 #define	_SOCKADDR_UNION_DEFINED
 /*
@@ -1021,9 +961,9 @@ alloc_sockaddr_aligned(struct linear_buffer *lb, int len)
 }
 #endif
 
-/*ARGSUSED*/
 static int
-route_output(struct mbuf *m, struct socket *so, ...)
+rts_send(struct socket *so, int flags, struct mbuf *m,
+    struct sockaddr *nam, struct mbuf *control, struct thread *td)
 {
 	struct rt_msghdr *rtm = NULL;
 	struct rt_addrinfo info;
@@ -1038,6 +978,13 @@ route_output(struct mbuf *m, struct socket *so, ...)
 	struct rib_cmd_info rc;
 	struct nhop_object *nh;
 
+	if ((flags & PRUS_OOB) || control != NULL) {
+		m_freem(m);
+		if (control != NULL)
+			m_freem(control);
+		return (EOPNOTSUPP);
+	}
+
 	fibnum = so->so_fibnum;
 #define senderr(e) { error = e; goto flush;}
 	if (m == NULL || ((m->m_len < sizeof(long)) &&
@@ -1231,7 +1178,7 @@ static void
 send_rtm_reply(struct socket *so, struct rt_msghdr *rtm, struct mbuf *m,
     sa_family_t saf, u_int fibnum, int rtm_errno)
 {
-	struct rawcb *rp = NULL;
+	struct rcb *rcb = NULL;
 
 	/*
 	 * Check to see if we don't want our own messages.
@@ -1244,7 +1191,7 @@ send_rtm_reply(struct socket *so, struct rt_msghdr *rtm, struct mbuf *m,
 			return;
 		}
 		/* There is another listener, so construct message */
-		rp = sotorawcb(so);
+		rcb = so->so_pcb;
 	}
 
 	if (rtm != NULL) {
@@ -1265,15 +1212,15 @@ send_rtm_reply(struct socket *so, struct rt_msghdr *rtm, struct mbuf *m,
 	if (m != NULL) {
 		M_SETFIB(m, fibnum);
 		m->m_flags |= RTS_FILTER_FIB;
-		if (rp) {
+		if (rcb) {
 			/*
 			 * XXX insure we don't get a copy by
 			 * invalidating our protocol
 			 */
-			unsigned short family = rp->rcb_proto.sp_family;
-			rp->rcb_proto.sp_family = 0;
+			sa_family_t family = rcb->rcb_family;
+			rcb->rcb_family = AF_UNSPEC;
 			rt_dispatch(m, saf);
-			rp->rcb_proto.sp_family = family;
+			rcb->rcb_family = family;
 		} else
 			rt_dispatch(m, saf);
 	}
@@ -2696,13 +2643,20 @@ static SYSCTL_NODE(_net, PF_ROUTE, routetable, CTLFLAG_RD | CTLFLAG_MPSAFE,
 
 static struct domain routedomain;		/* or at least forward */
 
+static struct pr_usrreqs route_usrreqs = {
+	.pru_abort =		rts_close,
+	.pru_attach =		rts_attach,
+	.pru_detach =		rts_detach,
+	.pru_send =		rts_send,
+	.pru_shutdown =		rts_shutdown,
+	.pru_close =		rts_close,
+};
+
 static struct protosw routesw[] = {
 {
 	.pr_type =		SOCK_RAW,
 	.pr_domain =		&routedomain,
 	.pr_flags =		PR_ATOMIC|PR_ADDR,
-	.pr_output =		route_output,
-	.pr_ctlinput =		raw_ctlinput,
 	.pr_usrreqs =		&route_usrreqs
 }
 };