Re: [RFC 09/12] netlink: Add host-side monitoring for late template interface binding

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 16505 bytes --]

On Sun, Dec 14, 2025 at 08:54:38PM -0500, Jon Maloy wrote:
> When pasta starts without an active template interface (e.g., WiFi
> not yet connected), it falls back to local mode. This change adds
> support for late binding: when the template interface gets an address
> later, pasta detects this via a host-side netlink socket and
> propagates the configuration to the namespace.
> 
> Late binding occurs when:
> - A specific interface is given via -I and it later gets an address

"-i"?   -I is the guest side interface name.

> - No interface is specified, and any interface gets an address.
>   In the latter case the first discovered interface is adopted as
>   template.
> 
> The key changes we make in this commit are:
> - We add a host-side netlink socket (nl_sock_linkaddr_host) to
>   monitor link and address changes on the template interface.
> - We add a nl_linkaddr_host_handler() to process these events
>   and propagate addresses to the namespace.
> - We add support for late binding: when ifi4/ifi6 are unset, we
>   adopt the interface that receives an address - either the one
>   specified via -I, or the first one to get an address if -I was
>   not given.
> - We bring the interface UP after first address is added via late
>   binding.
> - We retain CAP_NET_ADMIN in isolate_prefork() for pasta mode to allow
>   dynamic interface configuration after sandboxing.
> 
> Signed-off-by: Jon Maloy <jmaloy@redhat.com>
> ---
>  epoll_type.h |   4 +-
>  isolation.c  |   4 +
>  netlink.c    | 314 +++++++++++++++++++++++++++++++++++++++++++++++++++
>  netlink.h    |   3 +
>  passt.c      |   4 +
>  5 files changed, 328 insertions(+), 1 deletion(-)
> 
> diff --git a/epoll_type.h b/epoll_type.h
> index 0a16d94..8dc6b8a 100644
> --- a/epoll_type.h
> +++ b/epoll_type.h
> @@ -46,8 +46,10 @@ enum epoll_type {
>  	EPOLL_TYPE_REPAIR,
>  	/* Netlink neighbour subscription socket */
>  	EPOLL_TYPE_NL_NEIGH,
> -	/* Netlink link/address subscription socket */
> +	/* Netlink link/address subscription socket (namespace) */
>  	EPOLL_TYPE_NL_LINKADDR,
> +	/* Netlink link/address subscription socket (host, for template) */
> +	EPOLL_TYPE_NL_LINKADDR_HOST,
>  
>  	EPOLL_NUM_TYPES,
>  };
> diff --git a/isolation.c b/isolation.c
> index b25f349..633c396 100644
> --- a/isolation.c
> +++ b/isolation.c
> @@ -356,6 +356,10 @@ int isolate_prefork(const struct ctx *c)
>  	if (c->mode == MODE_PASTA) {
>  		/* Keep CAP_SYS_ADMIN, so we can enter the netns */
>  		ns_caps |= BIT(CAP_SYS_ADMIN);
> +		/* Keep CAP_NET_ADMIN for dynamic interface configuration
> +		 * (late binding when template interface comes up after start)
> +		 */
> +		ns_caps |= BIT(CAP_NET_ADMIN);
>  		/* Keep CAP_NET_BIND_SERVICE, so we can splice
>  		 * outbound connections to low port numbers
>  		 */
> diff --git a/netlink.c b/netlink.c
> index a8d3116..583ada8 100644
> --- a/netlink.c
> +++ b/netlink.c
> @@ -41,6 +41,9 @@
>  #include "netlink.h"
>  #include "epoll_ctl.h"
>  
> +/* Default namespace interface name from conf.c */
> +extern const char *pasta_default_ifn;
> +
>  /* Same as RTA_NEXT() but for nexthops: RTNH_NEXT() doesn't take 'attrlen' */
>  #define RTNH_NEXT_AND_DEC(rtnh, attrlen)				\
>  	((attrlen) -= RTNH_ALIGN((rtnh)->rtnh_len), RTNH_NEXT(rtnh))
> @@ -63,6 +66,7 @@ int nl_sock		 = -1;
>  int nl_sock_ns		 = -1;
>  static int nl_sock_neigh = -1;
>  static int nl_sock_linkaddr = -1;
> +static int nl_sock_linkaddr_host = -1;
>  static int nl_seq	 = 1;
>  
>  /**
> @@ -249,6 +253,175 @@ static bool nl_addr6_del(struct ctx *c, const struct in6_addr *addr)
>  	return true;
>  }
>  
> +/**
> + * nl_linkaddr_host_msg_read() - Handle host-side link/addr changes
> + * @c:		Execution context
> + * @nh:	Netlink message header
> + *
> + * Monitor template interface changes and propagate to namespace.
> + * Supports late binding: if no template was detected at startup,
> + * adopt the interface specified by -I when it gets an address.
> + */
> +static void nl_linkaddr_host_msg_read(struct ctx *c, const struct nlmsghdr *nh)
> +{
> +	if (nh->nlmsg_type == NLMSG_DONE || nh->nlmsg_type == NLMSG_ERROR)
> +		return;
> +
> +	if (nh->nlmsg_type == RTM_NEWADDR || nh->nlmsg_type == RTM_DELADDR) {
> +		bool is_new = (nh->nlmsg_type == RTM_NEWADDR);
> +		const struct ifaddrmsg *ifa = NLMSG_DATA(nh);
> +		struct rtattr *rta = IFA_RTA(ifa);
> +		size_t na = IFA_PAYLOAD(nh);
> +		bool late_binding = false;
> +		unsigned int template_ifi;
> +		char ifname[IFNAMSIZ];
> +		void *addr = NULL;
> +		bool is_default;
> +		bool is_match;
> +		bool unbound;
> +
> +		/* Get interface name for this message */
> +		if (!if_indextoname(ifa->ifa_index, ifname))
> +			snprintf(ifname, sizeof(ifname), "?");
> +
> +		/* Get template interface index, handling late binding.
> +		 * Late binding occurs when ifi4/ifi6 <= 0 (local mode) and either:
> +		 * - pasta_ifn is set and matches this interface, or
> +		 * - pasta_ifn contains the default name

I think this is trying to infer that we're in local mode from the
interface name.  That seems roundabout and fragile to me, better to
add a flag to struct ctx if we need it.  (If nothing else, using -I
would break this, wouldn't it?).

> +		 */
> +		if (ifa->ifa_family == AF_INET)
> +			template_ifi = c->ifi4;
> +		else if (ifa->ifa_family == AF_INET6)
> +			template_ifi = c->ifi6;
> +		else
> +			return;
> +
> +		/* Check for late binding conditions */
> +		is_default = !strcmp(c->pasta_ifn, pasta_default_ifn);
> +		is_match = !strcmp(ifname, c->pasta_ifn);
> +		unbound = (ifa->ifa_family == AF_INET) ?
> +			  (int)c->ifi4 <= 0 : (int)c->ifi6 <= 0;
> +
> +		if (unbound && (is_default || is_match)) {
> +			debug("Late binding: using %s as %s template", ifname,
> +			      ifa->ifa_family == AF_INET ? "IPv4" : "IPv6");
> +
> +			if (ifa->ifa_family == AF_INET) {
> +				c->ifi4 = ifa->ifa_index;
> +				template_ifi = c->ifi4;
> +			} else {
> +				c->ifi6 = ifa->ifa_index;
> +				template_ifi = c->ifi6;
> +			}
> +			late_binding = true;
> +
> +			if (is_default)
> +				snprintf(c->pasta_ifn, sizeof(c->pasta_ifn),
> +					 "%s", ifname);
> +		}
> +
> +		if (ifa->ifa_index != template_ifi)
> +			return;
> +
> +		/* Re-initialize rta/na for attribute parsing */
> +		rta = IFA_RTA(ifa);
> +		na = IFA_PAYLOAD(nh);
> +
> +		for (; RTA_OK(rta, na); rta = RTA_NEXT(rta, na)) {
> +			if (ifa->ifa_family == AF_INET &&
> +			    rta->rta_type == IFA_LOCAL) {
> +				addr = RTA_DATA(rta);
> +				break;
> +			} else if (ifa->ifa_family == AF_INET6 &&
> +				   rta->rta_type == IFA_ADDRESS) {
> +				addr = RTA_DATA(rta);
> +				break;
> +			}
> +		}
> +
> +		if (!addr) {
> +			info("No addr found in netlink linkaddr message");

Maybe best to check for this *before* touching important state
variables like c->ifi4?

> +			return;
> +		}
> +
> +		if (ifa->ifa_family == AF_INET) {
> +			struct in_addr *a = (struct in_addr *)addr;
> +			char buf[INET_ADDRSTRLEN];
> +			int rc;
> +
> +			inet_ntop(AF_INET, a, buf, sizeof(buf));
> +
> +			if (!is_new) {
> +				nl_addr4_del(c, a);
> +				nl_addr_del(nl_sock_ns, c->pasta_ifi,
> +					    AF_INET, a, ifa->ifa_prefixlen);

We only want to actually poke the guest if c->pasta_conf_ns.

> +				return;
> +			}
> +			rc = nl_addr_set(nl_sock_ns, c->pasta_ifi,
> +					 AF_INET, a,
> +					 ifa->ifa_prefixlen);
> +			if (rc < 0) {
> +				debug("Failed to add %s/%u to ns: %s", buf,
> +				      ifa->ifa_prefixlen, strerror_(-rc));
> +			} else {
> +				nl_addr4_add(c, a, ifa->ifa_prefixlen);

What's nl_addr4_add() and how does it differ from nl_addr_set()?

> +				c->ip4.addr_seen = *a;

This is a host side event, so we shouldn't be updating addr_seen.

> +				debug("Added %s/%u to namespace",
> +				      buf, ifa->ifa_prefixlen);
> +
> +				/* Bring interface UP on late binding */
> +				if (late_binding && !c->pasta_ifi_up) {
> +					nl_link_set_flags(nl_sock_ns,
> +							  c->pasta_ifi,
> +							  IFF_UP, IFF_UP);
> +					c->pasta_ifi_up = 1;
> +					debug("Brought interface up");
> +				}
> +				if (late_binding || c->pasta_ifi_up)

c->pasta_ifi_up must always be true at this point, no?

> +					arp_send_init_req(c);
> +			}
> +		} else if (ifa->ifa_family == AF_INET6) {
> +			struct in6_addr *a = (struct in6_addr *)addr;
> +			char buf[INET6_ADDRSTRLEN];
> +			int rc;
> +
> +			inet_ntop(AF_INET6, a, buf, sizeof(buf));
> +
> +			if (!is_new) {
> +				nl_addr6_del(c, a);
> +				nl_addr_del(nl_sock_ns, c->pasta_ifi,
> +					    AF_INET6, a, ifa->ifa_prefixlen);
> +				return;
> +			}
> +			rc = nl_addr_set(nl_sock_ns, c->pasta_ifi,
> +					 AF_INET6, a, ifa->ifa_prefixlen);
> +			if (rc < 0) {
> +				debug("Failed to add %s/%u to ns: %s",
> +				      buf, ifa->ifa_prefixlen,
> +				      strerror_(-rc));
> +			} else {
> +				nl_addr6_add(c, a, ifa->ifa_prefixlen);
> +				c->ip6.addr_seen = *a;
> +				debug("Added %s/%u to namespace",
> +				      buf, ifa->ifa_prefixlen);
> +
> +				/* Bring interface UP on late binding */
> +				if (late_binding && !c->pasta_ifi_up) {
> +					nl_link_set_flags(nl_sock_ns,
> +							  c->pasta_ifi,
> +							  IFF_UP, IFF_UP);
> +					c->pasta_ifi_up = 1;
> +					debug("Brought interface up");
> +				}
> +				if ((late_binding || c->pasta_ifi_up) &&
> +				    !c->no_ndp)
> +					ndp_send_init_req(c);
> +			}
> +		}
> +		return;
> +	}
> +}
> +
>  /**
>   * nl_linkaddr_msg_read() - Parse and log a netlink link/addr message
>   * @c:		Execution context
> @@ -432,6 +605,36 @@ void nl_linkaddr_notify_handler(struct ctx *c)
>  	}
>  }
>  
> +/**
> + * nl_linkaddr_host_handler() - Handle events from host link/addr notifier
> + * @c:		Execution context
> + *
> + * Monitor template interface changes and propagate to namespace
> + */
> +void nl_linkaddr_host_handler(struct ctx *c)
> +{
> +	char buf[NLBUFSIZ];
> +
> +	for (;;) {
> +		ssize_t n = recv(nl_sock_linkaddr_host, buf, sizeof(buf),
> +				 MSG_DONTWAIT);
> +		struct nlmsghdr *nh = (struct nlmsghdr *)buf;
> +
> +		if (n < 0) {
> +			if (errno == EINTR)
> +				continue;
> +			if (errno != EAGAIN)
> +				info("Host recv() error: %s", strerror_(errno));
> +			break;
> +		}
> +
> +		info("Host netlink: received %zd bytes", n);
> +
> +		for (; NLMSG_OK(nh, n); nh = NLMSG_NEXT(nh, n))
> +			nl_linkaddr_host_msg_read(c, nh);
> +	}
> +}
> +
>  /**
>   * nl_linkaddr_init_do() - Actually create and bind the netlink socket
>   * @arg:	Execution context (for namespace entry) or NULL
> @@ -464,6 +667,38 @@ static int nl_linkaddr_init_do(void *arg)
>  	return 0;
>  }
>  
> +/**
> + * nl_linkaddr_host_init_do() - Create host-side link/addr notifier socket
> + * @arg:	Unused
> + *
> + * Return: 0 on success, -1 on failure
> + */
> +static int nl_linkaddr_host_init_do(void *arg)

Why the void *?  This is host side, so you don't need an NS_CALL().

> +{
> +	struct sockaddr_nl addr = { .nl_family = AF_NETLINK,
> +		.nl_groups = RTMGRP_LINK | RTMGRP_IPV4_IFADDR | RTMGRP_IPV6_IFADDR };
> +
> +	(void)arg;
> +
> +	nl_sock_linkaddr_host = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC,
> +				       NETLINK_ROUTE);
> +	if (nl_sock_linkaddr_host < 0) {
> +		debug("socket() failed for host: %s", strerror_(errno));
> +		return -1;
> +	}
> +
> +	if (bind(nl_sock_linkaddr_host, (struct sockaddr *)&addr,
> +		 sizeof(addr)) < 0) {
> +		debug("bind() failed for host: %s", strerror_(errno));
> +		close(nl_sock_linkaddr_host);
> +		nl_sock_linkaddr_host = -1;
> +		return -1;
> +	}
> +
> +	debug("host socket fd=%d", nl_sock_linkaddr_host);
> +	return 0;
> +}
> +
>  /**
>   * nl_linkaddr_notify_init() - Initialize link/address change notifier
>   * @c:		Execution context
> @@ -502,6 +737,33 @@ int nl_linkaddr_notify_init(const struct ctx *c)
>  		return -1;
>  	}
>  
> +	debug("namespace socket fd=%d", nl_sock_linkaddr);

Looks like it belongs in an earlier patch.  Plus "namespace socket fd"
isn't very specific.

> +
> +	/* In PASTA mode, also create a host-side socket to monitor
> +	 * template interface changes
> +	 */
> +	if (c->mode == MODE_PASTA) {
> +		nl_linkaddr_host_init_do(NULL);

> +
> +		if (nl_sock_linkaddr_host < 0) {
> +			warn("Failed to create host link/addr notifier socket");
> +			/* Non-fatal - continue without host monitoring */
> +		} else {
> +			ref.type = EPOLL_TYPE_NL_LINKADDR_HOST;
> +			ev.data.u64 = ref.u64;
> +			if (epoll_ctl(c->epollfd, EPOLL_CTL_ADD,
> +				      nl_sock_linkaddr_host, &ev) == -1) {
> +				warn("epoll_ctl() failed on host notifier: %s",
> +				     strerror_(errno));
> +				close(nl_sock_linkaddr_host);
> +				nl_sock_linkaddr_host = -1;
> +			} else {
> +				info("Host netlink socket fd=%d, pasta_ifn=%s",
> +				     nl_sock_linkaddr_host, c->pasta_ifn);
> +			}
> +		}
> +	}
> +
>  	return 0;
>  }
>  /**
> @@ -1340,6 +1602,58 @@ int nl_addr_set(int s, unsigned int ifi, sa_family_t af,
>  	return nl_do(s, &req, RTM_NEWADDR, NLM_F_CREATE | NLM_F_EXCL, len);
>  }
>  
> +/**
> + * nl_addr_del() - Delete IP address from given interface
> + * @s:		Netlink socket
> + * @ifi:	Interface index
> + * @af:		Address family
> + * @addr:	Address to delete
> + * @prefix_len:	Prefix length
> + *
> + * Return: 0 on success, negative error code on failure
> + */
> +int nl_addr_del(int s, unsigned int ifi, sa_family_t af,
> +		const void *addr, int prefix_len)
> +{
> +	struct req_t {
> +		struct nlmsghdr nlh;
> +		struct ifaddrmsg ifa;
> +		union {
> +			struct {
> +				struct rtattr rta_l;
> +				struct in_addr l;
> +			} a4;
> +			struct {
> +				struct rtattr rta_l;
> +				struct in6_addr l;
> +			} a6;
> +		} del;
> +	} req = {
> +		.ifa.ifa_family    = af,
> +		.ifa.ifa_index     = ifi,
> +		.ifa.ifa_prefixlen = prefix_len,
> +	};
> +	ssize_t len;
> +
> +	if (af == AF_INET6) {
> +		size_t rta_len = RTA_LENGTH(sizeof(req.del.a6.l));
> +
> +		len = offsetof(struct req_t, del.a6) + sizeof(req.del.a6);
> +		memcpy(&req.del.a6.l, addr, sizeof(req.del.a6.l));
> +		req.del.a6.rta_l.rta_len = rta_len;
> +		req.del.a6.rta_l.rta_type = IFA_LOCAL;
> +	} else {
> +		size_t rta_len = RTA_LENGTH(sizeof(req.del.a4.l));
> +
> +		len = offsetof(struct req_t, del.a4) + sizeof(req.del.a4);
> +		memcpy(&req.del.a4.l, addr, sizeof(req.del.a4.l));
> +		req.del.a4.rta_l.rta_len = rta_len;
> +		req.del.a4.rta_l.rta_type = IFA_LOCAL;
> +	}
> +
> +	return nl_do(s, &req, RTM_DELADDR, 0, len);
> +}
> +
>  /**
>   * nl_addr_dup() - Copy IP addresses for given interface and address family
>   * @s_src:	Netlink socket in source network namespace
> diff --git a/netlink.h b/netlink.h
> index 1796a72..f65ae10 100644
> --- a/netlink.h
> +++ b/netlink.h
> @@ -35,5 +35,8 @@ void nl_neigh_notify_handler(const struct ctx *c);
>  
>  int nl_linkaddr_notify_init(const struct ctx *c);
>  void nl_linkaddr_notify_handler(struct ctx *c);
> +void nl_linkaddr_host_handler(struct ctx *c);
> +int nl_addr_del(int s, unsigned int ifi, sa_family_t af,
> +		const void *addr, int prefix_len);
>  
>  #endif /* NETLINK_H */
> diff --git a/passt.c b/passt.c
> index f274858..438dac8 100644
> --- a/passt.c
> +++ b/passt.c
> @@ -81,6 +81,7 @@ char *epoll_type_str[] = {
>  	[EPOLL_TYPE_REPAIR]		= "TCP_REPAIR helper socket",
>  	[EPOLL_TYPE_NL_NEIGH]		= "netlink neighbour notifier socket",
>  	[EPOLL_TYPE_NL_LINKADDR]	= "netlink link/address notifier socket",
> +	[EPOLL_TYPE_NL_LINKADDR_HOST]	= "netlink host link/address notifier socket",
>  };
>  static_assert(ARRAY_SIZE(epoll_type_str) == EPOLL_NUM_TYPES,
>  	      "epoll_type_str[] doesn't match enum epoll_type");
> @@ -308,6 +309,9 @@ static void passt_worker(void *opaque, int nfds, struct epoll_event *events)
>  		case EPOLL_TYPE_NL_LINKADDR:
>  			nl_linkaddr_notify_handler(c);
>  			break;
> +		case EPOLL_TYPE_NL_LINKADDR_HOST:
> +			nl_linkaddr_host_handler(c);
> +			break;
>  		default:
>  			/* Can't happen */
>  			ASSERT(0);
> -- 
> 2.51.1
> 

-- 
David Gibson (he or they)	| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you, not the other way
				| around.
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]