linux: store Linux Ethernet interface number in struct ifnet

The old approach where we go through the list of interfaces and count them
has bugs.  One obvious bug with this dynamic translation is that once an
Ethernet interface in the middle of the list goes away, all interfaces
following it would change their Linux names.

A bigger problem is the ifnet arrival and departure times.  For example
linsysfs has event handler for ifnet_arrival_event, and of course it wants
to resolve the name.  This accidentially works, due to a bug in
if_attach() where we call if_link_ifnet() before invoking all the event
handlers.  Once the bug is fixed linsysfs won't be able to resolve the old
way.  The other side is ifnet_departure_event, where there is no bug, the
eventhandlers are called after the if_unlink_ifnet().  This means old
translation won't work for departure event handlers.  One example is
netlink.  This change gives the Netlink a chance to emit a proper Linux
interface departure message.

However, there is another problem in Netlink, that the ifnet pointer is
lost in the Netlink translation layer.  Plug this with a cookie in netlink
writer structure that can be set by the route layer and used by the Netlink
Linux translation layer.  This part of the diff seems unrelated, but it is
hard to make it a separate change, as the old KPI goes away and to use the
new one we need the pointer.

Differential Revision:	https://reviews.freebsd.org/D54077

Author: glebius

sys/compat/linsysfs/linsysfs_net.c

@@ -90,7 +90,7 @@ linsysfs_if_addr(PFS_FILL_ARGS)
 
 	CURVNET_SET(TD_TO_VNET(td));
 	NET_EPOCH_ENTER(et);
-	ifp = ifname_linux_to_ifp(td, pn->pn_parent->pn_name);
+	ifp = ifname_linux_to_ifp(pn->pn_parent->pn_name);
 	if (ifp != NULL && (error = linux_ifhwaddr(ifp, &lsa)) == 0)
 		error = sbuf_printf(sb, "%02hhx:%02hhx:%02hhx:%02hhx:%02hhx:%02hhx\n",
 		    lsa.sa_data[0], lsa.sa_data[1], lsa.sa_data[2],
@@ -119,7 +119,7 @@ linsysfs_if_flags(PFS_FILL_ARGS)
 
 	CURVNET_SET(TD_TO_VNET(td));
 	NET_EPOCH_ENTER(et);
-	ifp = ifname_linux_to_ifp(td, pn->pn_parent->pn_name);
+	ifp = ifname_linux_to_ifp(pn->pn_parent->pn_name);
 	if (ifp != NULL)
 		error = sbuf_printf(sb, "0x%x\n", linux_ifflags(ifp));
 	else
@@ -138,7 +138,7 @@ linsysfs_if_ifindex(PFS_FILL_ARGS)
 
 	CURVNET_SET(TD_TO_VNET(td));
 	NET_EPOCH_ENTER(et);
-	ifp = ifname_linux_to_ifp(td, pn->pn_parent->pn_name);
+	ifp = ifname_linux_to_ifp(pn->pn_parent->pn_name);
 	if (ifp != NULL)
 		error = sbuf_printf(sb, "%u\n", if_getindex(ifp));
 	else
@@ -157,7 +157,7 @@ linsysfs_if_mtu(PFS_FILL_ARGS)
 
 	CURVNET_SET(TD_TO_VNET(td));
 	NET_EPOCH_ENTER(et);
-	ifp = ifname_linux_to_ifp(td, pn->pn_parent->pn_name);
+	ifp = ifname_linux_to_ifp( pn->pn_parent->pn_name);
 	if (ifp != NULL)
 		error = sbuf_printf(sb, "%u\n", if_getmtu(ifp));
 	else
@@ -186,7 +186,7 @@ linsysfs_if_type(PFS_FILL_ARGS)
 
 	CURVNET_SET(TD_TO_VNET(td));
 	NET_EPOCH_ENTER(et);
-	ifp = ifname_linux_to_ifp(td, pn->pn_parent->pn_name);
+	ifp = ifname_linux_to_ifp(pn->pn_parent->pn_name);
 	if (ifp != NULL && (error = linux_ifhwaddr(ifp, &lsa)) == 0)
 		error = sbuf_printf(sb, "%d\n", lsa.sa_family);
 	else
@@ -207,7 +207,7 @@ linsysfs_if_visible(PFS_VIS_ARGS)
 	visible = 0;
 	CURVNET_SET(TD_TO_VNET(td));
 	NET_EPOCH_ENTER(et);
-	ifp = ifname_linux_to_ifp(td, pn->pn_name);
+	ifp = ifname_linux_to_ifp(pn->pn_name);
 	if (ifp != NULL) {
 		TAILQ_FOREACH_SAFE(nq, &ifp_nodes_q, ifp_nodes_next, nq_tmp) {
 			if (nq->ifp == ifp && nq->vnet == curvnet) {

sys/compat/linux/linux.h

@@ -359,6 +359,8 @@ struct l_statx {
 	ktrstruct("l_sigset_t", (s), l)
 #endif
 
+void linux_ifnet_init(void);
+void linux_ifnet_uninit(void);
 void linux_netlink_register(void);
 void linux_netlink_deregister(void);
 

sys/compat/linux/linux_common.c

@@ -56,13 +56,15 @@ linux_common_modevent(module_t mod, int type, void *data)
 		linux_osd_jail_register();
 		SET_FOREACH(ldhp, linux_device_handler_set)
 			linux_device_register_handler(*ldhp);
+		linux_ifnet_init();
 		linux_netlink_register();
 		break;
 	case MOD_UNLOAD:
 		linux_dev_shm_destroy();
 		linux_osd_jail_deregister();
 		SET_FOREACH(ldhp, linux_device_handler_set)
 			linux_device_unregister_handler(*ldhp);
+		linux_ifnet_uninit();
 		linux_netlink_deregister();
 		break;
 	default:

sys/compat/linux/linux_common.h

@@ -28,10 +28,9 @@
 #ifndef _LINUX_COMMON_H_
 #define _LINUX_COMMON_H_
 
-int	ifname_bsd_to_linux_ifp(struct ifnet *, char *, size_t);
+int	ifname_bsd_to_linux_ifp(const struct ifnet *, char *, size_t);
 int	ifname_bsd_to_linux_idx(u_int, char *, size_t);
-int	ifname_bsd_to_linux_name(const char *, char *, size_t);
-struct ifnet *ifname_linux_to_ifp(struct thread *, const char *);
+struct ifnet *ifname_linux_to_ifp( const char *);
 int	ifname_linux_to_bsd(struct thread *, const char *, char *);
 
 unsigned short	linux_ifflags(struct ifnet *);

sys/compat/linux/linux_if.c

@@ -24,14 +24,18 @@
  */
 
 #include <sys/param.h>
+#include <sys/systm.h>
 #include <sys/ctype.h>
+#include <sys/eventhandler.h>
 #include <sys/jail.h>
 #include <sys/socket.h>
 #include <sys/sysctl.h>
 #include <net/if.h>
 #include <net/if_dl.h>
 #include <net/if_types.h>
 #include <net/if_var.h>
+#include <net/if_private.h>
+#include <net/vnet.h>
 
 #include <compat/linux/linux.h>
 #include <compat/linux/linux_common.h>
@@ -44,33 +48,67 @@ SYSCTL_BOOL(_compat_linux, OID_AUTO, use_real_ifnames, CTLFLAG_RWTUN,
     &use_real_ifnames, 0,
     "Use FreeBSD interface names instead of generating ethN aliases");
 
-/*
- * Criteria for interface name translation
- */
-#define	IFP_IS_ETH(ifp)		(if_gettype(ifp) == IFT_ETHER)
-#define	IFP_IS_LOOP(ifp)	(if_gettype(ifp) == IFT_LOOP)
+VNET_DEFINE_STATIC(struct unrhdr *, linux_eth_unr);
+#define	V_linux_eth_unr	VNET(linux_eth_unr)
 
-/*
- * Translate a FreeBSD interface name to a Linux interface name
- * by interface name, and return the number of bytes copied to lxname.
- */
-int
-ifname_bsd_to_linux_name(const char *bsdname, char *lxname, size_t len)
+static eventhandler_tag ifnet_arrival_tag;
+static eventhandler_tag ifnet_departure_tag;
+
+static void
+linux_ifnet_arrival(void *arg __unused, struct ifnet *ifp)
 {
-	struct epoch_tracker et;
-	struct ifnet *ifp;
-	int ret;
+	if (ifp->if_type == IFT_ETHER)
+		ifp->if_linux_ethno = alloc_unr(V_linux_eth_unr);
+}
 
-	CURVNET_ASSERT_SET();
+static void
+linux_ifnet_departure(void *arg __unused, struct ifnet *ifp)
+{
+	if (ifp->if_type == IFT_ETHER)
+		free_unr(V_linux_eth_unr, ifp->if_linux_ethno);
+}
 
-	ret = 0;
+void
+linux_ifnet_init(void)
+{
+	ifnet_arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
+	    linux_ifnet_arrival, NULL, EVENTHANDLER_PRI_FIRST);
+	ifnet_departure_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
+	    linux_ifnet_departure, NULL, EVENTHANDLER_PRI_LAST);
+}
+
+void
+linux_ifnet_uninit(void)
+{
+	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, ifnet_arrival_tag);
+	EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifnet_departure_tag);
+}
+
+static void
+linux_ifnet_vnet_init(void *arg __unused)
+{
+	struct epoch_tracker et;
+	struct if_iter it;
+	if_t ifp;
+
+	V_linux_eth_unr = new_unrhdr(0, INT_MAX, NULL);
 	NET_EPOCH_ENTER(et);
-	ifp = ifunit(bsdname);
-	if (ifp != NULL)
-		ret = ifname_bsd_to_linux_ifp(ifp, lxname, len);
+	for (ifp = if_iter_start(&it); ifp != NULL; ifp = if_iter_next(&it))
+		linux_ifnet_arrival(NULL, ifp);
 	NET_EPOCH_EXIT(et);
-	return (ret);
 }
+VNET_SYSINIT(linux_ifnet_vnet_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
+    linux_ifnet_vnet_init, NULL);
+
+#ifdef VIMAGE
+static void
+linux_ifnet_vnet_uninit(void *arg __unused)
+{
+	delete_unrhdr(V_linux_eth_unr);
+}
+VNET_SYSUNINIT(linux_ifnet_vnet_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
+    linux_ifnet_vnet_uninit, NULL);
+#endif
 
 /*
  * Translate a FreeBSD interface name to a Linux interface name
@@ -99,50 +137,23 @@ ifname_bsd_to_linux_idx(u_int idx, char *lxname, size_t len)
  * and return the number of bytes copied to lxname, 0 if interface
  * not found, -1 on error.
  */
-struct ifname_bsd_to_linux_ifp_cb_s {
-	struct ifnet	*ifp;
-	int		ethno;
-	char		*lxname;
-	size_t		len;
-};
-
-static int
-ifname_bsd_to_linux_ifp_cb(if_t ifp, void *arg)
-{
-	struct ifname_bsd_to_linux_ifp_cb_s *cbs = arg;
-
-	if (ifp == cbs->ifp)
-		return (snprintf(cbs->lxname, cbs->len, "eth%d", cbs->ethno));
-	if (IFP_IS_ETH(ifp))
-		cbs->ethno++;
-	return (0);
-}
-
 int
-ifname_bsd_to_linux_ifp(struct ifnet *ifp, char *lxname, size_t len)
+ifname_bsd_to_linux_ifp(const struct ifnet *ifp, char *lxname, size_t len)
 {
-	struct ifname_bsd_to_linux_ifp_cb_s arg = {
-		.ifp = ifp,
-		.ethno = 0,
-		.lxname = lxname,
-		.len = len,
-	};
-
-	NET_EPOCH_ASSERT();
-
 	/*
 	 * Linux loopback interface name is lo (not lo0),
 	 * we translate lo to lo0, loX to loX.
 	 */
-	if (IFP_IS_LOOP(ifp) && strncmp(if_name(ifp), "lo0", IFNAMSIZ) == 0)
+	if (ifp->if_type == IFT_LOOP &&
+	    strncmp(ifp->if_xname, "lo0", IFNAMSIZ) == 0)
 		return (strlcpy(lxname, "lo", len));
 
 	/* Short-circuit non ethernet interfaces. */
-	if (!IFP_IS_ETH(ifp) || use_real_ifnames)
-		return (strlcpy(lxname, if_name(ifp), len));
+	if (ifp->if_type != IFT_ETHER || use_real_ifnames)
+		return (strlcpy(lxname, ifp->if_xname, len));
 
 	/* Determine the (relative) unit number for ethernet interfaces. */
-	return (if_foreach(ifname_bsd_to_linux_ifp_cb, &arg));
+	return (snprintf(lxname, len, "eth%d", ifp->if_linux_ethno));
 }
 
 /*
@@ -154,7 +165,6 @@ ifname_bsd_to_linux_ifp(struct ifnet *ifp, char *lxname, size_t len)
 struct ifname_linux_to_ifp_cb_s {
 	bool		is_lo;
 	bool		is_eth;
-	int		ethno;
 	int		unit;
 	const char	*lxname;
 	if_t		ifp;
@@ -174,12 +184,11 @@ ifname_linux_to_ifp_cb(if_t ifp, void *arg)
 	 */
 	if (strncmp(if_name(ifp), cbs->lxname, LINUX_IFNAMSIZ) == 0)
 		goto out;
-	if (cbs->is_eth && IFP_IS_ETH(ifp) && cbs->unit == cbs->ethno)
+	if (cbs->is_eth && ifp->if_type == IFT_ETHER &&
+	    ifp->if_linux_ethno == cbs->unit)
 		goto out;
-	if (cbs->is_lo && IFP_IS_LOOP(ifp))
+	if (cbs->is_lo && ifp->if_type == IFT_LOOP)
 		goto out;
-	if (IFP_IS_ETH(ifp))
-		cbs->ethno++;
 	return (0);
 
 out:
@@ -188,12 +197,10 @@ ifname_linux_to_ifp_cb(if_t ifp, void *arg)
 }
 
 struct ifnet *
-ifname_linux_to_ifp(struct thread *td, const char *lxname)
+ifname_linux_to_ifp(const char *lxname)
 {
 	struct ifname_linux_to_ifp_cb_s arg = {
-		.ethno = 0,
 		.lxname = lxname,
-		.ifp = NULL,
 	};
 	int len;
 	char *ep;
@@ -228,7 +235,7 @@ ifname_linux_to_bsd(struct thread *td, const char *lxname, char *bsdname)
 
 	CURVNET_SET(TD_TO_VNET(td));
 	NET_EPOCH_ENTER(et);
-	ifp = ifname_linux_to_ifp(td, lxname);
+	ifp = ifname_linux_to_ifp(lxname);
 	if (ifp != NULL && bsdname != NULL)
 		strlcpy(bsdname, if_name(ifp), IFNAMSIZ);
 	NET_EPOCH_EXIT(et);
@@ -297,12 +304,12 @@ linux_ifhwaddr(struct ifnet *ifp, struct l_sockaddr *lsa)
 
 	NET_EPOCH_ASSERT();
 
-	if (IFP_IS_LOOP(ifp)) {
+	if (ifp->if_type == IFT_LOOP) {
 		bzero(lsa, sizeof(*lsa));
 		lsa->sa_family = LINUX_ARPHRD_LOOPBACK;
 		return (0);
 	}
-	if (!IFP_IS_ETH(ifp))
+	if (ifp->if_type != IFT_ETHER)
 		return (ENOENT);
 	if (if_foreach_addr_type(ifp, AF_LINK, linux_ifhwaddr_cb, lsa) > 0)
 		return (0);

sys/compat/linux/linux_netlink.c

@@ -249,9 +249,9 @@ nlmsg_translate_ifname_nla(struct nlattr *nla, struct nl_writer *nw)
 {
 	char ifname[LINUX_IFNAMSIZ];
 
-	if (ifname_bsd_to_linux_name((char *)(nla + 1), ifname,
-	    sizeof(ifname)) <= 0)
+	if (nw->ifp == NULL)
 		return (false);
+	(void)ifname_bsd_to_linux_ifp(nw->ifp, ifname, sizeof(ifname));
 	return (nlattr_add_string(nw, IFLA_IFNAME, ifname));
 }
 
@@ -564,7 +564,7 @@ nlmsg_to_linux(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw)
 }
 
 static struct nl_buf *
-nlmsgs_to_linux(struct nl_buf *orig, struct nlpcb *nlp)
+nlmsgs_to_linux(struct nl_buf *orig, struct nlpcb *nlp, const struct ifnet *ifp)
 {
 	struct nl_writer nw;
 	u_int offset, msglen;
@@ -573,6 +573,7 @@ nlmsgs_to_linux(struct nl_buf *orig, struct nlpcb *nlp)
 	    orig->datalen + SCRATCH_BUFFER_SIZE, nlp, false)))
 		return (NULL);
 
+	nw.ifp = ifp;
 	/* Assume correct headers. Buffer IS mutable */
 	for (offset = 0;
 	    offset + sizeof(struct nlmsghdr) <= orig->datalen;

sys/net/if_private.h

@@ -65,6 +65,7 @@ struct ifnet {
 	void	*if_linkmib;		/* link-type-specific MIB data */
 	size_t	if_linkmiblen;		/* length of above data */
 	u_int	if_refcount;		/* reference count */
+	u_int	if_linux_ethno;		/* linux name id for IFT_ETHER */
 
 	/* These fields are shared with struct if_data. */
 	uint8_t		if_type;	/* ethernet, tokenring, etc */

sys/netlink/netlink_io.c

@@ -217,7 +217,7 @@ nl_send(struct nl_writer *nw, struct nlpcb *nlp)
 	}
 
 	if (nlp->nl_linux && linux_netlink_p != NULL) {
-		nb = linux_netlink_p->msgs_to_linux(nw->buf, nlp);
+		nb = linux_netlink_p->msgs_to_linux(nw->buf, nlp, nw->ifp);
 		nl_buf_free(nw->buf);
 		nw->buf = NULL;
 		if (nb == NULL)

sys/netlink/netlink_linux.h

@@ -37,7 +37,8 @@ struct nlpcb;
 struct nl_pstate;
 struct nl_writer;
 
-typedef struct nl_buf * msgs_to_linux_cb_t(struct nl_buf *, struct nlpcb *);
+typedef struct nl_buf * msgs_to_linux_cb_t(struct nl_buf *, struct nlpcb *,
+			    const struct ifnet *);
 typedef int msg_from_linux_cb_t(int netlink_family, struct nlmsghdr **hdr,
     struct nl_pstate *npt);
 

sys/netlink/netlink_message_writer.h

@@ -39,6 +39,7 @@
 
 struct nl_buf;
 struct nl_writer;
+struct ifnet;
 typedef bool nl_writer_cb(struct nl_writer *nw);
 
 struct nl_writer {
@@ -53,6 +54,7 @@ struct nl_writer {
 			int		priv;
 		} group;
 	};
+	const struct ifnet *ifp;	/* Used by Linux translation only */
 	u_int		num_messages;	/* Number of messages in the buffer */
 	int		malloc_flag;	/* M_WAITOK or M_NOWAIT */
 	bool		ignore_limit;	/* If true, ignores RCVBUF limit */