diff options
| -rw-r--r-- | zebra/if_netlink.c | 15 | ||||
| -rw-r--r-- | zebra/if_netlink.h | 7 | ||||
| -rw-r--r-- | zebra/kernel_netlink.c | 195 | ||||
| -rw-r--r-- | zebra/kernel_netlink.h | 9 | ||||
| -rw-r--r-- | zebra/rt_netlink.c | 36 | ||||
| -rw-r--r-- | zebra/rt_netlink.h | 6 | ||||
| -rw-r--r-- | zebra/rule_netlink.c | 3 | ||||
| -rw-r--r-- | zebra/rule_netlink.h | 3 |
8 files changed, 160 insertions, 114 deletions
diff --git a/zebra/if_netlink.c b/zebra/if_netlink.c index 6c32407f7b..f153cc3510 100644 --- a/zebra/if_netlink.c +++ b/zebra/if_netlink.c @@ -557,10 +557,11 @@ static int netlink_bridge_interface(struct nlmsghdr *h, int len, ns_id_t ns_id, return 0; } -/* Called from interface_lookup_netlink(). This function is only used - during bootstrap. */ -static int netlink_interface(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +/* + * Called from interface_lookup_netlink(). This function is only used + * during bootstrap. + */ +static int netlink_interface(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct ifinfomsg *ifi; @@ -872,8 +873,7 @@ int kernel_address_delete_ipv6(struct interface *ifp, struct connected *ifc) return netlink_address(RTM_DELADDR, AF_INET6, ifp, ifc); } -int netlink_interface_addr(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +int netlink_interface_addr(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct ifaddrmsg *ifa; @@ -1076,8 +1076,7 @@ static void if_netlink_check_ifp_instance_consistency(uint16_t cmd, */ } -int netlink_link_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +int netlink_link_change(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct ifinfomsg *ifi; diff --git a/zebra/if_netlink.h b/zebra/if_netlink.h index 769c68a87e..65a266a519 100644 --- a/zebra/if_netlink.h +++ b/zebra/if_netlink.h @@ -23,10 +23,9 @@ #ifdef HAVE_NETLINK -extern int netlink_interface_addr(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup); -extern int netlink_link_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup); +extern int netlink_interface_addr(struct nlmsghdr *h, ns_id_t ns_id, + int startup); +extern int netlink_link_change(struct nlmsghdr *h, ns_id_t ns_id, int startup); extern int interface_lookup_netlink(struct zebra_ns *zns); #endif /* HAVE_NETLINK */ diff --git a/zebra/kernel_netlink.c b/zebra/kernel_netlink.c index ac5baba1cd..0e79b82533 100644 --- a/zebra/kernel_netlink.c +++ b/zebra/kernel_netlink.c @@ -129,11 +129,22 @@ extern uint32_t nl_rcvbufsize; extern struct zebra_privs_t zserv_privs; -int netlink_talk_filter(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +int netlink_talk_filter(struct nlmsghdr *h, ns_id_t ns_id, int startup) { - zlog_warn("netlink_talk: ignoring message type 0x%04x NS %u", - h->nlmsg_type, ns_id); + /* + * This is an error condition that must be handled during + * development. + * + * The netlink_talk_filter function is used for communication + * down the netlink_cmd pipe and we are expecting + * an ack being received. So if we get here + * then we did not receive the ack and instead + * received some other message in an unexpected + * way. + */ + zlog_err("%s: ignoring message type 0x%04x(%s) NS %u", + __PRETTY_FUNCTION__, h->nlmsg_type, + nl_msg_type_to_str(h->nlmsg_type), ns_id); return 0; } @@ -233,41 +244,50 @@ static int netlink_socket(struct nlsock *nl, unsigned long groups, return ret; } -static int netlink_information_fetch(struct sockaddr_nl *snl, - struct nlmsghdr *h, ns_id_t ns_id, +static int netlink_information_fetch(struct nlmsghdr *h, ns_id_t ns_id, int startup) { - /* JF: Ignore messages that aren't from the kernel */ - if (snl->nl_pid != 0) { - zlog_err("Ignoring message from pid %u", snl->nl_pid); - return 0; - } - + /* + * When we handle new message types here + * because we are starting to install them + * then lets check the netlink_install_filter + * and see if we should add the corresponding + * allow through entry there. + * Probably not needed to do but please + * think about it. + */ switch (h->nlmsg_type) { case RTM_NEWROUTE: - return netlink_route_change(snl, h, ns_id, startup); + return netlink_route_change(h, ns_id, startup); case RTM_DELROUTE: - return netlink_route_change(snl, h, ns_id, startup); + return netlink_route_change(h, ns_id, startup); case RTM_NEWLINK: - return netlink_link_change(snl, h, ns_id, startup); + return netlink_link_change(h, ns_id, startup); case RTM_DELLINK: - return netlink_link_change(snl, h, ns_id, startup); + return netlink_link_change(h, ns_id, startup); case RTM_NEWADDR: - return netlink_interface_addr(snl, h, ns_id, startup); + return netlink_interface_addr(h, ns_id, startup); case RTM_DELADDR: - return netlink_interface_addr(snl, h, ns_id, startup); + return netlink_interface_addr(h, ns_id, startup); case RTM_NEWNEIGH: - return netlink_neigh_change(snl, h, ns_id); + return netlink_neigh_change(h, ns_id); case RTM_DELNEIGH: - return netlink_neigh_change(snl, h, ns_id); + return netlink_neigh_change(h, ns_id); case RTM_NEWRULE: - return netlink_rule_change(snl, h, ns_id, startup); + return netlink_rule_change(h, ns_id, startup); case RTM_DELRULE: - return netlink_rule_change(snl, h, ns_id, startup); + return netlink_rule_change(h, ns_id, startup); default: - if (IS_ZEBRA_DEBUG_KERNEL) - zlog_debug("Unknown netlink nlmsg_type %d vrf %u\n", - h->nlmsg_type, ns_id); + /* + * If we have received this message then + * we have made a mistake during development + * and we need to write some code to handle + * this message type or not ask for + * it to be sent up to us + */ + zlog_err("Unknown netlink nlmsg_type %s(%d) vrf %u\n", + nl_msg_type_to_str(h->nlmsg_type), h->nlmsg_type, + ns_id); break; } return 0; @@ -284,31 +304,69 @@ static int kernel_read(struct thread *thread) return 0; } -/* Filter out messages from self that occur on listener socket, +/* + * Filter out messages from self that occur on listener socket, * caused by our actions on the command socket + * + * When we add new Netlink message types we probably + * do not need to add them here as that we are filtering + * on the routes we actually care to receive( which is rarer + * then the normal course of operations). We are intentionally + * allowing some messages from ourselves through + * ( I'm looking at you Interface based netlink messages ) + * so that we only had to write one way to handle incoming + * address add/delete changes. */ static void netlink_install_filter(int sock, __u32 pid) { + /* + * BPF_JUMP instructions and where you jump to are based upon + * 0 as being the next statement. So count from 0. Writing + * this down because every time I look at this I have to + * re-remember it. + */ struct sock_filter filter[] = { - /* 0: ldh [4] */ - BPF_STMT(BPF_LD | BPF_ABS | BPF_H, - offsetof(struct nlmsghdr, nlmsg_type)), - /* 1: jeq 0x18 jt 5 jf next */ - BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_NEWROUTE), 3, 0), - /* 2: jeq 0x19 jt 5 jf next */ - BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_DELROUTE), 2, 0), - /* 3: jeq 0x19 jt 5 jf next */ - BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_NEWNEIGH), 1, 0), - /* 4: jeq 0x19 jt 5 jf 8 */ - BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_DELNEIGH), 0, 3), - /* 5: ldw [12] */ + /* + * Logic: + * if (nlmsg_pid == pid) { + * if (the incoming nlmsg_type == + * RTM_NEWADDR | RTM_DELADDR) + * keep this message + * else + * skip this message + * } else + * keep this netlink message + */ + /* + * 0: Load the nlmsg_pid into the BPF register + */ BPF_STMT(BPF_LD | BPF_ABS | BPF_W, offsetof(struct nlmsghdr, nlmsg_pid)), - /* 6: jeq XX jt 7 jf 8 */ - BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htonl(pid), 0, 1), - /* 7: ret 0 (skip) */ + /* + * 1: Compare to pid + */ + BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htonl(pid), 0, 4), + /* + * 2: Load the nlmsg_type into BPF register + */ + BPF_STMT(BPF_LD | BPF_ABS | BPF_H, + offsetof(struct nlmsghdr, nlmsg_type)), + /* + * 3: Compare to RTM_NEWADDR + */ + BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_NEWADDR), 2, 0), + /* + * 4: Compare to RTM_DELADDR + */ + BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_DELADDR), 1, 0), + /* + * 5: This is the end state of we want to skip the + * message + */ BPF_STMT(BPF_RET | BPF_K, 0), - /* 8: ret 0xffff (keep) */ + /* 6: This is the end state of we want to keep + * the message + */ BPF_STMT(BPF_RET | BPF_K, 0xffff), }; @@ -453,8 +511,7 @@ const char *nl_rttype_to_str(uint8_t rttype) * startup -> Are we reading in under startup conditions? passed to * the filter. */ -int netlink_parse_info(int (*filter)(struct sockaddr_nl *, struct nlmsghdr *, - ns_id_t, int), +int netlink_parse_info(int (*filter)(struct nlmsghdr *, ns_id_t, int), struct nlsock *nl, struct zebra_ns *zns, int count, int startup) { @@ -621,24 +678,18 @@ int netlink_parse_info(int (*filter)(struct sockaddr_nl *, struct nlmsghdr *, h->nlmsg_type, h->nlmsg_len, h->nlmsg_seq, h->nlmsg_pid); - /* skip unsolicited messages originating from command - * socket - * linux sets the originators port-id for {NEW|DEL}ADDR - * messages, - * so this has to be checked here. */ - if (nl != &zns->netlink_cmd - && h->nlmsg_pid == zns->netlink_cmd.snl.nl_pid - && (h->nlmsg_type != RTM_NEWADDR - && h->nlmsg_type != RTM_DELADDR)) { - if (IS_ZEBRA_DEBUG_KERNEL) - zlog_debug( - "netlink_parse_info: %s packet comes from %s", - zns->netlink_cmd.name, - nl->name); + + /* + * Ignore messages that maybe sent from + * other actors besides the kernel + */ + if (snl.nl_pid != 0) { + zlog_err("Ignoring message from pid %u", + snl.nl_pid); continue; } - error = (*filter)(&snl, h, zns->ns_id, startup); + error = (*filter)(h, zns->ns_id, startup); if (error < 0) { zlog_err("%s filter function error", nl->name); ret = error; @@ -672,8 +723,7 @@ int netlink_parse_info(int (*filter)(struct sockaddr_nl *, struct nlmsghdr *, * startup -> Are we reading in under startup conditions * This is passed through eventually to filter. */ -int netlink_talk(int (*filter)(struct sockaddr_nl *, struct nlmsghdr *, ns_id_t, - int startup), +int netlink_talk(int (*filter)(struct nlmsghdr *, ns_id_t, int startup), struct nlmsghdr *n, struct nlsock *nl, struct zebra_ns *zns, int startup) { @@ -787,11 +837,24 @@ void kernel_init(struct zebra_ns *zns) { unsigned long groups; - /* Initialize netlink sockets */ - groups = RTMGRP_LINK | RTMGRP_IPV4_ROUTE | RTMGRP_IPV4_IFADDR - | RTMGRP_IPV6_ROUTE | RTMGRP_IPV6_IFADDR | RTMGRP_IPV4_MROUTE - | RTMGRP_NEIGH - | RTNLGRP_IPV4_RULE | RTNLGRP_IPV6_RULE; + /* + * Initialize netlink sockets + * + * If RTMGRP_XXX exists use that, but at some point + * I think the kernel developers realized that + * keeping track of all the different values would + * lead to confusion, so we need to convert the + * RTNLGRP_XXX to a bit position for ourself + */ + groups = RTMGRP_LINK | + RTMGRP_IPV4_ROUTE | + RTMGRP_IPV4_IFADDR | + RTMGRP_IPV6_ROUTE | + RTMGRP_IPV6_IFADDR | + RTMGRP_IPV4_MROUTE | + RTMGRP_NEIGH | + (1 << (RTNLGRP_IPV4_RULE - 1)) | + (1 << (RTNLGRP_IPV6_RULE - 1)); snprintf(zns->netlink.name, sizeof(zns->netlink.name), "netlink-listen (NS %u)", zns->ns_id); diff --git a/zebra/kernel_netlink.h b/zebra/kernel_netlink.h index dc075b9aff..3b4048ff69 100644 --- a/zebra/kernel_netlink.h +++ b/zebra/kernel_netlink.h @@ -45,14 +45,11 @@ extern const char *nl_rtproto_to_str(uint8_t rtproto); extern const char *nl_family_to_str(uint8_t family); extern const char *nl_rttype_to_str(uint8_t rttype); -extern int netlink_parse_info(int (*filter)(struct sockaddr_nl *, - struct nlmsghdr *, ns_id_t, int), +extern int netlink_parse_info(int (*filter)(struct nlmsghdr *, ns_id_t, int), struct nlsock *nl, struct zebra_ns *zns, int count, int startup); -extern int netlink_talk_filter(struct sockaddr_nl *, struct nlmsghdr *, ns_id_t, - int startup); -extern int netlink_talk(int (*filter)(struct sockaddr_nl *, struct nlmsghdr *, - ns_id_t, int startup), +extern int netlink_talk_filter(struct nlmsghdr *h, ns_id_t ns, int startup); +extern int netlink_talk(int (*filter)(struct nlmsghdr *, ns_id_t, int startup), struct nlmsghdr *n, struct nlsock *nl, struct zebra_ns *zns, int startup); extern int netlink_request(struct nlsock *nl, struct nlmsghdr *n); diff --git a/zebra/rt_netlink.c b/zebra/rt_netlink.c index 08c81d037a..e9b3e59d0e 100644 --- a/zebra/rt_netlink.c +++ b/zebra/rt_netlink.c @@ -247,8 +247,7 @@ static vrf_id_t vrf_lookup_by_table(uint32_t table_id, ns_id_t ns_id) } /* Looking up routing table by netlink interface. */ -static int netlink_route_change_read_unicast(struct sockaddr_nl *snl, - struct nlmsghdr *h, ns_id_t ns_id, +static int netlink_route_change_read_unicast(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; @@ -632,8 +631,7 @@ static int netlink_route_change_read_unicast(struct sockaddr_nl *snl, static struct mcast_route_data *mroute = NULL; -static int netlink_route_change_read_multicast(struct sockaddr_nl *snl, - struct nlmsghdr *h, +static int netlink_route_change_read_multicast(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; @@ -722,8 +720,7 @@ static int netlink_route_change_read_multicast(struct sockaddr_nl *snl, return 0; } -int netlink_route_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +int netlink_route_change(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct rtmsg *rtm; @@ -754,9 +751,9 @@ int netlink_route_change(struct sockaddr_nl *snl, struct nlmsghdr *h, return -1; if (rtm->rtm_type == RTN_MULTICAST) - netlink_route_change_read_multicast(snl, h, ns_id, startup); + netlink_route_change_read_multicast(h, ns_id, startup); else - netlink_route_change_read_unicast(snl, h, ns_id, startup); + netlink_route_change_read_unicast(h, ns_id, startup); return 0; } @@ -1815,8 +1812,7 @@ int kernel_del_vtep(vni_t vni, struct interface *ifp, struct in_addr *vtep_ip) ((struct rtattr *)(((char *)(r)) + NLMSG_ALIGN(sizeof(struct ndmsg)))) #endif -static int netlink_macfdb_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - int len, ns_id_t ns_id) +static int netlink_macfdb_change(struct nlmsghdr *h, int len, ns_id_t ns_id) { struct ndmsg *ndm; struct interface *ifp; @@ -1951,8 +1947,7 @@ static int netlink_macfdb_change(struct sockaddr_nl *snl, struct nlmsghdr *h, return zebra_vxlan_local_mac_del(ifp, br_if, &mac, vid); } -static int netlink_macfdb_table(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +static int netlink_macfdb_table(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct ndmsg *ndm; @@ -1970,7 +1965,7 @@ static int netlink_macfdb_table(struct sockaddr_nl *snl, struct nlmsghdr *h, if (ndm->ndm_family != AF_BRIDGE) return 0; - return netlink_macfdb_change(snl, h, len, ns_id); + return netlink_macfdb_change(h, len, ns_id); } /* Request for MAC FDB information from the kernel */ @@ -2126,8 +2121,7 @@ static int netlink_macfdb_update(struct interface *ifp, vlanid_t vid, (NUD_PERMANENT | NUD_NOARP | NUD_REACHABLE | NUD_PROBE | NUD_STALE \ | NUD_DELAY) -static int netlink_ipneigh_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - int len, ns_id_t ns_id) +static int netlink_ipneigh_change(struct nlmsghdr *h, int len, ns_id_t ns_id) { struct ndmsg *ndm; struct interface *ifp; @@ -2270,8 +2264,7 @@ static int netlink_ipneigh_change(struct sockaddr_nl *snl, struct nlmsghdr *h, return zebra_vxlan_handle_kernel_neigh_del(ifp, link_if, &ip); } -static int netlink_neigh_table(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +static int netlink_neigh_table(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct ndmsg *ndm; @@ -2289,7 +2282,7 @@ static int netlink_neigh_table(struct sockaddr_nl *snl, struct nlmsghdr *h, if (ndm->ndm_family != AF_INET && ndm->ndm_family != AF_INET6) return 0; - return netlink_neigh_change(snl, h, len); + return netlink_neigh_change(h, len); } /* Request for IP neighbor information from the kernel */ @@ -2349,8 +2342,7 @@ int netlink_neigh_read_for_vlan(struct zebra_ns *zns, struct interface *vlan_if) return ret; } -int netlink_neigh_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id) +int netlink_neigh_change(struct nlmsghdr *h, ns_id_t ns_id) { int len; struct ndmsg *ndm; @@ -2366,13 +2358,13 @@ int netlink_neigh_change(struct sockaddr_nl *snl, struct nlmsghdr *h, /* Is this a notification for the MAC FDB or IP neighbor table? */ ndm = NLMSG_DATA(h); if (ndm->ndm_family == AF_BRIDGE) - return netlink_macfdb_change(snl, h, len, ns_id); + return netlink_macfdb_change(h, len, ns_id); if (ndm->ndm_type != RTN_UNICAST) return 0; if (ndm->ndm_family == AF_INET || ndm->ndm_family == AF_INET6) - return netlink_ipneigh_change(snl, h, len, ns_id); + return netlink_ipneigh_change(h, len, ns_id); return 0; } diff --git a/zebra/rt_netlink.h b/zebra/rt_netlink.h index c8d918b1d2..c4f21d1504 100644 --- a/zebra/rt_netlink.h +++ b/zebra/rt_netlink.h @@ -59,12 +59,10 @@ void rt_netlink_init(void); extern int netlink_mpls_multipath(int cmd, zebra_lsp_t *lsp); -extern int netlink_route_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup); +extern int netlink_route_change(struct nlmsghdr *h, ns_id_t ns_id, int startup); extern int netlink_route_read(struct zebra_ns *zns); -extern int netlink_neigh_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id); +extern int netlink_neigh_change(struct nlmsghdr *h, ns_id_t ns_id); extern int netlink_macfdb_read(struct zebra_ns *zns); extern int netlink_macfdb_read_for_bridge(struct zebra_ns *zns, struct interface *ifp, diff --git a/zebra/rule_netlink.c b/zebra/rule_netlink.c index 310f0952fa..5f73545855 100644 --- a/zebra/rule_netlink.c +++ b/zebra/rule_netlink.c @@ -166,8 +166,7 @@ void kernel_del_pbr_rule(struct zebra_pbr_rule *rule) * notification of interest. The expectation is that if this corresponds * to a PBR rule added by FRR, it will be readded. */ -int netlink_rule_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup) +int netlink_rule_change(struct nlmsghdr *h, ns_id_t ns_id, int startup) { struct zebra_ns *zns; struct fib_rule_hdr *frh; diff --git a/zebra/rule_netlink.h b/zebra/rule_netlink.h index 3a9b51309e..4547a1bb3b 100644 --- a/zebra/rule_netlink.h +++ b/zebra/rule_netlink.h @@ -29,8 +29,7 @@ /* * Handle netlink notification informing a rule add or delete. */ -extern int netlink_rule_change(struct sockaddr_nl *snl, struct nlmsghdr *h, - ns_id_t ns_id, int startup); +extern int netlink_rule_change(struct nlmsghdr *h, ns_id_t ns_id, int startup); /* * Get to know existing PBR rules in the kernel - typically called at startup. |