diff --git a/sysdep/linux/netlink.c b/sysdep/linux/netlink.c index 71b290fd..29b744cb 100644 --- a/sysdep/linux/netlink.c +++ b/sysdep/linux/netlink.c @@ -69,6 +69,10 @@ #define RTA_ENCAP 22 #endif +#ifndef NETLINK_GET_STRICT_CHK +#define NETLINK_GET_STRICT_CHK 12 +#endif + #define krt_ipv4(p) ((p)->af == AF_INET) #define krt_ecmp6(p) ((p)->af == AF_INET6) @@ -158,11 +162,39 @@ nl_open_sock(struct nl_sock *nl) } static void -nl_set_strict_dump(struct nl_sock *nl, int strict) +nl_set_strict_dump(struct nl_sock *nl UNUSED, int strict UNUSED) { + /* + * Strict checking is not necessary, it improves behavior on newer kernels. + * If it is not available (missing SOL_NETLINK compile-time, or ENOPROTOOPT + * run-time), we can just ignore it. + */ +#ifdef SOL_NETLINK setsockopt(nl->fd, SOL_NETLINK, NETLINK_GET_STRICT_CHK, &strict, sizeof(strict)); +#endif } +static void +nl_set_rcvbuf(int fd, uint val) +{ + if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &val, sizeof(val)) < 0) + log(L_WARN "KRT: Cannot set netlink rx buffer size to %u: %m", val); +} + +static uint +nl_cfg_rx_buffer_size(struct config *cfg) +{ + uint bufsize = 0; + + struct proto_config *pc; + WALK_LIST(pc, cfg->protos) + if ((pc->protocol == &proto_unix_kernel) && !pc->disabled) + bufsize = MAX(bufsize, ((struct krt_config *) pc)->sys.netlink_rx_buffer); + + return bufsize; +} + + static void nl_open(void) { @@ -1965,6 +1997,8 @@ krt_do_scan(struct krt_proto *p UNUSED) /* CONFIG_ALL_TABLES_AT_ONCE => p is NUL static sock *nl_async_sk; /* BIRD socket for asynchronous notifications */ static byte *nl_async_rx_buffer; /* Receive buffer */ +static uint nl_async_bufsize; /* Kernel rx buffer size for the netlink socket */ +static struct config *nl_last_config; /* For tracking changes to nl_async_bufsize */ static void nl_async_msg(struct nlmsghdr *h) @@ -2100,6 +2134,32 @@ nl_open_async(void) bug("Netlink: sk_open failed"); } +static void +nl_update_async_bufsize(void) +{ + /* No async socket */ + if (!nl_async_sk) + return; + + /* Already reconfigured */ + if (nl_last_config == config) + return; + + /* Update netlink buffer size */ + uint bufsize = nl_cfg_rx_buffer_size(config); + if (bufsize && (bufsize != nl_async_bufsize)) + { + /* Log message for reconfigurations only */ + if (nl_last_config) + log(L_INFO "KRT: Changing netlink rx buffer size to %u", bufsize); + + nl_set_rcvbuf(nl_async_sk->fd, bufsize); + nl_async_bufsize = bufsize; + } + + nl_last_config = config; +} + /* * Interface to the UNIX krt module @@ -2128,6 +2188,7 @@ krt_sys_start(struct krt_proto *p) nl_open(); nl_open_async(); + nl_update_async_bufsize(); return 1; } @@ -2135,12 +2196,16 @@ krt_sys_start(struct krt_proto *p) void krt_sys_shutdown(struct krt_proto *p) { + nl_update_async_bufsize(); + HASH_REMOVE2(nl_table_map, RTH, krt_pool, p); } int krt_sys_reconfigure(struct krt_proto *p UNUSED, struct krt_config *n, struct krt_config *o) { + nl_update_async_bufsize(); + return (n->sys.table_id == o->sys.table_id) && (n->sys.metric == o->sys.metric); }