X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=89815006773717876c2220c58c0f46eb45823a04;hb=33b14e70e2c1c6dd6bf52ae4ec3099bc4456270e;hp=24a81bd86ca58c669d7ed32946097a82a23c8dd6;hpb=d3d8f1f7e51c1808471e081c46ec631af7375272;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index 24a81bd86..898150067 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -34,18 +34,21 @@ #include "bitmap.h" #include "dpif-provider.h" +#include "dynamic-string.h" +#include "flow.h" #include "netdev.h" #include "netdev-linux.h" #include "netdev-vport.h" +#include "netlink-notifier.h" #include "netlink-socket.h" #include "netlink.h" #include "odp-util.h" #include "ofpbuf.h" +#include "openvswitch/datapath-compat.h" #include "openvswitch/tunnel.h" #include "packets.h" #include "poll-loop.h" -#include "rtnetlink.h" -#include "rtnetlink-link.h" +#include "random.h" #include "shash.h" #include "sset.h" #include "unaligned.h" @@ -58,19 +61,25 @@ enum { LRU_MAX_PORTS = 1024 }; enum { LRU_MASK = LRU_MAX_PORTS - 1}; BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS)); +enum { N_UPCALL_SOCKS = 16 }; +BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS)); + +/* This ethtool flag was introduced in Linux 2.6.24, so it might be + * missing if we have old headers. */ +#define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */ + struct dpif_linux_dp { /* Generic Netlink header. */ uint8_t cmd; - /* struct odp_header. */ + /* struct ovs_header. */ int dp_ifindex; /* Attributes. */ - const char *name; /* ODP_DP_ATTR_NAME. */ - struct odp_stats stats; /* ODP_DP_ATTR_STATS. */ - enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */ - const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */ - uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */ + const char *name; /* OVS_DP_ATTR_NAME. */ + uint32_t upcall_pid; /* OVS_DP_UPCALL_PID. */ + struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */ + enum ovs_frag_handling ipv4_frags; /* OVS_DP_ATTR_IPV4_FRAGS. */ }; static void dpif_linux_dp_init(struct dpif_linux_dp *); @@ -87,7 +96,7 @@ struct dpif_linux_flow { /* Generic Netlink header. */ uint8_t cmd; - /* struct odp_header. */ + /* struct ovs_header. */ unsigned int nlmsg_flags; int dp_ifindex; @@ -97,16 +106,17 @@ struct dpif_linux_flow { * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to * access their values. * - * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in + * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in * the Netlink version of the command, even if actions_len is zero. */ - const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */ + const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */ size_t key_len; - const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */ + const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */ size_t actions_len; - const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */ - const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */ - const uint64_t *used; /* ODP_FLOW_ATTR_USED. */ - bool clear; /* ODP_FLOW_ATTR_CLEAR. */ + uint32_t upcall_pid; /* OVS_FLOW_ATTR_UPCALL_PID. */ + const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */ + const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */ + const uint64_t *used; /* OVS_FLOW_ATTR_USED. */ + bool clear; /* OVS_FLOW_ATTR_CLEAR. */ }; static void dpif_linux_flow_init(struct dpif_linux_flow *); @@ -125,14 +135,14 @@ struct dpif_linux { struct dpif dpif; int dp_ifindex; - /* Multicast group messages. */ - struct nl_sock *mc_sock; - uint32_t mcgroups[DPIF_N_UC_TYPES]; + /* Upcall messages. */ + struct nl_sock *upcall_socks[N_UPCALL_SOCKS]; + int last_read_upcall; unsigned int listen_mask; /* Change notification. */ struct sset changed_ports; /* Ports that have changed. */ - struct rtnetlink_notifier port_notifier; + struct nln_notifier *port_notifier; bool change_error; /* Queue of unused ports. */ @@ -144,19 +154,23 @@ struct dpif_linux { static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); -/* Generic Netlink family numbers for ODP. */ -static int odp_datapath_family; -static int odp_vport_family; -static int odp_flow_family; -static int odp_packet_family; +/* Generic Netlink family numbers for OVS. */ +static int ovs_datapath_family; +static int ovs_vport_family; +static int ovs_flow_family; +static int ovs_packet_family; /* Generic Netlink socket. */ static struct nl_sock *genl_sock; +static struct nln *nln = NULL; static int dpif_linux_init(void); -static int open_dpif(const struct dpif_linux_dp *, struct dpif **); -static void dpif_linux_port_changed(const struct rtnetlink_link_change *, - void *dpif); +static void open_dpif(const struct dpif_linux_dp *, struct dpif **); +static bool dpif_linux_nln_parse(struct ofpbuf *, void *); +static void dpif_linux_port_changed(const void *vport, void *dpif); +static uint32_t get_upcall_pid_port(struct dpif_linux *, uint32_t port); +static uint32_t get_upcall_pid_flow(struct dpif_linux *, + const struct nlattr *key, size_t key_len); static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, struct ofpbuf *); @@ -231,63 +245,60 @@ dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, /* Create or look up datapath. */ dpif_linux_dp_init(&dp_request); - dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET; + dp_request.cmd = create ? OVS_DP_CMD_NEW : OVS_DP_CMD_GET; dp_request.name = name; error = dpif_linux_dp_transact(&dp_request, &dp, &buf); if (error) { return error; } - error = open_dpif(&dp, dpifp); - ofpbuf_delete(buf); - return error; + open_dpif(&dp, dpifp); + ofpbuf_delete(buf); + return 0; } -static int +static void open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp) { struct dpif_linux *dpif; - int error; int i; - dpif = xmalloc(sizeof *dpif); - error = rtnetlink_link_notifier_register(&dpif->port_notifier, - dpif_linux_port_changed, dpif); - if (error) { - goto error_free; - } + dpif = xzalloc(sizeof *dpif); + dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed, + dpif); dpif_init(&dpif->dpif, &dpif_linux_class, dp->name, dp->dp_ifindex, dp->dp_ifindex); - dpif->mc_sock = NULL; - for (i = 0; i < DPIF_N_UC_TYPES; i++) { - dpif->mcgroups[i] = dp->mcgroups[i]; - } - dpif->listen_mask = 0; dpif->dp_ifindex = dp->dp_ifindex; sset_init(&dpif->changed_ports); - dpif->change_error = false; *dpifp = &dpif->dpif; - dpif->lru_head = dpif->lru_tail = 0; dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS); - bitmap_set1(dpif->lru_bitmap, ODPP_LOCAL); + bitmap_set1(dpif->lru_bitmap, OVSP_LOCAL); for (i = 1; i < LRU_MAX_PORTS; i++) { dpif_linux_push_port(dpif, i); } - return 0; +} -error_free: - free(dpif); - return error; +static void +destroy_upcall_socks(struct dpif_linux *dpif) +{ + int i; + + for (i = 0; i < N_UPCALL_SOCKS; i++) { + nl_sock_destroy(dpif->upcall_socks[i]); + dpif->upcall_socks[i] = NULL; + } } static void dpif_linux_close(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - rtnetlink_link_notifier_unregister(&dpif->port_notifier); + + nln_notifier_destroy(dpif->port_notifier); + destroy_upcall_socks(dpif); sset_destroy(&dpif->changed_ports); free(dpif->lru_bitmap); free(dpif); @@ -300,13 +311,29 @@ dpif_linux_destroy(struct dpif *dpif_) struct dpif_linux_dp dp; dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_CMD_DEL; + dp.cmd = OVS_DP_CMD_DEL; dp.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&dp, NULL, NULL); } +static void +dpif_linux_run(struct dpif *dpif OVS_UNUSED) +{ + if (nln) { + nln_run(nln); + } +} + +static void +dpif_linux_wait(struct dpif *dpif OVS_UNUSED) +{ + if (nln) { + nln_wait(nln); + } +} + static int -dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats) +dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) { struct dpif_linux_dp dp; struct ofpbuf *buf; @@ -314,7 +341,11 @@ dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats) error = dpif_linux_dp_get(dpif_, &dp, &buf); if (!error) { - *stats = dp.stats; + stats->n_frags = dp.stats.n_frags; + stats->n_hit = dp.stats.n_hit; + stats->n_missed = dp.stats.n_missed; + stats->n_lost = dp.stats.n_lost; + stats->n_flows = dp.stats.n_flows; ofpbuf_delete(buf); } return error; @@ -329,7 +360,7 @@ dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) error = dpif_linux_dp_get(dpif_, &dp, &buf); if (!error) { - *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP; + *drop_fragsp = dp.ipv4_frags == OVS_DP_FRAG_DROP; ofpbuf_delete(buf); } return error; @@ -342,9 +373,9 @@ dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) struct dpif_linux_dp dp; dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_CMD_SET; + dp.cmd = OVS_DP_CMD_SET; dp.dp_ifindex = dpif->dp_ifindex; - dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO; + dp.ipv4_frags = drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO; return dpif_linux_dp_transact(&dp, NULL, NULL); } @@ -361,10 +392,10 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, int error; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_CMD_NEW; + request.cmd = OVS_VPORT_CMD_NEW; request.dp_ifindex = dpif->dp_ifindex; request.type = netdev_vport_get_vport_type(netdev); - if (request.type == ODP_VPORT_TYPE_UNSPEC) { + if (request.type == OVS_VPORT_TYPE_UNSPEC) { VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has " "unsupported type `%s'", dpif_name(dpif_), name, type); @@ -378,13 +409,22 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, request.options_len = options->size; } + if (request.type == OVS_VPORT_TYPE_NETDEV) { + netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false); + } + /* Loop until we find a port that isn't used. */ do { request.port_no = dpif_linux_pop_port(dpif); + request.upcall_pid = get_upcall_pid_port(dpif, request.port_no); error = dpif_linux_vport_transact(&request, &reply, &buf); if (!error) { *port_nop = reply.port_no; + VLOG_DBG("%s: assigning port %"PRIu32" to netlink " + "pid %"PRIu32, + dpif_name(dpif_), request.port_no, + request.upcall_pid); } ofpbuf_delete(buf); } while (request.port_no != UINT32_MAX @@ -401,7 +441,7 @@ dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) int error; dpif_linux_vport_init(&vport); - vport.cmd = ODP_VPORT_CMD_DEL; + vport.cmd = OVS_VPORT_CMD_DEL; vport.dp_ifindex = dpif->dp_ifindex; vport.port_no = port_no; error = dpif_linux_vport_transact(&vport, NULL, NULL); @@ -422,7 +462,7 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, int error; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_CMD_GET; + request.cmd = OVS_VPORT_CMD_GET; request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex; request.port_no = port_no; request.name = port_name; @@ -432,12 +472,6 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, dpif_port->name = xstrdup(reply.name); dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply)); dpif_port->port_no = reply.port_no; - if (reply.stats) { - netdev_stats_from_rtnl_link_stats64(&dpif_port->stats, - reply.stats); - } else { - memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats); - } ofpbuf_delete(buf); } return error; @@ -472,7 +506,7 @@ dpif_linux_flow_flush(struct dpif *dpif_) struct dpif_linux_flow flow; dpif_linux_flow_init(&flow); - flow.cmd = ODP_FLOW_CMD_DEL; + flow.cmd = OVS_FLOW_CMD_DEL; flow.dp_ifindex = dpif->dp_ifindex; return dpif_linux_flow_transact(&flow, NULL, NULL); } @@ -496,7 +530,7 @@ dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep) state->complete = false; dpif_linux_vport_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; buf = ofpbuf_new(1024); @@ -533,11 +567,6 @@ dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_, dpif_port->name = (char *) vport.name; dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport); dpif_port->port_no = vport.port_no; - if (vport.stats) { - netdev_stats_from_rtnl_link_stats64(&dpif_port->stats, vport.stats); - } else { - memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats); - } return 0; } @@ -586,8 +615,6 @@ dpif_linux_port_poll_wait(const struct dpif *dpif_) struct dpif_linux *dpif = dpif_linux_cast(dpif_); if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) { poll_immediate_wake(); - } else { - rtnetlink_link_notifier_wait(); } } @@ -600,7 +627,7 @@ dpif_linux_flow_get__(const struct dpif *dpif_, struct dpif_linux_flow request; dpif_linux_flow_init(&request); - request.cmd = ODP_FLOW_CMD_GET; + request.cmd = OVS_FLOW_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; @@ -645,13 +672,14 @@ dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags, int error; dpif_linux_flow_init(&request); - request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET; + request.cmd = flags & DPIF_FP_CREATE ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET; request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; - /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */ + /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */ request.actions = actions ? actions : &dummy_action; request.actions_len = actions_len; + request.upcall_pid = get_upcall_pid_flow(dpif, key, key_len); if (flags & DPIF_FP_ZERO_STATS) { request.clear = true; } @@ -677,7 +705,7 @@ dpif_linux_flow_del(struct dpif *dpif_, int error; dpif_linux_flow_init(&request); - request.cmd = ODP_FLOW_CMD_DEL; + request.cmd = OVS_FLOW_CMD_DEL; request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; @@ -709,7 +737,7 @@ dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) *statep = state = xmalloc(sizeof *state); dpif_linux_flow_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; buf = ofpbuf_new(1024); @@ -783,31 +811,46 @@ dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) } static int -dpif_linux_execute(struct dpif *dpif_, - const struct nlattr *actions, size_t actions_len, - const struct ofpbuf *packet) +dpif_linux_execute__(int dp_ifindex, uint32_t upcall_pid, + const struct nlattr *key, size_t key_len, + const struct nlattr *actions, size_t actions_len, + const struct ofpbuf *packet) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct odp_header *execute; + struct ovs_header *execute; struct ofpbuf *buf; int error; buf = ofpbuf_new(128 + actions_len + packet->size); - nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST, - ODP_PACKET_CMD_EXECUTE, 1); + nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST, + OVS_PACKET_CMD_EXECUTE, 1); execute = ofpbuf_put_uninit(buf, sizeof *execute); - execute->dp_ifindex = dpif->dp_ifindex; + execute->dp_ifindex = dp_ifindex; - nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size); - nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len); + nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, packet->data, packet->size); + nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, key, key_len); + nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, actions, actions_len); + nl_msg_put_u32(buf, OVS_PACKET_ATTR_UPCALL_PID, upcall_pid); error = nl_sock_transact(genl_sock, buf, NULL); ofpbuf_delete(buf); return error; } +static int +dpif_linux_execute(struct dpif *dpif_, + const struct nlattr *key, size_t key_len, + const struct nlattr *actions, size_t actions_len, + const struct ofpbuf *packet) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + uint32_t upcall_pid = get_upcall_pid_flow(dpif, key, key_len); + + return dpif_linux_execute__(dpif->dp_ifindex, upcall_pid, key, key_len, + actions, actions_len, packet); +} + static int dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) { @@ -816,79 +859,123 @@ dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) return 0; } -static int -dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask) +static uint32_t +get_upcall_pid_port__(struct dpif_linux *dpif, uint32_t port) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int error; - int i; + int idx = port & (N_UPCALL_SOCKS - 1); + return nl_sock_pid(dpif->upcall_socks[idx]); +} - if (listen_mask == dpif->listen_mask) { - return 0; - } else if (!listen_mask) { - nl_sock_destroy(dpif->mc_sock); - dpif->mc_sock = NULL; - dpif->listen_mask = 0; +static uint32_t +get_upcall_pid_port(struct dpif_linux *dpif, uint32_t port) +{ + if (!(dpif->listen_mask & (1u << DPIF_UC_MISS))) { return 0; - } else if (!dpif->mc_sock) { - error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock); - if (error) { - return error; - } } - /* Unsubscribe from old groups. */ - for (i = 0; i < DPIF_N_UC_TYPES; i++) { - if (dpif->listen_mask & (1u << i)) { - nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]); - } - } + return get_upcall_pid_port__(dpif, port); +} - /* Update listen_mask. */ - dpif->listen_mask = listen_mask; +static uint32_t +get_upcall_pid_flow(struct dpif_linux *dpif, + const struct nlattr *key, size_t key_len) +{ + const struct nlattr *nla; + uint32_t port; - /* Subscribe to new groups. */ - error = 0; - for (i = 0; i < DPIF_N_UC_TYPES; i++) { - if (dpif->listen_mask & (1u << i)) { - int retval; + if (!(dpif->listen_mask & (1u << DPIF_UC_ACTION))) { + return 0; + } - retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]); - if (retval) { - error = retval; - } - } + nla = nl_attr_find__(key, key_len, OVS_KEY_ATTR_IN_PORT); + if (nla) { + port = nl_attr_get_u32(nla); + } else { + port = random_uint32(); } - return error; + + return get_upcall_pid_port__(dpif, port); } -static int -dpif_linux_get_sflow_probability(const struct dpif *dpif_, - uint32_t *probability) +static void +set_upcall_pids(struct dpif_linux *dpif) { - struct dpif_linux_dp dp; - struct ofpbuf *buf; + struct dpif_port port; + struct dpif_port_dump port_dump; + struct dpif_flow_dump flow_dump; + const struct nlattr *key; + size_t key_len; int error; - error = dpif_linux_dp_get(dpif_, &dp, &buf); - if (!error) { - *probability = dp.sampling ? *dp.sampling : 0; - ofpbuf_delete(buf); + DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) { + struct dpif_linux_vport vport_request; + + dpif_linux_vport_init(&vport_request); + vport_request.cmd = OVS_VPORT_CMD_SET; + vport_request.dp_ifindex = dpif->dp_ifindex; + vport_request.port_no = port.port_no; + vport_request.upcall_pid = get_upcall_pid_port(dpif, + vport_request.port_no); + error = dpif_linux_vport_transact(&vport_request, NULL, NULL); + if (!error) { + VLOG_DBG("%s: assigning port %"PRIu32" to netlink " + "pid %"PRIu32, + dpif_name(&dpif->dpif), vport_request.port_no, + vport_request.upcall_pid); + } else { + VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s", + dpif_name(&dpif->dpif), strerror(error)); + } } - return error; + + dpif_flow_dump_start(&flow_dump, &dpif->dpif); + while (dpif_flow_dump_next(&flow_dump, &key, &key_len, + NULL, NULL, NULL)) { + struct dpif_linux_flow flow_request; + + dpif_linux_flow_init(&flow_request); + flow_request.cmd = OVS_FLOW_CMD_SET; + flow_request.dp_ifindex = dpif->dp_ifindex; + flow_request.key = key; + flow_request.key_len = key_len; + flow_request.upcall_pid = get_upcall_pid_flow(dpif, key, key_len); + error = dpif_linux_flow_transact(&flow_request, NULL, NULL); + if (error) { + VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on flow: %s", + dpif_name(&dpif->dpif), strerror(error)); + } + } + dpif_flow_dump_done(&flow_dump); } static int -dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability) +dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_dp dp; - dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_CMD_SET; - dp.dp_ifindex = dpif->dp_ifindex; - dp.sampling = &probability; - return dpif_linux_dp_transact(&dp, NULL, NULL); + if (listen_mask == dpif->listen_mask) { + return 0; + } + + if (!listen_mask) { + destroy_upcall_socks(dpif); + } else if (!dpif->listen_mask) { + int i; + int error; + + for (i = 0; i < N_UPCALL_SOCKS; i++) { + error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]); + if (error) { + destroy_upcall_socks(dpif); + return error; + } + } + } + + dpif->listen_mask = listen_mask; + set_upcall_pids(dpif); + + return 0; } static int @@ -907,22 +994,18 @@ static int parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, int *dp_ifindex) { - static const struct nl_policy odp_packet_policy[] = { + static const struct nl_policy ovs_packet_policy[] = { /* Always present. */ - [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC, + [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC, .min_len = ETH_HEADER_LEN }, - [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, - - /* ODP_PACKET_CMD_ACTION only. */ - [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true }, + [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, - /* ODP_PACKET_CMD_SAMPLE only. */ - [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true }, - [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, + /* OVS_PACKET_CMD_ACTION only. */ + [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true }, }; - struct odp_header *odp_header; - struct nlattr *a[ARRAY_SIZE(odp_packet_policy)]; + struct ovs_header *ovs_header; + struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)]; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -932,17 +1015,16 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_packet_family - || !nl_policy_parse(&b, 0, odp_packet_policy, a, - ARRAY_SIZE(odp_packet_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_packet_family + || !nl_policy_parse(&b, 0, ovs_packet_policy, a, + ARRAY_SIZE(ovs_packet_policy))) { return EINVAL; } - type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS - : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION - : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE + type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS + : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION : -1); if (type < 0) { return EINVAL; @@ -951,22 +1033,14 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, memset(upcall, 0, sizeof *upcall); upcall->type = type; upcall->packet = buf; - upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]); - upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]); - upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]); - upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]); - upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA] - ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA]) + upcall->packet->data = (void *) nl_attr_get(a[OVS_PACKET_ATTR_PACKET]); + upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]); + upcall->key = (void *) nl_attr_get(a[OVS_PACKET_ATTR_KEY]); + upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]); + upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA] + ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA]) : 0); - upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL] - ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL]) - : 0); - if (a[ODP_PACKET_ATTR_ACTIONS]) { - upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]); - upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]); - } - - *dp_ifindex = odp_header->dp_ifindex; + *dp_ifindex = ovs_header->dp_ifindex; return 0; } @@ -975,32 +1049,49 @@ static int dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct ofpbuf *buf; - int error; int i; - - if (!dpif->mc_sock) { - return EAGAIN; - } - - for (i = 0; i < 50; i++) { - int dp_ifindex; - - error = nl_sock_recv(dpif->mc_sock, &buf, false); - if (error) { - return error; - } - - error = parse_odp_packet(buf, upcall, &dp_ifindex); - if (!error - && dp_ifindex == dpif->dp_ifindex - && dpif->listen_mask & (1u << upcall->type)) { - return 0; - } - - ofpbuf_delete(buf); - if (error) { - return error; + int read_tries = 0; + + if (!dpif->listen_mask) { + return EAGAIN; + } + + for (i = 0; i < N_UPCALL_SOCKS; i++) { + struct nl_sock *upcall_sock; + dpif->last_read_upcall = (dpif->last_read_upcall + 1) & + (N_UPCALL_SOCKS - 1); + upcall_sock = dpif->upcall_socks[dpif->last_read_upcall]; + + if (nl_sock_woke(upcall_sock)) { + int dp_ifindex; + + for (;;) { + struct ofpbuf *buf; + int error; + + if (++read_tries > 50) { + return EAGAIN; + } + + error = nl_sock_recv(upcall_sock, &buf, false); + if (error == EAGAIN) { + break; + } else if (error) { + return error; + } + + error = parse_odp_packet(buf, upcall, &dp_ifindex); + if (!error + && dp_ifindex == dpif->dp_ifindex + && dpif->listen_mask & (1u << upcall->type)) { + return 0; + } + + ofpbuf_delete(buf); + if (error) { + return error; + } + } } } @@ -1011,8 +1102,14 @@ static void dpif_linux_recv_wait(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (dpif->mc_sock) { - nl_sock_wait(dpif->mc_sock, POLLIN); + int i; + + if (!dpif->listen_mask) { + return; + } + + for (i = 0; i < N_UPCALL_SOCKS; i++) { + nl_sock_wait(dpif->upcall_socks[i], POLLIN); } } @@ -1020,20 +1117,25 @@ static void dpif_linux_recv_purge(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); + int i; + + if (!dpif->listen_mask) { + return; + } - if (dpif->mc_sock) { - nl_sock_drain(dpif->mc_sock); + for (i = 0; i < N_UPCALL_SOCKS; i++) { + nl_sock_drain(dpif->upcall_socks[i]); } } const struct dpif_class dpif_linux_class = { "system", - NULL, /* run */ - NULL, /* wait */ dpif_linux_enumerate, dpif_linux_open, dpif_linux_close, dpif_linux_destroy, + dpif_linux_run, + dpif_linux_wait, dpif_linux_get_stats, dpif_linux_get_drop_frags, dpif_linux_set_drop_frags, @@ -1057,8 +1159,6 @@ const struct dpif_class dpif_linux_class = { dpif_linux_execute, dpif_linux_recv_get_mask, dpif_linux_recv_set_mask, - dpif_linux_get_sflow_probability, - dpif_linux_set_sflow_probability, dpif_linux_queue_to_priority, dpif_linux_recv, dpif_linux_recv_wait, @@ -1071,26 +1171,38 @@ dpif_linux_init(void) static int error = -1; if (error < 0) { - error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY, - &odp_datapath_family); + unsigned int ovs_vport_mcgroup; + + error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY, + &ovs_datapath_family); if (error) { VLOG_ERR("Generic Netlink family '%s' does not exist. " "The Open vSwitch kernel module is probably not loaded.", - ODP_DATAPATH_FAMILY); + OVS_DATAPATH_FAMILY); } if (!error) { - error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family); + error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family); } if (!error) { - error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family); + error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family); } if (!error) { - error = nl_lookup_genl_family(ODP_PACKET_FAMILY, - &odp_packet_family); + error = nl_lookup_genl_family(OVS_PACKET_FAMILY, + &ovs_packet_family); } if (!error) { error = nl_sock_create(NETLINK_GENERIC, &genl_sock); } + if (!error) { + error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP, + &ovs_vport_mcgroup, + OVS_VPORT_MCGROUP_FALLBACK_ID); + } + if (!error) { + static struct dpif_linux_vport vport; + nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup, + dpif_linux_nln_parse, &vport); + } } return error; @@ -1111,58 +1223,59 @@ dpif_linux_is_internal_device(const char *name) name, strerror(error)); } - return reply.type == ODP_VPORT_TYPE_INTERNAL; + return reply.type == OVS_VPORT_TYPE_INTERNAL; } int dpif_linux_vport_send(int dp_ifindex, uint32_t port_no, const void *data, size_t size) { - struct odp_header *execute; - struct ofpbuf *buf; - size_t actions_ofs; - int error; - - buf = ofpbuf_new(128 + size); + struct ofpbuf actions, key, packet; + struct odputil_keybuf keybuf; + struct flow flow; + uint64_t action; - nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST, - ODP_PACKET_CMD_EXECUTE, 1); + ofpbuf_use_const(&packet, data, size); + flow_extract(&packet, htonll(0), 0, &flow); - execute = ofpbuf_put_uninit(buf, sizeof *execute); - execute->dp_ifindex = dp_ifindex; + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, &flow); - nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, data, size); + ofpbuf_use_stack(&actions, &action, sizeof action); + nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no); - actions_ofs = nl_msg_start_nested(buf, ODP_PACKET_ATTR_ACTIONS); - nl_msg_put_u32(buf, ODP_ACTION_ATTR_OUTPUT, port_no); - nl_msg_end_nested(buf, actions_ofs); + return dpif_linux_execute__(dp_ifindex, 0, key.data, key.size, + actions.data, actions.size, &packet); +} - error = nl_sock_transact(genl_sock, buf, NULL); - ofpbuf_delete(buf); - return error; +static bool +dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_) +{ + struct dpif_linux_vport *vport = vport_; + return dpif_linux_vport_from_ofpbuf(vport, buf) == 0; } static void -dpif_linux_port_changed(const struct rtnetlink_link_change *change, - void *dpif_) +dpif_linux_port_changed(const void *vport_, void *dpif_) { + const struct dpif_linux_vport *vport = vport_; struct dpif_linux *dpif = dpif_; - if (change) { - if (change->master_ifindex == dpif->dp_ifindex - && (change->nlmsg_type == RTM_NEWLINK - || change->nlmsg_type == RTM_DELLINK)) - { - /* Our datapath changed, either adding a new port or deleting an - * existing one. */ - sset_add(&dpif->changed_ports, change->ifname); + if (vport) { + if (vport->dp_ifindex == dpif->dp_ifindex + && (vport->cmd == OVS_VPORT_CMD_NEW + || vport->cmd == OVS_VPORT_CMD_DEL + || vport->cmd == OVS_VPORT_CMD_SET)) { + VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8, + dpif->dpif.full_name, vport->name, vport->cmd); + sset_add(&dpif->changed_ports, vport->name); } } else { dpif->change_error = true; } } -/* Parses the contents of 'buf', which contains a "struct odp_header" followed +/* Parses the contents of 'buf', which contains a "struct ovs_header" followed * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a * positive errno value. * @@ -1172,26 +1285,24 @@ static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, const struct ofpbuf *buf) { - static const struct nl_policy odp_vport_policy[] = { - [ODP_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 }, - [ODP_VPORT_ATTR_TYPE] = { .type = NL_A_U32 }, - [ODP_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, - [ODP_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC, - .min_len = sizeof(struct rtnl_link_stats64), - .max_len = sizeof(struct rtnl_link_stats64), + static const struct nl_policy ovs_vport_policy[] = { + [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 }, + [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 }, + [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, + [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 }, + [OVS_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC, + .min_len = sizeof(struct ovs_vport_stats), + .max_len = sizeof(struct ovs_vport_stats), .optional = true }, - [ODP_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC, + [OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC, .min_len = ETH_ADDR_LEN, .max_len = ETH_ADDR_LEN, .optional = true }, - [ODP_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true }, - [ODP_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, - [ODP_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true }, - [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true }, + [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, }; - struct nlattr *a[ARRAY_SIZE(odp_vport_policy)]; - struct odp_header *odp_header; + struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)]; + struct ovs_header *ovs_header; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -1201,95 +1312,77 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, ofpbuf_use_const(&b, buf->data, buf->size); nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_vport_family - || !nl_policy_parse(&b, 0, odp_vport_policy, a, - ARRAY_SIZE(odp_vport_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_vport_family + || !nl_policy_parse(&b, 0, ovs_vport_policy, a, + ARRAY_SIZE(ovs_vport_policy))) { return EINVAL; } vport->cmd = genl->cmd; - vport->dp_ifindex = odp_header->dp_ifindex; - vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); - vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]); - vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]); - if (a[ODP_VPORT_ATTR_STATS]) { - vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]); - } - if (a[ODP_VPORT_ATTR_ADDRESS]) { - vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]); - } - if (a[ODP_VPORT_ATTR_MTU]) { - vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]); - } else { - vport->mtu = INT_MAX; + vport->dp_ifindex = ovs_header->dp_ifindex; + vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); + vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]); + vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]); + if (a[OVS_VPORT_ATTR_UPCALL_PID]) { + vport->upcall_pid = nl_attr_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); } - if (a[ODP_VPORT_ATTR_OPTIONS]) { - vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]); - vport->options_len = nl_attr_get_size(a[ODP_VPORT_ATTR_OPTIONS]); + if (a[OVS_VPORT_ATTR_STATS]) { + vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]); } - if (a[ODP_VPORT_ATTR_IFINDEX]) { - vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]); + if (a[OVS_VPORT_ATTR_ADDRESS]) { + vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]); } - if (a[ODP_VPORT_ATTR_IFLINK]) { - vport->iflink = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFLINK]); + if (a[OVS_VPORT_ATTR_OPTIONS]) { + vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]); + vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]); } return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_header" +/* Appends to 'buf' (which must initially be empty) a "struct ovs_header" * followed by Netlink attributes corresponding to 'vport'. */ static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, struct ofpbuf *buf) { - struct odp_header *odp_header; + struct ovs_header *ovs_header; - nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO, + nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO, vport->cmd, 1); - odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_ifindex = vport->dp_ifindex; + ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); + ovs_header->dp_ifindex = vport->dp_ifindex; if (vport->port_no != UINT32_MAX) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no); + nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no); } - if (vport->type != ODP_VPORT_TYPE_UNSPEC) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_TYPE, vport->type); + if (vport->type != OVS_VPORT_TYPE_UNSPEC) { + nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type); } if (vport->name) { - nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name); + nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name); } + nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid); + if (vport->stats) { - nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS, + nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS, vport->stats, sizeof *vport->stats); } if (vport->address) { - nl_msg_put_unspec(buf, ODP_VPORT_ATTR_ADDRESS, + nl_msg_put_unspec(buf, OVS_VPORT_ATTR_ADDRESS, vport->address, ETH_ADDR_LEN); } - if (vport->mtu && vport->mtu != INT_MAX) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu); - } - if (vport->options) { - nl_msg_put_nested(buf, ODP_VPORT_ATTR_OPTIONS, + nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS, vport->options, vport->options_len); } - - if (vport->ifindex) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFINDEX, vport->ifindex); - } - - if (vport->iflink) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink); - } } /* Clears 'vport' to "empty" values. */ @@ -1303,7 +1396,7 @@ dpif_linux_vport_init(struct dpif_linux_vport *vport) /* Executes 'request' in the kernel datapath. If the command fails, returns a * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the - * result of the command is expected to be an odp_vport also, which is decoded + * result of the command is expected to be an ovs_vport also, which is decoded * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ int @@ -1353,13 +1446,13 @@ dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply, struct dpif_linux_vport request; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_CMD_GET; + request.cmd = OVS_VPORT_CMD_GET; request.name = name; return dpif_linux_vport_transact(&request, reply, bufp); } -/* Parses the contents of 'buf', which contains a "struct odp_header" followed +/* Parses the contents of 'buf', which contains a "struct ovs_header" followed * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a * positive errno value. * @@ -1368,19 +1461,17 @@ dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply, static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) { - static const struct nl_policy odp_datapath_policy[] = { - [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, - [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC, - .min_len = sizeof(struct odp_stats), - .max_len = sizeof(struct odp_stats), + static const struct nl_policy ovs_datapath_policy[] = { + [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, + [OVS_DP_ATTR_STATS] = { .type = NL_A_UNSPEC, + .min_len = sizeof(struct ovs_dp_stats), + .max_len = sizeof(struct ovs_dp_stats), .optional = true }, - [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true }, - [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true }, - [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true }, + [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true }, }; - struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)]; - struct odp_header *odp_header; + struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)]; + struct ovs_header *ovs_header; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -1390,56 +1481,25 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) ofpbuf_use_const(&b, buf->data, buf->size); nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_datapath_family - || !nl_policy_parse(&b, 0, odp_datapath_policy, a, - ARRAY_SIZE(odp_datapath_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_datapath_family + || !nl_policy_parse(&b, 0, ovs_datapath_policy, a, + ARRAY_SIZE(ovs_datapath_policy))) { return EINVAL; } dp->cmd = genl->cmd; - dp->dp_ifindex = odp_header->dp_ifindex; - dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]); - if (a[ODP_DP_ATTR_STATS]) { + dp->dp_ifindex = ovs_header->dp_ifindex; + dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]); + if (a[OVS_DP_ATTR_STATS]) { /* Can't use structure assignment because Netlink doesn't ensure * sufficient alignment for 64-bit members. */ - memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]), + memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]), sizeof dp->stats); } - if (a[ODP_DP_ATTR_IPV4_FRAGS]) { - dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); - } - if (a[ODP_DP_ATTR_SAMPLING]) { - dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]); - } - - if (a[ODP_DP_ATTR_MCGROUPS]) { - static const struct nl_policy odp_mcgroup_policy[] = { - [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true }, - [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true }, - [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true }, - }; - - struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)]; - - if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy, - mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) { - return EINVAL; - } - - if (mcgroups[ODP_PACKET_CMD_MISS]) { - dp->mcgroups[DPIF_UC_MISS] - = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]); - } - if (mcgroups[ODP_PACKET_CMD_ACTION]) { - dp->mcgroups[DPIF_UC_ACTION] - = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]); - } - if (mcgroups[ODP_PACKET_CMD_SAMPLE]) { - dp->mcgroups[DPIF_UC_SAMPLE] - = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]); - } + if (a[OVS_DP_ATTR_IPV4_FRAGS]) { + dp->ipv4_frags = nl_attr_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]); } return 0; @@ -1449,26 +1509,24 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) static void dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) { - struct odp_header *odp_header; + struct ovs_header *ovs_header; - nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family, + nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family, NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1); - odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_ifindex = dp->dp_ifindex; + ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); + ovs_header->dp_ifindex = dp->dp_ifindex; if (dp->name) { - nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name); + nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name); } - /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */ + nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, dp->upcall_pid); - if (dp->ipv4_frags) { - nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags); - } + /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */ - if (dp->sampling) { - nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling); + if (dp->ipv4_frags) { + nl_msg_put_u32(buf, OVS_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags); } } @@ -1486,7 +1544,7 @@ dpif_linux_dp_dump_start(struct nl_dump *dump) struct ofpbuf *buf; dpif_linux_dp_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; buf = ofpbuf_new(1024); dpif_linux_dp_to_ofpbuf(&request, buf); @@ -1538,13 +1596,13 @@ dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, struct dpif_linux_dp request; dpif_linux_dp_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&request, reply, bufp); } -/* Parses the contents of 'buf', which contains a "struct odp_header" followed +/* Parses the contents of 'buf', which contains a "struct ovs_header" followed * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a * positive errno value. * @@ -1554,20 +1612,21 @@ static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, const struct ofpbuf *buf) { - static const struct nl_policy odp_flow_policy[] = { - [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED }, - [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, - [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC, - .min_len = sizeof(struct odp_flow_stats), - .max_len = sizeof(struct odp_flow_stats), + static const struct nl_policy ovs_flow_policy[] = { + [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED }, + [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, + [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NL_A_U32 }, + [OVS_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC, + .min_len = sizeof(struct ovs_flow_stats), + .max_len = sizeof(struct ovs_flow_stats), .optional = true }, - [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, - [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, - /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */ + [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, + [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, + /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */ }; - struct nlattr *a[ARRAY_SIZE(odp_flow_policy)]; - struct odp_header *odp_header; + struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)]; + struct ovs_header *ovs_header; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -1577,65 +1636,70 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, ofpbuf_use_const(&b, buf->data, buf->size); nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_flow_family - || !nl_policy_parse(&b, 0, odp_flow_policy, a, - ARRAY_SIZE(odp_flow_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_flow_family + || !nl_policy_parse(&b, 0, ovs_flow_policy, a, + ARRAY_SIZE(ovs_flow_policy))) { return EINVAL; } flow->nlmsg_flags = nlmsg->nlmsg_flags; - flow->dp_ifindex = odp_header->dp_ifindex; - flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]); - flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]); - if (a[ODP_FLOW_ATTR_ACTIONS]) { - flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]); - flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]); + flow->dp_ifindex = ovs_header->dp_ifindex; + flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]); + flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]); + if (a[OVS_FLOW_ATTR_ACTIONS]) { + flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]); + flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]); } - if (a[ODP_FLOW_ATTR_STATS]) { - flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]); + if (a[OVS_FLOW_ATTR_UPCALL_PID]) { + flow->upcall_pid = nl_attr_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]); } - if (a[ODP_FLOW_ATTR_TCP_FLAGS]) { - flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]); + if (a[OVS_FLOW_ATTR_STATS]) { + flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]); } - if (a[ODP_FLOW_ATTR_USED]) { - flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]); + if (a[OVS_FLOW_ATTR_TCP_FLAGS]) { + flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]); + } + if (a[OVS_FLOW_ATTR_USED]) { + flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]); } return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_header" +/* Appends to 'buf' (which must initially be empty) a "struct ovs_header" * followed by Netlink attributes corresponding to 'flow'. */ static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, struct ofpbuf *buf) { - struct odp_header *odp_header; + struct ovs_header *ovs_header; - nl_msg_put_genlmsghdr(buf, 0, odp_flow_family, + nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family, NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags, flow->cmd, 1); - odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_ifindex = flow->dp_ifindex; + ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); + ovs_header->dp_ifindex = flow->dp_ifindex; if (flow->key_len) { - nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len); + nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len); } if (flow->actions || flow->actions_len) { - nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS, + nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS, flow->actions, flow->actions_len); } + nl_msg_put_u32(buf, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid); + /* We never need to send these to the kernel. */ assert(!flow->stats); assert(!flow->tcp_flags); assert(!flow->used); if (flow->clear) { - nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR); + nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR); } }