X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=c06bb897ff7deaa7a36d3f6c388a617556888da1;hb=c5dada75bb41781c9277f6aa24cb8e4a07c8a72f;hp=0f188c654a61ee6c79e9511185ca6b89c0b6f800;hpb=a24a65747a4aaf4929e8aa51f678d3ee7d57a1e3;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index 0f188c654..c06bb897f 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -28,7 +28,10 @@ #include #include #include +#include #include +#include +#include #include #include @@ -63,6 +66,7 @@ BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS)); enum { N_UPCALL_SOCKS = 16 }; BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS)); +BUILD_ASSERT_DECL(N_UPCALL_SOCKS <= 32); /* We use a 32-bit word as a mask. */ /* This ethtool flag was introduced in Linux 2.6.24, so it might be * missing if we have old headers. */ @@ -79,7 +83,6 @@ struct dpif_linux_dp { const char *name; /* OVS_DP_ATTR_NAME. */ const 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 *); @@ -102,9 +105,9 @@ struct dpif_linux_flow { /* Attributes. * - * The 'stats' and 'used' members point to 64-bit data that might only be - * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to - * access their values. + * The 'stats' member points to 64-bit data that might only be aligned on + * 32-bit boundaries, so get_unaligned_u64() should be used to access its + * values. * * 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. */ @@ -112,10 +115,9 @@ struct dpif_linux_flow { size_t key_len; const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */ size_t actions_len; - const 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. */ + const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */ bool clear; /* OVS_FLOW_ATTR_CLEAR. */ }; @@ -124,7 +126,7 @@ static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *, const struct ofpbuf *); static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *, struct ofpbuf *); -static int dpif_linux_flow_transact(const struct dpif_linux_flow *request, +static int dpif_linux_flow_transact(struct dpif_linux_flow *request, struct dpif_linux_flow *reply, struct ofpbuf **bufp); static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *, @@ -137,8 +139,8 @@ struct dpif_linux { /* Upcall messages. */ struct nl_sock *upcall_socks[N_UPCALL_SOCKS]; - int last_read_upcall; - unsigned int listen_mask; + uint32_t ready_mask; /* 1-bit for each sock with unread messages. */ + int epoll_fd; /* epoll fd that includes the upcall socks. */ /* Change notification. */ struct sset changed_ports; /* Ports that have changed. */ @@ -168,9 +170,7 @@ static int dpif_linux_init(void); 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 uint32_t dpif_linux_port_get_pid(const struct dpif *, uint16_t port_no); static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, struct ofpbuf *); @@ -236,6 +236,7 @@ dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, { struct dpif_linux_dp dp_request, dp; struct ofpbuf *buf; + uint32_t upcall_pid; int error; error = dpif_linux_init(); @@ -245,7 +246,13 @@ 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 ? OVS_DP_CMD_NEW : OVS_DP_CMD_GET; + if (create) { + dp_request.cmd = OVS_DP_CMD_NEW; + upcall_pid = 0; + dp_request.upcall_pid = &upcall_pid; + } else { + dp_request.cmd = OVS_DP_CMD_GET; + } dp_request.name = name; error = dpif_linux_dp_transact(&dp_request, &dp, &buf); if (error) { @@ -266,6 +273,7 @@ open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp) dpif = xzalloc(sizeof *dpif); dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed, dpif); + dpif->epoll_fd = -1; dpif_init(&dpif->dpif, &dpif_linux_class, dp->name, dp->dp_ifindex, dp->dp_ifindex); @@ -286,6 +294,10 @@ destroy_upcall_socks(struct dpif_linux *dpif) { int i; + if (dpif->epoll_fd >= 0) { + close(dpif->epoll_fd); + dpif->epoll_fd = -1; + } for (i = 0; i < N_UPCALL_SOCKS; i++) { nl_sock_destroy(dpif->upcall_socks[i]); dpif->upcall_socks[i] = NULL; @@ -341,7 +353,6 @@ dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) error = dpif_linux_dp_get(dpif_, &dp, &buf); if (!error) { - 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; @@ -351,34 +362,6 @@ dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) return error; } -static int -dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) -{ - struct dpif_linux_dp dp; - struct ofpbuf *buf; - int error; - - error = dpif_linux_dp_get(dpif_, &dp, &buf); - if (!error) { - *drop_fragsp = dp.ipv4_frags == OVS_DP_FRAG_DROP; - ofpbuf_delete(buf); - } - return error; -} - -static int -dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) -{ - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_dp dp; - - dpif_linux_dp_init(&dp); - dp.cmd = OVS_DP_CMD_SET; - dp.dp_ifindex = dpif->dp_ifindex; - dp.ipv4_frags = drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO; - return dpif_linux_dp_transact(&dp, NULL, NULL); -} - static int dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, uint16_t *port_nop) @@ -418,7 +401,7 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, uint32_t upcall_pid; request.port_no = dpif_linux_pop_port(dpif); - upcall_pid = get_upcall_pid_port(dpif, request.port_no); + upcall_pid = dpif_linux_port_get_pid(dpif_, request.port_no); request.upcall_pid = &upcall_pid; error = dpif_linux_vport_transact(&request, &reply, &buf); @@ -500,6 +483,19 @@ dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED) return 1024; } +static uint32_t +dpif_linux_port_get_pid(const struct dpif *dpif_, uint16_t port_no) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + if (dpif->epoll_fd < 0) { + return 0; + } else { + int idx = port_no & (N_UPCALL_SOCKS - 1); + return nl_sock_pid(dpif->upcall_socks[idx]); + } +} + static int dpif_linux_flow_flush(struct dpif *dpif_) { @@ -660,39 +656,42 @@ dpif_linux_flow_get(const struct dpif *dpif_, return error; } -static int -dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags, - const struct nlattr *key, size_t key_len, - const struct nlattr *actions, size_t actions_len, - struct dpif_flow_stats *stats) +static void +dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put, + struct dpif_linux_flow *request) { + static struct nlattr dummy_action; + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + dpif_linux_flow_init(request); + request->cmd = (put->flags & DPIF_FP_CREATE + ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET); + request->dp_ifindex = dpif->dp_ifindex; + request->key = put->key; + request->key_len = put->key_len; + /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */ + request->actions = put->actions ? put->actions : &dummy_action; + request->actions_len = put->actions_len; + if (put->flags & DPIF_FP_ZERO_STATS) { + request->clear = true; + } + request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; +} + +static int +dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put) +{ struct dpif_linux_flow request, reply; - struct nlattr dummy_action; - uint32_t upcall_pid; struct ofpbuf *buf; int error; - upcall_pid = get_upcall_pid_flow(dpif, key, key_len); - - dpif_linux_flow_init(&request); - 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 OVS_FLOW_ATTR_ACTIONS will always be included. */ - request.actions = actions ? actions : &dummy_action; - request.actions_len = actions_len; - request.upcall_pid = &upcall_pid; - if (flags & DPIF_FP_ZERO_STATS) { - request.clear = true; - } - request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; + dpif_linux_init_flow_put(dpif_, put, &request); error = dpif_linux_flow_transact(&request, - stats ? &reply : NULL, - stats ? &buf : NULL); - if (!error && stats) { - dpif_linux_flow_get_stats(&reply, stats); + put->stats ? &reply : NULL, + put->stats ? &buf : NULL); + if (!error && put->stats) { + dpif_linux_flow_get_stats(&reply, put->stats); ofpbuf_delete(buf); } return error; @@ -814,105 +813,132 @@ dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) return error; } -static int -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) +static struct ofpbuf * +dpif_linux_encode_execute(int dp_ifindex, + const struct dpif_execute *d_exec) { - struct ovs_header *execute; + struct ovs_header *k_exec; struct ofpbuf *buf; - int error; - buf = ofpbuf_new(128 + actions_len + packet->size); + buf = ofpbuf_new(128 + d_exec->actions_len + d_exec->packet->size); nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST, - OVS_PACKET_CMD_EXECUTE, 1); + OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION); - execute = ofpbuf_put_uninit(buf, sizeof *execute); - execute->dp_ifindex = dp_ifindex; + k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec); + k_exec->dp_ifindex = dp_ifindex; - 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); + nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, + d_exec->packet->data, d_exec->packet->size); + nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len); + nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, + d_exec->actions, d_exec->actions_len); - error = nl_sock_transact(genl_sock, buf, NULL); - ofpbuf_delete(buf); - return error; + return buf; } 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) +dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - uint32_t upcall_pid = get_upcall_pid_flow(dpif, key, key_len); + struct ofpbuf *request; + int error; + + request = dpif_linux_encode_execute(dp_ifindex, execute); + error = nl_sock_transact(genl_sock, request, NULL); + ofpbuf_delete(request); - return dpif_linux_execute__(dpif->dp_ifindex, upcall_pid, key, key_len, - actions, actions_len, packet); + return error; } static int -dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) +dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - *listen_mask = dpif->listen_mask; - return 0; -} -static uint32_t -get_upcall_pid_port__(struct dpif_linux *dpif, uint32_t port) -{ - int idx = port & (N_UPCALL_SOCKS - 1); - return nl_sock_pid(dpif->upcall_socks[idx]); + return dpif_linux_execute__(dpif->dp_ifindex, execute); } -static uint32_t -get_upcall_pid_port(struct dpif_linux *dpif, uint32_t port) +static void +dpif_linux_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) { - if (!(dpif->listen_mask & (1u << DPIF_UC_MISS))) { - return 0; + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct nl_transaction **txnsp; + struct nl_transaction *txns; + size_t i; + + txns = xmalloc(n_ops * sizeof *txns); + for (i = 0; i < n_ops; i++) { + struct nl_transaction *txn = &txns[i]; + struct dpif_op *op = ops[i]; + + if (op->type == DPIF_OP_FLOW_PUT) { + struct dpif_flow_put *put = &op->u.flow_put; + struct dpif_linux_flow request; + + dpif_linux_init_flow_put(dpif_, put, &request); + if (put->stats) { + request.nlmsg_flags |= NLM_F_ECHO; + } + txn->request = ofpbuf_new(1024); + dpif_linux_flow_to_ofpbuf(&request, txn->request); + } else if (op->type == DPIF_OP_EXECUTE) { + struct dpif_execute *execute = &op->u.execute; + + txn->request = dpif_linux_encode_execute(dpif->dp_ifindex, + execute); + } else { + NOT_REACHED(); + } } - return get_upcall_pid_port__(dpif, port); -} + txnsp = xmalloc(n_ops * sizeof *txnsp); + for (i = 0; i < n_ops; i++) { + txnsp[i] = &txns[i]; + } -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; + nl_sock_transact_multiple(genl_sock, txnsp, n_ops); - if (!(dpif->listen_mask & (1u << DPIF_UC_ACTION))) { - return 0; - } + free(txnsp); - nla = nl_attr_find__(key, key_len, OVS_KEY_ATTR_IN_PORT); - if (nla) { - port = nl_attr_get_u32(nla); - } else { - port = random_uint32(); - } + for (i = 0; i < n_ops; i++) { + struct nl_transaction *txn = &txns[i]; + struct dpif_op *op = ops[i]; + + if (op->type == DPIF_OP_FLOW_PUT) { + struct dpif_flow_put *put = &op->u.flow_put; + int error = txn->error; + + if (!error && put->stats) { + struct dpif_linux_flow reply; + + error = dpif_linux_flow_from_ofpbuf(&reply, txn->reply); + if (!error) { + dpif_linux_flow_get_stats(&reply, put->stats); + } + } + op->error = error; + } else if (op->type == DPIF_OP_EXECUTE) { + op->error = txn->error; + } else { + NOT_REACHED(); + } - return get_upcall_pid_port__(dpif, port); + ofpbuf_delete(txn->request); + ofpbuf_delete(txn->reply); + } + free(txns); } static void -set_upcall_pids(struct dpif_linux *dpif) +set_upcall_pids(struct dpif *dpif_) { - struct dpif_port port; + struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_port_dump port_dump; - struct dpif_flow_dump flow_dump; - const struct nlattr *key; - size_t key_len; + struct dpif_port port; int error; DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) { - uint32_t upcall_pid = get_upcall_pid_port(dpif, port.port_no); + uint32_t upcall_pid = dpif_linux_port_get_pid(dpif_, port.port_no); struct dpif_linux_vport vport_request; dpif_linux_vport_init(&vport_request); @@ -930,54 +956,52 @@ set_upcall_pids(struct dpif_linux *dpif) dpif_name(&dpif->dpif), strerror(error)); } } - - dpif_flow_dump_start(&flow_dump, &dpif->dpif); - while (dpif_flow_dump_next(&flow_dump, &key, &key_len, - NULL, NULL, NULL)) { - uint32_t upcall_pid = get_upcall_pid_flow(dpif, key, key_len); - 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 = &upcall_pid; - 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_recv_set_mask(struct dpif *dpif_, int listen_mask) +dpif_linux_recv_set(struct dpif *dpif_, bool enable) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (listen_mask == dpif->listen_mask) { + if ((dpif->epoll_fd >= 0) == enable) { return 0; } - if (!listen_mask) { + if (!enable) { destroy_upcall_socks(dpif); - } else if (!dpif->listen_mask) { + } else { int i; int error; + dpif->epoll_fd = epoll_create(N_UPCALL_SOCKS); + if (dpif->epoll_fd < 0) { + return errno; + } + for (i = 0; i < N_UPCALL_SOCKS; i++) { + struct epoll_event event; + error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]); if (error) { destroy_upcall_socks(dpif); return error; } + + memset(&event, 0, sizeof event); + event.events = EPOLLIN; + event.data.u32 = i; + if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, + nl_sock_fd(dpif->upcall_socks[i]), &event) < 0) { + error = errno; + destroy_upcall_socks(dpif); + return error; + } } + + dpif->ready_mask = 0; } - dpif->listen_mask = listen_mask; - set_upcall_pids(dpif); + set_upcall_pids(dpif_); return 0; } @@ -1053,48 +1077,60 @@ static int dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int i; int read_tries = 0; - if (!dpif->listen_mask) { + if (dpif->epoll_fd < 0) { 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 (!dpif->ready_mask) { + struct epoll_event events[N_UPCALL_SOCKS]; + int retval; + int i; - if (nl_sock_woke(upcall_sock)) { - int dp_ifindex; + do { + retval = epoll_wait(dpif->epoll_fd, events, N_UPCALL_SOCKS, 0); + } while (retval < 0 && errno == EINTR); + if (retval < 0) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); + VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno)); + } - for (;;) { - struct ofpbuf *buf; - int error; + for (i = 0; i < retval; i++) { + dpif->ready_mask |= 1u << events[i].data.u32; + } + } - if (++read_tries > 50) { - return EAGAIN; - } + while (dpif->ready_mask) { + int indx = ffs(dpif->ready_mask) - 1; + struct nl_sock *upcall_sock = dpif->upcall_socks[indx]; - error = nl_sock_recv(upcall_sock, &buf, false); - if (error == EAGAIN) { - break; - } else if (error) { - return error; - } + dpif->ready_mask &= ~(1u << indx); - error = parse_odp_packet(buf, upcall, &dp_ifindex); - if (!error - && dp_ifindex == dpif->dp_ifindex - && dpif->listen_mask & (1u << upcall->type)) { - return 0; - } + for (;;) { + struct ofpbuf *buf; + int dp_ifindex; + int error; - ofpbuf_delete(buf); - if (error) { - return 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) { + return 0; + } + + ofpbuf_delete(buf); + if (error) { + return error; } } } @@ -1106,15 +1142,12 @@ static void dpif_linux_recv_wait(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int i; - if (!dpif->listen_mask) { + if (dpif->epoll_fd < 0) { return; } - for (i = 0; i < N_UPCALL_SOCKS; i++) { - nl_sock_wait(dpif->upcall_socks[i], POLLIN); - } + poll_fd_wait(dpif->epoll_fd, POLLIN); } static void @@ -1123,7 +1156,7 @@ dpif_linux_recv_purge(struct dpif *dpif_) struct dpif_linux *dpif = dpif_linux_cast(dpif_); int i; - if (!dpif->listen_mask) { + if (dpif->epoll_fd < 0) { return; } @@ -1141,13 +1174,12 @@ const struct dpif_class dpif_linux_class = { dpif_linux_run, dpif_linux_wait, dpif_linux_get_stats, - dpif_linux_get_drop_frags, - dpif_linux_set_drop_frags, dpif_linux_port_add, dpif_linux_port_del, dpif_linux_port_query_by_number, dpif_linux_port_query_by_name, dpif_linux_get_max_ports, + dpif_linux_port_get_pid, dpif_linux_port_dump_start, dpif_linux_port_dump_next, dpif_linux_port_dump_done, @@ -1161,8 +1193,8 @@ const struct dpif_class dpif_linux_class = { dpif_linux_flow_dump_next, dpif_linux_flow_dump_done, dpif_linux_execute, - dpif_linux_recv_get_mask, - dpif_linux_recv_set_mask, + dpif_linux_operate, + dpif_linux_recv_set, dpif_linux_queue_to_priority, dpif_linux_recv, dpif_linux_recv_wait, @@ -1236,11 +1268,12 @@ dpif_linux_vport_send(int dp_ifindex, uint32_t port_no, { struct ofpbuf actions, key, packet; struct odputil_keybuf keybuf; + struct dpif_execute execute; struct flow flow; uint64_t action; ofpbuf_use_const(&packet, data, size); - flow_extract(&packet, htonll(0), 0, &flow); + flow_extract(&packet, 0, htonll(0), 0, &flow); ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); odp_flow_key_from_flow(&key, &flow); @@ -1248,8 +1281,12 @@ dpif_linux_vport_send(int dp_ifindex, uint32_t port_no, ofpbuf_use_stack(&actions, &action, sizeof action); nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no); - return dpif_linux_execute__(dp_ifindex, 0, key.data, key.size, - actions.data, actions.size, &packet); + execute.key = key.data; + execute.key_len = key.size; + execute.actions = actions.data; + execute.actions_len = actions.size; + execute.packet = &packet; + return dpif_linux_execute__(dp_ifindex, &execute); } static bool @@ -1294,9 +1331,7 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, [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), + [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats), .optional = true }, [OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC, .min_len = ETH_ADDR_LEN, @@ -1354,7 +1389,7 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, struct ovs_header *ovs_header; nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO, - vport->cmd, 1); + vport->cmd, OVS_VPORT_VERSION); ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); ovs_header->dp_ifindex = vport->dp_ifindex; @@ -1469,11 +1504,8 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) { 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), + [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats), .optional = true }, - [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true }, }; struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)]; @@ -1504,9 +1536,6 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]), sizeof dp->stats); } - if (a[OVS_DP_ATTR_IPV4_FRAGS]) { - dp->ipv4_frags = nl_attr_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]); - } return 0; } @@ -1518,7 +1547,8 @@ dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) struct ovs_header *ovs_header; nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family, - NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1); + NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, + OVS_DATAPATH_VERSION); ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); ovs_header->dp_ifindex = dp->dp_ifindex; @@ -1532,10 +1562,6 @@ dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) } /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */ - - if (dp->ipv4_frags) { - nl_msg_put_u32(buf, OVS_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags); - } } /* Clears 'dp' to "empty" values. */ @@ -1623,10 +1649,7 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, 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), + [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats), .optional = true }, [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, @@ -1660,9 +1683,6 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]); flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]); } - if (a[OVS_FLOW_ATTR_UPCALL_PID]) { - flow->upcall_pid = nl_attr_get(a[OVS_FLOW_ATTR_UPCALL_PID]); - } if (a[OVS_FLOW_ATTR_STATS]) { flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]); } @@ -1684,8 +1704,8 @@ dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, struct ovs_header *ovs_header; nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family, - NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags, - flow->cmd, 1); + NLM_F_REQUEST | flow->nlmsg_flags, + flow->cmd, OVS_FLOW_VERSION); ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); ovs_header->dp_ifindex = flow->dp_ifindex; @@ -1699,10 +1719,6 @@ dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, flow->actions, flow->actions_len); } - if (flow->upcall_pid) { - 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); @@ -1727,7 +1743,7 @@ dpif_linux_flow_init(struct dpif_linux_flow *flow) * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply * is no longer needed ('reply' will contain pointers into '*bufp'). */ static int -dpif_linux_flow_transact(const struct dpif_linux_flow *request, +dpif_linux_flow_transact(struct dpif_linux_flow *request, struct dpif_linux_flow *reply, struct ofpbuf **bufp) { struct ofpbuf *request_buf; @@ -1735,6 +1751,10 @@ dpif_linux_flow_transact(const struct dpif_linux_flow *request, assert((reply != NULL) == (bufp != NULL)); + if (reply) { + request->nlmsg_flags |= NLM_F_ECHO; + } + request_buf = ofpbuf_new(1024); dpif_linux_flow_to_ofpbuf(request, request_buf); error = nl_sock_transact(genl_sock, request_buf, bufp); @@ -1764,7 +1784,6 @@ dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow, stats->n_packets = 0; stats->n_bytes = 0; } - stats->used = flow->used ? get_unaligned_u64(flow->used) : 0; + stats->used = flow->used ? get_32aligned_u64(flow->used) : 0; stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0; } -