X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=c06bb897ff7deaa7a36d3f6c388a617556888da1;hb=6d4605fd85630eb3a9ec743763114f286373a544;hp=2466f9159e6dba75b1b15bdc05d49fec463a8c28;hpb=7257b535ab8e5fafd811c5f6788205eefdd44948;p=openvswitch diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index 2466f915..c06bb897 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. */ @@ -135,8 +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. */ @@ -166,9 +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 dpif_linux_port_get_pid__(const struct dpif *, - uint16_t port_no, - enum dpif_upcall_type); +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 *); @@ -271,6 +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); @@ -291,6 +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; @@ -394,8 +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 = dpif_linux_port_get_pid__(dpif_, request.port_no, - DPIF_UC_MISS); + upcall_pid = dpif_linux_port_get_pid(dpif_, request.port_no); request.upcall_pid = &upcall_pid; error = dpif_linux_vport_transact(&request, &reply, &buf); @@ -478,12 +484,11 @@ dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED) } static uint32_t -dpif_linux_port_get_pid__(const struct dpif *dpif_, uint16_t port_no, - enum dpif_upcall_type upcall_type) +dpif_linux_port_get_pid(const struct dpif *dpif_, uint16_t port_no) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (!(dpif->listen_mask & (1u << upcall_type))) { + if (dpif->epoll_fd < 0) { return 0; } else { int idx = port_no & (N_UPCALL_SOCKS - 1); @@ -491,12 +496,6 @@ dpif_linux_port_get_pid__(const struct dpif *dpif_, uint16_t port_no, } } -static uint32_t -dpif_linux_port_get_pid(const struct dpif *dpif, uint16_t port_no) -{ - return dpif_linux_port_get_pid__(dpif, port_no, DPIF_UC_ACTION); -} - static int dpif_linux_flow_flush(struct dpif *dpif_) { @@ -658,9 +657,7 @@ dpif_linux_flow_get(const struct dpif *dpif_, } static void -dpif_linux_init_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, +dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put, struct dpif_linux_flow *request) { static struct nlattr dummy_action; @@ -668,37 +665,33 @@ dpif_linux_init_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags, struct dpif_linux *dpif = dpif_linux_cast(dpif_); dpif_linux_flow_init(request); - request->cmd = (flags & DPIF_FP_CREATE + request->cmd = (put->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; + request->key = put->key; + request->key_len = put->key_len; /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */ - request->actions = actions ? actions : &dummy_action; - request->actions_len = actions_len; - if (flags & DPIF_FP_ZERO_STATS) { + 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 = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; + request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; } 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) +dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put) { struct dpif_linux_flow request, reply; struct ofpbuf *buf; int error; - dpif_linux_init_flow_put(dpif_, flags, key, key_len, actions, actions_len, - &request); + 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; @@ -822,39 +815,35 @@ dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) static struct ofpbuf * dpif_linux_encode_execute(int dp_ifindex, - const struct nlattr *key, size_t key_len, - const struct nlattr *actions, size_t actions_len, - const struct ofpbuf *packet) + const struct dpif_execute *d_exec) { - struct ovs_header *execute; + struct ovs_header *k_exec; struct ofpbuf *buf; - 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_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); return buf; } static int -dpif_linux_execute__(int dp_ifindex, 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 ofpbuf *request; int error; - request = dpif_linux_encode_execute(dp_ifindex, - key, key_len, actions, actions_len, - packet); + request = dpif_linux_encode_execute(dp_ifindex, execute); error = nl_sock_transact(genl_sock, request, NULL); ofpbuf_delete(request); @@ -862,19 +851,15 @@ dpif_linux_execute__(int dp_ifindex, const struct nlattr *key, size_t key_len, } 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(struct dpif *dpif_, const struct dpif_execute *execute) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - return dpif_linux_execute__(dpif->dp_ifindex, key, key_len, - actions, actions_len, packet); + return dpif_linux_execute__(dpif->dp_ifindex, execute); } static void -dpif_linux_operate(struct dpif *dpif_, union dpif_op **ops, size_t n_ops) +dpif_linux_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct nl_transaction **txnsp; @@ -884,26 +869,23 @@ dpif_linux_operate(struct dpif *dpif_, union dpif_op **ops, size_t n_ops) txns = xmalloc(n_ops * sizeof *txns); for (i = 0; i < n_ops; i++) { struct nl_transaction *txn = &txns[i]; - union dpif_op *op = ops[i]; + struct dpif_op *op = ops[i]; if (op->type == DPIF_OP_FLOW_PUT) { - struct dpif_flow_put *put = &op->flow_put; + struct dpif_flow_put *put = &op->u.flow_put; struct dpif_linux_flow request; - dpif_linux_init_flow_put(dpif_, put->flags, put->key, put->key_len, - put->actions, put->actions_len, - &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->execute; + struct dpif_execute *execute = &op->u.execute; - txn->request = dpif_linux_encode_execute( - dpif->dp_ifindex, execute->key, execute->key_len, - execute->actions, execute->actions_len, execute->packet); + txn->request = dpif_linux_encode_execute(dpif->dp_ifindex, + execute); } else { NOT_REACHED(); } @@ -920,10 +902,10 @@ dpif_linux_operate(struct dpif *dpif_, union dpif_op **ops, size_t n_ops) for (i = 0; i < n_ops; i++) { struct nl_transaction *txn = &txns[i]; - union dpif_op *op = ops[i]; + struct dpif_op *op = ops[i]; if (op->type == DPIF_OP_FLOW_PUT) { - struct dpif_flow_put *put = &op->flow_put; + struct dpif_flow_put *put = &op->u.flow_put; int error = txn->error; if (!error && put->stats) { @@ -934,11 +916,9 @@ dpif_linux_operate(struct dpif *dpif_, union dpif_op **ops, size_t n_ops) dpif_linux_flow_get_stats(&reply, put->stats); } } - put->error = error; + op->error = error; } else if (op->type == DPIF_OP_EXECUTE) { - struct dpif_execute *execute = &op->execute; - - execute->error = txn->error; + op->error = txn->error; } else { NOT_REACHED(); } @@ -949,14 +929,6 @@ dpif_linux_operate(struct dpif *dpif_, union dpif_op **ops, size_t n_ops) free(txns); } -static int -dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) -{ - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - *listen_mask = dpif->listen_mask; - return 0; -} - static void set_upcall_pids(struct dpif *dpif_) { @@ -966,8 +938,7 @@ set_upcall_pids(struct dpif *dpif_) int error; DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) { - uint32_t upcall_pid = dpif_linux_port_get_pid__(dpif_, port.port_no, - DPIF_UC_MISS); + 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); @@ -988,30 +959,48 @@ set_upcall_pids(struct dpif *dpif_) } 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_); return 0; @@ -1088,48 +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; } } } @@ -1141,15 +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 @@ -1158,7 +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; } @@ -1196,8 +1194,7 @@ const struct dpif_class dpif_linux_class = { dpif_linux_flow_dump_done, dpif_linux_execute, dpif_linux_operate, - dpif_linux_recv_get_mask, - dpif_linux_recv_set_mask, + dpif_linux_recv_set, dpif_linux_queue_to_priority, dpif_linux_recv, dpif_linux_recv_wait, @@ -1271,11 +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); @@ -1283,8 +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, 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 @@ -1329,9 +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, @@ -1389,7 +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; @@ -1504,9 +1504,7 @@ 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 }, }; @@ -1549,7 +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; @@ -1650,9 +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_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 }, @@ -1708,7 +1705,7 @@ dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family, NLM_F_REQUEST | flow->nlmsg_flags, - flow->cmd, 1); + flow->cmd, OVS_FLOW_VERSION); ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); ovs_header->dp_ifindex = flow->dp_ifindex;