X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=79ba67d8ff3f9f5833877a6148c2e2d5541a9da4;hb=05bfbcb693680794c5a0398587a7da6b46b3e3f5;hp=549c7e4463b97f7ca37cf0e68ec5ff6ff8c1106a;hpb=2ee6545f2bff7eb27e8c84965e3ff38dfa909bf6;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index 549c7e446..79ba67d8f 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -28,7 +28,10 @@ #include #include #include +#include #include +#include +#include #include #include @@ -44,9 +47,11 @@ #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 "random.h" #include "shash.h" #include "sset.h" #include "unaligned.h" @@ -59,6 +64,10 @@ 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)); +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. */ #define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */ @@ -72,10 +81,8 @@ struct dpif_linux_dp { /* Attributes. */ 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. */ - const uint32_t *sampling; /* OVS_DP_ATTR_SAMPLING. */ - uint32_t mcgroups[DPIF_N_UC_TYPES]; /* OVS_DP_ATTR_MCGROUPS. */ }; static void dpif_linux_dp_init(struct dpif_linux_dp *); @@ -98,9 +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. */ @@ -110,7 +117,7 @@ struct dpif_linux_flow { size_t actions_len; 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. */ }; @@ -119,7 +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 *, @@ -130,10 +137,11 @@ struct dpif_linux { struct dpif dpif; int dp_ifindex; - /* Multicast group messages. */ - struct nl_sock *mc_sock; - uint32_t mcgroups[DPIF_N_UC_TYPES]; - unsigned int listen_mask; + /* Upcall messages. */ + struct nl_sock *upcall_socks[N_UPCALL_SOCKS]; + uint32_t ready_mask; /* 1-bit for each sock with unread messages. */ + unsigned int listen_mask; /* Mask of DPIF_UC_* bits. */ + int epoll_fd; /* epoll fd that includes the upcall socks. */ /* Change notification. */ struct sset changed_ports; /* Ports that have changed. */ @@ -160,9 +168,12 @@ 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 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 void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, struct ofpbuf *); @@ -228,6 +239,7 @@ 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(); @@ -237,55 +249,62 @@ 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) { 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 i; - dpif = xmalloc(sizeof *dpif); + dpif = xzalloc(sizeof *dpif); dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed, dpif); - if (!dpif->port_notifier) { - goto error_free; - } + dpif->epoll_fd = -1; 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, OVSP_LOCAL); for (i = 1; i < LRU_MAX_PORTS; i++) { dpif_linux_push_port(dpif, i); } - return 0; +} -error_free: - free(dpif); - return EINVAL; +static void +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; + } } static void @@ -294,7 +313,7 @@ dpif_linux_close(struct dpif *dpif_) struct dpif_linux *dpif = dpif_linux_cast(dpif_); nln_notifier_destroy(dpif->port_notifier); - nl_sock_destroy(dpif->mc_sock); + destroy_upcall_socks(dpif); sset_destroy(&dpif->changed_ports); free(dpif->lru_bitmap); free(dpif); @@ -329,22 +348,7 @@ dpif_linux_wait(struct dpif *dpif OVS_UNUSED) } static int -dpif_linux_get_stats(const struct dpif *dpif_, struct ovs_dp_stats *stats) -{ - struct dpif_linux_dp dp; - struct ofpbuf *buf; - int error; - - error = dpif_linux_dp_get(dpif_, &dp, &buf); - if (!error) { - *stats = dp.stats; - ofpbuf_delete(buf); - } - return error; -} - -static int -dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) +dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) { struct dpif_linux_dp dp; struct ofpbuf *buf; @@ -352,25 +356,15 @@ 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 == OVS_DP_FRAG_DROP; + 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; } -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) @@ -407,11 +401,18 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, /* Loop until we find a port that isn't used. */ do { + 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); + request.upcall_pid = &upcall_pid; 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, upcall_pid); } ofpbuf_delete(buf); } while (request.port_no != UINT32_MAX @@ -456,9 +457,15 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, error = dpif_linux_vport_transact(&request, &reply, &buf); if (!error) { - 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.dp_ifindex != request.dp_ifindex) { + /* A query by name reported that 'port_name' is in some datapath + * other than 'dpif', but the caller wants to know about 'dpif'. */ + error = ENODEV; + } else { + dpif_port->name = xstrdup(reply.name); + dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply)); + dpif_port->port_no = reply.port_no; + } ofpbuf_delete(buf); } return error; @@ -486,6 +493,26 @@ 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, + enum dpif_upcall_type upcall_type) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + if (!(dpif->listen_mask & (1u << upcall_type))) { + return 0; + } else { + int idx = port_no & (N_UPCALL_SOCKS - 1); + return nl_sock_pid(dpif->upcall_socks[idx]); + } +} + +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_) { @@ -646,30 +673,43 @@ dpif_linux_flow_get(const struct dpif *dpif_, return error; } +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, + 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 = (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; + if (flags & DPIF_FP_ZERO_STATS) { + request->clear = true; + } + request->nlmsg_flags = 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) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow request, reply; - struct nlattr dummy_action; struct ofpbuf *buf; int error; - 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; - 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_, flags, key, key_len, actions, actions_len, + &request); error = dpif_linux_flow_transact(&request, stats ? &reply : NULL, stats ? &buf : NULL); @@ -796,20 +836,19 @@ dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) return error; } -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) +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) { struct ovs_header *execute; struct ofpbuf *buf; - int error; buf = ofpbuf_new(128 + actions_len + 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; @@ -818,8 +857,23 @@ dpif_linux_execute__(int dp_ifindex, nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, key, key_len); nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, actions, actions_len); - error = nl_sock_transact(genl_sock, buf, NULL); - ofpbuf_delete(buf); + 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) +{ + struct ofpbuf *request; + int error; + + request = dpif_linux_encode_execute(dp_ifindex, + key, key_len, actions, actions_len, + packet); + error = nl_sock_transact(genl_sock, request, NULL); + ofpbuf_delete(request); + return error; } @@ -835,6 +889,82 @@ dpif_linux_execute(struct dpif *dpif_, actions, actions_len, packet); } +static void +dpif_linux_operate(struct dpif *dpif_, union dpif_op **ops, size_t n_ops) +{ + 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]; + union dpif_op *op = ops[i]; + + if (op->type == DPIF_OP_FLOW_PUT) { + struct dpif_flow_put *put = &op->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); + 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; + + txn->request = dpif_linux_encode_execute( + dpif->dp_ifindex, execute->key, execute->key_len, + execute->actions, execute->actions_len, execute->packet); + } else { + NOT_REACHED(); + } + } + + txnsp = xmalloc(n_ops * sizeof *txnsp); + for (i = 0; i < n_ops; i++) { + txnsp[i] = &txns[i]; + } + + nl_sock_transact_multiple(genl_sock, txnsp, n_ops); + + free(txnsp); + + for (i = 0; i < n_ops; i++) { + struct nl_transaction *txn = &txns[i]; + union dpif_op *op = ops[i]; + + if (op->type == DPIF_OP_FLOW_PUT) { + struct dpif_flow_put *put = &op->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); + } + } + put->error = error; + } else if (op->type == DPIF_OP_EXECUTE) { + struct dpif_execute *execute = &op->execute; + + execute->error = txn->error; + } else { + NOT_REACHED(); + } + + ofpbuf_delete(txn->request); + ofpbuf_delete(txn->reply); + } + free(txns); +} + static int dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) { @@ -843,79 +973,83 @@ dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) return 0; } +static void +set_upcall_pids(struct dpif *dpif_) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_port_dump port_dump; + struct dpif_port port; + 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); + 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 = &upcall_pid; + 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, + upcall_pid); + } else { + VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s", + dpif_name(&dpif->dpif), strerror(error)); + } + } +} + static int dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int error; - int i; 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; - 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]); + if (!listen_mask) { + destroy_upcall_socks(dpif); + } else if (!dpif->listen_mask) { + int i; + int error; + + dpif->epoll_fd = epoll_create(N_UPCALL_SOCKS); + if (dpif->epoll_fd < 0) { + return errno; } - } - /* Update listen_mask. */ - dpif->listen_mask = listen_mask; + for (i = 0; i < N_UPCALL_SOCKS; i++) { + struct epoll_event event; - /* Subscribe to new groups. */ - error = 0; - for (i = 0; i < DPIF_N_UC_TYPES; i++) { - if (dpif->listen_mask & (1u << i)) { - int retval; + error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]); + if (error) { + destroy_upcall_socks(dpif); + return error; + } - retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]); - if (retval) { - error = retval; + 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; } } - } - return error; -} -static int -dpif_linux_get_sflow_probability(const struct dpif *dpif_, - uint32_t *probability) -{ - struct dpif_linux_dp dp; - struct ofpbuf *buf; - int error; - - error = dpif_linux_dp_get(dpif_, &dp, &buf); - if (!error) { - *probability = dp.sampling ? *dp.sampling : 0; - ofpbuf_delete(buf); + dpif->ready_mask = 0; } - return error; -} -static int -dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability) -{ - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_dp dp; + dpif->listen_mask = listen_mask; + set_upcall_pids(dpif_); - dpif_linux_dp_init(&dp); - dp.cmd = OVS_DP_CMD_SET; - dp.dp_ifindex = dpif->dp_ifindex; - dp.sampling = &probability; - return dpif_linux_dp_transact(&dp, NULL, NULL); + return 0; } static int @@ -942,10 +1076,6 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, /* OVS_PACKET_CMD_ACTION only. */ [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true }, - - /* OVS_PACKET_CMD_SAMPLE only. */ - [OVS_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true }, - [OVS_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, }; struct ovs_header *ovs_header; @@ -969,7 +1099,6 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION - : genl->cmd == OVS_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE : -1); if (type < 0) { return EINVAL; @@ -985,14 +1114,6 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA] ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA]) : 0); - upcall->sample_pool = (a[OVS_PACKET_ATTR_SAMPLE_POOL] - ? nl_attr_get_u32(a[OVS_PACKET_ATTR_SAMPLE_POOL]) - : 0); - if (a[OVS_PACKET_ATTR_ACTIONS]) { - upcall->actions = (void *) nl_attr_get(a[OVS_PACKET_ATTR_ACTIONS]); - upcall->actions_len = nl_attr_get_size(a[OVS_PACKET_ATTR_ACTIONS]); - } - *dp_ifindex = ovs_header->dp_ifindex; return 0; @@ -1002,32 +1123,63 @@ 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; + int read_tries = 0; - if (!dpif->mc_sock) { - return EAGAIN; + if (!dpif->listen_mask) { + return EAGAIN; } - for (i = 0; i < 50; i++) { - int dp_ifindex; + if (!dpif->ready_mask) { + struct epoll_event events[N_UPCALL_SOCKS]; + int retval; + int i; - error = nl_sock_recv(dpif->mc_sock, &buf, false); - if (error) { - return error; + 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)); } - error = parse_odp_packet(buf, upcall, &dp_ifindex); - if (!error - && dp_ifindex == dpif->dp_ifindex - && dpif->listen_mask & (1u << upcall->type)) { - return 0; + for (i = 0; i < retval; i++) { + dpif->ready_mask |= 1u << events[i].data.u32; } + } - ofpbuf_delete(buf); - if (error) { - return error; + while (dpif->ready_mask) { + int indx = ffs(dpif->ready_mask) - 1; + struct nl_sock *upcall_sock = dpif->upcall_socks[indx]; + + dpif->ready_mask &= ~(1u << indx); + + for (;;) { + struct ofpbuf *buf; + int dp_ifindex; + 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; + } } } @@ -1038,18 +1190,26 @@ 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); + + if (!dpif->listen_mask) { + return; } + + poll_fd_wait(dpif->epoll_fd, POLLIN); } static void dpif_linux_recv_purge(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); + int i; - if (dpif->mc_sock) { - nl_sock_drain(dpif->mc_sock); + if (!dpif->listen_mask) { + return; + } + + for (i = 0; i < N_UPCALL_SOCKS; i++) { + nl_sock_drain(dpif->upcall_socks[i]); } } @@ -1062,13 +1222,12 @@ 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, @@ -1082,10 +1241,9 @@ const struct dpif_class dpif_linux_class = { dpif_linux_flow_dump_next, dpif_linux_flow_dump_done, dpif_linux_execute, + dpif_linux_operate, 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, @@ -1122,7 +1280,8 @@ dpif_linux_init(void) } if (!error) { error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP, - &ovs_vport_mcgroup); + &ovs_vport_mcgroup, + OVS_VPORT_MCGROUP_FALLBACK_ID); } if (!error) { static struct dpif_linux_vport vport; @@ -1162,7 +1321,7 @@ dpif_linux_vport_send(int dp_ifindex, uint32_t port_no, 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); @@ -1215,16 +1374,14 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, [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_STATS] = { .type = NL_A_UNSPEC, - .min_len = sizeof(struct ovs_vport_stats), - .max_len = sizeof(struct ovs_vport_stats), + [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 }, + [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, .max_len = ETH_ADDR_LEN, .optional = true }, [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, - [OVS_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true }, }; struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)]; @@ -1251,6 +1408,9 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, 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(a[OVS_VPORT_ATTR_UPCALL_PID]); + } if (a[OVS_VPORT_ATTR_STATS]) { vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]); } @@ -1261,9 +1421,6 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]); vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]); } - if (a[OVS_VPORT_ATTR_IFINDEX]) { - vport->ifindex = nl_attr_get_u32(a[OVS_VPORT_ATTR_IFINDEX]); - } return 0; } @@ -1276,7 +1433,7 @@ 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; @@ -1293,6 +1450,10 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name); } + if (vport->upcall_pid) { + nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid); + } + if (vport->stats) { nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS, vport->stats, sizeof *vport->stats); @@ -1307,10 +1468,6 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS, vport->options, vport->options_len); } - - if (vport->ifindex) { - nl_msg_put_u32(buf, OVS_VPORT_ATTR_IFINDEX, vport->ifindex); - } } /* Clears 'vport' to "empty" values. */ @@ -1391,13 +1548,8 @@ 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 }, - [OVS_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true }, - [OVS_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true }, }; struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)]; @@ -1428,40 +1580,6 @@ 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]); - } - if (a[OVS_DP_ATTR_SAMPLING]) { - dp->sampling = nl_attr_get(a[OVS_DP_ATTR_SAMPLING]); - } - - if (a[OVS_DP_ATTR_MCGROUPS]) { - static const struct nl_policy ovs_mcgroup_policy[] = { - [OVS_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true }, - [OVS_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true }, - [OVS_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true }, - }; - - struct nlattr *mcgroups[ARRAY_SIZE(ovs_mcgroup_policy)]; - - if (!nl_parse_nested(a[OVS_DP_ATTR_MCGROUPS], ovs_mcgroup_policy, - mcgroups, ARRAY_SIZE(ovs_mcgroup_policy))) { - return EINVAL; - } - - if (mcgroups[OVS_PACKET_CMD_MISS]) { - dp->mcgroups[DPIF_UC_MISS] - = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_MISS]); - } - if (mcgroups[OVS_PACKET_CMD_ACTION]) { - dp->mcgroups[DPIF_UC_ACTION] - = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_ACTION]); - } - if (mcgroups[OVS_PACKET_CMD_SAMPLE]) { - dp->mcgroups[DPIF_UC_SAMPLE] - = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_SAMPLE]); - } - } return 0; } @@ -1473,7 +1591,8 @@ 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; @@ -1482,15 +1601,11 @@ dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name); } - /* 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); + if (dp->upcall_pid) { + nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid); } - if (dp->sampling) { - nl_msg_put_u32(buf, OVS_DP_ATTR_SAMPLING, *dp->sampling); - } + /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */ } /* Clears 'dp' to "empty" values. */ @@ -1578,9 +1693,7 @@ 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 }, @@ -1635,8 +1748,8 @@ 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; @@ -1674,7 +1787,7 @@ 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; @@ -1682,6 +1795,10 @@ 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); @@ -1711,7 +1828,6 @@ 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; } -