X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=f7f529266821786d104c550987a150c9fdc269c8;hb=bdd534291ff72077390f7b1745c4443ad3833d99;hp=0442f77364bbb793d38dfb86f99fab5951e8918b;hpb=fcd5d2301f7f58f55668333aa3fcc4128c0597c4;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index 0442f7736..f7f529266 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -25,8 +25,6 @@ #include #include #include -#include -#include #include #include #include @@ -73,6 +71,7 @@ struct dpif_linux_dp { /* Attributes. */ const char *name; /* OVS_DP_ATTR_NAME. */ const uint32_t *upcall_pid; /* OVS_DP_ATTR_UPCALL_PID. */ + uint32_t user_features; /* OVS_DP_ATTR_USER_FEATURES */ struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */ struct ovs_dp_megaflow_stats megaflow_stats; /* OVS_DP_ATTR_MEGAFLOW_STATS.*/ @@ -151,6 +150,7 @@ struct dpif_linux { /* Change notification. */ struct nl_sock *port_notifier; /* vport multicast group subscriber. */ + bool refresh_channels; }; static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); @@ -172,6 +172,7 @@ static int dpif_linux_init(void); static int open_dpif(const struct dpif_linux_dp *, struct dpif **); static uint32_t dpif_linux_port_get_pid(const struct dpif *, odp_port_t port_no); +static int dpif_linux_refresh_channels(struct dpif *); static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, struct ofpbuf *); @@ -229,9 +230,11 @@ dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, upcall_pid = 0; dp_request.upcall_pid = &upcall_pid; } else { - dp_request.cmd = OVS_DP_CMD_GET; + /* Use OVS_DP_CMD_SET to report user features */ + dp_request.cmd = OVS_DP_CMD_SET; } dp_request.name = name; + dp_request.user_features |= OVS_DP_F_UNALIGNED; error = dpif_linux_dp_transact(&dp_request, &dp, &buf); if (error) { return error; @@ -400,6 +403,16 @@ dpif_linux_destroy(struct dpif *dpif_) return dpif_linux_dp_transact(&dp, NULL, NULL); } +static void +dpif_linux_run(struct dpif *dpif_) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + if (dpif->refresh_channels) { + dpif->refresh_channels = false; + dpif_linux_refresh_channels(dpif_); + } +} + static int dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) { @@ -658,12 +671,6 @@ dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname, return dpif_linux_port_query__(dpif, 0, devname, dpif_port); } -static uint32_t -dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED) -{ - return MAX_PORTS; -} - static uint32_t dpif_linux_port_get_pid(const struct dpif *dpif_, odp_port_t port_no) { @@ -676,7 +683,8 @@ dpif_linux_port_get_pid(const struct dpif *dpif_, odp_port_t port_no) /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s * channel, since it is not heavily loaded. */ uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx; - pid = nl_sock_pid(dpif->channels[idx].sock); + const struct nl_sock *sock = dpif->channels[idx].sock; + pid = sock ? nl_sock_pid(sock) : 0; } ovs_mutex_unlock(&dpif->upcall_lock); @@ -699,46 +707,66 @@ struct dpif_linux_port_state { struct nl_dump dump; }; -static int -dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep) +static void +dpif_linux_port_dump_start__(const struct dpif *dpif_, struct nl_dump *dump) { const struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_port_state *state; struct dpif_linux_vport request; struct ofpbuf *buf; - *statep = state = xmalloc(sizeof *state); - dpif_linux_vport_init(&request); request.cmd = OVS_VPORT_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; buf = ofpbuf_new(1024); dpif_linux_vport_to_ofpbuf(&request, buf); - nl_dump_start(&state->dump, NETLINK_GENERIC, buf); + nl_dump_start(dump, NETLINK_GENERIC, buf); ofpbuf_delete(buf); +} + +static int +dpif_linux_port_dump_start(const struct dpif *dpif, void **statep) +{ + struct dpif_linux_port_state *state; + + *statep = state = xmalloc(sizeof *state); + dpif_linux_port_dump_start__(dpif, &state->dump); return 0; } static int -dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_, - struct dpif_port *dpif_port) +dpif_linux_port_dump_next__(const struct dpif *dpif_, struct nl_dump *dump, + struct dpif_linux_vport *vport) { - struct dpif_linux_port_state *state = state_; - struct dpif_linux_vport vport; + struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct ofpbuf buf; int error; - if (!nl_dump_next(&state->dump, &buf)) { + if (!nl_dump_next(dump, &buf)) { return EOF; } - error = dpif_linux_vport_from_ofpbuf(&vport, &buf); + error = dpif_linux_vport_from_ofpbuf(vport, &buf); if (error) { - return error; + VLOG_WARN_RL(&error_rl, "%s: failed to parse vport record (%s)", + dpif_name(&dpif->dpif), ovs_strerror(error)); } + return error; +} + +static int +dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_, + struct dpif_port *dpif_port) +{ + struct dpif_linux_port_state *state = state_; + struct dpif_linux_vport vport; + int error; + error = dpif_linux_port_dump_next__(dpif, &state->dump, &vport); + if (error) { + return error; + } dpif_port->name = CONST_CAST(char *, vport.name); dpif_port->type = CONST_CAST(char *, get_vport_type(&vport)); dpif_port->port_no = vport.port_no; @@ -801,6 +829,9 @@ dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep) || vport.cmd == OVS_VPORT_CMD_SET)) { VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8, dpif->dpif.full_name, vport.name, vport.cmd); + if (vport.cmd == OVS_VPORT_CMD_DEL) { + dpif->refresh_channels = true; + } *devnamep = xstrdup(vport.name); ofpbuf_uninit(&buf); return 0; @@ -981,7 +1012,7 @@ dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) } static int -dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_, +dpif_linux_flow_dump_next(const struct dpif *dpif_, void *state_, const struct nlattr **key, size_t *key_len, const struct nlattr **mask, size_t *mask_len, const struct nlattr **actions, size_t *actions_len, @@ -1051,10 +1082,11 @@ dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec, struct ofpbuf *buf) { struct ovs_header *k_exec; + size_t key_ofs; ofpbuf_prealloc_tailroom(buf, (64 + d_exec->packet->size - + d_exec->key_len + + ODP_KEY_METADATA_SIZE + d_exec->actions_len)); nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST, @@ -1065,7 +1097,11 @@ dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec, 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); + + key_ofs = nl_msg_start_nested(buf, OVS_PACKET_ATTR_KEY); + odp_key_from_pkt_metadata(buf, &d_exec->md); + nl_msg_end_nested(buf, key_ofs); + nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, d_exec->actions, d_exec->actions_len); } @@ -1086,7 +1122,7 @@ dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute) } static int -dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute) +dpif_linux_execute(struct dpif *dpif_, struct dpif_execute *execute) { const struct dpif_linux *dpif = dpif_linux_cast(dpif_); @@ -1157,7 +1193,7 @@ dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) break; default: - NOT_REACHED(); + OVS_NOT_REACHED(); } } @@ -1218,7 +1254,7 @@ dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) break; default: - NOT_REACHED(); + OVS_NOT_REACHED(); } ofpbuf_uninit(&aux->request); @@ -1237,45 +1273,59 @@ dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) } } +/* Synchronizes 'dpif->channels' with the set of vports currently in 'dpif' in + * the kernel, by adding a new channel for any kernel vport that lacks one and + * deleting any channels that have no backing kernel vports. */ static int -dpif_linux_recv_set__(struct dpif *dpif_, bool enable) +dpif_linux_refresh_channels(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); + unsigned long int *keep_channels; + struct dpif_linux_vport vport; + size_t keep_channels_nbits; + struct nl_dump dump; + int retval = 0; + size_t i; - if ((dpif->epoll_fd >= 0) == enable) { - return 0; + /* To start with, we need an epoll fd. */ + if (dpif->epoll_fd < 0) { + dpif->epoll_fd = epoll_create(10); + if (dpif->epoll_fd < 0) { + return errno; + } } - if (!enable) { - destroy_channels(dpif); - } else { - struct dpif_port_dump port_dump; - struct dpif_port port; + keep_channels_nbits = dpif->uc_array_size; + keep_channels = bitmap_allocate(keep_channels_nbits); - if (dpif->epoll_fd < 0) { - dpif->epoll_fd = epoll_create(10); - if (dpif->epoll_fd < 0) { - return errno; - } - } + dpif->n_events = dpif->event_offset = 0; - DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) { - struct dpif_linux_vport vport_request; - struct nl_sock *sock; - uint32_t upcall_pid; - int error; + dpif_linux_port_dump_start__(dpif_, &dump); + while (!dpif_linux_port_dump_next__(dpif_, &dump, &vport)) { + uint32_t port_no = odp_to_u32(vport.port_no); + struct nl_sock *sock = (port_no < dpif->uc_array_size + ? dpif->channels[port_no].sock + : NULL); + bool new_sock = !sock; + int error; + if (new_sock) { error = nl_sock_create(NETLINK_GENERIC, &sock); if (error) { - return error; + retval = error; + goto error; } + } - upcall_pid = nl_sock_pid(sock); + /* Configure the vport to deliver misses to 'sock'. */ + if (!vport.upcall_pid || *vport.upcall_pid != nl_sock_pid(sock)) { + uint32_t upcall_pid = nl_sock_pid(sock); + 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.port_no = vport.port_no; vport_request.upcall_pid = &upcall_pid; error = dpif_linux_vport_transact(&vport_request, NULL, NULL); if (!error) { @@ -1286,27 +1336,63 @@ dpif_linux_recv_set__(struct dpif *dpif_, bool enable) VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s", dpif_name(&dpif->dpif), ovs_strerror(error)); - nl_sock_destroy(sock); - if (error == ENODEV || error == ENOENT) { - /* This device isn't there, but keep trying the others. */ - continue; + if (error != ENODEV && error != ENOENT) { + retval = error; } else { - return error; + /* The vport isn't really there, even though the dump says + * it is. Probably we just hit a race after a port + * disappeared. */ } + goto error; } + } - error = add_channel(dpif, port.port_no, sock); + if (new_sock) { + error = add_channel(dpif, vport.port_no, sock); if (error) { VLOG_INFO("%s: could not add channel for port %s", - dpif_name(dpif_), port.name); - nl_sock_destroy(sock); - return error; + dpif_name(dpif_), vport.name); + retval = error; + goto error; } } + + if (port_no < keep_channels_nbits) { + bitmap_set1(keep_channels, port_no); + } + continue; + + error: + nl_sock_destroy(sock); } + nl_dump_done(&dump); - return 0; + /* Discard any saved channels that we didn't reuse. */ + for (i = 0; i < keep_channels_nbits; i++) { + if (!bitmap_is_set(keep_channels, i)) { + nl_sock_destroy(dpif->channels[i].sock); + dpif->channels[i].sock = NULL; + } + } + free(keep_channels); + + return retval; +} + +static int +dpif_linux_recv_set__(struct dpif *dpif_, bool enable) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + if ((dpif->epoll_fd >= 0) == enable) { + return 0; + } else if (!enable) { + destroy_channels(dpif); + return 0; + } else { + return dpif_linux_refresh_channels(dpif_); + } } static int @@ -1374,16 +1460,22 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, return EINVAL; } - memset(upcall, 0, sizeof *upcall); + /* (Re)set ALL fields of '*upcall' on successful return. */ upcall->type = type; - upcall->packet = buf; - upcall->packet->data = CONST_CAST(struct nlattr *, - nl_attr_get(a[OVS_PACKET_ATTR_PACKET])); - upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]); upcall->key = CONST_CAST(struct nlattr *, 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]; + + /* Allow overwriting the netlink attribute header without reallocating. */ + ofpbuf_use_stub(&upcall->packet, + CONST_CAST(struct nlattr *, + nl_attr_get(a[OVS_PACKET_ATTR_PACKET])) - 1, + nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]) + + sizeof(struct nlattr)); + upcall->packet.data = (char *)upcall->packet.data + sizeof(struct nlattr); + upcall->packet.size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]); + *dp_ifindex = ovs_header->dp_ifindex; return 0; @@ -1513,14 +1605,13 @@ const struct dpif_class dpif_linux_class = { dpif_linux_open, dpif_linux_close, dpif_linux_destroy, - NULL, /* run */ + dpif_linux_run, NULL, /* wait */ dpif_linux_get_stats, 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, @@ -1839,6 +1930,10 @@ dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid); } + if (dp->user_features) { + nl_msg_put_u32(buf, OVS_DP_ATTR_USER_FEATURES, dp->user_features); + } + /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */ } @@ -2097,7 +2192,7 @@ report_loss(struct dpif *dpif_, struct dpif_channel *ch) time_msec() - ch->last_poll); } - VLOG_WARN("%s: lost packet on channel %td%s", + VLOG_WARN("%s: lost packet on channel %"PRIdPTR"%s", dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s)); ds_destroy(&s); }