X-Git-Url: http://git.cascardo.eti.br/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=0f614532d1e51c8fa44a2d0656fea1b4e3cc4c51;hb=09538fdc5765afad5c5419483e688d1de4fd5aea;hp=d47ddf151852501fbf69997932ed1ad430d6ee14;hpb=b99d3ceeedd2c5533c13f9794616fc6d9c49c0de;p=cascardo%2Fovs.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index d47ddf151..0f614532d 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -18,7 +18,6 @@ #include "dpif-linux.h" -#include #include #include #include @@ -54,6 +53,7 @@ #include "random.h" #include "shash.h" #include "sset.h" +#include "timeval.h" #include "unaligned.h" #include "util.h" #include "vlog.h" @@ -61,10 +61,6 @@ VLOG_DEFINE_THIS_MODULE(dpif_linux); enum { MAX_PORTS = USHRT_MAX }; -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 */ @@ -129,23 +125,31 @@ static int dpif_linux_flow_transact(struct dpif_linux_flow *request, static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *, struct dpif_flow_stats *); +/* One of the dpif channels between the kernel and userspace. */ +struct dpif_channel { + struct nl_sock *sock; /* Netlink socket. */ + long long int last_poll; /* Last time this channel was polled. */ +}; + +static void report_loss(struct dpif *, struct dpif_channel *); + /* Datapath interface for the openvswitch Linux kernel module. */ struct dpif_linux { struct dpif dpif; int dp_ifindex; /* Upcall messages. */ - struct nl_sock *upcall_socks[N_UPCALL_SOCKS]; - uint32_t ready_mask; /* 1-bit for each sock with unread messages. */ - int epoll_fd; /* epoll fd that includes the upcall socks. */ + int uc_array_size; /* Size of 'channels' and 'epoll_events'. */ + struct dpif_channel *channels; + struct epoll_event *epoll_events; + int epoll_fd; /* epoll fd that includes channel socks. */ + int n_events; /* Num events returned by epoll_wait(). */ + int event_offset; /* Offset into 'epoll_events'. */ /* Change notification. */ struct sset changed_ports; /* Ports that have changed. */ struct nln_notifier *port_notifier; bool change_error; - - /* Port number allocation. */ - uint16_t alloc_port_no; }; static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); @@ -164,7 +168,7 @@ 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); +static uint32_t dpif_linux_port_get_pid(const struct dpif *, uint32_t port_no); static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, struct ofpbuf *); @@ -254,18 +258,110 @@ open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp) } static void -destroy_upcall_socks(struct dpif_linux *dpif) +destroy_channels(struct dpif_linux *dpif) { int i; - if (dpif->epoll_fd >= 0) { - close(dpif->epoll_fd); - dpif->epoll_fd = -1; + if (dpif->epoll_fd < 0) { + return; + } + + for (i = 0; i < dpif->uc_array_size; i++ ) { + struct dpif_linux_vport vport_request; + struct dpif_channel *ch = &dpif->channels[i]; + uint32_t upcall_pid = 0; + + if (!ch->sock) { + continue; + } + + /* Turn off upcalls. */ + dpif_linux_vport_init(&vport_request); + vport_request.cmd = OVS_VPORT_CMD_SET; + vport_request.dp_ifindex = dpif->dp_ifindex; + vport_request.port_no = i; + vport_request.upcall_pid = &upcall_pid; + dpif_linux_vport_transact(&vport_request, NULL, NULL); + + nl_sock_destroy(ch->sock); + } + + free(dpif->channels); + dpif->channels = NULL; + dpif->uc_array_size = 0; + + free(dpif->epoll_events); + dpif->epoll_events = NULL; + dpif->n_events = dpif->event_offset = 0; + + close(dpif->epoll_fd); + dpif->epoll_fd = -1; +} + +static int +add_channel(struct dpif_linux *dpif, uint32_t port_no, struct nl_sock *sock) +{ + struct epoll_event event; + + if (dpif->epoll_fd < 0) { + return 0; } - for (i = 0; i < N_UPCALL_SOCKS; i++) { - nl_sock_destroy(dpif->upcall_socks[i]); - dpif->upcall_socks[i] = NULL; + + /* We assume that the datapath densely chooses port numbers, which + * can therefore be used as an index into an array of channels. */ + if (port_no >= dpif->uc_array_size) { + int new_size = port_no + 1; + int i; + + if (new_size > 65535) { + VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big", + dpif_name(&dpif->dpif), port_no); + return EFBIG; + } + + dpif->channels = xrealloc(dpif->channels, + new_size * sizeof *dpif->channels); + for (i = dpif->uc_array_size; i < new_size; i++) { + dpif->channels[i].sock = NULL; + } + + dpif->epoll_events = xrealloc(dpif->epoll_events, + new_size * sizeof *dpif->epoll_events); + dpif->uc_array_size = new_size; + } + + memset(&event, 0, sizeof event); + event.events = EPOLLIN; + event.data.u32 = port_no; + if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock), + &event) < 0) { + return errno; } + + dpif->channels[port_no].sock = sock; + dpif->channels[port_no].last_poll = LLONG_MIN; + + return 0; +} + +static void +del_channel(struct dpif_linux *dpif, uint32_t port_no) +{ + struct dpif_channel *ch; + + if (dpif->epoll_fd < 0 || port_no >= dpif->uc_array_size) { + return; + } + + ch = &dpif->channels[port_no]; + if (!ch->sock) { + return; + } + + epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL); + + nl_sock_destroy(ch->sock); + ch->sock = NULL; } static void @@ -274,7 +370,7 @@ dpif_linux_close(struct dpif *dpif_) struct dpif_linux *dpif = dpif_linux_cast(dpif_); nln_notifier_destroy(dpif->port_notifier); - destroy_upcall_socks(dpif); + destroy_channels(dpif); sset_destroy(&dpif->changed_ports); free(dpif); } @@ -292,7 +388,7 @@ dpif_linux_destroy(struct dpif *dpif_) } static void -dpif_linux_run(struct dpif *dpif OVS_UNUSED) +dpif_linux_run(struct dpif *dpif_ OVS_UNUSED) { if (nln) { nln_run(nln); @@ -325,68 +421,149 @@ dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) return error; } +static const char * +get_vport_type(const struct dpif_linux_vport *vport) +{ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + + switch (vport->type) { + case OVS_VPORT_TYPE_NETDEV: + return "system"; + + case OVS_VPORT_TYPE_INTERNAL: + return "internal"; + + case OVS_VPORT_TYPE_GRE: + return "gre"; + + case OVS_VPORT_TYPE_GRE64: + return "gre64"; + + case OVS_VPORT_TYPE_VXLAN: + return "vxlan"; + + case OVS_VPORT_TYPE_UNSPEC: + case __OVS_VPORT_TYPE_MAX: + break; + } + + VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u", + vport->dp_ifindex, vport->name, (unsigned int) vport->type); + return "unknown"; +} + +static enum ovs_vport_type +netdev_to_ovs_vport_type(const struct netdev *netdev) +{ + const char *type = netdev_get_type(netdev); + + if (!strcmp(type, "tap") || !strcmp(type, "system")) { + return OVS_VPORT_TYPE_NETDEV; + } else if (!strcmp(type, "internal")) { + return OVS_VPORT_TYPE_INTERNAL; + } else if (strstr(type, "gre64")) { + return OVS_VPORT_TYPE_GRE64; + } else if (strstr(type, "gre")) { + return OVS_VPORT_TYPE_GRE; + } else if (!strcmp(type, "vxlan")) { + return OVS_VPORT_TYPE_VXLAN; + } else { + return OVS_VPORT_TYPE_UNSPEC; + } +} + static int dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, - uint16_t *port_nop) + uint32_t *port_nop) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - const char *name = netdev_get_name(netdev); + const struct netdev_tunnel_config *tnl_cfg; + const char *name = netdev_vport_get_dpif_port(netdev); const char *type = netdev_get_type(netdev); struct dpif_linux_vport request, reply; - const struct ofpbuf *options; + struct nl_sock *sock = NULL; + uint32_t upcall_pid; struct ofpbuf *buf; - int error, i = 0, max_ports = MAX_PORTS; + uint64_t options_stub[64 / 8]; + struct ofpbuf options; + int error; + + if (dpif->epoll_fd >= 0) { + error = nl_sock_create(NETLINK_GENERIC, &sock); + if (error) { + return error; + } + } dpif_linux_vport_init(&request); request.cmd = OVS_VPORT_CMD_NEW; request.dp_ifindex = dpif->dp_ifindex; - request.type = netdev_vport_get_vport_type(netdev); + request.type = netdev_to_ovs_vport_type(netdev); 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); + nl_sock_destroy(sock); return EINVAL; } request.name = name; - options = netdev_vport_get_options(netdev); - if (options && options->size) { - request.options = options->data; - 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 { - uint32_t upcall_pid; + tnl_cfg = netdev_get_tunnel_config(netdev); + if (tnl_cfg && tnl_cfg->dst_port != 0) { + ofpbuf_use_stack(&options, options_stub, sizeof options_stub); + nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT, + ntohs(tnl_cfg->dst_port)); + request.options = options.data; + request.options_len = options.size; + } - request.port_no = ++dpif->alloc_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); + request.port_no = *port_nop; + upcall_pid = sock ? nl_sock_pid(sock) : 0; + request.upcall_pid = &upcall_pid; - 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); - } else if (error == EFBIG) { - /* Older datapath has lower limit. */ - max_ports = dpif->alloc_port_no; - dpif->alloc_port_no = 0; + 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_), reply.port_no, upcall_pid); + } else { + if (error == EBUSY && *port_nop != UINT32_MAX) { + VLOG_INFO("%s: requested port %"PRIu32" is in use", + dpif_name(dpif_), *port_nop); } - + nl_sock_destroy(sock); ofpbuf_delete(buf); - } while ((i++ < max_ports) - && (error == EBUSY || error == EFBIG)); + return error; + } + ofpbuf_delete(buf); - return error; + if (sock) { + error = add_channel(dpif, *port_nop, sock); + if (error) { + VLOG_INFO("%s: could not add channel for port %s", + dpif_name(dpif_), name); + + /* Delete the port. */ + dpif_linux_vport_init(&request); + request.cmd = OVS_VPORT_CMD_DEL; + request.dp_ifindex = dpif->dp_ifindex; + request.port_no = *port_nop; + dpif_linux_vport_transact(&request, NULL, NULL); + + nl_sock_destroy(sock); + return error; + } + } + + return 0; } static int -dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) +dpif_linux_port_del(struct dpif *dpif_, uint32_t port_no) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_vport vport; @@ -398,6 +575,8 @@ dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) vport.port_no = port_no; error = dpif_linux_vport_transact(&vport, NULL, NULL); + del_channel(dpif, port_no); + return error; } @@ -422,9 +601,9 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, /* 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 { + } else if (dpif_port) { dpif_port->name = xstrdup(reply.name); - dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply)); + dpif_port->type = xstrdup(get_vport_type(&reply)); dpif_port->port_no = reply.port_no; } ofpbuf_delete(buf); @@ -433,7 +612,7 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, } static int -dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no, +dpif_linux_port_query_by_number(const struct dpif *dpif, uint32_t port_no, struct dpif_port *dpif_port) { return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port); @@ -453,15 +632,17 @@ 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) +dpif_linux_port_get_pid(const struct dpif *dpif_, uint32_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]); + /* The UINT32_MAX "reserved" port number uses the "ovs-system"'s + * channel, since it is not heavily loaded. */ + int idx = (port_no >= dpif->uc_array_size) ? 0 : port_no; + return nl_sock_pid(dpif->channels[idx].sock); } } @@ -521,8 +702,8 @@ dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_, return error; } - dpif_port->name = (char *) vport.name; - dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport); + 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; return 0; } @@ -594,7 +775,7 @@ dpif_linux_flow_get(const struct dpif *dpif_, dpif_linux_flow_get_stats(&reply, stats); } if (actionsp) { - buf->data = (void *) reply.actions; + buf->data = CONST_CAST(struct nlattr *, reply.actions); buf->size = reply.actions_len; *actionsp = buf; } else { @@ -767,14 +948,16 @@ dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) return error; } -static struct ofpbuf * -dpif_linux_encode_execute(int dp_ifindex, - const struct dpif_execute *d_exec) +static void +dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec, + struct ofpbuf *buf) { struct ovs_header *k_exec; - struct ofpbuf *buf; - buf = ofpbuf_new(128 + d_exec->actions_len + d_exec->packet->size); + ofpbuf_prealloc_tailroom(buf, (64 + + d_exec->packet->size + + d_exec->key_len + + d_exec->actions_len)); nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST, OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION); @@ -787,19 +970,19 @@ dpif_linux_encode_execute(int dp_ifindex, 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 dpif_execute *execute) { - struct ofpbuf *request; + uint64_t request_stub[1024 / 8]; + struct ofpbuf request; int error; - request = dpif_linux_encode_execute(dp_ifindex, execute); - error = nl_sock_transact(genl_sock, request, NULL); - ofpbuf_delete(request); + ofpbuf_use_stub(&request, request_stub, sizeof request_stub); + dpif_linux_encode_execute(dp_ifindex, execute, &request); + error = nl_sock_transact(genl_sock, &request, NULL); + ofpbuf_uninit(&request); return error; } @@ -812,48 +995,67 @@ dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute) return dpif_linux_execute__(dpif->dp_ifindex, execute); } +#define MAX_OPS 50 + static void -dpif_linux_operate(struct dpif *dpif_, struct 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; - struct nl_transaction *txns; + + struct op_auxdata { + struct nl_transaction txn; + + struct ofpbuf request; + uint64_t request_stub[1024 / 8]; + + struct ofpbuf reply; + uint64_t reply_stub[1024 / 8]; + } auxes[MAX_OPS]; + + struct nl_transaction *txnsp[MAX_OPS]; size_t i; - txns = xmalloc(n_ops * sizeof *txns); + ovs_assert(n_ops <= MAX_OPS); for (i = 0; i < n_ops; i++) { - struct nl_transaction *txn = &txns[i]; + struct op_auxdata *aux = &auxes[i]; struct dpif_op *op = ops[i]; struct dpif_flow_put *put; struct dpif_flow_del *del; struct dpif_execute *execute; - struct dpif_linux_flow request; + struct dpif_linux_flow flow; + + ofpbuf_use_stub(&aux->request, + aux->request_stub, sizeof aux->request_stub); + aux->txn.request = &aux->request; + + ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub); + aux->txn.reply = NULL; switch (op->type) { case DPIF_OP_FLOW_PUT: put = &op->u.flow_put; - dpif_linux_init_flow_put(dpif_, put, &request); + dpif_linux_init_flow_put(dpif_, put, &flow); if (put->stats) { - request.nlmsg_flags |= NLM_F_ECHO; + flow.nlmsg_flags |= NLM_F_ECHO; + aux->txn.reply = &aux->reply; } - txn->request = ofpbuf_new(1024); - dpif_linux_flow_to_ofpbuf(&request, txn->request); + dpif_linux_flow_to_ofpbuf(&flow, &aux->request); break; case DPIF_OP_FLOW_DEL: del = &op->u.flow_del; - dpif_linux_init_flow_del(dpif_, del, &request); + dpif_linux_init_flow_del(dpif_, del, &flow); if (del->stats) { - request.nlmsg_flags |= NLM_F_ECHO; + flow.nlmsg_flags |= NLM_F_ECHO; + aux->txn.reply = &aux->reply; } - txn->request = ofpbuf_new(1024); - dpif_linux_flow_to_ofpbuf(&request, txn->request); + dpif_linux_flow_to_ofpbuf(&flow, &aux->request); break; case DPIF_OP_EXECUTE: execute = &op->u.execute; - txn->request = dpif_linux_encode_execute(dpif->dp_ifindex, - execute); + dpif_linux_encode_execute(dpif->dp_ifindex, execute, + &aux->request); break; default: @@ -861,17 +1063,14 @@ dpif_linux_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) } } - txnsp = xmalloc(n_ops * sizeof *txnsp); for (i = 0; i < n_ops; i++) { - txnsp[i] = &txns[i]; + txnsp[i] = &auxes[i].txn; } - nl_sock_transact_multiple(genl_sock, txnsp, n_ops); - free(txnsp); - for (i = 0; i < n_ops; i++) { - struct nl_transaction *txn = &txns[i]; + struct op_auxdata *aux = &auxes[i]; + struct nl_transaction *txn = &auxes[i].txn; struct dpif_op *op = ops[i]; struct dpif_flow_put *put; struct dpif_flow_del *del; @@ -881,24 +1080,38 @@ dpif_linux_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) switch (op->type) { case DPIF_OP_FLOW_PUT: put = &op->u.flow_put; - if (!op->error && put->stats) { - struct dpif_linux_flow reply; - - op->error = dpif_linux_flow_from_ofpbuf(&reply, txn->reply); + if (put->stats) { if (!op->error) { - dpif_linux_flow_get_stats(&reply, put->stats); + struct dpif_linux_flow reply; + + op->error = dpif_linux_flow_from_ofpbuf(&reply, + txn->reply); + if (!op->error) { + dpif_linux_flow_get_stats(&reply, put->stats); + } + } + + if (op->error) { + memset(put->stats, 0, sizeof *put->stats); } } break; case DPIF_OP_FLOW_DEL: del = &op->u.flow_del; - if (!op->error && del->stats) { - struct dpif_linux_flow reply; - - op->error = dpif_linux_flow_from_ofpbuf(&reply, txn->reply); + if (del->stats) { if (!op->error) { - dpif_linux_flow_get_stats(&reply, del->stats); + struct dpif_linux_flow reply; + + op->error = dpif_linux_flow_from_ofpbuf(&reply, + txn->reply); + if (!op->error) { + dpif_linux_flow_get_stats(&reply, del->stats); + } + } + + if (op->error) { + memset(del->stats, 0, sizeof *del->stats); } } break; @@ -910,38 +1123,19 @@ dpif_linux_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) NOT_REACHED(); } - ofpbuf_delete(txn->request); - ofpbuf_delete(txn->reply); + ofpbuf_uninit(&aux->request); + ofpbuf_uninit(&aux->reply); } - free(txns); } static void -set_upcall_pids(struct dpif *dpif_) +dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) { - 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); - 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)); - } + while (n_ops > 0) { + size_t chunk = MIN(n_ops, MAX_OPS); + dpif_linux_operate__(dpif, ops, chunk); + ops += chunk; + n_ops -= chunk; } } @@ -955,41 +1149,63 @@ dpif_linux_recv_set(struct dpif *dpif_, bool enable) } if (!enable) { - destroy_upcall_socks(dpif); + destroy_channels(dpif); } else { - int i; - int error; + struct dpif_port_dump port_dump; + struct dpif_port port; - dpif->epoll_fd = epoll_create(N_UPCALL_SOCKS); + dpif->epoll_fd = epoll_create(10); if (dpif->epoll_fd < 0) { return errno; } - for (i = 0; i < N_UPCALL_SOCKS; i++) { - struct epoll_event event; + 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; - error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]); + error = nl_sock_create(NETLINK_GENERIC, &sock); 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); + upcall_pid = nl_sock_pid(sock); + + 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)); + nl_sock_destroy(sock); + + if (error == ENODEV || error == ENOENT) { + /* This device isn't there, but keep trying the others. */ + continue; + } else { + return error; + } + } + + error = add_channel(dpif, port.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->ready_mask = 0; } - set_upcall_pids(dpif_); - return 0; } @@ -1016,7 +1232,7 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, /* OVS_PACKET_CMD_ACTION only. */ - [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true }, + [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true }, }; struct ovs_header *ovs_header; @@ -1048,20 +1264,21 @@ 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[OVS_PACKET_ATTR_PACKET]); + 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 = (void *) nl_attr_get(a[OVS_PACKET_ATTR_KEY]); + 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] - ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA]) - : 0); + upcall->userdata = a[OVS_PACKET_ATTR_USERDATA]; *dp_ifindex = ovs_header->dp_ifindex; return 0; } static int -dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) +dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall, + struct ofpbuf *buf) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); int read_tries = 0; @@ -1070,32 +1287,30 @@ dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) return EAGAIN; } - if (!dpif->ready_mask) { - struct epoll_event events[N_UPCALL_SOCKS]; + if (dpif->event_offset >= dpif->n_events) { int retval; - int i; + + dpif->event_offset = dpif->n_events = 0; do { - retval = epoll_wait(dpif->epoll_fd, events, N_UPCALL_SOCKS, 0); + retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events, + dpif->uc_array_size, 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 (i = 0; i < retval; i++) { - dpif->ready_mask |= 1u << events[i].data.u32; + } else if (retval > 0) { + dpif->n_events = retval; } } - while (dpif->ready_mask) { - int indx = ffs(dpif->ready_mask) - 1; - struct nl_sock *upcall_sock = dpif->upcall_socks[indx]; + while (dpif->event_offset < dpif->n_events) { + int idx = dpif->epoll_events[dpif->event_offset].data.u32; + struct dpif_channel *ch = &dpif->channels[idx]; - dpif->ready_mask &= ~(1u << indx); + dpif->event_offset++; for (;;) { - struct ofpbuf *buf; int dp_ifindex; int error; @@ -1103,20 +1318,28 @@ dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) return EAGAIN; } - error = nl_sock_recv(upcall_sock, &buf, false); - if (error == EAGAIN) { - break; - } else if (error) { + error = nl_sock_recv(ch->sock, buf, false); + if (error == ENOBUFS) { + /* ENOBUFS typically means that we've received so many + * packets that the buffer overflowed. Try again + * immediately because there's almost certainly a packet + * waiting for us. */ + report_loss(dpif_, ch); + continue; + } + + ch->last_poll = time_msec(); + if (error) { + if (error == EAGAIN) { + break; + } return error; } error = parse_odp_packet(buf, upcall, &dp_ifindex); if (!error && dp_ifindex == dpif->dp_ifindex) { return 0; - } - - ofpbuf_delete(buf); - if (error) { + } else if (error) { return error; } } @@ -1141,20 +1364,23 @@ static void dpif_linux_recv_purge(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int i; + struct dpif_channel *ch; if (dpif->epoll_fd < 0) { return; } - for (i = 0; i < N_UPCALL_SOCKS; i++) { - nl_sock_drain(dpif->upcall_socks[i]); + for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; ch++) { + if (ch->sock) { + nl_sock_drain(ch->sock); + } } } const struct dpif_class dpif_linux_class = { "system", dpif_linux_enumerate, + NULL, dpif_linux_open, dpif_linux_close, dpif_linux_destroy, @@ -1249,33 +1475,6 @@ dpif_linux_is_internal_device(const char *name) 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 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, 0, htonll(0), 0, &flow); - - ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, &flow); - - ofpbuf_use_stack(&actions, &action, sizeof action); - nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no); - - 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 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_) { @@ -1320,10 +1519,6 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, [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 }, }; @@ -1357,9 +1552,6 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, if (a[OVS_VPORT_ATTR_STATS]) { vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]); } - if (a[OVS_VPORT_ATTR_ADDRESS]) { - vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]); - } 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]); @@ -1402,11 +1594,6 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, vport->stats, sizeof *vport->stats); } - if (vport->address) { - nl_msg_put_unspec(buf, OVS_VPORT_ATTR_ADDRESS, - vport->address, ETH_ADDR_LEN); - } - if (vport->options) { nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS, vport->options, vport->options_len); @@ -1435,7 +1622,7 @@ dpif_linux_vport_transact(const struct dpif_linux_vport *request, struct ofpbuf *request_buf; int error; - assert((reply != NULL) == (bufp != NULL)); + ovs_assert((reply != NULL) == (bufp != NULL)); error = dpif_linux_init(); if (error) { @@ -1586,7 +1773,7 @@ dpif_linux_dp_transact(const struct dpif_linux_dp *request, struct ofpbuf *request_buf; int error; - assert((reply != NULL) == (bufp != NULL)); + ovs_assert((reply != NULL) == (bufp != NULL)); request_buf = ofpbuf_new(1024); dpif_linux_dp_to_ofpbuf(request, request_buf); @@ -1707,9 +1894,9 @@ dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, } /* We never need to send these to the kernel. */ - assert(!flow->stats); - assert(!flow->tcp_flags); - assert(!flow->used); + ovs_assert(!flow->stats); + ovs_assert(!flow->tcp_flags); + ovs_assert(!flow->used); if (flow->clear) { nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR); @@ -1736,7 +1923,7 @@ dpif_linux_flow_transact(struct dpif_linux_flow *request, struct ofpbuf *request_buf; int error; - assert((reply != NULL) == (bufp != NULL)); + ovs_assert((reply != NULL) == (bufp != NULL)); if (reply) { request->nlmsg_flags |= NLM_F_ECHO; @@ -1774,3 +1961,27 @@ dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow, stats->used = flow->used ? get_32aligned_u64(flow->used) : 0; stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0; } + +/* Logs information about a packet that was recently lost in 'ch' (in + * 'dpif_'). */ +static void +report_loss(struct dpif *dpif_, struct dpif_channel *ch) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); + struct ds s; + + if (VLOG_DROP_WARN(&rl)) { + return; + } + + ds_init(&s); + if (ch->last_poll != LLONG_MIN) { + ds_put_format(&s, " (last polled %lld ms ago)", + time_msec() - ch->last_poll); + } + + VLOG_WARN("%s: lost packet on channel %td%s", + dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s)); + ds_destroy(&s); +}