X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=e2300d6c297987354d6aa3687f276133334f3842;hb=b3dcb73cc58a7ee089cfaec623818142196e1172;hp=e2c911cf967dc0428898ba65ba97ca7837cca644;hpb=640e1b20776f40633d33a5c2c2bba3e79eda0b64;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index e2c911cf9..e2300d6c2 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011 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 @@ -28,48 +27,53 @@ #include #include #include +#include #include +#include +#include #include #include #include "bitmap.h" #include "dpif-provider.h" +#include "dynamic-string.h" +#include "flow.h" #include "netdev.h" +#include "netdev-linux.h" #include "netdev-vport.h" +#include "netlink-notifier.h" #include "netlink-socket.h" #include "netlink.h" #include "odp-util.h" #include "ofpbuf.h" -#include "openvswitch/tunnel.h" #include "packets.h" #include "poll-loop.h" -#include "rtnetlink.h" -#include "rtnetlink-link.h" +#include "random.h" #include "shash.h" #include "sset.h" +#include "timeval.h" #include "unaligned.h" #include "util.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(dpif_linux); +enum { MAX_PORTS = USHRT_MAX }; -enum { LRU_MAX_PORTS = 1024 }; -enum { LRU_MASK = LRU_MAX_PORTS - 1}; -BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS)); +/* 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 */ struct dpif_linux_dp { /* Generic Netlink header. */ uint8_t cmd; - /* struct odp_header. */ + /* struct ovs_header. */ int dp_ifindex; /* Attributes. */ - const char *name; /* ODP_DP_ATTR_NAME. */ - struct odp_stats stats; /* ODP_DP_ATTR_STATS. */ - enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */ - const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */ - uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */ + 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. */ }; static void dpif_linux_dp_init(struct dpif_linux_dp *); @@ -86,26 +90,28 @@ struct dpif_linux_flow { /* Generic Netlink header. */ uint8_t cmd; - /* struct odp_header. */ + /* struct ovs_header. */ unsigned int nlmsg_flags; int dp_ifindex; /* 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 ODP_FLOW_ATTR_ACTIONS will be included in + * 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. */ - const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */ + const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */ size_t key_len; - const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */ + const struct nlattr *mask; /* OVS_FLOW_ATTR_MASK. */ + size_t mask_len; + const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */ size_t actions_len; - const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */ - const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */ - const uint64_t *used; /* ODP_FLOW_ATTR_USED. */ - bool clear; /* ODP_FLOW_ATTR_CLEAR. */ + const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */ + const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */ + const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */ + bool clear; /* OVS_FLOW_ATTR_CLEAR. */ }; static void dpif_linux_flow_init(struct dpif_linux_flow *); @@ -113,49 +119,57 @@ 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 *, 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; - /* Multicast group messages. */ - struct nl_sock *mc_sock; - uint32_t mcgroups[DPIF_N_UC_TYPES]; - unsigned int listen_mask; + /* Upcall messages. */ + struct ovs_mutex upcall_lock; + 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 rtnetlink_notifier port_notifier; - bool change_error; - - /* Queue of unused ports. */ - unsigned long *lru_bitmap; - uint16_t lru_ports[LRU_MAX_PORTS]; - size_t lru_head; - size_t lru_tail; + struct nl_sock *port_notifier; /* vport multicast group subscriber. */ }; static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); -/* Generic Netlink family numbers for ODP. */ -static int odp_datapath_family; -static int odp_vport_family; -static int odp_flow_family; -static int odp_packet_family; +/* Generic Netlink family numbers for OVS. + * + * Initialized by dpif_linux_init(). */ +static int ovs_datapath_family; +static int ovs_vport_family; +static int ovs_flow_family; +static int ovs_packet_family; -/* Generic Netlink socket. */ -static struct nl_sock *genl_sock; +/* Generic Netlink multicast groups for OVS. + * + * Initialized by dpif_linux_init(). */ +static unsigned int ovs_vport_mcgroup; static int dpif_linux_init(void); static int open_dpif(const struct dpif_linux_dp *, struct dpif **); -static void dpif_linux_port_changed(const struct rtnetlink_link_change *, - void *dpif); +static uint32_t dpif_linux_port_get_pid(const struct dpif *, + odp_port_t port_no); static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, struct ofpbuf *); @@ -169,29 +183,6 @@ dpif_linux_cast(const struct dpif *dpif) return CONTAINER_OF(dpif, struct dpif_linux, dpif); } -static void -dpif_linux_push_port(struct dpif_linux *dp, uint16_t port) -{ - if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) { - bitmap_set1(dp->lru_bitmap, port); - dp->lru_ports[dp->lru_head++ & LRU_MASK] = port; - } -} - -static uint32_t -dpif_linux_pop_port(struct dpif_linux *dp) -{ - uint16_t port; - - if (dp->lru_head == dp->lru_tail) { - return UINT32_MAX; - } - - port = dp->lru_ports[dp->lru_tail++ & LRU_MASK]; - bitmap_set0(dp->lru_bitmap, port); - return port; -} - static int dpif_linux_enumerate(struct sset *all_dps) { @@ -221,6 +212,7 @@ 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(); @@ -230,15 +222,21 @@ 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 ? ODP_DP_CMD_NEW : ODP_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; } @@ -246,49 +244,145 @@ static int open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp) { struct dpif_linux *dpif; - int error; - int i; - dpif = xmalloc(sizeof *dpif); - error = rtnetlink_link_notifier_register(&dpif->port_notifier, - dpif_linux_port_changed, dpif); - if (error) { - goto error_free; - } + dpif = xzalloc(sizeof *dpif); + dpif->port_notifier = NULL; + ovs_mutex_init(&dpif->upcall_lock); + 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, ODPP_LOCAL); - for (i = 1; i < LRU_MAX_PORTS; i++) { - dpif_linux_push_port(dpif, i); + return 0; +} + +static void +destroy_channels(struct dpif_linux *dpif) +{ + unsigned int i; + + 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; + } + + epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL); + + /* 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 = u32_to_odp(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; + + /* Don't close dpif->epoll_fd since that would cause other threads that + * call dpif_recv_wait(dpif) to wait on an arbitrary fd or a closed fd. */ +} + +static int +add_channel(struct dpif_linux *dpif, odp_port_t port_no, struct nl_sock *sock) +{ + struct epoll_event event; + uint32_t port_idx = odp_to_u32(port_no); + + if (dpif->epoll_fd < 0) { + return 0; + } + + /* We assume that the datapath densely chooses port numbers, which + * can therefore be used as an index into an array of channels. */ + if (port_idx >= dpif->uc_array_size) { + uint32_t new_size = port_idx + 1; + uint32_t i; + + if (new_size > MAX_PORTS) { + 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_idx; + if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock), + &event) < 0) { + return errno; } + + nl_sock_destroy(dpif->channels[port_idx].sock); + dpif->channels[port_idx].sock = sock; + dpif->channels[port_idx].last_poll = LLONG_MIN; + return 0; +} -error_free: - free(dpif); - return error; +static void +del_channel(struct dpif_linux *dpif, odp_port_t port_no) +{ + struct dpif_channel *ch; + uint32_t port_idx = odp_to_u32(port_no); + + if (dpif->epoll_fd < 0 || port_idx >= dpif->uc_array_size) { + return; + } + + ch = &dpif->channels[port_idx]; + if (!ch->sock) { + return; + } + + epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL); + dpif->event_offset = dpif->n_events = 0; + + nl_sock_destroy(ch->sock); + ch->sock = NULL; } static void dpif_linux_close(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - rtnetlink_link_notifier_unregister(&dpif->port_notifier); - sset_destroy(&dpif->changed_ports); - free(dpif->lru_bitmap); + + nl_sock_destroy(dpif->port_notifier); + destroy_channels(dpif); + if (dpif->epoll_fd >= 0) { + close(dpif->epoll_fd); + } + ovs_mutex_destroy(&dpif->upcall_lock); free(dpif); } @@ -299,13 +393,13 @@ dpif_linux_destroy(struct dpif *dpif_) struct dpif_linux_dp dp; dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_CMD_DEL; + dp.cmd = OVS_DP_CMD_DEL; dp.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&dp, NULL, NULL); } static int -dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats) +dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) { struct dpif_linux_dp dp; struct ofpbuf *buf; @@ -313,106 +407,210 @@ dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats) error = dpif_linux_dp_get(dpif_, &dp, &buf); if (!error) { - *stats = dp.stats; + 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_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) +static const char * +get_vport_type(const struct dpif_linux_vport *vport) { - struct dpif_linux_dp dp; - struct ofpbuf *buf; - int error; + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - error = dpif_linux_dp_get(dpif_, &dp, &buf); - if (!error) { - *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP; - ofpbuf_delete(buf); + 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_LISP: + return "lisp"; + + case OVS_VPORT_TYPE_UNSPEC: + case __OVS_VPORT_TYPE_MAX: + break; } - return error; + + VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u", + vport->dp_ifindex, vport->name, (unsigned int) vport->type); + return "unknown"; } -static int -dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) +static enum ovs_vport_type +netdev_to_ovs_vport_type(const struct netdev *netdev) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_dp dp; + const char *type = netdev_get_type(netdev); - dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_CMD_SET; - dp.dp_ifindex = dpif->dp_ifindex; - dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO; - return dpif_linux_dp_transact(&dp, NULL, NULL); + 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 if (!strcmp(type, "lisp")) { + return OVS_VPORT_TYPE_LISP; + } else { + return OVS_VPORT_TYPE_UNSPEC; + } } static int -dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, - uint16_t *port_nop) +dpif_linux_port_add__(struct dpif *dpif_, struct netdev *netdev, + odp_port_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; + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; + const char *name = netdev_vport_get_dpif_port(netdev, + namebuf, sizeof namebuf); 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; + 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 = ODP_VPORT_CMD_NEW; + request.cmd = OVS_VPORT_CMD_NEW; request.dp_ifindex = dpif->dp_ifindex; - request.type = netdev_vport_get_vport_type(netdev); - if (request.type == ODP_VPORT_TYPE_UNSPEC) { + 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 { - request.port_no = dpif_linux_pop_port(dpif); - error = dpif_linux_vport_transact(&request, &reply, &buf); + 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; + } - if (!error) { - *port_nop = reply.port_no; + request.port_no = *port_nop; + upcall_pid = sock ? nl_sock_pid(sock) : 0; + 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_), reply.port_no, upcall_pid); + } else { + if (error == EBUSY && *port_nop != ODPP_NONE) { + VLOG_INFO("%s: requested port %"PRIu32" is in use", + dpif_name(dpif_), *port_nop); } + nl_sock_destroy(sock); ofpbuf_delete(buf); - } while (request.port_no != UINT32_MAX - && (error == EBUSY || error == EFBIG)); + return error; + } + ofpbuf_delete(buf); + + 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_add(struct dpif *dpif_, struct netdev *netdev, + odp_port_t *port_nop) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + int error; + + ovs_mutex_lock(&dpif->upcall_lock); + error = dpif_linux_port_add__(dpif_, netdev, port_nop); + ovs_mutex_unlock(&dpif->upcall_lock); return error; } static int -dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) +dpif_linux_port_del__(struct dpif *dpif_, odp_port_t port_no) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_vport vport; int error; dpif_linux_vport_init(&vport); - vport.cmd = ODP_VPORT_CMD_DEL; + vport.cmd = OVS_VPORT_CMD_DEL; vport.dp_ifindex = dpif->dp_ifindex; vport.port_no = port_no; error = dpif_linux_vport_transact(&vport, NULL, NULL); - if (!error) { - dpif_linux_push_port(dpif, port_no); - } + del_channel(dpif, port_no); + return error; } static int -dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, +dpif_linux_port_del(struct dpif *dpif_, odp_port_t port_no) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + int error; + + ovs_mutex_lock(&dpif->upcall_lock); + error = dpif_linux_port_del__(dpif_, port_no); + ovs_mutex_unlock(&dpif->upcall_lock); + + return error; +} + +static int +dpif_linux_port_query__(const struct dpif *dpif, odp_port_t port_no, const char *port_name, struct dpif_port *dpif_port) { struct dpif_linux_vport request; @@ -421,23 +619,29 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, int error; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_CMD_GET; + request.cmd = OVS_VPORT_CMD_GET; request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex; request.port_no = port_no; request.name = port_name; 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 if (dpif_port) { + dpif_port->name = xstrdup(reply.name); + dpif_port->type = xstrdup(get_vport_type(&reply)); + dpif_port->port_no = reply.port_no; + } ofpbuf_delete(buf); } return error; } 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, odp_port_t port_no, struct dpif_port *dpif_port) { return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port); @@ -450,22 +654,39 @@ dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname, return dpif_linux_port_query__(dpif, 0, devname, dpif_port); } -static int +static uint32_t dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED) { - /* If the datapath increases its range of supported ports, then it should - * start reporting that. */ - return 1024; + return MAX_PORTS; +} + +static uint32_t +dpif_linux_port_get_pid(const struct dpif *dpif_, odp_port_t port_no) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + uint32_t port_idx = odp_to_u32(port_no); + uint32_t pid = 0; + + ovs_mutex_lock(&dpif->upcall_lock); + if (dpif->epoll_fd >= 0) { + /* 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); + } + ovs_mutex_unlock(&dpif->upcall_lock); + + return pid; } static int dpif_linux_flow_flush(struct dpif *dpif_) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow flow; dpif_linux_flow_init(&flow); - flow.cmd = ODP_FLOW_CMD_DEL; + flow.cmd = OVS_FLOW_CMD_DEL; flow.dp_ifindex = dpif->dp_ifindex; return dpif_linux_flow_transact(&flow, NULL, NULL); } @@ -477,7 +698,7 @@ struct dpif_linux_port_state { static int dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_port_state *state; struct dpif_linux_vport request; struct ofpbuf *buf; @@ -485,12 +706,12 @@ dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep) *statep = state = xmalloc(sizeof *state); dpif_linux_vport_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; buf = ofpbuf_new(1024); dpif_linux_vport_to_ofpbuf(&request, buf); - nl_dump_start(&state->dump, genl_sock, buf); + nl_dump_start(&state->dump, NETLINK_GENERIC, buf); ofpbuf_delete(buf); return 0; @@ -514,17 +735,18 @@ 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; } static int -dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) +dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_) { struct dpif_linux_port_state *state = state_; int error = nl_dump_done(&state->dump); + free(state); return error; } @@ -534,26 +756,75 @@ dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (dpif->change_error) { - dpif->change_error = false; - sset_clear(&dpif->changed_ports); + /* Lazily create the Netlink socket to listen for notifications. */ + if (!dpif->port_notifier) { + struct nl_sock *sock; + int error; + + error = nl_sock_create(NETLINK_GENERIC, &sock); + if (error) { + return error; + } + + error = nl_sock_join_mcgroup(sock, ovs_vport_mcgroup); + if (error) { + nl_sock_destroy(sock); + return error; + } + dpif->port_notifier = sock; + + /* We have no idea of the current state so report that everything + * changed. */ return ENOBUFS; - } else if (!sset_is_empty(&dpif->changed_ports)) { - *devnamep = sset_pop(&dpif->changed_ports); - return 0; - } else { - return EAGAIN; + } + + for (;;) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + uint64_t buf_stub[4096 / 8]; + struct ofpbuf buf; + int error; + + ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); + error = nl_sock_recv(dpif->port_notifier, &buf, false); + if (!error) { + struct dpif_linux_vport vport; + + error = dpif_linux_vport_from_ofpbuf(&vport, &buf); + if (!error) { + if (vport.dp_ifindex == dpif->dp_ifindex + && (vport.cmd == OVS_VPORT_CMD_NEW + || vport.cmd == OVS_VPORT_CMD_DEL + || vport.cmd == OVS_VPORT_CMD_SET)) { + VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8, + dpif->dpif.full_name, vport.name, vport.cmd); + *devnamep = xstrdup(vport.name); + ofpbuf_uninit(&buf); + return 0; + } + } + } else if (error != EAGAIN) { + VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)", + ovs_strerror(error)); + nl_sock_drain(dpif->port_notifier); + error = ENOBUFS; + } + + ofpbuf_uninit(&buf); + if (error) { + return error; + } } } static void dpif_linux_port_poll_wait(const struct dpif *dpif_) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) { - poll_immediate_wake(); + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + if (dpif->port_notifier) { + nl_sock_wait(dpif->port_notifier, POLLIN); } else { - rtnetlink_link_notifier_wait(); + poll_immediate_wake(); } } @@ -562,11 +833,11 @@ dpif_linux_flow_get__(const struct dpif *dpif_, const struct nlattr *key, size_t key_len, struct dpif_linux_flow *reply, struct ofpbuf **bufp) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow request; dpif_linux_flow_init(&request); - request.cmd = ODP_FLOW_CMD_GET; + request.cmd = OVS_FLOW_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; @@ -588,7 +859,7 @@ 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 { @@ -598,60 +869,77 @@ dpif_linux_flow_get(const struct dpif *dpif_, return error; } +static void +dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put, + struct dpif_linux_flow *request) +{ + static const struct nlattr dummy_action; + + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + dpif_linux_flow_init(request); + request->cmd = (put->flags & DPIF_FP_CREATE + ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET); + request->dp_ifindex = dpif->dp_ifindex; + request->key = put->key; + request->key_len = put->key_len; + request->mask = put->mask; + request->mask_len = put->mask_len; + /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */ + request->actions = (put->actions + ? put->actions + : CONST_CAST(struct nlattr *, &dummy_action)); + request->actions_len = put->actions_len; + if (put->flags & DPIF_FP_ZERO_STATS) { + request->clear = true; + } + request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; +} + static int -dpif_linux_flow_put(struct dpif *dpif_, 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 *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 ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET; - request.dp_ifindex = dpif->dp_ifindex; - request.key = key; - request.key_len = key_len; - /* Ensure that ODP_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_, 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; } +static void +dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del, + struct dpif_linux_flow *request) +{ + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + dpif_linux_flow_init(request); + request->cmd = OVS_FLOW_CMD_DEL; + request->dp_ifindex = dpif->dp_ifindex; + request->key = del->key; + request->key_len = del->key_len; +} + static int -dpif_linux_flow_del(struct dpif *dpif_, - const struct nlattr *key, size_t key_len, - struct dpif_flow_stats *stats) +dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow request, reply; struct ofpbuf *buf; int error; - dpif_linux_flow_init(&request); - request.cmd = ODP_FLOW_CMD_DEL; - request.dp_ifindex = dpif->dp_ifindex; - request.key = key; - request.key_len = key_len; + dpif_linux_init_flow_del(dpif_, del, &request); error = dpif_linux_flow_transact(&request, - stats ? &reply : NULL, - stats ? &buf : NULL); - if (!error && stats) { - dpif_linux_flow_get_stats(&reply, stats); + del->stats ? &reply : NULL, + del->stats ? &buf : NULL); + if (!error && del->stats) { + dpif_linux_flow_get_stats(&reply, del->stats); ofpbuf_delete(buf); } return error; @@ -667,7 +955,7 @@ struct dpif_linux_flow_state { static int dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow_state *state; struct dpif_linux_flow request; struct ofpbuf *buf; @@ -675,12 +963,12 @@ dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) *statep = state = xmalloc(sizeof *state); dpif_linux_flow_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; buf = ofpbuf_new(1024); dpif_linux_flow_to_ofpbuf(&request, buf); - nl_dump_start(&state->dump, genl_sock, buf); + nl_dump_start(&state->dump, NETLINK_GENERIC, buf); ofpbuf_delete(buf); state->buf = NULL; @@ -691,6 +979,7 @@ 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_, const struct nlattr **key, size_t *key_len, + const struct nlattr **mask, size_t *mask_len, const struct nlattr **actions, size_t *actions_len, const struct dpif_flow_stats **stats) { @@ -718,7 +1007,8 @@ dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_, if (error == ENOENT) { VLOG_DBG("dumped flow disappeared on get"); } else if (error) { - VLOG_WARN("error fetching dumped flow: %s", strerror(error)); + VLOG_WARN("error fetching dumped flow: %s", + ovs_strerror(error)); } } } while (error); @@ -731,6 +1021,10 @@ dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_, *key = state->flow.key; *key_len = state->flow.key_len; } + if (mask) { + *mask = state->flow.mask; + *mask_len = state->flow.mask ? state->flow.mask_len : 0; + } if (stats) { dpif_linux_flow_get_stats(&state->flow, &state->stats); *stats = &state->stats; @@ -748,113 +1042,280 @@ dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) return error; } -static int -dpif_linux_execute(struct dpif *dpif_, - const struct nlattr *actions, size_t actions_len, - const struct ofpbuf *packet) +static void +dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec, + struct ofpbuf *buf) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct odp_header *execute; - struct ofpbuf *buf; - int error; + struct ovs_header *k_exec; - buf = ofpbuf_new(128 + actions_len + 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, odp_packet_family, NLM_F_REQUEST, - ODP_PACKET_CMD_EXECUTE, 1); + nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST, + OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION); - execute = ofpbuf_put_uninit(buf, sizeof *execute); - execute->dp_ifindex = dpif->dp_ifindex; + k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec); + k_exec->dp_ifindex = dp_ifindex; - nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size); - nl_msg_put_unspec(buf, ODP_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); +} + +static int +dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute) +{ + uint64_t request_stub[1024 / 8]; + struct ofpbuf request; + int error; + + ofpbuf_use_stub(&request, request_stub, sizeof request_stub); + dpif_linux_encode_execute(dp_ifindex, execute, &request); + error = nl_transact(NETLINK_GENERIC, &request, NULL); + ofpbuf_uninit(&request); - error = nl_sock_transact(genl_sock, buf, NULL); - ofpbuf_delete(buf); return error; } static int -dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) +dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - *listen_mask = dpif->listen_mask; - return 0; + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + return dpif_linux_execute__(dpif->dp_ifindex, execute); } -static int -dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask) +#define MAX_OPS 50 + +static void +dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int error; - int i; + const struct dpif_linux *dpif = dpif_linux_cast(dpif_); + + 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; + + ovs_assert(n_ops <= MAX_OPS); + for (i = 0; i < n_ops; 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 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, &flow); + if (put->stats) { + flow.nlmsg_flags |= NLM_F_ECHO; + aux->txn.reply = &aux->reply; + } + 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, &flow); + if (del->stats) { + flow.nlmsg_flags |= NLM_F_ECHO; + aux->txn.reply = &aux->reply; + } + dpif_linux_flow_to_ofpbuf(&flow, &aux->request); + break; - 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; + case DPIF_OP_EXECUTE: + execute = &op->u.execute; + dpif_linux_encode_execute(dpif->dp_ifindex, execute, + &aux->request); + break; + + default: + NOT_REACHED(); } } - /* 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]); - } + for (i = 0; i < n_ops; i++) { + txnsp[i] = &auxes[i].txn; } + nl_transact_multiple(NETLINK_GENERIC, txnsp, n_ops); - /* Update listen_mask. */ - dpif->listen_mask = listen_mask; + for (i = 0; i < n_ops; 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; - /* Subscribe to new groups. */ - error = 0; - for (i = 0; i < DPIF_N_UC_TYPES; i++) { - if (dpif->listen_mask & (1u << i)) { - int retval; + op->error = txn->error; - retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]); - if (retval) { - error = retval; + switch (op->type) { + case DPIF_OP_FLOW_PUT: + put = &op->u.flow_put; + if (put->stats) { + if (!op->error) { + 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 (del->stats) { + if (!op->error) { + 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; + + case DPIF_OP_EXECUTE: + break; + + default: + NOT_REACHED(); } + + ofpbuf_uninit(&aux->request); + ofpbuf_uninit(&aux->reply); + } +} + +static void +dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) +{ + while (n_ops > 0) { + size_t chunk = MIN(n_ops, MAX_OPS); + dpif_linux_operate__(dpif, ops, chunk); + ops += chunk; + n_ops -= chunk; } - return error; } static int -dpif_linux_get_sflow_probability(const struct dpif *dpif_, - uint32_t *probability) +dpif_linux_recv_set__(struct dpif *dpif_, bool enable) { - struct dpif_linux_dp dp; - struct ofpbuf *buf; - int error; + struct dpif_linux *dpif = dpif_linux_cast(dpif_); - error = dpif_linux_dp_get(dpif_, &dp, &buf); - if (!error) { - *probability = dp.sampling ? *dp.sampling : 0; - ofpbuf_delete(buf); + if ((dpif->epoll_fd >= 0) == enable) { + return 0; } - return error; + + if (!enable) { + destroy_channels(dpif); + } else { + struct dpif_port_dump port_dump; + struct dpif_port port; + + if (dpif->epoll_fd < 0) { + dpif->epoll_fd = epoll_create(10); + if (dpif->epoll_fd < 0) { + return errno; + } + } + + 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, &sock); + if (error) { + return error; + } + + 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), ovs_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; + } + } + } + + return 0; } static int -dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability) +dpif_linux_recv_set(struct dpif *dpif_, bool enable) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_dp dp; + int error; - dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_CMD_SET; - dp.dp_ifindex = dpif->dp_ifindex; - dp.sampling = &probability; - return dpif_linux_dp_transact(&dp, NULL, NULL); + ovs_mutex_lock(&dpif->upcall_lock); + error = dpif_linux_recv_set__(dpif_, enable); + ovs_mutex_unlock(&dpif->upcall_lock); + + return error; } static int @@ -873,22 +1334,18 @@ static int parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, int *dp_ifindex) { - static const struct nl_policy odp_packet_policy[] = { + static const struct nl_policy ovs_packet_policy[] = { /* Always present. */ - [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC, + [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC, .min_len = ETH_HEADER_LEN }, - [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, - - /* ODP_PACKET_CMD_ACTION only. */ - [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true }, + [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, - /* ODP_PACKET_CMD_SAMPLE only. */ - [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true }, - [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, + /* OVS_PACKET_CMD_ACTION only. */ + [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true }, }; - struct odp_header *odp_header; - struct nlattr *a[ARRAY_SIZE(odp_packet_policy)]; + struct ovs_header *ovs_header; + struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)]; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -898,17 +1355,16 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_packet_family - || !nl_policy_parse(&b, 0, odp_packet_policy, a, - ARRAY_SIZE(odp_packet_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_packet_family + || !nl_policy_parse(&b, 0, ovs_packet_policy, a, + ARRAY_SIZE(ovs_packet_policy))) { return EINVAL; } - type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS - : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION - : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE + type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS + : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION : -1); if (type < 0) { return EINVAL; @@ -917,69 +1373,114 @@ 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[ODP_PACKET_ATTR_PACKET]); - upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]); - upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]); - upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]); - upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA] - ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA]) - : 0); - upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL] - ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL]) - : 0); - if (a[ODP_PACKET_ATTR_ACTIONS]) { - upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]); - upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]); - } - - *dp_ifindex = odp_header->dp_ifindex; + 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]; + *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_); - struct ofpbuf *buf; - int error; - int i; + int read_tries = 0; - if (!dpif->mc_sock) { - return EAGAIN; + if (dpif->epoll_fd < 0) { + return EAGAIN; } - for (i = 0; i < 50; i++) { - int dp_ifindex; + if (dpif->event_offset >= dpif->n_events) { + int retval; - error = nl_sock_recv(dpif->mc_sock, &buf, false); - if (error) { - return error; - } + dpif->event_offset = dpif->n_events = 0; - error = parse_odp_packet(buf, upcall, &dp_ifindex); - if (!error - && dp_ifindex == dpif->dp_ifindex - && dpif->listen_mask & (1u << upcall->type)) { - return 0; + do { + 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)", ovs_strerror(errno)); + } else if (retval > 0) { + dpif->n_events = retval; } + } - ofpbuf_delete(buf); - if (error) { - return error; + 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->event_offset++; + + for (;;) { + int dp_ifindex; + int error; + + if (++read_tries > 50) { + return EAGAIN; + } + + 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; + } else if (error) { + return error; + } } } return EAGAIN; } +static int +dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall, + struct ofpbuf *buf) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + int error; + + ovs_mutex_lock(&dpif->upcall_lock); + error = dpif_linux_recv__(dpif_, upcall, buf); + ovs_mutex_unlock(&dpif->upcall_lock); + + return error; +} + 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); + + ovs_mutex_lock(&dpif->upcall_lock); + if (dpif->epoll_fd >= 0) { + poll_fd_wait(dpif->epoll_fd, POLLIN); } + ovs_mutex_unlock(&dpif->upcall_lock); } static void @@ -987,27 +1488,36 @@ dpif_linux_recv_purge(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (dpif->mc_sock) { - nl_sock_drain(dpif->mc_sock); + ovs_mutex_lock(&dpif->upcall_lock); + if (dpif->epoll_fd >= 0) { + struct dpif_channel *ch; + + for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; + ch++) { + if (ch->sock) { + nl_sock_drain(ch->sock); + } + } } + ovs_mutex_unlock(&dpif->upcall_lock); } const struct dpif_class dpif_linux_class = { "system", dpif_linux_enumerate, + NULL, dpif_linux_open, dpif_linux_close, dpif_linux_destroy, NULL, /* run */ NULL, /* 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, @@ -1021,10 +1531,8 @@ const struct dpif_class dpif_linux_class = { dpif_linux_flow_dump_next, dpif_linux_flow_dump_done, dpif_linux_execute, - dpif_linux_recv_get_mask, - dpif_linux_recv_set_mask, - dpif_linux_get_sflow_probability, - dpif_linux_set_sflow_probability, + dpif_linux_operate, + dpif_linux_recv_set, dpif_linux_queue_to_priority, dpif_linux_recv, dpif_linux_recv_wait, @@ -1034,29 +1542,33 @@ const struct dpif_class dpif_linux_class = { static int dpif_linux_init(void) { - static int error = -1; + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static int error; - if (error < 0) { - error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY, - &odp_datapath_family); + if (ovsthread_once_start(&once)) { + error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY, + &ovs_datapath_family); if (error) { VLOG_ERR("Generic Netlink family '%s' does not exist. " "The Open vSwitch kernel module is probably not loaded.", - ODP_DATAPATH_FAMILY); + OVS_DATAPATH_FAMILY); } if (!error) { - error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family); + error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family); } if (!error) { - error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family); + error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family); } if (!error) { - error = nl_lookup_genl_family(ODP_PACKET_FAMILY, - &odp_packet_family); + error = nl_lookup_genl_family(OVS_PACKET_FAMILY, + &ovs_packet_family); } if (!error) { - error = nl_sock_create(NETLINK_GENERIC, &genl_sock); + error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP, + &ovs_vport_mcgroup); } + + ovsthread_once_done(&once); } return error; @@ -1074,61 +1586,13 @@ dpif_linux_is_internal_device(const char *name) ofpbuf_delete(buf); } else if (error != ENODEV && error != ENOENT) { VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)", - name, strerror(error)); + name, ovs_strerror(error)); } - return reply.type == ODP_VPORT_TYPE_INTERNAL; -} - -int -dpif_linux_vport_send(int dp_ifindex, uint32_t port_no, - const void *data, size_t size) -{ - struct odp_header *execute; - struct ofpbuf *buf; - size_t actions_ofs; - int error; - - buf = ofpbuf_new(128 + size); - - nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST, - ODP_PACKET_CMD_EXECUTE, 1); - - execute = ofpbuf_put_uninit(buf, sizeof *execute); - execute->dp_ifindex = dp_ifindex; - - nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, data, size); - - actions_ofs = nl_msg_start_nested(buf, ODP_PACKET_ATTR_ACTIONS); - nl_msg_put_u32(buf, ODP_ACTION_ATTR_OUTPUT, port_no); - nl_msg_end_nested(buf, actions_ofs); - - error = nl_sock_transact(genl_sock, buf, NULL); - ofpbuf_delete(buf); - return error; -} - -static void -dpif_linux_port_changed(const struct rtnetlink_link_change *change, - void *dpif_) -{ - struct dpif_linux *dpif = dpif_; - - if (change) { - if (change->master_ifindex == dpif->dp_ifindex - && (change->nlmsg_type == RTM_NEWLINK - || change->nlmsg_type == RTM_DELLINK)) - { - /* Our datapath changed, either adding a new port or deleting an - * existing one. */ - sset_add(&dpif->changed_ports, change->ifname); - } - } else { - dpif->change_error = true; - } + return reply.type == OVS_VPORT_TYPE_INTERNAL; } -/* Parses the contents of 'buf', which contains a "struct odp_header" followed +/* Parses the contents of 'buf', which contains a "struct ovs_header" followed * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a * positive errno value. * @@ -1138,26 +1602,18 @@ static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, const struct ofpbuf *buf) { - static const struct nl_policy odp_vport_policy[] = { - [ODP_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 }, - [ODP_VPORT_ATTR_TYPE] = { .type = NL_A_U32 }, - [ODP_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, - [ODP_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC, - .min_len = sizeof(struct rtnl_link_stats64), - .max_len = sizeof(struct rtnl_link_stats64), + static const struct nl_policy ovs_vport_policy[] = { + [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_UPCALL_PID] = { .type = NL_A_U32 }, + [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats), .optional = true }, - [ODP_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC, - .min_len = ETH_ADDR_LEN, - .max_len = ETH_ADDR_LEN, - .optional = true }, - [ODP_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true }, - [ODP_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, - [ODP_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true }, - [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true }, + [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, }; - struct nlattr *a[ARRAY_SIZE(odp_vport_policy)]; - struct odp_header *odp_header; + struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)]; + struct ovs_header *ovs_header; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -1167,95 +1623,71 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, ofpbuf_use_const(&b, buf->data, buf->size); nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_vport_family - || !nl_policy_parse(&b, 0, odp_vport_policy, a, - ARRAY_SIZE(odp_vport_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_vport_family + || !nl_policy_parse(&b, 0, ovs_vport_policy, a, + ARRAY_SIZE(ovs_vport_policy))) { return EINVAL; } vport->cmd = genl->cmd; - vport->dp_ifindex = odp_header->dp_ifindex; - vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); - vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]); - vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]); - if (a[ODP_VPORT_ATTR_STATS]) { - vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]); - } - if (a[ODP_VPORT_ATTR_ADDRESS]) { - vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]); - } - if (a[ODP_VPORT_ATTR_MTU]) { - vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]); - } else { - vport->mtu = INT_MAX; - } - if (a[ODP_VPORT_ATTR_OPTIONS]) { - vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]); - vport->options_len = nl_attr_get_size(a[ODP_VPORT_ATTR_OPTIONS]); + vport->dp_ifindex = ovs_header->dp_ifindex; + vport->port_no = nl_attr_get_odp_port(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[ODP_VPORT_ATTR_IFINDEX]) { - vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]); + if (a[OVS_VPORT_ATTR_STATS]) { + vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]); } - if (a[ODP_VPORT_ATTR_IFLINK]) { - vport->iflink = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFLINK]); + 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]); } return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_header" +/* Appends to 'buf' (which must initially be empty) a "struct ovs_header" * followed by Netlink attributes corresponding to 'vport'. */ static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, struct ofpbuf *buf) { - struct odp_header *odp_header; + struct ovs_header *ovs_header; - nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO, - vport->cmd, 1); + nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO, + vport->cmd, OVS_VPORT_VERSION); - odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_ifindex = vport->dp_ifindex; + ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); + ovs_header->dp_ifindex = vport->dp_ifindex; - if (vport->port_no != UINT32_MAX) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no); + if (vport->port_no != ODPP_NONE) { + nl_msg_put_odp_port(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no); } - if (vport->type != ODP_VPORT_TYPE_UNSPEC) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_TYPE, vport->type); + if (vport->type != OVS_VPORT_TYPE_UNSPEC) { + nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type); } if (vport->name) { - nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name); - } - - if (vport->stats) { - nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS, - vport->stats, sizeof *vport->stats); + nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name); } - if (vport->address) { - nl_msg_put_unspec(buf, ODP_VPORT_ATTR_ADDRESS, - vport->address, ETH_ADDR_LEN); + if (vport->upcall_pid) { + nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid); } - if (vport->mtu && vport->mtu != INT_MAX) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu); + if (vport->stats) { + nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS, + vport->stats, sizeof *vport->stats); } if (vport->options) { - nl_msg_put_nested(buf, ODP_VPORT_ATTR_OPTIONS, + nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS, vport->options, vport->options_len); } - - if (vport->ifindex) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFINDEX, vport->ifindex); - } - - if (vport->iflink) { - nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink); - } } /* Clears 'vport' to "empty" values. */ @@ -1263,13 +1695,13 @@ void dpif_linux_vport_init(struct dpif_linux_vport *vport) { memset(vport, 0, sizeof *vport); - vport->port_no = UINT32_MAX; + vport->port_no = ODPP_NONE; } /* Executes 'request' in the kernel datapath. If the command fails, returns a * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the - * result of the command is expected to be an odp_vport also, which is decoded + * result of the command is expected to be an ovs_vport also, which is decoded * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ int @@ -1280,7 +1712,7 @@ 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) { @@ -1293,7 +1725,7 @@ dpif_linux_vport_transact(const struct dpif_linux_vport *request, request_buf = ofpbuf_new(1024); dpif_linux_vport_to_ofpbuf(request, request_buf); - error = nl_sock_transact(genl_sock, request_buf, bufp); + error = nl_transact(NETLINK_GENERIC, request_buf, bufp); ofpbuf_delete(request_buf); if (reply) { @@ -1319,13 +1751,13 @@ dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply, struct dpif_linux_vport request; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_CMD_GET; + request.cmd = OVS_VPORT_CMD_GET; request.name = name; return dpif_linux_vport_transact(&request, reply, bufp); } -/* Parses the contents of 'buf', which contains a "struct odp_header" followed +/* Parses the contents of 'buf', which contains a "struct ovs_header" followed * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a * positive errno value. * @@ -1334,19 +1766,14 @@ dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply, static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) { - static const struct nl_policy odp_datapath_policy[] = { - [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, - [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC, - .min_len = sizeof(struct odp_stats), - .max_len = sizeof(struct odp_stats), + static const struct nl_policy ovs_datapath_policy[] = { + [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, + [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats), .optional = true }, - [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true }, - [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true }, - [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true }, }; - struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)]; - struct odp_header *odp_header; + struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)]; + struct ovs_header *ovs_header; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -1356,57 +1783,23 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) ofpbuf_use_const(&b, buf->data, buf->size); nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_datapath_family - || !nl_policy_parse(&b, 0, odp_datapath_policy, a, - ARRAY_SIZE(odp_datapath_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_datapath_family + || !nl_policy_parse(&b, 0, ovs_datapath_policy, a, + ARRAY_SIZE(ovs_datapath_policy))) { return EINVAL; } dp->cmd = genl->cmd; - dp->dp_ifindex = odp_header->dp_ifindex; - dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]); - if (a[ODP_DP_ATTR_STATS]) { + dp->dp_ifindex = ovs_header->dp_ifindex; + dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]); + if (a[OVS_DP_ATTR_STATS]) { /* Can't use structure assignment because Netlink doesn't ensure * sufficient alignment for 64-bit members. */ - memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]), + memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]), sizeof dp->stats); } - if (a[ODP_DP_ATTR_IPV4_FRAGS]) { - dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); - } - if (a[ODP_DP_ATTR_SAMPLING]) { - dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]); - } - - if (a[ODP_DP_ATTR_MCGROUPS]) { - static const struct nl_policy odp_mcgroup_policy[] = { - [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true }, - [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true }, - [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true }, - }; - - struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)]; - - if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy, - mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) { - return EINVAL; - } - - if (mcgroups[ODP_PACKET_CMD_MISS]) { - dp->mcgroups[DPIF_UC_MISS] - = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]); - } - if (mcgroups[ODP_PACKET_CMD_ACTION]) { - dp->mcgroups[DPIF_UC_ACTION] - = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]); - } - if (mcgroups[ODP_PACKET_CMD_SAMPLE]) { - dp->mcgroups[DPIF_UC_SAMPLE] - = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]); - } - } return 0; } @@ -1415,31 +1808,28 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) static void dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) { - struct odp_header *odp_header; + struct ovs_header *ovs_header; - nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family, - NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1); + nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family, + NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, + OVS_DATAPATH_VERSION); - odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_ifindex = dp->dp_ifindex; + ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); + ovs_header->dp_ifindex = dp->dp_ifindex; if (dp->name) { - nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name); + nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name); } - /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */ - - if (dp->ipv4_frags) { - nl_msg_put_u32(buf, ODP_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, ODP_DP_ATTR_SAMPLING, *dp->sampling); - } + /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */ } /* Clears 'dp' to "empty" values. */ -void +static void dpif_linux_dp_init(struct dpif_linux_dp *dp) { memset(dp, 0, sizeof *dp); @@ -1452,11 +1842,11 @@ dpif_linux_dp_dump_start(struct nl_dump *dump) struct ofpbuf *buf; dpif_linux_dp_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; buf = ofpbuf_new(1024); dpif_linux_dp_to_ofpbuf(&request, buf); - nl_dump_start(dump, genl_sock, buf); + nl_dump_start(dump, NETLINK_GENERIC, buf); ofpbuf_delete(buf); } @@ -1466,18 +1856,18 @@ dpif_linux_dp_dump_start(struct nl_dump *dump) * result of the command is expected to be of the same form, which is decoded * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ -int +static int dpif_linux_dp_transact(const struct dpif_linux_dp *request, struct dpif_linux_dp *reply, struct ofpbuf **bufp) { 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); - error = nl_sock_transact(genl_sock, request_buf, bufp); + error = nl_transact(NETLINK_GENERIC, request_buf, bufp); ofpbuf_delete(request_buf); if (reply) { @@ -1496,7 +1886,7 @@ dpif_linux_dp_transact(const struct dpif_linux_dp *request, /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'. * The caller must free '*bufp' when the reply is no longer needed ('reply' * will contain pointers into '*bufp'). */ -int +static int dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, struct ofpbuf **bufp) { @@ -1504,13 +1894,13 @@ dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, struct dpif_linux_dp request; dpif_linux_dp_init(&request); - request.cmd = ODP_DP_CMD_GET; + request.cmd = OVS_DP_CMD_GET; request.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&request, reply, bufp); } -/* Parses the contents of 'buf', which contains a "struct odp_header" followed +/* Parses the contents of 'buf', which contains a "struct ovs_header" followed * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a * positive errno value. * @@ -1520,20 +1910,19 @@ static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, const struct ofpbuf *buf) { - static const struct nl_policy odp_flow_policy[] = { - [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED }, - [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, - [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC, - .min_len = sizeof(struct odp_flow_stats), - .max_len = sizeof(struct odp_flow_stats), + static const struct nl_policy ovs_flow_policy[] = { + [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED }, + [OVS_FLOW_ATTR_MASK] = { .type = NL_A_NESTED, .optional = true }, + [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, + [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats), .optional = true }, - [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, - [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, - /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */ + [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, + [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, + /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */ }; - struct nlattr *a[ARRAY_SIZE(odp_flow_policy)]; - struct odp_header *odp_header; + struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)]; + struct ovs_header *ovs_header; struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; @@ -1543,70 +1932,79 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, ofpbuf_use_const(&b, buf->data, buf->size); nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); genl = ofpbuf_try_pull(&b, sizeof *genl); - odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); - if (!nlmsg || !genl || !odp_header - || nlmsg->nlmsg_type != odp_flow_family - || !nl_policy_parse(&b, 0, odp_flow_policy, a, - ARRAY_SIZE(odp_flow_policy))) { + ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); + if (!nlmsg || !genl || !ovs_header + || nlmsg->nlmsg_type != ovs_flow_family + || !nl_policy_parse(&b, 0, ovs_flow_policy, a, + ARRAY_SIZE(ovs_flow_policy))) { return EINVAL; } flow->nlmsg_flags = nlmsg->nlmsg_flags; - flow->dp_ifindex = odp_header->dp_ifindex; - flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]); - flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]); - if (a[ODP_FLOW_ATTR_ACTIONS]) { - flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]); - flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]); + flow->dp_ifindex = ovs_header->dp_ifindex; + flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]); + flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]); + + if (a[OVS_FLOW_ATTR_MASK]) { + flow->mask = nl_attr_get(a[OVS_FLOW_ATTR_MASK]); + flow->mask_len = nl_attr_get_size(a[OVS_FLOW_ATTR_MASK]); + } + if (a[OVS_FLOW_ATTR_ACTIONS]) { + flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]); + flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]); } - if (a[ODP_FLOW_ATTR_STATS]) { - flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]); + if (a[OVS_FLOW_ATTR_STATS]) { + flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]); } - if (a[ODP_FLOW_ATTR_TCP_FLAGS]) { - flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]); + if (a[OVS_FLOW_ATTR_TCP_FLAGS]) { + flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]); } - if (a[ODP_FLOW_ATTR_USED]) { - flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]); + if (a[OVS_FLOW_ATTR_USED]) { + flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]); } return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_header" +/* Appends to 'buf' (which must initially be empty) a "struct ovs_header" * followed by Netlink attributes corresponding to 'flow'. */ static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, struct ofpbuf *buf) { - struct odp_header *odp_header; + struct ovs_header *ovs_header; - nl_msg_put_genlmsghdr(buf, 0, odp_flow_family, - NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags, - flow->cmd, 1); + nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family, + NLM_F_REQUEST | flow->nlmsg_flags, + flow->cmd, OVS_FLOW_VERSION); - odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_ifindex = flow->dp_ifindex; + ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); + ovs_header->dp_ifindex = flow->dp_ifindex; if (flow->key_len) { - nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len); + nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len); + } + + if (flow->mask_len) { + nl_msg_put_unspec(buf, OVS_FLOW_ATTR_MASK, flow->mask, flow->mask_len); } if (flow->actions || flow->actions_len) { - nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS, + nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS, flow->actions, flow->actions_len); } /* 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, ODP_FLOW_ATTR_CLEAR); + nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR); } } /* Clears 'flow' to "empty" values. */ -void +static void dpif_linux_flow_init(struct dpif_linux_flow *flow) { memset(flow, 0, sizeof *flow); @@ -1618,18 +2016,22 @@ dpif_linux_flow_init(struct dpif_linux_flow *flow) * result of the command is expected to be a flow also, which is decoded and * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply * is no longer needed ('reply' will contain pointers into '*bufp'). */ -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) { 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; + } request_buf = ofpbuf_new(1024); dpif_linux_flow_to_ofpbuf(request, request_buf); - error = nl_sock_transact(genl_sock, request_buf, bufp); + error = nl_transact(NETLINK_GENERIC, request_buf, bufp); ofpbuf_delete(request_buf); if (reply) { @@ -1656,7 +2058,30 @@ 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; } + +/* 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); +}