X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-netdev.c;h=9a29b7eb3b76466a6d521808254f67e6fb5a1ab7;hb=cc36576070df622d0fc7a6e26ce01027e12b5b59;hp=1d05dd8d747bf21fb6a022df34ed4ebd1218848a;hpb=7257b535ab8e5fafd811c5f6788205eefdd44948;p=sliver-openvswitch.git diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 1d05dd8d7..9a29b7eb3 100644 --- a/lib/dpif-netdev.c +++ b/lib/dpif-netdev.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 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. @@ -17,7 +17,6 @@ #include #include "dpif.h" -#include #include #include #include @@ -41,7 +40,9 @@ #include "hmap.h" #include "list.h" #include "netdev.h" +#include "netdev-vport.h" #include "netlink.h" +#include "odp-execute.h" #include "odp-util.h" #include "ofp-print.h" #include "ofpbuf.h" @@ -49,6 +50,7 @@ #include "poll-loop.h" #include "random.h" #include "shash.h" +#include "sset.h" #include "timeval.h" #include "util.h" #include "vlog.h" @@ -69,8 +71,13 @@ enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */ enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 }; BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN)); +struct dp_netdev_upcall { + struct dpif_upcall upcall; /* Queued upcall information. */ + struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */ +}; + struct dp_netdev_queue { - struct dpif_upcall *upcalls[MAX_QUEUE_LEN]; + struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN]; unsigned int head, tail; }; @@ -80,6 +87,7 @@ struct dp_netdev { char *name; int open_cnt; bool destroyed; + int max_mtu; /* Maximum MTU of any port added so far. */ struct dp_netdev_queue queues[N_QUEUES]; struct hmap flow_table; /* Flow table. */ @@ -97,10 +105,12 @@ struct dp_netdev { /* A port in a netdev-based datapath. */ struct dp_netdev_port { - int port_no; /* Index into dp_netdev's 'ports'. */ + odp_port_t port_no; /* Index into dp_netdev's 'ports'. */ struct list node; /* Element in dp_netdev's 'port_list'. */ struct netdev *netdev; - bool internal; /* Internal port? */ + struct netdev_saved_flags *sf; + struct netdev_rx *rx; + char *type; /* Port type as requested by user. */ }; /* A flow in dp_netdev's 'flow_table'. */ @@ -112,7 +122,7 @@ struct dp_netdev_flow { long long int used; /* Last used time, in monotonic msecs. */ long long int packet_count; /* Number of packets matched. */ long long int byte_count; /* Number of bytes matched. */ - ovs_be16 tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */ + uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */ /* Actions. */ struct nlattr *actions; @@ -123,41 +133,39 @@ struct dp_netdev_flow { struct dpif_netdev { struct dpif dpif; struct dp_netdev *dp; - int listen_mask; unsigned int dp_serial; }; /* All netdev-based datapaths. */ static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs); -/* Maximum port MTU seen so far. */ -static int max_mtu = ETH_PAYLOAD_MAX; - -static int get_port_by_number(struct dp_netdev *, uint16_t port_no, +static int get_port_by_number(struct dp_netdev *, odp_port_t port_no, struct dp_netdev_port **portp); static int get_port_by_name(struct dp_netdev *, const char *devname, struct dp_netdev_port **portp); static void dp_netdev_free(struct dp_netdev *); static void dp_netdev_flow_flush(struct dp_netdev *); static int do_add_port(struct dp_netdev *, const char *devname, - const char *type, uint16_t port_no); -static int do_del_port(struct dp_netdev *, uint16_t port_no); + const char *type, odp_port_t port_no); +static int do_del_port(struct dp_netdev *, odp_port_t port_no); static int dpif_netdev_open(const struct dpif_class *, const char *name, bool create, struct dpif **); static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *, int queue_no, const struct flow *, - uint64_t arg); + const struct nlattr *userdata); static void dp_netdev_execute_actions(struct dp_netdev *, struct ofpbuf *, struct flow *, const struct nlattr *actions, size_t actions_len); - -static struct dpif_class dpif_dummy_class; +static void dp_netdev_port_input(struct dp_netdev *dp, + struct dp_netdev_port *port, + struct ofpbuf *packet, uint32_t skb_priority, + uint32_t skb_mark, const struct flow_tnl *tnl); static struct dpif_netdev * dpif_netdev_cast(const struct dpif *dpif) { - assert(dpif->dpif_class->open == dpif_netdev_open); + ovs_assert(dpif->dpif_class->open == dpif_netdev_open); return CONTAINER_OF(dpif, struct dpif_netdev, dpif); } @@ -167,6 +175,38 @@ get_dp_netdev(const struct dpif *dpif) return dpif_netdev_cast(dpif)->dp; } +static int +dpif_netdev_enumerate(struct sset *all_dps) +{ + struct shash_node *node; + + SHASH_FOR_EACH(node, &dp_netdevs) { + sset_add(all_dps, node->name); + } + return 0; +} + +static bool +dpif_netdev_class_is_dummy(const struct dpif_class *class) +{ + return class != &dpif_netdev_class; +} + +static bool +dpif_netdev_class_is_planetlab(const struct dpif_class *class) +{ + return class == &dpif_planetlab_class; +} + +static const char * +dpif_netdev_port_open_type(const struct dpif_class *class, const char *type) +{ + return strcmp(type, "internal") ? type + : dpif_netdev_class_is_planetlab(class) ? "pltap" + : dpif_netdev_class_is_dummy(class) ? "dummy" + : "tap"; +} + static struct dpif * create_dpif_netdev(struct dp_netdev *dp) { @@ -178,12 +218,53 @@ create_dpif_netdev(struct dp_netdev *dp) dpif = xmalloc(sizeof *dpif); dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id); dpif->dp = dp; - dpif->listen_mask = 0; dpif->dp_serial = dp->serial; return &dpif->dpif; } +/* Choose an unused, non-zero port number and return it on success. + * Return ODPP_NONE on failure. */ +static odp_port_t +choose_port(struct dp_netdev *dp, const char *name) +{ + uint32_t port_no; + + if (dp->class != &dpif_netdev_class && + dp->class != &dpif_planetlab_class) { + const char *p; + int start_no = 0; + + /* If the port name begins with "br", start the number search at + * 100 to make writing tests easier. */ + if (!strncmp(name, "br", 2)) { + start_no = 100; + } + + /* If the port name contains a number, try to assign that port number. + * This can make writing unit tests easier because port numbers are + * predictable. */ + for (p = name; *p != '\0'; p++) { + if (isdigit((unsigned char) *p)) { + port_no = start_no + strtol(p, NULL, 10); + if (port_no > 0 && port_no < MAX_PORTS + && !dp->ports[port_no]) { + return u32_to_odp(port_no); + } + break; + } + } + } + + for (port_no = 1; port_no < MAX_PORTS; port_no++) { + if (!dp->ports[port_no]) { + return u32_to_odp(port_no); + } + } + + return ODPP_NONE; +} + static int create_dp_netdev(const char *name, const struct dpif_class *class, struct dp_netdev **dpp) @@ -196,12 +277,14 @@ create_dp_netdev(const char *name, const struct dpif_class *class, dp->class = class; dp->name = xstrdup(name); dp->open_cnt = 0; + dp->max_mtu = ETH_PAYLOAD_MAX; for (i = 0; i < N_QUEUES; i++) { dp->queues[i].head = dp->queues[i].tail = 0; } hmap_init(&dp->flow_table); list_init(&dp->port_list); - error = do_add_port(dp, name, "internal", OVSP_LOCAL); + + error = do_add_port(dp, name, "internal", ODPP_LOCAL); if (error) { dp_netdev_free(dp); return error; @@ -228,7 +311,7 @@ dpif_netdev_open(const struct dpif_class *class, const char *name, if (error) { return error; } - assert(dp != NULL); + ovs_assert(dp != NULL); } } else { if (dp->class != class) { @@ -251,10 +334,8 @@ dp_netdev_purge_queues(struct dp_netdev *dp) struct dp_netdev_queue *q = &dp->queues[i]; while (q->tail != q->head) { - struct dpif_upcall *upcall = q->upcalls[q->tail++ & QUEUE_MASK]; - - ofpbuf_delete(upcall->packet); - free(upcall); + struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK]; + ofpbuf_uninit(&u->buf); } } } @@ -278,7 +359,7 @@ static void dpif_netdev_close(struct dpif *dpif) { struct dp_netdev *dp = get_dp_netdev(dpif); - assert(dp->open_cnt > 0); + ovs_assert(dp->open_cnt > 0); if (--dp->open_cnt == 0 && dp->destroyed) { shash_find_and_delete(&dp_netdevs, dp->name); dp_netdev_free(dp); @@ -307,48 +388,39 @@ dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) static int do_add_port(struct dp_netdev *dp, const char *devname, const char *type, - uint16_t port_no) + odp_port_t port_no) { + struct netdev_saved_flags *sf; struct dp_netdev_port *port; struct netdev *netdev; - bool internal; + struct netdev_rx *rx; + const char *open_type; int mtu; int error; /* XXX reject devices already in some dp_netdev. */ - if (type[0] == '\0' || !strcmp(type, "system") || !strcmp(type, "dummy")) { - internal = false; - } else if (!strcmp(type, "internal")) { - internal = true; - } else { - VLOG_WARN("%s: unsupported port type %s", devname, type); - return EINVAL; - } /* Open and validate network device. */ - if (dp->class == &dpif_dummy_class) { - type = "dummy"; - } else if (internal) { - type = "tap"; - } - - error = netdev_open(devname, type, &netdev); + open_type = dpif_netdev_port_open_type(dp->class, type); + error = netdev_open(devname, open_type, &netdev); if (error) { return error; } /* XXX reject loopback devices */ /* XXX reject non-Ethernet devices */ - error = netdev_listen(netdev); - if (error) { + error = netdev_rx_open(netdev, &rx); + if (error + && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) { VLOG_ERR("%s: cannot receive packets on this network device (%s)", - devname, strerror(errno)); + devname, ovs_strerror(errno)); netdev_close(netdev); return error; } - error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false); + error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf); if (error) { + netdev_rx_close(rx); netdev_close(netdev); return error; } @@ -356,15 +428,17 @@ do_add_port(struct dp_netdev *dp, const char *devname, const char *type, port = xmalloc(sizeof *port); port->port_no = port_no; port->netdev = netdev; - port->internal = internal; + port->sf = sf; + port->rx = rx; + port->type = xstrdup(type); error = netdev_get_mtu(netdev, &mtu); - if (!error) { - max_mtu = mtu; + if (!error && mtu > dp->max_mtu) { + dp->max_mtu = mtu; } list_push_back(&dp->port_list, &port->node); - dp->ports[port_no] = port; + dp->ports[odp_to_u32(port_no)] = port; dp->serial++; return 0; @@ -372,43 +446,55 @@ do_add_port(struct dp_netdev *dp, const char *devname, const char *type, static int dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev, - uint16_t *port_nop) + odp_port_t *port_nop) { struct dp_netdev *dp = get_dp_netdev(dpif); - int port_no; - - for (port_no = 0; port_no < MAX_PORTS; port_no++) { - if (!dp->ports[port_no]) { - *port_nop = port_no; - return do_add_port(dp, netdev_get_name(netdev), - netdev_get_type(netdev), port_no); + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; + const char *dpif_port; + odp_port_t port_no; + + dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); + if (*port_nop != ODPP_NONE) { + uint32_t port_idx = odp_to_u32(*port_nop); + if (port_idx >= MAX_PORTS) { + return EFBIG; + } else if (dp->ports[port_idx]) { + return EBUSY; } + port_no = *port_nop; + } else { + port_no = choose_port(dp, dpif_port); + } + if (port_no != ODPP_NONE) { + *port_nop = port_no; + return do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no); } return EFBIG; } static int -dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no) +dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no) { struct dp_netdev *dp = get_dp_netdev(dpif); - return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no); + return (port_no == ODPP_LOCAL ? + EINVAL : do_del_port(dp, port_no)); } static bool -is_valid_port_number(uint16_t port_no) +is_valid_port_number(odp_port_t port_no) { - return port_no < MAX_PORTS; + return odp_to_u32(port_no) < MAX_PORTS; } static int get_port_by_number(struct dp_netdev *dp, - uint16_t port_no, struct dp_netdev_port **portp) + odp_port_t port_no, struct dp_netdev_port **portp) { if (!is_valid_port_number(port_no)) { *portp = NULL; return EINVAL; } else { - *portp = dp->ports[port_no]; + *portp = dp->ports[odp_to_u32(port_no)]; return *portp ? 0 : ENOENT; } } @@ -429,10 +515,9 @@ get_port_by_name(struct dp_netdev *dp, } static int -do_del_port(struct dp_netdev *dp, uint16_t port_no) +do_del_port(struct dp_netdev *dp, odp_port_t port_no) { struct dp_netdev_port *port; - char *name; int error; error = get_port_by_number(dp, port_no, &port); @@ -441,13 +526,13 @@ do_del_port(struct dp_netdev *dp, uint16_t port_no) } list_remove(&port->node); - dp->ports[port->port_no] = NULL; + dp->ports[odp_to_u32(port_no)] = NULL; dp->serial++; - name = xstrdup(netdev_get_name(port->netdev)); netdev_close(port->netdev); - - free(name); + netdev_restore_flags(port->sf); + netdev_rx_close(port->rx); + free(port->type); free(port); return 0; @@ -458,12 +543,12 @@ answer_port_query(const struct dp_netdev_port *port, struct dpif_port *dpif_port) { dpif_port->name = xstrdup(netdev_get_name(port->netdev)); - dpif_port->type = xstrdup(port->internal ? "internal" : "system"); + dpif_port->type = xstrdup(port->type); dpif_port->port_no = port->port_no; } static int -dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no, +dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no, struct dpif_port *dpif_port) { struct dp_netdev *dp = get_dp_netdev(dpif); @@ -471,7 +556,7 @@ dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no, int error; error = get_port_by_number(dp, port_no, &port); - if (!error) { + if (!error && dpif_port) { answer_port_query(port, dpif_port); } return error; @@ -486,16 +571,16 @@ dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname, int error; error = get_port_by_name(dp, devname, &port); - if (!error) { + if (!error && dpif_port) { answer_port_query(port, dpif_port); } return error; } -static int +static odp_port_t dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED) { - return MAX_PORTS; + return u32_to_odp(MAX_PORTS); } static void @@ -525,7 +610,7 @@ dpif_netdev_flow_flush(struct dpif *dpif) } struct dp_netdev_port_state { - uint32_t port_no; + odp_port_t port_no; char *name; }; @@ -542,17 +627,18 @@ dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_, { struct dp_netdev_port_state *state = state_; struct dp_netdev *dp = get_dp_netdev(dpif); - uint32_t port_no; + uint32_t port_idx; - for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) { - struct dp_netdev_port *port = dp->ports[port_no]; + for (port_idx = odp_to_u32(state->port_no); + port_idx < MAX_PORTS; port_idx++) { + struct dp_netdev_port *port = dp->ports[port_idx]; if (port) { free(state->name); state->name = xstrdup(netdev_get_name(port->netdev)); dpif_port->name = state->name; - dpif_port->type = port->internal ? "internal" : "system"; + dpif_port->type = port->type; dpif_port->port_no = port->port_no; - state->port_no = port_no + 1; + state->port_no = u32_to_odp(port_idx + 1); return 0; } } @@ -608,14 +694,16 @@ get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats) stats->n_packets = flow->packet_count; stats->n_bytes = flow->byte_count; stats->used = flow->used; - stats->tcp_flags = TCP_FLAGS(flow->tcp_ctl); + stats->tcp_flags = flow->tcp_flags; } static int dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len, struct flow *flow) { - if (odp_flow_key_to_flow(key, key_len, flow)) { + odp_port_t in_port; + + if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) { /* This should not happen: it indicates that odp_flow_key_from_flow() * and odp_flow_key_to_flow() disagree on the acceptable form of a * flow. Log the problem as an error, with enough details to enable @@ -634,9 +722,8 @@ dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len, return EINVAL; } - if (flow->in_port < OFPP_MAX - ? flow->in_port >= MAX_PORTS - : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) { + in_port = flow->in_port.odp_port; + if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) { return EINVAL; } @@ -683,10 +770,9 @@ set_flow_actions(struct dp_netdev_flow *flow, } static int -add_flow(struct dpif *dpif, const struct flow *key, - const struct nlattr *actions, size_t actions_len) +dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key, + const struct nlattr *actions, size_t actions_len) { - struct dp_netdev *dp = get_dp_netdev(dpif); struct dp_netdev_flow *flow; int error; @@ -709,33 +795,31 @@ clear_stats(struct dp_netdev_flow *flow) flow->used = 0; flow->packet_count = 0; flow->byte_count = 0; - flow->tcp_ctl = 0; + flow->tcp_flags = 0; } static int -dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags, - const struct nlattr *nl_key, size_t nl_key_len, - const struct nlattr *actions, size_t actions_len, - struct dpif_flow_stats *stats) +dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put) { struct dp_netdev *dp = get_dp_netdev(dpif); struct dp_netdev_flow *flow; struct flow key; int error; - error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key); + error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key); if (error) { return error; } flow = dp_netdev_lookup_flow(dp, &key); if (!flow) { - if (flags & DPIF_FP_CREATE) { + if (put->flags & DPIF_FP_CREATE) { if (hmap_count(&dp->flow_table) < MAX_FLOWS) { - if (stats) { - memset(stats, 0, sizeof *stats); + if (put->stats) { + memset(put->stats, 0, sizeof *put->stats); } - return add_flow(dpif, &key, actions, actions_len); + return dp_netdev_flow_add(dp, &key, put->actions, + put->actions_len); } else { return EFBIG; } @@ -743,13 +827,13 @@ dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags, return ENOENT; } } else { - if (flags & DPIF_FP_MODIFY) { - int error = set_flow_actions(flow, actions, actions_len); + if (put->flags & DPIF_FP_MODIFY) { + int error = set_flow_actions(flow, put->actions, put->actions_len); if (!error) { - if (stats) { - get_dpif_flow_stats(flow, stats); + if (put->stats) { + get_dpif_flow_stats(flow, put->stats); } - if (flags & DPIF_FP_ZERO_STATS) { + if (put->flags & DPIF_FP_ZERO_STATS) { clear_stats(flow); } } @@ -761,24 +845,22 @@ dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags, } static int -dpif_netdev_flow_del(struct dpif *dpif, - const struct nlattr *nl_key, size_t nl_key_len, - struct dpif_flow_stats *stats) +dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del) { struct dp_netdev *dp = get_dp_netdev(dpif); struct dp_netdev_flow *flow; struct flow key; int error; - error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key); + error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key); if (error) { return error; } flow = dp_netdev_lookup_flow(dp, &key); if (flow) { - if (stats) { - get_dpif_flow_stats(flow, stats); + if (del->stats) { + get_dpif_flow_stats(flow, del->stats); } dp_netdev_free_flow(dp, flow); return 0; @@ -810,6 +892,7 @@ dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) static int dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_, const struct nlattr **key, size_t *key_len, + const struct nlattr **mask, size_t *mask_len, const struct nlattr **actions, size_t *actions_len, const struct dpif_flow_stats **stats) { @@ -829,12 +912,17 @@ dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_, struct ofpbuf buf; ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf); - odp_flow_key_from_flow(&buf, &flow->key); + odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port.odp_port); *key = buf.data; *key_len = buf.size; } + if (mask) { + *mask = NULL; + *mask_len = 0; + } + if (actions) { free(state->actions); state->actions = xmemdup(flow->actions, flow->actions_len); @@ -862,30 +950,29 @@ dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) } static int -dpif_netdev_execute(struct dpif *dpif, - const struct nlattr *key_attrs, size_t key_len, - const struct nlattr *actions, size_t actions_len, - const struct ofpbuf *packet) +dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute) { struct dp_netdev *dp = get_dp_netdev(dpif); struct ofpbuf copy; struct flow key; int error; - if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) { + if (execute->packet->size < ETH_HEADER_LEN || + execute->packet->size > UINT16_MAX) { return EINVAL; } /* Make a deep copy of 'packet', because we might modify its data. */ - ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size); + ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size); ofpbuf_reserve(©, DP_NETDEV_HEADROOM); - ofpbuf_put(©, packet->data, packet->size); + ofpbuf_put(©, execute->packet->data, execute->packet->size); - flow_extract(©, 0, -1, &key); - error = dpif_netdev_flow_from_nlattrs(key_attrs, key_len, &key); + flow_extract(©, 0, 0, NULL, NULL, &key); + error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len, + &key); if (!error) { dp_netdev_execute_actions(dp, ©, &key, - actions, actions_len); + execute->actions, execute->actions_len); } ofpbuf_uninit(©); @@ -893,32 +980,28 @@ dpif_netdev_execute(struct dpif *dpif, } static int -dpif_netdev_recv_get_mask(const struct dpif *dpif, int *listen_mask) +dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED) { - struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif); - *listen_mask = dpif_netdev->listen_mask; return 0; } static int -dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask) +dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED, + uint32_t queue_id, uint32_t *priority) { - struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif); - dpif_netdev->listen_mask = listen_mask; + *priority = queue_id; return 0; } static struct dp_netdev_queue * find_nonempty_queue(struct dpif *dpif) { - struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif); struct dp_netdev *dp = get_dp_netdev(dpif); - int mask = dpif_netdev->listen_mask; int i; for (i = 0; i < N_QUEUES; i++) { struct dp_netdev_queue *q = &dp->queues[i]; - if (q->head != q->tail && mask & (1u << i)) { + if (q->head != q->tail) { return q; } } @@ -926,13 +1009,18 @@ find_nonempty_queue(struct dpif *dpif) } static int -dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall) +dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall, + struct ofpbuf *buf) { struct dp_netdev_queue *q = find_nonempty_queue(dpif); if (q) { - struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK]; - *upcall = *u; - free(u); + struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK]; + + *upcall = u->upcall; + upcall->packet = buf; + + ofpbuf_uninit(buf); + *buf = u->buf; return 0; } else { @@ -959,38 +1047,37 @@ dpif_netdev_recv_purge(struct dpif *dpif) } static void -dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key, - const struct ofpbuf *packet) +dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet) { flow->used = time_msec(); flow->packet_count++; flow->byte_count += packet->size; - if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) { - struct tcp_header *th = packet->l4; - flow->tcp_ctl |= th->tcp_ctl; - } + flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key); } static void dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port, - struct ofpbuf *packet) + struct ofpbuf *packet, uint32_t skb_priority, + uint32_t skb_mark, const struct flow_tnl *tnl) { struct dp_netdev_flow *flow; struct flow key; + union flow_in_port in_port_; if (packet->size < ETH_HEADER_LEN) { return; } - flow_extract(packet, 0, port->port_no, &key); + in_port_.odp_port = port->port_no; + flow_extract(packet, skb_priority, skb_mark, tnl, &in_port_, &key); flow = dp_netdev_lookup_flow(dp, &key); if (flow) { - dp_netdev_flow_used(flow, &key, packet); + dp_netdev_flow_used(flow, packet); dp_netdev_execute_actions(dp, packet, &key, flow->actions, flow->actions_len); dp->n_hit++; } else { dp->n_missed++; - dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0); + dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL); } } @@ -1001,7 +1088,8 @@ dpif_netdev_run(struct dpif *dpif) struct dp_netdev_port *port; struct ofpbuf packet; - ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu); + ofpbuf_init(&packet, + DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu); LIST_FOR_EACH (port, node, &dp->port_list) { int error; @@ -1010,13 +1098,14 @@ dpif_netdev_run(struct dpif *dpif) ofpbuf_clear(&packet); ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM); - error = netdev_recv(port->netdev, &packet); + error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP; if (!error) { - dp_netdev_port_input(dp, port, &packet); + dp_netdev_port_input(dp, port, &packet, 0, 0, NULL); } else if (error != EAGAIN && error != EOPNOTSUPP) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_ERR_RL(&rl, "error receiving data from %s: %s", - netdev_get_name(port->netdev), strerror(error)); + netdev_get_name(port->netdev), ovs_strerror(error)); } } ofpbuf_uninit(&packet); @@ -1029,124 +1118,16 @@ dpif_netdev_wait(struct dpif *dpif) struct dp_netdev_port *port; LIST_FOR_EACH (port, node, &dp->port_list) { - netdev_recv_wait(port->netdev); - } -} - -static void -dp_netdev_pop_vlan(struct ofpbuf *packet) -{ - struct vlan_eth_header *veh = packet->l2; - if (packet->size >= sizeof *veh - && veh->veth_type == htons(ETH_TYPE_VLAN)) { - struct eth_header tmp; - - memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN); - memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN); - tmp.eth_type = veh->veth_next_type; - - ofpbuf_pull(packet, VLAN_HEADER_LEN); - packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN; - memcpy(packet->data, &tmp, sizeof tmp); - } -} - -static void -dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key) -{ - struct eth_header *eh = packet->l2; - - memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src); - memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst); -} - -static void -dp_netdev_set_ip_addr(struct ofpbuf *packet, ovs_be32 *addr, ovs_be32 new_addr) -{ - struct ip_header *nh = packet->l3; - - if (nh->ip_proto == IPPROTO_TCP && packet->l7) { - struct tcp_header *th = packet->l4; - th->tcp_csum = recalc_csum32(th->tcp_csum, *addr, new_addr); - } else if (nh->ip_proto == IPPROTO_UDP && packet->l7) { - struct udp_header *uh = packet->l4; - if (uh->udp_csum) { - uh->udp_csum = recalc_csum32(uh->udp_csum, *addr, new_addr); - if (!uh->udp_csum) { - uh->udp_csum = htons(0xffff); - } + if (port->rx) { + netdev_rx_wait(port->rx); } } - nh->ip_csum = recalc_csum32(nh->ip_csum, *addr, new_addr); - *addr = new_addr; } static void -dp_netdev_set_ip_tos(struct ip_header *nh, uint8_t new_tos) -{ - uint8_t *field = &nh->ip_tos; - - /* Set the DSCP bits and preserve the ECN bits. */ - uint8_t new = new_tos | (nh->ip_tos & IP_ECN_MASK); - - nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field), - htons((uint16_t) new)); - *field = new; -} - -static void -dp_netdev_set_ipv4(struct ofpbuf *packet, const struct ovs_key_ipv4 *ipv4_key) -{ - struct ip_header *nh = packet->l3; - - if (nh->ip_src != ipv4_key->ipv4_src) { - dp_netdev_set_ip_addr(packet, &nh->ip_src, ipv4_key->ipv4_src); - } - if (nh->ip_dst != ipv4_key->ipv4_dst) { - dp_netdev_set_ip_addr(packet, &nh->ip_dst, ipv4_key->ipv4_dst); - } - if (nh->ip_tos != ipv4_key->ipv4_tos) { - dp_netdev_set_ip_tos(nh, ipv4_key->ipv4_tos); - } -} - -static void -dp_netdev_set_port(ovs_be16 *port, ovs_be16 new_port, ovs_be16 *csum) -{ - *csum = recalc_csum16(*csum, *port, new_port); - *port = new_port; -} - -static void -dp_netdev_set_tcp_port(struct ofpbuf *packet, const struct ovs_key_tcp *tcp_key) -{ - struct tcp_header *th = packet->l4; - - if (th->tcp_src != tcp_key->tcp_src) { - dp_netdev_set_port(&th->tcp_src, tcp_key->tcp_src, &th->tcp_csum); - } - if (th->tcp_dst != tcp_key->tcp_dst) { - dp_netdev_set_port(&th->tcp_dst, tcp_key->tcp_dst, &th->tcp_csum); - } -} - -static void -dp_netdev_set_udp_port(struct ofpbuf *packet, const struct ovs_key_udp *udp_key) -{ - struct udp_header *uh = packet->l4; - - if (uh->udp_src != udp_key->udp_src) { - dp_netdev_set_port(&uh->udp_src, udp_key->udp_src, &uh->udp_csum); - } - if (uh->udp_dst != udp_key->udp_dst) { - dp_netdev_set_port(&uh->udp_dst, udp_key->udp_dst, &uh->udp_csum); - } -} - -static void -dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet, - uint16_t out_port) +dp_netdev_output_port(void *dp_, struct ofpbuf *packet, uint32_t out_port) { + struct dp_netdev *dp = dp_; struct dp_netdev_port *p = dp->ports[out_port]; if (p) { netdev_send(p->netdev, packet); @@ -1155,225 +1136,144 @@ dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet, static int dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet, - int queue_no, const struct flow *flow, uint64_t arg) + int queue_no, const struct flow *flow, + const struct nlattr *userdata) { struct dp_netdev_queue *q = &dp->queues[queue_no]; - struct dpif_upcall *upcall; - struct ofpbuf *buf; - size_t key_len; - - if (q->head - q->tail >= MAX_QUEUE_LEN) { - dp->n_lost++; - return ENOBUFS; - } - - buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size); - odp_flow_key_from_flow(buf, flow); - key_len = buf->size; - ofpbuf_pull(buf, key_len); - ofpbuf_reserve(buf, 2); - ofpbuf_put(buf, packet->data, packet->size); - - upcall = xzalloc(sizeof *upcall); - upcall->type = queue_no; - upcall->packet = buf; - upcall->key = buf->base; - upcall->key_len = key_len; - upcall->userdata = arg; - - q->upcalls[q->head++ & QUEUE_MASK] = upcall; - - return 0; -} - -static void -dp_netdev_sample(struct dp_netdev *dp, - struct ofpbuf *packet, struct flow *key, - const struct nlattr *action) -{ - const struct nlattr *subactions = NULL; - const struct nlattr *a; - size_t left; + if (q->head - q->tail < MAX_QUEUE_LEN) { + struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK]; + struct dpif_upcall *upcall = &u->upcall; + struct ofpbuf *buf = &u->buf; + size_t buf_size; + + upcall->type = queue_no; + + /* Allocate buffer big enough for everything. */ + buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size; + if (userdata) { + buf_size += NLA_ALIGN(userdata->nla_len); + } + ofpbuf_init(buf, buf_size); - NL_NESTED_FOR_EACH_UNSAFE (a, left, action) { - int type = nl_attr_type(a); + /* Put ODP flow. */ + odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port); + upcall->key = buf->data; + upcall->key_len = buf->size; - switch ((enum ovs_sample_attr) type) { - case OVS_SAMPLE_ATTR_PROBABILITY: - if (random_uint32() >= nl_attr_get_u32(a)) { - return; - } - break; + /* Put userdata. */ + if (userdata) { + upcall->userdata = ofpbuf_put(buf, userdata, + NLA_ALIGN(userdata->nla_len)); + } - case OVS_SAMPLE_ATTR_ACTIONS: - subactions = a; - break; + /* Put packet. + * + * We adjust 'data' and 'size' in 'buf' so that only the packet itself + * is visible in 'upcall->packet'. The ODP flow and (if present) + * userdata become part of the headroom. */ + ofpbuf_put_zeros(buf, 2); + buf->data = ofpbuf_put(buf, packet->data, packet->size); + buf->size = packet->size; + upcall->packet = buf; - case OVS_SAMPLE_ATTR_UNSPEC: - case __OVS_SAMPLE_ATTR_MAX: - default: - NOT_REACHED(); - } + return 0; + } else { + dp->n_lost++; + return ENOBUFS; } - - dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions), - nl_attr_get_size(subactions)); } static void -dp_netdev_action_userspace(struct dp_netdev *dp, - struct ofpbuf *packet, struct flow *key, - const struct nlattr *a) +dp_netdev_action_userspace(void *dp, struct ofpbuf *packet, + const struct flow *key, + const struct nlattr *userdata) { - const struct nlattr *userdata_attr; - uint64_t userdata; - - userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA); - userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0; dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata); } -static void -execute_set_action(struct ofpbuf *packet, const struct nlattr *a) -{ - enum ovs_key_attr type = nl_attr_type(a); - switch (type) { - case OVS_KEY_ATTR_TUN_ID: - break; - - case OVS_KEY_ATTR_ETHERNET: - dp_netdev_set_dl(packet, - nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet))); - break; - - case OVS_KEY_ATTR_IPV4: - dp_netdev_set_ipv4(packet, - nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4))); - break; - - case OVS_KEY_ATTR_TCP: - dp_netdev_set_tcp_port(packet, - nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp))); - break; - - case OVS_KEY_ATTR_UDP: - dp_netdev_set_udp_port(packet, - nl_attr_get_unspec(a, sizeof(struct ovs_key_udp))); - break; - - case OVS_KEY_ATTR_UNSPEC: - case OVS_KEY_ATTR_ETHERTYPE: - case OVS_KEY_ATTR_IPV6: - case OVS_KEY_ATTR_IN_PORT: - case OVS_KEY_ATTR_8021Q: - case OVS_KEY_ATTR_ICMP: - case OVS_KEY_ATTR_ICMPV6: - case OVS_KEY_ATTR_ARP: - case OVS_KEY_ATTR_ND: - case __OVS_KEY_ATTR_MAX: - default: - NOT_REACHED(); - } -} - static void dp_netdev_execute_actions(struct dp_netdev *dp, struct ofpbuf *packet, struct flow *key, const struct nlattr *actions, size_t actions_len) { - const struct nlattr *a; - unsigned int left; - - NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) { - const struct nlattr *nested; - const struct ovs_key_8021q *q_key; - int type = nl_attr_type(a); - - switch ((enum ovs_action_attr) type) { - case OVS_ACTION_ATTR_OUTPUT: - dp_netdev_output_port(dp, packet, nl_attr_get_u32(a)); - break; - - case OVS_ACTION_ATTR_USERSPACE: - dp_netdev_action_userspace(dp, packet, key, a); - break; - - case OVS_ACTION_ATTR_PUSH: - nested = nl_attr_get(a); - assert(nl_attr_type(nested) == OVS_KEY_ATTR_8021Q); - q_key = nl_attr_get_unspec(nested, sizeof(*q_key)); - eth_push_vlan(packet, q_key->q_tci); - break; - - case OVS_ACTION_ATTR_POP: - assert(nl_attr_get_u16(a) == OVS_KEY_ATTR_8021Q); - dp_netdev_pop_vlan(packet); - break; - - case OVS_ACTION_ATTR_SET: - execute_set_action(packet, nl_attr_get(a)); - break; - - case OVS_ACTION_ATTR_SAMPLE: - dp_netdev_sample(dp, packet, key, a); - break; - - case OVS_ACTION_ATTR_SET_PRIORITY: - case OVS_ACTION_ATTR_POP_PRIORITY: - /* not implemented */ - break; - - case OVS_ACTION_ATTR_UNSPEC: - case __OVS_ACTION_ATTR_MAX: - NOT_REACHED(); - } - } -} + odp_execute_actions(dp, packet, key, actions, actions_len, + dp_netdev_output_port, dp_netdev_action_userspace); +} + +#define DPIF_NETDEV_CLASS_FUNCTIONS \ + dpif_netdev_enumerate, \ + dpif_netdev_port_open_type, \ + dpif_netdev_open, \ + dpif_netdev_close, \ + dpif_netdev_destroy, \ + dpif_netdev_run, \ + dpif_netdev_wait, \ + dpif_netdev_get_stats, \ + dpif_netdev_port_add, \ + dpif_netdev_port_del, \ + dpif_netdev_port_query_by_number, \ + dpif_netdev_port_query_by_name, \ + dpif_netdev_get_max_ports, \ + NULL, /* port_get_pid */ \ + dpif_netdev_port_dump_start, \ + dpif_netdev_port_dump_next, \ + dpif_netdev_port_dump_done, \ + dpif_netdev_port_poll, \ + dpif_netdev_port_poll_wait, \ + dpif_netdev_flow_get, \ + dpif_netdev_flow_put, \ + dpif_netdev_flow_del, \ + dpif_netdev_flow_flush, \ + dpif_netdev_flow_dump_start, \ + dpif_netdev_flow_dump_next, \ + dpif_netdev_flow_dump_done, \ + dpif_netdev_execute, \ + NULL, /* operate */ \ + dpif_netdev_recv_set, \ + dpif_netdev_queue_to_priority, \ + dpif_netdev_recv, \ + dpif_netdev_recv_wait, \ + dpif_netdev_recv_purge, \ const struct dpif_class dpif_netdev_class = { "netdev", - NULL, /* enumerate */ - dpif_netdev_open, - dpif_netdev_close, - dpif_netdev_destroy, - dpif_netdev_run, - dpif_netdev_wait, - dpif_netdev_get_stats, - dpif_netdev_port_add, - dpif_netdev_port_del, - dpif_netdev_port_query_by_number, - dpif_netdev_port_query_by_name, - dpif_netdev_get_max_ports, - NULL, /* port_get_pid */ - dpif_netdev_port_dump_start, - dpif_netdev_port_dump_next, - dpif_netdev_port_dump_done, - dpif_netdev_port_poll, - dpif_netdev_port_poll_wait, - dpif_netdev_flow_get, - dpif_netdev_flow_put, - dpif_netdev_flow_del, - dpif_netdev_flow_flush, - dpif_netdev_flow_dump_start, - dpif_netdev_flow_dump_next, - dpif_netdev_flow_dump_done, - dpif_netdev_execute, - NULL, /* operate */ - dpif_netdev_recv_get_mask, - dpif_netdev_recv_set_mask, - NULL, /* queue_to_priority */ - dpif_netdev_recv, - dpif_netdev_recv_wait, - dpif_netdev_recv_purge, + DPIF_NETDEV_CLASS_FUNCTIONS +}; + +const struct dpif_class dpif_planetlab_class = { + "planetlab", + DPIF_NETDEV_CLASS_FUNCTIONS }; +static void +dpif_dummy_register__(const char *type) +{ + struct dpif_class *class; + + class = xmalloc(sizeof *class); + *class = dpif_netdev_class; + class->type = xstrdup(type); + dp_register_provider(class); +} + void -dpif_dummy_register(void) +dpif_dummy_register(bool override) { - if (!dpif_dummy_class.type) { - dpif_dummy_class = dpif_netdev_class; - dpif_dummy_class.type = "dummy"; - dp_register_provider(&dpif_dummy_class); + if (override) { + struct sset types; + const char *type; + + sset_init(&types); + dp_enumerate_types(&types); + SSET_FOR_EACH (type, &types) { + if (!dp_unregister_provider(type)) { + dpif_dummy_register__(type); + } + } + sset_destroy(&types); } + + dpif_dummy_register__("dummy"); } +