X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-netdev.c;h=d73050a3461480e68be2eb8df4c182a073c683ef;hb=2431be1b68d386bd616378d2c528242775c4d54a;hp=7b1bb5db22de9db1b521104804be2cf466265721;hpb=1e82e503c5358f8dce9eb2105448f0ec894d57bc;p=sliver-openvswitch.git diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 7b1bb5db2..36b6d4a05 100644 --- a/lib/dpif-netdev.c +++ b/lib/dpif-netdev.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010 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 @@ -25,6 +24,7 @@ #include #include #include +#include #include #include #include @@ -32,27 +32,34 @@ #include #include "csum.h" +#include "dpif.h" #include "dpif-provider.h" +#include "dummy.h" +#include "dynamic-string.h" #include "flow.h" #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" #include "packets.h" #include "poll-loop.h" -#include "queue.h" +#include "random.h" +#include "seq.h" +#include "shash.h" +#include "sset.h" #include "timeval.h" +#include "unixctl.h" #include "util.h" - #include "vlog.h" -#define THIS_MODULE VLM_dpif_netdev + +VLOG_DEFINE_THIS_MODULE(dpif_netdev); /* Configuration parameters. */ -enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */ -enum { MAX_QUEUE_LEN = 100 }; /* Maximum number of packets per queue. */ -enum { N_GROUPS = 16 }; /* Number of port groups. */ enum { MAX_PORTS = 256 }; /* Maximum number of ports. */ enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ @@ -60,91 +67,111 @@ enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ * headers to be aligned on a 4-byte boundary. */ enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN }; +/* Queues. */ +enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */ +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 dp_netdev_upcall upcalls[MAX_QUEUE_LEN]; + unsigned int head, tail; +}; + /* Datapath based on the network device interface from netdev.h. */ struct dp_netdev { - struct list node; - int dp_idx; + const struct dpif_class *class; + char *name; int open_cnt; bool destroyed; + int max_mtu; /* Maximum MTU of any port added so far. */ - bool drop_frags; /* Drop all IP fragments, if true. */ - struct ovs_queue queues[N_QUEUES]; /* Messages queued for dpif_recv(). */ + struct dp_netdev_queue queues[N_QUEUES]; struct hmap flow_table; /* Flow table. */ - struct odp_port_group groups[N_GROUPS]; + struct seq *queue_seq; /* Incremented whenever a packet is queued. */ /* Statistics. */ - long long int n_frags; /* Number of dropped IP fragments. */ long long int n_hit; /* Number of flow table matches. */ long long int n_missed; /* Number of flow table misses. */ long long int n_lost; /* Number of misses not passed to client. */ /* Ports. */ - int n_ports; struct dp_netdev_port *ports[MAX_PORTS]; struct list port_list; - unsigned int serial; + struct seq *port_seq; /* Incremented whenever a port changes. */ }; /* 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 (as ODP_PORT_INTERNAL)? */ + 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'. */ struct dp_netdev_flow { struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */ - flow_t key; + struct flow key; /* Statistics. */ - struct timeval used; /* Last used time, in milliseconds. */ - long long int packet_count; /* Number of packets matched. */ - long long int byte_count; /* Number of bytes matched. */ - uint8_t ip_tos; /* IP TOS value. */ - uint16_t tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */ + 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. */ + uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */ /* Actions. */ - union odp_action *actions; - unsigned int n_actions; + struct nlattr *actions; + size_t actions_len; }; /* Interface to netdev-based datapath. */ struct dpif_netdev { struct dpif dpif; struct dp_netdev *dp; - int listen_mask; - unsigned int dp_serial; + uint64_t last_port_seq; }; /* All netdev-based datapaths. */ -static struct dp_netdev *dp_netdevs[256]; -struct list dp_netdev_list = LIST_INITIALIZER(&dp_netdev_list); -enum { N_DP_NETDEVS = ARRAY_SIZE(dp_netdevs) }; +static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs); -/* Maximum port MTU seen so far. */ -static int max_mtu = ETH_PAYLOAD_MAX; +/* Global lock for all data. */ +static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER; -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, uint16_t flags, - uint16_t port_no); -static int do_del_port(struct dp_netdev *, uint16_t port_no); -static int dp_netdev_output_control(struct dp_netdev *, const struct ofpbuf *, - int queue_no, int port_no, uint32_t arg); -static int dp_netdev_execute_actions(struct dp_netdev *, - struct ofpbuf *, flow_t *, - const union odp_action *, int n); +static int do_add_port(struct dp_netdev *, const char *devname, + 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 *, + 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 void dp_netdev_port_input(struct dp_netdev *dp, + struct dp_netdev_port *port, + struct ofpbuf *packet, uint32_t skb_priority, + uint32_t pkt_mark, const struct flow_tnl *tnl); static struct dpif_netdev * dpif_netdev_cast(const struct dpif *dpif) { - dpif_assert_class(dpif, &dpif_netdev_class); + ovs_assert(dpif->dpif_class->open == dpif_netdev_open); return CONTAINER_OF(dpif, struct dpif_netdev, dpif); } @@ -155,127 +182,166 @@ get_dp_netdev(const struct dpif *dpif) } static int -name_to_dp_idx(const char *name) +dpif_netdev_enumerate(struct sset *all_dps) { - if (!strncmp(name, "dp", 2) && isdigit((unsigned char)name[2])) { - int dp_idx = atoi(name + 2); - if (dp_idx >= 0 && dp_idx < N_DP_NETDEVS) { - return dp_idx; - } + struct shash_node *node; + + ovs_mutex_lock(&dp_netdev_mutex); + SHASH_FOR_EACH(node, &dp_netdevs) { + sset_add(all_dps, node->name); } - return -1; + ovs_mutex_unlock(&dp_netdev_mutex); + + return 0; } -static struct dp_netdev * -find_dp_netdev(const char *name) +static bool +dpif_netdev_class_is_dummy(const struct dpif_class *class) { - int dp_idx; - size_t i; + return class != &dpif_netdev_class; +} - dp_idx = name_to_dp_idx(name); - if (dp_idx >= 0) { - return dp_netdevs[dp_idx]; - } +static bool +dpif_netdev_class_is_planetlab(const struct dpif_class *class) +{ + return class == &dpif_planetlab_class; +} - for (i = 0; i < N_DP_NETDEVS; i++) { - struct dp_netdev *dp = dp_netdevs[i]; - if (dp) { - struct dp_netdev_port *port; - if (!get_port_by_name(dp, name, &port)) { - return dp; - } - } - } - return NULL; +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) { + uint16_t netflow_id = hash_string(dp->name, 0); struct dpif_netdev *dpif; - char *dpname; dp->open_cnt++; - dpname = xasprintf("dp%d", dp->dp_idx); dpif = xmalloc(sizeof *dpif); - dpif_init(&dpif->dpif, &dpif_netdev_class, dpname, dp->dp_idx, dp->dp_idx); + 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; - free(dpname); + dpif->last_port_seq = seq_read(dp->port_seq); 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, int dp_idx, struct dpif **dpifp) +create_dp_netdev(const char *name, const struct dpif_class *class, + struct dp_netdev **dpp) { struct dp_netdev *dp; int error; int i; - if (dp_netdevs[dp_idx]) { - return EBUSY; - } - - /* Create datapath. */ - dp_netdevs[dp_idx] = dp = xzalloc(sizeof *dp); - list_push_back(&dp_netdev_list, &dp->node); - dp->dp_idx = dp_idx; + dp = xzalloc(sizeof *dp); + dp->class = class; + dp->name = xstrdup(name); dp->open_cnt = 0; - dp->drop_frags = false; + dp->max_mtu = ETH_PAYLOAD_MAX; for (i = 0; i < N_QUEUES; i++) { - queue_init(&dp->queues[i]); + dp->queues[i].head = dp->queues[i].tail = 0; } + dp->queue_seq = seq_create(); hmap_init(&dp->flow_table); - for (i = 0; i < N_GROUPS; i++) { - dp->groups[i].ports = NULL; - dp->groups[i].n_ports = 0; - dp->groups[i].group = i; - } list_init(&dp->port_list); - error = do_add_port(dp, name, ODP_PORT_INTERNAL, ODPP_LOCAL); + dp->port_seq = seq_create(); + + error = do_add_port(dp, name, "internal", ODPP_LOCAL); if (error) { dp_netdev_free(dp); - return ENODEV; + return error; } - *dpifp = create_dpif_netdev(dp); + shash_add(&dp_netdevs, name, dp); + + *dpp = dp; return 0; } static int -dpif_netdev_open(const char *name, const char *type OVS_UNUSED, bool create, - struct dpif **dpifp) +dpif_netdev_open(const struct dpif_class *class, const char *name, + bool create, struct dpif **dpifp) { - if (create) { - if (find_dp_netdev(name)) { - return EEXIST; - } else { - int dp_idx = name_to_dp_idx(name); - if (dp_idx >= 0) { - return create_dp_netdev(name, dp_idx, dpifp); - } else { - /* Scan for unused dp_idx number. */ - for (dp_idx = 0; dp_idx < N_DP_NETDEVS; dp_idx++) { - int error = create_dp_netdev(name, dp_idx, dpifp); - if (error != EBUSY) { - return error; - } - } + struct dp_netdev *dp; + int error; - /* All datapath numbers in use. */ - return ENOBUFS; - } - } + ovs_mutex_lock(&dp_netdev_mutex); + dp = shash_find_data(&dp_netdevs, name); + if (!dp) { + error = create ? create_dp_netdev(name, class, &dp) : ENODEV; } else { - struct dp_netdev *dp = find_dp_netdev(name); - if (dp) { - *dpifp = create_dpif_netdev(dp); - return 0; - } else { - return ENODEV; + error = (dp->class != class ? EINVAL + : create ? EEXIST + : 0); + } + if (!error) { + *dpifp = create_dpif_netdev(dp); + } + ovs_mutex_unlock(&dp_netdev_mutex); + + return error; +} + +static void +dp_netdev_purge_queues(struct dp_netdev *dp) +{ + int i; + + for (i = 0; i < N_QUEUES; i++) { + struct dp_netdev_queue *q = &dp->queues[i]; + + while (q->tail != q->head) { + struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK]; + ofpbuf_uninit(&u->buf); } } } @@ -283,23 +349,17 @@ dpif_netdev_open(const char *name, const char *type OVS_UNUSED, bool create, static void dp_netdev_free(struct dp_netdev *dp) { - int i; + struct dp_netdev_port *port, *next; dp_netdev_flow_flush(dp); - while (dp->n_ports > 0) { - struct dp_netdev_port *port = CONTAINER_OF( - dp->port_list.next, struct dp_netdev_port, node); + LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) { do_del_port(dp, port->port_no); } - for (i = 0; i < N_QUEUES; i++) { - queue_destroy(&dp->queues[i]); - } + dp_netdev_purge_queues(dp); + seq_destroy(dp->queue_seq); hmap_destroy(&dp->flow_table); - for (i = 0; i < N_GROUPS; i++) { - free(dp->groups[i].ports); - } - dp_netdevs[dp->dp_idx] = NULL; - list_remove(&dp->node); + seq_destroy(dp->port_seq); + free(dp->name); free(dp); } @@ -307,87 +367,90 @@ static void dpif_netdev_close(struct dpif *dpif) { struct dp_netdev *dp = get_dp_netdev(dpif); - assert(dp->open_cnt > 0); + + ovs_mutex_lock(&dp_netdev_mutex); + + 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); } free(dpif); + + ovs_mutex_unlock(&dp_netdev_mutex); } static int dpif_netdev_destroy(struct dpif *dpif) { struct dp_netdev *dp = get_dp_netdev(dpif); + + ovs_mutex_lock(&dp_netdev_mutex); dp->destroyed = true; + ovs_mutex_unlock(&dp_netdev_mutex); + return 0; } static int -dpif_netdev_get_stats(const struct dpif *dpif, struct odp_stats *stats) +dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) { struct dp_netdev *dp = get_dp_netdev(dpif); - memset(stats, 0, sizeof *stats); + + ovs_mutex_lock(&dp_netdev_mutex); stats->n_flows = hmap_count(&dp->flow_table); - stats->cur_capacity = hmap_capacity(&dp->flow_table); - stats->max_capacity = MAX_FLOWS; - stats->n_ports = dp->n_ports; - stats->max_ports = MAX_PORTS; - stats->max_groups = N_GROUPS; - stats->n_frags = dp->n_frags; stats->n_hit = dp->n_hit; stats->n_missed = dp->n_missed; stats->n_lost = dp->n_lost; - stats->max_miss_queue = MAX_QUEUE_LEN; - stats->max_action_queue = MAX_QUEUE_LEN; - return 0; -} - -static int -dpif_netdev_get_drop_frags(const struct dpif *dpif, bool *drop_fragsp) -{ - struct dp_netdev *dp = get_dp_netdev(dpif); - *drop_fragsp = dp->drop_frags; - return 0; -} + stats->n_masks = UINT64_MAX; + stats->n_mask_hit = UINT64_MAX; + ovs_mutex_unlock(&dp_netdev_mutex); -static int -dpif_netdev_set_drop_frags(struct dpif *dpif, bool drop_frags) -{ - struct dp_netdev *dp = get_dp_netdev(dpif); - dp->drop_frags = drop_frags; return 0; } static int -do_add_port(struct dp_netdev *dp, const char *devname, uint16_t flags, - uint16_t port_no) +do_add_port(struct dp_netdev *dp, const char *devname, const char *type, + odp_port_t port_no) { - bool internal = (flags & ODP_PORT_INTERNAL) != 0; + struct netdev_saved_flags *sf; struct dp_netdev_port *port; - struct netdev_options netdev_options; struct netdev *netdev; + struct netdev_rx *rx; + enum netdev_flags flags; + const char *open_type; int mtu; int error; /* XXX reject devices already in some dp_netdev. */ /* Open and validate network device. */ - memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = devname; - netdev_options.ethertype = NETDEV_ETH_TYPE_ANY; - if (internal) { - netdev_options.type = "tap"; - } - - error = netdev_open(&netdev_options, &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_turn_flags_on(netdev, NETDEV_PROMISC, false); + netdev_get_flags(netdev, &flags); + if (flags & NETDEV_LOOPBACK) { + VLOG_ERR("%s: cannot add a loopback device", devname); + netdev_close(netdev); + return EINVAL; + } + + 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, ovs_strerror(errno)); + netdev_close(netdev); + return error; + } + + error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf); if (error) { + netdev_rx_close(rx); netdev_close(netdev); return error; } @@ -395,59 +458,85 @@ do_add_port(struct dp_netdev *dp, const char *devname, uint16_t flags, 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); - netdev_get_mtu(netdev, &mtu); - if (mtu > max_mtu) { - max_mtu = mtu; + error = netdev_get_mtu(netdev, &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->n_ports++; - dp->serial++; + dp->ports[odp_to_u32(port_no)] = port; + seq_change(dp->port_seq); return 0; } static int -dpif_netdev_port_add(struct dpif *dpif, const char *devname, uint16_t flags, - uint16_t *port_nop) +dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev, + odp_port_t *port_nop) { struct dp_netdev *dp = get_dp_netdev(dpif); - int port_no; + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; + const char *dpif_port; + odp_port_t port_no; + int error; - for (port_no = 0; port_no < MAX_PORTS; port_no++) { - if (!dp->ports[port_no]) { - *port_nop = port_no; - return do_add_port(dp, devname, flags, port_no); + ovs_mutex_lock(&dp_netdev_mutex); + 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) { + error = EFBIG; + } else if (dp->ports[port_idx]) { + error = EBUSY; + } else { + error = 0; + port_no = *port_nop; } + } else { + port_no = choose_port(dp, dpif_port); + error = port_no == ODPP_NONE ? EFBIG : 0; } - return EFBIG; + if (!error) { + *port_nop = port_no; + error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no); + } + ovs_mutex_unlock(&dp_netdev_mutex); + + return error; } 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 == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no); + int error; + + ovs_mutex_lock(&dp_netdev_mutex); + error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no); + ovs_mutex_unlock(&dp_netdev_mutex); + + return error; } 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; } } @@ -458,7 +547,7 @@ get_port_by_name(struct dp_netdev *dp, { struct dp_netdev_port *port; - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { + LIST_FOR_EACH (port, node, &dp->port_list) { if (!strcmp(netdev_get_name(port->netdev), devname)) { *portp = port; return 0; @@ -468,10 +557,9 @@ 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); @@ -480,59 +568,69 @@ do_del_port(struct dp_netdev *dp, uint16_t port_no) } list_remove(&port->node); - dp->ports[port->port_no] = NULL; - dp->n_ports--; - dp->serial++; + dp->ports[odp_to_u32(port_no)] = NULL; + seq_change(dp->port_seq); - 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; } static void -answer_port_query(const struct dp_netdev_port *port, struct odp_port *odp_port) +answer_port_query(const struct dp_netdev_port *port, + struct dpif_port *dpif_port) { - memset(odp_port, 0, sizeof *odp_port); - ovs_strlcpy(odp_port->devname, netdev_get_name(port->netdev), - sizeof odp_port->devname); - odp_port->port = port->port_no; - odp_port->flags = port->internal ? ODP_PORT_INTERNAL : 0; + dpif_port->name = xstrdup(netdev_get_name(port->netdev)); + 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, - struct odp_port *odp_port) +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); struct dp_netdev_port *port; int error; + ovs_mutex_lock(&dp_netdev_mutex); error = get_port_by_number(dp, port_no, &port); - if (!error) { - answer_port_query(port, odp_port); + if (!error && dpif_port) { + answer_port_query(port, dpif_port); } + ovs_mutex_unlock(&dp_netdev_mutex); + return error; } static int dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname, - struct odp_port *odp_port) + struct dpif_port *dpif_port) { struct dp_netdev *dp = get_dp_netdev(dpif); struct dp_netdev_port *port; int error; + ovs_mutex_lock(&dp_netdev_mutex); error = get_port_by_name(dp, devname, &port); - if (!error) { - answer_port_query(port, odp_port); + if (!error && dpif_port) { + answer_port_query(port, dpif_port); } + ovs_mutex_unlock(&dp_netdev_mutex); + return error; } +static uint32_t +dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED) +{ + return MAX_PORTS; +} + static void dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow) { @@ -546,8 +644,7 @@ dp_netdev_flow_flush(struct dp_netdev *dp) { struct dp_netdev_flow *flow, *next; - HMAP_FOR_EACH_SAFE (flow, next, struct dp_netdev_flow, node, - &dp->flow_table) { + HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) { dp_netdev_free_flow(dp, flow); } } @@ -556,114 +653,100 @@ static int dpif_netdev_flow_flush(struct dpif *dpif) { struct dp_netdev *dp = get_dp_netdev(dpif); + + ovs_mutex_lock(&dp_netdev_mutex); dp_netdev_flow_flush(dp); + ovs_mutex_unlock(&dp_netdev_mutex); + return 0; } +struct dp_netdev_port_state { + odp_port_t port_no; + char *name; +}; + static int -dpif_netdev_port_list(const struct dpif *dpif, struct odp_port *ports, int n) +dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) { - struct dp_netdev *dp = get_dp_netdev(dpif); - struct dp_netdev_port *port; - int i; - - i = 0; - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { - struct odp_port *odp_port = &ports[i]; - if (i >= n) { - break; - } - answer_port_query(port, odp_port); - i++; - } - return dp->n_ports; + *statep = xzalloc(sizeof(struct dp_netdev_port_state)); + return 0; } static int -dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED) +dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_, + struct dpif_port *dpif_port) { - struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); - if (dpif->dp_serial != dpif->dp->serial) { - dpif->dp_serial = dpif->dp->serial; - return ENOBUFS; - } else { - return EAGAIN; - } -} + struct dp_netdev_port_state *state = state_; + struct dp_netdev *dp = get_dp_netdev(dpif); + uint32_t port_idx; + + ovs_mutex_lock(&dp_netdev_mutex); + 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->type; + dpif_port->port_no = port->port_no; + state->port_no = u32_to_odp(port_idx + 1); + ovs_mutex_unlock(&dp_netdev_mutex); -static void -dpif_netdev_port_poll_wait(const struct dpif *dpif_) -{ - struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); - if (dpif->dp_serial != dpif->dp->serial) { - poll_immediate_wake(); + return 0; + } } + ovs_mutex_unlock(&dp_netdev_mutex); + + return EOF; } static int -get_port_group(const struct dpif *dpif, int group_no, - struct odp_port_group **groupp) +dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) { - struct dp_netdev *dp = get_dp_netdev(dpif); - - if (group_no >= 0 && group_no < N_GROUPS) { - *groupp = &dp->groups[group_no]; - return 0; - } else { - *groupp = NULL; - return EINVAL; - } + struct dp_netdev_port_state *state = state_; + free(state->name); + free(state); + return 0; } static int -dpif_netdev_port_group_get(const struct dpif *dpif, int group_no, - uint16_t ports[], int n) +dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED) { - struct odp_port_group *group; + struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); + uint64_t new_port_seq; int error; - if (n < 0) { - return -EINVAL; - } - - error = get_port_group(dpif, group_no, &group); - if (!error) { - memcpy(ports, group->ports, MIN(n, group->n_ports) * sizeof *ports); - return group->n_ports; + ovs_mutex_lock(&dp_netdev_mutex); + new_port_seq = seq_read(dpif->dp->port_seq); + if (dpif->last_port_seq != new_port_seq) { + dpif->last_port_seq = new_port_seq; + error = ENOBUFS; } else { - return -error; + error = EAGAIN; } + ovs_mutex_unlock(&dp_netdev_mutex); + + return error; } -static int -dpif_netdev_port_group_set(struct dpif *dpif, int group_no, - const uint16_t ports[], int n) +static void +dpif_netdev_port_poll_wait(const struct dpif *dpif_) { - struct odp_port_group *group; - int error; - - if (n < 0 || n > MAX_PORTS) { - return EINVAL; - } + struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); - error = get_port_group(dpif, group_no, &group); - if (!error) { - free(group->ports); - group->ports = xmemdup(ports, n * sizeof *group->ports); - group->n_ports = n; - group->group = group_no; - } - return error; + ovs_mutex_lock(&dp_netdev_mutex); + seq_wait(dpif->dp->port_seq, dpif->last_port_seq); + ovs_mutex_unlock(&dp_netdev_mutex); } static struct dp_netdev_flow * -dp_netdev_lookup_flow(const struct dp_netdev *dp, const flow_t *key) +dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key) { struct dp_netdev_flow *flow; - assert(!key->reserved[0] && !key->reserved[1] && !key->reserved[2]); - HMAP_FOR_EACH_WITH_HASH (flow, struct dp_netdev_flow, node, - flow_hash(key, 0), &dp->flow_table) { + HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) { if (flow_equal(&flow->key, key)) { return flow; } @@ -672,147 +755,100 @@ dp_netdev_lookup_flow(const struct dp_netdev *dp, const flow_t *key) } static void -answer_flow_query(struct dp_netdev_flow *flow, uint32_t query_flags, - struct odp_flow *odp_flow) +get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats) { - if (flow) { - odp_flow->key = flow->key; - odp_flow->stats.n_packets = flow->packet_count; - odp_flow->stats.n_bytes = flow->byte_count; - odp_flow->stats.used_sec = flow->used.tv_sec; - odp_flow->stats.used_nsec = flow->used.tv_usec * 1000; - odp_flow->stats.tcp_flags = TCP_FLAGS(flow->tcp_ctl); - odp_flow->stats.ip_tos = flow->ip_tos; - odp_flow->stats.error = 0; - if (odp_flow->n_actions > 0) { - unsigned int n = MIN(odp_flow->n_actions, flow->n_actions); - memcpy(odp_flow->actions, flow->actions, - n * sizeof *odp_flow->actions); - odp_flow->n_actions = flow->n_actions; - } - - if (query_flags & ODPFF_ZERO_TCP_FLAGS) { - flow->tcp_ctl = 0; - } - - } else { - odp_flow->stats.error = ENOENT; - } + stats->n_packets = flow->packet_count; + stats->n_bytes = flow->byte_count; + stats->used = flow->used; + stats->tcp_flags = flow->tcp_flags; } static int -dpif_netdev_flow_get(const struct dpif *dpif, struct odp_flow flows[], int n) +dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len, + struct flow *flow) { - struct dp_netdev *dp = get_dp_netdev(dpif); - int i; + 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 + * debugging. */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + + if (!VLOG_DROP_ERR(&rl)) { + struct ds s; + + ds_init(&s); + odp_flow_key_format(key, key_len, &s); + VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s)); + ds_destroy(&s); + } - for (i = 0; i < n; i++) { - struct odp_flow *odp_flow = &flows[i]; - answer_flow_query(dp_netdev_lookup_flow(dp, &odp_flow->key), - odp_flow->flags, odp_flow); + return EINVAL; } - return 0; -} - -static int -dpif_netdev_validate_actions(const union odp_action *actions, int n_actions, - bool *mutates) -{ - unsigned int i; - - *mutates = false; - for (i = 0; i < n_actions; i++) { - const union odp_action *a = &actions[i]; - switch (a->type) { - case ODPAT_OUTPUT: - if (a->output.port >= MAX_PORTS) { - return EINVAL; - } - break; - - case ODPAT_OUTPUT_GROUP: - *mutates = true; - if (a->output_group.group >= N_GROUPS) { - return EINVAL; - } - break; - case ODPAT_CONTROLLER: - break; - - case ODPAT_SET_VLAN_VID: - *mutates = true; - if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) { - return EINVAL; - } - break; - - case ODPAT_SET_VLAN_PCP: - *mutates = true; - if (a->vlan_pcp.vlan_pcp & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) { - return EINVAL; - } - break; - - case ODPAT_SET_NW_TOS: - *mutates = true; - if (a->nw_tos.nw_tos & IP_ECN_MASK) { - return EINVAL; - } - break; - - case ODPAT_STRIP_VLAN: - case ODPAT_SET_DL_SRC: - case ODPAT_SET_DL_DST: - case ODPAT_SET_NW_SRC: - case ODPAT_SET_NW_DST: - case ODPAT_SET_TP_SRC: - case ODPAT_SET_TP_DST: - *mutates = true; - break; - - default: - return EOPNOTSUPP; - } + in_port = flow->in_port.odp_port; + if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) { + return EINVAL; } + return 0; } static int -set_flow_actions(struct dp_netdev_flow *flow, struct odp_flow *odp_flow) +dpif_netdev_flow_get(const struct dpif *dpif, + const struct nlattr *nl_key, size_t nl_key_len, + struct ofpbuf **actionsp, struct dpif_flow_stats *stats) { - size_t n_bytes; - bool mutates; + struct dp_netdev *dp = get_dp_netdev(dpif); + struct dp_netdev_flow *flow; + struct flow key; int error; - if (odp_flow->n_actions >= 4096 / sizeof *odp_flow->actions) { - return EINVAL; - } - error = dpif_netdev_validate_actions(odp_flow->actions, - odp_flow->n_actions, &mutates); + error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key); if (error) { return error; } - n_bytes = odp_flow->n_actions * sizeof *flow->actions; - flow->actions = xrealloc(flow->actions, n_bytes); - flow->n_actions = odp_flow->n_actions; - memcpy(flow->actions, odp_flow->actions, n_bytes); + ovs_mutex_lock(&dp_netdev_mutex); + flow = dp_netdev_lookup_flow(dp, &key); + if (flow) { + if (stats) { + get_dpif_flow_stats(flow, stats); + } + if (actionsp) { + *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len); + } + } else { + error = ENOENT; + } + ovs_mutex_unlock(&dp_netdev_mutex); + + return error; +} + +static int +set_flow_actions(struct dp_netdev_flow *flow, + const struct nlattr *actions, size_t actions_len) +{ + flow->actions = xrealloc(flow->actions, actions_len); + flow->actions_len = actions_len; + memcpy(flow->actions, actions, actions_len); return 0; } static int -add_flow(struct dpif *dpif, struct odp_flow *odp_flow) +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; flow = xzalloc(sizeof *flow); - flow->key = odp_flow->key; - memset(flow->key.reserved, 0, sizeof flow->key.reserved); + flow->key = *key; - error = set_flow_actions(flow, odp_flow); + error = set_flow_actions(flow, actions, actions_len); if (error) { free(flow); return error; @@ -825,150 +861,227 @@ add_flow(struct dpif *dpif, struct odp_flow *odp_flow) static void clear_stats(struct dp_netdev_flow *flow) { - flow->used.tv_sec = 0; - flow->used.tv_usec = 0; + flow->used = 0; flow->packet_count = 0; flow->byte_count = 0; - flow->ip_tos = 0; - flow->tcp_ctl = 0; + flow->tcp_flags = 0; } static int -dpif_netdev_flow_put(struct dpif *dpif, struct odp_flow_put *put) +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(put->key, put->key_len, &key); + if (error) { + return error; + } - flow = dp_netdev_lookup_flow(dp, &put->flow.key); + ovs_mutex_lock(&dp_netdev_mutex); + flow = dp_netdev_lookup_flow(dp, &key); if (!flow) { - if (put->flags & ODPPF_CREATE) { + if (put->flags & DPIF_FP_CREATE) { if (hmap_count(&dp->flow_table) < MAX_FLOWS) { - return add_flow(dpif, &put->flow); + if (put->stats) { + memset(put->stats, 0, sizeof *put->stats); + } + error = dp_netdev_flow_add(dp, &key, put->actions, + put->actions_len); } else { - return EFBIG; + error = EFBIG; } } else { - return ENOENT; + error = ENOENT; } } else { - if (put->flags & ODPPF_MODIFY) { - int error = set_flow_actions(flow, &put->flow); - if (!error && put->flags & ODPPF_ZERO_STATS) { - clear_stats(flow); + if (put->flags & DPIF_FP_MODIFY) { + error = set_flow_actions(flow, put->actions, put->actions_len); + if (!error) { + if (put->stats) { + get_dpif_flow_stats(flow, put->stats); + } + if (put->flags & DPIF_FP_ZERO_STATS) { + clear_stats(flow); + } } - return error; } else { - return EEXIST; + error = EEXIST; } } -} + ovs_mutex_unlock(&dp_netdev_mutex); + return error; +} static int -dpif_netdev_flow_del(struct dpif *dpif, struct odp_flow *odp_flow) +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(del->key, del->key_len, &key); + if (error) { + return error; + } - flow = dp_netdev_lookup_flow(dp, &odp_flow->key); + ovs_mutex_lock(&dp_netdev_mutex); + flow = dp_netdev_lookup_flow(dp, &key); if (flow) { - answer_flow_query(flow, 0, odp_flow); + if (del->stats) { + get_dpif_flow_stats(flow, del->stats); + } dp_netdev_free_flow(dp, flow); - return 0; } else { - return ENOENT; + error = ENOENT; } + ovs_mutex_unlock(&dp_netdev_mutex); + + return error; } +struct dp_netdev_flow_state { + uint32_t bucket; + uint32_t offset; + struct nlattr *actions; + struct odputil_keybuf keybuf; + struct dpif_flow_stats stats; +}; + static int -dpif_netdev_flow_list(const struct dpif *dpif, struct odp_flow flows[], int n) +dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) { + struct dp_netdev_flow_state *state; + + *statep = state = xmalloc(sizeof *state); + state->bucket = 0; + state->offset = 0; + state->actions = NULL; + return 0; +} + +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) +{ + struct dp_netdev_flow_state *state = state_; struct dp_netdev *dp = get_dp_netdev(dpif); struct dp_netdev_flow *flow; - int i; + struct hmap_node *node; - i = 0; - HMAP_FOR_EACH (flow, struct dp_netdev_flow, node, &dp->flow_table) { - if (i >= n) { - break; - } - answer_flow_query(flow, 0, &flows[i++]); + ovs_mutex_lock(&dp_netdev_mutex); + node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset); + if (!node) { + ovs_mutex_unlock(&dp_netdev_mutex); + return EOF; } - return hmap_count(&dp->flow_table); + + flow = CONTAINER_OF(node, struct dp_netdev_flow, node); + + if (key) { + struct ofpbuf buf; + + ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf); + 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); + + *actions = state->actions; + *actions_len = flow->actions_len; + } + + if (stats) { + get_dpif_flow_stats(flow, &state->stats); + *stats = &state->stats; + } + + ovs_mutex_unlock(&dp_netdev_mutex); + return 0; } static int -dpif_netdev_execute(struct dpif *dpif, uint16_t in_port, - const union odp_action actions[], int n_actions, - const struct ofpbuf *packet) +dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) +{ + struct dp_netdev_flow_state *state = state_; + + free(state->actions); + free(state); + return 0; +} + +static int +dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute) { struct dp_netdev *dp = get_dp_netdev(dpif); struct ofpbuf copy; - bool mutates; - flow_t flow; + 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; } - error = dpif_netdev_validate_actions(actions, n_actions, &mutates); - if (error) { - return error; - } + /* Make a deep copy of 'packet', because we might modify its data. */ + ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size); + ofpbuf_reserve(©, DP_NETDEV_HEADROOM); + ofpbuf_put(©, execute->packet->data, execute->packet->size); - if (mutates) { - /* We need a deep copy of 'packet' since we're going to modify its - * data. */ - ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size); - copy.data = (char*)copy.base + DP_NETDEV_HEADROOM; - ofpbuf_put(©, packet->data, packet->size); - } else { - /* We still need a shallow copy of 'packet', even though we won't - * modify its data, because flow_extract() modifies packet->l2, etc. - * We could probably get away with modifying those but it's more polite - * if we don't. */ - copy = *packet; - } - flow_extract(©, 0, in_port, &flow); - error = dp_netdev_execute_actions(dp, ©, &flow, actions, n_actions); - if (mutates) { - ofpbuf_uninit(©); + flow_extract(©, 0, 0, NULL, NULL, &key); + error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len, + &key); + if (!error) { + ovs_mutex_lock(&dp_netdev_mutex); + dp_netdev_execute_actions(dp, ©, &key, + execute->actions, execute->actions_len); + ovs_mutex_unlock(&dp_netdev_mutex); } + + ofpbuf_uninit(©); return error; } 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); - if (!(listen_mask & ~ODPL_ALL)) { - dpif_netdev->listen_mask = listen_mask; - return 0; - } else { - return EINVAL; - } + *priority = queue_id; + return 0; } -static struct ovs_queue * +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 ovs_queue *q = &dp->queues[i]; - if (q->n && mask & (1u << i)) { + struct dp_netdev_queue *q = &dp->queues[i]; + if (q->head != q->tail) { return q; } } @@ -976,417 +1089,347 @@ find_nonempty_queue(struct dpif *dpif) } static int -dpif_netdev_recv(struct dpif *dpif, struct ofpbuf **bufp) +dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall, + struct ofpbuf *buf) { - struct ovs_queue *q = find_nonempty_queue(dpif); + struct dp_netdev_queue *q; + int error; + + ovs_mutex_lock(&dp_netdev_mutex); + q = find_nonempty_queue(dpif); if (q) { - *bufp = queue_pop_head(q); - return 0; + struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK]; + + *upcall = u->upcall; + upcall->packet = buf; + + ofpbuf_uninit(buf); + *buf = u->buf; + + error = 0; } else { - return EAGAIN; + error = EAGAIN; } + ovs_mutex_unlock(&dp_netdev_mutex); + + return error; } static void dpif_netdev_recv_wait(struct dpif *dpif) { - struct ovs_queue *q = find_nonempty_queue(dpif); - if (q) { + struct dp_netdev *dp = get_dp_netdev(dpif); + uint64_t seq; + + ovs_mutex_lock(&dp_netdev_mutex); + seq = seq_read(dp->queue_seq); + if (find_nonempty_queue(dpif)) { poll_immediate_wake(); } else { - /* No messages ready to be received, and dp_wait() will ensure that we - * wake up to queue new messages, so there is nothing to do. */ + seq_wait(dp->queue_seq, seq); } + ovs_mutex_unlock(&dp_netdev_mutex); +} + +static void +dpif_netdev_recv_purge(struct dpif *dpif) +{ + struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif); + ovs_mutex_lock(&dp_netdev_mutex); + dp_netdev_purge_queues(dpif_netdev->dp); + ovs_mutex_unlock(&dp_netdev_mutex); } static void -dp_netdev_flow_used(struct dp_netdev_flow *flow, const flow_t *key, - const struct ofpbuf *packet) +dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet) { - time_timeval(&flow->used); + flow->used = time_msec(); flow->packet_count++; flow->byte_count += packet->size; - if (key->dl_type == htons(ETH_TYPE_IP)) { - struct ip_header *nh = packet->l3; - flow->ip_tos = nh->ip_tos; - - if (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 pkt_mark, const struct flow_tnl *tnl) { struct dp_netdev_flow *flow; - flow_t key; + struct flow key; + union flow_in_port in_port_; - if (flow_extract(packet, 0, port->port_no, &key) && dp->drop_frags) { - dp->n_frags++; + if (packet->size < ETH_HEADER_LEN) { return; } - + in_port_.odp_port = port->port_no; + flow_extract(packet, skb_priority, pkt_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->n_actions); + flow->actions, flow->actions_len); dp->n_hit++; } else { dp->n_missed++; - dp_netdev_output_control(dp, packet, _ODPL_MISS_NR, port->port_no, 0); + dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL); } } static void -dp_netdev_run(void) +dpif_netdev_run(struct dpif *dpif) { - struct ofpbuf packet; + struct dp_netdev_port *port; struct dp_netdev *dp; + struct ofpbuf packet; - ofpbuf_init(&packet, DP_NETDEV_HEADROOM + max_mtu); - LIST_FOR_EACH (dp, struct dp_netdev, node, &dp_netdev_list) { - struct dp_netdev_port *port; + ovs_mutex_lock(&dp_netdev_mutex); + dp = get_dp_netdev(dpif); + ofpbuf_init(&packet, + DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu); - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { - int error; + LIST_FOR_EACH (port, node, &dp->port_list) { + int error; - /* Reset packet contents. */ - packet.data = (char*)packet.base + DP_NETDEV_HEADROOM; - packet.size = 0; + /* Reset packet contents. */ + ofpbuf_clear(&packet); + ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM); - error = netdev_recv(port->netdev, &packet); - if (!error) { - dp_netdev_port_input(dp, port, &packet); - } else if (error != EAGAIN) { - 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)); - } + error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP; + if (!error) { + 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), ovs_strerror(error)); } } ofpbuf_uninit(&packet); + ovs_mutex_unlock(&dp_netdev_mutex); } static void -dp_netdev_wait(void) +dpif_netdev_wait(struct dpif *dpif) { - struct dp_netdev *dp; + struct dp_netdev_port *port; - LIST_FOR_EACH (dp, struct dp_netdev, node, &dp_netdev_list) { - struct dp_netdev_port *port; - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { - netdev_recv_wait(port->netdev); + /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and + * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before + * A makes it to poll_block(). + * + * But I think it doesn't matter: + * + * - In the dpif_port_add() case, A will not wake up when a packet + * arrives on the new port, but this would also happen if the + * ordering were reversed. + * + * - In the dpif_port_remove() case, A might wake up spuriously, but + * that is harmless. */ + + ovs_mutex_lock(&dp_netdev_mutex); + LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) { + if (port->rx) { + netdev_rx_wait(port->rx); } } + ovs_mutex_unlock(&dp_netdev_mutex); } - -/* Modify the TCI field of 'packet'. If a VLAN tag is not present, one - * is added with the TCI field set to 'tci'. If a VLAN tag is present, - * then 'mask' bits are cleared before 'tci' is logically OR'd into the - * TCI field. - * - * Note that the function does not ensure that 'tci' does not affect - * bits outside of 'mask'. - */ static void -dp_netdev_modify_vlan_tci(struct ofpbuf *packet, flow_t *key, - uint16_t tci, uint16_t mask) +dp_netdev_output_port(void *dp_, struct ofpbuf *packet, + const struct flow *flow OVS_UNUSED, odp_port_t out_port) { - struct vlan_eth_header *veh; - - if (key->dl_vlan != htons(ODP_VLAN_NONE)) { - /* Clear 'mask' bits, but maintain other TCI bits. */ - veh = packet->l2; - veh->veth_tci &= ~htons(mask); - veh->veth_tci |= htons(tci); - } else { - /* Insert new 802.1Q header. */ - struct eth_header *eh = packet->l2; - struct vlan_eth_header tmp; - memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN); - memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN); - tmp.veth_type = htons(ETH_TYPE_VLAN); - tmp.veth_tci = htons(tci); - tmp.veth_next_type = eh->eth_type; - - veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN); - memcpy(veh, &tmp, sizeof tmp); - packet->l2 = (char*)packet->l2 - VLAN_HEADER_LEN; + struct dp_netdev *dp = dp_; + struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)]; + if (p) { + netdev_send(p->netdev, packet); } - - key->dl_vlan = veh->veth_tci & htons(VLAN_VID_MASK); } -static void -dp_netdev_strip_vlan(struct ofpbuf *packet, flow_t *key) +static int +dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet, + int queue_no, const struct flow *flow, + const struct nlattr *userdata) { - struct vlan_eth_header *veh = packet->l2; - if (veh->veth_type == htons(ETH_TYPE_VLAN)) { - struct eth_header tmp; + struct dp_netdev_queue *q = &dp->queues[queue_no]; + 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); + + /* Put ODP flow. */ + odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port); + upcall->key = buf->data; + upcall->key_len = buf->size; + + /* Put userdata. */ + if (userdata) { + upcall->userdata = ofpbuf_put(buf, userdata, + NLA_ALIGN(userdata->nla_len)); + } - 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; + /* 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; - packet->size -= VLAN_HEADER_LEN; - packet->data = (char*)packet->data + VLAN_HEADER_LEN; - packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN; - memcpy(packet->data, &tmp, sizeof tmp); + seq_change(dp->queue_seq); - key->dl_vlan = htons(ODP_VLAN_NONE); + return 0; + } else { + dp->n_lost++; + return ENOBUFS; } } static void -dp_netdev_set_dl_src(struct ofpbuf *packet, flow_t *key, - const uint8_t dl_addr[ETH_ADDR_LEN]) +dp_netdev_action_userspace(void *dp, struct ofpbuf *packet, + const struct flow *key, + const struct nlattr *a) { - struct eth_header *eh = packet->l2; - memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src); - memcpy(key->dl_src, dl_addr, sizeof key->dl_src); + const struct nlattr *userdata; + + userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA); + dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata); } static void -dp_netdev_set_dl_dst(struct ofpbuf *packet, flow_t *key, - const uint8_t dl_addr[ETH_ADDR_LEN]) +dp_netdev_execute_actions(struct dp_netdev *dp, + struct ofpbuf *packet, struct flow *key, + const struct nlattr *actions, + size_t actions_len) { - struct eth_header *eh = packet->l2; - memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst); - memcpy(key->dl_dst, dl_addr, sizeof key->dl_dst); + odp_execute_actions(dp, packet, key, actions, actions_len, + dp_netdev_output_port, dp_netdev_action_userspace); } -static void -dp_netdev_set_nw_addr(struct ofpbuf *packet, flow_t *key, - const struct odp_action_nw_addr *a) -{ - if (key->dl_type == htons(ETH_TYPE_IP)) { - struct ip_header *nh = packet->l3; - uint32_t *field; - - field = a->type == ODPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst; - if (key->nw_proto == IP_TYPE_TCP) { - struct tcp_header *th = packet->l4; - th->tcp_csum = recalc_csum32(th->tcp_csum, *field, a->nw_addr); - } else if (key->nw_proto == IP_TYPE_UDP) { - struct udp_header *uh = packet->l4; - if (uh->udp_csum) { - uh->udp_csum = recalc_csum32(uh->udp_csum, *field, a->nw_addr); - if (!uh->udp_csum) { - uh->udp_csum = 0xffff; - } - } - } - nh->ip_csum = recalc_csum32(nh->ip_csum, *field, a->nw_addr); - *field = a->nw_addr; +#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, \ - if (a->type == ODPAT_SET_NW_SRC) { - key->nw_src = a->type; - } else { - key->nw_dst = a->type; - } - } -} +const struct dpif_class dpif_netdev_class = { + "netdev", + DPIF_NETDEV_CLASS_FUNCTIONS +}; + +const struct dpif_class dpif_planetlab_class = { + "planetlab", + DPIF_NETDEV_CLASS_FUNCTIONS +}; static void -dp_netdev_set_nw_tos(struct ofpbuf *packet, flow_t *key, - const struct odp_action_nw_tos *a) +dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[], void *aux OVS_UNUSED) { - if (key->dl_type == htons(ETH_TYPE_IP)) { - struct ip_header *nh = packet->l3; - uint8_t *field = &nh->ip_tos; - - /* Set the DSCP bits and preserve the ECN bits. */ - uint8_t new = a->nw_tos | (nh->ip_tos & IP_ECN_MASK); + struct dp_netdev_port *port; + struct dp_netdev *dp; + int port_no; - nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field), - htons((uint16_t)a->nw_tos)); - *field = new; - key->nw_tos = a->nw_tos; + dp = shash_find_data(&dp_netdevs, argv[1]); + if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { + unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); + return; } -} - -static void -dp_netdev_set_tp_port(struct ofpbuf *packet, flow_t *key, - const struct odp_action_tp_port *a) -{ - if (key->dl_type == htons(ETH_TYPE_IP)) { - uint16_t *field; - if (key->nw_proto == IPPROTO_TCP) { - struct tcp_header *th = packet->l4; - field = a->type == ODPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst; - th->tcp_csum = recalc_csum16(th->tcp_csum, *field, a->tp_port); - *field = a->tp_port; - } else if (key->nw_proto == IPPROTO_UDP) { - struct udp_header *uh = packet->l4; - field = a->type == ODPAT_SET_TP_SRC ? &uh->udp_src : &uh->udp_dst; - uh->udp_csum = recalc_csum16(uh->udp_csum, *field, a->tp_port); - *field = a->tp_port; - } else { - return; - } - if (a->type == ODPAT_SET_TP_SRC) { - key->tp_src = a->tp_port; - } else { - key->tp_dst = a->tp_port; - } + if (get_port_by_name(dp, argv[2], &port)) { + unixctl_command_reply_error(conn, "unknown port"); + return; } -} -static void -dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet, - uint16_t out_port) -{ - struct dp_netdev_port *p = dp->ports[out_port]; - if (p) { - netdev_send(p->netdev, packet); + port_no = atoi(argv[3]); + if (port_no <= 0 || port_no >= MAX_PORTS) { + unixctl_command_reply_error(conn, "bad port number"); + return; } + if (dp->ports[port_no]) { + unixctl_command_reply_error(conn, "port number already in use"); + return; + } + dp->ports[odp_to_u32(port->port_no)] = NULL; + dp->ports[port_no] = port; + port->port_no = u32_to_odp(port_no); + seq_change(dp->port_seq); + unixctl_command_reply(conn, NULL); } static void -dp_netdev_output_group(struct dp_netdev *dp, uint16_t group, uint16_t in_port, - struct ofpbuf *packet) +dpif_dummy_register__(const char *type) { - struct odp_port_group *g = &dp->groups[group]; - int i; + struct dpif_class *class; - for (i = 0; i < g->n_ports; i++) { - uint16_t out_port = g->ports[i]; - if (out_port != in_port) { - dp_netdev_output_port(dp, packet, out_port); - } - } + class = xmalloc(sizeof *class); + *class = dpif_netdev_class; + class->type = xstrdup(type); + dp_register_provider(class); } -static int -dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet, - int queue_no, int port_no, uint32_t arg) +void +dpif_dummy_register(bool override) { - struct ovs_queue *q = &dp->queues[queue_no]; - struct odp_msg *header; - struct ofpbuf *msg; - size_t msg_size; - - if (q->n >= MAX_QUEUE_LEN) { - dp->n_lost++; - return ENOBUFS; + 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); } - msg_size = sizeof *header + packet->size; - msg = ofpbuf_new(msg_size + DPIF_RECV_MSG_PADDING); - ofpbuf_reserve(msg, DPIF_RECV_MSG_PADDING); - header = ofpbuf_put_uninit(msg, sizeof *header); - header->type = queue_no; - header->length = msg_size; - header->port = port_no; - header->arg = arg; - ofpbuf_put(msg, packet->data, packet->size); - queue_push_tail(q, msg); - - return 0; -} + dpif_dummy_register__("dummy"); -static int -dp_netdev_execute_actions(struct dp_netdev *dp, - struct ofpbuf *packet, flow_t *key, - const union odp_action *actions, int n_actions) -{ - int i; - for (i = 0; i < n_actions; i++) { - const union odp_action *a = &actions[i]; - - switch (a->type) { - case ODPAT_OUTPUT: - dp_netdev_output_port(dp, packet, a->output.port); - break; - - case ODPAT_OUTPUT_GROUP: - dp_netdev_output_group(dp, a->output_group.group, key->in_port, - packet); - break; - - case ODPAT_CONTROLLER: - dp_netdev_output_control(dp, packet, _ODPL_ACTION_NR, - key->in_port, a->controller.arg); - break; - - case ODPAT_SET_VLAN_VID: - dp_netdev_modify_vlan_tci(packet, key, ntohs(a->vlan_vid.vlan_vid), - VLAN_VID_MASK); - break; - - case ODPAT_SET_VLAN_PCP: - dp_netdev_modify_vlan_tci( - packet, key, a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT, - VLAN_PCP_MASK); - break; - - case ODPAT_STRIP_VLAN: - dp_netdev_strip_vlan(packet, key); - break; - - case ODPAT_SET_DL_SRC: - dp_netdev_set_dl_src(packet, key, a->dl_addr.dl_addr); - break; - - case ODPAT_SET_DL_DST: - dp_netdev_set_dl_dst(packet, key, a->dl_addr.dl_addr); - break; - - case ODPAT_SET_NW_SRC: - case ODPAT_SET_NW_DST: - dp_netdev_set_nw_addr(packet, key, &a->nw_addr); - break; - - case ODPAT_SET_NW_TOS: - dp_netdev_set_nw_tos(packet, key, &a->nw_tos); - break; - - case ODPAT_SET_TP_SRC: - case ODPAT_SET_TP_DST: - dp_netdev_set_tp_port(packet, key, &a->tp_port); - break; - } - } - return 0; + unixctl_command_register("dpif-dummy/change-port-number", + "DP PORT NEW-NUMBER", + 3, 3, dpif_dummy_change_port_number, NULL); } -const struct dpif_class dpif_netdev_class = { - "netdev", - dp_netdev_run, - dp_netdev_wait, - NULL, /* enumerate */ - dpif_netdev_open, - dpif_netdev_close, - NULL, /* get_all_names */ - dpif_netdev_destroy, - dpif_netdev_get_stats, - dpif_netdev_get_drop_frags, - dpif_netdev_set_drop_frags, - dpif_netdev_port_add, - dpif_netdev_port_del, - dpif_netdev_port_query_by_number, - dpif_netdev_port_query_by_name, - dpif_netdev_port_list, - dpif_netdev_port_poll, - dpif_netdev_port_poll_wait, - dpif_netdev_port_group_get, - dpif_netdev_port_group_set, - dpif_netdev_flow_get, - dpif_netdev_flow_put, - dpif_netdev_flow_del, - dpif_netdev_flow_flush, - dpif_netdev_flow_list, - dpif_netdev_execute, - dpif_netdev_recv_get_mask, - dpif_netdev_recv_set_mask, - NULL, /* get_sflow_probability */ - NULL, /* set_sflow_probability */ - dpif_netdev_recv, - dpif_netdev_recv_wait, -};