X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=fe10952dcb726ff35b8863c5225a93dddf5f6e46;hb=37a1300c3ca05a97160d44dc34298c447b50ed9d;hp=e5ea31a5e8099a910750a70d891bc01077ed9d6c;hpb=1ba530f4b2cd5476a224dbbf87a3089a831a24b6;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index e5ea31a5e..fe10952dc 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -25,19 +25,19 @@ #include #include #include -#include #include #include #include #include -#include #include #include #include "dpif-provider.h" #include "netdev.h" #include "netdev-vport.h" +#include "netlink-socket.h" #include "netlink.h" +#include "odp-util.h" #include "ofpbuf.h" #include "openvswitch/tunnel.h" #include "packets.h" @@ -46,15 +46,79 @@ #include "rtnetlink-link.h" #include "shash.h" #include "svec.h" +#include "unaligned.h" #include "util.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(dpif_linux); +struct dpif_linux_dp { + /* Generic Netlink header. */ + uint8_t cmd; + + /* struct odp_header. */ + uint32_t dp_idx; + + /* Attributes. */ + const char *name; /* ODP_DP_ATTR_NAME. */ + struct odp_stats stats; /* ODP_DP_ATTR_STATS. */ + enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */ + const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */ + uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */ +}; + +static void dpif_linux_dp_init(struct dpif_linux_dp *); +static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *, + const struct ofpbuf *); +static void dpif_linux_dp_dump_start(struct nl_dump *); +static int dpif_linux_dp_transact(const struct dpif_linux_dp *request, + struct dpif_linux_dp *reply, + struct ofpbuf **bufp); +static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply, + struct ofpbuf **bufp); + +struct dpif_linux_flow { + /* Generic Netlink header. */ + uint8_t cmd; + + /* struct odp_header. */ + unsigned int nlmsg_flags; + uint32_t dp_idx; + + /* Attributes. + * + * The 'stats' and 'used' members point to 64-bit data that might only be + * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to + * access their values. */ + const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */ + size_t key_len; + const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */ + size_t actions_len; + const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */ + const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */ + const uint64_t *used; /* ODP_FLOW_ATTR_USED. */ + bool clear; /* ODP_FLOW_ATTR_CLEAR. */ +}; + +static void dpif_linux_flow_init(struct dpif_linux_flow *); +static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *, + const struct ofpbuf *); +static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *, + struct ofpbuf *); +static int dpif_linux_flow_transact(const struct dpif_linux_flow *request, + struct dpif_linux_flow *reply, + struct ofpbuf **bufp); +static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *, + struct dpif_flow_stats *); + /* Datapath interface for the openvswitch Linux kernel module. */ struct dpif_linux { struct dpif dpif; - int fd; + + /* Multicast group messages. */ + struct nl_sock *mc_sock; + uint32_t mcgroups[DPIF_N_UC_TYPES]; + unsigned int listen_mask; /* Used by dpif_linux_get_all_names(). */ char *local_ifname; @@ -69,16 +133,27 @@ struct dpif_linux { static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); -static int do_ioctl(const struct dpif *, int cmd, const void *arg); -static int open_dpif(const struct dpif_linux_vport *local_vport, +/* Generic Netlink family numbers for ODP. */ +static int odp_datapath_family; +static int odp_vport_family; +static int odp_flow_family; +static int odp_packet_family; + +/* Generic Netlink socket. */ +static struct nl_sock *genl_sock; + +static int dpif_linux_init(void); +static int open_dpif(const struct dpif_linux_dp *, + const struct dpif_linux_vport *local_vport, struct dpif **); -static int get_openvswitch_major(void); -static int create_minor(const char *name, int minor); -static int open_minor(int minor, int *fdp); -static int make_openvswitch_device(int minor, char **fnp); static void dpif_linux_port_changed(const struct rtnetlink_link_change *, void *dpif); +static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, + struct ofpbuf *); +static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *, + const struct ofpbuf *); + static struct dpif_linux * dpif_linux_cast(const struct dpif *dpif) { @@ -89,108 +164,86 @@ dpif_linux_cast(const struct dpif *dpif) static int dpif_linux_enumerate(struct svec *all_dps) { - int major; + struct nl_dump dump; + struct ofpbuf msg; int error; - int i; - /* Check that the Open vSwitch module is loaded. */ - major = get_openvswitch_major(); - if (major < 0) { - return -major; + error = dpif_linux_init(); + if (error) { + return error; } - error = 0; - for (i = 0; i < ODP_MAX; i++) { - struct dpif *dpif; - char devname[16]; - int retval; - - sprintf(devname, "dp%d", i); - retval = dpif_open(devname, "system", &dpif); - if (!retval) { - svec_add(all_dps, devname); - dpif_uninit(dpif, true); - } else if (retval != ENODEV && !error) { - error = retval; + dpif_linux_dp_dump_start(&dump); + while (nl_dump_next(&dump, &msg)) { + struct dpif_linux_dp dp; + + if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) { + svec_add(all_dps, dp.name); } } - return error; + return nl_dump_done(&dump); } static int dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, bool create, struct dpif **dpifp) { - struct dpif_linux_vport request, reply; + struct dpif_linux_vport vport_request, vport; + struct dpif_linux_dp dp_request, dp; struct ofpbuf *buf; int minor; int error; + error = dpif_linux_init(); + if (error) { + return error; + } + minor = !strncmp(name, "dp", 2) && isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1; - if (create) { - if (minor >= 0) { - error = create_minor(name, minor); - if (error) { - return error; - } - } else { - /* Scan for unused minor number. */ - for (minor = 0; ; minor++) { - if (minor >= ODP_MAX) { - /* All datapath numbers in use. */ - return ENOBUFS; - } - - error = create_minor(name, minor); - if (!error) { - break; - } else if (error != EBUSY) { - return error; - } - } - } - } - dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_GET; - request.port_no = ODPP_LOCAL; - if (minor >= 0) { - request.dp_idx = minor; - } else { - request.name = name; + /* Create or look up datapath. */ + dpif_linux_dp_init(&dp_request); + dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET; + dp_request.dp_idx = minor; + dp_request.name = minor < 0 ? name : NULL; + error = dpif_linux_dp_transact(&dp_request, &dp, &buf); + if (error) { + return error; } - - error = dpif_linux_vport_transact(&request, &reply, &buf); + ofpbuf_delete(buf); /* Pointers inside 'dp' are now invalid! */ + + /* Look up local port. */ + dpif_linux_vport_init(&vport_request); + vport_request.cmd = ODP_VPORT_CMD_GET; + vport_request.dp_idx = dp.dp_idx; + vport_request.port_no = ODPP_LOCAL; + vport_request.name = minor < 0 ? name : NULL; + error = dpif_linux_vport_transact(&vport_request, &vport, &buf); if (error) { return error; - } else if (reply.port_no != ODPP_LOCAL) { + } else if (vport.port_no != ODPP_LOCAL) { /* This is an Open vSwitch device but not the local port. We * intentionally support only using the name of the local port as the * name of a datapath; otherwise, it would be too difficult to * enumerate all the names of a datapath. */ error = EOPNOTSUPP; } else { - error = open_dpif(&reply, dpifp); + error = open_dpif(&dp, &vport, dpifp); } - ofpbuf_delete(buf); return error; } static int -open_dpif(const struct dpif_linux_vport *local_vport, struct dpif **dpifp) +open_dpif(const struct dpif_linux_dp *dp, + const struct dpif_linux_vport *local_vport, struct dpif **dpifp) { int dp_idx = local_vport->dp_idx; struct dpif_linux *dpif; char *name; int error; - int fd; - - error = open_minor(dp_idx, &fd); - if (error) { - goto error; - } + int i; dpif = xmalloc(sizeof *dpif); error = rtnetlink_link_notifier_register(&dpif->port_notifier, @@ -203,7 +256,11 @@ open_dpif(const struct dpif_linux_vport *local_vport, struct dpif **dpifp) dpif_init(&dpif->dpif, &dpif_linux_class, name, dp_idx, dp_idx); free(name); - dpif->fd = fd; + dpif->mc_sock = NULL; + for (i = 0; i < DPIF_N_UC_TYPES; i++) { + dpif->mcgroups[i] = dp->mcgroups[i]; + } + dpif->listen_mask = 0; dpif->local_ifname = xstrdup(local_vport->name); dpif->local_ifindex = local_vport->ifindex; dpif->minor = dp_idx; @@ -215,8 +272,6 @@ open_dpif(const struct dpif_linux_vport *local_vport, struct dpif **dpifp) error_free: free(dpif); - close(fd); -error: return error; } @@ -227,7 +282,6 @@ dpif_linux_close(struct dpif *dpif_) rtnetlink_link_notifier_unregister(&dpif->port_notifier); shash_destroy(&dpif->changed_ports); free(dpif->local_ifname); - close(dpif->fd); free(dpif); } @@ -244,25 +298,41 @@ dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names) static int dpif_linux_destroy(struct dpif *dpif_) { - return do_ioctl(dpif_, ODP_DP_DESTROY, NULL); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp dp; + + dpif_linux_dp_init(&dp); + dp.cmd = ODP_DP_CMD_DEL; + dp.dp_idx = dpif->minor; + return dpif_linux_dp_transact(&dp, NULL, NULL); } static int dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats) { - memset(stats, 0, sizeof *stats); - return do_ioctl(dpif_, ODP_DP_STATS, stats); + struct dpif_linux_dp dp; + struct ofpbuf *buf; + int error; + + error = dpif_linux_dp_get(dpif_, &dp, &buf); + if (!error) { + *stats = dp.stats; + ofpbuf_delete(buf); + } + return error; } static int dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) { - int drop_frags; + struct dpif_linux_dp dp; + struct ofpbuf *buf; int error; - error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags); + error = dpif_linux_dp_get(dpif_, &dp, &buf); if (!error) { - *drop_fragsp = drop_frags & 1; + *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP; + ofpbuf_delete(buf); } return error; } @@ -270,8 +340,14 @@ dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) static int dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) { - int drop_frags_int = drop_frags; - return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp dp; + + dpif_linux_dp_init(&dp); + dp.cmd = ODP_DP_CMD_SET; + dp.dp_idx = dpif->minor; + dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO; + return dpif_linux_dp_transact(&dp, NULL, NULL); } static int @@ -287,7 +363,7 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, int error; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_NEW; + request.cmd = ODP_VPORT_CMD_NEW; request.dp_idx = dpif->minor; request.type = netdev_vport_get_vport_type(netdev); if (request.type == ODP_VPORT_TYPE_UNSPEC) { @@ -320,7 +396,7 @@ dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) struct dpif_linux_vport vport; dpif_linux_vport_init(&vport); - vport.cmd = ODP_VPORT_DEL; + vport.cmd = ODP_VPORT_CMD_DEL; vport.dp_idx = dpif->minor; vport.port_no = port_no; return dpif_linux_vport_transact(&vport, NULL, NULL); @@ -336,7 +412,7 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, int error; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_GET; + request.cmd = ODP_VPORT_CMD_GET; request.dp_idx = dpif_linux_cast(dpif)->minor; request.port_no = port_no; request.name = port_name; @@ -365,60 +441,83 @@ dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname, return dpif_linux_port_query__(dpif, 0, devname, dpif_port); } +static int +dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED) +{ + /* If the datapath increases its range of supported ports, then it should + * start reporting that. */ + return 1024; +} + static int dpif_linux_flow_flush(struct dpif *dpif_) { - return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_flow flow; + + dpif_linux_flow_init(&flow); + flow.cmd = ODP_FLOW_CMD_DEL; + flow.dp_idx = dpif->minor; + return dpif_linux_flow_transact(&flow, NULL, NULL); } struct dpif_linux_port_state { - struct ofpbuf *buf; - uint32_t next; + struct nl_dump dump; }; static int -dpif_linux_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) +dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep) { - *statep = xzalloc(sizeof(struct dpif_linux_port_state)); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_port_state *state; + struct dpif_linux_vport request; + struct ofpbuf *buf; + + *statep = state = xmalloc(sizeof *state); + + dpif_linux_vport_init(&request); + request.cmd = ODP_DP_CMD_GET; + request.dp_idx = dpif->minor; + + buf = ofpbuf_new(1024); + dpif_linux_vport_to_ofpbuf(&request, buf); + nl_dump_start(&state->dump, genl_sock, buf); + ofpbuf_delete(buf); + return 0; } static int -dpif_linux_port_dump_next(const struct dpif *dpif, void *state_, +dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_, struct dpif_port *dpif_port) { struct dpif_linux_port_state *state = state_; - struct dpif_linux_vport request, reply; - struct ofpbuf *buf; + struct dpif_linux_vport vport; + struct ofpbuf buf; int error; - ofpbuf_delete(state->buf); - state->buf = NULL; + if (!nl_dump_next(&state->dump, &buf)) { + return EOF; + } - dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_DUMP; - request.dp_idx = dpif_linux_cast(dpif)->minor; - request.port_no = state->next; - error = dpif_linux_vport_transact(&request, &reply, &buf); + error = dpif_linux_vport_from_ofpbuf(&vport, &buf); if (error) { - return error == ENODEV ? EOF : error; - } else { - dpif_port->name = (char *) reply.name; - dpif_port->type = (char *) netdev_vport_get_netdev_type(&reply); - dpif_port->port_no = reply.port_no; - state->buf = buf; - state->next = reply.port_no + 1; - return 0; + return error; } + + dpif_port->name = (char *) vport.name; + dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport); + dpif_port->port_no = vport.port_no; + return 0; } static int dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) { struct dpif_linux_port_state *state = state_; - ofpbuf_delete(state->buf); + int error = nl_dump_done(&state->dump); free(state); - return 0; + return error; } static int @@ -451,89 +550,269 @@ dpif_linux_port_poll_wait(const struct dpif *dpif_) } static int -dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n) +dpif_linux_flow_get(const struct dpif *dpif_, + const struct nlattr *key, size_t key_len, + struct ofpbuf **actionsp, struct dpif_flow_stats *stats) { - struct odp_flowvec fv; - fv.flows = flows; - fv.n_flows = n; - return do_ioctl(dpif_, ODP_FLOW_GET, &fv); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_flow request, reply; + struct ofpbuf *buf; + int error; + + dpif_linux_flow_init(&request); + request.cmd = ODP_FLOW_CMD_GET; + request.dp_idx = dpif->minor; + request.key = key; + request.key_len = key_len; + error = dpif_linux_flow_transact(&request, &reply, &buf); + if (!error) { + if (stats) { + dpif_linux_flow_get_stats(&reply, stats); + } + if (actionsp) { + buf->data = (void *) reply.actions; + buf->size = reply.actions_len; + *actionsp = buf; + } else { + ofpbuf_delete(buf); + } + } + return error; } static int -dpif_linux_flow_put(struct dpif *dpif_, struct odp_flow_put *put) +dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags, + const struct nlattr *key, size_t key_len, + const struct nlattr *actions, size_t actions_len, + struct dpif_flow_stats *stats) { - return do_ioctl(dpif_, ODP_FLOW_PUT, put); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_flow request, reply; + struct ofpbuf *buf; + int error; + + dpif_linux_flow_init(&request); + request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET; + request.dp_idx = dpif->minor; + request.key = key; + request.key_len = key_len; + request.actions = actions; + request.actions_len = actions_len; + if (flags & DPIF_FP_ZERO_STATS) { + request.clear = true; + } + request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; + error = dpif_linux_flow_transact(&request, + stats ? &reply : NULL, + stats ? &buf : NULL); + if (!error && stats) { + dpif_linux_flow_get_stats(&reply, stats); + ofpbuf_delete(buf); + } + return error; } static int -dpif_linux_flow_del(struct dpif *dpif_, struct odp_flow *flow) +dpif_linux_flow_del(struct dpif *dpif_, + const struct nlattr *key, size_t key_len, + struct dpif_flow_stats *stats) { - return do_ioctl(dpif_, ODP_FLOW_DEL, flow); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_flow request, reply; + struct ofpbuf *buf; + int error; + + dpif_linux_flow_init(&request); + request.cmd = ODP_FLOW_CMD_DEL; + request.dp_idx = dpif->minor; + request.key = key; + request.key_len = key_len; + error = dpif_linux_flow_transact(&request, + stats ? &reply : NULL, + stats ? &buf : NULL); + if (!error && stats) { + dpif_linux_flow_get_stats(&reply, stats); + ofpbuf_delete(buf); + } + return error; } +struct dpif_linux_flow_state { + struct nl_dump dump; + struct dpif_linux_flow flow; + struct dpif_flow_stats stats; +}; + static int -dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) +dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) { - *statep = xzalloc(sizeof(struct odp_flow_dump)); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_flow_state *state; + struct dpif_linux_flow request; + struct ofpbuf *buf; + + *statep = state = xmalloc(sizeof *state); + + dpif_linux_flow_init(&request); + request.cmd = ODP_DP_CMD_GET; + request.dp_idx = dpif->minor; + + buf = ofpbuf_new(1024); + dpif_linux_flow_to_ofpbuf(&request, buf); + nl_dump_start(&state->dump, genl_sock, buf); + ofpbuf_delete(buf); + return 0; } static int -dpif_linux_flow_dump_next(const struct dpif *dpif, void *state, - struct odp_flow *flow) +dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_, + const struct nlattr **key, size_t *key_len, + const struct nlattr **actions, size_t *actions_len, + const struct dpif_flow_stats **stats) { - struct odp_flow_dump *dump = state; + struct dpif_linux_flow_state *state = state_; + struct ofpbuf buf; int error; - dump->flow = flow; - error = do_ioctl(dpif, ODP_FLOW_DUMP, dump); - return error ? error : flow->flags & ODPFF_EOF ? EOF : 0; + if (!nl_dump_next(&state->dump, &buf)) { + return EOF; + } + + error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf); + if (!error) { + if (key) { + *key = state->flow.key; + *key_len = state->flow.key_len; + } + if (actions) { + *actions = state->flow.actions; + *actions_len = state->flow.actions_len; + } + if (stats) { + dpif_linux_flow_get_stats(&state->flow, &state->stats); + *stats = &state->stats; + } + } + return error; } static int -dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state) +dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) { + struct dpif_linux_flow_state *state = state_; + int error = nl_dump_done(&state->dump); free(state); - return 0; + return error; } static int dpif_linux_execute(struct dpif *dpif_, const struct nlattr *actions, size_t actions_len, - const struct ofpbuf *buf) + const struct ofpbuf *packet) { - struct odp_execute execute; - memset(&execute, 0, sizeof execute); - execute.actions = (struct nlattr *) actions; - execute.actions_len = actions_len; - execute.data = buf->data; - execute.length = buf->size; - return do_ioctl(dpif_, ODP_EXECUTE, &execute); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct odp_header *execute; + struct ofpbuf *buf; + int error; + + buf = ofpbuf_new(128 + actions_len + packet->size); + + nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST, + ODP_PACKET_CMD_EXECUTE, 1); + + execute = ofpbuf_put_uninit(buf, sizeof *execute); + execute->dp_idx = dpif->minor; + + nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size); + nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len); + + error = nl_sock_transact(genl_sock, buf, NULL); + ofpbuf_delete(buf); + return error; } static int dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) { - return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + *listen_mask = dpif->listen_mask; + return 0; } static int dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask) { - return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + int error; + int i; + + if (listen_mask == dpif->listen_mask) { + return 0; + } else if (!listen_mask) { + nl_sock_destroy(dpif->mc_sock); + dpif->mc_sock = NULL; + dpif->listen_mask = 0; + return 0; + } else if (!dpif->mc_sock) { + error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock); + if (error) { + return error; + } + } + + /* Unsubscribe from old groups. */ + for (i = 0; i < DPIF_N_UC_TYPES; i++) { + if (dpif->listen_mask & (1u << i)) { + nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]); + } + } + + /* Update listen_mask. */ + dpif->listen_mask = listen_mask; + + /* Subscribe to new groups. */ + error = 0; + for (i = 0; i < DPIF_N_UC_TYPES; i++) { + if (dpif->listen_mask & (1u << i)) { + int retval; + + retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]); + if (retval) { + error = retval; + } + } + } + return error; } static int dpif_linux_get_sflow_probability(const struct dpif *dpif_, uint32_t *probability) { - return do_ioctl(dpif_, ODP_GET_SFLOW_PROBABILITY, probability); + struct dpif_linux_dp dp; + struct ofpbuf *buf; + int error; + + error = dpif_linux_dp_get(dpif_, &dp, &buf); + if (!error) { + *probability = dp.sampling ? *dp.sampling : 0; + ofpbuf_delete(buf); + } + return error; } static int dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability) { - return do_ioctl(dpif_, ODP_SET_SFLOW_PROBABILITY, &probability); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp dp; + + dpif_linux_dp_init(&dp); + dp.cmd = ODP_DP_CMD_SET; + dp.dp_idx = dpif->minor; + dp.sampling = &probability; + return dpif_linux_dp_transact(&dp, NULL, NULL); } static int @@ -549,33 +828,52 @@ dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED, } static int -parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall) +parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, + uint32_t *dp_idx) { static const struct nl_policy odp_packet_policy[] = { /* Always present. */ - [ODP_PACKET_ATTR_TYPE] = { .type = NL_A_U32 }, [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC, .min_len = ETH_HEADER_LEN }, [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, - /* _ODPL_ACTION_NR only. */ + /* ODP_PACKET_CMD_ACTION only. */ [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true }, - /* _ODPL_SFLOW_NR only. */ + /* ODP_PACKET_CMD_SAMPLE only. */ [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true }, [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, }; - struct odp_packet *odp_packet = buf->data; + struct odp_header *odp_header; struct nlattr *a[ARRAY_SIZE(odp_packet_policy)]; + struct nlmsghdr *nlmsg; + struct genlmsghdr *genl; + struct ofpbuf b; + int type; + + ofpbuf_use_const(&b, buf->data, buf->size); + + nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); + genl = ofpbuf_try_pull(&b, sizeof *genl); + odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); + if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_packet_family + || !nl_policy_parse(&b, 0, odp_packet_policy, a, + ARRAY_SIZE(odp_packet_policy))) { + return EINVAL; + } - if (!nl_policy_parse(buf, sizeof *odp_packet, odp_packet_policy, - a, ARRAY_SIZE(odp_packet_policy))) { + type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS + : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION + : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE + : -1); + if (type < 0) { return EINVAL; } memset(upcall, 0, sizeof *upcall); - upcall->type = nl_attr_get_u32(a[ODP_PACKET_ATTR_TYPE]); + upcall->type = type; upcall->packet = buf; upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]); upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]); @@ -592,6 +890,8 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall) upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]); } + *dp_idx = odp_header->dp_idx; + return 0; } @@ -600,75 +900,60 @@ dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct ofpbuf *buf; - int retval; int error; + int i; - buf = ofpbuf_new(65536); - retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf)); - if (retval < 0) { - error = errno; - if (error != EAGAIN) { - VLOG_WARN_RL(&error_rl, "%s: read failed: %s", - dpif_name(dpif_), strerror(error)); + if (!dpif->mc_sock) { + return EAGAIN; + } + + for (i = 0; i < 50; i++) { + uint32_t dp_idx; + + error = nl_sock_recv(dpif->mc_sock, &buf, false); + if (error) { + return error; } - } else if (retval >= sizeof(struct odp_packet)) { - struct odp_packet *odp_packet = buf->data; - buf->size += retval; - if (odp_packet->len <= retval) { - error = parse_odp_packet(buf, upcall); - } else { - VLOG_WARN_RL(&error_rl, "%s: discarding message truncated " - "from %"PRIu32" bytes to %d", - dpif_name(dpif_), odp_packet->len, retval); - error = ERANGE; + error = parse_odp_packet(buf, upcall, &dp_idx); + if (!error + && dp_idx == dpif->minor + && dpif->listen_mask & (1u << upcall->type)) { + return 0; } - } else if (!retval) { - VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif_)); - error = EPROTO; - } else { - VLOG_WARN_RL(&error_rl, "%s: discarding too-short message (%d bytes)", - dpif_name(dpif_), retval); - error = ERANGE; - } - if (error) { ofpbuf_delete(buf); + if (error) { + return error; + } } - return error; + + return EAGAIN; } static void dpif_linux_recv_wait(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - poll_fd_wait(dpif->fd, POLLIN); + if (dpif->mc_sock) { + nl_sock_wait(dpif->mc_sock, POLLIN); + } } static void dpif_linux_recv_purge(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int i; - /* This is somewhat bogus because it assumes that the following macros have - * fixed values, but it's going to go away later. */ -#define DP_N_QUEUES 3 -#define DP_MAX_QUEUE_LEN 100 - for (i = 0; i < DP_N_QUEUES * DP_MAX_QUEUE_LEN; i++) { - /* Reading even 1 byte discards a whole datagram and saves time. */ - char buffer; - - if (read(dpif->fd, &buffer, 1) != 1) { - break; - } + if (dpif->mc_sock) { + nl_sock_drain(dpif->mc_sock); } } const struct dpif_class dpif_linux_class = { "system", - NULL, - NULL, + NULL, /* run */ + NULL, /* wait */ dpif_linux_enumerate, dpif_linux_open, dpif_linux_close, @@ -681,6 +966,7 @@ const struct dpif_class dpif_linux_class = { dpif_linux_port_del, dpif_linux_port_query_by_number, dpif_linux_port_query_by_name, + dpif_linux_get_max_ports, dpif_linux_port_dump_start, dpif_linux_port_dump_next, dpif_linux_port_dump_done, @@ -704,14 +990,35 @@ const struct dpif_class dpif_linux_class = { dpif_linux_recv_purge, }; -static int get_openvswitch_major(void); -static int get_major(const char *target); - static int -do_ioctl(const struct dpif *dpif_, int cmd, const void *arg) +dpif_linux_init(void) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - return ioctl(dpif->fd, cmd, arg) ? errno : 0; + static int error = -1; + + if (error < 0) { + error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY, + &odp_datapath_family); + if (error) { + VLOG_ERR("Generic Netlink family '%s' does not exist. " + "The Open vSwitch kernel module is probably not loaded.", + ODP_DATAPATH_FAMILY); + } + if (!error) { + error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family); + } + if (!error) { + error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family); + } + if (!error) { + error = nl_lookup_genl_family(ODP_PACKET_FAMILY, + &odp_packet_family); + } + if (!error) { + error = nl_sock_create(NETLINK_GENERIC, &genl_sock); + } + } + + return error; } bool @@ -732,162 +1039,6 @@ dpif_linux_is_internal_device(const char *name) return reply.type == ODP_VPORT_TYPE_INTERNAL; } -static int -make_openvswitch_device(int minor, char **fnp) -{ - const char dirname[] = "/dev/net"; - int major; - dev_t dev; - struct stat s; - char fn[128]; - - *fnp = NULL; - - major = get_openvswitch_major(); - if (major < 0) { - return -major; - } - dev = makedev(major, minor); - - sprintf(fn, "%s/dp%d", dirname, minor); - if (!stat(fn, &s)) { - if (!S_ISCHR(s.st_mode)) { - VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing", - fn); - } else if (s.st_rdev != dev) { - VLOG_WARN_RL(&error_rl, - "%s is device %u:%u but should be %u:%u, fixing", - fn, major(s.st_rdev), minor(s.st_rdev), - major(dev), minor(dev)); - } else { - goto success; - } - if (unlink(fn)) { - VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)", - fn, strerror(errno)); - return errno; - } - } else if (errno == ENOENT) { - if (stat(dirname, &s)) { - if (errno == ENOENT) { - if (mkdir(dirname, 0755)) { - VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)", - dirname, strerror(errno)); - return errno; - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", - dirname, strerror(errno)); - return errno; - } - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno)); - return errno; - } - - /* The device needs to be created. */ - if (mknod(fn, S_IFCHR | 0700, dev)) { - VLOG_WARN_RL(&error_rl, - "%s: creating character device %u:%u failed (%s)", - fn, major(dev), minor(dev), strerror(errno)); - return errno; - } - -success: - *fnp = xstrdup(fn); - return 0; -} - -/* Return the major device number of the Open vSwitch device. If it - * cannot be determined, a negative errno is returned. */ -static int -get_openvswitch_major(void) -{ - static int openvswitch_major = -1; - if (openvswitch_major < 0) { - openvswitch_major = get_major("openvswitch"); - } - return openvswitch_major; -} - -static int -get_major(const char *target) -{ - const char fn[] = "/proc/devices"; - char line[128]; - FILE *file; - int ln; - - file = fopen(fn, "r"); - if (!file) { - VLOG_ERR("opening %s failed (%s)", fn, strerror(errno)); - return -errno; - } - - for (ln = 1; fgets(line, sizeof line, file); ln++) { - char name[64]; - int major; - - if (!strncmp(line, "Character", 9) || line[0] == '\0') { - /* Nothing to do. */ - } else if (!strncmp(line, "Block", 5)) { - /* We only want character devices, so skip the rest of the file. */ - break; - } else if (sscanf(line, "%d %63s", &major, name)) { - if (!strcmp(name, target)) { - fclose(file); - return major; - } - } else { - VLOG_WARN_ONCE("%s:%d: syntax error", fn, ln); - } - } - - fclose(file); - - VLOG_ERR("%s: %s major not found (is the module loaded?)", fn, target); - return -ENODEV; -} - -static int -create_minor(const char *name, int minor) -{ - int error; - int fd; - - error = open_minor(minor, &fd); - if (error) { - return error; - } - - error = ioctl(fd, ODP_DP_CREATE, name) ? errno : 0; - close(fd); - return error; -} - -static int -open_minor(int minor, int *fdp) -{ - int error; - char *fn; - - error = make_openvswitch_device(minor, &fn); - if (error) { - return error; - } - - *fdp = open(fn, O_RDONLY | O_NONBLOCK); - if (*fdp < 0) { - error = errno; - VLOG_WARN("%s: open failed (%s)", fn, strerror(error)); - free(fn); - return error; - } - free(fn); - return 0; -} - static void dpif_linux_port_changed(const struct rtnetlink_link_change *change, void *dpif_) @@ -908,7 +1059,7 @@ dpif_linux_port_changed(const struct rtnetlink_link_change *change, } } -/* Parses the contents of 'buf', which contains a "struct odp_vport" followed +/* Parses the contents of 'buf', which contains a "struct odp_header" followed * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a * positive errno value. * @@ -936,18 +1087,27 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true }, }; - struct odp_vport *odp_vport; struct nlattr *a[ARRAY_SIZE(odp_vport_policy)]; + struct odp_header *odp_header; + struct nlmsghdr *nlmsg; + struct genlmsghdr *genl; + struct ofpbuf b; dpif_linux_vport_init(vport); - if (!nl_policy_parse(buf, sizeof *odp_vport, odp_vport_policy, - a, ARRAY_SIZE(odp_vport_policy))) { + ofpbuf_use_const(&b, buf->data, buf->size); + nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); + genl = ofpbuf_try_pull(&b, sizeof *genl); + odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); + if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_vport_family + || !nl_policy_parse(&b, 0, odp_vport_policy, a, + ARRAY_SIZE(odp_vport_policy))) { return EINVAL; } - odp_vport = buf->data; - vport->dp_idx = odp_vport->dp_idx; + vport->cmd = genl->cmd; + vport->dp_idx = odp_header->dp_idx; vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]); vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]); @@ -973,15 +1133,19 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_vport" +/* Appends to 'buf' (which must initially be empty) a "struct odp_header" * followed by Netlink attributes corresponding to 'vport'. */ static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, struct ofpbuf *buf) { - struct odp_vport *odp_vport; + struct odp_header *odp_header; - ofpbuf_reserve(buf, sizeof odp_vport); + nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO, + vport->cmd, 1); + + odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); + odp_header->dp_idx = vport->dp_idx; if (vport->port_no != UINT32_MAX) { nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no); @@ -1021,11 +1185,6 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, if (vport->iflink) { nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink); } - - odp_vport = ofpbuf_push_uninit(buf, sizeof *odp_vport); - odp_vport->dp_idx = vport->dp_idx; - odp_vport->len = buf->size; - odp_vport->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data; } /* Clears 'vport' to "empty" values. */ @@ -1048,46 +1207,25 @@ dpif_linux_vport_transact(const struct dpif_linux_vport *request, struct dpif_linux_vport *reply, struct ofpbuf **bufp) { - static int dp0_fd = -1; - struct ofpbuf *buf = NULL; + struct ofpbuf *request_buf; int error; assert((reply != NULL) == (bufp != NULL)); - if (dp0_fd < 0) { - int fd; - error = open_minor(0, &fd); - if (error) { - goto error; - } - dp0_fd = fd; - } - - buf = ofpbuf_new(1024); - dpif_linux_vport_to_ofpbuf(request, buf); - - error = ioctl(dp0_fd, request->cmd, buf->data) ? errno : 0; - if (error) { - goto error; - } + request_buf = ofpbuf_new(1024); + dpif_linux_vport_to_ofpbuf(request, request_buf); + error = nl_sock_transact(genl_sock, request_buf, bufp); + ofpbuf_delete(request_buf); - if (bufp) { - buf->size = ((struct odp_vport *) buf->data)->len; - error = dpif_linux_vport_from_ofpbuf(reply, buf); + if (reply) { + if (!error) { + error = dpif_linux_vport_from_ofpbuf(reply, *bufp); + } if (error) { - goto error; + dpif_linux_vport_init(reply); + ofpbuf_delete(*bufp); + *bufp = NULL; } - *bufp = buf; - } else { - ofpbuf_delete(buf); - } - return 0; - -error: - ofpbuf_delete(buf); - if (bufp) { - memset(reply, 0, sizeof *reply); - *bufp = NULL; } return error; } @@ -1102,9 +1240,341 @@ dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply, struct dpif_linux_vport request; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_GET; + request.cmd = ODP_VPORT_CMD_GET; request.name = name; return dpif_linux_vport_transact(&request, reply, bufp); } + +/* Parses the contents of 'buf', which contains a "struct odp_header" followed + * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a + * positive errno value. + * + * 'dp' will contain pointers into 'buf', so the caller should not free 'buf' + * while 'dp' is still in use. */ +static int +dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) +{ + static const struct nl_policy odp_datapath_policy[] = { + [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, + [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC, + .min_len = sizeof(struct odp_stats), + .max_len = sizeof(struct odp_stats), + .optional = true }, + [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true }, + [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true }, + [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true }, + }; + + struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)]; + struct odp_header *odp_header; + struct nlmsghdr *nlmsg; + struct genlmsghdr *genl; + struct ofpbuf b; + + dpif_linux_dp_init(dp); + + ofpbuf_use_const(&b, buf->data, buf->size); + nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); + genl = ofpbuf_try_pull(&b, sizeof *genl); + odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); + if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_datapath_family + || !nl_policy_parse(&b, 0, odp_datapath_policy, a, + ARRAY_SIZE(odp_datapath_policy))) { + return EINVAL; + } + + dp->cmd = genl->cmd; + dp->dp_idx = odp_header->dp_idx; + dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]); + if (a[ODP_DP_ATTR_STATS]) { + /* Can't use structure assignment because Netlink doesn't ensure + * sufficient alignment for 64-bit members. */ + memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]), + sizeof dp->stats); + } + if (a[ODP_DP_ATTR_IPV4_FRAGS]) { + dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); + } + if (a[ODP_DP_ATTR_SAMPLING]) { + dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]); + } + + if (a[ODP_DP_ATTR_MCGROUPS]) { + static const struct nl_policy odp_mcgroup_policy[] = { + [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true }, + [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true }, + [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true }, + }; + + struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)]; + + if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy, + mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) { + return EINVAL; + } + + if (mcgroups[ODP_PACKET_CMD_MISS]) { + dp->mcgroups[DPIF_UC_MISS] + = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]); + } + if (mcgroups[ODP_PACKET_CMD_ACTION]) { + dp->mcgroups[DPIF_UC_ACTION] + = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]); + } + if (mcgroups[ODP_PACKET_CMD_SAMPLE]) { + dp->mcgroups[DPIF_UC_SAMPLE] + = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]); + } + } + + return 0; +} + +/* Appends to 'buf' the Generic Netlink message described by 'dp'. */ +static void +dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) +{ + struct odp_header *odp_header; + + nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family, + NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1); + + odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); + odp_header->dp_idx = dp->dp_idx; + + if (dp->name) { + nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name); + } + + /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */ + + if (dp->ipv4_frags) { + nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags); + } + + if (dp->sampling) { + nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling); + } +} + +/* Clears 'dp' to "empty" values. */ +void +dpif_linux_dp_init(struct dpif_linux_dp *dp) +{ + memset(dp, 0, sizeof *dp); + dp->dp_idx = -1; +} + +static void +dpif_linux_dp_dump_start(struct nl_dump *dump) +{ + struct dpif_linux_dp request; + struct ofpbuf *buf; + + dpif_linux_dp_init(&request); + request.cmd = ODP_DP_CMD_GET; + + buf = ofpbuf_new(1024); + dpif_linux_dp_to_ofpbuf(&request, buf); + nl_dump_start(dump, genl_sock, buf); + ofpbuf_delete(buf); +} + +/* Executes 'request' in the kernel datapath. If the command fails, returns a + * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 + * without doing anything else. If 'reply' and 'bufp' are nonnull, then the + * result of the command is expected to be of the same form, which is decoded + * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the + * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ +int +dpif_linux_dp_transact(const struct dpif_linux_dp *request, + struct dpif_linux_dp *reply, struct ofpbuf **bufp) +{ + struct ofpbuf *request_buf; + int error; + + assert((reply != NULL) == (bufp != NULL)); + + request_buf = ofpbuf_new(1024); + dpif_linux_dp_to_ofpbuf(request, request_buf); + error = nl_sock_transact(genl_sock, request_buf, bufp); + ofpbuf_delete(request_buf); + + if (reply) { + if (!error) { + error = dpif_linux_dp_from_ofpbuf(reply, *bufp); + } + if (error) { + dpif_linux_dp_init(reply); + ofpbuf_delete(*bufp); + *bufp = NULL; + } + } + return error; +} + +/* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'. + * The caller must free '*bufp' when the reply is no longer needed ('reply' + * will contain pointers into '*bufp'). */ +int +dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, + struct ofpbuf **bufp) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp request; + + dpif_linux_dp_init(&request); + request.cmd = ODP_DP_CMD_GET; + request.dp_idx = dpif->minor; + + return dpif_linux_dp_transact(&request, reply, bufp); +} + +/* Parses the contents of 'buf', which contains a "struct odp_header" followed + * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a + * positive errno value. + * + * 'flow' will contain pointers into 'buf', so the caller should not free 'buf' + * while 'flow' is still in use. */ +static int +dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, + const struct ofpbuf *buf) +{ + static const struct nl_policy odp_flow_policy[] = { + [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED }, + [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, + [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC, + .min_len = sizeof(struct odp_flow_stats), + .max_len = sizeof(struct odp_flow_stats), + .optional = true }, + [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, + [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, + /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */ + }; + + struct nlattr *a[ARRAY_SIZE(odp_flow_policy)]; + struct odp_header *odp_header; + struct nlmsghdr *nlmsg; + struct genlmsghdr *genl; + struct ofpbuf b; + + dpif_linux_flow_init(flow); + + ofpbuf_use_const(&b, buf->data, buf->size); + nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); + genl = ofpbuf_try_pull(&b, sizeof *genl); + odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); + if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_flow_family + || !nl_policy_parse(&b, 0, odp_flow_policy, a, + ARRAY_SIZE(odp_flow_policy))) { + return EINVAL; + } + + flow->nlmsg_flags = nlmsg->nlmsg_flags; + flow->dp_idx = odp_header->dp_idx; + flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]); + flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]); + if (a[ODP_FLOW_ATTR_ACTIONS]) { + flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]); + flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]); + } + if (a[ODP_FLOW_ATTR_STATS]) { + flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]); + } + if (a[ODP_FLOW_ATTR_TCP_FLAGS]) { + flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]); + } + return 0; +} + +/* Appends to 'buf' (which must initially be empty) a "struct odp_header" + * followed by Netlink attributes corresponding to 'flow'. */ +static void +dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, + struct ofpbuf *buf) +{ + struct odp_header *odp_header; + + nl_msg_put_genlmsghdr(buf, 0, odp_flow_family, + NLM_F_REQUEST | flow->nlmsg_flags, flow->cmd, 1); + + odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); + odp_header->dp_idx = flow->dp_idx; + + if (flow->key_len) { + nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len); + } + + if (flow->actions_len) { + nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS, + flow->actions, flow->actions_len); + } + + /* We never need to send these to the kernel. */ + assert(!flow->stats); + assert(!flow->tcp_flags); + assert(!flow->used); + + if (flow->clear) { + nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR); + } +} + +/* Clears 'flow' to "empty" values. */ +void +dpif_linux_flow_init(struct dpif_linux_flow *flow) +{ + memset(flow, 0, sizeof *flow); +} + +/* Executes 'request' in the kernel datapath. If the command fails, returns a + * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 + * without doing anything else. If 'reply' and 'bufp' are nonnull, then the + * result of the command is expected to be a flow also, which is decoded and + * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply + * is no longer needed ('reply' will contain pointers into '*bufp'). */ +int +dpif_linux_flow_transact(const struct dpif_linux_flow *request, + struct dpif_linux_flow *reply, struct ofpbuf **bufp) +{ + struct ofpbuf *request_buf; + int error; + + assert((reply != NULL) == (bufp != NULL)); + + request_buf = ofpbuf_new(1024); + dpif_linux_flow_to_ofpbuf(request, request_buf); + error = nl_sock_transact(genl_sock, request_buf, bufp); + ofpbuf_delete(request_buf); + + if (reply) { + if (!error) { + error = dpif_linux_flow_from_ofpbuf(reply, *bufp); + } + if (error) { + dpif_linux_flow_init(reply); + ofpbuf_delete(*bufp); + *bufp = NULL; + } + } + return error; +} + +static void +dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow, + struct dpif_flow_stats *stats) +{ + if (flow->stats) { + stats->n_packets = get_unaligned_u64(&flow->stats->n_packets); + stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes); + } else { + stats->n_packets = 0; + stats->n_bytes = 0; + } + stats->used = flow->used ? get_unaligned_u64(flow->used) : 0; + stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0; +}