X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=3c22b5567cc0a14eae659b02161e2e5248aeb835;hb=b3c01ed3308e7899e98e981bf465f74be86f5f12;hp=fe10952dcb726ff35b8863c5225a93dddf5f6e46;hpb=37a1300c3ca05a97160d44dc34298c447b50ed9d;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index fe10952dc..3c22b5567 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -45,6 +45,7 @@ #include "rtnetlink.h" #include "rtnetlink-link.h" #include "shash.h" +#include "sset.h" #include "svec.h" #include "unaligned.h" #include "util.h" @@ -57,7 +58,7 @@ struct dpif_linux_dp { uint8_t cmd; /* struct odp_header. */ - uint32_t dp_idx; + int dp_ifindex; /* Attributes. */ const char *name; /* ODP_DP_ATTR_NAME. */ @@ -83,13 +84,16 @@ struct dpif_linux_flow { /* struct odp_header. */ unsigned int nlmsg_flags; - uint32_t dp_idx; + int dp_ifindex; /* Attributes. * * The 'stats' and 'used' members point to 64-bit data that might only be * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to - * access their values. */ + * access their values. + * + * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in + * the Netlink version of the command, even if actions_len is zero. */ const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */ size_t key_len; const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */ @@ -114,19 +118,15 @@ static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *, /* Datapath interface for the openvswitch Linux kernel module. */ struct dpif_linux { struct dpif dpif; + int dp_ifindex; /* Multicast group messages. */ struct nl_sock *mc_sock; uint32_t mcgroups[DPIF_N_UC_TYPES]; unsigned int listen_mask; - /* Used by dpif_linux_get_all_names(). */ - char *local_ifname; - int minor; - /* Change notification. */ - int local_ifindex; /* Ifindex of local port. */ - struct shash changed_ports; /* Ports that have changed. */ + struct sset changed_ports; /* Ports that have changed. */ struct rtnetlink_notifier port_notifier; bool change_error; }; @@ -143,9 +143,7 @@ static int odp_packet_family; 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 open_dpif(const struct dpif_linux_dp *, struct dpif **); static void dpif_linux_port_changed(const struct rtnetlink_link_change *, void *dpif); @@ -188,10 +186,8 @@ static int dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, bool create, struct dpif **dpifp) { - 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(); @@ -199,49 +195,24 @@ dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, return error; } - minor = !strncmp(name, "dp", 2) - && isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1; - /* 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; + dp_request.name = name; error = dpif_linux_dp_transact(&dp_request, &dp, &buf); if (error) { return error; } - 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 (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(&dp, &vport, dpifp); - } + error = open_dpif(&dp, dpifp); ofpbuf_delete(buf); + return error; } static int -open_dpif(const struct dpif_linux_dp *dp, - const struct dpif_linux_vport *local_vport, struct dpif **dpifp) +open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp) { - int dp_idx = local_vport->dp_idx; struct dpif_linux *dpif; - char *name; int error; int i; @@ -252,19 +223,16 @@ open_dpif(const struct dpif_linux_dp *dp, goto error_free; } - name = xasprintf("dp%d", dp_idx); - dpif_init(&dpif->dpif, &dpif_linux_class, name, dp_idx, dp_idx); - free(name); + dpif_init(&dpif->dpif, &dpif_linux_class, dp->name, + dp->dp_ifindex, dp->dp_ifindex); dpif->mc_sock = NULL; for (i = 0; i < DPIF_N_UC_TYPES; i++) { dpif->mcgroups[i] = dp->mcgroups[i]; } dpif->listen_mask = 0; - dpif->local_ifname = xstrdup(local_vport->name); - dpif->local_ifindex = local_vport->ifindex; - dpif->minor = dp_idx; - shash_init(&dpif->changed_ports); + dpif->dp_ifindex = dp->dp_ifindex; + sset_init(&dpif->changed_ports); dpif->change_error = false; *dpifp = &dpif->dpif; @@ -280,21 +248,10 @@ dpif_linux_close(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); rtnetlink_link_notifier_unregister(&dpif->port_notifier); - shash_destroy(&dpif->changed_ports); - free(dpif->local_ifname); + sset_destroy(&dpif->changed_ports); free(dpif); } -static int -dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names) -{ - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - - svec_add_nocopy(all_names, xasprintf("dp%d", dpif->minor)); - svec_add(all_names, dpif->local_ifname); - return 0; -} - static int dpif_linux_destroy(struct dpif *dpif_) { @@ -303,7 +260,7 @@ dpif_linux_destroy(struct dpif *dpif_) dpif_linux_dp_init(&dp); dp.cmd = ODP_DP_CMD_DEL; - dp.dp_idx = dpif->minor; + dp.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&dp, NULL, NULL); } @@ -345,7 +302,7 @@ dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) dpif_linux_dp_init(&dp); dp.cmd = ODP_DP_CMD_SET; - dp.dp_idx = dpif->minor; + dp.dp_ifindex = dpif->dp_ifindex; dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO; return dpif_linux_dp_transact(&dp, NULL, NULL); } @@ -364,7 +321,7 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, dpif_linux_vport_init(&request); request.cmd = ODP_VPORT_CMD_NEW; - request.dp_idx = dpif->minor; + request.dp_ifindex = dpif->dp_ifindex; request.type = netdev_vport_get_vport_type(netdev); if (request.type == ODP_VPORT_TYPE_UNSPEC) { VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has " @@ -397,7 +354,7 @@ dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) dpif_linux_vport_init(&vport); vport.cmd = ODP_VPORT_CMD_DEL; - vport.dp_idx = dpif->minor; + vport.dp_ifindex = dpif->dp_ifindex; vport.port_no = port_no; return dpif_linux_vport_transact(&vport, NULL, NULL); } @@ -413,7 +370,7 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, dpif_linux_vport_init(&request); request.cmd = ODP_VPORT_CMD_GET; - request.dp_idx = dpif_linux_cast(dpif)->minor; + request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex; request.port_no = port_no; request.name = port_name; @@ -457,7 +414,7 @@ dpif_linux_flow_flush(struct dpif *dpif_) dpif_linux_flow_init(&flow); flow.cmd = ODP_FLOW_CMD_DEL; - flow.dp_idx = dpif->minor; + flow.dp_ifindex = dpif->dp_ifindex; return dpif_linux_flow_transact(&flow, NULL, NULL); } @@ -477,7 +434,7 @@ dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep) dpif_linux_vport_init(&request); request.cmd = ODP_DP_CMD_GET; - request.dp_idx = dpif->minor; + request.dp_ifindex = dpif->dp_ifindex; buf = ofpbuf_new(1024); dpif_linux_vport_to_ofpbuf(&request, buf); @@ -527,11 +484,10 @@ dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep) if (dpif->change_error) { dpif->change_error = false; - shash_clear(&dpif->changed_ports); + sset_clear(&dpif->changed_ports); return ENOBUFS; - } else if (!shash_is_empty(&dpif->changed_ports)) { - struct shash_node *node = shash_first(&dpif->changed_ports); - *devnamep = shash_steal(&dpif->changed_ports, node); + } else if (!sset_is_empty(&dpif->changed_ports)) { + *devnamep = sset_pop(&dpif->changed_ports); return 0; } else { return EAGAIN; @@ -542,7 +498,7 @@ static void dpif_linux_port_poll_wait(const struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (!shash_is_empty(&dpif->changed_ports) || dpif->change_error) { + if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) { poll_immediate_wake(); } else { rtnetlink_link_notifier_wait(); @@ -550,21 +506,31 @@ dpif_linux_port_poll_wait(const struct dpif *dpif_) } static int -dpif_linux_flow_get(const struct dpif *dpif_, - const struct nlattr *key, size_t key_len, - struct ofpbuf **actionsp, struct dpif_flow_stats *stats) +dpif_linux_flow_get__(const struct dpif *dpif_, + const struct nlattr *key, size_t key_len, + struct dpif_linux_flow *reply, struct ofpbuf **bufp) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_flow request, reply; - struct ofpbuf *buf; - int error; + struct dpif_linux_flow request; dpif_linux_flow_init(&request); request.cmd = ODP_FLOW_CMD_GET; - request.dp_idx = dpif->minor; + request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; - error = dpif_linux_flow_transact(&request, &reply, &buf); + return dpif_linux_flow_transact(&request, reply, bufp); +} + +static int +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 dpif_linux_flow reply; + struct ofpbuf *buf; + int error; + + error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf); if (!error) { if (stats) { dpif_linux_flow_get_stats(&reply, stats); @@ -588,15 +554,17 @@ dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags, { struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow request, reply; + struct nlattr dummy_action; struct ofpbuf *buf; int error; dpif_linux_flow_init(&request); request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET; - request.dp_idx = dpif->minor; + request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; - request.actions = actions; + /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */ + request.actions = actions ? actions : &dummy_action; request.actions_len = actions_len; if (flags & DPIF_FP_ZERO_STATS) { request.clear = true; @@ -624,7 +592,7 @@ dpif_linux_flow_del(struct dpif *dpif_, dpif_linux_flow_init(&request); request.cmd = ODP_FLOW_CMD_DEL; - request.dp_idx = dpif->minor; + request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; error = dpif_linux_flow_transact(&request, @@ -641,6 +609,7 @@ struct dpif_linux_flow_state { struct nl_dump dump; struct dpif_linux_flow flow; struct dpif_flow_stats stats; + struct ofpbuf *buf; }; static int @@ -655,13 +624,15 @@ dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) dpif_linux_flow_init(&request); request.cmd = ODP_DP_CMD_GET; - request.dp_idx = dpif->minor; + request.dp_ifindex = dpif->dp_ifindex; buf = ofpbuf_new(1024); dpif_linux_flow_to_ofpbuf(&request, buf); nl_dump_start(&state->dump, genl_sock, buf); ofpbuf_delete(buf); + state->buf = NULL; + return 0; } @@ -675,24 +646,42 @@ dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_, struct ofpbuf buf; int error; - if (!nl_dump_next(&state->dump, &buf)) { - return EOF; - } + do { + ofpbuf_delete(state->buf); + state->buf = NULL; - error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf); - if (!error) { - if (key) { - *key = state->flow.key; - *key_len = state->flow.key_len; + if (!nl_dump_next(&state->dump, &buf)) { + return EOF; } - if (actions) { - *actions = state->flow.actions; - *actions_len = state->flow.actions_len; + + error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf); + if (error) { + return error; } - if (stats) { - dpif_linux_flow_get_stats(&state->flow, &state->stats); - *stats = &state->stats; + + if (actions && !state->flow.actions) { + error = dpif_linux_flow_get__(dpif_, state->flow.key, + state->flow.key_len, + &state->flow, &state->buf); + if (error == ENOENT) { + VLOG_DBG("dumped flow disappeared on get"); + } else if (error) { + VLOG_WARN("error fetching dumped flow: %s", strerror(error)); + } } + } while (error); + + if (actions) { + *actions = state->flow.actions; + *actions_len = state->flow.actions_len; + } + if (key) { + *key = state->flow.key; + *key_len = state->flow.key_len; + } + if (stats) { + dpif_linux_flow_get_stats(&state->flow, &state->stats); + *stats = &state->stats; } return error; } @@ -702,6 +691,7 @@ 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); + ofpbuf_delete(state->buf); free(state); return error; } @@ -722,7 +712,7 @@ dpif_linux_execute(struct dpif *dpif_, ODP_PACKET_CMD_EXECUTE, 1); execute = ofpbuf_put_uninit(buf, sizeof *execute); - execute->dp_idx = dpif->minor; + execute->dp_ifindex = dpif->dp_ifindex; nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size); nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len); @@ -810,7 +800,7 @@ dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability) dpif_linux_dp_init(&dp); dp.cmd = ODP_DP_CMD_SET; - dp.dp_idx = dpif->minor; + dp.dp_ifindex = dpif->dp_ifindex; dp.sampling = &probability; return dpif_linux_dp_transact(&dp, NULL, NULL); } @@ -829,7 +819,7 @@ dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED, static int parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, - uint32_t *dp_idx) + int *dp_ifindex) { static const struct nl_policy odp_packet_policy[] = { /* Always present. */ @@ -890,7 +880,7 @@ 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; + *dp_ifindex = odp_header->dp_ifindex; return 0; } @@ -908,16 +898,16 @@ dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) } for (i = 0; i < 50; i++) { - uint32_t dp_idx; + int dp_ifindex; error = nl_sock_recv(dpif->mc_sock, &buf, false); if (error) { return error; } - error = parse_odp_packet(buf, upcall, &dp_idx); + error = parse_odp_packet(buf, upcall, &dp_ifindex); if (!error - && dp_idx == dpif->minor + && dp_ifindex == dpif->dp_ifindex && dpif->listen_mask & (1u << upcall->type)) { return 0; } @@ -957,7 +947,6 @@ const struct dpif_class dpif_linux_class = { dpif_linux_enumerate, dpif_linux_open, dpif_linux_close, - dpif_linux_get_all_names, dpif_linux_destroy, dpif_linux_get_stats, dpif_linux_get_drop_frags, @@ -1046,13 +1035,13 @@ dpif_linux_port_changed(const struct rtnetlink_link_change *change, struct dpif_linux *dpif = dpif_; if (change) { - if (change->master_ifindex == dpif->local_ifindex + if (change->master_ifindex == dpif->dp_ifindex && (change->nlmsg_type == RTM_NEWLINK || change->nlmsg_type == RTM_DELLINK)) { /* Our datapath changed, either adding a new port or deleting an * existing one. */ - shash_add_once(&dpif->changed_ports, change->ifname, NULL); + sset_add(&dpif->changed_ports, change->ifname); } } else { dpif->change_error = true; @@ -1107,7 +1096,7 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, } vport->cmd = genl->cmd; - vport->dp_idx = odp_header->dp_idx; + vport->dp_ifindex = odp_header->dp_ifindex; vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]); vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]); @@ -1119,6 +1108,8 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, } if (a[ODP_VPORT_ATTR_MTU]) { vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]); + } else { + vport->mtu = INT_MAX; } if (a[ODP_VPORT_ATTR_OPTIONS]) { vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]); @@ -1145,7 +1136,7 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, vport->cmd, 1); odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_idx = vport->dp_idx; + odp_header->dp_ifindex = vport->dp_ifindex; if (vport->port_no != UINT32_MAX) { nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no); @@ -1169,7 +1160,7 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, vport->address, ETH_ADDR_LEN); } - if (vport->mtu) { + if (vport->mtu && vport->mtu != INT_MAX) { nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu); } @@ -1192,7 +1183,6 @@ void dpif_linux_vport_init(struct dpif_linux_vport *vport) { memset(vport, 0, sizeof *vport); - vport->dp_idx = UINT32_MAX; vport->port_no = UINT32_MAX; } @@ -1286,7 +1276,7 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) } dp->cmd = genl->cmd; - dp->dp_idx = odp_header->dp_idx; + dp->dp_ifindex = odp_header->dp_ifindex; 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 @@ -1342,7 +1332,7 @@ dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) 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; + odp_header->dp_ifindex = dp->dp_ifindex; if (dp->name) { nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name); @@ -1364,7 +1354,6 @@ void dpif_linux_dp_init(struct dpif_linux_dp *dp) { memset(dp, 0, sizeof *dp); - dp->dp_idx = -1; } static void @@ -1427,7 +1416,7 @@ dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, dpif_linux_dp_init(&request); request.cmd = ODP_DP_CMD_GET; - request.dp_idx = dpif->minor; + request.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&request, reply, bufp); } @@ -1474,7 +1463,7 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, } flow->nlmsg_flags = nlmsg->nlmsg_flags; - flow->dp_idx = odp_header->dp_idx; + flow->dp_ifindex = odp_header->dp_ifindex; flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]); flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]); if (a[ODP_FLOW_ATTR_ACTIONS]) { @@ -1487,6 +1476,9 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, if (a[ODP_FLOW_ATTR_TCP_FLAGS]) { flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]); } + if (a[ODP_FLOW_ATTR_USED]) { + flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]); + } return 0; } @@ -1499,16 +1491,17 @@ dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, struct odp_header *odp_header; nl_msg_put_genlmsghdr(buf, 0, odp_flow_family, - NLM_F_REQUEST | flow->nlmsg_flags, flow->cmd, 1); + NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags, + flow->cmd, 1); odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); - odp_header->dp_idx = flow->dp_idx; + odp_header->dp_ifindex = flow->dp_ifindex; if (flow->key_len) { nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len); } - if (flow->actions_len) { + if (flow->actions || flow->actions_len) { nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS, flow->actions, flow->actions_len); }