X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=5c2c755031d665246c84333cbc2ca36871dda94c;hb=856081f683d3e7d5b5fa07af4233d285eb205c47;hp=ec8a952b3d1611773ca410cb47edbad0f78cb8dd;hpb=5136364f41b74de00f86873a6f5be4c8a19cb5ad;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index ec8a952b3..5c2c75503 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -35,15 +35,20 @@ #include "dpif-provider.h" #include "netdev.h" +#include "netdev-vport.h" +#include "netlink.h" #include "ofpbuf.h" +#include "openvswitch/tunnel.h" +#include "packets.h" #include "poll-loop.h" #include "rtnetlink.h" +#include "rtnetlink-link.h" #include "shash.h" #include "svec.h" #include "util.h" #include "vlog.h" -VLOG_DEFINE_THIS_MODULE(dpif_linux) +VLOG_DEFINE_THIS_MODULE(dpif_linux); /* Datapath interface for the openvswitch Linux kernel module. */ struct dpif_linux { @@ -70,7 +75,7 @@ static int get_openvswitch_major(void); static int create_minor(const char *name, int minor, struct dpif **dpifp); static int open_minor(int minor, struct dpif **dpifp); static int make_openvswitch_device(int minor, char **fnp); -static void dpif_linux_port_changed(const struct rtnetlink_change *, +static void dpif_linux_port_changed(const struct rtnetlink_link_change *, void *dpif); static struct dpif_linux * @@ -112,8 +117,8 @@ dpif_linux_enumerate(struct svec *all_dps) } static int -dpif_linux_open(const char *name, const char *type OVS_UNUSED, bool create, - struct dpif **dpifp) +dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, + bool create, struct dpif **dpifp) { int minor; @@ -156,7 +161,7 @@ dpif_linux_open(const char *name, const char *type OVS_UNUSED, bool create, * getting the local port's name. */ memset(&port, 0, sizeof port); port.port = ODPP_LOCAL; - if (ioctl(dpif->fd, ODP_PORT_QUERY, &port)) { + if (ioctl(dpif->fd, ODP_VPORT_QUERY, &port)) { error = errno; if (error != ENODEV) { VLOG_WARN("%s: probe returned unexpected error: %s", @@ -175,7 +180,7 @@ static void dpif_linux_close(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - rtnetlink_notifier_unregister(&dpif->port_notifier); + rtnetlink_link_notifier_unregister(&dpif->port_notifier); shash_destroy(&dpif->changed_ports); free(dpif->local_ifname); close(dpif->fd); @@ -195,28 +200,6 @@ dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names) static int dpif_linux_destroy(struct dpif *dpif_) { - struct odp_port *ports; - size_t n_ports; - int err; - int i; - - err = dpif_port_list(dpif_, &ports, &n_ports); - if (err) { - return err; - } - - for (i = 0; i < n_ports; i++) { - if (ports[i].port != ODPP_LOCAL) { - err = do_ioctl(dpif_, ODP_VPORT_DEL, ports[i].devname); - if (err) { - VLOG_WARN_RL(&error_rl, "%s: error deleting port %s (%s)", - dpif_name(dpif_), ports[i].devname, strerror(err)); - } - } - } - - free(ports); - return do_ioctl(dpif_, ODP_DP_DESTROY, NULL); } @@ -247,66 +230,91 @@ dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int); } +static void +translate_vport_type_to_netdev_type(struct odp_port *port) +{ + char *type = port->type; + + if (!strcmp(type, "netdev")) { + ovs_strlcpy(type, "system", sizeof port->type); + } else if (!strcmp(type, "gre")) { + const struct tnl_port_config *config; + + config = (struct tnl_port_config *)port->config; + if (config->flags & TNL_F_IPSEC) { + ovs_strlcpy(type, "ipsec_gre", sizeof port->type); + } + } +} + +static void +translate_netdev_type_to_vport_type(struct odp_port *port) +{ + char *type = port->type; + + if (!strcmp(type, "system")) { + ovs_strlcpy(type, "netdev", sizeof port->type); + } else if (!strcmp(type, "ipsec_gre")) { + ovs_strlcpy(type, "gre", sizeof port->type); + } +} + static int -dpif_linux_port_add(struct dpif *dpif_, const char *devname, uint16_t flags, - uint16_t *port_no) +dpif_linux_port_add(struct dpif *dpif, struct netdev *netdev, + uint16_t *port_nop) { + const char *name = netdev_get_name(netdev); + const char *type = netdev_get_type(netdev); struct odp_port port; int error; memset(&port, 0, sizeof port); - strncpy(port.devname, devname, sizeof port.devname); - port.flags = flags; - error = do_ioctl(dpif_, ODP_PORT_ATTACH, &port); + strncpy(port.devname, name, sizeof port.devname); + strncpy(port.type, type, sizeof port.type); + netdev_vport_get_config(netdev, port.config); + translate_netdev_type_to_vport_type(&port); + + error = do_ioctl(dpif, ODP_VPORT_ATTACH, &port); if (!error) { - *port_no = port.port; + *port_nop = port.port; } + return error; } static int -dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) +dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no_) { - int tmp = port_no; - int err; - struct odp_port port; - - err = dpif_port_query_by_number(dpif_, port_no, &port); - if (err) { - return err; - } - - err = do_ioctl(dpif_, ODP_PORT_DETACH, &tmp); - if (err) { - return err; - } + int port_no = port_no_; /* Kernel expects an "int". */ + return do_ioctl(dpif_, ODP_VPORT_DETACH, &port_no); +} - if (!netdev_is_open(port.devname)) { - /* Try deleting the port if no one has it open. This shouldn't - * actually be necessary unless the config changed while we weren't - * running but it won't hurt anything if the port is already gone. */ - do_ioctl(dpif_, ODP_VPORT_DEL, port.devname); +static int +dpif_linux_port_query__(const struct dpif *dpif, struct odp_port *port) +{ + int error = do_ioctl(dpif, ODP_VPORT_QUERY, port); + if (!error) { + translate_vport_type_to_netdev_type(port); } - - return 0; + return error; } static int -dpif_linux_port_query_by_number(const struct dpif *dpif_, uint16_t port_no, - struct odp_port *port) +dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no, + struct odp_port *port) { memset(port, 0, sizeof *port); port->port = port_no; - return do_ioctl(dpif_, ODP_PORT_QUERY, port); + return dpif_linux_port_query__(dpif, port); } static int -dpif_linux_port_query_by_name(const struct dpif *dpif_, const char *devname, +dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname, struct odp_port *port) { memset(port, 0, sizeof *port); strncpy(port->devname, devname, sizeof port->devname); - return do_ioctl(dpif_, ODP_PORT_QUERY, port); + return dpif_linux_port_query__(dpif, port); } static int @@ -319,12 +327,22 @@ static int dpif_linux_port_list(const struct dpif *dpif_, struct odp_port *ports, int n) { struct odp_portvec pv; + unsigned int i; int error; pv.ports = ports; pv.n_ports = n; - error = do_ioctl(dpif_, ODP_PORT_LIST, &pv); - return error ? -error : pv.n_ports; + error = do_ioctl(dpif_, ODP_VPORT_LIST, &pv); + if (error) { + return -error; + } + + for (i = 0; i < pv.n_ports; i++) { + struct odp_port *port = &pv.ports[i]; + + translate_vport_type_to_netdev_type(port); + } + return pv.n_ports; } static int @@ -338,8 +356,7 @@ dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep) return ENOBUFS; } else if (!shash_is_empty(&dpif->changed_ports)) { struct shash_node *node = shash_first(&dpif->changed_ports); - *devnamep = xstrdup(node->name); - shash_delete(&dpif->changed_ports, node); + *devnamep = shash_steal(&dpif->changed_ports, node); return 0; } else { return EAGAIN; @@ -353,38 +370,10 @@ dpif_linux_port_poll_wait(const struct dpif *dpif_) if (!shash_is_empty(&dpif->changed_ports) || dpif->change_error) { poll_immediate_wake(); } else { - rtnetlink_notifier_wait(); + rtnetlink_link_notifier_wait(); } } -static int -dpif_linux_port_group_get(const struct dpif *dpif_, int group, - uint16_t ports[], int n) -{ - struct odp_port_group pg; - int error; - - assert(n <= UINT16_MAX); - pg.group = group; - pg.ports = ports; - pg.n_ports = n; - error = do_ioctl(dpif_, ODP_PORT_GROUP_GET, &pg); - return error ? -error : pg.n_ports; -} - -static int -dpif_linux_port_group_set(struct dpif *dpif_, int group, - const uint16_t ports[], int n) -{ - struct odp_port_group pg; - - assert(n <= UINT16_MAX); - pg.group = group; - pg.ports = (uint16_t *) ports; - pg.n_ports = n; - return do_ioctl(dpif_, ODP_PORT_GROUP_SET, &pg); -} - static int dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n) { @@ -407,27 +396,40 @@ dpif_linux_flow_del(struct dpif *dpif_, struct odp_flow *flow) } static int -dpif_linux_flow_list(const struct dpif *dpif_, struct odp_flow flows[], int n) +dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) { - struct odp_flowvec fv; + *statep = xzalloc(sizeof(struct odp_flow_dump)); + return 0; +} + +static int +dpif_linux_flow_dump_next(const struct dpif *dpif, void *state, + struct odp_flow *flow) +{ + struct odp_flow_dump *dump = state; int error; - fv.flows = flows; - fv.n_flows = n; - error = do_ioctl(dpif_, ODP_FLOW_LIST, &fv); - return error ? -error : fv.n_flows; + dump->flow = flow; + error = do_ioctl(dpif, ODP_FLOW_DUMP, dump); + return error ? error : flow->flags & ODPFF_EOF ? EOF : 0; +} + +static int +dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state) +{ + free(state); + return 0; } static int -dpif_linux_execute(struct dpif *dpif_, uint16_t in_port, - const union odp_action actions[], int n_actions, +dpif_linux_execute(struct dpif *dpif_, + const struct nlattr *actions, size_t actions_len, const struct ofpbuf *buf) { struct odp_execute execute; memset(&execute, 0, sizeof execute); - execute.in_port = in_port; - execute.actions = (union odp_action *) actions; - execute.n_actions = n_actions; + 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); @@ -471,14 +473,61 @@ dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED, } static int -dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp) +parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall) +{ + 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_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true }, + + /* _ODPL_SFLOW_NR 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 nlattr *a[ARRAY_SIZE(odp_packet_policy)]; + + if (!nl_policy_parse(buf, sizeof *odp_packet, odp_packet_policy, + a, ARRAY_SIZE(odp_packet_policy))) { + return EINVAL; + } + + memset(upcall, 0, sizeof *upcall); + upcall->type = nl_attr_get_u32(a[ODP_PACKET_ATTR_TYPE]); + upcall->packet = buf; + upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]); + upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]); + upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]); + upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]); + upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA] + ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA]) + : 0); + upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL] + ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL]) + : 0); + if (a[ODP_PACKET_ATTR_ACTIONS]) { + upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]); + upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]); + } + + return 0; +} + +static int +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; - buf = ofpbuf_new_with_headroom(65536, DPIF_RECV_MSG_PADDING); + buf = ofpbuf_new(65536); retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf)); if (retval < 0) { error = errno; @@ -486,30 +535,30 @@ dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp) VLOG_WARN_RL(&error_rl, "%s: read failed: %s", dpif_name(dpif_), strerror(error)); } - } else if (retval >= sizeof(struct odp_msg)) { - struct odp_msg *msg = buf->data; - if (msg->length <= retval) { - buf->size += retval; - *bufp = buf; - return 0; + } 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_), msg->length, retval); + dpif_name(dpif_), odp_packet->len, retval); error = ERANGE; } } 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)", + VLOG_WARN_RL(&error_rl, "%s: discarding too-short message (%d bytes)", dpif_name(dpif_), retval); error = ERANGE; } - *bufp = NULL; - ofpbuf_delete(buf); + if (error) { + ofpbuf_delete(buf); + } return error; } @@ -539,13 +588,13 @@ const struct dpif_class dpif_linux_class = { dpif_linux_port_list, dpif_linux_port_poll, dpif_linux_port_poll_wait, - dpif_linux_port_group_get, - dpif_linux_port_group_set, dpif_linux_flow_get, dpif_linux_flow_put, dpif_linux_flow_del, dpif_linux_flow_flush, - dpif_linux_flow_list, + dpif_linux_flow_dump_start, + dpif_linux_flow_dump_next, + dpif_linux_flow_dump_done, dpif_linux_execute, dpif_linux_recv_get_mask, dpif_linux_recv_set_mask, @@ -787,8 +836,9 @@ open_minor(int minor, struct dpif **dpifp) fd = open(fn, O_RDONLY | O_NONBLOCK); if (fd >= 0) { struct dpif_linux *dpif = xmalloc(sizeof *dpif); - error = rtnetlink_notifier_register(&dpif->port_notifier, - dpif_linux_port_changed, dpif); + error = rtnetlink_link_notifier_register(&dpif->port_notifier, + dpif_linux_port_changed, + dpif); if (!error) { char *name; @@ -816,7 +866,8 @@ open_minor(int minor, struct dpif **dpifp) } static void -dpif_linux_port_changed(const struct rtnetlink_change *change, void *dpif_) +dpif_linux_port_changed(const struct rtnetlink_link_change *change, + void *dpif_) { struct dpif_linux *dpif = dpif_;